forked from FT_DSP/FT_DSP.gitlink.net
add some operators
This commit is contained in:
ccsuzzh
parent
a793985319
commit
4617c83979
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
|
|
@ -0,0 +1,314 @@
|
|||
Activation
|
||||
=================
|
||||
|
||||
对输入的数组中每一个元素执行激活函数计算，激活函数可选，具体函数见以下说明。
|
||||
|
||||
- ``Relu`` - 标准Relu函数。
|
||||
|
||||
.. math::
|
||||
|
||||
output_i = \max(0, input_i)
|
||||
|
||||
- ``Relu6`` - 在标准Relu函数的基础上进行输出上限限制。
|
||||
|
||||
.. math::
|
||||
|
||||
output_i = \min(\max(0, input_i),6)
|
||||
|
||||
- ``Clip`` - 将输入裁剪到区间 [min_val, max_val]
|
||||
|
||||
.. math::
|
||||
|
||||
output_i = \min(\max(input_i, \text{min_val}), \text{max_val})
|
||||
|
||||
- ``LRelu`` - 带泄露的线性整流单元（Leaky Rectified Linear Unit），它在输入为正时保持线性，在输入为负时也保留一个很小的斜率，以避免标准 ReLU 中的“死亡神经元”问题。
|
||||
|
||||
.. math::
|
||||
|
||||
output_i =
|
||||
\begin{cases}
|
||||
input_i, & input_i \ge 0 \\
|
||||
\alpha \cdot input_i, & input_i < 0
|
||||
\end{cases}
|
||||
|
||||
- ``Sigmoid`` - 常用的平滑非线性激活函数（又称逻辑函数），可以将任意实数映射到区间 :math:`(0, 1)`，常用于二分类问题的输出层，表示概率意义的结果。
|
||||
|
||||
.. math::
|
||||
|
||||
output_i = \frac{1}{1 + e^{-input_i}}
|
||||
|
||||
- ``Tanh`` - 双曲正切激活函数（Hyperbolic Tangent），其输出范围为 :math:`(-1, 1)` 。
|
||||
|
||||
.. math::
|
||||
|
||||
output_i = \tanh(input_i) = \frac{e^{input_i} - e^{-input_i}}{e^{input_i} + e^{-input_i}}
|
||||
|
||||
- ``HSigmoid`` - 硬 Sigmoid 激活函数（Hard Sigmoid），是 ``Sigmoid`` 函数的近似形式，计算简单、效率更高。
|
||||
|
||||
.. math::
|
||||
|
||||
output_i = \text{clip}\left(\frac{input_i + 3}{6}, 0, 1\right)
|
||||
|
||||
其中 ``clip(a, 0, 1)`` 表示将 ``a`` 限制在区间 :math:`[0, 1]` 内。
|
||||
- ``Swish`` - 自门控（Self-Gated）激活函数，由 Google 提出，结合了 ``Sigmoid`` 与线性特性，具有平滑且非单调的特点。
|
||||
|
||||
.. math::
|
||||
|
||||
output_i = input_i \cdot \sigma(input_i) = \frac{input_i}{1 + e^{-input_i}}
|
||||
|
||||
其中 :math:`\sigma(x)` 为标准 ``Sigmoid`` 函数。``Swish`` 在深层网络中通常表现优于 ``ReLU``。
|
||||
- ``HSwish`` - 硬 Swish 激活函数（Hard Swish），是 ``Swish`` 函数的近似形式，计算简单且在移动端模型（如 MobileNetV3）中被广泛采用。
|
||||
|
||||
.. math::
|
||||
|
||||
output_i = input_i \cdot \text{clip}\left(\frac{input_i + 3}{6}, 0, 1\right)
|
||||
|
||||
其中 ``clip(a, 0, 1)`` 表示将 ``a`` 限制在区间 :math:`[0, 1]` 内。
|
||||
- ``HardTanh`` - 硬双曲正切激活函数（Hard Tanh），是 ``Tanh`` 函数的分段线性近似形式，计算简单、梯度稳定，常用于量化或轻量网络中。
|
||||
|
||||
.. math::
|
||||
|
||||
output_i = \text{clip}(input_i, min\_val, max\_val)
|
||||
|
||||
其中 ``clip(x, min_val, max_val)`` 表示当 :math:`x < min\_val` 时输出 min_val，当 :math:`x > max\_val` 时输出 max_val，否则输出 :math:`x` 本身。
|
||||
|
||||
- ``Gelu`` - 高斯误差线性单元（Gaussian Error Linear Unit），是一种平滑的非线性激活函数，结合了 ``ReLU`` 与概率特性。 该函数支持精确计算及非近似计算模式，近似算法由 *Hendrycks & Gimpel (2016)* 提出，用以替代精确形式 :math:`output_i=x\Phi(x)` ，计算速度更快且精度损失极小。
|
||||
|
||||
.. math::
|
||||
|
||||
\begin{aligned}
|
||||
output_i =
|
||||
\begin{cases}
|
||||
0.5\,input_i \Bigl[ 1 + \tanh\!\Bigl(
|
||||
\sqrt{\frac{2}{\pi}}\,(input_i + 0.044715\,input_i^3)
|
||||
\Bigr) \Bigr], & flag = true, \\[6pt]
|
||||
input_i \,\Phi(input_i)
|
||||
= \tfrac{1}{2}x \Bigl[
|
||||
1 + \mathrm{erf}\!\Bigl(\tfrac{input_i}{\sqrt{2}}\Bigr)
|
||||
\Bigr], & flag = false.
|
||||
\end{cases}
|
||||
\end{aligned}
|
||||
|
||||
|
||||
其中 :math:`\Phi(x)` 为标准正态分布的累积分布函数。
|
||||
- ``Softplus`` - ``ReLU`` 的平滑近似形式，能在零点处保持可导性。
|
||||
|
||||
.. math::
|
||||
|
||||
output_i =
|
||||
\begin{cases}
|
||||
input_i, & input_i \gt 88.0 \\
|
||||
\ln(1 + e^{input_i}), & \text{otherwise}
|
||||
\end{cases}
|
||||
|
||||
- ``Elu`` - 在输入为正时保持线性，在输入为负时呈指数衰减，可缓解 ReLU 的“死亡神经元”问题。
|
||||
|
||||
.. math::
|
||||
|
||||
output_i =
|
||||
\begin{cases}
|
||||
input_i, & input_i \ge 0 \\
|
||||
\alpha (e^{input_i} - 1), & input_i < 0
|
||||
\end{cases}
|
||||
|
||||
其中 :math:`\alpha` 为超参数，通常取 :math:`\alpha = 1.0`。
|
||||
- ``Celu`` - 连续指数线性单元（Continuously Differentiable ELU），是 ``ELU`` 的改进版本，保证在零点处连续可导。
|
||||
|
||||
.. math::
|
||||
output_i =
|
||||
\begin{cases}
|
||||
input_i, & input_i \ge 0 \\
|
||||
\alpha (e^{\frac{input_i}{alpha}} - 1), & input_i < 0
|
||||
\end{cases}
|
||||
|
||||
其中 :math:`\alpha` 为可调超参数，用于控制负区间的平滑程度。
|
||||
- ``HardShrink`` - 硬收缩激活函数（Hard Shrinkage），用于稀疏化输出。
|
||||
|
||||
.. math::
|
||||
|
||||
output_i =
|
||||
\begin{cases}
|
||||
input_i, & \text{if } |input_i| > \lambda \\
|
||||
0, & \text{otherwise}
|
||||
\end{cases}
|
||||
|
||||
其中 :math:`\lambda` 为阈值常数。
|
||||
|
||||
- ``SoftShrink`` - 软收缩激活函数（Soft Shrinkage），与 ``HardShrink`` 类似，但收缩过程更加平滑。
|
||||
|
||||
.. math::
|
||||
|
||||
output_i =
|
||||
\begin{cases}
|
||||
input_i - \lambda, & \text{if } input_i > \lambda \\
|
||||
input_i + \lambda, & \text{if } input_i < -\lambda \\
|
||||
0, & \text{otherwise}
|
||||
\end{cases}
|
||||
|
||||
- ``SoftsignOpt`` - 优化的软符号函数（Optimized Softsign），是一种平滑的压缩函数，用于将输入映射到有限区间。
|
||||
|
||||
.. math::
|
||||
|
||||
output_i = \frac{input_i}{1 + |input_i|}
|
||||
|
||||
输入：
|
||||
- **Input0** - 输入数据地址。
|
||||
- **length** - 数组长度。
|
||||
- **args(部分激活函数)** - 激活函数计算参数（仅适用于部分函数）。
|
||||
- **core_mask(int, 可选)** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **output** - 计算结果地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持int8, fp32
|
||||
- MT7004 支持fp16, fp32
|
||||
|
||||
**共享存储版本：**
|
||||
|
||||
.. c:function:: void i8_relu_s(int8_t* Input0, int8_t* output,int length, int core_mask)
|
||||
.. c:function:: void fp_relu_s(float* Input0, float* output,int length, int core_mask)
|
||||
.. c:function:: void hp_relu_s(half* Input0, half* output,int length, int core_mask)
|
||||
.. c:function:: void i8_relu6_s(int8_t* Input0, int8_t* output,int length, int core_mask)
|
||||
.. c:function:: void fp_relu6_s(float* Input0, float* output,int length, int core_mask)
|
||||
.. c:function:: void hp_relu6_s(half* Input0, half* output,int length, int core_mask)
|
||||
.. c:function:: void i8_clip_s(int8_t* Input0, int8_t* output,int length, int8_t min_val, int8_t max_val, int core_mask)
|
||||
.. c:function:: void fp_clip_s(float* Input0, float* output,int length, float min_val, float max_val, int core_mask)
|
||||
.. c:function:: void hp_clip_s(half* Input0, half* output,int length, half min_val, half max_val, int core_mask)
|
||||
.. c:function:: void i8_lrelu_s(int8_t* Input0, int8_t* output,int length, float alpha, int core_mask)
|
||||
.. c:function:: void fp_lrelu_s(float* Input0, float* output,int length, float alpha, int core_mask)
|
||||
.. c:function:: void hp_lrelu_s(half* Input0, half* output,int length, half alpha, int core_mask)
|
||||
.. c:function:: void i8_sigmoid_s(int8_t* Input0, float* output,int length, int core_mask)
|
||||
.. c:function:: void fp_sigmoid_s(float* Input0, float* output,int length, int core_mask)
|
||||
.. c:function:: void hp_sigmoid_s(half* Input0, half* output,int length, int core_mask)
|
||||
.. c:function:: void i8_tanh_s(int8_t* Input0, float* output,int length, int core_mask)
|
||||
.. c:function:: void fp_tanh_s(float* Input0, float* output,int length, int core_mask)
|
||||
.. c:function:: void hp_tanh_s(half* Input0, half* output,int length, int core_mask)
|
||||
.. c:function:: void i8_hsigmoid_s(int8_t* Input0, float* output,int length, int core_mask)
|
||||
.. c:function:: void fp_hsigmoid_s(float* Input0, float* output,int length, int core_mask)
|
||||
.. c:function:: void hp_hsigmoid_s(half* Input0, half* output,int length, int core_mask)
|
||||
.. c:function:: void i8_swish_s(int8_t* Input0, float* output,int length, int core_mask)
|
||||
.. c:function:: void fp_swish_s(float* Input0, float* output,int length, int core_mask)
|
||||
.. c:function:: void hp_swish_s(half* Input0, half* output,int length, int core_mask)
|
||||
.. c:function:: void i8_hswish_s(int8_t* Input0, float* output,int length, int core_mask)
|
||||
.. c:function:: void fp_hswish_s(float* Input0, float* output,int length, int core_mask)
|
||||
.. c:function:: void hp_hswish_s(half* Input0, half* output,int length, int core_mask)
|
||||
.. c:function:: void i8_hardtanh_s(int8_t* Input0, int8_t* output,int length, int8_t min_val, int8_t max_val, int core_mask)
|
||||
.. c:function:: void fp_hardtanh_s(float* Input0, float* output,int length, float min_val, float max_val, int core_mask)
|
||||
.. c:function:: void hp_hardtanh_s(half* Input0, half* output,int length, half min_val, half max_val, int core_mask)
|
||||
.. c:function:: void i8_gelu_s(int8_t* Input0, float* output,int length, int approximate, int core_mask)
|
||||
.. c:function:: void fp_gelu_s(float* Input0, float* output,int length, int approximate, int core_mask)
|
||||
.. c:function:: void hp_gelu_s(half* Input0, half* output,int length, int approximate, int core_mask)
|
||||
.. c:function:: void i8_softplus_s(int8_t* Input0, float* output,int length, int core_mask)
|
||||
.. c:function:: void fp_softplus_s(float* Input0, float* output,int length, int core_mask)
|
||||
.. c:function:: void hp_softplus_s(half* Input0, half* output,int length, int core_mask)
|
||||
.. c:function:: void i8_elu_s(int8_t* Input0, float* output,int length, float alpha, int core_mask)
|
||||
.. c:function:: void fp_elu_s(float* Input0, float* output,int length, float alpha, int core_mask)
|
||||
.. c:function:: void hp_elu_s(half* Input0, half* output,int length, half alpha, int core_mask)
|
||||
.. c:function:: void i8_celu_s(int8_t* Input0, float* output,int length, float alpha, int core_mask)
|
||||
.. c:function:: void fp_celu_s(float* Input0, float* output,int length, float alpha, int core_mask)
|
||||
.. c:function:: void hp_celu_s(half* Input0, half* output,int length, half alpha, int core_mask)
|
||||
.. c:function:: void i8_hardshrink_s(int8_t* Input0, int8_t* output,int length, int8_t lambd, int core_mask)
|
||||
.. c:function:: void fp_hardshrink_s(float* Input0, float* output,int length, float lambd, int core_mask)
|
||||
.. c:function:: void hp_hardshrink_s(half* Input0, half* output,int length, half lambd, int core_mask)
|
||||
.. c:function:: void i8_softshrink_s(int8_t* Input0, int8_t* output,int length, int8_t lambd, int core_mask)
|
||||
.. c:function:: void fp_softshrink_s(float* Input0, float* output,int length, float lambd, int core_mask)
|
||||
.. c:function:: void hp_softshrink_s(half* Input0, half* output,int length, half lambd, int core_mask)
|
||||
.. c:function:: void i8_softsignopt_s(int8_t* Input0, float* output,int length, int core_mask)
|
||||
.. c:function:: void fp_softsignopt_s(float* Input0, float* output,int length, int core_mask)
|
||||
.. c:function:: void hp_softsignopt_s(half* Input0, half* output,int length, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 10
|
||||
|
||||
//FT78NE示例
|
||||
#include <stdio.h>
|
||||
#include <activation.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
float *input0 = (float *)0xA0000000; //input在DDR空间
|
||||
float *output = (float *)0xC0000000;
|
||||
int length = 1000;
|
||||
int core_mask = 0xff;
|
||||
fp_tanh_s(input0, output, length, core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
**私有存储版本：**
|
||||
|
||||
.. c:function:: void i8_relu_p(int8_t* Input0, int8_t* output,int length)
|
||||
.. c:function:: void fp_relu_p(float* Input0, float* output,int length)
|
||||
.. c:function:: void hp_relu_p(half* Input0, half* output,int length)
|
||||
.. c:function:: void i8_relu6_p(int8_t* Input0, int8_t* output,int length)
|
||||
.. c:function:: void fp_relu6_p(float* Input0, float* output,int length)
|
||||
.. c:function:: void hp_relu6_p(half* Input0, half* output,int length)
|
||||
.. c:function:: void i8_clip_p(int8_t* Input0, int8_t* output,int length, int8_t min_val, int8_t max_val)
|
||||
.. c:function:: void fp_clip_p(float* Input0, float* output,int length, float min_val, float max_val)
|
||||
.. c:function:: void hp_clip_p(half* Input0, half* output,int length, half min_val, half max_val)
|
||||
.. c:function:: void i8_lrelu_p(int8_t* Input0, int8_t* output,int length, float alpha)
|
||||
.. c:function:: void fp_lrelu_p(float* Input0, float* output,int length, float alpha)
|
||||
.. c:function:: void hp_lrelu_p(half* Input0, half* output,int length, half alpha)
|
||||
.. c:function:: void i8_sigmoid_p(int8_t* Input0, float* output,int length)
|
||||
.. c:function:: void fp_sigmoid_p(float* Input0, float* output,int length)
|
||||
.. c:function:: void hp_sigmoid_p(half* Input0, half* output,int length)
|
||||
.. c:function:: void i8_tanh_p(int8_t* Input0, float* output,int length)
|
||||
.. c:function:: void fp_tanh_p(float* Input0, float* output,int length)
|
||||
.. c:function:: void hp_tanh_p(half* Input0, half* output,int length)
|
||||
.. c:function:: void i8_hsigmoid_p(int8_t* Input0, float* output,int length)
|
||||
.. c:function:: void fp_hsigmoid_p(float* Input0, float* output,int length)
|
||||
.. c:function:: void hp_hsigmoid_p(half* Input0, half* output,int length)
|
||||
.. c:function:: void i8_swish_p(int8_t* Input0, float* output,int length)
|
||||
.. c:function:: void fp_swish_p(float* Input0, float* output,int length)
|
||||
.. c:function:: void hp_swish_p(half* Input0, half* output,int length)
|
||||
.. c:function:: void i8_hswish_p(int8_t* Input0, float* output,int length)
|
||||
.. c:function:: void fp_hswish_p(float* Input0, float* output,int length)
|
||||
.. c:function:: void hp_hswish_p(half* Input0, half* output,int length)
|
||||
.. c:function:: void i8_hardtanh_p(int8_t* Input0, int8_t* output,int length, int8_t min_val, int8_t max_val)
|
||||
.. c:function:: void fp_hardtanh_p(float* Input0, float* output,int length, float min_val, float max_val)
|
||||
.. c:function:: void hp_hardtanh_p(half* Input0, half* output,int length, half min_val, half max_val)
|
||||
.. c:function:: void i8_gelu_p(int8_t* Input0, float* output,int length, int approximate)
|
||||
.. c:function:: void fp_gelu_p(float* Input0, float* output,int length, int approximate)
|
||||
.. c:function:: void hp_gelu_p(half* Input0, half* output,int length, int approximate)
|
||||
.. c:function:: void i8_softplus_p(int8_t* Input0, float* output,int length)
|
||||
.. c:function:: void fp_softplus_p(float* Input0, float* output,int length)
|
||||
.. c:function:: void hp_softplus_p(half* Input0, half* output,int length)
|
||||
.. c:function:: void i8_elu_p(int8_t* Input0, float* output,int length, float alpha)
|
||||
.. c:function:: void fp_elu_p(float* Input0, float* output,int length, float alpha)
|
||||
.. c:function:: void hp_elu_p(half* Input0, half* output,int length, half alpha)
|
||||
.. c:function:: void i8_celu_p(int8_t* Input0, float* output,int length, float alpha)
|
||||
.. c:function:: void fp_celu_p(float* Input0, float* output,int length, float alpha)
|
||||
.. c:function:: void hp_celu_p(half* Input0, half* output,int length, half alpha)
|
||||
.. c:function:: void i8_hardshrink_p(int8_t* Input0, int8_t* output,int length, int8_t lambd)
|
||||
.. c:function:: void fp_hardshrink_p(float* Input0, float* output,int length, float lambd)
|
||||
.. c:function:: void hp_hardshrink_p(half* Input0, half* output,int length, half lambd)
|
||||
.. c:function:: void i8_softshrink_p(int8_t* Input0, int8_t* output,int length, int8_t lambd)
|
||||
.. c:function:: void fp_softshrink_p(float* Input0, float* output,int length, float lambd)
|
||||
.. c:function:: void hp_softshrink_p(half* Input0, half* output,int length, half lambd)
|
||||
.. c:function:: void i8_softsignopt_p(int8_t* Input0, float* output,int length)
|
||||
.. c:function:: void fp_softsignopt_p(float* Input0, float* output,int length)
|
||||
.. c:function:: void hp_softsignopt_p(half* Input0, half* output,int length)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 9
|
||||
|
||||
//FT78NE示例
|
||||
#include <stdio.h>
|
||||
#include <activation.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
float *input0 = (float *)0x10000000; //input在DDR空间
|
||||
float *output = (float *)0x10004000;
|
||||
int length = 1000;
|
||||
fp_tanh_p(input0, output, length);
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,108 @@
|
|||
AdamWeightDecay
|
||||
=================
|
||||
对权重张量执行 Adam Weight Decay 优化更新。
|
||||
|
||||
.. math::
|
||||
|
||||
\begin{aligned}
|
||||
m_t &= \beta_1 \cdot m_{t-1} + (1 - \beta_1) \cdot g_t \\
|
||||
v_t &= \beta_2 \cdot v_{t-1} + (1 - \beta_2) \cdot g_t^2 \\
|
||||
\hat{m}_t &= \frac{m_t}{\sqrt{v_t} + \epsilon} \\
|
||||
var_t &= var_{t-1} - lr \cdot (\hat{m}_t + decay \cdot var_{t-1})
|
||||
\end{aligned}
|
||||
|
||||
|
||||
输入：
|
||||
- **var** - 待更新权重张量首地址。
|
||||
- **m** - 一阶动量张量首地址。
|
||||
- **v** - 二阶动量张量首地址。
|
||||
- **gradient** - 梯度张量首地址。
|
||||
- **lr** - 学习率。
|
||||
- **beta1** - 一阶动量衰减系数。
|
||||
- **beta2** - 二阶动量衰减系数。
|
||||
- **epsilon** - 数值稳定项。
|
||||
- **decay** - 权重衰减系数。
|
||||
- **start** - 参与计算的起始索引（闭区间）。
|
||||
- **end** - 参与计算的结束索引（开区间）。
|
||||
- **core_mask(int, 可选)** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **var** - 原地写回更新后的权重张量。
|
||||
- **m** - 原地写回更新后的一阶动量张量。
|
||||
- **v** - 原地写回更新后的二阶动量张量。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持 fp32 数据类型。
|
||||
- MT7004 支持 fp16、fp32 数据类型。
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void hp_adamweightdecay_s(half *var, half *m, half *v, const half *gradient, float lr, float beta1, float beta2, float epsilon, float decay, int start, int end, int core_mask)
|
||||
.. c:function:: void fp_adamweightdecay_s(float *var, float *m, float *v, const float *gradient, float lr, float beta1, float beta2, float epsilon, float decay, int start, int end, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 17
|
||||
|
||||
// FT78NE 多核示例
|
||||
#include <stdio.h>
|
||||
|
||||
int main(void) {
|
||||
float *var = (float *)0xA0000000; // DDR 存储
|
||||
float *m = (float *)0xB0000000;
|
||||
float *v = (float *)0xC0000000;
|
||||
float *gradient = (float *)0xD0000000;
|
||||
int start = 0;
|
||||
int end = 4096;
|
||||
int core_mask = 0xff;
|
||||
float lr = 1e-3f;
|
||||
float beta1 = 0.9f;
|
||||
float beta2 = 0.999f;
|
||||
float epsilon = 1e-8f;
|
||||
float decay = 1e-2f;
|
||||
fp_adamweightdecay_s(var, m, v, gradient, lr,
|
||||
beta1, beta2, epsilon, decay,
|
||||
start, end, core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void hp_adamweightdecay_p(half *var, half *m, half *v, const half *gradient, float lr, float beta1, float beta2, float epsilon, float decay, int length)
|
||||
.. c:function:: void fp_adamweightdecay_p(float *var, float *m, float *v, const float *gradient, float lr, float beta1, float beta2, float epsilon, float decay, int length)
|
||||
|
||||
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 15
|
||||
|
||||
// MT7004 单核示例
|
||||
#include <stdio.h>
|
||||
|
||||
int main(void) {
|
||||
half *var = (half *)0x10000000; // L2 存储
|
||||
half *m = (half *)0x10002000;
|
||||
half *v = (half *)0x10004000;
|
||||
half *gradient = (half *)0x10006000;
|
||||
int length = 2048;
|
||||
float lr = 5e-4f;
|
||||
float beta1 = 0.9f;
|
||||
float beta2 = 0.999f;
|
||||
float epsilon = 1e-6f;
|
||||
float decay = 5e-3f;
|
||||
hp_adamweightdecay_p(var, m, v, gradient, lr,
|
||||
beta1, beta2, epsilon, decay,
|
||||
length);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
|
@ -0,0 +1,170 @@
|
|||
Adder
|
||||
=================
|
||||
|
||||
Adder是一种卷积替代算子，它使用L1距离度量（绝对差的和）代替传统卷积中的点积操作。与标准卷积不同，Adder通过计算特征与卷积核之间的绝对差的负和来进行特征提取。假定输入X，filter表示为F，它按以下公式计算：
|
||||
|
||||
.. math::
|
||||
|
||||
Y(m,n,t) = - \sum_{i=0}^{d} \sum_{j=0}^{d} \sum_{k=0}^{C_{in}} |X(m+i, n+j, k) - F(i,j,k,t)|
|
||||
|
||||
输入：
|
||||
- **input_x** - 输入数据的地址
|
||||
- **input_w** - 输入卷积核权重的地址
|
||||
- **bias** - 输入偏置的地址
|
||||
- **param** - 算子计算所需参数的结构体。其各成员见下述。
|
||||
- **core_mask** - 核掩码。
|
||||
|
||||
**AdderParameter定义：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
|
||||
typedef struct AdderParameter {
|
||||
void* workspace_; // 用于存放中间计算结果
|
||||
int output_batch_; // 输出数据总批次
|
||||
int input_batch_; // 输入数据总批次
|
||||
int input_h_; // 输入数据h维度大小
|
||||
int input_w_; // 输入数据w维度大小
|
||||
int output_h_; // 输出数据h维度大小
|
||||
int output_w_; // 输出数据w维度大小
|
||||
int input_channel_; // 输入数据通道数
|
||||
int output_channel_; // 输出数据通道数
|
||||
int kernel_h_; // 卷积核h维度大小
|
||||
int kernel_w_; // 卷积核w维度大小
|
||||
int group_; // 组数
|
||||
int pad_l_; // 左填充大小
|
||||
int pad_u_; // 上填充大小
|
||||
int dilation_h_; // 卷积核h维度膨胀尺寸大小
|
||||
int dilation_w_; // 卷积核w维度膨胀尺寸大小
|
||||
int stride_h_; // 卷积核h维度步长
|
||||
int stride_w_; // 卷积核w维度步长
|
||||
int buffer_size_; // 为分块计算所分配的缓存大小
|
||||
} AdderParameter;
|
||||
|
||||
输出：
|
||||
- **out_y** - 输出地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持int8, fp32
|
||||
- MT7004 支持fp16, fp32
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void i8_adder_s(int8_t* input_x, int8_t* input_w, int8_t* out_y, int* bias, AdderParameter *param, int core_mask)
|
||||
.. c:function:: void hp_adder_s(half* input_x, half* input_w, half* out_y, half* bias, AdderParameter *param, int core_mask)
|
||||
.. c:function:: void fp_adder_s(float* input_x, float* input_w, float* out_y, float* bias, AdderParameter *param, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 32
|
||||
|
||||
void TestAdderSMCFp32(int* input_shape, int* weight_shape, int* output_shape, int* stride, int* padding, int* dilation, int groups, float* bias, int core_mask) {
|
||||
int core_id = get_core_id();
|
||||
int logic_core_id = GetLogicCoreId(core_mask, core_id);
|
||||
int core_num = GetCoreNum(core_mask);
|
||||
float* input_data = (float*)0x88000000;
|
||||
float* weight = (float*)0x89000000;
|
||||
float* output_data = (float*)0x90000000;
|
||||
float* bias_data = (float*)0x91000000;
|
||||
AdderParameter* param = (AdderParameter*)0x92000000;
|
||||
if (logic_core_id == 0) {
|
||||
memcpy(bias_data, bias, sizeof(float) * output_shape[3]);
|
||||
param->dilation_h_ = dilation[0];
|
||||
param->dilation_w_ = dilation[1];
|
||||
param->group_ = groups;
|
||||
param->input_batch_ = input_shape[0];
|
||||
param->input_h_ = input_shape[1];
|
||||
param->input_w_ = input_shape[2];
|
||||
param->input_channel_ = input_shape[3];
|
||||
param->kernel_h_ = weight_shape[1];
|
||||
param->kernel_w_ = weight_shape[2];
|
||||
param->output_batch_ = output_shape[0];
|
||||
param->output_h_ = output_shape[1];
|
||||
param->output_w_ = output_shape[2];
|
||||
param->output_channel_ = output_shape[3];
|
||||
param->stride_h_ = stride[0];
|
||||
param->stride_w_ = stride[0];
|
||||
param->pad_u_ = padding[0];
|
||||
param->pad_l_ = padding[2];
|
||||
param->workspace_ = (float*)0x10000000; // workspace空间需分配在AM内，计算过程中会将数据搬运到workspace空间内进行计算
|
||||
}
|
||||
sys_bar(0, core_num); // 初始化参数完成后进行同步
|
||||
fp_adder_s(input_data, weight, output_data, bias_data, param, core_mask);
|
||||
}
|
||||
|
||||
void main(){
|
||||
int in_channel = 4;
|
||||
int out_channel = 4;
|
||||
int groups = 4;
|
||||
int input_shape[4] = {1, 30, 30, in_channel}; // NHWC
|
||||
int weight_shape[4] = {out_channel, 3, 3, in_channel / groups};
|
||||
int output_shape[4] = {1, 10, 10, out_channel}; // NHWC
|
||||
int stride[2] = {2, 2};
|
||||
int padding[4] = {1, 1, 1, 1};
|
||||
int dilation[2]= {2, 2};
|
||||
float bias[4] = {0, 0, 0, 0};
|
||||
int core_mask = 0b1111;
|
||||
TestAdderSMCFp32(input_shape, weight_shape, output_shape, stride, padding, dilation, groups, bias, core_mask);
|
||||
}
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void i8_adder_p(int8_t* input_x, int8_t* input_w, int8_t* out_y, int* bias, ConvParameter *conv_param, ConvQuantParameter quant_param, int core_mask)
|
||||
.. c:function:: void hp_adder_p(half* input_x, half* input_w, half* out_y, half* bias, ConvParameter *conv_param, int core_mask)
|
||||
.. c:function:: void fp_adder_p(float* input_x, float* input_w, float* out_y, float* bias, ConvParameter *conv_param, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 27
|
||||
|
||||
void TestAdderL2Fp32(int* input_shape, int* weight_shape, int* output_shape, int* stride, int* padding, int* dilation, int groups, float* bias, int core_mask) {
|
||||
float* input_data = (float*)0x10010000; // 私有存储版本地址设置在AM内
|
||||
float* weight = (float*)0x10020000;
|
||||
float* output_data = (float*)0x10030000;
|
||||
float* bias_data = (float*)0x10040000;
|
||||
AdderParameter* param = (AdderParameter*)0x10060000;
|
||||
memcpy(bias_data, bias, sizeof(float) * output_shape[3]);
|
||||
param->dilation_h_ = dilation[0];
|
||||
param->dilation_w_ = dilation[1];
|
||||
param->group_ = groups;
|
||||
param->input_batch_ = input_shape[0];
|
||||
param->input_h_ = input_shape[1];
|
||||
param->input_w_ = input_shape[2];
|
||||
param->input_channel_ = input_shape[3];
|
||||
param->kernel_h_ = weight_shape[1];
|
||||
param->kernel_w_ = weight_shape[2];
|
||||
param->output_batch_ = output_shape[0];
|
||||
param->output_h_ = output_shape[1];
|
||||
param->output_w_ = output_shape[2];
|
||||
param->output_channel_ = output_shape[3];
|
||||
param->stride_h_ = stride[0];
|
||||
param->stride_w_ = stride[0];
|
||||
param->pad_u_ = padding[0];
|
||||
param->pad_l_ = padding[2];
|
||||
param->workspace_ = (float*)0x10070000;
|
||||
param->buffer_size_ = 2048; // 私有存储版本中，必须设置该参数，用于确定分块计算的大小
|
||||
fp_adder_p(input_data, weight, output_data, bias_data, param, core_mask);
|
||||
}
|
||||
|
||||
void main(){
|
||||
int in_channel = 4;
|
||||
int out_channel = 4;
|
||||
int groups = 4;
|
||||
int input_shape[4] = {1, 30, 30, in_channel}; // NHWC
|
||||
int weight_shape[4] = {out_channel, 3, 3, in_channel / groups};
|
||||
int output_shape[4] = {1, 10, 10, out_channel}; // NHWC
|
||||
int stride[2] = {2, 2};
|
||||
int padding[4] = {1, 1, 1, 1};
|
||||
int dilation[2]= {2, 2};
|
||||
float bias[4] = {0, 0, 0, 0};
|
||||
int core_mask = 0b0001; // 私有存储版本只能设置为一个核心启动
|
||||
TestAdderL2Fp32(input_shape, weight_shape, output_shape, stride, padding, dilation, groups, bias, core_mask);
|
||||
}
|
||||
|
|
@ -0,0 +1,101 @@
|
|||
ApplyMomentum
|
||||
=================
|
||||
对权重张量执行 Momentum/改进动量优化更新。
|
||||
|
||||
.. math::
|
||||
|
||||
\begin{aligned}
|
||||
accu_t &= moment \cdot accu_{t-1} + g_t \\
|
||||
update_t &= \begin{cases}
|
||||
(accu_t \cdot moment + g_t), & \text{if nesterov = True} \\
|
||||
accu_t, & \text{otherwise}
|
||||
\end{cases} \\
|
||||
weight_t &= weight_{t-1} - learning\_rate \cdot update_t
|
||||
\end{aligned}
|
||||
|
||||
|
||||
输入：
|
||||
- **weight** - 待更新权重张量首地址。
|
||||
- **accumulate** - 动量累积张量首地址。
|
||||
- **gradient** - 梯度张量首地址。
|
||||
- **learning_rate** - 学习率。
|
||||
- **moment** - 动量系数。
|
||||
- **nesterov** - 是否启用 Nesterov 动量。
|
||||
- **start** - 参与计算的起始索引（闭区间）。
|
||||
- **end** - 参与计算的结束索引（开区间）。
|
||||
- **core_mask(int, 可选)** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **weight** - 原地写回更新后的权重张量。
|
||||
- **accumulate** - 原地写回更新后的动量张量。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持 fp32 数据类型。
|
||||
- MT7004 支持 fp16、fp32 数据类型。
|
||||
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void hp_applymomentum_s(half *weight, half *accumulate, const half *gradient, float learning_rate, float moment, bool nesterov, int start, int end, int core_mask)
|
||||
.. c:function:: void fp_applymomentum_s(float *weight, float *accumulate, const float *gradient, float learning_rate, float moment, bool nesterov, int start, int end, int core_mask)
|
||||
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 15
|
||||
|
||||
// FT78NE 多核示例
|
||||
#include <stdio.h>
|
||||
#include <stdbool.h>
|
||||
|
||||
int main(void) {
|
||||
float *weight = (float *)0xA0000000; // DDR 存储
|
||||
float *accumulate = (float *)0xB0000000;
|
||||
float *gradient = (float *)0xC0000000;
|
||||
int start = 0;
|
||||
int end = 4096;
|
||||
int core_mask = 0xff;
|
||||
float learning_rate = 1e-2f;
|
||||
float moment = 0.99f;
|
||||
bool nesterov = false;
|
||||
fp_applymomentum_s(weight, accumulate, gradient,
|
||||
learning_rate, moment, nesterov,
|
||||
start, end, core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void hp_applymomentum_p(half *weight, half *accumulate, const half *gradient, float learning_rate, float moment, bool nesterov, int length)
|
||||
.. c:function:: void fp_applymomentum_p(float *weight, float *accumulate, const float *gradient, float learning_rate, float moment, bool nesterov, int length)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 13
|
||||
|
||||
// MT7004 单核示例
|
||||
#include <stdio.h>
|
||||
#include <stdbool.h>
|
||||
|
||||
int main(void) {
|
||||
half *weight = (half *)0x10000000; // L2 存储
|
||||
half *accumulate = (half *)0x10002000;
|
||||
half *gradient = (half *)0x10004000;
|
||||
int length = 2048;
|
||||
float learning_rate = 5e-3f;
|
||||
float moment = 0.9f;
|
||||
bool nesterov = true;
|
||||
hp_applymomentum_p(weight, accumulate, gradient,
|
||||
learning_rate, moment, nesterov,
|
||||
length);
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,45 @@
|
|||
Assert
|
||||
=================
|
||||
|
||||
|
||||
.. c:function:: void assert(bool* Input, bool* output)
|
||||
|
||||
判断输入是否为 True。
|
||||
|
||||
.. math::
|
||||
|
||||
output_i = \begin{cases}
|
||||
\text{True}, & \text{if } Input_i = \text{True} \\
|
||||
\text{False}, & \text{if } Input_i = \text{False}
|
||||
\end{cases}
|
||||
|
||||
输入：
|
||||
- **Input** - 输入数据地址（布尔类型）。
|
||||
|
||||
输出：
|
||||
- **output** - 计算结果地址（布尔类型）。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- 本算子只有一个版本
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 10
|
||||
|
||||
// FT78NE/MT7004 示例（共享存储）
|
||||
#include <stdio.h>
|
||||
#include <stdbool.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
bool *input = (bool *)0xA0000000;
|
||||
bool *output = (bool *)0xC0000000;
|
||||
assert_s(input, output);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
|
@ -0,0 +1,79 @@
|
|||
Attention
|
||||
=================
|
||||
|
||||
多头缩放点积注意力机制（Scaled Dot-Product Attention）
|
||||
|
||||
.. math::
|
||||
|
||||
\text{Attention}(Q, K, V) = \operatorname{softmax}\left(\frac{Q K^\top}{\sqrt{d_k}}\right) V
|
||||
|
||||
输入：
|
||||
- **Q** - 查询矩阵地址（行优先），形状 :math:`[B, H, L, D]` 展平。
|
||||
- **K** - 键矩阵地址（行优先），形状 :math:`[B, H, L, D]` 展平。
|
||||
- **V** - 值矩阵地址（行优先），形状 :math:`[B, H, L, D]` 展平。
|
||||
- **batch_size (B)** - 批大小。
|
||||
- **seq_len (L)** - 序列长度。
|
||||
- **head_num (H)** - 多头数量。
|
||||
- **head_dim (D)** - 每头通道维数。
|
||||
- **QK, softmax_out** - 中间缓冲区地址，容量不小于 :math:`B\times H\times L\times L`。
|
||||
- **core_mask（可选）** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **output** - 输出地址（行优先），形状 :math:`[B, H, L, D]` 展平。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- 当前实现基于 fp32；输入/中间/输出缓冲区不应重叠。
|
||||
- 内存布局为行优先（row-major）。
|
||||
|
||||
**共享存储版本：**
|
||||
|
||||
.. c:function:: void fp_attention_s(float *Q, float *K, float *V, float *output, int batch_size, int seq_len, int head_num, int head_dim, float *QK, float *softmax_out, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 13
|
||||
|
||||
#include <stdio.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
int B = 2, L = 128, H = 8, D = 64;
|
||||
float *Q = (float *)0xA0000000; // DDR
|
||||
float *K = (float *)0xA1000000; // DDR
|
||||
float *V = (float *)0xA2000000; // DDR
|
||||
float *O = (float *)0xA3000000; // DDR
|
||||
float *QK = (float *)0xA4000000; // DDR
|
||||
float *SM = (float *)0xA5000000; // DDR
|
||||
int core_mask = 0xff;
|
||||
fp_attention_s(Q, K, V, O, B, L, H, D, QK, SM, core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
**私有存储版本：**
|
||||
|
||||
.. c:function:: void fp_attention_p(float *Q, float *K, float *V, float *output, int batch_size, int seq_len, int head_num, int head_dim, float *QK, float *softmax_out)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 12
|
||||
|
||||
#include <stdio.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
int B = 1, L = 64, H = 4, D = 32;
|
||||
float *Q = (float *)0x10000000; // L2
|
||||
float *K = (float *)0x10040000; // L2
|
||||
float *V = (float *)0x10080000; // L2
|
||||
float *O = (float *)0x100C0000; // L2
|
||||
float *QK = (float *)0x10100000; // L2
|
||||
float *SM = (float *)0x10200000; // L2
|
||||
fp_attention_p(Q, K, V, O, B, L, H, D, QK, SM);
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,120 @@
|
|||
AvgPoolingGrad
|
||||
=================
|
||||
根据输入梯度对平均池化前的特征图计算反向传播梯度。
|
||||
|
||||
.. math::
|
||||
|
||||
output_{b, x_h, x_w, c} += \frac{input_{b, y_h, y_w, c}}{\text{window}_h \cdot \text{window}_w}
|
||||
|
||||
其中 :math:`(y_h, y_w)` 与 :math:`(x_h, x_w)` 之间满足窗口与步长的映射关系。
|
||||
|
||||
输入：
|
||||
- **input** - 反向传播输入梯度张量首地址。
|
||||
- **output** - 反向传播输出梯度张量首地址。
|
||||
- **batch** - 批大小。
|
||||
- **output_h** - 池化输出高度。
|
||||
- **output_w** - 池化输出宽度。
|
||||
- **channel** - 通道数。
|
||||
- **input_w** - 池化输入宽度。
|
||||
- **input_h** - 池化输入高度。
|
||||
- **stride_w** - 水平步长。
|
||||
- **stride_h** - 垂直步长。
|
||||
- **pad_l** - 左侧填充大小。
|
||||
- **pad_u** - 上侧填充大小。
|
||||
- **window_w** - 池化窗口宽度。
|
||||
- **window_h** - 池化窗口高度。
|
||||
- **core_mask(int, 可选)** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **output** - 原地累加平均池化反向传播梯度。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持 fp32 数据类型。
|
||||
- MT7004 支持 fp16、fp32 数据类型。
|
||||
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void hp_avgpoolinggrad_s(const half *input, half *output, int batch, int output_h, int output_w, int channel, int input_w, int input_h, int stride_w, int stride_h, int pad_l, int pad_u, int window_w, int window_h, int core_mask)
|
||||
.. c:function:: void fp_avgpoolinggrad_s(const float *input, float *output, int batch, int output_h, int output_w, int channel, int input_w, int input_h, int stride_w, int stride_h, int pad_l, int pad_u, int window_w, int window_h, int core_mask)
|
||||
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 20
|
||||
|
||||
// FT78NE 多核示例
|
||||
#include <stdio.h>
|
||||
|
||||
int main(void) {
|
||||
const float *input = (const float *)0xA0000000; // DDR 存储
|
||||
float *output = (float *)0xB0000000;
|
||||
int batch = 16;
|
||||
int output_h = 7;
|
||||
int output_w = 7;
|
||||
int channel = 64;
|
||||
int input_w = 14;
|
||||
int input_h = 14;
|
||||
int stride_w = 2;
|
||||
int stride_h = 2;
|
||||
int pad_l = 0;
|
||||
int pad_u = 0;
|
||||
int window_w = 2;
|
||||
int window_h = 2;
|
||||
int core_mask = 0xff;
|
||||
fp_avgpoolinggrad_s(input, output, batch, output_h, output_w,
|
||||
channel, input_w, input_h,
|
||||
stride_w, stride_h,
|
||||
pad_l, pad_u,
|
||||
window_w, window_h,
|
||||
core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void hp_avgpoolinggrad_p(const half *input, half *output, int batch, int output_h, int output_w, int channel, int input_w, int input_h, int stride_w, int stride_h, int pad_l, int pad_u, int window_w, int window_h, int start_idx, int end_idx)
|
||||
.. c:function:: void fp_avgpoolinggrad_p(const float *input, float *output, int batch, int output_h, int output_w, int channel, int input_w, int input_h, int stride_w, int stride_h, int pad_l, int pad_u, int window_w, int window_h, int start_idx, int end_idx)
|
||||
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 21
|
||||
|
||||
// MT7004 单核示例
|
||||
#include <stdio.h>
|
||||
|
||||
int main(void) {
|
||||
const half *input = (const half *)0x10000000; // L2 存储
|
||||
half *output = (half *)0x10004000;
|
||||
int batch = 4;
|
||||
int output_h = 4;
|
||||
int output_w = 4;
|
||||
int channel = 128;
|
||||
int input_w = 8;
|
||||
int input_h = 8;
|
||||
int stride_w = 2;
|
||||
int stride_h = 2;
|
||||
int pad_l = 0;
|
||||
int pad_u = 0;
|
||||
int window_w = 2;
|
||||
int window_h = 2;
|
||||
int start_idx = 0;
|
||||
int end_idx = batch * output_h * output_w * channel;
|
||||
hp_avgpoolinggrad_p(input, output, batch, output_h, output_w,
|
||||
channel, input_w, input_h,
|
||||
stride_w, stride_h,
|
||||
pad_l, pad_u,
|
||||
window_w, window_h,
|
||||
start_idx, end_idx);
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,96 @@
|
|||
BatchToSpace
|
||||
=================
|
||||
将输入张量在批维度上分块并重新分布到空间维度，同时按照 ``crops`` 对输出空间范围进行裁剪。
|
||||
|
||||
.. math::
|
||||
|
||||
\begin{aligned}
|
||||
N_{\text{out}} &= \frac{N}{b_h \times b_w}, \\
|
||||
H_{\text{out}} &= b_h \times H - c_{\text{top}} - c_{\text{bottom}}, \\
|
||||
W_{\text{out}} &= b_w \times W - c_{\text{left}} - c_{\text{right}}, \\
|
||||
ext{output}[n, h, w, c] &= \text{input}[n', h', w', c]
|
||||
\end{aligned}
|
||||
|
||||
其中 :math:`N, H, W, C` 分别表示输入的 batch、高度、宽度和通道数；:math:`b_h, b_w` 为 ``block_size``；:math:`c_{*}` 来源于 ``crops``；:math:`n', h', w'` 由 ``BatchToSpace`` 映射关系确定。
|
||||
|
||||
输入：
|
||||
- **input** - 输入数据地址。
|
||||
- **input_shape** - 输入形状，格式为 ``[batch, height, width, channel]``。
|
||||
- **block_size** - 分块因子，格式为 ``[block_h, block_w]``。
|
||||
- **crops** - 裁剪参数，格式为 ``[top, bottom, left, right]``。
|
||||
- **core_mask(int, 可选)** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **output** - 输出数据地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持 fp32、fp64、cplx64、cplx128、int16、int8、int32 数据类型。
|
||||
- MT7004 支持 fp32、fp16、cplx64、int16、int32 数据类型。
|
||||
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void i8_batchtospace_s(int8_t *input, int8_t *output, const int *input_shape, const int *block_size, const int *crops, int data_size, int core_mask)
|
||||
.. c:function:: void i16_batchtospace_s(int16_t *input, int16_t *output, const int *input_shape, const int *block_size, const int *crops, int data_size, int core_mask)
|
||||
.. c:function:: void i32_batchtospace_s(int32_t *input, int32_t *output, const int *input_shape, const int *block_size, const int *crops, int data_size, int core_mask)
|
||||
.. c:function:: void hp_batchtospace_s(half *input, half *output, const int *input_shape, const int *block_size, const int *crops, int data_size, int core_mask)
|
||||
.. c:function:: void fp_batchtospace_s(float *input, float *output, const int *input_shape, const int *block_size, const int *crops, int data_size, int core_mask)
|
||||
.. c:function:: void dp_batchtospace_s(double *input, double *output, const int *input_shape, const int *block_size, const int *crops, int data_size, int core_mask)
|
||||
.. c:function:: void c64_batchtospace_s(float *input, float *output, const int *input_shape, const int *block_size, const int *crops, int data_size, int core_mask)
|
||||
.. c:function:: void c128_batchtospace_s(double *input, double *output, const int *input_shape, const int *block_size, const int *crops, int data_size, int core_mask)
|
||||
|
||||
|
||||
**C 调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 11
|
||||
|
||||
// 多核（共享存储）示例
|
||||
#include <stdio.h>
|
||||
|
||||
int main(int argc, char *argv[]) {
|
||||
float *input = (float *)0xA0000000; // 输入在 DDR 空间
|
||||
float *output = (float *)0xB0000000;
|
||||
int input_shape[4] = {400, 2, 2, 3};
|
||||
int block_size[2] = {2, 2};
|
||||
int crops[4] = {0, 0, 0, 0};
|
||||
int core_mask = 0xff;
|
||||
fp_batchtospace_s(input, output, input_shape, block_size, crops, sizeof(float), core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void i8_batchtospace_p(int8_t *input, int8_t *output, const int *input_shape, const int *block_size, const int *crops, int data_size)
|
||||
.. c:function:: void i16_batchtospace_p(int16_t *input, int16_t *output, const int *input_shape, const int *block_size, const int *crops, int data_size)
|
||||
.. c:function:: void i32_batchtospace_p(int32_t *input, int32_t *output, const int *input_shape, const int *block_size, const int *crops, int data_size)
|
||||
.. c:function:: void hp_batchtospace_p(half *input, half *output, const int *input_shape, const int *block_size, const int *crops, int data_size)
|
||||
.. c:function:: void fp_batchtospace_p(float *input, float *output, const int *input_shape, const int *block_size, const int *crops, int data_size)
|
||||
.. c:function:: void dp_batchtospace_p(double *input, double *output, const int *input_shape, const int *block_size, const int *crops, int data_size)
|
||||
.. c:function:: void c64_batchtospace_p(float *input, float *output, const int *input_shape, const int *block_size, const int *crops, int data_size)
|
||||
.. c:function:: void c128_batchtospace_p(double *input, double *output, const int *input_shape, const int *block_size, const int *crops, int data_size)
|
||||
|
||||
**C 调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 10
|
||||
|
||||
// 单核（私有存储）示例
|
||||
#include <stdio.h>
|
||||
|
||||
int main(int argc, char *argv[]) {
|
||||
float *input = (float *)0x10000000; // 输入在 L2 空间
|
||||
float *output = (float *)0x10010000;
|
||||
int input_shape[4] = {400, 2, 2, 3};
|
||||
int block_size[2] = {2, 2};
|
||||
int crops[4] = {0, 0, 0, 0};
|
||||
fp_batchtospace_p(input, output, input_shape, block_size, crops, sizeof(float));
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,91 @@
|
|||
BatchToSpaceND
|
||||
=================
|
||||
将输入张量的 batch 维度按 block 因子分解并重排到空间维度，随后根据 ``crops`` 参数对输出空间范围进行裁剪。
|
||||
|
||||
|
||||
- 输入形状： ``[batch, height, width, channel]``
|
||||
- ``block_size``： ``[block_h, block_w]``
|
||||
- ``crops``： ``[top, bottom, left, right]``
|
||||
|
||||
输入：
|
||||
- **input** - 输入数据地址。
|
||||
- **input_shape** - 输入形状，格式为 ``[batch, height, width, channel]``。
|
||||
- **block_size** - 分块因子，格式为 ``[block_h, block_w]``。
|
||||
- **crops** - 裁剪参数，格式为 ``[top, bottom, left, right]``。
|
||||
- **core_mask(int, 可选)** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **output** - 输出数据地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持的数据类型： fp32、fp64、cplx64、cplx128、int16、int8、int32。
|
||||
- MT7004 支持的数据类型： fp32、fp16、cplx64、int16、int32。
|
||||
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void i8_batchtospacend_s(int8_t *input, int8_t *output, const int *input_shape, const int *block_size, const int *crops, int data_size, int core_mask)
|
||||
.. c:function:: void i16_batchtospacend_s(int16_t *input, int16_t *output, const int *input_shape, const int *block_size, const int *crops, int data_size, int core_mask)
|
||||
.. c:function:: void i32_batchtospacend_s(int32_t *input, int32_t *output, const int *input_shape, const int *block_size, const int *crops, int data_size, int core_mask)
|
||||
.. c:function:: void hp_batchtospacend_s(half *input, half *output, const int *input_shape, const int *block_size, const int *crops, int data_size, int core_mask)
|
||||
.. c:function:: void fp_batchtospacend_s(float *input, float *output, const int *input_shape, const int *block_size, const int *crops, int data_size, int core_mask)
|
||||
.. c:function:: void dp_batchtospacend_s(double *input, double *output, const int *input_shape, const int *block_size, const int *crops, int data_size, int core_mask)
|
||||
.. c:function:: void c64_batchtospacend_s(float *input, float *output, const int *input_shape, const int *block_size, const int *crops, int data_size, int core_mask)
|
||||
.. c:function:: void c128_batchtospacend_s(double *input, double *output, const int *input_shape, const int *block_size, const int *crops, int data_size, int core_mask)
|
||||
|
||||
**C 调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 11
|
||||
|
||||
// FT78NE 多核示例
|
||||
#include <stdio.h>
|
||||
|
||||
int main(int argc, char *argv[]) {
|
||||
float *input = (float *)0xA0000000; // 多核版本：输入放在 DDR 地址 0xA0000000
|
||||
float *output = (float *)0xB0000000; // 多核版本：输出放在 DDR 地址 0xB0000000
|
||||
int input_shape[4] = {400, 2, 2, 3};
|
||||
int block_size[2] = {2, 2};
|
||||
int crops[4] = {0, 0, 0, 0};
|
||||
int core_mask = 0xff;
|
||||
fp_batchtospacend_s(input, output, input_shape, block_size, crops, sizeof(float), core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void i8_batchtospacend_p(int8_t *input, int8_t *output, const int *input_shape, const int *block_size, const int *crops, int data_size)
|
||||
.. c:function:: void i16_batchtospacend_p(int16_t *input, int16_t *output, const int *input_shape, const int *block_size, const int *crops, int data_size)
|
||||
.. c:function:: void i32_batchtospacend_p(int32_t *input, int32_t *output, const int *input_shape, const int *block_size, const int *crops, int data_size)
|
||||
.. c:function:: void hp_batchtospacend_p(half *input, half *output, const int *input_shape, const int *block_size, const int *crops, int data_size)
|
||||
.. c:function:: void fp_batchtospacend_p(float *input, float *output, const int *input_shape, const int *block_size, const int *crops, int data_size)
|
||||
.. c:function:: void dp_batchtospacend_p(double *input, double *output, const int *input_shape, const int *block_size, const int *crops, int data_size)
|
||||
.. c:function:: void c64_batchtospacend_p(float *input, float *output, const int *input_shape, const int *block_size, const int *crops, int data_size)
|
||||
.. c:function:: void c128_batchtospacend_p(double *input, double *output, const int *input_shape, const int *block_size, const int *crops, int data_size)
|
||||
|
||||
**C 调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 10
|
||||
|
||||
// FT78NE 单核示例
|
||||
#include <stdio.h>
|
||||
|
||||
int main(int argc, char *argv[]) {
|
||||
float *input = (float *)0x10000000; // 单核版本：输入放在 L2 地址 0x10000000
|
||||
float *output = (float *)0x10010000; // 单核版本：输出放在 L2 地址 0x10010000
|
||||
int input_shape[4] = {400, 2, 2, 3};
|
||||
int block_size[2] = {2, 2};
|
||||
int crops[4] = {0, 0, 0, 0};
|
||||
fp_batchtospacend_p(input, output, input_shape, block_size, crops, sizeof(float));
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
|
|
@ -0,0 +1,84 @@
|
|||
BroadcastTo
|
||||
=================
|
||||
将较小张量按 N-D 规则广播到目标形状并写入输出。
|
||||
|
||||
输入：
|
||||
- **input** - 输入数据地址。
|
||||
- **input_shape** - 输入形状数组。
|
||||
- **input_shape_size** - 输入形状长度。
|
||||
- **output_shape** - 目标输出形状数组。
|
||||
- **output_shape_size** - 输出形状长度。
|
||||
- **data_size** - 单个元素字节数（例如 sizeof(float)）。
|
||||
- **core_mask(int, 可选)** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **output** - 输出数据地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持的数据类型： fp32、fp64、cplx64、cplx128、int16、int8、int32。
|
||||
- MT7004 支持的数据类型： fp32、fp16、cplx64、int16、int32。
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void i8_broadcastto_s(int8_t *input, int8_t *output, const int *input_shape, int input_shape_size, const int *output_shape, int output_shape_size, int data_size, int core_mask)
|
||||
.. c:function:: void i16_broadcastto_s(int16_t *input, int16_t *output, const int *input_shape, int input_shape_size, const int *output_shape, int output_shape_size, int data_size, int core_mask)
|
||||
.. c:function:: void i32_broadcastto_s(int32_t *input, int32_t *output, const int *input_shape, int input_shape_size, const int *output_shape, int output_shape_size, int data_size, int core_mask)
|
||||
.. c:function:: void hp_broadcastto_s(half *input, half *output, const int *input_shape, int input_shape_size, const int *output_shape, int output_shape_size, int data_size, int core_mask)
|
||||
.. c:function:: void fp_broadcastto_s(float *input, float *output, const int *input_shape, int input_shape_size, const int *output_shape, int output_shape_size, int data_size, int core_mask)
|
||||
.. c:function:: void dp_broadcastto_s(double *input, double *output, const int *input_shape, int input_shape_size, const int *output_shape, int output_shape_size, int data_size, int core_mask)
|
||||
.. c:function:: void c64_broadcastto_s(float *input, float *output, const int *input_shape, int input_shape_size, const int *output_shape, int output_shape_size, int data_size, int core_mask)
|
||||
.. c:function:: void c128_broadcastto_s(double *input, double *output, const int *input_shape, int input_shape_size, const int *output_shape, int output_shape_size, int data_size, int core_mask)
|
||||
|
||||
**C 调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 10
|
||||
|
||||
// FT78NE 多核示例
|
||||
#include <stdio.h>
|
||||
|
||||
int main(int argc, char *argv[]) {
|
||||
float *input = (float *)0xA0000000; // 输入在 DDR 地址 0xA0000000
|
||||
float *output = (float *)0xB0000000; // 输出在 DDR 地址 0xB0000000
|
||||
int input_shape[4] = {1, 50, 1, 1};
|
||||
int output_shape[4] = {10, 50, 20, 1};
|
||||
int core_mask = 0xff;
|
||||
fp_broadcastto_s(input, output, input_shape, 4, output_shape, 4, sizeof(float), core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void i8_broadcastto_p(int8_t *input, int8_t *output, const int *input_shape, int input_shape_size, const int *output_shape, int output_shape_size, int data_size)
|
||||
.. c:function:: void i16_broadcastto_p(int16_t *input, int16_t *output, const int *input_shape, int input_shape_size, const int *output_shape, int output_shape_size, int data_size)
|
||||
.. c:function:: void i32_broadcastto_p(int32_t *input, int32_t *output, const int *input_shape, int input_shape_size, const int *output_shape, int output_shape_size, int data_size)
|
||||
.. c:function:: void hp_broadcastto_p(half *input, half *output, const int *input_shape, int input_shape_size, const int *output_shape, int output_shape_size, int data_size)
|
||||
.. c:function:: void fp_broadcastto_p(float *input, float *output, const int *input_shape, int input_shape_size, const int *output_shape, int output_shape_size, int data_size)
|
||||
.. c:function:: void dp_broadcastto_p(double *input, double *output, const int *input_shape, int input_shape_size, const int *output_shape, int output_shape_size, int data_size)
|
||||
.. c:function:: void c64_broadcastto_p(float *input, float *output, const int *input_shape, int input_shape_size, const int *output_shape, int output_shape_size, int data_size)
|
||||
.. c:function:: void c128_broadcastto_p(double *input, double *output, const int *input_shape, int input_shape_size, const int *output_shape, int output_shape_size, int data_size)
|
||||
|
||||
**C 调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 9
|
||||
|
||||
// FT78NE 单核示例
|
||||
#include <stdio.h>
|
||||
|
||||
int main(int argc, char *argv[]) {
|
||||
float *input = (float *)0x10000000; // 单核版本：输入放在 L2 地址 0x10000000
|
||||
float *output = (float *)0x10020000; // 单核版本：输出放在 L2 地址 0x10020000
|
||||
int input_shape[4] = {1, 50, 1, 1};
|
||||
int output_shape[4] = {10, 50, 20, 1};
|
||||
fp_broadcastto_p(input, output, input_shape, 4, output_shape, 4, sizeof(float));
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
|
@ -0,0 +1,192 @@
|
|||
Conv2d
|
||||
=================
|
||||
|
||||
对输入 Tensor 计算二维卷积，输入的 shape 为 :math:`(N, H_{in}, W_{in}, C_{in})`，其中 :math:`N` 为 batch size，:math:`C` 为通道数，:math:`H` 为特征图的高度，:math:`W` 为特征图的宽度。
|
||||
|
||||
根据以下公式计算输出：
|
||||
|
||||
.. math::
|
||||
|
||||
out(N_i, C_{out_j}) = bias(C_{out_j}) + \sum_{k=0}^{C_{in}-1} \text{ccor}(\text{weight}(C_{out_j}, k), X(N_i, k))
|
||||
|
||||
其中，:math:`bias` 为输出偏置，:math:`\text{ccor}` 为 cross-correlation 操作，:math:`weight` 为卷积核的值，:math:`X` 为输入的特征图。
|
||||
|
||||
- :math:`i` 对应 batch 数，其范围为 :math:`[0, N-1]`，其中 :math:`N` 为输入 batch。
|
||||
- :math:`j` 对应输出通道，其范围为 :math:`[0, C_{out}-1]`，其中 :math:`C_{out}` 为输出通道数，该值也等于卷积核的个数。
|
||||
- :math:`k` 对应输入通道数，其范围为 :math:`[0, C_{in}-1]`，其中 :math:`C_{in}` 为输入通道数，该值也等于卷积核的通道数。
|
||||
|
||||
因此，上面的公式中，:math:`bias(C_{out_j})` 为第 :math:`j` 个输出通道的偏置，:math:`weight(C_{out_j}, k)` 表示第 :math:`j` 个卷积核在第 :math:`k` 个输入通道的卷积核切片，:math:`X(N_i, k)` 为特征图第 :math:`i` 个 batch 第 :math:`k` 个输入通道的切片。卷积核 shape 为 :math:`(\text{kernel_size}[0], \text{kernel_size}[1])`，其中 kernel_size[0] 和 kernel_size[1] 是卷积核的高度和宽度。若考虑到输入输出通道以及 group，则完整卷积核的 shape 为 :math:`(C_{out}, \text{kernel_size}[0], \text{kernel_size}[1], C_{in}/\text{group})`，其中 group 是分组卷积时在通道上分割输入 :math:`x` 的组数。
|
||||
|
||||
输入：
|
||||
- **input_x** - 输入数据的地址
|
||||
- **input_w** - 输入卷积核权重的地址
|
||||
- **bias** - 输入偏置的地址
|
||||
- **conv_param** - 算子计算所需参数的结构体。其各成员见下述。
|
||||
- **quant_param** - 对int8类型进行量化计算所需参数的结构体。其各成员见下述。
|
||||
- **core_mask** - 核掩码。
|
||||
|
||||
**ConvParameter及ConvQuantParameter定义：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
|
||||
typedef struct ConvParameter {
|
||||
void* workspace_; // 用于存放中间计算结果
|
||||
int output_batch_; // 输出数据总批次
|
||||
int input_batch_; // 输入数据总批次
|
||||
int input_h_; // 输入数据h维度大小
|
||||
int input_w_; // 输入数据w维度大小
|
||||
int output_h_; // 输出数据h维度大小
|
||||
int output_w_; // 输出数据w维度大小
|
||||
int input_channel_; // 输入数据通道数
|
||||
int output_channel_; // 输出数据通道数
|
||||
int kernel_h_; // 卷积核h维度大小
|
||||
int kernel_w_; // 卷积核w维度大小
|
||||
int group_; // 组数
|
||||
int pad_l_; // 左填充大小
|
||||
int pad_u_; // 上填充大小
|
||||
int dilation_h_; // 卷积核h维度膨胀尺寸大小
|
||||
int dilation_w_; // 卷积核w维度膨胀尺寸大小
|
||||
int stride_h_; // 卷积核h维度步长
|
||||
int stride_w_; // 卷积核w维度步长
|
||||
int buffer_size_; // 为分块计算所分配的缓存大小
|
||||
} ConvParameter;
|
||||
|
||||
typedef struct ConvQuantParameter {
|
||||
int32_t* left_shift_;
|
||||
int32_t* right_shift_;
|
||||
int32_t* multiplier_;
|
||||
int32_t* filter_zp_ptr_;
|
||||
int32_t output_zp_;
|
||||
int32_t mini_;
|
||||
int32_t maxi_;
|
||||
int per_channel_;
|
||||
} ConvQuantParameter;
|
||||
|
||||
输出：
|
||||
- **out_y** - 输出地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持int8, fp32
|
||||
- MT7004 支持fp16, fp32
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void i8_conv2d_s(int8_t* input_x, int8_t* input_w, int8_t* out_y, int* bias, ConvParameter *conv_param, ConvQuantParameter quant_param, int core_mask)
|
||||
.. c:function:: void hp_conv2d_s(half* input_x, half* input_w, half* out_y, half* bias, ConvParameter *conv_param, int core_mask)
|
||||
.. c:function:: void fp_conv2d_s(float* input_x, float* input_w, float* out_y, float* bias, ConvParameter *conv_param, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 32
|
||||
|
||||
void TestConvSMCFp32(int* input_shape, int* weight_shape, int* output_shape, int* stride, int* padding, int* dilation, int groups, float* bias, int core_mask) {
|
||||
int core_id = get_core_id();
|
||||
int logic_core_id = GetLogicCoreId(core_mask, core_id);
|
||||
int core_num = GetCoreNum(core_mask);
|
||||
float* input_data = (float*)0x88000000;
|
||||
float* weight = (float*)0x89000000;
|
||||
float* output_data = (float*)0x90000000;
|
||||
float* bias_data = (float*)0x91000000;
|
||||
ConvParameter* param = (ConvParameter*)0x92000000;
|
||||
if (logic_core_id == 0) {
|
||||
memcpy(bias_data, bias, sizeof(float) * output_shape[3]);
|
||||
param->dilation_h_ = dilation[0];
|
||||
param->dilation_w_ = dilation[1];
|
||||
param->group_ = groups;
|
||||
param->input_batch_ = input_shape[0];
|
||||
param->input_h_ = input_shape[1];
|
||||
param->input_w_ = input_shape[2];
|
||||
param->input_channel_ = input_shape[3];
|
||||
param->kernel_h_ = weight_shape[1];
|
||||
param->kernel_w_ = weight_shape[2];
|
||||
param->output_batch_ = output_shape[0];
|
||||
param->output_h_ = output_shape[1];
|
||||
param->output_w_ = output_shape[2];
|
||||
param->output_channel_ = output_shape[3];
|
||||
param->stride_h_ = stride[0];
|
||||
param->stride_w_ = stride[0];
|
||||
param->pad_u_ = padding[0];
|
||||
param->pad_l_ = padding[2];
|
||||
param->workspace_ = (float*)0x10000000; // workspace空间需分配在AM内，计算过程中会将数据搬运到workspace空间内进行计算
|
||||
}
|
||||
sys_bar(0, core_num); // 初始化参数完成后进行同步
|
||||
fp_conv2d_s(input_data, weight, output_data, bias_data, param, core_mask);
|
||||
}
|
||||
|
||||
void main(){
|
||||
int in_channel = 4;
|
||||
int out_channel = 4;
|
||||
int groups = 4;
|
||||
int input_shape[4] = {1, 30, 30, in_channel}; // NHWC
|
||||
int weight_shape[4] = {out_channel, 3, 3, in_channel / groups};
|
||||
int output_shape[4] = {1, 10, 10, out_channel}; // NHWC
|
||||
int stride[2] = {2, 2};
|
||||
int padding[4] = {1, 1, 1, 1};
|
||||
int dilation[2]= {2, 2};
|
||||
float bias[4] = {0, 0, 0, 0};
|
||||
int core_mask = 0b1111;
|
||||
TestConvSMCFp32(input_shape, weight_shape, output_shape, stride, padding, dilation, groups, bias, core_mask);
|
||||
}
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void i8_conv2d_p(int8_t* input_x, int8_t* input_w, int8_t* out_y, int* bias, ConvParameter *conv_param, ConvQuantParameter quant_param, int core_mask)
|
||||
.. c:function:: void hp_conv2d_p(half* input_x, half* input_w, half* out_y, half* bias, ConvParameter *conv_param, int core_mask)
|
||||
.. c:function:: void fp_conv2d_p(float* input_x, float* input_w, float* out_y, float* bias, ConvParameter *conv_param, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 27
|
||||
|
||||
void TestConvL2Fp32(int* input_shape, int* weight_shape, int* output_shape, int* stride, int* padding, int* dilation, int groups, float* bias, int core_mask) {
|
||||
float* input_data = (float*)0x10010000; // 私有存储版本地址设置在AM内
|
||||
float* weight = (float*)0x10020000;
|
||||
float* output_data = (float*)0x10030000;
|
||||
float* bias_data = (float*)0x10040000;
|
||||
ConvParameter* param = (ConvParameter*)0x10060000;
|
||||
memcpy(bias_data, bias, sizeof(float) * output_shape[3]);
|
||||
param->dilation_h_ = dilation[0];
|
||||
param->dilation_w_ = dilation[1];
|
||||
param->group_ = groups;
|
||||
param->input_batch_ = input_shape[0];
|
||||
param->input_h_ = input_shape[1];
|
||||
param->input_w_ = input_shape[2];
|
||||
param->input_channel_ = input_shape[3];
|
||||
param->kernel_h_ = weight_shape[1];
|
||||
param->kernel_w_ = weight_shape[2];
|
||||
param->output_batch_ = output_shape[0];
|
||||
param->output_h_ = output_shape[1];
|
||||
param->output_w_ = output_shape[2];
|
||||
param->output_channel_ = output_shape[3];
|
||||
param->stride_h_ = stride[0];
|
||||
param->stride_w_ = stride[0];
|
||||
param->pad_u_ = padding[0];
|
||||
param->pad_l_ = padding[2];
|
||||
param->workspace_ = (float*)0x10070000;
|
||||
param->buffer_size_ = 2048; // 私有存储版本中，必须设置该参数，用于确定分块计算的大小
|
||||
fp_conv2d_p(input_data, weight, output_data, bias_data, param, core_mask);
|
||||
}
|
||||
|
||||
void main(){
|
||||
int in_channel = 4;
|
||||
int out_channel = 4;
|
||||
int groups = 4;
|
||||
int input_shape[4] = {1, 30, 30, in_channel}; // NHWC
|
||||
int weight_shape[4] = {out_channel, 3, 3, in_channel / groups};
|
||||
int output_shape[4] = {1, 10, 10, out_channel}; // NHWC
|
||||
int stride[2] = {2, 2};
|
||||
int padding[4] = {1, 1, 1, 1};
|
||||
int dilation[2]= {2, 2};
|
||||
float bias[4] = {0, 0, 0, 0};
|
||||
int core_mask = 0b0001; // 私有存储版本只能设置为一个核心启动
|
||||
TestConvL2Fp32(input_shape, weight_shape, output_shape, stride, padding, dilation, groups, bias, core_mask);
|
||||
}
|
||||
|
|
@ -0,0 +1,179 @@
|
|||
Conv2dTranspose
|
||||
=================
|
||||
|
||||
计算二维转置卷积，可以视为 Conv2d 对输入求梯度，也称为反卷积（实际不是真正的反卷积）。
|
||||
|
||||
输入的 shape 通常为 :math:`(N, H_{in}, W_{in}, C_{in})`，其中：
|
||||
|
||||
- :math:`N` 是 batch size
|
||||
- :math:`C_{in}` 是空间维度
|
||||
- :math:`H_{in}, W_{in}` 分别为特征层的高度和宽度
|
||||
|
||||
输入：
|
||||
- **input_x** - 输入数据的地址
|
||||
- **input_w** - 输入卷积核权重的地址
|
||||
- **bias** - 输入偏置的地址
|
||||
- **param** - 算子计算所需参数的结构体。其各成员见下述。
|
||||
- **core_mask** - 核掩码。
|
||||
|
||||
**ConvTransposeParameter定义：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
|
||||
typedef struct ConvTransposeParameter {
|
||||
void* workspace_; // 用于存放中间计算结果
|
||||
int output_batch_; // 输出数据总批次
|
||||
int input_batch_; // 输入数据总批次
|
||||
int input_h_; // 输入数据h维度大小
|
||||
int input_w_; // 输入数据w维度大小
|
||||
int output_h_; // 输出数据h维度大小
|
||||
int output_w_; // 输出数据w维度大小
|
||||
int input_channel_; // 输入数据通道数
|
||||
int output_channel_; // 输出数据通道数
|
||||
int kernel_h_; // 卷积核h维度大小
|
||||
int kernel_w_; // 卷积核w维度大小
|
||||
int group_; // 组数
|
||||
int pad_l_; // 左填充大小
|
||||
int pad_u_; // 上填充大小
|
||||
int dilation_h_; // 卷积核h维度膨胀尺寸大小
|
||||
int dilation_w_; // 卷积核w维度膨胀尺寸大小
|
||||
int stride_h_; // 卷积核h维度步长
|
||||
int stride_w_; // 卷积核w维度步长
|
||||
int buffer_size_; // 为分块计算所分配的缓存大小
|
||||
} ConvTransposeParameter;
|
||||
|
||||
输出：
|
||||
- **out_y** - 输出地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持int8, fp32
|
||||
- MT7004 支持fp16, fp32
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void i8_convtranspose_s(int8_t* input_x, int8_t* input_w, int8_t* out_y, int* bias, ConvTransposeParameter *conv_param, int core_mask)
|
||||
.. c:function:: void hp_convtranspose_s(half* input_x, half* input_w, half* out_y, half* bias, ConvTransposeParameter *conv_param, int core_mask)
|
||||
.. c:function:: void fp_convtranspose_s(float* input_x, float* input_w, float* out_y, float* bias, ConvTransposeParameter *conv_param, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 35
|
||||
|
||||
void TestConvTransposeSMCFp32(int* input_shape, int* weight_shape, int* output_shape, int* stride, int* padding, int* dilation, int groups, float* bias, int core_mask) {
|
||||
int core_id = get_core_id();
|
||||
int logic_core_id = GetLogicCoreId(core_mask, core_id);
|
||||
int core_num = GetCoreNum(core_mask);
|
||||
float* input_data = (float*)0x88000000;
|
||||
float* weight = (float*)0x89000000;
|
||||
float* output_data = (float*)0x90000000;
|
||||
float* bias_data = (float*)0x91000000;
|
||||
float* check = (float*)0x94000000;
|
||||
ConvTransposeParameter* param = (ConvTransposeParameter*)0x92000000;
|
||||
if (logic_core_id == 0) {
|
||||
memcpy(bias_data, bias, sizeof(float) * output_shape[3]);
|
||||
memset(output_data, 0, output_shape[0] * output_shape[1] * output_shape[2] * output_shape[3] * sizeof(float));
|
||||
memset(check, 0, output_shape[0] * output_shape[1] * output_shape[2] * output_shape[3] * sizeof(float));
|
||||
param->dilation_h_ = dilation[0];
|
||||
param->dilation_w_ = dilation[1];
|
||||
param->group_ = groups;
|
||||
param->input_batch_ = input_shape[0];
|
||||
param->input_h_ = input_shape[1];
|
||||
param->input_w_ = input_shape[2];
|
||||
param->input_channel_ = input_shape[3];
|
||||
param->kernel_h_ = weight_shape[1];
|
||||
param->kernel_w_ = weight_shape[2];
|
||||
param->output_batch_ = output_shape[0];
|
||||
param->output_h_ = output_shape[1];
|
||||
param->output_w_ = output_shape[2];
|
||||
param->output_channel_ = output_shape[3];
|
||||
param->stride_h_ = stride[0];
|
||||
param->stride_w_ = stride[0];
|
||||
param->pad_u_ = padding[0];
|
||||
param->pad_l_ = padding[2];
|
||||
param->workspace_ = (float*)0xA0000000;
|
||||
}
|
||||
sys_bar(0, core_num); // 初始化参数完成后进行同步
|
||||
fp_convtranspose_s(input_data, weight, output_data, bias_data, param, core_mask);
|
||||
}
|
||||
|
||||
void main(){
|
||||
int in_channel = 6;
|
||||
int out_channel = 6;
|
||||
int groups = 6;
|
||||
int input_shape[4] = {2, 5, 7, in_channel}; // NHWC
|
||||
int weight_shape[4] = {in_channel, 3, 3, out_channel / groups};
|
||||
int output_shape[4] = {2, 7, 9, out_channel}; // NHWC
|
||||
int stride[2] = {1, 1};
|
||||
int padding[4] = {0, 0, 0, 0};
|
||||
int dilation[2]= {1, 1};
|
||||
float bias[] = {0, 0, 0, 0, 0, 0};
|
||||
int core_mask = 0b1111;
|
||||
TestConvTransposeSMCFp32(input_shape, weight_shape, output_shape, stride, padding, dilation, groups, bias, core_mask);
|
||||
}
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void i8_convtranspose_p(int8_t* input_x, int8_t* input_w, int8_t* out_y, int* bias, ConvTransposeParameter *conv_param, int core_mask)
|
||||
.. c:function:: void hp_convtranspose_p(half* input_x, half* input_w, half* out_y, half* bias, ConvTransposeParameter *conv_param, int core_mask)
|
||||
.. c:function:: void fp_convtranspose_p(float* input_x, float* input_w, float* out_y, float* bias, ConvTransposeParameter *conv_param, int core_mask)
|
||||
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 30
|
||||
|
||||
void TestConvTransposeL2Fp32(int* input_shape, int* weight_shape, int* output_shape, int* stride, int* padding, int* dilation, int groups, float* bias, int core_mask) {
|
||||
float* input_data = (float*)0x10000000; // 私有存储版本地址设置在AM内
|
||||
float* weight = (float*)0x10001000;
|
||||
float* output_data = (float*)0x10002000;
|
||||
float* bias_data = (float*)0x10003000;
|
||||
float* check = (float*)0x10004000;
|
||||
ConvTransposeParameter* param = (ConvTransposeParameter*)0x10005000;
|
||||
memcpy(bias_data, bias, sizeof(float) * output_shape[3]);
|
||||
memset(output_data, 0, output_shape[0] * output_shape[1] * output_shape[2] * output_shape[3] * sizeof(float));
|
||||
memset(check, 0, output_shape[0] * output_shape[1] * output_shape[2] * output_shape[3] * sizeof(float));
|
||||
param->dilation_h_ = dilation[0];
|
||||
param->dilation_w_ = dilation[1];
|
||||
param->group_ = groups;
|
||||
param->input_batch_ = input_shape[0];
|
||||
param->input_h_ = input_shape[1];
|
||||
param->input_w_ = input_shape[2];
|
||||
param->input_channel_ = input_shape[3];
|
||||
param->kernel_h_ = weight_shape[1];
|
||||
param->kernel_w_ = weight_shape[2];
|
||||
param->output_batch_ = output_shape[0];
|
||||
param->output_h_ = output_shape[1];
|
||||
param->output_w_ = output_shape[2];
|
||||
param->output_channel_ = output_shape[3];
|
||||
param->stride_h_ = stride[0];
|
||||
param->stride_w_ = stride[0];
|
||||
param->pad_u_ = padding[0];
|
||||
param->pad_l_ = padding[2];
|
||||
param->workspace_ = (float*)0x10006000;
|
||||
param->buffer_size_ = 1024; // 私有存储版本中，必须设置该参数，用于确定分块计算的大小
|
||||
fp_convtranspose_p(input_data, weight, output_data, bias_data, param, core_mask);
|
||||
}
|
||||
|
||||
void main(){
|
||||
int in_channel = 6;
|
||||
int out_channel = 6;
|
||||
int groups = 6;
|
||||
int input_shape[4] = {2, 5, 7, in_channel}; // NHWC
|
||||
int weight_shape[4] = {in_channel, 3, 3, out_channel / groups};
|
||||
int output_shape[4] = {2, 7, 9, out_channel}; // NHWC
|
||||
int stride[2] = {1, 1};
|
||||
int padding[4] = {0, 0, 0, 0};
|
||||
int dilation[2]= {1, 1};
|
||||
float bias[] = {0, 0, 0, 0, 0, 0};
|
||||
int core_mask = 0b0001; // 私有存储版本只能设置为一个核心启动
|
||||
TestConvTransposeL2Fp32(input_shape, weight_shape, output_shape, stride, padding, dilation, groups, bias, core_mask);
|
||||
}
|
||||
|
|
@ -0,0 +1,139 @@
|
|||
Conv2DBackpropFilterFusion
|
||||
==========================
|
||||
计算二维卷积反向传播的权重梯度（Conv2D backprop filter fusion），支持常规卷积、Depthwise 卷积以及 1x1 优化路径，多核按批次与空间维度分块协同完成。
|
||||
|
||||
.. math::
|
||||
|
||||
dw = \text{Conv2DGradFilter}(x, dy)
|
||||
|
||||
输入：
|
||||
- **dy** - 输出梯度张量首地址，形状 ``[batch, out_h, out_w, out_channel]``。
|
||||
- **x** - 正向输入张量首地址，形状 ``[batch, in_h, in_w, in_channel]``。
|
||||
- **conv_param** - 卷积参数结构体地址，包含 ``stride``、``pad``、``dilation``、``group``、输入输出维度及共享工作空间指针等信息。
|
||||
|
||||
ConvParameter 字段说明：
|
||||
|
||||
- ``workspace_`` - 指向算子运行时使用的临时工作空间，需满足对齐与容量要求。
|
||||
- ``output_batch_`` - 输出梯度 ``dy`` 的批次数（通常等于输入批次数）。
|
||||
- ``input_batch_`` - 正向输入 ``x`` 的批次数，用于与 ``output_batch_`` 校验。
|
||||
- ``input_h_`` / ``input_w_`` - 正向输入特征图的高度与宽度。
|
||||
- ``output_h_`` / ``output_w_`` - 输出梯度特征图的高度与宽度。
|
||||
- ``input_channel_`` / ``output_channel_`` - 输入与输出通道数，需与 ``group_`` 配合满足整除关系。
|
||||
- ``kernel_h_`` / ``kernel_w_`` - 卷积核的高与宽。
|
||||
- ``group_`` - 组卷积数量，``group_ = 1`` 表示普通卷积。
|
||||
- ``pad_l_`` / ``pad_r_`` / ``pad_u_`` / ``pad_d_`` - 分别表示左右上下方向的填充大小。
|
||||
- ``dilation_h_`` / ``dilation_w_`` - 核心采样间隔（膨胀系数）。
|
||||
- ``stride_h_`` / ``stride_w_`` - 滑动窗口在高、宽方向的步长。
|
||||
- ``buffer_size_`` - 分配给 ``workspace_`` 的缓冲区字节数，在运行前需要正确设置。
|
||||
- ``nweights_`` - 卷积权重 ``w`` 的元素总数，用于内部分块和校验。
|
||||
|
||||
- **core_mask(int, 可选)** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **dw** - 卷积核梯度张量首地址，形状 ``[out_channel, in_channel/group, kernel_h, kernel_w]``。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持 fp32 数据类型。
|
||||
- MT7004 支持 fp16、fp32 数据类型。
|
||||
- 需在 ``conv_param->workspace_`` 中预先分配共享工作空间，长度不少于 ``conv_param->buffer_size_``。
|
||||
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void hp_conv2dbackpropfilterfusion_s(const half *dy, const half *x, half *dw, ConvParameter *conv_param, int core_mask)
|
||||
.. c:function:: void fp_conv2dbackpropfilterfusion_s(const float *dy, const float *x, float *dw, ConvParameter *conv_param, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 34
|
||||
|
||||
// FT78NE 多核示例
|
||||
#include <stdio.h>
|
||||
#include "conv_parameter.h"
|
||||
|
||||
int main(void) {
|
||||
const float *dy = (const float *)0xA0000000; // DDR 存储
|
||||
const float *x = (const float *)0xB0000000;
|
||||
float *dw = (float *)0xC0000000;
|
||||
ConvParameter *param = (ConvParameter *)0xB0001000;
|
||||
// 设置 ConvParameter 字段
|
||||
param->workspace_ = (void *)0xB0002000;
|
||||
param->buffer_size_ = 0x20000;
|
||||
param->input_batch_ = 1;
|
||||
param->input_h_ = 3;
|
||||
param->input_w_ = 3;
|
||||
param->input_channel_ = 4;
|
||||
param->output_batch_ = 1;
|
||||
param->output_h_ = 3;
|
||||
param->output_w_ = 3;
|
||||
param->output_channel_ = 4;
|
||||
param->kernel_h_ = 2;
|
||||
param->kernel_w_ = 2;
|
||||
param->group_ = 2;
|
||||
param->pad_u_ = 1;
|
||||
param->pad_d_ = 0;
|
||||
param->pad_l_ = 1;
|
||||
param->pad_r_ = 0;
|
||||
param->dilation_h_ = 1;
|
||||
param->dilation_w_ = 1;
|
||||
param->stride_h_ = 1;
|
||||
param->stride_w_ = 1;
|
||||
param->nweights_ = 4 * 2 * 2 * 2; // 示例值
|
||||
int core_mask = 0xff;
|
||||
fp_conv2dbackpropfilterfusion_s(dy, x, dw, param, core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void hp_conv2dbackpropfilterfusion_p(const half *dy, const half *x, half *dw, ConvParameter *conv_param)
|
||||
.. c:function:: void fp_conv2dbackpropfilterfusion_p(const float *dy, const float *x, float *dw, ConvParameter *conv_param)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 33
|
||||
|
||||
// MT7004 单核示例
|
||||
#include <stdio.h>
|
||||
#include "conv_parameter.h"
|
||||
|
||||
int main(void) {
|
||||
const half *dy = (const half *)0x10000000; // L2 存储
|
||||
const half *x = (const half *)0x10020000;
|
||||
half *dw = (half *)0x10040000;
|
||||
ConvParameter *param = (ConvParameter *)0x10060000;
|
||||
// 设置 ConvParameter 字段
|
||||
param->workspace_ = (void *)0x10070000;
|
||||
param->buffer_size_ = 0x10000;
|
||||
param->input_batch_ = 1;
|
||||
param->input_h_ = 3;
|
||||
param->input_w_ = 3;
|
||||
param->input_channel_ = 4;
|
||||
param->output_batch_ = 1;
|
||||
param->output_h_ = 3;
|
||||
param->output_w_ = 3;
|
||||
param->output_channel_ = 4;
|
||||
param->kernel_h_ = 2;
|
||||
param->kernel_w_ = 2;
|
||||
param->group_ = 2;
|
||||
param->pad_u_ = 1;
|
||||
param->pad_d_ = 0;
|
||||
param->pad_l_ = 1;
|
||||
param->pad_r_ = 0;
|
||||
param->dilation_h_ = 1;
|
||||
param->dilation_w_ = 1;
|
||||
param->stride_h_ = 1;
|
||||
param->stride_w_ = 1;
|
||||
param->nweights_ = 4 * 2 * 2 * 2; // 示例值
|
||||
hp_conv2dbackpropfilterfusion_p(dy, x, dw, param);
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,137 @@
|
|||
Conv2DBackpropInputFusion
|
||||
=========================
|
||||
计算二维卷积反向传播的输入梯度（Conv2D backprop input fusion），支持普通卷积、Depthwise 卷积以及 1x1 优化路径，多个核心通过核掩码协同完成批次并行。
|
||||
|
||||
.. math::
|
||||
|
||||
dx = \text{Conv2D}^\top(dy, w)
|
||||
|
||||
输入：
|
||||
- **dy** - 输出梯度张量首地址，形状 ``[batch, out_h, out_w, out_channel]``。
|
||||
- **w** - 卷积权重张量首地址，形状 ``[out_channel, kernel_h, kernel_w, in_channel/group]``。
|
||||
- **conv_param** - 卷积参数结构体地址，包含 ``stride``、``pad``、``dilation``、``group``、输入输出维度、批次数及共享工作空间指针等信息。
|
||||
|
||||
ConvParameter 字段说明：
|
||||
|
||||
- ``workspace_`` - 指向算子运行时使用的临时工作空间，需满足对齐与容量要求。
|
||||
- ``output_batch_`` - 输出梯度 ``dy`` 的批次数（通常等于输入批次数）。
|
||||
- ``input_batch_`` - 正向输入 ``x`` 的批次数，用于与 ``output_batch_`` 校验。
|
||||
- ``input_h_`` / ``input_w_`` - 正向输入特征图的高度与宽度。
|
||||
- ``output_h_`` / ``output_w_`` - 输出梯度特征图的高度与宽度。
|
||||
- ``input_channel_`` / ``output_channel_`` - 输入与输出通道数，需与 ``group_`` 配合满足整除关系。
|
||||
- ``kernel_h_`` / ``kernel_w_`` - 卷积核的高与宽。
|
||||
- ``group_`` - 组卷积数量，``group_ = 1`` 表示普通卷积。
|
||||
- ``pad_l_`` / ``pad_r_`` / ``pad_u_`` / ``pad_d_`` - 分别表示左右上下方向的填充大小。
|
||||
- ``dilation_h_`` / ``dilation_w_`` - 核心采样间隔（膨胀系数）。
|
||||
- ``stride_h_`` / ``stride_w_`` - 滑动窗口在高、宽方向的步长。
|
||||
- ``buffer_size_`` - 分配给 ``workspace_`` 的缓冲区字节数，在运行前需要正确设置。
|
||||
- ``nweights_`` - 卷积权重 ``w`` 的元素总数，用于内部分块和校验。
|
||||
|
||||
- **core_mask(int, 可选)** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **dx** - 输入梯度张量首地址，形状 ``[batch, in_h, in_w, in_channel]``。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持 fp32 数据类型。
|
||||
- MT7004 支持 fp16、fp32 数据类型。
|
||||
- 需在 ``conv_param->workspace_`` 中预先分配共享工作空间，并设置 ``conv_param->buffer_size_``。
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void hp_conv2dbackpropinputfusion_s(const half *dy, const half *w, half *dx, ConvParameter *conv_param, int core_mask)
|
||||
.. c:function:: void fp_conv2dbackpropinputfusion_s(const float *dy, const float *w, float *dx, ConvParameter *conv_param, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 34
|
||||
|
||||
// FT78NE 多核示例
|
||||
#include <stdio.h>
|
||||
#include "conv_parameter.h"
|
||||
|
||||
int main(void) {
|
||||
const float *dy = (const float *)0xA0000000; // DDR 存储
|
||||
const float *w = (const float *)0xB0000000;
|
||||
float *dx = (const float *)0xC0000000;
|
||||
ConvParameter *param = (ConvParameter *)0xB0001000; // 卷积参数共享区域
|
||||
// 设置 ConvParameter 字段
|
||||
param->workspace_ = (void *)0xB0002000; // 共享工作空间
|
||||
param->buffer_size_ = 0x20000;
|
||||
param->input_batch_ = 1;
|
||||
param->input_h_ = 5;
|
||||
param->input_w_ = 5;
|
||||
param->input_channel_ = 4;
|
||||
param->output_batch_ = 1;
|
||||
param->output_h_ = 3;
|
||||
param->output_w_ = 3;
|
||||
param->output_channel_ = 8;
|
||||
param->kernel_h_ = 3;
|
||||
param->kernel_w_ = 3;
|
||||
param->group_ = 1;
|
||||
param->pad_u_ = 1;
|
||||
param->pad_d_ = 1;
|
||||
param->pad_l_ = 1;
|
||||
param->pad_r_ = 1;
|
||||
param->dilation_h_ = 1;
|
||||
param->dilation_w_ = 1;
|
||||
param->stride_h_ = 2;
|
||||
param->stride_w_ = 2;
|
||||
param->nweights_ = 8 * 3 * 3 * 4; // 示例值
|
||||
int core_mask = 0xff;
|
||||
fp_conv2dbackpropinputfusion_s(dy, w, dx, param, core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void hp_conv2dbackpropinputfusion_p(const half *dy, const half *w, half *dx, ConvParameter *conv_param)
|
||||
.. c:function:: void fp_conv2dbackpropinputfusion_p(const float *dy, const float *w, float *dx, ConvParameter *conv_param)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 33
|
||||
|
||||
// MT7004 单核示例
|
||||
#include <stdio.h>
|
||||
#include "conv_parameter.h"
|
||||
|
||||
int main(void) {
|
||||
const half *dy = (const half *)0x10000000; // L2 存储
|
||||
const half *w = (const half *)0x10020000;
|
||||
half *dx = (half *)0x10040000;
|
||||
ConvParameter *param = (ConvParameter *)0x10060000;
|
||||
// 设置 ConvParameter 字段
|
||||
param->workspace_ = (void *)0x10070000;
|
||||
param->buffer_size_ = 0x10000;
|
||||
param->input_batch_ = 1;
|
||||
param->input_h_ = 5;
|
||||
param->input_w_ = 5;
|
||||
param->input_channel_ = 4;
|
||||
param->output_batch_ = 1;
|
||||
param->output_h_ = 3;
|
||||
param->output_w_ = 3;
|
||||
param->output_channel_ = 8;
|
||||
param->kernel_h_ = 3;
|
||||
param->kernel_w_ = 3;
|
||||
param->group_ = 1;
|
||||
param->pad_u_ = 1;
|
||||
param->pad_d_ = 1;
|
||||
param->pad_l_ = 1;
|
||||
param->pad_r_ = 1;
|
||||
param->dilation_h_ = 1;
|
||||
param->dilation_w_ = 1;
|
||||
param->stride_h_ = 2;
|
||||
param->stride_w_ = 2;
|
||||
param->nweights_ = 8 * 3 * 3 * 4; // 示例值
|
||||
hp_conv2dbackpropinputfusion_p(dy, w, dx, param);
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,64 @@
|
|||
Crop
|
||||
=================
|
||||
|
||||
类似于slice(张量切片)。不过只支持四维。
|
||||
|
||||
输入：
|
||||
- **input** - 输入数据地址。
|
||||
- **in_shape** - 输入张量形状。
|
||||
- **out_shape** - 输出张量形状。
|
||||
- **type_size** - 输入和输出张量数据类型的长度。
|
||||
- **offset** - 每一维度裁剪开始的偏移量
|
||||
- **axis** - 裁剪开始的维度
|
||||
- **core_mask** - 核掩码。
|
||||
|
||||
输出：
|
||||
- **output** - 输出地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持int8, fp32
|
||||
- MT7004 支持fp16, fp32
|
||||
|
||||
**共享/私有存储版本:**
|
||||
|
||||
.. c:function:: void anytype_crop_anycore(void *input, void *output, int *in_shape, int *out_shape, int type_size, int* offset, int axis, int core_mask)
|
||||
|
||||
各种数据类型、私有及共享空间版本均使用该函数。对于不同数据类型，改变type_size参数即可。
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 17
|
||||
|
||||
void TestCropSMCFp32(int* in_shape, int* out_shape, int axis, int* offset_, int core_mask) {
|
||||
int core_id = get_core_id();
|
||||
int logic_core_id = GetLogicCoreId(core_mask, core_id);
|
||||
int core_num = GetCoreNum(core_mask);
|
||||
float* input = (float*)0x88000000; // 测试私有空间时地址设置在私有空间内即可
|
||||
float* output = (float*)0x98000000;
|
||||
int* input_shape = (int*)0xA8000000;
|
||||
int* output_shape = (int*)0xA8200000;
|
||||
int type_size = sizeof(float);
|
||||
int* offset = (int*)0xA8410000;
|
||||
if (logic_core_id == 0) {
|
||||
memcpy(offset, offset_, sizeof(int) * (4 - axis));
|
||||
memcpy(input_shape, in_shape, sizeof(int) * 4);
|
||||
memcpy(output_shape, out_shape, sizeof(int) * 4);
|
||||
}
|
||||
sys_bar(0, core_num); // 初始化参数完成后进行同步
|
||||
anytype_crop_anycore(input, output, in_shape, out_shape, type_size, offset, axis, core_mask);
|
||||
}
|
||||
|
||||
void main(){
|
||||
int in_shape[4] = {2, 3, 3, 5};
|
||||
int out_shape[4] = {2, 2, 2, 5};
|
||||
int axis = 1;
|
||||
int offset[3] = {1, 1, 0};
|
||||
int core_mask = 0b1111; // 测试单核时核掩码设置为0b0001即可
|
||||
TestCropSMCFp32(in_shape, out_shape, axis, offset, core_mask);
|
||||
}
|
||||
|
|
@ -0,0 +1,94 @@
|
|||
CropAndResize
|
||||
=================
|
||||
|
||||
从输入图像Tensor中提取切片并调整其大小。仅支持双线性插值方法。
|
||||
|
||||
输入：
|
||||
- **src** - 输入数据的地址。
|
||||
- **box_idx** - boxes的索引，box_idx[i]的值表示第i个框的图像的值。
|
||||
- **boxes** - 第i行表示box_index[i]图像区域的坐标，并且坐标[y1,x1,y2,x2]是归一化后的值。归一化后的坐标值y，映射到图像y*(image_height-1)处，因此归一化后的图像高度范围为[0,1]，映射到实际图像高度范围为[0,image_height-1]。我们允许y1>y2，在这种情况下，视为原始图像的上下翻转变换。宽度尺寸的处理类似。坐标取值允许在[0,1]范围之外，在这种情况下，我们使用extrapolation_value外插值进行补齐。
|
||||
- **param** - 算子计算所需参数的结构体。其各成员见下述。
|
||||
- **extrapolation_value** - 外插值。
|
||||
- **core_mask** - 核掩码。
|
||||
|
||||
**CropAndResizeParameter定义：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
|
||||
typedef struct CropAndResizeParameter {
|
||||
int* input_shape_; // 输入张量形状
|
||||
int* output_shape_; // 输出张量形状
|
||||
int* x_lefts_; // 用于存储预处理结果
|
||||
int* x_rights_; // 用于存储预处理结果
|
||||
int* y_tops_; // 用于存储预处理结果
|
||||
int* y_bottoms_; // 用于存储预处理结果
|
||||
void* x_weights_; // 用于存储预处理结果
|
||||
void* y_weights_; // 用于存储预处理结果
|
||||
void* line_buffers_; // 用于存储中间结果
|
||||
} CropAndResizeParameter;
|
||||
|
||||
输出：
|
||||
- **output** - 输出地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持int8, fp32
|
||||
- MT7004 支持fp16, fp32
|
||||
|
||||
**共享/私有存储版本:**
|
||||
|
||||
.. c:function:: void i8_crop_and_resize_anycore(int8_t* src, int8_t* dst, int *box_idx, float *boxes, CropAndResizeParameter* param, float extrapolation_value, int core_mask)
|
||||
.. c:function:: void hp_crop_and_resize_anycore(half* src, half* dst, int *box_idx, float *boxes, CropAndResizeParameter* param, float extrapolation_value, int core_mask)
|
||||
.. c:function:: void fp_crop_and_resize_anycore(float* src, float* dst, int *box_idx, float *boxes, CropAndResizeParameter* param, half extrapolation_value, int core_mask)
|
||||
|
||||
私有及共享空间版本均使用这些函数。
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 28
|
||||
|
||||
void TestCropAndResizeSMCFp32(int* input_shape, int* output_shape, float* inp_boxes, int32_t* inp_box_idx, float extrapolation_value, int core_mask) {
|
||||
int core_id = get_core_id();
|
||||
int core_num = GetCoreNum(core_mask);
|
||||
int logic_core_id = GetLogicCoreId(core_mask, core_id);
|
||||
float* input = (float*)0x88000000; // 测试私有空间时地址设置在私有空间内即可
|
||||
float* output = (float*)0x89000000;
|
||||
float* boxes = (float*)0x8A000000;
|
||||
int* box_idx = (int*)0x8B000000;
|
||||
CropAndResizeParameter* param = (CropAndResizeParameter*)0x8C000000;
|
||||
if (logic_core_id == 0) {
|
||||
memcpy(boxes, inp_boxes, sizeof(float) * output_shape[0] * 4);
|
||||
memcpy(box_idx, inp_box_idx, sizeof(int) * output_shape[0]);
|
||||
param->input_shape_ = (int*)0x8D000000;
|
||||
memcpy(param->input_shape_, input_shape, sizeof(int) * 4);
|
||||
param->output_shape_ = (int*)0x8E000000;
|
||||
memcpy(param->output_shape_, output_shape, sizeof(int) * 4);
|
||||
param->line_buffers_ = (void*)0x8F000000;
|
||||
param->x_lefts_ = (int*)0x90000000;
|
||||
param->x_rights_ = (int*)0x91000000;
|
||||
param->y_bottoms_ = (int*)0x92000000;
|
||||
param->y_tops_ = (int*)0x93000000;
|
||||
param->x_weights_ = (void*)0x94000000;
|
||||
param->y_weights_ = (void*)0x95000000;
|
||||
PrepareCropAndResizeBilinear(param->input_shape_, boxes, param->output_shape_, param->y_bottoms_, param->y_tops_,
|
||||
param->x_lefts_, param->x_rights_, param->y_weights_, param->x_weights_); // 做预处理
|
||||
}
|
||||
sys_bar(0, core_num); // 初始化参数完成后进行同步
|
||||
fp_crop_and_resize_anycore(input, output, box_idx, boxes, param, extrapolation_value, core_mask);
|
||||
}
|
||||
|
||||
void main(){
|
||||
int input_shape[4] = {1, 4, 4, 4};
|
||||
int output_shape[4] = {1, 8, 8, 4};
|
||||
float boxes[4] = {0, 0, 0.5, 0.5};
|
||||
int box_idx[1] = {0};
|
||||
int core_mask = 0b1111; // 测试单核时核掩码设置为0b0001即可
|
||||
float extrapolation_value = 0.5;
|
||||
TestCropAndResizeSMCFp32(input_shape, output_shape, boxes, box_idx, extrapolation_value, core_mask);
|
||||
}
|
||||
|
|
@ -0,0 +1,84 @@
|
|||
DepthToSpace
|
||||
=================
|
||||
将输入张量的深度通道按 block_size 分解并重排到空间维度（Depth -> Space）。
|
||||
|
||||
输入：
|
||||
- **input** - 输入数据地址。
|
||||
- **in_shape** - 输入形状，格式为 ``[batch, height, width, channel]``。
|
||||
- **block_size** - block 因子（单个整数）。
|
||||
- **data_size** - 单个元素字节数（例如 sizeof(float)）。
|
||||
- **core_mask(int, 可选)** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **output** - 输出数据地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持的数据类型： fp32、fp64、cplx64、cplx128、int16、int8、int32。
|
||||
- MT7004 支持的数据类型： fp32、fp16、cplx64、int16、int32。
|
||||
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void i8_depthtospace_s(int8_t *input, int8_t *output, const int *in_shape, int block_size, int data_size, int core_mask)
|
||||
.. c:function:: void i16_depthtospace_s(int16_t *input, int16_t *output, const int *in_shape, int block_size, int data_size, int core_mask)
|
||||
.. c:function:: void i32_depthtospace_s(int32_t *input, int32_t *output, const int *in_shape, int block_size, int data_size, int core_mask)
|
||||
.. c:function:: void hp_depthtospace_s(half *input, half *output, const int *in_shape, int block_size, int data_size, int core_mask)
|
||||
.. c:function:: void fp_depthtospace_s(float *input, float *output, const int *in_shape, int block_size, int data_size, int core_mask)
|
||||
.. c:function:: void dp_depthtospace_s(double *input, double *output, const int *in_shape, int block_size, int data_size, int core_mask)
|
||||
.. c:function:: void c64_depthtospace_s(float *input, float *output, const int *in_shape, int block_size, int data_size, int core_mask)
|
||||
.. c:function:: void c128_depthtospace_s(double *input, double *output, const int *in_shape, int block_size, int data_size, int core_mask)
|
||||
|
||||
**C 调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 10
|
||||
|
||||
// FT78NE 多核示例
|
||||
#include <stdio.h>
|
||||
|
||||
int main(int argc, char *argv[]) {
|
||||
float *input = (float *)0xA0000000; // 多核版本：输入放在 DDR 地址 0xA0000000
|
||||
float *output = (float *)0xB0000000; // 多核版本：输出放在 DDR 地址 0xB0000000
|
||||
int in_shape[4] = {10, 16, 16, 4};
|
||||
int block_size = 2;
|
||||
int core_mask = 0xff;
|
||||
fp_depthtospace_s(input, output, in_shape, block_size, sizeof(float), core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void i8_depthtospace_p(int8_t *input, int8_t *output, const int *in_shape, int block_size, int data_size)
|
||||
.. c:function:: void i16_depthtospace_p(int16_t *input, int16_t *output, const int *in_shape, int block_size, int data_size)
|
||||
.. c:function:: void i32_depthtospace_p(int32_t *input, int32_t *output, const int *in_shape, int block_size, int data_size)
|
||||
.. c:function:: void hp_depthtospace_p(half *input, half *output, const int *in_shape, int block_size, int data_size)
|
||||
.. c:function:: void fp_depthtospace_p(float *input, float *output, const int *in_shape, int block_size, int data_size)
|
||||
.. c:function:: void dp_depthtospace_p(double *input, double *output, const int *in_shape, int block_size, int data_size)
|
||||
.. c:function:: void c64_depthtospace_p(float *input, float *output, const int *in_shape, int block_size, int data_size)
|
||||
.. c:function:: void c128_depthtospace_p(double *input, double *output, const int *in_shape, int block_size, int data_size)
|
||||
|
||||
**C 调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 9
|
||||
|
||||
// FT78NE 单核示例
|
||||
#include <stdio.h>
|
||||
|
||||
int main(int argc, char *argv[]) {
|
||||
float *input = (float *)0x10000000; // 单核版本：输入放在 L2 地址 0x10000000
|
||||
float *output = (float *)0x10040000; // 单核版本：输出放在 L2 地址 0x10040000
|
||||
int in_shape[4] = {10, 16, 16, 4};
|
||||
int block_size = 2;
|
||||
fp_depthtospace_p(input, output, in_shape, block_size, sizeof(float));
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
|
|
@ -3,5 +3,50 @@ DSP Library C API Reference
|
|||
|
||||
.. toctree::
|
||||
:maxdepth: 1
|
||||
|
||||
equal
|
||||
|
||||
equal
|
||||
activation
|
||||
expfusion
|
||||
reverse_sequence
|
||||
reversev2
|
||||
fillv2
|
||||
squeeze
|
||||
unsqueeze
|
||||
expand_dims
|
||||
leaky_relu
|
||||
lstm
|
||||
scatter_elements
|
||||
reduce
|
||||
resize
|
||||
crop_and_resize
|
||||
crop
|
||||
conv2d
|
||||
adder
|
||||
conv2d_transpose
|
||||
gru
|
||||
assert
|
||||
range
|
||||
raggedrange
|
||||
linspace
|
||||
matmulfusion
|
||||
attention
|
||||
floor
|
||||
floordiv
|
||||
embeddinglookup
|
||||
eltwise
|
||||
adamweightdecay
|
||||
applymomentum
|
||||
avgpoolinggrad
|
||||
batchtospace
|
||||
batchtospacend
|
||||
broadcastto
|
||||
depthtospace
|
||||
spacetodepth
|
||||
spacetobatch
|
||||
spacetobatchnd
|
||||
fusedbatchnorm
|
||||
groupnormfusion
|
||||
conv2dbackpropinputfusion
|
||||
conv2dbackpropfilterfusion
|
||||
sgd
|
||||
scalefusion
|
||||
|
|
|
|||
|
|
@ -0,0 +1,99 @@
|
|||
Eltwise
|
||||
=================
|
||||
|
||||
输入两个等长数组以及控制参数，根据控制参数的值决定对两个数组做对位相加、对位相乘或取最大值操作。
|
||||
|
||||
.. math::
|
||||
|
||||
\mathbf{output_i} =
|
||||
\begin{cases}
|
||||
\mathbf{Input0_i} \cdot \mathbf{Input1_i}, & \text{if } \text{eltwise_mode} = \text{Eltwise_PROD} \\[6pt]
|
||||
\mathbf{Input0_i} + \mathbf{Input1_i}, & \text{if } \text{eltwise_mode} = \text{Eltwise_SUM} \\[6pt]
|
||||
\max(\mathbf{Input0_i}, \mathbf{Input1_i}), & \text{if } \text{eltwise_mode} = \text{Eltwise_MAXIMUM}
|
||||
\end{cases}
|
||||
|
||||
输入：
|
||||
- **Input0** - 第一个输入数据地址。
|
||||
- **Input1** - 第二个输入数据地址。
|
||||
- **length** - 计算长度。
|
||||
- **core_mask(int, 可选)** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **output** - 计算结果地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持int8, int16, int32, fp32, fp64, cplx64(除最大值), cplx128(除最大值)
|
||||
- MT7004 支持fp16, fp32, int16, int32, cplx64(除最大值)
|
||||
|
||||
**共享存储版本：**
|
||||
|
||||
.. c:function:: void i8_eltwise_s(int8_t* Input0, int8_t* Input1, int8_t* output,int length, int eltwise_mode_, int core_mask)
|
||||
.. c:function:: void i16_eltwise_s(int16_t* Input0, int16_t* Input1, int16_t* output,int length, int eltwise_mode_, int core_mask)
|
||||
.. c:function:: void i32_eltwise_s(int* Input0, int* Input1, int* output,int length, int eltwise_mode_, int core_mask)
|
||||
.. c:function:: void hp_eltwise_s(half* Input0, half* Input1, half* output,int length, int eltwise_mode_, int core_mask)
|
||||
.. c:function:: void fp_eltwise_s(float* Input0, float* Input1, float* output,int length, int eltwise_mode_, int core_mask)
|
||||
.. c:function:: void dp_eltwise_s(double* Input0, double* Input1, double* output,int length, int eltwise_mode_, int core_mask)
|
||||
.. c:function:: void c64_eltwise_s(float* Input0, float* Input1, float* output,int length, int eltwise_mode_, int core_mask)
|
||||
.. c:function:: void c128_eltwise_s(double* Input0, double* Input1, double* output, int length, int eltwise_mode_, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 14
|
||||
|
||||
//FT78NE示例
|
||||
#include <stdio.h>
|
||||
#include <eltwise.h>
|
||||
#define Eltwise_PROD 0
|
||||
#define Eltwise_SUM 1
|
||||
#define Eltwise_MAXIMUM 2
|
||||
int main(int argc, char* argv[]) {
|
||||
float *input0 = (float *)0xA0000000; //input在DDR空间
|
||||
float *input1 = (float *)0xB0000000;
|
||||
float *output = (float *)0xC0000000;
|
||||
int length = 1000;
|
||||
int eltwise_mode_ = Eltwise_SUM;
|
||||
int core_mask = 0xff;
|
||||
fp_eltwise_s(input0, input1, output, length, eltwise_mode_, core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
**私有存储版本：**
|
||||
|
||||
.. c:function:: void i8_eltwise_p(int8_t *Input0, int8_t *Input1, int8_t *output, int eltwise_mode_, int length)
|
||||
.. c:function:: void i16_eltwise_p(int16_t *Input0, int16_t *Input1, int16_t *output, int eltwise_mode_, int length)
|
||||
.. c:function:: void i32_eltwise_p(int32_t *Input0, int32_t *Input1, int32_t *output, int eltwise_mode_, int length)
|
||||
.. c:function:: void hp_eltwise_p(half* Input0, half* Input1, bool* output, int eltwise_mode_,int length)
|
||||
.. c:function:: void fp_eltwise_p(float* Input0, float* Input1, float* output, int eltwise_mode_,int length)
|
||||
.. c:function:: void dp_eltwise_p(double* Input0, double* Input1, double* output, int eltwise_mode_,int length)
|
||||
.. c:function:: void c64_eltwise_p(float *Input0, float *Input1, float *output, int eltwise_mode_, int length)
|
||||
.. c:function:: void c128_eltwise_p(double *Input0, double *Input1, double *output, int eltwise_mode_, int length)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 14
|
||||
|
||||
//FT78NE示例
|
||||
#include <stdio.h>
|
||||
#include <eltwise.h>
|
||||
#define Eltwise_PROD 0
|
||||
#define Eltwise_SUM 1
|
||||
#define Eltwise_MAXIMUM 2
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
float *input0 = (float *)0x10810000; //input在L2空间
|
||||
float *input1 = (float *)0x10820000;
|
||||
float *output = (float *)0x10830000;
|
||||
int length = 1000;
|
||||
int eltwise_mode_ = Eltwise_SUM;
|
||||
fp_eltwise_p(input0, input1, output, eltwise_mode_, length);
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,101 @@
|
|||
EmbeddingLookup
|
||||
=================
|
||||
|
||||
传入一个矩阵和一组索引，根据给定的索引提取对应的行（向量），若此行不曾被标记为“已正则化”，则对此行进行正则化处理，将结果拼接输出，若已经正则化过则直接输出。
|
||||
|
||||
.. math::
|
||||
|
||||
\forall k \in [1, ids\_size], \quad
|
||||
\begin{cases}
|
||||
\text{if } \textbf{is_regulated}[i_k] = 0, &
|
||||
\begin{cases}
|
||||
\displaystyle X_{i_k} \leftarrow
|
||||
X_{i_k} \cdot \frac{\text{max_norm}}
|
||||
{\sum_{j=1}^{layer\_size\_} X_{i_k, j}} \\[10pt]
|
||||
\textbf{is_regulated}[i_k] \leftarrow 1
|
||||
\end{cases} \\[12pt]
|
||||
\text{输出向量 } Y_k \leftarrow X_{i_k}
|
||||
\end{cases}
|
||||
|
||||
输入：
|
||||
- **input_data** - 输入矩阵数据地址。
|
||||
- **ids** - 输入索引的存储地址。
|
||||
- **max_norm** - 最大范数约束。
|
||||
- **is_regulated** - 记录矩阵行是否被正则化的标志数组。
|
||||
- **ids_size_** - 输入索引个数。
|
||||
- **layer_size_** - 输入矩阵的列数。
|
||||
- **layer_num_** - 输入矩阵的行数。
|
||||
- **core_mask(int, 可选)** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **output** - 结果输出地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持fp32
|
||||
- MT7004 支持fp16, fp32
|
||||
|
||||
**共享存储版本：**
|
||||
|
||||
.. c:function:: void hp_embeddinglookup_s(half* input_data,int* ids, half* output, half max_norm_, bool* is_regulated, int ids_size_, int layer_size_, int layer_num_ , int core_mask)
|
||||
.. c:function:: void fp_embeddinglookup_s(float* input_data,int* ids, float* output, float max_norm_, bool* is_regulated, int ids_size_, int layer_size_, int layer_num_ , int core_mask)
|
||||
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 16
|
||||
|
||||
//FT78NE示例
|
||||
#include <stdio.h>
|
||||
#include <embeddinglookup.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
float *input_data = (float *)0xA0000000; //input在DDR空间
|
||||
float *output_data = (float *)0xA0872c00;
|
||||
int layer_size = 4;//列
|
||||
int layer_num = 5;//行
|
||||
float max_norm = 4.5;
|
||||
int ids[]={0,1,3};
|
||||
int ids_size = 3;//提取三行
|
||||
bool *output = (bool *)0xC0000000;
|
||||
bool is_regulated_[5] = {0};//layer_num
|
||||
int core_mask = 0xff;
|
||||
fp_embeddinglookup_s(input_data, ids, output_data, max_norm, is_regulated_, ids_size, layer_size, layer_num, core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
**私有存储版本：**
|
||||
|
||||
.. c:function:: void hp_embeddinglookup_p(half* Input0, half* Input1, bool* output,int length)
|
||||
.. c:function:: void fp_embeddinglookup_p(float* Input0, float* Input1, bool* output,int length)
|
||||
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 15
|
||||
|
||||
//FT78NE示例
|
||||
#include <stdio.h>
|
||||
#include <embeddinglookup.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
float *input_data = (float *)0x10810000; //input在DDR空间
|
||||
float *output_data = (float *)0x10820000;
|
||||
int layer_size = 4;//列
|
||||
int layer_num = 5;//行
|
||||
float max_norm = 4.5;
|
||||
int ids[]={0,1,3};
|
||||
int ids_size = 3;//提取三行
|
||||
bool *output = (bool *)0xC0000000;
|
||||
bool is_regulated_[5] = {0};//layer_num
|
||||
fp_embeddinglookup_s(input_data, ids, output_data, max_norm, is_regulated_, ids_size, layer_size, layer_num);
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -42,7 +42,7 @@ Equal
|
|||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 10
|
||||
:emphasize-lines: 11
|
||||
|
||||
//FT78NE示例
|
||||
#include <stdio.h>
|
||||
|
|
@ -74,16 +74,16 @@ Equal
|
|||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 9
|
||||
:emphasize-lines: 10
|
||||
|
||||
//FT78NE示例
|
||||
#include <stdio.h>
|
||||
#include <equal.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
float *input0 = (float *)0x10000000; //input在L2空间
|
||||
float *input1 = (float *)0x10001000;
|
||||
bool *output = (bool *)0xC0000000;
|
||||
float *input0 = (float *)0x10810000; //input在L2空间
|
||||
float *input1 = (float *)0x10820000;
|
||||
bool *output = (bool *)0x10830000;
|
||||
int length = 1000;
|
||||
fp_equal_p(input0, input1, output, length);
|
||||
return 0;
|
||||
|
|
|
|||
|
|
@ -0,0 +1,42 @@
|
|||
ExpandDims
|
||||
=================
|
||||
|
||||
对输入张量在给定的轴上添加额外维度。由于该算子仅改变张量形状，因此其DSP算子的作用是将数据从输入张量完整拷贝到输出张量。
|
||||
|
||||
输入：
|
||||
- **src** - 输入地址
|
||||
- **total_copy_size** - 计算得到的总共需拷贝的数据量，单位为字节。
|
||||
- **core_mask** - 核掩码。
|
||||
|
||||
输出：
|
||||
- **dst** - 输出地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持int8, int16, int32, fp32, fp64, cplx64, cplx128
|
||||
- MT7004 支持fp16, fp32, int16, int32, cplx64
|
||||
|
||||
**共享/私有存储版本:**
|
||||
|
||||
.. c:function:: void anytype_expand_dims_anycore(void* src, void* dst, int total_copy_size, int core_mask)
|
||||
|
||||
各种数据类型、私有及共享空间版本均使用该函数。
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 8
|
||||
|
||||
void main(){
|
||||
int core_mask = 0b1111; // 测试单核时核掩码设置为0b0001即可
|
||||
int core_num = GetCoreNum(core_mask);
|
||||
float* src = (float*)0x88000000; // 测试私有空间时地址设置在私有空间内即可
|
||||
float* dst = (float*)0x98000000;
|
||||
int shape[3] = {1, 10, 10};
|
||||
int total_copy_size = shape[0] * shape[1] * shape[2] * sizeof(float);
|
||||
anytype_expand_dims_anycore(src, dst, total_copy_size, core_mask);
|
||||
}
|
||||
|
|
@ -0,0 +1,101 @@
|
|||
ExpFusion
|
||||
=================
|
||||
|
||||
|
||||
|
||||
传入一个数组，逐元素计算其乘上输入因子（可选择）后的指数值，再将指数值乘上输出因子后输出。
|
||||
|
||||
.. math::
|
||||
|
||||
dst_i = \exp(src_i \cdot s_{in}) \cdot s_{out}
|
||||
\quad \text{where} \quad
|
||||
s_{in} =
|
||||
\begin{cases}
|
||||
1, & scale = 1 \\
|
||||
in\_scale, & scale \neq 1
|
||||
\end{cases}
|
||||
|
||||
\quad s_{out} = out\_scale
|
||||
|
||||
输入：
|
||||
- **src_data** - 输入数据地址。
|
||||
- **length** - 计算长度。
|
||||
- **in_scale** - 输入缩放因子，当scale != 1时启用。
|
||||
- **out_scale** - 输出缩放因子。
|
||||
- **scale** - 输入缩放因子启用控制。
|
||||
- **core_mask** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **dst_data** - 计算结果地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持int8, int16, int32, fp32, fp64, cplx64, cplx128
|
||||
- MT7004 支持fp16, fp32, int16, int32, cplx64
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void i8_expfusion_s(int8_t* src_data, int8_t* dst_data, int length, float in_scale, float out_scale, int scale, int core_mask)
|
||||
.. c:function:: void i16_expfusion_s(int16_t* src_data, half* dst_data, int length, float in_scale, float out_scale, int scale, int core_mask)
|
||||
.. c:function:: void i32_expfusion_s(int* src_data, float* dst_data, int length, float in_scale, float out_scale, int scale, int core_mask)
|
||||
.. c:function:: void hp_expfusion_s(half* src_data, half* dst_data, int length, float in_scale, float out_scale, int scale, int core_mask)
|
||||
.. c:function:: void fp_expfusion_s(float* src_data, float* dst_data, int length, float in_scale, float out_scale, int scale, int core_mask)
|
||||
.. c:function:: void dp_expfusion_s(double* src_data, double* dst_data, int length, float in_scale, float out_scale, int scale, int core_mask)
|
||||
.. c:function:: void c64_expfusion_s(float* src_data, float* dst_data, int length, float in_scale, float out_scale, int scale, int core_mask)
|
||||
.. c:function:: void c128_expfusion_s(double* src_data, double* dst_data, int length, float in_scale, float out_scale, int scale, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 12
|
||||
|
||||
//FT78NE示例
|
||||
#include <stdio.h>
|
||||
#include <expfusion.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
float *input0 = (float *)0xA0000000; //input在DDR空间
|
||||
float *output = (float *)0xC0000000;
|
||||
int length = 1000;
|
||||
float in_scale = 0.5, out_scale = 1.2;
|
||||
int scale = 1;
|
||||
int core_mask = 0xff;
|
||||
fp_expfusion_s( input0, output, length, in_scale, out_scale, scale,core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void i8_expfusion_p(int8_t* src_data, int8_t* dst_data, int length, float in_scale, float out_scale, int scale)
|
||||
.. c:function:: void i16_expfusion_p(int16_t* src_data, half* dst_data, int length, float in_scale, float out_scale, int scale)
|
||||
.. c:function:: void i32_expfusion_p(int* src_data, float* dst_data, int length, float in_scale, float out_scale, int scale)
|
||||
.. c:function:: void hp_expfusion_p(half* src_data, half* dst_data, int length, float in_scale, float out_scale, int scale)
|
||||
.. c:function:: void fp_expfusion_p(float* src_data, float* dst_data, int length, float in_scale, float out_scale, int scale)
|
||||
.. c:function:: void dp_expfusion_p(double* src_data, double* dst_data, int length, float in_scale, float out_scale, int scale)
|
||||
.. c:function:: void c64_expfusion_p(float* src_data, float* dst_data, int length, float in_scale, float out_scale, int scale)
|
||||
.. c:function:: void c128_expfusion_p(double* src_data, double* dst_data, int length, float in_scale, float out_scale, int scale)
|
||||
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 10
|
||||
|
||||
//FT78NE示例
|
||||
#include <stdio.h>
|
||||
#include <expfusion.h>
|
||||
int main(int argc, char* argv[]) {
|
||||
float *input0 = (float *)0x10810000; //input在L2空间
|
||||
float *output = (float *)0x10820000;
|
||||
int length = 1000;
|
||||
float in_scale = 0.5, out_scale = 1.2;
|
||||
int scale = 1;
|
||||
fp_expfusion_p( input0, output, length, in_scale, out_scale, scale);
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,71 @@
|
|||
FillV2
|
||||
=================
|
||||
|
||||
创建一个Tensor，根据指定的shape，将其值由value进行填充。
|
||||
|
||||
输入：
|
||||
- **value** - 填充值的地址
|
||||
- **length** - 由指定shape计算出的张量总长度。
|
||||
- **type_size** - 填充值的数据类型的长度
|
||||
- **core_mask** - 核掩码。
|
||||
|
||||
输出：
|
||||
- **output** - 输出地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持int8, int16, int32, fp32, fp64, cplx64, cplx128
|
||||
- MT7004 支持fp16, fp32, int16, int32, cplx64
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void anytype_fillv2_s(void* value, void* output, int length, int type_size, int core_mask)
|
||||
|
||||
对于不同数据类型，改变type_size参数即可。
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 15
|
||||
|
||||
void main() {
|
||||
float* input = (float*)0xA0000000;
|
||||
float value = 789.1;
|
||||
float* output = (float*)0x88000000;
|
||||
int i;
|
||||
int length = 1000;
|
||||
int core_mask = 0b1111;
|
||||
int core_id = get_core_id();
|
||||
int logic_core_id = GetLogicCoreId(core_mask, core_id);
|
||||
int core_num = GetCoreNum(core_mask);
|
||||
if (logic_core_id == 0) {
|
||||
*input = value;
|
||||
}
|
||||
sys_bar(0, core_num); // 初始化参数完成后进行同步
|
||||
anytype_fillv2_s(input, output, length, 4, core_mask);
|
||||
}
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void anytype_fillv2_p(void* value, void* output, int length, int type_size, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 9
|
||||
|
||||
void main() {
|
||||
float* input = (float*)0x10000000;
|
||||
float value = 789.1;
|
||||
float* output = (float*)0x10010000;
|
||||
int i;
|
||||
int length = 1000;
|
||||
int core_mask = 0b0001; // 要启动哪一个核，就将哪一位设置为1，只允许存在一个核心启动
|
||||
*input = value;
|
||||
anytype_fillv2_p(input, output, length, 4, core_mask);
|
||||
}
|
||||
|
|
@ -0,0 +1,78 @@
|
|||
Floor
|
||||
=================
|
||||
|
||||
|
||||
|
||||
传入一个浮点型数组，对于数组中每一个元素执行向下取整操作。
|
||||
|
||||
.. math::
|
||||
|
||||
dst_i = floorf(src_i )
|
||||
|
||||
输入：
|
||||
- **input** - 输入数据地址。
|
||||
- **length** - 计算长度。
|
||||
- **core_mask** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **output** - 计算结果地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持fp32, fp64
|
||||
- MT7004 支持fp16, fp32
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
|
||||
.. c:function:: void hp_floor_s(half* src_data, half* dst_data, int length, int core_mask)
|
||||
.. c:function:: void fp_floor_s(float* src_data, float* dst_data, int length, int core_mask)
|
||||
.. c:function:: void dp_floor_s(double* src_data, double* dst_data, int length, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 10
|
||||
|
||||
//FT78NE示例
|
||||
#include <stdio.h>
|
||||
#include <floor.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
float *input0 = (float *)0xA0000000; //input在DDR空间
|
||||
float *output = (float *)0xC0000000;
|
||||
int length = 1000;
|
||||
int core_mask = 0xff;
|
||||
fp_floor_s( input0, output, length,core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void hp_floor_p(half* src_data, half* dst_data, int length)
|
||||
.. c:function:: void fp_floor_p(float* src_data, float* dst_data, int length)
|
||||
.. c:function:: void dp_floor_p(double* src_data, double* dst_data, int length)
|
||||
|
||||
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 8
|
||||
|
||||
//FT78NE示例
|
||||
#include <stdio.h>
|
||||
#include <floor.h>
|
||||
int main(int argc, char* argv[]) {
|
||||
float *input0 = (float *)0x10000000; //input在L2空间
|
||||
float *output = (float *)0xC0000000;
|
||||
int length = 1000;
|
||||
fp_floor_p( input0, output, length);
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,81 @@
|
|||
FloorDiv
|
||||
=================
|
||||
|
||||
|
||||
|
||||
传入两个等长的浮点型数组，对于数组中对位元素做除法，再对除法所得结果执行向下取整操作。
|
||||
|
||||
.. math::
|
||||
|
||||
dst_i = floorf(\frac{input0_i}{input1_i})
|
||||
|
||||
输入：
|
||||
- **input0** - 被除数数据地址。
|
||||
- **input1** - 除数数据地址。
|
||||
- **length** - 计算长度。
|
||||
- **core_mask** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **output** - 计算结果地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持fp32, fp64
|
||||
- MT7004 支持fp16, fp32
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
|
||||
.. c:function:: void hp_floordiv_s(half* src_data0, half* src_data1, half* dst_data, int length, int core_mask)
|
||||
.. c:function:: void fp_floordiv_s(float* src_data0, float* src_data1, float* dst_data, int length, int core_mask)
|
||||
.. c:function:: void dp_floordiv_s(double* src_data0, double* src_data1, double* dst_data, int length, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 11
|
||||
|
||||
//FT78NE示例
|
||||
#include <stdio.h>
|
||||
#include <floordiv.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
float *input0 = (float *)0xA0000000; //input在DDR空间
|
||||
float *input1 = (float *)0xB0000000;
|
||||
float *output = (float *)0xC0000000;
|
||||
int length = 1000;
|
||||
int core_mask = 0xff;
|
||||
fp_floordiv_s( input0,input1, output, length,core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void hp_floordiv_p(half* src_data0, half* src_data1, half* dst_data, int length)
|
||||
.. c:function:: void fp_floordiv_p(float* src_data0, float* src_data1, float* dst_data, int length)
|
||||
.. c:function:: void dp_floordiv_p(double* src_data0, double* src_data1, double* dst_data, int length)
|
||||
|
||||
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 9
|
||||
|
||||
//FT78NE示例
|
||||
#include <stdio.h>
|
||||
#include <floordiv.h>
|
||||
int main(int argc, char* argv[]) {
|
||||
float *input0 = (float *)0x10810000; //input在L2空间
|
||||
float *input1 = (float *)0x10820000;
|
||||
float *output = (float *)0x10830000;
|
||||
int length = 100;
|
||||
fp_floordiv_p( input0, input1, output, length);
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,91 @@
|
|||
FusedBatchNorm
|
||||
=================
|
||||
对输入张量执行融合批归一化（Fused Batch Normalization），在多核间拆分批量单元并行完成归一化与仿射变换。
|
||||
|
||||
.. math::
|
||||
|
||||
\hat{x}_{b,c} = \frac{x_{b,c} - mean_c}{\sqrt{variance_c + \epsilon}}, \quad y_{b,c} = scale_c \cdot \hat{x}_{b,c} + offset_c
|
||||
|
||||
|
||||
输入：
|
||||
- **input** - 输入张量首地址，形状为 ``[unit, channel]``。
|
||||
- **scale** - 缩放系数数组首地址，长度为 ``channel``。
|
||||
- **offset** - 平移系数数组首地址，长度为 ``channel``。
|
||||
- **mean** - 归一化均值数组首地址，长度为 ``channel``。
|
||||
- **variance** - 归一化方差数组首地址，长度为 ``channel``。
|
||||
- **epsilon** - 数值稳定项。
|
||||
- **channel** - 通道数。
|
||||
- **unit** - 归一化单元数量（批量大小 × 高 × 宽）。
|
||||
- **core_mask(int, 可选)** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **output** - 写回融合批归一化计算结果的张量首地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持 fp32 数据类型。
|
||||
- MT7004 支持 fp16、fp32 数据类型。
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void hp_fusedbatchnorm_s(const half *input, const half *scale, const half *offset, const half *mean, const half *variance, float epsilon, int channel, int unit, int core_mask, half *output)
|
||||
.. c:function:: void fp_fusedbatchnorm_s(const float *input, const float *scale, const float *offset, const float *mean, const float *variance, float epsilon, int channel, int unit, int core_mask, float *output)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 15
|
||||
|
||||
// FT78NE 多核示例
|
||||
#include <stdio.h>
|
||||
|
||||
int main(void) {
|
||||
const float *input = (const float *)0xA0000000; // DDR 存储
|
||||
const float *scale = (const float *)0xB0000000;
|
||||
const float *offset = (const float *)0xB0001000;
|
||||
const float *mean = (const float *)0xB0002000;
|
||||
const float *variance = (const float *)0xB0003000;
|
||||
float *output = (float *)0xC0000000;
|
||||
int channel = 64;
|
||||
int unit = 1024;
|
||||
float epsilon = 1e-5f;
|
||||
int core_mask = 0xff;
|
||||
fp_fusedbatchnorm_s(input, scale, offset, mean, variance,
|
||||
epsilon, channel, unit, core_mask,
|
||||
output);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void hp_fusedbatchnorm_p(const half *input, const half *scale, const half *offset, const half *mean, const half *variance, float epsilon, int channel, int unit, half *output)
|
||||
.. c:function:: void fp_fusedbatchnorm_p(const float *input, const float *scale, const float *offset, const float *mean, const float *variance, float epsilon, int channel, int unit, float *output)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 14
|
||||
|
||||
// MT7004 单核示例
|
||||
#include <stdio.h>
|
||||
|
||||
int main(void) {
|
||||
const half *input = (const half *)0x10000000; // L2 存储
|
||||
const half *scale = (const half *)0x10004000;
|
||||
const half *offset = (const half *)0x10008000;
|
||||
const half *mean = (const half *)0x1000C000;
|
||||
const half *variance = (const half *)0x10010000;
|
||||
half *output = (half *)0x10014000;
|
||||
int channel = 32;
|
||||
int unit = 512;
|
||||
float epsilon = 1e-4f;
|
||||
hp_fusedbatchnorm_p(input, scale, offset, mean, variance,
|
||||
epsilon, channel, unit, output);
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,100 @@
|
|||
GroupNormFusion
|
||||
=================
|
||||
对输入张量执行分组归一化融合操作（Group Normalization Fusion），在多核环境中按批次拆分并行完成均值、方差与归一化计算。
|
||||
|
||||
.. math::
|
||||
|
||||
\hat{x}_{(b,u,c)} = \frac{x_{(b,u,c)} - \mu_{(b,g)}}{\sqrt{\sigma^2_{(b,g)} + \epsilon}}, \quad y_{(b,u,c)} = \hat{x}_{(b,u,c)} \cdot scale_c + offset_c
|
||||
|
||||
其中 :math:`(b,u,c)` 表示批次、空间位置与通道索引，:math:`g` 为通道所属分组。
|
||||
|
||||
输入：
|
||||
- **input** - 输入张量首地址，形状 ``[batch, unit, channel]``。
|
||||
- **scale** - 通道缩放系数首地址，长度为 ``channel``。
|
||||
- **offset** - 通道偏移系数首地址，长度为 ``channel``。
|
||||
- **mean** - 批次 × 分组的均值缓冲区首地址，长度为 ``batch * num_groups``。
|
||||
- **variance** - 批次 × 分组的方差缓冲区首地址，长度为 ``batch * num_groups``。
|
||||
- **epsilon** - 数值稳定项。
|
||||
- **num_groups** - 分组数。
|
||||
- **channel** - 通道总数。
|
||||
- **unit** - 每批次内的归一化单元数（H×W）。
|
||||
- **batch** - 批次数。
|
||||
- **core_mask(int, 可选)** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **output** - 写回分组归一化结果的张量首地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持 fp32 数据类型。
|
||||
- MT7004 支持 fp16、fp32 数据类型。
|
||||
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void hp_groupnormfusion_s(const half *input, const half *scale, const half *offset, half *mean, half *variance, float epsilon, int num_groups, int channel, int unit, int batch, int core_mask, half *output)
|
||||
.. c:function:: void fp_groupnormfusion_s(const float *input, const float *scale, const float *offset, float *mean, float *variance, float epsilon, int num_groups, int channel, int unit, int batch, int core_mask, float *output)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 17
|
||||
|
||||
// FT78NE 多核示例
|
||||
#include <stdio.h>
|
||||
|
||||
int main(void) {
|
||||
const float *input = (const float *)0xA0000000; // DDR 存储
|
||||
const float *scale = (const float *)0xB0000000;
|
||||
const float *offset = (const float *)0xB0001000;
|
||||
float *mean = (float *)0xB0002000;
|
||||
float *variance = (float *)0xB0003000;
|
||||
float *output = (float *)0xC0000000;
|
||||
int num_groups = 8;
|
||||
int channel = 64;
|
||||
int unit = 49;
|
||||
int batch = 32;
|
||||
float epsilon = 1e-5f;
|
||||
int core_mask = 0xff;
|
||||
fp_groupnormfusion_s(input, scale, offset, mean, variance,
|
||||
epsilon, num_groups, channel, unit,
|
||||
batch, core_mask, output);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void hp_groupnormfusion_p(const half *input, const half *scale, const half *offset, half *mean, half *variance, float epsilon, int num_groups, int channel, int unit, int batch, half *output)
|
||||
.. c:function:: void fp_groupnormfusion_p(const float *input, const float *scale, const float *offset, float *mean, float *variance, float epsilon, int num_groups, int channel, int unit, int batch, float *output)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 16
|
||||
|
||||
// MT7004 单核示例
|
||||
#include <stdio.h>
|
||||
|
||||
int main(void) {
|
||||
const half *input = (const half *)0x10000000; // L2 存储
|
||||
const half *scale = (const half *)0x10004000;
|
||||
const half *offset = (const half *)0x10008000;
|
||||
half *mean = (half *)0x1000C000;
|
||||
half *variance = (half *)0x10010000;
|
||||
half *output = (half *)0x10014000;
|
||||
int num_groups = 4;
|
||||
int channel = 32;
|
||||
int unit = 36;
|
||||
int batch = 16;
|
||||
float epsilon = 1e-4f;
|
||||
hp_groupnormfusion_p(input, scale, offset, mean, variance,
|
||||
epsilon, num_groups, channel, unit,
|
||||
batch, output);
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,201 @@
|
|||
GRU
|
||||
=================
|
||||
|
||||
将多层门控循环单元 (GRU) RNN 应用于输入序列。
|
||||
|
||||
GRU 网络模型中有两个门：更新门和重置门。将两个连续的时间节点表示为 :math:`t - 1` 和 :math:`t`。给定一个在时刻 :math:`t` 的输入 :math:`x_t`，一个隐藏状态 :math:`h_{t-1}`，在时刻 :math:`t` 的更新门和重置门使用门控制机制计算。更新门 :math:`z_t` 用于控制前一时刻的状态信息被带入到当前状态中的程度，重置门 :math:`r_t` 控制前一状态有多少信息被写入到当前候选集 :math:`n_t` 上
|
||||
|
||||
对于输入序列中的每个元素，每一层计算以下函数：
|
||||
|
||||
.. math::
|
||||
:nowrap:
|
||||
|
||||
\begin{align*}
|
||||
r_t &= \sigma(W_{ir}x_t + b_{ir} + W_{hr}h_{(t-1)} + b_{hr}) \\
|
||||
z_t &= \sigma(W_{iz}x_t + b_{iz} + W_{hz}h_{(t-1)} + b_{hz}) \\
|
||||
n_t &= \tanh(W_{in}x_t + b_{in} + r_t \odot (W_{hn}h_{(t-1)} + b_{hn})) \\
|
||||
h_t &= (1-z_t) \odot n_t + z_t \odot h_{(t-1)}
|
||||
\end{align*}
|
||||
|
||||
其中 :math:`\sigma` 是 sigmoid 激活函数，:math:`\odot` 是 Hadamard 积（逐元素乘积）。:math:`W, b` 是公式中输出和输入之间的可学习权重。例如，:math:`W_{ir}, b_{ir}` 是用于将输入 :math:`x_t` 转换为 :math:`r_t` 的权重和偏置。
|
||||
|
||||
注意，本算子中候选门 :math:`n_t` 的计算与原始论文和Mindspore框架略有不同。在原始实现中，:math:`r_t` 和上一隐藏状态 :math:`h_{(t-1)}` 之间的 Hadamard 积 (:math:`\odot`) 在与权重矩阵 :math:`W` 相乘和加上偏置之前进行：
|
||||
|
||||
.. math::
|
||||
|
||||
n_t = \tanh(W_{in}x_t + b_{in} + W_{hn}(r_t \odot h_{(t-1)}) + b_{hn})
|
||||
|
||||
本算子采用 PyTorch 实现方式，是在 :math:`W_{hn}h_{(t-1)}` 之后完成的：
|
||||
|
||||
.. math::
|
||||
|
||||
n_t = \tanh(W_{in}x_t + b_{in} + r_t \odot (W_{hn}h_{(t-1)} + b_{hn}))
|
||||
|
||||
输入：
|
||||
- **input** - 输入数据的地址。
|
||||
- **weight_g** - 可学习的输入-隐藏权重的地址。
|
||||
- **weight_r** - 可学习的隐藏-隐藏权重的地址。
|
||||
- **input_bias** - 可学习的输入-隐藏偏置的地址。
|
||||
- **state_bias** - 可学习的隐藏-隐藏偏置的地址。
|
||||
- **hidden_state** - 初始隐藏状态的地址。
|
||||
- **buffer** - 用于存储中间计算结果。
|
||||
- **gru_param** - 算子计算所需参数的结构体。其各成员见下述。
|
||||
- **core_mask** - 核掩码。
|
||||
|
||||
**GruParameter定义：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
|
||||
typedef struct GruParameter {
|
||||
int input_size_; // 输入input中预期特征的数量
|
||||
int hidden_size_; // 隐藏状态h中的特征数量
|
||||
int seq_len_; // 输入batch中每个序列的长度
|
||||
int batch_; // 总批次数
|
||||
int output_step_; // 每次循环中output步长
|
||||
int bidirectional_; // 是否为双向GRU
|
||||
int input_row_align_; // 输入行对齐值
|
||||
int input_col_align_; // 输入列对齐值
|
||||
int state_row_align_; // 隐藏状态行对齐值
|
||||
int state_col_align_; // 隐藏状态列对齐值
|
||||
int check_seq_len_; // 进行计算的序列长度
|
||||
} GruParameter;
|
||||
|
||||
输出：
|
||||
- **output** - 输出地址。
|
||||
- **hidden_state** - 最终的隐藏状态。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持int8, fp32
|
||||
- MT7004 支持fp16, fp32
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void i8_Gru_s(int8_t *output, int8_t *input, int8_t *weight_g, int8_t *weight_r, int8_t *input_bias, int8_t *state_bias, int8_t *hidden_state, int8_t *buffer[4], GruParameter *gru_param, int core_mask)
|
||||
.. c:function:: void hp_Gru_s(half *output, half *input, half *weight_g, half *weight_r, half *input_bias, half *state_bias, half *hidden_state, half *buffer[4], GruParameter *gru_param, int core_mask);
|
||||
.. c:function:: void fp_Gru_s(float *output, float *input, float *weight_g, float *weight_r, float *input_bias, float *state_bias, float *hidden_state, float *buffer[4], GruParameter *gru_param, int core_mask);
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 41
|
||||
|
||||
void TestGruSMCFp32(int check_seq_len, int seq_len, int batch_size, int input_size, int bidirectional, int hidden_size, int core_mask) {
|
||||
int core_id = get_core_id();
|
||||
int logic_core_id = GetLogicCoreId(core_mask, core_id);
|
||||
int core_num = GetCoreNum(core_mask);
|
||||
float *output = (void*)0x88000000;
|
||||
float *input = (void*)0x88100000;
|
||||
float *weight_g = (void*)0x88200000;
|
||||
float *weight_r = (void*)0x88300000;
|
||||
float *input_bias = (void*)0x88400000;
|
||||
float *state_bias = (void*)0x88500000;
|
||||
float *hidden_state = (void*)0x88600000;
|
||||
float** buffer = (float**)0x88700000;
|
||||
float *output_hidden_state = (void*)0x88800000;
|
||||
GruParameter* param = (GruParameter*)0x88900000;
|
||||
int hidden_state_batch = 1;
|
||||
int num_directions = 1;
|
||||
if (bidirectional) {
|
||||
hidden_state_batch = hidden_state_batch * 2;
|
||||
num_directions = num_directions * 2;
|
||||
}
|
||||
int input_col_align = hidden_size;
|
||||
int state_col_align = hidden_size;
|
||||
if (logic_core_id == 0) {
|
||||
memcpy(output_hidden_state, hidden_state, hidden_state_batch * batch_size * hidden_size * sizeof(float));
|
||||
memcpy(check_output_hidden_state, hidden_state, hidden_state_batch * batch_size * hidden_size * sizeof(float));
|
||||
buffer[0] = (void*)0x88A00000;
|
||||
buffer[1] = (void*)0x88B00000;
|
||||
buffer[2] = (void*)0x88C00000;
|
||||
buffer[3] = (void*)0x88D00000;
|
||||
param->batch_ = batch_size;
|
||||
param->bidirectional_ = bidirectional;
|
||||
param->hidden_size_ = hidden_size;
|
||||
param->input_col_align_ = input_col_align;
|
||||
param->input_size_ = input_size;
|
||||
param->output_step_ = batch_size * hidden_size * num_directions;
|
||||
param->seq_len_ = seq_len;
|
||||
param->state_col_align_ = state_col_align;
|
||||
param->check_seq_len_ = check_seq_len;
|
||||
}
|
||||
sys_bar(0, core_num); // 初始化参数完成后进行同步
|
||||
fp_Gru_s(output, input, weight_g, weight_r, input_bias, state_bias, output_hidden_state, buffer, param, core_mask);
|
||||
}
|
||||
|
||||
void main() {
|
||||
int check_seq_len = 2;
|
||||
int seq_len = 2;
|
||||
int batch_size = 2;
|
||||
int input_size = 2;
|
||||
int bidirectional = 0;
|
||||
int hidden_size = 2;
|
||||
int core_mask = 0b1111;
|
||||
TestGruSMCFp32(check_seq_len, seq_len, batch_size, input_size, bidirectional, hidden_size, core_mask);
|
||||
}
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void i8_Gru_p(int8_t *output, int8_t *input, int8_t *weight_g, int8_t *weight_r, int8_t *input_bias, int8_t *state_bias, int8_t *hidden_state, int8_t *buffer[4], GruParameter *gru_param, int core_mask)
|
||||
.. c:function:: void hp_Gru_p(half *output, half *input, half *weight_g, half *weight_r, half *input_bias, half *state_bias, half *hidden_state, half *buffer[4], GruParameter *gru_param, int core_mask);
|
||||
.. c:function:: void fp_Gru_p(float *output, float *input, float *weight_g, float *weight_r, float *input_bias, float *state_bias, float *hidden_state, float *buffer[4], GruParameter *gru_param, int core_mask);
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 35
|
||||
|
||||
void TestGruL2Fp32(int check_seq_len, int seq_len, int batch_size, int input_size, int bidirectional, int hidden_size, int core_mask) {
|
||||
float *output = (void*)0x10000000; // 私有存储版本地址设置在AM内
|
||||
float *input = (void*)0x10004000;
|
||||
float *weight_g = (void*)0x10008000;
|
||||
float *weight_r = (void*)0x1000C000;
|
||||
float *input_bias = (void*)0x10010000;
|
||||
float *state_bias = (void*)0x10014000;
|
||||
float *hidden_state = (void*)0x10018000;
|
||||
float** buffer = (float**)0x1001C000;
|
||||
float *output_hidden_state = (void*)0x10020000;
|
||||
GruParameter* param = (GruParameter*)0x10024000;
|
||||
int hidden_state_batch = 1;
|
||||
int num_directions = 1;
|
||||
if (bidirectional) {
|
||||
hidden_state_batch = hidden_state_batch * 2;
|
||||
num_directions = num_directions * 2;
|
||||
}
|
||||
int input_col_align = hidden_size;
|
||||
int state_col_align = hidden_size;
|
||||
memcpy(output_hidden_state, hidden_state, hidden_state_batch * batch_size * hidden_size * sizeof(float));
|
||||
memcpy(check_output_hidden_state, hidden_state, hidden_state_batch * batch_size * hidden_size * sizeof(float));
|
||||
buffer[0] = (void*)0x10030000;
|
||||
buffer[1] = (void*)0x10034000;
|
||||
buffer[2] = (void*)0x10038000;
|
||||
buffer[3] = (void*)0x1003C000;
|
||||
param->batch_ = batch_size;
|
||||
param->bidirectional_ = bidirectional;
|
||||
param->hidden_size_ = hidden_size;
|
||||
param->input_col_align_ = input_col_align;
|
||||
param->input_size_ = input_size;
|
||||
param->output_step_ = batch_size * hidden_size * num_directions;
|
||||
param->seq_len_ = seq_len;
|
||||
param->state_col_align_ = state_col_align;
|
||||
param->check_seq_len_ = check_seq_len;
|
||||
fp_Gru_p(output, input, weight_g, weight_r, input_bias, state_bias, output_hidden_state, buffer, param, core_mask);
|
||||
}
|
||||
|
||||
void main() {
|
||||
int check_seq_len = 2;
|
||||
int seq_len = 2;
|
||||
int batch_size = 2;
|
||||
int input_size = 2;
|
||||
int bidirectional = 0;
|
||||
int hidden_size = 2;
|
||||
int core_mask = 0b0001; // 私有存储版本只能设置为一个核心启动
|
||||
TestGruL2Fp32(check_seq_len, seq_len, batch_size, input_size, bidirectional, hidden_size, core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,75 @@
|
|||
LeakyReLu
|
||||
=================
|
||||
|
||||
Leaky ReLU激活函数。
|
||||
|
||||
该激活函数定义如下：
|
||||
|
||||
.. math::
|
||||
leaky\_relu(x) =
|
||||
\begin{cases}
|
||||
x, & \text{if } x \geq 0; \\
|
||||
\alpha \cdot x, & \text{otherwise.}
|
||||
\end{cases}
|
||||
|
||||
其中， :math:`\alpha` 表示 alpha 参数。
|
||||
|
||||
输入：
|
||||
- **input** - 输入数据的地址。
|
||||
- **elem_cnt** - 计算长度。
|
||||
- **alpha** - 公式中的alpha参数。
|
||||
- **core_mask** - 核掩码。
|
||||
|
||||
输出：
|
||||
- **output** - 输出地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持int8, fp32
|
||||
- MT7004 支持fp16, fp32
|
||||
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void fp_leaky_relu_s(float* input, float* output, int elem_cnt, float alpha, int core_mask)
|
||||
.. c:function:: void hp_leaky_relu_s(half* input, half* output, int elem_cnt, half alpha, int core_mask)
|
||||
.. c:function:: void i8_leaky_relu_s(const int8_t* input, int8_t* output, int elem_cnt, float alpha, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 7
|
||||
|
||||
void main() {
|
||||
float* input = (float*)0x82000000;
|
||||
float* output = (float*)0x88000000;
|
||||
int length = 1000;
|
||||
float alpha = 0.9;
|
||||
int core_mask = 0b1111;
|
||||
fp_leaky_relu_s(input, output, length, alpha, core_mask);
|
||||
}
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void fp_leaky_relu_p(float* input, float* output, int elem_cnt, float alpha, int core_mask)
|
||||
.. c:function:: void hp_leaky_relu_p(half* input, half* output, int elem_cnt, half alpha, int core_mask)
|
||||
.. c:function:: void i8_leaky_relu_p(const int8_t* input, int8_t* output, int elem_cnt, float alpha, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 7
|
||||
|
||||
void main() {
|
||||
float* input = (float*)0x10000000; // 私有存储版本地址设置在AM内
|
||||
float* output = (float*)0x10010000;
|
||||
int length = 1000;
|
||||
float alpha = 0.9;
|
||||
int core_mask = 0b0001; // 私有存储版本只能设置为一个核心启动
|
||||
fp_leaky_relu_p(input, output, length, alpha, core_mask);
|
||||
}
|
||||
|
|
@ -0,0 +1,68 @@
|
|||
LinSpace
|
||||
=================
|
||||
|
||||
生成从 ``start`` 到 ``end`` 的等间距序列，线性插值
|
||||
|
||||
.. math::
|
||||
|
||||
\text{output}_i = \text{start} + i \cdot \frac{\text{end} - \text{start}}{\text{length} - 1},\quad i = 0, 1, \dots, \text{length} - 1
|
||||
|
||||
输入：
|
||||
- **start** - 序列起始值。
|
||||
- **end** - 序列终止值（包含）。
|
||||
- **length** - 序列点数；要求 length ≥ 2。
|
||||
- **core_mask（可选）** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **output** - 输出序列地址（float32）。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- 多核版本按行数均匀分割，各核处理自己的段落。
|
||||
- 步长计算为 (end - start) / (length - 1)。
|
||||
|
||||
**共享存储版本：**
|
||||
|
||||
.. c:function:: void fp_linspace_s(float *output, float start, float end, int length, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 10
|
||||
|
||||
#include <stdio.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
float *output = (float *)0xA0000000; // DDR
|
||||
float start = 0.0f;
|
||||
float end = 100.0f;
|
||||
int length = 1001;
|
||||
int core_mask = 0xff;
|
||||
fp_linspace_s(output, start, end, length, core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
**私有存储版本：**
|
||||
|
||||
.. c:function:: void fp_linspace_p(float *output, float start, float step, int num)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 9
|
||||
|
||||
#include <stdio.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
float *output = (float *)0x10000000; // L2
|
||||
float start = 1.5f;
|
||||
float step = 0.5f;
|
||||
int num = 1000;
|
||||
fp_linspace_p(output, start, step, num);
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,181 @@
|
|||
LSTM
|
||||
=================
|
||||
一种常用的 循环神经网络（RNN） 模块，用于处理具有时序依赖特征的数据（如语音、文本、时间序列等）。每个时间步的公式化描述如下。
|
||||
|
||||
.. math::
|
||||
|
||||
\begin{aligned}
|
||||
i_t &= \sigma(W_{ii} x_t + W_{hi} h_{t-1} + b_i) && \text{(输入门)} \\[6pt]
|
||||
f_t &= \sigma(W_{if} x_t + W_{hf} h_{t-1} + b_f) && \text{(遗忘门)} \\[6pt]
|
||||
g_t &= \tanh(W_{ig} x_t + W_{hg} h_{t-1} + b_g) && \text{(候选状态)} \\[6pt]
|
||||
o_t &= \sigma(W_{io} x_t + W_{ho} h_{t-1} + b_o) && \text{(输出门)} \\[6pt]
|
||||
c_t &= f_t \odot c_{t-1} + i_t \odot g_t && \text{(细胞状态更新)} \\[6pt]
|
||||
h_t &= o_t \odot \tanh(c_t) && \text{(隐藏状态更新)}
|
||||
\end{aligned}
|
||||
|
||||
- :math:`x_t` : 当前时间步输入向量
|
||||
- :math:`h_{t-1}` : 上一时间步的隐藏状态
|
||||
- :math:`c_{t-1}` : 上一时间步的细胞状态
|
||||
- :math:`i_t, f_t, g_t, o_t` : 四个门（输入门、遗忘门、候选门、输出门）
|
||||
- :math:`W_*` : 对应的权重矩阵
|
||||
- :math:`b_*` : 偏置项
|
||||
- :math:`\sigma(\cdot)` : Sigmoid 函数
|
||||
- :math:`\odot` : 元素乘
|
||||
|
||||
|
||||
|
||||
输入：
|
||||
- **input** - 输入序列数据，形状为 :math:`(seq_len, batch, input_size)`，即每个时间步的输入特征。
|
||||
- **weight_i** - 输入到各门 :math:`(input、forget、cell、output)` 的权重矩阵，大小为 4 * hidden_size * input_size。
|
||||
- **weight_h** - 上一隐藏状态到各门的权重矩阵，大小为 :math:`4 * hidden_size * hidden_size`
|
||||
- **input_bias** - 输入部分的偏置项，对应 4 个门的偏置。
|
||||
- **state_bias** - 隐藏状态部分的偏置项（也是 :math:`4 * hidden_size`），与 input_bias 一起求和形成总偏置。
|
||||
- **hidden_state** - 当前批次初始隐藏状态输入（ :math:`h₀` ），执行后更新为最后时刻的隐藏状态输出（ :math:`hₜ`）
|
||||
- **cell_state** - 当前批次初始细胞状态输入（ :math:`c₀`），执行后更新为最后时刻的细胞状态输出（ :math:`cₜ`）。
|
||||
- **buffer** - 临时工作区指针数组（中间计算缓存，如门值、激活结果、临时矩阵等，用于优化性能）。
|
||||
- **LstmParameter** - LSTM 配置参数结构体，包含输入大小、隐藏层维度、序列长度、是否双向等信息。
|
||||
- **core_mask** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
**LstmParameter定义：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
|
||||
typedef struct LstmParameter {
|
||||
int input_size_;//每个时间步输入向量的维度（输入特征数）。
|
||||
int hidden_size_;//LSTM 隐藏状态的维度（每个门的内部计算大小）。
|
||||
int project_size_;//投影层输出维度（用于 LSTMP，有则在输出前线性压缩隐藏状态）。
|
||||
int output_size_;//实际输出维度，等于 hidden_size_ 或 project_size_（取决于是否使用投影层）。
|
||||
int seq_len_;//输入序列的时间步数（序列长度）。
|
||||
int batch_;//批次大小（一次处理的样本数量）。
|
||||
// other parameter
|
||||
int output_step_;//指定输出第几个时间步的结果（通常为最后一步或每步）。
|
||||
bool bidirectional_;//是否为双向 LSTM（true 表示前向和后向各一层）。
|
||||
float zoneout_cell_;//单元状态的 Zoneout 比例（防止过拟合的正则化参数）。
|
||||
float zoneout_hidden_;//隐藏状态的 Zoneout 比例（防止过拟合）。
|
||||
int input_row_align_;//输入张量的行对齐参数（用于 DMA 或 SIMD 加速的内存对齐）。
|
||||
int input_col_align_;//输入张量的列对齐参数。
|
||||
int state_row_align_;//状态张量（hidden/cell）的行对齐参数。
|
||||
int state_col_align_;//状态张量的列对齐参数。
|
||||
int proj_col_align_;//投影层矩阵的列对齐参数。
|
||||
bool has_bias_;//是否包含偏置项（true 表示使用 bias）。
|
||||
} LstmParameter;
|
||||
|
||||
输出：
|
||||
- **output** - 计算结果地址，存放 LSTM 每个时间步输出结果的缓冲区，维度通常为 :math:`(seq\_len, batch, output\_size)`
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持fp32
|
||||
- MT7004 支持fp32
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void fp_Lstm_s(float *output, const float *input, const float *weight_i, const float *weight_h, const float *input_bias,const float *state_bias, float *hidden_state, float *cell_state, float *buffer[9],const LstmParameter *lstm_param, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 40-42
|
||||
|
||||
//FT78NE示例
|
||||
#include <stdio.h>
|
||||
#include <lstm.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
LstmParameter *lstm_param = (LstmParameter *)0x90000000;
|
||||
lstm_param->seq_len_ = 20;
|
||||
lstm_param->batch_ = 1;
|
||||
lstm_param->input_size_ = 2000;
|
||||
lstm_param->hidden_size_ = 3;
|
||||
lstm_param->bidirectional_ = false;
|
||||
float * input = (float *)0xA0000000; //input在DDR空间
|
||||
float * weight_i = (float *)0xA1000000;
|
||||
float * weight_h = (float *)0xA3000000;
|
||||
float *input_bias_ =(float *) 0xB0900000;
|
||||
float * state_bias_ =(float *) 0xB0B00000;
|
||||
float * output_s = (float *)0xC0000000;
|
||||
float *hidden_state_s = (float *)0xC0100000;
|
||||
float *cell_state_s = (float *)0xC0200000;
|
||||
float *buffer[9];
|
||||
float * packed_input_ = (float *)0xB0000000;
|
||||
buffer[0] = packed_input_;
|
||||
float * gate = (float *)0xB0100000;
|
||||
buffer[1] = gate;
|
||||
float * packed_state = (float *)0xB0200000;
|
||||
buffer[2] = packed_state;
|
||||
float * state_gate = (float *)0xB0300000;
|
||||
buffer[3] = state_gate;
|
||||
float * cell_buffer = (float *)0xB0400000;
|
||||
buffer[4] = cell_buffer;
|
||||
float * hidden_buffer = (float *)0xB0500000;
|
||||
buffer[5] = hidden_buffer;
|
||||
float * packed_output = (float *)0xB0600000;
|
||||
buffer[6] = packed_output;
|
||||
float * left_matrix = (float *)0xB0700000;
|
||||
buffer[7] = left_matrix;
|
||||
float * packed_ptr = (float *)0xB0800000;
|
||||
buffer[8] = packed_ptr;
|
||||
int core_mask = 0xff;
|
||||
fp_Lstm_s(output_s, input, weight_i, weight_h, input_bias_,
|
||||
state_bias, hidden_state_s, cell_state_s, buffer,
|
||||
lstm_param, core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void fp_Lstm_p(float *output, const float *input, const float *weight_i, const float *weight_h, const float *input_bias, const float *state_bias, float *hidden_state, float *cell_state, float *buffer[9], const LstmParameter *lstm_param)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 38-40
|
||||
|
||||
//FT78NE示例
|
||||
#include <stdio.h>
|
||||
#include <lstm.h>
|
||||
int main(int argc, char* argv[]) {
|
||||
LstmParameter *lstm_param = (LstmParameter *)0x10000000;
|
||||
lstm_param->seq_len_ = 4;
|
||||
lstm_param->batch_ = 1;
|
||||
lstm_param->input_size_ = 2;
|
||||
lstm_param->hidden_size_ = 3;
|
||||
lstm_param->bidirectional_ = false;
|
||||
float * input = (float *)0x10000200; //input在DDR空间
|
||||
float * weight_i = (float *)0x10000400;
|
||||
float * weight_h = (float *)0x10000600;
|
||||
float *input_bias_ =(float *) 0x10000800;
|
||||
float * state_bias_ =(float *) 0x10000A00;
|
||||
float * output_s = (float *)0x10000C00;
|
||||
float *hidden_state_s = (float *)0x10000E00;
|
||||
float *cell_state_s = (float *)0x10001000;
|
||||
float *buffer[9];
|
||||
float * packed_input_ = (float *)0x10001200;
|
||||
buffer[0] = packed_input_;
|
||||
float * gate = (float *)0x10001400;
|
||||
buffer[1] = gate;
|
||||
float * packed_state = (float *)0x10001600;
|
||||
buffer[2] = packed_state;
|
||||
float * state_gate = (float *)0x10001800;
|
||||
buffer[3] = state_gate;
|
||||
float * cell_buffer = (float *)0x10001A00;
|
||||
buffer[4] = cell_buffer;
|
||||
float * hidden_buffer = (float *)0x10001C00;
|
||||
buffer[5] = hidden_buffer;
|
||||
float * packed_output = (float *)0x10001F00;
|
||||
buffer[6] = packed_output;
|
||||
float * left_matrix = (float *)0x10002000;
|
||||
buffer[7] = left_matrix;
|
||||
float * packed_ptr = (float *)0x10002200;
|
||||
buffer[8] = packed_ptr;
|
||||
fp_Lstm_p(output_s, input, weight_i, weight_h, input_bias_,
|
||||
state_bias, hidden_state_s, cell_state_s, buffer,
|
||||
lstm_param);
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,93 @@
|
|||
MatMulFusion
|
||||
=================
|
||||
|
||||
矩阵乘法融合（可选偏置和激活），计算：
|
||||
|
||||
.. math::
|
||||
|
||||
C = \operatorname{act}(A \times B + \text{bias})
|
||||
|
||||
其中 ``A`` 形状为 :math:`M\times K`，``B`` 为 :math:`K\times N`，``C`` 与可选的 ``bias`` 为 :math:`M\times N`。
|
||||
激活 ``act`` 支持：
|
||||
|
||||
- ``0``: 无激活（Identity）
|
||||
- ``1``: ReLU
|
||||
- ``2``: ReLU6
|
||||
|
||||
输入：
|
||||
- **A** - 输入矩阵 A（行优先，连续存储）。大小 M×K。
|
||||
- **B** - 输入矩阵 B（行优先，连续存储）。大小 K×N。
|
||||
- **bias** - 偏置矩阵（可为 NULL，大小 M×N）。
|
||||
- **M, N, K** - 维度参数。
|
||||
- **activation_type** - 激活类型，取值 {0,1,2}。
|
||||
- **core_mask（可选）** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **C** - 输出矩阵（行优先，大小 M×N）。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- 复数类型的激活逐分量应用于实部与虚部。
|
||||
- 请确保输入按行优先连续布局，且不发生类型范围溢出；int16/int32 计算未做饱和裁剪。
|
||||
- 共享存储版本要求地址位于 GSM/DDR/SMC 等共享可见空间。
|
||||
|
||||
**共享存储版本：**
|
||||
|
||||
.. c:function:: void i16_matmulfusion_s(int16_t *A, int16_t *B, int16_t *C, int16_t *bias, int M, int N, int K, int activation_type, int core_mask)
|
||||
.. c:function:: void i32_matmulfusion_s(int32_t *A, int32_t *B, int32_t *C, int32_t *bias, int M, int N, int K, int activation_type, int core_mask)
|
||||
.. c:function:: void fp_matmulfusion_s(float *A, float *B, float *C, float *bias, int M, int N, int K, int activation_type, int core_mask)
|
||||
.. c:function:: void dp_matmulfusion_s(double *A, double *B, double *C, double *bias, int M, int N, int K, int activation_type, int core_mask)
|
||||
.. c:function:: void c64_matmulfusion_s(float complex *A, float complex *B, float complex *C, float complex *bias, int M, int N, int K, int activation_type, int core_mask)
|
||||
.. c:function:: void c128_matmulfusion_s(double complex *A, double complex *B, double complex *C, double complex *bias, int M, int N, int K, int activation_type, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 12
|
||||
|
||||
#include <stdio.h>
|
||||
#include <stdbool.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
float *A = (float *)0xA0000000; // DDR
|
||||
float *B = (float *)0xA1000000; // DDR
|
||||
float *C = (float *)0xA2000000; // DDR
|
||||
float *bias = (float *)0xA3000000; // DDR，可为 NULL
|
||||
int M = 512, N = 512, K = 512;
|
||||
int activation_type = 1; // ReLU
|
||||
int core_mask = 0xff; // 8 核
|
||||
fp_matmulfusion_s(A, B, C, bias, M, N, K, activation_type, core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
**私有存储版本：**
|
||||
|
||||
.. c:function:: void i16_matmulfusion_p(int16_t *A, int16_t *B, int16_t *C, int16_t *bias, int M, int N, int K, int activation_type)
|
||||
.. c:function:: void i32_matmulfusion_p(int32_t *A, int32_t *B, int32_t *C, int32_t *bias, int M, int N, int K, int activation_type)
|
||||
.. c:function:: void fp_matmulfusion_p(float *A, float *B, float *C, float *bias, int M, int N, int K, int activation_type)
|
||||
.. c:function:: void dp_matmulfusion_p(double *A, double *B, double *C, double *bias, int M, int N, int K, int activation_type)
|
||||
.. c:function:: void c64_matmulfusion_p(float complex *A, float complex *B, float complex *C, float complex *bias, int M, int N, int K, int activation_type)
|
||||
.. c:function:: void c128_matmulfusion_p(double complex *A, double complex *B, double complex *C, double complex *bias, int M, int N, int K, int activation_type)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 10
|
||||
|
||||
#include <stdio.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
double *A = (double *)0x10000000; // L2
|
||||
double *B = (double *)0x10020000; // L2
|
||||
double *C = (double *)0x10040000; // L2/DDR
|
||||
double *bias = NULL; // 可为 NULL
|
||||
int M = 128, N = 128, K = 128;
|
||||
int activation_type = 0; // None
|
||||
dp_matmulfusion_p(A, B, C, bias, M, N, K, activation_type);
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,88 @@
|
|||
RaggedRange
|
||||
=================
|
||||
|
||||
为每个区间生成等差序列并拼接到一个一维输出，同时返回每段的边界索引。
|
||||
|
||||
.. math::
|
||||
|
||||
\begin{aligned}
|
||||
L_k &= \left\lceil \frac{\text{limit}_k - \text{start}_k}{\text{delta}_k} \right\rceil,\quad k=0,\dots,\text{range\_count}-1 \\
|
||||
\text{splits}[0] &= 0,\quad \text{splits}[k+1] = \text{splits}[k] + L_k \\
|
||||
\text{values}[\text{splits}[k] + j] &= \text{start}_k + j \cdot \text{delta}_k,\quad j=0,\dots,L_k-1
|
||||
\end{aligned}
|
||||
|
||||
输入：
|
||||
- **starts** - 各段起始值数组地址。
|
||||
- **limits** - 各段终止上界数组地址（半开区间 [start, limit) 语义）。
|
||||
- **deltas** - 各段步长数组地址，要求 ``delta_k != 0``，且方向与 ``limit_k - start_k`` 一致，否则该段长度视为 0。
|
||||
- **range_count** - 段数 K。
|
||||
- **core_mask（可选）** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **values** - 扁平化拼接的结果数组地址，长度为 ``splits[range_count]``。
|
||||
- **splits** - 长度为 ``range_count + 1`` 的整型数组，记录各段边界，且 ``splits[0] = 0``。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- 当前实现覆盖 int8、int16、int32、fp32、fp64 五种类型。
|
||||
- 不进行数值饱和裁剪；请确保序列不会在对应类型范围内溢出。
|
||||
- 共享存储版本要求输入/输出地址位于共享可见的存储空间（如 GSM/DDR/SMC 等）。
|
||||
|
||||
**共享存储版本：**
|
||||
|
||||
.. c:function:: void i8_raggedrange_s(int8_t *starts, int8_t *limits, int8_t *deltas, int range_count, int8_t *values, int *splits, int core_mask)
|
||||
.. c:function:: void i16_raggedrange_s(int16_t *starts, int16_t *limits, int16_t *deltas, int range_count, int16_t *values, int *splits, int core_mask)
|
||||
.. c:function:: void i32_raggedrange_s(int32_t *starts, int32_t *limits, int32_t *deltas, int range_count, int32_t *values, int *splits, int core_mask)
|
||||
.. c:function:: void fp_raggedrange_s(float *starts, float *limits, float *deltas, int range_count, float *values, int *splits, int core_mask)
|
||||
.. c:function:: void dp_raggedrange_s(double *starts, double *limits, double *deltas, int range_count, double *values, int *splits, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 12
|
||||
|
||||
#include <stdio.h>
|
||||
#include <stdbool.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
float *starts = (float *)0xA0000000; // DDR
|
||||
float *limits = (float *)0xA0100000; // DDR
|
||||
float *deltas = (float *)0xA0200000; // DDR
|
||||
float *values = (float *)0xA0300000; // DDR
|
||||
int *splits = (int *)0xA0400000; // DDR
|
||||
int range_count = 3;
|
||||
int core_mask = 0xff;
|
||||
fp_raggedrange_s(starts, limits, deltas, range_count, values, splits, core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
**私有存储版本：**
|
||||
|
||||
.. c:function:: void i8_raggedrange_p(int8_t *starts, int8_t *limits, int8_t *deltas, int range_count, int8_t *values, int *splits)
|
||||
.. c:function:: void i16_raggedrange_p(int16_t *starts, int16_t *limits, int16_t *deltas, int range_count, int16_t *values, int *splits)
|
||||
.. c:function:: void i32_raggedrange_p(int32_t *starts, int32_t *limits, int32_t *deltas, int range_count, int32_t *values, int *splits)
|
||||
.. c:function:: void fp_raggedrange_p(float *starts, float *limits, float *deltas, int range_count, float *values, int *splits)
|
||||
.. c:function:: void dp_raggedrange_p(double *starts, double *limits, double *deltas, int range_count, double *values, int *splits)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 10
|
||||
|
||||
#include <stdio.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
int32_t *starts = (int32_t *)0x10000000; // L2
|
||||
int32_t *limits = (int32_t *)0x10001000; // L2
|
||||
int32_t *deltas = (int32_t *)0x10002000; // L2
|
||||
int32_t *values = (int32_t *)0x10003000; // L2/DDR
|
||||
int *splits = (int *)0x10004000; // L2/DDR
|
||||
int range_count = 2;
|
||||
i32_raggedrange_p(starts, limits, deltas, range_count, values, splits);
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,78 @@
|
|||
Range
|
||||
=================
|
||||
|
||||
生成等差序列，逐元素写入输出：
|
||||
|
||||
.. math::
|
||||
|
||||
output_i = start + i \times delta,\quad i = 0, 1, \dots, \text{length} - 1
|
||||
|
||||
输入：
|
||||
- **start** - 序列起始值。
|
||||
- **delta** - 公差。
|
||||
- **length** - 序列长度。
|
||||
- **core_mask（可选）** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **output** - 输出数据地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持 int8, int16, int32, fp32, fp64。
|
||||
- MT7004 支持 fp16, fp32, int16, int32（如需 fp16，请使用相应 fp16 实现）。
|
||||
|
||||
**共享存储版本：**
|
||||
|
||||
.. c:function:: void i8_range_s(int8_t* output, int8_t start, int8_t delta, int length, int core_mask)
|
||||
.. c:function:: void i16_range_s(int16_t* output, int16_t start, int16_t delta, int length, int core_mask)
|
||||
.. c:function:: void i32_range_s(int32_t* output, int32_t start, int32_t delta, int length, int core_mask)
|
||||
.. c:function:: void fp_range_s(float* output, float start, float delta, int length, int core_mask)
|
||||
.. c:function:: void dp_range_s(double* output, double start, double delta, int length, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 11
|
||||
|
||||
#include <stdio.h>
|
||||
#include <stdbool.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
float *output = (float *)0xA0000000; // DDR
|
||||
float start = 1.5f;
|
||||
float delta = 0.5f;
|
||||
int length = 1000;
|
||||
int core_mask = 0xff;
|
||||
fp_range_s(output, start, delta, length, core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
**私有存储版本：**
|
||||
|
||||
.. c:function:: void i8_range_p(int8_t* output, int8_t start, int8_t delta, int length)
|
||||
.. c:function:: void i16_range_p(int16_t* output, int16_t start, int16_t delta, int length)
|
||||
.. c:function:: void i32_range_p(int32_t* output, int32_t start, int32_t delta, int length)
|
||||
.. c:function:: void fp_range_p(float* output, float start, float delta, int length)
|
||||
.. c:function:: void dp_range_p(double* output, double start, double delta, int length)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 10
|
||||
|
||||
#include <stdio.h>
|
||||
#include <stdbool.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
float *output = (float *)0x10000000; // L2
|
||||
float start = 1.5f;
|
||||
float delta = 0.5f;
|
||||
int length = 1000;
|
||||
fp_range_p(output, start, delta, length);
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,211 @@
|
|||
Reduce
|
||||
=================
|
||||
|
||||
对指定维度进行归约。
|
||||
|
||||
输入：
|
||||
- **src_data** - 输入数据的地址
|
||||
- **param** - 算子计算所需参数的结构体。其各成员见下述。
|
||||
- **core_mask** - 核掩码。
|
||||
|
||||
**ReduceParameter定义：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
|
||||
typedef struct ReduceParameter {
|
||||
void** data_buffers_; // 用于存储中间计算结果
|
||||
int* outer_sizes_; // 处理某个规约轴时，该轴之前所有轴的元素数
|
||||
int* inner_sizes_; // 某个规约轴之后的所有元素数
|
||||
int* axis_sizes_; // 规约轴的元素数
|
||||
int total_num_; // 输入张量的总元素数
|
||||
int num_axes_; // 待规约轴的数目
|
||||
int mode_; // 规约模式
|
||||
int output_num_; // 输出张量的总元素数
|
||||
/**该算子会根据ReduceParameter中的mode_参数选择实际规约所使用的方法。共有如下几种方法：
|
||||
Reduce_Mean=0,
|
||||
Reduce_Max=1,
|
||||
Reduce_Min=2,
|
||||
Reduce_Prod=3,
|
||||
Reduce_Sum=4,
|
||||
Reduce_SumSquare=5,
|
||||
Reduce_ASum=6,
|
||||
Reduce_L2Norm=7
|
||||
**/
|
||||
} ReduceParameter;
|
||||
|
||||
输出：
|
||||
- **dst_data** - 输出地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持int8, int16, int32, fp32, fp64
|
||||
- MT7004 支持fp16, fp32, int16, int32
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void i8_reduce_s(int8_t* src_data, int8_t* dst_data, ReduceParameter* param, int core_mask)
|
||||
.. c:function:: void i16_reduce_s(int16_t* src_data, half* dst_data, ReduceParameter* param, int core_mask)
|
||||
.. c:function:: void i32_reduce_s(int* src_data, float* dst_data, ReduceParameter* param, int core_mask)
|
||||
.. c:function:: void hp_reduce_s(half* src_data, half* dst_data, ReduceParameter* param, int core_mask)
|
||||
.. c:function:: void fp_reduce_s(float* src_data, float* dst_data, ReduceParameter* param, int core_mask)
|
||||
.. c:function:: void dp_reduce_s(double* src_data, double* dst_data, ReduceParameter* param, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 50
|
||||
|
||||
void PackParam(ReduceParameter* param, int ndim, int* input_shape, int num_axes, int* axes) {
|
||||
int tmp_input_shape[8];
|
||||
int total_num = 1;
|
||||
int i, j, k;
|
||||
for (i = 0; i < ndim; i++) {
|
||||
tmp_input_shape[i] = input_shape[i];
|
||||
total_num *= input_shape[i];
|
||||
}
|
||||
param->total_num_ = total_num;
|
||||
int offset_size = 0;
|
||||
for (i = 0; i < num_axes; ++i) {
|
||||
int axis = axes[i];
|
||||
int outer_size = 1;
|
||||
for (j = 0; j < axis; j++) {
|
||||
outer_size *= tmp_input_shape[j];
|
||||
}
|
||||
param->outer_sizes_[offset_size] = outer_size;
|
||||
int inner_size = 1;
|
||||
for (k = axis + 1; k < ndim; k++) {
|
||||
inner_size *= tmp_input_shape[k];
|
||||
}
|
||||
param->inner_sizes_[offset_size] = inner_size;
|
||||
param->axis_sizes_[offset_size] = tmp_input_shape[axis];
|
||||
offset_size++;
|
||||
tmp_input_shape[axis] = 1;
|
||||
}
|
||||
}
|
||||
|
||||
void TestReduceSMCFp32(int* input_shape, int ndim, int* axes, int num_axes, int mode, int keep_dims, int core_mask) {
|
||||
int core_id = get_core_id();
|
||||
int logic_core_id = GetLogicCoreId(core_mask, core_id);
|
||||
int core_num = GetCoreNum(core_mask);
|
||||
float* input = (float*)0x88000000;
|
||||
float* output = (float*)0x98000000;
|
||||
ReduceParameter* param = (ReduceParameter*)0xA8480000;
|
||||
if (logic_core_id == 0) {
|
||||
param->num_axes_ = num_axes;
|
||||
param->mode_ = mode;
|
||||
param->data_buffers_ = (void**)0xA8483000;
|
||||
param->inner_sizes_ = (int*)0xA8484000;
|
||||
param->outer_sizes_ = (int*)0xA8485000;
|
||||
param->axis_sizes_ = (int*)0xA8486000;
|
||||
int i;
|
||||
for (i = 0; i < num_axes - 1; i++) {
|
||||
param->data_buffers_[i] = (void*)(0xA8490000 + 0x1000000);
|
||||
}
|
||||
PackParam(param, ndim, input_shape, num_axes, axes);
|
||||
}
|
||||
sys_bar(0, core_num); // 初始化参数完成后进行同步
|
||||
fp_reduce_s(input, check, param, core_mask);
|
||||
}
|
||||
|
||||
void main(){
|
||||
int input_shape[3] = {4, 5, 5};
|
||||
int ndim = 3;
|
||||
int axes[1] = {1};
|
||||
int num_axes = 1;
|
||||
int mode = 7;
|
||||
int keep_dims = 1;
|
||||
int core_mask = 0b1111;
|
||||
TestReduceSMCFp32(input_shape, ndim, axes, num_axes, mode, keep_dims, core_mask);
|
||||
}
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void i8_reduce_p(int8_t* src_data, int8_t* dst_data, void* tmp_src_data, void* tmp_dst_data, ReduceParameter* param, int core_mask)
|
||||
.. c:function:: void i16_reduce_p(int16_t* src_data, half* dst_data, void* tmp_src_data, void* tmp_dst_data, ReduceParameter* param, int core_mask)
|
||||
.. c:function:: void i32_reduce_p(int* src_data, float* dst_data, void* tmp_src_data, void* tmp_dst_data, ReduceParameter* param, int core_mask)
|
||||
.. c:function:: void hp_reduce_p(half* src_data, half* dst_data, void* tmp_src_data, void* tmp_dst_data, ReduceParameter* param, int core_mask)
|
||||
.. c:function:: void fp_reduce_p(float* src_data, float* dst_data, void* tmp_src_data, void* tmp_dst_data, ReduceParameter* param, int core_mask)
|
||||
.. c:function:: void dp_reduce_p(double* src_data, double* dst_data, void* tmp_src_data, void* tmp_dst_data, ReduceParameter* param, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 59
|
||||
|
||||
void PackParam(ReduceParameter* param, int ndim, int* input_shape, int num_axes, int* axes) {
|
||||
int tmp_input_shape[8];
|
||||
int total_num = 1;
|
||||
int i, j, k;
|
||||
for (i = 0; i < ndim; i++) {
|
||||
tmp_input_shape[i] = input_shape[i];
|
||||
total_num *= input_shape[i];
|
||||
}
|
||||
param->total_num_ = total_num;
|
||||
int offset_size = 0;
|
||||
for (i = 0; i < num_axes; ++i) {
|
||||
int axis = axes[i];
|
||||
int outer_size = 1;
|
||||
for (j = 0; j < axis; j++) {
|
||||
outer_size *= tmp_input_shape[j];
|
||||
}
|
||||
param->outer_sizes_[offset_size] = outer_size;
|
||||
int inner_size = 1;
|
||||
for (k = axis + 1; k < ndim; k++) {
|
||||
inner_size *= tmp_input_shape[k];
|
||||
}
|
||||
param->inner_sizes_[offset_size] = inner_size;
|
||||
param->axis_sizes_[offset_size] = tmp_input_shape[axis];
|
||||
offset_size++;
|
||||
tmp_input_shape[axis] = 1;
|
||||
}
|
||||
}
|
||||
|
||||
void TestReduceL2Fp32(int* input_shape, int ndim, int* axes, int num_axes, int mode, int keep_dims, int core_mask) {
|
||||
float* input = (float*)0x10000000; // 原始输入输出数据需分配在AM中
|
||||
float* output = (float*)0x10010000;
|
||||
float* tmp_input = (float*)0x88000000; // 临时输入输出空间需分配在DDR或SMC中
|
||||
float* tmp_output = (float*)0x98000000;
|
||||
ReduceParameter* param = (ReduceParameter*)0x10020000;
|
||||
param->num_axes_ = num_axes;
|
||||
param->mode_ = mode;
|
||||
param->data_buffers_ = (void**)0x10021000;
|
||||
param->inner_sizes_ = (int*)0x10022000;
|
||||
param->outer_sizes_ = (int*)0x10023000;
|
||||
param->axis_sizes_ = (int*)0x10024000;
|
||||
int i, j;
|
||||
for (i = 0; i < ndim; i++) {
|
||||
int reduce_axis = 0;
|
||||
for (j = 0; j < num_axes; j++) {
|
||||
if (axes[j] == i) {
|
||||
reduce_axis = 1;
|
||||
break;
|
||||
}
|
||||
}
|
||||
if (!reduce_axis) {
|
||||
length *= input_shape[i];
|
||||
}
|
||||
}
|
||||
for (i = 0; i < num_axes - 1; i++) {
|
||||
param->data_buffers_[i] = (void*)(0xA8490000 + 0x1000000); // 每一个中间计算结果空间都需分配在DDR或SMC中
|
||||
}
|
||||
param->output_num_ = length;
|
||||
PackParam(param, ndim, input_shape, num_axes, axes);
|
||||
fp_reduce_p(input, check, param, core_mask);
|
||||
}
|
||||
|
||||
void main() {
|
||||
int input_shape[3] = {4, 5, 5};
|
||||
int ndim = 3;
|
||||
int axes[1] = {1};
|
||||
int num_axes = 1;
|
||||
int mode = 7;
|
||||
int keep_dims = 1;
|
||||
int core_mask = 0b0001; // 私有存储版本只能设置为一个核心启动
|
||||
TestReduceL2Fp32(input_shape, ndim, axes, num_axes, mode, keep_dims, core_mask);
|
||||
}
|
||||
|
|
@ -0,0 +1,96 @@
|
|||
Resize
|
||||
=================
|
||||
|
||||
对输入图像使用给定的插值方式去调整为给定的尺寸大小。
|
||||
|
||||
输入：
|
||||
- **input** - 输入数据的地址
|
||||
- **param** - 算子计算所需参数的结构体。其各成员见下述。
|
||||
- **core_mask** - 核掩码。
|
||||
|
||||
**ResizeParameter定义：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
|
||||
typedef struct ResizeParameter {
|
||||
int* input_shape_; // 输入张量形状
|
||||
int* output_shape_; // 输出张量形状
|
||||
int* x_lefts_; // 用于存储预处理结果
|
||||
int* x_rights_; // 用于存储预处理结果
|
||||
int* y_tops_; // 用于存储预处理结果
|
||||
int* y_bottoms_; // 用于存储预处理结果
|
||||
void* x_weights_; // 用于存储预处理结果
|
||||
void* y_weights_; // 用于存储预处理结果
|
||||
void* line_buffers_; // 用于存储中间结果
|
||||
int method_; // 所用的插值方法，0:最邻近插值，1:双线性插值，2:双三次插值
|
||||
int coordinate_transform_mode_; // 像素点对齐方式，0:非对称，1:中心对齐，2:偏移半像素
|
||||
float cubic_coeff_; // 一个仅在双三次插值中使用到的系数
|
||||
} ResizeParameter;
|
||||
|
||||
输出：
|
||||
- **output** - 输出地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持int8, fp32
|
||||
- MT7004 支持fp16, fp32
|
||||
|
||||
**共享/私有存储版本:**
|
||||
|
||||
.. c:function:: void i8_resize_anycore(int8_t* input, int8_t* output, ResizeParameter* param, int core_mask)
|
||||
.. c:function:: void hp_resize_anycore(half* input, half* output, ResizeParameter* param, int core_mask)
|
||||
.. c:function:: void fp_resize_anycore(float* input, float* output, ResizeParameter* param, int core_mask)
|
||||
|
||||
私有及共享空间版本均使用这些函数。
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 30
|
||||
|
||||
void TestResizeFp32SMC(int* input_shape, int* output_shape, ResizeMethod method, CoordinateTransformMode mode, float cubic_coeff, int core_mask) {
|
||||
int core_id = get_core_id();
|
||||
int core_num = GetCoreNum(core_mask);
|
||||
int logic_core_id = GetLogicCoreId(core_mask, core_id);
|
||||
float* input = (float*)0x84000000; // 测试私有空间时地址设置在私有空间内即可
|
||||
float* output = (float*)0x85000000;
|
||||
ResizeParameter* param = (ResizeParameter*)0x86000000;
|
||||
if (logic_core_id == 0) {
|
||||
param->coordinate_transform_mode_ = mode;
|
||||
param->method_ = method;
|
||||
param->cubic_coeff_ = cubic_coeff;
|
||||
param->input_shape_ = (int*)0x87000000;
|
||||
memcpy(param->input_shape_, input_shape, sizeof(int) * 4);
|
||||
param->output_shape_ = (int*)0x87100000;
|
||||
memcpy(param->output_shape_, output_shape, sizeof(int) * 4);
|
||||
param->line_buffers_ = (void*)0x88000000;
|
||||
param->x_lefts_ = (int*)0x89000000;
|
||||
param->x_rights_ = (int*)0x8A000000;
|
||||
param->y_bottoms_ = (int*)0x8B000000;
|
||||
param->y_tops_ = (int*)0x8C000000;
|
||||
param->x_weights_ = (void*)0x8D000000;
|
||||
param->y_weights_ = (void*)0x8E000000;
|
||||
if (method == BILINEAR) {
|
||||
PrepareResizeBilinear(param); // 做预处理
|
||||
} else if (method == CUBIC) {
|
||||
PrepareResizeBicubic(param, cubic_coeff); // 做预处理
|
||||
}
|
||||
}
|
||||
sys_bar(0, core_num); // 初始化参数完成后进行同步
|
||||
fp_resize_anycore(input, check, param, core_mask);
|
||||
}
|
||||
|
||||
void main(){
|
||||
int input_shape[4] = {2, 4, 4, 4};
|
||||
int output_shape[4] = {2, 8, 8, 4};
|
||||
ResizeMethod method = NEAREST;
|
||||
CoordinateTransformMode mode = ALIGN_CORNERS;
|
||||
float cubic_coeff = -0.75;
|
||||
int core_mask = 0b1111; // 测试单核时核掩码设置为0b0001即可
|
||||
TestResizeFp32SMC(input_shape, output_shape, method, mode, cubic_coeff, core_mask);
|
||||
}
|
||||
|
|
@ -0,0 +1,105 @@
|
|||
ReverseSequence
|
||||
=================
|
||||
|
||||
对输入序列进行部分反转。
|
||||
|
||||
输入：
|
||||
- **src** - 需反转数据的地址
|
||||
- **seq_lengths** - 指定反转长度，为一维向量。
|
||||
- **param** - 算子计算所需参数的结构体。其各成员见下述。
|
||||
- **core_mask** - 核掩码。
|
||||
|
||||
**ReverseSequenceParameter定义：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
|
||||
typedef struct ReverseSequenceParameter {
|
||||
int* shape_; // 输入和输出张量的形状
|
||||
int* strides_; // 一个记录张量每一维步长的数组
|
||||
int seq_dim_; // 指定反转的维度
|
||||
int batch_dim_; // 指定切片维度
|
||||
int ndim_; // 输入和输出张量的维度
|
||||
int type_size_; // 输入和输出张量数据类型的长度
|
||||
int copy_elem_num_; // 在inner_count循环中一次应被拷贝的元素数目
|
||||
int outer_stride_; // 外层元素间的步长
|
||||
int outer_count_; // 外层元素数
|
||||
int inner_stride_; // 内层元素间的步长
|
||||
int inner_count_; // 内层元素数
|
||||
} ReverseSequenceParameter;
|
||||
|
||||
输出：
|
||||
- **dst** - 输出地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持int8, int16, int32, fp32, fp64, cplx64, cplx128
|
||||
- MT7004 支持fp16, fp32, int16, int32, cplx64
|
||||
|
||||
**共享/私有存储版本:**
|
||||
|
||||
.. c:function:: void anytype_reverse_sequence_anycore(void* src, void* dst, int* seq_lengths, ReverseSequenceParameter* param, int core_mask)
|
||||
|
||||
各种数据类型、私有及共享空间版本均使用该函数。对于不同数据类型，改变param中的type_size_参数即可。
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 41
|
||||
|
||||
void Resize(ReverseSequenceParameter* param) {
|
||||
param->strides_ = (int*)0xA8020000;
|
||||
ComputeStrides(param->shape_, param->strides_, param->ndim_);
|
||||
int less_dim = param->batch_dim_ > param->seq_dim_ ? param->seq_dim_ : param->batch_dim_;
|
||||
int greater_dim = param->batch_dim_ < param->seq_dim_ ? param->seq_dim_ : param->batch_dim_;
|
||||
// calculate the size of elements should be copied at one time
|
||||
param->copy_elem_num_ = CountElementAfterDim(param->shape_, greater_dim, param->ndim_);
|
||||
// calculate the number of elements before the less axis and the stride
|
||||
param->outer_count_ = CountElementBeforeDim(param->shape_, less_dim);
|
||||
param->outer_stride_ = param->shape_[less_dim] * CountElementAfterDim(param->shape_, less_dim, param->ndim_);
|
||||
// calculate the number of elements between the less axis and the greater axis and the stride
|
||||
param->inner_count_ = 1;
|
||||
int i;
|
||||
for (i = less_dim + 1; i < greater_dim; i++) {
|
||||
param->inner_count_ *= param->shape_[i];
|
||||
}
|
||||
param->inner_stride_ = param->shape_[greater_dim] * CountElementAfterDim(param->shape_, greater_dim, param->ndim_);
|
||||
}
|
||||
|
||||
void TestReverseSequenceFp32(int* shape, int* orig_seq_lengths, int seq_dim, int batch_dim, int ndim, int core_mask) {
|
||||
int core_id = get_core_id();
|
||||
int core_num = GetCoreNum(core_mask);
|
||||
int logic_core_id = GetLogicCoreId(core_mask, core_id);
|
||||
float* input_data = (float*)0x88000000; // 测试私有空间时地址设置在私有空间内即可
|
||||
float* output_data = (float*)0x98000000;
|
||||
float* check = (float*)0xB8000000;
|
||||
int* seq_lengths = (int*)0xC8000000;
|
||||
ReverseSequenceParameter* param = (ReverseSequenceParameter*)0xA8000000;
|
||||
int i;
|
||||
if (logic_core_id == 0) {
|
||||
param->shape_ = (int*)0xA8010000;
|
||||
memcpy(param->shape_, shape, sizeof(int) * ndim);
|
||||
memcpy(seq_lengths, orig_seq_lengths, sizeof(int) * shape[batch_dim]);
|
||||
param->batch_dim_ = batch_dim;
|
||||
param->seq_dim_ = seq_dim;
|
||||
param->ndim_ = ndim;
|
||||
param->type_size_ = sizeof(float);
|
||||
Resize(param);
|
||||
}
|
||||
sys_bar(0, core_num); // 初始化参数完成后进行同步
|
||||
anytype_reverse_sequence_anycore(input_data, output_data, seq_lengths, param, core_mask);
|
||||
}
|
||||
|
||||
void main(){
|
||||
int shape[3] = {3, 4, 10};
|
||||
int seq_lengths[3] = {1, 2, 3};
|
||||
int ndim = 3;
|
||||
int seq_dim = 1;
|
||||
int batch_dim = 0;
|
||||
int core_mask = 0b1111; // 测试单核时核掩码设置为0b0001即可
|
||||
TestReverseSequenceFp32(shape, seq_lengths, seq_dim, batch_dim, ndim, core_mask);
|
||||
}
|
||||
|
|
@ -0,0 +1,109 @@
|
|||
ReverseV2
|
||||
=================
|
||||
|
||||
对输入Tensor按指定维度反转。
|
||||
|
||||
输入：
|
||||
- **src** - 需反转数据的地址
|
||||
- **param** - 算子计算所需参数的结构体。其各成员见下述。
|
||||
- **core_mask** - 核掩码。
|
||||
|
||||
**ReverseV2Parameter定义：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
|
||||
typedef struct ReverseV2Parameter {
|
||||
int* axis_flag_; // 用于存储需反转的轴，需反转的轴对应于数组索引的元素被置为1，其余元素为0
|
||||
int* input_shape_; // 输入张量的形状
|
||||
int* input_strides_; // 一个记录输入张量每一维步长的数组
|
||||
int** cur_coord_; // 二维数组，每个元素存储对应核心所使用的cur_coord数组地址，这个数组用于记录当前循环到的元素坐标
|
||||
int ndim_; // 输入和输出张量的维度
|
||||
int axis_ndim_; // 需反转的轴的数目
|
||||
int num_elem_; // 元素总数
|
||||
int type_size_; // 输入和输出张量数据类型的长度
|
||||
} ReverseV2Parameter;
|
||||
|
||||
输出：
|
||||
- **dst** - 输出地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持int8, int16, int32, fp32, fp64, cplx64, cplx128
|
||||
- MT7004 支持fp16, fp32, int16, int32, cplx64
|
||||
|
||||
**共享/私有存储版本:**
|
||||
|
||||
.. c:function:: void anytype_reversev2_anycore(void* src, void* dst, ReverseV2Parameter* param, int core_mask)
|
||||
|
||||
各种数据类型、私有及共享空间版本均使用该函数。对于不同数据类型，改变param中的type_size_参数即可。
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 50
|
||||
|
||||
void Resize(ReverseV2Parameter* param, int* axis) {
|
||||
int i;
|
||||
param->input_strides_[param->ndim_ - 1] = 1;
|
||||
for (i = param->ndim_ - 1; i > 0; i--) {
|
||||
param->input_strides_[i - 1] = param->input_strides_[i] * param->input_shape_[i];
|
||||
}
|
||||
for (i = 0; i < param->ndim_; i++) {
|
||||
// initialize axis_flag array by 0
|
||||
param->axis_flag_[i] = 0;
|
||||
}
|
||||
for (i = 0; i < param->axis_ndim_; i++) {
|
||||
param->axis_flag_[axis[i]] = 1;
|
||||
}
|
||||
param->num_elem_ = 1;
|
||||
for (i = 0; i < param->ndim_; i++) {
|
||||
param->num_elem_ *= param->input_shape_[i];
|
||||
}
|
||||
}
|
||||
|
||||
void TestReverseV2(int* shape, int ndim, int* axis, int axis_ndim, int core_mask) {
|
||||
int type_size = 4;
|
||||
int core_num = GetCoreNum(core_mask);
|
||||
int core_id = get_core_id();
|
||||
int logic_core_id = GetLogicCoreId(core_mask, core_id);
|
||||
float* input_data = (float*)0x88000000; // 测试私有空间时地址设置在私有空间内即可
|
||||
float* output_data = (float*)0xA8000000;
|
||||
float* check = (float*)0xC8000000;
|
||||
ReverseV2Parameter* param = (ReverseV2Parameter*)0x84000000;
|
||||
if (logic_core_id == 0) {
|
||||
int i, j;
|
||||
param->axis_ndim_ = axis_ndim;
|
||||
param->input_shape_ = (int*)0x84003000;
|
||||
for (i = 0; i < ndim; i++) {
|
||||
param->input_shape_[i] = shape[i];
|
||||
}
|
||||
param->ndim_ = ndim;
|
||||
param->type_size_ = type_size;
|
||||
param->input_strides_ = (int*)0x84004000;
|
||||
param->axis_flag_ = (int*)0x84005000;
|
||||
param->cur_coord_ = (int**)0x84006000;
|
||||
for (i = 0; i < 4; i++) {
|
||||
param->cur_coord_[i] = (int*)(0x84007000 + i * 0x1000LL);
|
||||
for (j = 0; j < ndim; j++) {
|
||||
param->cur_coord_[i][j] = 0;
|
||||
}
|
||||
}
|
||||
Resize(param, axis);
|
||||
}
|
||||
sys_bar(0, core_num); // 初始化参数完成后进行同步
|
||||
anytype_reversev2_anycore(input_data, output_data, param, core_mask);
|
||||
}
|
||||
|
||||
void main(){
|
||||
int shape[3] = {10, 10, 1000};
|
||||
int ndim = 3;
|
||||
int axis[3] = {0, 1, 2};
|
||||
int axis_ndim = 3;
|
||||
int core_mask = 0b1111;
|
||||
TestReverseV2(shape, ndim, axis, axis_ndim, core_mask);
|
||||
}
|
||||
|
|
@ -0,0 +1,89 @@
|
|||
ScaleFusion
|
||||
=================
|
||||
|
||||
|
||||
|
||||
传入一个数组，逐元素乘上因数，并加上偏置的值后输出。
|
||||
|
||||
.. math::
|
||||
|
||||
dst_i = src_i \cdot scale_i + bias_i
|
||||
|
||||
输入：
|
||||
- **src_data** - 输入数据地址。
|
||||
- **length** - 计算长度。
|
||||
- **scale** - 缩放因子数组首地址。
|
||||
- **bias** - 偏置数组首地址。
|
||||
- **core_mask** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **dst_data** - 计算结果地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持int8, int16, int32, fp32, fp64
|
||||
- MT7004 支持fp16, fp32, int16, int32
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void i8_scalefusion_s(int8_t* src_data, int8_t* dst_data, int length, float* scale, float* bias, int core_mask)
|
||||
.. c:function:: void i16_scalefusion_s(int16_t* src_data, int16_t* dst_data, int length, float* scale, float* bias, int core_mask)
|
||||
.. c:function:: void i32_scalefusion_s(int* src_data, int* dst_data, int length, float* scale, float* bias, int core_mask)
|
||||
.. c:function:: void hp_scalefusion_s(half* src_data, half* dst_data, int length, half* scale, half* bias, int core_mask)
|
||||
.. c:function:: void fp_scalefusion_s(float* src_data, float* dst_data, int length, float* scale, float* bias, int core_mask)
|
||||
.. c:function:: void dp_scalefusion_s(double* src_data, double* dst_data, int length, double* scale, double* bias, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 12
|
||||
|
||||
//FT78NE示例
|
||||
#include <stdio.h>
|
||||
#include <scalefusion.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
float *input0 = (float *)0xA0000000; //input在DDR空间
|
||||
float *output = (float *)0xC0000000;
|
||||
float *scale = (float *)0xB0000000; //scale在DDR空间
|
||||
float *bias = (float *)0xB1000000; //bias在DDR空间
|
||||
int length = 1000;
|
||||
int core_mask = 0xff;
|
||||
fp_scalefusion_s( input0, output, length, scale, bias, core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void i8_scalefusion_p(int8_t* src_data, int8_t* dst_data, int length, float* scale, float* bias)
|
||||
.. c:function:: void i16_scalefusion_p(int16_t* src_data, int16_t* dst_data, int length, float* scale, float* bias)
|
||||
.. c:function:: void i32_scalefusion_p(int* src_data, int* dst_data, int length, float* scale, float* bias)
|
||||
.. c:function:: void hp_scalefusion_p(half* src_data, half* dst_data, int length, half* scale, half* bias)
|
||||
.. c:function:: void fp_scalefusion_p(float* src_data, float* dst_data, int length, float* scale, float* bias)
|
||||
.. c:function:: void dp_scalefusion_p(double* src_data, double* dst_data, int length, double* scale, double* bias)
|
||||
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 10
|
||||
|
||||
//FT78NE示例
|
||||
#include <stdio.h>
|
||||
#include <scalefusion.h>
|
||||
int main(int argc, char* argv[]) {
|
||||
float *input0 = (float *)0x10810000; //input在L2空间
|
||||
float *output = (float *)0x10820000;
|
||||
float *scale = (float *)0x10830000; //scale在L2空间
|
||||
float *bias = (float *)0x10840000; //bias在L2空间
|
||||
int length = 1000;
|
||||
fp_scalefusion_p( input0, output, length, scale, bias);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
|
@ -0,0 +1,178 @@
|
|||
ScatterElements
|
||||
=================
|
||||
|
||||
返回一个新tensor，根据指定索引和更新值对input中的元素进行指定操作（替换或相加）。不支持隐式类型转换。举例：一个三维输入tensor的返回为:
|
||||
|
||||
.. code-block:: python
|
||||
|
||||
output[indices[i][j][k]][j][k] = updates[i][j][k] #if axis == 0, reduction == "none"
|
||||
output[i][indices[i][j][k]][k] += updates[i][j][k] #if axis == 1, reduction == "add"
|
||||
output[i][j][indices[i][j][k]] = updates[i][j][k] #if axis == 2, reduction == "none"
|
||||
|
||||
输入：
|
||||
- **input** - 输入数据的地址
|
||||
- **indices** - 指定索引。
|
||||
- **updates** - 更新值。
|
||||
- **param** - 算子计算所需参数的结构体。其各成员见下述。
|
||||
- **core_mask** - 核掩码。
|
||||
|
||||
**ScatterElementsParameter定义：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
|
||||
typedef struct ScatterElementsParameter {
|
||||
int* indices_stride_; // 对应于indices数组每一维度的步长
|
||||
int* output_stride_; // 对应于output数组每一维度的步长
|
||||
int input_dims_; // 输入张量的维度数
|
||||
int axis_; // 指定索引所在的轴
|
||||
int input_axis_size_; // 索引所在轴的元素数
|
||||
int indices_total_num_; // indices数组的总元素数
|
||||
int input_total_num_; // input数组的总元素数
|
||||
int reduction_type_; // 规约类型，0代表none，1代表add
|
||||
} ScatterElementsParameter;
|
||||
|
||||
输出：
|
||||
- **output** - 输出地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持int8, int16, int32, fp32, fp64, cplx64, cplx128
|
||||
- MT7004 支持fp16, fp32, int16, int32, cplx64
|
||||
- 如果 indices 中有多个索引向量对应于同一位置，则输出中该位置值是不确定的。
|
||||
- 如果 indices 的值超出 input 索引上下界，则相应的 updates 不会更新到 input，也不会抛出索引错误。
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void i8_scatter_elements_s(int8_t* input, int8_t* output, int* indices, int8_t* updates, ScatterElementsParameter* param, int core_mask)
|
||||
.. c:function:: void i16_scatter_elements_s(int16_t* input, int16_t* output, int* indices, int16_t* updates, ScatterElementsParameter* param, int core_mask)
|
||||
.. c:function:: void i32_scatter_elements_s(int* input, int* output, int* indices, int* updates, ScatterElementsParameter* param, int core_mask)
|
||||
.. c:function:: void hp_scatter_elements_s(half* input, half* output, int* indices, half* updates, ScatterElementsParameter* param, int core_mask)
|
||||
.. c:function:: void fp_scatter_elements_s(float* input, float* output, int* indices, float* updates, ScatterElementsParameter* param, int core_mask)
|
||||
.. c:function:: void dp_scatter_elements_s(double* input, double* output, int* indices, double* updates, ScatterElementsParameter* param, int core_mask)
|
||||
.. c:function:: void c64_scatter_elements_s(float* input, float* output, int* indices, float* updates, ScatterElementsParameter* param, int core_mask)
|
||||
.. c:function:: void c128_scatter_elements_s(double* input, double* output, int* indices, double* updates, ScatterElementsParameter* param, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 40
|
||||
|
||||
void PackParam(ScatterElementsParameter* param, int* indices_shape, int* input_shape) {
|
||||
param->indices_stride_[param->input_dims_ - 1] = 1;
|
||||
int i;
|
||||
for (i = param->input_dims_ - 1; i > 0; --i) {
|
||||
param->indices_stride_[i - 1] = param->indices_stride_[i] * indices_shape[i];
|
||||
}
|
||||
param->output_stride_[param->input_dims_ - 1] = 1;
|
||||
for (i = param->input_dims_ - 1; i > 0; --i) {
|
||||
param->output_stride_[i - 1] = param->output_stride_[i] * input_shape[i];
|
||||
}
|
||||
param->indices_total_num_ = 1;
|
||||
for (i = 0; i < param->input_dims_; i++) {
|
||||
param->indices_total_num_ *= indices_shape[i];
|
||||
}
|
||||
param->input_total_num_ = 1;
|
||||
for (i = 0; i < param->input_dims_; i++) {
|
||||
param->input_total_num_ *= input_shape[i];
|
||||
}
|
||||
param->input_axis_size_ = input_shape[param->axis_];
|
||||
}
|
||||
|
||||
void TestScatterElementsSMC(int* input_shape, int* indices_shape, int ndim, int axis, int reduction_type, int core_mask) {
|
||||
int core_num = GetCoreNum(core_mask);
|
||||
int core_id = get_core_id();
|
||||
int logic_core_id = GetLogicCoreId(core_mask, core_id);
|
||||
void* input_data = (void*)0x88000000;
|
||||
void* output_data = (void*)0x98000000;
|
||||
int* indices_data = (int*)0xA8000000;
|
||||
void* updates_data = (void*)0xB8000000;
|
||||
ScatterElementsParameter* param = (ScatterElementsParameter*)0xC8000000;
|
||||
if (logic_core_id == 0) {
|
||||
param->axis_ = axis;
|
||||
param->input_dims_ = ndim;
|
||||
param->indices_stride_ = (int*)0xC8020000;
|
||||
param->output_stride_ = (int*)0xC8040000;
|
||||
param->reduction_type_ = reduction_type;
|
||||
PackParam(param, indices_shape, input_shape);
|
||||
}
|
||||
sys_bar(0, core_num); // 初始化参数完成后进行同步
|
||||
fp_scatter_elements_s(input_data, output_data, indices_data, updates_data, param, core_mask);
|
||||
}
|
||||
|
||||
void main() {
|
||||
int input_shape[2] = {8, 30};
|
||||
int indices_shape[2] = {3, 3};
|
||||
int ndim = 2;
|
||||
int axis = 0;
|
||||
int reduction_type = 0;
|
||||
int core_mask = 0b1111;
|
||||
TestScatterElementsSMC(input_shape, indices_shape, ndim, axis, reduction_type, core_mask);
|
||||
}
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void i8_scatter_elements_p(int8_t* input, int8_t* output, int* indices, int8_t* updates, ScatterElementsParameter* param, int core_mask)
|
||||
.. c:function:: void i16_scatter_elements_p(int16_t* input, int16_t* output, int* indices, int16_t* updates, ScatterElementsParameter* param, int core_mask)
|
||||
.. c:function:: void i32_scatter_elements_p(int* input, int* output, int* indices, int* updates, ScatterElementsParameter* param, int core_mask)
|
||||
.. c:function:: void hp_scatter_elements_p(half* input, half* output, int* indices, half* updates, ScatterElementsParameter* param, int core_mask)
|
||||
.. c:function:: void fp_scatter_elements_p(float* input, float* output, int* indices, float* updates, ScatterElementsParameter* param, int core_mask)
|
||||
.. c:function:: void dp_scatter_elements_p(double* input, double* output, int* indices, double* updates, ScatterElementsParameter* param, int core_mask)
|
||||
.. c:function:: void c64_scatter_elements_p(float* input, float* output, int* indices, float* updates, ScatterElementsParameter* param, int core_mask)
|
||||
.. c:function:: void c128_scatter_elements_p(double* input, double* output, int* indices, double* updates, ScatterElementsParameter* param, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 34
|
||||
|
||||
void PackParam(ScatterElementsParameter* param, int* indices_shape, int* input_shape) {
|
||||
param->indices_stride_[param->input_dims_ - 1] = 1;
|
||||
int i;
|
||||
for (i = param->input_dims_ - 1; i > 0; --i) {
|
||||
param->indices_stride_[i - 1] = param->indices_stride_[i] * indices_shape[i];
|
||||
}
|
||||
param->output_stride_[param->input_dims_ - 1] = 1;
|
||||
for (i = param->input_dims_ - 1; i > 0; --i) {
|
||||
param->output_stride_[i - 1] = param->output_stride_[i] * input_shape[i];
|
||||
}
|
||||
param->indices_total_num_ = 1;
|
||||
for (i = 0; i < param->input_dims_; i++) {
|
||||
param->indices_total_num_ *= indices_shape[i];
|
||||
}
|
||||
param->input_total_num_ = 1;
|
||||
for (i = 0; i < param->input_dims_; i++) {
|
||||
param->input_total_num_ *= input_shape[i];
|
||||
}
|
||||
param->input_axis_size_ = input_shape[param->axis_];
|
||||
}
|
||||
|
||||
void TestScatterElementsL2(int* input_shape, int* indices_shape, int ndim, int axis, int reduction_type, int core_mask) {
|
||||
void* input_data = (void*)0x10000000; // 私有存储版本地址设置在AM内
|
||||
void* output_data = (void*)0x10001000;
|
||||
int* indices_data = (int*)0x10002000;
|
||||
void* updates_data = (void*)0x10003000;
|
||||
ScatterElementsParameter* param = (ScatterElementsParameter*)0x10004000;
|
||||
param->axis_ = axis;
|
||||
param->input_dims_ = ndim;
|
||||
param->indices_stride_ = (int*)0x10005000;
|
||||
param->output_stride_ = (int*)0x10006000;
|
||||
param->reduction_type_ = reduction_type;
|
||||
PackParam(param, indices_shape, input_shape);
|
||||
fp_scatter_elements_p(input_data, output_data, indices_data, updates_data, param, core_mask);
|
||||
}
|
||||
|
||||
void main() {
|
||||
int input_shape[2] = {8, 30};
|
||||
int indices_shape[2] = {3, 3};
|
||||
int ndim = 2;
|
||||
int axis = 0;
|
||||
int reduction_type = 0;
|
||||
int core_mask = 0b0001; // 私有存储版本只能设置为一个核心启动
|
||||
TestScatterElementsL2(input_shape, indices_shape, ndim, axis, reduction_type, core_mask);
|
||||
}
|
||||
|
|
@ -0,0 +1,109 @@
|
|||
SGD
|
||||
=================
|
||||
对权重张量执行带动量与权重衰减的随机梯度下降更新。
|
||||
|
||||
输入：
|
||||
- **weight** - 待更新权重张量首地址。
|
||||
- **accumulate** - 动量累积张量首地址。
|
||||
- **gradient** - 梯度张量首地址。
|
||||
- **learning_rate** - 学习率。
|
||||
- **dampening** - 动量阻尼系数。
|
||||
- **moment** - 动量系数。
|
||||
- **nesterov** - 是否启用 Nesterov 动量。
|
||||
- **weight_decay** - 权重衰减系数。
|
||||
- **start** - 参与计算的起始索引（闭区间）。
|
||||
- **end** - 参与计算的结束索引（开区间）。
|
||||
- **core_mask(int, 可选)** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **weight** - 原地写回更新后的权重张量。
|
||||
- **accumulate** - 原地写回更新后的动量张量。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持 fp32 数据类型。
|
||||
- MT7004 支持 fp16、fp32 数据类型。
|
||||
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void hp_sgd_s(half *weight, half *accumulate, const half *gradient, float learning_rate, float dampening, float moment, bool nesterov, float weight_decay, int start, int end, int core_mask)
|
||||
.. c:function:: void fp_sgd_s(float *weight, float *accumulate, const float *gradient, float learning_rate, float dampening, float moment, bool nesterov, float weight_decay, int start, int end, int core_mask)
|
||||
|
||||
|
||||
|
||||
.. math::
|
||||
|
||||
\begin{aligned}
|
||||
g'_t &= g_t + weight\_decay \cdot w_{t-1} \\
|
||||
m_t &= moment \cdot m_{t-1} + (1 - dampening) \cdot g'_t \\
|
||||
u_t &=
|
||||
\begin{cases}
|
||||
m_t \cdot moment + g'_t, & \text{if nesterov = True} \\
|
||||
m_t, & \text{otherwise}
|
||||
\end{cases} \\
|
||||
w_t &= w_{t-1} - learning\_rate \cdot u_t
|
||||
\end{aligned}
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 17
|
||||
|
||||
// FT78NE 多核示例
|
||||
#include <stdio.h>
|
||||
#include <stdbool.h>
|
||||
|
||||
int main(void) {
|
||||
float *weight = (float *)0xA0000000; // DDR 存储
|
||||
float *accumulate = (float *)0xB0000000;
|
||||
float *gradient = (float *)0xC0000000;
|
||||
int start = 0;
|
||||
int end = 4096;
|
||||
int core_mask = 0xff;
|
||||
float learning_rate = 1e-2f;
|
||||
float dampening = 0.0f;
|
||||
float moment = 0.9f;
|
||||
bool nesterov = true;
|
||||
float weight_decay = 1e-2f;
|
||||
fp_sgd_s(weight, accumulate, gradient, learning_rate,
|
||||
dampening, moment, nesterov, weight_decay,
|
||||
start, end, core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void hp_sgd_p(half *weight, half *accumulate, const half *gradient, float learning_rate, float dampening, float moment, bool nesterov, float weight_decay, int length)
|
||||
.. c:function:: void fp_sgd_p(float *weight, float *accumulate, const float *gradient, float learning_rate, float dampening, float moment, bool nesterov, float weight_decay, int length)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 15
|
||||
|
||||
// MT7004 单核示例
|
||||
#include <stdio.h>
|
||||
#include <stdbool.h>
|
||||
|
||||
int main(void) {
|
||||
half *weight = (half *)0x10000000; // L2 存储
|
||||
half *accumulate = (half *)0x10002000;
|
||||
half *gradient = (half *)0x10004000;
|
||||
int length = 2048;
|
||||
float learning_rate = 5e-3f;
|
||||
float dampening = 0.0f;
|
||||
float moment = 0.9f;
|
||||
bool nesterov = false;
|
||||
float weight_decay = 5e-3f;
|
||||
hp_sgd_p(weight, accumulate, gradient, learning_rate,
|
||||
dampening, moment, nesterov, weight_decay,
|
||||
length);
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,86 @@
|
|||
SpaceToBatch
|
||||
=================
|
||||
将输入在空间维度按照 block_size 分块并重新排列到 batch 维度，同时根据 paddings 执行必要的零填充。
|
||||
|
||||
输入：
|
||||
- **input** - 输入数据地址。
|
||||
- **block_size** - 分块因子，格式为 ``[block_h, block_w]``。
|
||||
- **paddings** - 填充参数，格式为 ``[top, bottom, left, right]``。
|
||||
- **input_shape** - 输入形状，格式为 ``[batch, height, width, channel]``。
|
||||
- **data_size** - 单个元素字节数（例如 sizeof(float)）。
|
||||
- **core_mask(int, 可选)** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **output** - 输出数据地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持的数据类型： fp32、fp64、cplx64、cplx128、int16、int8、int32。
|
||||
- MT7004 支持的数据类型： fp32、fp16、cplx64、int16、int32。
|
||||
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void i8_spacetobatch_s(int8_t *input, int8_t *output, const int *block_size, const int *paddings, const int *input_shape, int data_size, int core_mask)
|
||||
.. c:function:: void i16_spacetobatch_s(int16_t *input, int16_t *output, const int *block_size, const int *paddings, const int *input_shape, int data_size, int core_mask)
|
||||
.. c:function:: void i32_spacetobatch_s(int32_t *input, int32_t *output, const int *block_size, const int *paddings, const int *input_shape, int data_size, int core_mask)
|
||||
.. c:function:: void hp_spacetobatch_s(half *input, half *output, const int *block_size, const int *paddings, const int *input_shape, int data_size, int core_mask)
|
||||
.. c:function:: void fp_spacetobatch_s(float *input, float *output, const int *block_size, const int *paddings, const int *input_shape, int data_size, int core_mask)
|
||||
.. c:function:: void dp_spacetobatch_s(double *input, double *output, const int *block_size, const int *paddings, const int *input_shape, int data_size, int core_mask)
|
||||
.. c:function:: void c64_spacetobatch_s(float *input, float *output, const int *block_size, const int *paddings, const int *input_shape, int data_size, int core_mask)
|
||||
.. c:function:: void c128_spacetobatch_s(double *input, double *output, const int *block_size, const int *paddings, const int *input_shape, int data_size, int core_mask)
|
||||
|
||||
|
||||
**C 调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 11
|
||||
|
||||
// FT78NE 多核示例
|
||||
#include <stdio.h>
|
||||
|
||||
int main(int argc, char *argv[]) {
|
||||
float *input = (float *)0xA0000000; // 输入在 DDR 地址 0xA0000000
|
||||
float *output = (float *)0xB0000000; // 输出在 DDR 地址 0xB0000000
|
||||
int block_size[2] = {2, 2};
|
||||
int paddings[4] = {0, 0, 0, 0};
|
||||
int input_shape[4] = {1, 100, 10, 10};
|
||||
int core_mask = 0xff;
|
||||
fp_spacetobatch_s(input, output, block_size, paddings, input_shape, sizeof(float), core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void i8_spacetobatch_p(int8_t *input, int8_t *output, const int *block_size, const int *paddings, const int *input_shape, int data_size)
|
||||
.. c:function:: void i16_spacetobatch_p(int16_t *input, int16_t *output, const int *block_size, const int *paddings, const int *input_shape, int data_size)
|
||||
.. c:function:: void i32_spacetobatch_p(int32_t *input, int32_t *output, const int *block_size, const int *paddings, const int *input_shape, int data_size)
|
||||
.. c:function:: void hp_spacetobatch_p(half *input, half *output, const int *block_size, const int *paddings, const int *input_shape, int data_size)
|
||||
.. c:function:: void fp_spacetobatch_p(float *input, float *output, const int *block_size, const int *paddings, const int *input_shape, int data_size)
|
||||
.. c:function:: void dp_spacetobatch_p(double *input, double *output, const int *block_size, const int *paddings, const int *input_shape, int data_size)
|
||||
.. c:function:: void c64_spacetobatch_p(float *input, float *output, const int *block_size, const int *paddings, const int *input_shape, int data_size)
|
||||
.. c:function:: void c128_spacetobatch_p(double *input, double *output, const int *block_size, const int *paddings, const int *input_shape, int data_size)
|
||||
|
||||
**C 调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 10
|
||||
|
||||
// FT78NE 单核示例
|
||||
#include <stdio.h>
|
||||
|
||||
int main(int argc, char *argv[]) {
|
||||
float *input = (float *)0x10000000; // 单核版本：输入放在 L2 地址 0x10000000
|
||||
float *output = (float *)0x10040000; // 单核版本：输出放在 L2 地址 0x10040000
|
||||
int block_size[2] = {2, 2};
|
||||
int paddings[4] = {0, 0, 0, 0};
|
||||
int input_shape[4] = {1, 100, 10, 10};
|
||||
fp_spacetobatch_p(input, output, block_size, paddings, input_shape, sizeof(float));
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,87 @@
|
|||
SpaceToBatchND
|
||||
=================
|
||||
将输入在空间维度按照 block_size 分块并重新排列到 batch 维度，同时根据 paddings 执行必要的零填充。
|
||||
|
||||
输入：
|
||||
- **input** - 输入数据地址。
|
||||
- **block_size** - 分块因子，格式为 ``[block_h, block_w]``。
|
||||
- **paddings** - 填充参数，格式为 ``[top, bottom, left, right]``。
|
||||
- **input_shape** - 输入形状，格式为 ``[batch, height, width, channel]``。
|
||||
- **data_size** - 单个元素字节数。
|
||||
- **core_mask(int, 可选)** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **output** - 输出数据地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持的数据类型： fp32、fp64、cplx64、cplx128、int16、int8、int32。
|
||||
- MT7004 支持的数据类型： fp32、fp16、cplx64、int16、int32。
|
||||
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void i8_spacetobatchnd_s(int8_t *input, int8_t *output, const int *block_size, const int *paddings, const int *input_shape, int data_size, int core_mask)
|
||||
.. c:function:: void i16_spacetobatchnd_s(int16_t *input, int16_t *output, const int *block_size, const int *paddings, const int *input_shape, int data_size, int core_mask)
|
||||
.. c:function:: void i32_spacetobatchnd_s(int32_t *input, int32_t *output, const int *block_size, const int *paddings, const int *input_shape, int data_size, int core_mask)
|
||||
.. c:function:: void hp_spacetobatchnd_s(half *input, half *output, const int *block_size, const int *paddings, const int *input_shape, int data_size, int core_mask)
|
||||
.. c:function:: void fp_spacetobatchnd_s(float *input, float *output, const int *block_size, const int *paddings, const int *input_shape, int data_size, int core_mask)
|
||||
.. c:function:: void dp_spacetobatchnd_s(double *input, double *output, const int *block_size, const int *paddings, const int *input_shape, int data_size, int core_mask)
|
||||
.. c:function:: void c64_spacetobatchnd_s(float *input, float *output, const int *block_size, const int *paddings, const int *input_shape, int data_size, int core_mask)
|
||||
.. c:function:: void c128_spacetobatchnd_s(double *input, double *output, const int *block_size, const int *paddings, const int *input_shape, int data_size, int core_mask)
|
||||
|
||||
**C 调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 11
|
||||
|
||||
// FT78NE 多核示例
|
||||
#include <stdio.h>
|
||||
|
||||
int main(int argc, char *argv[]) {
|
||||
float *input = (float *)0xA0000000; // 输入在 DDR 地址 0xA0000000
|
||||
float *output = (float *)0xB0000000; // 输出在 DDR 地址 0xB0000000
|
||||
int block_size[2] = {2, 2};
|
||||
int paddings[4] = {0, 0, 0, 0};
|
||||
int input_shape[4] = {1, 100, 10, 10};
|
||||
int core_mask = 0xff;
|
||||
fp_spacetobatchnd_s(input, output, block_size, paddings, input_shape, sizeof(float), core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void i8_spacetobatchnd_p(int8_t *input, int8_t *output, const int *block_size, const int *paddings, const int *input_shape, int data_size)
|
||||
.. c:function:: void i16_spacetobatchnd_p(int16_t *input, int16_t *output, const int *block_size, const int *paddings, const int *input_shape, int data_size)
|
||||
.. c:function:: void i32_spacetobatchnd_p(int32_t *input, int32_t *output, const int *block_size, const int *paddings, const int *input_shape, int data_size)
|
||||
.. c:function:: void hp_spacetobatchnd_p(half *input, half *output, const int *block_size, const int *paddings, const int *input_shape, int data_size)
|
||||
.. c:function:: void fp_spacetobatchnd_p(float *input, float *output, const int *block_size, const int *paddings, const int *input_shape, int data_size)
|
||||
.. c:function:: void dp_spacetobatchnd_p(double *input, double *output, const int *block_size, const int *paddings, const int *input_shape, int data_size)
|
||||
.. c:function:: void c64_spacetobatchnd_p(float *input, float *output, const int *block_size, const int *paddings, const int *input_shape, int data_size)
|
||||
.. c:function:: void c128_spacetobatchnd_p(double *input, double *output, const int *block_size, const int *paddings, const int *input_shape, int data_size)
|
||||
|
||||
**C 调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 10
|
||||
|
||||
// FT78NE 单核示例
|
||||
#include <stdio.h>
|
||||
|
||||
int main(int argc, char *argv[]) {
|
||||
float *input = (float *)0x10000000; // 单核版本：输入放在 L2 地址 0x10000000
|
||||
float *output = (float *)0x10040000; // 单核版本：输出放在 L2 地址 0x10040000
|
||||
int block_size[2] = {2, 2};
|
||||
int paddings[4] = {0, 0, 0, 0};
|
||||
int input_shape[4] = {1, 100, 10, 10};
|
||||
fp_spacetobatchnd_p(input, output, block_size, paddings, input_shape, sizeof(float));
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
|
|
@ -0,0 +1,91 @@
|
|||
SpaceToDepth
|
||||
=================
|
||||
将输入张量的空间维按块(block)重组到深度通道（channel）中。
|
||||
|
||||
假设输入形状为: \[N, H, W, C\]，块大小为 B，则输出形状为:
|
||||
|
||||
.. math::
|
||||
|
||||
\text{out\_shape} = [N, H/ B, W / B, C * B * B]
|
||||
|
||||
对应元素映射为：
|
||||
|
||||
.. math::
|
||||
|
||||
output[n, h_{out}, w_{out}, c * B * B + (l * B + m)] = input[n, h_{out} * B + l, w_{out} * B + m, c]
|
||||
|
||||
输入：
|
||||
- **input** - 输入数据地址，按 NHWC 存储。
|
||||
- **in_shape** - 指向长度为4的数组，表示输入维度 \[N, H, W, C\]。
|
||||
- **block** - block 大小。
|
||||
- **data_size** - 每个元素的字节大小。
|
||||
- **core_mask(int, 可选)** - 核掩码（仅适用于共享存储版本）。
|
||||
|
||||
输出：
|
||||
- **output** - 输出数据地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持 fp32, fp64, cplx64, cplx128, int16, int8, int32
|
||||
- MT7004 支持 fp32, fp16, cplx64, int16, int32
|
||||
|
||||
**共享存储版本:**
|
||||
|
||||
.. c:function:: void i8_spacetodepth_s(int8_t* input, int8_t* output, const int* in_shape, int block, int data_size, int core_mask)
|
||||
.. c:function:: void i16_spacetodepth_s(int16_t* input, int16_t* output, const int* in_shape, int block, int data_size, int core_mask)
|
||||
.. c:function:: void i32_spacetodepth_s(int32_t* input, int32_t* output, const int* in_shape, int block, int data_size, int core_mask)
|
||||
.. c:function:: void hp_spacetodepth_s(half* input, half* output, const int* in_shape, int block, int data_size, int core_mask)
|
||||
.. c:function:: void fp_spacetodepth_s(float* input, float* output, const int* in_shape, int block, int data_size, int core_mask)
|
||||
.. c:function:: void dp_spacetodepth_s(double* input, double* output, const int* in_shape, int block, int data_size, int core_mask)
|
||||
.. c:function:: void c64_spacetodepth_s(float* input, float* output, const int* in_shape, int block, int data_size, int core_mask)
|
||||
.. c:function:: void c128_spacetodepth_s(double* input, double* output, const int* in_shape, int block, int data_size, int core_mask)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 9
|
||||
|
||||
#include <stdio.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
float *input = (float *)0xA0000000; // 多核版：输入在 DDR 区域
|
||||
float *output = (float *)0xB0000000; // 多核版：输出在 DDR 区域
|
||||
int in_shape[4] = {1, 8, 8, 1}; // N,H,W,C
|
||||
int block = 2;
|
||||
int core_mask = 0xff;
|
||||
fp_spacetodepth_s(input, output, in_shape, block, sizeof(float), core_mask);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
**私有存储版本:**
|
||||
|
||||
.. c:function:: void i8_spacetodepth_p(int8_t* input, int8_t* output, const int* in_shape, int block, int data_size)
|
||||
.. c:function:: void i16_spacetodepth_p(int16_t* input, int16_t* output, const int* in_shape, int block, int data_size)
|
||||
.. c:function:: void i32_spacetodepth_p(int32_t* input, int32_t* output, const int* in_shape, int block, int data_size)
|
||||
.. c:function:: void hp_spacetodepth_p(half* input, half* output, const int* in_shape, int block, int data_size)
|
||||
.. c:function:: void fp_spacetodepth_p(float* input, float* output, const int* in_shape, int block, int data_size)
|
||||
.. c:function:: void dp_spacetodepth_p(double* input, double* output, const int* in_shape, int block, int data_size)
|
||||
.. c:function:: void c64_spacetodepth_p(float* input, float* output, const int* in_shape, int block, int data_size)
|
||||
.. c:function:: void c128_spacetodepth_p(double* input, double* output, const int* in_shape, int block, int data_size)
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 8
|
||||
|
||||
#include <stdio.h>
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
float *input = (float *)0x10000000; // 单核版：输入在 L2
|
||||
float *output = (float *)0x10010000; // 单核版：输出在 L2
|
||||
int in_shape[4] = {1, 8, 8, 1};
|
||||
int block = 2;
|
||||
fp_spacetodepth_p(input, output, in_shape, block, sizeof(float));
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,42 @@
|
|||
Squeeze
|
||||
=================
|
||||
|
||||
返回删除指定axis中大小为1的维度后的Tensor。由于该算子仅改变张量形状，因此其DSP算子的作用是将数据从输入张量完整拷贝到输出张量。
|
||||
|
||||
输入：
|
||||
- **src** - 输入地址
|
||||
- **total_copy_size** - 计算得到的总共需拷贝的数据量，单位为字节。
|
||||
- **core_mask** - 核掩码。
|
||||
|
||||
输出：
|
||||
- **dst** - 输出地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持int8, int16, int32, fp32, fp64, cplx64, cplx128
|
||||
- MT7004 支持fp16, fp32, int16, int32, cplx64
|
||||
|
||||
**共享/私有存储版本:**
|
||||
|
||||
.. c:function:: void anytype_squeeze_anycore(void* src, void* dst, int total_copy_size, int core_mask)
|
||||
|
||||
各种数据类型、私有及共享空间版本均使用该函数。
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 8
|
||||
|
||||
void main(){
|
||||
int core_mask = 0b1111; // 测试单核时核掩码设置为0b0001即可
|
||||
int core_num = GetCoreNum(core_mask);
|
||||
float* src = (float*)0x88000000; // 测试私有空间时地址设置在私有空间内即可
|
||||
float* dst = (float*)0x98000000;
|
||||
int shape[3] = {1, 10, 10};
|
||||
int total_copy_size = shape[0] * shape[1] * shape[2] * sizeof(float);
|
||||
anytype_squeeze_anycore(src, dst, total_copy_size, core_mask);
|
||||
}
|
||||
|
|
@ -0,0 +1,42 @@
|
|||
UnSqueeze
|
||||
=================
|
||||
|
||||
对输入张量在给定的轴上添加额外维度。由于该算子仅改变张量形状，因此其DSP算子的作用是将数据从输入张量完整拷贝到输出张量。
|
||||
|
||||
输入：
|
||||
- **src** - 输入地址
|
||||
- **total_copy_size** - 计算得到的总共需拷贝的数据量，单位为字节。
|
||||
- **core_mask** - 核掩码。
|
||||
|
||||
输出：
|
||||
- **dst** - 输出地址。
|
||||
|
||||
支持平台：
|
||||
``FT78NE``
|
||||
``MT7004``
|
||||
|
||||
.. note::
|
||||
- FT78NE 支持int8, int16, int32, fp32, fp64, cplx64, cplx128
|
||||
- MT7004 支持fp16, fp32, int16, int32, cplx64
|
||||
|
||||
**共享/私有存储版本:**
|
||||
|
||||
.. c:function:: void anytype_unsqueeze_anycore(void* src, void* dst, int total_copy_size, int core_mask)
|
||||
|
||||
各种数据类型、私有及共享空间版本均使用该函数。
|
||||
|
||||
**C调用示例：**
|
||||
|
||||
.. code-block:: c
|
||||
:linenos:
|
||||
:emphasize-lines: 8
|
||||
|
||||
void main(){
|
||||
int core_mask = 0b1111; // 测试单核时核掩码设置为0b0001即可
|
||||
int core_num = GetCoreNum(core_mask);
|
||||
float* src = (float*)0x88000000; // 测试私有空间时地址设置在私有空间内即可
|
||||
float* dst = (float*)0x98000000;
|
||||
int shape[3] = {1, 10, 10};
|
||||
int total_copy_size = shape[0] * shape[1] * shape[2] * sizeof(float);
|
||||
anytype_unsqueeze_anycore(src, dst, total_copy_size, core_mask);
|
||||
}
|
||||
|
|
@ -0,0 +1,887 @@
|
|||
|
||||
|
||||
<!DOCTYPE html>
|
||||
<html class="writer-html5" lang="zh-CN" data-content_root="../../">
|
||||
<head>
|
||||
<meta charset="utf-8" /><meta name="viewport" content="width=device-width, initial-scale=1" />
|
||||
|
||||
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
|
||||
<title>Activation — MindSpore Signal+ 使用手册 alpha 文档</title>
|
||||
<link rel="stylesheet" type="text/css" href="../../_static/pygments.css?v=03e43079" />
|
||||
<link rel="stylesheet" type="text/css" href="../../_static/css/theme.css?v=e59714d7" />
|
||||
<link rel="stylesheet" type="text/css" href="../../_static/table.css?v=72b86d3e" />
|
||||
|
||||
|
||||
<script src="../../_static/jquery.js?v=5d32c60e"></script>
|
||||
<script src="../../_static/_sphinx_javascript_frameworks_compat.js?v=2cd50e6c"></script>
|
||||
<script src="../../_static/documentation_options.js?v=406e4f49"></script>
|
||||
<script src="../../_static/doctools.js?v=9bcbadda"></script>
|
||||
<script src="../../_static/sphinx_highlight.js?v=dc90522c"></script>
|
||||
<script src="../../_static/translations.js?v=beaddf03"></script>
|
||||
<script async="async" src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js"></script>
|
||||
<script src="../../_static/table.js?v=99ed15ba"></script>
|
||||
<script src="../../_static/js/theme.js"></script>
|
||||
<link rel="index" title="索引" href="../../genindex.html" />
|
||||
<link rel="search" title="搜索" href="../../search.html" />
|
||||
<link rel="next" title="ExpFusion" href="expfusion.html" />
|
||||
<link rel="prev" title="Equal" href="equal.html" />
|
||||
</head>
|
||||
|
||||
<body class="wy-body-for-nav">
|
||||
<div class="wy-grid-for-nav">
|
||||
<nav data-toggle="wy-nav-shift" class="wy-nav-side">
|
||||
<div class="wy-side-scroll">
|
||||
<div class="wy-side-nav-search" >
|
||||
|
||||
|
||||
|
||||
<a href="../../index.html" class="icon icon-home">
|
||||
MindSpore Signal+ 使用手册
|
||||
</a>
|
||||
<div role="search">
|
||||
<form id="rtd-search-form" class="wy-form" action="../../search.html" method="get">
|
||||
<input type="text" name="q" placeholder="搜索文档" aria-label="搜索文档" />
|
||||
<input type="hidden" name="check_keywords" value="yes" />
|
||||
<input type="hidden" name="area" value="default" />
|
||||
</form>
|
||||
</div>
|
||||
</div><div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="导航菜单">
|
||||
<ul class="current">
|
||||
<li class="toctree-l1"><a class="reference internal" href="../../quickstart/index.html">快速入门</a></li>
|
||||
<li class="toctree-l1"><a class="reference internal" href="../../appdevelop/index.html">应用开发示例</a></li>
|
||||
<li class="toctree-l1 current"><a class="reference internal" href="../index.html">算子库支持</a><ul class="current">
|
||||
<li class="toctree-l2"><a class="reference internal" href="../supported_op.html">算子库支持情况</a></li>
|
||||
<li class="toctree-l2"><a class="reference internal" href="../custom_op/index.html">自定义算子列表</a></li>
|
||||
<li class="toctree-l2 current"><a class="reference internal" href="dsplib_index.html">DSP Library C API Reference</a><ul class="current">
|
||||
<li class="toctree-l3"><a class="reference internal" href="equal.html">Equal</a></li>
|
||||
<li class="toctree-l3 current"><a class="current reference internal" href="#">Activation</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="expfusion.html">ExpFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="reverse_sequence.html">ReverseSequence</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="reversev2.html">ReverseV2</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="fillv2.html">FillV2</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="squeeze.html">Squeeze</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="unsqueeze.html">UnSqueeze</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="expand_dims.html">ExpandDims</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="leaky_relu.html">LeakyReLu</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="lstm.html">LSTM</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="scatter_elements.html">ScatterElements</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="reduce.html">Reduce</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="resize.html">Resize</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="crop_and_resize.html">CropAndResize</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="crop.html">Crop</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="conv2d.html">Conv2d</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="adder.html">Adder</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="conv2d_transpose.html">Conv2dTranspose</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="gru.html">GRU</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="assert.html">Assert</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="range.html">Range</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="raggedrange.html">RaggedRange</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="linspace.html">LinSpace</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="matmulfusion.html">MatMulFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="attention.html">Attention</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="floor.html">Floor</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="floordiv.html">FloorDiv</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="embeddinglookup.html">EmbeddingLookup</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="eltwise.html">Eltwise</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="adamweightdecay.html">AdamWeightDecay</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="applymomentum.html">ApplyMomentum</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="avgpoolinggrad.html">AvgPoolingGrad</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="batchtospace.html">BatchToSpace</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="batchtospacend.html">BatchToSpaceND</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="broadcastto.html">BroadcastTo</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="depthtospace.html">DepthToSpace</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="spacetodepth.html">SpaceToDepth</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="spacetobatch.html">SpaceToBatch</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="spacetobatchnd.html">SpaceToBatchND</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="fusedbatchnorm.html">FusedBatchNorm</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="groupnormfusion.html">GroupNormFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="conv2dbackpropinputfusion.html">Conv2DBackpropInputFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="conv2dbackpropfilterfusion.html">Conv2DBackpropFilterFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="sgd.html">SGD</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="scalefusion.html">ScaleFusion</a></li>
|
||||
</ul>
|
||||
</li>
|
||||
</ul>
|
||||
</li>
|
||||
<li class="toctree-l1"><a class="reference internal" href="../../refdoc/index.html">参考资料</a></li>
|
||||
</ul>
|
||||
|
||||
</div>
|
||||
</div>
|
||||
</nav>
|
||||
|
||||
<section data-toggle="wy-nav-shift" class="wy-nav-content-wrap"><nav class="wy-nav-top" aria-label="移动版导航菜单" >
|
||||
<i data-toggle="wy-nav-top" class="fa fa-bars"></i>
|
||||
<a href="../../index.html">MindSpore Signal+ 使用手册</a>
|
||||
</nav>
|
||||
|
||||
<div class="wy-nav-content">
|
||||
<div class="rst-content">
|
||||
<div role="navigation" aria-label="页面导航">
|
||||
<ul class="wy-breadcrumbs">
|
||||
<li><a href="../../index.html" class="icon icon-home" aria-label="Home"></a></li>
|
||||
<li class="breadcrumb-item"><a href="../index.html">算子库支持</a></li>
|
||||
<li class="breadcrumb-item"><a href="dsplib_index.html">DSP Library C API Reference</a></li>
|
||||
<li class="breadcrumb-item active">Activation</li>
|
||||
<li class="wy-breadcrumbs-aside">
|
||||
<a href="../../_sources/functionlib/dsplib/activation.rst.txt" rel="nofollow"> 查看页面源码</a>
|
||||
</li>
|
||||
</ul>
|
||||
<hr/>
|
||||
</div>
|
||||
<div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
|
||||
<div itemprop="articleBody">
|
||||
|
||||
<section id="activation">
|
||||
<h1>Activation<a class="headerlink" href="#activation" title="Link to this heading"></a></h1>
|
||||
<p>对输入的数组中每一个元素执行激活函数计算，激活函数可选，具体函数见以下说明。</p>
|
||||
<ul>
|
||||
<li><p><code class="docutils literal notranslate"><span class="pre">Relu</span></code> - 标准Relu函数。</p>
|
||||
<blockquote>
|
||||
<div><div class="math notranslate nohighlight">
|
||||
\[output_i = \max(0, input_i)\]</div>
|
||||
</div></blockquote>
|
||||
</li>
|
||||
<li><p><code class="docutils literal notranslate"><span class="pre">Relu6</span></code> - 在标准Relu函数的基础上进行输出上限限制。</p>
|
||||
<blockquote>
|
||||
<div><div class="math notranslate nohighlight">
|
||||
\[output_i = \min(\max(0, input_i),6)\]</div>
|
||||
</div></blockquote>
|
||||
</li>
|
||||
<li><p><code class="docutils literal notranslate"><span class="pre">Clip</span></code> - 将输入裁剪到区间 [min_val, max_val]</p>
|
||||
<blockquote>
|
||||
<div><div class="math notranslate nohighlight">
|
||||
\[output_i = \min(\max(input_i, \text{min_val}), \text{max_val})\]</div>
|
||||
</div></blockquote>
|
||||
</li>
|
||||
<li><p><code class="docutils literal notranslate"><span class="pre">LRelu</span></code> - 带泄露的线性整流单元（Leaky Rectified Linear Unit），它在输入为正时保持线性，在输入为负时也保留一个很小的斜率，以避免标准 ReLU 中的“死亡神经元”问题。</p>
|
||||
<blockquote>
|
||||
<div><div class="math notranslate nohighlight">
|
||||
\[\begin{split}output_i =
|
||||
\begin{cases}
|
||||
input_i, & input_i \ge 0 \\
|
||||
\alpha \cdot input_i, & input_i < 0
|
||||
\end{cases}\end{split}\]</div>
|
||||
</div></blockquote>
|
||||
</li>
|
||||
<li><p><code class="docutils literal notranslate"><span class="pre">Sigmoid</span></code> - 常用的平滑非线性激活函数（又称逻辑函数），可以将任意实数映射到区间 <span class="math notranslate nohighlight">\((0, 1)\)</span>，常用于二分类问题的输出层，表示概率意义的结果。</p>
|
||||
<blockquote>
|
||||
<div><div class="math notranslate nohighlight">
|
||||
\[output_i = \frac{1}{1 + e^{-input_i}}\]</div>
|
||||
</div></blockquote>
|
||||
</li>
|
||||
<li><p><code class="docutils literal notranslate"><span class="pre">Tanh</span></code> - 双曲正切激活函数（Hyperbolic Tangent），其输出范围为 <span class="math notranslate nohighlight">\((-1, 1)\)</span> 。</p>
|
||||
<blockquote>
|
||||
<div><div class="math notranslate nohighlight">
|
||||
\[output_i = \tanh(input_i) = \frac{e^{input_i} - e^{-input_i}}{e^{input_i} + e^{-input_i}}\]</div>
|
||||
</div></blockquote>
|
||||
</li>
|
||||
<li><p><code class="docutils literal notranslate"><span class="pre">HSigmoid</span></code> - 硬 Sigmoid 激活函数（Hard Sigmoid），是 <code class="docutils literal notranslate"><span class="pre">Sigmoid</span></code> 函数的近似形式，计算简单、效率更高。</p>
|
||||
<blockquote>
|
||||
<div><div class="math notranslate nohighlight">
|
||||
\[output_i = \text{clip}\left(\frac{input_i + 3}{6}, 0, 1\right)\]</div>
|
||||
<p>其中 <code class="docutils literal notranslate"><span class="pre">clip(a,</span> <span class="pre">0,</span> <span class="pre">1)</span></code> 表示将 <code class="docutils literal notranslate"><span class="pre">a</span></code> 限制在区间 <span class="math notranslate nohighlight">\([0, 1]\)</span> 内。</p>
|
||||
</div></blockquote>
|
||||
</li>
|
||||
<li><p><code class="docutils literal notranslate"><span class="pre">Swish</span></code> - 自门控（Self-Gated）激活函数，由 Google 提出，结合了 <code class="docutils literal notranslate"><span class="pre">Sigmoid</span></code> 与线性特性，具有平滑且非单调的特点。</p>
|
||||
<blockquote>
|
||||
<div><div class="math notranslate nohighlight">
|
||||
\[output_i = input_i \cdot \sigma(input_i) = \frac{input_i}{1 + e^{-input_i}}\]</div>
|
||||
<p>其中 <span class="math notranslate nohighlight">\(\sigma(x)\)</span> 为标准 <code class="docutils literal notranslate"><span class="pre">Sigmoid</span></code> 函数。<code class="docutils literal notranslate"><span class="pre">Swish</span></code> 在深层网络中通常表现优于 <code class="docutils literal notranslate"><span class="pre">ReLU</span></code>。</p>
|
||||
</div></blockquote>
|
||||
</li>
|
||||
<li><p><code class="docutils literal notranslate"><span class="pre">HSwish</span></code> - 硬 Swish 激活函数（Hard Swish），是 <code class="docutils literal notranslate"><span class="pre">Swish</span></code> 函数的近似形式，计算简单且在移动端模型（如 MobileNetV3）中被广泛采用。</p>
|
||||
<blockquote>
|
||||
<div><div class="math notranslate nohighlight">
|
||||
\[output_i = input_i \cdot \text{clip}\left(\frac{input_i + 3}{6}, 0, 1\right)\]</div>
|
||||
<p>其中 <code class="docutils literal notranslate"><span class="pre">clip(a,</span> <span class="pre">0,</span> <span class="pre">1)</span></code> 表示将 <code class="docutils literal notranslate"><span class="pre">a</span></code> 限制在区间 <span class="math notranslate nohighlight">\([0, 1]\)</span> 内。</p>
|
||||
</div></blockquote>
|
||||
</li>
|
||||
<li><p><code class="docutils literal notranslate"><span class="pre">HardTanh</span></code> - 硬双曲正切激活函数（Hard Tanh），是 <code class="docutils literal notranslate"><span class="pre">Tanh</span></code> 函数的分段线性近似形式，计算简单、梯度稳定，常用于量化或轻量网络中。</p>
|
||||
<blockquote>
|
||||
<div><div class="math notranslate nohighlight">
|
||||
\[output_i = \text{clip}(input_i, min\_val, max\_val)\]</div>
|
||||
<p>其中 <code class="docutils literal notranslate"><span class="pre">clip(x,</span> <span class="pre">min_val,</span> <span class="pre">max_val)</span></code> 表示当 <span class="math notranslate nohighlight">\(x < min\_val\)</span> 时输出 min_val，当 <span class="math notranslate nohighlight">\(x > max\_val\)</span> 时输出 max_val，否则输出 <span class="math notranslate nohighlight">\(x\)</span> 本身。</p>
|
||||
</div></blockquote>
|
||||
</li>
|
||||
<li><p><code class="docutils literal notranslate"><span class="pre">Gelu</span></code> - 高斯误差线性单元（Gaussian Error Linear Unit），是一种平滑的非线性激活函数，结合了 <code class="docutils literal notranslate"><span class="pre">ReLU</span></code> 与概率特性。 该函数支持精确计算及非近似计算模式，近似算法由 <em>Hendrycks & Gimpel (2016)</em> 提出，用以替代精确形式 <span class="math notranslate nohighlight">\(output_i=x\Phi(x)\)</span> ，计算速度更快且精度损失极小。</p>
|
||||
<blockquote>
|
||||
<div><div class="math notranslate nohighlight">
|
||||
\[\begin{split}\begin{aligned}
|
||||
output_i =
|
||||
\begin{cases}
|
||||
0.5\,input_i \Bigl[ 1 + \tanh\!\Bigl(
|
||||
\sqrt{\frac{2}{\pi}}\,(input_i + 0.044715\,input_i^3)
|
||||
\Bigr) \Bigr], & flag = true, \\[6pt]
|
||||
input_i \,\Phi(input_i)
|
||||
= \tfrac{1}{2}x \Bigl[
|
||||
1 + \mathrm{erf}\!\Bigl(\tfrac{input_i}{\sqrt{2}}\Bigr)
|
||||
\Bigr], & flag = false.
|
||||
\end{cases}
|
||||
\end{aligned}\end{split}\]</div>
|
||||
<p>其中 <span class="math notranslate nohighlight">\(\Phi(x)\)</span> 为标准正态分布的累积分布函数。</p>
|
||||
</div></blockquote>
|
||||
</li>
|
||||
<li><p><code class="docutils literal notranslate"><span class="pre">Softplus</span></code> - <code class="docutils literal notranslate"><span class="pre">ReLU</span></code> 的平滑近似形式，能在零点处保持可导性。</p>
|
||||
<blockquote>
|
||||
<div><div class="math notranslate nohighlight">
|
||||
\[\begin{split}output_i =
|
||||
\begin{cases}
|
||||
input_i, & input_i \gt 88.0 \\
|
||||
\ln(1 + e^{input_i}), & \text{otherwise}
|
||||
\end{cases}\end{split}\]</div>
|
||||
</div></blockquote>
|
||||
</li>
|
||||
<li><p><code class="docutils literal notranslate"><span class="pre">Elu</span></code> - 在输入为正时保持线性，在输入为负时呈指数衰减，可缓解 ReLU 的“死亡神经元”问题。</p>
|
||||
<blockquote>
|
||||
<div><div class="math notranslate nohighlight">
|
||||
\[\begin{split}output_i =
|
||||
\begin{cases}
|
||||
input_i, & input_i \ge 0 \\
|
||||
\alpha (e^{input_i} - 1), & input_i < 0
|
||||
\end{cases}\end{split}\]</div>
|
||||
</div></blockquote>
|
||||
<p>其中 <span class="math notranslate nohighlight">\(\alpha\)</span> 为超参数，通常取 <span class="math notranslate nohighlight">\(\alpha = 1.0\)</span>。</p>
|
||||
</li>
|
||||
<li><p><code class="docutils literal notranslate"><span class="pre">Celu</span></code> - 连续指数线性单元（Continuously Differentiable ELU），是 <code class="docutils literal notranslate"><span class="pre">ELU</span></code> 的改进版本，保证在零点处连续可导。</p>
|
||||
<blockquote>
|
||||
<div><div class="math notranslate nohighlight">
|
||||
\[\begin{split}output_i =
|
||||
\begin{cases}
|
||||
input_i, & input_i \ge 0 \\
|
||||
\alpha (e^{\frac{input_i}{alpha}} - 1), & input_i < 0
|
||||
\end{cases}\end{split}\]</div>
|
||||
<p>其中 <span class="math notranslate nohighlight">\(\alpha\)</span> 为可调超参数，用于控制负区间的平滑程度。</p>
|
||||
</div></blockquote>
|
||||
</li>
|
||||
<li><p><code class="docutils literal notranslate"><span class="pre">HardShrink</span></code> - 硬收缩激活函数（Hard Shrinkage），用于稀疏化输出。</p>
|
||||
<blockquote>
|
||||
<div><div class="math notranslate nohighlight">
|
||||
\[\begin{split}output_i =
|
||||
\begin{cases}
|
||||
input_i, & \text{if } |input_i| > \lambda \\
|
||||
0, & \text{otherwise}
|
||||
\end{cases}\end{split}\]</div>
|
||||
</div></blockquote>
|
||||
<p>其中 <span class="math notranslate nohighlight">\(\lambda\)</span> 为阈值常数。</p>
|
||||
</li>
|
||||
<li><p><code class="docutils literal notranslate"><span class="pre">SoftShrink</span></code> - 软收缩激活函数（Soft Shrinkage），与 <code class="docutils literal notranslate"><span class="pre">HardShrink</span></code> 类似，但收缩过程更加平滑。</p>
|
||||
<blockquote>
|
||||
<div><div class="math notranslate nohighlight">
|
||||
\[\begin{split}output_i =
|
||||
\begin{cases}
|
||||
input_i - \lambda, & \text{if } input_i > \lambda \\
|
||||
input_i + \lambda, & \text{if } input_i < -\lambda \\
|
||||
0, & \text{otherwise}
|
||||
\end{cases}\end{split}\]</div>
|
||||
</div></blockquote>
|
||||
</li>
|
||||
<li><p><code class="docutils literal notranslate"><span class="pre">SoftsignOpt</span></code> - 优化的软符号函数（Optimized Softsign），是一种平滑的压缩函数，用于将输入映射到有限区间。</p>
|
||||
<blockquote>
|
||||
<div><div class="math notranslate nohighlight">
|
||||
\[output_i = \frac{input_i}{1 + |input_i|}\]</div>
|
||||
</div></blockquote>
|
||||
</li>
|
||||
</ul>
|
||||
<dl class="simple">
|
||||
<dt>输入：</dt><dd><ul class="simple">
|
||||
<li><p><strong>Input0</strong> - 输入数据地址。</p></li>
|
||||
<li><p><strong>length</strong> - 数组长度。</p></li>
|
||||
<li><p><strong>args(部分激活函数)</strong> - 激活函数计算参数（仅适用于部分函数）。</p></li>
|
||||
<li><p><strong>core_mask(int, 可选)</strong> - 核掩码（仅适用于共享存储版本）。</p></li>
|
||||
</ul>
|
||||
</dd>
|
||||
<dt>输出：</dt><dd><ul class="simple">
|
||||
<li><p><strong>output</strong> - 计算结果地址。</p></li>
|
||||
</ul>
|
||||
</dd>
|
||||
<dt>支持平台：</dt><dd><p><code class="docutils literal notranslate"><span class="pre">FT78NE</span></code>
|
||||
<code class="docutils literal notranslate"><span class="pre">MT7004</span></code></p>
|
||||
</dd>
|
||||
</dl>
|
||||
<div class="admonition note">
|
||||
<p class="admonition-title">备注</p>
|
||||
<ul class="simple">
|
||||
<li><p>FT78NE 支持int8, fp32</p></li>
|
||||
<li><p>MT7004 支持fp16, fp32</p></li>
|
||||
</ul>
|
||||
</div>
|
||||
<p><strong>共享存储版本：</strong></p>
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_relu_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_relu_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_relu_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_relu_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_relu_s</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_relu_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_relu_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_relu_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_relu_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_relu6_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_relu6_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_relu6_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_relu6_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_relu6_s</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_relu6_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_relu6_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_relu6_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_relu6_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_clip_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_clip_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="n"><span class="pre">min_val</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="n"><span class="pre">max_val</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_clip_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_clip_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_clip_s</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">min_val</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">max_val</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_clip_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_clip_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_clip_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="n"><span class="pre">min_val</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="n"><span class="pre">max_val</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_clip_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_lrelu_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_lrelu_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">alpha</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_lrelu_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_lrelu_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_lrelu_s</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">alpha</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_lrelu_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_lrelu_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_lrelu_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="n"><span class="pre">alpha</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_lrelu_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_sigmoid_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_sigmoid_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_sigmoid_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_sigmoid_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_sigmoid_s</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_sigmoid_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_sigmoid_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_sigmoid_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_sigmoid_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_tanh_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_tanh_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_tanh_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_tanh_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_tanh_s</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_tanh_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_tanh_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_tanh_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_tanh_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_hsigmoid_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_hsigmoid_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_hsigmoid_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_hsigmoid_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_hsigmoid_s</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_hsigmoid_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_hsigmoid_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_hsigmoid_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_hsigmoid_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_swish_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_swish_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_swish_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_swish_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_swish_s</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_swish_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_swish_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_swish_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_swish_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_hswish_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_hswish_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_hswish_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_hswish_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_hswish_s</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_hswish_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_hswish_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_hswish_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_hswish_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_hardtanh_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_hardtanh_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="n"><span class="pre">min_val</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="n"><span class="pre">max_val</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_hardtanh_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_hardtanh_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_hardtanh_s</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">min_val</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">max_val</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_hardtanh_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_hardtanh_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_hardtanh_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="n"><span class="pre">min_val</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="n"><span class="pre">max_val</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_hardtanh_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_gelu_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_gelu_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">approximate</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_gelu_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_gelu_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_gelu_s</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">approximate</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_gelu_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_gelu_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_gelu_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">approximate</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_gelu_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_softplus_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_softplus_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_softplus_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_softplus_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_softplus_s</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_softplus_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_softplus_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_softplus_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_softplus_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_elu_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_elu_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">alpha</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_elu_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_elu_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_elu_s</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">alpha</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_elu_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_elu_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_elu_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="n"><span class="pre">alpha</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_elu_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_celu_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_celu_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">alpha</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_celu_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_celu_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_celu_s</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">alpha</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_celu_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_celu_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_celu_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="n"><span class="pre">alpha</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_celu_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_hardshrink_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_hardshrink_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="n"><span class="pre">lambd</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_hardshrink_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_hardshrink_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_hardshrink_s</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">lambd</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_hardshrink_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_hardshrink_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_hardshrink_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="n"><span class="pre">lambd</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_hardshrink_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_softshrink_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_softshrink_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="n"><span class="pre">lambd</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_softshrink_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_softshrink_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_softshrink_s</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">lambd</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_softshrink_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_softshrink_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_softshrink_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="n"><span class="pre">lambd</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_softshrink_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_softsignopt_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_softsignopt_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_softsignopt_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_softsignopt_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_softsignopt_s</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_softsignopt_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_softsignopt_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_softsignopt_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_softsignopt_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<p><strong>C调用示例：</strong></p>
|
||||
<blockquote>
|
||||
<div><div class="highlight-c notranslate"><div class="highlight"><pre><span></span><span class="linenos"> 1</span><span class="c1">//FT78NE示例</span>
|
||||
<span class="linenos"> 2</span><span class="cp">#include</span><span class="w"> </span><span class="cpf"><stdio.h></span>
|
||||
<span class="linenos"> 3</span><span class="cp">#include</span><span class="w"> </span><span class="cpf"><activation.h></span>
|
||||
<span class="linenos"> 4</span>
|
||||
<span class="linenos"> 5</span><span class="kt">int</span><span class="w"> </span><span class="nf">main</span><span class="p">(</span><span class="kt">int</span><span class="w"> </span><span class="n">argc</span><span class="p">,</span><span class="w"> </span><span class="kt">char</span><span class="o">*</span><span class="w"> </span><span class="n">argv</span><span class="p">[])</span><span class="w"> </span><span class="p">{</span>
|
||||
<span class="linenos"> 6</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="n">input0</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0xA0000000</span><span class="p">;</span><span class="w"> </span><span class="c1">//input在DDR空间</span>
|
||||
<span class="linenos"> 7</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="n">output</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0xC0000000</span><span class="p">;</span>
|
||||
<span class="linenos"> 8</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">length</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">1000</span><span class="p">;</span>
|
||||
<span class="linenos"> 9</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">core_mask</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mh">0xff</span><span class="p">;</span>
|
||||
<span class="hll"><span class="linenos">10</span><span class="w"> </span><span class="n">fp_tanh_s</span><span class="p">(</span><span class="n">input0</span><span class="p">,</span><span class="w"> </span><span class="n">output</span><span class="p">,</span><span class="w"> </span><span class="n">length</span><span class="p">,</span><span class="w"> </span><span class="n">core_mask</span><span class="p">);</span>
|
||||
</span><span class="linenos">11</span><span class="w"> </span><span class="k">return</span><span class="w"> </span><span class="mi">0</span><span class="p">;</span>
|
||||
<span class="linenos">12</span><span class="p">}</span>
|
||||
</pre></div>
|
||||
</div>
|
||||
</div></blockquote>
|
||||
<p><strong>私有存储版本：</strong></p>
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_relu_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_relu_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_relu_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_relu_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_relu_p</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_relu_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_relu_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_relu_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_relu_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_relu6_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_relu6_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_relu6_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_relu6_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_relu6_p</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_relu6_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_relu6_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_relu6_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_relu6_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_clip_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_clip_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="n"><span class="pre">min_val</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="n"><span class="pre">max_val</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_clip_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_clip_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_clip_p</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">min_val</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">max_val</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_clip_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_clip_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_clip_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="n"><span class="pre">min_val</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="n"><span class="pre">max_val</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_clip_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_lrelu_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_lrelu_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">alpha</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_lrelu_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_lrelu_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_lrelu_p</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">alpha</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_lrelu_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_lrelu_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_lrelu_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="n"><span class="pre">alpha</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_lrelu_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_sigmoid_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_sigmoid_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_sigmoid_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_sigmoid_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_sigmoid_p</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_sigmoid_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_sigmoid_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_sigmoid_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_sigmoid_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_tanh_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_tanh_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_tanh_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_tanh_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_tanh_p</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_tanh_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_tanh_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_tanh_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_tanh_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_hsigmoid_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_hsigmoid_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_hsigmoid_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_hsigmoid_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_hsigmoid_p</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_hsigmoid_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_hsigmoid_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_hsigmoid_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_hsigmoid_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_swish_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_swish_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_swish_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_swish_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_swish_p</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_swish_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_swish_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_swish_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_swish_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_hswish_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_hswish_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_hswish_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_hswish_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_hswish_p</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_hswish_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_hswish_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_hswish_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_hswish_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_hardtanh_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_hardtanh_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="n"><span class="pre">min_val</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="n"><span class="pre">max_val</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_hardtanh_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_hardtanh_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_hardtanh_p</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">min_val</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">max_val</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_hardtanh_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_hardtanh_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_hardtanh_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="n"><span class="pre">min_val</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="n"><span class="pre">max_val</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_hardtanh_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_gelu_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_gelu_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">approximate</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_gelu_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_gelu_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_gelu_p</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">approximate</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_gelu_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_gelu_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_gelu_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">approximate</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_gelu_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_softplus_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_softplus_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_softplus_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_softplus_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_softplus_p</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_softplus_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_softplus_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_softplus_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_softplus_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_elu_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_elu_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">alpha</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_elu_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_elu_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_elu_p</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">alpha</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_elu_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_elu_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_elu_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="n"><span class="pre">alpha</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_elu_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_celu_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_celu_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">alpha</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_celu_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_celu_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_celu_p</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">alpha</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_celu_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_celu_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_celu_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="n"><span class="pre">alpha</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_celu_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_hardshrink_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_hardshrink_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="n"><span class="pre">lambd</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_hardshrink_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_hardshrink_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_hardshrink_p</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">lambd</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_hardshrink_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_hardshrink_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_hardshrink_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="n"><span class="pre">lambd</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_hardshrink_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_softshrink_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_softshrink_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="n"><span class="pre">lambd</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_softshrink_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_softshrink_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_softshrink_p</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">lambd</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_softshrink_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_softshrink_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_softshrink_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="n"><span class="pre">lambd</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_softshrink_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_softsignopt_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_softsignopt_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_softsignopt_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_softsignopt_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_softsignopt_p</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_softsignopt_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_softsignopt_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_softsignopt_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input0</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_softsignopt_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<p><strong>C调用示例：</strong></p>
|
||||
<blockquote>
|
||||
<div><div class="highlight-c notranslate"><div class="highlight"><pre><span></span><span class="linenos"> 1</span><span class="c1">//FT78NE示例</span>
|
||||
<span class="linenos"> 2</span><span class="cp">#include</span><span class="w"> </span><span class="cpf"><stdio.h></span>
|
||||
<span class="linenos"> 3</span><span class="cp">#include</span><span class="w"> </span><span class="cpf"><activation.h></span>
|
||||
<span class="linenos"> 4</span>
|
||||
<span class="linenos"> 5</span><span class="kt">int</span><span class="w"> </span><span class="nf">main</span><span class="p">(</span><span class="kt">int</span><span class="w"> </span><span class="n">argc</span><span class="p">,</span><span class="w"> </span><span class="kt">char</span><span class="o">*</span><span class="w"> </span><span class="n">argv</span><span class="p">[])</span><span class="w"> </span><span class="p">{</span>
|
||||
<span class="linenos"> 6</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="n">input0</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0x10000000</span><span class="p">;</span><span class="w"> </span><span class="c1">//input在DDR空间</span>
|
||||
<span class="linenos"> 7</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="n">output</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0x10004000</span><span class="p">;</span>
|
||||
<span class="linenos"> 8</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">length</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">1000</span><span class="p">;</span>
|
||||
<span class="hll"><span class="linenos"> 9</span><span class="w"> </span><span class="n">fp_tanh_p</span><span class="p">(</span><span class="n">input0</span><span class="p">,</span><span class="w"> </span><span class="n">output</span><span class="p">,</span><span class="w"> </span><span class="n">length</span><span class="p">);</span>
|
||||
</span><span class="linenos">10</span><span class="w"> </span><span class="k">return</span><span class="w"> </span><span class="mi">0</span><span class="p">;</span>
|
||||
<span class="linenos">11</span><span class="p">}</span>
|
||||
</pre></div>
|
||||
</div>
|
||||
</div></blockquote>
|
||||
</section>
|
||||
|
||||
|
||||
</div>
|
||||
</div>
|
||||
<footer><div class="rst-footer-buttons" role="navigation" aria-label="页脚">
|
||||
<a href="equal.html" class="btn btn-neutral float-left" title="Equal" accesskey="p" rel="prev"><span class="fa fa-arrow-circle-left" aria-hidden="true"></span> 上一页</a>
|
||||
<a href="expfusion.html" class="btn btn-neutral float-right" title="ExpFusion" accesskey="n" rel="next">下一页 <span class="fa fa-arrow-circle-right" aria-hidden="true"></span></a>
|
||||
</div>
|
||||
|
||||
<hr/>
|
||||
|
||||
<div role="contentinfo">
|
||||
<p>© 版权所有 2025 - 2025, NUDT-674。</p>
|
||||
</div>
|
||||
|
||||
利用 <a href="https://www.sphinx-doc.org/">Sphinx</a> 构建，使用的
|
||||
<a href="https://github.com/readthedocs/sphinx_rtd_theme">主题</a>
|
||||
由 <a href="https://readthedocs.org">Read the Docs</a> 开发.
|
||||
|
||||
|
||||
</footer>
|
||||
</div>
|
||||
</div>
|
||||
</section>
|
||||
</div>
|
||||
<script>
|
||||
jQuery(function () {
|
||||
SphinxRtdTheme.Navigation.enable(true);
|
||||
});
|
||||
</script>
|
||||
|
||||
</body>
|
||||
</html>
|
||||
|
|
@ -0,0 +1,281 @@
|
|||
|
||||
|
||||
<!DOCTYPE html>
|
||||
<html class="writer-html5" lang="zh-CN" data-content_root="../../">
|
||||
<head>
|
||||
<meta charset="utf-8" /><meta name="viewport" content="width=device-width, initial-scale=1" />
|
||||
|
||||
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
|
||||
<title>AdamWeightDecay — MindSpore Signal+ 使用手册 alpha 文档</title>
|
||||
<link rel="stylesheet" type="text/css" href="../../_static/pygments.css?v=03e43079" />
|
||||
<link rel="stylesheet" type="text/css" href="../../_static/css/theme.css?v=e59714d7" />
|
||||
<link rel="stylesheet" type="text/css" href="../../_static/table.css?v=72b86d3e" />
|
||||
|
||||
|
||||
<script src="../../_static/jquery.js?v=5d32c60e"></script>
|
||||
<script src="../../_static/_sphinx_javascript_frameworks_compat.js?v=2cd50e6c"></script>
|
||||
<script src="../../_static/documentation_options.js?v=406e4f49"></script>
|
||||
<script src="../../_static/doctools.js?v=9bcbadda"></script>
|
||||
<script src="../../_static/sphinx_highlight.js?v=dc90522c"></script>
|
||||
<script src="../../_static/translations.js?v=beaddf03"></script>
|
||||
<script async="async" src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js"></script>
|
||||
<script src="../../_static/table.js?v=99ed15ba"></script>
|
||||
<script src="../../_static/js/theme.js"></script>
|
||||
<link rel="index" title="索引" href="../../genindex.html" />
|
||||
<link rel="search" title="搜索" href="../../search.html" />
|
||||
<link rel="next" title="ApplyMomentum" href="applymomentum.html" />
|
||||
<link rel="prev" title="Eltwise" href="eltwise.html" />
|
||||
</head>
|
||||
|
||||
<body class="wy-body-for-nav">
|
||||
<div class="wy-grid-for-nav">
|
||||
<nav data-toggle="wy-nav-shift" class="wy-nav-side">
|
||||
<div class="wy-side-scroll">
|
||||
<div class="wy-side-nav-search" >
|
||||
|
||||
|
||||
|
||||
<a href="../../index.html" class="icon icon-home">
|
||||
MindSpore Signal+ 使用手册
|
||||
</a>
|
||||
<div role="search">
|
||||
<form id="rtd-search-form" class="wy-form" action="../../search.html" method="get">
|
||||
<input type="text" name="q" placeholder="搜索文档" aria-label="搜索文档" />
|
||||
<input type="hidden" name="check_keywords" value="yes" />
|
||||
<input type="hidden" name="area" value="default" />
|
||||
</form>
|
||||
</div>
|
||||
</div><div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="导航菜单">
|
||||
<ul class="current">
|
||||
<li class="toctree-l1"><a class="reference internal" href="../../quickstart/index.html">快速入门</a></li>
|
||||
<li class="toctree-l1"><a class="reference internal" href="../../appdevelop/index.html">应用开发示例</a></li>
|
||||
<li class="toctree-l1 current"><a class="reference internal" href="../index.html">算子库支持</a><ul class="current">
|
||||
<li class="toctree-l2"><a class="reference internal" href="../supported_op.html">算子库支持情况</a></li>
|
||||
<li class="toctree-l2"><a class="reference internal" href="../custom_op/index.html">自定义算子列表</a></li>
|
||||
<li class="toctree-l2 current"><a class="reference internal" href="dsplib_index.html">DSP Library C API Reference</a><ul class="current">
|
||||
<li class="toctree-l3"><a class="reference internal" href="equal.html">Equal</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="activation.html">Activation</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="expfusion.html">ExpFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="reverse_sequence.html">ReverseSequence</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="reversev2.html">ReverseV2</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="fillv2.html">FillV2</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="squeeze.html">Squeeze</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="unsqueeze.html">UnSqueeze</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="expand_dims.html">ExpandDims</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="leaky_relu.html">LeakyReLu</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="lstm.html">LSTM</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="scatter_elements.html">ScatterElements</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="reduce.html">Reduce</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="resize.html">Resize</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="crop_and_resize.html">CropAndResize</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="crop.html">Crop</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="conv2d.html">Conv2d</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="adder.html">Adder</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="conv2d_transpose.html">Conv2dTranspose</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="gru.html">GRU</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="assert.html">Assert</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="range.html">Range</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="raggedrange.html">RaggedRange</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="linspace.html">LinSpace</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="matmulfusion.html">MatMulFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="attention.html">Attention</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="floor.html">Floor</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="floordiv.html">FloorDiv</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="embeddinglookup.html">EmbeddingLookup</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="eltwise.html">Eltwise</a></li>
|
||||
<li class="toctree-l3 current"><a class="current reference internal" href="#">AdamWeightDecay</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="applymomentum.html">ApplyMomentum</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="avgpoolinggrad.html">AvgPoolingGrad</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="batchtospace.html">BatchToSpace</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="batchtospacend.html">BatchToSpaceND</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="broadcastto.html">BroadcastTo</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="depthtospace.html">DepthToSpace</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="spacetodepth.html">SpaceToDepth</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="spacetobatch.html">SpaceToBatch</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="spacetobatchnd.html">SpaceToBatchND</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="fusedbatchnorm.html">FusedBatchNorm</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="groupnormfusion.html">GroupNormFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="conv2dbackpropinputfusion.html">Conv2DBackpropInputFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="conv2dbackpropfilterfusion.html">Conv2DBackpropFilterFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="sgd.html">SGD</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="scalefusion.html">ScaleFusion</a></li>
|
||||
</ul>
|
||||
</li>
|
||||
</ul>
|
||||
</li>
|
||||
<li class="toctree-l1"><a class="reference internal" href="../../refdoc/index.html">参考资料</a></li>
|
||||
</ul>
|
||||
|
||||
</div>
|
||||
</div>
|
||||
</nav>
|
||||
|
||||
<section data-toggle="wy-nav-shift" class="wy-nav-content-wrap"><nav class="wy-nav-top" aria-label="移动版导航菜单" >
|
||||
<i data-toggle="wy-nav-top" class="fa fa-bars"></i>
|
||||
<a href="../../index.html">MindSpore Signal+ 使用手册</a>
|
||||
</nav>
|
||||
|
||||
<div class="wy-nav-content">
|
||||
<div class="rst-content">
|
||||
<div role="navigation" aria-label="页面导航">
|
||||
<ul class="wy-breadcrumbs">
|
||||
<li><a href="../../index.html" class="icon icon-home" aria-label="Home"></a></li>
|
||||
<li class="breadcrumb-item"><a href="../index.html">算子库支持</a></li>
|
||||
<li class="breadcrumb-item"><a href="dsplib_index.html">DSP Library C API Reference</a></li>
|
||||
<li class="breadcrumb-item active">AdamWeightDecay</li>
|
||||
<li class="wy-breadcrumbs-aside">
|
||||
<a href="../../_sources/functionlib/dsplib/adamweightdecay.rst.txt" rel="nofollow"> 查看页面源码</a>
|
||||
</li>
|
||||
</ul>
|
||||
<hr/>
|
||||
</div>
|
||||
<div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
|
||||
<div itemprop="articleBody">
|
||||
|
||||
<section id="adamweightdecay">
|
||||
<h1>AdamWeightDecay<a class="headerlink" href="#adamweightdecay" title="Link to this heading"></a></h1>
|
||||
<p>对权重张量执行 Adam Weight Decay 优化更新。</p>
|
||||
<blockquote>
|
||||
<div><div class="math notranslate nohighlight">
|
||||
\[\begin{split}\begin{aligned}
|
||||
m_t &= \beta_1 \cdot m_{t-1} + (1 - \beta_1) \cdot g_t \\
|
||||
v_t &= \beta_2 \cdot v_{t-1} + (1 - \beta_2) \cdot g_t^2 \\
|
||||
\hat{m}_t &= \frac{m_t}{\sqrt{v_t} + \epsilon} \\
|
||||
var_t &= var_{t-1} - lr \cdot (\hat{m}_t + decay \cdot var_{t-1})
|
||||
\end{aligned}\end{split}\]</div>
|
||||
<dl class="simple">
|
||||
<dt>输入：</dt><dd><ul class="simple">
|
||||
<li><p><strong>var</strong> - 待更新权重张量首地址。</p></li>
|
||||
<li><p><strong>m</strong> - 一阶动量张量首地址。</p></li>
|
||||
<li><p><strong>v</strong> - 二阶动量张量首地址。</p></li>
|
||||
<li><p><strong>gradient</strong> - 梯度张量首地址。</p></li>
|
||||
<li><p><strong>lr</strong> - 学习率。</p></li>
|
||||
<li><p><strong>beta1</strong> - 一阶动量衰减系数。</p></li>
|
||||
<li><p><strong>beta2</strong> - 二阶动量衰减系数。</p></li>
|
||||
<li><p><strong>epsilon</strong> - 数值稳定项。</p></li>
|
||||
<li><p><strong>decay</strong> - 权重衰减系数。</p></li>
|
||||
<li><p><strong>start</strong> - 参与计算的起始索引（闭区间）。</p></li>
|
||||
<li><p><strong>end</strong> - 参与计算的结束索引（开区间）。</p></li>
|
||||
<li><p><strong>core_mask(int, 可选)</strong> - 核掩码（仅适用于共享存储版本）。</p></li>
|
||||
</ul>
|
||||
</dd>
|
||||
<dt>输出：</dt><dd><ul class="simple">
|
||||
<li><p><strong>var</strong> - 原地写回更新后的权重张量。</p></li>
|
||||
<li><p><strong>m</strong> - 原地写回更新后的一阶动量张量。</p></li>
|
||||
<li><p><strong>v</strong> - 原地写回更新后的二阶动量张量。</p></li>
|
||||
</ul>
|
||||
</dd>
|
||||
<dt>支持平台：</dt><dd><p><code class="docutils literal notranslate"><span class="pre">FT78NE</span></code>
|
||||
<code class="docutils literal notranslate"><span class="pre">MT7004</span></code></p>
|
||||
</dd>
|
||||
</dl>
|
||||
<div class="admonition note">
|
||||
<p class="admonition-title">备注</p>
|
||||
<ul class="simple">
|
||||
<li><p>FT78NE 支持 fp32 数据类型。</p></li>
|
||||
<li><p>MT7004 支持 fp16、fp32 数据类型。</p></li>
|
||||
</ul>
|
||||
</div>
|
||||
</div></blockquote>
|
||||
<p><strong>共享存储版本:</strong></p>
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_adamweightdecay_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_adamweightdecay_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">var</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">m</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">v</span></span>, <span class="k"><span class="pre">const</span></span><span class="w"> </span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">gradient</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">lr</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">beta1</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">beta2</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">epsilon</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">decay</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">start</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">end</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_adamweightdecay_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_adamweightdecay_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_adamweightdecay_s</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">var</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">m</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">v</span></span>, <span class="k"><span class="pre">const</span></span><span class="w"> </span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">gradient</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">lr</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">beta1</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">beta2</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">epsilon</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">decay</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">start</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">end</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_adamweightdecay_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd><p><strong>C调用示例：</strong></p>
|
||||
<div class="highlight-c notranslate"><div class="highlight"><pre><span></span><span class="linenos"> 1</span><span class="c1">// FT78NE 多核示例</span>
|
||||
<span class="linenos"> 2</span><span class="cp">#include</span><span class="w"> </span><span class="cpf"><stdio.h></span>
|
||||
<span class="linenos"> 3</span>
|
||||
<span class="linenos"> 4</span><span class="kt">int</span><span class="w"> </span><span class="nf">main</span><span class="p">(</span><span class="kt">void</span><span class="p">)</span><span class="w"> </span><span class="p">{</span>
|
||||
<span class="linenos"> 5</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="n">var</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0xA0000000</span><span class="p">;</span><span class="w"> </span><span class="c1">// DDR 存储</span>
|
||||
<span class="linenos"> 6</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="n">m</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0xB0000000</span><span class="p">;</span>
|
||||
<span class="linenos"> 7</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="n">v</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0xC0000000</span><span class="p">;</span>
|
||||
<span class="linenos"> 8</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="n">gradient</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0xD0000000</span><span class="p">;</span>
|
||||
<span class="linenos"> 9</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">start</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">0</span><span class="p">;</span>
|
||||
<span class="linenos">10</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">end</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">4096</span><span class="p">;</span>
|
||||
<span class="linenos">11</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">core_mask</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mh">0xff</span><span class="p">;</span>
|
||||
<span class="linenos">12</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="n">lr</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mf">1e-3f</span><span class="p">;</span>
|
||||
<span class="linenos">13</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="n">beta1</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mf">0.9f</span><span class="p">;</span>
|
||||
<span class="linenos">14</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="n">beta2</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mf">0.999f</span><span class="p">;</span>
|
||||
<span class="linenos">15</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="n">epsilon</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mf">1e-8f</span><span class="p">;</span>
|
||||
<span class="linenos">16</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="n">decay</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mf">1e-2f</span><span class="p">;</span>
|
||||
<span class="hll"><span class="linenos">17</span><span class="w"> </span><span class="n">fp_adamweightdecay_s</span><span class="p">(</span><span class="n">var</span><span class="p">,</span><span class="w"> </span><span class="n">m</span><span class="p">,</span><span class="w"> </span><span class="n">v</span><span class="p">,</span><span class="w"> </span><span class="n">gradient</span><span class="p">,</span><span class="w"> </span><span class="n">lr</span><span class="p">,</span>
|
||||
</span><span class="linenos">18</span><span class="w"> </span><span class="n">beta1</span><span class="p">,</span><span class="w"> </span><span class="n">beta2</span><span class="p">,</span><span class="w"> </span><span class="n">epsilon</span><span class="p">,</span><span class="w"> </span><span class="n">decay</span><span class="p">,</span>
|
||||
<span class="linenos">19</span><span class="w"> </span><span class="n">start</span><span class="p">,</span><span class="w"> </span><span class="n">end</span><span class="p">,</span><span class="w"> </span><span class="n">core_mask</span><span class="p">);</span>
|
||||
<span class="linenos">20</span><span class="w"> </span><span class="k">return</span><span class="w"> </span><span class="mi">0</span><span class="p">;</span>
|
||||
<span class="linenos">21</span><span class="p">}</span>
|
||||
</pre></div>
|
||||
</div>
|
||||
</dd></dl>
|
||||
|
||||
<p><strong>私有存储版本:</strong></p>
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_adamweightdecay_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_adamweightdecay_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">var</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">m</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">v</span></span>, <span class="k"><span class="pre">const</span></span><span class="w"> </span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">gradient</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">lr</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">beta1</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">beta2</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">epsilon</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">decay</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_adamweightdecay_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_adamweightdecay_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_adamweightdecay_p</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">var</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">m</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">v</span></span>, <span class="k"><span class="pre">const</span></span><span class="w"> </span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">gradient</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">lr</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">beta1</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">beta2</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">epsilon</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">decay</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_adamweightdecay_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd><p><strong>C调用示例：</strong></p>
|
||||
<div class="highlight-c notranslate"><div class="highlight"><pre><span></span><span class="linenos"> 1</span><span class="c1">// MT7004 单核示例</span>
|
||||
<span class="linenos"> 2</span><span class="cp">#include</span><span class="w"> </span><span class="cpf"><stdio.h></span>
|
||||
<span class="linenos"> 3</span>
|
||||
<span class="linenos"> 4</span><span class="kt">int</span><span class="w"> </span><span class="nf">main</span><span class="p">(</span><span class="kt">void</span><span class="p">)</span><span class="w"> </span><span class="p">{</span>
|
||||
<span class="linenos"> 5</span><span class="w"> </span><span class="n">half</span><span class="w"> </span><span class="o">*</span><span class="n">var</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="n">half</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0x10000000</span><span class="p">;</span><span class="w"> </span><span class="c1">// L2 存储</span>
|
||||
<span class="linenos"> 6</span><span class="w"> </span><span class="n">half</span><span class="w"> </span><span class="o">*</span><span class="n">m</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="n">half</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0x10002000</span><span class="p">;</span>
|
||||
<span class="linenos"> 7</span><span class="w"> </span><span class="n">half</span><span class="w"> </span><span class="o">*</span><span class="n">v</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="n">half</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0x10004000</span><span class="p">;</span>
|
||||
<span class="linenos"> 8</span><span class="w"> </span><span class="n">half</span><span class="w"> </span><span class="o">*</span><span class="n">gradient</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="n">half</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0x10006000</span><span class="p">;</span>
|
||||
<span class="linenos"> 9</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">length</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">2048</span><span class="p">;</span>
|
||||
<span class="linenos">10</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="n">lr</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mf">5e-4f</span><span class="p">;</span>
|
||||
<span class="linenos">11</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="n">beta1</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mf">0.9f</span><span class="p">;</span>
|
||||
<span class="linenos">12</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="n">beta2</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mf">0.999f</span><span class="p">;</span>
|
||||
<span class="linenos">13</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="n">epsilon</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mf">1e-6f</span><span class="p">;</span>
|
||||
<span class="linenos">14</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="n">decay</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mf">5e-3f</span><span class="p">;</span>
|
||||
<span class="hll"><span class="linenos">15</span><span class="w"> </span><span class="n">hp_adamweightdecay_p</span><span class="p">(</span><span class="n">var</span><span class="p">,</span><span class="w"> </span><span class="n">m</span><span class="p">,</span><span class="w"> </span><span class="n">v</span><span class="p">,</span><span class="w"> </span><span class="n">gradient</span><span class="p">,</span><span class="w"> </span><span class="n">lr</span><span class="p">,</span>
|
||||
</span><span class="linenos">16</span><span class="w"> </span><span class="n">beta1</span><span class="p">,</span><span class="w"> </span><span class="n">beta2</span><span class="p">,</span><span class="w"> </span><span class="n">epsilon</span><span class="p">,</span><span class="w"> </span><span class="n">decay</span><span class="p">,</span>
|
||||
<span class="linenos">17</span><span class="w"> </span><span class="n">length</span><span class="p">);</span>
|
||||
<span class="linenos">18</span><span class="w"> </span><span class="k">return</span><span class="w"> </span><span class="mi">0</span><span class="p">;</span>
|
||||
<span class="linenos">19</span><span class="p">}</span>
|
||||
</pre></div>
|
||||
</div>
|
||||
</dd></dl>
|
||||
|
||||
</section>
|
||||
|
||||
|
||||
</div>
|
||||
</div>
|
||||
<footer><div class="rst-footer-buttons" role="navigation" aria-label="页脚">
|
||||
<a href="eltwise.html" class="btn btn-neutral float-left" title="Eltwise" accesskey="p" rel="prev"><span class="fa fa-arrow-circle-left" aria-hidden="true"></span> 上一页</a>
|
||||
<a href="applymomentum.html" class="btn btn-neutral float-right" title="ApplyMomentum" accesskey="n" rel="next">下一页 <span class="fa fa-arrow-circle-right" aria-hidden="true"></span></a>
|
||||
</div>
|
||||
|
||||
<hr/>
|
||||
|
||||
<div role="contentinfo">
|
||||
<p>© 版权所有 2025 - 2025, NUDT-674。</p>
|
||||
</div>
|
||||
|
||||
利用 <a href="https://www.sphinx-doc.org/">Sphinx</a> 构建，使用的
|
||||
<a href="https://github.com/readthedocs/sphinx_rtd_theme">主题</a>
|
||||
由 <a href="https://readthedocs.org">Read the Docs</a> 开发.
|
||||
|
||||
|
||||
</footer>
|
||||
</div>
|
||||
</div>
|
||||
</section>
|
||||
</div>
|
||||
<script>
|
||||
jQuery(function () {
|
||||
SphinxRtdTheme.Navigation.enable(true);
|
||||
});
|
||||
</script>
|
||||
|
||||
</body>
|
||||
</html>
|
||||
|
|
@ -0,0 +1,352 @@
|
|||
|
||||
|
||||
<!DOCTYPE html>
|
||||
<html class="writer-html5" lang="zh-CN" data-content_root="../../">
|
||||
<head>
|
||||
<meta charset="utf-8" /><meta name="viewport" content="width=device-width, initial-scale=1" />
|
||||
|
||||
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
|
||||
<title>Adder — MindSpore Signal+ 使用手册 alpha 文档</title>
|
||||
<link rel="stylesheet" type="text/css" href="../../_static/pygments.css?v=03e43079" />
|
||||
<link rel="stylesheet" type="text/css" href="../../_static/css/theme.css?v=e59714d7" />
|
||||
<link rel="stylesheet" type="text/css" href="../../_static/table.css?v=72b86d3e" />
|
||||
|
||||
|
||||
<script src="../../_static/jquery.js?v=5d32c60e"></script>
|
||||
<script src="../../_static/_sphinx_javascript_frameworks_compat.js?v=2cd50e6c"></script>
|
||||
<script src="../../_static/documentation_options.js?v=406e4f49"></script>
|
||||
<script src="../../_static/doctools.js?v=9bcbadda"></script>
|
||||
<script src="../../_static/sphinx_highlight.js?v=dc90522c"></script>
|
||||
<script src="../../_static/translations.js?v=beaddf03"></script>
|
||||
<script async="async" src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js"></script>
|
||||
<script src="../../_static/table.js?v=99ed15ba"></script>
|
||||
<script src="../../_static/js/theme.js"></script>
|
||||
<link rel="index" title="索引" href="../../genindex.html" />
|
||||
<link rel="search" title="搜索" href="../../search.html" />
|
||||
<link rel="next" title="Conv2dTranspose" href="conv2d_transpose.html" />
|
||||
<link rel="prev" title="Conv2d" href="conv2d.html" />
|
||||
</head>
|
||||
|
||||
<body class="wy-body-for-nav">
|
||||
<div class="wy-grid-for-nav">
|
||||
<nav data-toggle="wy-nav-shift" class="wy-nav-side">
|
||||
<div class="wy-side-scroll">
|
||||
<div class="wy-side-nav-search" >
|
||||
|
||||
|
||||
|
||||
<a href="../../index.html" class="icon icon-home">
|
||||
MindSpore Signal+ 使用手册
|
||||
</a>
|
||||
<div role="search">
|
||||
<form id="rtd-search-form" class="wy-form" action="../../search.html" method="get">
|
||||
<input type="text" name="q" placeholder="搜索文档" aria-label="搜索文档" />
|
||||
<input type="hidden" name="check_keywords" value="yes" />
|
||||
<input type="hidden" name="area" value="default" />
|
||||
</form>
|
||||
</div>
|
||||
</div><div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="导航菜单">
|
||||
<ul class="current">
|
||||
<li class="toctree-l1"><a class="reference internal" href="../../quickstart/index.html">快速入门</a></li>
|
||||
<li class="toctree-l1"><a class="reference internal" href="../../appdevelop/index.html">应用开发示例</a></li>
|
||||
<li class="toctree-l1 current"><a class="reference internal" href="../index.html">算子库支持</a><ul class="current">
|
||||
<li class="toctree-l2"><a class="reference internal" href="../supported_op.html">算子库支持情况</a></li>
|
||||
<li class="toctree-l2"><a class="reference internal" href="../custom_op/index.html">自定义算子列表</a></li>
|
||||
<li class="toctree-l2 current"><a class="reference internal" href="dsplib_index.html">DSP Library C API Reference</a><ul class="current">
|
||||
<li class="toctree-l3"><a class="reference internal" href="equal.html">Equal</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="activation.html">Activation</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="expfusion.html">ExpFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="reverse_sequence.html">ReverseSequence</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="reversev2.html">ReverseV2</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="fillv2.html">FillV2</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="squeeze.html">Squeeze</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="unsqueeze.html">UnSqueeze</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="expand_dims.html">ExpandDims</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="leaky_relu.html">LeakyReLu</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="lstm.html">LSTM</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="scatter_elements.html">ScatterElements</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="reduce.html">Reduce</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="resize.html">Resize</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="crop_and_resize.html">CropAndResize</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="crop.html">Crop</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="conv2d.html">Conv2d</a></li>
|
||||
<li class="toctree-l3 current"><a class="current reference internal" href="#">Adder</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="conv2d_transpose.html">Conv2dTranspose</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="gru.html">GRU</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="assert.html">Assert</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="range.html">Range</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="raggedrange.html">RaggedRange</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="linspace.html">LinSpace</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="matmulfusion.html">MatMulFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="attention.html">Attention</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="floor.html">Floor</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="floordiv.html">FloorDiv</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="embeddinglookup.html">EmbeddingLookup</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="eltwise.html">Eltwise</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="adamweightdecay.html">AdamWeightDecay</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="applymomentum.html">ApplyMomentum</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="avgpoolinggrad.html">AvgPoolingGrad</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="batchtospace.html">BatchToSpace</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="batchtospacend.html">BatchToSpaceND</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="broadcastto.html">BroadcastTo</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="depthtospace.html">DepthToSpace</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="spacetodepth.html">SpaceToDepth</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="spacetobatch.html">SpaceToBatch</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="spacetobatchnd.html">SpaceToBatchND</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="fusedbatchnorm.html">FusedBatchNorm</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="groupnormfusion.html">GroupNormFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="conv2dbackpropinputfusion.html">Conv2DBackpropInputFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="conv2dbackpropfilterfusion.html">Conv2DBackpropFilterFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="sgd.html">SGD</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="scalefusion.html">ScaleFusion</a></li>
|
||||
</ul>
|
||||
</li>
|
||||
</ul>
|
||||
</li>
|
||||
<li class="toctree-l1"><a class="reference internal" href="../../refdoc/index.html">参考资料</a></li>
|
||||
</ul>
|
||||
|
||||
</div>
|
||||
</div>
|
||||
</nav>
|
||||
|
||||
<section data-toggle="wy-nav-shift" class="wy-nav-content-wrap"><nav class="wy-nav-top" aria-label="移动版导航菜单" >
|
||||
<i data-toggle="wy-nav-top" class="fa fa-bars"></i>
|
||||
<a href="../../index.html">MindSpore Signal+ 使用手册</a>
|
||||
</nav>
|
||||
|
||||
<div class="wy-nav-content">
|
||||
<div class="rst-content">
|
||||
<div role="navigation" aria-label="页面导航">
|
||||
<ul class="wy-breadcrumbs">
|
||||
<li><a href="../../index.html" class="icon icon-home" aria-label="Home"></a></li>
|
||||
<li class="breadcrumb-item"><a href="../index.html">算子库支持</a></li>
|
||||
<li class="breadcrumb-item"><a href="dsplib_index.html">DSP Library C API Reference</a></li>
|
||||
<li class="breadcrumb-item active">Adder</li>
|
||||
<li class="wy-breadcrumbs-aside">
|
||||
<a href="../../_sources/functionlib/dsplib/adder.rst.txt" rel="nofollow"> 查看页面源码</a>
|
||||
</li>
|
||||
</ul>
|
||||
<hr/>
|
||||
</div>
|
||||
<div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
|
||||
<div itemprop="articleBody">
|
||||
|
||||
<section id="adder">
|
||||
<h1>Adder<a class="headerlink" href="#adder" title="Link to this heading"></a></h1>
|
||||
<p>Adder是一种卷积替代算子，它使用L1距离度量（绝对差的和）代替传统卷积中的点积操作。与标准卷积不同，Adder通过计算特征与卷积核之间的绝对差的负和来进行特征提取。假定输入X，filter表示为F，它按以下公式计算：</p>
|
||||
<div class="math notranslate nohighlight">
|
||||
\[Y(m,n,t) = - \sum_{i=0}^{d} \sum_{j=0}^{d} \sum_{k=0}^{C_{in}} |X(m+i, n+j, k) - F(i,j,k,t)|\]</div>
|
||||
<dl class="simple">
|
||||
<dt>输入：</dt><dd><ul class="simple">
|
||||
<li><p><strong>input_x</strong> - 输入数据的地址</p></li>
|
||||
<li><p><strong>input_w</strong> - 输入卷积核权重的地址</p></li>
|
||||
<li><p><strong>bias</strong> - 输入偏置的地址</p></li>
|
||||
<li><p><strong>param</strong> - 算子计算所需参数的结构体。其各成员见下述。</p></li>
|
||||
<li><p><strong>core_mask</strong> - 核掩码。</p></li>
|
||||
</ul>
|
||||
</dd>
|
||||
</dl>
|
||||
<p><strong>AdderParameter定义：</strong></p>
|
||||
<div class="highlight-c notranslate"><div class="highlight"><pre><span></span><span class="linenos"> 1</span><span class="k">typedef</span><span class="w"> </span><span class="k">struct</span><span class="w"> </span><span class="nc">AdderParameter</span><span class="w"> </span><span class="p">{</span>
|
||||
<span class="linenos"> 2</span><span class="w"> </span><span class="kt">void</span><span class="o">*</span><span class="w"> </span><span class="n">workspace_</span><span class="p">;</span><span class="w"> </span><span class="c1">// 用于存放中间计算结果</span>
|
||||
<span class="linenos"> 3</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">output_batch_</span><span class="p">;</span><span class="w"> </span><span class="c1">// 输出数据总批次</span>
|
||||
<span class="linenos"> 4</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">input_batch_</span><span class="p">;</span><span class="w"> </span><span class="c1">// 输入数据总批次</span>
|
||||
<span class="linenos"> 5</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">input_h_</span><span class="p">;</span><span class="w"> </span><span class="c1">// 输入数据h维度大小</span>
|
||||
<span class="linenos"> 6</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">input_w_</span><span class="p">;</span><span class="w"> </span><span class="c1">// 输入数据w维度大小</span>
|
||||
<span class="linenos"> 7</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">output_h_</span><span class="p">;</span><span class="w"> </span><span class="c1">// 输出数据h维度大小</span>
|
||||
<span class="linenos"> 8</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">output_w_</span><span class="p">;</span><span class="w"> </span><span class="c1">// 输出数据w维度大小</span>
|
||||
<span class="linenos"> 9</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">input_channel_</span><span class="p">;</span><span class="w"> </span><span class="c1">// 输入数据通道数</span>
|
||||
<span class="linenos">10</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">output_channel_</span><span class="p">;</span><span class="w"> </span><span class="c1">// 输出数据通道数</span>
|
||||
<span class="linenos">11</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">kernel_h_</span><span class="p">;</span><span class="w"> </span><span class="c1">// 卷积核h维度大小</span>
|
||||
<span class="linenos">12</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">kernel_w_</span><span class="p">;</span><span class="w"> </span><span class="c1">// 卷积核w维度大小</span>
|
||||
<span class="linenos">13</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">group_</span><span class="p">;</span><span class="w"> </span><span class="c1">// 组数</span>
|
||||
<span class="linenos">14</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">pad_l_</span><span class="p">;</span><span class="w"> </span><span class="c1">// 左填充大小</span>
|
||||
<span class="linenos">15</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">pad_u_</span><span class="p">;</span><span class="w"> </span><span class="c1">// 上填充大小</span>
|
||||
<span class="linenos">16</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">dilation_h_</span><span class="p">;</span><span class="w"> </span><span class="c1">// 卷积核h维度膨胀尺寸大小</span>
|
||||
<span class="linenos">17</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">dilation_w_</span><span class="p">;</span><span class="w"> </span><span class="c1">// 卷积核w维度膨胀尺寸大小</span>
|
||||
<span class="linenos">18</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">stride_h_</span><span class="p">;</span><span class="w"> </span><span class="c1">// 卷积核h维度步长</span>
|
||||
<span class="linenos">19</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">stride_w_</span><span class="p">;</span><span class="w"> </span><span class="c1">// 卷积核w维度步长</span>
|
||||
<span class="linenos">20</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">buffer_size_</span><span class="p">;</span><span class="w"> </span><span class="c1">// 为分块计算所分配的缓存大小</span>
|
||||
<span class="linenos">21</span><span class="p">}</span><span class="w"> </span><span class="n">AdderParameter</span><span class="p">;</span>
|
||||
</pre></div>
|
||||
</div>
|
||||
<dl class="simple">
|
||||
<dt>输出：</dt><dd><ul class="simple">
|
||||
<li><p><strong>out_y</strong> - 输出地址。</p></li>
|
||||
</ul>
|
||||
</dd>
|
||||
<dt>支持平台：</dt><dd><p><code class="docutils literal notranslate"><span class="pre">FT78NE</span></code>
|
||||
<code class="docutils literal notranslate"><span class="pre">MT7004</span></code></p>
|
||||
</dd>
|
||||
</dl>
|
||||
<div class="admonition note">
|
||||
<p class="admonition-title">备注</p>
|
||||
<ul class="simple">
|
||||
<li><p>FT78NE 支持int8, fp32</p></li>
|
||||
<li><p>MT7004 支持fp16, fp32</p></li>
|
||||
</ul>
|
||||
</div>
|
||||
<p><strong>共享存储版本:</strong></p>
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_adder_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_adder_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">input_x</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">input_w</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">out_y</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">bias</span></span>, <span class="n"><span class="pre">AdderParameter</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">param</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_adder_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_adder_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_adder_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">input_x</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">input_w</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">out_y</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">bias</span></span>, <span class="n"><span class="pre">AdderParameter</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">param</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_adder_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_adder_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_adder_s</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">input_x</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">input_w</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">out_y</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">bias</span></span>, <span class="n"><span class="pre">AdderParameter</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">param</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_adder_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<p><strong>C调用示例：</strong></p>
|
||||
<div class="highlight-c notranslate"><div class="highlight"><pre><span></span><span class="linenos"> 1</span><span class="kt">void</span><span class="w"> </span><span class="nf">TestAdderSMCFp32</span><span class="p">(</span><span class="kt">int</span><span class="o">*</span><span class="w"> </span><span class="n">input_shape</span><span class="p">,</span><span class="w"> </span><span class="kt">int</span><span class="o">*</span><span class="w"> </span><span class="n">weight_shape</span><span class="p">,</span><span class="w"> </span><span class="kt">int</span><span class="o">*</span><span class="w"> </span><span class="n">output_shape</span><span class="p">,</span><span class="w"> </span><span class="kt">int</span><span class="o">*</span><span class="w"> </span><span class="n">stride</span><span class="p">,</span><span class="w"> </span><span class="kt">int</span><span class="o">*</span><span class="w"> </span><span class="n">padding</span><span class="p">,</span><span class="w"> </span><span class="kt">int</span><span class="o">*</span><span class="w"> </span><span class="n">dilation</span><span class="p">,</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">groups</span><span class="p">,</span><span class="w"> </span><span class="kt">float</span><span class="o">*</span><span class="w"> </span><span class="n">bias</span><span class="p">,</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">core_mask</span><span class="p">)</span><span class="w"> </span><span class="p">{</span>
|
||||
<span class="linenos"> 2</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">core_id</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">get_core_id</span><span class="p">();</span>
|
||||
<span class="linenos"> 3</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">logic_core_id</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">GetLogicCoreId</span><span class="p">(</span><span class="n">core_mask</span><span class="p">,</span><span class="w"> </span><span class="n">core_id</span><span class="p">);</span>
|
||||
<span class="linenos"> 4</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">core_num</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">GetCoreNum</span><span class="p">(</span><span class="n">core_mask</span><span class="p">);</span>
|
||||
<span class="linenos"> 5</span><span class="w"> </span><span class="kt">float</span><span class="o">*</span><span class="w"> </span><span class="n">input_data</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="o">*</span><span class="p">)</span><span class="mh">0x88000000</span><span class="p">;</span>
|
||||
<span class="linenos"> 6</span><span class="w"> </span><span class="kt">float</span><span class="o">*</span><span class="w"> </span><span class="n">weight</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="o">*</span><span class="p">)</span><span class="mh">0x89000000</span><span class="p">;</span>
|
||||
<span class="linenos"> 7</span><span class="w"> </span><span class="kt">float</span><span class="o">*</span><span class="w"> </span><span class="n">output_data</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="o">*</span><span class="p">)</span><span class="mh">0x90000000</span><span class="p">;</span>
|
||||
<span class="linenos"> 8</span><span class="w"> </span><span class="kt">float</span><span class="o">*</span><span class="w"> </span><span class="n">bias_data</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="o">*</span><span class="p">)</span><span class="mh">0x91000000</span><span class="p">;</span>
|
||||
<span class="linenos"> 9</span><span class="w"> </span><span class="n">AdderParameter</span><span class="o">*</span><span class="w"> </span><span class="n">param</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="n">AdderParameter</span><span class="o">*</span><span class="p">)</span><span class="mh">0x92000000</span><span class="p">;</span>
|
||||
<span class="linenos">10</span><span class="w"> </span><span class="k">if</span><span class="w"> </span><span class="p">(</span><span class="n">logic_core_id</span><span class="w"> </span><span class="o">==</span><span class="w"> </span><span class="mi">0</span><span class="p">)</span><span class="w"> </span><span class="p">{</span>
|
||||
<span class="linenos">11</span><span class="w"> </span><span class="n">memcpy</span><span class="p">(</span><span class="n">bias_data</span><span class="p">,</span><span class="w"> </span><span class="n">bias</span><span class="p">,</span><span class="w"> </span><span class="k">sizeof</span><span class="p">(</span><span class="kt">float</span><span class="p">)</span><span class="w"> </span><span class="o">*</span><span class="w"> </span><span class="n">output_shape</span><span class="p">[</span><span class="mi">3</span><span class="p">]);</span>
|
||||
<span class="linenos">12</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">dilation_h_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">dilation</span><span class="p">[</span><span class="mi">0</span><span class="p">];</span>
|
||||
<span class="linenos">13</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">dilation_w_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">dilation</span><span class="p">[</span><span class="mi">1</span><span class="p">];</span>
|
||||
<span class="linenos">14</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">group_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">groups</span><span class="p">;</span>
|
||||
<span class="linenos">15</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">input_batch_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">input_shape</span><span class="p">[</span><span class="mi">0</span><span class="p">];</span>
|
||||
<span class="linenos">16</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">input_h_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">input_shape</span><span class="p">[</span><span class="mi">1</span><span class="p">];</span>
|
||||
<span class="linenos">17</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">input_w_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">input_shape</span><span class="p">[</span><span class="mi">2</span><span class="p">];</span>
|
||||
<span class="linenos">18</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">input_channel_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">input_shape</span><span class="p">[</span><span class="mi">3</span><span class="p">];</span>
|
||||
<span class="linenos">19</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">kernel_h_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">weight_shape</span><span class="p">[</span><span class="mi">1</span><span class="p">];</span>
|
||||
<span class="linenos">20</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">kernel_w_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">weight_shape</span><span class="p">[</span><span class="mi">2</span><span class="p">];</span>
|
||||
<span class="linenos">21</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">output_batch_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">output_shape</span><span class="p">[</span><span class="mi">0</span><span class="p">];</span>
|
||||
<span class="linenos">22</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">output_h_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">output_shape</span><span class="p">[</span><span class="mi">1</span><span class="p">];</span>
|
||||
<span class="linenos">23</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">output_w_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">output_shape</span><span class="p">[</span><span class="mi">2</span><span class="p">];</span>
|
||||
<span class="linenos">24</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">output_channel_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">output_shape</span><span class="p">[</span><span class="mi">3</span><span class="p">];</span>
|
||||
<span class="linenos">25</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">stride_h_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">stride</span><span class="p">[</span><span class="mi">0</span><span class="p">];</span>
|
||||
<span class="linenos">26</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">stride_w_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">stride</span><span class="p">[</span><span class="mi">0</span><span class="p">];</span>
|
||||
<span class="linenos">27</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">pad_u_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">padding</span><span class="p">[</span><span class="mi">0</span><span class="p">];</span>
|
||||
<span class="linenos">28</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">pad_l_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">padding</span><span class="p">[</span><span class="mi">2</span><span class="p">];</span>
|
||||
<span class="linenos">29</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">workspace_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="o">*</span><span class="p">)</span><span class="mh">0x10000000</span><span class="p">;</span><span class="w"> </span><span class="c1">// workspace空间需分配在AM内，计算过程中会将数据搬运到workspace空间内进行计算</span>
|
||||
<span class="linenos">30</span><span class="w"> </span><span class="p">}</span>
|
||||
<span class="linenos">31</span><span class="w"> </span><span class="n">sys_bar</span><span class="p">(</span><span class="mi">0</span><span class="p">,</span><span class="w"> </span><span class="n">core_num</span><span class="p">);</span><span class="w"> </span><span class="c1">// 初始化参数完成后进行同步</span>
|
||||
<span class="hll"><span class="linenos">32</span><span class="w"> </span><span class="n">fp_adder_s</span><span class="p">(</span><span class="n">input_data</span><span class="p">,</span><span class="w"> </span><span class="n">weight</span><span class="p">,</span><span class="w"> </span><span class="n">output_data</span><span class="p">,</span><span class="w"> </span><span class="n">bias_data</span><span class="p">,</span><span class="w"> </span><span class="n">param</span><span class="p">,</span><span class="w"> </span><span class="n">core_mask</span><span class="p">);</span>
|
||||
</span><span class="linenos">33</span><span class="p">}</span>
|
||||
<span class="linenos">34</span>
|
||||
<span class="linenos">35</span><span class="kt">void</span><span class="w"> </span><span class="nf">main</span><span class="p">(){</span>
|
||||
<span class="linenos">36</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">in_channel</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">4</span><span class="p">;</span>
|
||||
<span class="linenos">37</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">out_channel</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">4</span><span class="p">;</span>
|
||||
<span class="linenos">38</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">groups</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">4</span><span class="p">;</span>
|
||||
<span class="linenos">39</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">input_shape</span><span class="p">[</span><span class="mi">4</span><span class="p">]</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">{</span><span class="mi">1</span><span class="p">,</span><span class="w"> </span><span class="mi">30</span><span class="p">,</span><span class="w"> </span><span class="mi">30</span><span class="p">,</span><span class="w"> </span><span class="n">in_channel</span><span class="p">};</span><span class="w"> </span><span class="c1">// NHWC</span>
|
||||
<span class="linenos">40</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">weight_shape</span><span class="p">[</span><span class="mi">4</span><span class="p">]</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">{</span><span class="n">out_channel</span><span class="p">,</span><span class="w"> </span><span class="mi">3</span><span class="p">,</span><span class="w"> </span><span class="mi">3</span><span class="p">,</span><span class="w"> </span><span class="n">in_channel</span><span class="w"> </span><span class="o">/</span><span class="w"> </span><span class="n">groups</span><span class="p">};</span>
|
||||
<span class="linenos">41</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">output_shape</span><span class="p">[</span><span class="mi">4</span><span class="p">]</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">{</span><span class="mi">1</span><span class="p">,</span><span class="w"> </span><span class="mi">10</span><span class="p">,</span><span class="w"> </span><span class="mi">10</span><span class="p">,</span><span class="w"> </span><span class="n">out_channel</span><span class="p">};</span><span class="w"> </span><span class="c1">// NHWC</span>
|
||||
<span class="linenos">42</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">stride</span><span class="p">[</span><span class="mi">2</span><span class="p">]</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">{</span><span class="mi">2</span><span class="p">,</span><span class="w"> </span><span class="mi">2</span><span class="p">};</span>
|
||||
<span class="linenos">43</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">padding</span><span class="p">[</span><span class="mi">4</span><span class="p">]</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">{</span><span class="mi">1</span><span class="p">,</span><span class="w"> </span><span class="mi">1</span><span class="p">,</span><span class="w"> </span><span class="mi">1</span><span class="p">,</span><span class="w"> </span><span class="mi">1</span><span class="p">};</span>
|
||||
<span class="linenos">44</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">dilation</span><span class="p">[</span><span class="mi">2</span><span class="p">]</span><span class="o">=</span><span class="w"> </span><span class="p">{</span><span class="mi">2</span><span class="p">,</span><span class="w"> </span><span class="mi">2</span><span class="p">};</span>
|
||||
<span class="linenos">45</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="n">bias</span><span class="p">[</span><span class="mi">4</span><span class="p">]</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">{</span><span class="mi">0</span><span class="p">,</span><span class="w"> </span><span class="mi">0</span><span class="p">,</span><span class="w"> </span><span class="mi">0</span><span class="p">,</span><span class="w"> </span><span class="mi">0</span><span class="p">};</span>
|
||||
<span class="linenos">46</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">core_mask</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mb">0b1111</span><span class="p">;</span>
|
||||
<span class="linenos">47</span><span class="w"> </span><span class="n">TestAdderSMCFp32</span><span class="p">(</span><span class="n">input_shape</span><span class="p">,</span><span class="w"> </span><span class="n">weight_shape</span><span class="p">,</span><span class="w"> </span><span class="n">output_shape</span><span class="p">,</span><span class="w"> </span><span class="n">stride</span><span class="p">,</span><span class="w"> </span><span class="n">padding</span><span class="p">,</span><span class="w"> </span><span class="n">dilation</span><span class="p">,</span><span class="w"> </span><span class="n">groups</span><span class="p">,</span><span class="w"> </span><span class="n">bias</span><span class="p">,</span><span class="w"> </span><span class="n">core_mask</span><span class="p">);</span>
|
||||
<span class="linenos">48</span><span class="p">}</span>
|
||||
</pre></div>
|
||||
</div>
|
||||
<p><strong>私有存储版本:</strong></p>
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.i8_adder_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">i8_adder_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">input_x</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">input_w</span></span>, <span class="n"><span class="pre">int8_t</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">out_y</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">bias</span></span>, <span class="n"><span class="pre">ConvParameter</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">conv_param</span></span>, <span class="n"><span class="pre">ConvQuantParameter</span></span><span class="w"> </span><span class="n"><span class="pre">quant_param</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.i8_adder_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_adder_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_adder_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">input_x</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">input_w</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">out_y</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">bias</span></span>, <span class="n"><span class="pre">ConvParameter</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">conv_param</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_adder_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_adder_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_adder_p</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">input_x</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">input_w</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">out_y</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">bias</span></span>, <span class="n"><span class="pre">ConvParameter</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">conv_param</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_adder_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<p><strong>C调用示例：</strong></p>
|
||||
<div class="highlight-c notranslate"><div class="highlight"><pre><span></span><span class="linenos"> 1</span><span class="kt">void</span><span class="w"> </span><span class="nf">TestAdderL2Fp32</span><span class="p">(</span><span class="kt">int</span><span class="o">*</span><span class="w"> </span><span class="n">input_shape</span><span class="p">,</span><span class="w"> </span><span class="kt">int</span><span class="o">*</span><span class="w"> </span><span class="n">weight_shape</span><span class="p">,</span><span class="w"> </span><span class="kt">int</span><span class="o">*</span><span class="w"> </span><span class="n">output_shape</span><span class="p">,</span><span class="w"> </span><span class="kt">int</span><span class="o">*</span><span class="w"> </span><span class="n">stride</span><span class="p">,</span><span class="w"> </span><span class="kt">int</span><span class="o">*</span><span class="w"> </span><span class="n">padding</span><span class="p">,</span><span class="w"> </span><span class="kt">int</span><span class="o">*</span><span class="w"> </span><span class="n">dilation</span><span class="p">,</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">groups</span><span class="p">,</span><span class="w"> </span><span class="kt">float</span><span class="o">*</span><span class="w"> </span><span class="n">bias</span><span class="p">,</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">core_mask</span><span class="p">)</span><span class="w"> </span><span class="p">{</span>
|
||||
<span class="linenos"> 2</span><span class="w"> </span><span class="kt">float</span><span class="o">*</span><span class="w"> </span><span class="n">input_data</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="o">*</span><span class="p">)</span><span class="mh">0x10010000</span><span class="p">;</span><span class="w"> </span><span class="c1">// 私有存储版本地址设置在AM内</span>
|
||||
<span class="linenos"> 3</span><span class="w"> </span><span class="kt">float</span><span class="o">*</span><span class="w"> </span><span class="n">weight</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="o">*</span><span class="p">)</span><span class="mh">0x10020000</span><span class="p">;</span>
|
||||
<span class="linenos"> 4</span><span class="w"> </span><span class="kt">float</span><span class="o">*</span><span class="w"> </span><span class="n">output_data</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="o">*</span><span class="p">)</span><span class="mh">0x10030000</span><span class="p">;</span>
|
||||
<span class="linenos"> 5</span><span class="w"> </span><span class="kt">float</span><span class="o">*</span><span class="w"> </span><span class="n">bias_data</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="o">*</span><span class="p">)</span><span class="mh">0x10040000</span><span class="p">;</span>
|
||||
<span class="linenos"> 6</span><span class="w"> </span><span class="n">AdderParameter</span><span class="o">*</span><span class="w"> </span><span class="n">param</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="n">AdderParameter</span><span class="o">*</span><span class="p">)</span><span class="mh">0x10060000</span><span class="p">;</span>
|
||||
<span class="linenos"> 7</span><span class="w"> </span><span class="n">memcpy</span><span class="p">(</span><span class="n">bias_data</span><span class="p">,</span><span class="w"> </span><span class="n">bias</span><span class="p">,</span><span class="w"> </span><span class="k">sizeof</span><span class="p">(</span><span class="kt">float</span><span class="p">)</span><span class="w"> </span><span class="o">*</span><span class="w"> </span><span class="n">output_shape</span><span class="p">[</span><span class="mi">3</span><span class="p">]);</span>
|
||||
<span class="linenos"> 8</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">dilation_h_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">dilation</span><span class="p">[</span><span class="mi">0</span><span class="p">];</span>
|
||||
<span class="linenos"> 9</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">dilation_w_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">dilation</span><span class="p">[</span><span class="mi">1</span><span class="p">];</span>
|
||||
<span class="linenos">10</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">group_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">groups</span><span class="p">;</span>
|
||||
<span class="linenos">11</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">input_batch_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">input_shape</span><span class="p">[</span><span class="mi">0</span><span class="p">];</span>
|
||||
<span class="linenos">12</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">input_h_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">input_shape</span><span class="p">[</span><span class="mi">1</span><span class="p">];</span>
|
||||
<span class="linenos">13</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">input_w_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">input_shape</span><span class="p">[</span><span class="mi">2</span><span class="p">];</span>
|
||||
<span class="linenos">14</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">input_channel_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">input_shape</span><span class="p">[</span><span class="mi">3</span><span class="p">];</span>
|
||||
<span class="linenos">15</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">kernel_h_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">weight_shape</span><span class="p">[</span><span class="mi">1</span><span class="p">];</span>
|
||||
<span class="linenos">16</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">kernel_w_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">weight_shape</span><span class="p">[</span><span class="mi">2</span><span class="p">];</span>
|
||||
<span class="linenos">17</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">output_batch_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">output_shape</span><span class="p">[</span><span class="mi">0</span><span class="p">];</span>
|
||||
<span class="linenos">18</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">output_h_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">output_shape</span><span class="p">[</span><span class="mi">1</span><span class="p">];</span>
|
||||
<span class="linenos">19</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">output_w_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">output_shape</span><span class="p">[</span><span class="mi">2</span><span class="p">];</span>
|
||||
<span class="linenos">20</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">output_channel_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">output_shape</span><span class="p">[</span><span class="mi">3</span><span class="p">];</span>
|
||||
<span class="linenos">21</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">stride_h_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">stride</span><span class="p">[</span><span class="mi">0</span><span class="p">];</span>
|
||||
<span class="linenos">22</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">stride_w_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">stride</span><span class="p">[</span><span class="mi">0</span><span class="p">];</span>
|
||||
<span class="linenos">23</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">pad_u_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">padding</span><span class="p">[</span><span class="mi">0</span><span class="p">];</span>
|
||||
<span class="linenos">24</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">pad_l_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">padding</span><span class="p">[</span><span class="mi">2</span><span class="p">];</span>
|
||||
<span class="linenos">25</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">workspace_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="o">*</span><span class="p">)</span><span class="mh">0x10070000</span><span class="p">;</span>
|
||||
<span class="linenos">26</span><span class="w"> </span><span class="n">param</span><span class="o">-></span><span class="n">buffer_size_</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">2048</span><span class="p">;</span><span class="w"> </span><span class="c1">// 私有存储版本中，必须设置该参数，用于确定分块计算的大小</span>
|
||||
<span class="hll"><span class="linenos">27</span><span class="w"> </span><span class="n">fp_adder_p</span><span class="p">(</span><span class="n">input_data</span><span class="p">,</span><span class="w"> </span><span class="n">weight</span><span class="p">,</span><span class="w"> </span><span class="n">output_data</span><span class="p">,</span><span class="w"> </span><span class="n">bias_data</span><span class="p">,</span><span class="w"> </span><span class="n">param</span><span class="p">,</span><span class="w"> </span><span class="n">core_mask</span><span class="p">);</span>
|
||||
</span><span class="linenos">28</span><span class="p">}</span>
|
||||
<span class="linenos">29</span>
|
||||
<span class="linenos">30</span><span class="kt">void</span><span class="w"> </span><span class="nf">main</span><span class="p">(){</span>
|
||||
<span class="linenos">31</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">in_channel</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">4</span><span class="p">;</span>
|
||||
<span class="linenos">32</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">out_channel</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">4</span><span class="p">;</span>
|
||||
<span class="linenos">33</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">groups</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">4</span><span class="p">;</span>
|
||||
<span class="linenos">34</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">input_shape</span><span class="p">[</span><span class="mi">4</span><span class="p">]</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">{</span><span class="mi">1</span><span class="p">,</span><span class="w"> </span><span class="mi">30</span><span class="p">,</span><span class="w"> </span><span class="mi">30</span><span class="p">,</span><span class="w"> </span><span class="n">in_channel</span><span class="p">};</span><span class="w"> </span><span class="c1">// NHWC</span>
|
||||
<span class="linenos">35</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">weight_shape</span><span class="p">[</span><span class="mi">4</span><span class="p">]</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">{</span><span class="n">out_channel</span><span class="p">,</span><span class="w"> </span><span class="mi">3</span><span class="p">,</span><span class="w"> </span><span class="mi">3</span><span class="p">,</span><span class="w"> </span><span class="n">in_channel</span><span class="w"> </span><span class="o">/</span><span class="w"> </span><span class="n">groups</span><span class="p">};</span>
|
||||
<span class="linenos">36</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">output_shape</span><span class="p">[</span><span class="mi">4</span><span class="p">]</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">{</span><span class="mi">1</span><span class="p">,</span><span class="w"> </span><span class="mi">10</span><span class="p">,</span><span class="w"> </span><span class="mi">10</span><span class="p">,</span><span class="w"> </span><span class="n">out_channel</span><span class="p">};</span><span class="w"> </span><span class="c1">// NHWC</span>
|
||||
<span class="linenos">37</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">stride</span><span class="p">[</span><span class="mi">2</span><span class="p">]</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">{</span><span class="mi">2</span><span class="p">,</span><span class="w"> </span><span class="mi">2</span><span class="p">};</span>
|
||||
<span class="linenos">38</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">padding</span><span class="p">[</span><span class="mi">4</span><span class="p">]</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">{</span><span class="mi">1</span><span class="p">,</span><span class="w"> </span><span class="mi">1</span><span class="p">,</span><span class="w"> </span><span class="mi">1</span><span class="p">,</span><span class="w"> </span><span class="mi">1</span><span class="p">};</span>
|
||||
<span class="linenos">39</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">dilation</span><span class="p">[</span><span class="mi">2</span><span class="p">]</span><span class="o">=</span><span class="w"> </span><span class="p">{</span><span class="mi">2</span><span class="p">,</span><span class="w"> </span><span class="mi">2</span><span class="p">};</span>
|
||||
<span class="linenos">40</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="n">bias</span><span class="p">[</span><span class="mi">4</span><span class="p">]</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">{</span><span class="mi">0</span><span class="p">,</span><span class="w"> </span><span class="mi">0</span><span class="p">,</span><span class="w"> </span><span class="mi">0</span><span class="p">,</span><span class="w"> </span><span class="mi">0</span><span class="p">};</span>
|
||||
<span class="linenos">41</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">core_mask</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mb">0b0001</span><span class="p">;</span><span class="w"> </span><span class="c1">// 私有存储版本只能设置为一个核心启动</span>
|
||||
<span class="linenos">42</span><span class="w"> </span><span class="n">TestAdderL2Fp32</span><span class="p">(</span><span class="n">input_shape</span><span class="p">,</span><span class="w"> </span><span class="n">weight_shape</span><span class="p">,</span><span class="w"> </span><span class="n">output_shape</span><span class="p">,</span><span class="w"> </span><span class="n">stride</span><span class="p">,</span><span class="w"> </span><span class="n">padding</span><span class="p">,</span><span class="w"> </span><span class="n">dilation</span><span class="p">,</span><span class="w"> </span><span class="n">groups</span><span class="p">,</span><span class="w"> </span><span class="n">bias</span><span class="p">,</span><span class="w"> </span><span class="n">core_mask</span><span class="p">);</span>
|
||||
<span class="linenos">43</span><span class="p">}</span>
|
||||
</pre></div>
|
||||
</div>
|
||||
</section>
|
||||
|
||||
|
||||
</div>
|
||||
</div>
|
||||
<footer><div class="rst-footer-buttons" role="navigation" aria-label="页脚">
|
||||
<a href="conv2d.html" class="btn btn-neutral float-left" title="Conv2d" accesskey="p" rel="prev"><span class="fa fa-arrow-circle-left" aria-hidden="true"></span> 上一页</a>
|
||||
<a href="conv2d_transpose.html" class="btn btn-neutral float-right" title="Conv2dTranspose" accesskey="n" rel="next">下一页 <span class="fa fa-arrow-circle-right" aria-hidden="true"></span></a>
|
||||
</div>
|
||||
|
||||
<hr/>
|
||||
|
||||
<div role="contentinfo">
|
||||
<p>© 版权所有 2025 - 2025, NUDT-674。</p>
|
||||
</div>
|
||||
|
||||
利用 <a href="https://www.sphinx-doc.org/">Sphinx</a> 构建，使用的
|
||||
<a href="https://github.com/readthedocs/sphinx_rtd_theme">主题</a>
|
||||
由 <a href="https://readthedocs.org">Read the Docs</a> 开发.
|
||||
|
||||
|
||||
</footer>
|
||||
</div>
|
||||
</div>
|
||||
</section>
|
||||
</div>
|
||||
<script>
|
||||
jQuery(function () {
|
||||
SphinxRtdTheme.Navigation.enable(true);
|
||||
});
|
||||
</script>
|
||||
|
||||
</body>
|
||||
</html>
|
||||
|
|
@ -0,0 +1,275 @@
|
|||
|
||||
|
||||
<!DOCTYPE html>
|
||||
<html class="writer-html5" lang="zh-CN" data-content_root="../../">
|
||||
<head>
|
||||
<meta charset="utf-8" /><meta name="viewport" content="width=device-width, initial-scale=1" />
|
||||
|
||||
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
|
||||
<title>ApplyMomentum — MindSpore Signal+ 使用手册 alpha 文档</title>
|
||||
<link rel="stylesheet" type="text/css" href="../../_static/pygments.css?v=03e43079" />
|
||||
<link rel="stylesheet" type="text/css" href="../../_static/css/theme.css?v=e59714d7" />
|
||||
<link rel="stylesheet" type="text/css" href="../../_static/table.css?v=72b86d3e" />
|
||||
|
||||
|
||||
<script src="../../_static/jquery.js?v=5d32c60e"></script>
|
||||
<script src="../../_static/_sphinx_javascript_frameworks_compat.js?v=2cd50e6c"></script>
|
||||
<script src="../../_static/documentation_options.js?v=406e4f49"></script>
|
||||
<script src="../../_static/doctools.js?v=9bcbadda"></script>
|
||||
<script src="../../_static/sphinx_highlight.js?v=dc90522c"></script>
|
||||
<script src="../../_static/translations.js?v=beaddf03"></script>
|
||||
<script async="async" src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js"></script>
|
||||
<script src="../../_static/table.js?v=99ed15ba"></script>
|
||||
<script src="../../_static/js/theme.js"></script>
|
||||
<link rel="index" title="索引" href="../../genindex.html" />
|
||||
<link rel="search" title="搜索" href="../../search.html" />
|
||||
<link rel="next" title="AvgPoolingGrad" href="avgpoolinggrad.html" />
|
||||
<link rel="prev" title="AdamWeightDecay" href="adamweightdecay.html" />
|
||||
</head>
|
||||
|
||||
<body class="wy-body-for-nav">
|
||||
<div class="wy-grid-for-nav">
|
||||
<nav data-toggle="wy-nav-shift" class="wy-nav-side">
|
||||
<div class="wy-side-scroll">
|
||||
<div class="wy-side-nav-search" >
|
||||
|
||||
|
||||
|
||||
<a href="../../index.html" class="icon icon-home">
|
||||
MindSpore Signal+ 使用手册
|
||||
</a>
|
||||
<div role="search">
|
||||
<form id="rtd-search-form" class="wy-form" action="../../search.html" method="get">
|
||||
<input type="text" name="q" placeholder="搜索文档" aria-label="搜索文档" />
|
||||
<input type="hidden" name="check_keywords" value="yes" />
|
||||
<input type="hidden" name="area" value="default" />
|
||||
</form>
|
||||
</div>
|
||||
</div><div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="导航菜单">
|
||||
<ul class="current">
|
||||
<li class="toctree-l1"><a class="reference internal" href="../../quickstart/index.html">快速入门</a></li>
|
||||
<li class="toctree-l1"><a class="reference internal" href="../../appdevelop/index.html">应用开发示例</a></li>
|
||||
<li class="toctree-l1 current"><a class="reference internal" href="../index.html">算子库支持</a><ul class="current">
|
||||
<li class="toctree-l2"><a class="reference internal" href="../supported_op.html">算子库支持情况</a></li>
|
||||
<li class="toctree-l2"><a class="reference internal" href="../custom_op/index.html">自定义算子列表</a></li>
|
||||
<li class="toctree-l2 current"><a class="reference internal" href="dsplib_index.html">DSP Library C API Reference</a><ul class="current">
|
||||
<li class="toctree-l3"><a class="reference internal" href="equal.html">Equal</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="activation.html">Activation</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="expfusion.html">ExpFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="reverse_sequence.html">ReverseSequence</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="reversev2.html">ReverseV2</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="fillv2.html">FillV2</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="squeeze.html">Squeeze</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="unsqueeze.html">UnSqueeze</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="expand_dims.html">ExpandDims</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="leaky_relu.html">LeakyReLu</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="lstm.html">LSTM</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="scatter_elements.html">ScatterElements</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="reduce.html">Reduce</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="resize.html">Resize</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="crop_and_resize.html">CropAndResize</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="crop.html">Crop</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="conv2d.html">Conv2d</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="adder.html">Adder</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="conv2d_transpose.html">Conv2dTranspose</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="gru.html">GRU</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="assert.html">Assert</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="range.html">Range</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="raggedrange.html">RaggedRange</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="linspace.html">LinSpace</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="matmulfusion.html">MatMulFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="attention.html">Attention</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="floor.html">Floor</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="floordiv.html">FloorDiv</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="embeddinglookup.html">EmbeddingLookup</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="eltwise.html">Eltwise</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="adamweightdecay.html">AdamWeightDecay</a></li>
|
||||
<li class="toctree-l3 current"><a class="current reference internal" href="#">ApplyMomentum</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="avgpoolinggrad.html">AvgPoolingGrad</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="batchtospace.html">BatchToSpace</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="batchtospacend.html">BatchToSpaceND</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="broadcastto.html">BroadcastTo</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="depthtospace.html">DepthToSpace</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="spacetodepth.html">SpaceToDepth</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="spacetobatch.html">SpaceToBatch</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="spacetobatchnd.html">SpaceToBatchND</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="fusedbatchnorm.html">FusedBatchNorm</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="groupnormfusion.html">GroupNormFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="conv2dbackpropinputfusion.html">Conv2DBackpropInputFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="conv2dbackpropfilterfusion.html">Conv2DBackpropFilterFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="sgd.html">SGD</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="scalefusion.html">ScaleFusion</a></li>
|
||||
</ul>
|
||||
</li>
|
||||
</ul>
|
||||
</li>
|
||||
<li class="toctree-l1"><a class="reference internal" href="../../refdoc/index.html">参考资料</a></li>
|
||||
</ul>
|
||||
|
||||
</div>
|
||||
</div>
|
||||
</nav>
|
||||
|
||||
<section data-toggle="wy-nav-shift" class="wy-nav-content-wrap"><nav class="wy-nav-top" aria-label="移动版导航菜单" >
|
||||
<i data-toggle="wy-nav-top" class="fa fa-bars"></i>
|
||||
<a href="../../index.html">MindSpore Signal+ 使用手册</a>
|
||||
</nav>
|
||||
|
||||
<div class="wy-nav-content">
|
||||
<div class="rst-content">
|
||||
<div role="navigation" aria-label="页面导航">
|
||||
<ul class="wy-breadcrumbs">
|
||||
<li><a href="../../index.html" class="icon icon-home" aria-label="Home"></a></li>
|
||||
<li class="breadcrumb-item"><a href="../index.html">算子库支持</a></li>
|
||||
<li class="breadcrumb-item"><a href="dsplib_index.html">DSP Library C API Reference</a></li>
|
||||
<li class="breadcrumb-item active">ApplyMomentum</li>
|
||||
<li class="wy-breadcrumbs-aside">
|
||||
<a href="../../_sources/functionlib/dsplib/applymomentum.rst.txt" rel="nofollow"> 查看页面源码</a>
|
||||
</li>
|
||||
</ul>
|
||||
<hr/>
|
||||
</div>
|
||||
<div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
|
||||
<div itemprop="articleBody">
|
||||
|
||||
<section id="applymomentum">
|
||||
<h1>ApplyMomentum<a class="headerlink" href="#applymomentum" title="Link to this heading"></a></h1>
|
||||
<p>对权重张量执行 Momentum/改进动量优化更新。</p>
|
||||
<blockquote>
|
||||
<div><div class="math notranslate nohighlight">
|
||||
\[\begin{split}\begin{aligned}
|
||||
accu_t &= moment \cdot accu_{t-1} + g_t \\
|
||||
update_t &= \begin{cases}
|
||||
(accu_t \cdot moment + g_t), & \text{if nesterov = True} \\
|
||||
accu_t, & \text{otherwise}
|
||||
\end{cases} \\
|
||||
weight_t &= weight_{t-1} - learning\_rate \cdot update_t
|
||||
\end{aligned}\end{split}\]</div>
|
||||
<dl class="simple">
|
||||
<dt>输入：</dt><dd><ul class="simple">
|
||||
<li><p><strong>weight</strong> - 待更新权重张量首地址。</p></li>
|
||||
<li><p><strong>accumulate</strong> - 动量累积张量首地址。</p></li>
|
||||
<li><p><strong>gradient</strong> - 梯度张量首地址。</p></li>
|
||||
<li><p><strong>learning_rate</strong> - 学习率。</p></li>
|
||||
<li><p><strong>moment</strong> - 动量系数。</p></li>
|
||||
<li><p><strong>nesterov</strong> - 是否启用 Nesterov 动量。</p></li>
|
||||
<li><p><strong>start</strong> - 参与计算的起始索引（闭区间）。</p></li>
|
||||
<li><p><strong>end</strong> - 参与计算的结束索引（开区间）。</p></li>
|
||||
<li><p><strong>core_mask(int, 可选)</strong> - 核掩码（仅适用于共享存储版本）。</p></li>
|
||||
</ul>
|
||||
</dd>
|
||||
<dt>输出：</dt><dd><ul class="simple">
|
||||
<li><p><strong>weight</strong> - 原地写回更新后的权重张量。</p></li>
|
||||
<li><p><strong>accumulate</strong> - 原地写回更新后的动量张量。</p></li>
|
||||
</ul>
|
||||
</dd>
|
||||
<dt>支持平台：</dt><dd><p><code class="docutils literal notranslate"><span class="pre">FT78NE</span></code>
|
||||
<code class="docutils literal notranslate"><span class="pre">MT7004</span></code></p>
|
||||
</dd>
|
||||
</dl>
|
||||
<div class="admonition note">
|
||||
<p class="admonition-title">备注</p>
|
||||
<ul class="simple">
|
||||
<li><p>FT78NE 支持 fp32 数据类型。</p></li>
|
||||
<li><p>MT7004 支持 fp16、fp32 数据类型。</p></li>
|
||||
</ul>
|
||||
</div>
|
||||
</div></blockquote>
|
||||
<p><strong>共享存储版本:</strong></p>
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_applymomentum_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_applymomentum_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">weight</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">accumulate</span></span>, <span class="k"><span class="pre">const</span></span><span class="w"> </span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">gradient</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">learning_rate</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">moment</span></span>, <span class="kt"><span class="pre">bool</span></span><span class="w"> </span><span class="n"><span class="pre">nesterov</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">start</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">end</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_applymomentum_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_applymomentum_s">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_applymomentum_s</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">weight</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">accumulate</span></span>, <span class="k"><span class="pre">const</span></span><span class="w"> </span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">gradient</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">learning_rate</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">moment</span></span>, <span class="kt"><span class="pre">bool</span></span><span class="w"> </span><span class="n"><span class="pre">nesterov</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">start</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">end</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_applymomentum_s" title="Link to this definition"></a><br /></dt>
|
||||
<dd><p><strong>C调用示例：</strong></p>
|
||||
<div class="highlight-c notranslate"><div class="highlight"><pre><span></span><span class="linenos"> 1</span><span class="c1">// FT78NE 多核示例</span>
|
||||
<span class="linenos"> 2</span><span class="cp">#include</span><span class="w"> </span><span class="cpf"><stdio.h></span>
|
||||
<span class="linenos"> 3</span><span class="cp">#include</span><span class="w"> </span><span class="cpf"><stdbool.h></span>
|
||||
<span class="linenos"> 4</span>
|
||||
<span class="linenos"> 5</span><span class="kt">int</span><span class="w"> </span><span class="nf">main</span><span class="p">(</span><span class="kt">void</span><span class="p">)</span><span class="w"> </span><span class="p">{</span>
|
||||
<span class="linenos"> 6</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="n">weight</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0xA0000000</span><span class="p">;</span><span class="w"> </span><span class="c1">// DDR 存储</span>
|
||||
<span class="linenos"> 7</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="n">accumulate</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0xB0000000</span><span class="p">;</span>
|
||||
<span class="linenos"> 8</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="n">gradient</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0xC0000000</span><span class="p">;</span>
|
||||
<span class="linenos"> 9</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">start</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">0</span><span class="p">;</span>
|
||||
<span class="linenos">10</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">end</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">4096</span><span class="p">;</span>
|
||||
<span class="linenos">11</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">core_mask</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mh">0xff</span><span class="p">;</span>
|
||||
<span class="linenos">12</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="n">learning_rate</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mf">1e-2f</span><span class="p">;</span>
|
||||
<span class="linenos">13</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="n">moment</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mf">0.99f</span><span class="p">;</span>
|
||||
<span class="linenos">14</span><span class="w"> </span><span class="kt">bool</span><span class="w"> </span><span class="n">nesterov</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="nb">false</span><span class="p">;</span>
|
||||
<span class="hll"><span class="linenos">15</span><span class="w"> </span><span class="n">fp_applymomentum_s</span><span class="p">(</span><span class="n">weight</span><span class="p">,</span><span class="w"> </span><span class="n">accumulate</span><span class="p">,</span><span class="w"> </span><span class="n">gradient</span><span class="p">,</span>
|
||||
</span><span class="linenos">16</span><span class="w"> </span><span class="n">learning_rate</span><span class="p">,</span><span class="w"> </span><span class="n">moment</span><span class="p">,</span><span class="w"> </span><span class="n">nesterov</span><span class="p">,</span>
|
||||
<span class="linenos">17</span><span class="w"> </span><span class="n">start</span><span class="p">,</span><span class="w"> </span><span class="n">end</span><span class="p">,</span><span class="w"> </span><span class="n">core_mask</span><span class="p">);</span>
|
||||
<span class="linenos">18</span><span class="w"> </span><span class="k">return</span><span class="w"> </span><span class="mi">0</span><span class="p">;</span>
|
||||
<span class="linenos">19</span><span class="p">}</span>
|
||||
</pre></div>
|
||||
</div>
|
||||
</dd></dl>
|
||||
|
||||
<p><strong>私有存储版本:</strong></p>
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.hp_applymomentum_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_applymomentum_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">weight</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">accumulate</span></span>, <span class="k"><span class="pre">const</span></span><span class="w"> </span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">gradient</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">learning_rate</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">moment</span></span>, <span class="kt"><span class="pre">bool</span></span><span class="w"> </span><span class="n"><span class="pre">nesterov</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_applymomentum_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.fp_applymomentum_p">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_applymomentum_p</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">weight</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">accumulate</span></span>, <span class="k"><span class="pre">const</span></span><span class="w"> </span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">gradient</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">learning_rate</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="n"><span class="pre">moment</span></span>, <span class="kt"><span class="pre">bool</span></span><span class="w"> </span><span class="n"><span class="pre">nesterov</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">length</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_applymomentum_p" title="Link to this definition"></a><br /></dt>
|
||||
<dd><p><strong>C调用示例：</strong></p>
|
||||
<div class="highlight-c notranslate"><div class="highlight"><pre><span></span><span class="linenos"> 1</span><span class="c1">// MT7004 单核示例</span>
|
||||
<span class="linenos"> 2</span><span class="cp">#include</span><span class="w"> </span><span class="cpf"><stdio.h></span>
|
||||
<span class="linenos"> 3</span><span class="cp">#include</span><span class="w"> </span><span class="cpf"><stdbool.h></span>
|
||||
<span class="linenos"> 4</span>
|
||||
<span class="linenos"> 5</span><span class="kt">int</span><span class="w"> </span><span class="nf">main</span><span class="p">(</span><span class="kt">void</span><span class="p">)</span><span class="w"> </span><span class="p">{</span>
|
||||
<span class="linenos"> 6</span><span class="w"> </span><span class="n">half</span><span class="w"> </span><span class="o">*</span><span class="n">weight</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="n">half</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0x10000000</span><span class="p">;</span><span class="w"> </span><span class="c1">// L2 存储</span>
|
||||
<span class="linenos"> 7</span><span class="w"> </span><span class="n">half</span><span class="w"> </span><span class="o">*</span><span class="n">accumulate</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="n">half</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0x10002000</span><span class="p">;</span>
|
||||
<span class="linenos"> 8</span><span class="w"> </span><span class="n">half</span><span class="w"> </span><span class="o">*</span><span class="n">gradient</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="n">half</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0x10004000</span><span class="p">;</span>
|
||||
<span class="linenos"> 9</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">length</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">2048</span><span class="p">;</span>
|
||||
<span class="linenos">10</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="n">learning_rate</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mf">5e-3f</span><span class="p">;</span>
|
||||
<span class="linenos">11</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="n">moment</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mf">0.9f</span><span class="p">;</span>
|
||||
<span class="linenos">12</span><span class="w"> </span><span class="kt">bool</span><span class="w"> </span><span class="n">nesterov</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="nb">true</span><span class="p">;</span>
|
||||
<span class="hll"><span class="linenos">13</span><span class="w"> </span><span class="n">hp_applymomentum_p</span><span class="p">(</span><span class="n">weight</span><span class="p">,</span><span class="w"> </span><span class="n">accumulate</span><span class="p">,</span><span class="w"> </span><span class="n">gradient</span><span class="p">,</span>
|
||||
</span><span class="linenos">14</span><span class="w"> </span><span class="n">learning_rate</span><span class="p">,</span><span class="w"> </span><span class="n">moment</span><span class="p">,</span><span class="w"> </span><span class="n">nesterov</span><span class="p">,</span>
|
||||
<span class="linenos">15</span><span class="w"> </span><span class="n">length</span><span class="p">);</span>
|
||||
<span class="linenos">16</span><span class="w"> </span><span class="k">return</span><span class="w"> </span><span class="mi">0</span><span class="p">;</span>
|
||||
<span class="linenos">17</span><span class="p">}</span>
|
||||
</pre></div>
|
||||
</div>
|
||||
</dd></dl>
|
||||
|
||||
</section>
|
||||
|
||||
|
||||
</div>
|
||||
</div>
|
||||
<footer><div class="rst-footer-buttons" role="navigation" aria-label="页脚">
|
||||
<a href="adamweightdecay.html" class="btn btn-neutral float-left" title="AdamWeightDecay" accesskey="p" rel="prev"><span class="fa fa-arrow-circle-left" aria-hidden="true"></span> 上一页</a>
|
||||
<a href="avgpoolinggrad.html" class="btn btn-neutral float-right" title="AvgPoolingGrad" accesskey="n" rel="next">下一页 <span class="fa fa-arrow-circle-right" aria-hidden="true"></span></a>
|
||||
</div>
|
||||
|
||||
<hr/>
|
||||
|
||||
<div role="contentinfo">
|
||||
<p>© 版权所有 2025 - 2025, NUDT-674。</p>
|
||||
</div>
|
||||
|
||||
利用 <a href="https://www.sphinx-doc.org/">Sphinx</a> 构建，使用的
|
||||
<a href="https://github.com/readthedocs/sphinx_rtd_theme">主题</a>
|
||||
由 <a href="https://readthedocs.org">Read the Docs</a> 开发.
|
||||
|
||||
|
||||
</footer>
|
||||
</div>
|
||||
</div>
|
||||
</section>
|
||||
</div>
|
||||
<script>
|
||||
jQuery(function () {
|
||||
SphinxRtdTheme.Navigation.enable(true);
|
||||
});
|
||||
</script>
|
||||
|
||||
</body>
|
||||
</html>
|
||||
|
|
@ -0,0 +1,213 @@
|
|||
|
||||
|
||||
<!DOCTYPE html>
|
||||
<html class="writer-html5" lang="zh-CN" data-content_root="../../">
|
||||
<head>
|
||||
<meta charset="utf-8" /><meta name="viewport" content="width=device-width, initial-scale=1" />
|
||||
|
||||
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
|
||||
<title>Assert — MindSpore Signal+ 使用手册 alpha 文档</title>
|
||||
<link rel="stylesheet" type="text/css" href="../../_static/pygments.css?v=03e43079" />
|
||||
<link rel="stylesheet" type="text/css" href="../../_static/css/theme.css?v=e59714d7" />
|
||||
<link rel="stylesheet" type="text/css" href="../../_static/table.css?v=72b86d3e" />
|
||||
|
||||
|
||||
<script src="../../_static/jquery.js?v=5d32c60e"></script>
|
||||
<script src="../../_static/_sphinx_javascript_frameworks_compat.js?v=2cd50e6c"></script>
|
||||
<script src="../../_static/documentation_options.js?v=406e4f49"></script>
|
||||
<script src="../../_static/doctools.js?v=9bcbadda"></script>
|
||||
<script src="../../_static/sphinx_highlight.js?v=dc90522c"></script>
|
||||
<script src="../../_static/translations.js?v=beaddf03"></script>
|
||||
<script async="async" src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js"></script>
|
||||
<script src="../../_static/table.js?v=99ed15ba"></script>
|
||||
<script src="../../_static/js/theme.js"></script>
|
||||
<link rel="index" title="索引" href="../../genindex.html" />
|
||||
<link rel="search" title="搜索" href="../../search.html" />
|
||||
<link rel="next" title="Range" href="range.html" />
|
||||
<link rel="prev" title="GRU" href="gru.html" />
|
||||
</head>
|
||||
|
||||
<body class="wy-body-for-nav">
|
||||
<div class="wy-grid-for-nav">
|
||||
<nav data-toggle="wy-nav-shift" class="wy-nav-side">
|
||||
<div class="wy-side-scroll">
|
||||
<div class="wy-side-nav-search" >
|
||||
|
||||
|
||||
|
||||
<a href="../../index.html" class="icon icon-home">
|
||||
MindSpore Signal+ 使用手册
|
||||
</a>
|
||||
<div role="search">
|
||||
<form id="rtd-search-form" class="wy-form" action="../../search.html" method="get">
|
||||
<input type="text" name="q" placeholder="搜索文档" aria-label="搜索文档" />
|
||||
<input type="hidden" name="check_keywords" value="yes" />
|
||||
<input type="hidden" name="area" value="default" />
|
||||
</form>
|
||||
</div>
|
||||
</div><div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="导航菜单">
|
||||
<ul class="current">
|
||||
<li class="toctree-l1"><a class="reference internal" href="../../quickstart/index.html">快速入门</a></li>
|
||||
<li class="toctree-l1"><a class="reference internal" href="../../appdevelop/index.html">应用开发示例</a></li>
|
||||
<li class="toctree-l1 current"><a class="reference internal" href="../index.html">算子库支持</a><ul class="current">
|
||||
<li class="toctree-l2"><a class="reference internal" href="../supported_op.html">算子库支持情况</a></li>
|
||||
<li class="toctree-l2"><a class="reference internal" href="../custom_op/index.html">自定义算子列表</a></li>
|
||||
<li class="toctree-l2 current"><a class="reference internal" href="dsplib_index.html">DSP Library C API Reference</a><ul class="current">
|
||||
<li class="toctree-l3"><a class="reference internal" href="equal.html">Equal</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="activation.html">Activation</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="expfusion.html">ExpFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="reverse_sequence.html">ReverseSequence</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="reversev2.html">ReverseV2</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="fillv2.html">FillV2</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="squeeze.html">Squeeze</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="unsqueeze.html">UnSqueeze</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="expand_dims.html">ExpandDims</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="leaky_relu.html">LeakyReLu</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="lstm.html">LSTM</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="scatter_elements.html">ScatterElements</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="reduce.html">Reduce</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="resize.html">Resize</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="crop_and_resize.html">CropAndResize</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="crop.html">Crop</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="conv2d.html">Conv2d</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="adder.html">Adder</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="conv2d_transpose.html">Conv2dTranspose</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="gru.html">GRU</a></li>
|
||||
<li class="toctree-l3 current"><a class="current reference internal" href="#">Assert</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="range.html">Range</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="raggedrange.html">RaggedRange</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="linspace.html">LinSpace</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="matmulfusion.html">MatMulFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="attention.html">Attention</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="floor.html">Floor</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="floordiv.html">FloorDiv</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="embeddinglookup.html">EmbeddingLookup</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="eltwise.html">Eltwise</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="adamweightdecay.html">AdamWeightDecay</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="applymomentum.html">ApplyMomentum</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="avgpoolinggrad.html">AvgPoolingGrad</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="batchtospace.html">BatchToSpace</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="batchtospacend.html">BatchToSpaceND</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="broadcastto.html">BroadcastTo</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="depthtospace.html">DepthToSpace</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="spacetodepth.html">SpaceToDepth</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="spacetobatch.html">SpaceToBatch</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="spacetobatchnd.html">SpaceToBatchND</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="fusedbatchnorm.html">FusedBatchNorm</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="groupnormfusion.html">GroupNormFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="conv2dbackpropinputfusion.html">Conv2DBackpropInputFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="conv2dbackpropfilterfusion.html">Conv2DBackpropFilterFusion</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="sgd.html">SGD</a></li>
|
||||
<li class="toctree-l3"><a class="reference internal" href="scalefusion.html">ScaleFusion</a></li>
|
||||
</ul>
|
||||
</li>
|
||||
</ul>
|
||||
</li>
|
||||
<li class="toctree-l1"><a class="reference internal" href="../../refdoc/index.html">参考资料</a></li>
|
||||
</ul>
|
||||
|
||||
</div>
|
||||
</div>
|
||||
</nav>
|
||||
|
||||
<section data-toggle="wy-nav-shift" class="wy-nav-content-wrap"><nav class="wy-nav-top" aria-label="移动版导航菜单" >
|
||||
<i data-toggle="wy-nav-top" class="fa fa-bars"></i>
|
||||
<a href="../../index.html">MindSpore Signal+ 使用手册</a>
|
||||
</nav>
|
||||
|
||||
<div class="wy-nav-content">
|
||||
<div class="rst-content">
|
||||
<div role="navigation" aria-label="页面导航">
|
||||
<ul class="wy-breadcrumbs">
|
||||
<li><a href="../../index.html" class="icon icon-home" aria-label="Home"></a></li>
|
||||
<li class="breadcrumb-item"><a href="../index.html">算子库支持</a></li>
|
||||
<li class="breadcrumb-item"><a href="dsplib_index.html">DSP Library C API Reference</a></li>
|
||||
<li class="breadcrumb-item active">Assert</li>
|
||||
<li class="wy-breadcrumbs-aside">
|
||||
<a href="../../_sources/functionlib/dsplib/assert.rst.txt" rel="nofollow"> 查看页面源码</a>
|
||||
</li>
|
||||
</ul>
|
||||
<hr/>
|
||||
</div>
|
||||
<div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
|
||||
<div itemprop="articleBody">
|
||||
|
||||
<section id="assert">
|
||||
<h1>Assert<a class="headerlink" href="#assert" title="Link to this heading"></a></h1>
|
||||
<dl class="c function">
|
||||
<dt class="sig sig-object c" id="c.assert">
|
||||
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">assert</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">bool</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">Input</span></span>, <span class="kt"><span class="pre">bool</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">output</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.assert" title="Link to this definition"></a><br /></dt>
|
||||
<dd></dd></dl>
|
||||
|
||||
<p>判断输入是否为 True。</p>
|
||||
<div class="math notranslate nohighlight">
|
||||
\[\begin{split}output_i = \begin{cases}
|
||||
\text{True}, & \text{if } Input_i = \text{True} \\
|
||||
\text{False}, & \text{if } Input_i = \text{False}
|
||||
\end{cases}\end{split}\]</div>
|
||||
<dl class="simple">
|
||||
<dt>输入：</dt><dd><ul class="simple">
|
||||
<li><p><strong>Input</strong> - 输入数据地址（布尔类型）。</p></li>
|
||||
</ul>
|
||||
</dd>
|
||||
<dt>输出：</dt><dd><ul class="simple">
|
||||
<li><p><strong>output</strong> - 计算结果地址（布尔类型）。</p></li>
|
||||
</ul>
|
||||
</dd>
|
||||
<dt>支持平台：</dt><dd><p><code class="docutils literal notranslate"><span class="pre">FT78NE</span></code>
|
||||
<code class="docutils literal notranslate"><span class="pre">MT7004</span></code></p>
|
||||
</dd>
|
||||
</dl>
|
||||
<div class="admonition note">
|
||||
<p class="admonition-title">备注</p>
|
||||
<ul class="simple">
|
||||
<li><p>本算子只有一个版本</p></li>
|
||||
</ul>
|
||||
</div>
|
||||
<p><strong>C调用示例：</strong></p>
|
||||
<div class="highlight-c notranslate"><div class="highlight"><pre><span></span><span class="linenos"> 1</span><span class="c1">// FT78NE/MT7004 示例（共享存储）</span>
|
||||
<span class="linenos"> 2</span><span class="cp">#include</span><span class="w"> </span><span class="cpf"><stdio.h></span>
|
||||
<span class="linenos"> 3</span><span class="cp">#include</span><span class="w"> </span><span class="cpf"><stdbool.h></span>
|
||||
<span class="linenos"> 4</span>
|
||||
<span class="linenos"> 5</span><span class="kt">int</span><span class="w"> </span><span class="nf">main</span><span class="p">(</span><span class="kt">int</span><span class="w"> </span><span class="n">argc</span><span class="p">,</span><span class="w"> </span><span class="kt">char</span><span class="o">*</span><span class="w"> </span><span class="n">argv</span><span class="p">[])</span><span class="w"> </span><span class="p">{</span>
|
||||
<span class="linenos"> 6</span><span class="w"> </span><span class="kt">bool</span><span class="w"> </span><span class="o">*</span><span class="n">input</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">bool</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0xA0000000</span><span class="p">;</span>
|
||||
<span class="linenos"> 7</span><span class="w"> </span><span class="kt">bool</span><span class="w"> </span><span class="o">*</span><span class="n">output</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">bool</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0xC0000000</span><span class="p">;</span>
|
||||
<span class="linenos"> 8</span><span class="w"> </span><span class="n">assert_s</span><span class="p">(</span><span class="n">input</span><span class="p">,</span><span class="w"> </span><span class="n">output</span><span class="p">);</span>
|
||||
<span class="linenos"> 9</span><span class="w"> </span><span class="k">return</span><span class="w"> </span><span class="mi">0</span><span class="p">;</span>
|
||||
<span class="hll"><span class="linenos">10</span><span class="p">}</span>
|
||||
</span></pre></div>
|
||||
</div>
|
||||
</section>
|
||||
|
||||
|
||||
</div>
|
||||
</div>
|
||||
<footer><div class="rst-footer-buttons" role="navigation" aria-label="页脚">
|
||||
<a href="gru.html" class="btn btn-neutral float-left" title="GRU" accesskey="p" rel="prev"><span class="fa fa-arrow-circle-left" aria-hidden="true"></span> 上一页</a>
|
||||
<a href="range.html" class="btn btn-neutral float-right" title="Range" accesskey="n" rel="next">下一页 <span class="fa fa-arrow-circle-right" aria-hidden="true"></span></a>
|
||||
</div>
|
||||
|
||||
<hr/>
|
||||
|
||||
<div role="contentinfo">
|
||||
<p>© 版权所有 2025 - 2025, NUDT-674。</p>
|
||||
</div>
|
||||
|
||||
利用 <a href="https://www.sphinx-doc.org/">Sphinx</a> 构建，使用的
|
||||
<a href="https://github.com/readthedocs/sphinx_rtd_theme">主题</a>
|
||||
由 <a href="https://readthedocs.org">Read the Docs</a> 开发.
|
||||
|
||||
|
||||
</footer>
|
||||
</div>
|
||||
</div>
|
||||
</section>
|
||||
</div>
|
||||
<script>
|
||||
jQuery(function () {
|
||||
SphinxRtdTheme.Navigation.enable(true);
|
||||
});
|
||||
</script>
|
||||
|
||||
</body>
|
||||
</html>
|
||||
Some files were not shown because too many files have changed in this diff Show More
Loading…
Reference in New Issue