natas_hw
/
openvino
0
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Pre.2021.1.submission (#2094) 

* fixed code and updated unit tests to accomodate auto-reshaping graphs, to unlock full validation

* [CPU][BF16] bf16 for Gemm or MatMul was enabled (#1920)

# Conflicts:
#	inference-engine/thirdparty/mkl-dnn

* Fuse EmbeddingBag

* [IE CLDNN] Fix result storing in leftover's branch (#2050)

Co-authored-by: Alexey Varyzgin <alexey.varyzgin@intel.com>
Co-authored-by: Vafin, Maxim <maxim.vafin@intel.com>
Co-authored-by: Ilya Znamenskiy <ilya.znamenskiy@intel.com>
This commit is contained in:
Maxim Shevtsov 2020-09-07 17:24:59 +03:00 committed by GitHub
parent 1e6ca0627a
commit c8b783f644
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 13 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
inference-engine/thirdparty/clDNN/kernel_selector/core/actual_kernels/softmax/softmax_kernel_items_class_optimized.cpp vendored
View File

@ -76,6 +76,7 @@ JitConstants SoftmaxKerneItemsClassOptimized::GetJitConstants(const softmax_para
auto jit = SoftmaxItemsClassKernelBase::GetJitConstants(params, kd);
jit.AddConstant(MakeJitConstant("WORKITEMS_PER_CLASSES", workitems_per_classes));
jit.AddConstant(MakeJitConstant("HAS_DRIVER_PROBLEMS", params.engineInfo.bIMADSupport));
return jit;
}

12
inference-engine/thirdparty/clDNN/kernel_selector/core/cl_kernels/softmax_gpu_items_class_optimized.cl vendored
View File

@ -63,12 +63,24 @@ KERNEL(softmax_items_class_optimized)(__global INPUT0_TYPE* input, __global OUTP
ACCUMULATOR_TYPE denominator = 0.0;
for (uint cls = 0; cls < FULL_ITERATIONS_NUM; cls++)
{
// This is a temporary solution for unresolved problem when ocl kernels compilation step doesn't produce actual binaries
// for current kernel but driver doesn't report any errors (JIRA CVS-32211)
#if HAS_DRIVER_PROBLEMS
data[cls] = data[cls] == max_value ? 1.0 : native_exp(data[cls] - max_value);
#else
data[cls] = native_exp(data[cls] - max_value);
#endif
denominator += data[cls];
}
if(simd_lane < LEFTOVERS)
{
// This is a temporary solution for unresolved problem when ocl kernels compilation step doesn't produce actual binaries
// for current kernel but driver doesn't report any errors (JIRA CVS-32211)
#if HAS_DRIVER_PROBLEMS
data[DATA_PER_WORKITEM-1] = data[DATA_PER_WORKITEM-1] == max_value ? 1.0 : native_exp(data[DATA_PER_WORKITEM-1] - max_value);
#else
data[DATA_PER_WORKITEM-1] = native_exp(data[DATA_PER_WORKITEM-1] - max_value);
#endif
denominator += data[DATA_PER_WORKITEM-1];
}