natas_hw
/
openvino
0
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
12,372 Commits 48 Branches 60 Tags 1.3 GiB
Commit Graph

1153 Commits

Author SHA1 Message Date
Andrew Kwangwoong Park 5604566795
[GPU] Minor fix to get correct input layout for dump layer (#19686) 
Signed-off-by: Andrew Park <andrew.park@intel.com>
 2023-09-11 12:21:09 +04:00
Andrew Kwangwoong Park 161ba14796
[GPU] Fix GatherND shape agnostic ref kernel (#19706) 
Signed-off-by: Andrew Park <andrew.park@intel.com>
 2023-09-11 12:20:10 +04:00
Kelvin Choi 2f4f76070f
[GPU] Update strided_slice for partially dynamic shape case (#19467) 2023-09-10 22:44:13 -07:00
Ilya Lavrenov 51d77cb59f
Migrate to ade v0.1.2c (#19714) 2023-09-11 07:15:59 +04:00
Ilya Lavrenov ed230cd879
Fixed build with oneDNN GPU in some Conan scenarios (#19711) 2023-09-09 01:52:58 +04:00
Ilya Lavrenov 1ef9cc70b5
Fixed compilation with gcc-13.2 (#19689) 2023-09-08 18:39:24 +04:00
Mateusz Tabaka a55b5381d3
Move BroadcastTransition to MOC (#19543) 
* Move BroadcastTransition to MOC

Broadcast that could be eliminated by BroadcastElementwiseFusion are moved down the graph
(by BroadcastTransition registered in the plugins). That prevents BroadcastElementwiseFusion
to eliminate them.

Ticket: CVS-118899

* dont count const layers

* remove virtual inheritance
 2023-09-08 11:05:54 +04:00
Jade Cho e2b553302b
[GPU] Use preferred output format if node impl type is onednn (#19601) 
+ Changes to use preferred format in both cldnn and onednn for gemm and
FC when shape inferencing.
 2023-09-08 15:56:40 +09:00
Sergey Shlyapnikov 4eb9c57424
[GPU] Add input feature leftovers processing for fully_connected_gpu_bs_f_bsv16_af8_vload kernel (#19650) 2023-09-07 13:20:11 +04:00
Taylor Yeonbok Lee 4124851d2b
Revert "[GPU] Fixed reordered memory cache not to contain original weight memory (#19465)" (#19659) 
This reverts commit e8f1df495c.
 2023-09-06 22:35:00 -07:00
David Nam cb479f4a5d
[GPU] No need to add reorder after strided_slice (#19411) 2023-09-06 11:43:51 -07:00
Sungeun Kim b7758b0504
[GPU] update the data-type of primitive from ops (#19302) 
* set the data-type of transpose by dt of ops.
* set output_data in calc_output_layouts
 2023-09-06 13:30:53 +09:00
Ilya Churaev 1d62f0141d
Rename cmake ie_ macros and ie_parallel script (#19606) 
* Rename cmake ie_ macros and ie_parallel script

* Add warning messages
 2023-09-05 19:31:52 +02:00
Vladimir Paramuzov 77600c7701
[GPU] Add FullyConnected custom op (#19539) 2023-09-05 16:47:05 +04:00
Mingyu Kim 3d679edf18
[GPU] Remove propagate_constants pass at pre_optimize_graph stage (#19323) 
Co-authored-by: Lyamin-Roman <roman.lyamin@intel.com>
 2023-09-05 16:37:31 +04:00
Vladimir Paramuzov 2f782b2131
[GPU] Add permute primitive instead of manual copy for deconv weights (#19516) 2023-09-04 17:03:49 +04:00
Ilya Lavrenov 3bc38695c5
A set of fixes for Conan C++ package manager (#19552) 2023-09-04 11:32:39 +04:00
Tomasz Jankowski 51df17912b
[Ref_Impl] Change namespace from nG to OV (#19363) 
* Use quotes for openvino includes

* Drop runtime from openvino::reference

* Drop runtime::reference

* Replace ngraph::reference with ov::reference - defs

* Replace ngraph::reference with ov::reference - uses

* Drop redundant nesting

* Fix non arch64 builds

* Move coordinate*pp files under openvino

* Move Coordinate... helpers under ov:: namespace

* Revert not needed changes

* Fix missing namespace scope

* Fix compilation

* Fix code style

* Use ov suppress deprecated macro instead of ngraph

---------

Co-authored-by: Raasz, Pawel <pawel.raasz@intel.com>
 2023-09-04 08:07:06 +04:00
Andrew Kwangwoong Park 92c6316e8e
[GPU] Fix input feature map indexing with pad and batch indices for ROIAlign (#19511) 
* [GPU] Fix input feature map indexing with pad and batch indices for ROIAlign

* Fix failed TCs for ov_gpu_func_tests

Signed-off-by: Andrew Park <andrew.park@intel.com>

* Fix to do batch interpretation for inconsistency between ROIALign input and const 1D tensor

Signed-off-by: Andrew Park <andrew.park@intel.com>

---------

Signed-off-by: Andrew Park <andrew.park@intel.com>
 2023-09-01 12:43:23 -07:00
Vladimir Paramuzov 38cad619af
[GPU] Allow simple attached mem as input memory for network (#19419) 2023-09-01 09:54:30 +04:00
Steve Yoo 05a24b1776
[GPU] Try to use softmax_ref when types are mismatched (#19209) 
* Remove support key for UINT8 and INT8
 2023-08-31 16:39:36 -07:00
Taylor Yeonbok Lee e8f1df495c
[GPU] Fixed reordered memory cache not to contain original weight memory (#19465) 
* Fixed reordered memory cache not to contain original weight memory

* Applied review comment

* Applied review comment
 2023-08-29 21:54:32 -07:00
Wilson Seok 94c21b53b3
fix build error by removing makeDynamicParam (#19431) 2023-08-28 12:10:05 +02:00
Wilson Seok 293c672064
add sqrt activation support in cpu_impl (#19421) 2023-08-25 12:10:01 -07:00
Wilson Seok f6dca869b2
fix reduce perferred format selection and layout for partial shape (#19319) 2023-08-25 12:09:25 -07:00
Min, Byungil bcedb0bb9b
[GPU] Resolve accuracy issue from clamp fused prims (#19409) 
+ Added condition when clamp activation is added to fused-ops for fp16 overflow
+ Added test-cases

Signed-off-by: Min, Byungil <byungil.min@intel.com>
 2023-08-25 11:21:09 -07:00
Wilson Seok f962511a84
[GPU] add check condition of input dynamic shape in conv fusing (#19219) 2023-08-25 16:13:53 +04:00
Vladimir Paramuzov a45e5e03c5
[GPU] Added some formats for pvc (#19388) 2023-08-25 15:09:42 +04:00
Tomasz Jankowski bcad953f5f
[Ref_Impl] Rename file paths to openvino relative (#19284) 
* Move files to new directories

* Use quotes for openvino includes

* Provide proxy calls for transition

of dependant components.

* Correct includes style

* Redo proxies

* Fix deprecated

* Move aliases to proxy files

* Apply code style
 2023-08-25 06:43:06 +04:00
Kelvin Choi ce47522165
[GPU] Memory reuse false for dynamic and null impl case (#19354) 2023-08-24 17:32:07 +09:00
Kelvin Choi c89b9edfe7
[GPU] 7-dimention only supports plain format (#19039) 
* Skip concat_input_order opt in case dependancy is dynamic

* Add plain 7d 8d case for jitter pitch size
 2023-08-23 20:00:02 -07:00
Paul Youngsoo Ahn 99cc3624b7
[GPU] Fix accuracy issue (#19351) 
- [scatter_update] Use input index for input buffer instead of output index
- [concat cpu impl] Sync input layout and mem_ptr when input host tensor creation
- Add unit tests for scatter_update and concat cpu impl
 2023-08-23 17:57:18 -07:00
Oleg Pipikin ab900606cd
Remove makeDynamicParams (#19226) 
* Remove makeDynamicParams

* Apply comments

* Fix1

* Fix2

* Fix3
 2023-08-23 18:57:29 +02:00
Vladimir Paramuzov 3b2e263879
[GPU] Fix reshape optimization (#19270) 2023-08-23 10:25:04 +00:00
Oleg Pipikin 7aa51d6775
Remove makeParams (#19306) 2023-08-23 11:39:05 +04:00
Oleg Pipikin de65abc6b3
Remove WA for vpu repo with CommonTestUtils namespace (#19275) 2023-08-22 13:03:45 +00:00
Sergey Shlyapnikov 7df8d1ca2d
[GPU] Add per iteration performance profiling mode (#18637) 2023-08-22 12:37:36 +02:00
Vladimir Paramuzov 526d76c81f
[GPU] New headers and namespaces in some parts (#19229) 2023-08-18 15:57:15 +04:00
Jade Cho 2394732055
[dGPU] Use cldnn concat if data type is fp32. (#19240) 2023-08-17 09:14:26 +00:00
Min, Byungil 1b9de79d0d
[GPU] Resolve fp16 overflow of mul (#19173) 
+ Fixed black output image by Nan output.
+ Resolved fp16 overflow of gemm primitive before softmax
+ Added fused post ops of clamp activation not to get inf which caused Nan output
+ Added new pass modify_fused_ops

Signed-off-by: Min, Byungil <byungil.min@intel.com>
 2023-08-17 18:09:10 +09:00
Andrei Gorbachev 49bbcb4cf6
[ПЗГ] fix coverity (#19204) 2023-08-17 12:46:35 +04:00
Wilson Seok b656feee57
[GPU] fix priorbox legacy code in CreateOp() and allow to use max_size is null (#19143) 2023-08-16 17:54:04 +04:00
Min, Byungil ef6c8c1d66
[GPU] Make gemm_tiled_opt support outer axis (#19210) 2023-08-16 16:43:47 +04:00
Andrew Kwangwoong Park 8a0a4df941
[GPU] Fix unintentional expansion of pad's rank into 2D (#19106) 
* [GPU] Fix unintentional expansion of pad's rank into 2D

* Add explicit 1d pad TCs for ov_gpu_unit_tests
 2023-08-16 14:29:00 +09:00
Roman Lyamin 188434e969
[GPU] image_2d_weights_winograd_6x3_s1 formats fix (#19205) 2023-08-16 08:59:43 +04:00
Sergey Shlyapnikov d13ae31a61
[GPU] Allow to use infinity value as a Pad's fill value (#19201) 2023-08-15 14:31:49 +00:00
Eddy Kim 0194cdce3b
[GPU] updated to add reorder after parameter when it is connected to quantize (#19182) 
* updated to add reorder after parameter when it is connected to quantize

* removed redundant code
 2023-08-14 23:29:49 -07:00
Roman Lyamin 9be28328bf
Added giy_xs_os_xsv2_osv8__ao32 format to to_weights_layout map (#19181) 2023-08-15 08:49:50 +04:00
Paul Youngsoo Ahn e2db808495
[GPU] Fix multiple output issue in get_output_layout(#19186) (#19186) 2023-08-14 20:49:21 +00:00
Sergey Shlyapnikov 86c4c6785d
[GPU] Don't invalidate users after reorder insertion (#19137) 2023-08-14 14:16:54 +04:00