aboutsummaryrefslogtreecommitdiff
path: root/src/core/CL/cl_kernels/depthwise_convolution.cl
AgeCommit message (Collapse)Author
2021-07-02Rework OpenCL Depthwise ConvolutionGian Marco Iodice
- Remove dedicated kernels for NCHW. Now we only use NHWC with permute - Remove specialized kernels for 3x3 NHWC - Simplify CLDepthwiseConvolutionLayer.cpp to call just the native implementation for both floating-point and quantized data types - Develop two parametric opencl kernels for depthwise convolution layer NHWC (floating-point and quantized) - Add support to export the weights to cl_image - Extend test for depthwise convolution on opencl Resolves COMPMID-4417 Change-Id: Ibe533f79c2860f9cac8e921895d5a8f947753a5c Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5893 Reviewed-by: Giorgio Arena <giorgio.arena@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2021-06-30Revert "Rework OpenCL Depthwise Convolution"Gian Marco Iodice
This reverts commit 561c176598cd14245e2e7918fdf136d1c888d1da. Reason for revert: <validation> Change-Id: I6f2d61c27520439bb538e9265736532104b24cf8 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5127 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2021-06-24Rework OpenCL Depthwise ConvolutionGian Marco Iodice
- Remove dedicated kernels for NCHW. Now we only use NHWC with permute - Remove specialized kernels for 3x3 NHWC - Simplify CLDepthwiseConvolutionLayer.cpp to call just the native implementation for both floating-point and quantized data types - Develop two parametric opencl kernels for depthwise convolution layer NHWC (floating-point and quantized) - Add support to export the weights to cl_image - Extend test for depthwise convolution on opencl Resolves COMPMID-4417 Change-Id: I253dd5d959a70783c82e62b1771a5e9f91621cb0 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5806 Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>
2021-04-20CLDepthwiseConvolutionLayer rework - Part 1Giorgio Arena
Remove the reshaped variant for CLDepthwiseConvolutionLayer 3x3 NHWC Quantized - Remove kernel selection by GPUTarget - Remove unused quantized support from the NHWC kernel - Remove CLDepthwiseConvolutionLayerReshapeWeightsKernel - Remove OpenCL kernels for reshaped dwc 3x3 quantized and weights reshape - Remove the "_bifrost" suffix in common OpenCL kernel - Remove the ICLDepthwiseConvolutionLayer3x3Kernel common interface Resolve COMPMID-3864, COMPMID-3907 Change-Id: Icfac0fb6c00e214985beb05dad7c0cdbbee7d830 Signed-off-by: Giorgio Arena <giorgio.arena@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5447 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2020-12-10[Review Shape] CLDepthwiseConvolutionLayer mismatchesGiorgio Arena
- Fixed a bug that corrected the number of dimensions of a TensorShape for added trailing 1s - Avoided adding offset_first_element for the Depthwise 3x3 NCHW OpenCL kernels, since it wouldn't align with the window which is based on the output - Adjusted padding requirements along the x for Depthwise 3x3 NCHW. The kernel should always add 2 * dilation_(x/y) to the num_elems_read_x/y - Adjusted the kernel's border_size given to the border handler at function level - Added the dataset that previously made the tests fail Resolves: COMPMID-4041 Change-Id: Ifab7d38b263f12173fcc96a5f0bd3375756c3c53 Signed-off-by: Giorgio Arena <giorgio.arena@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4673 Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: SiCong Li <sicong.li@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com>
2020-11-12COMPMID-3735 Remove OpenCL padding: CLSoftmaxLayerKernelGiorgio Arena
- Renamed SELECT_DATA_TYPE to SELECT_VEC_DATA_TYPE to reflect its usage with vectors. SELECT_DATA_TYPE(dt) will now return the primitive data type - Changed the interface of VEC_OFFS and V_OFFS in order to receive the primitive data type as a parameter rather than its vector form - Performed a general cleanup of the kernels, such as creating macro for sum and max reduces, remove reduntant macros, defines, variables, calculations, etc... - Using VEC_SIZE and VEC_SIZE_LEFTOVER in every kernel in order to allow computation for smaller shapes without adding paddings - Removed the actual padding from the kernel and adjusting its calculations accordingly. Added asserts for padding removal checks. Removed invalid Validate tests. Change-Id: If5ccbd5d34e255d38c7f6bfe8740e2b80b28e264 Signed-off-by: Giorgio Arena <giorgio.arena@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4277 Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: SiCong Li <sicong.li@arm.com> Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com>
2020-10-23COMPMID-3713 Remove OpenCL padding: CLDepthwiseConvolutionLayerNativeKernelGiorgio Arena
Signed-off-by: Giorgio Arena <giorgio.arena@arm.com> Change-Id: Ic43aba8a6a0a106fc4c1f665ff5cc3ccb31f403d Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4235 Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
2020-10-21COMPMID-3712 Remove OpenCL padding: CLDepthwiseConvolutionLayer3x3NHWCKernel ↵Giorgio Arena
FP16/32 Removed unused N from partial block loading macro Created utility to assert change in padding Signed-off-by: Giorgio Arena <giorgio.arena@arm.com> Change-Id: Ifdd30c66dbf5f2842c6b2d939000613d5011708e Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4192 Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2020-10-12COMPMID-3826 ArmNN Nightly failing for CLGiorgio Arena
Change-Id: I09f557b5cecafc669e12764e8592457212168d62 Signed-off-by: Giorgio Arena <giorgio.arena@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4131 Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com>
2020-07-20COMPMID-3532: Align data type support between doxygen and implementation - CLMichele Di Giorgio
Also removes some unused code. Change-Id: I85687c40999c3cdf9e6fccfcd020b0901a9515fe Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3581 Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com>
2020-07-09COMPMID-3324: Adjusting capitalization of Arm copyright claim to reflect Arm ↵Michele Di Giorgio
preferred presentation Change-Id: Ib7dcfcbb24b408999dfae366b9da396485aacf78 Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3525 Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com> Tested-by: Georgios Pinitas <georgios.pinitas@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2019-10-22COMPMID-2452: Fix CLDepthwiseConvolutionLayerNativeKernelMichele Di Giorgio
Change-Id: I284e4f4200a4b688bfd908e417b00973b547abda Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com> Reviewed-on: https://review.mlplatform.org/c/2140 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>
2019-10-15COMPMID-2600: Implement a new and generic depthwise convolution for CL ↵Michele Di Giorgio
QASYMM8 NHWC The NCHW case is supported at function level by permuting the inputs/outputs to NHWC. This patch also removes CLDirectConvolutionLayerOutputStageKernel which is deprecated and some kernels which were only used in the generic case of depthwise convolution. Change-Id: I91e0f02d0a2f4a4a352e08c248e648944137fe68 Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com> Reviewed-on: https://review.mlplatform.org/c/2056 Reviewed-by: Giorgio Arena <giorgio.arena@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
2019-09-12COMPMID-2599: Implement a new and generic depthwise convolution on OpenCL ↵Gian Marco Iodice
(Fp32/FP16-NHWC) Part 1 Change-Id: I5e1d27a7006199e9229e455a1df9bfc2ed4e8341 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://review.mlplatform.org/c/1898 Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Giorgio Arena <giorgio.arena@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com>
2019-08-13COMPMID-2580: Fix out of bound read in Depthwise Convolution layer (OpenCL)Gian Marco Iodice
Change-Id: I00e39ed21cc30034aa10ac58b64d533e833eafc8 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://review.mlplatform.org/c/1727 Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com>
2019-05-15COMPMID-2194: Refactor activation function macro in OpenCL. Change all ↵Usama Arif
activation calls to macro from activation_float_helpers.h The different kernels now call the macro from activation_float_helpers.h. activation_helpers.h is now removed. Change-Id: I2e1314c6bc891809e88590d99e048072541cca14 Signed-off-by: Usama Arif <usama.arif@arm.com> Reviewed-on: https://review.mlplatform.org/c/1123 Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com>
2019-05-03COMPMID-2108: Fuse Activation Layer in CLDepthwiseConvolutionLayer3x3Kernels ↵Manuel Bottini
for F32 Change-Id: I39dd23696b6d8573e172a59b9e327b6a69886f08 Signed-off-by: Manuel Bottini <manuel.bottini@arm.com> Reviewed-on: https://review.mlplatform.org/c/973 Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Usama Arif <usama.arif@arm.com> Reviewed-by: Giuseppe Rossini <giuseppe.rossini@arm.com>
2019-04-18COMPMID-2047: Add support for dilation in CLDepthwiseConvolution.Usama Arif
Change-Id: I3106aa34bd168985a56791613d95072756be6e9b Signed-off-by: Usama Arif <usama.arif@arm.com> Reviewed-on: https://review.mlplatform.org/c/958 Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Pablo Marquez <pablo.tello@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com>
2019-01-30COMPMID-1691: Optimize CLDepthwiseConvolutionKernel (QASYMM8/NHWC) for 3x3 ↵giuros01
kernels (stride=1 and stride=2) Change-Id: I7d0d2dc350feeb40d253d17f9ffd5051a8fb42ef Reviewed-on: https://review.mlplatform.org/511 Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com>
2018-11-06COMPMID-1703: Collapse the 4th dimensions in ↵Georgios Pinitas
CLDepthWiseConvolutionLayer3x3Kernel Change-Id: Ie274da79b15c03f86dfedc85bb721b3de34a0bb4
2018-11-02COMPMID-1581: Collapse windowsGeorgios Pinitas
Change-Id: Iec56c9a96d9736a63f13b65efa33311950f20661 Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/148572 Reviewed-by: Anthony Barbier <anthony.barbier@arm.com> Tested-by: bsgcomp <bsgcomp@arm.com>
2018-11-02COMPMID-1533 Implementing CLDepthWiseConvolutionLayer with FP16 (NHWC)Giorgio Arena
Change-Id: I46965aeb1fffba8cbf083cab7284c549b0e94d00 Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/145334 Tested-by: Jenkins <bsgcomp@arm.com> Reviewed-by: Michele DiGiorgio <michele.digiorgio@arm.com>
2018-11-02COMPMID-1383: OCLGrind failure in CLDepthwiseConvolution 3x3 stride 1 NHWCGeorgios Pinitas
Seems OCLGrind to operate wrongly on some intrinsics when there is a mixture of vectors and scalars passed to it. Change-Id: I9e3782e739603ec59bacc3c77d91a70b1899fe3e Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/139474 Tested-by: Jenkins <bsgcomp@arm.com> Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
2018-11-02COMPMID-970 : Remove QS8 / QS16 supportVidhya Sudhan Loganathan
Removed fixed point related code. Change-Id: I487acf138dace3b0450e0d72ca7071eaec254566 Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/137678 Tested-by: Jenkins <bsgcomp@arm.com> Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
2018-11-02COMPMID-811 Add NHWC data format support for CL depthwise convolutionGiorgio Arena
Change-Id: I574f7945f0be009c638d860028bce8b52b4120fd Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/136484 Tested-by: Jenkins <bsgcomp@arm.com> Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
2018-11-02COMPMID-1125: Add support for FP16 in CLDepthwiseConvolutionMichele Di Giorgio
Change-Id: I4838f5a8e4c33ed646cd05e0bb682fca635a29a3 Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/130469 Tested-by: Jenkins <bsgcomp@arm.com> Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>
2018-11-02COMPMID-926 Add depth multiplier support to NEON/CL/GLES depthwise convolutionGiorgio Arena
Change-Id: I03f32c62350e5ea43e77bb15fc5a832d83719e3b Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/126657 Tested-by: Jenkins <bsgcomp@arm.com> Reviewed-by: Michele DiGiorgio <michele.digiorgio@arm.com> Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
2018-11-02COMPMID-949: Optimizing CLDepthwiseConvolution3x3Kernel for FP16Michele Di Giorgio
Change-Id: I2af6544eab17004c5b3de56557cb2cc5efecc915 Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/122181 Tested-by: Jenkins <bsgcomp@arm.com> Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>
2018-11-02COMPMID-927: Adding support for FP16 in CLDepthwiseConvolutionLayer3x3Michele Di Giorgio
Change-Id: Ie5f299c7a7fbe3062cee22bb2b4ae5df818fe490 Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/121178 Tested-by: Jenkins <bsgcomp@arm.com> Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com> Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
2018-11-02COMPMID-856: CL Depthwise Convolution QASYMM8 supportGeorgios Pinitas
Change-Id: Ic6097e7cf160e8b829fb521b7b99d9a57d9799d3 Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/118774 Tested-by: Jenkins <bsgcomp@arm.com> Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
2018-11-02COMPMID-895 - Optimizing CLDepthwiseConvolution3x3KernelGian Marco
This patch brings the MACs utilisation up to 25 % when both stride_x and stride_y are equal to 1 Performance reported in the following confluence page: https://confluence.arm.com/display/MLENG/Depthwise+convolution+3x3+FP32+performance%3A+ACL+18.02 Change-Id: Ida1b64be9a88805902a3d90194559b58eb1224a3 Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/119068 Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com> Tested-by: Jenkins <bsgcomp@arm.com>
2018-11-02COMPMID-556 Improved indentation and error handling in format_doxygen.pyAnthony Barbier
Change-Id: I6f51ffe6c324d9da500716b52c97c344f2a2a164 Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/110486 Tested-by: BSG Visual Compute Jenkins server to access repositories on http://mpd-gerrit.cambridge.arm.com <bsgcomp@arm.com> Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
2018-11-02COMPMID-643: Add bias to CLDepthwiseConvolution.Georgios Pinitas
Change-Id: Ibfe7b8c1172d10cbcae7971fe86b82090519d31d Reviewed-on: http://mpd-gerrit.cambridge.arm.com/92798 Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com> Reviewed-by: Jaroslaw Rzepecki <jaroslaw.rzepecki@arm.com> Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
2018-11-02IVGCVSW-601: support for asymetric padding in cl conv and depthwise convJaroslaw Rzepecki
Change-Id: I5c6c95091ae77dba96459c0640f9f6167a988c8c Reviewed-on: http://mpd-gerrit.cambridge.arm.com/91700 Reviewed-by: Anthony Barbier <anthony.barbier@arm.com> Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
2018-11-02COMPMID-452 CL Generic Depthwise Convolution implementation.Giorgio Arena
Change-Id: I115e48fe6ce5e281f3791aa5d80fdc754cdd2b5e Reviewed-on: http://mpd-gerrit.cambridge.arm.com/85082 Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com> Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
2018-11-02COMPMID-452 CL Depthwise Separable Convolution Layer kernel implementation, ↵Giorgio Arena
validation and benchmarking for 3x3xC depthwise filter and DataType::F32. Change-Id: I95c0c87709763cdbf58d0de66025eac86e30791b Reviewed-on: http://mpd-gerrit.cambridge.arm.com/82768 Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com> Reviewed-by: Steven Niu <steven.niu@arm.com>
generated by cgit v1.2.1 (git 2.23.0) at 2024-05-09 06:48:41 +0000