aboutsummaryrefslogtreecommitdiff
path: root/src/gpu
AgeCommit message (Collapse)Author
2023-05-04Implement OpenCL MatMul heuristic for Arm® Mali™-G710Gian Marco Iodice
- Add heuristic for f32/f16 and int8 quantized data types - Include MatMul configuration selection in the CLMatMul operator Resolves COMPMID-5950, COMPMID-5957, COMPMID-5959, COMPMID-5925, COMPMID-5926, COMPMID-5927, COMPMID-5928 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Change-Id: Ic222148da0337b88d4d8c960e3b6ac31003d8bcb Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9564 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2023-05-02Fix export_to_cl_image issue in the fp16 GeMM implementationGian Marco Iodice
- The issue affects Fp16 GeMM on Arm® Mali™-G78 - The issue was caused by a missing fallback implementation for the case when export_to_cl_image cannot be used - The new implementation fixes this issues and make the GeMM implementation for M=1 also faster (4-5% on various networks with fully connected at the end of the model) - This patch also enables the H0=0 case in the GeMM examples Resolves COMPMID-5812, COMPMID-5688, and COMPMID-6147 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Change-Id: Ib7b355ae25337962598dd2ba21665b1a6b48686f Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/c/VisualCompute/ComputeLibrary/+/514664 Tested-by: bsgcomp <bsgcomp@arm.com> Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com> Comments-Addressed: bsgcomp <bsgcomp@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9526 Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2023-05-02Add fp16 GeMM heuristic for Arm® Mali™-G710Gian Marco Iodice
- Performance improvements on various networks between 5-20% Resolves COMPMID-6030 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Change-Id: Idcf7de57e6f5a94a6a94ec78229dd53c24de44f4 Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/c/VisualCompute/ComputeLibrary/+/514481 Tested-by: Viet-Hoa Do <viet-hoa.do@arm.com> Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com> Comments-Addressed: bsgcomp <bsgcomp@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9524 Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2023-04-27Add quantized CL MatMul kernel for LHS NT, RHS TJakub Sujak
Implement a native kernel for batched Matrix Multiplication for the quantized data types QASYMM8 and QASYMM8_SIGNED and with the MatMul attributes `adj_x = false, adj_y = true`. Resolves: COMPMID-5923 Change-Id: I477b2dd886edfe83beaba9efc7d6b05ed19f5da4 Signed-off-by: Jakub Sujak <jakub.sujak@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9467 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: SiCong Li <sicong.li@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2023-04-26Change fp16 GeMM heuristic for Arm® Mali™-G77Gian Marco Iodice
- Replace existing heuristic with look-up tables - Expected performance improvement is between 5-15% on various models Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Change-Id: Ie26ddf66895ede131aa06fde7b200ef94d2dd467 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9472 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2023-04-26Improve Winograd performance on OpenCLGian Marco Iodice
- Performs more output elements per work-item in the case of Fp16 computation in Winograd Input/Output transform Resolves COMPMID-6018 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Change-Id: If5e6f5182eff8c1f05a3505c437d0a997490f0bd Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9447 Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Jakub Sujak <jakub.sujak@arm.com> Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2023-04-20Implement CL kernel for a native batched matmul Quantized - LHS transposed, ↵Omar Al Khatib
RHS transposed Resolves: [COMPMID-5924] Signed-off-by: Omar Al Khatib <omar.alkhatib@arm.com> Change-Id: I9ba657737eb1e3a096c8341ad4ad311571f8edeb Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9454 Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: SiCong Li <sicong.li@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2023-04-17Add quantized CL MatMul kernels for Lhs NT/T, Rhs NTGunes Bayir
Implement OpenCL kernels for batched Matrix Multiplication for the quantized data types QASYMM8 and QASYMM8_SIGNED. Quantized MatMul is supported with the following MatMul attributes: * adj_x = false, adj_y = false * adj_x = true, adj_y = false We consider native format kernels only. In other words, no reshaping of the operand matrices is done. Resolves: COMPMID-5921, COMPMID-5922 Change-Id: I99e0f68054a2bd635c60ec2641acc2e7ff398473 Signed-off-by: Omar Al Khatib <omar.alkhatib@arm.com> Signed-off-by: Gunes Bayir <gunes.bayir@arm.com> Signed-off-by: Jakub Sujak <jakub.sujak@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9435 Reviewed-by: SiCong Li <sicong.li@arm.com> Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2023-04-14Align naming convention of ClMatMulJakub Sujak
Ensure naming of MatMul on GPU conforms to the naming convention <backend><operator><config> i.e. ClMatMul operator with the backend ClMatMulNativeKernel. Resolves: COMPMID-6015 Change-Id: I021d235b023ad17fe97bd6913e6a50d0ba4b194e Signed-off-by: Jakub Sujak <jakub.sujak@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9443 Reviewed-by: Gunes Bayir <gunes.bayir@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com>
2023-04-04Support dynamic weights for Fully Connected layers on GPUJakub Sujak
The fully connected function and operator running on GPU have been adapted to support dynamic weights. Dynamic weights require the reshape and data layout conversion of weight tensors at runtime in the prepare stage of the operator. The implementation for GPU is identical to the CPU implementation. This patch also deprecates the `are_weights_reshaped` option in Fully Connected. Resolves: COMPMID-5870 Change-Id: I28f967695879d82cc91a928d95308a4e0e52a597 Signed-off-by: Jakub Sujak <jakub.sujak@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9403 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2023-04-03Implement MatMul FunctionRamy Elgammal
Resolves: COMPMID-5949 Signed-off-by: Ramy Elgammal <ramy.elgammal@arm.com> Change-Id: Idd8cfe6ea94a14f0b23178f6781251b5f0955563 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9390 Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gunes Bayir <gunes.bayir@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2023-03-24Work around CLScale compiler-specific issueSiCong Li
Resolves COMPMID-5985 Signed-off-by: SiCong Li <sicong.li@arm.com> Change-Id: I0e789619f09e3adefe3655df347390f057300c0f Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9373 Reviewed-by: Gunes Bayir <gunes.bayir@arm.com> Reviewed-by: Pablo Marquez Tello <pablo.tello@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com>
2023-03-24Add Texture Pipe Support for Matmul Lhs T/NT Rhs NT kernelsGunes Bayir
Resolves: COMPMID-5945, COMPMID-5954 Change-Id: I7b27021d21f8e08c4896f6b1f595a75125064f9e Signed-off-by: Gunes Bayir <gunes.bayir@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9356 Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-by: SiCong Li <sicong.li@arm.com> Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2023-03-20Implement OpenCL MatMul for Lhs T Rhs T/NT FP32/16Gunes Bayir
- Implement opencl kernel for LHS transposed and RHS non-transposed - Implement opencl kernel for LHS transposed and RHS transposed - Add validation tests Resolves: COMPMID-5953, COMPMID-5955 Change-Id: I55589acbffe86c44e29807574975978a1ec09bad Signed-off-by: Gunes Bayir <gunes.bayir@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9345 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2023-03-17Implementation of RSQRT for quantized int8Ramy Elgammal
Resolves: COMPMID-5863 Change-Id: I9ff67face62826c1d335a6b941e8516be39bdac8 Signed-off-by: Ramy Elgammal <ramy.elgammal@arm.com> Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/c/VisualCompute/ComputeLibrary/+/488768 Tested-by: bsgcomp <bsgcomp@arm.com> Comments-Addressed: bsgcomp <bsgcomp@arm.com> Reviewed-by: Gunes Bayir <gunes.bayir@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9225 Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2023-03-17Implement OpenCL MatMul for Lhs NT Rhs T/NT FP32/16Ramy Elgammal
- Implement ClNativeMatMulKernel class - Implement opencl kernel for LHS non-transposed and RHS non-transposed - Implement opencl kernel for LHS non-transposed and RHS transposed - Add test fixture and dataset for matmul - Implement transpose_tensor() for reference implementation to transpose high dimensional tensors Resolves: COMPMID-5944, COMPMID-5951 Co-authored-by: Gunes Bayir <gunes.bayir@arm.com> Co-authored-by: Ramy Elgammal <ramy.elgammal@arm.com> Change-Id: I1d5b8978f41be27baddb3153ade880472141573f Signed-off-by: Gunes Bayir <gunes.bayir@arm.com> Signed-off-by: Ramy Elgammal <ramy.elgammal@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9333 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2023-03-06Fix LWS search space used by CLTunerSiCong Li
* Ensure CLTuner uses the real GWS used by run(), instead of the static GWS (which is usually changed at run time), by caching GWS in each kernel Note this is a somewhat inelegant workaround. The real issue stems from the fact that execution window and scheduler are very much coupled with our operator run() / run_op() method. (Please see COMPMID-5934) * Restrict LWS values to explore within GWS bound for exhaustive mode * Refactor gws_from_window() to include all the information required to calculate GWS * Log lws search space used for tuning * Fix ClDirectConv2dKernel config id Resolves COMPMID-5892 Signed-off-by: SiCong Li <sicong.li@arm.com> Change-Id: I420490d8b94d13ada2e44eb0a12078f883379334 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9193 Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2023-02-28Add an option to use lowest for max-poolingAdnan AlSinan
- Add a parameter in PoolingLayerInfo class to pick which value to use as min for max-pooling. Resolves: [ONCPUML-1166] Signed-off-by: Adnan AlSinan <adnan.alsinan@arm.com> Change-Id: I34e1cccc15176bbf31523c61e99f3188ddca23e1 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8989 Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: SiCong Li <sicong.li@arm.com> Reviewed-by: Pablo Marquez Tello <pablo.tello@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2023-01-18Add broadcast batched matmul validation casesSiCong Li
Related to: COMPMID-5660 Signed-off-by: SiCong Li <sicong.li@arm.com> Change-Id: I2314c8b21acc638402c77080d59db2f3fed58fe2 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8911 Reviewed-by: Jakub Sujak <jakub.sujak@arm.com> Reviewed-by: Mohmun02 <MohammedSuhail.Munshi@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2023-01-17Fix ClGemm crashes on unsupported data typesSiCong Li
Resolves COMPMID-5814 Change-Id: I09b206374cf3844c09aebd3c664daec9c2335e6d Signed-off-by: SiCong Li <sicong.li@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8953 Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2023-01-10Fix CL DirectConvolutionLayer validate testsSiCong Li
* Add missing activation infos * Remove faulty test "Shrink window" * Split the tests based on data layout * Fix ClDirectConv2dKernel::validate logic Fused activation in NCHW is not supported at all Resolves: COMPMID-5801 Change-Id: I64dfbd24b77bb02fb4a88b73d5ef84676d85b4fd Signed-off-by: SiCong Li <sicong.li@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8899 Reviewed-by: Jakub Sujak <jakub.sujak@arm.com> Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2023-01-10Extend cl image support to input and output tensorsGian Marco Iodice
- Add support for texture image to input and output of direct convolution - Extend T_LOAD2D_INDIRECT macro to read values from cl image storages Resolves COMPMID-5715 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Change-Id: Idb0410f53f6d0763cd9e39895a7cbf9bc826d33a Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8904 Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2022-12-29Optimize CL Scale/Resize Quantized by removing (de)quant. codeGunes Bayir
This patch removes the quant/dequant code in CLScale and the Resize operator in dynamic fusion. We don't support different quantization information for input and output and in this case the quantization and dequantization is not necessary. The very same optimization was delivered for cpu. It also moves the SCALE_X and SCALE_Y arguments to look-up table from build options in the template writer of Resize. Change-Id: Icd043c8671220c8feea935dd4b24a5b17c6c4ea4 Signed-off-by: Gunes Bayir <gunes.bayir@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8888 Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com> Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2022-12-29Update the ClConv2d heuristicGian Marco Iodice
- Update the ClConv2d heuristic to call indirect convolution on Arm® Mali™-G77 Gpus - Implement the indirect conv2d heuristic for selecting the block size Resolves COMPMID-5713 Change-Id: If6ad49124561207153685c6abd4f54950a376fbc Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8886 Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gunes Bayir <gunes.bayir@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2022-12-29Extend Transposed Conv. for tiles with N0>1Gunes Bayir
Partially Resolves: COMPMID-5724 Signed-off-by: Gunes Bayir <gunes.bayir@arm.com> Change-Id: I0aeddddcdd87c8c79f6dae9a76ffdc2ba0c08e17 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8883 Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2022-12-23Make CLReshape kernel window based on dst instead of srcRamy Elgammal
Signed-off-by: Ramy Elgammal <ramy.elgammal@arm.com> Partially-Resolves: COMPMID-5522 Change-Id: I1d90003079c3f24d081cc49f7b110eda753f6995 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8838 Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2022-12-14Optimize Transposed Convolution for CL backend (Quantized)Gunes Bayir
This patch optimizes transposed convolution for QASYMM and QASYMM8_SIGNED types, by extending the transposed convolution kernel written for FP32/16. Resolves: COMPMID-5723 Change-Id: Iab8f09231938adb949c506fd915ed45b885e5c7c Signed-off-by: Gunes Bayir <gunes.bayir@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8792 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2022-12-13Add CLAMP operator to Dynamic Fusion interfaceJakub Sujak
Add the CLAMP activation function for GPU backend with generic activation Component and TemplateWriter modules. CLAMP is internally implemented as LU_BOUNDED_RELU activation function with the alpha and beta variables swapped. We do NOT consider in-place computation cases in this patch. * CLAMP operator for GPU backend. * Activation Component and TemplateWriter for CL backend. * TemplateWriter generates tiled kernel code. * Supported data types: F16, F32. * Validation tests for CLAMP operation. Resolves: COMPMID-5519 Change-Id: Ieb097d6b1e6a7ed2b882518e88314454efb402f6 Signed-off-by: Jakub Sujak <jakub.sujak@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8762 Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gunes Bayir <gunes.bayir@arm.com> Reviewed-by: SiCong Li <sicong.li@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2022-12-12Fix build error resulting from incorrect header pathJakub Sujak
Resolves: COMPMID-5735 Signed-off-by: Jakub Sujak <jakub.sujak@arm.com> Change-Id: I3a3a3b103993be8c3413b55998b36df429be8260 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8780 Reviewed-by: Gunes Bayir <gunes.bayir@arm.com> Reviewed-by: Mohmun02 <MohammedSuhail.Munshi@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com>
2022-12-09Use heuristics for setting dynamic fusion direct conv2d tile sizesRamy Elgammal
Resolves: COMPMID-5735 Change-Id: I9958413b69c5052cfa205dd0e9457cc4953aaf35 Signed-off-by: Ramy Elgammal <ramy.elgammal@arm.com> Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/c/VisualCompute/ComputeLibrary/+/474818 Tested-by: bsgcomp <bsgcomp@arm.com> Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Comments-Addressed: bsgcomp <bsgcomp@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8724 Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2022-12-09Implement the OpenCL kernel to compute the indirect convolutionGian Marco Iodice
- Implement indirect convolution kernel - Add operator support - Add test Resolves COMPMID-5709 Change-Id: I9272304163471a5a40da7fdec204599f3c1d8e32 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8701 Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gunes Bayir <gunes.bayir@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2022-11-25Implement address precalculation for indirect conv2d - OpenCLGian Marco Iodice
- Implement kernel (ClIndirectConv2dAddressPrecalculationKernel) - Implement OpenCL kernel (indirect_convolution.cl) - Add test Resolves COMPMID-5708 Change-Id: If7408e37cbc6f9ad8506ff3334bc574e5d6763fb Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8661 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2022-11-22Remove dynamic fusion prototype with tests and examplesSiCong Li
Public headers of the new experimental dynamic fusion can be found in arm_compute/dynamic_fusion/ New examples on how to use the interface can be found in tests/validation/dynamic_fusion/gpu/Integration.cpp Resolves COMPMID-5683 Change-Id: I7ccb902a227fb487562df15fc3c30118d1d95bbd Signed-off-by: SiCong Li <sicong.li@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8671 Reviewed-by: Jakub Sujak <jakub.sujak@arm.com> Reviewed-by: Gunes Bayir <gunes.bayir@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com>
2022-11-14Optimize Transposed Convolution for CL backend (FP32/16)Gunes Bayir
This patch optimizes transposed convolution for CL backend by rewriting it in a single kernel instead of three (flip_kernel + upsample + conv). The new kernel skips the upsampling step which reduces the input space of convolution by stride_x * stride_y, resulting in significant performance improvement. It also skips the kernel flipping by traversing the weights accordingly, thus reduces the memory footprint. Resolves: COMPMID-5676 Signed-off-by: Gunes Bayir <gunes.bayir@arm.com> Change-Id: I8a333212dc7c5f7f0597aa58b0d56d44814baa14 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8588 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2022-11-01Rework direct convolution heuristic on OpenCLGian Marco Iodice
Resolves COMPMID-5634 Change-Id: I075de70d509d0c4430b4bcf3f218384e237a3a56 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/c/VisualCompute/ComputeLibrary/+/453708 Tested-by: bsgcomp <bsgcomp@arm.com> Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com> Comments-Addressed: bsgcomp <bsgcomp@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8473 Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
2022-10-06Rework DepthwiseConvolution heuristic on OpenCLGian Marco Iodice
Resolves COMPMID-5632 Change-Id: I2bdbe69a610ca2510fbd74d5d412842679299762 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8365 Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com> Reviewed-by: Jakub Sujak <jakub.sujak@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2022-10-06Improve start-up time in gemmlowp reshaped rhs only.Adnan AlSinan
- Does not Pass RESULT_MULTIPLIER, RESULT_SHIFT When PER_CHANNEL_QUANTIZATION is defined. Resolves COMPMID-5499 Signed-off-by: Adnan AlSinan <adnan.alsinan@arm.com> Change-Id: Ie433eaf83c003a6d5ccfeb89eb2783528dc2b48e Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8316 Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2022-10-04Update GEMM reshaped rhs only heuristicGian Marco Iodice
Resolves COMPMID-5631 Change-Id: I37d1d0d043f8d44d782d2225091af607ad131b58 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8364 Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com>
2022-10-03Force CL kernel compilation with 64 registersViet-Hoa Do
* For DDK version 30 and higher, force the CL compiler to use 64 registers for NHWC direct convolution. Resolves: COMPMID-5508 Signed-off-by: Viet-Hoa Do <viet-hoa.do@arm.com> Change-Id: I7d9ecc3b5a4eceaff44542cd26f6f05e30ab2c1f Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8351 Benchmark: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Pablo Marquez Tello <pablo.tello@arm.com>
2022-09-16Fix validation in validate_image2d_support_on_rhsGian Marco Iodice
Resolves COMPMID-5533 Change-Id: Ice3d9469c7486a700c58fb61fc692b13f368d202 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8148 Reviewed-by: Pablo Marquez Tello <pablo.tello@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2022-09-09Rework heuristic in ClConv2dGian Marco Iodice
The heuristic has been tweaked to call direct convolution when we think it can be faster than gemm-based convolution. The main change is affecting the selection of the convolution method on the first layer. In general, the question we should ask for the first convolution layer of a model is: when the execution time of im2col + gemm < direct?. Since im2col does not depend on the OFM, it means that when OFM is big enough, the contribution of im2col is small and the GEMM approach is preferable. From internal experiments, the OFM threshold is 64. Resolves COMPMID-5504, COMPMID-5504, COMPMID-5477 Change-Id: If1bd1fa93c185ffa874388e29866244e62ca3494 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8231 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gunes Bayir <gunes.bayir@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2022-09-09Add a macro guard in all OpenCL kernels in gemmlowp.clGian Marco Iodice
Resolves COMPMID-5498 Change-Id: I474f3f963257014255d082aab0ccbe3efe5aa067 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8222 Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Ramy Elgammal <ramy.elgammal@arm.com> Reviewed-by: Ramy Elgammal <ramy.elgammal@arm.com> Reviewed-by: Gunes Bayir <gunes.bayir@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2022-09-02Enable Winograd-based conv2d when IFM>=8 on GpuGian Marco Iodice
From an internal performance evaluation, it seems that Winograd-based Conv2D offers better performance than alternative methods such as direct convolution and gemm-based conv already from IFM=8. Before the condition was for IFM>=16 Resolves COMPMID-5532 Change-Id: I9ff04835d6fd07f5f0abeec9645c9d9cc913b6b7 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8147 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gunes Bayir <gunes.bayir@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2022-08-17Revert "Fix performance regression in ClConv2D"Ramy Elgammal
- Reverting commit e54d8c07e75d70baeb80fecbb43088027ea45658 because it has caused unexpected regressions. Resolves: COMPMID-5504 Signed-off-by: Ramy Elgammal <ramy.elgammal@arm.com> Change-Id: I2a0bcc6a311009a81f20a146079758ad138fff5b Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8092 Benchmark: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gunes Bayir <gunes.bayir@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com>
2022-08-16Fix performance regression in ClConv2DGian Marco Iodice
- Call gemm-based convolution when the kernel is large and the stride is unit Resolves: COMPMID-5504 Change-Id: I8dd83bd012000ed76824b96a8e37c98c861c59e4 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8081 Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Adnan AlSinan <adnan.alsinan@arm.com> Reviewed-by: Gunes Bayir <gunes.bayir@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2022-08-11Fix performance regression in Conv2D on OpenCLAdnan AlSinan
- Affect CL backend. - For FP32 datatype, affect different platforms. Resolves COMPMID-5479 Signed-off-by: Adnan AlSinan <adnan.alsinan@arm.com> Change-Id: I705d718bc9b7def218034958f7ef86f2c2abe06d Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8064 Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2022-08-11Disable unsafe FP optimizations in Winograd Output TransformGunes Bayir
The unsafe FP optimizations flag causes accuracy issues when certain conditions are met regarding the hardware type, data type and the activation function. Resolves: COMPMID-5375 Change-Id: I1b0b06549b8c108617962d006a20dd263d5e3c21 Signed-off-by: Gunes Bayir <gunes.bayir@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8061 Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2022-08-05Fix LeNet-f16 convolution regressionAdnan AlSinan
Resolves COMPMID-5420 Signed-off-by: Adnan AlSinan <adnan.alsinan@arm.com> Change-Id: I24dca916f49f82e7e5ec809500ae5fe32c8adc97 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8020 Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: SiCong Li <sicong.li@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2022-07-22Add GemmLowp MMUL Reshaped Only Rhs Support for QASYMM8/QASYMM8_SIGNEDFreddie Liardet
This patch introduces a GEMMLowp routine that is optimized for Arm(R) Mali(TM)-G715 and Arm(R) Mali(TM)-G615 Resolves: COMPMID-5398 Signed-off-by: Freddie Liardet <frederick.liardet@arm.com> Signed-off-by: Gunes Bayir <gunes.bayir@arm.com> Change-Id: I8d06453645688f3658b6c7c06f1ebc25a2505661 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7932 Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: SiCong Li <sicong.li@arm.com> Reviewed-by: Pablo Marquez Tello <pablo.tello@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2022-07-22Update ClConv2D heuristic to use direct convolutionAdnan AlSinan
Resolves COMPMID-5298 Change-Id: I4a7d788bc1f5f568bedcc22e7aca47ede6de71bf Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Signed-off-by: Adnan AlSinan <adnan.alsinan@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7891 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gunes Bayir <gunes.bayir@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
generated by cgit v1.2.1 (git 2.23.0) at 2024-06-28 10:58:30 +0000