aboutsummaryrefslogtreecommitdiff
path: root/src/core/CL/cl_kernels/nhwc
AgeCommit message (Collapse)Author
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>
2022-11-14Optimize 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>
2022-10-07Workaround CL compiler issue on FP16Viet-Hoa Do
Resolves: COMPMID-5600 Signed-off-by: Viet-Hoa Do <viet-hoa.do@arm.com> Change-Id: I5196d1639c48d0b8a116d47ed1d6c7334dc8f41e Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8374 Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Jakub Sujak <jakub.sujak@arm.com> Reviewed-by: Pablo Marquez Tello <pablo.tello@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@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>
2022-09-07Optimize depthwise convolution on OpenCLGian Marco Iodice
The optimization concerns the case where the depth multiplier is > 1. The depth multiplier for loop has been removed from the OpenCL kernel and the GWS has been mapped to the output shape. In this way, we can still perform a tile with N0 columns and improve the performance of depthwise conv over 80% when depth multiplier is > 1. Resolves COMPMID-5568 Change-Id: I604e287d4eeb31c54b9cc6c3072a698cd0e3e136 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8184 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gunes Bayir <gunes.bayir@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2022-07-21Fix direct convolution cases that were failing on OdroidAdnan AlSinan
- Affects OpenCL backend. - Resolves COMPMID-5416 Signed-off-by: Adnan AlSinan <adnan.alsinan@arm.com> Change-Id: I8953f9ac5c1ec9edf99399a651a544df4276ccf1 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7951 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>
2022-06-27Implement new Elementwise Dynamic Fusion Operators: Div, FloorMichalis Spyrou
Resolves: COMPMID-5355 Change-Id: I92f73fbe885f28bbe7b07965b90cfd807c93602f Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7745 Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: SiCong Li <sicong.li@arm.com>
2022-06-15Fix performance regression in Winograd Output Transform (OpenCL)Gian Marco Iodice
The regression was caused by NUM_TILES_X passed at runtime. Resolves COMPMID-5327 Change-Id: Id6ccd93784eda93af09f420c0d786050e2bbccd7 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7727 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
2022-04-19Add CLPool3d Int8 SupportMohammed Suhail Munshi
- Adds Qasymm8 and Qasymm8_signed support to the 3d pool operator Resolves: COMPMID-4669 Signed-off-by: Mohammed Suhail Munshi <MohammedSuhail.Munshi@arm.com> Change-Id: I36038c2b7c4f36baf67f7aae801356890e104538 Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/c/VisualCompute/ComputeLibrary/+/410496 Tested-by: bsgcomp <bsgcomp@arm.com> Reviewed-by: Sheri Zhang <sheri.zhang@arm.com> Comments-Addressed: bsgcomp <bsgcomp@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7391 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gunes Bayir <gunes.bayir@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2022-03-15Implementation of ClPooling3dramelg01
- For NDHWC layout - For F16 and F32 data types - Mixed Precision stil not supported Resolves: COMPMID-4670 Signed-off-by: ramy.elgammal@arm.com Change-Id: I0e14a13e4625569e8e5ee67e6033bd1efe0da469 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7262 Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: SiCong Li <sicong.li@arm.com> Reviewed-by: Gunes Bayir <gunes.bayir@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com>
2022-02-10Improve start-up time for winograd_output_transform_*_nhwcramelg01
- pass tensor's dimensions at runtime rather than compile time - Add guard macro to compile only kernel of internest Resolves: COMPMID-5120 Signed-off-by: Ramy Elgammal <ramy.elgammal@arm.com> Change-Id: I87c3b56ce0cd3c97ffdeabdd9c5d433f361bb005 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7101 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2022-02-09Remove deprecated remap functions.Adnan AlSinan
- Remove CLRemapKernel. - Remove NERemapKernel. Partially resolves COMPMID-4984 Signed-off-by: Adnan AlSinan <adnan.alsinan@arm.com> Change-Id: Ia61f9ac7447695d81178701cf0e9b7625a91eccc Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7056 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Pablo Marquez Tello <pablo.tello@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2022-02-09Improve start-up time for winograd_input_transform_*_nhwcramelg01
- pass tensor's dimensions at runtime rather than compile time - Add guard macro to compile only kernel(s) of internest Resolves: COMPMID-5119 Signed-off-by: Ramy Elgammal <ramy.elgammal@arm.com> Change-Id: Ib01098e397011a1201c2800c62a8954ec70e63e8 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7083 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Giorgio Arena <giorgio.arena@arm.com> Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2022-02-08Improve start-up time for winograd_filter_transform_*_nhwcramelg01
- pass tensor's dimensions at runtime rather than compile time - Add guard macro to compile only kernel of internest Resolves: COMPMID-5118 Signed-off-by: Ramy Elgammal <ramy.elgammal@arm.com> Change-Id: Ie42c3c07fdd817ce62e7cad354381bc22c6e9264 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7058 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2021-12-10Use #if directive instead of regular condition in CLDirectConv2DGiorgio Arena
Resolve COMPMID-5004 Signed-off-by: Giorgio Arena <giorgio.arena@arm.com> Change-Id: Ib3e1b5a891234316c411ea9825ec10c68c4ab5a3 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6788 Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Sheri Zhang <sheri.zhang@arm.com>
2021-12-01Improve start-up direct convolution on OpenCLGian Marco Iodice
- Pass arguments at runtime - Rework ClConv2D heuristic to select direct convolution when OFM < IFM also for small kernel sizes Resolves COMPMID-5000 Change-Id: I9b538e29093829bc366d24d1e904341c247fa22b Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6771 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Giorgio Arena <giorgio.arena@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2021-11-29Use loop unrolling only when the kernel height is less than 5Gian Marco Iodice
- In the dwc_native_fp_nhwc.cl, loop unrolling should only be enabled when kernel height is less than 5. - No performance regression experimented - The patch reduces the compilation time required for the kernel Resolves COMPMID-4887 Change-Id: I93188b9764cf7d1ad34ac164694f6f1fd37a90e8 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6744 Reviewed-by: Giorgio Arena <giorgio.arena@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com>
2021-11-17Improve start-up timer for ClIm2ColGiorgio Arena
Resolve COMPMID-4889 Signed-off-by: Giorgio Arena <giorgio.arena@arm.com> Change-Id: I4a88082b13865fdaeaba1b7216503cd640aa54df Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6680 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2021-11-17Improve start-up time for depthwise convolutionSheri Zhang
- Pass source and destination tensor dimension info at runtime Resolves: COMPMID-4887 Signed-off-by: Sheri Zhang <sheri.zhang@arm.com> Change-Id: Ib7c9f3ce6fb7cef600f7b0cd0fadafa4fa6888a1 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6635 Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
2021-11-09Improve start-up time for ClScaleAdnan AlSinan
- Add macro guard for different kernels in scale.cl - Rework TENSOR4D to the new format - Pass scale_x and scale_y at runtime Resolves COMPMID-4886 Signed-off-by: Adnan AlSinan <adnan.alsinan@arm.com> Change-Id: Ib904a703d511fb8260618057ac92e5ea9efeee2b Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6619 Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2021-10-20Implement CLDirectConv3DKernel - uint8/int8Giorgio Arena
Resolve COMPMID-4663 Signed-off-by: Giorgio Arena <giorgio.arena@arm.com> Change-Id: I5c3c1cffed5385c06b789543318f7f4d6096987e Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6468 Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Sheri Zhang <sheri.zhang@arm.com>
2021-10-15Fix CLConv3D filelist and commentsGiorgio Arena
Signed-off-by: Giorgio Arena <giorgio.arena@arm.com> Change-Id: I4d48f1b8eba6681a9de0ae5f1fd8a4ad1edf7fe8 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6439 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2021-10-14Implement CLDirectConv3D f32/f16Giorgio Arena
Resolve COMPMID-4660 Signed-off-by: Giorgio Arena <giorgio.arena@arm.com> Change-Id: Ibd66ec1eb6faa60086981b1e3a9c12561df3445f Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6420 Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
2021-09-14Optimize ClScaleKernel on NHWC (f32/f16/int8)Gian Marco Iodice
The new kernel performs the computation on multiples elements. The OpenCL kernel has been re-implemented using the new TILE macros Resolves COMPMID-4803,COMPMID-4804 Change-Id: Iac8fead65e21b64567a05dbc4fbaa61d362443f9 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6235 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Giorgio Arena <giorgio.arena@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
2021-08-23Remove padding from ClScaleKernelGiorgio Arena
- Merge quantized kernels with fp for bilinear interpolation (both NCHW and NHWC) - Pass dimensions at compile time rather than at run time - Use tile-based approach to rework the NCHW kernels - Remove unused functions/files Resolve COMPMID-4723 Signed-off-by: Giorgio Arena <giorgio.arena@arm.com> Change-Id: Ifcdf02beb9daa9f318395751b3c85eb2fe874082 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6138 Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
2021-07-25Reorganize the kernels into nhwc, nchw and common foldersAdnan AlSinan
The Following kernels have been split into nchw/nhwc kernels files: - batchnormalization_layer - batch_to_space - channel_shuffle - depth_to_space - dequantization_layer - im2col - normalization_layer - normalize_planar_yuv_layer - normalize_planar_yuv_layer_quantized - pooling_layer - pooling_layer_quantized - remap - reorg_layer - scale - scale_quantized - space_to_batch - space_to_depth - upsample_layer - winograd_filter_transform - winograd_input_transform - winograd_output_transform The following kernels have been moved to nchw folder: - direct_convolution1x1 - direct_convolution3x3 - direct_convolution5x5 - direct_convolution_quantized - prior_box_layer The following kernels have been moved to nhwc folder: - direct_convolution - dwc_native_fp_nhwc - dwc_native_quantized_nhwc The following kernels have been removed: - sobel_filter While the rest kerenls have been moved to the common folder. Partially resolves COMPMID-4453 Signed-off-by: Adnan AlSinan <adnan.alsinan@arm.com> Change-Id: Ic327ac935687ec351c610c65a3c6357f364a5a58 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5919 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
generated by cgit v1.2.1 (git 2.23.0) at 2024-08-05 04:34:33 +0000