From b03f7c5c780fe2df23eb8c5c1b4b1d65bd7f0339 Mon Sep 17 00:00:00 2001 From: Georgios Pinitas Date: Thu, 12 Jul 2018 10:49:53 +0100 Subject: COMPMID-1188 : Rename TNOX to Mali-G76 Change-Id: I136f7aa4bca268abd4fbe4f6ce4bcc2708ec3671 Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/139689 Tested-by: Jenkins Reviewed-by: Anthony Barbier --- arm_compute/core/GPUTarget.h | 2 +- src/core/CL/kernels/CLDirectConvolutionLayerKernel.cpp | 4 ++-- src/core/CL/kernels/CLGEMMMatrixAccumulateBiasesKernel.cpp | 2 +- src/core/CL/kernels/CLIm2ColKernel.cpp | 2 +- src/core/GPUTarget.cpp | 6 +++--- src/runtime/CL/functions/CLGEMM.cpp | 2 +- src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp | 2 +- src/runtime/CL/tuners/BifrostTuner.cpp | 12 ++++++------ tests/validation/UNIT/GPUTarget.cpp | 2 +- utils/TypePrinter.h | 4 ++-- 10 files changed, 19 insertions(+), 19 deletions(-) diff --git a/arm_compute/core/GPUTarget.h b/arm_compute/core/GPUTarget.h index 2292a7834c..b7178acb1a 100644 --- a/arm_compute/core/GPUTarget.h +++ b/arm_compute/core/GPUTarget.h @@ -45,7 +45,7 @@ enum class GPUTarget G51 = 0x230, G51BIG = 0x231, G51LIT = 0x232, - TNOX = 0x240, + G76 = 0x240, TTRX = 0x250, TBOX = 0x260 }; diff --git a/src/core/CL/kernels/CLDirectConvolutionLayerKernel.cpp b/src/core/CL/kernels/CLDirectConvolutionLayerKernel.cpp index dcb4ac1c5d..7f7437d6ef 100644 --- a/src/core/CL/kernels/CLDirectConvolutionLayerKernel.cpp +++ b/src/core/CL/kernels/CLDirectConvolutionLayerKernel.cpp @@ -114,7 +114,7 @@ std::pair validate_and_configure_window(ITensorInfo *input, ITen unsigned int num_elems_written_per_iteration_x = 0; unsigned int num_elems_written_per_iteration_y = 0; - if(gpu_target_is_in(target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G51, GPUTarget::G51BIG, GPUTarget::G51LIT, GPUTarget::TNOX) && (kernel_size <= 5) && (conv_stride_x == 1) + if(gpu_target_is_in(target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G51, GPUTarget::G51BIG, GPUTarget::G51LIT, GPUTarget::G76) && (kernel_size <= 5) && (conv_stride_x == 1) && (conv_stride_y == 1) && (data_type == DataType::F32)) { // Configure kernel window @@ -289,7 +289,7 @@ void CLDirectConvolutionLayerKernel::configure(const ICLTensor *input, const ICL CLBuildOptions build_options; build_options.add_option_if(_biases != nullptr, std::string("-DHAS_BIAS")); - if(gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G51, GPUTarget::G51BIG, GPUTarget::G51LIT, GPUTarget::TNOX) && (kernel_size <= 5) && (_conv_stride_x == 1) + if(gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G51, GPUTarget::G51BIG, GPUTarget::G51LIT, GPUTarget::G76) && (kernel_size <= 5) && (_conv_stride_x == 1) && (_conv_stride_y == 1) && (data_type == DataType::F32)) { build_options.add_option(std::string("-DWEIGHTS_DEPTH=" + support::cpp11::to_string(_weights->info()->dimension(2)))); diff --git a/src/core/CL/kernels/CLGEMMMatrixAccumulateBiasesKernel.cpp b/src/core/CL/kernels/CLGEMMMatrixAccumulateBiasesKernel.cpp index 04cf627818..ebe4013bf0 100644 --- a/src/core/CL/kernels/CLGEMMMatrixAccumulateBiasesKernel.cpp +++ b/src/core/CL/kernels/CLGEMMMatrixAccumulateBiasesKernel.cpp @@ -52,7 +52,7 @@ std::pair validate_and_configure_window(ITensorInfo *accum, ITen unsigned int &num_elems_processed_per_iteration) { // Select the vector size to use (8 for Bifrost; 16 for Midgard). - num_elems_processed_per_iteration = gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G51, GPUTarget::G51BIG, GPUTarget::G51LIT, GPUTarget::TNOX) ? 8 : 16; + num_elems_processed_per_iteration = gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G51, GPUTarget::G51BIG, GPUTarget::G51LIT, GPUTarget::G76) ? 8 : 16; // Configure kernel window Window win = calculate_max_window(*accum, Steps(num_elems_processed_per_iteration)); diff --git a/src/core/CL/kernels/CLIm2ColKernel.cpp b/src/core/CL/kernels/CLIm2ColKernel.cpp index b54575ae30..06ca005dd5 100644 --- a/src/core/CL/kernels/CLIm2ColKernel.cpp +++ b/src/core/CL/kernels/CLIm2ColKernel.cpp @@ -219,7 +219,7 @@ CLIm2ColKernel::configure_window(const ICLTensor *input, ICLTensor *output, cons } // Vector size optimized for the 11x11 AlexNet convolution on Bifrost. const GPUTarget gpu_target = get_target(); - if(gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G51, GPUTarget::G51BIG, GPUTarget::G51LIT, GPUTarget::TNOX) && kernel_dims.width == 11) + if(gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G51, GPUTarget::G51BIG, GPUTarget::G51LIT, GPUTarget::G76) && kernel_dims.width == 11) { vector_size = 8; } diff --git a/src/core/GPUTarget.cpp b/src/core/GPUTarget.cpp index 547456533a..a14a9c9971 100644 --- a/src/core/GPUTarget.cpp +++ b/src/core/GPUTarget.cpp @@ -51,9 +51,9 @@ arm_compute::GPUTarget get_bifrost_target(const std::string &version) { return arm_compute::GPUTarget::G51LIT; } - else if(version == "TNOX") + else if(version == "G76") { - return arm_compute::GPUTarget::TNOX; + return arm_compute::GPUTarget::G76; } else if(version == "TTRX") { @@ -106,7 +106,7 @@ const std::string &string_from_target(GPUTarget target) { GPUTarget::G51, "g51" }, { GPUTarget::G51BIG, "g51big" }, { GPUTarget::G51LIT, "g51lit" }, - { GPUTarget::TNOX, "tnox" }, + { GPUTarget::G76, "g76" }, { GPUTarget::TTRX, "ttrx" }, { GPUTarget::TBOX, "tbox" } }; diff --git a/src/runtime/CL/functions/CLGEMM.cpp b/src/runtime/CL/functions/CLGEMM.cpp index bb76872700..1f4df4f1a9 100644 --- a/src/runtime/CL/functions/CLGEMM.cpp +++ b/src/runtime/CL/functions/CLGEMM.cpp @@ -44,7 +44,7 @@ inline bool is_interleaved_transposed(int m, int n, int k, DataType data_type, b { bool flag = true; - if(gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72)) + if(gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G76)) { // COMPMID-852 if(k > 256 && m > 4 && is_data_type_float(data_type) && reshape_b_only_on_first_run) diff --git a/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp b/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp index 94dc0e071c..842ee73397 100644 --- a/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp +++ b/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp @@ -41,7 +41,7 @@ inline bool is_interleaved_transposed(int m, int n, int k, bool reshape_b_only_o { bool flag = true; - if(gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G51, GPUTarget::G51BIG, GPUTarget::G51LIT, GPUTarget::TNOX)) + if(gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G51, GPUTarget::G51BIG, GPUTarget::G51LIT, GPUTarget::G76)) { // COMPMID-852 if(k > 256 && m > 4 && reshape_b_only_on_first_run) diff --git a/src/runtime/CL/tuners/BifrostTuner.cpp b/src/runtime/CL/tuners/BifrostTuner.cpp index edd074ba08..fa67710cc8 100644 --- a/src/runtime/CL/tuners/BifrostTuner.cpp +++ b/src/runtime/CL/tuners/BifrostTuner.cpp @@ -132,7 +132,7 @@ void tune_col2im_kernel(CLCol2ImKernel &k) // Configure the local work size for Bifrost with a value obtained // via exhaustive autotuning over 30 representative tensor shapes. - if(gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G51, GPUTarget::G51BIG, GPUTarget::G51LIT, GPUTarget::TNOX)) + if(gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G51, GPUTarget::G51BIG, GPUTarget::G51LIT, GPUTarget::G76)) { if((k._convolved_dims.first == 7) || (k._convolved_dims.first == 14)) { @@ -153,7 +153,7 @@ void tune_im2col_kernel(CLIm2ColKernel &k) const GPUTarget gpu_target = k.get_target(); // Local work size optimized for the 11x11 AlexNet convolution on Bifrost. - if(gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G51, GPUTarget::G51BIG, GPUTarget::G51LIT, GPUTarget::TNOX) && k._kernel_dims.width == 11) + if(gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G51, GPUTarget::G51BIG, GPUTarget::G51LIT, GPUTarget::G76) && k._kernel_dims.width == 11) { const bool is_square_kernel = (k._kernel_dims.width == k._kernel_dims.height); if(!is_square_kernel && k._kernel_dims.width > 1 && !k._conv_info.has_padding()) @@ -171,7 +171,7 @@ void tune_depthwise_im2col_kernel(CLDepthwiseIm2ColKernel &k) // Configure the local work size for Bifrost with a value obtained // via exhaustive autotuning for the MobileNets tensor shapes. - if(gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G51, GPUTarget::G51BIG, GPUTarget::G51LIT, GPUTarget::TNOX)) + if(gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G51, GPUTarget::G51BIG, GPUTarget::G51LIT, GPUTarget::G76)) { lws_hint = cl::NDRange(1, 2, 1); } @@ -186,7 +186,7 @@ void tune_gemv_kernel(CLGEMMMatrixVectorMultiplyKernel &k) // Configure the local work size for Bifrost with a value obtained // via exhaustive autotuning for the MobileNets tensor shapes. - if(gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G51, GPUTarget::G51BIG, GPUTarget::G51LIT, GPUTarget::TNOX)) + if(gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G51, GPUTarget::G51BIG, GPUTarget::G51LIT, GPUTarget::G76)) { lws_hint = cl::NDRange(1, 1, 1); } @@ -207,7 +207,7 @@ void tune_gemm_kernel(CLGEMMMatrixMultiplyKernel &k) case GPUTarget::G51: case GPUTarget::G51BIG: case GPUTarget::G51LIT: - case GPUTarget::TNOX: + case GPUTarget::G76: if(k._input1->info()->dimension(1) == 24) { // LWS optimized for the 11x11 AlexNet convolution on Bifrost. @@ -240,7 +240,7 @@ void tune_pooling_kernel(CLPoolingLayerKernel &k) // invalid (e.g. exceeds the maximum workgroup size that the kernel can be launched with). if(k._input->info()->data_layout() == DataLayout::NCHW) { - if(gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G51, GPUTarget::G51BIG, GPUTarget::G51LIT, GPUTarget::TNOX)) + if(gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G51, GPUTarget::G51BIG, GPUTarget::G51LIT, GPUTarget::G76)) { cl::NDRange gws = ICLKernel::gws_from_window(k.window()); lws_hint = cl::NDRange(gws[0], gws[1], 1); diff --git a/tests/validation/UNIT/GPUTarget.cpp b/tests/validation/UNIT/GPUTarget.cpp index 62e0b1dbdc..eabfa1fb79 100644 --- a/tests/validation/UNIT/GPUTarget.cpp +++ b/tests/validation/UNIT/GPUTarget.cpp @@ -47,7 +47,7 @@ TEST_CASE(GetGPUTargetFromName, framework::DatasetMode::ALL) ARM_COMPUTE_EXPECT(get_target_from_name("Mali-G51") == GPUTarget::G51, framework::LogLevel::ERRORS); ARM_COMPUTE_EXPECT(get_target_from_name("Mali-G51BIG") == GPUTarget::G51BIG, framework::LogLevel::ERRORS); ARM_COMPUTE_EXPECT(get_target_from_name("Mali-G51LIT") == GPUTarget::G51LIT, framework::LogLevel::ERRORS); - ARM_COMPUTE_EXPECT(get_target_from_name("Mali-TNOX") == GPUTarget::TNOX, framework::LogLevel::ERRORS); + ARM_COMPUTE_EXPECT(get_target_from_name("Mali-G76") == GPUTarget::G76, framework::LogLevel::ERRORS); ARM_COMPUTE_EXPECT(get_target_from_name("Mali-TTRX") == GPUTarget::TTRX, framework::LogLevel::ERRORS); ARM_COMPUTE_EXPECT(get_target_from_name("Mali-TBOX") == GPUTarget::TBOX, framework::LogLevel::ERRORS); ARM_COMPUTE_EXPECT(get_target_from_name("Mali-T000") == GPUTarget::MIDGARD, framework::LogLevel::ERRORS); diff --git a/utils/TypePrinter.h b/utils/TypePrinter.h index 9bd8c2a667..eb03ff9b96 100644 --- a/utils/TypePrinter.h +++ b/utils/TypePrinter.h @@ -1525,8 +1525,8 @@ inline ::std::ostream &operator<<(::std::ostream &os, const GPUTarget &gpu_targe case GPUTarget::G51LIT: os << "G51LIT"; break; - case GPUTarget::TNOX: - os << "TNOX"; + case GPUTarget::G76: + os << "G76"; break; case GPUTarget::TTRX: os << "TTRX"; -- cgit v1.2.1