diff options
Diffstat (limited to 'src/core/CL')
| -rw-r--r-- | src/core/CL/kernels/CLDirectConvolutionLayerKernel.cpp | 4 | ||||
| -rw-r--r-- | src/core/CL/kernels/CLGEMMMatrixAccumulateBiasesKernel.cpp | 2 | ||||
| -rw-r--r-- | src/core/CL/kernels/CLIm2ColKernel.cpp | 2 |
3 files changed, 4 insertions, 4 deletions
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<Status, Window> 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<Status, Window> 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; } |