diff options
| author | Georgios Pinitas <georgios.pinitas@arm.com> | 2017-11-29 14:27:24 +0000 |
|---|---|---|
| committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:41:36 +0000 |
| commit | 6fdfaa856b1eb69d6afbef5727b99756912fc6fa (patch) | |
| tree | 1f1b916fc94a140ec16090a638f9d0531032d9bb /src | |
| parent | 47b5603a0db2797dda66a2b5dbbc451a740a5ecd (diff) | |
| download | ComputeLibrary-6fdfaa856b1eb69d6afbef5727b99756912fc6fa.tar.gz | |
COMPMID-713: Address failures in OCLGrind for CLDirectConvolution
-Changes way of clamping in the kernel side.
-Fills padding with quantized values
Change-Id: I94d17c341fd637fbb24390722162b551b62d16cb
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/111114
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Tested-by: BSG Visual Compute Jenkins server to access repositories on http://mpd-gerrit.cambridge.arm.com <bsgcomp@arm.com>
Diffstat (limited to 'src')
| -rw-r--r-- | src/core/CL/cl_kernels/direct_convolution_1x1_3x3_5x5_quantized.cl | 4 | ||||
| -rw-r--r-- | src/runtime/CL/functions/CLDirectConvolutionLayer.cpp | 8 |
2 files changed, 8 insertions, 4 deletions
diff --git a/src/core/CL/cl_kernels/direct_convolution_1x1_3x3_5x5_quantized.cl b/src/core/CL/cl_kernels/direct_convolution_1x1_3x3_5x5_quantized.cl index cbe826639d..d0cf032ac1 100644 --- a/src/core/CL/cl_kernels/direct_convolution_1x1_3x3_5x5_quantized.cl +++ b/src/core/CL/cl_kernels/direct_convolution_1x1_3x3_5x5_quantized.cl @@ -243,9 +243,7 @@ __kernel void direct_convolution_1x1_3x3_5x5_quantized( pixels0 = ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE(pixels0, output_multiplier, output_shift, 8); pixels0 = pixels0 + output_offset; - pixels0 = max(pixels0, 0); - pixels0 = min(pixels0, 255); - vstore8(convert_uchar8(pixels0), 0, (__global uchar *)dst.ptr); + vstore8(convert_uchar8_sat(pixels0), 0, (__global uchar *)dst.ptr); } #endif // defined(DATA_TYPE) && defined(STRIDE_X) && defined(WEIGHTS_DEPTH) diff --git a/src/runtime/CL/functions/CLDirectConvolutionLayer.cpp b/src/runtime/CL/functions/CLDirectConvolutionLayer.cpp index 8ef7068b88..759692505b 100644 --- a/src/runtime/CL/functions/CLDirectConvolutionLayer.cpp +++ b/src/runtime/CL/functions/CLDirectConvolutionLayer.cpp @@ -23,6 +23,7 @@ */ #include "arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h" +#include "arm_compute/core/CL/ICLTensor.h" #include "arm_compute/core/CL/kernels/CLDirectConvolutionLayerKernel.h" #include "arm_compute/core/PixelValue.h" #include "arm_compute/core/Utils.h" @@ -45,7 +46,12 @@ void CLDirectConvolutionLayer::configure(ICLTensor *input, const ICLTensor *weig _direct_conv_kernel.configure(input, weights, biases, output, conv_info); // Configure border handler - _input_border_handler.configure(input, _direct_conv_kernel.border_size(), BorderMode::CONSTANT, PixelValue(0)); + PixelValue &&zero_value(0.f); + if(is_data_type_quantized_asymmetric(input->info()->data_type())) + { + zero_value = PixelValue(static_cast<uint8_t>(input->info()->quantization_info().offset)); + } + _input_border_handler.configure(input, _direct_conv_kernel.border_size(), BorderMode::CONSTANT, zero_value); } Error CLDirectConvolutionLayer::validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info) |