diff options
| author | Michael Tyler <michael.tyler@arm.com> | 2023-07-07 12:01:32 +0100 |
|---|---|---|
| committer | michael.tyler <michael.tyler@arm.com> | 2023-07-13 12:21:29 +0000 |
| commit | 4c30de056afe8680b42723b26a2241811715b989 (patch) | |
| tree | 4f522a816a5ea1b58b51226eb685c786096f30e3 /src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_s8qs_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp | |
| parent | c8e1617807ef1985a39d8f8f5f69c113b758494d (diff) | |
| download | ComputeLibrary-4c30de056afe8680b42723b26a2241811715b989.tar.gz | |
Enable premultiplication for depthwise convolution
with fp16 and quantized types
Resolves: COMPMID-6337
Change-Id: I81542e51c9c0329f202ac8452f173b138e51a0f6
Signed-off-by: Michael Tyler <michael.tyler@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9883
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_s8qs_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_s8qs_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_s8qs_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp b/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_s8qs_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp index fc83aaf5d2..bd2941b3d6 100644 --- a/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_s8qs_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp +++ b/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_s8qs_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp @@ -64,7 +64,7 @@ class a64_s8qs_nhwc_3x3_s1_output2x2_dot_depthfirst : public DepthwiseDepthfirst ) const override { interleave_a64_s8q_3x3_dot::pack_parameters( - args.input_channels, buffer, reinterpret_cast<const int32_t *>(biases), + args.input_channels * args.channel_multiplier, buffer, reinterpret_cast<const int32_t *>(biases), reinterpret_cast<const int8_t *>(weights), qp, ld_weight_col, ld_weight_row ); } |