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/depthwise_fp32.cpp | |
| 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/depthwise_fp32.cpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_conv/depthwise/depthwise_fp32.cpp | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/src/core/NEON/kernels/arm_conv/depthwise/depthwise_fp32.cpp b/src/core/NEON/kernels/arm_conv/depthwise/depthwise_fp32.cpp index b0f606332b..760328f3ba 100644 --- a/src/core/NEON/kernels/arm_conv/depthwise/depthwise_fp32.cpp +++ b/src/core/NEON/kernels/arm_conv/depthwise/depthwise_fp32.cpp @@ -116,7 +116,7 @@ namespace { if (args.channel_multiplier > 1 && !prefer_premultiply(args)) { - return UINT32_MAX; + return std::numeric_limits<unsigned int>::max(); } // First-pass: compute the number of output pixels which will be computed. @@ -154,7 +154,7 @@ namespace unsigned int multiplier_cycle_estimate(const DepthwiseArgs &args, const Nothing &) { - return prefer_premultiply(args)? UINT32_MAX : 0; + return prefer_premultiply(args)? std::numeric_limits<unsigned int>::max() : 0; } unsigned int not_preferred(const DepthwiseArgs &, const Nothing &) |