diff options
author | Omar Al Khatib <omar.alkhatib@arm.com> | 2022-11-01 17:01:24 +0000 |
---|---|---|
committer | Omar Al Khatib <omar.alkhatib@arm.com> | 2022-11-07 15:45:21 +0000 |
commit | 605a928960c0c36384925a9064d12addc8f43a41 (patch) | |
tree | 71896f7d4683751c7c87c5606b95e0c9c421384a /src/core/NEON/wrapper/intrinsics/shr.h | |
parent | a2b131bf4680f83b10a7e3544b6183279d8c2691 (diff) | |
download | ComputeLibrary-605a928960c0c36384925a9064d12addc8f43a41.tar.gz |
Optimize CPU mul layer on quantized data
Resolves : [COMPMID-5461]
Signed-off-by: Omar Al Khatib <omar.alkhatib@arm.com>
Change-Id: I89b99d267c32b00ef44f9bb6e7c714dfe4a0d29d
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8420
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON/wrapper/intrinsics/shr.h')
-rw-r--r-- | src/core/NEON/wrapper/intrinsics/shr.h | 28 |
1 files changed, 22 insertions, 6 deletions
diff --git a/src/core/NEON/wrapper/intrinsics/shr.h b/src/core/NEON/wrapper/intrinsics/shr.h index d740091464..e41e9b8b31 100644 --- a/src/core/NEON/wrapper/intrinsics/shr.h +++ b/src/core/NEON/wrapper/intrinsics/shr.h @@ -25,21 +25,19 @@ #ifndef ARM_COMPUTE_WRAPPER_SHR_H #define ARM_COMPUTE_WRAPPER_SHR_H -#include <type_traits> #include <arm_neon.h> +#include <type_traits> namespace arm_compute { namespace wrapper { - #define VQRSHRN_IMPL(half_vtype, vtype, prefix, postfix) \ template <int b> \ inline half_vtype vqrshrn(const vtype &a) \ { \ return prefix##_##postfix(a, b); \ } - VQRSHRN_IMPL(int8x8_t, int16x8_t, vqrshrn_n, s16) VQRSHRN_IMPL(uint8x8_t, uint16x8_t, vqrshrn_n, u16) VQRSHRN_IMPL(int16x4_t, int32x4_t, vqrshrn_n, s32) @@ -77,20 +75,38 @@ VQRSHRN_SCALAR_IMPL(uint32_t, uint64_t, vqrshrnd_n, u64) { \ return prefix_signed##_##postfix(a, b); \ } \ - \ + \ template <int b, typename T> \ inline typename std::enable_if<std::is_integral<T>::value && !std::is_signed<T>::value, u##half_vtype>::type \ vqrshrn_ex(const vtype &a) \ { \ return prefix_unsigned##_##postfix(a, b); \ } - VQRSHRN_EX_IMPL(int8x8_t, int16x8_t, vqrshrn_n, vqrshrun_n, s16) VQRSHRN_EX_IMPL(int16x4_t, int32x4_t, vqrshrn_n, vqrshrun_n, s32) VQRSHRN_EX_IMPL(int32x2_t, int64x2_t, vqrshrn_n, vqrshrun_n, s64) - #undef VQRSHRN_EX_IMPL +#define VSHR_IMPL(vtype, prefix, postfix) \ + template <int b> \ + inline vtype vshr_n(const vtype &a) \ + { \ + return prefix##_##postfix(a, b); \ + } +VSHR_IMPL(uint8x8_t, vshr_n, u8) +VSHR_IMPL(int8x8_t, vshr_n, s8) +#undef VSHR_IMPL + +#define VSHRQ_IMPL(vtype, prefix, postfix) \ + template <int b> \ + inline vtype vshrq_n(const vtype &a) \ + { \ + return prefix##_##postfix(a, b); \ + } +VSHRQ_IMPL(uint32x4_t, vshrq_n, u32) +VSHRQ_IMPL(int32x4_t, vshrq_n, s32) +#undef VSHRQ_IMPL + #ifdef __aarch64__ #define VQRSHRN_EX_SCALAR_IMPL(half_vtype, vtype, prefix_signed, prefix_unsigned, postfix) \ template <int b, typename T> \ |