diff options
author | Anthony Barbier <anthony.barbier@arm.com> | 2018-07-03 16:22:02 +0100 |
---|---|---|
committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:54:10 +0000 |
commit | 5f707736413aeac77818c42838296966f8dc6761 (patch) | |
tree | b829ed3243ea5f3085f288836132416c78bc2e72 /src/core/NEON/kernels/arm_gemm/transforms/a32_transpose_interleave_8way_32bit.hpp | |
parent | 7485d5a62685cb745ab50e970adb722cb71557ac (diff) | |
download | ComputeLibrary-5f707736413aeac77818c42838296966f8dc6761.tar.gz |
COMPMID-1369: Revert accidental formatting of RSH's repo
Pulled latest fixes from David's repo:
commit f43ebe932c84083332b0b1a0348241b69dda63a7
Author: David Mansell <David.Mansell@arm.com>
Date: Tue Jul 3 18:09:01 2018 +0100
Whitespace tidying, fixed comment in gemv_batched imported from ACL.
Change-Id: Ie37a623f44e90d88072236cb853ac55ac82d5f51
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/138530
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-by: David Mansell <david.mansell@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/transforms/a32_transpose_interleave_8way_32bit.hpp')
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/transforms/a32_transpose_interleave_8way_32bit.hpp | 129 |
1 files changed, 70 insertions, 59 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/transforms/a32_transpose_interleave_8way_32bit.hpp b/src/core/NEON/kernels/arm_gemm/transforms/a32_transpose_interleave_8way_32bit.hpp index ea32c9665c..a7e17fa074 100644 --- a/src/core/NEON/kernels/arm_gemm/transforms/a32_transpose_interleave_8way_32bit.hpp +++ b/src/core/NEON/kernels/arm_gemm/transforms/a32_transpose_interleave_8way_32bit.hpp @@ -31,86 +31,97 @@ template <> template <typename T> inline void TransformImpl<8, 1, true, 4, 4>::Transform( - T *out, const T *const in, const int stride, - const int x0, const int xmax, const int k0, const int kmax) -{ - // Redirect to a 16x uint16_t specialisation - TransformImpl<16, 1, true, 2, 2>::Transform( - reinterpret_cast<uint16_t *>(out), - reinterpret_cast<const uint16_t *const>(in), - stride * 2, x0 * 2, xmax * 2, k0, kmax); + T* out, const T* const in, const int stride, + const int x0, const int xmax, const int k0, const int kmax +) { + // Redirect to a 16x uint16_t specialisation + TransformImpl<16, 1, true, 2, 2>::Transform( + reinterpret_cast<uint16_t *>(out), + reinterpret_cast<const uint16_t * const>(in), + stride*2, x0*2, xmax*2, k0, kmax + ); } // Generic 12x16-bit sized specialisation template <> template <typename T> inline void TransformImpl<16, 1, true, 2, 2>::Transform( - T *out, const T *const in, const int stride, - const int x0, const int xmax, const int k0, const int kmax) -{ - // Redirect to a uint16_t specialisation - Transform( - reinterpret_cast<uint16_t *>(out), - reinterpret_cast<const uint16_t *const>(in), - stride, x0, xmax, k0, kmax); + T* out, const T* const in, const int stride, + const int x0, const int xmax, const int k0, const int kmax +) { + // Redirect to a uint16_t specialisation + Transform( + reinterpret_cast<uint16_t *>(out), + reinterpret_cast<const uint16_t * const>(in), + stride, x0, xmax, k0, kmax + ); } // Specialised 16 x uint16_t version template <> -inline void TransposeInterleaveCommon<16, uint16_t, uint16_t>::moveblock_1x1(const uint16_t *&in0, uint16_t *out) -{ - __asm volatile( - "VLD1.32 {d0-d3}, [%[in0]]!\n" - "VST1.32 {d0-d3}, [%[out]]\n" ASM_PREFETCH("[%[in0], #192]") - : [in0] "+r"(in0), - [out] "+r"(out) - : - : "q0", "q1", "memory"); +inline void TransposeInterleaveCommon<16, uint16_t, uint16_t>::moveblock_1x1(const uint16_t *&in0, uint16_t *out) { + __asm volatile ( + "VLD1.32 {d0-d3}, [%[in0]]!\n" + "VST1.32 {d0-d3}, [%[out]]\n" + ASM_PREFETCH("[%[in0], #192]") + : [in0] "+r" (in0), + [out] "+r" (out) + : + : "q0", "q1", "memory" + ); } template <> -inline void TransposeInterleaveCommon<16, uint16_t, uint16_t>::moveblock_1x2(const uint16_t *&in0, const uint16_t *&in1, uint16_t *out) -{ - __asm volatile( - "VLD1.32 {d0-d3}, [%[in0]]!\n" - "VST1.32 {d0-d3}, [%[out]]!\n" ASM_PREFETCH("[%[in0], #192]") - "VLD1.32 {d0-d3}, [%[in1]]!\n" - "VST1.32 {d0-d3}, [%[out]]\n" ASM_PREFETCH("[%[in1], #192]") "SUB %[out], %[out], #32\n" - : [in0] "+r"(in0), - [in1] "+r"(in1), - [out] "+r"(out) - : - : "q0", "q1", "memory"); +inline void TransposeInterleaveCommon<16, uint16_t, uint16_t>::moveblock_1x2(const uint16_t *&in0, const uint16_t *&in1, uint16_t *out) { + __asm volatile ( + "VLD1.32 {d0-d3}, [%[in0]]!\n" + "VST1.32 {d0-d3}, [%[out]]!\n" + ASM_PREFETCH("[%[in0], #192]") + "VLD1.32 {d0-d3}, [%[in1]]!\n" + "VST1.32 {d0-d3}, [%[out]]\n" + ASM_PREFETCH("[%[in1], #192]") + "SUB %[out], %[out], #32\n" + : [in0] "+r" (in0), + [in1] "+r" (in1), + [out] "+r" (out) + : + : "q0", "q1", "memory" + ); } template <> -inline void TransposeInterleaveCommon<16, uint16_t, uint16_t>::moveblock_1x4(const uint16_t *&in0, const uint16_t *&in1, const uint16_t *&in2, const uint16_t *&in3, uint16_t *out) -{ - __asm __volatile( - "VLD1.32 {d0-d3}, [%[in0]]!\n" - "VST1.32 {d0-d3}, [%[out]]!\n" ASM_PREFETCH("[%[in0], #192]") - "VLD1.32 {d0-d3}, [%[in1]]!\n" - "VST1.32 {d0-d3}, [%[out]]!\n" ASM_PREFETCH("[%[in1], #192]") - "VLD1.32 {d0-d3}, [%[in2]]!\n" - "VST1.32 {d0-d3}, [%[out]]!\n" ASM_PREFETCH("[%[in2], #192]") - "VLD1.32 {d0-d3}, [%[in3]]!\n" - "VST1.32 {d0-d3}, [%[out]]\n" ASM_PREFETCH("[%[in3], #192]") "SUB %[out], %[out], #96\n" - : [in0] "+r"(in0), - [in1] "+r"(in1), - [in2] "+r"(in2), - [in3] "+r"(in3), - [out] "+r"(out) - : - : "q0", "q1", "memory"); +inline void TransposeInterleaveCommon<16, uint16_t, uint16_t>::moveblock_1x4(const uint16_t *&in0, const uint16_t *&in1, const uint16_t *&in2, const uint16_t *&in3, uint16_t *out) { + __asm __volatile ( + "VLD1.32 {d0-d3}, [%[in0]]!\n" + "VST1.32 {d0-d3}, [%[out]]!\n" + ASM_PREFETCH("[%[in0], #192]") + "VLD1.32 {d0-d3}, [%[in1]]!\n" + "VST1.32 {d0-d3}, [%[out]]!\n" + ASM_PREFETCH("[%[in1], #192]") + "VLD1.32 {d0-d3}, [%[in2]]!\n" + "VST1.32 {d0-d3}, [%[out]]!\n" + ASM_PREFETCH("[%[in2], #192]") + "VLD1.32 {d0-d3}, [%[in3]]!\n" + "VST1.32 {d0-d3}, [%[out]]\n" + ASM_PREFETCH("[%[in3], #192]") + "SUB %[out], %[out], #96\n" + : [in0] "+r" (in0), + [in1] "+r" (in1), + [in2] "+r" (in2), + [in3] "+r" (in3), + [out] "+r" (out) + : + : "q0", "q1", "memory" + ); } template <> template <> inline void TransformImpl<16, 1, true, 2, 2>::Transform( - uint16_t *out, const uint16_t *const in, const int stride, - const int x0, const int xmax, const int k0, const int kmax) -{ - TransposeInterleaveCommon<16, uint16_t, uint16_t>::Transform(out, in, stride, x0, xmax, k0, kmax); + uint16_t* out, const uint16_t* const in, const int stride, + const int x0, const int xmax, const int k0, const int kmax +) { + TransposeInterleaveCommon<16, uint16_t, uint16_t>::Transform(out, in, stride, x0, xmax, k0, kmax); } #endif // __arm__ |