From 74921eee924625426429044decefe3673561b174 Mon Sep 17 00:00:00 2001 From: Michael Tyler Date: Wed, 12 Apr 2023 17:43:17 +0100 Subject: Update CPU kernel implementations and guard directives Resolves COMPMID-6023 Change-Id: I868975d14c4f98af6716726feda22405a6a4c891 Signed-off-by: Michael Tyler Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9686 Tested-by: Arm Jenkins Reviewed-by: Viet-Hoa Do Comments-Addressed: Arm Jenkins Benchmark: Arm Jenkins --- .../arm_gemm/transforms/sve_transpose_interleave_6VL_2x4.hpp | 8 +++----- 1 file changed, 3 insertions(+), 5 deletions(-) (limited to 'src/core/NEON/kernels/arm_gemm/transforms/sve_transpose_interleave_6VL_2x4.hpp') diff --git a/src/core/NEON/kernels/arm_gemm/transforms/sve_transpose_interleave_6VL_2x4.hpp b/src/core/NEON/kernels/arm_gemm/transforms/sve_transpose_interleave_6VL_2x4.hpp index 2b5741a49c..910fc6cb02 100644 --- a/src/core/NEON/kernels/arm_gemm/transforms/sve_transpose_interleave_6VL_2x4.hpp +++ b/src/core/NEON/kernels/arm_gemm/transforms/sve_transpose_interleave_6VL_2x4.hpp @@ -24,8 +24,7 @@ #pragma once -#ifdef __ARM_FEATURE_SVE - +#if defined(ARM_COMPUTE_ENABLE_SVE) namespace { @@ -260,7 +259,6 @@ void sve_transpose_interleave_6VL_2x4(uint16_t *out, const uint16_t *in, size_t "bge 1b\n" "cbz %x[height], 12f\n" "6:" // Main loop skip - "7:" // Tail row loop: Head "mov x12, %x[in]\n" "add x11, x12, %x[in_stride]\n" @@ -386,7 +384,6 @@ void sve_transpose_interleave_6VL_2x4(uint16_t *out, const uint16_t *in, size_t "addvl %x[out], %x[out], #6\n" "bge 7b\n" "12:" // Done - : [height] "+&r" (height), [in] "+&r" (in), [out] "+&r" (out) : [in_stride] "r" (in_stride), [out_stride] "r" (out_stride), [pad_row] "r" (pad_row), [width] "r" (width) : "cc", "memory", "p0", "p1", "p2", "x9", "x10", "x11", "x12", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31" @@ -408,4 +405,5 @@ void Transform<6, 4, true, VLType::SVE>( ); } -#endif + +#endif // defined(ARM_COMPUTE_ENABLE_SVE) -- cgit v1.2.1