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 --- .../kernels/arm_gemm/transforms/a64_transpose_interleave_128.hpp | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-) (limited to 'src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_128.hpp') diff --git a/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_128.hpp b/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_128.hpp index e6186984e8..8574d89226 100644 --- a/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_128.hpp +++ b/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_128.hpp @@ -24,7 +24,7 @@ #pragma once -#ifdef __aarch64__ +#if defined(__aarch64__) namespace { @@ -193,7 +193,6 @@ void a64_transpose_interleave_128(uint32_t *out, const uint32_t *in, size_t widt "bge 1b\n" "cbz %x[height], 20f\n" "10:" // Main loop skip - "11:" // Tail row loop: Head "mov x20, %x[width]\n" "mov x25, %x[in]\n" @@ -264,7 +263,6 @@ void a64_transpose_interleave_128(uint32_t *out, const uint32_t *in, size_t widt "add %x[out], %x[out], #0x80\n" "bge 11b\n" "20:" // Done - : [height] "+&r" (height), [in] "+&r" (in), [out] "+&r" (out) : [in_stride] "r" (in_stride), [out_stride] "r" (out_stride), [width] "r" (width) : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x20", "x21", "x22", "x23", "x24", "x25" @@ -286,4 +284,5 @@ void Transform<32, 1, true, VLType::None>( ); } -#endif + +#endif // defined(__aarch64__) -- cgit v1.2.1