diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_24_bf16fp32.hpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_24_bf16fp32.hpp | 7 |
1 files changed, 3 insertions, 4 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_24_bf16fp32.hpp b/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_24_bf16fp32.hpp index 799a9cd91d..dcaf69d2a8 100644 --- a/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_24_bf16fp32.hpp +++ b/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_24_bf16fp32.hpp @@ -24,7 +24,7 @@ #pragma once -#ifdef __aarch64__ +#if defined(__aarch64__) namespace { @@ -198,7 +198,6 @@ void a64_transpose_interleave_24_bf16fp32(float *out, const bfloat16 *in, size_t "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" @@ -271,7 +270,6 @@ void a64_transpose_interleave_24_bf16fp32(float *out, const bfloat16 *in, size_t "add %x[out], %x[out], #0x30\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", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x20", "x21", "x22", "x23", "x24", "x25" @@ -292,4 +290,5 @@ void Transform<12, 1, true, VLType::None>( ); } -#endif + +#endif // defined(__aarch64__) |