diff options
author | Georgios Pinitas <georgios.pinitas@arm.com> | 2018-11-07 16:35:35 +0000 |
---|---|---|
committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2018-11-07 16:35:35 +0000 |
commit | d636bc5bdf8b319a5c0f301e0c6125c0268b36cf (patch) | |
tree | ca031e04c9ad68ca43208823334dda25ec081c18 /src/core/NEON/kernels/arm_gemm/transforms/a64_interleave_8way_half_to_float.hpp | |
parent | dd2619a777d8faaa17a7cd7c8f20c036903947ad (diff) | |
download | ComputeLibrary-d636bc5bdf8b319a5c0f301e0c6125c0268b36cf.tar.gz |
COMPMID-1451: Fixed zerobuff sizes and clobbers in interleave transforms.
Change-Id: If8fbd04d0817b9e654ffa9715879a2521de66963
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/transforms/a64_interleave_8way_half_to_float.hpp')
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/transforms/a64_interleave_8way_half_to_float.hpp | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/transforms/a64_interleave_8way_half_to_float.hpp b/src/core/NEON/kernels/arm_gemm/transforms/a64_interleave_8way_half_to_float.hpp index 773d56d913..88b40d7c1e 100644 --- a/src/core/NEON/kernels/arm_gemm/transforms/a64_interleave_8way_half_to_float.hpp +++ b/src/core/NEON/kernels/arm_gemm/transforms/a64_interleave_8way_half_to_float.hpp @@ -35,7 +35,7 @@ inline void TransformImpl<8, 1, false, 4, 2, false>::Transform(float *out, const float *outptr = out; const __fp16 *inptr = in; - __fp16 zerobuff[8]; + __fp16 zerobuff[16]; // 8 for asm loop plus up to 7 for overflow loop for (int y=y0; y<ymax; y+=8) { const __fp16 *inptr0 = inptr + y * ldin + k0; @@ -172,7 +172,7 @@ inline void TransformImpl<8, 1, false, 4, 2, false>::Transform(float *out, const [inptr4] "+r" (inptr4), [inptr5] "+r" (inptr5), [inptr6] "+r" (inptr6), [inptr7] "+r" (inptr7), [outptr] "+r" (outptr) : : "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", - "v13", "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23" + "v13", "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "memory" ); } |