Browse Source

Merge pull request #3379 from martin-frbg/issue3369-2

Add casts to fix compiler warnings for SkylakeX sasum/dasum
tags/v0.3.18
Martin Kroeker GitHub 4 years ago
parent
commit
80346b8813
No known key found for this signature in database GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 6 additions and 6 deletions
  1. +4
    -4
      kernel/x86_64/dasum_microk_skylakex-2.c
  2. +2
    -2
      kernel/x86_64/sasum_microk_skylakex-2.c

+ 4
- 4
kernel/x86_64/dasum_microk_skylakex-2.c View File

@@ -58,10 +58,10 @@ static FLOAT dasum_kernel(BLASLONG n, FLOAT *x1)

__m128i abs_mask2 = _mm_set1_epi64x(0x7fffffffffffffff);
for (i = tail_index_AVX512; i < tail_index_SSE; i += 8) {
accum_20 += (__m128d)_mm_and_si128(_mm_loadu_si128(&x1[i + 0]), abs_mask2);
accum_21 += (__m128d)_mm_and_si128(_mm_loadu_si128(&x1[i + 2]), abs_mask2);
accum_22 += (__m128d)_mm_and_si128(_mm_loadu_si128(&x1[i + 4]), abs_mask2);
accum_23 += (__m128d)_mm_and_si128(_mm_loadu_si128(&x1[i + 6]), abs_mask2);
accum_20 += (__m128d)_mm_and_si128(_mm_loadu_si128((__m128i*)&x1[i + 0]), abs_mask2);
accum_21 += (__m128d)_mm_and_si128(_mm_loadu_si128((__m128i*)&x1[i + 2]), abs_mask2);
accum_22 += (__m128d)_mm_and_si128(_mm_loadu_si128((__m128i*)&x1[i + 4]), abs_mask2);
accum_23 += (__m128d)_mm_and_si128(_mm_loadu_si128((__m128i*)&x1[i + 6]), abs_mask2);
}

accum_20 = accum_20 + accum_21 + accum_22 + accum_23;


+ 2
- 2
kernel/x86_64/sasum_microk_skylakex-2.c View File

@@ -53,8 +53,8 @@ static FLOAT sasum_kernel(BLASLONG n, FLOAT *x1)

__m128i abs_mask2 = _mm_set1_epi32(0x7fffffff);
for (i = tail_index_AVX512; i < tail_index_SSE; i += 8) {
accum_20 += (__m128)_mm_and_si128(_mm_loadu_si128(&x1[i + 0]), abs_mask2);
accum_21 += (__m128)_mm_and_si128(_mm_loadu_si128(&x1[i + 4]), abs_mask2);
accum_20 += (__m128)_mm_and_si128(_mm_loadu_si128((__m128i*)&x1[i + 0]), abs_mask2);
accum_21 += (__m128)_mm_and_si128(_mm_loadu_si128((__m128i*)&x1[i + 4]), abs_mask2);
}
accum_20 += accum_21;


Loading…
Cancel
Save