Add casts

This commit is contained in:
Martin Kroeker 2021-09-14 21:41:53 +02:00 committed by GitHub
parent af8843875a
commit ce036a2fc0
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 6 additions and 6 deletions

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;

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;