Skip to content

Commit 80346b8

Browse files
authored
Merge pull request #3379 from martin-frbg/issue3369-2
Add casts to fix compiler warnings for SkylakeX sasum/dasum
2 parents 13182b2 + ce036a2 commit 80346b8

File tree

2 files changed

+6
-6
lines changed

2 files changed

+6
-6
lines changed

kernel/x86_64/dasum_microk_skylakex-2.c

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -58,10 +58,10 @@ static FLOAT dasum_kernel(BLASLONG n, FLOAT *x1)
5858

5959
__m128i abs_mask2 = _mm_set1_epi64x(0x7fffffffffffffff);
6060
for (i = tail_index_AVX512; i < tail_index_SSE; i += 8) {
61-
accum_20 += (__m128d)_mm_and_si128(_mm_loadu_si128(&x1[i + 0]), abs_mask2);
62-
accum_21 += (__m128d)_mm_and_si128(_mm_loadu_si128(&x1[i + 2]), abs_mask2);
63-
accum_22 += (__m128d)_mm_and_si128(_mm_loadu_si128(&x1[i + 4]), abs_mask2);
64-
accum_23 += (__m128d)_mm_and_si128(_mm_loadu_si128(&x1[i + 6]), abs_mask2);
61+
accum_20 += (__m128d)_mm_and_si128(_mm_loadu_si128((__m128i*)&x1[i + 0]), abs_mask2);
62+
accum_21 += (__m128d)_mm_and_si128(_mm_loadu_si128((__m128i*)&x1[i + 2]), abs_mask2);
63+
accum_22 += (__m128d)_mm_and_si128(_mm_loadu_si128((__m128i*)&x1[i + 4]), abs_mask2);
64+
accum_23 += (__m128d)_mm_and_si128(_mm_loadu_si128((__m128i*)&x1[i + 6]), abs_mask2);
6565
}
6666

6767
accum_20 = accum_20 + accum_21 + accum_22 + accum_23;

kernel/x86_64/sasum_microk_skylakex-2.c

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -53,8 +53,8 @@ static FLOAT sasum_kernel(BLASLONG n, FLOAT *x1)
5353

5454
__m128i abs_mask2 = _mm_set1_epi32(0x7fffffff);
5555
for (i = tail_index_AVX512; i < tail_index_SSE; i += 8) {
56-
accum_20 += (__m128)_mm_and_si128(_mm_loadu_si128(&x1[i + 0]), abs_mask2);
57-
accum_21 += (__m128)_mm_and_si128(_mm_loadu_si128(&x1[i + 4]), abs_mask2);
56+
accum_20 += (__m128)_mm_and_si128(_mm_loadu_si128((__m128i*)&x1[i + 0]), abs_mask2);
57+
accum_21 += (__m128)_mm_and_si128(_mm_loadu_si128((__m128i*)&x1[i + 4]), abs_mask2);
5858
}
5959

6060
accum_20 += accum_21;

0 commit comments

Comments
 (0)