8000 Fix typo and compiler issue. · pytorch/pytorch@5dfa3c6 · GitHub
[go: up one dir, main page]

Skip to content

Commit 5dfa3c6

Browse files
committed
Fix typo and compiler issue.
- retrigger the CI by ammending the commit message due to the modification of 09e8ff9 4days ago.
1 parent 032dcf0 commit 5dfa3c6

File tree

2 files changed

+3
-3
lines changed

2 files changed

+3
-3
lines changed

aten/src/ATen/cpu/vec/vec256/vec256_float.h

+1-1
Original file line numberDiff line numberDiff line change
@@ -269,7 +269,7 @@ template <> class Vectorized<float> {
269269
// Set to +inf where x > ln(FLT_MAX)
270270
result = _mm256_blendv_ps(result, vec_inf, mask_too_large);
271271
// final interpretation to float
272-
return result;
272+
return result;
273273
}
274274

275275
Vectorized<float> exp_u20() const {

aten/src/ATen/cpu/vec/vec512/vec512_float.h

+2-2
Original file line numberDiff line numberDiff line change
@@ -286,15 +286,15 @@ template <> class Vectorized<float> {
286286
auto tmp = _mm512_fmadd_ps(vec_a, vec_src, vec_b);
287287
// headache bis - we loose precision with the cast but it "fits", but ok after f32 -> f16 later
288288
__m512i casted_integer = _mm512_cvttps_epi32(tmp);
289-
// boundary condition, lower than the min -> 0
289+
// boundary condition, lower than the min -> 0
290290
casted_integer = _mm512_mask_mov_epi32(casted_integer, min_mask, vec_zero);
291291
// boundary condition, larger than the max -> +oo
292292
casted_integer = _mm512_mask_mov_epi32(casted_integer, max_mask, vec_infinity);
293293
// final interpretation to float
294294
return _mm512_castsi512_ps(casted_integer);
295295
}
296296
Vectorized<float> exp_u20() const {
297-
// A faster version o exp with ULP=20
297+
// A faster version of exp with ULP=20
298298
const __m512 vec_factorial_1 =
299299
_mm512_set1_ps(0.999999701f); // 1/factorial(1)
300300
const __m512 vec_factorial_2 =

0 commit comments

Comments
 (0)
0