File tree 2 files changed +3
-3
lines changed
2 files changed +3
-3
lines changed Original file line number Diff line number Diff line change @@ -269,7 +269,7 @@ template <> class Vectorized<float> {
269
269
// Set to +inf where x > ln(FLT_MAX)
270
270
result = _mm256_blendv_ps (result, vec_inf, mask_too_large);
271
271
// final interpretation to float
272
- return result;
272
+ return result;
273
273
}
274
274
275
275
Vectorized<float > exp_u20 () const {
Original file line number Diff line number Diff line change @@ -286,15 +286,15 @@ template <> class Vectorized<float> {
286
286
auto tmp = _mm512_fmadd_ps (vec_a, vec_src, vec_b);
287
287
// headache bis - we loose precision with the cast but it "fits", but ok after f32 -> f16 later
288
288
__m512i casted_integer = _mm512_cvttps_epi32 (tmp);
289
- // boundary condition, lower than the min -> 0
289
+ // boundary condition, lower than the min -> 0
290
290
casted_integer = _mm512_mask_mov_epi32 (casted_integer, min_mask, vec_zero);
291
291
// boundary condition, larger than the max -> +oo
292
292
casted_integer = _mm512_mask_mov_epi32 (casted_integer, max_mask, vec_infinity);
293
293
// final interpretation to float
294
294
return _mm512_castsi512_ps (casted_integer);
295
295
}
296
296
Vectorized<float > exp_u20 () const {
297
- // A faster version o exp with ULP=20
297
+ // A faster version of exp with ULP=20
298
298
const __m512 vec_factorial_1 =
299
299
_mm512_set1_ps (0 .999999701f ); // 1/factorial(1)
300
300
const __m512 vec_factorial_2 =
You can’t perform that action at this time.
0 commit comments