fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-25 19:30:11 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions

nir/lower_int64: Fix [iu]mul_high handling

Browse source 
e551040c60, which added a new mechanism for 64-bit imul which is more
efficient on BDW and later Intel hardware also introduced a bug where we
weren't properly walking both X and Y.  No idea how testing didn't find
this.

Fixes: e551040c60 ("nir/glsl: Add another way of doing lower_imul64 for gen8+"
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6306
Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15829>
This commit is contained in:
Jason Ekstrand 2022-04-08 15:06:11 -05:00 • committed by Marge Bot
parent 48ae404b42
commit d0ace28790
1 changed files with 2 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
src/compiler/nir/nir_lower_int64.c
View file

@ -455,7 +455,7 @@ lower_mul_high64(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y,
for (unsigned i = 0; i < 4; i++) {
nir_ssa_def *carry = NULL;
for (unsigned j = 0; j < 4; j++) {
/* The maximum values of x32[i] and y32[i] are UINT32_MAX so the
/* The maximum values of x32[i] and y32[j] are UINT32_MAX so the
* maximum value of tmp is UINT32_MAX * UINT32_MAX. The maximum
* value that will fit in tmp is
*
@ -466,7 +466,7 @@ lower_mul_high64(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y,
* so we're guaranteed that we can add in two more 32-bit values
* without overflowing tmp.
*/
nir_ssa_def *tmp = nir_umul_2x32_64(b, x32[i], y32[i]);
nir_ssa_def *tmp = nir_umul_2x32_64(b, x32[i], y32[j]);
if (res[i + j])
tmp = nir_iadd(b, tmp, nir_u2u64(b, res[i + j]));