i965/fs: Don't propagate conditional modifiers from integer compares to adds

No shader-db changes on any Intel platform... which probably explains
why no bugs have been bisected to this problem since it landed in Mesa
18.1. :( The commit mentioned below is in 18.2, so 18.1 would need a
slightly different fix (due to code refactoring).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Fixes: 77f269bb56 "i965/fs: Refactor propagation of conditional modifiers from compares to adds"
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com> (reviewed the original patch)
Cc: Matt Turner <mattst88@gmail.com> (reviewed the original patch)
This commit is contained in:
Ian Romanick
2018-09-12 17:16:50 -07:00
parent 0dd8189f15
commit df9dbc03d3

View File

@@ -211,9 +211,17 @@ opt_cmod_propagation_local(const gen_device_info *devinfo, bblock_t *block)
/* A CMP with a second source of zero can match with anything. A CMP
* with a second source that is not zero can only match with an ADD
* instruction.
*
* Only apply this optimization to float-point sources. It can fail for
* integers. For inputs a = 0x80000000, b = 4, int(0x80000000) < 4, but
* int(0x80000000) - 4 overflows and results in 0x7ffffffc. that's not
* less than zero, so the flags get set differently than for (a < b).
*/
if (inst->opcode == BRW_OPCODE_CMP && !inst->src[1].is_zero()) {
progress = cmod_propagate_cmp_to_add(devinfo, block, inst) || progress;
if (brw_reg_type_is_floating_point(inst->src[0].type) &&
cmod_propagate_cmp_to_add(devinfo, block, inst))
progress = true;
continue;
}