fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-05 09:38:07 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

aco: don't use uniform continues if exec might be empty

Browse source 
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31143>
This commit is contained in:
Rhys Perry 2024-10-10 14:17:34 +01:00 • committed by Marge Bot
parent aa0ede751d
commit 63b0692eac
2 changed files with 56 additions and 7 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

25
src/amd/compiler/aco_instruction_selection.cpp
View file

@ -10249,6 +10249,14 @@ emit_loop_jump(isel_context* ctx, bool is_break)
append_logical_end(ctx->block);
unsigned idx = ctx->block->index;
/* If exec is empty inside uniform control flow in a loop, we can assume that all invocations
* of the loop are inactive. Breaking from the loop is the right thing to do in that case.
* We shouldn't perform a uniform continue, or else we might never reach a break.
*/
bool potentially_empty_exec = ctx->cf_info.exec.potentially_empty_discard ||
ctx->cf_info.exec.potentially_empty_break ||
ctx->cf_info.exec.potentially_empty_continue;
if (is_break) {
logical_target = ctx->cf_info.parent_loop.exit;
add_logical_edge(idx, logical_target);
@ -10274,7 +10282,7 @@ emit_loop_jump(isel_context* ctx, bool is_break)
add_logical_edge(idx, logical_target);
ctx->block->kind |= block_kind_continue;
if (!ctx->cf_info.parent_if.is_divergent) {
if (!ctx->cf_info.parent_if.is_divergent && !potentially_empty_exec) {
/* uniform continue - directly jump to the loop header */
ctx->block->kind |= block_kind_uniform;
ctx->cf_info.has_branch = true;
@ -10283,14 +10291,17 @@ emit_loop_jump(isel_context* ctx, bool is_break)
return;
}
/* for potential uniform breaks after this continue,
we must ensure that they are handled correctly */
ctx->cf_info.parent_loop.has_divergent_continue = true;
ctx->cf_info.parent_loop.has_divergent_branch = true;
if (!ctx->cf_info.exec.potentially_empty_continue) {
ctx->cf_info.exec.potentially_empty_continue = true;
ctx->cf_info.exec.potentially_empty_continue_depth = ctx->block->loop_nest_depth;
if (ctx->cf_info.parent_if.is_divergent) {
/* for potential uniform breaks after this continue,
we must ensure that they are handled correctly */
ctx->cf_info.parent_loop.has_divergent_continue = true;
if (!ctx->cf_info.exec.potentially_empty_continue) {
ctx->cf_info.exec.potentially_empty_continue = true;
ctx->cf_info.exec.potentially_empty_continue_depth = ctx->block->loop_nest_depth;
}
}
}

38
src/amd/compiler/tests/test_isel.cpp
View file

@ -1741,3 +1741,41 @@ BEGIN_TEST(isel.cf.empty_exec.repair_ssa)
finish_isel_test();
END_TEST
/*
* loop {
* if (uniform) {
* terminate_if
* //potentially empty
* continue
* }
* //potentially empty
* break
* }
*/
BEGIN_TEST(isel.cf.empty_exec.loop_uniform_continue)
if (!setup_nir_cs(GFX11))
return;
nir_push_loop(nb);
{
//>> BB1
//! /* logical preds: BB0, BB2, / linear preds: BB0, BB3, / kind: uniform, loop-header, */
nir_push_if(nb, nir_unit_test_uniform_amd(nb, 1, 1, .base = 0));
{
//>> BB2
//! /* logical preds: BB1, / linear preds: BB1, / kind: continue, discard, */
nir_terminate_if(nb, nir_unit_test_uniform_amd(nb, 1, 1, .base = 1));
nir_jump(nb, nir_jump_continue);
}
nir_pop_if(nb, NULL);
//>> BB6
//! /* logical preds: BB5, / linear preds: BB4, BB5, / kind: uniform, break, */
nir_jump(nb, nir_jump_break);
}
nir_pop_loop(nb, NULL);
finish_isel_test();
END_TEST