fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-06 09:28:07 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

aco/gfx12: fix s_wait_event immediate

Browse source 
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Acked-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29466>
This commit is contained in:
Rhys Perry 2024-05-27 14:23:38 +01:00 • committed by Marge Bot
parent 26c981b2b8
commit 2a4424425a
5 changed files with 17 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

3
src/amd/compiler/aco_instruction_selection.cpp
View file

@ -11498,7 +11498,8 @@ pops_await_overlapped_waves(isel_context* ctx)
/* GFX11+ - waiting for the export from the overlapped waves.
* Await the export_ready event (bit wait_event_imm_dont_wait_export_ready clear).
*/
bld.sopp(aco_opcode::s_wait_event, 0);
bld.sopp(aco_opcode::s_wait_event,
ctx->program->gfx_level >= GFX12 ? wait_event_imm_wait_export_ready_gfx12 : 0);
return;
}

11
src/amd/compiler/aco_ir.cpp
View file

@ -209,6 +209,14 @@ init_program(Program* program, Stage stage, const struct aco_shader_info* info,
program->next_fp_mode.round32 = fp_round_ne;
}
bool
is_wait_export_ready(amd_gfx_level gfx_level, const Instruction* instr)
{
return instr->opcode == aco_opcode::s_wait_event &&
(gfx_level >= GFX12 ? (instr->salu().imm & wait_event_imm_wait_export_ready_gfx12)
: !(instr->salu().imm & wait_event_imm_dont_wait_export_ready_gfx11));
}
memory_sync_info
get_sync_info(const Instruction* instr)
{
@ -216,8 +224,7 @@ get_sync_info(const Instruction* instr)
* overlapping waves in the queue family.
*/
if (instr->opcode == aco_opcode::p_pops_gfx9_overlapped_wave_wait_done ||
(instr->opcode == aco_opcode::s_wait_event &&
!(instr->salu().imm & wait_event_imm_dont_wait_export_ready))) {
instr->opcode == aco_opcode::s_wait_event) {
return memory_sync_info(storage_buffer | storage_image, semantic_acquire, scope_queuefamily);
} else if (instr->opcode == aco_opcode::p_pops_gfx9_ordered_section_done) {
return memory_sync_info(storage_buffer | storage_image, semantic_release, scope_queuefamily);

4
src/amd/compiler/aco_ir.h
View file

@ -240,7 +240,8 @@ enum wait_event_imm : uint16_t {
* their ordered sections (by performing the `done` export), and that the current wave may enter
* its ordered section.
*/
wait_event_imm_dont_wait_export_ready = 0x1,
wait_event_imm_dont_wait_export_ready_gfx11 = 0x1,
wait_event_imm_wait_export_ready_gfx12 = 0x2,
};
constexpr Format
@ -1727,6 +1728,7 @@ is_phi(aco_ptr<Instruction>& instr)
return is_phi(instr.get());
}
bool is_wait_export_ready(amd_gfx_level gfx_level, const Instruction* instr);
memory_sync_info get_sync_info(const Instruction* instr);
inline bool

4
src/amd/compiler/aco_print_ir.cpp
View file

@ -395,8 +395,8 @@ print_instr_format_specific(enum amd_gfx_level gfx_level, const Instruction* ins
break;
}
case aco_opcode::s_wait_event: {
if (!(imm & wait_event_imm_dont_wait_export_ready))
fprintf(output, " export_ready");
if (is_wait_export_ready(gfx_level, instr))
fprintf(output, " wait_export_ready");
break;
}
default: {

3
src/amd/compiler/aco_scheduler.cpp
View file

@ -562,8 +562,7 @@ perform_hazard_query(hazard_query* query, Instruction* instr, bool upwards)
*/
if (upwards) {
if (instr->opcode == aco_opcode::p_pops_gfx9_add_exiting_wave_id ||
(instr->opcode == aco_opcode::s_wait_event &&
!(instr->salu().imm & wait_event_imm_dont_wait_export_ready))) {
is_wait_export_ready(query->gfx_level, instr)) {
return hazard_fail_unreorderable;
}
} else {