AlexIndustrial/mesa
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

aco: add wait_imm::unpack and wait_imm::max

Browse Source 
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28981>
This commit is contained in:
Rhys Perry
2024-05-03 11:19:57 +01:00
committed by Marge Bot
parent c894c9ab1b
commit 75532d8687
7 changed files with 104 additions and 86 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/amd/compiler/aco_insert_waitcnt.cpp
+5 -19
View File
@@ -228,16 +228,15 @@ struct wait_entry {
};
struct target_info {
uint16_t max_cnt[wait_type_num] = {};
wait_imm max_cnt;
uint32_t events[wait_type_num] = {};
uint16_t unordered_events;
target_info(enum amd_gfx_level gfx_level)
{
max_cnt[wait_type_vm] = gfx_level >= GFX9 ? 62 : 14;
max_cnt[wait_type_exp] = 6;
max_cnt[wait_type_lgkm] = gfx_level >= GFX10 ? 62 : 14;
max_cnt[wait_type_vs] = gfx_level >= GFX10 ? 62 : 0;
max_cnt = wait_imm::max(gfx_level);
for (unsigned i = 0; i < wait_type_num; i++)
max_cnt[i] = max_cnt[i] ? max_cnt[i] - 1 : 0;
events[wait_type_exp] = event_exp_pos | event_exp_param | event_exp_mrt_null |
event_gds_gpr_lock | event_vmem_gpr_lock | event_ldsdir;
@@ -402,19 +401,6 @@ check_instr(wait_ctx& ctx, wait_imm& wait, alu_delay_info& delay, Instruction* i
}
}
bool
parse_wait_instr(wait_ctx& ctx, wait_imm& imm, Instruction* instr)
{
if (instr->opcode == aco_opcode::s_waitcnt_vscnt && instr->operands[0].physReg() == sgpr_null) {
imm.vs = std::min<uint8_t>(imm.vs, instr->salu().imm);
return true;
} else if (instr->opcode == aco_opcode::s_waitcnt) {
imm.combine(wait_imm(ctx.gfx_level, instr->salu().imm));
return true;
}
return false;
}
bool
parse_delay_alu(wait_ctx& ctx, alu_delay_info& delay, Instruction* instr)
{
@@ -962,7 +948,7 @@ handle_block(Program* program, Block& block, wait_ctx& ctx)
for (size_t i = 0; i < block.instructions.size(); i++) {
aco_ptr<Instruction>& instr = block.instructions[i];
bool is_wait = parse_wait_instr(ctx, queued_imm, instr.get());
bool is_wait = queued_imm.unpack(ctx.gfx_level, instr.get());
bool is_delay_alu = parse_delay_alu(ctx, queued_delay, instr.get());
memory_sync_info sync_info = get_sync_info(instr.get());