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

aco/insert_exec: exit shader using exec for top level discard

Browse Source 
Totals from 14538 (18.31% of 79395) affected shaders:
no changes

Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <None>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32731>
This commit is contained in:
Georg Lehmann
2024-12-19 20:24:28 +01:00
committed by Marge Bot
parent 6b35d6f75b
commit 42512208d8
2 changed files with 24 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/amd/compiler/aco_insert_exec_mask.cpp
+21 -24
View File
@@ -441,36 +441,36 @@ process_instructions(exec_ctx& ctx, Block* block, std::vector<aco_ptr<Instructio
info.exec[1].type &= ~mask_type_global;
}
Temp cond, exit_cond;
Temp cond;
if (instr->operands[0].isConstant()) {
assert(instr->operands[0].constantValue() == -1u);
/* save condition and set exec to zero */
exit_cond = bld.tmp(s1);
cond =
bld.sop1(Builder::s_and_saveexec, bld.def(bld.lm), bld.scc(Definition(exit_cond)),
Definition(exec, bld.lm), Operand::zero(), Operand(exec, bld.lm));
cond = bld.sop1(Builder::s_and_saveexec, bld.def(bld.lm), bld.def(s1, scc),
Definition(exec, bld.lm), Operand::zero(), Operand(exec, bld.lm));
} else {
cond = instr->operands[0].getTemp();
/* discard from current exec */
exit_cond = bld.sop2(Builder::s_andn2, Definition(exec, bld.lm), bld.def(s1, scc),
current_exec, cond)
.def(1)
.getTemp();
bld.sop2(Builder::s_andn2, Definition(exec, bld.lm), bld.def(s1, scc), current_exec,
cond);
}
/* discard from inner to outer exec mask on stack */
int num = info.exec.size() - 2;
for (int i = num; i >= 0; i--) {
Instruction* andn2 =
bld.sop2(Builder::s_andn2, bld.def(bld.lm), bld.def(s1, scc), info.exec[i].op, cond);
info.exec[i].op = Operand(andn2->definitions[0].getTemp());
exit_cond = andn2->definitions[1].getTemp();
if (info.exec.size() == 1) {
instr->operands[0] = Operand(exec, bld.lm);
} else {
/* discard from inner to outer exec mask on stack */
int num = info.exec.size() - 2;
Temp exit_cond;
for (int i = num; i >= 0; i--) {
Instruction* andn2 = bld.sop2(Builder::s_andn2, bld.def(bld.lm), bld.def(s1, scc),
info.exec[i].op, cond);
info.exec[i].op = Operand(andn2->definitions[0].getTemp());
exit_cond = andn2->definitions[1].getTemp();
}
instr->operands[0] = bld.scc(exit_cond);
}
instr->opcode = aco_opcode::p_exit_early_if_not;
instr->operands[0] = bld.scc(exit_cond);
assert(!ctx.handle_wqm || (info.exec[0].type & mask_type_wqm) == 0);
} else if (instr->opcode == aco_opcode::p_is_helper) {
Definition dst = instr->definitions[0];
assert(dst.size() == bld.lm.size());
@@ -511,21 +511,18 @@ process_instructions(exec_ctx& ctx, Block* block, std::vector<aco_ptr<Instructio
Definition def = state == Exact ? Definition(exec, bld.lm) : bld.def(bld.lm);
Operand src = instr->operands[0].isConstant() ? Operand(exec, bld.lm) : instr->operands[0];
Definition exit_cond =
bld.sop2(Builder::s_andn2, def, bld.def(s1, scc), info.exec[0].op, src).def(1);
bld.sop2(Builder::s_andn2, def, bld.def(s1, scc), info.exec[0].op, src);
info.exec[0].op = def.isTemp() ? Operand(def.getTemp()) : Operand(exec, bld.lm);
/* Update global WQM mask and store in exec. */
if (state == WQM) {
assert(info.exec.size() > 1);
exit_cond =
bld.sop1(Builder::s_wqm, Definition(exec, bld.lm), bld.def(s1, scc), def.getTemp())
.def(1);
bld.sop1(Builder::s_wqm, Definition(exec, bld.lm), bld.def(s1, scc), def.getTemp());
}
/* End shader if global mask is zero. */
instr->opcode = aco_opcode::p_exit_early_if_not;
instr->operands[0] = bld.scc(exit_cond.getTemp());
instr->operands[0] = Operand(exec, bld.lm);
bld.insert(std::move(instr));
/* Update all other exec masks. */
src/amd/compiler/tests/test_d3d11_derivs.cpp
+3 -2
View File
@@ -115,8 +115,9 @@ BEGIN_TEST(d3d11_derivs.discard)
pbld.add_vsfs(vs, fs);
/* The discard gets emitted as demote_if. */
//>> s2: %_:exec, s1: %cond:scc = s_wqm_b64 %_
//>> p_exit_early_if_not (kill)%cond:scc
//>> s2: %global, s1: (kill)%_:scc = s_andn2_b64 (kill)%_, (kill)%_
//>> s2: %_:exec, s1: (kill)%_:scc = s_wqm_b64 %global
//>> p_exit_early_if_not %_:exec
//>> v4: %_ = image_sample (kill)%_, (kill)%_, v1: undef, (kill)%_, (kill)%_ 2d
pbld.print_ir(VK_SHADER_STAGE_FRAGMENT_BIT, "ACO IR");
END_TEST