fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-24 19:40:10 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

aco: fix long-jump version of discard early exit

Browse source 
It isn't safe to modify the exec mask before the discard block, and the
definition interferes with GFX11 NOP insertion.

Just use s[0:1] instead, since we won't be using it.

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/18125>
This commit is contained in:
Rhys Perry 2022-08-18 14:31:06 +01:00 • committed by Marge Bot
parent 1a250eeae3
commit fb13ed6ff0
4 changed files with 50 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

20
src/amd/compiler/aco_assembler.cpp
View file

@ -902,10 +902,18 @@ emit_long_jump(asm_context& ctx, SOPP_instruction* branch, bool backwards,
{
Builder bld(ctx.program);
Definition def_tmp_lo(branch->definitions[0].physReg(), s1);
Operand op_tmp_lo(branch->definitions[0].physReg(), s1);
Definition def_tmp_hi(branch->definitions[0].physReg().advance(4), s1);
Operand op_tmp_hi(branch->definitions[0].physReg().advance(4), s1);
Definition def;
if (branch->definitions.empty()) {
assert(ctx.program->blocks[branch->block].kind & block_kind_discard_early_exit);
def = Definition(PhysReg(0), s2); /* The discard early exit block doesn't use SGPRs. */
} else {
def = branch->definitions[0];
}
Definition def_tmp_lo(def.physReg(), s1);
Operand op_tmp_lo(def.physReg(), s1);
Definition def_tmp_hi(def.physReg().advance(4), s1);
Operand op_tmp_hi(def.physReg().advance(4), s1);
aco_ptr<Instruction> instr;
@ -926,7 +934,7 @@ emit_long_jump(asm_context& ctx, SOPP_instruction* branch, bool backwards,
}
/* create the new PC and stash SCC in the LSB */
instr.reset(bld.sop1(aco_opcode::s_getpc_b64, branch->definitions[0]).instr);
instr.reset(bld.sop1(aco_opcode::s_getpc_b64, def).instr);
emit_instruction(ctx, out, instr.get());
instr.reset(
@ -944,7 +952,7 @@ emit_long_jump(asm_context& ctx, SOPP_instruction* branch, bool backwards,
/* create the s_setpc_b64 to jump */
instr.reset(
bld.sop1(aco_opcode::s_setpc_b64, Operand(branch->definitions[0].physReg(), s2)).instr);
bld.sop1(aco_opcode::s_setpc_b64, Operand(def.physReg(), s2)).instr);
emit_instruction(ctx, out, instr.get());
}

1
src/amd/compiler/aco_ir.h
View file

@ -1836,6 +1836,7 @@ enum block_kind {
block_kind_branch = 1 << 8,
block_kind_merge = 1 << 9,
block_kind_invert = 1 << 10,
block_kind_discard_early_exit = 1 << 11,
block_kind_uses_discard = 1 << 12,
block_kind_needs_lowering = 1 << 13,
block_kind_export_end = 1 << 15,

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

@ -2121,6 +2121,7 @@ lower_to_hw_instr(Program* program)
if (!discard_block) {
discard_block = program->create_and_insert_block();
discard_block->kind = block_kind_discard_early_exit;
block = &program->blocks[block_idx];
bld.reset(discard_block);
@ -2133,8 +2134,7 @@ lower_to_hw_instr(Program* program)
}
assert(instr->operands[0].physReg() == scc);
bld.sopp(aco_opcode::s_cbranch_scc0, Definition(exec, s2), instr->operands[0],
discard_block->index);
bld.sopp(aco_opcode::s_cbranch_scc0, instr->operands[0], discard_block->index);
discard_block->linear_preds.push_back(block->index);
block->linear_succs.push_back(discard_block->index);

33
src/amd/compiler/tests/test_assembler.cpp
View file

@ -226,6 +226,39 @@ BEGIN_TEST(assembler.long_jump.constaddr)
finish_assembler_test();
END_TEST
BEGIN_TEST(assembler.long_jump.discard_early_exit)
if (!setup_cs(NULL, (amd_gfx_level)GFX10))
return;
//! BB0:
//! s_cbranch_scc1 BB1 ; bf850006
//! s_getpc_b64 s[0:1] ; be801f00
//! s_addc_u32 s0, s0, 0x20014 ; 8200ff00 00020014
//! s_bitcmp1_b32 s0, 0 ; bf0d8000
//! s_bitset0_b32 s0, 0 ; be801b80
//! s_setpc_b64 s[0:1] ; be802000
bld.sopp(aco_opcode::s_cbranch_scc0, 2);
bld.reset(program->create_and_insert_block());
//! BB1:
//! s_nop 1 ; bf800001
//!(then repeated 32766 times)
//! s_endpgm ; bf810000
for (unsigned i = 0; i < INT16_MAX; i++)
bld.sopp(aco_opcode::s_nop, -1, 1);
//! BB2:
//! s_endpgm ; bf810000
bld.reset(program->create_and_insert_block());
program->blocks[1].linear_preds.push_back(0u);
program->blocks[2].linear_preds.push_back(0u);
program->blocks[2].kind = block_kind_discard_early_exit;
finish_assembler_test();
END_TEST
BEGIN_TEST(assembler.v_add3)
for (unsigned i = GFX9; i <= GFX10; i++) {
if (!setup_cs(NULL, (amd_gfx_level)i))