diff options
| author | Samuel Pitoiset <samuel.pitoiset@gmail.com> | 2020-11-17 17:14:49 +0100 |
|---|---|---|
| committer | Dylan Baker <dylan.c.baker@intel.com> | 2020-12-14 09:42:43 -0800 |
| commit | 4c5d644bdb75ce69e69aa8a742860e2d8835e7f2 (patch) | |
| tree | c11670e641dcc6cc6465ba32bf276c964075037b | |
| parent | 60e001e95ff9acb6871c4b66ab6eb08eb82936ad (diff) | |
aco: fix combining max(-min(a, b), c) if a or b uses the neg modifier
No fossils-db changes.
Cc: 20.2, 20.3
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7657>
(cherry picked from commit 0fcd379184d658285f3313c5c4026253e0ec6930)
| -rw-r--r-- | src/amd/compiler/aco_optimizer.cpp | 8 | ||||
| -rw-r--r-- | src/amd/compiler/tests/test_optimizer.cpp | 23 |
2 files changed, 27 insertions, 4 deletions
diff --git a/src/amd/compiler/aco_optimizer.cpp b/src/amd/compiler/aco_optimizer.cpp index 8aef3809cec..9bb1ad698c7 100644 --- a/src/amd/compiler/aco_optimizer.cpp +++ b/src/amd/compiler/aco_optimizer.cpp @@ -2023,8 +2023,8 @@ bool combine_minmax(opt_ctx& ctx, aco_ptr<Instruction>& instr, aco_opcode opposi uint64_t omod_clamp = ctx.info[instr->definitions[0].tempId()].label & (label_omod_success | label_clamp_success); - /* min(-max(a, b), c) -> min3(-a, -b, c) * - * max(-min(a, b), c) -> max3(-a, -b, c) */ + /* min(-max(a, b), c) -> min3(c, -a, -b) * + * max(-min(a, b), c) -> max3(c, -a, -b) */ for (unsigned swap = 0; swap < 2; swap++) { Operand operands[3]; bool neg[3], abs[3], clamp, precise; @@ -2036,8 +2036,8 @@ bool combine_minmax(opt_ctx& ctx, aco_ptr<Instruction>& instr, aco_opcode opposi &clamp, &omod, &inbetween_neg, NULL, NULL, &precise) && inbetween_neg) { ctx.uses[instr->operands[swap].tempId()]--; - neg[1] = true; - neg[2] = true; + neg[1] = !neg[1]; + neg[2] = !neg[2]; create_vop3_for_op3(ctx, minmax3, instr, operands, neg, abs, opsel, clamp, omod); if (omod_clamp & label_omod_success) ctx.info[instr->definitions[0].tempId()].set_omod_success(instr.get()); diff --git a/src/amd/compiler/tests/test_optimizer.cpp b/src/amd/compiler/tests/test_optimizer.cpp index ad704a5134d..2c856e8775e 100644 --- a/src/amd/compiler/tests/test_optimizer.cpp +++ b/src/amd/compiler/tests/test_optimizer.cpp @@ -209,3 +209,26 @@ BEGIN_TEST(optimize.add3) finish_opt_test(); END_TEST + +BEGIN_TEST(optimize.minmax) + for (unsigned i = GFX8; i <= GFX10; i++) { + //>> v1: %a, s2: %_:exec = p_startpgm + if (!setup_cs("v1", (chip_class)i)) + continue; + + //! v1: %res0 = v_max3_f32 0, -0, %a + //! p_unit_test 0, %res0 + Temp xor0 = bld.vop2(aco_opcode::v_xor_b32, bld.def(v1), Operand(0x80000000u), Operand(inputs[0])); + Temp min = bld.vop2(aco_opcode::v_min_f32, bld.def(v1), Operand(0u), xor0); + Temp xor1 = bld.vop2(aco_opcode::v_xor_b32, bld.def(v1), Operand(0x80000000u), min); + writeout(0, bld.vop2(aco_opcode::v_max_f32, bld.def(v1), Operand(0u), xor1)); + + //! v1: %res1 = v_max3_f32 0, -0, -%a + //! p_unit_test 1, %res1 + min = bld.vop2(aco_opcode::v_min_f32, bld.def(v1), Operand(0u), Operand(inputs[0])); + xor1 = bld.vop2(aco_opcode::v_xor_b32, bld.def(v1), Operand(0x80000000u), min); + writeout(1, bld.vop2(aco_opcode::v_max_f32, bld.def(v1), Operand(0u), xor1)); + + finish_opt_test(); + } +END_TEST |