Mesa (master): aco: fix combining max(-min(a, b), c) if a or b uses the neg modifier

[GitLab Mirror]

Module: Mesa
Branch: master
Commit: 0fcd379184d658285f3313c5c4026253e0ec6930
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=0fcd379184d658285f3313c5c4026253e0ec6930

Author: Samuel Pitoiset <samuel.pitoiset at gmail.com>
Date:   Tue Nov 17 17:14:49 2020 +0100

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 at gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02 at gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7657>

---

 src/amd/compiler/aco_optimizer.cpp        |  8 ++++----
 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 1d796d4ad03..5d810b306bb 100644
--- a/src/amd/compiler/aco_optimizer.cpp
+++ b/src/amd/compiler/aco_optimizer.cpp
@@ -2106,8 +2106,8 @@ bool combine_minmax(opt_ctx& ctx, aco_ptr<Instruction>& instr, aco_opcode opposi
    if (combine_three_valu_op(ctx, instr, instr->opcode, minmax3, "012", 1 | 2))
       return true;
 
-   /* 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;
@@ -2119,8 +2119,8 @@ 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);
          return true;
       }
diff --git a/src/amd/compiler/tests/test_optimizer.cpp b/src/amd/compiler/tests/test_optimizer.cpp
index 84f77aeaed8..8d6805febb8 100644
--- a/src/amd/compiler/tests/test_optimizer.cpp
+++ b/src/amd/compiler/tests/test_optimizer.cpp
@@ -700,3 +700,26 @@ 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