From 5d8f5615d0a584f463b86678be833f0da7f9d22a Mon Sep 17 00:00:00 2001
From: Rhys Perry <pendingchaos02@gmail.com>
Date: Fri, 29 Apr 2022 17:23:20 +0100
Subject: [PATCH] aco: ignore precise flag when optimizing integer clamps
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

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/16296>
---
 src/amd/compiler/aco_optimizer.cpp        |  3 ++-
 src/amd/compiler/tests/test_optimizer.cpp | 29 ++++++++++++-----------
 2 files changed, 17 insertions(+), 15 deletions(-)

diff --git a/src/amd/compiler/aco_optimizer.cpp b/src/amd/compiler/aco_optimizer.cpp
index 71497cb7267..26cb5caf121 100644
--- a/src/amd/compiler/aco_optimizer.cpp
+++ b/src/amd/compiler/aco_optimizer.cpp
@@ -2851,7 +2851,8 @@ combine_clamp(opt_ctx& ctx, aco_ptr<Instruction>& instr, aco_opcode min, aco_opc
          /* max(min(src, upper), lower) returns upper if src is NaN, but
           * med3(src, lower, upper) returns lower.
           */
-         if (precise && instr->opcode != min)
+         if (precise && instr->opcode != min &&
+             (min == aco_opcode::v_min_f16 || min == aco_opcode::v_min_f32))
             continue;
 
          int const0_idx = -1, const1_idx = -1;
diff --git a/src/amd/compiler/tests/test_optimizer.cpp b/src/amd/compiler/tests/test_optimizer.cpp
index 1684b9ad824..8afb56c4cb5 100644
--- a/src/amd/compiler/tests/test_optimizer.cpp
+++ b/src/amd/compiler/tests/test_optimizer.cpp
@@ -530,21 +530,22 @@ BEGIN_TEST(optimize.clamp)
                            bld.vop2(cfg.max, bld.def(v1), inputs[2], inputs[0])));
 
       /* correct NaN behaviour with precise */
+      if (cfg.min == aco_opcode::v_min_f16 || cfg.min == aco_opcode::v_min_f32) {
+         //~f(16|32)! v1: %res7 = @med3 @ub, @lb, %a
+         //~f(16|32)! p_unit_test 7, %res7
+         Builder::Result max = bld.vop2(cfg.max, bld.def(v1), cfg.lb, inputs[0]);
+         max.def(0).setPrecise(true);
+         Builder::Result min = bld.vop2(cfg.min, bld.def(v1), cfg.ub, max);
+         max.def(0).setPrecise(true);
+         writeout(7, min);
 
-      //! v1: %res7 = @med3 @ub, @lb, %a
-      //! p_unit_test 7, %res7
-      Builder::Result max = bld.vop2(cfg.max, bld.def(v1), cfg.lb, inputs[0]);
-      max.def(0).setPrecise(true);
-      Builder::Result min = bld.vop2(cfg.min, bld.def(v1), cfg.ub, max);
-      max.def(0).setPrecise(true);
-      writeout(7, min);
-
-      //! v1: (precise)%res8_tmp = @min @ub, %a
-      //! v1: %res8 = @max @lb, %res8_tmp
-      //! p_unit_test 8, %res8
-      min = bld.vop2(cfg.min, bld.def(v1), cfg.ub, inputs[0]);
-      min.def(0).setPrecise(true);
-      writeout(8, bld.vop2(cfg.max, bld.def(v1), cfg.lb, min));
+         //~f(16|32)! v1: (precise)%res8_tmp = @min @ub, %a
+         //~f(16|32)! v1: %res8 = @max @lb, %res8_tmp
+         //~f(16|32)! p_unit_test 8, %res8
+         min = bld.vop2(cfg.min, bld.def(v1), cfg.ub, inputs[0]);
+         min.def(0).setPrecise(true);
+         writeout(8, bld.vop2(cfg.max, bld.def(v1), cfg.lb, min));
+      }
 
       finish_opt_test();
    }