Mesa (master): aco: fix gfx10_wave64_bpermute

[GitLab Mirror]

Module: Mesa
Branch: master
Commit: 20eb1acb6f404ffa4e502e7de8dec8ac83e7a8a8
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=20eb1acb6f404ffa4e502e7de8dec8ac83e7a8a8

Author: Rhys Perry <pendingchaos02 at gmail.com>
Date:   Mon Feb  3 17:54:07 2020 +0000

aco: fix gfx10_wave64_bpermute

Since 9254fb4fc72, the pass replaced the SCC clobber with the scalar
identity temporary. Just skip most of the temporary setup, since we don't
need it for gfx10_wave64_bpermute.

Although shuffles are disabled on GFX10, Detroit: Become Human seems to
use them anyway.

Signed-off-by: Rhys Perry <pendingchaos02 at gmail.com>
Reviewed-By: Timur Kristóf <timur.kristof at gmail.com>
Fixes: 9254fb4fc72ed289ffded28ef067b4582973e90c ('aco: don't use a scalar
       temporary for reductions on GFX10')

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3683>

---

 src/amd/compiler/aco_reduce_assign.cpp | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/src/amd/compiler/aco_reduce_assign.cpp b/src/amd/compiler/aco_reduce_assign.cpp
index 58c64cfb019..96846e926d4 100644
--- a/src/amd/compiler/aco_reduce_assign.cpp
+++ b/src/amd/compiler/aco_reduce_assign.cpp
@@ -114,6 +114,11 @@ void setup_reduce_temp(Program* program)
             }
          }
 
+         if (op == gfx10_wave64_bpermute) {
+            instr->operands[1] = Operand(reduceTmp);
+            continue;
+         }
+
          /* same as before, except for the vector temporary instead of the reduce temporary */
          unsigned cluster_size = static_cast<Pseudo_reduction_instruction *>(instr)->cluster_size;
          bool need_vtmp = op == imul32 || op == fadd64 || op == fmul64 ||
@@ -121,7 +126,7 @@ void setup_reduce_temp(Program* program)
                           op == umax64 || op == imin64 || op == imax64 ||
                           op == imul64;
 
-         if (program->chip_class >= GFX10 && cluster_size == 64 && op != gfx10_wave64_bpermute)
+         if (program->chip_class >= GFX10 && cluster_size == 64)
             need_vtmp = true;
          if (program->chip_class >= GFX10 && op == iadd64)
             need_vtmp = true;