Mesa (main): aco: don't create unnecessary addition in indirect get_sampler_desc()

[GitLab Mirror]

Module: Mesa
Branch: main
Commit: 786d434397c2e3566404b5efe2492cd08e0e6c86
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=786d434397c2e3566404b5efe2492cd08e0e6c86

Author: Rhys Perry <pendingchaos02 at gmail.com>
Date:   Fri Dec  3 13:42:25 2021 +0000

aco: don't create unnecessary addition in indirect get_sampler_desc()

I don't think this has any effect on GFX9+ because the addition is
combined into the load.

fossil-db (polaris10):
Totals from 12595 (9.29% of 135627) affected shaders:
SGPRs: 1054348 -> 1054860 (+0.05%); split: -0.02%, +0.07%
VGPRs: 667240 -> 667320 (+0.01%); split: -0.01%, +0.02%
CodeSize: 82761508 -> 82512816 (-0.30%); split: -0.30%, +0.00%
MaxWaves: 62182 -> 62181 (-0.00%)
Instrs: 16072934 -> 16010764 (-0.39%); split: -0.39%, +0.00%
Latency: 582819635 -> 582287964 (-0.09%); split: -0.13%, +0.04%
InvThroughput: 276460536 -> 276417613 (-0.02%); split: -0.06%, +0.05%
VClause: 261656 -> 261654 (-0.00%); split: -0.01%, +0.01%
SClause: 680952 -> 680854 (-0.01%); split: -0.05%, +0.04%
Copies: 1727202 -> 1727742 (+0.03%); split: -0.12%, +0.15%
Branches: 547050 -> 547033 (-0.00%); split: -0.01%, +0.00%

Signed-off-by: Rhys Perry <pendingchaos02 at gmail.com>
Reviewed-by: Daniel Schürmann <daniel at schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14043>

---

 src/amd/compiler/aco_instruction_selection.cpp | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/src/amd/compiler/aco_instruction_selection.cpp b/src/amd/compiler/aco_instruction_selection.cpp
index ad496f7e714..7b786c85667 100644
--- a/src/amd/compiler/aco_instruction_selection.cpp
+++ b/src/amd/compiler/aco_instruction_selection.cpp
@@ -5865,9 +5865,10 @@ get_sampler_desc(isel_context* ctx, nir_deref_instr* deref_instr,
    if (!index_set) {
       off = bld.copy(bld.def(s1), Operand::c32(offset));
    } else {
-      off = Operand(
-         (Temp)bld.sop2(aco_opcode::s_add_i32, bld.def(s1), bld.def(s1, scc), Operand::c32(offset),
-                        bld.sop2(aco_opcode::s_mul_i32, bld.def(s1), Operand::c32(stride), index)));
+      off = bld.sop2(aco_opcode::s_mul_i32, bld.def(s1), Operand::c32(stride), index);
+      if (offset)
+         off = bld.sop2(aco_opcode::s_add_i32, bld.def(s1), bld.def(s1, scc), Operand::c32(offset),
+                        off);
    }
 
    Temp res = bld.smem(opcode, bld.def(type), list, off);