From ba3fd5dc5739a208033bc4cec47acc5e492cce0e Mon Sep 17 00:00:00 2001
From: Caio Oliveira <caio.oliveira@intel.com>
Date: Sat, 3 Aug 2024 20:38:22 -0700
Subject: [PATCH] intel/brw: Don't retype load_subgroup_invocation result to
 signed

The values are small unsigned integers, so their signed representation
will be the same -- the sign conversion is not needed.  As a result the
extra MOV can be elided by the optimizations.

Fossil-db results for DG2

```
Totals:
Instrs: 151779000 -> 151761591 (-0.01%)
Cycle count: 12743968649 -> 12742826024 (-0.01%); split: -0.01%, +0.00%
Max live registers: 31834993 -> 31834996 (+0.00%)

Totals from 17018 (2.70% of 631450) affected shaders:
Instrs: 2381740 -> 2364331 (-0.73%)
Cycle count: 76798588 -> 75655963 (-1.49%); split: -1.70%, +0.22%
Max live registers: 378921 -> 378924 (+0.00%)
```

and TGL

```
Totals:
Instrs: 149812033 -> 149794080 (-0.01%); split: -0.01%, +0.00%
Cycle count: 11534727002 -> 11534929834 (+0.00%); split: -0.01%, +0.01%
Spill count: 42510 -> 42511 (+0.00%); split: -0.00%, +0.01%
Fill count: 75100 -> 75101 (+0.00%); split: -0.00%, +0.00%
Max live registers: 31727318 -> 31727321 (+0.00%)

Totals from 17421 (2.76% of 630458) affected shaders:
Instrs: 3092614 -> 3074661 (-0.58%); split: -0.58%, +0.00%
Cycle count: 286061417 -> 286264249 (+0.07%); split: -0.32%, +0.39%
Spill count: 11538 -> 11539 (+0.01%); split: -0.02%, +0.03%
Fill count: 21359 -> 21360 (+0.00%); split: -0.01%, +0.01%
Max live registers: 418954 -> 418957 (+0.00%)
```

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30508>
---
 src/intel/compiler/brw_fs_nir.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/intel/compiler/brw_fs_nir.cpp b/src/intel/compiler/brw_fs_nir.cpp
index f4f6ce65de5..5666d32f82a 100644
--- a/src/intel/compiler/brw_fs_nir.cpp
+++ b/src/intel/compiler/brw_fs_nir.cpp
@@ -7333,7 +7333,7 @@ fs_nir_emit_intrinsic(nir_to_brw_state &ntb,
       break;
 
    case nir_intrinsic_load_subgroup_invocation:
-      bld.MOV(retype(dest, BRW_TYPE_D),
+      bld.MOV(retype(dest, BRW_TYPE_UD),
               ntb.system_values[SYSTEM_VALUE_SUBGROUP_INVOCATION]);
       break;