From c1ecf08e0d0090f25e9b7f0fea99d2cc95c7b91c Mon Sep 17 00:00:00 2001
From: Mary Guillemard <mary.guillemard@collabora.com>
Date: Tue, 17 Dec 2024 15:11:28 +0100
Subject: [PATCH] nak: Fix 8-bit selection for vectors

This fix at least permutation issues on vec16 of 8-bits values for
cooperative matrix.

Fixes: 9e84e9e44b1 ("nak: Add base support for 8 and 16-bit types")
Suggested-by: M Henning <drawoc@darkrefraction.com>
Signed-off-by: Mary Guillemard <mary.guillemard@collabora.com>
Reviewed-by: M Henning <drawoc@darkrefraction.com>
(cherry picked from commit 979dfaf0bb34eb566acabbb04b9d84eb31039559)

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32852>
---
 .pick_status.json                    |  2 +-
 src/nouveau/compiler/nak/from_nir.rs | 25 +++++++++++++++++++++----
 2 files changed, 22 insertions(+), 5 deletions(-)

diff --git a/.pick_status.json b/.pick_status.json
index fcaca36af54..4a3a25de075 100644
--- a/.pick_status.json
+++ b/.pick_status.json
@@ -2144,7 +2144,7 @@
         "description": "nak: Fix 8-bit selection for vectors",
         "nominated": true,
         "nomination_type": 2,
-        "resolution": 0,
+        "resolution": 1,
         "main_sha": null,
         "because_sha": "9e84e9e44b111a6afe8a346fb0bb74f9c597af61",
         "notes": null
diff --git a/src/nouveau/compiler/nak/from_nir.rs b/src/nouveau/compiler/nak/from_nir.rs
index 775d782c6af..1c6bf049d45 100644
--- a/src/nouveau/compiler/nak/from_nir.rs
+++ b/src/nouveau/compiler/nak/from_nir.rs
@@ -543,24 +543,41 @@ impl<'a> ShaderFromNir<'a> {
                     }
                     8 => {
                         for dc in 0..bits.div_ceil(32) {
-                            let mut psrc = [Src::new_zero(); 4];
+                            let mut psrc = [None; 4];
                             let mut psel = [0_u8; 4];
 
                             for b in 0..4 {
                                 let sc = dc * 4 + b;
                                 if sc < srcs.len() {
                                     let (ssa, byte) = srcs[sc];
+                                    // Deduplicate psrc entries
                                     for i in 0..4_u8 {
                                         let psrc_i = &mut psrc[usize::from(i)];
-                                        if *psrc_i == Src::new_zero() {
-                                            *psrc_i = ssa.into();
-                                        } else if *psrc_i != Src::from(ssa) {
+                                        if psrc_i.is_none() {
+                                            *psrc_i = Some(ssa.into());
+                                        } else if *psrc_i
+                                            != Some(Src::from(ssa))
+                                        {
                                             continue;
                                         }
                                         psel[b] = i * 4 + byte;
+                                        break;
                                     }
                                 }
                             }
+
+                            let psrc = {
+                                let mut res = [Src::new_zero(); 4];
+
+                                for (idx, src) in psrc.iter().enumerate() {
+                                    if let Some(src) = src {
+                                        res[idx] = *src;
+                                    }
+                                }
+
+                                res
+                            };
+
                             comps.push(b.prmt4(psrc, psel)[0]);
                         }
                     }