From 1006283732abe2e3074e2c5fa7b97a616eea3929 Mon Sep 17 00:00:00 2001
From: Faith Ekstrand <faith.ekstrand@collabora.com>
Date: Sat, 20 Jul 2024 17:46:11 -0500
Subject: [PATCH] nak: Fix shl64 for pre-Volta

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30275>
---
 src/nouveau/compiler/nak/builder.rs | 41 ++++++++++++++++++++++-------
 1 file changed, 31 insertions(+), 10 deletions(-)

diff --git a/src/nouveau/compiler/nak/builder.rs b/src/nouveau/compiler/nak/builder.rs
index 827fa618a05..8cd109412c5 100644
--- a/src/nouveau/compiler/nak/builder.rs
+++ b/src/nouveau/compiler/nak/builder.rs
@@ -148,16 +148,37 @@ pub trait SSABuilder: Builder {
         debug_assert!(shift.src_mod.is_none());
 
         let dst = self.alloc_ssa(RegFile::GPR, 2);
-        self.push_op(OpShf {
-            dst: dst[0].into(),
-            low: x[0].into(),
-            high: 0.into(),
-            shift,
-            right: false,
-            wrap: true,
-            data_type: IntType::U64,
-            dst_high: false,
-        });
+        if self.sm() >= 70 {
+            self.push_op(OpShf {
+                dst: dst[0].into(),
+                low: x[0].into(),
+                high: 0.into(),
+                shift,
+                right: false,
+                wrap: true,
+                data_type: IntType::U64,
+                dst_high: false,
+            });
+        } else {
+            // On Maxwell and earlier, shf.l doesn't work without .high so we
+            // have to use a regular 32-bit shift here.  32-bit shift doesn't
+            // have the NIR wrap semantics so we need to wrap manually.
+            let shift = if let SrcRef::Imm32(imm) = shift.src_ref {
+                (imm & 0x3f).into()
+            } else {
+                self.lop2(LogicOp2::And, shift, 0x3f.into()).into()
+            };
+            self.push_op(OpShf {
+                dst: dst[0].into(),
+                low: 0.into(),
+                high: x[0].into(),
+                shift,
+                right: false,
+                wrap: false,
+                data_type: IntType::U32,
+                dst_high: true,
+            });
+        }
         self.push_op(OpShf {
             dst: dst[1].into(),
             low: x[0].into(),