From 2f2031d2b22d64d015625db2dd11e44e7ab37091 Mon Sep 17 00:00:00 2001
From: Ralf Jung <post@ralfj.de>
Date: Sat, 8 Jun 2024 18:05:44 +0200
Subject: [PATCH] simd packed types: update outdated check, extend codegen test

---
 compiler/rustc_codegen_llvm/src/intrinsic.rs | 10 +++++----
 tests/codegen/simd/packed-simd.rs            | 23 ++++++++++++++++----
 2 files changed, 25 insertions(+), 8 deletions(-)

diff --git a/compiler/rustc_codegen_llvm/src/intrinsic.rs b/compiler/rustc_codegen_llvm/src/intrinsic.rs
index 7b1038d5617..ad4e753fe9d 100644
--- a/compiler/rustc_codegen_llvm/src/intrinsic.rs
+++ b/compiler/rustc_codegen_llvm/src/intrinsic.rs
@@ -1109,10 +1109,12 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
         tcx.normalize_erasing_late_bound_regions(ty::ParamEnv::reveal_all(), callee_ty.fn_sig(tcx));
     let arg_tys = sig.inputs();
 
-    // Vectors must be immediates (non-power-of-2 #[repr(packed)] are not)
-    for (ty, arg) in arg_tys.iter().zip(args) {
-        if ty.is_simd() && !matches!(arg.val, OperandValue::Immediate(_)) {
-            return_error!(InvalidMonomorphization::SimdArgument { span, name, ty: *ty });
+    // Sanity-check: all vector arguments must be immediates.
+    if cfg!(debug_assertions) {
+        for (ty, arg) in arg_tys.iter().zip(args) {
+            if ty.is_simd() {
+                assert!(matches!(arg.val, OperandValue::Immediate(_)));
+            }
         }
     }
 
diff --git a/tests/codegen/simd/packed-simd.rs b/tests/codegen/simd/packed-simd.rs
index f0911b6e360..1df09c96e6c 100644
--- a/tests/codegen/simd/packed-simd.rs
+++ b/tests/codegen/simd/packed-simd.rs
@@ -9,10 +9,11 @@ use core::intrinsics::simd as intrinsics;
 use core::{mem, ptr};
 
 // Test codegen for not only "packed" but also "fully aligned" SIMD types, and conversion between
-// A repr(packed,simd) type with 3 elements can't exceed its element alignment,
-// whereas the same type as repr(simd) will instead have padding.
+// them. A repr(packed,simd) type with 3 elements can't exceed its element alignment, whereas the
+// same type as repr(simd) will instead have padding.
 
 #[repr(simd, packed)]
+#[derive(Copy, Clone)]
 pub struct Simd<T, const N: usize>([T; N]);
 
 #[repr(simd)]
@@ -28,11 +29,11 @@ fn load<T, const N: usize>(v: Simd<T, N>) -> FullSimd<T, N> {
     }
 }
 
-// CHECK-LABEL: square_packed
+// CHECK-LABEL: square_packed_full
 // CHECK-SAME: ptr{{[a-z_ ]*}} sret([[RET_TYPE:[^)]+]]) [[RET_ALIGN:align (8|16)]]{{[^%]*}} [[RET_VREG:%[_0-9]*]]
 // CHECK-SAME: ptr{{[a-z_ ]*}} align 4
 #[no_mangle]
-pub fn square_packed(x: Simd<f32, 3>) -> FullSimd<f32, 3> {
+pub fn square_packed_full(x: Simd<f32, 3>) -> FullSimd<f32, 3> {
     // CHECK-NEXT: start
     // noopt: alloca [[RET_TYPE]], [[RET_ALIGN]]
     // CHECK: load <3 x float>
@@ -42,3 +43,17 @@ pub fn square_packed(x: Simd<f32, 3>) -> FullSimd<f32, 3> {
     // CHECK-NEXT: ret void
     unsafe { intrinsics::simd_mul(x, x) }
 }
+
+// CHECK-LABEL: square_packed
+// CHECK-SAME: ptr{{[a-z_ ]*}} sret([[RET_TYPE:[^)]+]]) [[RET_ALIGN:align 4]]{{[^%]*}} [[RET_VREG:%[_0-9]*]]
+// CHECK-SAME: ptr{{[a-z_ ]*}} align 4
+#[no_mangle]
+pub fn square_packed(x: Simd<f32, 3>) -> Simd<f32, 3> {
+    // CHECK-NEXT: start
+    // CHECK-NEXT: load <3 x float>
+    // noopt-NEXT: load <3 x float>
+    // CHECK-NEXT: [[VREG:%[a-z0-9_]+]] = fmul <3 x float>
+    // CHECK-NEXT: store <3 x float> [[VREG]], ptr [[RET_VREG]], [[RET_ALIGN]]
+    // CHECK-NEXT: ret void
+    unsafe { intrinsics::simd_mul(x, x) }
+}