[llvm-branch-commits] [llvm] [SelectionDAG][X86] Widen <2 x T> vector types for atomic load (PR #120598)

[via llvm-branch-commits]

================
@@ -146,6 +146,47 @@ define <1 x i64> @atomic_vec1_i64_align(ptr %x) nounwind {
   ret <1 x i64> %ret
 }
 
+define <2 x i8> @atomic_vec2_i8(ptr %x) {
+; CHECK3-LABEL: atomic_vec2_i8:
+; CHECK3:       ## %bb.0:
+; CHECK3-NEXT:    movzwl (%rdi), %eax
+; CHECK3-NEXT:    retq
+;
+; CHECK0-LABEL: atomic_vec2_i8:
+; CHECK0:       ## %bb.0:
+; CHECK0-NEXT:    movw (%rdi), %ax
+; CHECK0-NEXT:    retq
+  %ret = load atomic <2 x i8>, ptr %x acquire, align 4
+  ret <2 x i8> %ret
+}
+
+define <2 x i16> @atomic_vec2_i16(ptr %x) {
+; CHECK-LABEL: atomic_vec2_i16:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    movl (%rdi), %eax
+; CHECK-NEXT:    retq
+  %ret = load atomic <2 x i16>, ptr %x acquire, align 4
+  ret <2 x i16> %ret
+}
----------------
jofrn wrote:

https://github.com/llvm/llvm-project/pull/120640/files contains <2,4 x half> and <2,4 x bloat>

https://github.com/llvm/llvm-project/pull/120598