[llvm] [MemoryLocation] Size Scalable Masked MemOps (PR #154785)

[David Sherwood via llvm-commits]

================
@@ -0,0 +1,196 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt < %s -aa-pipeline=basic-aa -passes=dse -S | FileCheck %s
+
+define <vscale x 4 x float> @dead_scalable_store(i32 %0, ptr %1) {
+; CHECK-LABEL: define <vscale x 4 x float> @dead_scalable_store(
+; CHECK: call void @llvm.masked.store.nxv4f32.p0(<vscale x 4 x float> %load.1.16, ptr nonnull %gep.arr.16, i32 1, <vscale x 4 x i1> %mask)
+; CHECK-NOT: call void @llvm.masked.store.nxv4f32.p0(<vscale x 4 x float> %load.1.32, ptr nonnull %gep.arr.32, i32 1, <vscale x 4 x i1> %mask)
+; CHECK: call void @llvm.masked.store.nxv4f32.p0(<vscale x 4 x float> %load.1.48, ptr nonnull %gep.arr.48, i32 1, <vscale x 4 x i1> %mask)
+;
+  %arr = alloca [64 x i32], align 4
+  %mask = call <vscale x 4 x i1> @llvm.get.active.lane.mask.nxv4i1.i32(i32 0, i32 4)
+
+  %gep.1.16 = getelementptr inbounds nuw i8, ptr %1, i64 16
+  %gep.1.32 = getelementptr inbounds nuw i8, ptr %1, i64 32
+  %gep.1.48 = getelementptr inbounds nuw i8, ptr %1, i64 48
+  %gep.arr.16 = getelementptr inbounds nuw i8, ptr %arr, i64 16
+  %gep.arr.32 = getelementptr inbounds nuw i8, ptr %arr, i64 32
+  %gep.arr.48 = getelementptr inbounds nuw i8, ptr %arr, i64 48
+
+  %load.1.16 = tail call <vscale x 4 x float> @llvm.masked.load.nxv4f32.p0(ptr nonnull %gep.1.16, i32 1, <vscale x 4 x i1> %mask, <vscale x 4 x float> zeroinitializer)
+  call void @llvm.masked.store.nxv4f32.p0(<vscale x 4 x float> %load.1.16, ptr nonnull %gep.arr.16, i32 1, <vscale x 4 x i1> %mask)
+
+  %load.1.32 = tail call <vscale x 4 x float> @llvm.masked.load.nxv4f32.p0(ptr nonnull %gep.1.32, i32 1, <vscale x 4 x i1> %mask, <vscale x 4 x float> zeroinitializer)
----------------
david-arm wrote:

Is it safe to add `tail` to memory intrinsic calls? Might be good to remove from all masked load/store intrinsics.

https://github.com/llvm/llvm-project/pull/154785