[llvm] [MemoryLocation] Teach MemoryLocation about llvm.experimental.memset.pattern (PR #120421)

[Philip Reames via llvm-commits]

================
@@ -66,6 +66,71 @@ define void @memset_and_store_2(ptr %ptr, i64 %len) {
   ret void
 }
 
+define void @memset_pattern_equal_size_values(ptr %ptr, i64 %len) {
+; CHECK-LABEL: @memset_pattern_equal_size_values(
+; CHECK-NEXT:    call void @llvm.experimental.memset.pattern.p0.i8.i64(ptr align 1 [[PTR:%.*]], i8 0, i64 [[LEN:%.*]], i1 false)
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.experimental.memset.pattern.p0.i8.i64(ptr align 1 %ptr, i8 0, i64 %len, i1 false)
+  call void @llvm.experimental.memset.pattern.p0.i8.i64(ptr align 1 %ptr, i8 0, i64 %len, i1 false)
+  ret void
+}
+
+define void @memset_pattern_different_size_values_1(ptr %ptr, i64 %len.1, i64 %len.2) {
+; CHECK-LABEL: @memset_pattern_different_size_values_1(
+; CHECK-NEXT:    call void @llvm.experimental.memset.pattern.p0.i8.i64(ptr align 1 [[PTR:%.*]], i8 0, i64 [[LEN_1:%.*]], i1 false)
----------------
preames wrote:

A general enhancement we could do for this case would be to fold the calls to a length which is the umin of the two lengths.  This is generic and not related to this change.

https://github.com/llvm/llvm-project/pull/120421