diff --git a/llvm/lib/Transforms/Utils/LowerMemIntrinsics.cpp b/llvm/lib/Transforms/Utils/LowerMemIntrinsics.cpp
index c75de8687879d..88934a34d2e2a 100644
--- a/llvm/lib/Transforms/Utils/LowerMemIntrinsics.cpp
+++ b/llvm/lib/Transforms/Utils/LowerMemIntrinsics.cpp
@@ -13,6 +13,7 @@
 #include "llvm/IR/IntrinsicInst.h"
 #include "llvm/IR/MDBuilder.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/MathExtras.h"
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 #include <optional>
 
@@ -155,6 +156,26 @@ void llvm::createMemCpyLoopKnownSize(
          "Bytes copied should match size in the call!");
 }
 
+// \returns \p Len udiv \p OpSize, checking for optimization opportunities.
+static Value *getRuntimeLoopCount(const DataLayout &DL, IRBuilderBase &B,
+                                  Value *Len, Value *OpSize,
+                                  unsigned OpSizeVal) {
+  // For powers of 2, we can lshr by log2 instead of using udiv.
+  if (isPowerOf2_32(OpSizeVal))
+    return B.CreateLShr(Len, Log2_32(OpSizeVal));
+  return B.CreateUDiv(Len, OpSize);
+}
+
+// \returns \p Len urem \p OpSize, checking for optimization opportunities.
+static Value *getRuntimeLoopRemainder(const DataLayout &DL, IRBuilderBase &B,
+                                      Value *Len, Value *OpSize,
+                                      unsigned OpSizeVal) {
+  // For powers of 2, we can and by (OpSizeVal - 1) instead of using urem.
+  if (isPowerOf2_32(OpSizeVal))
+    return B.CreateAnd(Len, OpSizeVal - 1);
+  return B.CreateURem(Len, OpSize);
+}
+
 void llvm::createMemCpyLoopUnknownSize(
     Instruction *InsertBefore, Value *SrcAddr, Value *DstAddr, Value *CopyLen,
     Align SrcAlign, Align DstAlign, bool SrcIsVolatile, bool DstIsVolatile,
@@ -194,9 +215,11 @@ void llvm::createMemCpyLoopUnknownSize(
   Type *Int8Type = Type::getInt8Ty(Ctx);
   bool LoopOpIsInt8 = LoopOpType == Int8Type;
   ConstantInt *CILoopOpSize = ConstantInt::get(ILengthType, LoopOpSize);
-  Value *RuntimeLoopCount = LoopOpIsInt8 ?
-                            CopyLen :
-                            PLBuilder.CreateUDiv(CopyLen, CILoopOpSize);
+  Value *RuntimeLoopCount = LoopOpIsInt8
+                                ? CopyLen
+                                : getRuntimeLoopCount(DL, PLBuilder, CopyLen,
+                                                      CILoopOpSize, LoopOpSize);
+
   BasicBlock *LoopBB =
       BasicBlock::Create(Ctx, "loop-memcpy-expansion", ParentFunc, PostLoopBB);
   IRBuilder<> LoopBuilder(LoopBB);
@@ -239,8 +262,8 @@ void llvm::createMemCpyLoopUnknownSize(
     assert((ResLoopOpSize == AtomicElementSize ? *AtomicElementSize : 1) &&
            "Store size is expected to match type size");
 
-    // Add in the
-    Value *RuntimeResidual = PLBuilder.CreateURem(CopyLen, CILoopOpSize);
+    Value *RuntimeResidual = getRuntimeLoopRemainder(DL, PLBuilder, CopyLen,
+                                                     CILoopOpSize, LoopOpSize);
     Value *RuntimeBytesCopied = PLBuilder.CreateSub(CopyLen, RuntimeResidual);
 
     // Loop body for the residual copy.
diff --git a/llvm/test/CodeGen/AMDGPU/lower-mem-intrinsics.ll b/llvm/test/CodeGen/AMDGPU/lower-mem-intrinsics.ll
index 8018e0cbf1cc3..0f4e790a6976f 100644
--- a/llvm/test/CodeGen/AMDGPU/lower-mem-intrinsics.ll
+++ b/llvm/test/CodeGen/AMDGPU/lower-mem-intrinsics.ll
@@ -189,8 +189,8 @@ define amdgpu_kernel void @min_size_large_static_memset_caller0(ptr addrspace(1)
 
 define amdgpu_kernel void @variable_memcpy_caller0(ptr addrspace(1) %dst, ptr addrspace(1) %src, i64 %n) #0 {
 ; OPT-LABEL: @variable_memcpy_caller0(
-; OPT-NEXT:    [[TMP1:%.*]] = udiv i64 [[N:%.*]], 16
-; OPT-NEXT:    [[TMP2:%.*]] = urem i64 [[N]], 16
+; OPT-NEXT:    [[TMP1:%.*]] = lshr i64 [[N:%.*]], 4
+; OPT-NEXT:    [[TMP2:%.*]] = and i64 [[N]], 15
 ; OPT-NEXT:    [[TMP3:%.*]] = sub i64 [[N]], [[TMP2]]
 ; OPT-NEXT:    [[TMP4:%.*]] = icmp ne i64 [[TMP1]], 0
 ; OPT-NEXT:    br i1 [[TMP4]], label [[LOOP_MEMCPY_EXPANSION:%.*]], label [[LOOP_MEMCPY_RESIDUAL_HEADER:%.*]]
@@ -225,8 +225,8 @@ define amdgpu_kernel void @variable_memcpy_caller0(ptr addrspace(1) %dst, ptr ad
 
 define amdgpu_kernel void @variable_memcpy_caller1(ptr addrspace(1) %dst, ptr addrspace(1) %src, i64 %n) #0 {
 ; OPT-LABEL: @variable_memcpy_caller1(
-; OPT-NEXT:    [[TMP1:%.*]] = udiv i64 [[N:%.*]], 16
-; OPT-NEXT:    [[TMP2:%.*]] = urem i64 [[N]], 16
+; OPT-NEXT:    [[TMP1:%.*]] = lshr i64 [[N:%.*]], 4
+; OPT-NEXT:    [[TMP2:%.*]] = and i64 [[N]], 15
 ; OPT-NEXT:    [[TMP3:%.*]] = sub i64 [[N]], [[TMP2]]
 ; OPT-NEXT:    [[TMP4:%.*]] = icmp ne i64 [[TMP1]], 0
 ; OPT-NEXT:    br i1 [[TMP4]], label [[LOOP_MEMCPY_EXPANSION:%.*]], label [[LOOP_MEMCPY_RESIDUAL_HEADER:%.*]]
@@ -261,8 +261,8 @@ define amdgpu_kernel void @variable_memcpy_caller1(ptr addrspace(1) %dst, ptr ad
 
 define amdgpu_kernel void @memcpy_multi_use_one_function(ptr addrspace(1) %dst0, ptr addrspace(1) %dst1, ptr addrspace(1) %src, i64 %n, i64 %m) #0 {
 ; OPT-LABEL: @memcpy_multi_use_one_function(
-; OPT-NEXT:    [[TMP1:%.*]] = udiv i64 [[N:%.*]], 16
-; OPT-NEXT:    [[TMP2:%.*]] = urem i64 [[N]], 16
+; OPT-NEXT:    [[TMP1:%.*]] = lshr i64 [[N:%.*]], 4
+; OPT-NEXT:    [[TMP2:%.*]] = and i64 [[N]], 15
 ; OPT-NEXT:    [[TMP3:%.*]] = sub i64 [[N]], [[TMP2]]
 ; OPT-NEXT:    [[TMP4:%.*]] = icmp ne i64 [[TMP1]], 0
 ; OPT-NEXT:    br i1 [[TMP4]], label [[LOOP_MEMCPY_EXPANSION2:%.*]], label [[LOOP_MEMCPY_RESIDUAL_HEADER5:%.*]]
@@ -286,8 +286,8 @@ define amdgpu_kernel void @memcpy_multi_use_one_function(ptr addrspace(1) %dst0,
 ; OPT-NEXT:    [[TMP15:%.*]] = icmp ult i64 [[TMP14]], [[TMP2]]
 ; OPT-NEXT:    br i1 [[TMP15]], label [[LOOP_MEMCPY_RESIDUAL4]], label [[POST_LOOP_MEMCPY_EXPANSION1:%.*]]
 ; OPT:       post-loop-memcpy-expansion1:
-; OPT-NEXT:    [[TMP16:%.*]] = udiv i64 [[M:%.*]], 16
-; OPT-NEXT:    [[TMP17:%.*]] = urem i64 [[M]], 16
+; OPT-NEXT:    [[TMP16:%.*]] = lshr i64 [[M:%.*]], 4
+; OPT-NEXT:    [[TMP17:%.*]] = and i64 [[M]], 15
 ; OPT-NEXT:    [[TMP18:%.*]] = sub i64 [[M]], [[TMP17]]
 ; OPT-NEXT:    [[TMP19:%.*]] = icmp ne i64 [[TMP16]], 0
 ; OPT-NEXT:    br i1 [[TMP19]], label [[LOOP_MEMCPY_EXPANSION:%.*]], label [[LOOP_MEMCPY_RESIDUAL_HEADER:%.*]]
@@ -326,8 +326,8 @@ define amdgpu_kernel void @memcpy_multi_use_one_function(ptr addrspace(1) %dst0,
 
 define amdgpu_kernel void @memcpy_alt_type(ptr addrspace(1) %dst, ptr addrspace(3) %src, i32 %n) #0 {
 ; OPT-LABEL: @memcpy_alt_type(
-; OPT-NEXT:    [[TMP1:%.*]] = udiv i32 [[N:%.*]], 8
-; OPT-NEXT:    [[TMP2:%.*]] = urem i32 [[N]], 8
+; OPT-NEXT:    [[TMP1:%.*]] = lshr i32 [[N:%.*]], 3
+; OPT-NEXT:    [[TMP2:%.*]] = and i32 [[N]], 7
 ; OPT-NEXT:    [[TMP3:%.*]] = sub i32 [[N]], [[TMP2]]
 ; OPT-NEXT:    [[TMP4:%.*]] = icmp ne i32 [[TMP1]], 0
 ; OPT-NEXT:    br i1 [[TMP4]], label [[LOOP_MEMCPY_EXPANSION:%.*]], label [[LOOP_MEMCPY_RESIDUAL_HEADER:%.*]]
@@ -363,8 +363,8 @@ define amdgpu_kernel void @memcpy_alt_type(ptr addrspace(1) %dst, ptr addrspace(
 ; One of the uses in the function should be expanded, the other left alone.
 define amdgpu_kernel void @memcpy_multi_use_one_function_keep_small(ptr addrspace(1) %dst0, ptr addrspace(1) %dst1, ptr addrspace(1) %src, i64 %n) #0 {
 ; MAX1024-LABEL: @memcpy_multi_use_one_function_keep_small(
-; MAX1024-NEXT:    [[TMP1:%.*]] = udiv i64 [[N:%.*]], 16
-; MAX1024-NEXT:    [[TMP2:%.*]] = urem i64 [[N]], 16
+; MAX1024-NEXT:    [[TMP1:%.*]] = lshr i64 [[N:%.*]], 4
+; MAX1024-NEXT:    [[TMP2:%.*]] = and i64 [[N]], 15
 ; MAX1024-NEXT:    [[TMP3:%.*]] = sub i64 [[N]], [[TMP2]]
 ; MAX1024-NEXT:    [[TMP4:%.*]] = icmp ne i64 [[TMP1]], 0
 ; MAX1024-NEXT:    br i1 [[TMP4]], label [[LOOP_MEMCPY_EXPANSION:%.*]], label [[LOOP_MEMCPY_RESIDUAL_HEADER:%.*]]
@@ -395,8 +395,8 @@ define amdgpu_kernel void @memcpy_multi_use_one_function_keep_small(ptr addrspac
 ; MAX1024-NEXT:    br i1 [[TMP16]], label [[LOOP_MEMCPY_RESIDUAL]], label [[POST_LOOP_MEMCPY_EXPANSION]]
 ;
 ; ALL-LABEL: @memcpy_multi_use_one_function_keep_small(
-; ALL-NEXT:    [[TMP1:%.*]] = udiv i64 [[N:%.*]], 16
-; ALL-NEXT:    [[TMP2:%.*]] = urem i64 [[N]], 16
+; ALL-NEXT:    [[TMP1:%.*]] = lshr i64 [[N:%.*]], 4
+; ALL-NEXT:    [[TMP2:%.*]] = and i64 [[N]], 15
 ; ALL-NEXT:    [[TMP3:%.*]] = sub i64 [[N]], [[TMP2]]
 ; ALL-NEXT:    [[TMP4:%.*]] = icmp ne i64 [[TMP1]], 0
 ; ALL-NEXT:    br i1 [[TMP4]], label [[LOOP_MEMCPY_EXPANSION:%.*]], label [[LOOP_MEMCPY_RESIDUAL_HEADER:%.*]]
@@ -909,8 +909,8 @@ define amdgpu_kernel void @memcpy_private_align2_private_align2_1027(ptr addrspa
 
 define amdgpu_kernel void @memcpy_global_align4_global_align4_variable(ptr addrspace(1) %dst, ptr addrspace(1) %src, i64 %n) #0 {
 ; OPT-LABEL: @memcpy_global_align4_global_align4_variable(
-; OPT-NEXT:    [[TMP1:%.*]] = udiv i64 [[N:%.*]], 16
-; OPT-NEXT:    [[TMP2:%.*]] = urem i64 [[N]], 16
+; OPT-NEXT:    [[TMP1:%.*]] = lshr i64 [[N:%.*]], 4
+; OPT-NEXT:    [[TMP2:%.*]] = and i64 [[N]], 15
 ; OPT-NEXT:    [[TMP3:%.*]] = sub i64 [[N]], [[TMP2]]
 ; OPT-NEXT:    [[TMP4:%.*]] = icmp ne i64 [[TMP1]], 0
 ; OPT-NEXT:    br i1 [[TMP4]], label [[LOOP_MEMCPY_EXPANSION:%.*]], label [[LOOP_MEMCPY_RESIDUAL_HEADER:%.*]]
@@ -945,8 +945,8 @@ define amdgpu_kernel void @memcpy_global_align4_global_align4_variable(ptr addrs
 
 define amdgpu_kernel void @memcpy_global_align2_global_align2_variable(ptr addrspace(1) %dst, ptr addrspace(1) %src, i64 %n) #0 {
 ; OPT-LABEL: @memcpy_global_align2_global_align2_variable(
-; OPT-NEXT:    [[TMP1:%.*]] = udiv i64 [[N:%.*]], 2
-; OPT-NEXT:    [[TMP2:%.*]] = urem i64 [[N]], 2
+; OPT-NEXT:    [[TMP1:%.*]] = lshr i64 [[N:%.*]], 1
+; OPT-NEXT:    [[TMP2:%.*]] = and i64 [[N]], 1
 ; OPT-NEXT:    [[TMP3:%.*]] = sub i64 [[N]], [[TMP2]]
 ; OPT-NEXT:    [[TMP4:%.*]] = icmp ne i64 [[TMP1]], 0
 ; OPT-NEXT:    br i1 [[TMP4]], label [[LOOP_MEMCPY_EXPANSION:%.*]], label [[LOOP_MEMCPY_RESIDUAL_HEADER:%.*]]
@@ -981,8 +981,8 @@ define amdgpu_kernel void @memcpy_global_align2_global_align2_variable(ptr addrs
 
 define amdgpu_kernel void @memcpy_global_align1_global_align1_variable(ptr addrspace(1) %dst, ptr addrspace(1) %src, i64 %n) #0 {
 ; OPT-LABEL: @memcpy_global_align1_global_align1_variable(
-; OPT-NEXT:    [[TMP1:%.*]] = udiv i64 [[N:%.*]], 16
-; OPT-NEXT:    [[TMP2:%.*]] = urem i64 [[N]], 16
+; OPT-NEXT:    [[TMP1:%.*]] = lshr i64 [[N:%.*]], 4
+; OPT-NEXT:    [[TMP2:%.*]] = and i64 [[N]], 15
 ; OPT-NEXT:    [[TMP3:%.*]] = sub i64 [[N]], [[TMP2]]
 ; OPT-NEXT:    [[TMP4:%.*]] = icmp ne i64 [[TMP1]], 0
 ; OPT-NEXT:    br i1 [[TMP4]], label [[LOOP_MEMCPY_EXPANSION:%.*]], label [[LOOP_MEMCPY_RESIDUAL_HEADER:%.*]]
@@ -1017,8 +1017,8 @@ define amdgpu_kernel void @memcpy_global_align1_global_align1_variable(ptr addrs
 
 define amdgpu_kernel void @memcpy_local_align4_local_align4_variable(ptr addrspace(3) %dst, ptr addrspace(3) %src, i32 %n) #0 {
 ; OPT-LABEL: @memcpy_local_align4_local_align4_variable(
-; OPT-NEXT:    [[TMP1:%.*]] = udiv i32 [[N:%.*]], 8
-; OPT-NEXT:    [[TMP2:%.*]] = urem i32 [[N]], 8
+; OPT-NEXT:    [[TMP1:%.*]] = lshr i32 [[N:%.*]], 3
+; OPT-NEXT:    [[TMP2:%.*]] = and i32 [[N]], 7
 ; OPT-NEXT:    [[TMP3:%.*]] = sub i32 [[N]], [[TMP2]]
 ; OPT-NEXT:    [[TMP4:%.*]] = icmp ne i32 [[TMP1]], 0
 ; OPT-NEXT:    br i1 [[TMP4]], label [[LOOP_MEMCPY_EXPANSION:%.*]], label [[LOOP_MEMCPY_RESIDUAL_HEADER:%.*]]
@@ -1053,8 +1053,8 @@ define amdgpu_kernel void @memcpy_local_align4_local_align4_variable(ptr addrspa
 
 define amdgpu_kernel void @memcpy_local_align2_local_align2_variable(ptr addrspace(3) %dst, ptr addrspace(3) %src, i32 %n) #0 {
 ; OPT-LABEL: @memcpy_local_align2_local_align2_variable(
-; OPT-NEXT:    [[TMP1:%.*]] = udiv i32 [[N:%.*]], 2
-; OPT-NEXT:    [[TMP2:%.*]] = urem i32 [[N]], 2
+; OPT-NEXT:    [[TMP1:%.*]] = lshr i32 [[N:%.*]], 1
+; OPT-NEXT:    [[TMP2:%.*]] = and i32 [[N]], 1
 ; OPT-NEXT:    [[TMP3:%.*]] = sub i32 [[N]], [[TMP2]]
 ; OPT-NEXT:    [[TMP4:%.*]] = icmp ne i32 [[TMP1]], 0
 ; OPT-NEXT:    br i1 [[TMP4]], label [[LOOP_MEMCPY_EXPANSION:%.*]], label [[LOOP_MEMCPY_RESIDUAL_HEADER:%.*]]
@@ -1089,8 +1089,8 @@ define amdgpu_kernel void @memcpy_local_align2_local_align2_variable(ptr addrspa
 
 define amdgpu_kernel void @memcpy_local_align1_local_align1_variable(ptr addrspace(3) %dst, ptr addrspace(3) %src, i32 %n) #0 {
 ; OPT-LABEL: @memcpy_local_align1_local_align1_variable(
-; OPT-NEXT:    [[TMP1:%.*]] = udiv i32 [[N:%.*]], 8
-; OPT-NEXT:    [[TMP2:%.*]] = urem i32 [[N]], 8
+; OPT-NEXT:    [[TMP1:%.*]] = lshr i32 [[N:%.*]], 3
+; OPT-NEXT:    [[TMP2:%.*]] = and i32 [[N]], 7
 ; OPT-NEXT:    [[TMP3:%.*]] = sub i32 [[N]], [[TMP2]]
 ; OPT-NEXT:    [[TMP4:%.*]] = icmp ne i32 [[TMP1]], 0
 ; OPT-NEXT:    br i1 [[TMP4]], label [[LOOP_MEMCPY_EXPANSION:%.*]], label [[LOOP_MEMCPY_RESIDUAL_HEADER:%.*]]
@@ -1125,8 +1125,8 @@ define amdgpu_kernel void @memcpy_local_align1_local_align1_variable(ptr addrspa
 
 define amdgpu_kernel void @memcpy_local_align4_global_align4_variable(ptr addrspace(3) %dst, ptr addrspace(1) %src, i32 %n) #0 {
 ; OPT-LABEL: @memcpy_local_align4_global_align4_variable(
-; OPT-NEXT:    [[TMP1:%.*]] = udiv i32 [[N:%.*]], 8
-; OPT-NEXT:    [[TMP2:%.*]] = urem i32 [[N]], 8
+; OPT-NEXT:    [[TMP1:%.*]] = lshr i32 [[N:%.*]], 3
+; OPT-NEXT:    [[TMP2:%.*]] = and i32 [[N]], 7
 ; OPT-NEXT:    [[TMP3:%.*]] = sub i32 [[N]], [[TMP2]]
 ; OPT-NEXT:    [[TMP4:%.*]] = icmp ne i32 [[TMP1]], 0
 ; OPT-NEXT:    br i1 [[TMP4]], label [[LOOP_MEMCPY_EXPANSION:%.*]], label [[LOOP_MEMCPY_RESIDUAL_HEADER:%.*]]
@@ -1161,8 +1161,8 @@ define amdgpu_kernel void @memcpy_local_align4_global_align4_variable(ptr addrsp
 
 define amdgpu_kernel void @memcpy_global_align4_local_align4_variable(ptr addrspace(1) %dst, ptr addrspace(3) %src, i32 %n) #0 {
 ; OPT-LABEL: @memcpy_global_align4_local_align4_variable(
-; OPT-NEXT:    [[TMP1:%.*]] = udiv i32 [[N:%.*]], 8
-; OPT-NEXT:    [[TMP2:%.*]] = urem i32 [[N]], 8
+; OPT-NEXT:    [[TMP1:%.*]] = lshr i32 [[N:%.*]], 3
+; OPT-NEXT:    [[TMP2:%.*]] = and i32 [[N]], 7
 ; OPT-NEXT:    [[TMP3:%.*]] = sub i32 [[N]], [[TMP2]]
 ; OPT-NEXT:    [[TMP4:%.*]] = icmp ne i32 [[TMP1]], 0
 ; OPT-NEXT:    br i1 [[TMP4]], label [[LOOP_MEMCPY_EXPANSION:%.*]], label [[LOOP_MEMCPY_RESIDUAL_HEADER:%.*]]
@@ -1488,9 +1488,9 @@ define amdgpu_kernel void @memmove_private_align1_global_align1(ptr addrspace(5)
 ; ALL:       load-store-loop:
 ; ALL-NEXT:    [[LOOP_INDEX:%.*]] = phi i64 [ 0, [[TMP0:%.*]] ], [ [[TMP4:%.*]], [[LOAD_STORE_LOOP]] ]
 ; ALL-NEXT:    [[TMP1:%.*]] = getelementptr inbounds <4 x i32>, ptr addrspace(1) [[SRC:%.*]], i64 [[LOOP_INDEX]]
-; ALL-NEXT:    [[TMP2:%.*]] = load <4 x i32>, ptr addrspace(1) [[TMP1]], align 1, !alias.scope !0
+; ALL-NEXT:    [[TMP2:%.*]] = load <4 x i32>, ptr addrspace(1) [[TMP1]], align 1, !alias.scope [[META0:![0-9]+]]
 ; ALL-NEXT:    [[TMP3:%.*]] = getelementptr inbounds <4 x i32>, ptr addrspace(5) [[DST:%.*]], i64 [[LOOP_INDEX]]
-; ALL-NEXT:    store <4 x i32> [[TMP2]], ptr addrspace(5) [[TMP3]], align 1, !noalias !0
+; ALL-NEXT:    store <4 x i32> [[TMP2]], ptr addrspace(5) [[TMP3]], align 1, !noalias [[META0]]
 ; ALL-NEXT:    [[TMP4]] = add i64 [[LOOP_INDEX]], 1
 ; ALL-NEXT:    [[TMP5:%.*]] = icmp ult i64 [[TMP4]], 16
 ; ALL-NEXT:    br i1 [[TMP5]], label [[LOAD_STORE_LOOP]], label [[MEMCPY_SPLIT:%.*]]
@@ -1511,9 +1511,9 @@ define amdgpu_kernel void @memmove_global_align1_private_align1(ptr addrspace(1)
 ; ALL:       load-store-loop:
 ; ALL-NEXT:    [[LOOP_INDEX:%.*]] = phi i64 [ 0, [[TMP0:%.*]] ], [ [[TMP4:%.*]], [[LOAD_STORE_LOOP]] ]
 ; ALL-NEXT:    [[TMP1:%.*]] = getelementptr inbounds <4 x i32>, ptr addrspace(5) [[SRC:%.*]], i64 [[LOOP_INDEX]]
-; ALL-NEXT:    [[TMP2:%.*]] = load <4 x i32>, ptr addrspace(5) [[TMP1]], align 1, !alias.scope !3
+; ALL-NEXT:    [[TMP2:%.*]] = load <4 x i32>, ptr addrspace(5) [[TMP1]], align 1, !alias.scope [[META3:![0-9]+]]
 ; ALL-NEXT:    [[TMP3:%.*]] = getelementptr inbounds <4 x i32>, ptr addrspace(1) [[DST:%.*]], i64 [[LOOP_INDEX]]
-; ALL-NEXT:    store <4 x i32> [[TMP2]], ptr addrspace(1) [[TMP3]], align 1, !noalias !3
+; ALL-NEXT:    store <4 x i32> [[TMP2]], ptr addrspace(1) [[TMP3]], align 1, !noalias [[META3]]
 ; ALL-NEXT:    [[TMP4]] = add i64 [[LOOP_INDEX]], 1
 ; ALL-NEXT:    [[TMP5:%.*]] = icmp ult i64 [[TMP4]], 16
 ; ALL-NEXT:    br i1 [[TMP5]], label [[LOAD_STORE_LOOP]], label [[MEMCPY_SPLIT:%.*]]
@@ -1561,9 +1561,9 @@ define amdgpu_kernel void @memmove_local_align1_private_align1(ptr addrspace(3)
 ; ALL:       load-store-loop:
 ; ALL-NEXT:    [[LOOP_INDEX:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP4:%.*]], [[LOAD_STORE_LOOP]] ]
 ; ALL-NEXT:    [[TMP1:%.*]] = getelementptr inbounds <2 x i32>, ptr addrspace(5) [[SRC:%.*]], i32 [[LOOP_INDEX]]
-; ALL-NEXT:    [[TMP2:%.*]] = load <2 x i32>, ptr addrspace(5) [[TMP1]], align 1, !alias.scope !6
+; ALL-NEXT:    [[TMP2:%.*]] = load <2 x i32>, ptr addrspace(5) [[TMP1]], align 1, !alias.scope [[META6:![0-9]+]]
 ; ALL-NEXT:    [[TMP3:%.*]] = getelementptr inbounds <2 x i32>, ptr addrspace(3) [[DST:%.*]], i32 [[LOOP_INDEX]]
-; ALL-NEXT:    store <2 x i32> [[TMP2]], ptr addrspace(3) [[TMP3]], align 1, !noalias !6
+; ALL-NEXT:    store <2 x i32> [[TMP2]], ptr addrspace(3) [[TMP3]], align 1, !noalias [[META6]]
 ; ALL-NEXT:    [[TMP4]] = add i32 [[LOOP_INDEX]], 1
 ; ALL-NEXT:    [[TMP5:%.*]] = icmp ult i32 [[TMP4]], 32
 ; ALL-NEXT:    br i1 [[TMP5]], label [[LOAD_STORE_LOOP]], label [[MEMCPY_SPLIT:%.*]]
@@ -1576,17 +1576,17 @@ define amdgpu_kernel void @memmove_local_align1_private_align1(ptr addrspace(3)
 
 define amdgpu_kernel void @memmove_local_align1_private_align1_unknown_size(ptr addrspace(3) %dst, ptr addrspace(5) %src, i32 %size) {
 ; MAX1024-LABEL: @memmove_local_align1_private_align1_unknown_size(
-; MAX1024-NEXT:    [[TMP1:%.*]] = udiv i32 [[SIZE:%.*]], 8
-; MAX1024-NEXT:    [[TMP2:%.*]] = urem i32 [[SIZE]], 8
+; MAX1024-NEXT:    [[TMP1:%.*]] = lshr i32 [[SIZE:%.*]], 3
+; MAX1024-NEXT:    [[TMP2:%.*]] = and i32 [[SIZE]], 7
 ; MAX1024-NEXT:    [[TMP3:%.*]] = sub i32 [[SIZE]], [[TMP2]]
 ; MAX1024-NEXT:    [[TMP4:%.*]] = icmp ne i32 [[TMP1]], 0
 ; MAX1024-NEXT:    br i1 [[TMP4]], label [[LOOP_MEMCPY_EXPANSION:%.*]], label [[LOOP_MEMCPY_RESIDUAL_HEADER:%.*]]
 ; MAX1024:       loop-memcpy-expansion:
 ; MAX1024-NEXT:    [[LOOP_INDEX:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP8:%.*]], [[LOOP_MEMCPY_EXPANSION]] ]
 ; MAX1024-NEXT:    [[TMP5:%.*]] = getelementptr inbounds <2 x i32>, ptr addrspace(5) [[SRC:%.*]], i32 [[LOOP_INDEX]]
-; MAX1024-NEXT:    [[TMP6:%.*]] = load <2 x i32>, ptr addrspace(5) [[TMP5]], align 1, !alias.scope !0
+; MAX1024-NEXT:    [[TMP6:%.*]] = load <2 x i32>, ptr addrspace(5) [[TMP5]], align 1, !alias.scope [[META0:![0-9]+]]
 ; MAX1024-NEXT:    [[TMP7:%.*]] = getelementptr inbounds <2 x i32>, ptr addrspace(3) [[DST:%.*]], i32 [[LOOP_INDEX]]
-; MAX1024-NEXT:    store <2 x i32> [[TMP6]], ptr addrspace(3) [[TMP7]], align 1, !noalias !0
+; MAX1024-NEXT:    store <2 x i32> [[TMP6]], ptr addrspace(3) [[TMP7]], align 1, !noalias [[META0]]
 ; MAX1024-NEXT:    [[TMP8]] = add i32 [[LOOP_INDEX]], 1
 ; MAX1024-NEXT:    [[TMP9:%.*]] = icmp ult i32 [[TMP8]], [[TMP1]]
 ; MAX1024-NEXT:    br i1 [[TMP9]], label [[LOOP_MEMCPY_EXPANSION]], label [[LOOP_MEMCPY_RESIDUAL_HEADER]]
@@ -1594,9 +1594,9 @@ define amdgpu_kernel void @memmove_local_align1_private_align1_unknown_size(ptr
 ; MAX1024-NEXT:    [[RESIDUAL_LOOP_INDEX:%.*]] = phi i32 [ 0, [[LOOP_MEMCPY_RESIDUAL_HEADER]] ], [ [[TMP14:%.*]], [[LOOP_MEMCPY_RESIDUAL:%.*]] ]
 ; MAX1024-NEXT:    [[TMP10:%.*]] = add i32 [[TMP3]], [[RESIDUAL_LOOP_INDEX]]
 ; MAX1024-NEXT:    [[TMP11:%.*]] = getelementptr inbounds i8, ptr addrspace(5) [[SRC]], i32 [[TMP10]]
-; MAX1024-NEXT:    [[TMP12:%.*]] = load i8, ptr addrspace(5) [[TMP11]], align 1, !alias.scope !0
+; MAX1024-NEXT:    [[TMP12:%.*]] = load i8, ptr addrspace(5) [[TMP11]], align 1, !alias.scope [[META0]]
 ; MAX1024-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[DST]], i32 [[TMP10]]
-; MAX1024-NEXT:    store i8 [[TMP12]], ptr addrspace(3) [[TMP13]], align 1, !noalias !0
+; MAX1024-NEXT:    store i8 [[TMP12]], ptr addrspace(3) [[TMP13]], align 1, !noalias [[META0]]
 ; MAX1024-NEXT:    [[TMP14]] = add i32 [[RESIDUAL_LOOP_INDEX]], 1
 ; MAX1024-NEXT:    [[TMP15:%.*]] = icmp ult i32 [[TMP14]], [[TMP2]]
 ; MAX1024-NEXT:    br i1 [[TMP15]], label [[LOOP_MEMCPY_RESIDUAL]], label [[POST_LOOP_MEMCPY_EXPANSION:%.*]]
@@ -1607,17 +1607,17 @@ define amdgpu_kernel void @memmove_local_align1_private_align1_unknown_size(ptr
 ; MAX1024-NEXT:    br i1 [[TMP16]], label [[LOOP_MEMCPY_RESIDUAL]], label [[POST_LOOP_MEMCPY_EXPANSION]]
 ;
 ; ALL-LABEL: @memmove_local_align1_private_align1_unknown_size(
-; ALL-NEXT:    [[TMP1:%.*]] = udiv i32 [[SIZE:%.*]], 8
-; ALL-NEXT:    [[TMP2:%.*]] = urem i32 [[SIZE]], 8
+; ALL-NEXT:    [[TMP1:%.*]] = lshr i32 [[SIZE:%.*]], 3
+; ALL-NEXT:    [[TMP2:%.*]] = and i32 [[SIZE]], 7
 ; ALL-NEXT:    [[TMP3:%.*]] = sub i32 [[SIZE]], [[TMP2]]
 ; ALL-NEXT:    [[TMP4:%.*]] = icmp ne i32 [[TMP1]], 0
 ; ALL-NEXT:    br i1 [[TMP4]], label [[LOOP_MEMCPY_EXPANSION:%.*]], label [[LOOP_MEMCPY_RESIDUAL_HEADER:%.*]]
 ; ALL:       loop-memcpy-expansion:
 ; ALL-NEXT:    [[LOOP_INDEX:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP8:%.*]], [[LOOP_MEMCPY_EXPANSION]] ]
 ; ALL-NEXT:    [[TMP5:%.*]] = getelementptr inbounds <2 x i32>, ptr addrspace(5) [[SRC:%.*]], i32 [[LOOP_INDEX]]
-; ALL-NEXT:    [[TMP6:%.*]] = load <2 x i32>, ptr addrspace(5) [[TMP5]], align 1, !alias.scope !9
+; ALL-NEXT:    [[TMP6:%.*]] = load <2 x i32>, ptr addrspace(5) [[TMP5]], align 1, !alias.scope [[META9:![0-9]+]]
 ; ALL-NEXT:    [[TMP7:%.*]] = getelementptr inbounds <2 x i32>, ptr addrspace(3) [[DST:%.*]], i32 [[LOOP_INDEX]]
-; ALL-NEXT:    store <2 x i32> [[TMP6]], ptr addrspace(3) [[TMP7]], align 1, !noalias !9
+; ALL-NEXT:    store <2 x i32> [[TMP6]], ptr addrspace(3) [[TMP7]], align 1, !noalias [[META9]]
 ; ALL-NEXT:    [[TMP8]] = add i32 [[LOOP_INDEX]], 1
 ; ALL-NEXT:    [[TMP9:%.*]] = icmp ult i32 [[TMP8]], [[TMP1]]
 ; ALL-NEXT:    br i1 [[TMP9]], label [[LOOP_MEMCPY_EXPANSION]], label [[LOOP_MEMCPY_RESIDUAL_HEADER]]
@@ -1625,9 +1625,9 @@ define amdgpu_kernel void @memmove_local_align1_private_align1_unknown_size(ptr
 ; ALL-NEXT:    [[RESIDUAL_LOOP_INDEX:%.*]] = phi i32 [ 0, [[LOOP_MEMCPY_RESIDUAL_HEADER]] ], [ [[TMP14:%.*]], [[LOOP_MEMCPY_RESIDUAL:%.*]] ]
 ; ALL-NEXT:    [[TMP10:%.*]] = add i32 [[TMP3]], [[RESIDUAL_LOOP_INDEX]]
 ; ALL-NEXT:    [[TMP11:%.*]] = getelementptr inbounds i8, ptr addrspace(5) [[SRC]], i32 [[TMP10]]
-; ALL-NEXT:    [[TMP12:%.*]] = load i8, ptr addrspace(5) [[TMP11]], align 1, !alias.scope !9
+; ALL-NEXT:    [[TMP12:%.*]] = load i8, ptr addrspace(5) [[TMP11]], align 1, !alias.scope [[META9]]
 ; ALL-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[DST]], i32 [[TMP10]]
-; ALL-NEXT:    store i8 [[TMP12]], ptr addrspace(3) [[TMP13]], align 1, !noalias !9
+; ALL-NEXT:    store i8 [[TMP12]], ptr addrspace(3) [[TMP13]], align 1, !noalias [[META9]]
 ; ALL-NEXT:    [[TMP14]] = add i32 [[RESIDUAL_LOOP_INDEX]], 1
 ; ALL-NEXT:    [[TMP15:%.*]] = icmp ult i32 [[TMP14]], [[TMP2]]
 ; ALL-NEXT:    br i1 [[TMP15]], label [[LOOP_MEMCPY_RESIDUAL]], label [[POST_LOOP_MEMCPY_EXPANSION:%.*]]
@@ -1651,9 +1651,9 @@ define amdgpu_kernel void @memmove_private_align1_local_align1(ptr addrspace(5)
 ; ALL:       load-store-loop:
 ; ALL-NEXT:    [[LOOP_INDEX:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP4:%.*]], [[LOAD_STORE_LOOP]] ]
 ; ALL-NEXT:    [[TMP1:%.*]] = getelementptr inbounds <2 x i32>, ptr addrspace(3) [[SRC:%.*]], i32 [[LOOP_INDEX]]
-; ALL-NEXT:    [[TMP2:%.*]] = load <2 x i32>, ptr addrspace(3) [[TMP1]], align 1, !alias.scope !12
+; ALL-NEXT:    [[TMP2:%.*]] = load <2 x i32>, ptr addrspace(3) [[TMP1]], align 1, !alias.scope [[META12:![0-9]+]]
 ; ALL-NEXT:    [[TMP3:%.*]] = getelementptr inbounds <2 x i32>, ptr addrspace(5) [[DST:%.*]], i32 [[LOOP_INDEX]]
-; ALL-NEXT:    store <2 x i32> [[TMP2]], ptr addrspace(5) [[TMP3]], align 1, !noalias !12
+; ALL-NEXT:    store <2 x i32> [[TMP2]], ptr addrspace(5) [[TMP3]], align 1, !noalias [[META12]]
 ; ALL-NEXT:    [[TMP4]] = add i32 [[LOOP_INDEX]], 1
 ; ALL-NEXT:    [[TMP5:%.*]] = icmp ult i32 [[TMP4]], 32
 ; ALL-NEXT:    br i1 [[TMP5]], label [[LOAD_STORE_LOOP]], label [[MEMCPY_SPLIT:%.*]]
@@ -1666,17 +1666,17 @@ define amdgpu_kernel void @memmove_private_align1_local_align1(ptr addrspace(5)
 
 define amdgpu_kernel void @memmove_private_align1_local_align1_unknown_size(ptr addrspace(5) %dst, ptr addrspace(3) %src, i32 %size) {
 ; MAX1024-LABEL: @memmove_private_align1_local_align1_unknown_size(
-; MAX1024-NEXT:    [[TMP1:%.*]] = udiv i32 [[SIZE:%.*]], 8
-; MAX1024-NEXT:    [[TMP2:%.*]] = urem i32 [[SIZE]], 8
+; MAX1024-NEXT:    [[TMP1:%.*]] = lshr i32 [[SIZE:%.*]], 3
+; MAX1024-NEXT:    [[TMP2:%.*]] = and i32 [[SIZE]], 7
 ; MAX1024-NEXT:    [[TMP3:%.*]] = sub i32 [[SIZE]], [[TMP2]]
 ; MAX1024-NEXT:    [[TMP4:%.*]] = icmp ne i32 [[TMP1]], 0
 ; MAX1024-NEXT:    br i1 [[TMP4]], label [[LOOP_MEMCPY_EXPANSION:%.*]], label [[LOOP_MEMCPY_RESIDUAL_HEADER:%.*]]
 ; MAX1024:       loop-memcpy-expansion:
 ; MAX1024-NEXT:    [[LOOP_INDEX:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP8:%.*]], [[LOOP_MEMCPY_EXPANSION]] ]
 ; MAX1024-NEXT:    [[TMP5:%.*]] = getelementptr inbounds <2 x i32>, ptr addrspace(3) [[SRC:%.*]], i32 [[LOOP_INDEX]]
-; MAX1024-NEXT:    [[TMP6:%.*]] = load <2 x i32>, ptr addrspace(3) [[TMP5]], align 1, !alias.scope !3
+; MAX1024-NEXT:    [[TMP6:%.*]] = load <2 x i32>, ptr addrspace(3) [[TMP5]], align 1, !alias.scope [[META3:![0-9]+]]
 ; MAX1024-NEXT:    [[TMP7:%.*]] = getelementptr inbounds <2 x i32>, ptr addrspace(5) [[DST:%.*]], i32 [[LOOP_INDEX]]
-; MAX1024-NEXT:    store <2 x i32> [[TMP6]], ptr addrspace(5) [[TMP7]], align 1, !noalias !3
+; MAX1024-NEXT:    store <2 x i32> [[TMP6]], ptr addrspace(5) [[TMP7]], align 1, !noalias [[META3]]
 ; MAX1024-NEXT:    [[TMP8]] = add i32 [[LOOP_INDEX]], 1
 ; MAX1024-NEXT:    [[TMP9:%.*]] = icmp ult i32 [[TMP8]], [[TMP1]]
 ; MAX1024-NEXT:    br i1 [[TMP9]], label [[LOOP_MEMCPY_EXPANSION]], label [[LOOP_MEMCPY_RESIDUAL_HEADER]]
@@ -1684,9 +1684,9 @@ define amdgpu_kernel void @memmove_private_align1_local_align1_unknown_size(ptr
 ; MAX1024-NEXT:    [[RESIDUAL_LOOP_INDEX:%.*]] = phi i32 [ 0, [[LOOP_MEMCPY_RESIDUAL_HEADER]] ], [ [[TMP14:%.*]], [[LOOP_MEMCPY_RESIDUAL:%.*]] ]
 ; MAX1024-NEXT:    [[TMP10:%.*]] = add i32 [[TMP3]], [[RESIDUAL_LOOP_INDEX]]
 ; MAX1024-NEXT:    [[TMP11:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[SRC]], i32 [[TMP10]]
-; MAX1024-NEXT:    [[TMP12:%.*]] = load i8, ptr addrspace(3) [[TMP11]], align 1, !alias.scope !3
+; MAX1024-NEXT:    [[TMP12:%.*]] = load i8, ptr addrspace(3) [[TMP11]], align 1, !alias.scope [[META3]]
 ; MAX1024-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i8, ptr addrspace(5) [[DST]], i32 [[TMP10]]
-; MAX1024-NEXT:    store i8 [[TMP12]], ptr addrspace(5) [[TMP13]], align 1, !noalias !3
+; MAX1024-NEXT:    store i8 [[TMP12]], ptr addrspace(5) [[TMP13]], align 1, !noalias [[META3]]
 ; MAX1024-NEXT:    [[TMP14]] = add i32 [[RESIDUAL_LOOP_INDEX]], 1
 ; MAX1024-NEXT:    [[TMP15:%.*]] = icmp ult i32 [[TMP14]], [[TMP2]]
 ; MAX1024-NEXT:    br i1 [[TMP15]], label [[LOOP_MEMCPY_RESIDUAL]], label [[POST_LOOP_MEMCPY_EXPANSION:%.*]]
@@ -1697,17 +1697,17 @@ define amdgpu_kernel void @memmove_private_align1_local_align1_unknown_size(ptr
 ; MAX1024-NEXT:    br i1 [[TMP16]], label [[LOOP_MEMCPY_RESIDUAL]], label [[POST_LOOP_MEMCPY_EXPANSION]]
 ;
 ; ALL-LABEL: @memmove_private_align1_local_align1_unknown_size(
-; ALL-NEXT:    [[TMP1:%.*]] = udiv i32 [[SIZE:%.*]], 8
-; ALL-NEXT:    [[TMP2:%.*]] = urem i32 [[SIZE]], 8
+; ALL-NEXT:    [[TMP1:%.*]] = lshr i32 [[SIZE:%.*]], 3
+; ALL-NEXT:    [[TMP2:%.*]] = and i32 [[SIZE]], 7
 ; ALL-NEXT:    [[TMP3:%.*]] = sub i32 [[SIZE]], [[TMP2]]
 ; ALL-NEXT:    [[TMP4:%.*]] = icmp ne i32 [[TMP1]], 0
 ; ALL-NEXT:    br i1 [[TMP4]], label [[LOOP_MEMCPY_EXPANSION:%.*]], label [[LOOP_MEMCPY_RESIDUAL_HEADER:%.*]]
 ; ALL:       loop-memcpy-expansion:
 ; ALL-NEXT:    [[LOOP_INDEX:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP8:%.*]], [[LOOP_MEMCPY_EXPANSION]] ]
 ; ALL-NEXT:    [[TMP5:%.*]] = getelementptr inbounds <2 x i32>, ptr addrspace(3) [[SRC:%.*]], i32 [[LOOP_INDEX]]
-; ALL-NEXT:    [[TMP6:%.*]] = load <2 x i32>, ptr addrspace(3) [[TMP5]], align 1, !alias.scope !15
+; ALL-NEXT:    [[TMP6:%.*]] = load <2 x i32>, ptr addrspace(3) [[TMP5]], align 1, !alias.scope [[META15:![0-9]+]]
 ; ALL-NEXT:    [[TMP7:%.*]] = getelementptr inbounds <2 x i32>, ptr addrspace(5) [[DST:%.*]], i32 [[LOOP_INDEX]]
-; ALL-NEXT:    store <2 x i32> [[TMP6]], ptr addrspace(5) [[TMP7]], align 1, !noalias !15
+; ALL-NEXT:    store <2 x i32> [[TMP6]], ptr addrspace(5) [[TMP7]], align 1, !noalias [[META15]]
 ; ALL-NEXT:    [[TMP8]] = add i32 [[LOOP_INDEX]], 1
 ; ALL-NEXT:    [[TMP9:%.*]] = icmp ult i32 [[TMP8]], [[TMP1]]
 ; ALL-NEXT:    br i1 [[TMP9]], label [[LOOP_MEMCPY_EXPANSION]], label [[LOOP_MEMCPY_RESIDUAL_HEADER]]
@@ -1715,9 +1715,9 @@ define amdgpu_kernel void @memmove_private_align1_local_align1_unknown_size(ptr
 ; ALL-NEXT:    [[RESIDUAL_LOOP_INDEX:%.*]] = phi i32 [ 0, [[LOOP_MEMCPY_RESIDUAL_HEADER]] ], [ [[TMP14:%.*]], [[LOOP_MEMCPY_RESIDUAL:%.*]] ]
 ; ALL-NEXT:    [[TMP10:%.*]] = add i32 [[TMP3]], [[RESIDUAL_LOOP_INDEX]]
 ; ALL-NEXT:    [[TMP11:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[SRC]], i32 [[TMP10]]
-; ALL-NEXT:    [[TMP12:%.*]] = load i8, ptr addrspace(3) [[TMP11]], align 1, !alias.scope !15
+; ALL-NEXT:    [[TMP12:%.*]] = load i8, ptr addrspace(3) [[TMP11]], align 1, !alias.scope [[META15]]
 ; ALL-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i8, ptr addrspace(5) [[DST]], i32 [[TMP10]]
-; ALL-NEXT:    store i8 [[TMP12]], ptr addrspace(5) [[TMP13]], align 1, !noalias !15
+; ALL-NEXT:    store i8 [[TMP12]], ptr addrspace(5) [[TMP13]], align 1, !noalias [[META15]]
 ; ALL-NEXT:    [[TMP14]] = add i32 [[RESIDUAL_LOOP_INDEX]], 1
 ; ALL-NEXT:    [[TMP15:%.*]] = icmp ult i32 [[TMP14]], [[TMP2]]
 ; ALL-NEXT:    br i1 [[TMP15]], label [[LOOP_MEMCPY_RESIDUAL]], label [[POST_LOOP_MEMCPY_EXPANSION:%.*]]
@@ -1731,5 +1731,49 @@ define amdgpu_kernel void @memmove_private_align1_local_align1_unknown_size(ptr
   ret void
 }
 
+define void @test_umin(i64 %0, i64 %idxprom, ptr %x, ptr %y) {
+; OPT-LABEL: @test_umin(
+; OPT-NEXT:  entry:
+; OPT-NEXT:    [[ARRAYIDX:%.*]] = getelementptr [32 x [8 x i64]], ptr [[Y:%.*]], i64 0, i64 [[IDXPROM:%.*]]
+; OPT-NEXT:    [[SPEC_SELECT:%.*]] = tail call i64 @llvm.umin.i64(i64 sub (i64 ptrtoint (ptr addrspacecast (ptr addrspace(4) inttoptr (i64 32 to ptr addrspace(4)) to ptr) to i64), i64 ptrtoint (ptr addrspacecast (ptr addrspace(4) null to ptr) to i64)), i64 56)
+; OPT-NEXT:    [[TMP1:%.*]] = lshr i64 [[SPEC_SELECT]], 4
+; OPT-NEXT:    [[TMP2:%.*]] = and i64 [[SPEC_SELECT]], 15
+; OPT-NEXT:    [[TMP3:%.*]] = sub i64 [[SPEC_SELECT]], [[TMP2]]
+; OPT-NEXT:    [[TMP4:%.*]] = icmp ne i64 [[TMP1]], 0
+; OPT-NEXT:    br i1 [[TMP4]], label [[LOOP_MEMCPY_EXPANSION:%.*]], label [[LOOP_MEMCPY_RESIDUAL_HEADER:%.*]]
+; OPT:       loop-memcpy-expansion:
+; OPT-NEXT:    [[LOOP_INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[TMP8:%.*]], [[LOOP_MEMCPY_EXPANSION]] ]
+; OPT-NEXT:    [[TMP5:%.*]] = getelementptr inbounds <4 x i32>, ptr [[X:%.*]], i64 [[LOOP_INDEX]]
+; OPT-NEXT:    [[TMP6:%.*]] = load <4 x i32>, ptr [[TMP5]], align 1
+; OPT-NEXT:    [[TMP7:%.*]] = getelementptr inbounds <4 x i32>, ptr [[ARRAYIDX]], i64 [[LOOP_INDEX]]
+; OPT-NEXT:    store <4 x i32> [[TMP6]], ptr [[TMP7]], align 1
+; OPT-NEXT:    [[TMP8]] = add i64 [[LOOP_INDEX]], 1
+; OPT-NEXT:    [[TMP9:%.*]] = icmp ult i64 [[TMP8]], [[TMP1]]
+; OPT-NEXT:    br i1 [[TMP9]], label [[LOOP_MEMCPY_EXPANSION]], label [[LOOP_MEMCPY_RESIDUAL_HEADER]]
+; OPT:       loop-memcpy-residual:
+; OPT-NEXT:    [[RESIDUAL_LOOP_INDEX:%.*]] = phi i64 [ 0, [[LOOP_MEMCPY_RESIDUAL_HEADER]] ], [ [[TMP14:%.*]], [[LOOP_MEMCPY_RESIDUAL:%.*]] ]
+; OPT-NEXT:    [[TMP10:%.*]] = add i64 [[TMP3]], [[RESIDUAL_LOOP_INDEX]]
+; OPT-NEXT:    [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[X]], i64 [[TMP10]]
+; OPT-NEXT:    [[TMP12:%.*]] = load i8, ptr [[TMP11]], align 1
+; OPT-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i8, ptr [[ARRAYIDX]], i64 [[TMP10]]
+; OPT-NEXT:    store i8 [[TMP12]], ptr [[TMP13]], align 1
+; OPT-NEXT:    [[TMP14]] = add i64 [[RESIDUAL_LOOP_INDEX]], 1
+; OPT-NEXT:    [[TMP15:%.*]] = icmp ult i64 [[TMP14]], [[TMP2]]
+; OPT-NEXT:    br i1 [[TMP15]], label [[LOOP_MEMCPY_RESIDUAL]], label [[POST_LOOP_MEMCPY_EXPANSION:%.*]]
+; OPT:       post-loop-memcpy-expansion:
+; OPT-NEXT:    ret void
+; OPT:       loop-memcpy-residual-header:
+; OPT-NEXT:    [[TMP16:%.*]] = icmp ne i64 [[TMP2]], 0
+; OPT-NEXT:    br i1 [[TMP16]], label [[LOOP_MEMCPY_RESIDUAL]], label [[POST_LOOP_MEMCPY_EXPANSION]]
+;
+entry:
+  %arrayidx = getelementptr [32 x [8 x i64]], ptr %y, i64 0, i64 %idxprom
+  %spec.select =  tail call i64 @llvm.umin.i64(i64 sub (i64 ptrtoint (ptr addrspacecast (ptr addrspace(4) inttoptr (i64 32 to ptr addrspace(4)) to ptr) to i64), i64 ptrtoint (ptr addrspacecast (ptr addrspace(4) null to ptr) to i64)), i64 56)
+  tail call void @llvm.memcpy.p0.p0.i64(ptr %arrayidx, ptr %x, i64 %spec.select, i1 false)
+  ret void
+}
+
+declare i64 @llvm.umin.i64(i64, i64)
+
 attributes #0 = { nounwind }
 attributes #1 = { argmemonly nounwind }
diff --git a/llvm/test/CodeGen/AMDGPU/memcpy-crash-issue63986.ll b/llvm/test/CodeGen/AMDGPU/memcpy-crash-issue63986.ll
index 8cf70eaf29e85..1dd18b4228fe5 100644
--- a/llvm/test/CodeGen/AMDGPU/memcpy-crash-issue63986.ll
+++ b/llvm/test/CodeGen/AMDGPU/memcpy-crash-issue63986.ll
@@ -59,8 +59,9 @@ define void @issue63986(i64 %0, i64 %idxprom) {
 ; CHECK-NEXT:    flat_store_byte v[8:9], v25 offset:12
 ; CHECK-NEXT:    s_cbranch_vccz .LBB0_2
 ; CHECK-NEXT:  ; %bb.3: ; %loop-memcpy-residual-header
-; CHECK-NEXT:    s_and_b32 s4, 32, 15
-; CHECK-NEXT:    s_mov_b32 s5, 0
+; CHECK-NEXT:    s_mov_b32 s4, 0
+; CHECK-NEXT:    s_mov_b32 s5, s4
+; CHECK-NEXT:    s_cmp_lg_u64 s[4:5], 0
 ; CHECK-NEXT:    s_cbranch_scc0 .LBB0_5
 ; CHECK-NEXT:  ; %bb.4:
 ; CHECK-NEXT:    ; implicit-def: $vgpr2_vgpr3