diff --git a/llvm/test/CodeGen/Thumb2/LowOverheadLoops/unpredload.ll b/llvm/test/CodeGen/Thumb2/LowOverheadLoops/unpredload.ll
index de741c088877e6..6b5b6b2b1b677e 100644
--- a/llvm/test/CodeGen/Thumb2/LowOverheadLoops/unpredload.ll
+++ b/llvm/test/CodeGen/Thumb2/LowOverheadLoops/unpredload.ll
@@ -5,7 +5,7 @@ define void @arm_cmplx_mag_squared_q15_mve(ptr %pSrc, ptr %pDst, i32 %blockSize)
 ; CHECK-LABEL: arm_cmplx_mag_squared_q15_mve:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    push {r7, lr}
-; CHECK-NEXT:    subs.w r12, r2, #8
+; CHECK-NEXT:    subs.w r3, r2, #8
 ; CHECK-NEXT:    dlstp.16 lr, r2
 ; CHECK-NEXT:  .LBB0_1: @ %do.body
 ; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
diff --git a/llvm/test/Instrumentation/HWAddressSanitizer/use-after-scope.ll b/llvm/test/Instrumentation/HWAddressSanitizer/use-after-scope.ll
index 35a20b14f2f918..82be1cc4496870 100644
--- a/llvm/test/Instrumentation/HWAddressSanitizer/use-after-scope.ll
+++ b/llvm/test/Instrumentation/HWAddressSanitizer/use-after-scope.ll
@@ -81,21 +81,21 @@ define dso_local i32 @standard_lifetime() local_unnamed_addr sanitize_hwaddress
 ; AARCH64-SCOPE-LABEL: @standard_lifetime(
 ; AARCH64-SCOPE-NEXT:    [[TMP1:%.*]] = call ptr @llvm.thread.pointer()
 ; AARCH64-SCOPE-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 48
-; AARCH64-SCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 4
+; AARCH64-SCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 ; AARCH64-SCOPE-NEXT:    [[TMP4:%.*]] = ashr i64 [[TMP3]], 3
-; AARCH64-SCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META1:![0-9]+]])
+; AARCH64-SCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META2:![0-9]+]])
 ; AARCH64-SCOPE-NEXT:    [[TMP6:%.*]] = call ptr @llvm.frameaddress.p0(i32 0)
 ; AARCH64-SCOPE-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[TMP6]] to i64
 ; AARCH64-SCOPE-NEXT:    [[TMP8:%.*]] = shl i64 [[TMP7]], 44
 ; AARCH64-SCOPE-NEXT:    [[TMP9:%.*]] = or i64 [[TMP5]], [[TMP8]]
 ; AARCH64-SCOPE-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP3]] to ptr
-; AARCH64-SCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 4
+; AARCH64-SCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 8
 ; AARCH64-SCOPE-NEXT:    [[TMP11:%.*]] = ashr i64 [[TMP3]], 56
 ; AARCH64-SCOPE-NEXT:    [[TMP12:%.*]] = shl nuw nsw i64 [[TMP11]], 12
 ; AARCH64-SCOPE-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], -1
 ; AARCH64-SCOPE-NEXT:    [[TMP14:%.*]] = add i64 [[TMP3]], 8
 ; AARCH64-SCOPE-NEXT:    [[TMP15:%.*]] = and i64 [[TMP14]], [[TMP13]]
-; AARCH64-SCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 4
+; AARCH64-SCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 8
 ; AARCH64-SCOPE-NEXT:    [[TMP16:%.*]] = or i64 [[TMP3]], 4294967295
 ; AARCH64-SCOPE-NEXT:    [[HWASAN_SHADOW:%.*]] = add i64 [[TMP16]], 1
 ; AARCH64-SCOPE-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[HWASAN_SHADOW]] to ptr
@@ -133,21 +133,21 @@ define dso_local i32 @standard_lifetime() local_unnamed_addr sanitize_hwaddress
 ; AARCH64-NOSCOPE-LABEL: @standard_lifetime(
 ; AARCH64-NOSCOPE-NEXT:    [[TMP1:%.*]] = call ptr @llvm.thread.pointer()
 ; AARCH64-NOSCOPE-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 48
-; AARCH64-NOSCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 4
+; AARCH64-NOSCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 ; AARCH64-NOSCOPE-NEXT:    [[TMP4:%.*]] = ashr i64 [[TMP3]], 3
-; AARCH64-NOSCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META1:![0-9]+]])
+; AARCH64-NOSCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META2:![0-9]+]])
 ; AARCH64-NOSCOPE-NEXT:    [[TMP6:%.*]] = call ptr @llvm.frameaddress.p0(i32 0)
 ; AARCH64-NOSCOPE-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[TMP6]] to i64
 ; AARCH64-NOSCOPE-NEXT:    [[TMP8:%.*]] = shl i64 [[TMP7]], 44
 ; AARCH64-NOSCOPE-NEXT:    [[TMP9:%.*]] = or i64 [[TMP5]], [[TMP8]]
 ; AARCH64-NOSCOPE-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP3]] to ptr
-; AARCH64-NOSCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 4
+; AARCH64-NOSCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 8
 ; AARCH64-NOSCOPE-NEXT:    [[TMP11:%.*]] = ashr i64 [[TMP3]], 56
 ; AARCH64-NOSCOPE-NEXT:    [[TMP12:%.*]] = shl nuw nsw i64 [[TMP11]], 12
 ; AARCH64-NOSCOPE-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], -1
 ; AARCH64-NOSCOPE-NEXT:    [[TMP14:%.*]] = add i64 [[TMP3]], 8
 ; AARCH64-NOSCOPE-NEXT:    [[TMP15:%.*]] = and i64 [[TMP14]], [[TMP13]]
-; AARCH64-NOSCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 4
+; AARCH64-NOSCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 8
 ; AARCH64-NOSCOPE-NEXT:    [[TMP16:%.*]] = or i64 [[TMP3]], 4294967295
 ; AARCH64-NOSCOPE-NEXT:    [[HWASAN_SHADOW:%.*]] = add i64 [[TMP16]], 1
 ; AARCH64-NOSCOPE-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[HWASAN_SHADOW]] to ptr
@@ -183,21 +183,21 @@ define dso_local i32 @standard_lifetime() local_unnamed_addr sanitize_hwaddress
 ; AARCH64-SHORT-SCOPE-LABEL: @standard_lifetime(
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP1:%.*]] = call ptr @llvm.thread.pointer()
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 48
-; AARCH64-SHORT-SCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 4
+; AARCH64-SHORT-SCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP4:%.*]] = ashr i64 [[TMP3]], 3
-; AARCH64-SHORT-SCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META1:![0-9]+]])
+; AARCH64-SHORT-SCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META2:![0-9]+]])
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP6:%.*]] = call ptr @llvm.frameaddress.p0(i32 0)
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[TMP6]] to i64
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP8:%.*]] = shl i64 [[TMP7]], 44
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP9:%.*]] = or i64 [[TMP5]], [[TMP8]]
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP3]] to ptr
-; AARCH64-SHORT-SCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 4
+; AARCH64-SHORT-SCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 8
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP11:%.*]] = ashr i64 [[TMP3]], 56
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP12:%.*]] = shl nuw nsw i64 [[TMP11]], 12
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], -1
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP14:%.*]] = add i64 [[TMP3]], 8
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP15:%.*]] = and i64 [[TMP14]], [[TMP13]]
-; AARCH64-SHORT-SCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 4
+; AARCH64-SHORT-SCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 8
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP16:%.*]] = or i64 [[TMP3]], 4294967295
 ; AARCH64-SHORT-SCOPE-NEXT:    [[HWASAN_SHADOW:%.*]] = add i64 [[TMP16]], 1
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[HWASAN_SHADOW]] to ptr
@@ -238,21 +238,21 @@ define dso_local i32 @standard_lifetime() local_unnamed_addr sanitize_hwaddress
 ; AARCH64-SHORT-NOSCOPE-LABEL: @standard_lifetime(
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP1:%.*]] = call ptr @llvm.thread.pointer()
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 48
-; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 4
+; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP4:%.*]] = ashr i64 [[TMP3]], 3
-; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META1:![0-9]+]])
+; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META2:![0-9]+]])
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP6:%.*]] = call ptr @llvm.frameaddress.p0(i32 0)
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[TMP6]] to i64
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP8:%.*]] = shl i64 [[TMP7]], 44
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP9:%.*]] = or i64 [[TMP5]], [[TMP8]]
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP3]] to ptr
-; AARCH64-SHORT-NOSCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 4
+; AARCH64-SHORT-NOSCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 8
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP11:%.*]] = ashr i64 [[TMP3]], 56
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP12:%.*]] = shl nuw nsw i64 [[TMP11]], 12
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], -1
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP14:%.*]] = add i64 [[TMP3]], 8
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP15:%.*]] = and i64 [[TMP14]], [[TMP13]]
-; AARCH64-SHORT-NOSCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 4
+; AARCH64-SHORT-NOSCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 8
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP16:%.*]] = or i64 [[TMP3]], 4294967295
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[HWASAN_SHADOW:%.*]] = add i64 [[TMP16]], 1
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[HWASAN_SHADOW]] to ptr
@@ -362,21 +362,21 @@ define dso_local i32 @standard_lifetime_optnone() local_unnamed_addr optnone noi
 ; AARCH64-SCOPE-LABEL: @standard_lifetime_optnone(
 ; AARCH64-SCOPE-NEXT:    [[TMP1:%.*]] = call ptr @llvm.thread.pointer()
 ; AARCH64-SCOPE-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 48
-; AARCH64-SCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 4
+; AARCH64-SCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 ; AARCH64-SCOPE-NEXT:    [[TMP4:%.*]] = ashr i64 [[TMP3]], 3
-; AARCH64-SCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META1]])
+; AARCH64-SCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META2]])
 ; AARCH64-SCOPE-NEXT:    [[TMP6:%.*]] = call ptr @llvm.frameaddress.p0(i32 0)
 ; AARCH64-SCOPE-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[TMP6]] to i64
 ; AARCH64-SCOPE-NEXT:    [[TMP8:%.*]] = shl i64 [[TMP7]], 44
 ; AARCH64-SCOPE-NEXT:    [[TMP9:%.*]] = or i64 [[TMP5]], [[TMP8]]
 ; AARCH64-SCOPE-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP3]] to ptr
-; AARCH64-SCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 4
+; AARCH64-SCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 8
 ; AARCH64-SCOPE-NEXT:    [[TMP11:%.*]] = ashr i64 [[TMP3]], 56
 ; AARCH64-SCOPE-NEXT:    [[TMP12:%.*]] = shl nuw nsw i64 [[TMP11]], 12
 ; AARCH64-SCOPE-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], -1
 ; AARCH64-SCOPE-NEXT:    [[TMP14:%.*]] = add i64 [[TMP3]], 8
 ; AARCH64-SCOPE-NEXT:    [[TMP15:%.*]] = and i64 [[TMP14]], [[TMP13]]
-; AARCH64-SCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 4
+; AARCH64-SCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 8
 ; AARCH64-SCOPE-NEXT:    [[TMP16:%.*]] = or i64 [[TMP3]], 4294967295
 ; AARCH64-SCOPE-NEXT:    [[HWASAN_SHADOW:%.*]] = add i64 [[TMP16]], 1
 ; AARCH64-SCOPE-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[HWASAN_SHADOW]] to ptr
@@ -414,21 +414,21 @@ define dso_local i32 @standard_lifetime_optnone() local_unnamed_addr optnone noi
 ; AARCH64-NOSCOPE-LABEL: @standard_lifetime_optnone(
 ; AARCH64-NOSCOPE-NEXT:    [[TMP1:%.*]] = call ptr @llvm.thread.pointer()
 ; AARCH64-NOSCOPE-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 48
-; AARCH64-NOSCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 4
+; AARCH64-NOSCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 ; AARCH64-NOSCOPE-NEXT:    [[TMP4:%.*]] = ashr i64 [[TMP3]], 3
-; AARCH64-NOSCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META1]])
+; AARCH64-NOSCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META2]])
 ; AARCH64-NOSCOPE-NEXT:    [[TMP6:%.*]] = call ptr @llvm.frameaddress.p0(i32 0)
 ; AARCH64-NOSCOPE-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[TMP6]] to i64
 ; AARCH64-NOSCOPE-NEXT:    [[TMP8:%.*]] = shl i64 [[TMP7]], 44
 ; AARCH64-NOSCOPE-NEXT:    [[TMP9:%.*]] = or i64 [[TMP5]], [[TMP8]]
 ; AARCH64-NOSCOPE-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP3]] to ptr
-; AARCH64-NOSCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 4
+; AARCH64-NOSCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 8
 ; AARCH64-NOSCOPE-NEXT:    [[TMP11:%.*]] = ashr i64 [[TMP3]], 56
 ; AARCH64-NOSCOPE-NEXT:    [[TMP12:%.*]] = shl nuw nsw i64 [[TMP11]], 12
 ; AARCH64-NOSCOPE-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], -1
 ; AARCH64-NOSCOPE-NEXT:    [[TMP14:%.*]] = add i64 [[TMP3]], 8
 ; AARCH64-NOSCOPE-NEXT:    [[TMP15:%.*]] = and i64 [[TMP14]], [[TMP13]]
-; AARCH64-NOSCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 4
+; AARCH64-NOSCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 8
 ; AARCH64-NOSCOPE-NEXT:    [[TMP16:%.*]] = or i64 [[TMP3]], 4294967295
 ; AARCH64-NOSCOPE-NEXT:    [[HWASAN_SHADOW:%.*]] = add i64 [[TMP16]], 1
 ; AARCH64-NOSCOPE-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[HWASAN_SHADOW]] to ptr
@@ -464,21 +464,21 @@ define dso_local i32 @standard_lifetime_optnone() local_unnamed_addr optnone noi
 ; AARCH64-SHORT-SCOPE-LABEL: @standard_lifetime_optnone(
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP1:%.*]] = call ptr @llvm.thread.pointer()
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 48
-; AARCH64-SHORT-SCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 4
+; AARCH64-SHORT-SCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP4:%.*]] = ashr i64 [[TMP3]], 3
-; AARCH64-SHORT-SCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META1]])
+; AARCH64-SHORT-SCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META2]])
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP6:%.*]] = call ptr @llvm.frameaddress.p0(i32 0)
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[TMP6]] to i64
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP8:%.*]] = shl i64 [[TMP7]], 44
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP9:%.*]] = or i64 [[TMP5]], [[TMP8]]
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP3]] to ptr
-; AARCH64-SHORT-SCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 4
+; AARCH64-SHORT-SCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 8
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP11:%.*]] = ashr i64 [[TMP3]], 56
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP12:%.*]] = shl nuw nsw i64 [[TMP11]], 12
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], -1
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP14:%.*]] = add i64 [[TMP3]], 8
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP15:%.*]] = and i64 [[TMP14]], [[TMP13]]
-; AARCH64-SHORT-SCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 4
+; AARCH64-SHORT-SCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 8
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP16:%.*]] = or i64 [[TMP3]], 4294967295
 ; AARCH64-SHORT-SCOPE-NEXT:    [[HWASAN_SHADOW:%.*]] = add i64 [[TMP16]], 1
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[HWASAN_SHADOW]] to ptr
@@ -519,21 +519,21 @@ define dso_local i32 @standard_lifetime_optnone() local_unnamed_addr optnone noi
 ; AARCH64-SHORT-NOSCOPE-LABEL: @standard_lifetime_optnone(
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP1:%.*]] = call ptr @llvm.thread.pointer()
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 48
-; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 4
+; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP4:%.*]] = ashr i64 [[TMP3]], 3
-; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META1]])
+; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META2]])
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP6:%.*]] = call ptr @llvm.frameaddress.p0(i32 0)
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[TMP6]] to i64
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP8:%.*]] = shl i64 [[TMP7]], 44
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP9:%.*]] = or i64 [[TMP5]], [[TMP8]]
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP3]] to ptr
-; AARCH64-SHORT-NOSCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 4
+; AARCH64-SHORT-NOSCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 8
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP11:%.*]] = ashr i64 [[TMP3]], 56
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP12:%.*]] = shl nuw nsw i64 [[TMP11]], 12
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], -1
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP14:%.*]] = add i64 [[TMP3]], 8
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP15:%.*]] = and i64 [[TMP14]], [[TMP13]]
-; AARCH64-SHORT-NOSCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 4
+; AARCH64-SHORT-NOSCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 8
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP16:%.*]] = or i64 [[TMP3]], 4294967295
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[HWASAN_SHADOW:%.*]] = add i64 [[TMP16]], 1
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[HWASAN_SHADOW]] to ptr
@@ -633,21 +633,21 @@ define dso_local i32 @multiple_lifetimes() local_unnamed_addr sanitize_hwaddress
 ; AARCH64-SCOPE-LABEL: @multiple_lifetimes(
 ; AARCH64-SCOPE-NEXT:    [[TMP1:%.*]] = call ptr @llvm.thread.pointer()
 ; AARCH64-SCOPE-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 48
-; AARCH64-SCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 4
+; AARCH64-SCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 ; AARCH64-SCOPE-NEXT:    [[TMP4:%.*]] = ashr i64 [[TMP3]], 3
-; AARCH64-SCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META1]])
+; AARCH64-SCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META2]])
 ; AARCH64-SCOPE-NEXT:    [[TMP6:%.*]] = call ptr @llvm.frameaddress.p0(i32 0)
 ; AARCH64-SCOPE-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[TMP6]] to i64
 ; AARCH64-SCOPE-NEXT:    [[TMP8:%.*]] = shl i64 [[TMP7]], 44
 ; AARCH64-SCOPE-NEXT:    [[TMP9:%.*]] = or i64 [[TMP5]], [[TMP8]]
 ; AARCH64-SCOPE-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP3]] to ptr
-; AARCH64-SCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 4
+; AARCH64-SCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 8
 ; AARCH64-SCOPE-NEXT:    [[TMP11:%.*]] = ashr i64 [[TMP3]], 56
 ; AARCH64-SCOPE-NEXT:    [[TMP12:%.*]] = shl nuw nsw i64 [[TMP11]], 12
 ; AARCH64-SCOPE-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], -1
 ; AARCH64-SCOPE-NEXT:    [[TMP14:%.*]] = add i64 [[TMP3]], 8
 ; AARCH64-SCOPE-NEXT:    [[TMP15:%.*]] = and i64 [[TMP14]], [[TMP13]]
-; AARCH64-SCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 4
+; AARCH64-SCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 8
 ; AARCH64-SCOPE-NEXT:    [[TMP16:%.*]] = or i64 [[TMP3]], 4294967295
 ; AARCH64-SCOPE-NEXT:    [[HWASAN_SHADOW:%.*]] = add i64 [[TMP16]], 1
 ; AARCH64-SCOPE-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[HWASAN_SHADOW]] to ptr
@@ -679,21 +679,21 @@ define dso_local i32 @multiple_lifetimes() local_unnamed_addr sanitize_hwaddress
 ; AARCH64-NOSCOPE-LABEL: @multiple_lifetimes(
 ; AARCH64-NOSCOPE-NEXT:    [[TMP1:%.*]] = call ptr @llvm.thread.pointer()
 ; AARCH64-NOSCOPE-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 48
-; AARCH64-NOSCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 4
+; AARCH64-NOSCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 ; AARCH64-NOSCOPE-NEXT:    [[TMP4:%.*]] = ashr i64 [[TMP3]], 3
-; AARCH64-NOSCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META1]])
+; AARCH64-NOSCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META2]])
 ; AARCH64-NOSCOPE-NEXT:    [[TMP6:%.*]] = call ptr @llvm.frameaddress.p0(i32 0)
 ; AARCH64-NOSCOPE-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[TMP6]] to i64
 ; AARCH64-NOSCOPE-NEXT:    [[TMP8:%.*]] = shl i64 [[TMP7]], 44
 ; AARCH64-NOSCOPE-NEXT:    [[TMP9:%.*]] = or i64 [[TMP5]], [[TMP8]]
 ; AARCH64-NOSCOPE-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP3]] to ptr
-; AARCH64-NOSCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 4
+; AARCH64-NOSCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 8
 ; AARCH64-NOSCOPE-NEXT:    [[TMP11:%.*]] = ashr i64 [[TMP3]], 56
 ; AARCH64-NOSCOPE-NEXT:    [[TMP12:%.*]] = shl nuw nsw i64 [[TMP11]], 12
 ; AARCH64-NOSCOPE-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], -1
 ; AARCH64-NOSCOPE-NEXT:    [[TMP14:%.*]] = add i64 [[TMP3]], 8
 ; AARCH64-NOSCOPE-NEXT:    [[TMP15:%.*]] = and i64 [[TMP14]], [[TMP13]]
-; AARCH64-NOSCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 4
+; AARCH64-NOSCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 8
 ; AARCH64-NOSCOPE-NEXT:    [[TMP16:%.*]] = or i64 [[TMP3]], 4294967295
 ; AARCH64-NOSCOPE-NEXT:    [[HWASAN_SHADOW:%.*]] = add i64 [[TMP16]], 1
 ; AARCH64-NOSCOPE-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[HWASAN_SHADOW]] to ptr
@@ -725,21 +725,21 @@ define dso_local i32 @multiple_lifetimes() local_unnamed_addr sanitize_hwaddress
 ; AARCH64-SHORT-SCOPE-LABEL: @multiple_lifetimes(
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP1:%.*]] = call ptr @llvm.thread.pointer()
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 48
-; AARCH64-SHORT-SCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 4
+; AARCH64-SHORT-SCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP4:%.*]] = ashr i64 [[TMP3]], 3
-; AARCH64-SHORT-SCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META1]])
+; AARCH64-SHORT-SCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META2]])
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP6:%.*]] = call ptr @llvm.frameaddress.p0(i32 0)
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[TMP6]] to i64
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP8:%.*]] = shl i64 [[TMP7]], 44
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP9:%.*]] = or i64 [[TMP5]], [[TMP8]]
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP3]] to ptr
-; AARCH64-SHORT-SCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 4
+; AARCH64-SHORT-SCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 8
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP11:%.*]] = ashr i64 [[TMP3]], 56
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP12:%.*]] = shl nuw nsw i64 [[TMP11]], 12
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], -1
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP14:%.*]] = add i64 [[TMP3]], 8
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP15:%.*]] = and i64 [[TMP14]], [[TMP13]]
-; AARCH64-SHORT-SCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 4
+; AARCH64-SHORT-SCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 8
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP16:%.*]] = or i64 [[TMP3]], 4294967295
 ; AARCH64-SHORT-SCOPE-NEXT:    [[HWASAN_SHADOW:%.*]] = add i64 [[TMP16]], 1
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[HWASAN_SHADOW]] to ptr
@@ -774,21 +774,21 @@ define dso_local i32 @multiple_lifetimes() local_unnamed_addr sanitize_hwaddress
 ; AARCH64-SHORT-NOSCOPE-LABEL: @multiple_lifetimes(
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP1:%.*]] = call ptr @llvm.thread.pointer()
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 48
-; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 4
+; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP4:%.*]] = ashr i64 [[TMP3]], 3
-; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META1]])
+; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META2]])
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP6:%.*]] = call ptr @llvm.frameaddress.p0(i32 0)
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[TMP6]] to i64
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP8:%.*]] = shl i64 [[TMP7]], 44
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP9:%.*]] = or i64 [[TMP5]], [[TMP8]]
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP3]] to ptr
-; AARCH64-SHORT-NOSCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 4
+; AARCH64-SHORT-NOSCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 8
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP11:%.*]] = ashr i64 [[TMP3]], 56
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP12:%.*]] = shl nuw nsw i64 [[TMP11]], 12
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], -1
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP14:%.*]] = add i64 [[TMP3]], 8
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP15:%.*]] = and i64 [[TMP14]], [[TMP13]]
-; AARCH64-SHORT-NOSCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 4
+; AARCH64-SHORT-NOSCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 8
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP16:%.*]] = or i64 [[TMP3]], 4294967295
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[HWASAN_SHADOW:%.*]] = add i64 [[TMP16]], 1
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[HWASAN_SHADOW]] to ptr
@@ -893,21 +893,21 @@ define dso_local i32 @unreachable_exit() local_unnamed_addr sanitize_hwaddress {
 ; AARCH64-SCOPE-LABEL: @unreachable_exit(
 ; AARCH64-SCOPE-NEXT:    [[TMP1:%.*]] = call ptr @llvm.thread.pointer()
 ; AARCH64-SCOPE-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 48
-; AARCH64-SCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 4
+; AARCH64-SCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 ; AARCH64-SCOPE-NEXT:    [[TMP4:%.*]] = ashr i64 [[TMP3]], 3
-; AARCH64-SCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META1]])
+; AARCH64-SCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META2]])
 ; AARCH64-SCOPE-NEXT:    [[TMP6:%.*]] = call ptr @llvm.frameaddress.p0(i32 0)
 ; AARCH64-SCOPE-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[TMP6]] to i64
 ; AARCH64-SCOPE-NEXT:    [[TMP8:%.*]] = shl i64 [[TMP7]], 44
 ; AARCH64-SCOPE-NEXT:    [[TMP9:%.*]] = or i64 [[TMP5]], [[TMP8]]
 ; AARCH64-SCOPE-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP3]] to ptr
-; AARCH64-SCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 4
+; AARCH64-SCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 8
 ; AARCH64-SCOPE-NEXT:    [[TMP11:%.*]] = ashr i64 [[TMP3]], 56
 ; AARCH64-SCOPE-NEXT:    [[TMP12:%.*]] = shl nuw nsw i64 [[TMP11]], 12
 ; AARCH64-SCOPE-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], -1
 ; AARCH64-SCOPE-NEXT:    [[TMP14:%.*]] = add i64 [[TMP3]], 8
 ; AARCH64-SCOPE-NEXT:    [[TMP15:%.*]] = and i64 [[TMP14]], [[TMP13]]
-; AARCH64-SCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 4
+; AARCH64-SCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 8
 ; AARCH64-SCOPE-NEXT:    [[TMP16:%.*]] = or i64 [[TMP3]], 4294967295
 ; AARCH64-SCOPE-NEXT:    [[HWASAN_SHADOW:%.*]] = add i64 [[TMP16]], 1
 ; AARCH64-SCOPE-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[HWASAN_SHADOW]] to ptr
@@ -950,21 +950,21 @@ define dso_local i32 @unreachable_exit() local_unnamed_addr sanitize_hwaddress {
 ; AARCH64-NOSCOPE-LABEL: @unreachable_exit(
 ; AARCH64-NOSCOPE-NEXT:    [[TMP1:%.*]] = call ptr @llvm.thread.pointer()
 ; AARCH64-NOSCOPE-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 48
-; AARCH64-NOSCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 4
+; AARCH64-NOSCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 ; AARCH64-NOSCOPE-NEXT:    [[TMP4:%.*]] = ashr i64 [[TMP3]], 3
-; AARCH64-NOSCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META1]])
+; AARCH64-NOSCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META2]])
 ; AARCH64-NOSCOPE-NEXT:    [[TMP6:%.*]] = call ptr @llvm.frameaddress.p0(i32 0)
 ; AARCH64-NOSCOPE-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[TMP6]] to i64
 ; AARCH64-NOSCOPE-NEXT:    [[TMP8:%.*]] = shl i64 [[TMP7]], 44
 ; AARCH64-NOSCOPE-NEXT:    [[TMP9:%.*]] = or i64 [[TMP5]], [[TMP8]]
 ; AARCH64-NOSCOPE-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP3]] to ptr
-; AARCH64-NOSCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 4
+; AARCH64-NOSCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 8
 ; AARCH64-NOSCOPE-NEXT:    [[TMP11:%.*]] = ashr i64 [[TMP3]], 56
 ; AARCH64-NOSCOPE-NEXT:    [[TMP12:%.*]] = shl nuw nsw i64 [[TMP11]], 12
 ; AARCH64-NOSCOPE-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], -1
 ; AARCH64-NOSCOPE-NEXT:    [[TMP14:%.*]] = add i64 [[TMP3]], 8
 ; AARCH64-NOSCOPE-NEXT:    [[TMP15:%.*]] = and i64 [[TMP14]], [[TMP13]]
-; AARCH64-NOSCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 4
+; AARCH64-NOSCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 8
 ; AARCH64-NOSCOPE-NEXT:    [[TMP16:%.*]] = or i64 [[TMP3]], 4294967295
 ; AARCH64-NOSCOPE-NEXT:    [[HWASAN_SHADOW:%.*]] = add i64 [[TMP16]], 1
 ; AARCH64-NOSCOPE-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[HWASAN_SHADOW]] to ptr
@@ -1006,21 +1006,21 @@ define dso_local i32 @unreachable_exit() local_unnamed_addr sanitize_hwaddress {
 ; AARCH64-SHORT-SCOPE-LABEL: @unreachable_exit(
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP1:%.*]] = call ptr @llvm.thread.pointer()
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 48
-; AARCH64-SHORT-SCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 4
+; AARCH64-SHORT-SCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP4:%.*]] = ashr i64 [[TMP3]], 3
-; AARCH64-SHORT-SCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META1]])
+; AARCH64-SHORT-SCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META2]])
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP6:%.*]] = call ptr @llvm.frameaddress.p0(i32 0)
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[TMP6]] to i64
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP8:%.*]] = shl i64 [[TMP7]], 44
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP9:%.*]] = or i64 [[TMP5]], [[TMP8]]
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP3]] to ptr
-; AARCH64-SHORT-SCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 4
+; AARCH64-SHORT-SCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 8
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP11:%.*]] = ashr i64 [[TMP3]], 56
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP12:%.*]] = shl nuw nsw i64 [[TMP11]], 12
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], -1
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP14:%.*]] = add i64 [[TMP3]], 8
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP15:%.*]] = and i64 [[TMP14]], [[TMP13]]
-; AARCH64-SHORT-SCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 4
+; AARCH64-SHORT-SCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 8
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP16:%.*]] = or i64 [[TMP3]], 4294967295
 ; AARCH64-SHORT-SCOPE-NEXT:    [[HWASAN_SHADOW:%.*]] = add i64 [[TMP16]], 1
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[HWASAN_SHADOW]] to ptr
@@ -1066,21 +1066,21 @@ define dso_local i32 @unreachable_exit() local_unnamed_addr sanitize_hwaddress {
 ; AARCH64-SHORT-NOSCOPE-LABEL: @unreachable_exit(
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP1:%.*]] = call ptr @llvm.thread.pointer()
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 48
-; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 4
+; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP4:%.*]] = ashr i64 [[TMP3]], 3
-; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META1]])
+; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META2]])
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP6:%.*]] = call ptr @llvm.frameaddress.p0(i32 0)
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[TMP6]] to i64
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP8:%.*]] = shl i64 [[TMP7]], 44
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP9:%.*]] = or i64 [[TMP5]], [[TMP8]]
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP3]] to ptr
-; AARCH64-SHORT-NOSCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 4
+; AARCH64-SHORT-NOSCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 8
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP11:%.*]] = ashr i64 [[TMP3]], 56
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP12:%.*]] = shl nuw nsw i64 [[TMP11]], 12
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], -1
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP14:%.*]] = add i64 [[TMP3]], 8
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP15:%.*]] = and i64 [[TMP14]], [[TMP13]]
-; AARCH64-SHORT-NOSCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 4
+; AARCH64-SHORT-NOSCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 8
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP16:%.*]] = or i64 [[TMP3]], 4294967295
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[HWASAN_SHADOW:%.*]] = add i64 [[TMP16]], 1
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[HWASAN_SHADOW]] to ptr
@@ -1201,21 +1201,21 @@ define dso_local i32 @diamond_lifetime() local_unnamed_addr sanitize_hwaddress {
 ; AARCH64-SCOPE-LABEL: @diamond_lifetime(
 ; AARCH64-SCOPE-NEXT:    [[TMP1:%.*]] = call ptr @llvm.thread.pointer()
 ; AARCH64-SCOPE-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 48
-; AARCH64-SCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 4
+; AARCH64-SCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 ; AARCH64-SCOPE-NEXT:    [[TMP4:%.*]] = ashr i64 [[TMP3]], 3
-; AARCH64-SCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META1]])
+; AARCH64-SCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META2]])
 ; AARCH64-SCOPE-NEXT:    [[TMP6:%.*]] = call ptr @llvm.frameaddress.p0(i32 0)
 ; AARCH64-SCOPE-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[TMP6]] to i64
 ; AARCH64-SCOPE-NEXT:    [[TMP8:%.*]] = shl i64 [[TMP7]], 44
 ; AARCH64-SCOPE-NEXT:    [[TMP9:%.*]] = or i64 [[TMP5]], [[TMP8]]
 ; AARCH64-SCOPE-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP3]] to ptr
-; AARCH64-SCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 4
+; AARCH64-SCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 8
 ; AARCH64-SCOPE-NEXT:    [[TMP11:%.*]] = ashr i64 [[TMP3]], 56
 ; AARCH64-SCOPE-NEXT:    [[TMP12:%.*]] = shl nuw nsw i64 [[TMP11]], 12
 ; AARCH64-SCOPE-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], -1
 ; AARCH64-SCOPE-NEXT:    [[TMP14:%.*]] = add i64 [[TMP3]], 8
 ; AARCH64-SCOPE-NEXT:    [[TMP15:%.*]] = and i64 [[TMP14]], [[TMP13]]
-; AARCH64-SCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 4
+; AARCH64-SCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 8
 ; AARCH64-SCOPE-NEXT:    [[TMP16:%.*]] = or i64 [[TMP3]], 4294967295
 ; AARCH64-SCOPE-NEXT:    [[HWASAN_SHADOW:%.*]] = add i64 [[TMP16]], 1
 ; AARCH64-SCOPE-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[HWASAN_SHADOW]] to ptr
@@ -1262,21 +1262,21 @@ define dso_local i32 @diamond_lifetime() local_unnamed_addr sanitize_hwaddress {
 ; AARCH64-NOSCOPE-LABEL: @diamond_lifetime(
 ; AARCH64-NOSCOPE-NEXT:    [[TMP1:%.*]] = call ptr @llvm.thread.pointer()
 ; AARCH64-NOSCOPE-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 48
-; AARCH64-NOSCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 4
+; AARCH64-NOSCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 ; AARCH64-NOSCOPE-NEXT:    [[TMP4:%.*]] = ashr i64 [[TMP3]], 3
-; AARCH64-NOSCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META1]])
+; AARCH64-NOSCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META2]])
 ; AARCH64-NOSCOPE-NEXT:    [[TMP6:%.*]] = call ptr @llvm.frameaddress.p0(i32 0)
 ; AARCH64-NOSCOPE-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[TMP6]] to i64
 ; AARCH64-NOSCOPE-NEXT:    [[TMP8:%.*]] = shl i64 [[TMP7]], 44
 ; AARCH64-NOSCOPE-NEXT:    [[TMP9:%.*]] = or i64 [[TMP5]], [[TMP8]]
 ; AARCH64-NOSCOPE-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP3]] to ptr
-; AARCH64-NOSCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 4
+; AARCH64-NOSCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 8
 ; AARCH64-NOSCOPE-NEXT:    [[TMP11:%.*]] = ashr i64 [[TMP3]], 56
 ; AARCH64-NOSCOPE-NEXT:    [[TMP12:%.*]] = shl nuw nsw i64 [[TMP11]], 12
 ; AARCH64-NOSCOPE-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], -1
 ; AARCH64-NOSCOPE-NEXT:    [[TMP14:%.*]] = add i64 [[TMP3]], 8
 ; AARCH64-NOSCOPE-NEXT:    [[TMP15:%.*]] = and i64 [[TMP14]], [[TMP13]]
-; AARCH64-NOSCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 4
+; AARCH64-NOSCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 8
 ; AARCH64-NOSCOPE-NEXT:    [[TMP16:%.*]] = or i64 [[TMP3]], 4294967295
 ; AARCH64-NOSCOPE-NEXT:    [[HWASAN_SHADOW:%.*]] = add i64 [[TMP16]], 1
 ; AARCH64-NOSCOPE-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[HWASAN_SHADOW]] to ptr
@@ -1314,21 +1314,21 @@ define dso_local i32 @diamond_lifetime() local_unnamed_addr sanitize_hwaddress {
 ; AARCH64-SHORT-SCOPE-LABEL: @diamond_lifetime(
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP1:%.*]] = call ptr @llvm.thread.pointer()
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 48
-; AARCH64-SHORT-SCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 4
+; AARCH64-SHORT-SCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP4:%.*]] = ashr i64 [[TMP3]], 3
-; AARCH64-SHORT-SCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META1]])
+; AARCH64-SHORT-SCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META2]])
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP6:%.*]] = call ptr @llvm.frameaddress.p0(i32 0)
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[TMP6]] to i64
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP8:%.*]] = shl i64 [[TMP7]], 44
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP9:%.*]] = or i64 [[TMP5]], [[TMP8]]
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP3]] to ptr
-; AARCH64-SHORT-SCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 4
+; AARCH64-SHORT-SCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 8
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP11:%.*]] = ashr i64 [[TMP3]], 56
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP12:%.*]] = shl nuw nsw i64 [[TMP11]], 12
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], -1
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP14:%.*]] = add i64 [[TMP3]], 8
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP15:%.*]] = and i64 [[TMP14]], [[TMP13]]
-; AARCH64-SHORT-SCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 4
+; AARCH64-SHORT-SCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 8
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP16:%.*]] = or i64 [[TMP3]], 4294967295
 ; AARCH64-SHORT-SCOPE-NEXT:    [[HWASAN_SHADOW:%.*]] = add i64 [[TMP16]], 1
 ; AARCH64-SHORT-SCOPE-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[HWASAN_SHADOW]] to ptr
@@ -1378,21 +1378,21 @@ define dso_local i32 @diamond_lifetime() local_unnamed_addr sanitize_hwaddress {
 ; AARCH64-SHORT-NOSCOPE-LABEL: @diamond_lifetime(
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP1:%.*]] = call ptr @llvm.thread.pointer()
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 48
-; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 4
+; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP4:%.*]] = ashr i64 [[TMP3]], 3
-; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META1]])
+; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP5:%.*]] = call i64 @llvm.read_register.i64(metadata [[META2]])
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP6:%.*]] = call ptr @llvm.frameaddress.p0(i32 0)
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[TMP6]] to i64
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP8:%.*]] = shl i64 [[TMP7]], 44
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP9:%.*]] = or i64 [[TMP5]], [[TMP8]]
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP3]] to ptr
-; AARCH64-SHORT-NOSCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 4
+; AARCH64-SHORT-NOSCOPE-NEXT:    store i64 [[TMP9]], ptr [[TMP10]], align 8
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP11:%.*]] = ashr i64 [[TMP3]], 56
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP12:%.*]] = shl nuw nsw i64 [[TMP11]], 12
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], -1
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP14:%.*]] = add i64 [[TMP3]], 8
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP15:%.*]] = and i64 [[TMP14]], [[TMP13]]
-; AARCH64-SHORT-NOSCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 4
+; AARCH64-SHORT-NOSCOPE-NEXT:    store i64 [[TMP15]], ptr [[TMP2]], align 8
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP16:%.*]] = or i64 [[TMP3]], 4294967295
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[HWASAN_SHADOW:%.*]] = add i64 [[TMP16]], 1
 ; AARCH64-SHORT-NOSCOPE-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[HWASAN_SHADOW]] to ptr
diff --git a/llvm/test/Transforms/InterleavedAccess/RISCV/zve32x.ll b/llvm/test/Transforms/InterleavedAccess/RISCV/zve32x.ll
index b76fbdcef64329..ac3cab6638b8c8 100644
--- a/llvm/test/Transforms/InterleavedAccess/RISCV/zve32x.ll
+++ b/llvm/test/Transforms/InterleavedAccess/RISCV/zve32x.ll
@@ -13,10 +13,10 @@ define <4 x i1> @load_large_vector(ptr %p) {
 ; ZVE32X-NEXT:    ret <4 x i1> [[RET]]
 ;
 ; ZVE64X-LABEL: @load_large_vector(
-; ZVE64X-NEXT:    [[L:%.*]] = load <12 x ptr>, ptr [[P:%.*]], align 128
-; ZVE64X-NEXT:    [[S1:%.*]] = shufflevector <12 x ptr> [[L]], <12 x ptr> poison, <4 x i32> <i32 0, i32 3, i32 6, i32 9>
-; ZVE64X-NEXT:    [[S2:%.*]] = shufflevector <12 x ptr> [[L]], <12 x ptr> poison, <4 x i32> <i32 1, i32 4, i32 7, i32 10>
-; ZVE64X-NEXT:    [[RET:%.*]] = icmp ne <4 x ptr> [[S1]], [[S2]]
+; ZVE64X-NEXT:    [[TMP1:%.*]] = call { <4 x ptr>, <4 x ptr>, <4 x ptr> } @llvm.riscv.seg3.load.v4p0.p0.i64(ptr [[P:%.*]], i64 4)
+; ZVE64X-NEXT:    [[TMP2:%.*]] = extractvalue { <4 x ptr>, <4 x ptr>, <4 x ptr> } [[TMP1]], 1
+; ZVE64X-NEXT:    [[TMP3:%.*]] = extractvalue { <4 x ptr>, <4 x ptr>, <4 x ptr> } [[TMP1]], 0
+; ZVE64X-NEXT:    [[RET:%.*]] = icmp ne <4 x ptr> [[TMP3]], [[TMP2]]
 ; ZVE64X-NEXT:    ret <4 x i1> [[RET]]
 ;
   %l = load <12 x ptr>, ptr %p
diff --git a/llvm/test/Transforms/InterleavedAccess/RISCV/zvl32b.ll b/llvm/test/Transforms/InterleavedAccess/RISCV/zvl32b.ll
index 0e8afacdf463bb..a94e6a70e79e68 100644
--- a/llvm/test/Transforms/InterleavedAccess/RISCV/zvl32b.ll
+++ b/llvm/test/Transforms/InterleavedAccess/RISCV/zvl32b.ll
@@ -16,11 +16,11 @@ define {<16 x i32>, <16 x i32>} @load_factor2_large(ptr %ptr) {
 ; ZVL32B-NEXT:    ret { <16 x i32>, <16 x i32> } [[RES1]]
 ;
 ; ZVL128B-LABEL: @load_factor2_large(
-; ZVL128B-NEXT:    [[INTERLEAVED_VEC:%.*]] = load <32 x i32>, ptr [[PTR:%.*]], align 128
-; ZVL128B-NEXT:    [[V0:%.*]] = shufflevector <32 x i32> [[INTERLEAVED_VEC]], <32 x i32> poison, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
-; ZVL128B-NEXT:    [[V1:%.*]] = shufflevector <32 x i32> [[INTERLEAVED_VEC]], <32 x i32> poison, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
-; ZVL128B-NEXT:    [[RES0:%.*]] = insertvalue { <16 x i32>, <16 x i32> } undef, <16 x i32> [[V0]], 0
-; ZVL128B-NEXT:    [[RES1:%.*]] = insertvalue { <16 x i32>, <16 x i32> } [[RES0]], <16 x i32> [[V1]], 1
+; ZVL128B-NEXT:    [[TMP1:%.*]] = call { <16 x i32>, <16 x i32> } @llvm.riscv.seg2.load.v16i32.p0.i32(ptr [[PTR:%.*]], i32 16)
+; ZVL128B-NEXT:    [[TMP2:%.*]] = extractvalue { <16 x i32>, <16 x i32> } [[TMP1]], 1
+; ZVL128B-NEXT:    [[TMP3:%.*]] = extractvalue { <16 x i32>, <16 x i32> } [[TMP1]], 0
+; ZVL128B-NEXT:    [[RES0:%.*]] = insertvalue { <16 x i32>, <16 x i32> } undef, <16 x i32> [[TMP3]], 0
+; ZVL128B-NEXT:    [[RES1:%.*]] = insertvalue { <16 x i32>, <16 x i32> } [[RES0]], <16 x i32> [[TMP2]], 1
 ; ZVL128B-NEXT:    ret { <16 x i32>, <16 x i32> } [[RES1]]
 ;
   %interleaved.vec = load <32 x i32>, ptr %ptr
diff --git a/llvm/test/Transforms/SLPVectorizer/AArch64/transpose-inseltpoison.ll b/llvm/test/Transforms/SLPVectorizer/AArch64/transpose-inseltpoison.ll
index 1bd63b79b0f5ca..1304bb248e96ec 100644
--- a/llvm/test/Transforms/SLPVectorizer/AArch64/transpose-inseltpoison.ll
+++ b/llvm/test/Transforms/SLPVectorizer/AArch64/transpose-inseltpoison.ll
@@ -96,13 +96,17 @@ define <4 x i32> @build_vec_v4i32(<4 x i32> %v0, <4 x i32> %v1) {
 
 define <4 x i32> @build_vec_v4i32_reuse_0(<2 x i32> %v0, <2 x i32> %v1) {
 ; CHECK-LABEL: @build_vec_v4i32_reuse_0(
-; CHECK-NEXT:    [[TMP1:%.*]] = add <2 x i32> [[V0:%.*]], [[V1:%.*]]
-; CHECK-NEXT:    [[TMP2:%.*]] = sub <2 x i32> [[V0]], [[V1]]
-; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP2]], <2 x i32> <i32 1, i32 2>
-; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP2]], <2 x i32> <i32 0, i32 3>
-; CHECK-NEXT:    [[TMP5:%.*]] = add <2 x i32> [[TMP4]], [[TMP3]]
-; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <2 x i32> [[TMP5]], <2 x i32> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
-; CHECK-NEXT:    ret <4 x i32> [[TMP6]]
+; CHECK-NEXT:    [[TMP1:%.*]] = sub <2 x i32> [[V0:%.*]], [[V1:%.*]]
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <2 x i32> [[V1]], <2 x i32> [[V0]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP2]])
+; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x i32> [[TMP1]], i64 0
+; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x i32> [[TMP1]], i64 1
+; CHECK-NEXT:    [[TMP2_1:%.*]] = add i32 [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP3_0:%.*]] = insertelement <4 x i32> poison, i32 [[TMP3]], i64 0
+; CHECK-NEXT:    [[TMP3_1:%.*]] = insertelement <4 x i32> [[TMP3_0]], i32 [[TMP2_1]], i64 1
+; CHECK-NEXT:    [[TMP3_2:%.*]] = insertelement <4 x i32> [[TMP3_1]], i32 [[TMP3]], i64 2
+; CHECK-NEXT:    [[TMP3_3:%.*]] = insertelement <4 x i32> [[TMP3_2]], i32 [[TMP2_1]], i64 3
+; CHECK-NEXT:    ret <4 x i32> [[TMP3_3]]
 ;
   %v0.0 = extractelement <2 x i32> %v0, i32 0
   %v0.1 = extractelement <2 x i32> %v0, i32 1
@@ -158,16 +162,20 @@ define <4 x i32> @build_vec_v4i32_reuse_1(<2 x i32> %v0, <2 x i32> %v1) {
 
 define <4 x i32> @build_vec_v4i32_3_binops(<2 x i32> %v0, <2 x i32> %v1) {
 ; CHECK-LABEL: @build_vec_v4i32_3_binops(
-; CHECK-NEXT:    [[TMP1:%.*]] = add <2 x i32> [[V0:%.*]], [[V1:%.*]]
-; CHECK-NEXT:    [[TMP2:%.*]] = mul <2 x i32> [[V0]], [[V1]]
-; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP2]], <2 x i32> <i32 1, i32 2>
-; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP2]], <2 x i32> <i32 0, i32 3>
-; CHECK-NEXT:    [[TMP5:%.*]] = add <2 x i32> [[TMP4]], [[TMP3]]
+; CHECK-NEXT:    [[TMP1:%.*]] = mul <2 x i32> [[V0:%.*]], [[V1:%.*]]
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <2 x i32> [[V1]], <2 x i32> [[V0]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP2]])
+; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x i32> [[TMP1]], i64 0
+; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x i32> [[TMP1]], i64 1
+; CHECK-NEXT:    [[TMP2_1:%.*]] = add i32 [[TMP4]], [[TMP5]]
 ; CHECK-NEXT:    [[TMP6:%.*]] = xor <2 x i32> [[V0]], [[V1]]
 ; CHECK-NEXT:    [[TMP7:%.*]] = shufflevector <2 x i32> [[TMP6]], <2 x i32> poison, <2 x i32> <i32 1, i32 0>
 ; CHECK-NEXT:    [[TMP8:%.*]] = xor <2 x i32> [[V0]], [[V1]]
 ; CHECK-NEXT:    [[TMP9:%.*]] = add <2 x i32> [[TMP7]], [[TMP8]]
-; CHECK-NEXT:    [[TMP3_31:%.*]] = shufflevector <2 x i32> [[TMP5]], <2 x i32> [[TMP9]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT:    [[TMP3_0:%.*]] = insertelement <4 x i32> poison, i32 [[TMP3]], i64 0
+; CHECK-NEXT:    [[TMP3_1:%.*]] = insertelement <4 x i32> [[TMP3_0]], i32 [[TMP2_1]], i64 1
+; CHECK-NEXT:    [[TMP10:%.*]] = shufflevector <2 x i32> [[TMP9]], <2 x i32> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP3_31:%.*]] = shufflevector <4 x i32> [[TMP3_1]], <4 x i32> [[TMP10]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
 ; CHECK-NEXT:    ret <4 x i32> [[TMP3_31]]
 ;
   %v0.0 = extractelement <2 x i32> %v0, i32 0
diff --git a/llvm/test/Transforms/SLPVectorizer/AArch64/transpose.ll b/llvm/test/Transforms/SLPVectorizer/AArch64/transpose.ll
index f99f6ecd333823..66998fd0f8898f 100644
--- a/llvm/test/Transforms/SLPVectorizer/AArch64/transpose.ll
+++ b/llvm/test/Transforms/SLPVectorizer/AArch64/transpose.ll
@@ -96,13 +96,17 @@ define <4 x i32> @build_vec_v4i32(<4 x i32> %v0, <4 x i32> %v1) {
 
 define <4 x i32> @build_vec_v4i32_reuse_0(<2 x i32> %v0, <2 x i32> %v1) {
 ; CHECK-LABEL: @build_vec_v4i32_reuse_0(
-; CHECK-NEXT:    [[TMP1:%.*]] = add <2 x i32> [[V0:%.*]], [[V1:%.*]]
-; CHECK-NEXT:    [[TMP2:%.*]] = sub <2 x i32> [[V0]], [[V1]]
-; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP2]], <2 x i32> <i32 1, i32 2>
-; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP2]], <2 x i32> <i32 0, i32 3>
-; CHECK-NEXT:    [[TMP5:%.*]] = add <2 x i32> [[TMP4]], [[TMP3]]
-; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <2 x i32> [[TMP5]], <2 x i32> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
-; CHECK-NEXT:    ret <4 x i32> [[TMP6]]
+; CHECK-NEXT:    [[TMP1:%.*]] = sub <2 x i32> [[V0:%.*]], [[V1:%.*]]
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <2 x i32> [[V1]], <2 x i32> [[V0]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP2]])
+; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x i32> [[TMP1]], i64 0
+; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x i32> [[TMP1]], i64 1
+; CHECK-NEXT:    [[TMP2_1:%.*]] = add i32 [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP3_0:%.*]] = insertelement <4 x i32> poison, i32 [[TMP3]], i64 0
+; CHECK-NEXT:    [[TMP3_1:%.*]] = insertelement <4 x i32> [[TMP3_0]], i32 [[TMP2_1]], i64 1
+; CHECK-NEXT:    [[TMP3_2:%.*]] = insertelement <4 x i32> [[TMP3_1]], i32 [[TMP3]], i64 2
+; CHECK-NEXT:    [[TMP3_3:%.*]] = insertelement <4 x i32> [[TMP3_2]], i32 [[TMP2_1]], i64 3
+; CHECK-NEXT:    ret <4 x i32> [[TMP3_3]]
 ;
   %v0.0 = extractelement <2 x i32> %v0, i32 0
   %v0.1 = extractelement <2 x i32> %v0, i32 1
@@ -158,16 +162,20 @@ define <4 x i32> @build_vec_v4i32_reuse_1(<2 x i32> %v0, <2 x i32> %v1) {
 
 define <4 x i32> @build_vec_v4i32_3_binops(<2 x i32> %v0, <2 x i32> %v1) {
 ; CHECK-LABEL: @build_vec_v4i32_3_binops(
-; CHECK-NEXT:    [[TMP1:%.*]] = add <2 x i32> [[V0:%.*]], [[V1:%.*]]
-; CHECK-NEXT:    [[TMP2:%.*]] = mul <2 x i32> [[V0]], [[V1]]
-; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP2]], <2 x i32> <i32 1, i32 2>
-; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP2]], <2 x i32> <i32 0, i32 3>
-; CHECK-NEXT:    [[TMP5:%.*]] = add <2 x i32> [[TMP4]], [[TMP3]]
+; CHECK-NEXT:    [[TMP1:%.*]] = mul <2 x i32> [[V0:%.*]], [[V1:%.*]]
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <2 x i32> [[V1]], <2 x i32> [[V0]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP2]])
+; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x i32> [[TMP1]], i64 0
+; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x i32> [[TMP1]], i64 1
+; CHECK-NEXT:    [[TMP2_1:%.*]] = add i32 [[TMP4]], [[TMP5]]
 ; CHECK-NEXT:    [[TMP6:%.*]] = xor <2 x i32> [[V0]], [[V1]]
 ; CHECK-NEXT:    [[TMP7:%.*]] = shufflevector <2 x i32> [[TMP6]], <2 x i32> poison, <2 x i32> <i32 1, i32 0>
 ; CHECK-NEXT:    [[TMP8:%.*]] = xor <2 x i32> [[V0]], [[V1]]
 ; CHECK-NEXT:    [[TMP9:%.*]] = add <2 x i32> [[TMP7]], [[TMP8]]
-; CHECK-NEXT:    [[TMP3_31:%.*]] = shufflevector <2 x i32> [[TMP5]], <2 x i32> [[TMP9]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT:    [[TMP3_0:%.*]] = insertelement <4 x i32> poison, i32 [[TMP3]], i64 0
+; CHECK-NEXT:    [[TMP3_1:%.*]] = insertelement <4 x i32> [[TMP3_0]], i32 [[TMP2_1]], i64 1
+; CHECK-NEXT:    [[TMP10:%.*]] = shufflevector <2 x i32> [[TMP9]], <2 x i32> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP3_31:%.*]] = shufflevector <4 x i32> [[TMP3_1]], <4 x i32> [[TMP10]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
 ; CHECK-NEXT:    ret <4 x i32> [[TMP3_31]]
 ;
   %v0.0 = extractelement <2 x i32> %v0, i32 0
diff --git a/llvm/test/Transforms/VectorCombine/RISCV/vpintrin-scalarization.ll b/llvm/test/Transforms/VectorCombine/RISCV/vpintrin-scalarization.ll
index 908006d43ac9e8..57c92524be58b4 100644
--- a/llvm/test/Transforms/VectorCombine/RISCV/vpintrin-scalarization.ll
+++ b/llvm/test/Transforms/VectorCombine/RISCV/vpintrin-scalarization.ll
@@ -37,14 +37,14 @@ declare <1 x i64> @llvm.vp.mul.v1i64(<1 x i64>, <1 x i64>, <1 x i1>, i32)
 declare <4 x i64> @llvm.vp.mul.v4i64(<4 x i64>, <4 x i64>, <4 x i1>, i32)
 
 define <vscale x 1 x i64> @add_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %y, i32 zeroext %evl) {
-; VEC-COMBINE-LABEL: @add_nxv1i64_allonesmask(
-; VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
-; VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
-; VEC-COMBINE-NEXT:    [[TMP1:%.*]] = add i64 [[Y:%.*]], 42
-; VEC-COMBINE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
-; VEC-COMBINE-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
-; VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
-; VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
+; VEC-COMBINE-64-LABEL: @add_nxv1i64_allonesmask(
+; VEC-COMBINE-64-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-64-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP1:%.*]] = add i64 [[Y:%.*]], 42
+; VEC-COMBINE-64-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
+; VEC-COMBINE-64-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-64-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
 ;
 ; NO-VEC-COMBINE-LABEL: @add_nxv1i64_allonesmask(
 ; NO-VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
@@ -54,6 +54,15 @@ define <vscale x 1 x i64> @add_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %y
 ; NO-VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.add.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
 ; NO-VEC-COMBINE-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
 ; NO-VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
+;
+; VEC-COMBINE-32-LABEL: @add_nxv1i64_allonesmask(
+; VEC-COMBINE-32-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-32-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[Y:%.*]], i64 0
+; VEC-COMBINE-32-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[TMP1]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.add.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-32-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
+; VEC-COMBINE-32-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
 ;
   %splat = insertelement <vscale x 1 x i1> poison, i1 -1, i32 0
   %mask = shufflevector <vscale x 1 x i1> %splat, <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
@@ -80,14 +89,14 @@ define <vscale x 1 x i64> @add_nxv1i64_anymask(<vscale x 1 x i64> %x, i64 %y, <v
 }
 
 define <vscale x 1 x i64> @sub_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %y, i32 zeroext %evl) {
-; VEC-COMBINE-LABEL: @sub_nxv1i64_allonesmask(
-; VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
-; VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
-; VEC-COMBINE-NEXT:    [[TMP1:%.*]] = sub i64 [[Y:%.*]], 42
-; VEC-COMBINE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
-; VEC-COMBINE-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
-; VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
-; VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
+; VEC-COMBINE-64-LABEL: @sub_nxv1i64_allonesmask(
+; VEC-COMBINE-64-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-64-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP1:%.*]] = sub i64 [[Y:%.*]], 42
+; VEC-COMBINE-64-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
+; VEC-COMBINE-64-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-64-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
 ;
 ; NO-VEC-COMBINE-LABEL: @sub_nxv1i64_allonesmask(
 ; NO-VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
@@ -97,6 +106,15 @@ define <vscale x 1 x i64> @sub_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %y
 ; NO-VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.sub.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
 ; NO-VEC-COMBINE-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
 ; NO-VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
+;
+; VEC-COMBINE-32-LABEL: @sub_nxv1i64_allonesmask(
+; VEC-COMBINE-32-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-32-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[Y:%.*]], i64 0
+; VEC-COMBINE-32-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[TMP1]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.sub.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-32-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
+; VEC-COMBINE-32-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
 ;
   %splat = insertelement <vscale x 1 x i1> poison, i1 -1, i32 0
   %mask = shufflevector <vscale x 1 x i1> %splat, <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
@@ -123,14 +141,14 @@ define <vscale x 1 x i64> @sub_nxv1i64_anymask(<vscale x 1 x i64> %x, i64 %y, <v
 }
 
 define <vscale x 1 x i64> @mul_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %y, i32 zeroext %evl) {
-; VEC-COMBINE-LABEL: @mul_nxv1i64_allonesmask(
-; VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
-; VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
-; VEC-COMBINE-NEXT:    [[TMP1:%.*]] = mul i64 [[Y:%.*]], 42
-; VEC-COMBINE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
-; VEC-COMBINE-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
-; VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
-; VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
+; VEC-COMBINE-64-LABEL: @mul_nxv1i64_allonesmask(
+; VEC-COMBINE-64-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-64-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP1:%.*]] = mul i64 [[Y:%.*]], 42
+; VEC-COMBINE-64-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
+; VEC-COMBINE-64-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-64-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
 ;
 ; NO-VEC-COMBINE-LABEL: @mul_nxv1i64_allonesmask(
 ; NO-VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
@@ -140,6 +158,15 @@ define <vscale x 1 x i64> @mul_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %y
 ; NO-VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
 ; NO-VEC-COMBINE-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
 ; NO-VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
+;
+; VEC-COMBINE-32-LABEL: @mul_nxv1i64_allonesmask(
+; VEC-COMBINE-32-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-32-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[Y:%.*]], i64 0
+; VEC-COMBINE-32-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[TMP1]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-32-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
+; VEC-COMBINE-32-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
 ;
   %splat = insertelement <vscale x 1 x i1> poison, i1 -1, i32 0
   %mask = shufflevector <vscale x 1 x i1> %splat, <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
@@ -166,14 +193,32 @@ define <vscale x 1 x i64> @mul_nxv1i64_anymask(<vscale x 1 x i64> %x, i64 %y, <v
 }
 
 define <vscale x 1 x i64> @sdiv_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %y, i32 zeroext %evl) {
-; ALL-LABEL: @sdiv_nxv1i64_allonesmask(
-; ALL-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
-; ALL-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[Y:%.*]], i64 0
-; ALL-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[TMP1]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.sdiv.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
-; ALL-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
-; ALL-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
+; VEC-COMBINE-64-LABEL: @sdiv_nxv1i64_allonesmask(
+; VEC-COMBINE-64-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-64-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP1:%.*]] = sdiv i64 [[Y:%.*]], 42
+; VEC-COMBINE-64-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
+; VEC-COMBINE-64-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-64-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
+;
+; NO-VEC-COMBINE-LABEL: @sdiv_nxv1i64_allonesmask(
+; NO-VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; NO-VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[Y:%.*]], i64 0
+; NO-VEC-COMBINE-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[TMP1]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.sdiv.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; NO-VEC-COMBINE-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
+; NO-VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
+;
+; VEC-COMBINE-32-LABEL: @sdiv_nxv1i64_allonesmask(
+; VEC-COMBINE-32-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-32-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[Y:%.*]], i64 0
+; VEC-COMBINE-32-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[TMP1]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.sdiv.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-32-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
+; VEC-COMBINE-32-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
 ;
   %splat = insertelement <vscale x 1 x i1> poison, i1 -1, i32 0
   %mask = shufflevector <vscale x 1 x i1> %splat, <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
@@ -221,14 +266,32 @@ define <vscale x 1 x i64> @sdiv_nxv1i64_unspeculatable(i64 %x, i64 %y, i32 zeroe
 }
 
 define <vscale x 1 x i64> @udiv_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %y, i32 zeroext %evl) {
-; ALL-LABEL: @udiv_nxv1i64_allonesmask(
-; ALL-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
-; ALL-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[Y:%.*]], i64 0
-; ALL-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[TMP1]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.udiv.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
-; ALL-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
-; ALL-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
+; VEC-COMBINE-64-LABEL: @udiv_nxv1i64_allonesmask(
+; VEC-COMBINE-64-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-64-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP1:%.*]] = udiv i64 [[Y:%.*]], 42
+; VEC-COMBINE-64-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
+; VEC-COMBINE-64-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-64-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
+;
+; NO-VEC-COMBINE-LABEL: @udiv_nxv1i64_allonesmask(
+; NO-VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; NO-VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[Y:%.*]], i64 0
+; NO-VEC-COMBINE-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[TMP1]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.udiv.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; NO-VEC-COMBINE-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
+; NO-VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
+;
+; VEC-COMBINE-32-LABEL: @udiv_nxv1i64_allonesmask(
+; VEC-COMBINE-32-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-32-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[Y:%.*]], i64 0
+; VEC-COMBINE-32-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[TMP1]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.udiv.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-32-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
+; VEC-COMBINE-32-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
 ;
   %splat = insertelement <vscale x 1 x i1> poison, i1 -1, i32 0
   %mask = shufflevector <vscale x 1 x i1> %splat, <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
@@ -276,14 +339,32 @@ define <vscale x 1 x i64> @udiv_nxv1i64_unspeculatable(i64 %x, i64 %y, i32 zeroe
 }
 
 define <vscale x 1 x i64> @srem_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %y, i32 zeroext %evl) {
-; ALL-LABEL: @srem_nxv1i64_allonesmask(
-; ALL-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
-; ALL-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[Y:%.*]], i64 0
-; ALL-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[TMP1]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.srem.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
-; ALL-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
-; ALL-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
+; VEC-COMBINE-64-LABEL: @srem_nxv1i64_allonesmask(
+; VEC-COMBINE-64-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-64-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP1:%.*]] = srem i64 [[Y:%.*]], 42
+; VEC-COMBINE-64-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
+; VEC-COMBINE-64-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-64-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
+;
+; NO-VEC-COMBINE-LABEL: @srem_nxv1i64_allonesmask(
+; NO-VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; NO-VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[Y:%.*]], i64 0
+; NO-VEC-COMBINE-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[TMP1]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.srem.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; NO-VEC-COMBINE-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
+; NO-VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
+;
+; VEC-COMBINE-32-LABEL: @srem_nxv1i64_allonesmask(
+; VEC-COMBINE-32-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-32-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[Y:%.*]], i64 0
+; VEC-COMBINE-32-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[TMP1]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.srem.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-32-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
+; VEC-COMBINE-32-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
 ;
   %splat = insertelement <vscale x 1 x i1> poison, i1 -1, i32 0
   %mask = shufflevector <vscale x 1 x i1> %splat, <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
@@ -331,14 +412,32 @@ define <vscale x 1 x i64> @srem_nxv1i64_unspeculatable(i64 %x, i64 %y, i32 zeroe
 }
 
 define <vscale x 1 x i64> @urem_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %y, i32 zeroext %evl) {
-; ALL-LABEL: @urem_nxv1i64_allonesmask(
-; ALL-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
-; ALL-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[Y:%.*]], i64 0
-; ALL-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[TMP1]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.urem.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
-; ALL-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
-; ALL-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
+; VEC-COMBINE-64-LABEL: @urem_nxv1i64_allonesmask(
+; VEC-COMBINE-64-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-64-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP1:%.*]] = urem i64 [[Y:%.*]], 42
+; VEC-COMBINE-64-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
+; VEC-COMBINE-64-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-64-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
+;
+; NO-VEC-COMBINE-LABEL: @urem_nxv1i64_allonesmask(
+; NO-VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; NO-VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[Y:%.*]], i64 0
+; NO-VEC-COMBINE-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[TMP1]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.urem.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; NO-VEC-COMBINE-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
+; NO-VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
+;
+; VEC-COMBINE-32-LABEL: @urem_nxv1i64_allonesmask(
+; VEC-COMBINE-32-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-32-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[Y:%.*]], i64 0
+; VEC-COMBINE-32-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[TMP1]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.urem.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-32-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
+; VEC-COMBINE-32-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
 ;
   %splat = insertelement <vscale x 1 x i1> poison, i1 -1, i32 0
   %mask = shufflevector <vscale x 1 x i1> %splat, <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
@@ -386,15 +485,21 @@ define <vscale x 1 x i64> @urem_nxv1i64_unspeculatable(i64 %x, i64 %y, i32 zeroe
 }
 
 define <vscale x 1 x i64> @sdiv_nxv1i64_allonesmask_knownvl(i64 %x, i64 %y) {
-; ALL-LABEL: @sdiv_nxv1i64_allonesmask_knownvl(
-; ALL-NEXT:    [[MASK_HEAD:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
-; ALL-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[MASK_HEAD]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[X_HEAD:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[X:%.*]], i64 0
-; ALL-NEXT:    [[X_SPLAT:%.*]] = shufflevector <vscale x 1 x i64> [[X_HEAD]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[Y_HEAD:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[X]], i64 0
-; ALL-NEXT:    [[Y_SPLAT:%.*]] = shufflevector <vscale x 1 x i64> [[Y_HEAD]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[RES:%.*]] = call <vscale x 1 x i64> @llvm.vp.sdiv.nxv1i64(<vscale x 1 x i64> [[X_SPLAT]], <vscale x 1 x i64> [[Y_SPLAT]], <vscale x 1 x i1> [[MASK]], i32 4)
-; ALL-NEXT:    ret <vscale x 1 x i64> [[RES]]
+; VEC-COMBINE-LABEL: @sdiv_nxv1i64_allonesmask_knownvl(
+; VEC-COMBINE-NEXT:    [[TMP1:%.*]] = sdiv i64 [[X:%.*]], [[X]]
+; VEC-COMBINE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
+; VEC-COMBINE-NEXT:    [[RES:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[RES]]
+;
+; NO-VEC-COMBINE-LABEL: @sdiv_nxv1i64_allonesmask_knownvl(
+; NO-VEC-COMBINE-NEXT:    [[MASK_HEAD:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; NO-VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[MASK_HEAD]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[X_HEAD:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[X:%.*]], i64 0
+; NO-VEC-COMBINE-NEXT:    [[X_SPLAT:%.*]] = shufflevector <vscale x 1 x i64> [[X_HEAD]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[Y_HEAD:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[X]], i64 0
+; NO-VEC-COMBINE-NEXT:    [[Y_SPLAT:%.*]] = shufflevector <vscale x 1 x i64> [[Y_HEAD]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[RES:%.*]] = call <vscale x 1 x i64> @llvm.vp.sdiv.nxv1i64(<vscale x 1 x i64> [[X_SPLAT]], <vscale x 1 x i64> [[Y_SPLAT]], <vscale x 1 x i1> [[MASK]], i32 4)
+; NO-VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[RES]]
 ;
   %mask.head = insertelement <vscale x 1 x i1> poison, i1 -1, i32 0
   %mask = shufflevector <vscale x 1 x i1> %mask.head, <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
@@ -424,15 +529,21 @@ define <vscale x 1 x i64> @sdiv_nxv1i64_anymask_knownvl(i64 %x, i64 %y, <vscale
 }
 
 define <vscale x 1 x i64> @udiv_nxv1i64_allonesmask_knownvl(i64 %x, i64 %y) {
-; ALL-LABEL: @udiv_nxv1i64_allonesmask_knownvl(
-; ALL-NEXT:    [[MASK_HEAD:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
-; ALL-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[MASK_HEAD]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[X_HEAD:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[X:%.*]], i64 0
-; ALL-NEXT:    [[X_SPLAT:%.*]] = shufflevector <vscale x 1 x i64> [[X_HEAD]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[Y_HEAD:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[X]], i64 0
-; ALL-NEXT:    [[Y_SPLAT:%.*]] = shufflevector <vscale x 1 x i64> [[Y_HEAD]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[RES:%.*]] = call <vscale x 1 x i64> @llvm.vp.udiv.nxv1i64(<vscale x 1 x i64> [[X_SPLAT]], <vscale x 1 x i64> [[Y_SPLAT]], <vscale x 1 x i1> [[MASK]], i32 4)
-; ALL-NEXT:    ret <vscale x 1 x i64> [[RES]]
+; VEC-COMBINE-LABEL: @udiv_nxv1i64_allonesmask_knownvl(
+; VEC-COMBINE-NEXT:    [[TMP1:%.*]] = udiv i64 [[X:%.*]], [[X]]
+; VEC-COMBINE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
+; VEC-COMBINE-NEXT:    [[RES:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[RES]]
+;
+; NO-VEC-COMBINE-LABEL: @udiv_nxv1i64_allonesmask_knownvl(
+; NO-VEC-COMBINE-NEXT:    [[MASK_HEAD:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; NO-VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[MASK_HEAD]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[X_HEAD:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[X:%.*]], i64 0
+; NO-VEC-COMBINE-NEXT:    [[X_SPLAT:%.*]] = shufflevector <vscale x 1 x i64> [[X_HEAD]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[Y_HEAD:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[X]], i64 0
+; NO-VEC-COMBINE-NEXT:    [[Y_SPLAT:%.*]] = shufflevector <vscale x 1 x i64> [[Y_HEAD]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[RES:%.*]] = call <vscale x 1 x i64> @llvm.vp.udiv.nxv1i64(<vscale x 1 x i64> [[X_SPLAT]], <vscale x 1 x i64> [[Y_SPLAT]], <vscale x 1 x i1> [[MASK]], i32 4)
+; NO-VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[RES]]
 ;
   %mask.head = insertelement <vscale x 1 x i1> poison, i1 -1, i32 0
   %mask = shufflevector <vscale x 1 x i1> %mask.head, <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
@@ -462,15 +573,21 @@ define <vscale x 1 x i64> @udiv_nxv1i64_anymask_knownvl(i64 %x, i64 %y, <vscale
 }
 
 define <vscale x 1 x i64> @srem_nxv1i64_allonesmask_knownvl(i64 %x, i64 %y) {
-; ALL-LABEL: @srem_nxv1i64_allonesmask_knownvl(
-; ALL-NEXT:    [[MASK_HEAD:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
-; ALL-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[MASK_HEAD]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[X_HEAD:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[X:%.*]], i64 0
-; ALL-NEXT:    [[X_SPLAT:%.*]] = shufflevector <vscale x 1 x i64> [[X_HEAD]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[Y_HEAD:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[X]], i64 0
-; ALL-NEXT:    [[Y_SPLAT:%.*]] = shufflevector <vscale x 1 x i64> [[Y_HEAD]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[RES:%.*]] = call <vscale x 1 x i64> @llvm.vp.srem.nxv1i64(<vscale x 1 x i64> [[X_SPLAT]], <vscale x 1 x i64> [[Y_SPLAT]], <vscale x 1 x i1> [[MASK]], i32 4)
-; ALL-NEXT:    ret <vscale x 1 x i64> [[RES]]
+; VEC-COMBINE-LABEL: @srem_nxv1i64_allonesmask_knownvl(
+; VEC-COMBINE-NEXT:    [[TMP1:%.*]] = srem i64 [[X:%.*]], [[X]]
+; VEC-COMBINE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
+; VEC-COMBINE-NEXT:    [[RES:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[RES]]
+;
+; NO-VEC-COMBINE-LABEL: @srem_nxv1i64_allonesmask_knownvl(
+; NO-VEC-COMBINE-NEXT:    [[MASK_HEAD:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; NO-VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[MASK_HEAD]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[X_HEAD:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[X:%.*]], i64 0
+; NO-VEC-COMBINE-NEXT:    [[X_SPLAT:%.*]] = shufflevector <vscale x 1 x i64> [[X_HEAD]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[Y_HEAD:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[X]], i64 0
+; NO-VEC-COMBINE-NEXT:    [[Y_SPLAT:%.*]] = shufflevector <vscale x 1 x i64> [[Y_HEAD]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[RES:%.*]] = call <vscale x 1 x i64> @llvm.vp.srem.nxv1i64(<vscale x 1 x i64> [[X_SPLAT]], <vscale x 1 x i64> [[Y_SPLAT]], <vscale x 1 x i1> [[MASK]], i32 4)
+; NO-VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[RES]]
 ;
   %mask.head = insertelement <vscale x 1 x i1> poison, i1 -1, i32 0
   %mask = shufflevector <vscale x 1 x i1> %mask.head, <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
@@ -500,15 +617,21 @@ define <vscale x 1 x i64> @srem_nxv1i64_anymask_knownvl(i64 %x, i64 %y, <vscale
 }
 
 define <vscale x 1 x i64> @urem_nxv1i64_allonesmask_knownvl(i64 %x, i64 %y) {
-; ALL-LABEL: @urem_nxv1i64_allonesmask_knownvl(
-; ALL-NEXT:    [[MASK_HEAD:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
-; ALL-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[MASK_HEAD]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[X_HEAD:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[X:%.*]], i64 0
-; ALL-NEXT:    [[X_SPLAT:%.*]] = shufflevector <vscale x 1 x i64> [[X_HEAD]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[Y_HEAD:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[X]], i64 0
-; ALL-NEXT:    [[Y_SPLAT:%.*]] = shufflevector <vscale x 1 x i64> [[Y_HEAD]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[RES:%.*]] = call <vscale x 1 x i64> @llvm.vp.urem.nxv1i64(<vscale x 1 x i64> [[X_SPLAT]], <vscale x 1 x i64> [[Y_SPLAT]], <vscale x 1 x i1> [[MASK]], i32 4)
-; ALL-NEXT:    ret <vscale x 1 x i64> [[RES]]
+; VEC-COMBINE-LABEL: @urem_nxv1i64_allonesmask_knownvl(
+; VEC-COMBINE-NEXT:    [[TMP1:%.*]] = urem i64 [[X:%.*]], [[X]]
+; VEC-COMBINE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
+; VEC-COMBINE-NEXT:    [[RES:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[RES]]
+;
+; NO-VEC-COMBINE-LABEL: @urem_nxv1i64_allonesmask_knownvl(
+; NO-VEC-COMBINE-NEXT:    [[MASK_HEAD:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; NO-VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[MASK_HEAD]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[X_HEAD:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[X:%.*]], i64 0
+; NO-VEC-COMBINE-NEXT:    [[X_SPLAT:%.*]] = shufflevector <vscale x 1 x i64> [[X_HEAD]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[Y_HEAD:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[X]], i64 0
+; NO-VEC-COMBINE-NEXT:    [[Y_SPLAT:%.*]] = shufflevector <vscale x 1 x i64> [[Y_HEAD]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[RES:%.*]] = call <vscale x 1 x i64> @llvm.vp.urem.nxv1i64(<vscale x 1 x i64> [[X_SPLAT]], <vscale x 1 x i64> [[Y_SPLAT]], <vscale x 1 x i1> [[MASK]], i32 4)
+; NO-VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[RES]]
 ;
   %mask.head = insertelement <vscale x 1 x i1> poison, i1 -1, i32 0
   %mask = shufflevector <vscale x 1 x i1> %mask.head, <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
@@ -538,14 +661,14 @@ define <vscale x 1 x i64> @urem_nxv1i64_anymask_knownvl(i64 %x, i64 %y, <vscale
 }
 
 define <vscale x 1 x i64> @ashr_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %y, i32 zeroext %evl) {
-; VEC-COMBINE-LABEL: @ashr_nxv1i64_allonesmask(
-; VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
-; VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
-; VEC-COMBINE-NEXT:    [[TMP1:%.*]] = ashr i64 [[Y:%.*]], 42
-; VEC-COMBINE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
-; VEC-COMBINE-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
-; VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
-; VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
+; VEC-COMBINE-64-LABEL: @ashr_nxv1i64_allonesmask(
+; VEC-COMBINE-64-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-64-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP1:%.*]] = ashr i64 [[Y:%.*]], 42
+; VEC-COMBINE-64-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
+; VEC-COMBINE-64-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-64-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
 ;
 ; NO-VEC-COMBINE-LABEL: @ashr_nxv1i64_allonesmask(
 ; NO-VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
@@ -555,6 +678,15 @@ define <vscale x 1 x i64> @ashr_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %
 ; NO-VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.ashr.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
 ; NO-VEC-COMBINE-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
 ; NO-VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
+;
+; VEC-COMBINE-32-LABEL: @ashr_nxv1i64_allonesmask(
+; VEC-COMBINE-32-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-32-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[Y:%.*]], i64 0
+; VEC-COMBINE-32-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[TMP1]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.ashr.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-32-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
+; VEC-COMBINE-32-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
 ;
   %splat = insertelement <vscale x 1 x i1> poison, i1 -1, i32 0
   %mask = shufflevector <vscale x 1 x i1> %splat, <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
@@ -581,14 +713,14 @@ define <vscale x 1 x i64> @ashr_nxv1i64_anymask(<vscale x 1 x i64> %x, i64 %y, <
 }
 
 define <vscale x 1 x i64> @lshr_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %y, i32 zeroext %evl) {
-; VEC-COMBINE-LABEL: @lshr_nxv1i64_allonesmask(
-; VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
-; VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
-; VEC-COMBINE-NEXT:    [[TMP1:%.*]] = lshr i64 [[Y:%.*]], 42
-; VEC-COMBINE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
-; VEC-COMBINE-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
-; VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
-; VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
+; VEC-COMBINE-64-LABEL: @lshr_nxv1i64_allonesmask(
+; VEC-COMBINE-64-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-64-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP1:%.*]] = lshr i64 [[Y:%.*]], 42
+; VEC-COMBINE-64-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
+; VEC-COMBINE-64-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-64-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
 ;
 ; NO-VEC-COMBINE-LABEL: @lshr_nxv1i64_allonesmask(
 ; NO-VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
@@ -598,6 +730,15 @@ define <vscale x 1 x i64> @lshr_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %
 ; NO-VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.lshr.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
 ; NO-VEC-COMBINE-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
 ; NO-VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
+;
+; VEC-COMBINE-32-LABEL: @lshr_nxv1i64_allonesmask(
+; VEC-COMBINE-32-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-32-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[Y:%.*]], i64 0
+; VEC-COMBINE-32-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[TMP1]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.lshr.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-32-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
+; VEC-COMBINE-32-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
 ;
   %splat = insertelement <vscale x 1 x i1> poison, i1 -1, i32 0
   %mask = shufflevector <vscale x 1 x i1> %splat, <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
@@ -624,14 +765,14 @@ define <vscale x 1 x i64> @lshr_nxv1i64_anymask(<vscale x 1 x i64> %x, i64 %y, <
 }
 
 define <vscale x 1 x i64> @shl_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %y, i32 zeroext %evl) {
-; VEC-COMBINE-LABEL: @shl_nxv1i64_allonesmask(
-; VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
-; VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
-; VEC-COMBINE-NEXT:    [[TMP1:%.*]] = shl i64 [[Y:%.*]], 42
-; VEC-COMBINE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
-; VEC-COMBINE-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
-; VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
-; VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
+; VEC-COMBINE-64-LABEL: @shl_nxv1i64_allonesmask(
+; VEC-COMBINE-64-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-64-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP1:%.*]] = shl i64 [[Y:%.*]], 42
+; VEC-COMBINE-64-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
+; VEC-COMBINE-64-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-64-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
 ;
 ; NO-VEC-COMBINE-LABEL: @shl_nxv1i64_allonesmask(
 ; NO-VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
@@ -641,6 +782,15 @@ define <vscale x 1 x i64> @shl_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %y
 ; NO-VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.shl.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
 ; NO-VEC-COMBINE-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
 ; NO-VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
+;
+; VEC-COMBINE-32-LABEL: @shl_nxv1i64_allonesmask(
+; VEC-COMBINE-32-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-32-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[Y:%.*]], i64 0
+; VEC-COMBINE-32-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[TMP1]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.shl.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-32-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
+; VEC-COMBINE-32-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
 ;
   %splat = insertelement <vscale x 1 x i1> poison, i1 -1, i32 0
   %mask = shufflevector <vscale x 1 x i1> %splat, <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
@@ -667,14 +817,14 @@ define <vscale x 1 x i64> @shl_nxv1i64_anymask(<vscale x 1 x i64> %x, i64 %y, <v
 }
 
 define <vscale x 1 x i64> @or_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %y, i32 zeroext %evl) {
-; VEC-COMBINE-LABEL: @or_nxv1i64_allonesmask(
-; VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
-; VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
-; VEC-COMBINE-NEXT:    [[TMP1:%.*]] = or i64 [[Y:%.*]], 42
-; VEC-COMBINE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
-; VEC-COMBINE-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
-; VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
-; VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
+; VEC-COMBINE-64-LABEL: @or_nxv1i64_allonesmask(
+; VEC-COMBINE-64-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-64-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP1:%.*]] = or i64 [[Y:%.*]], 42
+; VEC-COMBINE-64-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
+; VEC-COMBINE-64-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-64-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
 ;
 ; NO-VEC-COMBINE-LABEL: @or_nxv1i64_allonesmask(
 ; NO-VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
@@ -684,6 +834,15 @@ define <vscale x 1 x i64> @or_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %y,
 ; NO-VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.or.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
 ; NO-VEC-COMBINE-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
 ; NO-VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
+;
+; VEC-COMBINE-32-LABEL: @or_nxv1i64_allonesmask(
+; VEC-COMBINE-32-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-32-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[Y:%.*]], i64 0
+; VEC-COMBINE-32-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[TMP1]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.or.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-32-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
+; VEC-COMBINE-32-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
 ;
   %splat = insertelement <vscale x 1 x i1> poison, i1 -1, i32 0
   %mask = shufflevector <vscale x 1 x i1> %splat, <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
@@ -710,14 +869,14 @@ define <vscale x 1 x i64> @or_nxv1i64_anymask(<vscale x 1 x i64> %x, i64 %y, <vs
 }
 
 define <vscale x 1 x i64> @and_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %y, i32 zeroext %evl) {
-; VEC-COMBINE-LABEL: @and_nxv1i64_allonesmask(
-; VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
-; VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
-; VEC-COMBINE-NEXT:    [[TMP1:%.*]] = and i64 [[Y:%.*]], 42
-; VEC-COMBINE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
-; VEC-COMBINE-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
-; VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
-; VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
+; VEC-COMBINE-64-LABEL: @and_nxv1i64_allonesmask(
+; VEC-COMBINE-64-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-64-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP1:%.*]] = and i64 [[Y:%.*]], 42
+; VEC-COMBINE-64-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
+; VEC-COMBINE-64-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-64-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
 ;
 ; NO-VEC-COMBINE-LABEL: @and_nxv1i64_allonesmask(
 ; NO-VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
@@ -727,6 +886,15 @@ define <vscale x 1 x i64> @and_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %y
 ; NO-VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.and.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
 ; NO-VEC-COMBINE-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
 ; NO-VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
+;
+; VEC-COMBINE-32-LABEL: @and_nxv1i64_allonesmask(
+; VEC-COMBINE-32-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-32-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[Y:%.*]], i64 0
+; VEC-COMBINE-32-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[TMP1]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.and.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-32-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
+; VEC-COMBINE-32-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
 ;
   %splat = insertelement <vscale x 1 x i1> poison, i1 -1, i32 0
   %mask = shufflevector <vscale x 1 x i1> %splat, <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
@@ -753,14 +921,14 @@ define <vscale x 1 x i64> @and_nxv1i64_anymask(<vscale x 1 x i64> %x, i64 %y, <v
 }
 
 define <vscale x 1 x i64> @xor_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %y, i32 zeroext %evl) {
-; VEC-COMBINE-LABEL: @xor_nxv1i64_allonesmask(
-; VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
-; VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
-; VEC-COMBINE-NEXT:    [[TMP1:%.*]] = xor i64 [[Y:%.*]], 42
-; VEC-COMBINE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
-; VEC-COMBINE-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
-; VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
-; VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
+; VEC-COMBINE-64-LABEL: @xor_nxv1i64_allonesmask(
+; VEC-COMBINE-64-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-64-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP1:%.*]] = xor i64 [[Y:%.*]], 42
+; VEC-COMBINE-64-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP1]], i64 0
+; VEC-COMBINE-64-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-64-NEXT:    ret <vscale x 1 x i64> [[TMP3]]
 ;
 ; NO-VEC-COMBINE-LABEL: @xor_nxv1i64_allonesmask(
 ; NO-VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
@@ -770,6 +938,15 @@ define <vscale x 1 x i64> @xor_nxv1i64_allonesmask(<vscale x 1 x i64> %x, i64 %y
 ; NO-VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.xor.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
 ; NO-VEC-COMBINE-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
 ; NO-VEC-COMBINE-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
+;
+; VEC-COMBINE-32-LABEL: @xor_nxv1i64_allonesmask(
+; VEC-COMBINE-32-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-32-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[Y:%.*]], i64 0
+; VEC-COMBINE-32-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x i64> [[TMP1]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x i64> @llvm.vp.xor.nxv1i64(<vscale x 1 x i64> [[TMP2]], <vscale x 1 x i64> shufflevector (<vscale x 1 x i64> insertelement (<vscale x 1 x i64> poison, i64 42, i64 0), <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-32-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.vp.mul.nxv1i64(<vscale x 1 x i64> [[X:%.*]], <vscale x 1 x i64> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
+; VEC-COMBINE-32-NEXT:    ret <vscale x 1 x i64> [[TMP4]]
 ;
   %splat = insertelement <vscale x 1 x i1> poison, i1 -1, i32 0
   %mask = shufflevector <vscale x 1 x i1> %splat, <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
@@ -1082,14 +1259,23 @@ define <vscale x 1 x float> @fdiv_nxv1f32_anymask(<vscale x 1 x float> %x, float
 }
 
 define <vscale x 1 x float> @frem_nxv1f32_allonesmask(<vscale x 1 x float> %x, float %y, i32 zeroext %evl) {
-; ALL-LABEL: @frem_nxv1f32_allonesmask(
-; ALL-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
-; ALL-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x float> poison, float [[Y:%.*]], i64 0
-; ALL-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x float> [[TMP1]], <vscale x 1 x float> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x float> @llvm.vp.frem.nxv1f32(<vscale x 1 x float> [[TMP2]], <vscale x 1 x float> shufflevector (<vscale x 1 x float> insertelement (<vscale x 1 x float> poison, float 4.200000e+01, i64 0), <vscale x 1 x float> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
-; ALL-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x float> @llvm.vp.fadd.nxv1f32(<vscale x 1 x float> [[X:%.*]], <vscale x 1 x float> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
-; ALL-NEXT:    ret <vscale x 1 x float> [[TMP4]]
+; VEC-COMBINE-LABEL: @frem_nxv1f32_allonesmask(
+; VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-NEXT:    [[TMP1:%.*]] = frem float [[Y:%.*]], 4.200000e+01
+; VEC-COMBINE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x float> poison, float [[TMP1]], i64 0
+; VEC-COMBINE-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x float> [[DOTSPLATINSERT]], <vscale x 1 x float> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x float> @llvm.vp.fadd.nxv1f32(<vscale x 1 x float> [[X:%.*]], <vscale x 1 x float> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-NEXT:    ret <vscale x 1 x float> [[TMP3]]
+;
+; NO-VEC-COMBINE-LABEL: @frem_nxv1f32_allonesmask(
+; NO-VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; NO-VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x float> poison, float [[Y:%.*]], i64 0
+; NO-VEC-COMBINE-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x float> [[TMP1]], <vscale x 1 x float> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x float> @llvm.vp.frem.nxv1f32(<vscale x 1 x float> [[TMP2]], <vscale x 1 x float> shufflevector (<vscale x 1 x float> insertelement (<vscale x 1 x float> poison, float 4.200000e+01, i64 0), <vscale x 1 x float> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; NO-VEC-COMBINE-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x float> @llvm.vp.fadd.nxv1f32(<vscale x 1 x float> [[X:%.*]], <vscale x 1 x float> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 [[EVL]])
+; NO-VEC-COMBINE-NEXT:    ret <vscale x 1 x float> [[TMP4]]
 ;
 
 
@@ -1166,14 +1352,23 @@ define <vscale x 1 x float> @fdiv_nxv1f32_anymask_knownvl(<vscale x 1 x float> %
 }
 
 define <vscale x 1 x float> @frem_nxv1f32_allonesmask_knownvl(<vscale x 1 x float> %x, float %y) {
-; ALL-LABEL: @frem_nxv1f32_allonesmask_knownvl(
-; ALL-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
-; ALL-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x float> poison, float [[Y:%.*]], i64 0
-; ALL-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x float> [[TMP1]], <vscale x 1 x float> poison, <vscale x 1 x i32> zeroinitializer
-; ALL-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x float> @llvm.vp.frem.nxv1f32(<vscale x 1 x float> [[TMP2]], <vscale x 1 x float> shufflevector (<vscale x 1 x float> insertelement (<vscale x 1 x float> poison, float 4.200000e+01, i64 0), <vscale x 1 x float> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 4)
-; ALL-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x float> @llvm.vp.fadd.nxv1f32(<vscale x 1 x float> [[X:%.*]], <vscale x 1 x float> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 4)
-; ALL-NEXT:    ret <vscale x 1 x float> [[TMP4]]
+; VEC-COMBINE-LABEL: @frem_nxv1f32_allonesmask_knownvl(
+; VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-NEXT:    [[TMP1:%.*]] = frem float [[Y:%.*]], 4.200000e+01
+; VEC-COMBINE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 1 x float> poison, float [[TMP1]], i64 0
+; VEC-COMBINE-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x float> [[DOTSPLATINSERT]], <vscale x 1 x float> poison, <vscale x 1 x i32> zeroinitializer
+; VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x float> @llvm.vp.fadd.nxv1f32(<vscale x 1 x float> [[X:%.*]], <vscale x 1 x float> [[TMP2]], <vscale x 1 x i1> [[MASK]], i32 4)
+; VEC-COMBINE-NEXT:    ret <vscale x 1 x float> [[TMP3]]
+;
+; NO-VEC-COMBINE-LABEL: @frem_nxv1f32_allonesmask_knownvl(
+; NO-VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <vscale x 1 x i1> poison, i1 true, i32 0
+; NO-VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <vscale x 1 x i1> [[SPLAT]], <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[TMP1:%.*]] = insertelement <vscale x 1 x float> poison, float [[Y:%.*]], i64 0
+; NO-VEC-COMBINE-NEXT:    [[TMP2:%.*]] = shufflevector <vscale x 1 x float> [[TMP1]], <vscale x 1 x float> poison, <vscale x 1 x i32> zeroinitializer
+; NO-VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <vscale x 1 x float> @llvm.vp.frem.nxv1f32(<vscale x 1 x float> [[TMP2]], <vscale x 1 x float> shufflevector (<vscale x 1 x float> insertelement (<vscale x 1 x float> poison, float 4.200000e+01, i64 0), <vscale x 1 x float> poison, <vscale x 1 x i32> zeroinitializer), <vscale x 1 x i1> [[MASK]], i32 4)
+; NO-VEC-COMBINE-NEXT:    [[TMP4:%.*]] = call <vscale x 1 x float> @llvm.vp.fadd.nxv1f32(<vscale x 1 x float> [[X:%.*]], <vscale x 1 x float> [[TMP3]], <vscale x 1 x i1> [[MASK]], i32 4)
+; NO-VEC-COMBINE-NEXT:    ret <vscale x 1 x float> [[TMP4]]
 ;
 
 
@@ -1417,14 +1612,14 @@ define <vscale x 8 x half> @fadd_nxv8f16_anymask(<vscale x 8 x half> %x, half %y
 }
 
 define <1 x i64> @add_v1i64_allonesmask(<1 x i64> %x, i64 %y, i32 zeroext %evl) {
-; VEC-COMBINE-LABEL: @add_v1i64_allonesmask(
-; VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <1 x i1> poison, i1 true, i32 0
-; VEC-COMBINE-NEXT:    [[MASK:%.*]] = shufflevector <1 x i1> [[SPLAT]], <1 x i1> poison, <1 x i32> zeroinitializer
-; VEC-COMBINE-NEXT:    [[TMP1:%.*]] = add i64 [[Y:%.*]], 42
-; VEC-COMBINE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <1 x i64> poison, i64 [[TMP1]], i64 0
-; VEC-COMBINE-NEXT:    [[TMP2:%.*]] = shufflevector <1 x i64> [[DOTSPLATINSERT]], <1 x i64> poison, <1 x i32> zeroinitializer
-; VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <1 x i64> @llvm.vp.mul.v1i64(<1 x i64> [[X:%.*]], <1 x i64> [[TMP2]], <1 x i1> [[MASK]], i32 [[EVL:%.*]])
-; VEC-COMBINE-NEXT:    ret <1 x i64> [[TMP3]]
+; VEC-COMBINE-64-LABEL: @add_v1i64_allonesmask(
+; VEC-COMBINE-64-NEXT:    [[SPLAT:%.*]] = insertelement <1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-64-NEXT:    [[MASK:%.*]] = shufflevector <1 x i1> [[SPLAT]], <1 x i1> poison, <1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP1:%.*]] = add i64 [[Y:%.*]], 42
+; VEC-COMBINE-64-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <1 x i64> poison, i64 [[TMP1]], i64 0
+; VEC-COMBINE-64-NEXT:    [[TMP2:%.*]] = shufflevector <1 x i64> [[DOTSPLATINSERT]], <1 x i64> poison, <1 x i32> zeroinitializer
+; VEC-COMBINE-64-NEXT:    [[TMP3:%.*]] = call <1 x i64> @llvm.vp.mul.v1i64(<1 x i64> [[X:%.*]], <1 x i64> [[TMP2]], <1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-64-NEXT:    ret <1 x i64> [[TMP3]]
 ;
 ; NO-VEC-COMBINE-LABEL: @add_v1i64_allonesmask(
 ; NO-VEC-COMBINE-NEXT:    [[SPLAT:%.*]] = insertelement <1 x i1> poison, i1 true, i32 0
@@ -1434,6 +1629,15 @@ define <1 x i64> @add_v1i64_allonesmask(<1 x i64> %x, i64 %y, i32 zeroext %evl)
 ; NO-VEC-COMBINE-NEXT:    [[TMP3:%.*]] = call <1 x i64> @llvm.vp.add.v1i64(<1 x i64> [[TMP2]], <1 x i64> <i64 42>, <1 x i1> [[MASK]], i32 [[EVL:%.*]])
 ; NO-VEC-COMBINE-NEXT:    [[TMP4:%.*]] = call <1 x i64> @llvm.vp.mul.v1i64(<1 x i64> [[X:%.*]], <1 x i64> [[TMP3]], <1 x i1> [[MASK]], i32 [[EVL]])
 ; NO-VEC-COMBINE-NEXT:    ret <1 x i64> [[TMP4]]
+;
+; VEC-COMBINE-32-LABEL: @add_v1i64_allonesmask(
+; VEC-COMBINE-32-NEXT:    [[SPLAT:%.*]] = insertelement <1 x i1> poison, i1 true, i32 0
+; VEC-COMBINE-32-NEXT:    [[MASK:%.*]] = shufflevector <1 x i1> [[SPLAT]], <1 x i1> poison, <1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP1:%.*]] = insertelement <1 x i64> poison, i64 [[Y:%.*]], i64 0
+; VEC-COMBINE-32-NEXT:    [[TMP2:%.*]] = shufflevector <1 x i64> [[TMP1]], <1 x i64> poison, <1 x i32> zeroinitializer
+; VEC-COMBINE-32-NEXT:    [[TMP3:%.*]] = call <1 x i64> @llvm.vp.add.v1i64(<1 x i64> [[TMP2]], <1 x i64> <i64 42>, <1 x i1> [[MASK]], i32 [[EVL:%.*]])
+; VEC-COMBINE-32-NEXT:    [[TMP4:%.*]] = call <1 x i64> @llvm.vp.mul.v1i64(<1 x i64> [[X:%.*]], <1 x i64> [[TMP3]], <1 x i1> [[MASK]], i32 [[EVL]])
+; VEC-COMBINE-32-NEXT:    ret <1 x i64> [[TMP4]]
 ;
   %splat = insertelement <1 x i1> poison, i1 -1, i32 0
   %mask = shufflevector <1 x i1> %splat, <1 x i1> poison, <1 x i32> zeroinitializer
@@ -1501,6 +1705,3 @@ define <4 x i64> @add_v4i64_anymask(<4 x i64> %x, i64 %y, <4 x i1> %mask, i32 ze
   %4 = call <4 x i64> @llvm.vp.mul.v4i64(<4 x i64> %x, <4 x i64> %3, <4 x i1> %mask, i32 %evl)
   ret <4 x i64> %4
 }
-;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
-; VEC-COMBINE-32: {{.*}}
-; VEC-COMBINE-64: {{.*}}