[DAGCombine] Move AND nodes to multiple load leaves

Search from AND nodes to find whether they can be propagated back to
loads, so that the AND and load can be combined into a narrow load.
We search through OR, XOR and other AND nodes and all bar one of the
leaves are required to be loads or constants. The exception node then
needs to be masked off meaning that the 'and' isn't removed, but the
loads(s) are narrowed still.

Differential Revision: https://reviews.llvm.org/D39604

llvm-svn: 319773

1 files changed, 238 insertions, 355 deletions

diff --git a/llvm/test/CodeGen/ARM/and-load-combine.ll b/llvm/test/CodeGen/ARM/and-load-combine.ll
index 6f0c1235959..f8e5d411c97 100644
--- a/llvm/test/CodeGen/ARM/and-load-combine.ll
+++ b/llvm/test/CodeGen/ARM/and-load-combine.ll
@@ -5,34 +5,30 @@
 ; RUN: llc -mtriple=thumbv8m.main %s -o - | FileCheck %s --check-prefix=THUMB2
 
 define arm_aapcscc zeroext i1 @cmp_xor8_short_short(i16* nocapture readonly %a,
+                                                    i16* nocapture readonly %b) {
 ; ARM-LABEL: cmp_xor8_short_short:
-; ARM:       @ %bb.0: @ %entry
-; ARM-NEXT:    ldrh r0, [r0]
-; ARM-NEXT:    ldrh r1, [r1]
-; ARM-NEXT:    eor r1, r1, r0
+; ARM:         ldrb r2, [r0]
 ; ARM-NEXT:    mov r0, #0
-; ARM-NEXT:    tst r1, #255
+; ARM-NEXT:    ldrb r1, [r1]
+; ARM-NEXT:    teq r1, r2
 ; ARM-NEXT:    movweq r0, #1
 ; ARM-NEXT:    bx lr
 ;
 ; ARMEB-LABEL: cmp_xor8_short_short:
-; ARMEB:       @ %bb.0: @ %entry
-; ARMEB-NEXT:    ldrh r0, [r0]
-; ARMEB-NEXT:    ldrh r1, [r1]
-; ARMEB-NEXT:    eor r1, r1, r0
+; ARMEB:        ldrb r2, [r0, #1]
 ; ARMEB-NEXT:    mov r0, #0
-; ARMEB-NEXT:    tst r1, #255
+; ARMEB-NEXT:    ldrb r1, [r1, #1]
+; ARMEB-NEXT:    teq r1, r2
 ; ARMEB-NEXT:    movweq r0, #1
 ; ARMEB-NEXT:    bx lr
 ;
 ; THUMB1-LABEL: cmp_xor8_short_short:
-; THUMB1:       @ %bb.0: @ %entry
-; THUMB1-NEXT:    ldrh r0, [r0]
-; THUMB1-NEXT:    ldrh r2, [r1]
+; THUMB1:         ldrb r0, [r0]
+; THUMB1-NEXT:    ldrb r2, [r1]
 ; THUMB1-NEXT:    eors r2, r0
 ; THUMB1-NEXT:    movs r0, #1
 ; THUMB1-NEXT:    movs r1, #0
-; THUMB1-NEXT:    lsls r2, r2, #24
+; THUMB1-NEXT:    cmp r2, #0
 ; THUMB1-NEXT:    beq .LBB0_2
 ; THUMB1-NEXT:  @ %bb.1: @ %entry
 ; THUMB1-NEXT:    mov r0, r1
@@ -40,16 +36,13 @@ define arm_aapcscc zeroext i1 @cmp_xor8_short_short(i16* nocapture readonly %a,
 ; THUMB1-NEXT:    bx lr
 ;
 ; THUMB2-LABEL: cmp_xor8_short_short:
-; THUMB2:       @ %bb.0: @ %entry
-; THUMB2-NEXT:    ldrh r0, [r0]
-; THUMB2-NEXT:    ldrh r1, [r1]
-; THUMB2-NEXT:    eors r0, r1
-; THUMB2-NEXT:    lsls r0, r0, #24
-; THUMB2-NEXT:    mov.w r0, #0
+; THUMB2:         ldrb r2, [r0]
+; THUMB2-NEXT:    movs r0, #0
+; THUMB2-NEXT:    ldrb r1, [r1]
+; THUMB2-NEXT:    teq.w r1, r2
 ; THUMB2-NEXT:    it eq
 ; THUMB2-NEXT:    moveq r0, #1
 ; THUMB2-NEXT:    bx lr
-                                                    i16* nocapture readonly %b) {
 entry:
   %0 = load i16, i16* %a, align 2
   %1 = load i16, i16* %b, align 2
@@ -60,34 +53,30 @@ entry:
 }
 
 define arm_aapcscc zeroext i1 @cmp_xor8_short_int(i16* nocapture readonly %a,
+                                                  i32* nocapture readonly %b) {
 ; ARM-LABEL: cmp_xor8_short_int:
-; ARM:       @ %bb.0: @ %entry
-; ARM-NEXT:    ldrh r0, [r0]
-; ARM-NEXT:    ldr r1, [r1]
-; ARM-NEXT:    eor r1, r1, r0
+; ARM:         ldrb r2, [r0]
 ; ARM-NEXT:    mov r0, #0
-; ARM-NEXT:    tst r1, #255
+; ARM-NEXT:    ldrb r1, [r1]
+; ARM-NEXT:    teq r1, r2
 ; ARM-NEXT:    movweq r0, #1
 ; ARM-NEXT:    bx lr
 ;
 ; ARMEB-LABEL: cmp_xor8_short_int:
-; ARMEB:       @ %bb.0: @ %entry
-; ARMEB-NEXT:    ldrh r0, [r0]
-; ARMEB-NEXT:    ldr r1, [r1]
-; ARMEB-NEXT:    eor r1, r1, r0
+; ARMEB:         ldrb r2, [r0, #1]
 ; ARMEB-NEXT:    mov r0, #0
-; ARMEB-NEXT:    tst r1, #255
+; ARMEB-NEXT:    ldrb r1, [r1, #3]
+; ARMEB-NEXT:    teq r1, r2
 ; ARMEB-NEXT:    movweq r0, #1
 ; ARMEB-NEXT:    bx lr
 ;
 ; THUMB1-LABEL: cmp_xor8_short_int:
-; THUMB1:       @ %bb.0: @ %entry
-; THUMB1-NEXT:    ldrh r0, [r0]
-; THUMB1-NEXT:    ldr r2, [r1]
+; THUMB1:         ldrb r0, [r0]
+; THUMB1-NEXT:    ldrb r2, [r1]
 ; THUMB1-NEXT:    eors r2, r0
 ; THUMB1-NEXT:    movs r0, #1
 ; THUMB1-NEXT:    movs r1, #0
-; THUMB1-NEXT:    lsls r2, r2, #24
+; THUMB1-NEXT:    cmp r2, #0
 ; THUMB1-NEXT:    beq .LBB1_2
 ; THUMB1-NEXT:  @ %bb.1: @ %entry
 ; THUMB1-NEXT:    mov r0, r1
@@ -95,16 +84,13 @@ define arm_aapcscc zeroext i1 @cmp_xor8_short_int(i16* nocapture readonly %a,
 ; THUMB1-NEXT:    bx lr
 ;
 ; THUMB2-LABEL: cmp_xor8_short_int:
-; THUMB2:       @ %bb.0: @ %entry
-; THUMB2-NEXT:    ldrh r0, [r0]
-; THUMB2-NEXT:    ldr r1, [r1]
-; THUMB2-NEXT:    eors r0, r1
-; THUMB2-NEXT:    lsls r0, r0, #24
-; THUMB2-NEXT:    mov.w r0, #0
+; THUMB2:         ldrb r2, [r0]
+; THUMB2-NEXT:    movs r0, #0
+; THUMB2-NEXT:    ldrb r1, [r1]
+; THUMB2-NEXT:    teq.w r1, r2
 ; THUMB2-NEXT:    it eq
 ; THUMB2-NEXT:    moveq r0, #1
 ; THUMB2-NEXT:    bx lr
-                                                  i32* nocapture readonly %b) {
 entry:
   %0 = load i16, i16* %a, align 2
   %conv = zext i16 %0 to i32
@@ -116,34 +102,30 @@ entry:
 }
 
 define arm_aapcscc zeroext i1 @cmp_xor8_int_int(i32* nocapture readonly %a,
+                                                i32* nocapture readonly %b) {
 ; ARM-LABEL: cmp_xor8_int_int:
-; ARM:       @ %bb.0: @ %entry
-; ARM-NEXT:    ldr r0, [r0]
-; ARM-NEXT:    ldr r1, [r1]
-; ARM-NEXT:    eor r1, r1, r0
+; ARM:         ldrb r2, [r0]
 ; ARM-NEXT:    mov r0, #0
-; ARM-NEXT:    tst r1, #255
+; ARM-NEXT:    ldrb r1, [r1]
+; ARM-NEXT:    teq r1, r2
 ; ARM-NEXT:    movweq r0, #1
 ; ARM-NEXT:    bx lr
 ;
 ; ARMEB-LABEL: cmp_xor8_int_int:
-; ARMEB:       @ %bb.0: @ %entry
-; ARMEB-NEXT:    ldr r0, [r0]
-; ARMEB-NEXT:    ldr r1, [r1]
-; ARMEB-NEXT:    eor r1, r1, r0
+; ARMEB:         ldrb r2, [r0, #3]
 ; ARMEB-NEXT:    mov r0, #0
-; ARMEB-NEXT:    tst r1, #255
+; ARMEB-NEXT:    ldrb r1, [r1, #3]
+; ARMEB-NEXT:    teq r1, r2
 ; ARMEB-NEXT:    movweq r0, #1
 ; ARMEB-NEXT:    bx lr
 ;
 ; THUMB1-LABEL: cmp_xor8_int_int:
-; THUMB1:       @ %bb.0: @ %entry
-; THUMB1-NEXT:    ldr r0, [r0]
-; THUMB1-NEXT:    ldr r2, [r1]
+; THUMB1:         ldrb r0, [r0]
+; THUMB1-NEXT:    ldrb r2, [r1]
 ; THUMB1-NEXT:    eors r2, r0
 ; THUMB1-NEXT:    movs r0, #1
 ; THUMB1-NEXT:    movs r1, #0
-; THUMB1-NEXT:    lsls r2, r2, #24
+; THUMB1-NEXT:    cmp r2, #0
 ; THUMB1-NEXT:    beq .LBB2_2
 ; THUMB1-NEXT:  @ %bb.1: @ %entry
 ; THUMB1-NEXT:    mov r0, r1
@@ -151,16 +133,13 @@ define arm_aapcscc zeroext i1 @cmp_xor8_int_int(i32* nocapture readonly %a,
 ; THUMB1-NEXT:    bx lr
 ;
 ; THUMB2-LABEL: cmp_xor8_int_int:
-; THUMB2:       @ %bb.0: @ %entry
-; THUMB2-NEXT:    ldr r0, [r0]
-; THUMB2-NEXT:    ldr r1, [r1]
-; THUMB2-NEXT:    eors r0, r1
-; THUMB2-NEXT:    lsls r0, r0, #24
-; THUMB2-NEXT:    mov.w r0, #0
+; THUMB2:         ldrb r2, [r0]
+; THUMB2-NEXT:    movs r0, #0
+; THUMB2-NEXT:    ldrb r1, [r1]
+; THUMB2-NEXT:    teq.w r1, r2
 ; THUMB2-NEXT:    it eq
 ; THUMB2-NEXT:    moveq r0, #1
 ; THUMB2-NEXT:    bx lr
-                                                i32* nocapture readonly %b) {
 entry:
   %0 = load i32, i32* %a, align 4
   %1 = load i32, i32* %b, align 4
@@ -171,36 +150,30 @@ entry:
 }
 
 define arm_aapcscc zeroext i1 @cmp_xor16(i32* nocapture readonly %a,
+                                         i32* nocapture readonly %b) {
 ; ARM-LABEL: cmp_xor16:
-; ARM:       @ %bb.0: @ %entry
-; ARM-NEXT:    ldr r0, [r0]
-; ARM-NEXT:    movw r2, #65535
-; ARM-NEXT:    ldr r1, [r1]
-; ARM-NEXT:    eor r1, r1, r0
+; ARM:         ldrh r2, [r0]
 ; ARM-NEXT:    mov r0, #0
-; ARM-NEXT:    tst r1, r2
+; ARM-NEXT:    ldrh r1, [r1]
+; ARM-NEXT:    teq r1, r2
 ; ARM-NEXT:    movweq r0, #1
 ; ARM-NEXT:    bx lr
 ;
 ; ARMEB-LABEL: cmp_xor16:
-; ARMEB:       @ %bb.0: @ %entry
-; ARMEB-NEXT:    ldr r0, [r0]
-; ARMEB-NEXT:    movw r2, #65535
-; ARMEB-NEXT:    ldr r1, [r1]
-; ARMEB-NEXT:    eor r1, r1, r0
+; ARMEB:         ldrh r2, [r0, #2]
 ; ARMEB-NEXT:    mov r0, #0
-; ARMEB-NEXT:    tst r1, r2
+; ARMEB-NEXT:    ldrh r1, [r1, #2]
+; ARMEB-NEXT:    teq r1, r2
 ; ARMEB-NEXT:    movweq r0, #1
 ; ARMEB-NEXT:    bx lr
 ;
 ; THUMB1-LABEL: cmp_xor16:
-; THUMB1:       @ %bb.0: @ %entry
-; THUMB1-NEXT:    ldr r0, [r0]
-; THUMB1-NEXT:    ldr r2, [r1]
+; THUMB1:         ldrh r0, [r0]
+; THUMB1-NEXT:    ldrh r2, [r1]
 ; THUMB1-NEXT:    eors r2, r0
 ; THUMB1-NEXT:    movs r0, #1
 ; THUMB1-NEXT:    movs r1, #0
-; THUMB1-NEXT:    lsls r2, r2, #16
+; THUMB1-NEXT:    cmp r2, #0
 ; THUMB1-NEXT:    beq .LBB3_2
 ; THUMB1-NEXT:  @ %bb.1: @ %entry
 ; THUMB1-NEXT:    mov r0, r1
@@ -208,16 +181,13 @@ define arm_aapcscc zeroext i1 @cmp_xor16(i32* nocapture readonly %a,
 ; THUMB1-NEXT:    bx lr
 ;
 ; THUMB2-LABEL: cmp_xor16:
-; THUMB2:       @ %bb.0: @ %entry
-; THUMB2-NEXT:    ldr r0, [r0]
-; THUMB2-NEXT:    ldr r1, [r1]
-; THUMB2-NEXT:    eors r0, r1
-; THUMB2-NEXT:    lsls r0, r0, #16
-; THUMB2-NEXT:    mov.w r0, #0
+; THUMB2:         ldrh r2, [r0]
+; THUMB2-NEXT:    movs r0, #0
+; THUMB2-NEXT:    ldrh r1, [r1]
+; THUMB2-NEXT:    teq.w r1, r2
 ; THUMB2-NEXT:    it eq
 ; THUMB2-NEXT:    moveq r0, #1
 ; THUMB2-NEXT:    bx lr
-                                         i32* nocapture readonly %b) {
 entry:
   %0 = load i32, i32* %a, align 4
   %1 = load i32, i32* %b, align 4
@@ -228,34 +198,30 @@ entry:
 }
 
 define arm_aapcscc zeroext i1 @cmp_or8_short_short(i16* nocapture readonly %a,
+                                                   i16* nocapture readonly %b) {
 ; ARM-LABEL: cmp_or8_short_short:
-; ARM:       @ %bb.0: @ %entry
-; ARM-NEXT:    ldrh r0, [r0]
-; ARM-NEXT:    ldrh r1, [r1]
-; ARM-NEXT:    orr r1, r1, r0
+; ARM:         ldrb r0, [r0]
+; ARM-NEXT:    ldrb r1, [r1]
+; ARM-NEXT:    orrs r0, r1, r0
 ; ARM-NEXT:    mov r0, #0
-; ARM-NEXT:    tst r1, #255
 ; ARM-NEXT:    movweq r0, #1
 ; ARM-NEXT:    bx lr
 ;
 ; ARMEB-LABEL: cmp_or8_short_short:
-; ARMEB:       @ %bb.0: @ %entry
-; ARMEB-NEXT:    ldrh r0, [r0]
-; ARMEB-NEXT:    ldrh r1, [r1]
-; ARMEB-NEXT:    orr r1, r1, r0
+; ARMEB:         ldrb r0, [r0, #1]
+; ARMEB-NEXT:    ldrb r1, [r1, #1]
+; ARMEB-NEXT:    orrs r0, r1, r0
 ; ARMEB-NEXT:    mov r0, #0
-; ARMEB-NEXT:    tst r1, #255
 ; ARMEB-NEXT:    movweq r0, #1
 ; ARMEB-NEXT:    bx lr
 ;
 ; THUMB1-LABEL: cmp_or8_short_short:
-; THUMB1:       @ %bb.0: @ %entry
-; THUMB1-NEXT:    ldrh r0, [r0]
-; THUMB1-NEXT:    ldrh r2, [r1]
+; THUMB1:         ldrb r0, [r0]
+; THUMB1-NEXT:    ldrb r2, [r1]
 ; THUMB1-NEXT:    orrs r2, r0
 ; THUMB1-NEXT:    movs r0, #1
 ; THUMB1-NEXT:    movs r1, #0
-; THUMB1-NEXT:    lsls r2, r2, #24
+; THUMB1-NEXT:    cmp r2, #0
 ; THUMB1-NEXT:    beq .LBB4_2
 ; THUMB1-NEXT:  @ %bb.1: @ %entry
 ; THUMB1-NEXT:    mov r0, r1
@@ -263,16 +229,13 @@ define arm_aapcscc zeroext i1 @cmp_or8_short_short(i16* nocapture readonly %a,
 ; THUMB1-NEXT:    bx lr
 ;
 ; THUMB2-LABEL: cmp_or8_short_short:
-; THUMB2:       @ %bb.0: @ %entry
-; THUMB2-NEXT:    ldrh r0, [r0]
-; THUMB2-NEXT:    ldrh r1, [r1]
+; THUMB2:         ldrb r0, [r0]
+; THUMB2-NEXT:    ldrb r1, [r1]
 ; THUMB2-NEXT:    orrs r0, r1
-; THUMB2-NEXT:    lsls r0, r0, #24
 ; THUMB2-NEXT:    mov.w r0, #0
 ; THUMB2-NEXT:    it eq
 ; THUMB2-NEXT:    moveq r0, #1
 ; THUMB2-NEXT:    bx lr
-                                                   i16* nocapture readonly %b) {
 entry:
   %0 = load i16, i16* %a, align 2
   %1 = load i16, i16* %b, align 2
@@ -283,34 +246,30 @@ entry:
 }
 
 define arm_aapcscc zeroext i1 @cmp_or8_short_int(i16* nocapture readonly %a,
+                                                 i32* nocapture readonly %b) {
 ; ARM-LABEL: cmp_or8_short_int:
-; ARM:       @ %bb.0: @ %entry
-; ARM-NEXT:    ldrh r0, [r0]
-; ARM-NEXT:    ldr r1, [r1]
-; ARM-NEXT:    orr r1, r1, r0
+; ARM:         ldrb r0, [r0]
+; ARM-NEXT:    ldrb r1, [r1]
+; ARM-NEXT:    orrs r0, r1, r0
 ; ARM-NEXT:    mov r0, #0
-; ARM-NEXT:    tst r1, #255
 ; ARM-NEXT:    movweq r0, #1
 ; ARM-NEXT:    bx lr
 ;
 ; ARMEB-LABEL: cmp_or8_short_int:
-; ARMEB:       @ %bb.0: @ %entry
-; ARMEB-NEXT:    ldrh r0, [r0]
-; ARMEB-NEXT:    ldr r1, [r1]
-; ARMEB-NEXT:    orr r1, r1, r0
+; ARMEB:         ldrb r0, [r0, #1]
+; ARMEB-NEXT:    ldrb r1, [r1, #3]
+; ARMEB-NEXT:    orrs r0, r1, r0
 ; ARMEB-NEXT:    mov r0, #0
-; ARMEB-NEXT:    tst r1, #255
 ; ARMEB-NEXT:    movweq r0, #1
 ; ARMEB-NEXT:    bx lr
 ;
 ; THUMB1-LABEL: cmp_or8_short_int:
-; THUMB1:       @ %bb.0: @ %entry
-; THUMB1-NEXT:    ldrh r0, [r0]
-; THUMB1-NEXT:    ldr r2, [r1]
+; THUMB1:         ldrb r0, [r0]
+; THUMB1-NEXT:    ldrb r2, [r1]
 ; THUMB1-NEXT:    orrs r2, r0
 ; THUMB1-NEXT:    movs r0, #1
 ; THUMB1-NEXT:    movs r1, #0
-; THUMB1-NEXT:    lsls r2, r2, #24
+; THUMB1-NEXT:    cmp r2, #0
 ; THUMB1-NEXT:    beq .LBB5_2
 ; THUMB1-NEXT:  @ %bb.1: @ %entry
 ; THUMB1-NEXT:    mov r0, r1
@@ -318,16 +277,13 @@ define arm_aapcscc zeroext i1 @cmp_or8_short_int(i16* nocapture readonly %a,
 ; THUMB1-NEXT:    bx lr
 ;
 ; THUMB2-LABEL: cmp_or8_short_int:
-; THUMB2:       @ %bb.0: @ %entry
-; THUMB2-NEXT:    ldrh r0, [r0]
-; THUMB2-NEXT:    ldr r1, [r1]
+; THUMB2:         ldrb r0, [r0]
+; THUMB2-NEXT:    ldrb r1, [r1]
 ; THUMB2-NEXT:    orrs r0, r1
-; THUMB2-NEXT:    lsls r0, r0, #24
 ; THUMB2-NEXT:    mov.w r0, #0
 ; THUMB2-NEXT:    it eq
 ; THUMB2-NEXT:    moveq r0, #1
 ; THUMB2-NEXT:    bx lr
-                                                 i32* nocapture readonly %b) {
 entry:
   %0 = load i16, i16* %a, align 2
   %conv = zext i16 %0 to i32
@@ -339,34 +295,30 @@ entry:
 }
 
 define arm_aapcscc zeroext i1 @cmp_or8_int_int(i32* nocapture readonly %a,
+                                               i32* nocapture readonly %b) {
 ; ARM-LABEL: cmp_or8_int_int:
-; ARM:       @ %bb.0: @ %entry
-; ARM-NEXT:    ldr r0, [r0]
-; ARM-NEXT:    ldr r1, [r1]
-; ARM-NEXT:    orr r1, r1, r0
+; ARM:         ldrb r0, [r0]
+; ARM-NEXT:    ldrb r1, [r1]
+; ARM-NEXT:    orrs r0, r1, r0
 ; ARM-NEXT:    mov r0, #0
-; ARM-NEXT:    tst r1, #255
 ; ARM-NEXT:    movweq r0, #1
 ; ARM-NEXT:    bx lr
 ;
 ; ARMEB-LABEL: cmp_or8_int_int:
-; ARMEB:       @ %bb.0: @ %entry
-; ARMEB-NEXT:    ldr r0, [r0]
-; ARMEB-NEXT:    ldr r1, [r1]
-; ARMEB-NEXT:    orr r1, r1, r0
+; ARMEB:         ldrb r0, [r0, #3]
+; ARMEB-NEXT:    ldrb r1, [r1, #3]
+; ARMEB-NEXT:    orrs r0, r1, r0
 ; ARMEB-NEXT:    mov r0, #0
-; ARMEB-NEXT:    tst r1, #255
 ; ARMEB-NEXT:    movweq r0, #1
 ; ARMEB-NEXT:    bx lr
 ;
 ; THUMB1-LABEL: cmp_or8_int_int:
-; THUMB1:       @ %bb.0: @ %entry
-; THUMB1-NEXT:    ldr r0, [r0]
-; THUMB1-NEXT:    ldr r2, [r1]
+; THUMB1:         ldrb r0, [r0]
+; THUMB1-NEXT:    ldrb r2, [r1]
 ; THUMB1-NEXT:    orrs r2, r0
 ; THUMB1-NEXT:    movs r0, #1
 ; THUMB1-NEXT:    movs r1, #0
-; THUMB1-NEXT:    lsls r2, r2, #24
+; THUMB1-NEXT:    cmp r2, #0
 ; THUMB1-NEXT:    beq .LBB6_2
 ; THUMB1-NEXT:  @ %bb.1: @ %entry
 ; THUMB1-NEXT:    mov r0, r1
@@ -374,16 +326,13 @@ define arm_aapcscc zeroext i1 @cmp_or8_int_int(i32* nocapture readonly %a,
 ; THUMB1-NEXT:    bx lr
 ;
 ; THUMB2-LABEL: cmp_or8_int_int:
-; THUMB2:       @ %bb.0: @ %entry
-; THUMB2-NEXT:    ldr r0, [r0]
-; THUMB2-NEXT:    ldr r1, [r1]
+; THUMB2:         ldrb r0, [r0]
+; THUMB2-NEXT:    ldrb r1, [r1]
 ; THUMB2-NEXT:    orrs r0, r1
-; THUMB2-NEXT:    lsls r0, r0, #24
 ; THUMB2-NEXT:    mov.w r0, #0
 ; THUMB2-NEXT:    it eq
 ; THUMB2-NEXT:    moveq r0, #1
 ; THUMB2-NEXT:    bx lr
-                                               i32* nocapture readonly %b) {
 entry:
   %0 = load i32, i32* %a, align 4
   %1 = load i32, i32* %b, align 4
@@ -394,36 +343,30 @@ entry:
 }
 
 define arm_aapcscc zeroext i1 @cmp_or16(i32* nocapture readonly %a,
+                                        i32* nocapture readonly %b) {
 ; ARM-LABEL: cmp_or16:
-; ARM:       @ %bb.0: @ %entry
-; ARM-NEXT:    ldr r0, [r0]
-; ARM-NEXT:    movw r2, #65535
-; ARM-NEXT:    ldr r1, [r1]
-; ARM-NEXT:    orr r1, r1, r0
+; ARM:         ldrh r0, [r0]
+; ARM-NEXT:    ldrh r1, [r1]
+; ARM-NEXT:    orrs r0, r1, r0
 ; ARM-NEXT:    mov r0, #0
-; ARM-NEXT:    tst r1, r2
 ; ARM-NEXT:    movweq r0, #1
 ; ARM-NEXT:    bx lr
 ;
 ; ARMEB-LABEL: cmp_or16:
-; ARMEB:       @ %bb.0: @ %entry
-; ARMEB-NEXT:    ldr r0, [r0]
-; ARMEB-NEXT:    movw r2, #65535
-; ARMEB-NEXT:    ldr r1, [r1]
-; ARMEB-NEXT:    orr r1, r1, r0
+; ARMEB:         ldrh r0, [r0, #2]
+; ARMEB-NEXT:    ldrh r1, [r1, #2]
+; ARMEB-NEXT:    orrs r0, r1, r0
 ; ARMEB-NEXT:    mov r0, #0
-; ARMEB-NEXT:    tst r1, r2
 ; ARMEB-NEXT:    movweq r0, #1
 ; ARMEB-NEXT:    bx lr
 ;
 ; THUMB1-LABEL: cmp_or16:
-; THUMB1:       @ %bb.0: @ %entry
-; THUMB1-NEXT:    ldr r0, [r0]
-; THUMB1-NEXT:    ldr r2, [r1]
+; THUMB1:         ldrh r0, [r0]
+; THUMB1-NEXT:    ldrh r2, [r1]
 ; THUMB1-NEXT:    orrs r2, r0
 ; THUMB1-NEXT:    movs r0, #1
 ; THUMB1-NEXT:    movs r1, #0
-; THUMB1-NEXT:    lsls r2, r2, #16
+; THUMB1-NEXT:    cmp r2, #0
 ; THUMB1-NEXT:    beq .LBB7_2
 ; THUMB1-NEXT:  @ %bb.1: @ %entry
 ; THUMB1-NEXT:    mov r0, r1
@@ -431,16 +374,13 @@ define arm_aapcscc zeroext i1 @cmp_or16(i32* nocapture readonly %a,
 ; THUMB1-NEXT:    bx lr
 ;
 ; THUMB2-LABEL: cmp_or16:
-; THUMB2:       @ %bb.0: @ %entry
-; THUMB2-NEXT:    ldr r0, [r0]
-; THUMB2-NEXT:    ldr r1, [r1]
+; THUMB2:         ldrh r0, [r0]
+; THUMB2-NEXT:    ldrh r1, [r1]
 ; THUMB2-NEXT:    orrs r0, r1
-; THUMB2-NEXT:    lsls r0, r0, #16
 ; THUMB2-NEXT:    mov.w r0, #0
 ; THUMB2-NEXT:    it eq
 ; THUMB2-NEXT:    moveq r0, #1
 ; THUMB2-NEXT:    bx lr
-                                        i32* nocapture readonly %b) {
 entry:
   %0 = load i32, i32* %a, align 4
   %1 = load i32, i32* %b, align 4
@@ -451,34 +391,29 @@ entry:
 }
 
 define arm_aapcscc zeroext i1 @cmp_and8_short_short(i16* nocapture readonly %a,
+                                                    i16* nocapture readonly %b) {
 ; ARM-LABEL: cmp_and8_short_short:
-; ARM:       @ %bb.0: @ %entry
-; ARM-NEXT:    ldrh r1, [r1]
-; ARM-NEXT:    ldrh r0, [r0]
-; ARM-NEXT:    and r1, r0, r1
+; ARM:         ldrb r2, [r0]
 ; ARM-NEXT:    mov r0, #0
-; ARM-NEXT:    tst r1, #255
+; ARM-NEXT:    ldrb r1, [r1]
+; ARM-NEXT:    tst r2, r1
 ; ARM-NEXT:    movweq r0, #1
 ; ARM-NEXT:    bx lr
 ;
 ; ARMEB-LABEL: cmp_and8_short_short:
-; ARMEB:       @ %bb.0: @ %entry
-; ARMEB-NEXT:    ldrh r1, [r1]
-; ARMEB-NEXT:    ldrh r0, [r0]
-; ARMEB-NEXT:    and r1, r0, r1
+; ARMEB:         ldrb r2, [r0, #1]
 ; ARMEB-NEXT:    mov r0, #0
-; ARMEB-NEXT:    tst r1, #255
+; ARMEB-NEXT:    ldrb r1, [r1, #1]
+; ARMEB-NEXT:    tst r2, r1
 ; ARMEB-NEXT:    movweq r0, #1
 ; ARMEB-NEXT:    bx lr
 ;
 ; THUMB1-LABEL: cmp_and8_short_short:
-; THUMB1:       @ %bb.0: @ %entry
-; THUMB1-NEXT:    ldrh r1, [r1]
-; THUMB1-NEXT:    ldrh r2, [r0]
-; THUMB1-NEXT:    ands r2, r1
+; THUMB1:         ldrb r2, [r1]
+; THUMB1-NEXT:    ldrb r3, [r0]
 ; THUMB1-NEXT:    movs r0, #1
 ; THUMB1-NEXT:    movs r1, #0
-; THUMB1-NEXT:    lsls r2, r2, #24
+; THUMB1-NEXT:    tst r3, r2
 ; THUMB1-NEXT:    beq .LBB8_2
 ; THUMB1-NEXT:  @ %bb.1: @ %entry
 ; THUMB1-NEXT:    mov r0, r1
@@ -486,16 +421,13 @@ define arm_aapcscc zeroext i1 @cmp_and8_short_short(i16* nocapture readonly %a,
 ; THUMB1-NEXT:    bx lr
 ;
 ; THUMB2-LABEL: cmp_and8_short_short:
-; THUMB2:       @ %bb.0: @ %entry
-; THUMB2-NEXT:    ldrh r1, [r1]
-; THUMB2-NEXT:    ldrh r0, [r0]
-; THUMB2-NEXT:    ands r0, r1
-; THUMB2-NEXT:    lsls r0, r0, #24
-; THUMB2-NEXT:    mov.w r0, #0
+; THUMB2:         ldrb r2, [r0]
+; THUMB2-NEXT:    movs r0, #0
+; THUMB2-NEXT:    ldrb r1, [r1]
+; THUMB2-NEXT:    tst r2, r1
 ; THUMB2-NEXT:    it eq
 ; THUMB2-NEXT:    moveq r0, #1
 ; THUMB2-NEXT:    bx lr
-                                                    i16* nocapture readonly %b) {
 entry:
   %0 = load i16, i16* %a, align 2
   %1 = load i16, i16* %b, align 2
@@ -506,34 +438,29 @@ entry:
 }
 
 define arm_aapcscc zeroext i1 @cmp_and8_short_int(i16* nocapture readonly %a,
+                                                  i32* nocapture readonly %b) {
 ; ARM-LABEL: cmp_and8_short_int:
-; ARM:       @ %bb.0: @ %entry
-; ARM-NEXT:    ldrh r0, [r0]
-; ARM-NEXT:    ldr r1, [r1]
-; ARM-NEXT:    and r1, r1, r0
+; ARM:         ldrb r2, [r0]
 ; ARM-NEXT:    mov r0, #0
-; ARM-NEXT:    tst r1, #255
+; ARM-NEXT:    ldrb r1, [r1]
+; ARM-NEXT:    tst r1, r2
 ; ARM-NEXT:    movweq r0, #1
 ; ARM-NEXT:    bx lr
 ;
 ; ARMEB-LABEL: cmp_and8_short_int:
-; ARMEB:       @ %bb.0: @ %entry
-; ARMEB-NEXT:    ldrh r0, [r0]
-; ARMEB-NEXT:    ldr r1, [r1]
-; ARMEB-NEXT:    and r1, r1, r0
+; ARMEB:         ldrb r2, [r0, #1]
 ; ARMEB-NEXT:    mov r0, #0
-; ARMEB-NEXT:    tst r1, #255
+; ARMEB-NEXT:    ldrb r1, [r1, #3]
+; ARMEB-NEXT:    tst r1, r2
 ; ARMEB-NEXT:    movweq r0, #1
 ; ARMEB-NEXT:    bx lr
 ;
 ; THUMB1-LABEL: cmp_and8_short_int:
-; THUMB1:       @ %bb.0: @ %entry
-; THUMB1-NEXT:    ldrh r0, [r0]
-; THUMB1-NEXT:    ldr r2, [r1]
-; THUMB1-NEXT:    ands r2, r0
+; THUMB1:         ldrb r2, [r0]
+; THUMB1-NEXT:    ldrb r3, [r1]
 ; THUMB1-NEXT:    movs r0, #1
 ; THUMB1-NEXT:    movs r1, #0
-; THUMB1-NEXT:    lsls r2, r2, #24
+; THUMB1-NEXT:    tst r3, r2
 ; THUMB1-NEXT:    beq .LBB9_2
 ; THUMB1-NEXT:  @ %bb.1: @ %entry
 ; THUMB1-NEXT:    mov r0, r1
@@ -541,16 +468,13 @@ define arm_aapcscc zeroext i1 @cmp_and8_short_int(i16* nocapture readonly %a,
 ; THUMB1-NEXT:    bx lr
 ;
 ; THUMB2-LABEL: cmp_and8_short_int:
-; THUMB2:       @ %bb.0: @ %entry
-; THUMB2-NEXT:    ldrh r0, [r0]
-; THUMB2-NEXT:    ldr r1, [r1]
-; THUMB2-NEXT:    ands r0, r1
-; THUMB2-NEXT:    lsls r0, r0, #24
-; THUMB2-NEXT:    mov.w r0, #0
+; THUMB2:         ldrb r2, [r0]
+; THUMB2-NEXT:    movs r0, #0
+; THUMB2-NEXT:    ldrb r1, [r1]
+; THUMB2-NEXT:    tst r1, r2
 ; THUMB2-NEXT:    it eq
 ; THUMB2-NEXT:    moveq r0, #1
 ; THUMB2-NEXT:    bx lr
-                                                  i32* nocapture readonly %b) {
 entry:
   %0 = load i16, i16* %a, align 2
   %1 = load i32, i32* %b, align 4
@@ -562,34 +486,29 @@ entry:
 }
 
 define arm_aapcscc zeroext i1 @cmp_and8_int_int(i32* nocapture readonly %a,
+                                                i32* nocapture readonly %b) {
 ; ARM-LABEL: cmp_and8_int_int:
-; ARM:       @ %bb.0: @ %entry
-; ARM-NEXT:    ldr r1, [r1]
-; ARM-NEXT:    ldr r0, [r0]
-; ARM-NEXT:    and r1, r0, r1
+; ARM:         ldrb r2, [r0]
 ; ARM-NEXT:    mov r0, #0
-; ARM-NEXT:    tst r1, #255
+; ARM-NEXT:    ldrb r1, [r1]
+; ARM-NEXT:    tst r2, r1
 ; ARM-NEXT:    movweq r0, #1
 ; ARM-NEXT:    bx lr
 ;
 ; ARMEB-LABEL: cmp_and8_int_int:
-; ARMEB:       @ %bb.0: @ %entry
-; ARMEB-NEXT:    ldr r1, [r1]
-; ARMEB-NEXT:    ldr r0, [r0]
-; ARMEB-NEXT:    and r1, r0, r1
+; ARMEB:         ldrb r2, [r0, #3]
 ; ARMEB-NEXT:    mov r0, #0
-; ARMEB-NEXT:    tst r1, #255
+; ARMEB-NEXT:    ldrb r1, [r1, #3]
+; ARMEB-NEXT:    tst r2, r1
 ; ARMEB-NEXT:    movweq r0, #1
 ; ARMEB-NEXT:    bx lr
 ;
 ; THUMB1-LABEL: cmp_and8_int_int:
-; THUMB1:       @ %bb.0: @ %entry
-; THUMB1-NEXT:    ldr r1, [r1]
-; THUMB1-NEXT:    ldr r2, [r0]
-; THUMB1-NEXT:    ands r2, r1
+; THUMB1:         ldrb r2, [r1]
+; THUMB1-NEXT:    ldrb r3, [r0]
 ; THUMB1-NEXT:    movs r0, #1
 ; THUMB1-NEXT:    movs r1, #0
-; THUMB1-NEXT:    lsls r2, r2, #24
+; THUMB1-NEXT:    tst r3, r2
 ; THUMB1-NEXT:    beq .LBB10_2
 ; THUMB1-NEXT:  @ %bb.1: @ %entry
 ; THUMB1-NEXT:    mov r0, r1
@@ -597,16 +516,13 @@ define arm_aapcscc zeroext i1 @cmp_and8_int_int(i32* nocapture readonly %a,
 ; THUMB1-NEXT:    bx lr
 ;
 ; THUMB2-LABEL: cmp_and8_int_int:
-; THUMB2:       @ %bb.0: @ %entry
-; THUMB2-NEXT:    ldr r1, [r1]
-; THUMB2-NEXT:    ldr r0, [r0]
-; THUMB2-NEXT:    ands r0, r1
-; THUMB2-NEXT:    lsls r0, r0, #24
-; THUMB2-NEXT:    mov.w r0, #0
+; THUMB2:         ldrb r2, [r0]
+; THUMB2-NEXT:    movs r0, #0
+; THUMB2-NEXT:    ldrb r1, [r1]
+; THUMB2-NEXT:    tst r2, r1
 ; THUMB2-NEXT:    it eq
 ; THUMB2-NEXT:    moveq r0, #1
 ; THUMB2-NEXT:    bx lr
-                                                i32* nocapture readonly %b) {
 entry:
   %0 = load i32, i32* %a, align 4
   %1 = load i32, i32* %b, align 4
@@ -617,36 +533,29 @@ entry:
 }
 
 define arm_aapcscc zeroext i1 @cmp_and16(i32* nocapture readonly %a,
+                                         i32* nocapture readonly %b) {
 ; ARM-LABEL: cmp_and16:
-; ARM:       @ %bb.0: @ %entry
-; ARM-NEXT:    ldr r1, [r1]
-; ARM-NEXT:    movw r2, #65535
-; ARM-NEXT:    ldr r0, [r0]
-; ARM-NEXT:    and r1, r0, r1
+; ARM:         ldrh r2, [r0]
 ; ARM-NEXT:    mov r0, #0
-; ARM-NEXT:    tst r1, r2
+; ARM-NEXT:    ldrh r1, [r1]
+; ARM-NEXT:    tst r2, r1
 ; ARM-NEXT:    movweq r0, #1
 ; ARM-NEXT:    bx lr
 ;
 ; ARMEB-LABEL: cmp_and16:
-; ARMEB:       @ %bb.0: @ %entry
-; ARMEB-NEXT:    ldr r1, [r1]
-; ARMEB-NEXT:    movw r2, #65535
-; ARMEB-NEXT:    ldr r0, [r0]
-; ARMEB-NEXT:    and r1, r0, r1
+; ARMEB:         ldrh r2, [r0, #2]
 ; ARMEB-NEXT:    mov r0, #0
-; ARMEB-NEXT:    tst r1, r2
+; ARMEB-NEXT:    ldrh r1, [r1, #2]
+; ARMEB-NEXT:    tst r2, r1
 ; ARMEB-NEXT:    movweq r0, #1
 ; ARMEB-NEXT:    bx lr
 ;
 ; THUMB1-LABEL: cmp_and16:
-; THUMB1:       @ %bb.0: @ %entry
-; THUMB1-NEXT:    ldr r1, [r1]
-; THUMB1-NEXT:    ldr r2, [r0]
-; THUMB1-NEXT:    ands r2, r1
+; THUMB1:         ldrh r2, [r1]
+; THUMB1-NEXT:    ldrh r3, [r0]
 ; THUMB1-NEXT:    movs r0, #1
 ; THUMB1-NEXT:    movs r1, #0
-; THUMB1-NEXT:    lsls r2, r2, #16
+; THUMB1-NEXT:    tst r3, r2
 ; THUMB1-NEXT:    beq .LBB11_2
 ; THUMB1-NEXT:  @ %bb.1: @ %entry
 ; THUMB1-NEXT:    mov r0, r1
@@ -654,16 +563,13 @@ define arm_aapcscc zeroext i1 @cmp_and16(i32* nocapture readonly %a,
 ; THUMB1-NEXT:    bx lr
 ;
 ; THUMB2-LABEL: cmp_and16:
-; THUMB2:       @ %bb.0: @ %entry
-; THUMB2-NEXT:    ldr r1, [r1]
-; THUMB2-NEXT:    ldr r0, [r0]
-; THUMB2-NEXT:    ands r0, r1
-; THUMB2-NEXT:    lsls r0, r0, #16
-; THUMB2-NEXT:    mov.w r0, #0
+; THUMB2:         ldrh r2, [r0]
+; THUMB2-NEXT:    movs r0, #0
+; THUMB2-NEXT:    ldrh r1, [r1]
+; THUMB2-NEXT:    tst r2, r1
 ; THUMB2-NEXT:    it eq
 ; THUMB2-NEXT:    moveq r0, #1
 ; THUMB2-NEXT:    bx lr
-                                         i32* nocapture readonly %b) {
 entry:
   %0 = load i32, i32* %a, align 4
   %1 = load i32, i32* %b, align 4
@@ -675,35 +581,31 @@ entry:
 
 define arm_aapcscc i32 @add_and16(i32* nocapture readonly %a, i32 %y, i32 %z) {
 ; ARM-LABEL: add_and16:
-; ARM:       @ %bb.0: @ %entry
-; ARM-NEXT:    ldr r0, [r0]
-; ARM-NEXT:    add r1, r1, r2
+; ARM:         add r1, r1, r2
+; ARM-NEXT:    ldrh r0, [r0]
+; ARM-NEXT:    uxth r1, r1
 ; ARM-NEXT:    orr r0, r0, r1
-; ARM-NEXT:    uxth r0, r0
 ; ARM-NEXT:    bx lr
 ;
 ; ARMEB-LABEL: add_and16:
-; ARMEB:       @ %bb.0: @ %entry
-; ARMEB-NEXT:    ldr r0, [r0]
-; ARMEB-NEXT:    add r1, r1, r2
+; ARMEB:         add r1, r1, r2
+; ARMEB-NEXT:    ldrh r0, [r0, #2]
+; ARMEB-NEXT:    uxth r1, r1
 ; ARMEB-NEXT:    orr r0, r0, r1
-; ARMEB-NEXT:    uxth r0, r0
 ; ARMEB-NEXT:    bx lr
 ;
 ; THUMB1-LABEL: add_and16:
-; THUMB1:       @ %bb.0: @ %entry
-; THUMB1-NEXT:    adds r1, r1, r2
-; THUMB1-NEXT:    ldr r0, [r0]
+; THUMB1:         adds r1, r1, r2
+; THUMB1-NEXT:    uxth r1, r1
+; THUMB1-NEXT:    ldrh r0, [r0]
 ; THUMB1-NEXT:    orrs r0, r1
-; THUMB1-NEXT:    uxth r0, r0
 ; THUMB1-NEXT:    bx lr
 ;
 ; THUMB2-LABEL: add_and16:
-; THUMB2:       @ %bb.0: @ %entry
-; THUMB2-NEXT:    ldr r0, [r0]
-; THUMB2-NEXT:    add r1, r2
+; THUMB2:         add r1, r2
+; THUMB2-NEXT:    ldrh r0, [r0]
+; THUMB2-NEXT:    uxth r1, r1
 ; THUMB2-NEXT:    orrs r0, r1
-; THUMB2-NEXT:    uxth r0, r0
 ; THUMB2-NEXT:    bx lr
 entry:
   %x = load i32, i32* %a, align 4
@@ -715,43 +617,39 @@ entry:
 
 define arm_aapcscc i32 @test1(i32* %a, i32* %b, i32 %x, i32 %y) {
 ; ARM-LABEL: test1:
-; ARM:       @ %bb.0: @ %entry
-; ARM-NEXT:    mul r2, r2, r3
-; ARM-NEXT:    ldr r1, [r1]
-; ARM-NEXT:    ldr r0, [r0]
+; ARM:         mul r2, r2, r3
+; ARM-NEXT:    ldrh r1, [r1]
+; ARM-NEXT:    ldrh r0, [r0]
 ; ARM-NEXT:    eor r0, r0, r1
-; ARM-NEXT:    orr r0, r0, r2
-; ARM-NEXT:    uxth r0, r0
+; ARM-NEXT:    uxth r1, r2
+; ARM-NEXT:    orr r0, r0, r1
 ; ARM-NEXT:    bx lr
 ;
 ; ARMEB-LABEL: test1:
-; ARMEB:       @ %bb.0: @ %entry
-; ARMEB-NEXT:    mul r2, r2, r3
-; ARMEB-NEXT:    ldr r1, [r1]
-; ARMEB-NEXT:    ldr r0, [r0]
+; ARMEB:         mul r2, r2, r3
+; ARMEB-NEXT:    ldrh r1, [r1, #2]
+; ARMEB-NEXT:    ldrh r0, [r0, #2]
 ; ARMEB-NEXT:    eor r0, r0, r1
-; ARMEB-NEXT:    orr r0, r0, r2
-; ARMEB-NEXT:    uxth r0, r0
+; ARMEB-NEXT:    uxth r1, r2
+; ARMEB-NEXT:    orr r0, r0, r1
 ; ARMEB-NEXT:    bx lr
 ;
 ; THUMB1-LABEL: test1:
-; THUMB1:       @ %bb.0: @ %entry
+; THUMB1:         ldrh r1, [r1]
+; THUMB1-NEXT:    ldrh r4, [r0]
+; THUMB1-NEXT:    eors r4, r1
 ; THUMB1-NEXT:    muls r2, r3, r2
-; THUMB1-NEXT:    ldr r1, [r1]
-; THUMB1-NEXT:    ldr r0, [r0]
-; THUMB1-NEXT:    eors r0, r1
-; THUMB1-NEXT:    orrs r0, r2
-; THUMB1-NEXT:    uxth r0, r0
-; THUMB1-NEXT:    bx lr
+; THUMB1-NEXT:    uxth r0, r2
+; THUMB1-NEXT:    orrs r0, r4
+; THUMB1-NEXT:    pop
 ;
 ; THUMB2-LABEL: test1:
-; THUMB2:       @ %bb.0: @ %entry
-; THUMB2-NEXT:    muls r2, r3, r2
-; THUMB2-NEXT:    ldr r1, [r1]
-; THUMB2-NEXT:    ldr r0, [r0]
+; THUMB2:         ldrh r1, [r1]
+; THUMB2-NEXT:    ldrh r0, [r0]
 ; THUMB2-NEXT:    eors r0, r1
-; THUMB2-NEXT:    orrs r0, r2
-; THUMB2-NEXT:    uxth r0, r0
+; THUMB2-NEXT:    mul r1, r2, r3
+; THUMB2-NEXT:    uxth r1, r1
+; THUMB2-NEXT:    orrs r0, r1
 ; THUMB2-NEXT:    bx lr
 entry:
   %0 = load i32, i32* %a, align 4
@@ -765,8 +663,7 @@ entry:
 
 define arm_aapcscc i32 @test2(i32* %a, i32* %b, i32 %x, i32 %y) {
 ; ARM-LABEL: test2:
-; ARM:       @ %bb.0: @ %entry
-; ARM-NEXT:    ldr r1, [r1]
+; ARM:         ldr r1, [r1]
 ; ARM-NEXT:    ldr r0, [r0]
 ; ARM-NEXT:    mul r1, r2, r1
 ; ARM-NEXT:    eor r0, r0, r3
@@ -775,8 +672,7 @@ define arm_aapcscc i32 @test2(i32* %a, i32* %b, i32 %x, i32 %y) {
 ; ARM-NEXT:    bx lr
 ;
 ; ARMEB-LABEL: test2:
-; ARMEB:       @ %bb.0: @ %entry
-; ARMEB-NEXT:    ldr r1, [r1]
+; ARMEB:         ldr r1, [r1]
 ; ARMEB-NEXT:    ldr r0, [r0]
 ; ARMEB-NEXT:    mul r1, r2, r1
 ; ARMEB-NEXT:    eor r0, r0, r3
@@ -785,8 +681,7 @@ define arm_aapcscc i32 @test2(i32* %a, i32* %b, i32 %x, i32 %y) {
 ; ARMEB-NEXT:    bx lr
 ;
 ; THUMB1-LABEL: test2:
-; THUMB1:       @ %bb.0: @ %entry
-; THUMB1-NEXT:    ldr r1, [r1]
+; THUMB1:         ldr r1, [r1]
 ; THUMB1-NEXT:    muls r1, r2, r1
 ; THUMB1-NEXT:    ldr r0, [r0]
 ; THUMB1-NEXT:    eors r0, r3
@@ -795,8 +690,7 @@ define arm_aapcscc i32 @test2(i32* %a, i32* %b, i32 %x, i32 %y) {
 ; THUMB1-NEXT:    bx lr
 ;
 ; THUMB2-LABEL: test2:
-; THUMB2:       @ %bb.0: @ %entry
-; THUMB2-NEXT:    ldr r1, [r1]
+; THUMB2:         ldr r1, [r1]
 ; THUMB2-NEXT:    ldr r0, [r0]
 ; THUMB2-NEXT:    muls r1, r2, r1
 ; THUMB2-NEXT:    eors r0, r3
@@ -815,8 +709,7 @@ entry:
 
 define arm_aapcscc i32 @test3(i32* %a, i32* %b, i32 %x, i16* %y) {
 ; ARM-LABEL: test3:
-; ARM:       @ %bb.0: @ %entry
-; ARM-NEXT:    ldr r0, [r0]
+; ARM:         ldr r0, [r0]
 ; ARM-NEXT:    mul r1, r2, r0
 ; ARM-NEXT:    ldrh r2, [r3]
 ; ARM-NEXT:    eor r0, r0, r2
@@ -825,8 +718,7 @@ define arm_aapcscc i32 @test3(i32* %a, i32* %b, i32 %x, i16* %y) {
 ; ARM-NEXT:    bx lr
 ;
 ; ARMEB-LABEL: test3:
-; ARMEB:       @ %bb.0: @ %entry
-; ARMEB-NEXT:    ldr r0, [r0]
+; ARMEB:         ldr r0, [r0]
 ; ARMEB-NEXT:    mul r1, r2, r0
 ; ARMEB-NEXT:    ldrh r2, [r3]
 ; ARMEB-NEXT:    eor r0, r0, r2
@@ -835,8 +727,7 @@ define arm_aapcscc i32 @test3(i32* %a, i32* %b, i32 %x, i16* %y) {
 ; ARMEB-NEXT:    bx lr
 ;
 ; THUMB1-LABEL: test3:
-; THUMB1:       @ %bb.0: @ %entry
-; THUMB1-NEXT:    ldr r0, [r0]
+; THUMB1:         ldr r0, [r0]
 ; THUMB1-NEXT:    muls r2, r0, r2
 ; THUMB1-NEXT:    ldrh r1, [r3]
 ; THUMB1-NEXT:    eors r1, r0
@@ -845,8 +736,7 @@ define arm_aapcscc i32 @test3(i32* %a, i32* %b, i32 %x, i16* %y) {
 ; THUMB1-NEXT:    bx lr
 ;
 ; THUMB2-LABEL: test3:
-; THUMB2:       @ %bb.0: @ %entry
-; THUMB2-NEXT:    ldr r0, [r0]
+; THUMB2:         ldr r0, [r0]
 ; THUMB2-NEXT:    mul r1, r2, r0
 ; THUMB2-NEXT:    ldrh r2, [r3]
 ; THUMB2-NEXT:    eors r0, r2
@@ -866,43 +756,39 @@ entry:
 
 define arm_aapcscc i32 @test4(i32* %a, i32* %b, i32 %x, i32 %y) {
 ; ARM-LABEL: test4:
-; ARM:       @ %bb.0: @ %entry
-; ARM-NEXT:    mul r2, r2, r3
-; ARM-NEXT:    ldr r1, [r1]
-; ARM-NEXT:    ldr r0, [r0]
+; ARM:         mul r2, r2, r3
+; ARM-NEXT:    ldrh r1, [r1]
+; ARM-NEXT:    ldrh r0, [r0]
 ; ARM-NEXT:    eor r0, r0, r1
-; ARM-NEXT:    orr r0, r0, r2
-; ARM-NEXT:    uxth r0, r0
+; ARM-NEXT:    uxth r1, r2
+; ARM-NEXT:    orr r0, r0, r1
 ; ARM-NEXT:    bx lr
 ;
 ; ARMEB-LABEL: test4:
-; ARMEB:       @ %bb.0: @ %entry
-; ARMEB-NEXT:    mul r2, r2, r3
-; ARMEB-NEXT:    ldr r1, [r1]
-; ARMEB-NEXT:    ldr r0, [r0]
+; ARMEB:         mul r2, r2, r3
+; ARMEB-NEXT:    ldrh r1, [r1, #2]
+; ARMEB-NEXT:    ldrh r0, [r0, #2]
 ; ARMEB-NEXT:    eor r0, r0, r1
-; ARMEB-NEXT:    orr r0, r0, r2
-; ARMEB-NEXT:    uxth r0, r0
+; ARMEB-NEXT:    uxth r1, r2
+; ARMEB-NEXT:    orr r0, r0, r1
 ; ARMEB-NEXT:    bx lr
 ;
 ; THUMB1-LABEL: test4:
-; THUMB1:       @ %bb.0: @ %entry
+; THUMB1:         ldrh r1, [r1]
+; THUMB1-NEXT:    ldrh r4, [r0]
+; THUMB1-NEXT:    eors r4, r1
 ; THUMB1-NEXT:    muls r2, r3, r2
-; THUMB1-NEXT:    ldr r1, [r1]
-; THUMB1-NEXT:    ldr r0, [r0]
-; THUMB1-NEXT:    eors r0, r1
-; THUMB1-NEXT:    orrs r0, r2
-; THUMB1-NEXT:    uxth r0, r0
-; THUMB1-NEXT:    bx lr
+; THUMB1-NEXT:    uxth r0, r2
+; THUMB1-NEXT:    orrs r0, r4
+; THUMB1-NEXT:    pop
 ;
 ; THUMB2-LABEL: test4:
-; THUMB2:       @ %bb.0: @ %entry
-; THUMB2-NEXT:    muls r2, r3, r2
-; THUMB2-NEXT:    ldr r1, [r1]
-; THUMB2-NEXT:    ldr r0, [r0]
+; THUMB2:         ldrh r1, [r1]
+; THUMB2-NEXT:    ldrh r0, [r0]
 ; THUMB2-NEXT:    eors r0, r1
-; THUMB2-NEXT:    orrs r0, r2
-; THUMB2-NEXT:    uxth r0, r0
+; THUMB2-NEXT:    mul r1, r2, r3
+; THUMB2-NEXT:    uxth r1, r1
+; THUMB2-NEXT:    orrs r0, r1
 ; THUMB2-NEXT:    bx lr
 entry:
   %0 = load i32, i32* %a, align 4
@@ -916,43 +802,39 @@ entry:
 
 define arm_aapcscc i32 @test5(i32* %a, i32* %b, i32 %x, i16 zeroext %y) {
 ; ARM-LABEL: test5:
-; ARM:       @ %bb.0: @ %entry
-; ARM-NEXT:    ldr r1, [r1]
-; ARM-NEXT:    ldr r0, [r0]
+; ARM:         ldr r1, [r1]
+; ARM-NEXT:    ldrh r0, [r0]
 ; ARM-NEXT:    mul r1, r2, r1
 ; ARM-NEXT:    eor r0, r0, r3
+; ARM-NEXT:    uxth r1, r1
 ; ARM-NEXT:    orr r0, r0, r1
-; ARM-NEXT:    uxth r0, r0
 ; ARM-NEXT:    bx lr
 ;
 ; ARMEB-LABEL: test5:
-; ARMEB:       @ %bb.0: @ %entry
-; ARMEB-NEXT:    ldr r1, [r1]
-; ARMEB-NEXT:    ldr r0, [r0]
+; ARMEB:         ldr r1, [r1]
+; ARMEB-NEXT:    ldrh r0, [r0, #2]
 ; ARMEB-NEXT:    mul r1, r2, r1
 ; ARMEB-NEXT:    eor r0, r0, r3
+; ARMEB-NEXT:    uxth r1, r1
 ; ARMEB-NEXT:    orr r0, r0, r1
-; ARMEB-NEXT:    uxth r0, r0
 ; ARMEB-NEXT:    bx lr
 ;
 ; THUMB1-LABEL: test5:
-; THUMB1:       @ %bb.0: @ %entry
-; THUMB1-NEXT:    ldr r1, [r1]
-; THUMB1-NEXT:    muls r1, r2, r1
-; THUMB1-NEXT:    ldr r0, [r0]
-; THUMB1-NEXT:    eors r0, r3
-; THUMB1-NEXT:    orrs r0, r1
+; THUMB1:         ldrh r4, [r0]
+; THUMB1-NEXT:    eors r4, r3
+; THUMB1-NEXT:    ldr r0, [r1]
+; THUMB1-NEXT:    muls r0, r2, r0
 ; THUMB1-NEXT:    uxth r0, r0
-; THUMB1-NEXT:    bx lr
+; THUMB1-NEXT:    orrs r0, r4
+; THUMB1-NEXT:    pop
 ;
 ; THUMB2-LABEL: test5:
-; THUMB2:       @ %bb.0: @ %entry
-; THUMB2-NEXT:    ldr r1, [r1]
-; THUMB2-NEXT:    ldr r0, [r0]
+; THUMB2:         ldr r1, [r1]
+; THUMB2-NEXT:    ldrh r0, [r0]
 ; THUMB2-NEXT:    muls r1, r2, r1
 ; THUMB2-NEXT:    eors r0, r3
+; THUMB2-NEXT:    uxth r1, r1
 ; THUMB2-NEXT:    orrs r0, r1
-; THUMB2-NEXT:    uxth r0, r0
 ; THUMB2-NEXT:    bx lr
 entry:
   %0 = load i32, i32* %a, align 4
@@ -964,3 +846,4 @@ entry:
   %and = and i32 %or, 65535
   ret i32 %and
 }
+