[Tests] Update to newer ISA

There are some issues w/missed opts on older platforms, but that's not the purpose of this test.  Using a newer API points out that some TODOs are already handled, and allows addition of tests to exercise other issues (future patch.)

llvm-svn: 356473

diff --git a/llvm/test/CodeGen/X86/atomic-unordered.ll b/llvm/test/CodeGen/X86/atomic-unordered.ll
index ba77d53..aef99b7 100644 (file)
--- a/llvm/test/CodeGen/X86/atomic-unordered.ll
+++ b/llvm/test/CodeGen/X86/atomic-unordered.ll
@@ -1,6 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -O0 < %s -mtriple=x86_64-linux-generic -verify-machineinstrs -mattr=sse2 | FileCheck --check-prefix=CHECK-O0 %s
-; RUN: llc -O3 < %s -mtriple=x86_64-linux-generic -verify-machineinstrs -mattr=sse2 | FileCheck --check-prefix=CHECK-O3 %s
+; RUN: llc -O0 < %s -mtriple=x86_64-linux-generic -verify-machineinstrs -mcpu=skylake | FileCheck --check-prefix=CHECK-O0 %s
+; RUN: llc -O3 < %s -mtriple=x86_64-linux-generic -verify-machineinstrs -mcpu=skylake | FileCheck --check-prefix=CHECK-O3 %s
 
 define i8 @load_i8(i8* %ptr) {
 ; CHECK-O0-LABEL: load_i8:
@@ -276,18 +276,16 @@ define void @widen_broadcast_unaligned(i32* %p0, i32 %v) {
 define void @vec_store(i32* %p0, <2 x i32> %vec) {
 ; CHECK-O0-LABEL: vec_store:
 ; CHECK-O0:       # %bb.0:
-; CHECK-O0-NEXT:    movd %xmm0, %eax
-; CHECK-O0-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; CHECK-O0-NEXT:    movd %xmm0, %ecx
+; CHECK-O0-NEXT:    vmovd %xmm0, %eax
+; CHECK-O0-NEXT:    vpextrd $2, %xmm0, %ecx
 ; CHECK-O0-NEXT:    movl %eax, (%rdi)
 ; CHECK-O0-NEXT:    movl %ecx, 4(%rdi)
 ; CHECK-O0-NEXT:    retq
 ;
 ; CHECK-O3-LABEL: vec_store:
 ; CHECK-O3:       # %bb.0:
-; CHECK-O3-NEXT:    movd %xmm0, %eax
-; CHECK-O3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; CHECK-O3-NEXT:    movd %xmm0, %ecx
+; CHECK-O3-NEXT:    vmovd %xmm0, %eax
+; CHECK-O3-NEXT:    vpextrd $2, %xmm0, %ecx
 ; CHECK-O3-NEXT:    movl %eax, (%rdi)
 ; CHECK-O3-NEXT:    movl %ecx, 4(%rdi)
 ; CHECK-O3-NEXT:    retq
@@ -303,18 +301,16 @@ define void @vec_store(i32* %p0, <2 x i32> %vec) {
 define void @vec_store_unaligned(i32* %p0, <2 x i32> %vec) {
 ; CHECK-O0-LABEL: vec_store_unaligned:
 ; CHECK-O0:       # %bb.0:
-; CHECK-O0-NEXT:    movd %xmm0, %eax
-; CHECK-O0-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; CHECK-O0-NEXT:    movd %xmm0, %ecx
+; CHECK-O0-NEXT:    vmovd %xmm0, %eax
+; CHECK-O0-NEXT:    vpextrd $2, %xmm0, %ecx
 ; CHECK-O0-NEXT:    movl %eax, (%rdi)
 ; CHECK-O0-NEXT:    movl %ecx, 4(%rdi)
 ; CHECK-O0-NEXT:    retq
 ;
 ; CHECK-O3-LABEL: vec_store_unaligned:
 ; CHECK-O3:       # %bb.0:
-; CHECK-O3-NEXT:    movd %xmm0, %eax
-; CHECK-O3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; CHECK-O3-NEXT:    movd %xmm0, %ecx
+; CHECK-O3-NEXT:    vmovd %xmm0, %eax
+; CHECK-O3-NEXT:    vpextrd $2, %xmm0, %ecx
 ; CHECK-O3-NEXT:    movl %eax, (%rdi)
 ; CHECK-O3-NEXT:    movl %ecx, 4(%rdi)
 ; CHECK-O3-NEXT:    retq
@@ -333,14 +329,14 @@ define void @vec_store_unaligned(i32* %p0, <2 x i32> %vec) {
 define void @widen_broadcast2(i32* %p0, <2 x i32> %vec) {
 ; CHECK-O0-LABEL: widen_broadcast2:
 ; CHECK-O0:       # %bb.0:
-; CHECK-O0-NEXT:    movd %xmm0, %eax
+; CHECK-O0-NEXT:    vmovd %xmm0, %eax
 ; CHECK-O0-NEXT:    movl %eax, (%rdi)
 ; CHECK-O0-NEXT:    movl %eax, 4(%rdi)
 ; CHECK-O0-NEXT:    retq
 ;
 ; CHECK-O3-LABEL: widen_broadcast2:
 ; CHECK-O3:       # %bb.0:
-; CHECK-O3-NEXT:    movd %xmm0, %eax
+; CHECK-O3-NEXT:    vmovd %xmm0, %eax
 ; CHECK-O3-NEXT:    movl %eax, (%rdi)
 ; CHECK-O3-NEXT:    movl %eax, 4(%rdi)
 ; CHECK-O3-NEXT:    retq
@@ -355,14 +351,14 @@ define void @widen_broadcast2(i32* %p0, <2 x i32> %vec) {
 define void @widen_broadcast2_unaligned(i32* %p0, <2 x i32> %vec) {
 ; CHECK-O0-LABEL: widen_broadcast2_unaligned:
 ; CHECK-O0:       # %bb.0:
-; CHECK-O0-NEXT:    movd %xmm0, %eax
+; CHECK-O0-NEXT:    vmovd %xmm0, %eax
 ; CHECK-O0-NEXT:    movl %eax, (%rdi)
 ; CHECK-O0-NEXT:    movl %eax, 4(%rdi)
 ; CHECK-O0-NEXT:    retq
 ;
 ; CHECK-O3-LABEL: widen_broadcast2_unaligned:
 ; CHECK-O3:       # %bb.0:
-; CHECK-O3-NEXT:    movd %xmm0, %eax
+; CHECK-O3-NEXT:    vmovd %xmm0, %eax
 ; CHECK-O3-NEXT:    movl %eax, (%rdi)
 ; CHECK-O3-NEXT:    movl %eax, 4(%rdi)
 ; CHECK-O3-NEXT:    retq
@@ -622,9 +618,20 @@ define i64 @load_fold_sdiv2(i64* %p, i64 %v2) {
 ; CHECK-O3-LABEL: load_fold_sdiv2:
 ; CHECK-O3:       # %bb.0:
 ; CHECK-O3-NEXT:    movq (%rdi), %rax
+; CHECK-O3-NEXT:    movq %rax, %rcx
+; CHECK-O3-NEXT:    orq %rsi, %rcx
+; CHECK-O3-NEXT:    shrq $32, %rcx
+; CHECK-O3-NEXT:    je .LBB31_1
+; CHECK-O3-NEXT:  # %bb.2:
 ; CHECK-O3-NEXT:    cqto
 ; CHECK-O3-NEXT:    idivq %rsi
 ; CHECK-O3-NEXT:    retq
+; CHECK-O3-NEXT:  .LBB31_1:
+; CHECK-O3-NEXT:    # kill: def $eax killed $eax killed $rax
+; CHECK-O3-NEXT:    xorl %edx, %edx
+; CHECK-O3-NEXT:    divl %esi
+; CHECK-O3-NEXT:    # kill: def $eax killed $eax def $rax
+; CHECK-O3-NEXT:    retq
   %v = load atomic i64, i64* %p unordered, align 8
   %ret = sdiv i64 %v, %v2
   ret i64 %ret
@@ -641,8 +648,20 @@ define i64 @load_fold_sdiv3(i64* %p1, i64* %p2) {
 ; CHECK-O3-LABEL: load_fold_sdiv3:
 ; CHECK-O3:       # %bb.0:
 ; CHECK-O3-NEXT:    movq (%rdi), %rax
+; CHECK-O3-NEXT:    movq (%rsi), %rcx
+; CHECK-O3-NEXT:    movq %rax, %rdx
+; CHECK-O3-NEXT:    orq %rcx, %rdx
+; CHECK-O3-NEXT:    shrq $32, %rdx
+; CHECK-O3-NEXT:    je .LBB32_1
+; CHECK-O3-NEXT:  # %bb.2:
 ; CHECK-O3-NEXT:    cqto
-; CHECK-O3-NEXT:    idivq (%rsi)
+; CHECK-O3-NEXT:    idivq %rcx
+; CHECK-O3-NEXT:    retq
+; CHECK-O3-NEXT:  .LBB32_1:
+; CHECK-O3-NEXT:    # kill: def $eax killed $eax killed $rax
+; CHECK-O3-NEXT:    xorl %edx, %edx
+; CHECK-O3-NEXT:    divl %ecx
+; CHECK-O3-NEXT:    # kill: def $eax killed $eax def $rax
 ; CHECK-O3-NEXT:    retq
   %v = load atomic i64, i64* %p1 unordered, align 8
   %v2 = load atomic i64, i64* %p2 unordered, align 8
@@ -686,9 +705,20 @@ define i64 @load_fold_udiv2(i64* %p, i64 %v2) {
 ; CHECK-O3-LABEL: load_fold_udiv2:
 ; CHECK-O3:       # %bb.0:
 ; CHECK-O3-NEXT:    movq (%rdi), %rax
+; CHECK-O3-NEXT:    movq %rax, %rcx
+; CHECK-O3-NEXT:    orq %rsi, %rcx
+; CHECK-O3-NEXT:    shrq $32, %rcx
+; CHECK-O3-NEXT:    je .LBB34_1
+; CHECK-O3-NEXT:  # %bb.2:
 ; CHECK-O3-NEXT:    xorl %edx, %edx
 ; CHECK-O3-NEXT:    divq %rsi
 ; CHECK-O3-NEXT:    retq
+; CHECK-O3-NEXT:  .LBB34_1:
+; CHECK-O3-NEXT:    # kill: def $eax killed $eax killed $rax
+; CHECK-O3-NEXT:    xorl %edx, %edx
+; CHECK-O3-NEXT:    divl %esi
+; CHECK-O3-NEXT:    # kill: def $eax killed $eax def $rax
+; CHECK-O3-NEXT:    retq
   %v = load atomic i64, i64* %p unordered, align 8
   %ret = udiv i64 %v, %v2
   ret i64 %ret
@@ -706,8 +736,20 @@ define i64 @load_fold_udiv3(i64* %p1, i64* %p2) {
 ; CHECK-O3-LABEL: load_fold_udiv3:
 ; CHECK-O3:       # %bb.0:
 ; CHECK-O3-NEXT:    movq (%rdi), %rax
+; CHECK-O3-NEXT:    movq (%rsi), %rcx
+; CHECK-O3-NEXT:    movq %rax, %rdx
+; CHECK-O3-NEXT:    orq %rcx, %rdx
+; CHECK-O3-NEXT:    shrq $32, %rdx
+; CHECK-O3-NEXT:    je .LBB35_1
+; CHECK-O3-NEXT:  # %bb.2:
+; CHECK-O3-NEXT:    xorl %edx, %edx
+; CHECK-O3-NEXT:    divq %rcx
+; CHECK-O3-NEXT:    retq
+; CHECK-O3-NEXT:  .LBB35_1:
+; CHECK-O3-NEXT:    # kill: def $eax killed $eax killed $rax
 ; CHECK-O3-NEXT:    xorl %edx, %edx
-; CHECK-O3-NEXT:    divq (%rsi)
+; CHECK-O3-NEXT:    divl %ecx
+; CHECK-O3-NEXT:    # kill: def $eax killed $eax def $rax
 ; CHECK-O3-NEXT:    retq
   %v = load atomic i64, i64* %p1 unordered, align 8
   %v2 = load atomic i64, i64* %p2 unordered, align 8
@@ -760,10 +802,21 @@ define i64 @load_fold_srem2(i64* %p, i64 %v2) {
 ; CHECK-O3-LABEL: load_fold_srem2:
 ; CHECK-O3:       # %bb.0:
 ; CHECK-O3-NEXT:    movq (%rdi), %rax
+; CHECK-O3-NEXT:    movq %rax, %rcx
+; CHECK-O3-NEXT:    orq %rsi, %rcx
+; CHECK-O3-NEXT:    shrq $32, %rcx
+; CHECK-O3-NEXT:    je .LBB37_1
+; CHECK-O3-NEXT:  # %bb.2:
 ; CHECK-O3-NEXT:    cqto
 ; CHECK-O3-NEXT:    idivq %rsi
 ; CHECK-O3-NEXT:    movq %rdx, %rax
 ; CHECK-O3-NEXT:    retq
+; CHECK-O3-NEXT:  .LBB37_1:
+; CHECK-O3-NEXT:    # kill: def $eax killed $eax killed $rax
+; CHECK-O3-NEXT:    xorl %edx, %edx
+; CHECK-O3-NEXT:    divl %esi
+; CHECK-O3-NEXT:    movl %edx, %eax
+; CHECK-O3-NEXT:    retq
   %v = load atomic i64, i64* %p unordered, align 8
   %ret = srem i64 %v, %v2
   ret i64 %ret
@@ -781,10 +834,22 @@ define i64 @load_fold_srem3(i64* %p1, i64* %p2) {
 ; CHECK-O3-LABEL: load_fold_srem3:
 ; CHECK-O3:       # %bb.0:
 ; CHECK-O3-NEXT:    movq (%rdi), %rax
+; CHECK-O3-NEXT:    movq (%rsi), %rcx
+; CHECK-O3-NEXT:    movq %rax, %rdx
+; CHECK-O3-NEXT:    orq %rcx, %rdx
+; CHECK-O3-NEXT:    shrq $32, %rdx
+; CHECK-O3-NEXT:    je .LBB38_1
+; CHECK-O3-NEXT:  # %bb.2:
 ; CHECK-O3-NEXT:    cqto
-; CHECK-O3-NEXT:    idivq (%rsi)
+; CHECK-O3-NEXT:    idivq %rcx
 ; CHECK-O3-NEXT:    movq %rdx, %rax
 ; CHECK-O3-NEXT:    retq
+; CHECK-O3-NEXT:  .LBB38_1:
+; CHECK-O3-NEXT:    # kill: def $eax killed $eax killed $rax
+; CHECK-O3-NEXT:    xorl %edx, %edx
+; CHECK-O3-NEXT:    divl %ecx
+; CHECK-O3-NEXT:    movl %edx, %eax
+; CHECK-O3-NEXT:    retq
   %v = load atomic i64, i64* %p1 unordered, align 8
   %v2 = load atomic i64, i64* %p2 unordered, align 8
   %ret = srem i64 %v, %v2
@@ -834,10 +899,21 @@ define i64 @load_fold_urem2(i64* %p, i64 %v2) {
 ; CHECK-O3-LABEL: load_fold_urem2:
 ; CHECK-O3:       # %bb.0:
 ; CHECK-O3-NEXT:    movq (%rdi), %rax
+; CHECK-O3-NEXT:    movq %rax, %rcx
+; CHECK-O3-NEXT:    orq %rsi, %rcx
+; CHECK-O3-NEXT:    shrq $32, %rcx
+; CHECK-O3-NEXT:    je .LBB40_1
+; CHECK-O3-NEXT:  # %bb.2:
 ; CHECK-O3-NEXT:    xorl %edx, %edx
 ; CHECK-O3-NEXT:    divq %rsi
 ; CHECK-O3-NEXT:    movq %rdx, %rax
 ; CHECK-O3-NEXT:    retq
+; CHECK-O3-NEXT:  .LBB40_1:
+; CHECK-O3-NEXT:    # kill: def $eax killed $eax killed $rax
+; CHECK-O3-NEXT:    xorl %edx, %edx
+; CHECK-O3-NEXT:    divl %esi
+; CHECK-O3-NEXT:    movl %edx, %eax
+; CHECK-O3-NEXT:    retq
   %v = load atomic i64, i64* %p unordered, align 8
   %ret = urem i64 %v, %v2
   ret i64 %ret
@@ -856,10 +932,22 @@ define i64 @load_fold_urem3(i64* %p1, i64* %p2) {
 ; CHECK-O3-LABEL: load_fold_urem3:
 ; CHECK-O3:       # %bb.0:
 ; CHECK-O3-NEXT:    movq (%rdi), %rax
+; CHECK-O3-NEXT:    movq (%rsi), %rcx
+; CHECK-O3-NEXT:    movq %rax, %rdx
+; CHECK-O3-NEXT:    orq %rcx, %rdx
+; CHECK-O3-NEXT:    shrq $32, %rdx
+; CHECK-O3-NEXT:    je .LBB41_1
+; CHECK-O3-NEXT:  # %bb.2:
 ; CHECK-O3-NEXT:    xorl %edx, %edx
-; CHECK-O3-NEXT:    divq (%rsi)
+; CHECK-O3-NEXT:    divq %rcx
 ; CHECK-O3-NEXT:    movq %rdx, %rax
 ; CHECK-O3-NEXT:    retq
+; CHECK-O3-NEXT:  .LBB41_1:
+; CHECK-O3-NEXT:    # kill: def $eax killed $eax killed $rax
+; CHECK-O3-NEXT:    xorl %edx, %edx
+; CHECK-O3-NEXT:    divl %ecx
+; CHECK-O3-NEXT:    movl %edx, %eax
+; CHECK-O3-NEXT:    retq
   %v = load atomic i64, i64* %p1 unordered, align 8
   %v2 = load atomic i64, i64* %p2 unordered, align 8
   %ret = urem i64 %v, %v2
@@ -885,7 +973,6 @@ define i64 @load_fold_shl1(i64* %p) {
   ret i64 %ret
 }
 
-; Legal to fold (TODO)
 define i64 @load_fold_shl2(i64* %p, i64 %v2) {
 ; CHECK-O0-LABEL: load_fold_shl2:
 ; CHECK-O0:       # %bb.0:
@@ -898,17 +985,13 @@ define i64 @load_fold_shl2(i64* %p, i64 %v2) {
 ;
 ; CHECK-O3-LABEL: load_fold_shl2:
 ; CHECK-O3:       # %bb.0:
-; CHECK-O3-NEXT:    movq %rsi, %rcx
-; CHECK-O3-NEXT:    movq (%rdi), %rax
-; CHECK-O3-NEXT:    # kill: def $cl killed $cl killed $rcx
-; CHECK-O3-NEXT:    shlq %cl, %rax
+; CHECK-O3-NEXT:    shlxq %rsi, (%rdi), %rax
 ; CHECK-O3-NEXT:    retq
   %v = load atomic i64, i64* %p unordered, align 8
   %ret = shl i64 %v, %v2
   ret i64 %ret
 }
 
-; Legal to fold (TODO)
 define i64 @load_fold_shl3(i64* %p1, i64* %p2) {
 ; CHECK-O0-LABEL: load_fold_shl3:
 ; CHECK-O0:       # %bb.0:
@@ -921,10 +1004,8 @@ define i64 @load_fold_shl3(i64* %p1, i64* %p2) {
 ;
 ; CHECK-O3-LABEL: load_fold_shl3:
 ; CHECK-O3:       # %bb.0:
-; CHECK-O3-NEXT:    movq (%rdi), %rax
-; CHECK-O3-NEXT:    movq (%rsi), %rcx
-; CHECK-O3-NEXT:    # kill: def $cl killed $cl killed $rcx
-; CHECK-O3-NEXT:    shlq %cl, %rax
+; CHECK-O3-NEXT:    movq (%rsi), %rax
+; CHECK-O3-NEXT:    shlxq %rax, (%rdi), %rax
 ; CHECK-O3-NEXT:    retq
   %v = load atomic i64, i64* %p1 unordered, align 8
   %v2 = load atomic i64, i64* %p2 unordered, align 8
@@ -951,7 +1032,6 @@ define i64 @load_fold_lshr1(i64* %p) {
   ret i64 %ret
 }
 
-; Legal to fold (TODO)
 define i64 @load_fold_lshr2(i64* %p, i64 %v2) {
 ; CHECK-O0-LABEL: load_fold_lshr2:
 ; CHECK-O0:       # %bb.0:
@@ -964,17 +1044,13 @@ define i64 @load_fold_lshr2(i64* %p, i64 %v2) {
 ;
 ; CHECK-O3-LABEL: load_fold_lshr2:
 ; CHECK-O3:       # %bb.0:
-; CHECK-O3-NEXT:    movq %rsi, %rcx
-; CHECK-O3-NEXT:    movq (%rdi), %rax
-; CHECK-O3-NEXT:    # kill: def $cl killed $cl killed $rcx
-; CHECK-O3-NEXT:    shrq %cl, %rax
+; CHECK-O3-NEXT:    shrxq %rsi, (%rdi), %rax
 ; CHECK-O3-NEXT:    retq
   %v = load atomic i64, i64* %p unordered, align 8
   %ret = lshr i64 %v, %v2
   ret i64 %ret
 }
 
-; Legal to fold (TODO)
 define i64 @load_fold_lshr3(i64* %p1, i64* %p2) {
 ; CHECK-O0-LABEL: load_fold_lshr3:
 ; CHECK-O0:       # %bb.0:
@@ -987,10 +1063,8 @@ define i64 @load_fold_lshr3(i64* %p1, i64* %p2) {
 ;
 ; CHECK-O3-LABEL: load_fold_lshr3:
 ; CHECK-O3:       # %bb.0:
-; CHECK-O3-NEXT:    movq (%rdi), %rax
-; CHECK-O3-NEXT:    movq (%rsi), %rcx
-; CHECK-O3-NEXT:    # kill: def $cl killed $cl killed $rcx
-; CHECK-O3-NEXT:    shrq %cl, %rax
+; CHECK-O3-NEXT:    movq (%rsi), %rax
+; CHECK-O3-NEXT:    shrxq %rax, (%rdi), %rax
 ; CHECK-O3-NEXT:    retq
   %v = load atomic i64, i64* %p1 unordered, align 8
   %v2 = load atomic i64, i64* %p2 unordered, align 8
@@ -1017,7 +1091,6 @@ define i64 @load_fold_ashr1(i64* %p) {
   ret i64 %ret
 }
 
-; Legal to fold (TODO)
 define i64 @load_fold_ashr2(i64* %p, i64 %v2) {
 ; CHECK-O0-LABEL: load_fold_ashr2:
 ; CHECK-O0:       # %bb.0:
@@ -1030,17 +1103,13 @@ define i64 @load_fold_ashr2(i64* %p, i64 %v2) {
 ;
 ; CHECK-O3-LABEL: load_fold_ashr2:
 ; CHECK-O3:       # %bb.0:
-; CHECK-O3-NEXT:    movq %rsi, %rcx
-; CHECK-O3-NEXT:    movq (%rdi), %rax
-; CHECK-O3-NEXT:    # kill: def $cl killed $cl killed $rcx
-; CHECK-O3-NEXT:    sarq %cl, %rax
+; CHECK-O3-NEXT:    sarxq %rsi, (%rdi), %rax
 ; CHECK-O3-NEXT:    retq
   %v = load atomic i64, i64* %p unordered, align 8
   %ret = ashr i64 %v, %v2
   ret i64 %ret
 }
 
-; Legal to fold (TODO)
 define i64 @load_fold_ashr3(i64* %p1, i64* %p2) {
 ; CHECK-O0-LABEL: load_fold_ashr3:
 ; CHECK-O0:       # %bb.0:
@@ -1053,10 +1122,8 @@ define i64 @load_fold_ashr3(i64* %p1, i64* %p2) {
 ;
 ; CHECK-O3-LABEL: load_fold_ashr3:
 ; CHECK-O3:       # %bb.0:
-; CHECK-O3-NEXT:    movq (%rdi), %rax
-; CHECK-O3-NEXT:    movq (%rsi), %rcx
-; CHECK-O3-NEXT:    # kill: def $cl killed $cl killed $rcx
-; CHECK-O3-NEXT:    sarq %cl, %rax
+; CHECK-O3-NEXT:    movq (%rsi), %rax
+; CHECK-O3-NEXT:    sarxq %rax, (%rdi), %rax
 ; CHECK-O3-NEXT:    retq
   %v = load atomic i64, i64* %p1 unordered, align 8
   %v2 = load atomic i64, i64* %p2 unordered, align 8
@@ -1467,10 +1534,22 @@ define void @rmw_fold_sdiv2(i64* %p, i64 %v) {
 ; CHECK-O3-LABEL: rmw_fold_sdiv2:
 ; CHECK-O3:       # %bb.0:
 ; CHECK-O3-NEXT:    movq (%rdi), %rax
+; CHECK-O3-NEXT:    movq %rax, %rcx
+; CHECK-O3-NEXT:    orq %rsi, %rcx
+; CHECK-O3-NEXT:    shrq $32, %rcx
+; CHECK-O3-NEXT:    je .LBB70_1
+; CHECK-O3-NEXT:  # %bb.2:
 ; CHECK-O3-NEXT:    cqto
 ; CHECK-O3-NEXT:    idivq %rsi
 ; CHECK-O3-NEXT:    movq %rax, (%rdi)
 ; CHECK-O3-NEXT:    retq
+; CHECK-O3-NEXT:  .LBB70_1:
+; CHECK-O3-NEXT:    # kill: def $eax killed $eax killed $rax
+; CHECK-O3-NEXT:    xorl %edx, %edx
+; CHECK-O3-NEXT:    divl %esi
+; CHECK-O3-NEXT:    # kill: def $eax killed $eax def $rax
+; CHECK-O3-NEXT:    movq %rax, (%rdi)
+; CHECK-O3-NEXT:    retq
   %prev = load atomic i64, i64* %p unordered, align 8
   %val = sdiv i64 %prev, %v
   store atomic i64 %val, i64* %p unordered, align 8
@@ -1517,10 +1596,22 @@ define void @rmw_fold_udiv2(i64* %p, i64 %v) {
 ; CHECK-O3-LABEL: rmw_fold_udiv2:
 ; CHECK-O3:       # %bb.0:
 ; CHECK-O3-NEXT:    movq (%rdi), %rax
+; CHECK-O3-NEXT:    movq %rax, %rcx
+; CHECK-O3-NEXT:    orq %rsi, %rcx
+; CHECK-O3-NEXT:    shrq $32, %rcx
+; CHECK-O3-NEXT:    je .LBB72_1
+; CHECK-O3-NEXT:  # %bb.2:
 ; CHECK-O3-NEXT:    xorl %edx, %edx
 ; CHECK-O3-NEXT:    divq %rsi
 ; CHECK-O3-NEXT:    movq %rax, (%rdi)
 ; CHECK-O3-NEXT:    retq
+; CHECK-O3-NEXT:  .LBB72_1:
+; CHECK-O3-NEXT:    # kill: def $eax killed $eax killed $rax
+; CHECK-O3-NEXT:    xorl %edx, %edx
+; CHECK-O3-NEXT:    divl %esi
+; CHECK-O3-NEXT:    # kill: def $eax killed $eax def $rax
+; CHECK-O3-NEXT:    movq %rax, (%rdi)
+; CHECK-O3-NEXT:    retq
   %prev = load atomic i64, i64* %p unordered, align 8
   %val = udiv i64 %prev, %v
   store atomic i64 %val, i64* %p unordered, align 8
@@ -1583,10 +1674,22 @@ define void @rmw_fold_srem2(i64* %p, i64 %v) {
 ; CHECK-O3-LABEL: rmw_fold_srem2:
 ; CHECK-O3:       # %bb.0:
 ; CHECK-O3-NEXT:    movq (%rdi), %rax
+; CHECK-O3-NEXT:    movq %rax, %rcx
+; CHECK-O3-NEXT:    orq %rsi, %rcx
+; CHECK-O3-NEXT:    shrq $32, %rcx
+; CHECK-O3-NEXT:    je .LBB74_1
+; CHECK-O3-NEXT:  # %bb.2:
 ; CHECK-O3-NEXT:    cqto
 ; CHECK-O3-NEXT:    idivq %rsi
 ; CHECK-O3-NEXT:    movq %rdx, (%rdi)
 ; CHECK-O3-NEXT:    retq
+; CHECK-O3-NEXT:  .LBB74_1:
+; CHECK-O3-NEXT:    # kill: def $eax killed $eax killed $rax
+; CHECK-O3-NEXT:    xorl %edx, %edx
+; CHECK-O3-NEXT:    divl %esi
+; CHECK-O3-NEXT:    # kill: def $edx killed $edx def $rdx
+; CHECK-O3-NEXT:    movq %rdx, (%rdi)
+; CHECK-O3-NEXT:    retq
   %prev = load atomic i64, i64* %p unordered, align 8
   %val = srem i64 %prev, %v
   store atomic i64 %val, i64* %p unordered, align 8
@@ -1642,10 +1745,22 @@ define void @rmw_fold_urem2(i64* %p, i64 %v) {
 ; CHECK-O3-LABEL: rmw_fold_urem2:
 ; CHECK-O3:       # %bb.0:
 ; CHECK-O3-NEXT:    movq (%rdi), %rax
+; CHECK-O3-NEXT:    movq %rax, %rcx
+; CHECK-O3-NEXT:    orq %rsi, %rcx
+; CHECK-O3-NEXT:    shrq $32, %rcx
+; CHECK-O3-NEXT:    je .LBB76_1
+; CHECK-O3-NEXT:  # %bb.2:
 ; CHECK-O3-NEXT:    xorl %edx, %edx
 ; CHECK-O3-NEXT:    divq %rsi
 ; CHECK-O3-NEXT:    movq %rdx, (%rdi)
 ; CHECK-O3-NEXT:    retq
+; CHECK-O3-NEXT:  .LBB76_1:
+; CHECK-O3-NEXT:    # kill: def $eax killed $eax killed $rax
+; CHECK-O3-NEXT:    xorl %edx, %edx
+; CHECK-O3-NEXT:    divl %esi
+; CHECK-O3-NEXT:    # kill: def $edx killed $edx def $rdx
+; CHECK-O3-NEXT:    movq %rdx, (%rdi)
+; CHECK-O3-NEXT:    retq
   %prev = load atomic i64, i64* %p unordered, align 8
   %val = urem i64 %prev, %v
   store atomic i64 %val, i64* %p unordered, align 8
@@ -1680,16 +1795,15 @@ define void @rmw_fold_shl2(i64* %p, i64 %v) {
 ; CHECK-O0:       # %bb.0:
 ; CHECK-O0-NEXT:    movq (%rdi), %rax
 ; CHECK-O0-NEXT:    movb %sil, %cl
-; CHECK-O0-NEXT:    shlq %cl, %rax
+; CHECK-O0-NEXT:    # implicit-def: $rsi
+; CHECK-O0-NEXT:    movb %cl, %sil
+; CHECK-O0-NEXT:    shlxq %rsi, %rax, %rax
 ; CHECK-O0-NEXT:    movq %rax, (%rdi)
 ; CHECK-O0-NEXT:    retq
 ;
 ; CHECK-O3-LABEL: rmw_fold_shl2:
 ; CHECK-O3:       # %bb.0:
-; CHECK-O3-NEXT:    movq %rsi, %rcx
-; CHECK-O3-NEXT:    movq (%rdi), %rax
-; CHECK-O3-NEXT:    # kill: def $cl killed $cl killed $rcx
-; CHECK-O3-NEXT:    shlq %cl, %rax
+; CHECK-O3-NEXT:    shlxq %rsi, (%rdi), %rax
 ; CHECK-O3-NEXT:    movq %rax, (%rdi)
 ; CHECK-O3-NEXT:    retq
   %prev = load atomic i64, i64* %p unordered, align 8
@@ -1726,16 +1840,15 @@ define void @rmw_fold_lshr2(i64* %p, i64 %v) {
 ; CHECK-O0:       # %bb.0:
 ; CHECK-O0-NEXT:    movq (%rdi), %rax
 ; CHECK-O0-NEXT:    movb %sil, %cl
-; CHECK-O0-NEXT:    shrq %cl, %rax
+; CHECK-O0-NEXT:    # implicit-def: $rsi
+; CHECK-O0-NEXT:    movb %cl, %sil
+; CHECK-O0-NEXT:    shrxq %rsi, %rax, %rax
 ; CHECK-O0-NEXT:    movq %rax, (%rdi)
 ; CHECK-O0-NEXT:    retq
 ;
 ; CHECK-O3-LABEL: rmw_fold_lshr2:
 ; CHECK-O3:       # %bb.0:
-; CHECK-O3-NEXT:    movq %rsi, %rcx
-; CHECK-O3-NEXT:    movq (%rdi), %rax
-; CHECK-O3-NEXT:    # kill: def $cl killed $cl killed $rcx
-; CHECK-O3-NEXT:    shrq %cl, %rax
+; CHECK-O3-NEXT:    shrxq %rsi, (%rdi), %rax
 ; CHECK-O3-NEXT:    movq %rax, (%rdi)
 ; CHECK-O3-NEXT:    retq
   %prev = load atomic i64, i64* %p unordered, align 8
@@ -1772,16 +1885,15 @@ define void @rmw_fold_ashr2(i64* %p, i64 %v) {
 ; CHECK-O0:       # %bb.0:
 ; CHECK-O0-NEXT:    movq (%rdi), %rax
 ; CHECK-O0-NEXT:    movb %sil, %cl
-; CHECK-O0-NEXT:    sarq %cl, %rax
+; CHECK-O0-NEXT:    # implicit-def: $rsi
+; CHECK-O0-NEXT:    movb %cl, %sil
+; CHECK-O0-NEXT:    sarxq %rsi, %rax, %rax
 ; CHECK-O0-NEXT:    movq %rax, (%rdi)
 ; CHECK-O0-NEXT:    retq
 ;
 ; CHECK-O3-LABEL: rmw_fold_ashr2:
 ; CHECK-O3:       # %bb.0:
-; CHECK-O3-NEXT:    movq %rsi, %rcx
-; CHECK-O3-NEXT:    movq (%rdi), %rax
-; CHECK-O3-NEXT:    # kill: def $cl killed $cl killed $rcx
-; CHECK-O3-NEXT:    sarq %cl, %rax
+; CHECK-O3-NEXT:    sarxq %rsi, (%rdi), %rax
 ; CHECK-O3-NEXT:    movq %rax, (%rdi)
 ; CHECK-O3-NEXT:    retq
   %prev = load atomic i64, i64* %p unordered, align 8