[X86] Add codegen tests for slow-shld scalar funnel shifts

llvm-svn: 347195

2 files changed, 521 insertions, 198 deletions

diff --git a/llvm/test/CodeGen/X86/fshl.ll b/llvm/test/CodeGen/X86/fshl.ll
index 1b03fd41309..ece79f68acc 100644
--- a/llvm/test/CodeGen/X86/fshl.ll
+++ b/llvm/test/CodeGen/X86/fshl.ll
@@ -1,6 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=i686-unknown-unknown | FileCheck %s --check-prefixes=CHECK,X86
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s --check-prefixes=CHECK,X64
+; RUN: llc < %s -mtriple=i686-unknown-unknown | FileCheck %s --check-prefixes=CHECK,X86,X86-FAST
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+slow-shld | FileCheck %s --check-prefixes=CHECK,X86,X86-SLOW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s --check-prefixes=CHECK,X64,X64-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+slow-shld | FileCheck %s --check-prefixes=CHECK,X64,X64-SLOW
 
 declare i8 @llvm.fshl.i8(i8, i8, i8) nounwind readnone
 declare i16 @llvm.fshl.i16(i16, i16, i16) nounwind readnone
@@ -54,65 +56,149 @@ define i8 @var_shift_i8(i8 %x, i8 %y, i8 %z) nounwind {
 }
 
 define i16 @var_shift_i16(i16 %x, i16 %y, i16 %z) nounwind {
-; X86-LABEL: var_shift_i16:
-; X86:       # %bb.0:
-; X86-NEXT:    pushl %esi
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %esi
-; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    andl $15, %ecx
-; X86-NEXT:    movl %eax, %edx
-; X86-NEXT:    shldw %cl, %si, %dx
-; X86-NEXT:    testw %cx, %cx
-; X86-NEXT:    je .LBB1_2
-; X86-NEXT:  # %bb.1:
-; X86-NEXT:    movl %edx, %eax
-; X86-NEXT:  .LBB1_2:
-; X86-NEXT:    # kill: def $ax killed $ax killed $eax
-; X86-NEXT:    popl %esi
-; X86-NEXT:    retl
+; X86-FAST-LABEL: var_shift_i16:
+; X86-FAST:       # %bb.0:
+; X86-FAST-NEXT:    pushl %esi
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-FAST-NEXT:    movzwl {{[0-9]+}}(%esp), %esi
+; X86-FAST-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-FAST-NEXT:    andl $15, %ecx
+; X86-FAST-NEXT:    movl %eax, %edx
+; X86-FAST-NEXT:    shldw %cl, %si, %dx
+; X86-FAST-NEXT:    testw %cx, %cx
+; X86-FAST-NEXT:    je .LBB1_2
+; X86-FAST-NEXT:  # %bb.1:
+; X86-FAST-NEXT:    movl %edx, %eax
+; X86-FAST-NEXT:  .LBB1_2:
+; X86-FAST-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-FAST-NEXT:    popl %esi
+; X86-FAST-NEXT:    retl
 ;
-; X64-LABEL: var_shift_i16:
-; X64:       # %bb.0:
-; X64-NEXT:    movl %edx, %ecx
-; X64-NEXT:    andl $15, %ecx
-; X64-NEXT:    movl %edi, %eax
-; X64-NEXT:    shldw %cl, %si, %ax
-; X64-NEXT:    testw %cx, %cx
-; X64-NEXT:    cmovel %edi, %eax
-; X64-NEXT:    # kill: def $ax killed $ax killed $eax
-; X64-NEXT:    retq
+; X86-SLOW-LABEL: var_shift_i16:
+; X86-SLOW:       # %bb.0:
+; X86-SLOW-NEXT:    pushl %edi
+; X86-SLOW-NEXT:    pushl %esi
+; X86-SLOW-NEXT:    movzwl {{[0-9]+}}(%esp), %esi
+; X86-SLOW-NEXT:    movzwl {{[0-9]+}}(%esp), %edx
+; X86-SLOW-NEXT:    andl $15, %edx
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SLOW-NEXT:    movl %eax, %edi
+; X86-SLOW-NEXT:    movl %edx, %ecx
+; X86-SLOW-NEXT:    shll %cl, %edi
+; X86-SLOW-NEXT:    movl $16, %ecx
+; X86-SLOW-NEXT:    subl %edx, %ecx
+; X86-SLOW-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-SLOW-NEXT:    shrl %cl, %esi
+; X86-SLOW-NEXT:    testw %dx, %dx
+; X86-SLOW-NEXT:    je .LBB1_2
+; X86-SLOW-NEXT:  # %bb.1:
+; X86-SLOW-NEXT:    orl %esi, %edi
+; X86-SLOW-NEXT:    movl %edi, %eax
+; X86-SLOW-NEXT:  .LBB1_2:
+; X86-SLOW-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-SLOW-NEXT:    popl %esi
+; X86-SLOW-NEXT:    popl %edi
+; X86-SLOW-NEXT:    retl
+;
+; X64-FAST-LABEL: var_shift_i16:
+; X64-FAST:       # %bb.0:
+; X64-FAST-NEXT:    movl %edx, %ecx
+; X64-FAST-NEXT:    andl $15, %ecx
+; X64-FAST-NEXT:    movl %edi, %eax
+; X64-FAST-NEXT:    shldw %cl, %si, %ax
+; X64-FAST-NEXT:    testw %cx, %cx
+; X64-FAST-NEXT:    cmovel %edi, %eax
+; X64-FAST-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-FAST-NEXT:    retq
+;
+; X64-SLOW-LABEL: var_shift_i16:
+; X64-SLOW:       # %bb.0:
+; X64-SLOW-NEXT:    movzwl %si, %eax
+; X64-SLOW-NEXT:    andl $15, %edx
+; X64-SLOW-NEXT:    movl %edi, %esi
+; X64-SLOW-NEXT:    movl %edx, %ecx
+; X64-SLOW-NEXT:    shll %cl, %esi
+; X64-SLOW-NEXT:    movl $16, %ecx
+; X64-SLOW-NEXT:    subl %edx, %ecx
+; X64-SLOW-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-SLOW-NEXT:    shrl %cl, %eax
+; X64-SLOW-NEXT:    orl %esi, %eax
+; X64-SLOW-NEXT:    testw %dx, %dx
+; X64-SLOW-NEXT:    cmovel %edi, %eax
+; X64-SLOW-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-SLOW-NEXT:    retq
   %tmp = tail call i16 @llvm.fshl.i16(i16 %x, i16 %y, i16 %z)
   ret i16 %tmp
 }
 
 define i32 @var_shift_i32(i32 %x, i32 %y, i32 %z) nounwind {
-; X86-LABEL: var_shift_i32:
-; X86:       # %bb.0:
-; X86-NEXT:    pushl %esi
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    andl $31, %ecx
-; X86-NEXT:    movl %eax, %edx
-; X86-NEXT:    shldl %cl, %esi, %edx
-; X86-NEXT:    testl %ecx, %ecx
-; X86-NEXT:    je .LBB2_2
-; X86-NEXT:  # %bb.1:
-; X86-NEXT:    movl %edx, %eax
-; X86-NEXT:  .LBB2_2:
-; X86-NEXT:    popl %esi
-; X86-NEXT:    retl
+; X86-FAST-LABEL: var_shift_i32:
+; X86-FAST:       # %bb.0:
+; X86-FAST-NEXT:    pushl %esi
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-FAST-NEXT:    andl $31, %ecx
+; X86-FAST-NEXT:    movl %eax, %edx
+; X86-FAST-NEXT:    shldl %cl, %esi, %edx
+; X86-FAST-NEXT:    testl %ecx, %ecx
+; X86-FAST-NEXT:    je .LBB2_2
+; X86-FAST-NEXT:  # %bb.1:
+; X86-FAST-NEXT:    movl %edx, %eax
+; X86-FAST-NEXT:  .LBB2_2:
+; X86-FAST-NEXT:    popl %esi
+; X86-FAST-NEXT:    retl
 ;
-; X64-LABEL: var_shift_i32:
-; X64:       # %bb.0:
-; X64-NEXT:    movl %edx, %ecx
-; X64-NEXT:    andl $31, %ecx
-; X64-NEXT:    movl %edi, %eax
-; X64-NEXT:    shldl %cl, %esi, %eax
-; X64-NEXT:    testl %ecx, %ecx
-; X64-NEXT:    cmovel %edi, %eax
-; X64-NEXT:    retq
+; X86-SLOW-LABEL: var_shift_i32:
+; X86-SLOW:       # %bb.0:
+; X86-SLOW-NEXT:    pushl %edi
+; X86-SLOW-NEXT:    pushl %esi
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-SLOW-NEXT:    andl $31, %edx
+; X86-SLOW-NEXT:    movl %eax, %edi
+; X86-SLOW-NEXT:    movl %edx, %ecx
+; X86-SLOW-NEXT:    shll %cl, %edi
+; X86-SLOW-NEXT:    movl %edx, %ecx
+; X86-SLOW-NEXT:    negl %ecx
+; X86-SLOW-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-SLOW-NEXT:    shrl %cl, %esi
+; X86-SLOW-NEXT:    testl %edx, %edx
+; X86-SLOW-NEXT:    je .LBB2_2
+; X86-SLOW-NEXT:  # %bb.1:
+; X86-SLOW-NEXT:    orl %esi, %edi
+; X86-SLOW-NEXT:    movl %edi, %eax
+; X86-SLOW-NEXT:  .LBB2_2:
+; X86-SLOW-NEXT:    popl %esi
+; X86-SLOW-NEXT:    popl %edi
+; X86-SLOW-NEXT:    retl
+;
+; X64-FAST-LABEL: var_shift_i32:
+; X64-FAST:       # %bb.0:
+; X64-FAST-NEXT:    movl %edx, %ecx
+; X64-FAST-NEXT:    andl $31, %ecx
+; X64-FAST-NEXT:    movl %edi, %eax
+; X64-FAST-NEXT:    shldl %cl, %esi, %eax
+; X64-FAST-NEXT:    testl %ecx, %ecx
+; X64-FAST-NEXT:    cmovel %edi, %eax
+; X64-FAST-NEXT:    retq
+;
+; X64-SLOW-LABEL: var_shift_i32:
+; X64-SLOW:       # %bb.0:
+; X64-SLOW-NEXT:    movl %esi, %eax
+; X64-SLOW-NEXT:    andl $31, %edx
+; X64-SLOW-NEXT:    movl %edi, %esi
+; X64-SLOW-NEXT:    movl %edx, %ecx
+; X64-SLOW-NEXT:    shll %cl, %esi
+; X64-SLOW-NEXT:    movl %edx, %ecx
+; X64-SLOW-NEXT:    negl %ecx
+; X64-SLOW-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-SLOW-NEXT:    shrl %cl, %eax
+; X64-SLOW-NEXT:    orl %esi, %eax
+; X64-SLOW-NEXT:    testl %edx, %edx
+; X64-SLOW-NEXT:    cmovel %edi, %eax
+; X64-SLOW-NEXT:    retq
   %tmp = tail call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 %z)
   ret i32 %tmp
 }
@@ -174,15 +260,31 @@ define i64 @var_shift_i64(i64 %x, i64 %y, i64 %z) nounwind {
 ; X86-NEXT:    popl %ebp
 ; X86-NEXT:    retl
 ;
-; X64-LABEL: var_shift_i64:
-; X64:       # %bb.0:
-; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    andl $63, %ecx
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    shldq %cl, %rsi, %rax
-; X64-NEXT:    testq %rcx, %rcx
-; X64-NEXT:    cmoveq %rdi, %rax
-; X64-NEXT:    retq
+; X64-FAST-LABEL: var_shift_i64:
+; X64-FAST:       # %bb.0:
+; X64-FAST-NEXT:    movq %rdx, %rcx
+; X64-FAST-NEXT:    andl $63, %ecx
+; X64-FAST-NEXT:    movq %rdi, %rax
+; X64-FAST-NEXT:    shldq %cl, %rsi, %rax
+; X64-FAST-NEXT:    testq %rcx, %rcx
+; X64-FAST-NEXT:    cmoveq %rdi, %rax
+; X64-FAST-NEXT:    retq
+;
+; X64-SLOW-LABEL: var_shift_i64:
+; X64-SLOW:       # %bb.0:
+; X64-SLOW-NEXT:    movq %rsi, %rax
+; X64-SLOW-NEXT:    andl $63, %edx
+; X64-SLOW-NEXT:    movq %rdi, %rsi
+; X64-SLOW-NEXT:    movl %edx, %ecx
+; X64-SLOW-NEXT:    shlq %cl, %rsi
+; X64-SLOW-NEXT:    movl %edx, %ecx
+; X64-SLOW-NEXT:    negl %ecx
+; X64-SLOW-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-SLOW-NEXT:    shrq %cl, %rax
+; X64-SLOW-NEXT:    orq %rsi, %rax
+; X64-SLOW-NEXT:    testq %rdx, %rdx
+; X64-SLOW-NEXT:    cmoveq %rdi, %rax
+; X64-SLOW-NEXT:    retq
   %tmp = tail call i64 @llvm.fshl.i64(i64 %x, i64 %y, i64 %z)
   ret i64 %tmp
 }
@@ -214,55 +316,115 @@ define i8 @const_shift_i8(i8 %x, i8 %y) nounwind {
 }
 
 define i16 @const_shift_i16(i16 %x, i16 %y) nounwind {
-; X86-LABEL: const_shift_i16:
-; X86:       # %bb.0:
-; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    shldw $7, %cx, %ax
-; X86-NEXT:    retl
+; X86-FAST-LABEL: const_shift_i16:
+; X86-FAST:       # %bb.0:
+; X86-FAST-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-FAST-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
+; X86-FAST-NEXT:    shldw $7, %cx, %ax
+; X86-FAST-NEXT:    retl
 ;
-; X64-LABEL: const_shift_i16:
-; X64:       # %bb.0:
-; X64-NEXT:    movl %edi, %eax
-; X64-NEXT:    shldw $7, %si, %ax
-; X64-NEXT:    # kill: def $ax killed $ax killed $eax
-; X64-NEXT:    retq
+; X86-SLOW-LABEL: const_shift_i16:
+; X86-SLOW:       # %bb.0:
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SLOW-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-SLOW-NEXT:    shrl $9, %ecx
+; X86-SLOW-NEXT:    shll $7, %eax
+; X86-SLOW-NEXT:    orl %ecx, %eax
+; X86-SLOW-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-SLOW-NEXT:    retl
+;
+; X64-FAST-LABEL: const_shift_i16:
+; X64-FAST:       # %bb.0:
+; X64-FAST-NEXT:    movl %edi, %eax
+; X64-FAST-NEXT:    shldw $7, %si, %ax
+; X64-FAST-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-FAST-NEXT:    retq
+;
+; X64-SLOW-LABEL: const_shift_i16:
+; X64-SLOW:       # %bb.0:
+; X64-SLOW-NEXT:    movzwl %si, %eax
+; X64-SLOW-NEXT:    shll $7, %edi
+; X64-SLOW-NEXT:    shrl $9, %eax
+; X64-SLOW-NEXT:    orl %edi, %eax
+; X64-SLOW-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-SLOW-NEXT:    retq
   %tmp = tail call i16 @llvm.fshl.i16(i16 %x, i16 %y, i16 7)
   ret i16 %tmp
 }
 
 define i32 @const_shift_i32(i32 %x, i32 %y) nounwind {
-; X86-LABEL: const_shift_i32:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    shldl $7, %ecx, %eax
-; X86-NEXT:    retl
+; X86-FAST-LABEL: const_shift_i32:
+; X86-FAST:       # %bb.0:
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-FAST-NEXT:    shldl $7, %ecx, %eax
+; X86-FAST-NEXT:    retl
 ;
-; X64-LABEL: const_shift_i32:
-; X64:       # %bb.0:
-; X64-NEXT:    movl %edi, %eax
-; X64-NEXT:    shldl $7, %esi, %eax
-; X64-NEXT:    retq
+; X86-SLOW-LABEL: const_shift_i32:
+; X86-SLOW:       # %bb.0:
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-SLOW-NEXT:    shrl $25, %ecx
+; X86-SLOW-NEXT:    shll $7, %eax
+; X86-SLOW-NEXT:    orl %ecx, %eax
+; X86-SLOW-NEXT:    retl
+;
+; X64-FAST-LABEL: const_shift_i32:
+; X64-FAST:       # %bb.0:
+; X64-FAST-NEXT:    movl %edi, %eax
+; X64-FAST-NEXT:    shldl $7, %esi, %eax
+; X64-FAST-NEXT:    retq
+;
+; X64-SLOW-LABEL: const_shift_i32:
+; X64-SLOW:       # %bb.0:
+; X64-SLOW-NEXT:    # kill: def $esi killed $esi def $rsi
+; X64-SLOW-NEXT:    # kill: def $edi killed $edi def $rdi
+; X64-SLOW-NEXT:    shrl $25, %esi
+; X64-SLOW-NEXT:    shll $7, %edi
+; X64-SLOW-NEXT:    leal (%rdi,%rsi), %eax
+; X64-SLOW-NEXT:    retq
   %tmp = tail call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 7)
   ret i32 %tmp
 }
 
 define i64 @const_shift_i64(i64 %x, i64 %y) nounwind {
-; X86-LABEL: const_shift_i64:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NEXT:    shrdl $25, %ecx, %eax
-; X86-NEXT:    shldl $7, %ecx, %edx
-; X86-NEXT:    retl
+; X86-FAST-LABEL: const_shift_i64:
+; X86-FAST:       # %bb.0:
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-FAST-NEXT:    shrdl $25, %ecx, %eax
+; X86-FAST-NEXT:    shldl $7, %ecx, %edx
+; X86-FAST-NEXT:    retl
 ;
-; X64-LABEL: const_shift_i64:
-; X64:       # %bb.0:
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    shldq $7, %rsi, %rax
-; X64-NEXT:    retq
+; X86-SLOW-LABEL: const_shift_i64:
+; X86-SLOW:       # %bb.0:
+; X86-SLOW-NEXT:    pushl %esi
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-SLOW-NEXT:    shrl $25, %esi
+; X86-SLOW-NEXT:    movl %ecx, %eax
+; X86-SLOW-NEXT:    shll $7, %eax
+; X86-SLOW-NEXT:    orl %esi, %eax
+; X86-SLOW-NEXT:    shrl $25, %ecx
+; X86-SLOW-NEXT:    shll $7, %edx
+; X86-SLOW-NEXT:    orl %ecx, %edx
+; X86-SLOW-NEXT:    popl %esi
+; X86-SLOW-NEXT:    retl
+;
+; X64-FAST-LABEL: const_shift_i64:
+; X64-FAST:       # %bb.0:
+; X64-FAST-NEXT:    movq %rdi, %rax
+; X64-FAST-NEXT:    shldq $7, %rsi, %rax
+; X64-FAST-NEXT:    retq
+;
+; X64-SLOW-LABEL: const_shift_i64:
+; X64-SLOW:       # %bb.0:
+; X64-SLOW-NEXT:    shrq $57, %rsi
+; X64-SLOW-NEXT:    shlq $7, %rdi
+; X64-SLOW-NEXT:    leaq (%rdi,%rsi), %rax
+; X64-SLOW-NEXT:    retq
   %tmp = tail call i64 @llvm.fshl.i64(i64 %x, i64 %y, i64 7)
   ret i64 %tmp
 }
diff --git a/llvm/test/CodeGen/X86/fshr.ll b/llvm/test/CodeGen/X86/fshr.ll
index 0fba48d198c..09daeff0a98 100644
--- a/llvm/test/CodeGen/X86/fshr.ll
+++ b/llvm/test/CodeGen/X86/fshr.ll
@@ -1,6 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=i686-unknown-unknown | FileCheck %s --check-prefixes=CHECK,X86
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s --check-prefixes=CHECK,X64
+; RUN: llc < %s -mtriple=i686-unknown-unknown | FileCheck %s --check-prefixes=CHECK,X86,X86-FAST
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+slow-shld | FileCheck %s --check-prefixes=CHECK,X86,X86-SLOW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s --check-prefixes=CHECK,X64,X64-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+slow-shld | FileCheck %s --check-prefixes=CHECK,X64,X64-SLOW
 
 declare i8 @llvm.fshr.i8(i8, i8, i8) nounwind readnone
 declare i16 @llvm.fshr.i16(i16, i16, i16) nounwind readnone
@@ -54,65 +56,148 @@ define i8 @var_shift_i8(i8 %x, i8 %y, i8 %z) nounwind {
 }
 
 define i16 @var_shift_i16(i16 %x, i16 %y, i16 %z) nounwind {
-; X86-LABEL: var_shift_i16:
-; X86:       # %bb.0:
-; X86-NEXT:    pushl %esi
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %esi
-; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    andl $15, %ecx
-; X86-NEXT:    movl %eax, %edx
-; X86-NEXT:    shrdw %cl, %si, %dx
-; X86-NEXT:    testw %cx, %cx
-; X86-NEXT:    je .LBB1_2
-; X86-NEXT:  # %bb.1:
-; X86-NEXT:    movl %edx, %eax
-; X86-NEXT:  .LBB1_2:
-; X86-NEXT:    # kill: def $ax killed $ax killed $eax
-; X86-NEXT:    popl %esi
-; X86-NEXT:    retl
+; X86-FAST-LABEL: var_shift_i16:
+; X86-FAST:       # %bb.0:
+; X86-FAST-NEXT:    pushl %esi
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-FAST-NEXT:    movzwl {{[0-9]+}}(%esp), %esi
+; X86-FAST-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-FAST-NEXT:    andl $15, %ecx
+; X86-FAST-NEXT:    movl %eax, %edx
+; X86-FAST-NEXT:    shrdw %cl, %si, %dx
+; X86-FAST-NEXT:    testw %cx, %cx
+; X86-FAST-NEXT:    je .LBB1_2
+; X86-FAST-NEXT:  # %bb.1:
+; X86-FAST-NEXT:    movl %edx, %eax
+; X86-FAST-NEXT:  .LBB1_2:
+; X86-FAST-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-FAST-NEXT:    popl %esi
+; X86-FAST-NEXT:    retl
 ;
-; X64-LABEL: var_shift_i16:
-; X64:       # %bb.0:
-; X64-NEXT:    movl %edx, %ecx
-; X64-NEXT:    andl $15, %ecx
-; X64-NEXT:    movl %esi, %eax
-; X64-NEXT:    shrdw %cl, %di, %ax
-; X64-NEXT:    testw %cx, %cx
-; X64-NEXT:    cmovel %esi, %eax
-; X64-NEXT:    # kill: def $ax killed $ax killed $eax
-; X64-NEXT:    retq
+; X86-SLOW-LABEL: var_shift_i16:
+; X86-SLOW:       # %bb.0:
+; X86-SLOW-NEXT:    pushl %edi
+; X86-SLOW-NEXT:    pushl %esi
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-SLOW-NEXT:    movzwl {{[0-9]+}}(%esp), %edx
+; X86-SLOW-NEXT:    andl $15, %edx
+; X86-SLOW-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
+; X86-SLOW-NEXT:    movl %eax, %edi
+; X86-SLOW-NEXT:    movl %edx, %ecx
+; X86-SLOW-NEXT:    shrl %cl, %edi
+; X86-SLOW-NEXT:    movl $16, %ecx
+; X86-SLOW-NEXT:    subl %edx, %ecx
+; X86-SLOW-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-SLOW-NEXT:    shll %cl, %esi
+; X86-SLOW-NEXT:    testw %dx, %dx
+; X86-SLOW-NEXT:    je .LBB1_2
+; X86-SLOW-NEXT:  # %bb.1:
+; X86-SLOW-NEXT:    orl %edi, %esi
+; X86-SLOW-NEXT:    movl %esi, %eax
+; X86-SLOW-NEXT:  .LBB1_2:
+; X86-SLOW-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-SLOW-NEXT:    popl %esi
+; X86-SLOW-NEXT:    popl %edi
+; X86-SLOW-NEXT:    retl
+;
+; X64-FAST-LABEL: var_shift_i16:
+; X64-FAST:       # %bb.0:
+; X64-FAST-NEXT:    movl %edx, %ecx
+; X64-FAST-NEXT:    andl $15, %ecx
+; X64-FAST-NEXT:    movl %esi, %eax
+; X64-FAST-NEXT:    shrdw %cl, %di, %ax
+; X64-FAST-NEXT:    testw %cx, %cx
+; X64-FAST-NEXT:    cmovel %esi, %eax
+; X64-FAST-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-FAST-NEXT:    retq
+;
+; X64-SLOW-LABEL: var_shift_i16:
+; X64-SLOW:       # %bb.0:
+; X64-SLOW-NEXT:    movzwl %si, %eax
+; X64-SLOW-NEXT:    andl $15, %edx
+; X64-SLOW-NEXT:    movl %edx, %ecx
+; X64-SLOW-NEXT:    shrl %cl, %eax
+; X64-SLOW-NEXT:    movl $16, %ecx
+; X64-SLOW-NEXT:    subl %edx, %ecx
+; X64-SLOW-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-SLOW-NEXT:    shll %cl, %edi
+; X64-SLOW-NEXT:    orl %edi, %eax
+; X64-SLOW-NEXT:    testw %dx, %dx
+; X64-SLOW-NEXT:    cmovel %esi, %eax
+; X64-SLOW-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-SLOW-NEXT:    retq
   %tmp = tail call i16 @llvm.fshr.i16(i16 %x, i16 %y, i16 %z)
   ret i16 %tmp
 }
 
 define i32 @var_shift_i32(i32 %x, i32 %y, i32 %z) nounwind {
-; X86-LABEL: var_shift_i32:
-; X86:       # %bb.0:
-; X86-NEXT:    pushl %esi
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    andl $31, %ecx
-; X86-NEXT:    movl %eax, %edx
-; X86-NEXT:    shrdl %cl, %esi, %edx
-; X86-NEXT:    testl %ecx, %ecx
-; X86-NEXT:    je .LBB2_2
-; X86-NEXT:  # %bb.1:
-; X86-NEXT:    movl %edx, %eax
-; X86-NEXT:  .LBB2_2:
-; X86-NEXT:    popl %esi
-; X86-NEXT:    retl
+; X86-FAST-LABEL: var_shift_i32:
+; X86-FAST:       # %bb.0:
+; X86-FAST-NEXT:    pushl %esi
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-FAST-NEXT:    andl $31, %ecx
+; X86-FAST-NEXT:    movl %eax, %edx
+; X86-FAST-NEXT:    shrdl %cl, %esi, %edx
+; X86-FAST-NEXT:    testl %ecx, %ecx
+; X86-FAST-NEXT:    je .LBB2_2
+; X86-FAST-NEXT:  # %bb.1:
+; X86-FAST-NEXT:    movl %edx, %eax
+; X86-FAST-NEXT:  .LBB2_2:
+; X86-FAST-NEXT:    popl %esi
+; X86-FAST-NEXT:    retl
 ;
-; X64-LABEL: var_shift_i32:
-; X64:       # %bb.0:
-; X64-NEXT:    movl %edx, %ecx
-; X64-NEXT:    andl $31, %ecx
-; X64-NEXT:    movl %esi, %eax
-; X64-NEXT:    shrdl %cl, %edi, %eax
-; X64-NEXT:    testl %ecx, %ecx
-; X64-NEXT:    cmovel %esi, %eax
-; X64-NEXT:    retq
+; X86-SLOW-LABEL: var_shift_i32:
+; X86-SLOW:       # %bb.0:
+; X86-SLOW-NEXT:    pushl %edi
+; X86-SLOW-NEXT:    pushl %esi
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-SLOW-NEXT:    andl $31, %edx
+; X86-SLOW-NEXT:    movl %eax, %edi
+; X86-SLOW-NEXT:    movl %edx, %ecx
+; X86-SLOW-NEXT:    shrl %cl, %edi
+; X86-SLOW-NEXT:    movl %edx, %ecx
+; X86-SLOW-NEXT:    negl %ecx
+; X86-SLOW-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-SLOW-NEXT:    shll %cl, %esi
+; X86-SLOW-NEXT:    testl %edx, %edx
+; X86-SLOW-NEXT:    je .LBB2_2
+; X86-SLOW-NEXT:  # %bb.1:
+; X86-SLOW-NEXT:    orl %edi, %esi
+; X86-SLOW-NEXT:    movl %esi, %eax
+; X86-SLOW-NEXT:  .LBB2_2:
+; X86-SLOW-NEXT:    popl %esi
+; X86-SLOW-NEXT:    popl %edi
+; X86-SLOW-NEXT:    retl
+;
+; X64-FAST-LABEL: var_shift_i32:
+; X64-FAST:       # %bb.0:
+; X64-FAST-NEXT:    movl %edx, %ecx
+; X64-FAST-NEXT:    andl $31, %ecx
+; X64-FAST-NEXT:    movl %esi, %eax
+; X64-FAST-NEXT:    shrdl %cl, %edi, %eax
+; X64-FAST-NEXT:    testl %ecx, %ecx
+; X64-FAST-NEXT:    cmovel %esi, %eax
+; X64-FAST-NEXT:    retq
+;
+; X64-SLOW-LABEL: var_shift_i32:
+; X64-SLOW:       # %bb.0:
+; X64-SLOW-NEXT:    movl %edi, %eax
+; X64-SLOW-NEXT:    andl $31, %edx
+; X64-SLOW-NEXT:    movl %esi, %edi
+; X64-SLOW-NEXT:    movl %edx, %ecx
+; X64-SLOW-NEXT:    shrl %cl, %edi
+; X64-SLOW-NEXT:    movl %edx, %ecx
+; X64-SLOW-NEXT:    negl %ecx
+; X64-SLOW-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-SLOW-NEXT:    shll %cl, %eax
+; X64-SLOW-NEXT:    orl %edi, %eax
+; X64-SLOW-NEXT:    testl %edx, %edx
+; X64-SLOW-NEXT:    cmovel %esi, %eax
+; X64-SLOW-NEXT:    retq
   %tmp = tail call i32 @llvm.fshr.i32(i32 %x, i32 %y, i32 %z)
   ret i32 %tmp
 }
@@ -170,15 +255,31 @@ define i64 @var_shift_i64(i64 %x, i64 %y, i64 %z) nounwind {
 ; X86-NEXT:    popl %ebp
 ; X86-NEXT:    retl
 ;
-; X64-LABEL: var_shift_i64:
-; X64:       # %bb.0:
-; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    andl $63, %ecx
-; X64-NEXT:    movq %rsi, %rax
-; X64-NEXT:    shrdq %cl, %rdi, %rax
-; X64-NEXT:    testq %rcx, %rcx
-; X64-NEXT:    cmoveq %rsi, %rax
-; X64-NEXT:    retq
+; X64-FAST-LABEL: var_shift_i64:
+; X64-FAST:       # %bb.0:
+; X64-FAST-NEXT:    movq %rdx, %rcx
+; X64-FAST-NEXT:    andl $63, %ecx
+; X64-FAST-NEXT:    movq %rsi, %rax
+; X64-FAST-NEXT:    shrdq %cl, %rdi, %rax
+; X64-FAST-NEXT:    testq %rcx, %rcx
+; X64-FAST-NEXT:    cmoveq %rsi, %rax
+; X64-FAST-NEXT:    retq
+;
+; X64-SLOW-LABEL: var_shift_i64:
+; X64-SLOW:       # %bb.0:
+; X64-SLOW-NEXT:    movq %rdi, %rax
+; X64-SLOW-NEXT:    andl $63, %edx
+; X64-SLOW-NEXT:    movq %rsi, %rdi
+; X64-SLOW-NEXT:    movl %edx, %ecx
+; X64-SLOW-NEXT:    shrq %cl, %rdi
+; X64-SLOW-NEXT:    movl %edx, %ecx
+; X64-SLOW-NEXT:    negl %ecx
+; X64-SLOW-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-SLOW-NEXT:    shlq %cl, %rax
+; X64-SLOW-NEXT:    orq %rdi, %rax
+; X64-SLOW-NEXT:    testq %rdx, %rdx
+; X64-SLOW-NEXT:    cmoveq %rsi, %rax
+; X64-SLOW-NEXT:    retq
   %tmp = tail call i64 @llvm.fshr.i64(i64 %x, i64 %y, i64 %z)
   ret i64 %tmp
 }
@@ -210,55 +311,115 @@ define i8 @const_shift_i8(i8 %x, i8 %y) nounwind {
 }
 
 define i16 @const_shift_i16(i16 %x, i16 %y) nounwind {
-; X86-LABEL: const_shift_i16:
-; X86:       # %bb.0:
-; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    shldw $9, %cx, %ax
-; X86-NEXT:    retl
+; X86-FAST-LABEL: const_shift_i16:
+; X86-FAST:       # %bb.0:
+; X86-FAST-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-FAST-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
+; X86-FAST-NEXT:    shldw $9, %cx, %ax
+; X86-FAST-NEXT:    retl
 ;
-; X64-LABEL: const_shift_i16:
-; X64:       # %bb.0:
-; X64-NEXT:    movl %edi, %eax
-; X64-NEXT:    shldw $9, %si, %ax
-; X64-NEXT:    # kill: def $ax killed $ax killed $eax
-; X64-NEXT:    retq
+; X86-SLOW-LABEL: const_shift_i16:
+; X86-SLOW:       # %bb.0:
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SLOW-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-SLOW-NEXT:    shrl $7, %ecx
+; X86-SLOW-NEXT:    shll $9, %eax
+; X86-SLOW-NEXT:    orl %ecx, %eax
+; X86-SLOW-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-SLOW-NEXT:    retl
+;
+; X64-FAST-LABEL: const_shift_i16:
+; X64-FAST:       # %bb.0:
+; X64-FAST-NEXT:    movl %edi, %eax
+; X64-FAST-NEXT:    shldw $9, %si, %ax
+; X64-FAST-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-FAST-NEXT:    retq
+;
+; X64-SLOW-LABEL: const_shift_i16:
+; X64-SLOW:       # %bb.0:
+; X64-SLOW-NEXT:    movzwl %si, %eax
+; X64-SLOW-NEXT:    shll $9, %edi
+; X64-SLOW-NEXT:    shrl $7, %eax
+; X64-SLOW-NEXT:    orl %edi, %eax
+; X64-SLOW-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-SLOW-NEXT:    retq
   %tmp = tail call i16 @llvm.fshr.i16(i16 %x, i16 %y, i16 7)
   ret i16 %tmp
 }
 
 define i32 @const_shift_i32(i32 %x, i32 %y) nounwind {
-; X86-LABEL: const_shift_i32:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    shldl $25, %ecx, %eax
-; X86-NEXT:    retl
+; X86-FAST-LABEL: const_shift_i32:
+; X86-FAST:       # %bb.0:
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-FAST-NEXT:    shldl $25, %ecx, %eax
+; X86-FAST-NEXT:    retl
 ;
-; X64-LABEL: const_shift_i32:
-; X64:       # %bb.0:
-; X64-NEXT:    movl %edi, %eax
-; X64-NEXT:    shldl $25, %esi, %eax
-; X64-NEXT:    retq
+; X86-SLOW-LABEL: const_shift_i32:
+; X86-SLOW:       # %bb.0:
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-SLOW-NEXT:    shrl $7, %ecx
+; X86-SLOW-NEXT:    shll $25, %eax
+; X86-SLOW-NEXT:    orl %ecx, %eax
+; X86-SLOW-NEXT:    retl
+;
+; X64-FAST-LABEL: const_shift_i32:
+; X64-FAST:       # %bb.0:
+; X64-FAST-NEXT:    movl %edi, %eax
+; X64-FAST-NEXT:    shldl $25, %esi, %eax
+; X64-FAST-NEXT:    retq
+;
+; X64-SLOW-LABEL: const_shift_i32:
+; X64-SLOW:       # %bb.0:
+; X64-SLOW-NEXT:    # kill: def $esi killed $esi def $rsi
+; X64-SLOW-NEXT:    # kill: def $edi killed $edi def $rdi
+; X64-SLOW-NEXT:    shrl $7, %esi
+; X64-SLOW-NEXT:    shll $25, %edi
+; X64-SLOW-NEXT:    leal (%rdi,%rsi), %eax
+; X64-SLOW-NEXT:    retq
   %tmp = tail call i32 @llvm.fshr.i32(i32 %x, i32 %y, i32 7)
   ret i32 %tmp
 }
 
 define i64 @const_shift_i64(i64 %x, i64 %y) nounwind {
-; X86-LABEL: const_shift_i64:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NEXT:    shldl $25, %ecx, %edx
-; X86-NEXT:    shrdl $7, %ecx, %eax
-; X86-NEXT:    retl
+; X86-FAST-LABEL: const_shift_i64:
+; X86-FAST:       # %bb.0:
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-FAST-NEXT:    shldl $25, %ecx, %edx
+; X86-FAST-NEXT:    shrdl $7, %ecx, %eax
+; X86-FAST-NEXT:    retl
 ;
-; X64-LABEL: const_shift_i64:
-; X64:       # %bb.0:
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    shldq $57, %rsi, %rax
-; X64-NEXT:    retq
+; X86-SLOW-LABEL: const_shift_i64:
+; X86-SLOW:       # %bb.0:
+; X86-SLOW-NEXT:    pushl %esi
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-SLOW-NEXT:    movl %ecx, %esi
+; X86-SLOW-NEXT:    shll $25, %esi
+; X86-SLOW-NEXT:    shrl $7, %eax
+; X86-SLOW-NEXT:    orl %esi, %eax
+; X86-SLOW-NEXT:    shrl $7, %ecx
+; X86-SLOW-NEXT:    shll $25, %edx
+; X86-SLOW-NEXT:    orl %ecx, %edx
+; X86-SLOW-NEXT:    popl %esi
+; X86-SLOW-NEXT:    retl
+;
+; X64-FAST-LABEL: const_shift_i64:
+; X64-FAST:       # %bb.0:
+; X64-FAST-NEXT:    movq %rdi, %rax
+; X64-FAST-NEXT:    shldq $57, %rsi, %rax
+; X64-FAST-NEXT:    retq
+;
+; X64-SLOW-LABEL: const_shift_i64:
+; X64-SLOW:       # %bb.0:
+; X64-SLOW-NEXT:    shrq $7, %rsi
+; X64-SLOW-NEXT:    shlq $57, %rdi
+; X64-SLOW-NEXT:    leaq (%rdi,%rsi), %rax
+; X64-SLOW-NEXT:    retq
   %tmp = tail call i64 @llvm.fshr.i64(i64 %x, i64 %y, i64 7)
   ret i64 %tmp
 }