[X86] autogen segmented stacks tests (NFC)

[erikdesjardins]

.

[craig.topper]

LGTM

llvm/test/CodeGen/X86/segmented-stacks-dynamic.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mcpu=generic -mtriple=i686-linux -verify-machineinstrs | FileCheck %s -check-prefix=X86
 ; RUN: llc < %s -mcpu=generic -mtriple=x86_64-linux -verify-machineinstrs | FileCheck %s -check-prefix=X64
 ; RUN: llc < %s -mcpu=generic -mtriple=x86_64-linux-gnux32 -verify-machineinstrs | FileCheck %s -check-prefix=X32ABI
 ; RUN: llc < %s -mcpu=generic -mtriple=i686-linux -filetype=obj
 ; RUN: llc < %s -mcpu=generic -mtriple=x86_64-linux -filetype=obj
 ; RUN: llc < %s -mcpu=generic -mtriple=x86_64-linux-gnux32 -filetype=obj
 
 ; Just to prevent the alloca from being optimized away
 declare void @dummy_use(i32*, i32)
 
 define i32 @test_basic(i32 %l) #0 {
+; X86-LABEL: test_basic:
+; X86:       # %bb.0:
+; X86-NEXT:    cmpl %gs:48, %esp
+; X86-NEXT:    jbe .LBB0_1
+; X86-NEXT:  .LBB0_2:
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    .cfi_def_cfa_offset 8
+; X86-NEXT:    .cfi_offset %ebp, -8
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    .cfi_def_cfa_register %ebp
+; X86-NEXT:    pushl %esi
+; X86-NEXT:    pushl %eax
+; X86-NEXT:    .cfi_offset %esi, -12
+; X86-NEXT:    movl 8(%ebp), %esi
+; X86-NEXT:    leal 15(,%esi,4), %ecx
+; X86-NEXT:    andl $-16, %ecx
+; X86-NEXT:    movl %esp, %eax
+; X86-NEXT:    subl %ecx, %eax
+; X86-NEXT:    cmpl %eax, %gs:48
+; X86-NEXT:    jg .LBB0_4
+; X86-NEXT:  # %bb.3:
+; X86-NEXT:    movl %eax, %esp
+; X86-NEXT:    jmp .LBB0_5
+; X86-NEXT:  .LBB0_4:
+; X86-NEXT:    subl $12, %esp
+; X86-NEXT:    pushl %ecx
+; X86-NEXT:    calll __morestack_allocate_stack_space
+; X86-NEXT:    addl $16, %esp
+; X86-NEXT:  .LBB0_5:
+; X86-NEXT:    subl $8, %esp
+; X86-NEXT:    pushl %esi
+; X86-NEXT:    pushl %eax
+; X86-NEXT:    calll dummy_use@PLT
+; X86-NEXT:    addl $16, %esp
+; X86-NEXT:    testl %esi, %esi
+; X86-NEXT:    je .LBB0_6
+; X86-NEXT:  # %bb.8: # %false
+; X86-NEXT:    addl $-1, %esi
+; X86-NEXT:    subl $12, %esp
+; X86-NEXT:    pushl %esi
+; X86-NEXT:    calll test_basic@PLT
+; X86-NEXT:    jmp .LBB0_7
+; X86-NEXT:  .LBB0_6: # %true
+; X86-NEXT:    xorl %eax, %eax
+; X86-NEXT:  .LBB0_7: # %true
+; X86-NEXT:    leal -4(%ebp), %esp
+; X86-NEXT:    popl %esi
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    .cfi_def_cfa %esp, 4
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_1:
+; X86-NEXT:    .cfi_restore %ebp
+; X86-NEXT:    .cfi_restore %esi
+; X86-NEXT:    pushl $4
+; X86-NEXT:    pushl $12
+; X86-NEXT:    calll __morestack
+; X86-NEXT:    retl
+; X86-NEXT:    jmp .LBB0_2
+;
+; X64-LABEL: test_basic:
+; X64:       # %bb.0:
+; X64-NEXT:    cmpq %fs:112, %rsp
+; X64-NEXT:    jbe .LBB0_1
+; X64-NEXT:  .LBB0_2:
+; X64-NEXT:    pushq %rbp
+; X64-NEXT:    .cfi_def_cfa_offset 16
+; X64-NEXT:    .cfi_offset %rbp, -16
+; X64-NEXT:    movq %rsp, %rbp
+; X64-NEXT:    .cfi_def_cfa_register %rbp
+; X64-NEXT:    pushq %rbx
+; X64-NEXT:    pushq %rax
+; X64-NEXT:    .cfi_offset %rbx, -24
+; X64-NEXT:    movl %edi, %ebx
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    leaq 15(,%rax,4), %rax
+; X64-NEXT:    andq $-16, %rax
+; X64-NEXT:    movq %rsp, %rdi
+; X64-NEXT:    subq %rax, %rdi
+; X64-NEXT:    cmpq %rdi, %fs:112
+; X64-NEXT:    jg .LBB0_4
+; X64-NEXT:  # %bb.3:
+; X64-NEXT:    movq %rdi, %rsp
+; X64-NEXT:    jmp .LBB0_5
+; X64-NEXT:  .LBB0_4:
+; X64-NEXT:    movq %rax, %rdi
+; X64-NEXT:    callq __morestack_allocate_stack_space
+; X64-NEXT:    movq %rax, %rdi
+; X64-NEXT:  .LBB0_5:
+; X64-NEXT:    movl %ebx, %esi
+; X64-NEXT:    callq dummy_use@PLT
+; X64-NEXT:    testl %ebx, %ebx
+; X64-NEXT:    je .LBB0_6
+; X64-NEXT:  # %bb.8: # %false
+; X64-NEXT:    addl $-1, %ebx
+; X64-NEXT:    movl %ebx, %edi
+; X64-NEXT:    callq test_basic@PLT
+; X64-NEXT:    jmp .LBB0_7
+; X64-NEXT:  .LBB0_6: # %true
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:  .LBB0_7: # %true
+; X64-NEXT:    leaq -8(%rbp), %rsp
+; X64-NEXT:    popq %rbx
+; X64-NEXT:    popq %rbp
+; X64-NEXT:    .cfi_def_cfa %rsp, 8
+; X64-NEXT:    retq
+; X64-NEXT:  .LBB0_1:
+; X64-NEXT:    .cfi_restore %rbx
+; X64-NEXT:    .cfi_restore %rbp
+; X64-NEXT:    movabsq $24, %r10
+; X64-NEXT:    movabsq $0, %r11
+; X64-NEXT:    callq __morestack
+; X64-NEXT:    retq
+; X64-NEXT:    jmp .LBB0_2
+;
+; X32ABI-LABEL: test_basic:
+; X32ABI:       # %bb.0:
+; X32ABI-NEXT:    cmpl %fs:64, %esp
+; X32ABI-NEXT:    jbe .LBB0_1
+; X32ABI-NEXT:  .LBB0_2:
+; X32ABI-NEXT:    pushq %rbp
+; X32ABI-NEXT:    .cfi_def_cfa_offset 16
+; X32ABI-NEXT:    .cfi_offset %rbp, -16
+; X32ABI-NEXT:    movl %esp, %ebp
+; X32ABI-NEXT:    .cfi_def_cfa_register %rbp
+; X32ABI-NEXT:    pushq %rbx
+; X32ABI-NEXT:    pushq %rax
+; X32ABI-NEXT:    .cfi_offset %rbx, -24
+; X32ABI-NEXT:    movl %edi, %ebx
+; X32ABI-NEXT:    leal 15(,%rbx,4), %eax
+; X32ABI-NEXT:    andl $-16, %eax
+; X32ABI-NEXT:    movl %esp, %edi
+; X32ABI-NEXT:    subl %eax, %edi
+; X32ABI-NEXT:    cmpl %edi, %fs:64
+; X32ABI-NEXT:    jg .LBB0_4
+; X32ABI-NEXT:  # %bb.3:
+; X32ABI-NEXT:    movl %edi, %esp
+; X32ABI-NEXT:    jmp .LBB0_5
+; X32ABI-NEXT:  .LBB0_4:
+; X32ABI-NEXT:    movl %eax, %edi
+; X32ABI-NEXT:    callq __morestack_allocate_stack_space
+; X32ABI-NEXT:    movl %eax, %edi
+; X32ABI-NEXT:  .LBB0_5:
+; X32ABI-NEXT:    movl %ebx, %esi
+; X32ABI-NEXT:    callq dummy_use@PLT
+; X32ABI-NEXT:    testl %ebx, %ebx
+; X32ABI-NEXT:    je .LBB0_6
+; X32ABI-NEXT:  # %bb.8: # %false
+; X32ABI-NEXT:    addl $-1, %ebx
+; X32ABI-NEXT:    movl %ebx, %edi
+; X32ABI-NEXT:    callq test_basic@PLT
+; X32ABI-NEXT:    jmp .LBB0_7
+; X32ABI-NEXT:  .LBB0_6: # %true
+; X32ABI-NEXT:    xorl %eax, %eax
+; X32ABI-NEXT:  .LBB0_7: # %true
+; X32ABI-NEXT:    leal -8(%ebp), %esp
+; X32ABI-NEXT:    popq %rbx
+; X32ABI-NEXT:    popq %rbp
+; X32ABI-NEXT:    .cfi_def_cfa %rsp, 8
+; X32ABI-NEXT:    retq
+; X32ABI-NEXT:  .LBB0_1:
+; X32ABI-NEXT:    .cfi_def_cfa_register 4294967294
+; X32ABI-NEXT:    .cfi_restore %rbx
+; X32ABI-NEXT:    .cfi_restore %rbp
+; X32ABI-NEXT:    movl $24, %r10d
+; X32ABI-NEXT:    movl $0, %r11d
+; X32ABI-NEXT:    callq __morestack
+; X32ABI-NEXT:    retq
+; X32ABI-NEXT:    jmp .LBB0_2
         %mem = alloca i32, i32 %l
         call void @dummy_use (i32* %mem, i32 %l)
         %terminate = icmp eq i32 %l, 0
         br i1 %terminate, label %true, label %false
 
 true:
         ret i32 0
 
 false:
         %newlen = sub i32 %l, 1
         %retvalue = call i32 @test_basic(i32 %newlen)
         ret i32 %retvalue
 
-; X86-LABEL:      test_basic:
 
-; X86:      cmpl %gs:48, %esp
-; X86-NEXT: jbe .LBB0_1
 
-; X86:      movl %esp, %eax
-; X86:      subl %ecx, %eax
-; X86-NEXT: cmpl %eax, %gs:48
 
-; X86:      movl %eax, %esp
 
-; X86:      subl $12, %esp
-; X86-NEXT: pushl %ecx
-; X86-NEXT: calll __morestack_allocate_stack_space
-; X86-NEXT: addl $16, %esp
 
-; X86:      pushl $4
-; X86-NEXT: pushl $12
-; X86-NEXT: calll __morestack
-; X86-NEXT: ret
 
-; X64-LABEL:      test_basic:
 
-; X64:      cmpq %fs:112, %rsp
-; X64-NEXT: jbe      .LBB0_1
 
-; X64:      movq %rsp, %[[RDI:rdi|rax]]
-; X64:      subq %{{.*}}, %[[RDI]]
-; X64-NEXT: cmpq %[[RDI]], %fs:112
 
-; X64:      movq %[[RDI]], %rsp
 
-; X64:      movq %{{.*}}, %rdi
-; X64-NEXT: callq __morestack_allocate_stack_space
-; X64:      movq %rax, %rdi
 
-; X64:      movabsq $24, %r10
-; X64-NEXT: movabsq $0, %r11
-; X64-NEXT: callq __morestack
-; X64-NEXT: ret
 
-; X32ABI-LABEL:      test_basic:
 
-; X32ABI:      cmpl %fs:64, %esp
-; X32ABI-NEXT: jbe      .LBB0_1
 
-; X32ABI:      movl %esp, %[[EDI:edi|eax]]
-; X32ABI:      subl %{{.*}}, %[[EDI]]
-; X32ABI-NEXT: cmpl %[[EDI]], %fs:64
 
-; X32ABI:      movl %[[EDI]], %esp
 
-; X32ABI:      movl %{{.*}}, %edi
-; X32ABI-NEXT: callq __morestack_allocate_stack_space
-; X32ABI:      movl %eax, %edi
 
-; X32ABI:      movl $24, %r10d
-; X32ABI-NEXT: movl $0, %r11d
-; X32ABI-NEXT: callq __morestack
-; X32ABI-NEXT: ret
 
 }
 
 attributes #0 = { "split-stack" }

llvm/test/CodeGen/X86/segmented-stacks-standalone.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mcpu=generic -mtriple=i686-linux -verify-machineinstrs | FileCheck %s
 ; RUN: llc < %s -mcpu=generic -mtriple=x86_64-linux  -verify-machineinstrs | FileCheck %s
 
 ; This test is standalone because segmented-stacks.ll generates
 ; object-files with both .note.GNU-split-stack (for the split-stack
 ; functions) and .note.GNU-no-split-stack sections (for the
 ; non-split-stack functions). But a split-stack function without a
 ; stack frame should have a .note.GNU-split-stack section regardless
 ; of any other contents of the compilation unit.
 
 define void @test_nostack() #0 {
+; CHECK-LABEL: test_nostack:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    ret{{[l|q]}}
   ret void
 }
 
 attributes #0 = { "split-stack" }
 
 ; CHECK: .section ".note.GNU-split-stack","",@progbits
 ; CHECK: .section ".note.GNU-no-split-stack","",@progbits

llvm/test/CodeGen/X86/segmented-stacks.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mcpu=generic -mtriple=i686-linux -verify-machineinstrs | FileCheck %s -check-prefix=X86-Linux
 ; RUN: llc < %s -mcpu=generic -mtriple=x86_64-linux  -verify-machineinstrs | FileCheck %s -check-prefix=X64-Linux
 ; RUN: llc < %s -mcpu=generic -mtriple=x86_64-linux -code-model=large -verify-machineinstrs | FileCheck %s -check-prefix=X64-Linux-Large
 ; RUN: llc < %s -mcpu=generic -mtriple=x86_64-linux-gnux32 -verify-machineinstrs | FileCheck %s -check-prefix=X32ABI
 ; RUN: llc < %s -mcpu=generic -mtriple=i686-darwin -verify-machineinstrs | FileCheck %s -check-prefix=X86-Darwin
 ; RUN: llc < %s -mcpu=generic -mtriple=x86_64-darwin -verify-machineinstrs | FileCheck %s -check-prefix=X64-Darwin
 ; RUN: llc < %s -mcpu=generic -mtriple=i686-mingw32 -verify-machineinstrs | FileCheck %s -check-prefix=X86-MinGW
 ; RUN: llc < %s -mcpu=generic -mtriple=x86_64-freebsd -verify-machineinstrs | FileCheck %s -check-prefix=X64-FreeBSD

 
 ; X64-Solaris: Segmented stacks not supported on this platform
 ; X86-FreeBSD: Segmented stacks not supported on FreeBSD i386
 
 ; Just to prevent the alloca from being optimized away
 declare void @dummy_use(i32*, i32)
 
 define void @test_basic() #0 {
-        %mem = alloca i32, i32 10
-        call void @dummy_use (i32* %mem, i32 10)
-  ret void
-
-; X86-Linux-LABEL:       test_basic:
-
-; X86-Linux:       cmpl %gs:48, %esp
-; X86-Linux-NEXT:  jbe  .LBB0_1
-
-; X86-Linux:       pushl $0
-; X86-Linux-NEXT:  pushl $44
-; X86-Linux-NEXT:  calll __morestack
-; X86-Linux-NEXT:  ret
-
-; X64-Linux-LABEL:       test_basic:
-
-; X64-Linux:       cmpq %fs:112, %rsp
-; X64-Linux-NEXT:  jbe  .LBB0_1
-
-; X64-Linux:       movabsq $40, %r10
-; X64-Linux-NEXT:  movabsq $0, %r11
-; X64-Linux-NEXT:  callq __morestack
-; X64-Linux-NEXT:  ret
-
-; X64-Linux-Large-LABEL:       test_basic:
-
-; X64-Linux-Large:       cmpq %fs:112, %rsp
-; X64-Linux-Large-NEXT:  jbe  .LBB0_1
-
-; X64-Linux-Large:       movabsq $40, %r10
-; X64-Linux-Large-NEXT:  movabsq $0, %r11
-; X64-Linux-Large-NEXT:  callq *__morestack_addr(%rip)
-; X64-Linux-Large-NEXT:  ret
-
-; X32ABI-LABEL:       test_basic:
-
-; X32ABI:       cmpl %fs:64, %esp
-; X32ABI-NEXT:  jbe .LBB0_1
-
-; X32ABI:       movl $40, %r10d
-; X32ABI-NEXT:  movl $0, %r11d
-; X32ABI-NEXT:  callq __morestack
-; X32ABI-NEXT:  ret
-
-; X86-Darwin-LABEL:      test_basic:
-
-; X86-Darwin:      movl $432, %ecx
-; X86-Darwin-NEXT: cmpl %gs:(%ecx), %esp
-; X86-Darwin-NEXT: jbe  LBB0_1
-
-; X86-Darwin:      pushl $0
-; X86-Darwin-NEXT: pushl $60
-; X86-Darwin-NEXT: calll ___morestack
-; X86-Darwin-NEXT: ret
-
-; X64-Darwin-LABEL:      test_basic:
-
-; X64-Darwin:      cmpq %gs:816, %rsp
-; X64-Darwin-NEXT: jbe  LBB0_1
-
-; X64-Darwin:      movabsq $40, %r10
-; X64-Darwin-NEXT: movabsq $0, %r11
-; X64-Darwin-NEXT: callq ___morestack
-; X64-Darwin-NEXT: ret
-
-; X86-MinGW-LABEL:       test_basic:
-
-; X86-MinGW:       cmpl %fs:20, %esp
-; X86-MinGW-NEXT:  jbe      LBB0_1
-
-; X86-MinGW:       pushl $0
-; X86-MinGW-NEXT:  pushl $40
-; X86-MinGW-NEXT:  calll ___morestack
-; X86-MinGW-NEXT:  ret
-
-; X64-MinGW-LABEL:       test_basic:
-
-; X64-MinGW:       cmpq %gs:40, %rsp
-; X64-MinGW-NEXT:  jbe      .LBB0_1
-
-; X64-MinGW:       movabsq $72, %r10
-; X64-MinGW-NEXT:  movabsq $32, %r11
-; X64-MinGW-NEXT:  callq __morestack
-; X64-MinGW-NEXT:  retq
-
-; X64-FreeBSD-LABEL:       test_basic:
-
-; X64-FreeBSD:       cmpq %fs:24, %rsp
-; X64-FreeBSD-NEXT:  jbe      .LBB0_1
-
-; X64-FreeBSD:       movabsq $40, %r10
-; X64-FreeBSD-NEXT:  movabsq $0, %r11
-; X64-FreeBSD-NEXT:  callq __morestack
-; X64-FreeBSD-NEXT:  ret
-
-; X86-DFlyBSD-LABEL:       test_basic:
-
-; X86-DFlyBSD:       cmpl %fs:16, %esp
-; X86-DFlyBSD-NEXT:  jbe      .LBB0_1
-
-; X86-DFlyBSD:       pushl $0
-; X86-DFlyBSD-NEXT:  pushl $40
-; X86-DFlyBSD-NEXT:  calll __morestack
-; X86-DFlyBSD-NEXT:  ret
-
-; X64-DFlyBSD-LABEL:       test_basic:
-
-; X64-DFlyBSD:       cmpq %fs:32, %rsp
-; X64-DFlyBSD-NEXT:  jbe      .LBB0_1
-
-; X64-DFlyBSD:       movabsq $40, %r10
-; X64-DFlyBSD-NEXT:  movabsq $0, %r11
-; X64-DFlyBSD-NEXT:  callq __morestack
-; X64-DFlyBSD-NEXT:  ret
-
-}
-
-define i32 @test_nested(i32 * nest %closure, i32 %other) #0 {
-       %addend = load i32 , i32 * %closure
-       %result = add i32 %other, %addend
+; X86-Linux-LABEL: test_basic:
+; X86-Linux:       # %bb.0:
+; X86-Linux-NEXT:    cmpl %gs:48, %esp
+; X86-Linux-NEXT:    jbe .LBB0_1
+; X86-Linux-NEXT:  .LBB0_2:
+; X86-Linux-NEXT:    subl $52, %esp
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset 52
+; X86-Linux-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X86-Linux-NEXT:    pushl $10
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-Linux-NEXT:    pushl %eax
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-Linux-NEXT:    calll dummy_use@PLT
+; X86-Linux-NEXT:    addl $60, %esp
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset -60
+; X86-Linux-NEXT:    retl
+; X86-Linux-NEXT:  .LBB0_1:
+; X86-Linux-NEXT:    pushl $0
+; X86-Linux-NEXT:    pushl $44
+; X86-Linux-NEXT:    calll __morestack
+; X86-Linux-NEXT:    retl
+; X86-Linux-NEXT:    jmp .LBB0_2
+;
+; X64-Linux-LABEL: test_basic:
+; X64-Linux:       # %bb.0:
+; X64-Linux-NEXT:    cmpq %fs:112, %rsp
+; X64-Linux-NEXT:    jbe .LBB0_1
+; X64-Linux-NEXT:  .LBB0_2:
+; X64-Linux-NEXT:    subq $40, %rsp
+; X64-Linux-NEXT:    .cfi_def_cfa_offset 48
+; X64-Linux-NEXT:    movq %rsp, %rdi
+; X64-Linux-NEXT:    movl $10, %esi
+; X64-Linux-NEXT:    callq dummy_use@PLT
+; X64-Linux-NEXT:    addq $40, %rsp
+; X64-Linux-NEXT:    .cfi_def_cfa_offset 8
+; X64-Linux-NEXT:    retq
+; X64-Linux-NEXT:  .LBB0_1:
+; X64-Linux-NEXT:    movabsq $40, %r10
+; X64-Linux-NEXT:    movabsq $0, %r11
+; X64-Linux-NEXT:    callq __morestack
+; X64-Linux-NEXT:    retq
+; X64-Linux-NEXT:    jmp .LBB0_2
+;
+; X64-Linux-Large-LABEL: test_basic:
+; X64-Linux-Large:       # %bb.0:
+; X64-Linux-Large-NEXT:    cmpq %fs:112, %rsp
+; X64-Linux-Large-NEXT:    jbe .LBB0_1
+; X64-Linux-Large-NEXT:  .LBB0_2:
+; X64-Linux-Large-NEXT:    subq $40, %rsp
+; X64-Linux-Large-NEXT:    .cfi_def_cfa_offset 48
+; X64-Linux-Large-NEXT:    movabsq $dummy_use, %rax
+; X64-Linux-Large-NEXT:    movq %rsp, %rdi
+; X64-Linux-Large-NEXT:    movl $10, %esi
+; X64-Linux-Large-NEXT:    callq *%rax
+; X64-Linux-Large-NEXT:    addq $40, %rsp
+; X64-Linux-Large-NEXT:    .cfi_def_cfa_offset 8
+; X64-Linux-Large-NEXT:    retq
+; X64-Linux-Large-NEXT:  .LBB0_1:
+; X64-Linux-Large-NEXT:    movabsq $40, %r10
+; X64-Linux-Large-NEXT:    movabsq $0, %r11
+; X64-Linux-Large-NEXT:    callq *__morestack_addr(%rip)
+; X64-Linux-Large-NEXT:    retq
+; X64-Linux-Large-NEXT:    jmp .LBB0_2
+;
+; X32ABI-LABEL: test_basic:
+; X32ABI:       # %bb.0:
+; X32ABI-NEXT:    cmpl %fs:64, %esp
+; X32ABI-NEXT:    jbe .LBB0_1
+; X32ABI-NEXT:  .LBB0_2:
+; X32ABI-NEXT:    subl $40, %esp
+; X32ABI-NEXT:    .cfi_def_cfa_offset 48
+; X32ABI-NEXT:    movl %esp, %edi
+; X32ABI-NEXT:    movl $10, %esi
+; X32ABI-NEXT:    callq dummy_use@PLT
+; X32ABI-NEXT:    addl $40, %esp
+; X32ABI-NEXT:    .cfi_def_cfa_offset 8
+; X32ABI-NEXT:    retq
+; X32ABI-NEXT:  .LBB0_1:
+; X32ABI-NEXT:    movl $40, %r10d
+; X32ABI-NEXT:    movl $0, %r11d
+; X32ABI-NEXT:    callq __morestack
+; X32ABI-NEXT:    retq
+; X32ABI-NEXT:    jmp .LBB0_2
+;
+; X86-Darwin-LABEL: test_basic:
+; X86-Darwin:       ## %bb.0:
+; X86-Darwin-NEXT:    movl $432, %ecx ## imm = 0x1B0
+; X86-Darwin-NEXT:    cmpl %gs:(%ecx), %esp
+; X86-Darwin-NEXT:    jbe LBB0_1
+; X86-Darwin-NEXT:  LBB0_2:
+; X86-Darwin-NEXT:    subl $60, %esp
+; X86-Darwin-NEXT:    .cfi_def_cfa_offset 64
+; X86-Darwin-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X86-Darwin-NEXT:    movl %eax, (%esp)
+; X86-Darwin-NEXT:    movl $10, {{[0-9]+}}(%esp)
+; X86-Darwin-NEXT:    calll _dummy_use
+; X86-Darwin-NEXT:    addl $60, %esp
+; X86-Darwin-NEXT:    retl
+; X86-Darwin-NEXT:  LBB0_1:
+; X86-Darwin-NEXT:    pushl $0
+; X86-Darwin-NEXT:    pushl $60
+; X86-Darwin-NEXT:    calll ___morestack
+; X86-Darwin-NEXT:    retl
+; X86-Darwin-NEXT:    jmp LBB0_2
+;
+; X64-Darwin-LABEL: test_basic:
+; X64-Darwin:       ## %bb.0:
+; X64-Darwin-NEXT:    cmpq %gs:816, %rsp
+; X64-Darwin-NEXT:    jbe LBB0_1
+; X64-Darwin-NEXT:  LBB0_2:
+; X64-Darwin-NEXT:    subq $40, %rsp
+; X64-Darwin-NEXT:    .cfi_def_cfa_offset 48
+; X64-Darwin-NEXT:    movq %rsp, %rdi
+; X64-Darwin-NEXT:    movl $10, %esi
+; X64-Darwin-NEXT:    callq _dummy_use
+; X64-Darwin-NEXT:    addq $40, %rsp
+; X64-Darwin-NEXT:    retq
+; X64-Darwin-NEXT:  LBB0_1:
+; X64-Darwin-NEXT:    movabsq $40, %r10
+; X64-Darwin-NEXT:    movabsq $0, %r11
+; X64-Darwin-NEXT:    callq ___morestack
+; X64-Darwin-NEXT:    retq
+; X64-Darwin-NEXT:    jmp LBB0_2
+;
+; X86-MinGW-LABEL: test_basic:
+; X86-MinGW:       # %bb.0:
+; X86-MinGW-NEXT:    cmpl %fs:20, %esp
+; X86-MinGW-NEXT:    jbe LBB0_1
+; X86-MinGW-NEXT:  LBB0_2:
+; X86-MinGW-NEXT:    subl $40, %esp
+; X86-MinGW-NEXT:    .cfi_def_cfa_offset 44
+; X86-MinGW-NEXT:    movl %esp, %eax
+; X86-MinGW-NEXT:    pushl $10
+; X86-MinGW-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-MinGW-NEXT:    pushl %eax
+; X86-MinGW-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-MinGW-NEXT:    calll _dummy_use
+; X86-MinGW-NEXT:    addl $48, %esp
+; X86-MinGW-NEXT:    .cfi_adjust_cfa_offset -48
+; X86-MinGW-NEXT:    retl
+; X86-MinGW-NEXT:  LBB0_1:
+; X86-MinGW-NEXT:    pushl $0
+; X86-MinGW-NEXT:    pushl $40
+; X86-MinGW-NEXT:    calll ___morestack
+; X86-MinGW-NEXT:    retl
+; X86-MinGW-NEXT:    jmp LBB0_2
+;
+; X64-FreeBSD-LABEL: test_basic:
+; X64-FreeBSD:       # %bb.0:
+; X64-FreeBSD-NEXT:    cmpq %fs:24, %rsp
+; X64-FreeBSD-NEXT:    jbe .LBB0_1
+; X64-FreeBSD-NEXT:  .LBB0_2:
+; X64-FreeBSD-NEXT:    subq $40, %rsp
+; X64-FreeBSD-NEXT:    .cfi_def_cfa_offset 48
+; X64-FreeBSD-NEXT:    movq %rsp, %rdi
+; X64-FreeBSD-NEXT:    movl $10, %esi
+; X64-FreeBSD-NEXT:    callq dummy_use@PLT
+; X64-FreeBSD-NEXT:    addq $40, %rsp
+; X64-FreeBSD-NEXT:    .cfi_def_cfa_offset 8
+; X64-FreeBSD-NEXT:    retq
+; X64-FreeBSD-NEXT:  .LBB0_1:
+; X64-FreeBSD-NEXT:    movabsq $40, %r10
+; X64-FreeBSD-NEXT:    movabsq $0, %r11
+; X64-FreeBSD-NEXT:    callq __morestack
+; X64-FreeBSD-NEXT:    retq
+; X64-FreeBSD-NEXT:    jmp .LBB0_2
+;
+; X86-DFlyBSD-LABEL: test_basic:
+; X86-DFlyBSD:       # %bb.0:
+; X86-DFlyBSD-NEXT:    cmpl %fs:16, %esp
+; X86-DFlyBSD-NEXT:    jbe .LBB0_1
+; X86-DFlyBSD-NEXT:  .LBB0_2:
+; X86-DFlyBSD-NEXT:    subl $40, %esp
+; X86-DFlyBSD-NEXT:    .cfi_def_cfa_offset 44
+; X86-DFlyBSD-NEXT:    movl %esp, %eax
+; X86-DFlyBSD-NEXT:    pushl $10
+; X86-DFlyBSD-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-DFlyBSD-NEXT:    pushl %eax
+; X86-DFlyBSD-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-DFlyBSD-NEXT:    calll dummy_use@PLT
+; X86-DFlyBSD-NEXT:    addl $48, %esp
+; X86-DFlyBSD-NEXT:    .cfi_adjust_cfa_offset -48
+; X86-DFlyBSD-NEXT:    retl
+; X86-DFlyBSD-NEXT:  .LBB0_1:
+; X86-DFlyBSD-NEXT:    pushl $0
+; X86-DFlyBSD-NEXT:    pushl $40
+; X86-DFlyBSD-NEXT:    calll __morestack
+; X86-DFlyBSD-NEXT:    retl
+; X86-DFlyBSD-NEXT:    jmp .LBB0_2
+;
+; X64-DFlyBSD-LABEL: test_basic:
+; X64-DFlyBSD:       # %bb.0:
+; X64-DFlyBSD-NEXT:    cmpq %fs:32, %rsp
+; X64-DFlyBSD-NEXT:    jbe .LBB0_1
+; X64-DFlyBSD-NEXT:  .LBB0_2:
+; X64-DFlyBSD-NEXT:    subq $40, %rsp
+; X64-DFlyBSD-NEXT:    .cfi_def_cfa_offset 48
+; X64-DFlyBSD-NEXT:    movq %rsp, %rdi
+; X64-DFlyBSD-NEXT:    movl $10, %esi
+; X64-DFlyBSD-NEXT:    callq dummy_use@PLT
+; X64-DFlyBSD-NEXT:    addq $40, %rsp
+; X64-DFlyBSD-NEXT:    .cfi_def_cfa_offset 8
+; X64-DFlyBSD-NEXT:    retq
+; X64-DFlyBSD-NEXT:  .LBB0_1:
+; X64-DFlyBSD-NEXT:    movabsq $40, %r10
+; X64-DFlyBSD-NEXT:    movabsq $0, %r11
+; X64-DFlyBSD-NEXT:    callq __morestack
+; X64-DFlyBSD-NEXT:    retq
+; X64-DFlyBSD-NEXT:    jmp .LBB0_2
+;
+; X64-MinGW-LABEL: test_basic:
+; X64-MinGW:       # %bb.0:
+; X64-MinGW-NEXT:    cmpq %gs:40, %rsp
+; X64-MinGW-NEXT:    jbe .LBB0_1
+; X64-MinGW-NEXT:  .LBB0_2:
+; X64-MinGW-NEXT:    subq $72, %rsp
+; X64-MinGW-NEXT:    .seh_stackalloc 72
+; X64-MinGW-NEXT:    .seh_endprologue
+; X64-MinGW-NEXT:    leaq {{[0-9]+}}(%rsp), %rcx
+; X64-MinGW-NEXT:    movl $10, %edx
+; X64-MinGW-NEXT:    callq dummy_use
+; X64-MinGW-NEXT:    nop
+; X64-MinGW-NEXT:    addq $72, %rsp
+; X64-MinGW-NEXT:    retq
+; X64-MinGW-NEXT:  .LBB0_1:
+; X64-MinGW-NEXT:    movabsq $72, %r10
+; X64-MinGW-NEXT:    movabsq $32, %r11
+; X64-MinGW-NEXT:    callq __morestack
+; X64-MinGW-NEXT:    retq
+; X64-MinGW-NEXT:    jmp .LBB0_2
+; X64-MinGW-NEXT:    .seh_endproc
         %mem = alloca i32, i32 10
         call void @dummy_use (i32* %mem, i32 10)
-       ret i32 %result
+  ret void
+}
 
-; X86-Linux:       cmpl %gs:48, %esp
-; X86-Linux-NEXT:  jbe  .LBB1_1
-
-; X86-Linux:       pushl $4
-; X86-Linux-NEXT:  pushl $44
-; X86-Linux-NEXT:  calll __morestack
-; X86-Linux-NEXT:  ret
-
-; X64-Linux:       cmpq %fs:112, %rsp
-; X64-Linux-NEXT:  jbe  .LBB1_1
-
-; X64-Linux:       movq %r10, %rax
-; X64-Linux-NEXT:  movabsq $56, %r10
-; X64-Linux-NEXT:  movabsq $0, %r11
-; X64-Linux-NEXT:  callq __morestack
-; X64-Linux-NEXT:  ret
-; X64-Linux-NEXT:  movq %rax, %r10
-
-; X32ABI:       cmpl %fs:64, %esp
-; X32ABI-NEXT:  jbe .LBB1_1
-
-; X32ABI:       movl %r10d, %eax
-; X32ABI-NEXT:  movl $56, %r10d
-; X32ABI-NEXT:  movl $0, %r11d
-; X32ABI-NEXT:  callq __morestack
-; X32ABI-NEXT:  ret
-; X32ABI-NEXT:  movq %rax, %r10
-
-; X86-Darwin:      movl $432, %edx
-; X86-Darwin-NEXT: cmpl %gs:(%edx), %esp
-; X86-Darwin-NEXT: jbe  LBB1_1
-
-; X86-Darwin:      pushl $4
-; X86-Darwin-NEXT: pushl $60
-; X86-Darwin-NEXT: calll ___morestack
-; X86-Darwin-NEXT: ret
-
-; X64-Darwin:      cmpq %gs:816, %rsp
-; X64-Darwin-NEXT: jbe  LBB1_1
-
-; X64-Darwin:      movq %r10, %rax
-; X64-Darwin-NEXT: movabsq $56, %r10
-; X64-Darwin-NEXT: movabsq $0, %r11
-; X64-Darwin-NEXT: callq ___morestack
-; X64-Darwin-NEXT: ret
-; X64-Darwin-NEXT: movq %rax, %r10
-
-; X86-MinGW:       cmpl %fs:20, %esp
-; X86-MinGW-NEXT:  jbe      LBB1_1
-
-; X86-MinGW:       pushl $4
-; X86-MinGW-NEXT:  pushl $44
-; X86-MinGW-NEXT:  calll ___morestack
-; X86-MinGW-NEXT:  ret
-
-; X64-MinGW-LABEL: test_nested:
-; X64-MinGW:       cmpq %gs:40, %rsp
-; X64-MinGW-NEXT:  jbe      .LBB1_1
-
-; X64-MinGW:       movq %r10, %rax
-; X64-MinGW-NEXT:  movabsq $88, %r10
-; X64-MinGW-NEXT:  movabsq $32, %r11
-; X64-MinGW-NEXT:  callq __morestack
-; X64-MinGW-NEXT:  retq
-; X64-MinGW-NEXT:  movq %rax, %r10
-
-; X64-FreeBSD:       cmpq %fs:24, %rsp
-; X64-FreeBSD-NEXT:  jbe      .LBB1_1
-
-; X64-FreeBSD:       movq %r10, %rax
-; X64-FreeBSD-NEXT:  movabsq $56, %r10
-; X64-FreeBSD-NEXT:  movabsq $0, %r11
-; X64-FreeBSD-NEXT:  callq __morestack
-; X64-FreeBSD-NEXT:  ret
-; X64-FreeBSD-NEXT:  movq %rax, %r10
-
-; X86-DFlyBSD:       cmpl %fs:16, %esp
-; X86-DFlyBSD-NEXT:  jbe      .LBB1_1
-
-; X86-DFlyBSD:       pushl $4
-; X86-DFlyBSD-NEXT:  pushl $44
-; X86-DFlyBSD-NEXT:  calll __morestack
-; X86-DFlyBSD-NEXT:  ret
-
-; X64-DFlyBSD:       cmpq %fs:32, %rsp
-; X64-DFlyBSD-NEXT:  jbe      .LBB1_1
-
-; X64-DFlyBSD:       movq %r10, %rax
-; X64-DFlyBSD-NEXT:  movabsq $56, %r10
-; X64-DFlyBSD-NEXT:  movabsq $0, %r11
-; X64-DFlyBSD-NEXT:  callq __morestack
-; X64-DFlyBSD-NEXT:  ret
-; X64-DFlyBSD-NEXT:  movq %rax, %r10
-
+define i32 @test_nested(i32 * nest %closure, i32 %other) #0 {
+; X86-Linux-LABEL: test_nested:
+; X86-Linux:       # %bb.0:
+; X86-Linux-NEXT:    cmpl %gs:48, %esp
+; X86-Linux-NEXT:    jbe .LBB1_1
+; X86-Linux-NEXT:  .LBB1_2:
+; X86-Linux-NEXT:    pushl %esi
+; X86-Linux-NEXT:    .cfi_def_cfa_offset 8
+; X86-Linux-NEXT:    subl $40, %esp
+; X86-Linux-NEXT:    .cfi_def_cfa_offset 48
+; X86-Linux-NEXT:    .cfi_offset %esi, -8
+; X86-Linux-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-Linux-NEXT:    addl (%ecx), %esi
+; X86-Linux-NEXT:    subl $8, %esp
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset 8
+; X86-Linux-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X86-Linux-NEXT:    pushl $10
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-Linux-NEXT:    pushl %eax
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-Linux-NEXT:    calll dummy_use@PLT
+; X86-Linux-NEXT:    addl $16, %esp
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset -16
+; X86-Linux-NEXT:    movl %esi, %eax
+; X86-Linux-NEXT:    addl $40, %esp
+; X86-Linux-NEXT:    .cfi_def_cfa_offset 8
+; X86-Linux-NEXT:    popl %esi
+; X86-Linux-NEXT:    .cfi_def_cfa_offset 4
+; X86-Linux-NEXT:    retl
+; X86-Linux-NEXT:  .LBB1_1:
+; X86-Linux-NEXT:    .cfi_restore %esi
+; X86-Linux-NEXT:    pushl $4
+; X86-Linux-NEXT:    pushl $44
+; X86-Linux-NEXT:    calll __morestack
+; X86-Linux-NEXT:    retl
+; X86-Linux-NEXT:    jmp .LBB1_2
+;
+; X64-Linux-LABEL: test_nested:
+; X64-Linux:       # %bb.0:
+; X64-Linux-NEXT:    cmpq %fs:112, %rsp
+; X64-Linux-NEXT:    jbe .LBB1_1
+; X64-Linux-NEXT:  .LBB1_2:
+; X64-Linux-NEXT:    pushq %rbx
+; X64-Linux-NEXT:    .cfi_def_cfa_offset 16
+; X64-Linux-NEXT:    subq $48, %rsp
+; X64-Linux-NEXT:    .cfi_def_cfa_offset 64
+; X64-Linux-NEXT:    .cfi_offset %rbx, -16
+; X64-Linux-NEXT:    movl %edi, %ebx
+; X64-Linux-NEXT:    addl (%r10), %ebx
+; X64-Linux-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; X64-Linux-NEXT:    movl $10, %esi
+; X64-Linux-NEXT:    callq dummy_use@PLT
+; X64-Linux-NEXT:    movl %ebx, %eax
+; X64-Linux-NEXT:    addq $48, %rsp
+; X64-Linux-NEXT:    .cfi_def_cfa_offset 16
+; X64-Linux-NEXT:    popq %rbx
+; X64-Linux-NEXT:    .cfi_def_cfa_offset 8
+; X64-Linux-NEXT:    retq
+; X64-Linux-NEXT:  .LBB1_1:
+; X64-Linux-NEXT:    .cfi_restore %rbx
+; X64-Linux-NEXT:    movq %r10, %rax
+; X64-Linux-NEXT:    movabsq $56, %r10
+; X64-Linux-NEXT:    movabsq $0, %r11
+; X64-Linux-NEXT:    callq __morestack
+; X64-Linux-NEXT:    retq
+; X64-Linux-NEXT:    movq %rax, %r10
+; X64-Linux-NEXT:    jmp .LBB1_2
+;
+; X64-Linux-Large-LABEL: test_nested:
+; X64-Linux-Large:       # %bb.0:
+; X64-Linux-Large-NEXT:    cmpq %fs:112, %rsp
+; X64-Linux-Large-NEXT:    jbe .LBB1_1
+; X64-Linux-Large-NEXT:  .LBB1_2:
+; X64-Linux-Large-NEXT:    pushq %rbx
+; X64-Linux-Large-NEXT:    .cfi_def_cfa_offset 16
+; X64-Linux-Large-NEXT:    subq $48, %rsp
+; X64-Linux-Large-NEXT:    .cfi_def_cfa_offset 64
+; X64-Linux-Large-NEXT:    .cfi_offset %rbx, -16
+; X64-Linux-Large-NEXT:    movl %edi, %ebx
+; X64-Linux-Large-NEXT:    addl (%r10), %ebx
+; X64-Linux-Large-NEXT:    movabsq $dummy_use, %rax
+; X64-Linux-Large-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; X64-Linux-Large-NEXT:    movl $10, %esi
+; X64-Linux-Large-NEXT:    callq *%rax
+; X64-Linux-Large-NEXT:    movl %ebx, %eax
+; X64-Linux-Large-NEXT:    addq $48, %rsp
+; X64-Linux-Large-NEXT:    .cfi_def_cfa_offset 16
+; X64-Linux-Large-NEXT:    popq %rbx
+; X64-Linux-Large-NEXT:    .cfi_def_cfa_offset 8
+; X64-Linux-Large-NEXT:    retq
+; X64-Linux-Large-NEXT:  .LBB1_1:
+; X64-Linux-Large-NEXT:    .cfi_restore %rbx
+; X64-Linux-Large-NEXT:    movq %r10, %rax
+; X64-Linux-Large-NEXT:    movabsq $56, %r10
+; X64-Linux-Large-NEXT:    movabsq $0, %r11
+; X64-Linux-Large-NEXT:    callq *__morestack_addr(%rip)
+; X64-Linux-Large-NEXT:    retq
+; X64-Linux-Large-NEXT:    movq %rax, %r10
+; X64-Linux-Large-NEXT:    jmp .LBB1_2
+;
+; X32ABI-LABEL: test_nested:
+; X32ABI:       # %bb.0:
+; X32ABI-NEXT:    cmpl %fs:64, %esp
+; X32ABI-NEXT:    jbe .LBB1_1
+; X32ABI-NEXT:  .LBB1_2:
+; X32ABI-NEXT:    pushq %rbx
+; X32ABI-NEXT:    .cfi_def_cfa_offset 16
+; X32ABI-NEXT:    subl $48, %esp
+; X32ABI-NEXT:    .cfi_def_cfa_offset 64
+; X32ABI-NEXT:    .cfi_offset %rbx, -16
+; X32ABI-NEXT:    movl %edi, %ebx
+; X32ABI-NEXT:    addl (%r10d), %ebx
+; X32ABI-NEXT:    leal {{[0-9]+}}(%rsp), %edi
+; X32ABI-NEXT:    movl $10, %esi
+; X32ABI-NEXT:    callq dummy_use@PLT
+; X32ABI-NEXT:    movl %ebx, %eax
+; X32ABI-NEXT:    addl $48, %esp
+; X32ABI-NEXT:    .cfi_def_cfa_offset 16
+; X32ABI-NEXT:    popq %rbx
+; X32ABI-NEXT:    .cfi_def_cfa_offset 8
+; X32ABI-NEXT:    retq
+; X32ABI-NEXT:  .LBB1_1:
+; X32ABI-NEXT:    .cfi_restore %rbx
+; X32ABI-NEXT:    movl %r10d, %eax
+; X32ABI-NEXT:    movl $56, %r10d
+; X32ABI-NEXT:    movl $0, %r11d
+; X32ABI-NEXT:    callq __morestack
+; X32ABI-NEXT:    retq
+; X32ABI-NEXT:    movq %rax, %r10
+; X32ABI-NEXT:    jmp .LBB1_2
+;
+; X86-Darwin-LABEL: test_nested:
+; X86-Darwin:       ## %bb.0:
+; X86-Darwin-NEXT:    movl $432, %edx ## imm = 0x1B0
+; X86-Darwin-NEXT:    cmpl %gs:(%edx), %esp
+; X86-Darwin-NEXT:    jbe LBB1_1
+; X86-Darwin-NEXT:  LBB1_2:
+; X86-Darwin-NEXT:    pushl %esi
+; X86-Darwin-NEXT:    .cfi_def_cfa_offset 8
+; X86-Darwin-NEXT:    subl $56, %esp
+; X86-Darwin-NEXT:    .cfi_def_cfa_offset 64
+; X86-Darwin-NEXT:    .cfi_offset %esi, -8
+; X86-Darwin-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-Darwin-NEXT:    addl (%ecx), %esi
+; X86-Darwin-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X86-Darwin-NEXT:    movl %eax, (%esp)
+; X86-Darwin-NEXT:    movl $10, {{[0-9]+}}(%esp)
+; X86-Darwin-NEXT:    calll _dummy_use
+; X86-Darwin-NEXT:    movl %esi, %eax
+; X86-Darwin-NEXT:    addl $56, %esp
+; X86-Darwin-NEXT:    popl %esi
+; X86-Darwin-NEXT:    retl
+; X86-Darwin-NEXT:  LBB1_1:
+; X86-Darwin-NEXT:    pushl $4
+; X86-Darwin-NEXT:    pushl $60
+; X86-Darwin-NEXT:    calll ___morestack
+; X86-Darwin-NEXT:    retl
+; X86-Darwin-NEXT:    jmp LBB1_2
+;
+; X64-Darwin-LABEL: test_nested:
+; X64-Darwin:       ## %bb.0:
+; X64-Darwin-NEXT:    cmpq %gs:816, %rsp
+; X64-Darwin-NEXT:    jbe LBB1_1
+; X64-Darwin-NEXT:  LBB1_2:
+; X64-Darwin-NEXT:    pushq %rbx
+; X64-Darwin-NEXT:    .cfi_def_cfa_offset 16
+; X64-Darwin-NEXT:    subq $48, %rsp
+; X64-Darwin-NEXT:    .cfi_def_cfa_offset 64
+; X64-Darwin-NEXT:    .cfi_offset %rbx, -16
+; X64-Darwin-NEXT:    movl %edi, %ebx
+; X64-Darwin-NEXT:    addl (%r10), %ebx
+; X64-Darwin-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; X64-Darwin-NEXT:    movl $10, %esi
+; X64-Darwin-NEXT:    callq _dummy_use
+; X64-Darwin-NEXT:    movl %ebx, %eax
+; X64-Darwin-NEXT:    addq $48, %rsp
+; X64-Darwin-NEXT:    popq %rbx
+; X64-Darwin-NEXT:    retq
+; X64-Darwin-NEXT:  LBB1_1:
+; X64-Darwin-NEXT:    movq %r10, %rax
+; X64-Darwin-NEXT:    movabsq $56, %r10
+; X64-Darwin-NEXT:    movabsq $0, %r11
+; X64-Darwin-NEXT:    callq ___morestack
+; X64-Darwin-NEXT:    retq
+; X64-Darwin-NEXT:    movq %rax, %r10
+; X64-Darwin-NEXT:    jmp LBB1_2
+;
+; X86-MinGW-LABEL: test_nested:
+; X86-MinGW:       # %bb.0:
+; X86-MinGW-NEXT:    cmpl %fs:20, %esp
+; X86-MinGW-NEXT:    jbe LBB1_1
+; X86-MinGW-NEXT:  LBB1_2:
+; X86-MinGW-NEXT:    pushl %esi
+; X86-MinGW-NEXT:    .cfi_def_cfa_offset 8
+; X86-MinGW-NEXT:    subl $40, %esp
+; X86-MinGW-NEXT:    .cfi_def_cfa_offset 48
+; X86-MinGW-NEXT:    .cfi_offset %esi, -8
+; X86-MinGW-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-MinGW-NEXT:    addl (%ecx), %esi
+; X86-MinGW-NEXT:    movl %esp, %eax
+; X86-MinGW-NEXT:    pushl $10
+; X86-MinGW-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-MinGW-NEXT:    pushl %eax
+; X86-MinGW-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-MinGW-NEXT:    calll _dummy_use
+; X86-MinGW-NEXT:    addl $8, %esp
+; X86-MinGW-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-MinGW-NEXT:    movl %esi, %eax
+; X86-MinGW-NEXT:    addl $40, %esp
+; X86-MinGW-NEXT:    popl %esi
+; X86-MinGW-NEXT:    retl
+; X86-MinGW-NEXT:  LBB1_1:
+; X86-MinGW-NEXT:    .cfi_def_cfa_offset 4
+; X86-MinGW-NEXT:    .cfi_restore %esi
+; X86-MinGW-NEXT:    pushl $4
+; X86-MinGW-NEXT:    pushl $44
+; X86-MinGW-NEXT:    calll ___morestack
+; X86-MinGW-NEXT:    retl
+; X86-MinGW-NEXT:    jmp LBB1_2
+;
+; X64-FreeBSD-LABEL: test_nested:
+; X64-FreeBSD:       # %bb.0:
+; X64-FreeBSD-NEXT:    cmpq %fs:24, %rsp
+; X64-FreeBSD-NEXT:    jbe .LBB1_1
+; X64-FreeBSD-NEXT:  .LBB1_2:
+; X64-FreeBSD-NEXT:    pushq %rbx
+; X64-FreeBSD-NEXT:    .cfi_def_cfa_offset 16
+; X64-FreeBSD-NEXT:    subq $48, %rsp
+; X64-FreeBSD-NEXT:    .cfi_def_cfa_offset 64
+; X64-FreeBSD-NEXT:    .cfi_offset %rbx, -16
+; X64-FreeBSD-NEXT:    movl %edi, %ebx
+; X64-FreeBSD-NEXT:    addl (%r10), %ebx
+; X64-FreeBSD-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; X64-FreeBSD-NEXT:    movl $10, %esi
+; X64-FreeBSD-NEXT:    callq dummy_use@PLT
+; X64-FreeBSD-NEXT:    movl %ebx, %eax
+; X64-FreeBSD-NEXT:    addq $48, %rsp
+; X64-FreeBSD-NEXT:    .cfi_def_cfa_offset 16
+; X64-FreeBSD-NEXT:    popq %rbx
+; X64-FreeBSD-NEXT:    .cfi_def_cfa_offset 8
+; X64-FreeBSD-NEXT:    retq
+; X64-FreeBSD-NEXT:  .LBB1_1:
+; X64-FreeBSD-NEXT:    .cfi_restore %rbx
+; X64-FreeBSD-NEXT:    movq %r10, %rax
+; X64-FreeBSD-NEXT:    movabsq $56, %r10
+; X64-FreeBSD-NEXT:    movabsq $0, %r11
+; X64-FreeBSD-NEXT:    callq __morestack
+; X64-FreeBSD-NEXT:    retq
+; X64-FreeBSD-NEXT:    movq %rax, %r10
+; X64-FreeBSD-NEXT:    jmp .LBB1_2
+;
+; X86-DFlyBSD-LABEL: test_nested:
+; X86-DFlyBSD:       # %bb.0:
+; X86-DFlyBSD-NEXT:    cmpl %fs:16, %esp
+; X86-DFlyBSD-NEXT:    jbe .LBB1_1
+; X86-DFlyBSD-NEXT:  .LBB1_2:
+; X86-DFlyBSD-NEXT:    pushl %esi
+; X86-DFlyBSD-NEXT:    .cfi_def_cfa_offset 8
+; X86-DFlyBSD-NEXT:    subl $40, %esp
+; X86-DFlyBSD-NEXT:    .cfi_def_cfa_offset 48
+; X86-DFlyBSD-NEXT:    .cfi_offset %esi, -8
+; X86-DFlyBSD-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-DFlyBSD-NEXT:    addl (%ecx), %esi
+; X86-DFlyBSD-NEXT:    movl %esp, %eax
+; X86-DFlyBSD-NEXT:    pushl $10
+; X86-DFlyBSD-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-DFlyBSD-NEXT:    pushl %eax
+; X86-DFlyBSD-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-DFlyBSD-NEXT:    calll dummy_use@PLT
+; X86-DFlyBSD-NEXT:    addl $8, %esp
+; X86-DFlyBSD-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-DFlyBSD-NEXT:    movl %esi, %eax
+; X86-DFlyBSD-NEXT:    addl $40, %esp
+; X86-DFlyBSD-NEXT:    .cfi_def_cfa_offset 8
+; X86-DFlyBSD-NEXT:    popl %esi
+; X86-DFlyBSD-NEXT:    .cfi_def_cfa_offset 4
+; X86-DFlyBSD-NEXT:    retl
+; X86-DFlyBSD-NEXT:  .LBB1_1:
+; X86-DFlyBSD-NEXT:    .cfi_restore %esi
+; X86-DFlyBSD-NEXT:    pushl $4
+; X86-DFlyBSD-NEXT:    pushl $44
+; X86-DFlyBSD-NEXT:    calll __morestack
+; X86-DFlyBSD-NEXT:    retl
+; X86-DFlyBSD-NEXT:    jmp .LBB1_2
+;
+; X64-DFlyBSD-LABEL: test_nested:
+; X64-DFlyBSD:       # %bb.0:
+; X64-DFlyBSD-NEXT:    cmpq %fs:32, %rsp
+; X64-DFlyBSD-NEXT:    jbe .LBB1_1
+; X64-DFlyBSD-NEXT:  .LBB1_2:
+; X64-DFlyBSD-NEXT:    pushq %rbx
+; X64-DFlyBSD-NEXT:    .cfi_def_cfa_offset 16
+; X64-DFlyBSD-NEXT:    subq $48, %rsp
+; X64-DFlyBSD-NEXT:    .cfi_def_cfa_offset 64
+; X64-DFlyBSD-NEXT:    .cfi_offset %rbx, -16
+; X64-DFlyBSD-NEXT:    movl %edi, %ebx
+; X64-DFlyBSD-NEXT:    addl (%r10), %ebx
+; X64-DFlyBSD-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; X64-DFlyBSD-NEXT:    movl $10, %esi
+; X64-DFlyBSD-NEXT:    callq dummy_use@PLT
+; X64-DFlyBSD-NEXT:    movl %ebx, %eax
+; X64-DFlyBSD-NEXT:    addq $48, %rsp
+; X64-DFlyBSD-NEXT:    .cfi_def_cfa_offset 16
+; X64-DFlyBSD-NEXT:    popq %rbx
+; X64-DFlyBSD-NEXT:    .cfi_def_cfa_offset 8
+; X64-DFlyBSD-NEXT:    retq
+; X64-DFlyBSD-NEXT:  .LBB1_1:
+; X64-DFlyBSD-NEXT:    .cfi_restore %rbx
+; X64-DFlyBSD-NEXT:    movq %r10, %rax
+; X64-DFlyBSD-NEXT:    movabsq $56, %r10
+; X64-DFlyBSD-NEXT:    movabsq $0, %r11
+; X64-DFlyBSD-NEXT:    callq __morestack
+; X64-DFlyBSD-NEXT:    retq
+; X64-DFlyBSD-NEXT:    movq %rax, %r10
+; X64-DFlyBSD-NEXT:    jmp .LBB1_2
+;
+; X64-MinGW-LABEL: test_nested:
+; X64-MinGW:       # %bb.0:
+; X64-MinGW-NEXT:    cmpq %gs:40, %rsp
+; X64-MinGW-NEXT:    jbe .LBB1_1
+; X64-MinGW-NEXT:  .LBB1_2:
+; X64-MinGW-NEXT:    pushq %rsi
+; X64-MinGW-NEXT:    .seh_pushreg %rsi
+; X64-MinGW-NEXT:    subq $80, %rsp
+; X64-MinGW-NEXT:    .seh_stackalloc 80
+; X64-MinGW-NEXT:    .seh_endprologue
+; X64-MinGW-NEXT:    movl %ecx, %esi
+; X64-MinGW-NEXT:    addl (%r10), %esi
+; X64-MinGW-NEXT:    leaq {{[0-9]+}}(%rsp), %rcx
+; X64-MinGW-NEXT:    movl $10, %edx
+; X64-MinGW-NEXT:    callq dummy_use
+; X64-MinGW-NEXT:    movl %esi, %eax
+; X64-MinGW-NEXT:    addq $80, %rsp
+; X64-MinGW-NEXT:    popq %rsi
+; X64-MinGW-NEXT:    retq
+; X64-MinGW-NEXT:  .LBB1_1:
+; X64-MinGW-NEXT:    movq %r10, %rax
+; X64-MinGW-NEXT:    movabsq $88, %r10
+; X64-MinGW-NEXT:    movabsq $32, %r11
+; X64-MinGW-NEXT:    callq __morestack
+; X64-MinGW-NEXT:    retq
+; X64-MinGW-NEXT:    movq %rax, %r10
+; X64-MinGW-NEXT:    jmp .LBB1_2
+; X64-MinGW-NEXT:    .seh_endproc
+       %addend = load i32 , i32 * %closure
+       %result = add i32 %other, %addend
+       %mem = alloca i32, i32 10
+       call void @dummy_use (i32* %mem, i32 10)
+       ret i32 %result
 }
 
 define void @test_large() #0 {
-        %mem = alloca i32, i32 10000
-        call void @dummy_use (i32* %mem, i32 3)
-        ret void
-
-; X86-Linux-LABEL:       test_large:
-
-; X86-Linux:       leal -40012(%esp), %ecx
-; X86-Linux-NEXT:  cmpl %gs:48, %ecx
-; X86-Linux-NEXT:  jbe  .LBB2_1
-
-; X86-Linux:       pushl $0
-; X86-Linux-NEXT:  pushl $40012
-; X86-Linux-NEXT:  calll __morestack
-; X86-Linux-NEXT:  ret
-
-; X64-Linux:       leaq -40008(%rsp), %r11
-; X64-Linux-NEXT:  cmpq %fs:112, %r11
-; X64-Linux-NEXT:  jbe  .LBB2_1
-
-; X64-Linux:       movabsq $40008, %r10
-; X64-Linux-NEXT:  movabsq $0, %r11
-; X64-Linux-NEXT:  callq __morestack
-; X64-Linux-NEXT:  ret
-
-; X32ABI:       leal -40008(%rsp), %r11d
-; X32ABI-NEXT:  cmpl %fs:64, %r11d
-; X32ABI-NEXT:  jbe .LBB2_1
-
-; X32ABI:       movl $40008, %r10d
-; X32ABI-NEXT:  movl $0, %r11d
-; X32ABI-NEXT:  callq __morestack
-; X32ABI-NEXT:  ret
-
-; X86-Darwin:      leal -40012(%esp), %ecx
-; X86-Darwin-NEXT: movl $432, %eax
-; X86-Darwin-NEXT: cmpl %gs:(%eax), %ecx
-; X86-Darwin-NEXT: jbe  LBB2_1
-
-; X86-Darwin:      pushl $0
-; X86-Darwin-NEXT: pushl $40012
-; X86-Darwin-NEXT: calll ___morestack
-; X86-Darwin-NEXT: ret
-
-; X64-Darwin:      leaq -40008(%rsp), %r11
-; X64-Darwin-NEXT: cmpq %gs:816, %r11
-; X64-Darwin-NEXT: jbe      LBB2_1
-
-; X64-Darwin:      movabsq $40008, %r10
-; X64-Darwin-NEXT: movabsq $0, %r11
-; X64-Darwin-NEXT: callq ___morestack
-; X64-Darwin-NEXT: ret
-
-; X86-MinGW:       leal -40000(%esp), %ecx
-; X86-MinGW-NEXT:  cmpl %fs:20, %ecx
-; X86-MinGW-NEXT:  jbe      LBB2_1
-
-; X86-MinGW:       pushl $0
-; X86-MinGW-NEXT:  pushl $40000
-; X86-MinGW-NEXT:  calll ___morestack
-; X86-MinGW-NEXT:  ret
-
-; X64-MinGW-LABEL: test_large:
-; X64-MinGW:       leaq -40040(%rsp), %r11
-; X64-MinGW-NEXT:  cmpq %gs:40, %r11
-; X64-MinGW-NEXT:  jbe      .LBB2_1
-
-; X64-MinGW:       movabsq $40040, %r10
-; X64-MinGW-NEXT:  movabsq $32, %r11
-; X64-MinGW-NEXT:  callq __morestack
-; X64-MinGW-NEXT:  retq
-
-; X64-FreeBSD:       leaq -40008(%rsp), %r11
-; X64-FreeBSD-NEXT:  cmpq %fs:24, %r11
-; X64-FreeBSD-NEXT:  jbe      .LBB2_1
-
-; X64-FreeBSD:       movabsq $40008, %r10
-; X64-FreeBSD-NEXT:  movabsq $0, %r11
-; X64-FreeBSD-NEXT:  callq __morestack
-; X64-FreeBSD-NEXT:  ret
-
-; X86-DFlyBSD:       leal -40000(%esp), %ecx
-; X86-DFlyBSD-NEXT:  cmpl %fs:16, %ecx
-; X86-DFlyBSD-NEXT:  jbe      .LBB2_1
-
-; X86-DFlyBSD:       pushl $0
-; X86-DFlyBSD-NEXT:  pushl $40000
-; X86-DFlyBSD-NEXT:  calll __morestack
-; X86-DFlyBSD-NEXT:  ret
-
-; X64-DFlyBSD:       leaq -40008(%rsp), %r11
-; X64-DFlyBSD-NEXT:  cmpq %fs:32, %r11
-; X64-DFlyBSD-NEXT:  jbe      .LBB2_1
-
-; X64-DFlyBSD:       movabsq $40008, %r10
-; X64-DFlyBSD-NEXT:  movabsq $0, %r11
-; X64-DFlyBSD-NEXT:  callq __morestack
-; X64-DFlyBSD-NEXT:  ret
-
+; X86-Linux-LABEL: test_large:
+; X86-Linux:       # %bb.0:
+; X86-Linux-NEXT:    leal -{{[0-9]+}}(%esp), %ecx
+; X86-Linux-NEXT:    cmpl %gs:48, %ecx
+; X86-Linux-NEXT:    jbe .LBB2_1
+; X86-Linux-NEXT:  .LBB2_2:
+; X86-Linux-NEXT:    subl $40020, %esp # imm = 0x9C54
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset 40020
+; X86-Linux-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X86-Linux-NEXT:    pushl $3
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-Linux-NEXT:    pushl %eax
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-Linux-NEXT:    calll dummy_use@PLT
+; X86-Linux-NEXT:    addl $40028, %esp # imm = 0x9C5C
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset -40028
+; X86-Linux-NEXT:    retl
+; X86-Linux-NEXT:  .LBB2_1:
+; X86-Linux-NEXT:    pushl $0
+; X86-Linux-NEXT:    pushl $40012 # imm = 0x9C4C
+; X86-Linux-NEXT:    calll __morestack
+; X86-Linux-NEXT:    retl
+; X86-Linux-NEXT:    jmp .LBB2_2
+;
+; X64-Linux-LABEL: test_large:
+; X64-Linux:       # %bb.0:
+; X64-Linux-NEXT:    leaq -{{[0-9]+}}(%rsp), %r11
+; X64-Linux-NEXT:    cmpq %fs:112, %r11
+; X64-Linux-NEXT:    jbe .LBB2_1
+; X64-Linux-NEXT:  .LBB2_2:
+; X64-Linux-NEXT:    subq $40008, %rsp # imm = 0x9C48
+; X64-Linux-NEXT:    .cfi_def_cfa_offset 40016
+; X64-Linux-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; X64-Linux-NEXT:    movl $3, %esi
+; X64-Linux-NEXT:    callq dummy_use@PLT
+; X64-Linux-NEXT:    addq $40008, %rsp # imm = 0x9C48
+; X64-Linux-NEXT:    .cfi_def_cfa_offset 8
+; X64-Linux-NEXT:    retq
+; X64-Linux-NEXT:  .LBB2_1:
+; X64-Linux-NEXT:    movabsq $40008, %r10 # imm = 0x9C48
+; X64-Linux-NEXT:    movabsq $0, %r11
+; X64-Linux-NEXT:    callq __morestack
+; X64-Linux-NEXT:    retq
+; X64-Linux-NEXT:    jmp .LBB2_2
+;
+; X64-Linux-Large-LABEL: test_large:
+; X64-Linux-Large:       # %bb.0:
+; X64-Linux-Large-NEXT:    leaq -{{[0-9]+}}(%rsp), %r11
+; X64-Linux-Large-NEXT:    cmpq %fs:112, %r11
+; X64-Linux-Large-NEXT:    jbe .LBB2_1
+; X64-Linux-Large-NEXT:  .LBB2_2:
+; X64-Linux-Large-NEXT:    subq $40008, %rsp # imm = 0x9C48
+; X64-Linux-Large-NEXT:    .cfi_def_cfa_offset 40016
+; X64-Linux-Large-NEXT:    movabsq $dummy_use, %rax
+; X64-Linux-Large-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; X64-Linux-Large-NEXT:    movl $3, %esi
+; X64-Linux-Large-NEXT:    callq *%rax
+; X64-Linux-Large-NEXT:    addq $40008, %rsp # imm = 0x9C48
+; X64-Linux-Large-NEXT:    .cfi_def_cfa_offset 8
+; X64-Linux-Large-NEXT:    retq
+; X64-Linux-Large-NEXT:  .LBB2_1:
+; X64-Linux-Large-NEXT:    movabsq $40008, %r10 # imm = 0x9C48
+; X64-Linux-Large-NEXT:    movabsq $0, %r11
+; X64-Linux-Large-NEXT:    callq *__morestack_addr(%rip)
+; X64-Linux-Large-NEXT:    retq
+; X64-Linux-Large-NEXT:    jmp .LBB2_2
+;
+; X32ABI-LABEL: test_large:
+; X32ABI:       # %bb.0:
+; X32ABI-NEXT:    leal -{{[0-9]+}}(%rsp), %r11d
+; X32ABI-NEXT:    cmpl %fs:64, %r11d
+; X32ABI-NEXT:    jbe .LBB2_1
+; X32ABI-NEXT:  .LBB2_2:
+; X32ABI-NEXT:    subl $40008, %esp # imm = 0x9C48
+; X32ABI-NEXT:    .cfi_def_cfa_offset 40016
+; X32ABI-NEXT:    leal {{[0-9]+}}(%rsp), %edi
+; X32ABI-NEXT:    movl $3, %esi
+; X32ABI-NEXT:    callq dummy_use@PLT
+; X32ABI-NEXT:    addl $40008, %esp # imm = 0x9C48
+; X32ABI-NEXT:    .cfi_def_cfa_offset 8
+; X32ABI-NEXT:    retq
+; X32ABI-NEXT:  .LBB2_1:
+; X32ABI-NEXT:    movl $40008, %r10d # imm = 0x9C48
+; X32ABI-NEXT:    movl $0, %r11d
+; X32ABI-NEXT:    callq __morestack
+; X32ABI-NEXT:    retq
+; X32ABI-NEXT:    jmp .LBB2_2
+;
+; X86-Darwin-LABEL: test_large:
+; X86-Darwin:       ## %bb.0:
+; X86-Darwin-NEXT:    leal -{{[0-9]+}}(%esp), %ecx
+; X86-Darwin-NEXT:    movl $432, %eax ## imm = 0x1B0
+; X86-Darwin-NEXT:    cmpl %gs:(%eax), %ecx
+; X86-Darwin-NEXT:    jbe LBB2_1
+; X86-Darwin-NEXT:  LBB2_2:
+; X86-Darwin-NEXT:    subl $40012, %esp ## imm = 0x9C4C
+; X86-Darwin-NEXT:    .cfi_def_cfa_offset 40016
+; X86-Darwin-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X86-Darwin-NEXT:    movl %eax, (%esp)
+; X86-Darwin-NEXT:    movl $3, {{[0-9]+}}(%esp)
+; X86-Darwin-NEXT:    calll _dummy_use
+; X86-Darwin-NEXT:    addl $40012, %esp ## imm = 0x9C4C
+; X86-Darwin-NEXT:    retl
+; X86-Darwin-NEXT:  LBB2_1:
+; X86-Darwin-NEXT:    pushl $0
+; X86-Darwin-NEXT:    pushl $40012 ## imm = 0x9C4C
+; X86-Darwin-NEXT:    calll ___morestack
+; X86-Darwin-NEXT:    retl
+; X86-Darwin-NEXT:    jmp LBB2_2
+;
+; X64-Darwin-LABEL: test_large:
+; X64-Darwin:       ## %bb.0:
+; X64-Darwin-NEXT:    leaq -{{[0-9]+}}(%rsp), %r11
+; X64-Darwin-NEXT:    cmpq %gs:816, %r11
+; X64-Darwin-NEXT:    jbe LBB2_1
+; X64-Darwin-NEXT:  LBB2_2:
+; X64-Darwin-NEXT:    subq $40008, %rsp ## imm = 0x9C48
+; X64-Darwin-NEXT:    .cfi_def_cfa_offset 40016
+; X64-Darwin-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; X64-Darwin-NEXT:    movl $3, %esi
+; X64-Darwin-NEXT:    callq _dummy_use
+; X64-Darwin-NEXT:    addq $40008, %rsp ## imm = 0x9C48
+; X64-Darwin-NEXT:    retq
+; X64-Darwin-NEXT:  LBB2_1:
+; X64-Darwin-NEXT:    movabsq $40008, %r10 ## imm = 0x9C48
+; X64-Darwin-NEXT:    movabsq $0, %r11
+; X64-Darwin-NEXT:    callq ___morestack
+; X64-Darwin-NEXT:    retq
+; X64-Darwin-NEXT:    jmp LBB2_2
+;
+; X86-MinGW-LABEL: test_large:
+; X86-MinGW:       # %bb.0:
+; X86-MinGW-NEXT:    leal -{{[0-9]+}}(%esp), %ecx
+; X86-MinGW-NEXT:    cmpl %fs:20, %ecx
+; X86-MinGW-NEXT:    jbe LBB2_1
+; X86-MinGW-NEXT:  LBB2_2:
+; X86-MinGW-NEXT:    movl $40000, %eax # imm = 0x9C40
+; X86-MinGW-NEXT:    calll __alloca
+; X86-MinGW-NEXT:    .cfi_def_cfa_offset 40004
+; X86-MinGW-NEXT:    movl %esp, %eax
+; X86-MinGW-NEXT:    pushl $3
+; X86-MinGW-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-MinGW-NEXT:    pushl %eax
+; X86-MinGW-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-MinGW-NEXT:    calll _dummy_use
+; X86-MinGW-NEXT:    addl $40008, %esp # imm = 0x9C48
+; X86-MinGW-NEXT:    .cfi_adjust_cfa_offset -40008
+; X86-MinGW-NEXT:    retl
+; X86-MinGW-NEXT:  LBB2_1:
+; X86-MinGW-NEXT:    pushl $0
+; X86-MinGW-NEXT:    pushl $40000 # imm = 0x9C40
+; X86-MinGW-NEXT:    calll ___morestack
+; X86-MinGW-NEXT:    retl
+; X86-MinGW-NEXT:    jmp LBB2_2
+;
+; X64-FreeBSD-LABEL: test_large:
+; X64-FreeBSD:       # %bb.0:
+; X64-FreeBSD-NEXT:    leaq -{{[0-9]+}}(%rsp), %r11
+; X64-FreeBSD-NEXT:    cmpq %fs:24, %r11
+; X64-FreeBSD-NEXT:    jbe .LBB2_1
+; X64-FreeBSD-NEXT:  .LBB2_2:
+; X64-FreeBSD-NEXT:    subq $40008, %rsp # imm = 0x9C48
+; X64-FreeBSD-NEXT:    .cfi_def_cfa_offset 40016
+; X64-FreeBSD-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; X64-FreeBSD-NEXT:    movl $3, %esi
+; X64-FreeBSD-NEXT:    callq dummy_use@PLT
+; X64-FreeBSD-NEXT:    addq $40008, %rsp # imm = 0x9C48
+; X64-FreeBSD-NEXT:    .cfi_def_cfa_offset 8
+; X64-FreeBSD-NEXT:    retq
+; X64-FreeBSD-NEXT:  .LBB2_1:
+; X64-FreeBSD-NEXT:    movabsq $40008, %r10 # imm = 0x9C48
+; X64-FreeBSD-NEXT:    movabsq $0, %r11
+; X64-FreeBSD-NEXT:    callq __morestack
+; X64-FreeBSD-NEXT:    retq
+; X64-FreeBSD-NEXT:    jmp .LBB2_2
+;
+; X86-DFlyBSD-LABEL: test_large:
+; X86-DFlyBSD:       # %bb.0:
+; X86-DFlyBSD-NEXT:    leal -{{[0-9]+}}(%esp), %ecx
+; X86-DFlyBSD-NEXT:    cmpl %fs:16, %ecx
+; X86-DFlyBSD-NEXT:    jbe .LBB2_1
+; X86-DFlyBSD-NEXT:  .LBB2_2:
+; X86-DFlyBSD-NEXT:    subl $40000, %esp # imm = 0x9C40
+; X86-DFlyBSD-NEXT:    .cfi_def_cfa_offset 40004
+; X86-DFlyBSD-NEXT:    movl %esp, %eax
+; X86-DFlyBSD-NEXT:    pushl $3
+; X86-DFlyBSD-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-DFlyBSD-NEXT:    pushl %eax
+; X86-DFlyBSD-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-DFlyBSD-NEXT:    calll dummy_use@PLT
+; X86-DFlyBSD-NEXT:    addl $40008, %esp # imm = 0x9C48
+; X86-DFlyBSD-NEXT:    .cfi_adjust_cfa_offset -40008
+; X86-DFlyBSD-NEXT:    retl
+; X86-DFlyBSD-NEXT:  .LBB2_1:
+; X86-DFlyBSD-NEXT:    pushl $0
+; X86-DFlyBSD-NEXT:    pushl $40000 # imm = 0x9C40
+; X86-DFlyBSD-NEXT:    calll __morestack
+; X86-DFlyBSD-NEXT:    retl
+; X86-DFlyBSD-NEXT:    jmp .LBB2_2
+;
+; X64-DFlyBSD-LABEL: test_large:
+; X64-DFlyBSD:       # %bb.0:
+; X64-DFlyBSD-NEXT:    leaq -{{[0-9]+}}(%rsp), %r11
+; X64-DFlyBSD-NEXT:    cmpq %fs:32, %r11
+; X64-DFlyBSD-NEXT:    jbe .LBB2_1
+; X64-DFlyBSD-NEXT:  .LBB2_2:
+; X64-DFlyBSD-NEXT:    subq $40008, %rsp # imm = 0x9C48
+; X64-DFlyBSD-NEXT:    .cfi_def_cfa_offset 40016
+; X64-DFlyBSD-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; X64-DFlyBSD-NEXT:    movl $3, %esi
+; X64-DFlyBSD-NEXT:    callq dummy_use@PLT
+; X64-DFlyBSD-NEXT:    addq $40008, %rsp # imm = 0x9C48
+; X64-DFlyBSD-NEXT:    .cfi_def_cfa_offset 8
+; X64-DFlyBSD-NEXT:    retq
+; X64-DFlyBSD-NEXT:  .LBB2_1:
+; X64-DFlyBSD-NEXT:    movabsq $40008, %r10 # imm = 0x9C48
+; X64-DFlyBSD-NEXT:    movabsq $0, %r11
+; X64-DFlyBSD-NEXT:    callq __morestack
+; X64-DFlyBSD-NEXT:    retq
+; X64-DFlyBSD-NEXT:    jmp .LBB2_2
+;
+; X64-MinGW-LABEL: test_large:
+; X64-MinGW:       # %bb.0:
+; X64-MinGW-NEXT:    leaq -{{[0-9]+}}(%rsp), %r11
+; X64-MinGW-NEXT:    cmpq %gs:40, %r11
+; X64-MinGW-NEXT:    jbe .LBB2_1
+; X64-MinGW-NEXT:  .LBB2_2:
+; X64-MinGW-NEXT:    movl $40040, %eax # imm = 0x9C68
+; X64-MinGW-NEXT:    callq ___chkstk_ms
+; X64-MinGW-NEXT:    subq %rax, %rsp
+; X64-MinGW-NEXT:    .seh_stackalloc 40040
+; X64-MinGW-NEXT:    .seh_endprologue
+; X64-MinGW-NEXT:    leaq {{[0-9]+}}(%rsp), %rcx
+; X64-MinGW-NEXT:    movl $3, %edx
+; X64-MinGW-NEXT:    callq dummy_use
+; X64-MinGW-NEXT:    nop
+; X64-MinGW-NEXT:    addq $40040, %rsp # imm = 0x9C68
+; X64-MinGW-NEXT:    retq
+; X64-MinGW-NEXT:  .LBB2_1:
+; X64-MinGW-NEXT:    movabsq $40040, %r10 # imm = 0x9C68
+; X64-MinGW-NEXT:    movabsq $32, %r11
+; X64-MinGW-NEXT:    callq __morestack
+; X64-MinGW-NEXT:    retq
+; X64-MinGW-NEXT:    jmp .LBB2_2
+; X64-MinGW-NEXT:    .seh_endproc
+        %mem = alloca i32, i32 10000
+        call void @dummy_use (i32* %mem, i32 3)
+        ret void
 }
 
 define fastcc void @test_fastcc() #0 {
-        %mem = alloca i32, i32 10
-        call void @dummy_use (i32* %mem, i32 10)
-        ret void
-
-; X86-Linux-LABEL:       test_fastcc:
-
-; X86-Linux:       cmpl %gs:48, %esp
-; X86-Linux-NEXT:  jbe  .LBB3_1
-
-; X86-Linux:       pushl $0
-; X86-Linux-NEXT:  pushl $44
-; X86-Linux-NEXT:  calll __morestack
-; X86-Linux-NEXT:  ret
-
-; X64-Linux-LABEL:       test_fastcc:
-
-; X64-Linux:       cmpq %fs:112, %rsp
-; X64-Linux-NEXT:  jbe  .LBB3_1
-
-; X64-Linux:       movabsq $40, %r10
-; X64-Linux-NEXT:  movabsq $0, %r11
-; X64-Linux-NEXT:  callq __morestack
-; X64-Linux-NEXT:  ret
-
-; X32ABI-LABEL:       test_fastcc:
-
-; X32ABI:       cmpl %fs:64, %esp
-; X32ABI-NEXT:  jbe .LBB3_1
-
-; X32ABI:       movl $40, %r10d
-; X32ABI-NEXT:  movl $0, %r11d
-; X32ABI-NEXT:  callq __morestack
-; X32ABI-NEXT:  ret
-
-; X86-Darwin-LABEL:      test_fastcc:
-
-; X86-Darwin:      movl $432, %eax
-; X86-Darwin-NEXT: cmpl %gs:(%eax), %esp
-; X86-Darwin-NEXT: jbe  LBB3_1
-
-; X86-Darwin:      pushl $0
-; X86-Darwin-NEXT: pushl $60
-; X86-Darwin-NEXT: calll ___morestack
-; X86-Darwin-NEXT: ret
-
-; X64-Darwin-LABEL:      test_fastcc:
-
-; X64-Darwin:      cmpq %gs:816, %rsp
-; X64-Darwin-NEXT: jbe  LBB3_1
-
-; X64-Darwin:      movabsq $40, %r10
-; X64-Darwin-NEXT: movabsq $0, %r11
-; X64-Darwin-NEXT: callq ___morestack
-; X64-Darwin-NEXT: ret
-
-; X86-MinGW-LABEL:       test_fastcc:
-
-; X86-MinGW:       cmpl %fs:20, %esp
-; X86-MinGW-NEXT:  jbe      LBB3_1
-
-; X86-MinGW:       pushl $0
-; X86-MinGW-NEXT:  pushl $40
-; X86-MinGW-NEXT:  calll ___morestack
-; X86-MinGW-NEXT:  ret
-
-; X64-MinGW-LABEL:       test_fastcc:
-
-; X64-MinGW:       cmpq %gs:40, %rsp
-; X64-MinGW-NEXT:  jbe      .LBB3_1
-
-; X64-MinGW:       movabsq $72, %r10
-; X64-MinGW-NEXT:  movabsq $32, %r11
-; X64-MinGW-NEXT:  callq __morestack
-; X64-MinGW-NEXT:  retq
-
-; X64-FreeBSD-LABEL:       test_fastcc:
-
-; X64-FreeBSD:       cmpq %fs:24, %rsp
-; X64-FreeBSD-NEXT:  jbe    .LBB3_1
-
-; X64-FreeBSD:       movabsq $40, %r10
-; X64-FreeBSD-NEXT:  movabsq $0, %r11
-; X64-FreeBSD-NEXT:  callq __morestack
-; X64-FreeBSD-NEXT:  ret
-
-; X86-DFlyBSD-LABEL:       test_fastcc:
-
-; X86-DFlyBSD:       cmpl %fs:16, %esp
-; X86-DFlyBSD-NEXT:  jbe     .LBB3_1
-
-; X86-DFlyBSD:       pushl $0
-; X86-DFlyBSD-NEXT:  pushl $40
-; X86-DFlyBSD-NEXT:  calll __morestack
-; X86-DFlyBSD-NEXT:  ret
-
-; X64-DFlyBSD-LABEL:       test_fastcc:
-
-; X64-DFlyBSD:       cmpq %fs:32, %rsp
-; X64-DFlyBSD-NEXT:  jbe      .LBB3_1
-
-; X64-DFlyBSD:       movabsq $40, %r10
-; X64-DFlyBSD-NEXT:  movabsq $0, %r11
-; X64-DFlyBSD-NEXT:  callq __morestack
-; X64-DFlyBSD-NEXT:  ret
-
+; X86-Linux-LABEL: test_fastcc:
+; X86-Linux:       # %bb.0:
+; X86-Linux-NEXT:    cmpl %gs:48, %esp
+; X86-Linux-NEXT:    jbe .LBB3_1
+; X86-Linux-NEXT:  .LBB3_2:
+; X86-Linux-NEXT:    subl $52, %esp
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset 52
+; X86-Linux-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X86-Linux-NEXT:    pushl $10
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-Linux-NEXT:    pushl %eax
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-Linux-NEXT:    calll dummy_use@PLT
+; X86-Linux-NEXT:    addl $60, %esp
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset -60
+; X86-Linux-NEXT:    retl
+; X86-Linux-NEXT:  .LBB3_1:
+; X86-Linux-NEXT:    pushl $0
+; X86-Linux-NEXT:    pushl $44
+; X86-Linux-NEXT:    calll __morestack
+; X86-Linux-NEXT:    retl
+; X86-Linux-NEXT:    jmp .LBB3_2
+;
+; X64-Linux-LABEL: test_fastcc:
+; X64-Linux:       # %bb.0:
+; X64-Linux-NEXT:    cmpq %fs:112, %rsp
+; X64-Linux-NEXT:    jbe .LBB3_1
+; X64-Linux-NEXT:  .LBB3_2:
+; X64-Linux-NEXT:    subq $40, %rsp
+; X64-Linux-NEXT:    .cfi_def_cfa_offset 48
+; X64-Linux-NEXT:    movq %rsp, %rdi
+; X64-Linux-NEXT:    movl $10, %esi
+; X64-Linux-NEXT:    callq dummy_use@PLT
+; X64-Linux-NEXT:    addq $40, %rsp
+; X64-Linux-NEXT:    .cfi_def_cfa_offset 8
+; X64-Linux-NEXT:    retq
+; X64-Linux-NEXT:  .LBB3_1:
+; X64-Linux-NEXT:    movabsq $40, %r10
+; X64-Linux-NEXT:    movabsq $0, %r11
+; X64-Linux-NEXT:    callq __morestack
+; X64-Linux-NEXT:    retq
+; X64-Linux-NEXT:    jmp .LBB3_2
+;
+; X64-Linux-Large-LABEL: test_fastcc:
+; X64-Linux-Large:       # %bb.0:
+; X64-Linux-Large-NEXT:    cmpq %fs:112, %rsp
+; X64-Linux-Large-NEXT:    jbe .LBB3_1
+; X64-Linux-Large-NEXT:  .LBB3_2:
+; X64-Linux-Large-NEXT:    subq $40, %rsp
+; X64-Linux-Large-NEXT:    .cfi_def_cfa_offset 48
+; X64-Linux-Large-NEXT:    movabsq $dummy_use, %rax
+; X64-Linux-Large-NEXT:    movq %rsp, %rdi
+; X64-Linux-Large-NEXT:    movl $10, %esi
+; X64-Linux-Large-NEXT:    callq *%rax
+; X64-Linux-Large-NEXT:    addq $40, %rsp
+; X64-Linux-Large-NEXT:    .cfi_def_cfa_offset 8
+; X64-Linux-Large-NEXT:    retq
+; X64-Linux-Large-NEXT:  .LBB3_1:
+; X64-Linux-Large-NEXT:    movabsq $40, %r10
+; X64-Linux-Large-NEXT:    movabsq $0, %r11
+; X64-Linux-Large-NEXT:    callq *__morestack_addr(%rip)
+; X64-Linux-Large-NEXT:    retq
+; X64-Linux-Large-NEXT:    jmp .LBB3_2
+;
+; X32ABI-LABEL: test_fastcc:
+; X32ABI:       # %bb.0:
+; X32ABI-NEXT:    cmpl %fs:64, %esp
+; X32ABI-NEXT:    jbe .LBB3_1
+; X32ABI-NEXT:  .LBB3_2:
+; X32ABI-NEXT:    subl $40, %esp
+; X32ABI-NEXT:    .cfi_def_cfa_offset 48
+; X32ABI-NEXT:    movl %esp, %edi
+; X32ABI-NEXT:    movl $10, %esi
+; X32ABI-NEXT:    callq dummy_use@PLT
+; X32ABI-NEXT:    addl $40, %esp
+; X32ABI-NEXT:    .cfi_def_cfa_offset 8
+; X32ABI-NEXT:    retq
+; X32ABI-NEXT:  .LBB3_1:
+; X32ABI-NEXT:    movl $40, %r10d
+; X32ABI-NEXT:    movl $0, %r11d
+; X32ABI-NEXT:    callq __morestack
+; X32ABI-NEXT:    retq
+; X32ABI-NEXT:    jmp .LBB3_2
+;
+; X86-Darwin-LABEL: test_fastcc:
+; X86-Darwin:       ## %bb.0:
+; X86-Darwin-NEXT:    movl $432, %eax ## imm = 0x1B0
+; X86-Darwin-NEXT:    cmpl %gs:(%eax), %esp
+; X86-Darwin-NEXT:    jbe LBB3_1
+; X86-Darwin-NEXT:  LBB3_2:
+; X86-Darwin-NEXT:    subl $60, %esp
+; X86-Darwin-NEXT:    .cfi_def_cfa_offset 64
+; X86-Darwin-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X86-Darwin-NEXT:    movl %eax, (%esp)
+; X86-Darwin-NEXT:    movl $10, {{[0-9]+}}(%esp)
+; X86-Darwin-NEXT:    calll _dummy_use
+; X86-Darwin-NEXT:    addl $60, %esp
+; X86-Darwin-NEXT:    retl
+; X86-Darwin-NEXT:  LBB3_1:
+; X86-Darwin-NEXT:    pushl $0
+; X86-Darwin-NEXT:    pushl $60
+; X86-Darwin-NEXT:    calll ___morestack
+; X86-Darwin-NEXT:    retl
+; X86-Darwin-NEXT:    jmp LBB3_2
+;
+; X64-Darwin-LABEL: test_fastcc:
+; X64-Darwin:       ## %bb.0:
+; X64-Darwin-NEXT:    cmpq %gs:816, %rsp
+; X64-Darwin-NEXT:    jbe LBB3_1
+; X64-Darwin-NEXT:  LBB3_2:
+; X64-Darwin-NEXT:    subq $40, %rsp
+; X64-Darwin-NEXT:    .cfi_def_cfa_offset 48
+; X64-Darwin-NEXT:    movq %rsp, %rdi
+; X64-Darwin-NEXT:    movl $10, %esi
+; X64-Darwin-NEXT:    callq _dummy_use
+; X64-Darwin-NEXT:    addq $40, %rsp
+; X64-Darwin-NEXT:    retq
+; X64-Darwin-NEXT:  LBB3_1:
+; X64-Darwin-NEXT:    movabsq $40, %r10
+; X64-Darwin-NEXT:    movabsq $0, %r11
+; X64-Darwin-NEXT:    callq ___morestack
+; X64-Darwin-NEXT:    retq
+; X64-Darwin-NEXT:    jmp LBB3_2
+;
+; X86-MinGW-LABEL: test_fastcc:
+; X86-MinGW:       # %bb.0:
+; X86-MinGW-NEXT:    cmpl %fs:20, %esp
+; X86-MinGW-NEXT:    jbe LBB3_1
+; X86-MinGW-NEXT:  LBB3_2:
+; X86-MinGW-NEXT:    subl $40, %esp
+; X86-MinGW-NEXT:    .cfi_def_cfa_offset 44
+; X86-MinGW-NEXT:    movl %esp, %eax
+; X86-MinGW-NEXT:    pushl $10
+; X86-MinGW-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-MinGW-NEXT:    pushl %eax
+; X86-MinGW-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-MinGW-NEXT:    calll _dummy_use
+; X86-MinGW-NEXT:    addl $48, %esp
+; X86-MinGW-NEXT:    .cfi_adjust_cfa_offset -48
+; X86-MinGW-NEXT:    retl
+; X86-MinGW-NEXT:  LBB3_1:
+; X86-MinGW-NEXT:    pushl $0
+; X86-MinGW-NEXT:    pushl $40
+; X86-MinGW-NEXT:    calll ___morestack
+; X86-MinGW-NEXT:    retl
+; X86-MinGW-NEXT:    jmp LBB3_2
+;
+; X64-FreeBSD-LABEL: test_fastcc:
+; X64-FreeBSD:       # %bb.0:
+; X64-FreeBSD-NEXT:    cmpq %fs:24, %rsp
+; X64-FreeBSD-NEXT:    jbe .LBB3_1
+; X64-FreeBSD-NEXT:  .LBB3_2:
+; X64-FreeBSD-NEXT:    subq $40, %rsp
+; X64-FreeBSD-NEXT:    .cfi_def_cfa_offset 48
+; X64-FreeBSD-NEXT:    movq %rsp, %rdi
+; X64-FreeBSD-NEXT:    movl $10, %esi
+; X64-FreeBSD-NEXT:    callq dummy_use@PLT
+; X64-FreeBSD-NEXT:    addq $40, %rsp
+; X64-FreeBSD-NEXT:    .cfi_def_cfa_offset 8
+; X64-FreeBSD-NEXT:    retq
+; X64-FreeBSD-NEXT:  .LBB3_1:
+; X64-FreeBSD-NEXT:    movabsq $40, %r10
+; X64-FreeBSD-NEXT:    movabsq $0, %r11
+; X64-FreeBSD-NEXT:    callq __morestack
+; X64-FreeBSD-NEXT:    retq
+; X64-FreeBSD-NEXT:    jmp .LBB3_2
+;
+; X86-DFlyBSD-LABEL: test_fastcc:
+; X86-DFlyBSD:       # %bb.0:
+; X86-DFlyBSD-NEXT:    cmpl %fs:16, %esp
+; X86-DFlyBSD-NEXT:    jbe .LBB3_1
+; X86-DFlyBSD-NEXT:  .LBB3_2:
+; X86-DFlyBSD-NEXT:    subl $40, %esp
+; X86-DFlyBSD-NEXT:    .cfi_def_cfa_offset 44
+; X86-DFlyBSD-NEXT:    movl %esp, %eax
+; X86-DFlyBSD-NEXT:    pushl $10
+; X86-DFlyBSD-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-DFlyBSD-NEXT:    pushl %eax
+; X86-DFlyBSD-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-DFlyBSD-NEXT:    calll dummy_use@PLT
+; X86-DFlyBSD-NEXT:    addl $48, %esp
+; X86-DFlyBSD-NEXT:    .cfi_adjust_cfa_offset -48
+; X86-DFlyBSD-NEXT:    retl
+; X86-DFlyBSD-NEXT:  .LBB3_1:
+; X86-DFlyBSD-NEXT:    pushl $0
+; X86-DFlyBSD-NEXT:    pushl $40
+; X86-DFlyBSD-NEXT:    calll __morestack
+; X86-DFlyBSD-NEXT:    retl
+; X86-DFlyBSD-NEXT:    jmp .LBB3_2
+;
+; X64-DFlyBSD-LABEL: test_fastcc:
+; X64-DFlyBSD:       # %bb.0:
+; X64-DFlyBSD-NEXT:    cmpq %fs:32, %rsp
+; X64-DFlyBSD-NEXT:    jbe .LBB3_1
+; X64-DFlyBSD-NEXT:  .LBB3_2:
+; X64-DFlyBSD-NEXT:    subq $40, %rsp
+; X64-DFlyBSD-NEXT:    .cfi_def_cfa_offset 48
+; X64-DFlyBSD-NEXT:    movq %rsp, %rdi
+; X64-DFlyBSD-NEXT:    movl $10, %esi
+; X64-DFlyBSD-NEXT:    callq dummy_use@PLT
+; X64-DFlyBSD-NEXT:    addq $40, %rsp
+; X64-DFlyBSD-NEXT:    .cfi_def_cfa_offset 8
+; X64-DFlyBSD-NEXT:    retq
+; X64-DFlyBSD-NEXT:  .LBB3_1:
+; X64-DFlyBSD-NEXT:    movabsq $40, %r10
+; X64-DFlyBSD-NEXT:    movabsq $0, %r11
+; X64-DFlyBSD-NEXT:    callq __morestack
+; X64-DFlyBSD-NEXT:    retq
+; X64-DFlyBSD-NEXT:    jmp .LBB3_2
+;
+; X64-MinGW-LABEL: test_fastcc:
+; X64-MinGW:       # %bb.0:
+; X64-MinGW-NEXT:    cmpq %gs:40, %rsp
+; X64-MinGW-NEXT:    jbe .LBB3_1
+; X64-MinGW-NEXT:  .LBB3_2:
+; X64-MinGW-NEXT:    subq $72, %rsp
+; X64-MinGW-NEXT:    .seh_stackalloc 72
+; X64-MinGW-NEXT:    .seh_endprologue
+; X64-MinGW-NEXT:    leaq {{[0-9]+}}(%rsp), %rcx
+; X64-MinGW-NEXT:    movl $10, %edx
+; X64-MinGW-NEXT:    callq dummy_use
+; X64-MinGW-NEXT:    nop
+; X64-MinGW-NEXT:    addq $72, %rsp
+; X64-MinGW-NEXT:    retq
+; X64-MinGW-NEXT:  .LBB3_1:
+; X64-MinGW-NEXT:    movabsq $72, %r10
+; X64-MinGW-NEXT:    movabsq $32, %r11
+; X64-MinGW-NEXT:    callq __morestack
+; X64-MinGW-NEXT:    retq
+; X64-MinGW-NEXT:    jmp .LBB3_2
+; X64-MinGW-NEXT:    .seh_endproc
+        %mem = alloca i32, i32 10
+        call void @dummy_use (i32* %mem, i32 10)
+        ret void
 }
 
 define fastcc void @test_fastcc_large() #0 {
-        %mem = alloca i32, i32 10000
-        call void @dummy_use (i32* %mem, i32 3)
-        ret void
-
-; X86-Linux-LABEL:       test_fastcc_large:
-
-; X86-Linux:       leal -40012(%esp), %eax
-; X86-Linux-NEXT:  cmpl %gs:48, %eax
-; X86-Linux-NEXT:  jbe  .LBB4_1
-
-; X86-Linux:       pushl $0
-; X86-Linux-NEXT:  pushl $40012
-; X86-Linux-NEXT:  calll __morestack
-; X86-Linux-NEXT:  ret
-
-; X64-Linux-LABEL:       test_fastcc_large:
-
-; X64-Linux:       leaq -40008(%rsp), %r11
-; X64-Linux-NEXT:  cmpq %fs:112, %r11
-; X64-Linux-NEXT:  jbe  .LBB4_1
-
-; X64-Linux:       movabsq $40008, %r10
-; X64-Linux-NEXT:  movabsq $0, %r11
-; X64-Linux-NEXT:  callq __morestack
-; X64-Linux-NEXT:  ret
-
-; X32ABI-LABEL:       test_fastcc_large:
-
-; X32ABI:       leal -40008(%rsp), %r11d
-; X32ABI-NEXT:  cmpl %fs:64, %r11d
-; X32ABI-NEXT:  jbe .LBB4_1
-
-; X32ABI:       movl $40008, %r10d
-; X32ABI-NEXT:  movl $0, %r11d
-; X32ABI-NEXT:  callq __morestack
-; X32ABI-NEXT:  ret
-
-; X86-Darwin-LABEL:      test_fastcc_large:
-
-; X86-Darwin:      leal -40012(%esp), %eax
-; X86-Darwin-NEXT: movl $432, %ecx
-; X86-Darwin-NEXT: cmpl %gs:(%ecx), %eax
-; X86-Darwin-NEXT: jbe  LBB4_1
-
-; X86-Darwin:      pushl $0
-; X86-Darwin-NEXT: pushl $40012
-; X86-Darwin-NEXT: calll ___morestack
-; X86-Darwin-NEXT: ret
-
-; X64-Darwin-LABEL:      test_fastcc_large:
-
-; X64-Darwin:      leaq -40008(%rsp), %r11
-; X64-Darwin-NEXT: cmpq %gs:816, %r11
-; X64-Darwin-NEXT: jbe  LBB4_1
-
-; X64-Darwin:      movabsq $40008, %r10
-; X64-Darwin-NEXT: movabsq $0, %r11
-; X64-Darwin-NEXT: callq ___morestack
-; X64-Darwin-NEXT: ret
-
-; X86-MinGW-LABEL:       test_fastcc_large:
-
-; X86-MinGW:       leal -40000(%esp), %eax
-; X86-MinGW-NEXT:  cmpl %fs:20, %eax
-; X86-MinGW-NEXT:  jbe      LBB4_1
-
-; X86-MinGW:       pushl $0
-; X86-MinGW-NEXT:  pushl $40000
-; X86-MinGW-NEXT:  calll ___morestack
-; X86-MinGW-NEXT:  ret
-
-; X64-MinGW-LABEL:       test_fastcc_large:
-
-; X64-MinGW:       leaq -40040(%rsp), %r11
-; X64-MinGW-NEXT:  cmpq %gs:40, %r11
-; X64-MinGW-NEXT:  jbe      .LBB4_1
-
-; X64-MinGW:       movabsq $40040, %r10
-; X64-MinGW-NEXT:  movabsq $32, %r11
-; X64-MinGW-NEXT:  callq __morestack
-; X64-MinGW-NEXT:  retq
-
-; X64-FreeBSD-LABEL:       test_fastcc_large:
-
-; X64-FreeBSD:       leaq -40008(%rsp), %r11
-; X64-FreeBSD-NEXT:  cmpq %fs:24, %r11
-; X64-FreeBSD-NEXT:  jbe     .LBB4_1
-
-; X64-FreeBSD:       movabsq $40008, %r10
-; X64-FreeBSD-NEXT:  movabsq $0, %r11
-; X64-FreeBSD-NEXT:  callq __morestack
-; X64-FreeBSD-NEXT:  ret
-
-; X86-DFlyBSD-LABEL:       test_fastcc_large:
-
-; X86-DFlyBSD:       leal -40000(%esp), %eax
-; X86-DFlyBSD-NEXT:  cmpl %fs:16, %eax
-; X86-DFlyBSD-NEXT:  jbe      .LBB4_1
-
-; X86-DFlyBSD:       pushl $0
-; X86-DFlyBSD-NEXT:  pushl $40000
-; X86-DFlyBSD-NEXT:  calll __morestack
-; X86-DFlyBSD-NEXT:  ret
-
-; X64-DFlyBSD-LABEL:       test_fastcc_large:
-
-; X64-DFlyBSD:       leaq -40008(%rsp), %r11
-; X64-DFlyBSD-NEXT:  cmpq %fs:32, %r11
-; X64-DFlyBSD-NEXT:  jbe      .LBB4_1
-
-; X64-DFlyBSD:       movabsq $40008, %r10
-; X64-DFlyBSD-NEXT:  movabsq $0, %r11
-; X64-DFlyBSD-NEXT:  callq __morestack
-; X64-DFlyBSD-NEXT:  ret
-
-}
-
-define fastcc void @test_fastcc_large_with_ecx_arg(i32 %a) #0 {
+; X86-Linux-LABEL: test_fastcc_large:
+; X86-Linux:       # %bb.0:
+; X86-Linux-NEXT:    leal -{{[0-9]+}}(%esp), %eax
+; X86-Linux-NEXT:    cmpl %gs:48, %eax
+; X86-Linux-NEXT:    jbe .LBB4_1
+; X86-Linux-NEXT:  .LBB4_2:
+; X86-Linux-NEXT:    subl $40020, %esp # imm = 0x9C54
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset 40020
+; X86-Linux-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X86-Linux-NEXT:    pushl $3
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-Linux-NEXT:    pushl %eax
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-Linux-NEXT:    calll dummy_use@PLT
+; X86-Linux-NEXT:    addl $40028, %esp # imm = 0x9C5C
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset -40028
+; X86-Linux-NEXT:    retl
+; X86-Linux-NEXT:  .LBB4_1:
+; X86-Linux-NEXT:    pushl $0
+; X86-Linux-NEXT:    pushl $40012 # imm = 0x9C4C
+; X86-Linux-NEXT:    calll __morestack
+; X86-Linux-NEXT:    retl
+; X86-Linux-NEXT:    jmp .LBB4_2
+;
+; X64-Linux-LABEL: test_fastcc_large:
+; X64-Linux:       # %bb.0:
+; X64-Linux-NEXT:    leaq -{{[0-9]+}}(%rsp), %r11
+; X64-Linux-NEXT:    cmpq %fs:112, %r11
+; X64-Linux-NEXT:    jbe .LBB4_1
+; X64-Linux-NEXT:  .LBB4_2:
+; X64-Linux-NEXT:    subq $40008, %rsp # imm = 0x9C48
+; X64-Linux-NEXT:    .cfi_def_cfa_offset 40016
+; X64-Linux-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; X64-Linux-NEXT:    movl $3, %esi
+; X64-Linux-NEXT:    callq dummy_use@PLT
+; X64-Linux-NEXT:    addq $40008, %rsp # imm = 0x9C48
+; X64-Linux-NEXT:    .cfi_def_cfa_offset 8
+; X64-Linux-NEXT:    retq
+; X64-Linux-NEXT:  .LBB4_1:
+; X64-Linux-NEXT:    movabsq $40008, %r10 # imm = 0x9C48
+; X64-Linux-NEXT:    movabsq $0, %r11
+; X64-Linux-NEXT:    callq __morestack
+; X64-Linux-NEXT:    retq
+; X64-Linux-NEXT:    jmp .LBB4_2
+;
+; X64-Linux-Large-LABEL: test_fastcc_large:
+; X64-Linux-Large:       # %bb.0:
+; X64-Linux-Large-NEXT:    leaq -{{[0-9]+}}(%rsp), %r11
+; X64-Linux-Large-NEXT:    cmpq %fs:112, %r11
+; X64-Linux-Large-NEXT:    jbe .LBB4_1
+; X64-Linux-Large-NEXT:  .LBB4_2:
+; X64-Linux-Large-NEXT:    subq $40008, %rsp # imm = 0x9C48
+; X64-Linux-Large-NEXT:    .cfi_def_cfa_offset 40016
+; X64-Linux-Large-NEXT:    movabsq $dummy_use, %rax
+; X64-Linux-Large-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; X64-Linux-Large-NEXT:    movl $3, %esi
+; X64-Linux-Large-NEXT:    callq *%rax
+; X64-Linux-Large-NEXT:    addq $40008, %rsp # imm = 0x9C48
+; X64-Linux-Large-NEXT:    .cfi_def_cfa_offset 8
+; X64-Linux-Large-NEXT:    retq
+; X64-Linux-Large-NEXT:  .LBB4_1:
+; X64-Linux-Large-NEXT:    movabsq $40008, %r10 # imm = 0x9C48
+; X64-Linux-Large-NEXT:    movabsq $0, %r11
+; X64-Linux-Large-NEXT:    callq *__morestack_addr(%rip)
+; X64-Linux-Large-NEXT:    retq
+; X64-Linux-Large-NEXT:    jmp .LBB4_2
+;
+; X32ABI-LABEL: test_fastcc_large:
+; X32ABI:       # %bb.0:
+; X32ABI-NEXT:    leal -{{[0-9]+}}(%rsp), %r11d
+; X32ABI-NEXT:    cmpl %fs:64, %r11d
+; X32ABI-NEXT:    jbe .LBB4_1
+; X32ABI-NEXT:  .LBB4_2:
+; X32ABI-NEXT:    subl $40008, %esp # imm = 0x9C48
+; X32ABI-NEXT:    .cfi_def_cfa_offset 40016
+; X32ABI-NEXT:    leal {{[0-9]+}}(%rsp), %edi
+; X32ABI-NEXT:    movl $3, %esi
+; X32ABI-NEXT:    callq dummy_use@PLT
+; X32ABI-NEXT:    addl $40008, %esp # imm = 0x9C48
+; X32ABI-NEXT:    .cfi_def_cfa_offset 8
+; X32ABI-NEXT:    retq
+; X32ABI-NEXT:  .LBB4_1:
+; X32ABI-NEXT:    movl $40008, %r10d # imm = 0x9C48
+; X32ABI-NEXT:    movl $0, %r11d
+; X32ABI-NEXT:    callq __morestack
+; X32ABI-NEXT:    retq
+; X32ABI-NEXT:    jmp .LBB4_2
+;
+; X86-Darwin-LABEL: test_fastcc_large:
+; X86-Darwin:       ## %bb.0:
+; X86-Darwin-NEXT:    leal -{{[0-9]+}}(%esp), %eax
+; X86-Darwin-NEXT:    movl $432, %ecx ## imm = 0x1B0
+; X86-Darwin-NEXT:    cmpl %gs:(%ecx), %eax
+; X86-Darwin-NEXT:    jbe LBB4_1
+; X86-Darwin-NEXT:  LBB4_2:
+; X86-Darwin-NEXT:    subl $40012, %esp ## imm = 0x9C4C
+; X86-Darwin-NEXT:    .cfi_def_cfa_offset 40016
+; X86-Darwin-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X86-Darwin-NEXT:    movl %eax, (%esp)
+; X86-Darwin-NEXT:    movl $3, {{[0-9]+}}(%esp)
+; X86-Darwin-NEXT:    calll _dummy_use
+; X86-Darwin-NEXT:    addl $40012, %esp ## imm = 0x9C4C
+; X86-Darwin-NEXT:    retl
+; X86-Darwin-NEXT:  LBB4_1:
+; X86-Darwin-NEXT:    pushl $0
+; X86-Darwin-NEXT:    pushl $40012 ## imm = 0x9C4C
+; X86-Darwin-NEXT:    calll ___morestack
+; X86-Darwin-NEXT:    retl
+; X86-Darwin-NEXT:    jmp LBB4_2
+;
+; X64-Darwin-LABEL: test_fastcc_large:
+; X64-Darwin:       ## %bb.0:
+; X64-Darwin-NEXT:    leaq -{{[0-9]+}}(%rsp), %r11
+; X64-Darwin-NEXT:    cmpq %gs:816, %r11
+; X64-Darwin-NEXT:    jbe LBB4_1
+; X64-Darwin-NEXT:  LBB4_2:
+; X64-Darwin-NEXT:    subq $40008, %rsp ## imm = 0x9C48
+; X64-Darwin-NEXT:    .cfi_def_cfa_offset 40016
+; X64-Darwin-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; X64-Darwin-NEXT:    movl $3, %esi
+; X64-Darwin-NEXT:    callq _dummy_use
+; X64-Darwin-NEXT:    addq $40008, %rsp ## imm = 0x9C48
+; X64-Darwin-NEXT:    retq
+; X64-Darwin-NEXT:  LBB4_1:
+; X64-Darwin-NEXT:    movabsq $40008, %r10 ## imm = 0x9C48
+; X64-Darwin-NEXT:    movabsq $0, %r11
+; X64-Darwin-NEXT:    callq ___morestack
+; X64-Darwin-NEXT:    retq
+; X64-Darwin-NEXT:    jmp LBB4_2
+;
+; X86-MinGW-LABEL: test_fastcc_large:
+; X86-MinGW:       # %bb.0:
+; X86-MinGW-NEXT:    leal -{{[0-9]+}}(%esp), %eax
+; X86-MinGW-NEXT:    cmpl %fs:20, %eax
+; X86-MinGW-NEXT:    jbe LBB4_1
+; X86-MinGW-NEXT:  LBB4_2:
+; X86-MinGW-NEXT:    movl $40000, %eax # imm = 0x9C40
+; X86-MinGW-NEXT:    calll __alloca
+; X86-MinGW-NEXT:    .cfi_def_cfa_offset 40004
+; X86-MinGW-NEXT:    movl %esp, %eax
+; X86-MinGW-NEXT:    pushl $3
+; X86-MinGW-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-MinGW-NEXT:    pushl %eax
+; X86-MinGW-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-MinGW-NEXT:    calll _dummy_use
+; X86-MinGW-NEXT:    addl $40008, %esp # imm = 0x9C48
+; X86-MinGW-NEXT:    .cfi_adjust_cfa_offset -40008
+; X86-MinGW-NEXT:    retl
+; X86-MinGW-NEXT:  LBB4_1:
+; X86-MinGW-NEXT:    pushl $0
+; X86-MinGW-NEXT:    pushl $40000 # imm = 0x9C40
+; X86-MinGW-NEXT:    calll ___morestack
+; X86-MinGW-NEXT:    retl
+; X86-MinGW-NEXT:    jmp LBB4_2
+;
+; X64-FreeBSD-LABEL: test_fastcc_large:
+; X64-FreeBSD:       # %bb.0:
+; X64-FreeBSD-NEXT:    leaq -{{[0-9]+}}(%rsp), %r11
+; X64-FreeBSD-NEXT:    cmpq %fs:24, %r11
+; X64-FreeBSD-NEXT:    jbe .LBB4_1
+; X64-FreeBSD-NEXT:  .LBB4_2:
+; X64-FreeBSD-NEXT:    subq $40008, %rsp # imm = 0x9C48
+; X64-FreeBSD-NEXT:    .cfi_def_cfa_offset 40016
+; X64-FreeBSD-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; X64-FreeBSD-NEXT:    movl $3, %esi
+; X64-FreeBSD-NEXT:    callq dummy_use@PLT
+; X64-FreeBSD-NEXT:    addq $40008, %rsp # imm = 0x9C48
+; X64-FreeBSD-NEXT:    .cfi_def_cfa_offset 8
+; X64-FreeBSD-NEXT:    retq
+; X64-FreeBSD-NEXT:  .LBB4_1:
+; X64-FreeBSD-NEXT:    movabsq $40008, %r10 # imm = 0x9C48
+; X64-FreeBSD-NEXT:    movabsq $0, %r11
+; X64-FreeBSD-NEXT:    callq __morestack
+; X64-FreeBSD-NEXT:    retq
+; X64-FreeBSD-NEXT:    jmp .LBB4_2
+;
+; X86-DFlyBSD-LABEL: test_fastcc_large:
+; X86-DFlyBSD:       # %bb.0:
+; X86-DFlyBSD-NEXT:    leal -{{[0-9]+}}(%esp), %eax
+; X86-DFlyBSD-NEXT:    cmpl %fs:16, %eax
+; X86-DFlyBSD-NEXT:    jbe .LBB4_1
+; X86-DFlyBSD-NEXT:  .LBB4_2:
+; X86-DFlyBSD-NEXT:    subl $40000, %esp # imm = 0x9C40
+; X86-DFlyBSD-NEXT:    .cfi_def_cfa_offset 40004
+; X86-DFlyBSD-NEXT:    movl %esp, %eax
+; X86-DFlyBSD-NEXT:    pushl $3
+; X86-DFlyBSD-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-DFlyBSD-NEXT:    pushl %eax
+; X86-DFlyBSD-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-DFlyBSD-NEXT:    calll dummy_use@PLT
+; X86-DFlyBSD-NEXT:    addl $40008, %esp # imm = 0x9C48
+; X86-DFlyBSD-NEXT:    .cfi_adjust_cfa_offset -40008
+; X86-DFlyBSD-NEXT:    retl
+; X86-DFlyBSD-NEXT:  .LBB4_1:
+; X86-DFlyBSD-NEXT:    pushl $0
+; X86-DFlyBSD-NEXT:    pushl $40000 # imm = 0x9C40
+; X86-DFlyBSD-NEXT:    calll __morestack
+; X86-DFlyBSD-NEXT:    retl
+; X86-DFlyBSD-NEXT:    jmp .LBB4_2
+;
+; X64-DFlyBSD-LABEL: test_fastcc_large:
+; X64-DFlyBSD:       # %bb.0:
+; X64-DFlyBSD-NEXT:    leaq -{{[0-9]+}}(%rsp), %r11
+; X64-DFlyBSD-NEXT:    cmpq %fs:32, %r11
+; X64-DFlyBSD-NEXT:    jbe .LBB4_1
+; X64-DFlyBSD-NEXT:  .LBB4_2:
+; X64-DFlyBSD-NEXT:    subq $40008, %rsp # imm = 0x9C48
+; X64-DFlyBSD-NEXT:    .cfi_def_cfa_offset 40016
+; X64-DFlyBSD-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; X64-DFlyBSD-NEXT:    movl $3, %esi
+; X64-DFlyBSD-NEXT:    callq dummy_use@PLT
+; X64-DFlyBSD-NEXT:    addq $40008, %rsp # imm = 0x9C48
+; X64-DFlyBSD-NEXT:    .cfi_def_cfa_offset 8
+; X64-DFlyBSD-NEXT:    retq
+; X64-DFlyBSD-NEXT:  .LBB4_1:
+; X64-DFlyBSD-NEXT:    movabsq $40008, %r10 # imm = 0x9C48
+; X64-DFlyBSD-NEXT:    movabsq $0, %r11
+; X64-DFlyBSD-NEXT:    callq __morestack
+; X64-DFlyBSD-NEXT:    retq
+; X64-DFlyBSD-NEXT:    jmp .LBB4_2
+;
+; X64-MinGW-LABEL: test_fastcc_large:
+; X64-MinGW:       # %bb.0:
+; X64-MinGW-NEXT:    leaq -{{[0-9]+}}(%rsp), %r11
+; X64-MinGW-NEXT:    cmpq %gs:40, %r11
+; X64-MinGW-NEXT:    jbe .LBB4_1
+; X64-MinGW-NEXT:  .LBB4_2:
+; X64-MinGW-NEXT:    movl $40040, %eax # imm = 0x9C68
+; X64-MinGW-NEXT:    callq ___chkstk_ms
+; X64-MinGW-NEXT:    subq %rax, %rsp
+; X64-MinGW-NEXT:    .seh_stackalloc 40040
+; X64-MinGW-NEXT:    .seh_endprologue
+; X64-MinGW-NEXT:    leaq {{[0-9]+}}(%rsp), %rcx
+; X64-MinGW-NEXT:    movl $3, %edx
+; X64-MinGW-NEXT:    callq dummy_use
+; X64-MinGW-NEXT:    nop
+; X64-MinGW-NEXT:    addq $40040, %rsp # imm = 0x9C68
+; X64-MinGW-NEXT:    retq
+; X64-MinGW-NEXT:  .LBB4_1:
+; X64-MinGW-NEXT:    movabsq $40040, %r10 # imm = 0x9C68
+; X64-MinGW-NEXT:    movabsq $32, %r11
+; X64-MinGW-NEXT:    callq __morestack
+; X64-MinGW-NEXT:    retq
+; X64-MinGW-NEXT:    jmp .LBB4_2
+; X64-MinGW-NEXT:    .seh_endproc
         %mem = alloca i32, i32 10000
-        call void @dummy_use (i32* %mem, i32 %a)
+        call void @dummy_use (i32* %mem, i32 3)
         ret void
+}
 
 ; This is testing that the Mac implementation preserves ecx
 
-; X86-Darwin-LABEL:      test_fastcc_large_with_ecx_arg:
-
-; X86-Darwin:      leal -40012(%esp), %eax
-; X86-Darwin-NEXT: pushl %ecx
-; X86-Darwin-NEXT: movl $432, %ecx
-; X86-Darwin-NEXT: cmpl %gs:(%ecx), %eax
-; X86-Darwin-NEXT: popl %ecx
-; X86-Darwin-NEXT: jbe  LBB5_1
-
-; X86-Darwin:      pushl $0
-; X86-Darwin-NEXT: pushl $40012
-; X86-Darwin-NEXT: calll ___morestack
-; X86-Darwin-NEXT: ret
-
+define fastcc void @test_fastcc_large_with_ecx_arg(i32 %a) #0 {
+; X86-Linux-LABEL: test_fastcc_large_with_ecx_arg:
+; X86-Linux:       # %bb.0:
+; X86-Linux-NEXT:    leal -{{[0-9]+}}(%esp), %eax
+; X86-Linux-NEXT:    cmpl %gs:48, %eax
+; X86-Linux-NEXT:    jbe .LBB5_1
+; X86-Linux-NEXT:  .LBB5_2:
+; X86-Linux-NEXT:    subl $40020, %esp # imm = 0x9C54
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset 40020
+; X86-Linux-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X86-Linux-NEXT:    pushl %ecx
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-Linux-NEXT:    pushl %eax
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-Linux-NEXT:    calll dummy_use@PLT
+; X86-Linux-NEXT:    addl $40028, %esp # imm = 0x9C5C
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset -40028
+; X86-Linux-NEXT:    retl
+; X86-Linux-NEXT:  .LBB5_1:
+; X86-Linux-NEXT:    pushl $0
+; X86-Linux-NEXT:    pushl $40012 # imm = 0x9C4C
+; X86-Linux-NEXT:    calll __morestack
+; X86-Linux-NEXT:    retl
+; X86-Linux-NEXT:    jmp .LBB5_2
+;
+; X64-Linux-LABEL: test_fastcc_large_with_ecx_arg:
+; X64-Linux:       # %bb.0:
+; X64-Linux-NEXT:    leaq -{{[0-9]+}}(%rsp), %r11
+; X64-Linux-NEXT:    cmpq %fs:112, %r11
+; X64-Linux-NEXT:    jbe .LBB5_1
+; X64-Linux-NEXT:  .LBB5_2:
+; X64-Linux-NEXT:    subq $40008, %rsp # imm = 0x9C48
+; X64-Linux-NEXT:    .cfi_def_cfa_offset 40016
+; X64-Linux-NEXT:    movl %edi, %esi
+; X64-Linux-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; X64-Linux-NEXT:    callq dummy_use@PLT
+; X64-Linux-NEXT:    addq $40008, %rsp # imm = 0x9C48
+; X64-Linux-NEXT:    .cfi_def_cfa_offset 8
+; X64-Linux-NEXT:    retq
+; X64-Linux-NEXT:  .LBB5_1:
+; X64-Linux-NEXT:    movabsq $40008, %r10 # imm = 0x9C48
+; X64-Linux-NEXT:    movabsq $0, %r11
+; X64-Linux-NEXT:    callq __morestack
+; X64-Linux-NEXT:    retq
+; X64-Linux-NEXT:    jmp .LBB5_2
+;
+; X64-Linux-Large-LABEL: test_fastcc_large_with_ecx_arg:
+; X64-Linux-Large:       # %bb.0:
+; X64-Linux-Large-NEXT:    leaq -{{[0-9]+}}(%rsp), %r11
+; X64-Linux-Large-NEXT:    cmpq %fs:112, %r11
+; X64-Linux-Large-NEXT:    jbe .LBB5_1
+; X64-Linux-Large-NEXT:  .LBB5_2:
+; X64-Linux-Large-NEXT:    subq $40008, %rsp # imm = 0x9C48
+; X64-Linux-Large-NEXT:    .cfi_def_cfa_offset 40016
+; X64-Linux-Large-NEXT:    movl %edi, %esi
+; X64-Linux-Large-NEXT:    movabsq $dummy_use, %rax
+; X64-Linux-Large-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; X64-Linux-Large-NEXT:    callq *%rax
+; X64-Linux-Large-NEXT:    addq $40008, %rsp # imm = 0x9C48
+; X64-Linux-Large-NEXT:    .cfi_def_cfa_offset 8
+; X64-Linux-Large-NEXT:    retq
+; X64-Linux-Large-NEXT:  .LBB5_1:
+; X64-Linux-Large-NEXT:    movabsq $40008, %r10 # imm = 0x9C48
+; X64-Linux-Large-NEXT:    movabsq $0, %r11
+; X64-Linux-Large-NEXT:    callq *__morestack_addr(%rip)
+; X64-Linux-Large-NEXT:    retq
+; X64-Linux-Large-NEXT:    jmp .LBB5_2
+;
+; X32ABI-LABEL: test_fastcc_large_with_ecx_arg:
+; X32ABI:       # %bb.0:
+; X32ABI-NEXT:    leal -{{[0-9]+}}(%rsp), %r11d
+; X32ABI-NEXT:    cmpl %fs:64, %r11d
+; X32ABI-NEXT:    jbe .LBB5_1
+; X32ABI-NEXT:  .LBB5_2:
+; X32ABI-NEXT:    subl $40008, %esp # imm = 0x9C48
+; X32ABI-NEXT:    .cfi_def_cfa_offset 40016
+; X32ABI-NEXT:    movl %edi, %esi
+; X32ABI-NEXT:    leal {{[0-9]+}}(%rsp), %edi
+; X32ABI-NEXT:    callq dummy_use@PLT
+; X32ABI-NEXT:    addl $40008, %esp # imm = 0x9C48
+; X32ABI-NEXT:    .cfi_def_cfa_offset 8
+; X32ABI-NEXT:    retq
+; X32ABI-NEXT:  .LBB5_1:
+; X32ABI-NEXT:    movl $40008, %r10d # imm = 0x9C48
+; X32ABI-NEXT:    movl $0, %r11d
+; X32ABI-NEXT:    callq __morestack
+; X32ABI-NEXT:    retq
+; X32ABI-NEXT:    jmp .LBB5_2
+;
+; X86-Darwin-LABEL: test_fastcc_large_with_ecx_arg:
+; X86-Darwin:       ## %bb.0:
+; X86-Darwin-NEXT:    leal -{{[0-9]+}}(%esp), %eax
+; X86-Darwin-NEXT:    pushl %ecx
+; X86-Darwin-NEXT:    movl $432, %ecx ## imm = 0x1B0
+; X86-Darwin-NEXT:    cmpl %gs:(%ecx), %eax
+; X86-Darwin-NEXT:    popl %ecx
+; X86-Darwin-NEXT:    jbe LBB5_1
+; X86-Darwin-NEXT:  LBB5_2:
+; X86-Darwin-NEXT:    subl $40012, %esp ## imm = 0x9C4C
+; X86-Darwin-NEXT:    .cfi_def_cfa_offset 40016
+; X86-Darwin-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
+; X86-Darwin-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X86-Darwin-NEXT:    movl %eax, (%esp)
+; X86-Darwin-NEXT:    calll _dummy_use
+; X86-Darwin-NEXT:    addl $40012, %esp ## imm = 0x9C4C
+; X86-Darwin-NEXT:    retl
+; X86-Darwin-NEXT:  LBB5_1:
+; X86-Darwin-NEXT:    pushl $0
+; X86-Darwin-NEXT:    pushl $40012 ## imm = 0x9C4C
+; X86-Darwin-NEXT:    calll ___morestack
+; X86-Darwin-NEXT:    retl
+; X86-Darwin-NEXT:    jmp LBB5_2
+;
+; X64-Darwin-LABEL: test_fastcc_large_with_ecx_arg:
+; X64-Darwin:       ## %bb.0:
+; X64-Darwin-NEXT:    leaq -{{[0-9]+}}(%rsp), %r11
+; X64-Darwin-NEXT:    cmpq %gs:816, %r11
+; X64-Darwin-NEXT:    jbe LBB5_1
+; X64-Darwin-NEXT:  LBB5_2:
+; X64-Darwin-NEXT:    subq $40008, %rsp ## imm = 0x9C48
+; X64-Darwin-NEXT:    .cfi_def_cfa_offset 40016
+; X64-Darwin-NEXT:    movl %edi, %esi
+; X64-Darwin-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; X64-Darwin-NEXT:    callq _dummy_use
+; X64-Darwin-NEXT:    addq $40008, %rsp ## imm = 0x9C48
+; X64-Darwin-NEXT:    retq
+; X64-Darwin-NEXT:  LBB5_1:
+; X64-Darwin-NEXT:    movabsq $40008, %r10 ## imm = 0x9C48
+; X64-Darwin-NEXT:    movabsq $0, %r11
+; X64-Darwin-NEXT:    callq ___morestack
+; X64-Darwin-NEXT:    retq
+; X64-Darwin-NEXT:    jmp LBB5_2
+;
+; X86-MinGW-LABEL: test_fastcc_large_with_ecx_arg:
+; X86-MinGW:       # %bb.0:
+; X86-MinGW-NEXT:    leal -{{[0-9]+}}(%esp), %eax
+; X86-MinGW-NEXT:    cmpl %fs:20, %eax
+; X86-MinGW-NEXT:    jbe LBB5_1
+; X86-MinGW-NEXT:  LBB5_2:
+; X86-MinGW-NEXT:    movl $40000, %eax # imm = 0x9C40
+; X86-MinGW-NEXT:    calll __alloca
+; X86-MinGW-NEXT:    .cfi_def_cfa_offset 40004
+; X86-MinGW-NEXT:    movl %esp, %eax
+; X86-MinGW-NEXT:    pushl %ecx
+; X86-MinGW-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-MinGW-NEXT:    pushl %eax
+; X86-MinGW-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-MinGW-NEXT:    calll _dummy_use
+; X86-MinGW-NEXT:    addl $40008, %esp # imm = 0x9C48
+; X86-MinGW-NEXT:    .cfi_adjust_cfa_offset -40008
+; X86-MinGW-NEXT:    retl
+; X86-MinGW-NEXT:  LBB5_1:
+; X86-MinGW-NEXT:    pushl $0
+; X86-MinGW-NEXT:    pushl $40000 # imm = 0x9C40
+; X86-MinGW-NEXT:    calll ___morestack
+; X86-MinGW-NEXT:    retl
+; X86-MinGW-NEXT:    jmp LBB5_2
+;
+; X64-FreeBSD-LABEL: test_fastcc_large_with_ecx_arg:
+; X64-FreeBSD:       # %bb.0:
+; X64-FreeBSD-NEXT:    leaq -{{[0-9]+}}(%rsp), %r11
+; X64-FreeBSD-NEXT:    cmpq %fs:24, %r11
+; X64-FreeBSD-NEXT:    jbe .LBB5_1
+; X64-FreeBSD-NEXT:  .LBB5_2:
+; X64-FreeBSD-NEXT:    subq $40008, %rsp # imm = 0x9C48
+; X64-FreeBSD-NEXT:    .cfi_def_cfa_offset 40016
+; X64-FreeBSD-NEXT:    movl %edi, %esi
+; X64-FreeBSD-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; X64-FreeBSD-NEXT:    callq dummy_use@PLT
+; X64-FreeBSD-NEXT:    addq $40008, %rsp # imm = 0x9C48
+; X64-FreeBSD-NEXT:    .cfi_def_cfa_offset 8
+; X64-FreeBSD-NEXT:    retq
+; X64-FreeBSD-NEXT:  .LBB5_1:
+; X64-FreeBSD-NEXT:    movabsq $40008, %r10 # imm = 0x9C48
+; X64-FreeBSD-NEXT:    movabsq $0, %r11
+; X64-FreeBSD-NEXT:    callq __morestack
+; X64-FreeBSD-NEXT:    retq
+; X64-FreeBSD-NEXT:    jmp .LBB5_2
+;
+; X86-DFlyBSD-LABEL: test_fastcc_large_with_ecx_arg:
+; X86-DFlyBSD:       # %bb.0:
+; X86-DFlyBSD-NEXT:    leal -{{[0-9]+}}(%esp), %eax
+; X86-DFlyBSD-NEXT:    cmpl %fs:16, %eax
+; X86-DFlyBSD-NEXT:    jbe .LBB5_1
+; X86-DFlyBSD-NEXT:  .LBB5_2:
+; X86-DFlyBSD-NEXT:    subl $40000, %esp # imm = 0x9C40
+; X86-DFlyBSD-NEXT:    .cfi_def_cfa_offset 40004
+; X86-DFlyBSD-NEXT:    movl %esp, %eax
+; X86-DFlyBSD-NEXT:    pushl %ecx
+; X86-DFlyBSD-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-DFlyBSD-NEXT:    pushl %eax
+; X86-DFlyBSD-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-DFlyBSD-NEXT:    calll dummy_use@PLT
+; X86-DFlyBSD-NEXT:    addl $40008, %esp # imm = 0x9C48
+; X86-DFlyBSD-NEXT:    .cfi_adjust_cfa_offset -40008
+; X86-DFlyBSD-NEXT:    retl
+; X86-DFlyBSD-NEXT:  .LBB5_1:
+; X86-DFlyBSD-NEXT:    pushl $0
+; X86-DFlyBSD-NEXT:    pushl $40000 # imm = 0x9C40
+; X86-DFlyBSD-NEXT:    calll __morestack
+; X86-DFlyBSD-NEXT:    retl
+; X86-DFlyBSD-NEXT:    jmp .LBB5_2
+;
+; X64-DFlyBSD-LABEL: test_fastcc_large_with_ecx_arg:
+; X64-DFlyBSD:       # %bb.0:
+; X64-DFlyBSD-NEXT:    leaq -{{[0-9]+}}(%rsp), %r11
+; X64-DFlyBSD-NEXT:    cmpq %fs:32, %r11
+; X64-DFlyBSD-NEXT:    jbe .LBB5_1
+; X64-DFlyBSD-NEXT:  .LBB5_2:
+; X64-DFlyBSD-NEXT:    subq $40008, %rsp # imm = 0x9C48
+; X64-DFlyBSD-NEXT:    .cfi_def_cfa_offset 40016
+; X64-DFlyBSD-NEXT:    movl %edi, %esi
+; X64-DFlyBSD-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; X64-DFlyBSD-NEXT:    callq dummy_use@PLT
+; X64-DFlyBSD-NEXT:    addq $40008, %rsp # imm = 0x9C48
+; X64-DFlyBSD-NEXT:    .cfi_def_cfa_offset 8
+; X64-DFlyBSD-NEXT:    retq
+; X64-DFlyBSD-NEXT:  .LBB5_1:
+; X64-DFlyBSD-NEXT:    movabsq $40008, %r10 # imm = 0x9C48
+; X64-DFlyBSD-NEXT:    movabsq $0, %r11
+; X64-DFlyBSD-NEXT:    callq __morestack
+; X64-DFlyBSD-NEXT:    retq
+; X64-DFlyBSD-NEXT:    jmp .LBB5_2
+;
+; X64-MinGW-LABEL: test_fastcc_large_with_ecx_arg:
+; X64-MinGW:       # %bb.0:
+; X64-MinGW-NEXT:    leaq -{{[0-9]+}}(%rsp), %r11
+; X64-MinGW-NEXT:    cmpq %gs:40, %r11
+; X64-MinGW-NEXT:    jbe .LBB5_1
+; X64-MinGW-NEXT:  .LBB5_2:
+; X64-MinGW-NEXT:    movl $40040, %eax # imm = 0x9C68
+; X64-MinGW-NEXT:    callq ___chkstk_ms
+; X64-MinGW-NEXT:    subq %rax, %rsp
+; X64-MinGW-NEXT:    .seh_stackalloc 40040
+; X64-MinGW-NEXT:    .seh_endprologue
+; X64-MinGW-NEXT:    movl %ecx, %edx
+; X64-MinGW-NEXT:    leaq {{[0-9]+}}(%rsp), %rcx
+; X64-MinGW-NEXT:    callq dummy_use
+; X64-MinGW-NEXT:    nop
+; X64-MinGW-NEXT:    addq $40040, %rsp # imm = 0x9C68
+; X64-MinGW-NEXT:    retq
+; X64-MinGW-NEXT:  .LBB5_1:
+; X64-MinGW-NEXT:    movabsq $40040, %r10 # imm = 0x9C68
+; X64-MinGW-NEXT:    movabsq $32, %r11
+; X64-MinGW-NEXT:    callq __morestack
+; X64-MinGW-NEXT:    retq
+; X64-MinGW-NEXT:    jmp .LBB5_2
+; X64-MinGW-NEXT:    .seh_endproc
+        %mem = alloca i32, i32 10000
+        call void @dummy_use (i32* %mem, i32 %a)
+        ret void
 }
 
 define void @test_nostack() #0 {
-  ret void
-
 ; X86-Linux-LABEL: test_nostack:
-; X86-Linux-NOT:   calll __morestack
-
+; X86-Linux:       # %bb.0:
+; X86-Linux-NEXT:    retl
+;
 ; X64-Linux-LABEL: test_nostack:
-; X64-Linux-NOT:   callq __morestack
-
+; X64-Linux:       # %bb.0:
+; X64-Linux-NEXT:    retq
+;
+; X64-Linux-Large-LABEL: test_nostack:
+; X64-Linux-Large:       # %bb.0:
+; X64-Linux-Large-NEXT:    retq
+;
 ; X32ABI-LABEL: test_nostack:
-; X32ABI-NOT:   callq __morestack
-
+; X32ABI:       # %bb.0:
+; X32ABI-NEXT:    retq
+;
 ; X86-Darwin-LABEL: test_nostack:
-; X86-Darwin-NOT:   calll __morestack
-
+; X86-Darwin:       ## %bb.0:
+; X86-Darwin-NEXT:    retl
+;
 ; X64-Darwin-LABEL: test_nostack:
-; X64-Darwin-NOT:   callq __morestack
-
+; X64-Darwin:       ## %bb.0:
+; X64-Darwin-NEXT:    retq
+;
 ; X86-MinGW-LABEL: test_nostack:
-; X86-MinGW-NOT:   calll __morestack
-
-; X64-MinGW-LABEL: test_nostack:
-; X64-MinGW-NOT:   callq __morestack
-
+; X86-MinGW:       # %bb.0:
+; X86-MinGW-NEXT:    retl
+;
 ; X64-FreeBSD-LABEL: test_nostack:
-; X64-FreeBSD-NOT:   callq __morestack
-
+; X64-FreeBSD:       # %bb.0:
+; X64-FreeBSD-NEXT:    retq
+;
 ; X86-DFlyBSD-LABEL: test_nostack:
-; X86-DFlyBSD-NOT:   calll __morestack
-
+; X86-DFlyBSD:       # %bb.0:
+; X86-DFlyBSD-NEXT:    retl
+;
 ; X64-DFlyBSD-LABEL: test_nostack:
-; X64-DFlyBSD-NOT:   callq __morestack
+; X64-DFlyBSD:       # %bb.0:
+; X64-DFlyBSD-NEXT:    retq
+;
+; X64-MinGW-LABEL: test_nostack:
+; X64-MinGW:       # %bb.0:
+; X64-MinGW-NEXT:    retq
+  ret void
 }
 
 define void @test_nosplitstck() {
+; X86-Linux-LABEL: test_nosplitstck:
+; X86-Linux:       # %bb.0:
+; X86-Linux-NEXT:    retl
+;
+; X64-Linux-LABEL: test_nosplitstck:
+; X64-Linux:       # %bb.0:
+; X64-Linux-NEXT:    retq
+;
+; X64-Linux-Large-LABEL: test_nosplitstck:
+; X64-Linux-Large:       # %bb.0:
+; X64-Linux-Large-NEXT:    retq
+;
+; X32ABI-LABEL: test_nosplitstck:
+; X32ABI:       # %bb.0:
+; X32ABI-NEXT:    retq
+;
+; X86-Darwin-LABEL: test_nosplitstck:
+; X86-Darwin:       ## %bb.0:
+; X86-Darwin-NEXT:    retl
+;
+; X64-Darwin-LABEL: test_nosplitstck:
+; X64-Darwin:       ## %bb.0:
+; X64-Darwin-NEXT:    retq
+;
+; X86-MinGW-LABEL: test_nosplitstck:
+; X86-MinGW:       # %bb.0:
+; X86-MinGW-NEXT:    retl
+;
+; X64-FreeBSD-LABEL: test_nosplitstck:
+; X64-FreeBSD:       # %bb.0:
+; X64-FreeBSD-NEXT:    retq
+;
+; X86-DFlyBSD-LABEL: test_nosplitstck:
+; X86-DFlyBSD:       # %bb.0:
+; X86-DFlyBSD-NEXT:    retl
+;
+; X64-DFlyBSD-LABEL: test_nosplitstck:
+; X64-DFlyBSD:       # %bb.0:
+; X64-DFlyBSD-NEXT:    retq
+;
+; X64-MinGW-LABEL: test_nosplitstck:
+; X64-MinGW:       # %bb.0:
+; X64-MinGW-NEXT:    retq
   ret void
 }
 
 ; Test to make sure that a morestack call is generated if there is a
 ; sibling call, even if the function in question has no stack frame
 ; (PR37807).
 
 declare i32 @callee(i32)
 
 define i32 @test_sibling_call_empty_frame(i32 %x) #0 {
-  %call = tail call i32 @callee(i32 %x) #0
-  ret i32 %call
-
-; X86-Linux-LABEL:       test_sibling_call_empty_frame:
-; X86-Linux:  calll __morestack
-
-; X64-Linux-LABEL:       test_sibling_call_empty_frame:
-; X64-Linux:  callq __morestack
-
-; X64-Linux-Large-LABEL:       test_sibling_call_empty_frame:
-; X64-Linux-Large:  callq *__morestack_addr(%rip)
-
-; X32ABI-LABEL:       test_sibling_call_empty_frame:
-; X32ABI:  callq __morestack
-
-; X86-Darwin-LABEL:      test_sibling_call_empty_frame:
-; X86-Darwin: calll ___morestack
-
-; X64-Darwin-LABEL:      test_sibling_call_empty_frame:
-; X64-Darwin: callq ___morestack
-
-; X86-MinGW-LABEL:       test_sibling_call_empty_frame:
-; X86-MinGW:  calll ___morestack
-
-; X64-MinGW-LABEL:       test_sibling_call_empty_frame:
-; X64-MinGW:  callq __morestack
-
-; X64-FreeBSD-LABEL:       test_sibling_call_empty_frame:
-; X64-FreeBSD:  callq __morestack
-
-; X86-DFlyBSD-LABEL:       test_sibling_call_empty_frame:
-; X86-DFlyBSD:  calll __morestack
-; X86-DFlyBSD-NEXT:  ret
-
-; X64-DFlyBSD-LABEL:       test_sibling_call_empty_frame:
-; X64-DFlyBSD:  callq __morestack
-
+; X86-Linux-LABEL: test_sibling_call_empty_frame:
+; X86-Linux:       # %bb.0:
+; X86-Linux-NEXT:    cmpl %gs:48, %esp
+; X86-Linux-NEXT:    jbe .LBB8_1
+; X86-Linux-NEXT:  # %bb.2:
+; X86-Linux-NEXT:    jmp callee@PLT # TAILCALL
+; X86-Linux-NEXT:  .LBB8_1:
+; X86-Linux-NEXT:    pushl $4
+; X86-Linux-NEXT:    pushl $0
+; X86-Linux-NEXT:    calll __morestack
+; X86-Linux-NEXT:    retl
+; X86-Linux-NEXT:    jmp callee@PLT # TAILCALL
+;
+; X64-Linux-LABEL: test_sibling_call_empty_frame:
+; X64-Linux:       # %bb.0:
+; X64-Linux-NEXT:    cmpq %fs:112, %rsp
+; X64-Linux-NEXT:    jbe .LBB8_1
+; X64-Linux-NEXT:  # %bb.2:
+; X64-Linux-NEXT:    jmp callee@PLT # TAILCALL
+; X64-Linux-NEXT:  .LBB8_1:
+; X64-Linux-NEXT:    movabsq $0, %r10
+; X64-Linux-NEXT:    movabsq $0, %r11
+; X64-Linux-NEXT:    callq __morestack
+; X64-Linux-NEXT:    retq
+; X64-Linux-NEXT:    jmp callee@PLT # TAILCALL
+;
+; X64-Linux-Large-LABEL: test_sibling_call_empty_frame:
+; X64-Linux-Large:       # %bb.0:
+; X64-Linux-Large-NEXT:    cmpq %fs:112, %rsp
+; X64-Linux-Large-NEXT:    jbe .LBB8_1
+; X64-Linux-Large-NEXT:  # %bb.2:
+; X64-Linux-Large-NEXT:    movabsq $callee, %rax
+; X64-Linux-Large-NEXT:    jmpq *%rax # TAILCALL
+; X64-Linux-Large-NEXT:  .LBB8_1:
+; X64-Linux-Large-NEXT:    movabsq $0, %r10
+; X64-Linux-Large-NEXT:    movabsq $0, %r11
+; X64-Linux-Large-NEXT:    callq *__morestack_addr(%rip)
+; X64-Linux-Large-NEXT:    retq
+; X64-Linux-Large-NEXT:    movabsq $callee, %rax
+; X64-Linux-Large-NEXT:    jmpq *%rax # TAILCALL
+;
+; X32ABI-LABEL: test_sibling_call_empty_frame:
+; X32ABI:       # %bb.0:
+; X32ABI-NEXT:    cmpl %fs:64, %esp
+; X32ABI-NEXT:    jbe .LBB8_1
+; X32ABI-NEXT:  # %bb.2:
+; X32ABI-NEXT:    jmp callee@PLT # TAILCALL
+; X32ABI-NEXT:  .LBB8_1:
+; X32ABI-NEXT:    movl $0, %r10d
+; X32ABI-NEXT:    movl $0, %r11d
+; X32ABI-NEXT:    callq __morestack
+; X32ABI-NEXT:    retq
+; X32ABI-NEXT:    jmp callee@PLT # TAILCALL
+;
+; X86-Darwin-LABEL: test_sibling_call_empty_frame:
+; X86-Darwin:       ## %bb.0:
+; X86-Darwin-NEXT:    movl $432, %ecx ## imm = 0x1B0
+; X86-Darwin-NEXT:    cmpl %gs:(%ecx), %esp
+; X86-Darwin-NEXT:    jbe LBB8_1
+; X86-Darwin-NEXT:  ## %bb.2:
+; X86-Darwin-NEXT:    jmp _callee ## TAILCALL
+; X86-Darwin-NEXT:  LBB8_1:
+; X86-Darwin-NEXT:    pushl $4
+; X86-Darwin-NEXT:    pushl $0
+; X86-Darwin-NEXT:    calll ___morestack
+; X86-Darwin-NEXT:    retl
+; X86-Darwin-NEXT:    jmp _callee ## TAILCALL
+;
+; X64-Darwin-LABEL: test_sibling_call_empty_frame:
+; X64-Darwin:       ## %bb.0:
+; X64-Darwin-NEXT:    cmpq %gs:816, %rsp
+; X64-Darwin-NEXT:    jbe LBB8_1
+; X64-Darwin-NEXT:  ## %bb.2:
+; X64-Darwin-NEXT:    jmp _callee ## TAILCALL
+; X64-Darwin-NEXT:  LBB8_1:
+; X64-Darwin-NEXT:    movabsq $0, %r10
+; X64-Darwin-NEXT:    movabsq $0, %r11
+; X64-Darwin-NEXT:    callq ___morestack
+; X64-Darwin-NEXT:    retq
+; X64-Darwin-NEXT:    jmp _callee ## TAILCALL
+;
+; X86-MinGW-LABEL: test_sibling_call_empty_frame:
+; X86-MinGW:       # %bb.0:
+; X86-MinGW-NEXT:    cmpl %fs:20, %esp
+; X86-MinGW-NEXT:    jbe LBB8_1
+; X86-MinGW-NEXT:  # %bb.2:
+; X86-MinGW-NEXT:    jmp _callee # TAILCALL
+; X86-MinGW-NEXT:  LBB8_1:
+; X86-MinGW-NEXT:    pushl $4
+; X86-MinGW-NEXT:    pushl $0
+; X86-MinGW-NEXT:    calll ___morestack
+; X86-MinGW-NEXT:    retl
+; X86-MinGW-NEXT:    jmp _callee # TAILCALL
+;
+; X64-FreeBSD-LABEL: test_sibling_call_empty_frame:
+; X64-FreeBSD:       # %bb.0:
+; X64-FreeBSD-NEXT:    cmpq %fs:24, %rsp
+; X64-FreeBSD-NEXT:    jbe .LBB8_1
+; X64-FreeBSD-NEXT:  # %bb.2:
+; X64-FreeBSD-NEXT:    jmp callee@PLT # TAILCALL
+; X64-FreeBSD-NEXT:  .LBB8_1:
+; X64-FreeBSD-NEXT:    movabsq $0, %r10
+; X64-FreeBSD-NEXT:    movabsq $0, %r11
+; X64-FreeBSD-NEXT:    callq __morestack
+; X64-FreeBSD-NEXT:    retq
+; X64-FreeBSD-NEXT:    jmp callee@PLT # TAILCALL
+;
+; X86-DFlyBSD-LABEL: test_sibling_call_empty_frame:
+; X86-DFlyBSD:       # %bb.0:
+; X86-DFlyBSD-NEXT:    cmpl %fs:16, %esp
+; X86-DFlyBSD-NEXT:    jbe .LBB8_1
+; X86-DFlyBSD-NEXT:  # %bb.2:
+; X86-DFlyBSD-NEXT:    jmp callee@PLT # TAILCALL
+; X86-DFlyBSD-NEXT:  .LBB8_1:
+; X86-DFlyBSD-NEXT:    pushl $4
+; X86-DFlyBSD-NEXT:    pushl $0
+; X86-DFlyBSD-NEXT:    calll __morestack
+; X86-DFlyBSD-NEXT:    retl
+; X86-DFlyBSD-NEXT:    jmp callee@PLT # TAILCALL
+;
+; X64-DFlyBSD-LABEL: test_sibling_call_empty_frame:
+; X64-DFlyBSD:       # %bb.0:
+; X64-DFlyBSD-NEXT:    cmpq %fs:32, %rsp
+; X64-DFlyBSD-NEXT:    jbe .LBB8_1
+; X64-DFlyBSD-NEXT:  # %bb.2:
+; X64-DFlyBSD-NEXT:    jmp callee@PLT # TAILCALL
+; X64-DFlyBSD-NEXT:  .LBB8_1:
+; X64-DFlyBSD-NEXT:    movabsq $0, %r10
+; X64-DFlyBSD-NEXT:    movabsq $0, %r11
+; X64-DFlyBSD-NEXT:    callq __morestack
+; X64-DFlyBSD-NEXT:    retq
+; X64-DFlyBSD-NEXT:    jmp callee@PLT # TAILCALL
+;
+; X64-MinGW-LABEL: test_sibling_call_empty_frame:
+; X64-MinGW:       # %bb.0:
+; X64-MinGW-NEXT:    cmpq %gs:40, %rsp
+; X64-MinGW-NEXT:    jbe .LBB8_1
+; X64-MinGW-NEXT:  # %bb.2:
+; X64-MinGW-NEXT:    jmp callee # TAILCALL
+; X64-MinGW-NEXT:  .LBB8_1:
+; X64-MinGW-NEXT:    movabsq $0, %r10
+; X64-MinGW-NEXT:    movabsq $32, %r11
+; X64-MinGW-NEXT:    callq __morestack
+; X64-MinGW-NEXT:    retq
+; X64-MinGW-NEXT:    jmp callee # TAILCALL
+  %call = tail call i32 @callee(i32 %x) #0
+  ret i32 %call
 }
 
 ; Test that unused nested argument doesn't need saving/restoring.
 
 define i32 @test_nested_unused(i32 * nest %unused) #0 {
-       %mem = alloca i32, i32 10
-       call void @dummy_use (i32* %mem, i32 10)
-       ret i32 123
-
+; X86-Linux-LABEL: test_nested_unused:
+; X86-Linux:       # %bb.0:
+; X86-Linux-NEXT:    cmpl %gs:48, %esp
+; X86-Linux-NEXT:    jbe .LBB9_1
+; X86-Linux-NEXT:  .LBB9_2:
+; X86-Linux-NEXT:    subl $52, %esp
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset 52
+; X86-Linux-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X86-Linux-NEXT:    pushl $10
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-Linux-NEXT:    pushl %eax
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-Linux-NEXT:    calll dummy_use@PLT
+; X86-Linux-NEXT:    addl $16, %esp
+; X86-Linux-NEXT:    .cfi_adjust_cfa_offset -16
+; X86-Linux-NEXT:    movl $123, %eax
+; X86-Linux-NEXT:    addl $44, %esp
+; X86-Linux-NEXT:    .cfi_def_cfa_offset 4
+; X86-Linux-NEXT:    retl
+; X86-Linux-NEXT:  .LBB9_1:
+; X86-Linux-NEXT:    pushl $0
+; X86-Linux-NEXT:    pushl $44
+; X86-Linux-NEXT:    calll __morestack
+; X86-Linux-NEXT:    retl
+; X86-Linux-NEXT:    jmp .LBB9_2
+;
 ; X64-Linux-LABEL: test_nested_unused:
-; X64-Linux-NOT:   movq %r10, %rax
-; X64-Linux:       callq __morestack
-; X64-Linux-NOT:   movq %rax, %r10
-
+; X64-Linux:       # %bb.0:
+; X64-Linux-NEXT:    cmpq %fs:112, %rsp
+; X64-Linux-NEXT:    jbe .LBB9_1
+; X64-Linux-NEXT:  .LBB9_2:
+; X64-Linux-NEXT:    subq $40, %rsp
+; X64-Linux-NEXT:    .cfi_def_cfa_offset 48
+; X64-Linux-NEXT:    movq %rsp, %rdi
+; X64-Linux-NEXT:    movl $10, %esi
+; X64-Linux-NEXT:    callq dummy_use@PLT
+; X64-Linux-NEXT:    movl $123, %eax
+; X64-Linux-NEXT:    addq $40, %rsp
+; X64-Linux-NEXT:    .cfi_def_cfa_offset 8
+; X64-Linux-NEXT:    retq
+; X64-Linux-NEXT:  .LBB9_1:
+; X64-Linux-NEXT:    movabsq $40, %r10
+; X64-Linux-NEXT:    movabsq $0, %r11
+; X64-Linux-NEXT:    callq __morestack
+; X64-Linux-NEXT:    retq
+; X64-Linux-NEXT:    jmp .LBB9_2
+;
+; X64-Linux-Large-LABEL: test_nested_unused:
+; X64-Linux-Large:       # %bb.0:
+; X64-Linux-Large-NEXT:    cmpq %fs:112, %rsp
+; X64-Linux-Large-NEXT:    jbe .LBB9_1
+; X64-Linux-Large-NEXT:  .LBB9_2:
+; X64-Linux-Large-NEXT:    subq $40, %rsp
+; X64-Linux-Large-NEXT:    .cfi_def_cfa_offset 48
+; X64-Linux-Large-NEXT:    movabsq $dummy_use, %rax
+; X64-Linux-Large-NEXT:    movq %rsp, %rdi
+; X64-Linux-Large-NEXT:    movl $10, %esi
+; X64-Linux-Large-NEXT:    callq *%rax
+; X64-Linux-Large-NEXT:    movl $123, %eax
+; X64-Linux-Large-NEXT:    addq $40, %rsp
+; X64-Linux-Large-NEXT:    .cfi_def_cfa_offset 8
+; X64-Linux-Large-NEXT:    retq
+; X64-Linux-Large-NEXT:  .LBB9_1:
+; X64-Linux-Large-NEXT:    movabsq $40, %r10
+; X64-Linux-Large-NEXT:    movabsq $0, %r11
+; X64-Linux-Large-NEXT:    callq *__morestack_addr(%rip)
+; X64-Linux-Large-NEXT:    retq
+; X64-Linux-Large-NEXT:    jmp .LBB9_2
+;
+; X32ABI-LABEL: test_nested_unused:
+; X32ABI:       # %bb.0:
+; X32ABI-NEXT:    cmpl %fs:64, %esp
+; X32ABI-NEXT:    jbe .LBB9_1
+; X32ABI-NEXT:  .LBB9_2:
+; X32ABI-NEXT:    subl $40, %esp
+; X32ABI-NEXT:    .cfi_def_cfa_offset 48
+; X32ABI-NEXT:    movl %esp, %edi
+; X32ABI-NEXT:    movl $10, %esi
+; X32ABI-NEXT:    callq dummy_use@PLT
+; X32ABI-NEXT:    movl $123, %eax
+; X32ABI-NEXT:    addl $40, %esp
+; X32ABI-NEXT:    .cfi_def_cfa_offset 8
+; X32ABI-NEXT:    retq
+; X32ABI-NEXT:  .LBB9_1:
+; X32ABI-NEXT:    movl $40, %r10d
+; X32ABI-NEXT:    movl $0, %r11d
+; X32ABI-NEXT:    callq __morestack
+; X32ABI-NEXT:    retq
+; X32ABI-NEXT:    jmp .LBB9_2
+;
+; X86-Darwin-LABEL: test_nested_unused:
+; X86-Darwin:       ## %bb.0:
+; X86-Darwin-NEXT:    movl $432, %ecx ## imm = 0x1B0
+; X86-Darwin-NEXT:    cmpl %gs:(%ecx), %esp
+; X86-Darwin-NEXT:    jbe LBB9_1
+; X86-Darwin-NEXT:  LBB9_2:
+; X86-Darwin-NEXT:    subl $60, %esp
+; X86-Darwin-NEXT:    .cfi_def_cfa_offset 64
+; X86-Darwin-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X86-Darwin-NEXT:    movl %eax, (%esp)
+; X86-Darwin-NEXT:    movl $10, {{[0-9]+}}(%esp)
+; X86-Darwin-NEXT:    calll _dummy_use
+; X86-Darwin-NEXT:    movl $123, %eax
+; X86-Darwin-NEXT:    addl $60, %esp
+; X86-Darwin-NEXT:    retl
+; X86-Darwin-NEXT:  LBB9_1:
+; X86-Darwin-NEXT:    pushl $0
+; X86-Darwin-NEXT:    pushl $60
+; X86-Darwin-NEXT:    calll ___morestack
+; X86-Darwin-NEXT:    retl
+; X86-Darwin-NEXT:    jmp LBB9_2
+;
 ; X64-Darwin-LABEL: test_nested_unused:
-; X64-Darwin-NOT:   movq %r10, %rax
-; X64-Darwin:       callq ___morestack
-; X64-Darwin-NOT:   movq %rax, %r10
-
+; X64-Darwin:       ## %bb.0:
+; X64-Darwin-NEXT:    cmpq %gs:816, %rsp
+; X64-Darwin-NEXT:    jbe LBB9_1
+; X64-Darwin-NEXT:  LBB9_2:
+; X64-Darwin-NEXT:    subq $40, %rsp
+; X64-Darwin-NEXT:    .cfi_def_cfa_offset 48
+; X64-Darwin-NEXT:    movq %rsp, %rdi
+; X64-Darwin-NEXT:    movl $10, %esi
+; X64-Darwin-NEXT:    callq _dummy_use
+; X64-Darwin-NEXT:    movl $123, %eax
+; X64-Darwin-NEXT:    addq $40, %rsp
+; X64-Darwin-NEXT:    retq
+; X64-Darwin-NEXT:  LBB9_1:
+; X64-Darwin-NEXT:    movabsq $40, %r10
+; X64-Darwin-NEXT:    movabsq $0, %r11
+; X64-Darwin-NEXT:    callq ___morestack
+; X64-Darwin-NEXT:    retq
+; X64-Darwin-NEXT:    jmp LBB9_2
+;
+; X86-MinGW-LABEL: test_nested_unused:
+; X86-MinGW:       # %bb.0:
+; X86-MinGW-NEXT:    cmpl %fs:20, %esp
+; X86-MinGW-NEXT:    jbe LBB9_1
+; X86-MinGW-NEXT:  LBB9_2:
+; X86-MinGW-NEXT:    subl $40, %esp
+; X86-MinGW-NEXT:    .cfi_def_cfa_offset 44
+; X86-MinGW-NEXT:    movl %esp, %eax
+; X86-MinGW-NEXT:    pushl $10
+; X86-MinGW-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-MinGW-NEXT:    pushl %eax
+; X86-MinGW-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-MinGW-NEXT:    calll _dummy_use
+; X86-MinGW-NEXT:    addl $8, %esp
+; X86-MinGW-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-MinGW-NEXT:    movl $123, %eax
+; X86-MinGW-NEXT:    addl $40, %esp
+; X86-MinGW-NEXT:    retl
+; X86-MinGW-NEXT:  LBB9_1:
+; X86-MinGW-NEXT:    .cfi_def_cfa_offset 4
+; X86-MinGW-NEXT:    pushl $0
+; X86-MinGW-NEXT:    pushl $40
+; X86-MinGW-NEXT:    calll ___morestack
+; X86-MinGW-NEXT:    retl
+; X86-MinGW-NEXT:    jmp LBB9_2
+;
 ; X64-FreeBSD-LABEL: test_nested_unused:
-; X64-FreeBSD-NOT:   movq %r10, %rax
-; X64-FreeBSD:       callq __morestack
-; X64-FreeBSD-NOT:   movq %rax, %r10
-
+; X64-FreeBSD:       # %bb.0:
+; X64-FreeBSD-NEXT:    cmpq %fs:24, %rsp
+; X64-FreeBSD-NEXT:    jbe .LBB9_1
+; X64-FreeBSD-NEXT:  .LBB9_2:
+; X64-FreeBSD-NEXT:    subq $40, %rsp
+; X64-FreeBSD-NEXT:    .cfi_def_cfa_offset 48
+; X64-FreeBSD-NEXT:    movq %rsp, %rdi
+; X64-FreeBSD-NEXT:    movl $10, %esi
+; X64-FreeBSD-NEXT:    callq dummy_use@PLT
+; X64-FreeBSD-NEXT:    movl $123, %eax
+; X64-FreeBSD-NEXT:    addq $40, %rsp
+; X64-FreeBSD-NEXT:    .cfi_def_cfa_offset 8
+; X64-FreeBSD-NEXT:    retq
+; X64-FreeBSD-NEXT:  .LBB9_1:
+; X64-FreeBSD-NEXT:    movabsq $40, %r10
+; X64-FreeBSD-NEXT:    movabsq $0, %r11
+; X64-FreeBSD-NEXT:    callq __morestack
+; X64-FreeBSD-NEXT:    retq
+; X64-FreeBSD-NEXT:    jmp .LBB9_2
+;
+; X86-DFlyBSD-LABEL: test_nested_unused:
+; X86-DFlyBSD:       # %bb.0:
+; X86-DFlyBSD-NEXT:    cmpl %fs:16, %esp
+; X86-DFlyBSD-NEXT:    jbe .LBB9_1
+; X86-DFlyBSD-NEXT:  .LBB9_2:
+; X86-DFlyBSD-NEXT:    subl $40, %esp
+; X86-DFlyBSD-NEXT:    .cfi_def_cfa_offset 44
+; X86-DFlyBSD-NEXT:    movl %esp, %eax
+; X86-DFlyBSD-NEXT:    pushl $10
+; X86-DFlyBSD-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-DFlyBSD-NEXT:    pushl %eax
+; X86-DFlyBSD-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-DFlyBSD-NEXT:    calll dummy_use@PLT
+; X86-DFlyBSD-NEXT:    addl $8, %esp
+; X86-DFlyBSD-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-DFlyBSD-NEXT:    movl $123, %eax
+; X86-DFlyBSD-NEXT:    addl $40, %esp
+; X86-DFlyBSD-NEXT:    .cfi_def_cfa_offset 4
+; X86-DFlyBSD-NEXT:    retl
+; X86-DFlyBSD-NEXT:  .LBB9_1:
+; X86-DFlyBSD-NEXT:    pushl $0
+; X86-DFlyBSD-NEXT:    pushl $40
+; X86-DFlyBSD-NEXT:    calll __morestack
+; X86-DFlyBSD-NEXT:    retl
+; X86-DFlyBSD-NEXT:    jmp .LBB9_2
+;
 ; X64-DFlyBSD-LABEL: test_nested_unused:
-; X64-DFlyBSD-NOT:   movq %r10, %rax
-; X64-DFlyBSD:       callq __morestack
-; X64-DFlyBSD-NOT:   movq %rax, %r10
-
+; X64-DFlyBSD:       # %bb.0:
+; X64-DFlyBSD-NEXT:    cmpq %fs:32, %rsp
+; X64-DFlyBSD-NEXT:    jbe .LBB9_1
+; X64-DFlyBSD-NEXT:  .LBB9_2:
+; X64-DFlyBSD-NEXT:    subq $40, %rsp
+; X64-DFlyBSD-NEXT:    .cfi_def_cfa_offset 48
+; X64-DFlyBSD-NEXT:    movq %rsp, %rdi
+; X64-DFlyBSD-NEXT:    movl $10, %esi
+; X64-DFlyBSD-NEXT:    callq dummy_use@PLT
+; X64-DFlyBSD-NEXT:    movl $123, %eax
+; X64-DFlyBSD-NEXT:    addq $40, %rsp
+; X64-DFlyBSD-NEXT:    .cfi_def_cfa_offset 8
+; X64-DFlyBSD-NEXT:    retq
+; X64-DFlyBSD-NEXT:  .LBB9_1:
+; X64-DFlyBSD-NEXT:    movabsq $40, %r10
+; X64-DFlyBSD-NEXT:    movabsq $0, %r11
+; X64-DFlyBSD-NEXT:    callq __morestack
+; X64-DFlyBSD-NEXT:    retq
+; X64-DFlyBSD-NEXT:    jmp .LBB9_2
+;
 ; X64-MinGW-LABEL: test_nested_unused:
-; X64-MinGW-NOT:   movq %r10, %rax
-; X64-MinGW:       callq __morestack
-; X64-MinGW-NOT:   movq %rax, %r10
-
-; X32ABI-LABEL: test_nested_unused:
-; X32ABI-NOT:   movl %r10d, %eax
-; X32ABI:       callq __morestack
-; X32ABI-NOT:   movq %rax, %r10
-
+; X64-MinGW:       # %bb.0:
+; X64-MinGW-NEXT:    cmpq %gs:40, %rsp
+; X64-MinGW-NEXT:    jbe .LBB9_1
+; X64-MinGW-NEXT:  .LBB9_2:
+; X64-MinGW-NEXT:    subq $72, %rsp
+; X64-MinGW-NEXT:    .seh_stackalloc 72
+; X64-MinGW-NEXT:    .seh_endprologue
+; X64-MinGW-NEXT:    leaq {{[0-9]+}}(%rsp), %rcx
+; X64-MinGW-NEXT:    movl $10, %edx
+; X64-MinGW-NEXT:    callq dummy_use
+; X64-MinGW-NEXT:    movl $123, %eax
+; X64-MinGW-NEXT:    addq $72, %rsp
+; X64-MinGW-NEXT:    retq
+; X64-MinGW-NEXT:  .LBB9_1:
+; X64-MinGW-NEXT:    movabsq $72, %r10
+; X64-MinGW-NEXT:    movabsq $32, %r11
+; X64-MinGW-NEXT:    callq __morestack
+; X64-MinGW-NEXT:    retq
+; X64-MinGW-NEXT:    jmp .LBB9_2
+; X64-MinGW-NEXT:    .seh_endproc
+       %mem = alloca i32, i32 10
+       call void @dummy_use (i32* %mem, i32 10)
+       ret i32 123
 }
 
 attributes #0 = { "split-stack" }
 
 ; X64-Linux-Large: .rodata
 ; X64-Linux-Large-NEXT: __morestack_addr:
 ; X64-Linux-Large-NEXT: .quad __morestack