[X86][MS] Change the alignment of f80 to 16 bytes on Windows 32bits to match with ICC

[pengfei]

Summary

MSVC currently doesn't support 80 bits long double. ICC supports it when
the option /Qlong-double is specified. Changing the alignment of f80
to 16 bytes so that we can be compatible with ICC's option.

[pengfei]

This is split from D115441. Add a targeted LLVM test. Thanks @rnk for the advise.

[rnk]

Thanks, looks good. I have a minor code refactoring suggestion, but feel free to resolve it as you see fit and land it.

clang/lib/Basic/Targets/X86.h

534–535

I think it would be simpler to make the MSVC mode data layout change here.

It is also my understanding that, prior to rGb214cbc7852fa1719c5d0b, only long-lived string constants could be passed to resetDataLayout. Since 2016, that is unneeded. resetDataLayout now copies its string argument. (@jyknight @thakis, who made the changes here).

I suggest structuring the code like this:

bool IsWinCOFF = getTriple()...
bool IsMSVC = getTriple()...
StringRef Mangling = IsWinCOFF ? "m:x" : "m:e";
StringRef F80 = IsMSVC ? "f80:128" : "f80:32";
std::string Layout = "e-" + Mangling + "-p:32..." + F80 + "-n8:...";
resetDataLayout(Layout, IsWinCOFF ? "_" : "");

[craig.topper]

Does this have the same autoupgrade issues as @efriedma raised in https://reviews.llvm.org/D86310

[rnk]

Yeah, let's try to reach some resolution on that.

In the mean time, I discovered the alignstack parameter attribute:
https://llvm.org/docs/LangRef.html#parameter-attributes

Could that be used to solve this problem in the frontend instead?

[pengfei]

In D115942#3226101, @craig.topper wrote:

Does this have the same autoupgrade issues as @efriedma raised in https://reviews.llvm.org/D86310

No. The differences are that i128 can be generated by front end while f80 cannot without D115441 in the given datalayout (i.e., MSVC).
I also did some experiments, e.g., generating a bc file (without f80 type) with existing compiler and compile it with the one with this patch. There's no error during compilation. So I think we don't bother to do autoupgrade at all.

[pengfei]

In D115942#3226146, @rnk wrote:

Yeah, let's try to reach some resolution on that.

The things are different. We don't support f80 type on Windows 32 bits previously. It means we don't have the burden to upgrade, since there's no global/load/store/GEP/etc for f80.

In the mean time, I discovered the alignstack parameter attribute:
https://llvm.org/docs/LangRef.html#parameter-attributes

Could that be used to solve this problem in the frontend instead?

It might be feasible, but I don't think it's a good idea. It looks to me more like a language specific alignment hint instead of medium that carrying target specific information.

[pengfei]

Structure the layout spelling.

[rnk]

In D115942#3227047, @pengfei wrote:

In D115942#3226146, @rnk wrote:

Yeah, let's try to reach some resolution on that.

The things are different. We don't support f80 type on Windows 32 bits previously. It means we don't have the burden to upgrade, since there's no global/load/store/GEP/etc for f80.

Good point, I buy that argument.

[rnk]

However, do we need to make changes to llvm::UpgradeDataLayoutString? Otherwise, I believe old bitcode for MSVC targets will no longer be able to be loaded for LTO.

[pengfei]

However, do we need to make changes to llvm::UpgradeDataLayoutString? Otherwise, I believe old bitcode for MSVC targets will no longer be able to be loaded for LTO.

Makes sense. Done, thanks!

[pengfei]

Missed one unittest.

[rnk]

I'm happy with this, but I'd like to get an ack from @craig.topper or @efriedma about the backwards compatibility concern. I think it's addresed.

llvm/lib/IR/AutoUpgrade.cpp

4583–4589

I think the early return here is not helping readability. Please add a comment here about why this upgrade is being done, something like:

// For 32-bit MSVC targets, raise the alignment of f80 values to 16 bytes. Raising the alignment is safe because Clang did not produce f80 values in the MSVC environment before this upgrade was added.
if (T.isWindowsMSVCEnvironment() && T.isArch32Bit()) {
 ....

[pengfei]

Add comments in code. @rnk, thanks for your patient review!

llvm/lib/IR/AutoUpgrade.cpp

4583–4589

That's better. Thanks for the suggestion.

[craig.topper]

LGTM

[pengfei]

Thanks Craig!

[fhahn]

It looks like this may be causing the following bot to fail https://lab.llvm.org/buildbot/#/builders/16/builds/22138 with

******************** TEST 'lld :: COFF/lto-lazy-reference.ll' FAILED ********************
Script:
--
: 'RUN: at line 2';   /b/1/llvm-clang-x86_64-expensive-checks-debian/build/bin/llc -mtriple=i686-pc-windows-msvc -filetype=obj -o /b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference-quadruple.obj /b/1/llvm-clang-x86_64-expensive-checks-debian/llvm-project/lld/test/COFF/Inputs/lto-lazy-reference-quadruple.ll
: 'RUN: at line 3';   /b/1/llvm-clang-x86_64-expensive-checks-debian/build/bin/llvm-as -o /b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference-dummy.bc /b/1/llvm-clang-x86_64-expensive-checks-debian/llvm-project/lld/test/COFF/Inputs/lto-lazy-reference-dummy.ll
: 'RUN: at line 4';   rm -f /b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference.ll.tmp.lib
: 'RUN: at line 5';   llvm-ar cru /b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference.ll.tmp.lib /b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference-quadruple.obj /b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference-dummy.bc
: 'RUN: at line 6';   /b/1/llvm-clang-x86_64-expensive-checks-debian/build/bin/llvm-as -o /b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference.ll.tmp.obj /b/1/llvm-clang-x86_64-expensive-checks-debian/llvm-project/lld/test/COFF/lto-lazy-reference.ll
: 'RUN: at line 7';   /b/1/llvm-clang-x86_64-expensive-checks-debian/build/bin/lld-link /out:/b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference.ll.tmp.exe /entry:main /subsystem:console /b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference.ll.tmp.obj /b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference.ll.tmp.lib
--
Exit Code: 134
Command Output (stderr):
--
+ : 'RUN: at line 2'
+ /b/1/llvm-clang-x86_64-expensive-checks-debian/build/bin/llc -mtriple=i686-pc-windows-msvc -filetype=obj -o /b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference-quadruple.obj /b/1/llvm-clang-x86_64-expensive-checks-debian/llvm-project/lld/test/COFF/Inputs/lto-lazy-reference-quadruple.ll
+ : 'RUN: at line 3'
+ /b/1/llvm-clang-x86_64-expensive-checks-debian/build/bin/llvm-as -o /b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference-dummy.bc /b/1/llvm-clang-x86_64-expensive-checks-debian/llvm-project/lld/test/COFF/Inputs/lto-lazy-reference-dummy.ll
+ : 'RUN: at line 4'
+ rm -f /b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference.ll.tmp.lib
+ : 'RUN: at line 5'
+ llvm-ar cru /b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference.ll.tmp.lib /b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference-quadruple.obj /b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference-dummy.bc
+ : 'RUN: at line 6'
+ /b/1/llvm-clang-x86_64-expensive-checks-debian/build/bin/llvm-as -o /b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference.ll.tmp.obj /b/1/llvm-clang-x86_64-expensive-checks-debian/llvm-project/lld/test/COFF/lto-lazy-reference.ll
+ : 'RUN: at line 7'
+ /b/1/llvm-clang-x86_64-expensive-checks-debian/build/bin/lld-link /out:/b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference.ll.tmp.exe /entry:main /subsystem:console /b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference.ll.tmp.obj /b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference.ll.tmp.lib
lld-link: /b/1/llvm-clang-x86_64-expensive-checks-debian/llvm-project/llvm/lib/CodeGen/MachineFunction.cpp:207: void llvm::MachineFunction::init(): Assertion `Target.isCompatibleDataLayout(getDataLayout()) && "Can't create a MachineFunction using a Module with a " "Target-incompatible DataLayout attached\n"' failed.
/b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference.ll.script: line 6: 2469401 Aborted                 /b/1/llvm-clang-x86_64-expensive-checks-debian/build/bin/lld-link /out:/b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference.ll.tmp.exe /entry:main /subsystem:console /b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference.ll.tmp.obj /b/1/llvm-clang-x86_64-expensive-checks-debian/build/tools/lld/test/COFF/Output/lto-lazy-reference.ll.tmp.lib
--

[pengfei]

Thanks @fhahn ! I saw the failure. I think it just needs to update the datalayout in the lit tests. I am checking on this.

[pengfei]

Should be fixed by rG9b43237128da.

[efriedma]

Not sure I agree with the "fix" to the lld tests. If the linker is crashing, clearly there's a bug, not just an incorrect test.

[rnk]

Right, the auto-upgrade should upgrade and prevent the need to change the bitcode data layout. LTO is the main use case for auto-upgrading.

[pengfei]

I see, I will investigate the reason and do a follow up. Thanks @efriedma and @rnk !

[pengfei]

It caused by bug in AutoUpgrade.cpp, fixed on D117270.

[pengfei]

Fix AutoUpgrade bug catched by buildbot.

clang/lib/Basic/Targets/X86.h

 
 // x86-32 Windows target
 class LLVM_LIBRARY_VISIBILITY WindowsX86_32TargetInfo
     : public WindowsTargetInfo<X86_32TargetInfo> {
 public:
   WindowsX86_32TargetInfo(const llvm::Triple &Triple, const TargetOptions &Opts)
       : WindowsTargetInfo<X86_32TargetInfo>(Triple, Opts) {
     DoubleAlign = LongLongAlign = 64;
     bool IsWinCOFF =
         getTriple().isOSWindows() && getTriple().isOSBinFormatCOFF();
-    resetDataLayout(IsWinCOFF ? "e-m:x-p:32:32-p270:32:32-p271:32:32-p272:64:"
-                                "64-i64:64-f80:32-n8:16:32-a:0:32-S32"
-                              : "e-m:e-p:32:32-p270:32:32-p271:32:32-p272:64:"
-                                "64-i64:64-f80:32-n8:16:32-a:0:32-S32",
-                    IsWinCOFF ? "_" : "");
+    bool IsMSVC = getTriple().isWindowsMSVCEnvironment();
+    std::string Layout = IsWinCOFF ? "e-m:x" : "e-m:e";
+    Layout += "-p:32:32-p270:32:32-p271:32:32-p272:64:64-i64:64-";
+    Layout += IsMSVC ? "f80:128" : "f80:32";
+    Layout += "-n8:16:32-a:0:32-S32";
+    resetDataLayout(Layout, IsWinCOFF ? "_" : "");
   }
 };
 
 // x86-32 Windows Visual Studio target
 class LLVM_LIBRARY_VISIBILITY MicrosoftX86_32TargetInfo
     : public WindowsX86_32TargetInfo {
 public:
   MicrosoftX86_32TargetInfo(const llvm::Triple &Triple,

clang/test/CodeGen/target-data.c

 // RUN: %clang_cc1 -triple i686-unknown-unknown -emit-llvm -o - %s | \
 // RUN:     FileCheck --check-prefix=I686-UNKNOWN %s
 // I686-UNKNOWN: target datalayout = "e-m:e-p:32:32-p270:32:32-p271:32:32-p272:64:64-f64:32:64-f80:32-n8:16:32-S128"
 
 // RUN: %clang_cc1 -triple i686-apple-darwin9 -emit-llvm -o - %s | \
 // RUN:     FileCheck --check-prefix=I686-DARWIN %s
 // I686-DARWIN: target datalayout = "e-m:o-p:32:32-p270:32:32-p271:32:32-p272:64:64-f64:32:64-f80:128-n8:16:32-S128"
 
 // RUN: %clang_cc1 -triple i686-unknown-win32 -emit-llvm -o - %s | \
 // RUN:     FileCheck --check-prefix=I686-WIN32 %s
-// I686-WIN32: target datalayout = "e-m:x-p:32:32-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:32-n8:16:32-a:0:32-S32"
+// I686-WIN32: target datalayout = "e-m:x-p:32:32-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32-a:0:32-S32"
 
 // RUN: %clang_cc1 -triple i686-unknown-cygwin -emit-llvm -o - %s | \
 // RUN:     FileCheck --check-prefix=I686-CYGWIN %s
 // I686-CYGWIN: target datalayout = "e-m:x-p:32:32-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:32-n8:16:32-a:0:32-S32"
 
 // RUN: %clang_cc1 -triple i686-pc-macho -emit-llvm -o - %s | \
 // RUN:     FileCheck --check-prefix=I686-MACHO %s
 // I686-MACHO: target datalayout = "e-m:o-p:32:32-p270:32:32-p271:32:32-p272:64:64-f64:32:64-f80:32-n8:16:32-S128"

llvm/lib/IR/AutoUpgrade.cpp

 std::string llvm::UpgradeDataLayoutString(StringRef DL, StringRef TT) {
   Triple T(TT);
   // For AMDGPU we uprgrade older DataLayouts to include the default globals
   // address space of 1.
   if (T.isAMDGPU() && !DL.contains("-G") && !DL.startswith("G")) {
     return DL.empty() ? std::string("G1") : (DL + "-G1").str();
   }
 
+  std::string Res = DL.str();
   std::string AddrSpaces = "-p270:32:32-p271:32:32-p272:64:64";
   // If X86, and the datalayout matches the expected format, add pointer size
   // address spaces to the datalayout.
   if (!T.isX86() || DL.contains(AddrSpaces))
-    return std::string(DL);
+    return Res;
 
   SmallVector<StringRef, 4> Groups;
   Regex R("(e-m:[a-z](-p:32:32)?)(-[if]64:.*$)");
-  if (!R.match(DL, &Groups))
-    return std::string(DL);
+  if (R.match(DL, &Groups))
+    Res = (Groups[1] + AddrSpaces + Groups[3]).str();
+  if (T.isArch64Bit() || !T.isWindowsMSVCEnvironment())
+    return Res;
 
-  return (Groups[1] + AddrSpaces + Groups[3]).str();
+  StringRef Ref = Res;
+  auto I = Ref.find("-f80:32-");
+  if (I != StringRef::npos)
+    Res = (Ref.take_front(I) + "-f80:128-" + Ref.drop_front(I + 8)).str();
+
+  return Res;
 }
 
 void llvm::UpgradeAttributes(AttrBuilder &B) {
   StringRef FramePointer;
   if (B.contains("no-frame-pointer-elim")) {
     // The value can be "true" or "false".
     for (const auto &I : B.td_attrs())
       if (I.first == "no-frame-pointer-elim")

llvm/lib/Target/X86/X86TargetMachine.cpp

static std::string computeDataLayout(const Triple &TT) {
   else if (TT.isOSIAMCU())
     Ret += "-i64:32-f64:32";
   else
     Ret += "-f64:32:64";
 
   // Some ABIs align long double to 128 bits, others to 32.
   if (TT.isOSNaCl() || TT.isOSIAMCU())
     ; // No f80
-  else if (TT.isArch64Bit() || TT.isOSDarwin())
+  else if (TT.isArch64Bit() || TT.isOSDarwin() || TT.isWindowsMSVCEnvironment())
     Ret += "-f80:128";
   else
     Ret += "-f80:32";
 
   if (TT.isOSIAMCU())
     Ret += "-f128:32";
 
   // The registers can hold 8, 16, 32 or, in x86-64, 64 bits.

llvm/test/Bitcode/upgrade-datalayout3.ll

 ; Test to make sure datalayout is automatically upgraded.
 ;
 ; RUN: llvm-as %s -o - | llvm-dis - | FileCheck %s
 
 target datalayout = "e-m:w-p:32:32-i64:64-f80:32-n8:16:32-S32"
 target triple = "i686-pc-windows-msvc"
 
-; CHECK: target datalayout = "e-m:w-p:32:32-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:32-n8:16:32-S32"
+; CHECK: target datalayout = "e-m:w-p:32:32-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32-S32"

llvm/test/CodeGen/X86/long-double-abi-align.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --no_x86_scrub_sp --no_x86_scrub_mem_shuffle
 ; RUN: llc < %s -mtriple=i386-pc-windows-msvc | FileCheck %s --check-prefix=MSVC
 ; RUN: llc < %s -mtriple=i386-pc-mingw32 | FileCheck %s --check-prefix=MINGW
 ; RUN: llc < %s -mtriple=i386-pc-linux-gnu | FileCheck %s --check-prefix=LINUX
 ; RUN: llc < %s -mtriple=i386--apple-darwin  | FileCheck %s --check-prefix=DARWIN
 
 define void @foo(i32 %0, x86_fp80 %1, i32 %2) nounwind {
 ; MSVC-LABEL: foo:
 ; MSVC:       # %bb.0:
 ; MSVC-NEXT:    pushl %ebp
 ; MSVC-NEXT:    movl %esp, %ebp
 ; MSVC-NEXT:    andl $-16, %esp
 ; MSVC-NEXT:    subl $32, %esp
-; MSVC-NEXT:    fldt 12(%ebp)
+; MSVC-NEXT:    fldt 24(%ebp)
 ; MSVC-NEXT:    fstpt (%esp)
 ; MSVC-NEXT:    leal 8(%ebp), %eax
 ; MSVC-NEXT:    pushl %eax
 ; MSVC-NEXT:    calll _escape
 ; MSVC-NEXT:    addl $4, %esp
 ; MSVC-NEXT:    movl %esp, %eax
 ; MSVC-NEXT:    pushl %eax
 ; MSVC-NEXT:    calll _escape
 ; MSVC-NEXT:    addl $4, %esp
-; MSVC-NEXT:    leal 24(%ebp), %eax
+; MSVC-NEXT:    leal 40(%ebp), %eax
 ; MSVC-NEXT:    pushl %eax
 ; MSVC-NEXT:    calll _escape
 ; MSVC-NEXT:    addl $4, %esp
 ; MSVC-NEXT:    movl %ebp, %esp
 ; MSVC-NEXT:    popl %ebp
 ; MSVC-NEXT:    retl
 ;
 ; MINGW-LABEL: foo:

llvm/test/CodeGen/X86/scalar-fp-to-i32.ll

; X87-NEXT:    retl
   ret i32 %r
 }
 
 define i32 @x_to_u32(x86_fp80 %a) nounwind {
 ; X86-AVX512-WIN-LABEL: x_to_u32:
 ; X86-AVX512-WIN:       # %bb.0:
 ; X86-AVX512-WIN-NEXT:    pushl %ebp
 ; X86-AVX512-WIN-NEXT:    movl %esp, %ebp
-; X86-AVX512-WIN-NEXT:    andl $-8, %esp
-; X86-AVX512-WIN-NEXT:    subl $8, %esp
+; X86-AVX512-WIN-NEXT:    andl $-16, %esp
+; X86-AVX512-WIN-NEXT:    subl $16, %esp
 ; X86-AVX512-WIN-NEXT:    fldt 8(%ebp)
 ; X86-AVX512-WIN-NEXT:    fisttpll (%esp)
 ; X86-AVX512-WIN-NEXT:    movl (%esp), %eax
 ; X86-AVX512-WIN-NEXT:    movl %ebp, %esp
 ; X86-AVX512-WIN-NEXT:    popl %ebp
 ; X86-AVX512-WIN-NEXT:    retl
 ;
 ; X86-AVX512-LIN-LABEL: x_to_u32:

 ; X64-AVX512-LIN-NEXT:    fisttpll -{{[0-9]+}}(%rsp)
 ; X64-AVX512-LIN-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
 ; X64-AVX512-LIN-NEXT:    retq
 ;
 ; X86-SSE3-WIN-LABEL: x_to_u32:
 ; X86-SSE3-WIN:       # %bb.0:
 ; X86-SSE3-WIN-NEXT:    pushl %ebp
 ; X86-SSE3-WIN-NEXT:    movl %esp, %ebp
-; X86-SSE3-WIN-NEXT:    andl $-8, %esp
-; X86-SSE3-WIN-NEXT:    subl $8, %esp
+; X86-SSE3-WIN-NEXT:    andl $-16, %esp
+; X86-SSE3-WIN-NEXT:    subl $16, %esp
 ; X86-SSE3-WIN-NEXT:    fldt 8(%ebp)
 ; X86-SSE3-WIN-NEXT:    fisttpll (%esp)
 ; X86-SSE3-WIN-NEXT:    movl (%esp), %eax
 ; X86-SSE3-WIN-NEXT:    movl %ebp, %esp
 ; X86-SSE3-WIN-NEXT:    popl %ebp
 ; X86-SSE3-WIN-NEXT:    retl
 ;
 ; X86-SSE3-LIN-LABEL: x_to_u32:

 ; X64-SSE3-LIN-NEXT:    fisttpll -{{[0-9]+}}(%rsp)
 ; X64-SSE3-LIN-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
 ; X64-SSE3-LIN-NEXT:    retq
 ;
 ; X86-SSE2-WIN-LABEL: x_to_u32:
 ; X86-SSE2-WIN:       # %bb.0:
 ; X86-SSE2-WIN-NEXT:    pushl %ebp
 ; X86-SSE2-WIN-NEXT:    movl %esp, %ebp
-; X86-SSE2-WIN-NEXT:    andl $-8, %esp
-; X86-SSE2-WIN-NEXT:    subl $16, %esp
+; X86-SSE2-WIN-NEXT:    andl $-16, %esp
+; X86-SSE2-WIN-NEXT:    subl $32, %esp
 ; X86-SSE2-WIN-NEXT:    fldt 8(%ebp)
 ; X86-SSE2-WIN-NEXT:    fnstcw {{[0-9]+}}(%esp)
 ; X86-SSE2-WIN-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
 ; X86-SSE2-WIN-NEXT:    orl $3072, %eax # imm = 0xC00
 ; X86-SSE2-WIN-NEXT:    movw %ax, {{[0-9]+}}(%esp)
 ; X86-SSE2-WIN-NEXT:    fldcw {{[0-9]+}}(%esp)
 ; X86-SSE2-WIN-NEXT:    fistpll {{[0-9]+}}(%esp)
 ; X86-SSE2-WIN-NEXT:    fldcw {{[0-9]+}}(%esp)

 ; X64-SSE2-LIN-NEXT:    fldcw -{{[0-9]+}}(%rsp)
 ; X64-SSE2-LIN-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
 ; X64-SSE2-LIN-NEXT:    retq
 ;
 ; X86-SSE1-WIN-LABEL: x_to_u32:
 ; X86-SSE1-WIN:       # %bb.0:
 ; X86-SSE1-WIN-NEXT:    pushl %ebp
 ; X86-SSE1-WIN-NEXT:    movl %esp, %ebp
-; X86-SSE1-WIN-NEXT:    andl $-8, %esp
-; X86-SSE1-WIN-NEXT:    subl $16, %esp
+; X86-SSE1-WIN-NEXT:    andl $-16, %esp
+; X86-SSE1-WIN-NEXT:    subl $32, %esp
 ; X86-SSE1-WIN-NEXT:    fldt 8(%ebp)
 ; X86-SSE1-WIN-NEXT:    fnstcw {{[0-9]+}}(%esp)
 ; X86-SSE1-WIN-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
 ; X86-SSE1-WIN-NEXT:    orl $3072, %eax # imm = 0xC00
 ; X86-SSE1-WIN-NEXT:    movw %ax, {{[0-9]+}}(%esp)
 ; X86-SSE1-WIN-NEXT:    fldcw {{[0-9]+}}(%esp)
 ; X86-SSE1-WIN-NEXT:    fistpll {{[0-9]+}}(%esp)
 ; X86-SSE1-WIN-NEXT:    fldcw {{[0-9]+}}(%esp)

 ; X86-SSE1-LIN-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-SSE1-LIN-NEXT:    addl $20, %esp
 ; X86-SSE1-LIN-NEXT:    retl
 ;
 ; X87-WIN-LABEL: x_to_u32:
 ; X87-WIN:       # %bb.0:
 ; X87-WIN-NEXT:    pushl %ebp
 ; X87-WIN-NEXT:    movl %esp, %ebp
-; X87-WIN-NEXT:    andl $-8, %esp
-; X87-WIN-NEXT:    subl $16, %esp
+; X87-WIN-NEXT:    andl $-16, %esp
+; X87-WIN-NEXT:    subl $32, %esp
 ; X87-WIN-NEXT:    fldt 8(%ebp)
 ; X87-WIN-NEXT:    fnstcw {{[0-9]+}}(%esp)
 ; X87-WIN-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
 ; X87-WIN-NEXT:    orl $3072, %eax # imm = 0xC00
 ; X87-WIN-NEXT:    movw %ax, {{[0-9]+}}(%esp)
 ; X87-WIN-NEXT:    fldcw {{[0-9]+}}(%esp)
 ; X87-WIN-NEXT:    fistpll {{[0-9]+}}(%esp)
 ; X87-WIN-NEXT:    fldcw {{[0-9]+}}(%esp)

 ; X87-LIN-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X87-LIN-NEXT:    addl $20, %esp
 ; X87-LIN-NEXT:    retl
   %r = fptoui x86_fp80 %a to i32
   ret i32 %r
 }
 
 define i32 @x_to_s32(x86_fp80 %a) nounwind {
-; X86-AVX512-LABEL: x_to_s32:
-; X86-AVX512:       # %bb.0:
-; X86-AVX512-NEXT:    pushl %eax
-; X86-AVX512-NEXT:    fldt {{[0-9]+}}(%esp)
-; X86-AVX512-NEXT:    fisttpl (%esp)
-; X86-AVX512-NEXT:    movl (%esp), %eax
-; X86-AVX512-NEXT:    popl %ecx
-; X86-AVX512-NEXT:    retl
+; X86-AVX512-WIN-LABEL: x_to_s32:
+; X86-AVX512-WIN:       # %bb.0:
+; X86-AVX512-WIN-NEXT:    pushl %ebp
+; X86-AVX512-WIN-NEXT:    movl %esp, %ebp
+; X86-AVX512-WIN-NEXT:    andl $-16, %esp
+; X86-AVX512-WIN-NEXT:    subl $16, %esp
+; X86-AVX512-WIN-NEXT:    fldt 8(%ebp)
+; X86-AVX512-WIN-NEXT:    fisttpl {{[0-9]+}}(%esp)
+; X86-AVX512-WIN-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-AVX512-WIN-NEXT:    movl %ebp, %esp
+; X86-AVX512-WIN-NEXT:    popl %ebp
+; X86-AVX512-WIN-NEXT:    retl
+;
+; X86-AVX512-LIN-LABEL: x_to_s32:
+; X86-AVX512-LIN:       # %bb.0:
+; X86-AVX512-LIN-NEXT:    pushl %eax
+; X86-AVX512-LIN-NEXT:    fldt {{[0-9]+}}(%esp)
+; X86-AVX512-LIN-NEXT:    fisttpl (%esp)
+; X86-AVX512-LIN-NEXT:    movl (%esp), %eax
+; X86-AVX512-LIN-NEXT:    popl %ecx
+; X86-AVX512-LIN-NEXT:    retl
 ;
 ; X64-AVX512-WIN-LABEL: x_to_s32:
 ; X64-AVX512-WIN:       # %bb.0:
 ; X64-AVX512-WIN-NEXT:    pushq %rax
 ; X64-AVX512-WIN-NEXT:    fldt (%rcx)
 ; X64-AVX512-WIN-NEXT:    fisttpl {{[0-9]+}}(%rsp)
 ; X64-AVX512-WIN-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; X64-AVX512-WIN-NEXT:    popq %rcx
 ; X64-AVX512-WIN-NEXT:    retq
 ;
 ; X64-AVX512-LIN-LABEL: x_to_s32:
 ; X64-AVX512-LIN:       # %bb.0:
 ; X64-AVX512-LIN-NEXT:    fldt {{[0-9]+}}(%rsp)
 ; X64-AVX512-LIN-NEXT:    fisttpl -{{[0-9]+}}(%rsp)
 ; X64-AVX512-LIN-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
 ; X64-AVX512-LIN-NEXT:    retq
 ;
-; X86-SSE3-LABEL: x_to_s32:
-; X86-SSE3:       # %bb.0:
-; X86-SSE3-NEXT:    pushl %eax
-; X86-SSE3-NEXT:    fldt {{[0-9]+}}(%esp)
-; X86-SSE3-NEXT:    fisttpl (%esp)
-; X86-SSE3-NEXT:    movl (%esp), %eax
-; X86-SSE3-NEXT:    popl %ecx
-; X86-SSE3-NEXT:    retl
+; X86-SSE3-WIN-LABEL: x_to_s32:
+; X86-SSE3-WIN:       # %bb.0:
+; X86-SSE3-WIN-NEXT:    pushl %ebp
+; X86-SSE3-WIN-NEXT:    movl %esp, %ebp
+; X86-SSE3-WIN-NEXT:    andl $-16, %esp
+; X86-SSE3-WIN-NEXT:    subl $16, %esp
+; X86-SSE3-WIN-NEXT:    fldt 8(%ebp)
+; X86-SSE3-WIN-NEXT:    fisttpl {{[0-9]+}}(%esp)
+; X86-SSE3-WIN-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SSE3-WIN-NEXT:    movl %ebp, %esp
+; X86-SSE3-WIN-NEXT:    popl %ebp
+; X86-SSE3-WIN-NEXT:    retl
+;
+; X86-SSE3-LIN-LABEL: x_to_s32:
+; X86-SSE3-LIN:       # %bb.0:
+; X86-SSE3-LIN-NEXT:    pushl %eax
+; X86-SSE3-LIN-NEXT:    fldt {{[0-9]+}}(%esp)
+; X86-SSE3-LIN-NEXT:    fisttpl (%esp)
+; X86-SSE3-LIN-NEXT:    movl (%esp), %eax
+; X86-SSE3-LIN-NEXT:    popl %ecx
+; X86-SSE3-LIN-NEXT:    retl
 ;
 ; X64-SSE3-WIN-LABEL: x_to_s32:
 ; X64-SSE3-WIN:       # %bb.0:
 ; X64-SSE3-WIN-NEXT:    pushq %rax
 ; X64-SSE3-WIN-NEXT:    fldt (%rcx)
 ; X64-SSE3-WIN-NEXT:    fisttpl {{[0-9]+}}(%rsp)
 ; X64-SSE3-WIN-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; X64-SSE3-WIN-NEXT:    popq %rcx
 ; X64-SSE3-WIN-NEXT:    retq
 ;
 ; X64-SSE3-LIN-LABEL: x_to_s32:
 ; X64-SSE3-LIN:       # %bb.0:
 ; X64-SSE3-LIN-NEXT:    fldt {{[0-9]+}}(%rsp)
 ; X64-SSE3-LIN-NEXT:    fisttpl -{{[0-9]+}}(%rsp)
 ; X64-SSE3-LIN-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
 ; X64-SSE3-LIN-NEXT:    retq
 ;
-; X86-SSE2-LABEL: x_to_s32:
-; X86-SSE2:       # %bb.0:
-; X86-SSE2-NEXT:    subl $8, %esp
-; X86-SSE2-NEXT:    fldt {{[0-9]+}}(%esp)
-; X86-SSE2-NEXT:    fnstcw (%esp)
-; X86-SSE2-NEXT:    movzwl (%esp), %eax
-; X86-SSE2-NEXT:    orl $3072, %eax # imm = 0xC00
-; X86-SSE2-NEXT:    movw %ax, {{[0-9]+}}(%esp)
-; X86-SSE2-NEXT:    fldcw {{[0-9]+}}(%esp)
-; X86-SSE2-NEXT:    fistpl {{[0-9]+}}(%esp)
-; X86-SSE2-NEXT:    fldcw (%esp)
-; X86-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SSE2-NEXT:    addl $8, %esp
-; X86-SSE2-NEXT:    retl
+; X86-SSE2-WIN-LABEL: x_to_s32:
+; X86-SSE2-WIN:       # %bb.0:
+; X86-SSE2-WIN-NEXT:    pushl %ebp
+; X86-SSE2-WIN-NEXT:    movl %esp, %ebp
+; X86-SSE2-WIN-NEXT:    andl $-16, %esp
+; X86-SSE2-WIN-NEXT:    subl $16, %esp
+; X86-SSE2-WIN-NEXT:    fldt 8(%ebp)
+; X86-SSE2-WIN-NEXT:    fnstcw (%esp)
+; X86-SSE2-WIN-NEXT:    movzwl (%esp), %eax
+; X86-SSE2-WIN-NEXT:    orl $3072, %eax # imm = 0xC00
+; X86-SSE2-WIN-NEXT:    movw %ax, {{[0-9]+}}(%esp)
+; X86-SSE2-WIN-NEXT:    fldcw {{[0-9]+}}(%esp)
+; X86-SSE2-WIN-NEXT:    fistpl {{[0-9]+}}(%esp)
+; X86-SSE2-WIN-NEXT:    fldcw (%esp)
+; X86-SSE2-WIN-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SSE2-WIN-NEXT:    movl %ebp, %esp
+; X86-SSE2-WIN-NEXT:    popl %ebp
+; X86-SSE2-WIN-NEXT:    retl
+;
+; X86-SSE2-LIN-LABEL: x_to_s32:
+; X86-SSE2-LIN:       # %bb.0:
+; X86-SSE2-LIN-NEXT:    subl $8, %esp
+; X86-SSE2-LIN-NEXT:    fldt {{[0-9]+}}(%esp)
+; X86-SSE2-LIN-NEXT:    fnstcw (%esp)
+; X86-SSE2-LIN-NEXT:    movzwl (%esp), %eax
+; X86-SSE2-LIN-NEXT:    orl $3072, %eax # imm = 0xC00
+; X86-SSE2-LIN-NEXT:    movw %ax, {{[0-9]+}}(%esp)
+; X86-SSE2-LIN-NEXT:    fldcw {{[0-9]+}}(%esp)
+; X86-SSE2-LIN-NEXT:    fistpl {{[0-9]+}}(%esp)
+; X86-SSE2-LIN-NEXT:    fldcw (%esp)
+; X86-SSE2-LIN-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SSE2-LIN-NEXT:    addl $8, %esp
+; X86-SSE2-LIN-NEXT:    retl
 ;
 ; X64-SSE2-WIN-LABEL: x_to_s32:
 ; X64-SSE2-WIN:       # %bb.0:
 ; X64-SSE2-WIN-NEXT:    pushq %rax
 ; X64-SSE2-WIN-NEXT:    fldt (%rcx)
 ; X64-SSE2-WIN-NEXT:    fnstcw (%rsp)
 ; X64-SSE2-WIN-NEXT:    movzwl (%rsp), %eax
 ; X64-SSE2-WIN-NEXT:    orl $3072, %eax # imm = 0xC00

 ; X64-SSE2-LIN-NEXT:    orl $3072, %eax # imm = 0xC00
 ; X64-SSE2-LIN-NEXT:    movw %ax, -{{[0-9]+}}(%rsp)
 ; X64-SSE2-LIN-NEXT:    fldcw -{{[0-9]+}}(%rsp)
 ; X64-SSE2-LIN-NEXT:    fistpl -{{[0-9]+}}(%rsp)
 ; X64-SSE2-LIN-NEXT:    fldcw -{{[0-9]+}}(%rsp)
 ; X64-SSE2-LIN-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
 ; X64-SSE2-LIN-NEXT:    retq
 ;
-; X86-SSE1-LABEL: x_to_s32:
-; X86-SSE1:       # %bb.0:
-; X86-SSE1-NEXT:    subl $8, %esp
-; X86-SSE1-NEXT:    fldt {{[0-9]+}}(%esp)
-; X86-SSE1-NEXT:    fnstcw (%esp)
-; X86-SSE1-NEXT:    movzwl (%esp), %eax
-; X86-SSE1-NEXT:    orl $3072, %eax # imm = 0xC00
-; X86-SSE1-NEXT:    movw %ax, {{[0-9]+}}(%esp)
-; X86-SSE1-NEXT:    fldcw {{[0-9]+}}(%esp)
-; X86-SSE1-NEXT:    fistpl {{[0-9]+}}(%esp)
-; X86-SSE1-NEXT:    fldcw (%esp)
-; X86-SSE1-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SSE1-NEXT:    addl $8, %esp
-; X86-SSE1-NEXT:    retl
+; X86-SSE1-WIN-LABEL: x_to_s32:
+; X86-SSE1-WIN:       # %bb.0:
+; X86-SSE1-WIN-NEXT:    pushl %ebp
+; X86-SSE1-WIN-NEXT:    movl %esp, %ebp
+; X86-SSE1-WIN-NEXT:    andl $-16, %esp
+; X86-SSE1-WIN-NEXT:    subl $16, %esp
+; X86-SSE1-WIN-NEXT:    fldt 8(%ebp)
+; X86-SSE1-WIN-NEXT:    fnstcw (%esp)
+; X86-SSE1-WIN-NEXT:    movzwl (%esp), %eax
+; X86-SSE1-WIN-NEXT:    orl $3072, %eax # imm = 0xC00
+; X86-SSE1-WIN-NEXT:    movw %ax, {{[0-9]+}}(%esp)
+; X86-SSE1-WIN-NEXT:    fldcw {{[0-9]+}}(%esp)
+; X86-SSE1-WIN-NEXT:    fistpl {{[0-9]+}}(%esp)
+; X86-SSE1-WIN-NEXT:    fldcw (%esp)
+; X86-SSE1-WIN-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SSE1-WIN-NEXT:    movl %ebp, %esp
+; X86-SSE1-WIN-NEXT:    popl %ebp
+; X86-SSE1-WIN-NEXT:    retl
 ;
-; X87-LABEL: x_to_s32:
-; X87:       # %bb.0:
-; X87-NEXT:    subl $8, %esp
-; X87-NEXT:    fldt {{[0-9]+}}(%esp)
-; X87-NEXT:    fnstcw (%esp)
-; X87-NEXT:    movzwl (%esp), %eax
-; X87-NEXT:    orl $3072, %eax # imm = 0xC00
-; X87-NEXT:    movw %ax, {{[0-9]+}}(%esp)
-; X87-NEXT:    fldcw {{[0-9]+}}(%esp)
-; X87-NEXT:    fistpl {{[0-9]+}}(%esp)
-; X87-NEXT:    fldcw (%esp)
-; X87-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X87-NEXT:    addl $8, %esp
-; X87-NEXT:    retl
+; X86-SSE1-LIN-LABEL: x_to_s32:
+; X86-SSE1-LIN:       # %bb.0:
+; X86-SSE1-LIN-NEXT:    subl $8, %esp
+; X86-SSE1-LIN-NEXT:    fldt {{[0-9]+}}(%esp)
+; X86-SSE1-LIN-NEXT:    fnstcw (%esp)
+; X86-SSE1-LIN-NEXT:    movzwl (%esp), %eax
+; X86-SSE1-LIN-NEXT:    orl $3072, %eax # imm = 0xC00
+; X86-SSE1-LIN-NEXT:    movw %ax, {{[0-9]+}}(%esp)
+; X86-SSE1-LIN-NEXT:    fldcw {{[0-9]+}}(%esp)
+; X86-SSE1-LIN-NEXT:    fistpl {{[0-9]+}}(%esp)
+; X86-SSE1-LIN-NEXT:    fldcw (%esp)
+; X86-SSE1-LIN-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SSE1-LIN-NEXT:    addl $8, %esp
+; X86-SSE1-LIN-NEXT:    retl
+;
+; X87-WIN-LABEL: x_to_s32:
+; X87-WIN:       # %bb.0:
+; X87-WIN-NEXT:    pushl %ebp
+; X87-WIN-NEXT:    movl %esp, %ebp
+; X87-WIN-NEXT:    andl $-16, %esp
+; X87-WIN-NEXT:    subl $16, %esp
+; X87-WIN-NEXT:    fldt 8(%ebp)
+; X87-WIN-NEXT:    fnstcw (%esp)
+; X87-WIN-NEXT:    movzwl (%esp), %eax
+; X87-WIN-NEXT:    orl $3072, %eax # imm = 0xC00
+; X87-WIN-NEXT:    movw %ax, {{[0-9]+}}(%esp)
+; X87-WIN-NEXT:    fldcw {{[0-9]+}}(%esp)
+; X87-WIN-NEXT:    fistpl {{[0-9]+}}(%esp)
+; X87-WIN-NEXT:    fldcw (%esp)
+; X87-WIN-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X87-WIN-NEXT:    movl %ebp, %esp
+; X87-WIN-NEXT:    popl %ebp
+; X87-WIN-NEXT:    retl
+;
+; X87-LIN-LABEL: x_to_s32:
+; X87-LIN:       # %bb.0:
+; X87-LIN-NEXT:    subl $8, %esp
+; X87-LIN-NEXT:    fldt {{[0-9]+}}(%esp)
+; X87-LIN-NEXT:    fnstcw (%esp)
+; X87-LIN-NEXT:    movzwl (%esp), %eax
+; X87-LIN-NEXT:    orl $3072, %eax # imm = 0xC00
+; X87-LIN-NEXT:    movw %ax, {{[0-9]+}}(%esp)
+; X87-LIN-NEXT:    fldcw {{[0-9]+}}(%esp)
+; X87-LIN-NEXT:    fistpl {{[0-9]+}}(%esp)
+; X87-LIN-NEXT:    fldcw (%esp)
+; X87-LIN-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X87-LIN-NEXT:    addl $8, %esp
+; X87-LIN-NEXT:    retl
   %r = fptosi x86_fp80 %a to i32
   ret i32 %r
 }
 
 define i32 @t_to_u32(fp128 %a) nounwind {
 ; X86-AVX512-WIN-LABEL: t_to_u32:
 ; X86-AVX512-WIN:       # %bb.0:
 ; X86-AVX512-WIN-NEXT:    subl $16, %esp

llvm/test/CodeGen/X86/scalar-fp-to-i64.ll

; X87-LIN-NEXT:    retl
   ret i64 %r
 }
 
 define i64 @x_to_u64(x86_fp80 %a) nounwind {
 ; X86-AVX512-WIN-LABEL: x_to_u64:
 ; X86-AVX512-WIN:       # %bb.0:
 ; X86-AVX512-WIN-NEXT:    pushl %ebp
 ; X86-AVX512-WIN-NEXT:    movl %esp, %ebp
-; X86-AVX512-WIN-NEXT:    andl $-8, %esp
-; X86-AVX512-WIN-NEXT:    subl $8, %esp
+; X86-AVX512-WIN-NEXT:    andl $-16, %esp
+; X86-AVX512-WIN-NEXT:    subl $16, %esp
 ; X86-AVX512-WIN-NEXT:    fldt 8(%ebp)
 ; X86-AVX512-WIN-NEXT:    flds __real@5f000000
 ; X86-AVX512-WIN-NEXT:    xorl %edx, %edx
 ; X86-AVX512-WIN-NEXT:    fucomi %st(1), %st
 ; X86-AVX512-WIN-NEXT:    fldz
 ; X86-AVX512-WIN-NEXT:    fcmovbe %st(1), %st
 ; X86-AVX512-WIN-NEXT:    fstp %st(1)
 ; X86-AVX512-WIN-NEXT:    fsubrp %st, %st(1)

 ; X64-AVX512-LIN-NEXT:    shlq $63, %rax
 ; X64-AVX512-LIN-NEXT:    xorq -{{[0-9]+}}(%rsp), %rax
 ; X64-AVX512-LIN-NEXT:    retq
 ;
 ; X86-SSE3-WIN-LABEL: x_to_u64:
 ; X86-SSE3-WIN:       # %bb.0:
 ; X86-SSE3-WIN-NEXT:    pushl %ebp
 ; X86-SSE3-WIN-NEXT:    movl %esp, %ebp
-; X86-SSE3-WIN-NEXT:    andl $-8, %esp
-; X86-SSE3-WIN-NEXT:    subl $8, %esp
+; X86-SSE3-WIN-NEXT:    andl $-16, %esp
+; X86-SSE3-WIN-NEXT:    subl $16, %esp
 ; X86-SSE3-WIN-NEXT:    fldt 8(%ebp)
 ; X86-SSE3-WIN-NEXT:    flds __real@5f000000
 ; X86-SSE3-WIN-NEXT:    xorl %edx, %edx
 ; X86-SSE3-WIN-NEXT:    fucomi %st(1), %st
 ; X86-SSE3-WIN-NEXT:    fldz
 ; X86-SSE3-WIN-NEXT:    fcmovbe %st(1), %st
 ; X86-SSE3-WIN-NEXT:    fstp %st(1)
 ; X86-SSE3-WIN-NEXT:    fsubrp %st, %st(1)

 ; X64-SSE3-LIN-NEXT:    shlq $63, %rax
 ; X64-SSE3-LIN-NEXT:    xorq -{{[0-9]+}}(%rsp), %rax
 ; X64-SSE3-LIN-NEXT:    retq
 ;
 ; X86-SSE2-WIN-LABEL: x_to_u64:
 ; X86-SSE2-WIN:       # %bb.0:
 ; X86-SSE2-WIN-NEXT:    pushl %ebp
 ; X86-SSE2-WIN-NEXT:    movl %esp, %ebp
-; X86-SSE2-WIN-NEXT:    andl $-8, %esp
-; X86-SSE2-WIN-NEXT:    subl $16, %esp
+; X86-SSE2-WIN-NEXT:    andl $-16, %esp
+; X86-SSE2-WIN-NEXT:    subl $32, %esp
 ; X86-SSE2-WIN-NEXT:    fldt 8(%ebp)
 ; X86-SSE2-WIN-NEXT:    flds __real@5f000000
 ; X86-SSE2-WIN-NEXT:    xorl %edx, %edx
 ; X86-SSE2-WIN-NEXT:    fucomi %st(1), %st
 ; X86-SSE2-WIN-NEXT:    setbe %dl
 ; X86-SSE2-WIN-NEXT:    fldz
 ; X86-SSE2-WIN-NEXT:    fcmovbe %st(1), %st
 ; X86-SSE2-WIN-NEXT:    fstp %st(1)

 ; X64-SSE2-LIN-NEXT:    shlq $63, %rax
 ; X64-SSE2-LIN-NEXT:    xorq -{{[0-9]+}}(%rsp), %rax
 ; X64-SSE2-LIN-NEXT:    retq
 ;
 ; X87-WIN-LABEL: x_to_u64:
 ; X87-WIN:       # %bb.0:
 ; X87-WIN-NEXT:    pushl %ebp
 ; X87-WIN-NEXT:    movl %esp, %ebp
-; X87-WIN-NEXT:    andl $-8, %esp
-; X87-WIN-NEXT:    subl $16, %esp
+; X87-WIN-NEXT:    andl $-16, %esp
+; X87-WIN-NEXT:    subl $32, %esp
 ; X87-WIN-NEXT:    fldt 8(%ebp)
 ; X87-WIN-NEXT:    flds __real@5f000000
 ; X87-WIN-NEXT:    fucom %st(1)
 ; X87-WIN-NEXT:    fnstsw %ax
 ; X87-WIN-NEXT:    xorl %edx, %edx
 ; X87-WIN-NEXT:    # kill: def $ah killed $ah killed $ax
 ; X87-WIN-NEXT:    sahf
 ; X87-WIN-NEXT:    setbe %al
; X87-LIN-NEXT:    retl
   ret i64 %r
 }
 
 define i64 @x_to_s64(x86_fp80 %a) nounwind {
 ; X86-AVX512-WIN-LABEL: x_to_s64:
 ; X86-AVX512-WIN:       # %bb.0:
 ; X86-AVX512-WIN-NEXT:    pushl %ebp
 ; X86-AVX512-WIN-NEXT:    movl %esp, %ebp
-; X86-AVX512-WIN-NEXT:    andl $-8, %esp
-; X86-AVX512-WIN-NEXT:    subl $8, %esp
+; X86-AVX512-WIN-NEXT:    andl $-16, %esp
+; X86-AVX512-WIN-NEXT:    subl $16, %esp
 ; X86-AVX512-WIN-NEXT:    fldt 8(%ebp)
 ; X86-AVX512-WIN-NEXT:    fisttpll (%esp)
 ; X86-AVX512-WIN-NEXT:    movl (%esp), %eax
 ; X86-AVX512-WIN-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-AVX512-WIN-NEXT:    movl %ebp, %esp
 ; X86-AVX512-WIN-NEXT:    popl %ebp
 ; X86-AVX512-WIN-NEXT:    retl
 ;

 ; X64-AVX512-LIN-NEXT:    fisttpll -{{[0-9]+}}(%rsp)
 ; X64-AVX512-LIN-NEXT:    movq -{{[0-9]+}}(%rsp), %rax
 ; X64-AVX512-LIN-NEXT:    retq
 ;
 ; X86-SSE3-WIN-LABEL: x_to_s64:
 ; X86-SSE3-WIN:       # %bb.0:
 ; X86-SSE3-WIN-NEXT:    pushl %ebp
 ; X86-SSE3-WIN-NEXT:    movl %esp, %ebp
-; X86-SSE3-WIN-NEXT:    andl $-8, %esp
-; X86-SSE3-WIN-NEXT:    subl $8, %esp
+; X86-SSE3-WIN-NEXT:    andl $-16, %esp
+; X86-SSE3-WIN-NEXT:    subl $16, %esp
 ; X86-SSE3-WIN-NEXT:    fldt 8(%ebp)
 ; X86-SSE3-WIN-NEXT:    fisttpll (%esp)
 ; X86-SSE3-WIN-NEXT:    movl (%esp), %eax
 ; X86-SSE3-WIN-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-SSE3-WIN-NEXT:    movl %ebp, %esp
 ; X86-SSE3-WIN-NEXT:    popl %ebp
 ; X86-SSE3-WIN-NEXT:    retl
 ;

 ; X64-SSE3-LIN-NEXT:    fisttpll -{{[0-9]+}}(%rsp)
 ; X64-SSE3-LIN-NEXT:    movq -{{[0-9]+}}(%rsp), %rax
 ; X64-SSE3-LIN-NEXT:    retq
 ;
 ; X86-SSE2-WIN-LABEL: x_to_s64:
 ; X86-SSE2-WIN:       # %bb.0:
 ; X86-SSE2-WIN-NEXT:    pushl %ebp
 ; X86-SSE2-WIN-NEXT:    movl %esp, %ebp
-; X86-SSE2-WIN-NEXT:    andl $-8, %esp
-; X86-SSE2-WIN-NEXT:    subl $16, %esp
+; X86-SSE2-WIN-NEXT:    andl $-16, %esp
+; X86-SSE2-WIN-NEXT:    subl $32, %esp
 ; X86-SSE2-WIN-NEXT:    fldt 8(%ebp)
 ; X86-SSE2-WIN-NEXT:    fnstcw {{[0-9]+}}(%esp)
 ; X86-SSE2-WIN-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
 ; X86-SSE2-WIN-NEXT:    orl $3072, %eax # imm = 0xC00
 ; X86-SSE2-WIN-NEXT:    movw %ax, {{[0-9]+}}(%esp)
 ; X86-SSE2-WIN-NEXT:    fldcw {{[0-9]+}}(%esp)
 ; X86-SSE2-WIN-NEXT:    fistpll {{[0-9]+}}(%esp)
 ; X86-SSE2-WIN-NEXT:    fldcw {{[0-9]+}}(%esp)

 ; X64-SSE2-LIN-NEXT:    fldcw -{{[0-9]+}}(%rsp)
 ; X64-SSE2-LIN-NEXT:    movq -{{[0-9]+}}(%rsp), %rax
 ; X64-SSE2-LIN-NEXT:    retq
 ;
 ; X87-WIN-LABEL: x_to_s64:
 ; X87-WIN:       # %bb.0:
 ; X87-WIN-NEXT:    pushl %ebp
 ; X87-WIN-NEXT:    movl %esp, %ebp
-; X87-WIN-NEXT:    andl $-8, %esp
-; X87-WIN-NEXT:    subl $16, %esp
+; X87-WIN-NEXT:    andl $-16, %esp
+; X87-WIN-NEXT:    subl $32, %esp
 ; X87-WIN-NEXT:    fldt 8(%ebp)
 ; X87-WIN-NEXT:    fnstcw {{[0-9]+}}(%esp)
 ; X87-WIN-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
 ; X87-WIN-NEXT:    orl $3072, %eax # imm = 0xC00
 ; X87-WIN-NEXT:    movw %ax, {{[0-9]+}}(%esp)
 ; X87-WIN-NEXT:    fldcw {{[0-9]+}}(%esp)
 ; X87-WIN-NEXT:    fistpll {{[0-9]+}}(%esp)
 ; X87-WIN-NEXT:    fldcw {{[0-9]+}}(%esp)