RPCSX/llvm
1
0
Fork 0
mirror of https://github.com/RPCSX/llvm.git synced 2025-02-25 15:41:05 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Avoid unnecessary stack realignment in musttail thunks with SSE2 enabled

Browse Source 
The X86 musttail implementation finds register parameters to forward by
running the calling convention algorithm until a non-register location
is returned. However, assigning a vector memory location has the side
effect of increasing the function's stack alignment. We shouldn't
increase the stack alignment when we are only looking for register
parameters, so this change conditionalizes it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@258442 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Reid Kleckner 2016-01-21 22:23:22 +00:00
parent 180abd8ba0
commit 1264f61388
3 changed files with 12 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
include/llvm/CodeGen/CallingConvLower.h
View File

@ -195,6 +195,7 @@ class CCState {
private:
CallingConv::ID CallingConv;
bool IsVarArg;
bool AnalyzingMustTailForwardedRegs = false;
MachineFunction &MF;
const TargetRegisterInfo &TRI;
SmallVectorImpl<CCValAssign> &Locs;
@ -416,10 +417,15 @@ public:
unsigned Result = StackOffset;
StackOffset += Size;
MaxStackArgAlign = std::max(Align, MaxStackArgAlign);
MF.getFrameInfo()->ensureMaxAlignment(Align);
ensureMaxAlignment(Align);
return Result;
}
void ensureMaxAlignment(unsigned Align) {
if (!AnalyzingMustTailForwardedRegs)
MF.getFrameInfo()->ensureMaxAlignment(Align);
}
/// Version of AllocateStack with extra register to be shadowed.
unsigned AllocateStack(unsigned Size, unsigned Align, unsigned ShadowReg) {
MarkAllocated(ShadowReg);

3
lib/CodeGen/CallingConvLower.cpp
View File

@ -51,7 +51,7 @@ void CCState::HandleByVal(unsigned ValNo, MVT ValVT,
Size = MinSize;
if (MinAlign > (int)Align)
Align = MinAlign;
MF.getFrameInfo()->ensureMaxAlignment(Align);
ensureMaxAlignment(Align);
MF.getSubtarget().getTargetLowering()->HandleByVal(this, Size, Align);
Size = unsigned(alignTo(Size, MinAlign));
unsigned Offset = AllocateStack(Size, Align);
@ -236,6 +236,7 @@ void CCState::analyzeMustTailForwardedRegisters(
// variadic functions, so we need to assume we're not variadic so that we get
// all the registers that might be used in a non-variadic call.
SaveAndRestore<bool> SavedVarArg(IsVarArg, false);
SaveAndRestore<bool> SavedMustTail(AnalyzingMustTailForwardedRegs, true);
for (MVT RegVT : RegParmTypes) {
SmallVector<MCPhysReg, 8> RemainingRegs;

3
test/CodeGen/X86/musttail-varargs.ll
View File

@ -2,6 +2,7 @@
; RUN: llc < %s -enable-tail-merge=0 -mtriple=x86_64-linux-gnux32 | FileCheck %s --check-prefix=LINUX-X32
; RUN: llc < %s -enable-tail-merge=0 -mtriple=x86_64-windows | FileCheck %s --check-prefix=WINDOWS
; RUN: llc < %s -enable-tail-merge=0 -mtriple=i686-windows | FileCheck %s --check-prefix=X86
; RUN: llc < %s -enable-tail-merge=0 -mtriple=i686-windows -mattr=+sse2 | FileCheck %s --check-prefix=X86
; Test that we actually spill and reload all arguments in the variadic argument
; pack. Doing a normal call will clobber all argument registers, and we will
@ -136,6 +137,8 @@ define void @g_thunk(i8* %fptr_i8, ...) {
; WINDOWS: jmpq *%rcx # TAILCALL
; X86-LABEL: _g_thunk:
; X86-NOT: push %ebp
; X86-NOT: andl {{.*}}, %esp
; X86: jmpl *%eax # TAILCALL
; Do a simple multi-exit multi-bb test.