maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

Revert "Re-commit r265039 "[X86] Merge adjacent stack adjustments in eliminateCallFramePseudoInstr (PR27140)"" 

It seems to be causing ASan tests to crash, probably due to
miscompiling the run-time somehow.

llvm-svn: 265551
This commit is contained in:
Hans Wennborg 2016-04-06 16:10:20 +00:00
parent bd90c60afb
commit a7e396b5ef
9 changed files with 32 additions and 103 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

39
llvm/lib/Target/X86/X86FrameLowering.cpp
View File

@ -378,16 +378,11 @@ int X86FrameLowering::mergeSPUpdates(MachineBasicBlock &MBB,
if ((Opc == X86::ADD64ri32 || Opc == X86::ADD64ri8 ||
Opc == X86::ADD32ri || Opc == X86::ADD32ri8) &&
PI->getOperand(0).getReg() == StackPtr){
assert(PI->getOperand(1).getReg() == StackPtr);
Offset += PI->getOperand(2).getImm();
MBB.erase(PI);
if (!doMergeWithPrevious) MBBI = NI;
} else if ((Opc == X86::LEA32r || Opc == X86::LEA64_32r) &&
PI->getOperand(0).getReg() == StackPtr &&
PI->getOperand(1).getReg() == StackPtr &&
PI->getOperand(2).getImm() == 1 &&
PI->getOperand(3).getReg() == X86::NoRegister &&
PI->getOperand(5).getReg() == X86::NoRegister) {
PI->getOperand(0).getReg() == StackPtr) {
// For LEAs we have: def = lea SP, FI, noreg, Offset, noreg.
Offset += PI->getOperand(4).getImm();
MBB.erase(PI);
@ -395,7 +390,6 @@ int X86FrameLowering::mergeSPUpdates(MachineBasicBlock &MBB,
} else if ((Opc == X86::SUB64ri32 || Opc == X86::SUB64ri8 ||
Opc == X86::SUB32ri || Opc == X86::SUB32ri8) &&
PI->getOperand(0).getReg() == StackPtr) {
assert(PI->getOperand(1).getReg() == StackPtr);
Offset -= PI->getOperand(2).getImm();
MBB.erase(PI);
if (!doMergeWithPrevious) MBBI = NI;
@ -2539,22 +2533,13 @@ eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
BuildCFI(MBB, I, DL,
MCCFIInstruction::createAdjustCfaOffset(nullptr, -InternalAmt));
// Add Amount to SP to destroy a frame, or subtract to setup.
int64_t StackAdjustment = isDestroy ? Amount : -Amount;
if (Amount) {
// Add Amount to SP to destroy a frame, and subtract to setup.
int Offset = isDestroy ? Amount : -Amount;
if (StackAdjustment) {
// Merge with any previous or following adjustment instruction.
StackAdjustment += mergeSPUpdates(MBB, I, true);
StackAdjustment += mergeSPUpdates(MBB, I, false);
if (!StackAdjustment) {
// This and the merged instruction canceled out each other.
return I;
}
if (!(Fn->optForMinSize() &&
adjustStackWithPops(MBB, I, DL, StackAdjustment)))
BuildStackAdjustment(MBB, I, DL, StackAdjustment, /*InEpilogue=*/false);
if (!(Fn->optForMinSize() &&
adjustStackWithPops(MBB, I, DL, Offset)))
BuildStackAdjustment(MBB, I, DL, Offset, /*InEpilogue=*/false);
}
if (DwarfCFI && !hasFP(MF)) {
@ -2564,12 +2549,14 @@ eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
// CFI only for EH purposes or for debugging. EH only requires the CFA
// offset to be correct at each call site, while for debugging we want
// it to be more precise.
int CFAOffset = Amount;
// TODO: When not using precise CFA, we also need to adjust for the
// InternalAmt here.
if (StackAdjustment) {
BuildCFI(MBB, I, DL, MCCFIInstruction::createAdjustCfaOffset(
nullptr, -StackAdjustment));
if (CFAOffset) {
CFAOffset = isDestroy ? -CFAOffset : CFAOffset;
BuildCFI(MBB, I, DL,
MCCFIInstruction::createAdjustCfaOffset(nullptr, CFAOffset));
}
}

2
llvm/test/CodeGen/X86/2006-05-02-InstrSched1.ll
View File

@ -1,6 +1,6 @@
; REQUIRES: asserts
; RUN: llc < %s -mtriple=i686-unknown-linux -relocation-model=static -stats 2>&1 | \
; RUN: grep asm-printer | grep 14
; RUN: grep asm-printer | grep 15
;
; It's possible to schedule this in 14 instructions by avoiding
; callee-save registers, but the scheduler isn't currently that

2
llvm/test/CodeGen/X86/fold-push.ll
View File

@ -14,7 +14,7 @@ define void @test(i32 %a, i32 %b) optsize nounwind {
; SLM: movl (%esp), [[RELOAD:%e..]]
; SLM-NEXT: pushl [[RELOAD]]
; CHECK: calll
; CHECK-NEXT: addl $8, %esp
; CHECK-NEXT: addl $4, %esp
%c = add i32 %a, %b
call void @foo(i32 %c)
call void asm sideeffect "nop", "~{ax},~{bx},~{cx},~{dx},~{bp},~{si},~{di}"()

14
llvm/test/CodeGen/X86/force-align-stack-alloca.ll
View File

@ -32,21 +32,15 @@ define i64 @g(i32 %i) nounwind {
; CHECK: movl %{{...}}, %esp
; CHECK-NOT: {{[^ ,]*}}, %esp
;
; Next we set up the memset call.
; Next we set up the memset call, and then undo it.
; CHECK: subl $20, %esp
; CHECK-NOT: {{[^ ,]*}}, %esp
; CHECK: pushl
; CHECK: pushl
; CHECK: pushl
; CHECK: calll memset
;
; Deallocating 32 bytes of outgoing call frame for memset and
; allocating 28 bytes for calling f yields a 4-byte adjustment:
; CHECK-NEXT: addl $4, %esp
; CHECK-NEXT: addl $32, %esp
; CHECK-NOT: {{[^ ,]*}}, %esp
;
; And move on to call 'f', and then restore the stack.
; CHECK: pushl
; Next we set up the call to 'f'.
; CHECK: subl $28, %esp
; CHECK-NOT: {{[^ ,]*}}, %esp
; CHECK: calll f
; CHECK-NEXT: addl $32, %esp

3
llvm/test/CodeGen/X86/localescape.ll
View File

@ -137,5 +137,6 @@ define void @alloc_func_no_frameaddr() {
; X86: movl $13, (%esp)
; X86: pushl $0
; X86: calll _print_framealloc_from_fp
; X86: addl $12, %esp
; X86: addl $4, %esp
; X86: addl $8, %esp
; X86: retl

3
llvm/test/CodeGen/X86/memset-2.ll
View File

@ -6,7 +6,8 @@ declare void @llvm.memset.i32(i8*, i8, i32, i32) nounwind
define fastcc void @t1() nounwind {
; CHECK-LABEL: t1:
; CHECK: ## BB#0: ## %entry
; CHECK-NEXT: subl $16, %esp
; CHECK-NEXT: subl $12, %esp
; CHECK-NEXT: subl $4, %esp
; CHECK-NEXT: pushl $188
; CHECK-NEXT: pushl $0
; CHECK-NEXT: pushl $0

64
llvm/test/CodeGen/X86/movtopush.ll
View File

@ -2,7 +2,6 @@
; RUN: llc < %s -mtriple=i686-windows -no-x86-call-frame-opt | FileCheck %s -check-prefix=NOPUSH
; RUN: llc < %s -mtriple=x86_64-windows | FileCheck %s -check-prefix=X64
; RUN: llc < %s -mtriple=i686-windows -stackrealign -stack-alignment=32 | FileCheck %s -check-prefix=ALIGNED
; RUN: llc < %s -mtriple=i686-pc-linux | FileCheck %s -check-prefix=LINUX
%class.Class = type { i32 }
%struct.s = type { i64 }
@ -13,10 +12,6 @@ declare x86_thiscallcc void @thiscall(%class.Class* %class, i32 %a, i32 %b, i32
declare void @oneparam(i32 %a)
declare void @eightparams(i32 %a, i32 %b, i32 %c, i32 %d, i32 %e, i32 %f, i32 %g, i32 %h)
declare void @struct(%struct.s* byval %a, i32 %b, i32 %c, i32 %d)
declare void @inalloca(<{ %struct.s }>* inalloca)
declare i8* @llvm.stacksave()
declare void @llvm.stackrestore(i8*)
; We should get pushes for x86, even though there is a reserved call frame.
; Make sure we don't touch x86-64, and that turning it off works.
@ -228,7 +223,8 @@ entry:
; NORMAL-NEXT: pushl $2
; NORMAL-NEXT: pushl $1
; NORMAL-NEXT: call
; NORMAL-NEXT: subl $4, %esp
; NORMAL-NEXT: addl $16, %esp
; NORMAL-NEXT: subl $20, %esp
; NORMAL-NEXT: movl 20(%esp), [[E1:%e..]]
; NORMAL-NEXT: movl 24(%esp), [[E2:%e..]]
; NORMAL-NEXT: movl [[E2]], 4(%esp)
@ -265,7 +261,7 @@ entry:
; NORMAL-NEXT: pushl $2
; NORMAL-NEXT: pushl $1
; NORMAL-NEXT: calll *16(%esp)
; NORMAL-NEXT: addl $24, %esp
; NORMAL-NEXT: addl $16, %esp
define void @test10() optsize {
%stack_fptr = alloca void (i32, i32, i32, i32)*
store void (i32, i32, i32, i32)* @good, void (i32, i32, i32, i32)** %stack_fptr
@ -318,7 +314,8 @@ entry:
; NORMAL-NEXT: pushl $2
; NORMAL-NEXT: pushl $1
; NORMAL-NEXT: calll _good
; NORMAL-NEXT: subl $4, %esp
; NORMAL-NEXT: addl $16, %esp
; NORMAL-NEXT: subl $20, %esp
; NORMAL: movl $8, 16(%esp)
; NORMAL-NEXT: movl $7, 12(%esp)
; NORMAL-NEXT: movl $6, 8(%esp)
@ -361,54 +358,3 @@ entry:
call void @good(i32 %val1, i32 %val2, i32 %val3, i32 %add)
ret i32* %ptr3
}
; Make sure to fold adjacent stack adjustments.
; LINUX-LABEL: pr27140:
; LINUX: subl $12, %esp
; LINUX: .cfi_def_cfa_offset 16
; LINUX-NOT: sub
; LINUX: pushl $4
; LINUX: .cfi_adjust_cfa_offset 4
; LINUX: pushl $3
; LINUX: .cfi_adjust_cfa_offset 4
; LINUX: pushl $2
; LINUX: .cfi_adjust_cfa_offset 4
; LINUX: pushl $1
; LINUX: .cfi_adjust_cfa_offset 4
; LINUX: calll good
; LINUX: addl $28, %esp
; LINUX: .cfi_adjust_cfa_offset -28
; LINUX-NOT: add
; LINUX: retl
define void @pr27140() optsize {
entry:
tail call void @good(i32 1, i32 2, i32 3, i32 4)
ret void
}
; Check that a stack restore (leal -4(%ebp), %esp) doesn't get merged with a
; stack adjustment (addl $12, %esp). Just because it's a lea doesn't mean it's
; simply decreasing the stack pointer.
; NORMAL-LABEL: test14:
; NORMAL: calll _B_func
; NORMAL: leal -4(%ebp), %esp
; NORMAL-NOT: %esp
; NORMAL: retl
%struct.A = type { i32, i32 }
%struct.B = type { i8 }
declare x86_thiscallcc %struct.B* @B_ctor(%struct.B* returned, %struct.A* byval)
declare void @B_func(%struct.B* sret, %struct.B*, i32)
define void @test14(%struct.A* %a) {
entry:
%ref.tmp = alloca %struct.B, align 1
%agg.tmp = alloca i64, align 4
%tmpcast = bitcast i64* %agg.tmp to %struct.A*
%tmp = alloca %struct.B, align 1
%0 = bitcast %struct.A* %a to i64*
%1 = load i64, i64* %0, align 4
store i64 %1, i64* %agg.tmp, align 4
%call = call x86_thiscallcc %struct.B* @B_ctor(%struct.B* %ref.tmp, %struct.A* byval %tmpcast)
%2 = getelementptr inbounds %struct.B, %struct.B* %tmp, i32 0, i32 0
call void @B_func(%struct.B* sret %tmp, %struct.B* %ref.tmp, i32 1)
ret void
}

4
llvm/test/CodeGen/X86/push-cfi-debug.ll
View File

@ -23,8 +23,8 @@ declare x86_stdcallcc void @stdfoo(i32, i32) #0
; CHECK: .cfi_adjust_cfa_offset 4
; CHECK: calll stdfoo
; CHECK: .cfi_adjust_cfa_offset -8
; CHECK: addl $20, %esp
; CHECK: .cfi_adjust_cfa_offset -20
; CHECK: addl $8, %esp
; CHECK: .cfi_adjust_cfa_offset -8
define void @test1() #0 !dbg !4 {
entry:
tail call void @foo(i32 1, i32 2) #1, !dbg !10

4
llvm/test/CodeGen/X86/push-cfi.ll
View File

@ -82,8 +82,8 @@ cleanup:
; LINUX-NEXT: Ltmp{{[0-9]+}}:
; LINUX-NEXT: .cfi_adjust_cfa_offset 4
; LINUX-NEXT: call
; LINUX-NEXT: addl $28, %esp
; LINUX: .cfi_adjust_cfa_offset -28
; LINUX-NEXT: addl $16, %esp
; LINUX: .cfi_adjust_cfa_offset -16
; DARWIN-NOT: .cfi_escape
; DARWIN-NOT: pushl
define void @test2_nofp() #0 personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {