This is an archive of the discontinued LLVM Phabricator instance.

Differential D133946

[CodeGen] Don't zero callee-save registers with zero-call-used-regs (PR57692)
ClosedPublic

Authored by nikic on Sep 15 2022, 8:07 AM.

Details

Reviewers
void
nickdesaulniers
efriedma
craig.topper
MaskRay
Commits
rGb4309800e9dc: [CodeGen] Don't zero callee-save registers with zero-call-used-regs (PR57692)
Summary

Callee save registers must be preserved, so -fzero-call-used-regs should not be zeroing them. The previous implementation only did not zero callee save registers that were saved&restored inside the function, but we need preserve all of them

Fixes https://github.com/llvm/llvm-project/issues/57692.

Diff Detail

Unit TestsFailed

TimeTest
90 msx64 debian > lld.MachO::alias-symbols.s

Event Timeline

nikic created this revision.Sep 15 2022, 8:07 AM
Herald added a project: Restricted Project. · View Herald TranscriptSep 15 2022, 8:07 AM
Herald added subscribers: StephenFan, pengfei, hiraditya. · View Herald Transcript
nikic requested review of this revision.Sep 15 2022, 8:07 AM
Herald added a project: Restricted Project. · View Herald TranscriptSep 15 2022, 8:07 AM
Herald added a subscriber: llvm-commits. · View Herald Transcript
Harbormaster completed remote builds in B186868: Diff 460415.Sep 15 2022, 9:10 AM
MaskRay accepted this revision.Sep 15 2022, 10:02 AM
This revision is now accepted and ready to land.Sep 15 2022, 10:02 AM
This revision was landed with ongoing or failed builds.Sep 16 2022, 2:52 AM
Closed by commit rGb4309800e9dc: [CodeGen] Don't zero callee-save registers with zero-call-used-regs (PR57692) (authored by nikic). · Explain Why
This revision was automatically updated to reflect the committed changes.
nikic added a commit: rGb4309800e9dc: [CodeGen] Don't zero callee-save registers with zero-call-used-regs (PR57692).
nickdesaulniers mentioned this in D110869: [X86] Implement -fzero-call-used-regs option.Sep 16 2022, 5:42 AM

Revision Contents

PathSize
llvm/
lib/
CodeGen/
7 lines
test/
CodeGen/
AArch64/
32 lines
X86/
20 lines

Diff 460415

llvm/lib/CodeGen/PrologEpilogInserter.cpp

Show First 20 Lines • Show All 1,267 Lines • ▼ Show 20 Lines for (MachineBasicBlock::const_iterator I = MBBI, E = MBB.end(); I != E;
for (const MCPhysReg &Reg : for (const MCPhysReg &Reg :
TRI.sub_and_superregs_inclusive(MO.getReg())) TRI.sub_and_superregs_inclusive(MO.getReg()))
RegsToZero.reset(Reg); RegsToZero.reset(Reg);
} }
} }
} }
// Don't clear registers that are reset before exiting. // Don't clear registers that must be preserved.
for (const CalleeSavedInfo &CSI : MF.getFrameInfo().getCalleeSavedInfo()) for (const MCPhysReg *CSRegs = TRI.getCalleeSavedRegs(&MF);
for (MCRegister Reg : TRI.sub_and_superregs_inclusive(CSI.getReg())) MCPhysReg CSReg = *CSRegs; ++CSRegs)
for (MCRegister Reg : TRI.sub_and_superregs_inclusive(CSReg))
RegsToZero.reset(Reg); RegsToZero.reset(Reg);
const TargetFrameLowering &TFI = *MF.getSubtarget().getFrameLowering(); const TargetFrameLowering &TFI = *MF.getSubtarget().getFrameLowering();
for (MachineBasicBlock &MBB : MF) for (MachineBasicBlock &MBB : MF)
if (MBB.isReturnBlock()) if (MBB.isReturnBlock())
TFI.emitZeroCallUsedRegs(RegsToZero, MBB); TFI.emitZeroCallUsedRegs(RegsToZero, MBB);
} }
▲ Show 20 Lines • Show All 194 Lines • Show Last 20 Lines

llvm/test/CodeGen/AArch64/zero-call-used-regs.ll

Show First 20 Lines • Show All 214 Lines • ▼ Show 20 Lines
; DEFAULT-NEXT: movi v0.2d, #0000000000000000 ; DEFAULT-NEXT: movi v0.2d, #0000000000000000
; DEFAULT-NEXT: movi v1.2d, #0000000000000000 ; DEFAULT-NEXT: movi v1.2d, #0000000000000000
; DEFAULT-NEXT: movi v2.2d, #0000000000000000 ; DEFAULT-NEXT: movi v2.2d, #0000000000000000
; DEFAULT-NEXT: movi v3.2d, #0000000000000000 ; DEFAULT-NEXT: movi v3.2d, #0000000000000000
; DEFAULT-NEXT: movi v4.2d, #0000000000000000 ; DEFAULT-NEXT: movi v4.2d, #0000000000000000
; DEFAULT-NEXT: movi v5.2d, #0000000000000000 ; DEFAULT-NEXT: movi v5.2d, #0000000000000000
; DEFAULT-NEXT: movi v6.2d, #0000000000000000 ; DEFAULT-NEXT: movi v6.2d, #0000000000000000
; DEFAULT-NEXT: movi v7.2d, #0000000000000000 ; DEFAULT-NEXT: movi v7.2d, #0000000000000000
; DEFAULT-NEXT: movi v8.2d, #0000000000000000
; DEFAULT-NEXT: movi v9.2d, #0000000000000000
; DEFAULT-NEXT: movi v10.2d, #0000000000000000
; DEFAULT-NEXT: movi v11.2d, #0000000000000000
; DEFAULT-NEXT: movi v12.2d, #0000000000000000
; DEFAULT-NEXT: movi v13.2d, #0000000000000000
; DEFAULT-NEXT: movi v14.2d, #0000000000000000
; DEFAULT-NEXT: movi v15.2d, #0000000000000000
; DEFAULT-NEXT: movi v16.2d, #0000000000000000 ; DEFAULT-NEXT: movi v16.2d, #0000000000000000
; DEFAULT-NEXT: movi v17.2d, #0000000000000000 ; DEFAULT-NEXT: movi v17.2d, #0000000000000000
; DEFAULT-NEXT: movi v18.2d, #0000000000000000 ; DEFAULT-NEXT: movi v18.2d, #0000000000000000
; DEFAULT-NEXT: movi v19.2d, #0000000000000000 ; DEFAULT-NEXT: movi v19.2d, #0000000000000000
; DEFAULT-NEXT: movi v20.2d, #0000000000000000 ; DEFAULT-NEXT: movi v20.2d, #0000000000000000
; DEFAULT-NEXT: movi v21.2d, #0000000000000000 ; DEFAULT-NEXT: movi v21.2d, #0000000000000000
; DEFAULT-NEXT: movi v22.2d, #0000000000000000 ; DEFAULT-NEXT: movi v22.2d, #0000000000000000
; DEFAULT-NEXT: movi v23.2d, #0000000000000000 ; DEFAULT-NEXT: movi v23.2d, #0000000000000000
Show All 32 Lines
; SVE-NEXT: mov z0.d, #0 // =0x0 ; SVE-NEXT: mov z0.d, #0 // =0x0
; SVE-NEXT: mov z1.d, #0 // =0x0 ; SVE-NEXT: mov z1.d, #0 // =0x0
; SVE-NEXT: mov z2.d, #0 // =0x0 ; SVE-NEXT: mov z2.d, #0 // =0x0
; SVE-NEXT: mov z3.d, #0 // =0x0 ; SVE-NEXT: mov z3.d, #0 // =0x0
; SVE-NEXT: mov z4.d, #0 // =0x0 ; SVE-NEXT: mov z4.d, #0 // =0x0
; SVE-NEXT: mov z5.d, #0 // =0x0 ; SVE-NEXT: mov z5.d, #0 // =0x0
; SVE-NEXT: mov z6.d, #0 // =0x0 ; SVE-NEXT: mov z6.d, #0 // =0x0
; SVE-NEXT: mov z7.d, #0 // =0x0 ; SVE-NEXT: mov z7.d, #0 // =0x0
; SVE-NEXT: mov z8.d, #0 // =0x0
; SVE-NEXT: mov z9.d, #0 // =0x0
; SVE-NEXT: mov z10.d, #0 // =0x0
; SVE-NEXT: mov z11.d, #0 // =0x0
; SVE-NEXT: mov z12.d, #0 // =0x0
; SVE-NEXT: mov z13.d, #0 // =0x0
; SVE-NEXT: mov z14.d, #0 // =0x0
; SVE-NEXT: mov z15.d, #0 // =0x0
; SVE-NEXT: mov z16.d, #0 // =0x0 ; SVE-NEXT: mov z16.d, #0 // =0x0
; SVE-NEXT: mov z17.d, #0 // =0x0 ; SVE-NEXT: mov z17.d, #0 // =0x0
; SVE-NEXT: mov z18.d, #0 // =0x0 ; SVE-NEXT: mov z18.d, #0 // =0x0
; SVE-NEXT: mov z19.d, #0 // =0x0 ; SVE-NEXT: mov z19.d, #0 // =0x0
; SVE-NEXT: mov z20.d, #0 // =0x0 ; SVE-NEXT: mov z20.d, #0 // =0x0
; SVE-NEXT: mov z21.d, #0 // =0x0 ; SVE-NEXT: mov z21.d, #0 // =0x0
; SVE-NEXT: mov z22.d, #0 // =0x0 ; SVE-NEXT: mov z22.d, #0 // =0x0
; SVE-NEXT: mov z23.d, #0 // =0x0 ; SVE-NEXT: mov z23.d, #0 // =0x0
▲ Show 20 Lines • Show All 248 Lines • ▼ Show 20 Lines
; DEFAULT-NEXT: mov x18, #0 ; DEFAULT-NEXT: mov x18, #0
; DEFAULT-NEXT: movi v1.2d, #0000000000000000 ; DEFAULT-NEXT: movi v1.2d, #0000000000000000
; DEFAULT-NEXT: movi v2.2d, #0000000000000000 ; DEFAULT-NEXT: movi v2.2d, #0000000000000000
; DEFAULT-NEXT: movi v3.2d, #0000000000000000 ; DEFAULT-NEXT: movi v3.2d, #0000000000000000
; DEFAULT-NEXT: movi v4.2d, #0000000000000000 ; DEFAULT-NEXT: movi v4.2d, #0000000000000000
; DEFAULT-NEXT: movi v5.2d, #0000000000000000 ; DEFAULT-NEXT: movi v5.2d, #0000000000000000
; DEFAULT-NEXT: movi v6.2d, #0000000000000000 ; DEFAULT-NEXT: movi v6.2d, #0000000000000000
; DEFAULT-NEXT: movi v7.2d, #0000000000000000 ; DEFAULT-NEXT: movi v7.2d, #0000000000000000
; DEFAULT-NEXT: movi v8.2d, #0000000000000000
; DEFAULT-NEXT: movi v9.2d, #0000000000000000
; DEFAULT-NEXT: movi v10.2d, #0000000000000000
; DEFAULT-NEXT: movi v11.2d, #0000000000000000
; DEFAULT-NEXT: movi v12.2d, #0000000000000000
; DEFAULT-NEXT: movi v13.2d, #0000000000000000
; DEFAULT-NEXT: movi v14.2d, #0000000000000000
; DEFAULT-NEXT: movi v15.2d, #0000000000000000
; DEFAULT-NEXT: movi v16.2d, #0000000000000000 ; DEFAULT-NEXT: movi v16.2d, #0000000000000000
; DEFAULT-NEXT: movi v17.2d, #0000000000000000 ; DEFAULT-NEXT: movi v17.2d, #0000000000000000
; DEFAULT-NEXT: movi v18.2d, #0000000000000000 ; DEFAULT-NEXT: movi v18.2d, #0000000000000000
; DEFAULT-NEXT: movi v19.2d, #0000000000000000 ; DEFAULT-NEXT: movi v19.2d, #0000000000000000
; DEFAULT-NEXT: movi v20.2d, #0000000000000000 ; DEFAULT-NEXT: movi v20.2d, #0000000000000000
; DEFAULT-NEXT: movi v21.2d, #0000000000000000 ; DEFAULT-NEXT: movi v21.2d, #0000000000000000
; DEFAULT-NEXT: movi v22.2d, #0000000000000000 ; DEFAULT-NEXT: movi v22.2d, #0000000000000000
; DEFAULT-NEXT: movi v23.2d, #0000000000000000 ; DEFAULT-NEXT: movi v23.2d, #0000000000000000
Show All 32 Lines
; SVE-NEXT: mov x18, #0 ; SVE-NEXT: mov x18, #0
; SVE-NEXT: mov z1.d, #0 // =0x0 ; SVE-NEXT: mov z1.d, #0 // =0x0
; SVE-NEXT: mov z2.d, #0 // =0x0 ; SVE-NEXT: mov z2.d, #0 // =0x0
; SVE-NEXT: mov z3.d, #0 // =0x0 ; SVE-NEXT: mov z3.d, #0 // =0x0
; SVE-NEXT: mov z4.d, #0 // =0x0 ; SVE-NEXT: mov z4.d, #0 // =0x0
; SVE-NEXT: mov z5.d, #0 // =0x0 ; SVE-NEXT: mov z5.d, #0 // =0x0
; SVE-NEXT: mov z6.d, #0 // =0x0 ; SVE-NEXT: mov z6.d, #0 // =0x0
; SVE-NEXT: mov z7.d, #0 // =0x0 ; SVE-NEXT: mov z7.d, #0 // =0x0
; SVE-NEXT: mov z8.d, #0 // =0x0
; SVE-NEXT: mov z9.d, #0 // =0x0
; SVE-NEXT: mov z10.d, #0 // =0x0
; SVE-NEXT: mov z11.d, #0 // =0x0
; SVE-NEXT: mov z12.d, #0 // =0x0
; SVE-NEXT: mov z13.d, #0 // =0x0
; SVE-NEXT: mov z14.d, #0 // =0x0
; SVE-NEXT: mov z15.d, #0 // =0x0
; SVE-NEXT: mov z16.d, #0 // =0x0 ; SVE-NEXT: mov z16.d, #0 // =0x0
; SVE-NEXT: mov z17.d, #0 // =0x0 ; SVE-NEXT: mov z17.d, #0 // =0x0
; SVE-NEXT: mov z18.d, #0 // =0x0 ; SVE-NEXT: mov z18.d, #0 // =0x0
; SVE-NEXT: mov z19.d, #0 // =0x0 ; SVE-NEXT: mov z19.d, #0 // =0x0
; SVE-NEXT: mov z20.d, #0 // =0x0 ; SVE-NEXT: mov z20.d, #0 // =0x0
; SVE-NEXT: mov z21.d, #0 // =0x0 ; SVE-NEXT: mov z21.d, #0 // =0x0
; SVE-NEXT: mov z22.d, #0 // =0x0 ; SVE-NEXT: mov z22.d, #0 // =0x0
; SVE-NEXT: mov z23.d, #0 // =0x0 ; SVE-NEXT: mov z23.d, #0 // =0x0
▲ Show 20 Lines • Show All 44 Lines • Show Last 20 Lines

llvm/test/CodeGen/X86/zero-call-used-regs.ll

Show First 20 Lines • Show All 104 Lines • ▼ Show 20 Lines
entry: entry:
ret i32 %x ret i32 %x
} }
define dso_local i32 @all_gpr(i32 returned %x) local_unnamed_addr #0 "zero-call-used-regs"="all-gpr" { define dso_local i32 @all_gpr(i32 returned %x) local_unnamed_addr #0 "zero-call-used-regs"="all-gpr" {
; I386-LABEL: all_gpr: ; I386-LABEL: all_gpr:
; I386: # %bb.0: # %entry ; I386: # %bb.0: # %entry
; I386-NEXT: movl {{[0-9]+}}(%esp), %eax ; I386-NEXT: movl {{[0-9]+}}(%esp), %eax
; I386-NEXT: xorl %ebp, %ebp
; I386-NEXT: xorl %ebx, %ebx
; I386-NEXT: xorl %ecx, %ecx ; I386-NEXT: xorl %ecx, %ecx
; I386-NEXT: xorl %edi, %edi
; I386-NEXT: xorl %edx, %edx ; I386-NEXT: xorl %edx, %edx
; I386-NEXT: xorl %esi, %esi
; I386-NEXT: retl ; I386-NEXT: retl
; ;
; X86-64-LABEL: all_gpr: ; X86-64-LABEL: all_gpr:
; X86-64: # %bb.0: # %entry ; X86-64: # %bb.0: # %entry
; X86-64-NEXT: movl %edi, %eax ; X86-64-NEXT: movl %edi, %eax
; X86-64-NEXT: xorl %ebp, %ebp
; X86-64-NEXT: xorl %ebx, %ebx
; X86-64-NEXT: xorl %ecx, %ecx ; X86-64-NEXT: xorl %ecx, %ecx
; X86-64-NEXT: xorl %edi, %edi ; X86-64-NEXT: xorl %edi, %edi
; X86-64-NEXT: xorl %edx, %edx ; X86-64-NEXT: xorl %edx, %edx
; X86-64-NEXT: xorl %esi, %esi ; X86-64-NEXT: xorl %esi, %esi
; X86-64-NEXT: xorl %r8d, %r8d ; X86-64-NEXT: xorl %r8d, %r8d
; X86-64-NEXT: xorl %r9d, %r9d ; X86-64-NEXT: xorl %r9d, %r9d
; X86-64-NEXT: xorl %r10d, %r10d ; X86-64-NEXT: xorl %r10d, %r10d
; X86-64-NEXT: xorl %r11d, %r11d ; X86-64-NEXT: xorl %r11d, %r11d
; X86-64-NEXT: xorl %r12d, %r12d
; X86-64-NEXT: xorl %r13d, %r13d
; X86-64-NEXT: xorl %r14d, %r14d
; X86-64-NEXT: xorl %r15d, %r15d
; X86-64-NEXT: retq ; X86-64-NEXT: retq
entry: entry:
ret i32 %x ret i32 %x
} }
define dso_local i32 @all_arg(i32 returned %x) local_unnamed_addr #0 "zero-call-used-regs"="all-arg" { define dso_local i32 @all_arg(i32 returned %x) local_unnamed_addr #0 "zero-call-used-regs"="all-arg" {
; I386-LABEL: all_arg: ; I386-LABEL: all_arg:
Show All 39 Lines
; I386-NEXT: fldz ; I386-NEXT: fldz
; I386-NEXT: fstp %st(0) ; I386-NEXT: fstp %st(0)
; I386-NEXT: fstp %st(0) ; I386-NEXT: fstp %st(0)
; I386-NEXT: fstp %st(0) ; I386-NEXT: fstp %st(0)
; I386-NEXT: fstp %st(0) ; I386-NEXT: fstp %st(0)
; I386-NEXT: fstp %st(0) ; I386-NEXT: fstp %st(0)
; I386-NEXT: fstp %st(0) ; I386-NEXT: fstp %st(0)
; I386-NEXT: fstp %st(0) ; I386-NEXT: fstp %st(0)
; I386-NEXT: xorl %ebp, %ebp
; I386-NEXT: xorl %ebx, %ebx
; I386-NEXT: xorl %ecx, %ecx ; I386-NEXT: xorl %ecx, %ecx
; I386-NEXT: xorl %edi, %edi
; I386-NEXT: xorl %edx, %edx ; I386-NEXT: xorl %edx, %edx
; I386-NEXT: xorl %esi, %esi
; I386-NEXT: xorps %xmm0, %xmm0 ; I386-NEXT: xorps %xmm0, %xmm0
; I386-NEXT: xorps %xmm1, %xmm1 ; I386-NEXT: xorps %xmm1, %xmm1
; I386-NEXT: xorps %xmm2, %xmm2 ; I386-NEXT: xorps %xmm2, %xmm2
; I386-NEXT: xorps %xmm3, %xmm3 ; I386-NEXT: xorps %xmm3, %xmm3
; I386-NEXT: xorps %xmm4, %xmm4 ; I386-NEXT: xorps %xmm4, %xmm4
; I386-NEXT: xorps %xmm5, %xmm5 ; I386-NEXT: xorps %xmm5, %xmm5
; I386-NEXT: xorps %xmm6, %xmm6 ; I386-NEXT: xorps %xmm6, %xmm6
; I386-NEXT: xorps %xmm7, %xmm7 ; I386-NEXT: xorps %xmm7, %xmm7
Show All 13 Lines
; X86-64-NEXT: fstp %st(0) ; X86-64-NEXT: fstp %st(0)
; X86-64-NEXT: fstp %st(0) ; X86-64-NEXT: fstp %st(0)
; X86-64-NEXT: fstp %st(0) ; X86-64-NEXT: fstp %st(0)
; X86-64-NEXT: fstp %st(0) ; X86-64-NEXT: fstp %st(0)
; X86-64-NEXT: fstp %st(0) ; X86-64-NEXT: fstp %st(0)
; X86-64-NEXT: fstp %st(0) ; X86-64-NEXT: fstp %st(0)
; X86-64-NEXT: fstp %st(0) ; X86-64-NEXT: fstp %st(0)
; X86-64-NEXT: fstp %st(0) ; X86-64-NEXT: fstp %st(0)
; X86-64-NEXT: xorl %ebp, %ebp
; X86-64-NEXT: xorl %ebx, %ebx
; X86-64-NEXT: xorl %ecx, %ecx ; X86-64-NEXT: xorl %ecx, %ecx
; X86-64-NEXT: xorl %edi, %edi ; X86-64-NEXT: xorl %edi, %edi
; X86-64-NEXT: xorl %edx, %edx ; X86-64-NEXT: xorl %edx, %edx
; X86-64-NEXT: xorl %esi, %esi ; X86-64-NEXT: xorl %esi, %esi
; X86-64-NEXT: xorl %r8d, %r8d ; X86-64-NEXT: xorl %r8d, %r8d
; X86-64-NEXT: xorl %r9d, %r9d ; X86-64-NEXT: xorl %r9d, %r9d
; X86-64-NEXT: xorl %r10d, %r10d ; X86-64-NEXT: xorl %r10d, %r10d
; X86-64-NEXT: xorl %r11d, %r11d ; X86-64-NEXT: xorl %r11d, %r11d
; X86-64-NEXT: xorl %r12d, %r12d
; X86-64-NEXT: xorl %r13d, %r13d
; X86-64-NEXT: xorl %r14d, %r14d
; X86-64-NEXT: xorl %r15d, %r15d
; X86-64-NEXT: xorps %xmm0, %xmm0 ; X86-64-NEXT: xorps %xmm0, %xmm0
; X86-64-NEXT: xorps %xmm1, %xmm1 ; X86-64-NEXT: xorps %xmm1, %xmm1
; X86-64-NEXT: xorps %xmm2, %xmm2 ; X86-64-NEXT: xorps %xmm2, %xmm2
; X86-64-NEXT: xorps %xmm3, %xmm3 ; X86-64-NEXT: xorps %xmm3, %xmm3
; X86-64-NEXT: xorps %xmm4, %xmm4 ; X86-64-NEXT: xorps %xmm4, %xmm4
; X86-64-NEXT: xorps %xmm5, %xmm5 ; X86-64-NEXT: xorps %xmm5, %xmm5
; X86-64-NEXT: xorps %xmm6, %xmm6 ; X86-64-NEXT: xorps %xmm6, %xmm6
; X86-64-NEXT: xorps %xmm7, %xmm7 ; X86-64-NEXT: xorps %xmm7, %xmm7
Show All 32 Lines