mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-23 03:02:36 +01:00
[IPRA][ARM] Disable no-CSR optimisation for ARM
This optimisation isn't generally profitable for ARM, because we can save/restore many registers in the prologue and epilogue using the PUSH and POP instructions, but mostly use individual LDR/STR instructions for other spills. Differential revision: https://reviews.llvm.org/D64910 llvm-svn: 367670
This commit is contained in:
parent
c00402ed88
commit
fac519c9d7
@ -378,6 +378,11 @@ public:
|
||||
return true;
|
||||
}
|
||||
|
||||
/// Check if the no-CSR optimisation is profitable for the given function.
|
||||
virtual bool isProfitableForNoCSROpt(const Function &F) const {
|
||||
return true;
|
||||
}
|
||||
|
||||
/// Return initial CFA offset value i.e. the one valid at the beginning of the
|
||||
/// function (before any stack operations).
|
||||
virtual int getInitialCFAOffset(const MachineFunction &MF) const;
|
||||
|
@ -171,7 +171,8 @@ bool RegUsageInfoCollector::runOnMachineFunction(MachineFunction &MF) {
|
||||
SetRegAsDefined(PReg);
|
||||
}
|
||||
|
||||
if (TargetFrameLowering::isSafeForNoCSROpt(F)) {
|
||||
if (TargetFrameLowering::isSafeForNoCSROpt(F) &&
|
||||
MF.getSubtarget().getFrameLowering()->isProfitableForNoCSROpt(F)) {
|
||||
++NumCSROpt;
|
||||
LLVM_DEBUG(dbgs() << MF.getName()
|
||||
<< " function optimized for not having CSR.\n");
|
||||
|
@ -71,7 +71,9 @@ void TargetFrameLowering::determineCalleeSaves(MachineFunction &MF,
|
||||
|
||||
// When interprocedural register allocation is enabled caller saved registers
|
||||
// are preferred over callee saved registers.
|
||||
if (MF.getTarget().Options.EnableIPRA && isSafeForNoCSROpt(MF.getFunction()))
|
||||
if (MF.getTarget().Options.EnableIPRA &&
|
||||
isSafeForNoCSROpt(MF.getFunction()) &&
|
||||
isProfitableForNoCSROpt(MF.getFunction()))
|
||||
return;
|
||||
|
||||
// Get the callee saved register list...
|
||||
|
@ -63,6 +63,11 @@ public:
|
||||
bool enableShrinkWrapping(const MachineFunction &MF) const override {
|
||||
return true;
|
||||
}
|
||||
bool isProfitableForNoCSROpt(const Function &F) const override {
|
||||
// The no-CSR optimisation is bad for code size on ARM, because we can save
|
||||
// many registers with a single PUSH/POP pair.
|
||||
return false;
|
||||
}
|
||||
|
||||
private:
|
||||
void emitPushInst(MachineBasicBlock &MBB, MachineBasicBlock::iterator MI,
|
||||
|
22
test/CodeGen/ARM/ipra-no-csr.ll
Normal file
22
test/CodeGen/ARM/ipra-no-csr.ll
Normal file
@ -0,0 +1,22 @@
|
||||
; RUN: llc -mtriple armv7a--none-eabi < %s | FileCheck %s
|
||||
; RUN: llc -mtriple armv7a--none-eabi < %s -enable-ipra | FileCheck %s
|
||||
|
||||
; Other targets disable callee-saved registers for internal functions when
|
||||
; using IPRA, but that isn't profitable for ARM because the PUSH/POP
|
||||
; instructions can more efficiently save registers than using individual
|
||||
; LDR/STRs in the caller.
|
||||
|
||||
define internal void @callee() norecurse {
|
||||
; CHECK-LABEL: callee:
|
||||
entry:
|
||||
; CHECK: push {r4, lr}
|
||||
; CHECK: pop {r4, pc}
|
||||
tail call void asm sideeffect "", "~{r4}"()
|
||||
ret void
|
||||
}
|
||||
|
||||
define void @caller() {
|
||||
entry:
|
||||
call void @callee()
|
||||
ret void
|
||||
}
|
Loading…
Reference in New Issue
Block a user