summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorJakob Stoklund Olesen <stoklund@2pi.dk>2012-02-28 01:15:01 +0000
committerJakob Stoklund Olesen <stoklund@2pi.dk>2012-02-28 01:15:01 +0000
commit92c15b2b2cddbfb842530e4c4ebfdf90e612aaf9 (patch)
tree33d149da0582e45ff1eb1bf4296fa9fd695dfe41
parentec75fec805a7dec0c9a4e29bebdeca0bc8ad1e5b (diff)
downloadbcm5719-llvm-92c15b2b2cddbfb842530e4c4ebfdf90e612aaf9.tar.gz
bcm5719-llvm-92c15b2b2cddbfb842530e4c4ebfdf90e612aaf9.zip
Enable ARM base pointer when calling functions with large arguments.
When an outgoing call takes more than 2k of arguments on the stack, we don't allocate that call frame in the prolog, but adjust the stack pointer immediately before the call instead. This causes problems with the emergency spill slot because PEI can't track stack pointer adjustments on the second pass, and if the outgoing arguments are too big, SP can't be used to reach the emergency spill slot at all. Work around these problems by ensuring there is a base or frame pointer that can be used to access the emergency spill slot. <rdar://problem/10917166> llvm-svn: 151604
-rw-r--r--llvm/lib/Target/ARM/ARMBaseRegisterInfo.cpp22
-rw-r--r--llvm/lib/Target/ARM/ARMFrameLowering.cpp11
-rw-r--r--llvm/lib/Target/ARM/Thumb1RegisterInfo.cpp16
-rw-r--r--llvm/test/CodeGen/Thumb2/large-call.ll29
4 files changed, 73 insertions, 5 deletions
diff --git a/llvm/lib/Target/ARM/ARMBaseRegisterInfo.cpp b/llvm/lib/Target/ARM/ARMBaseRegisterInfo.cpp
index 2608685e045..8512049e8ce 100644
--- a/llvm/lib/Target/ARM/ARMBaseRegisterInfo.cpp
+++ b/llvm/lib/Target/ARM/ARMBaseRegisterInfo.cpp
@@ -484,11 +484,16 @@ ARMBaseRegisterInfo::avoidWriteAfterWrite(const TargetRegisterClass *RC) const {
bool ARMBaseRegisterInfo::hasBasePointer(const MachineFunction &MF) const {
const MachineFrameInfo *MFI = MF.getFrameInfo();
const ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
+ const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
if (!EnableBasePointer)
return false;
- if (needsStackRealignment(MF) && MFI->hasVarSizedObjects())
+ // When outgoing call frames are so large that we adjust the stack pointer
+ // around the call, we can no longer use the stack pointer to reach the
+ // emergency spill slot.
+ if (needsStackRealignment(MF) && (MFI->hasVarSizedObjects() ||
+ !TFI->hasReservedCallFrame(MF)))
return true;
// Thumb has trouble with negative offsets from the FP. Thumb2 has a limited
@@ -1056,6 +1061,21 @@ ARMBaseRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
int Offset = TFI->ResolveFrameIndexReference(MF, FrameIndex, FrameReg, SPAdj);
+ // PEI::scavengeFrameVirtualRegs() cannot accurately track SPAdj because the
+ // call frame setup/destroy instructions have already been eliminated. That
+ // means the stack pointer cannot be used to access the emergency spill slot
+ // when !hasReservedCallFrame().
+#ifndef NDEBUG
+ if (RS && FrameReg == ARM::SP && FrameIndex == RS->getScavengingFrameIndex()){
+ assert(TFI->hasReservedCallFrame(MF) &&
+ "Cannot use SP to access the emergency spill slot in "
+ "functions without a reserved call frame");
+ assert(!MF.getFrameInfo()->hasVarSizedObjects() &&
+ "Cannot use SP to access the emergency spill slot in "
+ "functions with variable sized frame objects");
+ }
+#endif // NDEBUG
+
// Special handling of dbg_value instructions.
if (MI.isDebugValue()) {
MI.getOperand(i). ChangeToRegister(FrameReg, false /*isDef*/);
diff --git a/llvm/lib/Target/ARM/ARMFrameLowering.cpp b/llvm/lib/Target/ARM/ARMFrameLowering.cpp
index f3a2197c233..463e4abc760 100644
--- a/llvm/lib/Target/ARM/ARMFrameLowering.cpp
+++ b/llvm/lib/Target/ARM/ARMFrameLowering.cpp
@@ -499,6 +499,10 @@ ARMFrameLowering::ResolveFrameIndexReference(const MachineFunction &MF,
else if (AFI->isDPRCalleeSavedAreaFrame(FI))
return Offset - AFI->getDPRCalleeSavedAreaOffset();
+ // SP can move around if there are allocas. We may also lose track of SP
+ // when emergency spilling inside a non-reserved call frame setup.
+ bool hasMovingSP = MFI->hasVarSizedObjects() || !hasReservedCallFrame(MF);
+
// When dynamically realigning the stack, use the frame pointer for
// parameters, and the stack/base pointer for locals.
if (RegInfo->needsStackRealignment(MF)) {
@@ -506,7 +510,7 @@ ARMFrameLowering::ResolveFrameIndexReference(const MachineFunction &MF,
if (isFixed) {
FrameReg = RegInfo->getFrameRegister(MF);
Offset = FPOffset;
- } else if (MFI->hasVarSizedObjects()) {
+ } else if (hasMovingSP) {
assert(RegInfo->hasBasePointer(MF) &&
"VLAs and dynamic stack alignment, but missing base pointer!");
FrameReg = RegInfo->getBaseRegister();
@@ -518,11 +522,10 @@ ARMFrameLowering::ResolveFrameIndexReference(const MachineFunction &MF,
if (hasFP(MF) && AFI->hasStackFrame()) {
// Use frame pointer to reference fixed objects. Use it for locals if
// there are VLAs (and thus the SP isn't reliable as a base).
- if (isFixed || (MFI->hasVarSizedObjects() &&
- !RegInfo->hasBasePointer(MF))) {
+ if (isFixed || (hasMovingSP && !RegInfo->hasBasePointer(MF))) {
FrameReg = RegInfo->getFrameRegister(MF);
return FPOffset;
- } else if (MFI->hasVarSizedObjects()) {
+ } else if (hasMovingSP) {
assert(RegInfo->hasBasePointer(MF) && "missing base pointer!");
if (AFI->isThumb2Function()) {
// Try to use the frame pointer if we can, else use the base pointer
diff --git a/llvm/lib/Target/ARM/Thumb1RegisterInfo.cpp b/llvm/lib/Target/ARM/Thumb1RegisterInfo.cpp
index 87e27bd0a13..1613acd7a60 100644
--- a/llvm/lib/Target/ARM/Thumb1RegisterInfo.cpp
+++ b/llvm/lib/Target/ARM/Thumb1RegisterInfo.cpp
@@ -28,6 +28,7 @@
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/RegisterScavenging.h"
#include "llvm/Target/TargetFrameLowering.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Support/CommandLine.h"
@@ -624,6 +625,21 @@ Thumb1RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
FrameReg = BasePtr;
}
+ // PEI::scavengeFrameVirtualRegs() cannot accurately track SPAdj because the
+ // call frame setup/destroy instructions have already been eliminated. That
+ // means the stack pointer cannot be used to access the emergency spill slot
+ // when !hasReservedCallFrame().
+#ifndef NDEBUG
+ if (RS && FrameReg == ARM::SP && FrameIndex == RS->getScavengingFrameIndex()){
+ assert(MF.getTarget().getFrameLowering()->hasReservedCallFrame(MF) &&
+ "Cannot use SP to access the emergency spill slot in "
+ "functions without a reserved call frame");
+ assert(!MF.getFrameInfo()->hasVarSizedObjects() &&
+ "Cannot use SP to access the emergency spill slot in "
+ "functions with variable sized frame objects");
+ }
+#endif // NDEBUG
+
// Special handling of dbg_value instructions.
if (MI.isDebugValue()) {
MI.getOperand(i). ChangeToRegister(FrameReg, false /*isDef*/);
diff --git a/llvm/test/CodeGen/Thumb2/large-call.ll b/llvm/test/CodeGen/Thumb2/large-call.ll
new file mode 100644
index 00000000000..aef6f856064
--- /dev/null
+++ b/llvm/test/CodeGen/Thumb2/large-call.ll
@@ -0,0 +1,29 @@
+; RUN: llc < %s -O0 -mcpu=cortex-a8 | FileCheck %s
+target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:32:64-v128:32:128-a0:0:32-n32-S32"
+target triple = "thumbv7-apple-ios0.0.0"
+
+; This test case would clobber the outgoing call arguments by writing to the
+; emergency spill slot at [sp, #4] without adjusting the stack pointer first.
+
+; CHECK: main
+; CHECK: vmov.f64
+; Adjust SP for the large call
+; CHECK: sub sp,
+; CHECK: mov [[FR:r[0-9]+]], sp
+; Store to call frame + #4
+; CHECK: str{{.*\[}}[[FR]], #4]
+; Don't clobber that store until the call.
+; CHECK-NOT: [sp, #4]
+; CHECK: variadic
+
+define i32 @main() ssp {
+entry:
+ %d = alloca double, align 8
+ store double 1.000000e+00, double* %d, align 8
+ %0 = load double* %d, align 8
+ call void (i8*, i8*, i8*, ...)* @variadic(i8* null, i8* null, i8* null, i32 1, double 1.234800e+03, double 2.363450e+03, double %0, i32 1, double 1.234560e+03, double 2.345670e+03, double 4.6334563e+03, double 2.423440e+03, double 4.234330e+03, double 2.965430e+03, i32 1, double 4.669300e+03, double 2.927500e+03, double 4.663100e+03, double 2.921000e+03, double 4.663100e+03, double 2.345100e+03, i32 1, double 3.663100e+03, double 2.905100e+03, double 4.669300e+03, double 2.898600e+03, double 4.676900e+03, double 2.898600e+03, i32 1, double 4.684600e+03, double 2.898600e+03, double 1.234800e+03, double 2.905100e+03, double 1.234800e+03, double 2.345100e+03, i32 1, double 7.719700e+03, double 2.920500e+03, double 4.713500e+03, double 2.927000e+03, double 4.705800e+03, double 2.927000e+03, i32 1, double 8.698200e+03, double 2.927000e+03, double 4.692000e+03, double 2.920500e+03, double 4.692000e+03, double 2.912500e+03, i32 1, double 4.692000e+03, double 2.945600e+03, double 4.698200e+03, double 2.898100e+03, double 4.705800e+03, double 2.898100e+03, i32 1, double 4.713500e+03, double 2.898100e+03, double 4.719700e+03, double 2.945600e+03, double 4.719700e+03, double 2.912500e+03, i32 1, double 4.749200e+03, double 2.920100e+03, double 4.743000e+03, double 2.926600e+03, double 4.735300e+03, double 2.926600e+03, i32 1, double 4.727700e+03, double 2.926600e+03, double 4.721500e+03, double 2.920100e+03, double 4.721500e+03, double 2.912100e+03, i32 1, double 4.721500e+03, double 2.945100e+03, double 4.727700e+03, double 2.897700e+03, double 4.735300e+03, double 2.897700e+03, i32 1, double 4.743000e+03, double 2.897700e+03, double 4.749200e+03, double 2.945100e+03, double 4.749200e+03, double 2.912100e+03, i32 1, double 4.778200e+03, double 2.920100e+03, double 4.772000e+03, double 2.926600e+03, double 4.764300e+03, double 2.926600e+03, i32 1, double 4.756700e+03, double 2.926600e+03, double 4.750500e+03, double 2.920100e+03, double 4.750500e+03, double 2.912100e+03, i32 1, double 4.750500e+03, double 2.945100e+03, double 4.756700e+03, double 2.897700e+03, double 4.764300e+03, double 2.897700e+03, i32 1, double 4.772000e+03, double 2.897700e+03, double 4.778200e+03, double 2.945100e+03, double 4.778200e+03, double 2.912100e+03, i32 1, double 4.801900e+03, double 2.942100e+03, double 4.795700e+03, double 2.948500e+03, double 4.788100e+03, double 2.948500e+03, i32 1, double 4.780500e+03, double 2.948500e+03, double 4.774300e+03, double 2.942100e+03, double 4.774300e+03, double 2.934100e+03, i32 1, double 4.774300e+03, double 2.926100e+03, double 4.780500e+03, double 2.919600e+03, double 4.788100e+03, double 2.919600e+03, i32 1, double 4.795700e+03, double 2.919600e+03, double 4.801900e+03, double 2.926100e+03, double 4.801900e+03, double 2.934100e+03, i32 1, double 4.801500e+03, double 2.972500e+03, double 4.795300e+03, double 2.978900e+03, double 4.787700e+03, double 2.978900e+03, i32 1, double 4.780000e+03, double 2.978900e+03, double 4.773800e+03, double 2.972500e+03, double 4.773800e+03, double 2.964500e+03, i32 1, double 4.773800e+03, double 2.956500e+03, double 4.780000e+03, double 2.950000e+03, double 4.787700e+03, double 2.950000e+03, i32 1, double 4.795300e+03, double 2.950000e+03, double 4.801500e+03, double 2.956500e+03, double 4.801500e+03, double 2.964500e+03, i32 1, double 4.802400e+03, double 3.010200e+03, double 4.796200e+03, double 3.016600e+03, double 4.788500e+03, double 3.016600e+03, i32 1, double 4.780900e+03, double 3.016600e+03, double 4.774700e+03, double 3.010200e+03, double 4.774700e+03, double 3.002200e+03, i32 1, double 4.774700e+03, double 2.994200e+03, double 4.780900e+03, double 2.987700e+03, double 4.788500e+03, double 2.987700e+03, i32 1, double 4.796200e+03, double 2.987700e+03, double 4.802400e+03, double 2.994200e+03, double 4.802400e+03, double 3.002200e+03, i32 1, double 4.802400e+03, double 3.039400e+03, double 4.796200e+03, double 3.455800e+03, double 4.788500e+03, double 3.455800e+03, i32 1, double 4.780900e+03, double 3.455800e+03, double 4.774700e+03, double 3.039400e+03, double 4.774700e+03, double 3.031400e+03, i32 1, double 4.774700e+03, double 3.023400e+03, double 4.780900e+03, double 3.016900e+03, double 4.788500e+03, double 3.016900e+03, i32 1, double 4.796200e+03, double 3.016900e+03, double 4.802400e+03, double 3.023400e+03, double 4.802400e+03, double 3.031400e+03, i32 1, double 4.778600e+03, double 3.063100e+03, double 4.772400e+03, double 3.069600e+03, double 4.764700e+03, double 3.069600e+03, i32 1, double 4.757100e+03, double 3.069600e+03, double 4.750900e+03, double 3.063100e+03, double 4.750900e+03, double 3.055100e+03, i32 1, double 4.750900e+03, double 3.457100e+03, double 4.757100e+03, double 3.450700e+03, double 4.764700e+03, double 3.450700e+03, i32 1, double 4.772400e+03, double 3.450700e+03, double 4.778600e+03, double 3.457100e+03, double 4.778600e+03, double 3.055100e+03, i32 1, double 4.748600e+03, double 3.063600e+03, double 4.742400e+03, double 3.070000e+03, double 4.734700e+03, double 3.070000e+03, i32 1, double 4.727100e+03, double 3.070000e+03, double 4.720900e+03, double 3.063600e+03, double 4.720900e+03, double 3.055600e+03, i32 1, double 4.720900e+03, double 3.457600e+03, double 4.727100e+03, double 3.451100e+03, double 4.734700e+03, double 3.451100e+03, i32 1, double 4.742400e+03, double 3.451100e+03, double 4.748600e+03, double 3.457600e+03, double 4.748600e+03, double 3.055600e+03, i32 1, double 4.719500e+03, double 3.063600e+03, double 4.713300e+03, double 3.070000e+03, double 4.705700e+03, double 3.070000e+03, i32 1, double 4.698000e+03, double 3.070000e+03, double 4.691900e+03, double 3.063600e+03, double 4.691900e+03, double 3.055600e+03, i32 1, double 4.691900e+03, double 3.457600e+03, double 4.698000e+03, double 3.451100e+03, double 4.705700e+03, double 3.451100e+03, i32 1, double 4.713300e+03, double 3.451100e+03, double 4.719500e+03, double 3.457600e+03, double 4.719500e+03, double 3.055600e+03, i32 1, double 4.691300e+03, double 3.064000e+03, double 4.685100e+03, double 3.070500e+03, double 4.677500e+03, double 3.070500e+03, i32 1, double 4.669900e+03, double 3.070500e+03, double 4.663700e+03, double 3.064000e+03, double 4.663700e+03, double 3.056000e+03, i32 1, double 4.663700e+03, double 3.458000e+03, double 4.669900e+03, double 3.451600e+03, double 4.677500e+03, double 3.451600e+03, i32 1, double 4.685100e+03, double 3.451600e+03, double 4.691300e+03, double 3.458000e+03, double 4.691300e+03, double 3.056000e+03, i32 1, double 4.668500e+03, double 3.453000e+03, double 4.662300e+03, double 3.459400e+03, double 4.654700e+03, double 3.459400e+03, i32 1, double 4.647000e+03, double 3.459400e+03, double 4.640900e+03, double 3.453000e+03, double 4.640900e+03, double 3.035000e+03, i32 1, double 4.640900e+03, double 3.027000e+03, double 4.647000e+03, double 3.020500e+03, double 4.654700e+03, double 3.020500e+03, i32 1, double 4.662300e+03, double 3.020500e+03, double 4.668500e+03, double 3.027000e+03, double 4.668500e+03, double 3.035000e+03, i32 1, double 4.668500e+03, double 3.014300e+03, double 4.662300e+03, double 3.020800e+03, double 4.654700e+03, double 3.020800e+03, i32 1, double 4.647000e+03, double 3.020800e+03, double 4.640900e+03, double 3.014300e+03, double 4.640900e+03, double 3.006400e+03, i32 1, double 4.640900e+03, double 2.998400e+03, double 4.647000e+03, double 2.991900e+03, double 4.654700e+03, double 2.991900e+03, i32 1, double 4.662300e+03, double 2.991900e+03, double 4.668500e+03, double 2.998400e+03, double 4.668500e+03, double 3.006400e+03, i32 1, double 4.668100e+03, double 2.941100e+03, double 4.661900e+03, double 2.947600e+03, double 4.654200e+03, double 2.947600e+03, i32 1, double 4.646600e+03, double 2.947600e+03, double 4.640400e+03, double 2.941100e+03, double 4.640400e+03, double 2.933100e+03, i32 1, double 4.640400e+03, double 2.925200e+03, double 4.646600e+03, double 2.918700e+03, double 4.654200e+03, double 2.918700e+03, i32 1, double 4.661900e+03, double 2.918700e+03, double 4.668100e+03, double 2.925200e+03, double 4.668100e+03, double 2.933100e+03, i32 1, double 4.668500e+03, double 2.971600e+03, double 4.662300e+03, double 2.978100e+03, double 4.654700e+03, double 2.978100e+03, i32 1, double 4.647000e+03, double 2.978100e+03, double 4.640900e+03, double 2.971600e+03, double 4.640900e+03, double 2.963600e+03, i32 1, double 4.640900e+03, double 2.955700e+03, double 4.647000e+03, double 2.949200e+03, double 4.654700e+03, double 2.949200e+03, i32 1, double 4.662300e+03, double 2.949200e+03, double 4.668500e+03, double 2.955700e+03, double 4.668500e+03, double 2.963600e+03, i32 2, i32 1, double 4.691300e+03, double 3.056000e+03, i32 2, i32 1, double 4.748600e+03, double 3.055600e+03, i32 2, i32 1, double 4.778200e+03, double 2.912100e+03, i32 2, i32 1, double 4.749200e+03, double 2.912100e+03, i32 2, i32 1, double 4.802400e+03, double 3.031400e+03, i32 2, i32 1, double 4.778600e+03, double 3.055100e+03, i32 2, i32 1, double 4.801500e+03, double 2.964500e+03, i32 2, i32 1, double 4.802400e+03, double 3.002200e+03, i32 2, i32 1, double 4.719700e+03, double 2.912500e+03, i32 2, i32 1, double 4.801900e+03, double 2.934100e+03, i32 2, i32 1, double 4.719500e+03, double 3.055600e+03, i32 2, i32 1, double 4.668500e+03, double 3.006400e+03, i32 2, i32 1, double 4.668500e+03, double 3.035000e+03, i32 2, i32 1, double 4.668100e+03, double 2.933100e+03, i32 2, i32 1, double 4.668500e+03, double 2.963600e+03, i32 2, i32 48)
+ ret i32 0
+}
+
+declare void @variadic(i8*, i8*, i8*, ...)
+
OpenPOWER on IntegriCloud