summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorEvan Cheng <evan.cheng@apple.com>2011-11-08 21:21:09 +0000
committerEvan Cheng <evan.cheng@apple.com>2011-11-08 21:21:09 +0000
commit3568a1051efb9a9edbd4914b04b44e9d7bc1b004 (patch)
tree871918870b9c3a8712b540a382dc734b5a65a83d
parent66dc8ca04b719f3ab4aa650609dbd56b055ecb34 (diff)
downloadllvm-3568a1051efb9a9edbd4914b04b44e9d7bc1b004.tar.gz
llvm-3568a1051efb9a9edbd4914b04b44e9d7bc1b004.tar.bz2
llvm-3568a1051efb9a9edbd4914b04b44e9d7bc1b004.tar.xz
Add workaround for Cortex-M3 errata 602117 by replacing ldrd x, y, [x] with ldm or ldr pairs.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@144123 91177308-0d34-0410-b5e6-96231b3b80d8
-rw-r--r--lib/Target/ARM/ARMLoadStoreOptimizer.cpp12
-rw-r--r--test/CodeGen/ARM/ldrd.ll23
2 files changed, 21 insertions, 14 deletions
diff --git a/lib/Target/ARM/ARMLoadStoreOptimizer.cpp b/lib/Target/ARM/ARMLoadStoreOptimizer.cpp
index faa8ba7684..a871ed7d44 100644
--- a/lib/Target/ARM/ARMLoadStoreOptimizer.cpp
+++ b/lib/Target/ARM/ARMLoadStoreOptimizer.cpp
@@ -62,6 +62,7 @@ namespace {
const TargetInstrInfo *TII;
const TargetRegisterInfo *TRI;
+ const ARMSubtarget *STI;
ARMFunctionInfo *AFI;
RegScavenger *RS;
bool isThumb2;
@@ -1071,11 +1072,17 @@ bool ARMLoadStoreOpt::FixInvalidRegPairOp(MachineBasicBlock &MBB,
unsigned Opcode = MI->getOpcode();
if (Opcode == ARM::LDRD || Opcode == ARM::STRD ||
Opcode == ARM::t2LDRDi8 || Opcode == ARM::t2STRDi8) {
+ const MachineOperand &BaseOp = MI->getOperand(2);
+ unsigned BaseReg = BaseOp.getReg();
unsigned EvenReg = MI->getOperand(0).getReg();
unsigned OddReg = MI->getOperand(1).getReg();
unsigned EvenRegNum = TRI->getDwarfRegNum(EvenReg, false);
unsigned OddRegNum = TRI->getDwarfRegNum(OddReg, false);
- if ((EvenRegNum & 1) == 0 && (EvenRegNum + 1) == OddRegNum)
+ // ARM errata 602117: LDRD with base in list may result in incorrect base
+ // register when interrupted or faulted.
+ bool Errata602117 = EvenReg == BaseReg && STI->getCPUString() == "cortex-m3";
+ if (!Errata602117 &&
+ ((EvenRegNum & 1) == 0 && (EvenRegNum + 1) == OddRegNum))
return false;
MachineBasicBlock::iterator NewBBI = MBBI;
@@ -1087,8 +1094,6 @@ bool ARMLoadStoreOpt::FixInvalidRegPairOp(MachineBasicBlock &MBB,
bool OddDeadKill = isLd ?
MI->getOperand(1).isDead() : MI->getOperand(1).isKill();
bool OddUndef = MI->getOperand(1).isUndef();
- const MachineOperand &BaseOp = MI->getOperand(2);
- unsigned BaseReg = BaseOp.getReg();
bool BaseKill = BaseOp.isKill();
bool BaseUndef = BaseOp.isUndef();
bool OffKill = isT2 ? false : MI->getOperand(3).isKill();
@@ -1380,6 +1385,7 @@ bool ARMLoadStoreOpt::runOnMachineFunction(MachineFunction &Fn) {
AFI = Fn.getInfo<ARMFunctionInfo>();
TII = TM.getInstrInfo();
TRI = TM.getRegisterInfo();
+ STI = &TM.getSubtarget<ARMSubtarget>();
RS = new RegScavenger();
isThumb2 = AFI->isThumb2Function();
diff --git a/test/CodeGen/ARM/ldrd.ll b/test/CodeGen/ARM/ldrd.ll
index 8010f20689..d72e9bfc1d 100644
--- a/test/CodeGen/ARM/ldrd.ll
+++ b/test/CodeGen/ARM/ldrd.ll
@@ -1,21 +1,22 @@
-; RUN: llc < %s -mtriple=armv6-apple-darwin -regalloc=linearscan | FileCheck %s -check-prefix=V6
-; RUN: llc < %s -mtriple=armv5-apple-darwin -regalloc=linearscan | FileCheck %s -check-prefix=V5
-; RUN: llc < %s -mtriple=armv6-eabi -regalloc=linearscan | FileCheck %s -check-prefix=EABI
-; rdar://r6949835
+; RUN: llc < %s -mtriple=thumbv7-apple-ios -mcpu=cortex-a8 -regalloc=fast | FileCheck %s -check-prefix=A8
+; RUN: llc < %s -mtriple=thumbv7-apple-ios -mcpu=cortex-m3 -regalloc=fast | FileCheck %s -check-prefix=M3
+; rdar://6949835
-; Magic ARM pair hints works best with linearscan.
+; Magic ARM pair hints works best with linearscan / fast.
+
+; Cortex-M3 errata 602117: LDRD with base in list may result in incorrect base
+; register when interrupted or faulted.
@b = external global i64*
define i64 @t(i64 %a) nounwind readonly {
entry:
-;V6: ldrd r2, r3, [r2]
-
-;V5: ldr r{{[0-9]+}}, [r2]
-;V5: ldr r{{[0-9]+}}, [r2, #4]
+; A8: t:
+; A8: ldrd r2, r3, [r2]
-;EABI: ldr r{{[0-9]+}}, [r2]
-;EABI: ldr r{{[0-9]+}}, [r2, #4]
+; M3: t:
+; M3-NOT: ldrd
+; M3: ldm.w r2, {r2, r3}
%0 = load i64** @b, align 4
%1 = load i64* %0, align 4