ARM has a peephole optimization which looks for a def / use pair. The def

produces a 32-bit immediate which is consumed by the use. It tries to 
fold the immediate by breaking it into two parts and fold them into the
immmediate fields of two uses. e.g
       movw    r2, #40885
       movt    r3, #46540
       add     r0, r0, r3
=>
       add.w   r0, r0, #3019898880
       add.w   r0, r0, #30146560
;
However, this transformation is incorrect if the user produces a flag. e.g.
       movw    r2, #40885
       movt    r3, #46540
       adds    r0, r0, r3
=>
       add.w   r0, r0, #3019898880
       adds.w  r0, r0, #30146560
Note the adds.w may not set the carry flag even if the original sequence
would.

rdar://11116189

llvm-svn: 153484

2 files changed, 52 insertions, 0 deletions

diff --git a/llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp b/llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp
index 6c97261cf90..17ff7192b1e 100644
--- a/llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp
+++ b/llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp
@@ -1916,6 +1916,25 @@ bool ARMBaseInstrInfo::FoldImmediate(MachineInstr *UseMI,
   if (!MRI->hasOneNonDBGUse(Reg))
     return false;
 
+  const MCInstrDesc &DefMCID = DefMI->getDesc();
+  if (DefMCID.hasOptionalDef()) {
+    unsigned NumOps = DefMCID.getNumOperands();
+    const MachineOperand &MO = DefMI->getOperand(NumOps-1);
+    if (MO.getReg() == ARM::CPSR && !MO.isDead())
+      // If DefMI defines CPSR and it is not dead, it's obviously not safe
+      // to delete DefMI.
+      return false;
+  }
+
+  const MCInstrDesc &UseMCID = UseMI->getDesc();
+  if (UseMCID.hasOptionalDef()) {
+    unsigned NumOps = UseMCID.getNumOperands();
+    if (UseMI->getOperand(NumOps-1).getReg() == ARM::CPSR)
+      // If the instruction sets the flag, do not attempt this optimization
+      // since it may change the semantics of the code.
+      return false;
+  }
+
   unsigned UseOpc = UseMI->getOpcode();
   unsigned NewUseOpc = 0;
   uint32_t ImmVal = (uint32_t)DefMI->getOperand(1).getImm();
diff --git a/llvm/test/CodeGen/ARM/2012-03-26-FoldImmBug.ll b/llvm/test/CodeGen/ARM/2012-03-26-FoldImmBug.ll
new file mode 100644
index 00000000000..0ff4f510eb3
--- /dev/null
+++ b/llvm/test/CodeGen/ARM/2012-03-26-FoldImmBug.ll
@@ -0,0 +1,33 @@
+; RUN: llc < %s -mtriple=thumbv7-apple-ios -mcpu=cortex-a8 | FileCheck %s
+
+; ARM has a peephole optimization which looks for a def / use pair. The def
+; produces a 32-bit immediate which is consumed by the use. It tries to 
+; fold the immediate by breaking it into two parts and fold them into the
+; immmediate fields of two uses. e.g
+;        movw    r2, #40885
+;        movt    r3, #46540
+;        add     r0, r0, r3
+; =>
+;        add.w   r0, r0, #3019898880
+;        add.w   r0, r0, #30146560
+;
+; However, this transformation is incorrect if the user produces a flag. e.g.
+;        movw    r2, #40885
+;        movt    r3, #46540
+;        adds    r0, r0, r3
+; =>
+;        add.w   r0, r0, #3019898880
+;        adds.w  r0, r0, #30146560
+; Note the adds.w may not set the carry flag even if the original sequence
+; would.
+;
+; rdar://11116189
+define i64 @t(i64 %aInput) nounwind {
+; CHECK: t:
+; CHECK: movs [[REG:(r[0-9]+)]], #0
+; CHECK: movt [[REG]], #46540
+; CHECK: adds r{{[0-9]+}}, r{{[0-9]+}}, [[REG]]
+  %1 = mul i64 %aInput, 1000000
+  %2 = add i64 %1, -7952618389194932224
+  ret i64 %2
+}