summaryrefslogtreecommitdiffstats
path: root/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp')
-rw-r--r--llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp16
1 files changed, 14 insertions, 2 deletions
diff --git a/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp b/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp
index cb9d7f7c59d..e7511e6a457 100644
--- a/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp
+++ b/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp
@@ -94,8 +94,15 @@ static bool canShrink(MachineInstr &MI, const SIInstrInfo *TII,
// is vcc. We should handle this the same way we handle vopc, by addding
// a register allocation hint pre-regalloc and then do the shrining
// post-regalloc.
- if (Src2)
- return false;
+ if (Src2) {
+ if (MI.getOpcode() != AMDGPU::V_MAC_F32_e64)
+ return false;
+
+ const MachineOperand *Src2Mod =
+ TII->getNamedOperand(MI, AMDGPU::OpName::src2_modifiers);
+ if (!isVGPR(Src2, TRI, MRI) || (Src2Mod && Src2Mod->getImm() != 0))
+ return false;
+ }
const MachineOperand *Src1 = TII->getNamedOperand(MI, AMDGPU::OpName::src1);
const MachineOperand *Src1Mod =
@@ -259,6 +266,11 @@ bool SIShrinkInstructions::runOnMachineFunction(MachineFunction &MF) {
if (Src1)
Inst32.addOperand(*Src1);
+ const MachineOperand *Src2 =
+ TII->getNamedOperand(MI, AMDGPU::OpName::src2);
+ if (Src2)
+ Inst32.addOperand(*Src2);
+
++NumInstructionsShrunk;
MI.eraseFromParent();
OpenPOWER on IntegriCloud