aboutsummaryrefslogtreecommitdiffstats
path: root/lib/Target
diff options
context:
space:
mode:
authorVincent Lejeune <vljn@ovi.com>2013-11-16 16:24:41 +0000
committerVincent Lejeune <vljn@ovi.com>2013-11-16 16:24:41 +0000
commit411079785388290738049dd099bff8755e6a2c8d (patch)
treee79a958d491354edc7924d409e6b31962f9f063a /lib/Target
parent0fb32eb56aabb24950bbfecc927596a3fffabcb1 (diff)
downloadexternal_llvm-411079785388290738049dd099bff8755e6a2c8d.zip
external_llvm-411079785388290738049dd099bff8755e6a2c8d.tar.gz
external_llvm-411079785388290738049dd099bff8755e6a2c8d.tar.bz2
R600: Make dot_4 instructions predicable
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194927 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'lib/Target')
-rw-r--r--lib/Target/R600/R600InstrInfo.cpp19
1 files changed, 19 insertions, 0 deletions
diff --git a/lib/Target/R600/R600InstrInfo.cpp b/lib/Target/R600/R600InstrInfo.cpp
index 8436d5f..1f47416 100644
--- a/lib/Target/R600/R600InstrInfo.cpp
+++ b/lib/Target/R600/R600InstrInfo.cpp
@@ -1009,6 +1009,20 @@ R600InstrInfo::PredicateInstruction(MachineInstr *MI,
return true;
}
+ if (MI->getOpcode() == AMDGPU::DOT_4) {
+ MI->getOperand(getOperandIdx(*MI, AMDGPU::OpName::pred_sel_X))
+ .setReg(Pred[2].getReg());
+ MI->getOperand(getOperandIdx(*MI, AMDGPU::OpName::pred_sel_Y))
+ .setReg(Pred[2].getReg());
+ MI->getOperand(getOperandIdx(*MI, AMDGPU::OpName::pred_sel_Z))
+ .setReg(Pred[2].getReg());
+ MI->getOperand(getOperandIdx(*MI, AMDGPU::OpName::pred_sel_W))
+ .setReg(Pred[2].getReg());
+ MachineInstrBuilder MIB(*MI->getParent()->getParent(), MI);
+ MIB.addReg(AMDGPU::PREDICATE_BIT, RegState::Implicit);
+ return true;
+ }
+
if (PIdx != -1) {
MachineOperand &PMO = MI->getOperand(PIdx);
PMO.setReg(Pred[2].getReg());
@@ -1217,6 +1231,11 @@ MachineInstr *R600InstrInfo::buildSlotOfVectorInstruction(
AMDGPU::OpName::src1_sel,
};
+ MachineOperand &MO = MI->getOperand(getOperandIdx(MI->getOpcode(),
+ getSlotedOps(AMDGPU::OpName::pred_sel, Slot)));
+ MIB->getOperand(getOperandIdx(Opcode, AMDGPU::OpName::pred_sel))
+ .setReg(MO.getReg());
+
for (unsigned i = 0; i < 14; i++) {
MachineOperand &MO = MI->getOperand(
getOperandIdx(MI->getOpcode(), getSlotedOps(Operands[i], Slot)));