radeon/llvm: Fix encoding of V_CNDMASK_B32

The CodeEmitter was not setting the VGPR bit for src0, because the
instruction definition had the VCC register in the src0 slot, instead of
the actual src0 register.  This has been fixed by moving the VCC
register to the end of the operand list.
This commit is contained in:
Tom Stellard 2012-08-31 16:11:38 -04:00
parent f73ffacbf0
commit f9fede884b
2 changed files with 4 additions and 4 deletions

View file

@ -242,9 +242,9 @@ void SITargetLowering::LowerSI_V_CNDLT(MachineInstr *MI, MachineBasicBlock &BB,
BuildMI(BB, I, BB.findDebugLoc(I), TII->get(AMDGPU::V_CNDMASK_B32))
.addOperand(MI->getOperand(0))
.addReg(AMDGPU::VCC)
.addOperand(MI->getOperand(2))
.addOperand(MI->getOperand(3));
.addOperand(MI->getOperand(3))
.addReg(AMDGPU::VCC);
MI->eraseFromParent();
}

View file

@ -670,7 +670,7 @@ def S_WAITCNT : SOPP <0x0000000c, (ins i32imm:$simm16), "S_WAITCNT $simm16",
/* XXX: No VOP3 version of this instruction yet */
def V_CNDMASK_B32 : VOP2 <0x00000000, (outs VReg_32:$dst),
(ins VCCReg:$vcc, AllReg_32:$src0, VReg_32:$src1), "V_CNDMASK_B32",
(ins AllReg_32:$src0, VReg_32:$src1, VCCReg:$vcc), "V_CNDMASK_B32",
[(set (i32 VReg_32:$dst),
(select VCCReg:$vcc, AllReg_32:$src0, VReg_32:$src1))] > {
@ -680,7 +680,7 @@ def V_CNDMASK_B32 : VOP2 <0x00000000, (outs VReg_32:$dst),
//f32 pattern for V_CNDMASK_B32
def : Pat <
(f32 (select VCCReg:$vcc, AllReg_32:$src0, VReg_32:$src1)),
(V_CNDMASK_B32 VCCReg:$vcc, AllReg_32:$src0, VReg_32:$src1)
(V_CNDMASK_B32 AllReg_32:$src0, VReg_32:$src1, VCCReg:$vcc)
>;
defm V_READLANE_B32 : VOP2_32 <0x00000001, "V_READLANE_B32", []>;