6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2025-11-07 12:18:32 +00:00
Code Issues Projects Releases Wiki Activity

R600/SI: Remove i1 pseudo VALU ops

Browse Source 
Select i1 logical ops directly to 64-bit SALU instructions.
Vector i1 values are always really in SGPRs, with each
bit for each item in the wave. This saves about 4 instructions
when and/or/xoring any condition, and also helps write conditions
that need to be passed in vcc.

This should work correctly now that the SGPR live range
fixing pass works. More work is needed to eliminate the VReg_1
pseudo regclass and possibly the entire SILowerI1Copies pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223206 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Matt Arsenault
2014-12-03 05:22:35 +00:00
parent bd5f9f45d1
commit ec0a7cd15a
10 changed files with 339 additions and 124 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
lib/Target/R600/SIInstrInfo.td
View File

@@ -131,6 +131,10 @@ def as_i32imm: SDNodeXForm<imm, [{
return CurDAG->getTargetConstant(N->getSExtValue(), MVT::i32);
}]>;
def as_i64imm: SDNodeXForm<imm, [{
return CurDAG->getTargetConstant(N->getSExtValue(), MVT::i64);
}]>;
def IMM8bit : PatLeaf <(imm),
[{return isUInt<8>(N->getZExtValue());}]
>;

53
lib/Target/R600/SIInstructions.td
View File

@@ -1686,30 +1686,8 @@ defm V_TRIG_PREOP_F64 : VOP3Inst <
//===----------------------------------------------------------------------===//
// Pseudo Instructions
//===----------------------------------------------------------------------===//
let isCodeGenOnly = 1, isPseudo = 1 in {
def V_MOV_I1 : InstSI <
(outs VReg_1:$dst),
(ins i1imm:$src),
"", [(set i1:$dst, (imm:$src))]
>;
def V_AND_I1 : InstSI <
(outs VReg_1:$dst), (ins VReg_1:$src0, VReg_1:$src1), "",
[(set i1:$dst, (and i1:$src0, i1:$src1))]
>;
def V_OR_I1 : InstSI <
(outs VReg_1:$dst), (ins VReg_1:$src0, VReg_1:$src1), "",
[(set i1:$dst, (or i1:$src0, i1:$src1))]
>;
def V_XOR_I1 : InstSI <
(outs VReg_1:$dst), (ins VReg_1:$src0, VReg_1:$src1), "",
[(set i1:$dst, (xor i1:$src0, i1:$src1))]
>;
let hasSideEffects = 1 in {
def SGPR_USE : InstSI <(outs),(ins), "", []>;
}
@@ -2495,6 +2473,14 @@ def : Pat <
(S_MOV_B64 InlineImm<i64>:$imm)
>;
// XXX - Should this use a s_cmp to set SCC?
// Set to sign-extended 64-bit value (true = -1, false = 0)
def : Pat <
(i1 imm:$imm),
(S_MOV_B64 (i64 (as_i64imm $imm)))
>;
/********** ===================== **********/
/********** Interpolation Paterns **********/
/********** ===================== **********/
@@ -3045,6 +3031,27 @@ def : Pat <
(V_CNDMASK_B32_e64 0, -1, $src), sub1)
>;
// If we need to perform a logical operation on i1 values, we need to
// use vector comparisons since there is only one SCC register. Vector
// comparisions still write to a pair of SGPRs, so treat these as
// 64-bit comparisons. When legalizing SGPR copies, instructions
// resulting in the copies from SCC to these instructions will be
// moved to the VALU.
def : Pat <
(i1 (and i1:$src0, i1:$src1)),
(S_AND_B64 $src0, $src1)
>;
def : Pat <
(i1 (or i1:$src0, i1:$src1)),
(S_OR_B64 $src0, $src1)
>;
def : Pat <
(i1 (xor i1:$src0, i1:$src1)),
(S_XOR_B64 $src0, $src1)
>;
def : Pat <
(f32 (sint_to_fp i1:$src)),
(V_CNDMASK_B32_e64 (i32 0), CONST.FP32_NEG_ONE, $src)
@@ -3057,7 +3064,7 @@ def : Pat <
def : Pat <
(f64 (sint_to_fp i1:$src)),
(V_CVT_F64_I32_e32 (V_CNDMASK_B32_e64 (i32 0), (i32 -1), $src))
(V_CVT_F64_I32_e32 (V_CNDMASK_B32_e64 (i32 0), (i32 -1), $src))
>;
def : Pat <

76
lib/Target/R600/SILowerI1Copies.cpp
View File

@@ -85,30 +85,6 @@ bool SILowerI1Copies::runOnMachineFunction(MachineFunction &MF) {
Next = std::next(I);
MachineInstr &MI = *I;
if (MI.getOpcode() == AMDGPU::V_MOV_I1) {
I1Defs.push_back(MI.getOperand(0).getReg());
MI.setDesc(TII->get(AMDGPU::V_MOV_B32_e32));
continue;
}
if (MI.getOpcode() == AMDGPU::V_AND_I1) {
I1Defs.push_back(MI.getOperand(0).getReg());
MI.setDesc(TII->get(AMDGPU::V_AND_B32_e64));
continue;
}
if (MI.getOpcode() == AMDGPU::V_OR_I1) {
I1Defs.push_back(MI.getOperand(0).getReg());
MI.setDesc(TII->get(AMDGPU::V_OR_B32_e64));
continue;
}
if (MI.getOpcode() == AMDGPU::V_XOR_I1) {
I1Defs.push_back(MI.getOperand(0).getReg());
MI.setDesc(TII->get(AMDGPU::V_XOR_B32_e64));
continue;
}
if (MI.getOpcode() == AMDGPU::IMPLICIT_DEF) {
unsigned Reg = MI.getOperand(0).getReg();
const TargetRegisterClass *RC = MRI.getRegClass(Reg);
@@ -117,32 +93,52 @@ bool SILowerI1Copies::runOnMachineFunction(MachineFunction &MF) {
continue;
}
if (MI.getOpcode() != AMDGPU::COPY ||
!TargetRegisterInfo::isVirtualRegister(MI.getOperand(0).getReg()) ||
!TargetRegisterInfo::isVirtualRegister(MI.getOperand(1).getReg()))
if (MI.getOpcode() != AMDGPU::COPY)
continue;
const MachineOperand &Dst = MI.getOperand(0);
const MachineOperand &Src = MI.getOperand(1);
const TargetRegisterClass *DstRC =
MRI.getRegClass(MI.getOperand(0).getReg());
const TargetRegisterClass *SrcRC =
MRI.getRegClass(MI.getOperand(1).getReg());
if (!TargetRegisterInfo::isVirtualRegister(Src.getReg()) ||
!TargetRegisterInfo::isVirtualRegister(Dst.getReg()))
continue;
const TargetRegisterClass *DstRC = MRI.getRegClass(Dst.getReg());
const TargetRegisterClass *SrcRC = MRI.getRegClass(Src.getReg());
if (DstRC == &AMDGPU::VReg_1RegClass &&
TRI->getCommonSubClass(SrcRC, &AMDGPU::SGPR_64RegClass)) {
I1Defs.push_back(MI.getOperand(0).getReg());
BuildMI(MBB, &MI, MI.getDebugLoc(), TII->get(AMDGPU::V_CNDMASK_B32_e64))
.addOperand(MI.getOperand(0))
.addImm(0)
.addImm(-1)
.addOperand(MI.getOperand(1));
I1Defs.push_back(Dst.getReg());
DebugLoc DL = MI.getDebugLoc();
MachineInstr *DefInst = MRI.getUniqueVRegDef(Src.getReg());
if (DefInst->getOpcode() == AMDGPU::S_MOV_B64) {
if (DefInst->getOperand(1).isImm()) {
I1Defs.push_back(Dst.getReg());
int64_t Val = DefInst->getOperand(1).getImm();
assert(Val == 0 || Val == -1);
BuildMI(MBB, &MI, DL, TII->get(AMDGPU::V_MOV_B32_e32))
.addOperand(Dst)
.addImm(Val);
MI.eraseFromParent();
continue;
}
}
BuildMI(MBB, &MI, DL, TII->get(AMDGPU::V_CNDMASK_B32_e64))
.addOperand(Dst)
.addImm(0)
.addImm(-1)
.addOperand(Src);
MI.eraseFromParent();
} else if (TRI->getCommonSubClass(DstRC, &AMDGPU::SGPR_64RegClass) &&
SrcRC == &AMDGPU::VReg_1RegClass) {
BuildMI(MBB, &MI, MI.getDebugLoc(), TII->get(AMDGPU::V_CMP_NE_I32_e64))
.addOperand(MI.getOperand(0))
.addOperand(MI.getOperand(1))
.addImm(0);
.addOperand(Dst)
.addOperand(Src)
.addImm(0);
MI.eraseFromParent();
}
}