6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2024-07-08 18:30:04 +00:00
Code Issues Projects Releases Wiki Activity

R600/SI: Initial support for LDS/GDS instructions

Browse Source 
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@186009 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Michel Danzer 2013-07-10 16:36:43 +00:00
parent 0a9aaacd72
commit 7740daa8ba
5 changed files with 68 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
lib/Target/R600/SIInsertWaits.cpp
View File

@ -134,6 +134,8 @@ Counters SIInsertWaits::getHwCounts(MachineInstr &MI) {
if (TSFlags & SIInstrFlags::LGKM_CNT) {
MachineOperand &Op = MI.getOperand(0);
if (!Op.isReg())
Op = MI.getOperand(1);
assert(Op.isReg() && "First LGKM operand must be a register!");
unsigned Reg = Op.getReg();

24
lib/Target/R600/SIInstrFormats.td
View File

@ -281,6 +281,30 @@ class VINTRP <bits <2> op, dag outs, dag ins, string asm, list<dag> pattern> :
let Uses = [EXEC] in {
class DS <bits<8> op, dag outs, dag ins, string asm, list<dag> pattern> :
Enc64 <outs, ins, asm, pattern> {
bits<8> vdst;
bits<1> gds;
bits<8> addr;
bits<8> data0;
bits<8> data1;
bits<8> offset0;
bits<8> offset1;
let Inst{7-0} = offset0;
let Inst{15-8} = offset1;
let Inst{17} = gds;
let Inst{25-18} = op;
let Inst{31-26} = 0x36; //encoding
let Inst{39-32} = addr;
let Inst{47-40} = data0;
let Inst{55-48} = data1;
let Inst{63-56} = vdst;
let LGKM_CNT = 1;
}
class MUBUF <bits<7> op, dag outs, dag ins, string asm, list<dag> pattern> :
Enc64<outs, ins, asm, pattern> {

23
lib/Target/R600/SIInstrInfo.td
View File

@ -287,6 +287,29 @@ class VOP3_64 <bits<9> op, string opName, list<dag> pattern> : VOP3 <
// Vector I/O classes
//===----------------------------------------------------------------------===//
class DS_Load_Helper <bits<8> op, string asm, RegisterClass regClass> : DS <
op,
(outs regClass:$vdst),
(ins i1imm:$gds, VReg_32:$addr, VReg_32:$data0, VReg_32:$data1,
i8imm:$offset0, i8imm:$offset1),
asm#" $vdst, $gds, $addr, $data0, $data1, $offset0, $offset1, [M0]",
[]> {
let mayLoad = 1;
let mayStore = 0;
}
class DS_Store_Helper <bits<8> op, string asm, RegisterClass regClass> : DS <
op,
(outs),
(ins i1imm:$gds, VReg_32:$addr, VReg_32:$data0, VReg_32:$data1,
i8imm:$offset0, i8imm:$offset1),
asm#" $gds, $addr, $data0, $data1, $offset0, $offset1, [M0]",
[]> {
let mayStore = 1;
let mayLoad = 0;
let vdst = 0;
}
class MTBUF_Store_Helper <bits<3> op, string asm, RegisterClass regClass> : MTBUF <
op,
(outs),

3
lib/Target/R600/SIInstructions.td
View File

@ -391,6 +391,9 @@ defm V_CMPX_CLASS_F64 : VOPC_64 <0x000000b8, "V_CMPX_CLASS_F64">;
} // End isCompare = 1
def DS_WRITE_B32 : DS_Store_Helper <0x0000000d, "DS_WRITE_B32", VReg_32>;
def DS_READ_B32 : DS_Load_Helper <0x00000036, "DS_READ_B32", VReg_32>;
//def BUFFER_LOAD_FORMAT_X : MUBUF_ <0x00000000, "BUFFER_LOAD_FORMAT_X", []>;
//def BUFFER_LOAD_FORMAT_XY : MUBUF_ <0x00000001, "BUFFER_LOAD_FORMAT_XY", []>;
//def BUFFER_LOAD_FORMAT_XYZ : MUBUF_ <0x00000002, "BUFFER_LOAD_FORMAT_XYZ", []>;

16
lib/Target/R600/SILowerControlFlow.cpp
View File

@ -411,6 +411,7 @@ bool SILowerControlFlowPass::runOnMachineFunction(MachineFunction &MF) {
TRI = MF.getTarget().getRegisterInfo();
bool HaveKill = false;
bool NeedM0 = false;
bool NeedWQM = false;
unsigned Depth = 0;
@ -482,6 +483,13 @@ bool SILowerControlFlowPass::runOnMachineFunction(MachineFunction &MF) {
IndirectDst(MI);
break;
case AMDGPU::DS_READ_B32:
NeedWQM = true;
// Fall through
case AMDGPU::DS_WRITE_B32:
NeedM0 = true;
break;
case AMDGPU::V_INTERP_P1_F32:
case AMDGPU::V_INTERP_P2_F32:
case AMDGPU::V_INTERP_MOV_F32:
@ -492,6 +500,14 @@ bool SILowerControlFlowPass::runOnMachineFunction(MachineFunction &MF) {
}
}
if (NeedM0) {
MachineBasicBlock &MBB = MF.front();
// Initialize M0 to a value that won't cause LDS access to be discarded
// due to offset clamping
BuildMI(MBB, MBB.getFirstNonPHI(), DebugLoc(), TII->get(AMDGPU::S_MOV_B32),
AMDGPU::M0).addImm(0xffffffff);
}
if (NeedWQM) {
MachineBasicBlock &MBB = MF.front();
BuildMI(MBB, MBB.getFirstNonPHI(), DebugLoc(), TII->get(AMDGPU::S_WQM_B64),