aboutsummaryrefslogtreecommitdiff
path: root/lib/Target/R600/MCTargetDesc/R600MCCodeEmitter.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'lib/Target/R600/MCTargetDesc/R600MCCodeEmitter.cpp')
-rw-r--r--lib/Target/R600/MCTargetDesc/R600MCCodeEmitter.cpp201
1 files changed, 101 insertions, 100 deletions
diff --git a/lib/Target/R600/MCTargetDesc/R600MCCodeEmitter.cpp b/lib/Target/R600/MCTargetDesc/R600MCCodeEmitter.cpp
index d20716000d..6ef4d40934 100644
--- a/lib/Target/R600/MCTargetDesc/R600MCCodeEmitter.cpp
+++ b/lib/Target/R600/MCTargetDesc/R600MCCodeEmitter.cpp
@@ -66,8 +66,6 @@ private:
void EmitSrcISA(const MCInst &MI, unsigned RegOpIdx, unsigned SelOpIdx,
raw_ostream &OS) const;
void EmitDst(const MCInst &MI, raw_ostream &OS) const;
- void EmitTexInstr(const MCInst &MI, SmallVectorImpl<MCFixup> &Fixups,
- raw_ostream &OS) const;
void EmitFCInstr(const MCInst &MI, raw_ostream &OS) const;
void EmitNullBytes(unsigned int byteCount, raw_ostream &OS) const;
@@ -103,7 +101,8 @@ enum InstrTypes {
INSTR_FC,
INSTR_NATIVE,
INSTR_VTX,
- INSTR_EXPORT
+ INSTR_EXPORT,
+ INSTR_CFALU
};
enum FCInstr {
@@ -140,9 +139,7 @@ MCCodeEmitter *llvm::createR600MCCodeEmitter(const MCInstrInfo &MCII,
void R600MCCodeEmitter::EncodeInstruction(const MCInst &MI, raw_ostream &OS,
SmallVectorImpl<MCFixup> &Fixups) const {
- if (isTexOp(MI.getOpcode())) {
- EmitTexInstr(MI, Fixups, OS);
- } else if (isFCOp(MI.getOpcode())){
+ if (isFCOp(MI.getOpcode())){
EmitFCInstr(MI, OS);
} else if (MI.getOpcode() == AMDGPU::RETURN ||
MI.getOpcode() == AMDGPU::BUNDLE ||
@@ -150,6 +147,10 @@ void R600MCCodeEmitter::EncodeInstruction(const MCInst &MI, raw_ostream &OS,
return;
} else {
switch(MI.getOpcode()) {
+ case AMDGPU::STACK_SIZE: {
+ EmitByte(MI.getOperand(0).getImm(), OS);
+ break;
+ }
case AMDGPU::RAT_WRITE_CACHELESS_32_eg:
case AMDGPU::RAT_WRITE_CACHELESS_128_eg: {
uint64_t inst = getBinaryCodeForInstr(MI, Fixups);
@@ -175,6 +176,77 @@ void R600MCCodeEmitter::EncodeInstruction(const MCInst &MI, raw_ostream &OS,
Emit(InstWord2, OS);
break;
}
+ case AMDGPU::TEX_LD:
+ case AMDGPU::TEX_GET_TEXTURE_RESINFO:
+ case AMDGPU::TEX_SAMPLE:
+ case AMDGPU::TEX_SAMPLE_C:
+ case AMDGPU::TEX_SAMPLE_L:
+ case AMDGPU::TEX_SAMPLE_C_L:
+ case AMDGPU::TEX_SAMPLE_LB:
+ case AMDGPU::TEX_SAMPLE_C_LB:
+ case AMDGPU::TEX_SAMPLE_G:
+ case AMDGPU::TEX_SAMPLE_C_G:
+ case AMDGPU::TEX_GET_GRADIENTS_H:
+ case AMDGPU::TEX_GET_GRADIENTS_V:
+ case AMDGPU::TEX_SET_GRADIENTS_H:
+ case AMDGPU::TEX_SET_GRADIENTS_V: {
+ unsigned Opcode = MI.getOpcode();
+ bool HasOffsets = (Opcode == AMDGPU::TEX_LD);
+ unsigned OpOffset = HasOffsets ? 3 : 0;
+ int64_t Sampler = MI.getOperand(OpOffset + 3).getImm();
+ int64_t TextureType = MI.getOperand(OpOffset + 4).getImm();
+
+ uint32_t SrcSelect[4] = {0, 1, 2, 3};
+ uint32_t Offsets[3] = {0, 0, 0};
+ uint64_t CoordType[4] = {1, 1, 1, 1};
+
+ if (HasOffsets)
+ for (unsigned i = 0; i < 3; i++)
+ Offsets[i] = MI.getOperand(i + 2).getImm();
+
+ if (TextureType == TEXTURE_RECT ||
+ TextureType == TEXTURE_SHADOWRECT) {
+ CoordType[ELEMENT_X] = 0;
+ CoordType[ELEMENT_Y] = 0;
+ }
+
+ if (TextureType == TEXTURE_1D_ARRAY ||
+ TextureType == TEXTURE_SHADOW1D_ARRAY) {
+ if (Opcode == AMDGPU::TEX_SAMPLE_C_L ||
+ Opcode == AMDGPU::TEX_SAMPLE_C_LB) {
+ CoordType[ELEMENT_Y] = 0;
+ } else {
+ CoordType[ELEMENT_Z] = 0;
+ SrcSelect[ELEMENT_Z] = ELEMENT_Y;
+ }
+ } else if (TextureType == TEXTURE_2D_ARRAY ||
+ TextureType == TEXTURE_SHADOW2D_ARRAY) {
+ CoordType[ELEMENT_Z] = 0;
+ }
+
+
+ if ((TextureType == TEXTURE_SHADOW1D ||
+ TextureType == TEXTURE_SHADOW2D ||
+ TextureType == TEXTURE_SHADOWRECT ||
+ TextureType == TEXTURE_SHADOW1D_ARRAY) &&
+ Opcode != AMDGPU::TEX_SAMPLE_C_L &&
+ Opcode != AMDGPU::TEX_SAMPLE_C_LB) {
+ SrcSelect[ELEMENT_W] = ELEMENT_Z;
+ }
+
+ uint64_t Word01 = getBinaryCodeForInstr(MI, Fixups) |
+ CoordType[ELEMENT_X] << 60 | CoordType[ELEMENT_Y] << 61 |
+ CoordType[ELEMENT_Z] << 62 | CoordType[ELEMENT_W] << 63;
+ uint32_t Word2 = Sampler << 15 | SrcSelect[ELEMENT_X] << 20 |
+ SrcSelect[ELEMENT_Y] << 23 | SrcSelect[ELEMENT_Z] << 26 |
+ SrcSelect[ELEMENT_W] << 29 | Offsets[0] << 0 | Offsets[1] << 5 |
+ Offsets[2] << 10;
+
+ EmitByte(INSTR_TEX, OS);
+ Emit(Word01, OS);
+ Emit(Word2, OS);
+ break;
+ }
case AMDGPU::EG_ExportSwz:
case AMDGPU::R600_ExportSwz:
case AMDGPU::EG_ExportBuf:
@@ -184,7 +256,29 @@ void R600MCCodeEmitter::EncodeInstruction(const MCInst &MI, raw_ostream &OS,
Emit(Inst, OS);
break;
}
-
+ case AMDGPU::CF_ALU:
+ case AMDGPU::CF_ALU_PUSH_BEFORE: {
+ uint64_t Inst = getBinaryCodeForInstr(MI, Fixups);
+ EmitByte(INSTR_CFALU, OS);
+ Emit(Inst, OS);
+ break;
+ }
+ case AMDGPU::CF_TC:
+ case AMDGPU::CF_VC:
+ case AMDGPU::CF_CALL_FS:
+ return;
+ case AMDGPU::WHILE_LOOP:
+ case AMDGPU::END_LOOP:
+ case AMDGPU::LOOP_BREAK:
+ case AMDGPU::CF_CONTINUE:
+ case AMDGPU::CF_JUMP:
+ case AMDGPU::CF_ELSE:
+ case AMDGPU::POP: {
+ uint64_t Inst = getBinaryCodeForInstr(MI, Fixups);
+ EmitByte(INSTR_NATIVE, OS);
+ Emit(Inst, OS);
+ break;
+ }
default:
EmitALUInstr(MI, Fixups, OS);
break;
@@ -334,99 +428,6 @@ void R600MCCodeEmitter::EmitSrcISA(const MCInst &MI, unsigned RegOpIdx,
Emit(InlineConstant.i, OS);
}
-void R600MCCodeEmitter::EmitTexInstr(const MCInst &MI,
- SmallVectorImpl<MCFixup> &Fixups,
- raw_ostream &OS) const {
-
- unsigned Opcode = MI.getOpcode();
- bool hasOffsets = (Opcode == AMDGPU::TEX_LD);
- unsigned OpOffset = hasOffsets ? 3 : 0;
- int64_t Resource = MI.getOperand(OpOffset + 2).getImm();
- int64_t Sampler = MI.getOperand(OpOffset + 3).getImm();
- int64_t TextureType = MI.getOperand(OpOffset + 4).getImm();
- unsigned srcSelect[4] = {0, 1, 2, 3};
-
- // Emit instruction type
- EmitByte(1, OS);
-
- // Emit instruction
- EmitByte(getBinaryCodeForInstr(MI, Fixups), OS);
-
- // Emit resource id
- EmitByte(Resource, OS);
-
- // Emit source register
- EmitByte(getHWReg(MI.getOperand(1).getReg()), OS);
-
- // XXX: Emit src isRelativeAddress
- EmitByte(0, OS);
-
- // Emit destination register
- EmitByte(getHWReg(MI.getOperand(0).getReg()), OS);
-
- // XXX: Emit dst isRealtiveAddress
- EmitByte(0, OS);
-
- // XXX: Emit dst select
- EmitByte(0, OS); // X
- EmitByte(1, OS); // Y
- EmitByte(2, OS); // Z
- EmitByte(3, OS); // W
-
- // XXX: Emit lod bias
- EmitByte(0, OS);
-
- // XXX: Emit coord types
- unsigned coordType[4] = {1, 1, 1, 1};
-
- if (TextureType == TEXTURE_RECT
- || TextureType == TEXTURE_SHADOWRECT) {
- coordType[ELEMENT_X] = 0;
- coordType[ELEMENT_Y] = 0;
- }
-
- if (TextureType == TEXTURE_1D_ARRAY
- || TextureType == TEXTURE_SHADOW1D_ARRAY) {
- if (Opcode == AMDGPU::TEX_SAMPLE_C_L || Opcode == AMDGPU::TEX_SAMPLE_C_LB) {
- coordType[ELEMENT_Y] = 0;
- } else {
- coordType[ELEMENT_Z] = 0;
- srcSelect[ELEMENT_Z] = ELEMENT_Y;
- }
- } else if (TextureType == TEXTURE_2D_ARRAY
- || TextureType == TEXTURE_SHADOW2D_ARRAY) {
- coordType[ELEMENT_Z] = 0;
- }
-
- for (unsigned i = 0; i < 4; i++) {
- EmitByte(coordType[i], OS);
- }
-
- // XXX: Emit offsets
- if (hasOffsets)
- for (unsigned i = 2; i < 5; i++)
- EmitByte(MI.getOperand(i).getImm()<<1, OS);
- else
- EmitNullBytes(3, OS);
-
- // Emit sampler id
- EmitByte(Sampler, OS);
-
- // XXX:Emit source select
- if ((TextureType == TEXTURE_SHADOW1D
- || TextureType == TEXTURE_SHADOW2D
- || TextureType == TEXTURE_SHADOWRECT
- || TextureType == TEXTURE_SHADOW1D_ARRAY)
- && Opcode != AMDGPU::TEX_SAMPLE_C_L
- && Opcode != AMDGPU::TEX_SAMPLE_C_LB) {
- srcSelect[ELEMENT_W] = ELEMENT_Z;
- }
-
- for (unsigned i = 0; i < 4; i++) {
- EmitByte(srcSelect[i], OS);
- }
-}
-
void R600MCCodeEmitter::EmitFCInstr(const MCInst &MI, raw_ostream &OS) const {
// Emit instruction type
generated by cgit v1.2.3-18-g5258 (git 2.32.0) at 2024-11-21 16:41:01 +0000