void Vc4Shader::Emit_Mov(CInstruction &Inst)
{
assert(this->uShaderType == D3D10_SB_PIXEL_SHADER ||
this->uShaderType == D3D10_SB_VERTEX_SHADER);
VC4_ASSERT(Inst.m_NumOperands == 2);
{
for (uint8_t i = 0, aCurrent = D3D10_SB_OPERAND_4_COMPONENT_MASK_X; i < 4; i++)
{
if (Inst.m_Operands[0].m_WriteMask & aCurrent)
{
Vc4Register dst = Find_Vc4Register_M(Inst.m_Operands[0], aCurrent);
Vc4Register src[1];
Setup_SourceRegisters(Inst, 1, ARRAYSIZE(src), i , src);
{
Vc4Instruction Vc4Inst;
Vc4Inst.Vc4_m_MOV(dst, src[0]);
Vc4Inst.Vc4_m_Pack(dst.GetPack(i));
Vc4Inst.Emit(CurrentStorage);
}
}
aCurrent <<= 1;
}
}
{ // Emit a NOP
Vc4Instruction Vc4Inst;
Vc4Inst.Emit(CurrentStorage);
}
}
void Vc4Shader::Emit_with_Mul_pipe(CInstruction &Inst)
{
assert(this->uShaderType == D3D10_SB_PIXEL_SHADER ||
this->uShaderType == D3D10_SB_VERTEX_SHADER);
VC4_ASSERT(Inst.m_NumOperands == 3);
{
for (uint8_t i = 0, aCurrent = D3D10_SB_OPERAND_4_COMPONENT_MASK_X; i < 4; i++)
{
if (Inst.m_Operands[0].m_WriteMask & aCurrent)
{
Vc4Register dst = Find_Vc4Register_M(Inst.m_Operands[0], aCurrent);
uint8_t pack = VC4_QPU_PACK_A_32;
if (dst.GetFlags().packed)
{
pack = VC4_QPU_PACK_MUL_8a + i;
}
Vc4Register src[2];
this->Setup_SourceRegisters(Inst, 1, ARRAYSIZE(src), i, src);
{
Vc4Instruction Vc4Inst;
switch (Inst.m_OpCode)
{
case D3D10_SB_OPCODE_MUL:
Vc4Inst.Vc4_m_FMUL(dst, src[0], src[1]);
break;
default:
VC4_ASSERT(false);
}
Vc4Inst.Vc4_m_Pack(pack);
Vc4Inst.Emit(CurrentStorage);
}
}
aCurrent <<= 1;
}
}
{ // Emit a NOP
Vc4Instruction Vc4Inst;
Vc4Inst.Emit(CurrentStorage);
}
}
HRESULT Vc4Shader::Translate_VS()
{
assert(this->uShaderType == D3D10_SB_VERTEX_SHADER);
this->SetCurrentStorage(this->ShaderStorage, this->ShaderUniform);
this->HLSL_ParseDecl();
this->HLSL_Link_PS();
this->Emit_Prologue_VS();
{
CInstruction Inst;
assert(Inst.m_bSaturate == false); // saturate is not supported.
while (HLSL_GetShaderInstruction(this->HLSLParser, Inst))
{
// Need to add support for D3D10_SB_OPCODE_IADD - Issue #38
switch (Inst.m_OpCode)
{
case D3D10_SB_OPCODE_ADD:
case D3D10_SB_OPCODE_MAX:
case D3D10_SB_OPCODE_MIN:
case D3D10_SB_OPCODE_IADD:
this->Emit_with_Add_pipe(Inst);
break;
case D3D10_SB_OPCODE_DP2:
case D3D10_SB_OPCODE_DP3:
case D3D10_SB_OPCODE_DP4:
this->Emit_DPx(Inst);
break;
case D3D10_SB_OPCODE_MAD:
this->Emit_Mad(Inst);
break;
case D3D10_SB_OPCODE_MOV:
this->Emit_Mov(Inst);
break;
case D3D10_SB_OPCODE_MUL:
this->Emit_with_Mul_pipe(Inst);
break;
case D3D10_SB_OPCODE_RET:
break;
default:
VC4_ASSERT(false);
}
}
}
this->ShaderStorageAux->CopyFrom(*this->ShaderStorage); // Copy VS to CS.
this->ShaderUniformAux->CopyFrom(*this->ShaderUniform); // Copy VS uniform to CS.
this->Emit_ShaderOutput_VS(true); // VS
this->Emit_Epilogue(); // VS
this->SetCurrentStorage(this->ShaderStorageAux, this->ShaderUniformAux); // switch to CS storage.
this->Emit_ShaderOutput_VS(false); // CS
this->Emit_Epilogue(); // CS
return S_OK;
}
HRESULT Vc4Shader::Translate_PS()
{
assert(this->uShaderType == D3D10_SB_PIXEL_SHADER);
this->SetCurrentStorage(this->ShaderStorage, this->ShaderUniform);
this->HLSL_ParseDecl();
this->Emit_Prologue_PS();
{
CInstruction Inst;
assert(Inst.m_bSaturate == false); // saturate is not supported.
while (HLSL_GetShaderInstruction(this->HLSLParser, Inst))
{
switch (Inst.m_OpCode)
{
case D3D10_SB_OPCODE_ADD:
case D3D10_SB_OPCODE_MAX:
case D3D10_SB_OPCODE_MIN:
this->Emit_with_Add_pipe(Inst);
break;
case D3D10_SB_OPCODE_DP2:
case D3D10_SB_OPCODE_DP3:
case D3D10_SB_OPCODE_DP4:
this->Emit_DPx(Inst);
break;
case D3D10_SB_OPCODE_MAD:
this->Emit_Mad(Inst);
break;
case D3D10_SB_OPCODE_MOV:
this->Emit_Mov(Inst);
break;
case D3D10_SB_OPCODE_MUL:
this->Emit_with_Mul_pipe(Inst);
break;
case D3D10_SB_OPCODE_RET:
break;
case D3D10_SB_OPCODE_SAMPLE:
this->Emit_Sample(Inst);
break;
default:
VC4_ASSERT(false);
}
}
}
this->Emit_Epilogue();
return S_OK;
}
void Vc4Shader::HLSL_Link_PS()
{
assert(this->uShaderType == D3D10_SB_VERTEX_SHADER);
assert(this->HLSLDownstreamParser.IsValid());
boolean bDone = false, bFound = false;
CInstruction Inst;
while (HLSL_GetShaderInstruction(this->HLSLDownstreamParser, Inst) && !bDone)
{
switch (Inst.m_OpCode)
{
case D3D10_SB_OPCODE_DCL_INPUT_PS:
VC4_ASSERT(Inst.m_InputPSDecl.InterpolationMode == D3D10_SB_INTERPOLATION_LINEAR); // PS input must be linear.
VC4_ASSERT(Inst.m_NumOperands == 1);
VC4_ASSERT(Inst.m_Operands[0].m_ComponentSelection == D3D10_SB_OPERAND_4_COMPONENT_MASK_MODE);
VC4_ASSERT(Inst.m_Operands[0].m_IndexDimension == D3D10_SB_OPERAND_INDEX_1D);
VC4_ASSERT(Inst.m_Operands[0].m_IndexType[0] == D3D10_SB_OPERAND_INDEX_IMMEDIATE32);
VC4_ASSERT(Inst.m_Operands[0].m_Index[0].m_RegIndex < 8);
VC4_ASSERT(Inst.m_Operands[0].m_WriteMask & D3D10_SB_OPERAND_4_COMPONENT_MASK_MASK);
for (uint8_t i = 0, aCurrent = D3D10_SB_OPERAND_4_COMPONENT_MASK_X; i < 4; i++)
{
if (Inst.m_Operands[0].m_WriteMask & aCurrent)
{
VC4_ASSERT(this->OutputRegister[Inst.m_Operands[0].m_Index[0].m_RegIndex][i].flags.valid);
this->OutputRegister[Inst.m_Operands[0].m_Index[0].m_RegIndex][i].flags.linkage = true;
}
aCurrent <<= 1;
}
bFound = true;
break;
default:
if (bFound)
{
bDone = true;
}
}
}
}
void Vc4Shader::Emit_Prologue_VS()
{
assert(this->uShaderType == D3D10_SB_VERTEX_SHADER);
VC4_ASSERT(cInput < 16); // VR_SETUP:NUM limitation, vpm only can read up to 16 values.
{
Vc4Instruction Vc4Inst(vc4_load_immediate_32);
Vc4Register vr_setup(VC4_QPU_ALU_REG_A, VC4_QPU_WADDR_VPMVCD_RD_SETUP);
Vc4Register value; value.SetImmediateI(MAKE_VR_SETUP(cInput, 1, true, false, VC4_QPU_32BIT_VECTOR, 0));
Vc4Inst.Vc4_a_LOAD32(vr_setup, value);
Vc4Inst.Emit(CurrentStorage);
}
{
Vc4Instruction Vc4Inst(vc4_load_immediate_32);
Vc4Register vw_setup(VC4_QPU_ALU_REG_B, VC4_QPU_WADDR_VPMVCD_WR_SETUP);
Vc4Register value; value.SetImmediateI(MAKE_VW_SETUP(1, true, false, VC4_QPU_32BIT_VECTOR, 0));
Vc4Inst.Vc4_a_LOAD32(vw_setup, value);
Vc4Inst.Emit(CurrentStorage);
}
for (uint8_t iRegUsed = 0, iRegIndex = 0; iRegUsed < this->cInput; iRegIndex++)
{
Vc4Instruction Vc4Inst;
Vc4Register raX = this->InputRegister[iRegIndex / 4][iRegIndex % 4];
if (raX.GetFlags().valid)
{
assert(raX.GetMux() == VC4_QPU_ALU_REG_A || raX.GetMux() == VC4_QPU_ALU_REG_B);
Vc4Register vpm(raX.GetMux(), VC4_QPU_RADDR_VPM);
Vc4Inst.Vc4_a_MOV(raX, vpm);
Vc4Inst.Emit(CurrentStorage);
iRegUsed++;
}
}
{ // Emit a NOP
Vc4Instruction Vc4Inst;
Vc4Inst.Emit(CurrentStorage);
}
}
void Vc4Shader::HLSL_ParseDecl()
{
assert(this->uShaderType == D3D10_SB_PIXEL_SHADER ||
this->uShaderType == D3D10_SB_VERTEX_SHADER);
assert(this->HLSLParser.IsValid());
boolean bDone = false;
D3D10_SB_OPCODE_TYPE OpCode;
while (HLSL_PeekShaderInstructionOpCode(this->HLSLParser, OpCode) && !bDone)
{
CInstruction Inst;
switch (OpCode)
{
case D3D10_SB_OPCODE_DCL_RESOURCE:
HLSL_GetShaderInstruction(this->HLSLParser, Inst);
VC4_ASSERT(Inst.m_Operands[0].m_Index[0].m_RegIndex < ARRAYSIZE(this->ResourceDimension));
this->ResourceDimension[Inst.m_Operands[0].m_Index[0].m_RegIndex] = Inst.m_ResourceDecl.SRVInfo.Dimension;
cResources++;
break;
case D3D10_SB_OPCODE_DCL_CONSTANT_BUFFER:
HLSL_GetShaderInstruction(this->HLSLParser, Inst);
// Issue 37: VC4 Find_Vc4Register_I needs to implement dynamic index support for constant buffers
VC4_ASSERT(Inst.m_ResourceDecl.CBInfo.AccessPattern == D3D10_SB_CONSTANT_BUFFER_IMMEDIATE_INDEXED);
cConstants++;
break;
case D3D10_SB_OPCODE_DCL_SAMPLER:
HLSL_GetShaderInstruction(this->HLSLParser, Inst);
cSampler++;
break;
case D3D10_SB_OPCODE_DCL_INPUT:
case D3D10_SB_OPCODE_DCL_INPUT_PS:
HLSL_GetShaderInstruction(this->HLSLParser, Inst);
if (Inst.m_OpCode == D3D10_SB_OPCODE_DCL_INPUT_PS)
{
VC4_ASSERT(Inst.m_InputPSDecl.InterpolationMode == D3D10_SB_INTERPOLATION_LINEAR); // PS input must be linear.
}
VC4_ASSERT(Inst.m_NumOperands == 1);
VC4_ASSERT(Inst.m_Operands[0].m_ComponentSelection == D3D10_SB_OPERAND_4_COMPONENT_MASK_MODE);
VC4_ASSERT(Inst.m_Operands[0].m_IndexDimension == D3D10_SB_OPERAND_INDEX_1D);
VC4_ASSERT(Inst.m_Operands[0].m_IndexType[0] == D3D10_SB_OPERAND_INDEX_IMMEDIATE32);
VC4_ASSERT(Inst.m_Operands[0].m_Index[0].m_RegIndex < 8);
VC4_ASSERT(Inst.m_Operands[0].m_WriteMask & D3D10_SB_OPERAND_4_COMPONENT_MASK_MASK);
for (uint8_t i = 0, aCurrent = D3D10_SB_OPERAND_4_COMPONENT_MASK_X; i < 4; i++)
{
if (Inst.m_Operands[0].m_WriteMask & aCurrent)
{
VC4_ASSERT((ROS_VC4_INPUT_REGISTER_FILE_START + cInput) <= ROS_VC4_INPUT_REGISTER_FILE_END);
this->InputRegister[Inst.m_Operands[0].m_Index[0].m_RegIndex][i].flags.valid = true;
this->InputRegister[Inst.m_Operands[0].m_Index[0].m_RegIndex][i].flags.require_linear_conversion = (Inst.m_OpCode == D3D10_SB_OPCODE_DCL_INPUT_PS ? true : false);
this->InputRegister[Inst.m_Operands[0].m_Index[0].m_RegIndex][i].addr = ROS_VC4_INPUT_REGISTER_FILE_START + cInput++;
this->InputRegister[Inst.m_Operands[0].m_Index[0].m_RegIndex][i].mux = ROS_VC4_INPUT_REGISTER_FILE;
this->InputRegister[Inst.m_Operands[0].m_Index[0].m_RegIndex][i].swizzleMask = aCurrent;
}
aCurrent <<= 1;
}
break;
case D3D10_SB_OPCODE_DCL_OUTPUT:
case D3D10_SB_OPCODE_DCL_OUTPUT_SIV:
HLSL_GetShaderInstruction(this->HLSLParser, Inst);
VC4_ASSERT(Inst.m_NumOperands == 1);
VC4_ASSERT(Inst.m_Operands[0].m_ComponentSelection == D3D10_SB_OPERAND_4_COMPONENT_MASK_MODE);
VC4_ASSERT(Inst.m_Operands[0].m_IndexDimension == D3D10_SB_OPERAND_INDEX_1D);
VC4_ASSERT(Inst.m_Operands[0].m_IndexType[0] == D3D10_SB_OPERAND_INDEX_IMMEDIATE32);
if (this->uShaderType == D3D10_SB_PIXEL_SHADER)
{
// only support single 8888RGBA colour output from pixel shader.
VC4_ASSERT(Inst.m_Operands[0].m_WriteMask == (D3D10_SB_OPERAND_4_COMPONENT_MASK_R | D3D10_SB_OPERAND_4_COMPONENT_MASK_G | D3D10_SB_OPERAND_4_COMPONENT_MASK_B | D3D10_SB_OPERAND_4_COMPONENT_MASK_A));
VC4_ASSERT(Inst.m_Operands[0].m_Index[0].m_RegIndex == 0);
VC4_ASSERT(cOutput == 0);
this->OutputRegister[0][0].flags.valid = true;
this->OutputRegister[0][0].flags.color = true;
this->OutputRegister[0][0].flags.packed = true; // RGBA components are packed in single register (see above WriteMask assert).
this->OutputRegister[0][0].addr = ROS_VC4_OUTPUT_REGISTER_FILE_START;
this->OutputRegister[0][0].mux = ROS_VC4_OUTPUT_REGISTER_FILE;
this->OutputRegister[0][0].swizzleMask = (uint8_t)(Inst.m_Operands[0].m_WriteMask & D3D10_SB_OPERAND_4_COMPONENT_MASK_MASK);
// TODO: more generic color channel swizzle support.
DXGI_FORMAT texFormat = UmdCompiler->GetRenderTargetFormat(0);
VC4_ASSERT((texFormat == DXGI_FORMAT_B8G8R8A8_UNORM) || (texFormat == DXGI_FORMAT_R8G8B8A8_UNORM));
this->OutputRegister[0][0].flags.swap_color_channel = (texFormat != DXGI_FORMAT_R8G8B8A8_UNORM);
cOutput++;
}
else
{
VC4_ASSERT(this->uShaderType == D3D10_SB_VERTEX_SHADER);
VC4_ASSERT(Inst.m_Operands[0].m_Index[0].m_RegIndex < 8);
bool bPos;
uint8_t aMask;
if ((Inst.m_OpCode == D3D10_SB_OPCODE_DCL_OUTPUT_SIV) && (Inst.m_InputDeclSIV.Name == D3D10_SB_NAME_POSITION))
{
bPos = true;
aMask = D3D10_SB_OPERAND_4_COMPONENT_MASK_MASK;
}
else
{
bPos = false;
aMask = (Inst.m_Operands[0].m_WriteMask & D3D10_SB_OPERAND_4_COMPONENT_MASK_MASK);
VC4_ASSERT(aMask);
}
for (uint8_t i = 0, aCurrent = D3D10_SB_OPERAND_4_COMPONENT_MASK_X; i < 4; i++)
{
if (aMask & aCurrent)
{
VC4_ASSERT((ROS_VC4_OUTPUT_REGISTER_FILE_START + cOutput) <= ROS_VC4_OUTPUT_REGISTER_FILE_END);
this->OutputRegister[Inst.m_Operands[0].m_Index[0].m_RegIndex][i].flags.valid = true;
this->OutputRegister[Inst.m_Operands[0].m_Index[0].m_RegIndex][i].flags.position = bPos;
this->OutputRegister[Inst.m_Operands[0].m_Index[0].m_RegIndex][i].addr = ROS_VC4_OUTPUT_REGISTER_FILE_START + cOutput++;
this->OutputRegister[Inst.m_Operands[0].m_Index[0].m_RegIndex][i].mux = ROS_VC4_OUTPUT_REGISTER_FILE;
this->OutputRegister[Inst.m_Operands[0].m_Index[0].m_RegIndex][i].swizzleMask = aCurrent;
}
aCurrent <<= 1;
}
}
break;
case D3D10_SB_OPCODE_DCL_TEMPS:
HLSL_GetShaderInstruction(this->HLSLParser, Inst);
// Temp register doesn't have swizzle mask, so assume all 4 components to be used.
// TODO: AllocateRegister(); Currently temps are allocated at ra16~ra31.
// since currently reserve temp to ra16~31, so only upto 4 temps are allowed.
VC4_ASSERT(Inst.m_TempsDecl.NumTemps <= 4);
for (uint8_t i = 0; i < Inst.m_TempsDecl.NumTemps * 4; i++)
{
VC4_ASSERT((ROS_VC4_TEMP_REGISTER_FILE_START + cTemp) <= ROS_VC4_TEMP_REGISTER_FILE_END);
this->TempRegister[i / 4][i % 4].flags.valid = true;
this->TempRegister[i / 4][i % 4].flags.temp = true;
this->TempRegister[i / 4][i % 4].addr = ROS_VC4_TEMP_REGISTER_FILE_START + cTemp++;
this->TempRegister[i / 4][i % 4].mux = ROS_VC4_TEMP_REGISTER_FILE;
this->TempRegister[i / 4][i % 4].swizzleMask = D3D10_SB_OPERAND_4_COMPONENT_MASK_X << (i % 4);
}
break;
case D3D10_SB_OPCODE_DCL_GLOBAL_FLAGS:
HLSL_GetShaderInstruction(this->HLSLParser, Inst);
// TODO:
break;
case D3D10_SB_OPCODE_DCL_INDEX_RANGE:
case D3D10_SB_OPCODE_DCL_GS_OUTPUT_PRIMITIVE_TOPOLOGY:
case D3D10_SB_OPCODE_DCL_GS_INPUT_PRIMITIVE:
case D3D10_SB_OPCODE_DCL_MAX_OUTPUT_VERTEX_COUNT:
case D3D10_SB_OPCODE_DCL_INPUT_SGV:
case D3D10_SB_OPCODE_DCL_INPUT_SIV:
case D3D10_SB_OPCODE_DCL_INPUT_PS_SGV:
case D3D10_SB_OPCODE_DCL_INPUT_PS_SIV:
case D3D10_SB_OPCODE_DCL_OUTPUT_SGV:
case D3D10_SB_OPCODE_DCL_INDEXABLE_TEMP:
HLSL_GetShaderInstruction(this->HLSLParser, Inst);
VC4_ASSERT(false); // TODO
__fallthrough;
default:
if (OpCode >= D3D10_SB_OPCODE_RESERVED0)
{
VC4_ASSERT(false); // only 10.0 opcode is supported.
}
bDone = true;
}
}
}
void Vc4Shader::Emit_Sample(CInstruction &Inst)
{
assert(this->uShaderType == D3D10_SB_PIXEL_SHADER);
VC4_ASSERT(Inst.m_NumOperands == 4);
boolean bUnpack = false;
Vc4Register o[4];
VC4_ASSERT(Inst.m_Operands[0].m_NumComponents == D3D10_SB_OPERAND_4_COMPONENT);
VC4_ASSERT(Inst.m_Operands[0].m_IndexDimension == D3D10_SB_OPERAND_INDEX_1D);
VC4_ASSERT(Inst.m_Operands[0].m_IndexType[0] == D3D10_SB_OPERAND_INDEX_IMMEDIATE32);
VC4_ASSERT(Inst.m_Operands[0].m_ComponentSelection == D3D10_SB_OPERAND_4_COMPONENT_MASK_MODE);
VC4_ASSERT(Inst.m_Operands[0].m_WriteMask == (D3D10_SB_OPERAND_4_COMPONENT_MASK_R | D3D10_SB_OPERAND_4_COMPONENT_MASK_G | D3D10_SB_OPERAND_4_COMPONENT_MASK_B | D3D10_SB_OPERAND_4_COMPONENT_MASK_A));
switch (Inst.m_Operands[0].m_Type)
{
case D3D10_SB_OPERAND_TYPE_OUTPUT:
o[0] = Find_Vc4Register_M(Inst.m_Operands[0], (Inst.m_Operands[0].m_WriteMask & D3D10_SB_OPERAND_4_COMPONENT_MASK_MASK));
VC4_ASSERT(o[0].GetFlags().packed);
break;
case D3D10_SB_OPERAND_TYPE_TEMP:
for (uint8_t i = 0, aCurrent = D3D10_SB_OPERAND_4_COMPONENT_MASK_X; i < 4; i++)
{
if (Inst.m_Operands[0].m_WriteMask & aCurrent)
{
o[i] = Find_Vc4Register_M(Inst.m_Operands[0], aCurrent);
}
aCurrent <<= 1;
}
bUnpack = true;
break;
default:
VC4_ASSERT(false);
}
// Resource
VC4_ASSERT(Inst.m_Operands[2].m_Type == D3D10_SB_OPERAND_TYPE_RESOURCE);
VC4_ASSERT(Inst.m_Operands[2].m_NumComponents == D3D10_SB_OPERAND_4_COMPONENT);
VC4_ASSERT(Inst.m_Operands[2].m_IndexDimension == D3D10_SB_OPERAND_INDEX_1D);
VC4_ASSERT(Inst.m_Operands[2].m_IndexType[0] == D3D10_SB_OPERAND_INDEX_IMMEDIATE32);
uint32_t resourceIndex = Inst.m_Operands[2].m_Index[0].m_RegIndex;
uint32_t texDimension = this->ResourceDimension[resourceIndex];
DXGI_FORMAT texFormat = UmdCompiler->GetShaderResourceFormat((uint8_t)resourceIndex);
VC4_ASSERT((texFormat == DXGI_FORMAT_B8G8R8A8_UNORM) || (texFormat == DXGI_FORMAT_R8G8B8A8_UNORM));
// TODO: more generic color channel swizzle support.
boolean bSwapColorChannel = (texFormat != DXGI_FORMAT_R8G8B8A8_UNORM);
// Texture coordinate
VC4_ASSERT(Inst.m_Operands[1].m_NumComponents == D3D10_SB_OPERAND_4_COMPONENT);
VC4_ASSERT(Inst.m_Operands[1].m_IndexDimension == D3D10_SB_OPERAND_INDEX_1D);
VC4_ASSERT(Inst.m_Operands[1].m_IndexType[0] == D3D10_SB_OPERAND_INDEX_IMMEDIATE32);
VC4_ASSERT(Inst.m_Operands[1].m_ComponentSelection == D3D10_SB_OPERAND_4_COMPONENT_SWIZZLE_MODE);
Vc4Register s;
Vc4Register t;
Vc4Register r;
switch (texDimension)
{
case D3D10_SB_RESOURCE_DIMENSION_TEXTURECUBE:
r = Find_Vc4Register_M(Inst.m_Operands[1], D3D10_SB_OPERAND_4_COMPONENT_MASK(Inst.m_Operands[1].m_Swizzle[2]));
__fallthrough;
case D3D10_SB_RESOURCE_DIMENSION_TEXTURE2D:
t = Find_Vc4Register_M(Inst.m_Operands[1], D3D10_SB_OPERAND_4_COMPONENT_MASK(Inst.m_Operands[1].m_Swizzle[1]));
__fallthrough;
case D3D10_SB_RESOURCE_DIMENSION_TEXTURE1D:
s = Find_Vc4Register_M(Inst.m_Operands[1], D3D10_SB_OPERAND_4_COMPONENT_MASK(Inst.m_Operands[1].m_Swizzle[0]));
break;
case D3D10_SB_RESOURCE_DIMENSION_BUFFER:
case D3D10_SB_RESOURCE_DIMENSION_TEXTURE3D:
case D3D10_SB_RESOURCE_DIMENSION_TEXTURE2DMS:
case D3D10_SB_RESOURCE_DIMENSION_TEXTURE1DARRAY:
case D3D10_SB_RESOURCE_DIMENSION_TEXTURE2DARRAY:
case D3D10_SB_RESOURCE_DIMENSION_TEXTURE2DMSARRAY:
default:
assert(false);
}
// Sampler
VC4_ASSERT(Inst.m_Operands[3].m_Type == D3D10_SB_OPERAND_TYPE_SAMPLER);
VC4_ASSERT(Inst.m_Operands[3].m_IndexDimension == D3D10_SB_OPERAND_INDEX_1D);
VC4_ASSERT(Inst.m_Operands[3].m_IndexType[0] == D3D10_SB_OPERAND_INDEX_IMMEDIATE32);
uint32_t samplerIndex = Inst.m_Operands[3].m_Index[0].m_RegIndex;
// texture address : z
if (r.GetFlags().valid)
{
Vc4Instruction Vc4Inst;
Vc4Register tmu0_r(VC4_QPU_ALU_REG_A, VC4_QPU_WADDR_TMU0_R);
Vc4Inst.Vc4_a_MOV(tmu0_r, r);
Vc4Inst.Emit(CurrentStorage);
}
// texture address : y
if (t.GetFlags().valid)
{
Vc4Instruction Vc4Inst;
Vc4Register tmu0_t(VC4_QPU_ALU_REG_A, VC4_QPU_WADDR_TMU0_T);
Vc4Inst.Vc4_a_MOV(tmu0_t, t);
Vc4Inst.Emit(CurrentStorage);
}
// texture address : x and must write 's' at last.
assert(s.GetFlags().valid);
{
Vc4Instruction Vc4Inst;
Vc4Register tmu0_s(VC4_QPU_ALU_REG_A, VC4_QPU_WADDR_TMU0_S);
Vc4Inst.Vc4_a_MOV(tmu0_s, s);
Vc4Inst.Emit(CurrentStorage);
}
// add uniform references.
{
{
VC4_UNIFORM_FORMAT u;
u.Type = VC4_UNIFORM_TYPE_SAMPLER_CONFIG_P0;
u.samplerConfiguration.samplerIndex = samplerIndex;
u.samplerConfiguration.resourceIndex = resourceIndex;
this->AddUniformReference(u);
}
{
VC4_UNIFORM_FORMAT u;
u.Type = VC4_UNIFORM_TYPE_SAMPLER_CONFIG_P1;
u.samplerConfiguration.samplerIndex = samplerIndex;
u.samplerConfiguration.resourceIndex = resourceIndex;
this->AddUniformReference(u);
}
if (r.GetFlags().valid) // only cube needs P2 config.
{
VC4_UNIFORM_FORMAT u;
u.Type = VC4_UNIFORM_TYPE_SAMPLER_CONFIG_P2;
u.samplerConfiguration.samplerIndex = samplerIndex;
u.samplerConfiguration.resourceIndex = resourceIndex;
this->AddUniformReference(u);
}
}
// Sample texture, result come up in r4.
{
Vc4Instruction Vc4Inst;
Vc4Inst.Vc4_Sig(VC4_QPU_SIG_LOAD_TMU0);
Vc4Inst.Emit(CurrentStorage);
}
// Sample result is now at r4.
Vc4Register r4(VC4_QPU_ALU_R4);
// Move result at r4 to output register.
if (Inst.m_Operands[0].m_Type == D3D10_SB_OPERAND_TYPE_OUTPUT)
{
if (bSwapColorChannel == o[0].GetFlags().swap_color_channel)
{
Vc4Instruction Vc4Inst;
Vc4Inst.Vc4_a_MOV(o[0], r4);
Vc4Inst.Emit(CurrentStorage);
}
else
{
// R, G, B channel
for (uint8_t i = 0; i < 3; i++)
{
Vc4Instruction Vc4Inst;
Vc4Inst.Vc4_m_MOV(o[0], r4);
Vc4Inst.Vc4_m_Pack(VC4_QPU_PACK_MUL_8c - i);
Vc4Inst.Vc4_m_Unpack(VC4_QPU_UNPACK_8a + i, true); // Use R4 unpack.
Vc4Inst.Emit(CurrentStorage);
}
// A channel
{
Vc4Instruction Vc4Inst;
Vc4Inst.Vc4_m_MOV(o[0], r4);
Vc4Inst.Vc4_m_Pack(VC4_QPU_PACK_MUL_8d);
Vc4Inst.Vc4_m_Unpack(VC4_QPU_UNPACK_8d, true); // Use R4 unpack.
Vc4Inst.Emit(CurrentStorage);
}
}
}
else
{
// Move each color channel at r4 to o[i].
// R, G, B channel
for (uint8_t i = 0; i < 3; i++)
{
Vc4Register out = bSwapColorChannel ? o[2 - i] : o[i];
if (out.GetFlags().valid)
{
Vc4Instruction Vc4Inst;
Vc4Inst.Vc4_m_MOV(out, r4);
Vc4Inst.Vc4_m_Unpack(VC4_QPU_UNPACK_8a + i, true); // Use R4 unpack.
Vc4Inst.Emit(CurrentStorage);
}
}
// A channel
if (o[3].GetFlags().valid)
{
Vc4Instruction Vc4Inst;
Vc4Inst.Vc4_m_MOV(o[3], r4);
Vc4Inst.Vc4_m_Unpack(VC4_QPU_UNPACK_8d, true); // Use R4 unpack.
Vc4Inst.Emit(CurrentStorage);
}
}
{ // Emit a NOP
Vc4Instruction Vc4Inst;
Vc4Inst.Emit(CurrentStorage);
}
}
void Vc4Shader::Emit_with_Add_pipe(CInstruction &Inst)
{
assert(this->uShaderType == D3D10_SB_PIXEL_SHADER ||
this->uShaderType == D3D10_SB_VERTEX_SHADER);
VC4_ASSERT(Inst.m_NumOperands == 3);
{
for (uint8_t i = 0, aCurrent = D3D10_SB_OPERAND_4_COMPONENT_MASK_X; i < 4; i++)
{
if (Inst.m_Operands[0].m_WriteMask & aCurrent)
{
Vc4Register dst = Find_Vc4Register_M(Inst.m_Operands[0], aCurrent);
Vc4Register src[2];
this->Setup_SourceRegisters(Inst, 1, ARRAYSIZE(src), i, src);
Vc4Register _dst;
if (dst.GetFlags().packed)
{
// pack has to be done at mul pipe, so result to r3,
// then use mul pipe to move to final dst (with pack).
Vc4Register r3(VC4_QPU_ALU_R3, VC4_QPU_WADDR_ACC3);
_dst = r3;
}
else
{
_dst = dst;
}
{
Vc4Instruction Vc4Inst;
switch (Inst.m_OpCode)
{
case D3D10_SB_OPCODE_ADD:
Vc4Inst.Vc4_a_FADD(_dst, src[0], src[1]);
break;
case D3D10_SB_OPCODE_MAX:
Vc4Inst.Vc4_a_FMAX(_dst, src[0], src[1]);
break;
case D3D10_SB_OPCODE_MIN:
Vc4Inst.Vc4_a_FMIN(_dst, src[0], src[1]);
break;
case D3D10_SB_OPCODE_IADD:
Vc4Inst.Vc4_a_IADD(_dst, src[0], src[1]);
break;
default:
VC4_ASSERT(false);
}
Vc4Inst.Emit(CurrentStorage);
}
if (dst.GetFlags().packed)
{
Vc4Instruction Vc4Inst;
Vc4Inst.Vc4_m_MOV(dst, _dst);
Vc4Inst.Vc4_m_Pack(dst.GetPack(i));
Vc4Inst.Emit(CurrentStorage);
}
}
aCurrent <<= 1;
}
}
{ // Emit a NOP
Vc4Instruction Vc4Inst;
Vc4Inst.Emit(CurrentStorage);
}
}
void Vc4Shader::Emit_DPx(CInstruction &Inst)
{
assert(this->uShaderType == D3D10_SB_PIXEL_SHADER ||
this->uShaderType == D3D10_SB_VERTEX_SHADER);
VC4_ASSERT(Inst.m_NumOperands == 3);
{
// DP2 loop 2 times.
// DP3 loop 3 times.
// DP4 loop 4 times.
uint8_t c = (uint8_t)(Inst.m_OpCode - 13);
// where to accumulate result of mul.
Vc4Register accum(VC4_QPU_ALU_R3, VC4_QPU_WADDR_ACC3);
{
Vc4Register zero(VC4_QPU_ALU_REG_B, 0); // 0 as small immediate in raddr_b
Vc4Instruction Vc4Inst(vc4_alu_small_immediate);
Vc4Inst.Vc4_m_MOV(accum, zero);
Vc4Inst.Emit(CurrentStorage);
}
for(uint8_t i = 0; i < c; i++)
{
Vc4Register temp(VC4_QPU_ALU_R1, VC4_QPU_WADDR_ACC1);
Vc4Register src[2];
Setup_SourceRegisters(Inst, 1, ARRAYSIZE(src), i, src);
{
Vc4Instruction Vc4Inst;
Vc4Inst.Vc4_m_FMUL(temp, src[0], src[1]);
if (i > 0)
{
Vc4Inst.Vc4_a_FADD(accum, accum, temp);
}
Vc4Inst.Emit(CurrentStorage);
}
if (i+1 == c)
{
Vc4Instruction Vc4Inst;
Vc4Inst.Vc4_a_FADD(accum, accum, temp);
Vc4Inst.Emit(CurrentStorage);
}
}
// replicate ouput where specified.
for (uint8_t i = 0, aCurrent = D3D10_SB_OPERAND_4_COMPONENT_MASK_X; i < 4; i++)
{
if (Inst.m_Operands[0].m_WriteMask & aCurrent)
{
Vc4Register dst = Find_Vc4Register_M(Inst.m_Operands[0], aCurrent);
Vc4Instruction Vc4Inst;
Vc4Inst.Vc4_m_MOV(dst, accum);
Vc4Inst.Vc4_m_Pack(dst.GetPack(i));
Vc4Inst.Emit(CurrentStorage);
}
aCurrent <<= 1;
}
}
{ // Emit a NOP
Vc4Instruction Vc4Inst;
Vc4Inst.Emit(CurrentStorage);
}
}
void Vc4Shader::Emit_Mad(CInstruction &Inst)
{
assert(this->uShaderType == D3D10_SB_PIXEL_SHADER ||
this->uShaderType == D3D10_SB_VERTEX_SHADER);
VC4_ASSERT(Inst.m_NumOperands == 4);
{
for (uint8_t i = 0, aCurrent = D3D10_SB_OPERAND_4_COMPONENT_MASK_X; i < 4; i++)
{
if (Inst.m_Operands[0].m_WriteMask & aCurrent)
{
Vc4Register accum(VC4_QPU_ALU_R3, VC4_QPU_WADDR_ACC3);
Vc4Register dst = Find_Vc4Register_M(Inst.m_Operands[0], aCurrent);
// perform mul first 2 operands
{
Vc4Register src[2];
this->Setup_SourceRegisters(Inst, 1, ARRAYSIZE(src), i, src);
{
Vc4Instruction Vc4Inst;
Vc4Inst.Vc4_m_FMUL(accum, src[0], src[1]);
Vc4Inst.Emit(CurrentStorage);
}
}
Vc4Register _dst;
if (dst.GetFlags().packed)
{
// pack has to be done at mul pipe, so result to r3,
// then use mul pipe to move to final dst (with pack).
_dst = accum;
}
else
{
_dst = dst;
}
// perform add with 3rd operand.
{
Vc4Register src[1];
this->Setup_SourceRegisters(Inst, 3, ARRAYSIZE(src), i, src);
{
Vc4Instruction Vc4Inst;
Vc4Inst.Vc4_a_FADD(_dst, accum, src[0]);
Vc4Inst.Emit(CurrentStorage);
}
}
// move to destination (with packing).
if (dst.GetFlags().packed)
{
Vc4Instruction Vc4Inst;
Vc4Inst.Vc4_m_MOV(dst, accum);
Vc4Inst.Vc4_m_Pack(dst.GetPack(i));
Vc4Inst.Emit(CurrentStorage);
}
}
aCurrent <<= 1;
}
}
{ // Emit a NOP
Vc4Instruction Vc4Inst;
Vc4Inst.Emit(CurrentStorage);
}
}