ShuriZma
/
xenia
mirror of https://github.com/xenia-project/xenia.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

[D3D12] DXBC: Use the same code for dp4/dp3 as add/mul/max/min

This commit is contained in:
Triang3l 2018-09-08 20:16:00 +03:00
parent b91866fa49
commit b56ff6e325
1 changed files with 3 additions and 28 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
src/xenia/gpu/dxbc_shader_translator.cc
View File

@ -2311,6 +2311,9 @@ void DxbcShaderTranslator::ProcessVectorAluInstruction(
// max is commonly used as mov, but probably better not to convert it to // max is commonly used as mov, but probably better not to convert it to
// make sure things like flusing denormals aren't affected. // make sure things like flusing denormals aren't affected.
case AluVectorOpcode::kMin: case AluVectorOpcode::kMin:
case AluVectorOpcode::kDp4:
case AluVectorOpcode::kDp3:
// dp4 and dp3 replicate the result implicitly.
shader_code_.push_back(ENCODE_D3D10_SB_OPCODE_TYPE( shader_code_.push_back(ENCODE_D3D10_SB_OPCODE_TYPE(
kCoreOpcodes[uint32_t(instr.vector_opcode)]) | kCoreOpcodes[uint32_t(instr.vector_opcode)]) |
ENCODE_D3D10_SB_TOKENIZED_INSTRUCTION_LENGTH( ENCODE_D3D10_SB_TOKENIZED_INSTRUCTION_LENGTH(
@ -2438,34 +2441,6 @@ void DxbcShaderTranslator::ProcessVectorAluInstruction(
++stat_.movc_instruction_count; ++stat_.movc_instruction_count;
break; break;
case AluVectorOpcode::kDp4:
// Replicated implicitly.
shader_code_.push_back(ENCODE_D3D10_SB_OPCODE_TYPE(D3D10_SB_OPCODE_DP4) |
ENCODE_D3D10_SB_TOKENIZED_INSTRUCTION_LENGTH(
3 + operand_length_sums[1]));
shader_code_.push_back(
EncodeVectorMaskedOperand(D3D10_SB_OPERAND_TYPE_TEMP, 0b1111, 1));
shader_code_.push_back(system_temp_pv_);
UseDxbcSourceOperand(dxbc_operands[0]);
UseDxbcSourceOperand(dxbc_operands[1]);
++stat_.instruction_count;
++stat_.float_instruction_count;
break;
case AluVectorOpcode::kDp3:
// Replicated implicitly.
shader_code_.push_back(ENCODE_D3D10_SB_OPCODE_TYPE(D3D10_SB_OPCODE_DP3) |
ENCODE_D3D10_SB_TOKENIZED_INSTRUCTION_LENGTH(
3 + operand_length_sums[1]));
shader_code_.push_back(
EncodeVectorMaskedOperand(D3D10_SB_OPERAND_TYPE_TEMP, 0b1111, 1));
shader_code_.push_back(system_temp_pv_);
UseDxbcSourceOperand(dxbc_operands[0]);
UseDxbcSourceOperand(dxbc_operands[1]);
++stat_.instruction_count;
++stat_.float_instruction_count;
break;
case AluVectorOpcode::kDp2Add: case AluVectorOpcode::kDp2Add:
// (dot(src0.xy, src1.xy) + src2.x).xxxx // (dot(src0.xy, src1.xy) + src2.x).xxxx
replicate_result = true; replicate_result = true;