// Copyright 2018 yuzu Emulator Project // Licensed under GPLv2 or any later version // Refer to the license.txt file included. #include "common/assert.h" #include "common/common_types.h" #include "video_core/engines/shader_bytecode.h" #include "video_core/shader/shader_ir.h" namespace VideoCommon::Shader { using Tegra::Shader::Instruction; using Tegra::Shader::OpCode; using Tegra::Shader::Register; u32 ShaderIR::DecodeArithmeticInteger(BasicBlock& bb, u32 pc) { const Instruction instr = {program_code[pc]}; const auto opcode = OpCode::Decode(instr); Node op_a = GetRegister(instr.gpr8); Node op_b = [&]() { if (instr.is_b_imm) { return Immediate(instr.alu.GetSignedImm20_20()); } else if (instr.is_b_gpr) { return GetRegister(instr.gpr20); } else { return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.offset); } }(); switch (opcode->get().GetId()) { case OpCode::Id::IADD_C: case OpCode::Id::IADD_R: case OpCode::Id::IADD_IMM: { UNIMPLEMENTED_IF_MSG(instr.generates_cc, "Condition codes generation in IADD is not implemented"); UNIMPLEMENTED_IF_MSG(instr.alu.saturate_d, "IADD saturation not implemented"); op_a = GetOperandAbsNegInteger(op_a, false, instr.alu_integer.negate_a, true); op_b = GetOperandAbsNegInteger(op_b, false, instr.alu_integer.negate_b, true); SetRegister(bb, instr.gpr0, Operation(OperationCode::IAdd, PRECISE, op_a, op_b)); break; } case OpCode::Id::ISCADD_C: case OpCode::Id::ISCADD_R: case OpCode::Id::ISCADD_IMM: { UNIMPLEMENTED_IF_MSG(instr.generates_cc, "Condition codes generation in ISCADD is not implemented"); op_a = GetOperandAbsNegInteger(op_a, false, instr.alu_integer.negate_a, true); op_b = GetOperandAbsNegInteger(op_b, false, instr.alu_integer.negate_b, true); const Node shift = Immediate(static_cast(instr.alu_integer.shift_amount)); const Node shifted_a = Operation(OperationCode::ILogicalShiftLeft, NO_PRECISE, op_a, shift); const Node value = Operation(OperationCode::IAdd, NO_PRECISE, shifted_a, op_b); SetRegister(bb, instr.gpr0, value); break; } case OpCode::Id::SEL_C: case OpCode::Id::SEL_R: case OpCode::Id::SEL_IMM: { const Node condition = GetPredicate(instr.sel.pred, instr.sel.neg_pred != 0); const Node value = Operation(OperationCode::Select, PRECISE, condition, op_a, op_b); SetRegister(bb, instr.gpr0, value); break; } case OpCode::Id::LOP_C: case OpCode::Id::LOP_R: case OpCode::Id::LOP_IMM: { UNIMPLEMENTED_IF_MSG(instr.generates_cc, "Condition codes generation in LOP is not implemented"); if (instr.alu.lop.invert_a) op_a = Operation(OperationCode::IBitwiseNot, NO_PRECISE, op_a); if (instr.alu.lop.invert_b) op_b = Operation(OperationCode::IBitwiseNot, NO_PRECISE, op_b); WriteLogicOperation(bb, instr.gpr0, instr.alu.lop.operation, op_a, op_b, instr.alu.lop.pred_result_mode, instr.alu.lop.pred48); break; } case OpCode::Id::LOP3_C: case OpCode::Id::LOP3_R: case OpCode::Id::LOP3_IMM: { UNIMPLEMENTED_IF_MSG(instr.generates_cc, "Condition codes generation in LOP3 is not implemented"); const Node op_c = GetRegister(instr.gpr39); const Node lut = [&]() { if (opcode->get().GetId() == OpCode::Id::LOP3_R) { return Immediate(instr.alu.lop3.GetImmLut28()); } else { return Immediate(instr.alu.lop3.GetImmLut48()); } }(); WriteLop3Instruction(bb, instr.gpr0, op_a, op_b, op_c, lut); break; } case OpCode::Id::IMNMX_C: case OpCode::Id::IMNMX_R: case OpCode::Id::IMNMX_IMM: { UNIMPLEMENTED_IF(instr.imnmx.exchange != Tegra::Shader::IMinMaxExchange::None); UNIMPLEMENTED_IF_MSG(instr.generates_cc, "Condition codes generation in IMNMX is not implemented"); const bool is_signed = instr.imnmx.is_signed; const Node condition = GetPredicate(instr.imnmx.pred, instr.imnmx.negate_pred != 0); const Node min = SignedOperation(OperationCode::IMin, is_signed, NO_PRECISE, op_a, op_b); const Node max = SignedOperation(OperationCode::IMax, is_signed, NO_PRECISE, op_a, op_b); const Node value = Operation(OperationCode::Select, NO_PRECISE, condition, min, max); SetRegister(bb, instr.gpr0, value); break; } default: UNIMPLEMENTED_MSG("Unhandled ArithmeticInteger instruction: {}", opcode->get().GetName()); } return pc; } void ShaderIR::WriteLop3Instruction(BasicBlock& bb, Register dest, Node op_a, Node op_b, Node op_c, Node imm_lut) { constexpr u32 lop_iterations = 32; const Node one = Immediate(1); const Node two = Immediate(2); Node value{}; for (u32 i = 0; i < lop_iterations; ++i) { const Node shift_amount = Immediate(i); const Node a = Operation(OperationCode::ILogicalShiftRight, NO_PRECISE, op_c, shift_amount); const Node pack_0 = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, a, one); const Node b = Operation(OperationCode::ILogicalShiftRight, NO_PRECISE, op_b, shift_amount); const Node c = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, b, one); const Node pack_1 = Operation(OperationCode::ILogicalShiftLeft, NO_PRECISE, c, one); const Node d = Operation(OperationCode::ILogicalShiftRight, NO_PRECISE, op_a, shift_amount); const Node e = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, d, one); const Node pack_2 = Operation(OperationCode::ILogicalShiftLeft, NO_PRECISE, e, two); const Node pack_01 = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, pack_0, pack_1); const Node pack_012 = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, pack_01, pack_2); const Node shifted_bit = Operation(OperationCode::ILogicalShiftRight, NO_PRECISE, imm_lut, pack_012); const Node bit = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, shifted_bit, one); const Node right = Operation(OperationCode::ILogicalShiftLeft, NO_PRECISE, bit, shift_amount); if (i > 0) { value = Operation(OperationCode::IBitwiseOr, NO_PRECISE, value, right); } else { value = right; } } SetRegister(bb, dest, value); } } // namespace VideoCommon::Shader