diff options
Diffstat (limited to 'src/video_core/shader')
-rw-r--r-- | src/video_core/shader/decode/bfe.cpp | 69 | ||||
-rw-r--r-- | src/video_core/shader/decode/xmad.cpp | 63 | ||||
-rw-r--r-- | src/video_core/shader/node_helper.cpp | 2 |
3 files changed, 98 insertions, 36 deletions
diff --git a/src/video_core/shader/decode/bfe.cpp b/src/video_core/shader/decode/bfe.cpp index e02bcd097..8e3b46e8e 100644 --- a/src/video_core/shader/decode/bfe.cpp +++ b/src/video_core/shader/decode/bfe.cpp @@ -17,33 +17,60 @@ u32 ShaderIR::DecodeBfe(NodeBlock& bb, u32 pc) { const Instruction instr = {program_code[pc]}; const auto opcode = OpCode::Decode(instr); - UNIMPLEMENTED_IF(instr.bfe.negate_b); - Node op_a = GetRegister(instr.gpr8); - op_a = GetOperandAbsNegInteger(op_a, false, instr.bfe.negate_a, false); - - switch (opcode->get().GetId()) { - case OpCode::Id::BFE_IMM: { - UNIMPLEMENTED_IF_MSG(instr.generates_cc, - "Condition codes generation in BFE is not implemented"); + Node op_b = [&] { + switch (opcode->get().GetId()) { + case OpCode::Id::BFE_R: + return GetRegister(instr.gpr20); + case OpCode::Id::BFE_C: + return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()); + case OpCode::Id::BFE_IMM: + return Immediate(instr.alu.GetSignedImm20_20()); + default: + UNREACHABLE(); + return Immediate(0); + } + }(); - const Node inner_shift_imm = Immediate(static_cast<u32>(instr.bfe.GetLeftShiftValue())); - const Node outer_shift_imm = - Immediate(static_cast<u32>(instr.bfe.GetLeftShiftValue() + instr.bfe.shift_position)); + UNIMPLEMENTED_IF_MSG(instr.bfe.rd_cc, "Condition codes in BFE is not implemented"); - const Node inner_shift = - Operation(OperationCode::ILogicalShiftLeft, NO_PRECISE, op_a, inner_shift_imm); - const Node outer_shift = - Operation(OperationCode::ILogicalShiftRight, NO_PRECISE, inner_shift, outer_shift_imm); + const bool is_signed = instr.bfe.is_signed; - SetInternalFlagsFromInteger(bb, outer_shift, instr.generates_cc); - SetRegister(bb, instr.gpr0, outer_shift); - break; - } - default: - UNIMPLEMENTED_MSG("Unhandled BFE instruction: {}", opcode->get().GetName()); + // using reverse parallel method in + // https://graphics.stanford.edu/~seander/bithacks.html#ReverseParallel + // note for later if possible to implement faster method. + if (instr.bfe.brev) { + const auto swap = [&](u32 s, u32 mask) { + Node v1 = + SignedOperation(OperationCode::ILogicalShiftRight, is_signed, op_a, Immediate(s)); + if (mask != 0) { + v1 = SignedOperation(OperationCode::IBitwiseAnd, is_signed, std::move(v1), + Immediate(mask)); + } + Node v2 = op_a; + if (mask != 0) { + v2 = SignedOperation(OperationCode::IBitwiseAnd, is_signed, std::move(v2), + Immediate(mask)); + } + v2 = SignedOperation(OperationCode::ILogicalShiftLeft, is_signed, std::move(v2), + Immediate(s)); + return SignedOperation(OperationCode::IBitwiseOr, is_signed, std::move(v1), + std::move(v2)); + }; + op_a = swap(1, 0x55555555U); + op_a = swap(2, 0x33333333U); + op_a = swap(4, 0x0F0F0F0FU); + op_a = swap(8, 0x00FF00FFU); + op_a = swap(16, 0); } + const auto offset = SignedOperation(OperationCode::IBitfieldExtract, is_signed, op_b, + Immediate(0), Immediate(8)); + const auto bits = SignedOperation(OperationCode::IBitfieldExtract, is_signed, op_b, + Immediate(8), Immediate(8)); + auto result = SignedOperation(OperationCode::IBitfieldExtract, is_signed, op_a, offset, bits); + SetRegister(bb, instr.gpr0, std::move(result)); + return pc; } diff --git a/src/video_core/shader/decode/xmad.cpp b/src/video_core/shader/decode/xmad.cpp index 206961909..fbd7e9a17 100644 --- a/src/video_core/shader/decode/xmad.cpp +++ b/src/video_core/shader/decode/xmad.cpp @@ -12,6 +12,7 @@ namespace VideoCommon::Shader { using Tegra::Shader::Instruction; using Tegra::Shader::OpCode; +using Tegra::Shader::PredCondition; u32 ShaderIR::DecodeXmad(NodeBlock& bb, u32 pc) { const Instruction instr = {program_code[pc]}; @@ -63,15 +64,18 @@ u32 ShaderIR::DecodeXmad(NodeBlock& bb, u32 pc) { } }(); - op_a = BitfieldExtract(op_a, instr.xmad.high_a ? 16 : 0, 16); + op_a = SignedOperation(OperationCode::IBitfieldExtract, is_signed_a, std::move(op_a), + instr.xmad.high_a ? Immediate(16) : Immediate(0), Immediate(16)); const Node original_b = op_b; - op_b = BitfieldExtract(op_b, is_high_b ? 16 : 0, 16); + op_b = SignedOperation(OperationCode::IBitfieldExtract, is_signed_b, std::move(op_b), + is_high_b ? Immediate(16) : Immediate(0), Immediate(16)); - // TODO(Rodrigo): Use an appropiate sign for this operation - Node product = Operation(OperationCode::IMul, NO_PRECISE, op_a, op_b); + // we already check sign_a and sign_b is difference or not before so just use one in here. + Node product = SignedOperation(OperationCode::IMul, is_signed_a, op_a, op_b); if (is_psl) { - product = Operation(OperationCode::ILogicalShiftLeft, NO_PRECISE, product, Immediate(16)); + product = + SignedOperation(OperationCode::ILogicalShiftLeft, is_signed_a, product, Immediate(16)); } SetTemporary(bb, 0, product); product = GetTemporary(0); @@ -88,12 +92,40 @@ u32 ShaderIR::DecodeXmad(NodeBlock& bb, u32 pc) { return BitfieldExtract(original_c, 16, 16); case Tegra::Shader::XmadMode::CBcc: { const Node shifted_b = SignedOperation(OperationCode::ILogicalShiftLeft, is_signed_b, - NO_PRECISE, original_b, Immediate(16)); - return SignedOperation(OperationCode::IAdd, is_signed_c, NO_PRECISE, original_c, - shifted_b); + original_b, Immediate(16)); + return SignedOperation(OperationCode::IAdd, is_signed_c, original_c, shifted_b); + } + case Tegra::Shader::XmadMode::CSfu: { + const Node comp_a = GetPredicateComparisonInteger(PredCondition::Equal, is_signed_a, + op_a, Immediate(0)); + const Node comp_b = GetPredicateComparisonInteger(PredCondition::Equal, is_signed_b, + op_b, Immediate(0)); + const Node comp = Operation(OperationCode::LogicalOr, comp_a, comp_b); + + const Node comp_minus_a = GetPredicateComparisonInteger( + PredCondition::NotEqual, is_signed_a, + SignedOperation(OperationCode::IBitwiseAnd, is_signed_a, op_a, + Immediate(0x80000000)), + Immediate(0)); + const Node comp_minus_b = GetPredicateComparisonInteger( + PredCondition::NotEqual, is_signed_b, + SignedOperation(OperationCode::IBitwiseAnd, is_signed_b, op_b, + Immediate(0x80000000)), + Immediate(0)); + + Node new_c = Operation( + OperationCode::Select, comp_minus_a, + SignedOperation(OperationCode::IAdd, is_signed_c, original_c, Immediate(-65536)), + original_c); + new_c = Operation( + OperationCode::Select, comp_minus_b, + SignedOperation(OperationCode::IAdd, is_signed_c, new_c, Immediate(-65536)), + std::move(new_c)); + + return Operation(OperationCode::Select, comp, original_c, std::move(new_c)); } default: - UNIMPLEMENTED_MSG("Unhandled XMAD mode: {}", static_cast<u32>(instr.xmad.mode.Value())); + UNREACHABLE(); return Immediate(0); } }(); @@ -102,18 +134,19 @@ u32 ShaderIR::DecodeXmad(NodeBlock& bb, u32 pc) { op_c = GetTemporary(1); // TODO(Rodrigo): Use an appropiate sign for this operation - Node sum = Operation(OperationCode::IAdd, product, op_c); + Node sum = SignedOperation(OperationCode::IAdd, is_signed_a, product, std::move(op_c)); SetTemporary(bb, 2, sum); sum = GetTemporary(2); if (is_merge) { - const Node a = BitfieldExtract(sum, 0, 16); - const Node b = - Operation(OperationCode::ILogicalShiftLeft, NO_PRECISE, original_b, Immediate(16)); - sum = Operation(OperationCode::IBitwiseOr, NO_PRECISE, a, b); + const Node a = SignedOperation(OperationCode::IBitfieldExtract, is_signed_a, std::move(sum), + Immediate(0), Immediate(16)); + const Node b = SignedOperation(OperationCode::ILogicalShiftLeft, is_signed_b, original_b, + Immediate(16)); + sum = SignedOperation(OperationCode::IBitwiseOr, is_signed_a, a, b); } SetInternalFlagsFromInteger(bb, sum, instr.generates_cc); - SetRegister(bb, instr.gpr0, sum); + SetRegister(bb, instr.gpr0, std::move(sum)); return pc; } diff --git a/src/video_core/shader/node_helper.cpp b/src/video_core/shader/node_helper.cpp index b3dcd291c..76c56abb5 100644 --- a/src/video_core/shader/node_helper.cpp +++ b/src/video_core/shader/node_helper.cpp @@ -68,6 +68,8 @@ OperationCode SignedToUnsignedCode(OperationCode operation_code, bool is_signed) return OperationCode::UBitwiseXor; case OperationCode::IBitwiseNot: return OperationCode::UBitwiseNot; + case OperationCode::IBitfieldExtract: + return OperationCode::UBitfieldExtract; case OperationCode::IBitfieldInsert: return OperationCode::UBitfieldInsert; case OperationCode::IBitCount: |