suyu/src/video_core/shader/decode/video.cpp
ReinUsesLisp 76f178ba6e shader/video: Partially implement VMNMX
Implements the common usages for VMNMX. Inputs with a different size
than 32 bits are not supported and sign mismatches aren't supported
either.

VMNMX works as follows:
It grabs Ra and Rb and applies a maximum/minimum on them (this is
defined by .MX), having in mind the input sign. This result can then be
saturated. After the intermediate result is calculated, it applies
another operation on it using Rc. These operations are merges,
accumulations or another min/max pass.

This instruction allows to implement with a more flexible approach GCN's
min3 and max3 instructions (for instance).
2020-04-12 00:34:42 -03:00

171 lines
6.4 KiB
C++

// 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/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using std::move;
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
using Tegra::Shader::Pred;
using Tegra::Shader::VideoType;
using Tegra::Shader::VmadShr;
using Tegra::Shader::VmnmxOperation;
using Tegra::Shader::VmnmxType;
u32 ShaderIR::DecodeVideo(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
if (opcode->get().GetId() == OpCode::Id::VMNMX) {
DecodeVMNMX(bb, instr);
return pc;
}
const Node op_a =
GetVideoOperand(GetRegister(instr.gpr8), instr.video.is_byte_chunk_a, instr.video.signed_a,
instr.video.type_a, instr.video.byte_height_a);
const Node op_b = [this, instr] {
if (instr.video.use_register_b) {
return GetVideoOperand(GetRegister(instr.gpr20), instr.video.is_byte_chunk_b,
instr.video.signed_b, instr.video.type_b,
instr.video.byte_height_b);
}
if (instr.video.signed_b) {
const auto imm = static_cast<s16>(instr.alu.GetImm20_16());
return Immediate(static_cast<u32>(imm));
} else {
return Immediate(instr.alu.GetImm20_16());
}
}();
switch (opcode->get().GetId()) {
case OpCode::Id::VMAD: {
const bool result_signed = instr.video.signed_a == 1 || instr.video.signed_b == 1;
const Node op_c = GetRegister(instr.gpr39);
Node value = SignedOperation(OperationCode::IMul, result_signed, NO_PRECISE, op_a, op_b);
value = SignedOperation(OperationCode::IAdd, result_signed, NO_PRECISE, value, op_c);
if (instr.vmad.shr == VmadShr::Shr7 || instr.vmad.shr == VmadShr::Shr15) {
const Node shift = Immediate(instr.vmad.shr == VmadShr::Shr7 ? 7 : 15);
value =
SignedOperation(OperationCode::IArithmeticShiftRight, result_signed, value, shift);
}
SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
SetRegister(bb, instr.gpr0, value);
break;
}
case OpCode::Id::VSETP: {
// We can't use the constant predicate as destination.
ASSERT(instr.vsetp.pred3 != static_cast<u64>(Pred::UnusedIndex));
const bool sign = instr.video.signed_a == 1 || instr.video.signed_b == 1;
const Node first_pred = GetPredicateComparisonInteger(instr.vsetp.cond, sign, op_a, op_b);
const Node second_pred = GetPredicate(instr.vsetp.pred39, false);
const OperationCode combiner = GetPredicateCombiner(instr.vsetp.op);
// Set the primary predicate to the result of Predicate OP SecondPredicate
SetPredicate(bb, instr.vsetp.pred3, Operation(combiner, first_pred, second_pred));
if (instr.vsetp.pred0 != static_cast<u64>(Pred::UnusedIndex)) {
// Set the secondary predicate to the result of !Predicate OP SecondPredicate,
// if enabled
const Node negate_pred = Operation(OperationCode::LogicalNegate, first_pred);
SetPredicate(bb, instr.vsetp.pred0, Operation(combiner, negate_pred, second_pred));
}
break;
}
default:
UNIMPLEMENTED_MSG("Unhandled video instruction: {}", opcode->get().GetName());
}
return pc;
}
Node ShaderIR::GetVideoOperand(Node op, bool is_chunk, bool is_signed,
Tegra::Shader::VideoType type, u64 byte_height) {
if (!is_chunk) {
return BitfieldExtract(op, static_cast<u32>(byte_height * 8), 8);
}
const Node zero = Immediate(0);
switch (type) {
case Tegra::Shader::VideoType::Size16_Low:
return BitfieldExtract(op, 0, 16);
case Tegra::Shader::VideoType::Size16_High:
return BitfieldExtract(op, 16, 16);
case Tegra::Shader::VideoType::Size32:
// TODO(Rodrigo): From my hardware tests it becomes a bit "mad" when this type is used
// (1 * 1 + 0 == 0x5b800000). Until a better explanation is found: abort.
UNIMPLEMENTED();
return zero;
case Tegra::Shader::VideoType::Invalid:
UNREACHABLE_MSG("Invalid instruction encoding");
return zero;
default:
UNREACHABLE();
return zero;
}
}
void ShaderIR::DecodeVMNMX(NodeBlock& bb, Tegra::Shader::Instruction instr) {
UNIMPLEMENTED_IF(!instr.vmnmx.is_op_b_register);
UNIMPLEMENTED_IF(instr.vmnmx.SourceFormatA() != VmnmxType::Bits32);
UNIMPLEMENTED_IF(instr.vmnmx.SourceFormatB() != VmnmxType::Bits32);
UNIMPLEMENTED_IF(instr.vmnmx.is_src_a_signed != instr.vmnmx.is_src_b_signed);
UNIMPLEMENTED_IF(instr.vmnmx.sat);
UNIMPLEMENTED_IF(instr.generates_cc);
Node op_a = GetRegister(instr.gpr8);
Node op_b = GetRegister(instr.gpr20);
Node op_c = GetRegister(instr.gpr39);
const bool is_oper1_signed = instr.vmnmx.is_src_a_signed; // Stubbed
const bool is_oper2_signed = instr.vmnmx.is_dest_signed;
const auto operation_a = instr.vmnmx.mx ? OperationCode::IMax : OperationCode::IMin;
Node value = SignedOperation(operation_a, is_oper1_signed, move(op_a), move(op_b));
switch (instr.vmnmx.operation) {
case VmnmxOperation::Mrg_16H:
value = BitfieldInsert(move(op_c), move(value), 16, 16);
break;
case VmnmxOperation::Mrg_16L:
value = BitfieldInsert(move(op_c), move(value), 0, 16);
break;
case VmnmxOperation::Mrg_8B0:
value = BitfieldInsert(move(op_c), move(value), 0, 8);
break;
case VmnmxOperation::Mrg_8B2:
value = BitfieldInsert(move(op_c), move(value), 16, 8);
break;
case VmnmxOperation::Acc:
value = Operation(OperationCode::IAdd, move(value), move(op_c));
break;
case VmnmxOperation::Min:
value = SignedOperation(OperationCode::IMin, is_oper2_signed, move(value), move(op_c));
break;
case VmnmxOperation::Max:
value = SignedOperation(OperationCode::IMax, is_oper2_signed, move(value), move(op_c));
break;
case VmnmxOperation::Nop:
break;
default:
UNREACHABLE();
break;
}
SetRegister(bb, instr.gpr0, move(value));
}
} // namespace VideoCommon::Shader