1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
|
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <cstring>
#include <set>
#include <fmt/format.h>
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/engines/shader_header.h"
#include "video_core/shader/control_flow.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
namespace {
/**
* Returns whether the instruction at the specified offset is a 'sched' instruction.
* Sched instructions always appear before a sequence of 3 instructions.
*/
constexpr bool IsSchedInstruction(u32 offset, u32 main_offset) {
constexpr u32 SchedPeriod = 4;
u32 absolute_offset = offset - main_offset;
return (absolute_offset % SchedPeriod) == 0;
}
} // namespace
void ShaderIR::Decode() {
std::memcpy(&header, program_code.data(), sizeof(Tegra::Shader::Header));
ShaderCharacteristics shader_info{};
bool can_proceed = ScanFlow(program_code, program_code.size(), main_offset, shader_info);
if (can_proceed) {
coverage_begin = shader_info.start;
coverage_end = shader_info.end;
if (shader_info.decompilable) {
return;
}
// we can't decompile it, fallback to standard method
for (const auto& block : shader_info.blocks) {
basic_blocks.insert({block.start, DecodeRange(block.start, block.end + 1)});
}
return;
}
LOG_WARNING(HW_GPU, "Flow Analysis failed, falling back to brute force compiling");
// Now we need to deal with an undecompilable shader. We need to brute force
// a shader that captures every position.
coverage_begin = shader_info.start;
const u32 shader_end = static_cast<u32>(program_size / sizeof(u64));
coverage_end = shader_end;
for (u32 label = main_offset; label < shader_end; label++) {
basic_blocks.insert({label, DecodeRange(label, label + 1)});
}
return;
}
NodeBlock ShaderIR::DecodeRange(u32 begin, u32 end) {
NodeBlock basic_block;
for (u32 pc = begin; pc < (begin > end ? MAX_PROGRAM_LENGTH : end);) {
pc = DecodeInstr(basic_block, pc);
}
return basic_block;
}
u32 ShaderIR::DecodeInstr(NodeBlock& bb, u32 pc) {
// Ignore sched instructions when generating code.
if (IsSchedInstruction(pc, main_offset)) {
return pc + 1;
}
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
// Decoding failure
if (!opcode) {
UNIMPLEMENTED_MSG("Unhandled instruction: {0:x}", instr.value);
return pc + 1;
}
bb.push_back(
Comment(fmt::format("{}: {} (0x{:016x})", pc, opcode->get().GetName(), instr.value)));
using Tegra::Shader::Pred;
UNIMPLEMENTED_IF_MSG(instr.pred.full_pred == Pred::NeverExecute,
"NeverExecute predicate not implemented");
static const std::map<OpCode::Type, u32 (ShaderIR::*)(NodeBlock&, u32)> decoders = {
{OpCode::Type::Arithmetic, &ShaderIR::DecodeArithmetic},
{OpCode::Type::ArithmeticImmediate, &ShaderIR::DecodeArithmeticImmediate},
{OpCode::Type::Bfe, &ShaderIR::DecodeBfe},
{OpCode::Type::Bfi, &ShaderIR::DecodeBfi},
{OpCode::Type::Shift, &ShaderIR::DecodeShift},
{OpCode::Type::ArithmeticInteger, &ShaderIR::DecodeArithmeticInteger},
{OpCode::Type::ArithmeticIntegerImmediate, &ShaderIR::DecodeArithmeticIntegerImmediate},
{OpCode::Type::ArithmeticHalf, &ShaderIR::DecodeArithmeticHalf},
{OpCode::Type::ArithmeticHalfImmediate, &ShaderIR::DecodeArithmeticHalfImmediate},
{OpCode::Type::Ffma, &ShaderIR::DecodeFfma},
{OpCode::Type::Hfma2, &ShaderIR::DecodeHfma2},
{OpCode::Type::Conversion, &ShaderIR::DecodeConversion},
{OpCode::Type::Memory, &ShaderIR::DecodeMemory},
{OpCode::Type::Texture, &ShaderIR::DecodeTexture},
{OpCode::Type::Image, &ShaderIR::DecodeImage},
{OpCode::Type::FloatSetPredicate, &ShaderIR::DecodeFloatSetPredicate},
{OpCode::Type::IntegerSetPredicate, &ShaderIR::DecodeIntegerSetPredicate},
{OpCode::Type::HalfSetPredicate, &ShaderIR::DecodeHalfSetPredicate},
{OpCode::Type::PredicateSetRegister, &ShaderIR::DecodePredicateSetRegister},
{OpCode::Type::PredicateSetPredicate, &ShaderIR::DecodePredicateSetPredicate},
{OpCode::Type::RegisterSetPredicate, &ShaderIR::DecodeRegisterSetPredicate},
{OpCode::Type::FloatSet, &ShaderIR::DecodeFloatSet},
{OpCode::Type::IntegerSet, &ShaderIR::DecodeIntegerSet},
{OpCode::Type::HalfSet, &ShaderIR::DecodeHalfSet},
{OpCode::Type::Video, &ShaderIR::DecodeVideo},
{OpCode::Type::Xmad, &ShaderIR::DecodeXmad},
};
std::vector<Node> tmp_block;
if (const auto decoder = decoders.find(opcode->get().GetType()); decoder != decoders.end()) {
pc = (this->*decoder->second)(tmp_block, pc);
} else {
pc = DecodeOther(tmp_block, pc);
}
// Some instructions (like SSY) don't have a predicate field, they are always unconditionally
// executed.
const bool can_be_predicated = OpCode::IsPredicatedInstruction(opcode->get().GetId());
const auto pred_index = static_cast<u32>(instr.pred.pred_index);
if (can_be_predicated && pred_index != static_cast<u32>(Pred::UnusedIndex)) {
const Node conditional =
Conditional(GetPredicate(pred_index, instr.negate_pred != 0), std::move(tmp_block));
global_code.push_back(conditional);
bb.push_back(conditional);
} else {
for (auto& node : tmp_block) {
global_code.push_back(node);
bb.push_back(node);
}
}
return pc + 1;
}
} // namespace VideoCommon::Shader
|
