N-archive/suyu
1
0
Fork 0
mirror of https://git.suyu.dev/suyu/suyu.git synced 2024-11-23 15:22:45 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions

shader_ir: Correct parsing of scheduling instructions and correct sizing

Browse source
This commit is contained in:
Fernando Sahmkow 2019-06-26 12:19:43 -04:00 committed by FernandoS27
parent cfb3db1a32
commit 34357b110c
2 changed files with 30 additions and 13 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

41
src/video_core/shader/control_flow.cpp
View file

@ -20,10 +20,10 @@ using Tegra::Shader::OpCode;
constexpr s32 unassigned_branch = -2; constexpr s32 unassigned_branch = -2;
/*** /**
* 'ControlStack' represents a static stack of control jumps such as SSY and PBK * 'ControlStack' represents a static stack of control jumps such as SSY and PBK
* stacks in Maxwell. * stacks in Maxwell.
***/ **/
struct ControlStack { struct ControlStack {
static constexpr std::size_t stack_fixed_size = 20; static constexpr std::size_t stack_fixed_size = 20;
std::array<u32, stack_fixed_size> stack{}; std::array<u32, stack_fixed_size> stack{};
@ -105,9 +105,11 @@ struct BlockInfo {
}; };
struct CFGRebuildState { struct CFGRebuildState {
explicit CFGRebuildState(const ProgramCode& program_code, const std::size_t program_size) explicit CFGRebuildState(const ProgramCode& program_code, const std::size_t program_size,
: program_code{program_code}, program_size{program_size} {} const u32 start)
: program_code{program_code}, program_size{program_size}, start{start} {}
u32 start{};
std::vector<BlockInfo> block_info{}; std::vector<BlockInfo> block_info{};
std::list<u32> inspect_queries{}; std::list<u32> inspect_queries{};
std::list<Query> queries{}; std::list<Query> queries{};
@ -120,7 +122,7 @@ struct CFGRebuildState {
const std::size_t program_size; const std::size_t program_size;
}; };
enum class BlockCollision : u32 { None = 0, Found = 1, Inside = 2 }; enum class BlockCollision : u32 { None, Found, Inside };
std::pair<BlockCollision, std::vector<BlockInfo>::iterator> TryGetBlock(CFGRebuildState& state, std::pair<BlockCollision, std::vector<BlockInfo>::iterator> TryGetBlock(CFGRebuildState& state,
u32 address) { u32 address) {
@ -155,15 +157,26 @@ Pred GetPredicate(u32 index, bool negated) {
return static_cast<Pred>(index + (negated ? 8 : 0)); return static_cast<Pred>(index + (negated ? 8 : 0));
} }
/**
* 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;
}
enum class ParseResult : u32 { enum class ParseResult : u32 {
ControlCaught = 0, ControlCaught,
BlockEnd = 1, BlockEnd,
AbnormalFlow = 2, AbnormalFlow,
}; };
ParseResult ParseCode(CFGRebuildState& state, u32 address, ParseInfo& parse_info) { ParseResult ParseCode(CFGRebuildState& state, u32 address, ParseInfo& parse_info) {
u32 offset = static_cast<u32>(address); u32 offset = static_cast<u32>(address);
const u32 end_address = static_cast<u32>(state.program_size / 8U); const u32 end_address = static_cast<u32>(state.program_size / sizeof(Instruction));
const auto insert_label = ([](CFGRebuildState& state, u32 address) { const auto insert_label = ([](CFGRebuildState& state, u32 address) {
auto pair = state.labels.emplace(address); auto pair = state.labels.emplace(address);
@ -183,6 +196,10 @@ ParseResult ParseCode(CFGRebuildState& state, u32 address, ParseInfo& parse_info
parse_info.branch_info.ignore = true; parse_info.branch_info.ignore = true;
break; break;
} }
if (IsSchedInstruction(offset, state.start)) {
offset++;
continue;
}
const Instruction instr = {state.program_code[offset]}; const Instruction instr = {state.program_code[offset]};
const auto opcode = OpCode::Decode(instr); const auto opcode = OpCode::Decode(instr);
if (!opcode || opcode->get().GetType() != OpCode::Type::Flow) { if (!opcode || opcode->get().GetType() != OpCode::Type::Flow) {
@ -447,11 +464,11 @@ bool TryQuery(CFGRebuildState& state) {
std::optional<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 program_size, std::optional<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 program_size,
u32 start_address) { u32 start_address) {
CFGRebuildState state{program_code, program_size}; CFGRebuildState state{program_code, program_size, start_address};
// Inspect Code and generate blocks // Inspect Code and generate blocks
state.labels.clear(); state.labels.clear();
state.labels.emplace(start_address); state.labels.emplace(start_address);
state.inspect_queries.push_back(start_address); state.inspect_queries.push_back(state.start);
while (!state.inspect_queries.empty()) { while (!state.inspect_queries.empty()) {
if (!TryInspectAddress(state)) { if (!TryInspectAddress(state)) {
return {}; return {};
@ -459,7 +476,7 @@ std::optional<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u
} }
// Decompile Stacks // Decompile Stacks
Query start_query{}; Query start_query{};
start_query.address = start_address; start_query.address = state.start;
state.queries.push_back(start_query); state.queries.push_back(start_query);
bool decompiled = true; bool decompiled = true;
while (!state.queries.empty()) { while (!state.queries.empty()) {

2
src/video_core/shader/decode.cpp
View file

@ -39,7 +39,7 @@ void ShaderIR::Decode() {
std::memcpy(&header, program_code.data(), sizeof(Tegra::Shader::Header)); std::memcpy(&header, program_code.data(), sizeof(Tegra::Shader::Header));
disable_flow_stack = false; disable_flow_stack = false;
const auto info = ScanFlow(program_code, program_size, main_offset); const auto info = ScanFlow(program_code, MAX_PROGRAM_LENGTH * sizeof(u64), main_offset);
if (info) { if (info) {
const auto& shader_info = *info; const auto& shader_info = *info;
coverage_begin = shader_info.start; coverage_begin = shader_info.start;