Compare commits

...

2 Commits

Author SHA1 Message Date
lizzie 2df264fa48 [shader_recompiler] use reusable stable_vector<> instead of object pools
Signed-off-by: lizzie <lizzie@eden-emu.dev>
2026-07-31 19:10:43 +02:00
simply0001 54046ac60e [video_core/macro] check HLE hashes before compiling (#4236)
- [x] I have read and followed the [Contribution Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/CONTRIBUTING.md#code-contributions).
- [x] I have read and followed the [AI Policy](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/AI.md)
- [x] I have read and followed the [Coding Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/Coding.md) to the best of my ability.

-------------------

Known HLE macros are identified by a hash, but MacroEngine compiled them first and and afterwards it threw the compiled program away when the hash matched. This fix makes it so it checks the hash first and caches the HLE implementation directly, so it only compiles when the hash is unknown or if HLE is disabled.

Cached macros were also constantly checking the hash again and walking through each `std::get_if` until their variant matched. So I dispatched them through `std::visit` instead, and keep one resolved code span for hashing, compiling, and dumping so mid-method uploads use the right range.

Continues the macro hot path work from [#4067](https://git.eden-emu.dev/eden-emu/eden/pulls/4067)

Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4236
Reviewed-by: Shinmegumi <shinmegumi@eden-emu.dev>
Reviewed-by: Lizzie <lizzie@eden-emu.dev>
2026-07-30 06:25:43 +02:00
18 changed files with 394 additions and 568 deletions
+1 -2
View File
@@ -234,13 +234,12 @@ add_library(shader_recompiler STATIC
ir_opt/texture_pass.cpp
ir_opt/vendor_workaround_pass.cpp
ir_opt/verification_pass.cpp
object_pool.h
profile.h
program_header.h
runtime_info.h
shader_info.h
varying_state.h
shader_pool.h
)
target_link_libraries(shader_recompiler PUBLIC common fmt::fmt sirit::sirit)
@@ -14,7 +14,7 @@
namespace Shader::IR {
Block::Block(ObjectPool<Inst>& inst_pool_) : inst_pool{&inst_pool_} {}
Block::Block(boost::container::stable_vector<Inst>& inst_pool_) : inst_pool{&inst_pool_} {}
Block::~Block() = default;
@@ -23,13 +23,12 @@ void Block::AppendNewInst(Opcode op, std::initializer_list<Value> args) {
}
Block::iterator Block::PrependNewInst(iterator insertion_point, const Inst& base_inst) {
Inst* const inst{inst_pool->Create(base_inst)};
Inst* const inst{&inst_pool->emplace_back(base_inst)};
return instructions.insert(insertion_point, *inst);
}
Block::iterator Block::PrependNewInst(iterator insertion_point, Opcode op,
std::initializer_list<Value> args, u32 flags) {
Inst* const inst{inst_pool->Create(op, flags)};
Block::iterator Block::PrependNewInst(iterator insertion_point, Opcode op, std::initializer_list<Value> args, u32 flags) {
Inst* const inst{&inst_pool->emplace_back(op, flags)};
const auto result_it{instructions.insert(insertion_point, *inst)};
if (inst->NumArgs() != args.size()) {
@@ -53,12 +52,10 @@ void Block::AddBranch(Block* block) {
block->imm_predecessors.push_back(this);
}
static std::string BlockToIndex(const std::map<const Block*, size_t>& block_to_index,
Block* block) {
if (const auto it{block_to_index.find(block)}; it != block_to_index.end()) {
static std::string BlockToIndex(const std::map<const Block*, size_t>& block_to_index, Block* block) {
if (const auto it{block_to_index.find(block)}; it != block_to_index.end())
return fmt::format("{{Block ${}}}", it->second);
}
return fmt::format("$<unknown block {:016x}>", reinterpret_cast<u64>(block));
return fmt::format("$<unknown block {:016x}>", u64(block));
}
static size_t InstIndex(std::map<const Inst*, size_t>& inst_to_index, size_t& inst_index,
@@ -1,4 +1,4 @@
// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
@@ -13,11 +13,11 @@
#include <bit>
#include <numeric>
#include <boost/intrusive/list.hpp>
#include <boost/container/stable_vector.hpp>
#include "common/common_types.h"
#include "shader_recompiler/frontend/ir/condition.h"
#include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/object_pool.h"
namespace Shader::IR {
@@ -30,7 +30,7 @@ public:
using reverse_iterator = InstructionList::reverse_iterator;
using const_reverse_iterator = InstructionList::const_reverse_iterator;
explicit Block(ObjectPool<Inst>& inst_pool_);
explicit Block(boost::container::stable_vector<Inst>& inst_pool_);
~Block();
Block(const Block&) = delete;
@@ -170,7 +170,7 @@ public:
private:
/// Memory pool for instruction list
ObjectPool<Inst>* inst_pool;
boost::container::stable_vector<Inst>* inst_pool;
/// List of instructions in this block
InstructionList instructions;
@@ -1,4 +1,4 @@
// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
@@ -174,11 +174,10 @@ bool Block::Contains(Location pc) const noexcept {
return pc >= begin && pc < end;
}
Function::Function(ObjectPool<Block>& block_pool, Location start_address)
: entrypoint{start_address} {
Function::Function(boost::container::stable_vector<Block>& block_pool, Location start_address) : entrypoint{start_address} {
Label& label{labels.emplace_back()};
label.address = start_address;
label.block = block_pool.Create(Block{});
label.block = &block_pool.emplace_back(Block{});
label.block->begin = start_address;
label.block->end = start_address;
label.block->end_class = EndClass::Branch;
@@ -187,12 +186,13 @@ Function::Function(ObjectPool<Block>& block_pool, Location start_address)
label.block->branch_false = nullptr;
}
CFG::CFG(Environment& env_, ObjectPool<Block>& block_pool_, Location start_address,
bool exits_to_dispatcher_)
: env{env_}, block_pool{block_pool_}, program_start{start_address}, exits_to_dispatcher{
exits_to_dispatcher_} {
CFG::CFG(Environment& env_, boost::container::stable_vector<Block>& block_pool_, Location start_address, bool exits_to_dispatcher_)
: env{env_}
, block_pool{block_pool_}
, program_start{start_address}
, exits_to_dispatcher{exits_to_dispatcher_} {
if (exits_to_dispatcher) {
dispatch_block = block_pool.Create(Block{});
dispatch_block = &block_pool.emplace_back(Block{});
dispatch_block->begin = {};
dispatch_block->end = {};
dispatch_block->end_class = EndClass::Exit;
@@ -371,7 +371,7 @@ void CFG::AnalyzeCondInst(Block* block, FunctionId function_id, Location pc,
return;
}
// Create a virtual block and a conditional block
Block* const conditional_block{block_pool.Create()};
Block* const conditional_block{&block_pool.emplace_back()};
Block virtual_block{};
virtual_block.begin = block->begin.Virtual();
virtual_block.end = block->begin.Virtual();
@@ -546,7 +546,7 @@ Block* CFG::AddLabel(Block* block, Stack stack, Location pc, FunctionId function
if (label_it != function.labels.end()) {
return label_it->block;
}
Block* const new_block{block_pool.Create()};
Block* const new_block{&block_pool.emplace_back()};
new_block->begin = pc;
new_block->end = pc;
new_block->end_class = EndClass::Branch;
@@ -1,3 +1,6 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
@@ -9,6 +12,7 @@
#include <vector>
#include <boost/container/small_vector.hpp>
#include <boost/container/stable_vector.hpp>
#include <boost/intrusive/set.hpp>
#include "shader_recompiler/environment.h"
@@ -17,7 +21,6 @@
#include "shader_recompiler/frontend/maxwell/instruction.h"
#include "shader_recompiler/frontend/maxwell/location.h"
#include "shader_recompiler/frontend/maxwell/opcodes.h"
#include "shader_recompiler/object_pool.h"
namespace Shader::Maxwell::Flow {
@@ -96,7 +99,7 @@ struct Label {
};
struct Function {
explicit Function(ObjectPool<Block>& block_pool, Location start_address);
explicit Function(boost::container::stable_vector<Block>& block_pool, Location start_address);
Location entrypoint;
boost::container::small_vector<Label, 16> labels;
@@ -110,8 +113,7 @@ class CFG {
};
public:
explicit CFG(Environment& env, ObjectPool<Block>& block_pool, Location start_address,
bool exits_to_dispatcher = false);
explicit CFG(Environment& env, boost::container::stable_vector<Block>& block_pool, Location start_address, bool exits_to_dispatcher = false);
CFG& operator=(const CFG&) = delete;
CFG(const CFG&) = delete;
@@ -138,27 +140,20 @@ private:
/// Inspect already visited blocks.
/// Return true when the block has already been visited
bool InspectVisitedBlocks(FunctionId function_id, const Label& label);
AnalysisState AnalyzeInst(Block* block, FunctionId function_id, Location pc);
void AnalyzeCondInst(Block* block, FunctionId function_id, Location pc, EndClass insn_end_class,
IR::Condition cond);
void AnalyzeCondInst(Block* block, FunctionId function_id, Location pc, EndClass insn_end_class, IR::Condition cond);
/// Return true when the branch instruction is confirmed to be a branch
bool AnalyzeBranch(Block* block, FunctionId function_id, Location pc, Instruction inst,
Opcode opcode);
void AnalyzeBRA(Block* block, FunctionId function_id, Location pc, Instruction inst,
bool is_absolute);
AnalysisState AnalyzeBRX(Block* block, Location pc, Instruction inst, bool is_absolute,
FunctionId function_id);
bool AnalyzeBranch(Block* block, FunctionId function_id, Location pc, Instruction inst, Opcode opcode);
void AnalyzeBRA(Block* block, FunctionId function_id, Location pc, Instruction inst, bool is_absolute);
AnalysisState AnalyzeBRX(Block* block, Location pc, Instruction inst, bool is_absolute, FunctionId function_id);
AnalysisState AnalyzeEXIT(Block* block, FunctionId function_id, Location pc, Instruction inst);
/// Return the branch target block id
Block* AddLabel(Block* block, Stack stack, Location pc, FunctionId function_id);
Environment& env;
ObjectPool<Block>& block_pool;
boost::container::stable_vector<Block>& block_pool;
boost::container::small_vector<Function, 1> functions;
Location program_start;
bool exits_to_dispatcher{};
@@ -13,143 +13,19 @@
#include <fmt/ranges.h>
#include <boost/intrusive/list.hpp>
#include <ranges>
#include "shader_recompiler/shader_pool.h"
#include "shader_recompiler/environment.h"
#include "shader_recompiler/frontend/ir/basic_block.h"
#include "shader_recompiler/frontend/ir/ir_emitter.h"
#include "shader_recompiler/frontend/maxwell/structured_control_flow.h"
#include "shader_recompiler/frontend/maxwell/translate/translate.h"
#include "shader_recompiler/frontend/maxwell/translate_program.h"
#include "shader_recompiler/frontend/maxwell/control_flow.h"
#include "shader_recompiler/host_translate_info.h"
#include "shader_recompiler/object_pool.h"
namespace Shader::Maxwell {
namespace {
struct Statement;
// Use normal_link because we are not guaranteed to destroy the tree in order
using ListBaseHook =
boost::intrusive::list_base_hook<boost::intrusive::link_mode<boost::intrusive::normal_link>>;
using Tree = boost::intrusive::list<Statement,
// Allow using Statement without a definition
boost::intrusive::base_hook<ListBaseHook>,
// Avoid linear complexity on splice, size is never called
boost::intrusive::constant_time_size<false>>;
using Node = Tree::iterator;
enum class StatementType {
Code,
Goto,
Label,
If,
Loop,
Break,
Return,
Kill,
Unreachable,
Function,
Identity,
Not,
Or,
SetVariable,
SetIndirectBranchVariable,
Variable,
IndirectBranchCond,
};
bool HasChildren(StatementType type) {
switch (type) {
case StatementType::If:
case StatementType::Loop:
case StatementType::Function:
return true;
default:
return false;
}
}
struct Goto {};
struct Label {};
struct If {};
struct Loop {};
struct Break {};
struct Return {};
struct Kill {};
struct Unreachable {};
struct FunctionTag {};
struct Identity {};
struct Not {};
struct Or {};
struct SetVariable {};
struct SetIndirectBranchVariable {};
struct Variable {};
struct IndirectBranchCond {};
#ifdef _MSC_VER
#pragma warning(push)
#pragma warning(disable : 26495) // Always initialize a member variable, expected in Statement
#endif
struct Statement : ListBaseHook {
Statement(const Flow::Block* block_, Statement* up_)
: block{block_}, up{up_}, type{StatementType::Code} {}
Statement(Goto, Statement* cond_, Node label_, Statement* up_)
: label{label_}, cond{cond_}, up{up_}, type{StatementType::Goto} {}
Statement(Label, u32 id_, Statement* up_) : id{id_}, up{up_}, type{StatementType::Label} {}
Statement(If, Statement* cond_, Tree&& children_, Statement* up_)
: children{std::move(children_)}, cond{cond_}, up{up_}, type{StatementType::If} {}
Statement(Loop, Statement* cond_, Tree&& children_, Statement* up_)
: children{std::move(children_)}, cond{cond_}, up{up_}, type{StatementType::Loop} {}
Statement(Break, Statement* cond_, Statement* up_)
: cond{cond_}, up{up_}, type{StatementType::Break} {}
Statement(Return, Statement* up_) : up{up_}, type{StatementType::Return} {}
Statement(Kill, Statement* up_) : up{up_}, type{StatementType::Kill} {}
Statement(Unreachable, Statement* up_) : up{up_}, type{StatementType::Unreachable} {}
Statement(FunctionTag) : children{}, type{StatementType::Function} {}
Statement(Identity, IR::Condition cond_, Statement* up_)
: guest_cond{cond_}, up{up_}, type{StatementType::Identity} {}
Statement(Not, Statement* op_, Statement* up_) : op{op_}, up{up_}, type{StatementType::Not} {}
Statement(Or, Statement* op_a_, Statement* op_b_, Statement* up_)
: op_a{op_a_}, op_b{op_b_}, up{up_}, type{StatementType::Or} {}
Statement(SetVariable, u32 id_, Statement* op_, Statement* up_)
: op{op_}, id{id_}, up{up_}, type{StatementType::SetVariable} {}
Statement(SetIndirectBranchVariable, IR::Reg branch_reg_, s32 branch_offset_, Statement* up_)
: branch_offset{branch_offset_},
branch_reg{branch_reg_}, up{up_}, type{StatementType::SetIndirectBranchVariable} {}
Statement(Variable, u32 id_, Statement* up_)
: id{id_}, up{up_}, type{StatementType::Variable} {}
Statement(IndirectBranchCond, u32 location_, Statement* up_)
: location{location_}, up{up_}, type{StatementType::IndirectBranchCond} {}
~Statement() {
if (HasChildren(type)) {
std::destroy_at(&children);
}
}
union {
const Flow::Block* block;
Node label;
Tree children;
IR::Condition guest_cond;
Statement* op;
Statement* op_a;
u32 location;
s32 branch_offset;
};
union {
Statement* cond;
Statement* op_b;
u32 id;
IR::Reg branch_reg;
};
Statement* up{};
StatementType type;
};
#ifdef _MSC_VER
#pragma warning(pop)
#endif
std::string DumpExpr(const Statement* stmt) {
switch (stmt->type) {
@@ -314,7 +190,9 @@ bool NeedsLift(Node goto_stmt, Node label_stmt) noexcept {
class GotoPass {
public:
explicit GotoPass(Flow::CFG& cfg, ObjectPool<Statement>& stmt_pool) : pool{stmt_pool} {
explicit GotoPass(Flow::CFG& cfg, ShaderPools& pools_)
: pools{pools_}
{
std::vector gotos{BuildTree(cfg)};
const auto end{gotos.rend()};
for (auto goto_stmt = gotos.rbegin(); goto_stmt != end; ++goto_stmt) {
@@ -385,16 +263,14 @@ private:
return gotos;
}
void BuildTree(Flow::CFG& cfg, Flow::Function& function, u32& label_id,
std::vector<Node>& gotos, Node function_insert_point,
std::optional<Node> return_label) {
Statement* const false_stmt{pool.Create(Identity{}, IR::Condition{false}, &root_stmt)};
void BuildTree(Flow::CFG& cfg, Flow::Function& function, u32& label_id, std::vector<Node>& gotos, Node function_insert_point, std::optional<Node> return_label) {
Statement* const false_stmt{&pools.stmt.emplace_back(Identity{}, IR::Condition{false}, &root_stmt)};
Tree& root{root_stmt.children};
ankerl::unordered_dense::map<Flow::Block*, Node> local_labels;
local_labels.reserve(function.blocks.size());
for (Flow::Block& block : function.blocks) {
Statement* const label{pool.Create(Label{}, label_id, &root_stmt)};
Statement* const label{&pools.stmt.emplace_back(Label{}, label_id, &root_stmt)};
const Node label_it{root.insert(function_insert_point, *label)};
local_labels.emplace(&block, label_it);
++label_id;
@@ -406,46 +282,39 @@ private:
// Reset goto variables before the first block and after its respective label
const auto make_reset_variable{[&]() -> Statement& {
return *pool.Create(SetVariable{}, label->id, false_stmt, &root_stmt);
return pools.stmt.emplace_back(SetVariable{}, label->id, false_stmt, &root_stmt);
}};
root.push_front(make_reset_variable());
root.insert(ip, make_reset_variable());
root.insert(ip, *pool.Create(&block, &root_stmt));
root.insert(ip, pools.stmt.emplace_back(&block, &root_stmt));
switch (block.end_class) {
case Flow::EndClass::Branch: {
Statement* const always_cond{
pool.Create(Identity{}, IR::Condition{true}, &root_stmt)};
Statement* const always_cond{&pools.stmt.emplace_back(Identity{}, IR::Condition{true}, &root_stmt)};
if (block.cond == IR::Condition{true}) {
const Node true_label{local_labels.at(block.branch_true)};
gotos.push_back(
root.insert(ip, *pool.Create(Goto{}, always_cond, true_label, &root_stmt)));
gotos.push_back(root.insert(ip, pools.stmt.emplace_back(Goto{}, always_cond, true_label, &root_stmt)));
} else if (block.cond == IR::Condition{false}) {
const Node false_label{local_labels.at(block.branch_false)};
gotos.push_back(root.insert(
ip, *pool.Create(Goto{}, always_cond, false_label, &root_stmt)));
gotos.push_back(root.insert(ip, pools.stmt.emplace_back(Goto{}, always_cond, false_label, &root_stmt)));
} else {
const Node true_label{local_labels.at(block.branch_true)};
const Node false_label{local_labels.at(block.branch_false)};
Statement* const true_cond{pool.Create(Identity{}, block.cond, &root_stmt)};
gotos.push_back(
root.insert(ip, *pool.Create(Goto{}, true_cond, true_label, &root_stmt)));
gotos.push_back(root.insert(
ip, *pool.Create(Goto{}, always_cond, false_label, &root_stmt)));
Statement* const true_cond{&pools.stmt.emplace_back(Identity{}, block.cond, &root_stmt)};
gotos.push_back(root.insert(ip, pools.stmt.emplace_back(Goto{}, true_cond, true_label, &root_stmt)));
gotos.push_back(root.insert(ip, pools.stmt.emplace_back(Goto{}, always_cond, false_label, &root_stmt)));
}
break;
}
case Flow::EndClass::IndirectBranch:
root.insert(ip, *pool.Create(SetIndirectBranchVariable{}, block.branch_reg,
block.branch_offset, &root_stmt));
root.insert(ip, pools.stmt.emplace_back(SetIndirectBranchVariable{}, block.branch_reg, block.branch_offset, &root_stmt));
for (const Flow::IndirectBranch& indirect : block.indirect_branches) {
const Node indirect_label{local_labels.at(indirect.block)};
Statement* cond{
pool.Create(IndirectBranchCond{}, indirect.address, &root_stmt)};
Statement* goto_stmt{pool.Create(Goto{}, cond, indirect_label, &root_stmt)};
Statement* cond{&pools.stmt.emplace_back(IndirectBranchCond{}, indirect.address, &root_stmt)};
Statement* goto_stmt{&pools.stmt.emplace_back(Goto{}, cond, indirect_label, &root_stmt)};
gotos.push_back(root.insert(ip, *goto_stmt));
}
root.insert(ip, *pool.Create(Unreachable{}, &root_stmt));
root.insert(ip, pools.stmt.emplace_back(Unreachable{}, &root_stmt));
break;
case Flow::EndClass::Call: {
Flow::Function& call{cfg.Functions()[block.function_call]};
@@ -454,16 +323,16 @@ private:
break;
}
case Flow::EndClass::Exit:
root.insert(ip, *pool.Create(Return{}, &root_stmt));
root.insert(ip, pools.stmt.emplace_back(Return{}, &root_stmt));
break;
case Flow::EndClass::Return: {
Statement* const always_cond{pool.Create(Identity{}, block.cond, &root_stmt)};
auto goto_stmt{pool.Create(Goto{}, always_cond, return_label.value(), &root_stmt)};
Statement* const always_cond{&pools.stmt.emplace_back(Identity{}, block.cond, &root_stmt)};
auto goto_stmt{&pools.stmt.emplace_back(Goto{}, always_cond, return_label.value(), &root_stmt)};
gotos.push_back(root.insert(ip, *goto_stmt));
break;
}
case Flow::EndClass::Kill:
root.insert(ip, *pool.Create(Kill{}, &root_stmt));
root.insert(ip, pools.stmt.emplace_back(Kill{}, &root_stmt));
break;
}
}
@@ -479,8 +348,8 @@ private:
Tree& body{goto_stmt->up->children};
Tree if_body;
if_body.splice(if_body.begin(), body, std::next(goto_stmt), label_stmt);
Statement* const cond{pool.Create(Not{}, goto_stmt->cond, &root_stmt)};
Statement* const if_stmt{pool.Create(If{}, cond, std::move(if_body), goto_stmt->up)};
Statement* const cond{&pools.stmt.emplace_back(Not{}, goto_stmt->cond, &root_stmt)};
Statement* const if_stmt{&pools.stmt.emplace_back(If{}, cond, std::move(if_body), goto_stmt->up)};
UpdateTreeUp(if_stmt);
body.insert(goto_stmt, *if_stmt);
body.erase(goto_stmt);
@@ -491,7 +360,7 @@ private:
Tree loop_body;
loop_body.splice(loop_body.begin(), body, label_stmt, goto_stmt);
Statement* const cond{goto_stmt->cond};
Statement* const loop{pool.Create(Loop{}, cond, std::move(loop_body), goto_stmt->up)};
Statement* const loop{&pools.stmt.emplace_back(Loop{}, cond, std::move(loop_body), goto_stmt->up)};
UpdateTreeUp(loop);
body.insert(goto_stmt, *loop);
body.erase(goto_stmt);
@@ -516,15 +385,15 @@ private:
const u32 label_id{label->id};
Statement* const goto_cond{goto_stmt->cond};
Statement* const set_var{pool.Create(SetVariable{}, label_id, goto_cond, parent)};
Statement* const set_var{&pools.stmt.emplace_back(SetVariable{}, label_id, goto_cond, parent)};
body.insert(goto_stmt, *set_var);
Tree if_body;
if_body.splice(if_body.begin(), body, std::next(goto_stmt), label_nested_stmt);
Statement* const variable{pool.Create(Variable{}, label_id, &root_stmt)};
Statement* const neg_var{pool.Create(Not{}, variable, &root_stmt)};
Statement* const variable{&pools.stmt.emplace_back(Variable{}, label_id, &root_stmt)};
Statement* const neg_var{&pools.stmt.emplace_back(Not{}, variable, &root_stmt)};
if (!if_body.empty()) {
Statement* const if_stmt{pool.Create(If{}, neg_var, std::move(if_body), parent)};
Statement* const if_stmt{&pools.stmt.emplace_back(If{}, neg_var, std::move(if_body), parent)};
UpdateTreeUp(if_stmt);
body.insert(goto_stmt, *if_stmt);
}
@@ -533,8 +402,7 @@ private:
switch (label_nested_stmt->type) {
case StatementType::If:
// Update nested if condition
label_nested_stmt->cond =
pool.Create(Or{}, variable, label_nested_stmt->cond, &root_stmt);
label_nested_stmt->cond = &pools.stmt.emplace_back(Or{}, variable, label_nested_stmt->cond, &root_stmt);
break;
case StatementType::Loop:
break;
@@ -542,7 +410,7 @@ private:
throw LogicError("Invalid inward movement");
}
Tree& nested_tree{label_nested_stmt->children};
Statement* const new_goto{pool.Create(Goto{}, variable, label, &*label_nested_stmt)};
Statement* const new_goto{&pools.stmt.emplace_back(Goto{}, variable, label, &*label_nested_stmt)};
return nested_tree.insert(nested_tree.begin(), *new_goto);
}
@@ -556,16 +424,16 @@ private:
Tree loop_body;
loop_body.splice(loop_body.begin(), body, label_nested_stmt, goto_stmt);
SanitizeNoBreaks(loop_body);
Statement* const variable{pool.Create(Variable{}, label_id, &root_stmt)};
Statement* const loop_stmt{pool.Create(Loop{}, variable, std::move(loop_body), parent)};
Statement* const variable{&pools.stmt.emplace_back(Variable{}, label_id, &root_stmt)};
Statement* const loop_stmt{&pools.stmt.emplace_back(Loop{}, variable, std::move(loop_body), parent)};
UpdateTreeUp(loop_stmt);
body.insert(goto_stmt, *loop_stmt);
Statement* const new_goto{pool.Create(Goto{}, variable, label, loop_stmt)};
Statement* const new_goto{&pools.stmt.emplace_back(Goto{}, variable, label, loop_stmt)};
loop_stmt->children.push_front(*new_goto);
const Node new_goto_node{loop_stmt->children.begin()};
Statement* const set_var{pool.Create(SetVariable{}, label_id, goto_stmt->cond, loop_stmt)};
Statement* const set_var{&pools.stmt.emplace_back(SetVariable{}, label_id, goto_stmt->cond, loop_stmt)};
loop_stmt->children.push_back(*set_var);
body.erase(goto_stmt);
@@ -577,22 +445,22 @@ private:
Tree& body{parent->children};
const u32 label_id{goto_stmt->label->id};
Statement* const goto_cond{goto_stmt->cond};
Statement* const set_goto_var{pool.Create(SetVariable{}, label_id, goto_cond, &*parent)};
Statement* const set_goto_var{&pools.stmt.emplace_back(SetVariable{}, label_id, goto_cond, &*parent)};
body.insert(goto_stmt, *set_goto_var);
Tree if_body;
if_body.splice(if_body.begin(), body, std::next(goto_stmt), body.end());
if_body.pop_front();
Statement* const cond{pool.Create(Variable{}, label_id, &root_stmt)};
Statement* const neg_cond{pool.Create(Not{}, cond, &root_stmt)};
Statement* const if_stmt{pool.Create(If{}, neg_cond, std::move(if_body), &*parent)};
Statement* const cond{&pools.stmt.emplace_back(Variable{}, label_id, &root_stmt)};
Statement* const neg_cond{&pools.stmt.emplace_back(Not{}, cond, &root_stmt)};
Statement* const if_stmt{&pools.stmt.emplace_back(If{}, neg_cond, std::move(if_body), &*parent)};
UpdateTreeUp(if_stmt);
body.insert(goto_stmt, *if_stmt);
body.erase(goto_stmt);
Statement* const new_cond{pool.Create(Variable{}, label_id, &root_stmt)};
Statement* const new_goto{pool.Create(Goto{}, new_cond, goto_stmt->label, parent->up)};
Statement* const new_cond{&pools.stmt.emplace_back(Variable{}, label_id, &root_stmt)};
Statement* const new_goto{&pools.stmt.emplace_back(Goto{}, new_cond, goto_stmt->label, parent->up)};
Tree& parent_tree{parent->up->children};
return parent_tree.insert(std::next(parent), *new_goto);
}
@@ -602,21 +470,21 @@ private:
Tree& body{parent->children};
const u32 label_id{goto_stmt->label->id};
Statement* const goto_cond{goto_stmt->cond};
Statement* const set_goto_var{pool.Create(SetVariable{}, label_id, goto_cond, parent)};
Statement* const cond{pool.Create(Variable{}, label_id, &root_stmt)};
Statement* const break_stmt{pool.Create(Break{}, cond, parent)};
Statement* const set_goto_var{&pools.stmt.emplace_back(SetVariable{}, label_id, goto_cond, parent)};
Statement* const cond{&pools.stmt.emplace_back(Variable{}, label_id, &root_stmt)};
Statement* const break_stmt{&pools.stmt.emplace_back(Break{}, cond, parent)};
body.insert(goto_stmt, *set_goto_var);
body.insert(goto_stmt, *break_stmt);
body.erase(goto_stmt);
const Node loop{Tree::s_iterator_to(*goto_stmt->up)};
Statement* const new_goto_cond{pool.Create(Variable{}, label_id, &root_stmt)};
Statement* const new_goto{pool.Create(Goto{}, new_goto_cond, goto_stmt->label, loop->up)};
Statement* const new_goto_cond{&pools.stmt.emplace_back(Variable{}, label_id, &root_stmt)};
Statement* const new_goto{&pools.stmt.emplace_back(Goto{}, new_goto_cond, goto_stmt->label, loop->up)};
Tree& parent_tree{loop->up->children};
return parent_tree.insert(std::next(loop), *new_goto);
}
ObjectPool<Statement>& pool;
ShaderPools& pools;
Statement root_stmt{FunctionTag{}};
};
@@ -652,11 +520,11 @@ private:
class TranslatePass {
public:
TranslatePass(ObjectPool<IR::Inst>& inst_pool_, ObjectPool<IR::Block>& block_pool_,
ObjectPool<Statement>& stmt_pool_, Environment& env_, Statement& root_stmt,
IR::AbstractSyntaxList& syntax_list_, const HostTranslateInfo& host_info)
: stmt_pool{stmt_pool_}, inst_pool{inst_pool_}, block_pool{block_pool_}, env{env_},
syntax_list{syntax_list_} {
TranslatePass(ShaderPools& pools_, Environment& env_, Statement& root_stmt, IR::AbstractSyntaxList& syntax_list_, const HostTranslateInfo& host_info)
: pools{pools_}
, env{env_}
, syntax_list{syntax_list_}
{
Visit(root_stmt, nullptr, nullptr);
IR::Block& first_block{*syntax_list.front().data.block};
@@ -674,7 +542,7 @@ private:
if (current_block) {
return;
}
current_block = block_pool.Create(inst_pool);
current_block = &pools.block.emplace_back(pools.inst);
auto& node{syntax_list.emplace_back()};
node.type = IR::AbstractSyntaxNode::Type::Block;
node.data.block = current_block;
@@ -740,7 +608,7 @@ private:
break;
}
case StatementType::Loop: {
IR::Block* const loop_header_block{block_pool.Create(inst_pool)};
IR::Block* const loop_header_block{&pools.block.emplace_back(pools.inst)};
if (current_block) {
current_block->AddBranch(loop_header_block);
}
@@ -748,7 +616,7 @@ private:
header_node.type = IR::AbstractSyntaxNode::Type::Block;
header_node.data.block = loop_header_block;
IR::Block* const continue_block{block_pool.Create(inst_pool)};
IR::Block* const continue_block{&pools.block.emplace_back(pools.inst)};
IR::Block* const merge_block{MergeBlock(parent, stmt)};
const size_t loop_node_index{syntax_list.size()};
@@ -814,7 +682,7 @@ private:
}
case StatementType::Return: {
ensure_block();
IR::Block* return_block{block_pool.Create(inst_pool)};
IR::Block* return_block{&pools.block.emplace_back(pools.inst)};
IR::IREmitter{*return_block}.Epilogue();
current_block->AddBranch(return_block);
@@ -862,10 +730,10 @@ private:
Statement* merge_stmt{TryFindForwardBlock(stmt)};
if (!merge_stmt) {
// Create a merge block we can visit later
merge_stmt = stmt_pool.Create(&dummy_flow_block, &parent);
merge_stmt = &pools.stmt.emplace_back(&dummy_flow_block, &parent);
parent.children.insert(std::next(Tree::s_iterator_to(stmt)), *merge_stmt);
}
return block_pool.Create(inst_pool);
return &pools.block.emplace_back(pools.inst);
}
void DemoteCombinationPass() {
@@ -973,9 +841,7 @@ private:
asl.insert(next_it_2, demote_if_node);
}
ObjectPool<Statement>& stmt_pool;
ObjectPool<IR::Inst>& inst_pool;
ObjectPool<IR::Block>& block_pool;
ShaderPools& pools;
Environment& env;
IR::AbstractSyntaxList& syntax_list;
bool uses_demote_to_helper{};
@@ -983,15 +849,12 @@ private:
};
} // Anonymous namespace
IR::AbstractSyntaxList BuildASL(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Block>& block_pool,
Environment& env, Flow::CFG& cfg,
const HostTranslateInfo& host_info) {
ObjectPool<Statement> stmt_pool{64};
GotoPass goto_pass{cfg, stmt_pool};
IR::AbstractSyntaxList BuildASL(ShaderPools& pools, Environment& env, Flow::CFG& cfg, const HostTranslateInfo& host_info) {
GotoPass goto_pass{cfg, pools};
Statement& root{goto_pass.RootStatement()};
IR::AbstractSyntaxList syntax_list;
TranslatePass{inst_pool, block_pool, stmt_pool, env, root, syntax_list, host_info};
stmt_pool.ReleaseContents();
TranslatePass pass{pools, env, root, syntax_list, host_info};
pools.stmt.clear();
return syntax_list;
}
@@ -1,3 +1,6 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
@@ -8,15 +11,13 @@
#include "shader_recompiler/frontend/ir/basic_block.h"
#include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/frontend/maxwell/control_flow.h"
#include "shader_recompiler/object_pool.h"
namespace Shader {
struct HostTranslateInfo;
namespace Maxwell {
[[nodiscard]] IR::AbstractSyntaxList BuildASL(ObjectPool<IR::Inst>& inst_pool,
ObjectPool<IR::Block>& block_pool, Environment& env,
Flow::CFG& cfg, const HostTranslateInfo& host_info);
struct ShaderPools;
[[nodiscard]] IR::AbstractSyntaxList BuildASL(ShaderPools& pools, Environment& env, Flow::CFG& cfg, const HostTranslateInfo& host_info);
} // namespace Maxwell
} // namespace Shader
@@ -10,6 +10,7 @@
#include <queue>
#include "common/settings.h"
#include "shader_recompiler/shader_pool.h"
#include "shader_recompiler/exception.h"
#include "shader_recompiler/frontend/ir/basic_block.h"
#include "shader_recompiler/frontend/ir/ir_emitter.h"
@@ -234,13 +235,12 @@ void LowerGeometryPassthrough(const IR::Program& program, const HostTranslateInf
} // Anonymous namespace
IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Block>& block_pool,
Environment& env, Flow::CFG& cfg, const HostTranslateInfo& host_info) {
IR::Program TranslateProgram(ShaderPools& pools, Environment& env, Flow::CFG& cfg, const HostTranslateInfo& host_info) {
HostTranslateInfo normalized_host_info{host_info};
normalized_host_info.ApplyDescriptorLimitPolicy();
IR::Program program;
program.syntax_list = BuildASL(inst_pool, block_pool, env, cfg, normalized_host_info);
program.syntax_list = BuildASL(pools, env, cfg, host_info);
program.blocks = GenerateBlocks(program.syntax_list);
program.post_order_blocks = PostOrder(program.syntax_list.front());
program.stage = env.ShaderStage();
@@ -410,11 +410,7 @@ void ConvertLegacyToGeneric(IR::Program& program, const Shader::RuntimeInfo& run
}
}
IR::Program GenerateGeometryPassthrough(ObjectPool<IR::Inst>& inst_pool,
ObjectPool<IR::Block>& block_pool,
const HostTranslateInfo& host_info,
IR::Program& source_program,
Shader::OutputTopology output_topology) {
IR::Program GenerateGeometryPassthrough(ShaderPools& pools, const HostTranslateInfo& host_info, IR::Program& source_program, Shader::OutputTopology output_topology) {
IR::Program program;
program.stage = Stage::Geometry;
program.output_topology = output_topology;
@@ -426,16 +422,15 @@ IR::Program GenerateGeometryPassthrough(ObjectPool<IR::Inst>& inst_pool,
program.info.stores.Set(IR::Attribute::Layer, true);
program.info.stores.Set(source_program.info.emulated_layer, false);
IR::Block* current_block = block_pool.Create(inst_pool);
IR::Block* current_block = &pools.block.emplace_back(pools.inst);
auto& node{program.syntax_list.emplace_back()};
node.type = IR::AbstractSyntaxNode::Type::Block;
node.data.block = current_block;
IR::IREmitter ir{*current_block};
EmitGeometryPassthrough(ir, program, program.info.stores, true,
source_program.info.emulated_layer);
EmitGeometryPassthrough(ir, program, program.info.stores, true, source_program.info.emulated_layer);
IR::Block* return_block{block_pool.Create(inst_pool)};
IR::Block* return_block{&pools.block.emplace_back(pools.inst)};
IR::IREmitter{*return_block}.Epilogue();
current_block->AddBranch(return_block);
@@ -1,3 +1,6 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
@@ -7,7 +10,7 @@
#include "shader_recompiler/frontend/ir/basic_block.h"
#include "shader_recompiler/frontend/ir/program.h"
#include "shader_recompiler/frontend/maxwell/control_flow.h"
#include "shader_recompiler/object_pool.h"
#include "shader_recompiler/frontend/maxwell/structured_control_flow.h"
#include "shader_recompiler/runtime_info.h"
namespace Shader {
@@ -16,22 +19,17 @@ struct HostTranslateInfo;
namespace Shader::Maxwell {
[[nodiscard]] IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool,
ObjectPool<IR::Block>& block_pool, Environment& env,
Flow::CFG& cfg, const HostTranslateInfo& host_info);
struct ShaderPools;
[[nodiscard]] IR::Program MergeDualVertexPrograms(IR::Program& vertex_a, IR::Program& vertex_b,
Environment& env_vertex_b);
[[nodiscard]] IR::Program TranslateProgram(ShaderPools& pools, Environment& env, Flow::CFG& cfg, const HostTranslateInfo& host_info);
[[nodiscard]] IR::Program MergeDualVertexPrograms(IR::Program& vertex_a, IR::Program& vertex_b, Environment& env_vertex_b);
void ConvertLegacyToGeneric(IR::Program& program, const RuntimeInfo& runtime_info);
// Maxwell v1 and older Nvidia cards don't support setting gl_Layer from non-geometry stages.
// This creates a workaround by setting the layer as a generic output and creating a
// passthrough geometry shader that reads the generic and sets the layer.
[[nodiscard]] IR::Program GenerateGeometryPassthrough(ObjectPool<IR::Inst>& inst_pool,
ObjectPool<IR::Block>& block_pool,
const HostTranslateInfo& host_info,
IR::Program& source_program,
Shader::OutputTopology output_topology);
[[nodiscard]] IR::Program GenerateGeometryPassthrough(ShaderPools& pools, const HostTranslateInfo& host_info, IR::Program& source_program, Shader::OutputTopology output_topology);
} // namespace Shader::Maxwell
-106
View File
@@ -1,106 +0,0 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <memory>
#include <type_traits>
#include <utility>
namespace Shader {
template <typename T>
requires std::is_destructible_v<T>
class ObjectPool {
public:
explicit ObjectPool(size_t chunk_size = 8192) : new_chunk_size{chunk_size} {
node = &chunks.emplace_back(new_chunk_size);
}
template <typename... Args>
requires std::is_constructible_v<T, Args...>
[[nodiscard]] T* Create(Args&&... args) {
return std::construct_at(Memory(), std::forward<Args>(args)...);
}
void ReleaseContents() {
if (chunks.empty()) {
return;
}
Chunk& root{chunks.front()};
if (root.used_objects == root.num_objects) {
// Root chunk has been filled, squash allocations into it
const size_t total_objects{root.num_objects + new_chunk_size * (chunks.size() - 1)};
chunks.clear();
chunks.emplace_back(total_objects);
} else {
root.Release();
chunks.resize(1);
}
chunks.shrink_to_fit();
node = &chunks.front();
}
private:
struct NonTrivialDummy {
NonTrivialDummy() noexcept {}
};
union Storage {
Storage() noexcept {}
~Storage() noexcept {}
NonTrivialDummy dummy{};
T object;
};
struct Chunk {
explicit Chunk() = default;
explicit Chunk(size_t size)
: num_objects{size}, storage{std::make_unique<Storage[]>(size)} {}
Chunk& operator=(Chunk&& rhs) noexcept {
Release();
used_objects = std::exchange(rhs.used_objects, 0);
num_objects = std::exchange(rhs.num_objects, 0);
storage = std::move(rhs.storage);
return *this;
}
Chunk(Chunk&& rhs) noexcept
: used_objects{std::exchange(rhs.used_objects, 0)},
num_objects{std::exchange(rhs.num_objects, 0)}, storage{std::move(rhs.storage)} {}
~Chunk() {
Release();
}
void Release() {
std::destroy_n(storage.get(), used_objects);
used_objects = 0;
}
size_t used_objects{};
size_t num_objects{};
std::unique_ptr<Storage[]> storage;
};
[[nodiscard]] T* Memory() {
Chunk* const chunk{FreeChunk()};
return &chunk->storage[chunk->used_objects++].object;
}
[[nodiscard]] Chunk* FreeChunk() {
if (node->used_objects != node->num_objects) {
return node;
}
node = &chunks.emplace_back(new_chunk_size);
return node;
}
Chunk* node{};
std::vector<Chunk> chunks;
size_t new_chunk_size{};
};
} // namespace Shader
+134
View File
@@ -0,0 +1,134 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#pragma once
#include <boost/container/stable_vector.hpp>
#include <boost/intrusive/list.hpp>
#include "shader_recompiler/frontend/maxwell/control_flow.h"
namespace Shader::Maxwell {
struct Statement;
// Use normal_link because we are not guaranteed to destroy the tree in order
using ListBaseHook = boost::intrusive::list_base_hook<boost::intrusive::link_mode<boost::intrusive::normal_link>>;
using Tree = boost::intrusive::list<Statement,
// Allow using Statement without a definition
boost::intrusive::base_hook<ListBaseHook>,
// Avoid linear complexity on splice, size is never called
boost::intrusive::constant_time_size<false>>;
using Node = Tree::iterator;
enum class StatementType {
Code,
Goto,
Label,
If,
Loop,
Break,
Return,
Kill,
Unreachable,
Function,
Identity,
Not,
Or,
SetVariable,
SetIndirectBranchVariable,
Variable,
IndirectBranchCond,
};
[[nodiscard]] inline bool HasChildren(StatementType type) {
switch (type) {
case StatementType::If:
case StatementType::Loop:
case StatementType::Function:
return true;
default:
return false;
}
}
struct Goto {};
struct Label {};
struct If {};
struct Loop {};
struct Break {};
struct Return {};
struct Kill {};
struct Unreachable {};
struct FunctionTag {};
struct Identity {};
struct Not {};
struct Or {};
struct SetVariable {};
struct SetIndirectBranchVariable {};
struct Variable {};
struct IndirectBranchCond {};
#ifdef _MSC_VER
#pragma warning(push)
#pragma warning(disable : 26495) // Always initialize a member variable, expected in Statement
#endif
struct Statement : ListBaseHook {
Statement(const Flow::Block* block_, Statement* up_) : block{block_}, up{up_}, type{StatementType::Code} {}
Statement(Goto, Statement* cond_, Node label_, Statement* up_) : label{label_}, cond{cond_}, up{up_}, type{StatementType::Goto} {}
Statement(Label, u32 id_, Statement* up_) : id{id_}, up{up_}, type{StatementType::Label} {}
Statement(If, Statement* cond_, Tree&& children_, Statement* up_) : children{std::move(children_)}, cond{cond_}, up{up_}, type{StatementType::If} {}
Statement(Loop, Statement* cond_, Tree&& children_, Statement* up_) : children{std::move(children_)}, cond{cond_}, up{up_}, type{StatementType::Loop} {}
Statement(Break, Statement* cond_, Statement* up_) : cond{cond_}, up{up_}, type{StatementType::Break} {}
Statement(Return, Statement* up_) : up{up_}, type{StatementType::Return} {}
Statement(Kill, Statement* up_) : up{up_}, type{StatementType::Kill} {}
Statement(Unreachable, Statement* up_) : up{up_}, type{StatementType::Unreachable} {}
Statement(FunctionTag) : children{}, type{StatementType::Function} {}
Statement(Identity, IR::Condition cond_, Statement* up_) : guest_cond{cond_}, up{up_}, type{StatementType::Identity} {}
Statement(Not, Statement* op_, Statement* up_) : op{op_}, up{up_}, type{StatementType::Not} {}
Statement(Or, Statement* op_a_, Statement* op_b_, Statement* up_) : op_a{op_a_}, op_b{op_b_}, up{up_}, type{StatementType::Or} {}
Statement(SetVariable, u32 id_, Statement* op_, Statement* up_) : op{op_}, id{id_}, up{up_}, type{StatementType::SetVariable} {}
Statement(SetIndirectBranchVariable, IR::Reg branch_reg_, s32 branch_offset_, Statement* up_) : branch_offset{branch_offset_}, branch_reg{branch_reg_}, up{up_}, type{StatementType::SetIndirectBranchVariable} {}
Statement(Variable, u32 id_, Statement* up_) : id{id_}, up{up_}, type{StatementType::Variable} {}
Statement(IndirectBranchCond, u32 location_, Statement* up_) : location{location_}, up{up_}, type{StatementType::IndirectBranchCond} {}
~Statement() {
if (HasChildren(type)) {
std::destroy_at(&children);
}
}
union {
const Flow::Block* block;
Node label;
Tree children;
IR::Condition guest_cond;
Statement* op;
Statement* op_a;
u32 location;
s32 branch_offset;
};
union {
Statement* cond;
Statement* op_b;
u32 id;
IR::Reg branch_reg;
};
Statement* up{};
StatementType type;
};
#ifdef _MSC_VER
#pragma warning(pop)
#endif
struct ShaderPools {
void clear() {
flow_block.clear();
block.clear();
inst.clear();
stmt.clear();
}
boost::container::stable_vector<Shader::IR::Inst> inst{};
boost::container::stable_vector<Shader::IR::Block> block{};
boost::container::stable_vector<Shader::Maxwell::Flow::Block> flow_block{};
boost::container::stable_vector<Shader::Maxwell::Statement> stmt;
};
}
+83 -79
View File
@@ -417,14 +417,6 @@ void HLE_TransformFeedbackSetup::Execute(Core::System& system, Engines::Maxwell3
default: return std::monostate{};
}
}
[[nodiscard]] inline bool CanBeHLEProgram(u64 hash) noexcept {
switch (hash) {
#define HLE_MACRO_ELEM(HASH, TY, VAL) case HASH: return true;
HLE_MACRO_LIST
#undef HLE_MACRO_ELEM
default: return false;
}
}
void MacroInterpreterImpl::Execute(Core::System& system, Engines::Maxwell3D& maxwell3d, std::span<const u32> params, u32 method) {
Reset();
@@ -1345,80 +1337,92 @@ static void Dump(u64 hash, std::span<const u32> code, bool decompiled = false) {
macro_file.write(reinterpret_cast<const char*>(code.data()), code.size_bytes());
}
void MacroEngine::Execute(Core::System& system, Engines::Maxwell3D& maxwell3d, u32 method, std::span<const u32> parameters) {
auto const execute_variant = [&system, &maxwell3d, &parameters, method](AnyCachedMacro& acm) {
if (auto a = std::get_if<HLE_DrawArraysIndirect>(&acm))
return a->Execute(system, maxwell3d, parameters, method);
if (auto a = std::get_if<HLE_DrawIndexedIndirect>(&acm))
return a->Execute(system, maxwell3d, parameters, method);
if (auto a = std::get_if<HLE_MultiDrawIndexedIndirectCount>(&acm))
return a->Execute(system, maxwell3d, parameters, method);
if (auto a = std::get_if<HLE_MultiLayerClear>(&acm))
return a->Execute(system, maxwell3d, parameters, method);
if (auto a = std::get_if<HLE_C713C83D8F63CCF3>(&acm))
return a->Execute(system, maxwell3d, parameters, method);
if (auto a = std::get_if<HLE_D7333D26E0A93EDE>(&acm))
return a->Execute(system, maxwell3d, parameters, method);
if (auto a = std::get_if<HLE_BindShader>(&acm))
return a->Execute(system, maxwell3d, parameters, method);
if (auto a = std::get_if<HLE_SetRasterBoundingBox>(&acm))
return a->Execute(system, maxwell3d, parameters, method);
if (auto a = std::get_if<HLE_ClearConstBuffer>(&acm))
return a->Execute(system, maxwell3d, parameters, method);
if (auto a = std::get_if<HLE_ClearMemory>(&acm))
return a->Execute(system, maxwell3d, parameters, method);
if (auto a = std::get_if<HLE_TransformFeedbackSetup>(&acm))
return a->Execute(system, maxwell3d, parameters, method);
if (auto a = std::get_if<HLE_DrawIndirectByteCount>(&acm))
return a->Execute(system, maxwell3d, parameters, method);
if (auto a = std::get_if<MacroInterpreterImpl>(&acm))
return a->Execute(system, maxwell3d, parameters, method);
if (auto a = std::get_if<std::unique_ptr<DynamicCachedMacro>>(&acm))
return a->get()->Execute(system, maxwell3d, parameters, method);
};
if (auto const it = macro_cache.find(method); it != macro_cache.end()) {
auto& ci = it->second;
if (!CanBeHLEProgram(ci.hash) || Settings::values.disable_macro_hle)
maxwell3d.RefreshParameters(); //LLE must reload parameters
execute_variant(ci.program);
} else {
// Macro not compiled, check if it's uploaded and if so, compile it
std::optional<u32> mid_method;
const auto macro_code = uploaded_macro_code.find(method);
if (macro_code == uploaded_macro_code.end()) {
for (const auto& [method_base, code] : uploaded_macro_code) {
if (method >= method_base && (method - method_base) < code.size()) {
mid_method = method_base;
break;
}
}
if (!mid_method.has_value()) {
ASSERT_MSG(false, "Macro 0x{0:x} was not uploaded", method);
return;
}
}
auto& ci = macro_cache[method];
if (mid_method) {
const auto& macro_cached = uploaded_macro_code[mid_method.value()];
const auto rebased_method = method - mid_method.value();
auto& code = uploaded_macro_code[method];
code.resize(macro_cached.size() - rebased_method);
std::memcpy(code.data(), macro_cached.data() + rebased_method, code.size() * sizeof(u32));
ci.hash = Common::HashValue(code);
ci.program = Compile(system, maxwell3d, code);
} else {
ci.program = Compile(system, maxwell3d, macro_code->second);
ci.hash = Common::HashValue(macro_code->second);
}
if (CanBeHLEProgram(ci.hash) && !Settings::values.disable_macro_hle) {
ci.program = GetHLEProgram(ci.hash);
} else {
void MacroEngine::Execute(Core::System& system, Engines::Maxwell3D& maxwell3d, u32 method,
std::span<const u32> parameters) {
const auto execute_variant = [&system, &maxwell3d, &parameters,
method](AnyCachedMacro& cached) {
if (std::holds_alternative<MacroInterpreterImpl>(cached) ||
std::holds_alternative<std::unique_ptr<DynamicCachedMacro>>(cached) ||
Settings::values.disable_macro_hle) {
maxwell3d.RefreshParameters();
}
execute_variant(ci.program);
if (Settings::values.dump_macros) {
Dump(ci.hash, macro_code->second, !std::holds_alternative<std::monostate>(ci.program));
if (auto program = std::get_if<HLE_DrawArraysIndirect>(&cached))
return program->Execute(system, maxwell3d, parameters, method);
if (auto program = std::get_if<HLE_DrawIndexedIndirect>(&cached))
return program->Execute(system, maxwell3d, parameters, method);
if (auto program = std::get_if<HLE_MultiDrawIndexedIndirectCount>(&cached))
return program->Execute(system, maxwell3d, parameters, method);
if (auto program = std::get_if<HLE_MultiLayerClear>(&cached))
return program->Execute(system, maxwell3d, parameters, method);
if (auto program = std::get_if<HLE_C713C83D8F63CCF3>(&cached))
return program->Execute(system, maxwell3d, parameters, method);
if (auto program = std::get_if<HLE_D7333D26E0A93EDE>(&cached))
return program->Execute(system, maxwell3d, parameters, method);
if (auto program = std::get_if<HLE_BindShader>(&cached))
return program->Execute(system, maxwell3d, parameters, method);
if (auto program = std::get_if<HLE_SetRasterBoundingBox>(&cached))
return program->Execute(system, maxwell3d, parameters, method);
if (auto program = std::get_if<HLE_ClearConstBuffer>(&cached))
return program->Execute(system, maxwell3d, parameters, method);
if (auto program = std::get_if<HLE_ClearMemory>(&cached))
return program->Execute(system, maxwell3d, parameters, method);
if (auto program = std::get_if<HLE_TransformFeedbackSetup>(&cached))
return program->Execute(system, maxwell3d, parameters, method);
if (auto program = std::get_if<HLE_DrawIndirectByteCount>(&cached))
return program->Execute(system, maxwell3d, parameters, method);
if (auto program = std::get_if<MacroInterpreterImpl>(&cached))
return program->Execute(system, maxwell3d, parameters, method);
if (auto program = std::get_if<std::unique_ptr<DynamicCachedMacro>>(&cached))
return program->get()->Execute(system, maxwell3d, parameters, method);
UNREACHABLE();
};
if (auto const it = macro_cache.find(method); it != macro_cache.end()) {
execute_variant(it->second.program);
return;
}
// Macro not compiled, check if it's uploaded and if so, compile it
std::span<const u32> code;
auto macro_code = uploaded_macro_code.find(method);
if (macro_code == uploaded_macro_code.end()) {
std::optional<u32> mid_method;
for (const auto& [method_base, uploaded_code] : uploaded_macro_code) {
if (method >= method_base && (method - method_base) < uploaded_code.size()) {
mid_method = method_base;
break;
}
}
if (!mid_method) {
ASSERT_MSG(false, "Macro 0x{0:x} was not uploaded", method);
return;
}
const auto source = uploaded_macro_code.find(*mid_method);
ASSERT(source != uploaded_macro_code.end());
const auto rebased_method = method - *mid_method;
std::vector<u32> rebased_code(source->second.begin() + rebased_method,
source->second.end());
const auto [it, inserted] = uploaded_macro_code.emplace(method, std::move(rebased_code));
ASSERT(inserted);
code = it->second;
} else {
code = macro_code->second;
}
auto& ci = macro_cache[method];
ci.hash = Common::HashRange(code.begin(), code.end());
if (!Settings::values.disable_macro_hle) {
ci.program = GetHLEProgram(ci.hash);
}
if (std::holds_alternative<std::monostate>(ci.program)) {
ci.program = Compile(system, maxwell3d, code);
}
execute_variant(ci.program);
if (Settings::values.dump_macros) {
Dump(ci.hash, code, !std::holds_alternative<std::monostate>(ci.program));
}
}
@@ -43,7 +43,6 @@ using Shader::Backend::GLASM::EmitGLASM;
using Shader::Backend::GLSL::EmitGLSL;
using Shader::Backend::SPIRV::EmitSPIRV;
using Shader::Maxwell::ConvertLegacyToGeneric;
using Shader::Maxwell::GenerateGeometryPassthrough;
using Shader::Maxwell::MergeDualVertexPrograms;
using Shader::Maxwell::TranslateProgram;
using VideoCommon::ComputeEnvironment;
@@ -316,7 +315,7 @@ void ShaderCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading,
ComputePipelineKey key;
file.read(reinterpret_cast<char*>(&key), sizeof(key));
queue_work([this, key, env_ = std::move(env), &state, &callback](Context* ctx) mutable {
ctx->pools.ReleaseContents();
ctx->pools.clear();
auto pipeline{CreateComputePipeline(ctx->pools, key, env_, true)};
std::scoped_lock lock{state.mutex};
if (pipeline) {
@@ -337,7 +336,7 @@ void ShaderCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading,
for (auto& env : envs_) {
env_ptrs.push_back(&env);
}
ctx->pools.ReleaseContents();
ctx->pools.clear();
auto pipeline{CreateGraphicsPipeline(ctx->pools, key, MakeSpan(env_ptrs), false, true)};
std::scoped_lock lock{state.mutex};
if (pipeline) {
@@ -447,9 +446,8 @@ std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline() {
GraphicsEnvironments environments;
GetGraphicsEnvironments(environments, graphics_key.unique_hashes);
main_pools.ReleaseContents();
auto pipeline{CreateGraphicsPipeline(main_pools, graphics_key, environments.Span(),
use_asynchronous_shaders)};
main_pools.clear();
auto pipeline{CreateGraphicsPipeline(main_pools, graphics_key, environments.Span(), use_asynchronous_shaders)};
if (!pipeline || shader_cache_filename.empty()) {
return pipeline;
}
@@ -463,10 +461,7 @@ std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline() {
return pipeline;
}
std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline(
ShaderContext::ShaderPools& pools, const GraphicsPipelineKey& key,
std::span<Shader::Environment* const> envs, bool use_shader_workers,
bool force_context_flush) try {
std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline(Shader::Maxwell::ShaderPools& pools, const GraphicsPipelineKey& key, std::span<Shader::Environment* const> envs, bool use_shader_workers, bool force_context_flush) try {
auto hash = key.Hash();
LOG_INFO(Render_OpenGL, "{:#016x}", hash);
size_t env_index{};
@@ -479,12 +474,10 @@ std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline(
Shader::IR::Program* layer_source_program{};
for (size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
const bool is_emulated_stage = layer_source_program != nullptr
&& index == u32(Maxwell::ShaderType::Geometry);
const bool is_emulated_stage = layer_source_program != nullptr && index == u32(Maxwell::ShaderType::Geometry);
if (key.unique_hashes[index] == 0 && is_emulated_stage) {
auto topology = MaxwellToOutputTopology(key.gs_input_topology);
programs[index] = GenerateGeometryPassthrough(pools.inst, pools.block, host_info,
*layer_source_program, topology);
programs[index] = Shader::Maxwell::GenerateGeometryPassthrough(pools, host_info, *layer_source_program, topology);
continue;
}
if (key.unique_hashes[index] == 0) {
@@ -502,13 +495,13 @@ std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline(
if (!uses_vertex_a || index != 1) {
// Normal path
programs[index] = TranslateProgram(pools.inst, pools.block, env, cfg, host_info);
programs[index] = TranslateProgram(pools, env, cfg, host_info);
total_storage_buffers += Shader::NumDescriptors(programs[index].info.storage_buffers_descriptors);
} else {
// VertexB path when VertexA is present.
auto& program_va{programs[0]};
auto program_vb{TranslateProgram(pools.inst, pools.block, env, cfg, host_info)};
auto program_vb{TranslateProgram(pools, env, cfg, host_info)};
total_storage_buffers += Shader::NumDescriptors(program_vb.info.storage_buffers_descriptors);
programs[index] = MergeDualVertexPrograms(program_va, program_vb, env);
}
@@ -575,7 +568,7 @@ std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline(
ComputeEnvironment env{*kepler_compute, *gpu_memory, program_base, qmd.program_start};
env.SetCachedSize(shader->size_bytes);
main_pools.ReleaseContents();
main_pools.clear();
auto pipeline{CreateComputePipeline(main_pools, key, env)};
if (!pipeline || shader_cache_filename.empty()) {
return pipeline;
@@ -585,9 +578,7 @@ std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline(
return pipeline;
}
std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline(
ShaderContext::ShaderPools& pools, const ComputePipelineKey& key, Shader::Environment& env,
bool force_context_flush) try {
std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline(Shader::Maxwell::ShaderPools& pools, const ComputePipelineKey& key, Shader::Environment& env, bool force_context_flush) try {
auto hash = key.Hash();
LOG_INFO(Render_OpenGL, "{:#016x}", hash);
@@ -597,7 +588,7 @@ std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline(
env.Dump(hash, key.unique_hash);
}
auto program{TranslateProgram(pools.inst, pools.block, env, cfg, host_info)};
auto program{TranslateProgram(pools, env, cfg, host_info)};
const u32 num_storage_buffers{Shader::NumDescriptors(program.info.storage_buffers_descriptors)};
Shader::RuntimeInfo info;
info.glasm_use_storage_buffers = num_storage_buffers <= device.GetMaxGLASMStorageBufferBlocks();
@@ -12,6 +12,7 @@
#include "common/common_types.h"
#include "common/thread_worker.h"
#include "shader_recompiler/host_translate_info.h"
#include "shader_recompiler/shader_pool.h"
#include "shader_recompiler/profile.h"
#include "video_core/renderer_opengl/gl_compute_pipeline.h"
#include "video_core/renderer_opengl/gl_graphics_pipeline.h"
@@ -51,20 +52,9 @@ private:
[[nodiscard]] GraphicsPipeline* BuiltPipeline(GraphicsPipeline* pipeline) const noexcept;
std::unique_ptr<GraphicsPipeline> CreateGraphicsPipeline();
std::unique_ptr<GraphicsPipeline> CreateGraphicsPipeline(
ShaderContext::ShaderPools& pools, const GraphicsPipelineKey& key,
std::span<Shader::Environment* const> envs, bool use_shader_workers,
bool force_context_flush = false);
std::unique_ptr<ComputePipeline> CreateComputePipeline(const ComputePipelineKey& key,
const VideoCommon::ShaderInfo* shader);
std::unique_ptr<ComputePipeline> CreateComputePipeline(ShaderContext::ShaderPools& pools,
const ComputePipelineKey& key,
Shader::Environment& env,
bool force_context_flush = false);
std::unique_ptr<GraphicsPipeline> CreateGraphicsPipeline(Shader::Maxwell::ShaderPools& pools, const GraphicsPipelineKey& key, std::span<Shader::Environment* const> envs, bool use_shader_workers, bool force_context_flush = false);
std::unique_ptr<ComputePipeline> CreateComputePipeline(const ComputePipelineKey& key, const VideoCommon::ShaderInfo* shader);
std::unique_ptr<ComputePipeline> CreateComputePipeline(Shader::Maxwell::ShaderPools& pools, const ComputePipelineKey& key, Shader::Environment& env, bool force_context_flush = false);
std::unique_ptr<ShaderWorker> CreateWorkers() const;
Core::Frontend::EmuWindow& emu_window;
@@ -80,7 +70,7 @@ private:
GraphicsPipelineKey graphics_key{};
GraphicsPipeline* current_pipeline{};
ShaderContext::ShaderPools main_pools;
Shader::Maxwell::ShaderPools main_pools;
ankerl::unordered_dense::map<GraphicsPipelineKey, std::unique_ptr<GraphicsPipeline>> graphics_cache;
ankerl::unordered_dense::map<ComputePipelineKey, std::unique_ptr<ComputePipeline>> compute_cache;
@@ -1,33 +1,26 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <boost/container/stable_vector.hpp>
#include "core/frontend/emu_window.h"
#include "core/frontend/graphics_context.h"
#include "shader_recompiler/frontend/ir/basic_block.h"
#include "shader_recompiler/frontend/maxwell/control_flow.h"
#include "shader_recompiler/frontend/maxwell/structured_control_flow.h"
#include "shader_recompiler/frontend/maxwell/translate_program.h"
namespace OpenGL::ShaderContext {
struct ShaderPools {
void ReleaseContents() {
flow_block.ReleaseContents();
block.ReleaseContents();
inst.ReleaseContents();
}
Shader::ObjectPool<Shader::IR::Inst> inst{8192};
Shader::ObjectPool<Shader::IR::Block> block{32};
Shader::ObjectPool<Shader::Maxwell::Flow::Block> flow_block{32};
};
struct Context {
explicit Context(Core::Frontend::EmuWindow& emu_window)
: gl_context{emu_window.CreateSharedContext()}, scoped{*gl_context} {}
explicit Context(Core::Frontend::EmuWindow& emu_window) : gl_context{emu_window.CreateSharedContext()}, scoped{*gl_context} {}
std::unique_ptr<Core::Frontend::GraphicsContext> gl_context;
Core::Frontend::GraphicsContext::Scoped scoped;
ShaderPools pools;
Shader::Maxwell::ShaderPools pools;
};
} // namespace OpenGL::ShaderContext
@@ -52,15 +52,6 @@
namespace Vulkan {
namespace {
using Shader::Backend::SPIRV::EmitSPIRV;
using Shader::Maxwell::ConvertLegacyToGeneric;
using Shader::Maxwell::GenerateGeometryPassthrough;
using Shader::Maxwell::MergeDualVertexPrograms;
using Shader::Maxwell::TranslateProgram;
using VideoCommon::ComputeEnvironment;
using VideoCommon::FileEnvironment;
using VideoCommon::GenericEnvironment;
using VideoCommon::GraphicsEnvironment;
constexpr u32 CACHE_VERSION = 18;
constexpr std::array<char, 8> VULKAN_CACHE_MAGIC_NUMBER{'y', 'u', 'z', 'u', 'v', 'k', 'c', 'h'};
@@ -611,12 +602,12 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading
if (device.IsKhrPipelineExecutablePropertiesEnabled()) {
state.statistics = std::make_unique<PipelineStatistics>(device);
}
const auto load_compute{[&](std::ifstream& file, FileEnvironment env) {
const auto load_compute{[&](std::ifstream& file, VideoCommon::FileEnvironment env) {
ComputePipelineCacheKey key;
file.read(reinterpret_cast<char*>(&key), sizeof(key));
workers.QueueWork([this, key, env_ = std::move(env), &state, &callback]() mutable {
ShaderPools pools;
Shader::Maxwell::ShaderPools pools;
auto pipeline{CreateComputePipeline(pools, key, env_, state.statistics.get(), false)};
std::scoped_lock lock{state.mutex};
if (pipeline) {
@@ -629,7 +620,7 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading
});
++state.total;
}};
const auto load_graphics{[&](std::ifstream& file, std::vector<FileEnvironment> envs) {
const auto load_graphics{[&](std::ifstream& file, std::vector<VideoCommon::FileEnvironment> envs) {
GraphicsPipelineCacheKey key;
file.read(reinterpret_cast<char*>(&key), sizeof(key));
@@ -662,7 +653,7 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading
}
workers.QueueWork([this, key, envs_ = std::move(envs), &state, &callback]() mutable {
ShaderPools pools;
Shader::Maxwell::ShaderPools pools;
boost::container::static_vector<Shader::Environment*, 5> env_ptrs;
for (auto& env : envs_) {
env_ptrs.push_back(&env);
@@ -736,10 +727,7 @@ GraphicsPipeline* PipelineCache::BuiltPipeline(GraphicsPipeline* pipeline) const
return nullptr;
}
std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
ShaderPools& pools, const GraphicsPipelineCacheKey& key,
std::span<Shader::Environment* const> envs, PipelineStatistics* statistics,
bool build_in_parallel) try {
std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(Shader::Maxwell::ShaderPools& pools, const GraphicsPipelineCacheKey& key, std::span<Shader::Environment* const> envs, PipelineStatistics* statistics, bool build_in_parallel) try {
auto hash = key.Hash();
LOG_INFO(Render_Vulkan, "{:#016x}", hash);
size_t env_index{0};
@@ -751,12 +739,10 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
Shader::IR::Program* layer_source_program{};
for (size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
const bool is_emulated_stage = layer_source_program != nullptr &&
index == static_cast<u32>(Maxwell::ShaderType::Geometry);
const bool is_emulated_stage = layer_source_program != nullptr && index == u32(Maxwell::ShaderType::Geometry);
if (key.unique_hashes[index] == 0 && is_emulated_stage) {
auto topology = MaxwellToOutputTopology(key.state.topology);
programs[index] = GenerateGeometryPassthrough(pools.inst, pools.block, host_info,
*layer_source_program, topology);
programs[index] = Shader::Maxwell::GenerateGeometryPassthrough(pools, host_info, *layer_source_program, topology);
continue;
}
if (key.unique_hashes[index] == 0) {
@@ -765,16 +751,16 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
Shader::Environment& env{*envs[env_index]};
++env_index;
const u32 cfg_offset{static_cast<u32>(env.StartAddress() + sizeof(Shader::ProgramHeader))};
const u32 cfg_offset{u32(env.StartAddress() + sizeof(Shader::ProgramHeader))};
Shader::Maxwell::Flow::CFG cfg(env, pools.flow_block, cfg_offset, index == 0);
if (!uses_vertex_a || index != 1) {
// Normal path
programs[index] = TranslateProgram(pools.inst, pools.block, env, cfg, host_info);
programs[index] = Shader::Maxwell::TranslateProgram(pools, env, cfg, host_info);
} else {
// VertexB path when VertexA is present.
auto& program_va{programs[0]};
auto program_vb{TranslateProgram(pools.inst, pools.block, env, cfg, host_info)};
programs[index] = MergeDualVertexPrograms(program_va, program_vb, env);
auto program_vb{Shader::Maxwell::TranslateProgram(pools, env, cfg, host_info)};
programs[index] = Shader::Maxwell::MergeDualVertexPrograms(program_va, program_vb, env);
}
if (Settings::values.dump_guest_shaders) {
@@ -804,8 +790,8 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
infos[stage_index] = &program.info;
const auto runtime_info{MakeRuntimeInfo(programs, key, program, previous_stage, device)};
ConvertLegacyToGeneric(program, runtime_info);
const std::vector<u32> code{EmitSPIRV(profile, runtime_info, program, binding)};
Shader::Maxwell::ConvertLegacyToGeneric(program, runtime_info);
const std::vector<u32> code{Shader::Backend::SPIRV::EmitSPIRV(profile, runtime_info, program, binding)};
device.SaveShader(code);
modules[stage_index] = BuildShader(device, code);
@@ -861,14 +847,14 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline() {
GraphicsEnvironments environments;
GetGraphicsEnvironments(environments, graphics_key.unique_hashes);
main_pools.ReleaseContents();
main_pools.clear();
auto pipeline{
CreateGraphicsPipeline(main_pools, graphics_key, environments.Span(), nullptr, true)};
if (!pipeline || pipeline_cache_filename.empty()) {
return pipeline;
}
serialization_thread.QueueWork([this, key = graphics_key, envs = std::move(environments.envs)] {
boost::container::static_vector<const GenericEnvironment*, Maxwell::MaxShaderProgram>
boost::container::static_vector<const VideoCommon::GenericEnvironment*, Maxwell::MaxShaderProgram>
env_ptrs;
for (size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
if (key.unique_hashes[index] != 0) {
@@ -884,24 +870,22 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
const ComputePipelineCacheKey& key, const ShaderInfo* shader) {
const GPUVAddr program_base{kepler_compute->regs.code_loc.Address()};
const auto& qmd{kepler_compute->launch_description};
ComputeEnvironment env{*kepler_compute, *gpu_memory, program_base, qmd.program_start};
VideoCommon::ComputeEnvironment env{*kepler_compute, *gpu_memory, program_base, qmd.program_start};
env.SetCachedSize(shader->size_bytes);
main_pools.ReleaseContents();
main_pools.clear();
auto pipeline{CreateComputePipeline(main_pools, key, env, nullptr, true)};
if (!pipeline || pipeline_cache_filename.empty()) {
return pipeline;
}
serialization_thread.QueueWork([this, key, env_ = std::move(env)] {
SerializePipeline(key, std::array<const GenericEnvironment*, 1>{&env_},
SerializePipeline(key, std::array<const VideoCommon::GenericEnvironment*, 1>{&env_},
pipeline_cache_filename, CACHE_VERSION);
});
return pipeline;
}
std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
ShaderPools& pools, const ComputePipelineCacheKey& key, Shader::Environment& env,
PipelineStatistics* statistics, bool build_in_parallel) try {
std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(Shader::Maxwell::ShaderPools& pools, const ComputePipelineCacheKey& key, Shader::Environment& env, PipelineStatistics* statistics, bool build_in_parallel) try {
auto hash = key.Hash();
if (device.HasBrokenCompute()) {
LOG_ERROR(Render_Vulkan, "Skipping {:#016x}", hash);
@@ -917,11 +901,9 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
env.Dump(hash, key.unique_hash);
}
auto program{TranslateProgram(pools.inst, pools.block, env, cfg, host_info)};
auto program{Shader::Maxwell::TranslateProgram(pools, env, cfg, host_info)};
const VkDriverIdKHR driver_id = device.GetDriverID();
const bool needs_shared_mem_clamp =
driver_id == VK_DRIVER_ID_QUALCOMM_PROPRIETARY ||
driver_id == VK_DRIVER_ID_ARM_PROPRIETARY;
const bool needs_shared_mem_clamp = driver_id == VK_DRIVER_ID_QUALCOMM_PROPRIETARY || driver_id == VK_DRIVER_ID_ARM_PROPRIETARY;
const u32 max_shared_memory = device.GetMaxComputeSharedMemorySize();
if (needs_shared_mem_clamp && program.shared_memory_size > max_shared_memory) {
LOG_WARNING(Render_Vulkan,
@@ -931,7 +913,7 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
max_shared_memory / 1024);
program.shared_memory_size = max_shared_memory;
}
const std::vector<u32> code{EmitSPIRV(profile, program)};
const std::vector<u32> code{Shader::Backend::SPIRV::EmitSPIRV(profile, program)};
device.SaveShader(code);
vk::ShaderModule spv_module{BuildShader(device, code)};
@@ -11,17 +11,20 @@
#include <filesystem>
#include <memory>
#include <type_traits>
#include <ankerl/unordered_dense.h>
#include <vector>
#include <ankerl/unordered_dense.h>
#include <boost/container/stable_vector.hpp>
#include "common/common_types.h"
#include "common/thread_worker.h"
#include "shader_recompiler/frontend/ir/basic_block.h"
#include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/frontend/maxwell/control_flow.h"
#include "shader_recompiler/frontend/maxwell/structured_control_flow.h"
#include "shader_recompiler/frontend/maxwell/translate_program.h"
#include "shader_recompiler/host_translate_info.h"
#include "shader_recompiler/object_pool.h"
#include "shader_recompiler/profile.h"
#include "shader_recompiler/shader_pool.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/host1x/gpu_device_memory_manager.h"
#include "video_core/renderer_vulkan/fixed_pipeline_state.h"
@@ -88,18 +91,6 @@ class Scheduler;
using VideoCommon::ShaderInfo;
struct ShaderPools {
void ReleaseContents() {
flow_block.ReleaseContents();
block.ReleaseContents();
inst.ReleaseContents();
}
Shader::ObjectPool<Shader::IR::Inst> inst{8192};
Shader::ObjectPool<Shader::IR::Block> block{32};
Shader::ObjectPool<Shader::Maxwell::Flow::Block> flow_block{32};
};
class PipelineCache : public VideoCommon::ShaderCache {
public:
explicit PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_, const Device& device,
@@ -124,14 +115,14 @@ private:
std::unique_ptr<GraphicsPipeline> CreateGraphicsPipeline();
std::unique_ptr<GraphicsPipeline> CreateGraphicsPipeline(
ShaderPools& pools, const GraphicsPipelineCacheKey& key,
Shader::Maxwell::ShaderPools& pools, const GraphicsPipelineCacheKey& key,
std::span<Shader::Environment* const> envs, PipelineStatistics* statistics,
bool build_in_parallel);
std::unique_ptr<ComputePipeline> CreateComputePipeline(const ComputePipelineCacheKey& key,
const ShaderInfo* shader);
std::unique_ptr<ComputePipeline> CreateComputePipeline(ShaderPools& pools,
std::unique_ptr<ComputePipeline> CreateComputePipeline(Shader::Maxwell::ShaderPools& pools,
const ComputePipelineCacheKey& key,
Shader::Environment& env,
PipelineStatistics* statistics,
@@ -160,7 +151,7 @@ private:
ankerl::unordered_dense::map<ComputePipelineCacheKey, std::unique_ptr<ComputePipeline>> compute_cache;
ankerl::unordered_dense::map<GraphicsPipelineCacheKey, std::unique_ptr<GraphicsPipeline>> graphics_cache;
ShaderPools main_pools;
Shader::Maxwell::ShaderPools main_pools;
Shader::Profile profile;
Shader::HostTranslateInfo host_info;
+1 -2
View File
@@ -11,7 +11,6 @@
#include "common/assert.h"
#include "shader_recompiler/frontend/maxwell/control_flow.h"
#include "shader_recompiler/object_pool.h"
#include "video_core/control/channel_state.h"
#include "video_core/dirty_flags.h"
#include "video_core/engines/kepler_compute.h"
@@ -236,7 +235,7 @@ const ShaderInfo* ShaderCache::MakeShaderInfo(GenericEnvironment& env, VAddr cpu
} else {
// Slow path, not really hit on commercial games
// Build a control flow graph to get the real shader size
Shader::ObjectPool<Shader::Maxwell::Flow::Block> flow_block;
boost::container::stable_vector<Shader::Maxwell::Flow::Block> flow_block;
Shader::Maxwell::Flow::CFG cfg{env, flow_block, env.StartAddress()};
info->unique_hash = env.CalculateHash();
info->size_bytes = env.ReadSizeBytes();