hdkv/yuzu
1
Fork 0
Code Issues Pull Requests Releases 1 Activity

early-access version 1870

Browse Source
This commit is contained in:
pineappleEA
2021-07-12 20:42:10 +02:00
parent c28ecd0f43
commit f8a45c1e72
36 changed files with 362 additions and 150 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/shader_recompiler/backend/glasm/emit_context.h
View File

@@ -59,7 +59,7 @@ public:
}
std::string code;
RegAlloc reg_alloc{*this};
RegAlloc reg_alloc{};
const Info& info;
const Profile& profile;
const RuntimeInfo& runtime_info;

7
src/shader_recompiler/backend/glasm/emit_glasm.cpp
View File

@@ -2,7 +2,7 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <ranges>
#include <algorithm>
#include <string>
#include <tuple>
@@ -196,7 +196,10 @@ void PrecolorInst(IR::Inst& phi) {
void Precolor(const IR::Program& program) {
for (IR::Block* const block : program.blocks) {
for (IR::Inst& phi : block->Instructions() | std::views::take_while(IR::IsPhi)) {
for (IR::Inst& phi : block->Instructions()) {
if (!IR::IsPhi(phi)) {
break;
}
PrecolorInst(phi);
}
}

3
src/shader_recompiler/backend/glasm/reg_alloc.h
View File

@@ -86,7 +86,7 @@ struct ScalarF64 : Value {};
class RegAlloc {
public:
RegAlloc(EmitContext& ctx_) : ctx{ctx_} {}
RegAlloc() = default;
Register Define(IR::Inst& inst);
@@ -142,7 +142,6 @@ private:
void Free(Id id);
EmitContext& ctx;
size_t num_used_registers{};
size_t num_used_long_registers{};
std::bitset<NUM_REGS> register_use{};

20
src/shader_recompiler/backend/glsl/emit_glsl.cpp
View File

@@ -2,8 +2,10 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <ranges>
#include <algorithm>
#include <string>
#include <tuple>
#include <type_traits>
#include "common/div_ceil.h"
#include "common/settings.h"
@@ -120,7 +122,10 @@ void PrecolorInst(IR::Inst& phi) {
void Precolor(const IR::Program& program) {
for (IR::Block* const block : program.blocks) {
for (IR::Inst& phi : block->Instructions() | std::views::take_while(IR::IsPhi)) {
for (IR::Inst& phi : block->Instructions()) {
if (!IR::IsPhi(phi)) {
break;
}
PrecolorInst(phi);
}
}
@@ -218,8 +223,15 @@ std::string EmitGLSL(const Profile& profile, const RuntimeInfo& runtime_info, IR
const std::string version{fmt::format("#version 450{}\n", GlslVersionSpecifier(ctx))};
ctx.header.insert(0, version);
if (program.shared_memory_size > 0) {
ctx.header +=
fmt::format("shared uint smem[{}];", Common::DivCeil(program.shared_memory_size, 4U));
const auto requested_size{program.shared_memory_size};
const auto max_size{profile.gl_max_compute_smem_size};
const bool needs_clamp{requested_size > max_size};
if (needs_clamp) {
LOG_WARNING(Shader_GLSL, "Requested shared memory size ({}) exceeds device limit ({})",
requested_size, max_size);
}
const auto smem_size{needs_clamp ? max_size : requested_size};
ctx.header += fmt::format("shared uint smem[{}];", Common::DivCeil(smem_size, 4U));
}
ctx.header += "void main(){\n";
if (program.local_memory_size > 0) {

2
src/shader_recompiler/backend/glsl/emit_glsl_floating_point.cpp
View File

@@ -22,7 +22,7 @@ void Compare(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string
}
bool IsPrecise(const IR::Inst& inst) {
return {inst.Flags<IR::FpControl>().no_contraction};
return inst.Flags<IR::FpControl>().no_contraction;
}
} // Anonymous namespace

15
src/shader_recompiler/backend/spirv/emit_spirv_image.cpp
View File

@@ -109,7 +109,7 @@ private:
return;
}
if (offset.IsImmediate()) {
Add(spv::ImageOperandsMask::ConstOffset, ctx.SConst(offset.U32()));
Add(spv::ImageOperandsMask::ConstOffset, ctx.SConst(static_cast<s32>(offset.U32())));
return;
}
IR::Inst* const inst{offset.InstRecursive()};
@@ -117,16 +117,21 @@ private:
switch (inst->GetOpcode()) {
case IR::Opcode::CompositeConstructU32x2:
Add(spv::ImageOperandsMask::ConstOffset,
ctx.SConst(inst->Arg(0).U32(), inst->Arg(1).U32()));
ctx.SConst(static_cast<s32>(inst->Arg(0).U32()),
static_cast<s32>(inst->Arg(1).U32())));
return;
case IR::Opcode::CompositeConstructU32x3:
Add(spv::ImageOperandsMask::ConstOffset,
ctx.SConst(inst->Arg(0).U32(), inst->Arg(1).U32(), inst->Arg(2).U32()));
ctx.SConst(static_cast<s32>(inst->Arg(0).U32()),
static_cast<s32>(inst->Arg(1).U32()),
static_cast<s32>(inst->Arg(2).U32())));
return;
case IR::Opcode::CompositeConstructU32x4:
Add(spv::ImageOperandsMask::ConstOffset,
ctx.SConst(inst->Arg(0).U32(), inst->Arg(1).U32(), inst->Arg(2).U32(),
inst->Arg(3).U32()));
ctx.SConst(static_cast<s32>(inst->Arg(0).U32()),
static_cast<s32>(inst->Arg(1).U32()),
static_cast<s32>(inst->Arg(2).U32()),
static_cast<s32>(inst->Arg(3).U32())));
return;
default:
break;

3
src/shader_recompiler/frontend/ir/opcodes.h
View File

@@ -67,7 +67,8 @@ constexpr OpcodeMeta META_TABLE[]{
};
constexpr size_t CalculateNumArgsOf(Opcode op) {
const auto& arg_types{META_TABLE[static_cast<size_t>(op)].arg_types};
return std::distance(arg_types.begin(), std::ranges::find(arg_types, Type::Void));
return static_cast<size_t>(
std::distance(arg_types.begin(), std::ranges::find(arg_types, Type::Void)));
}
constexpr u8 NUM_ARGS[]{

13
src/shader_recompiler/frontend/maxwell/control_flow.cpp
View File

@@ -5,7 +5,6 @@
#include <algorithm>
#include <array>
#include <optional>
#include <ranges>
#include <string>
#include <utility>
@@ -151,18 +150,18 @@ std::pair<Location, Stack> Stack::Pop(Token token) const {
}
std::optional<Location> Stack::Peek(Token token) const {
const auto reverse_entries{entries | std::views::reverse};
const auto it{std::ranges::find(reverse_entries, token, &StackEntry::token)};
if (it == reverse_entries.end()) {
const auto it{std::find_if(entries.rbegin(), entries.rend(),
[token](const auto& entry) { return entry.token == token; })};
if (it == entries.rend()) {
return std::nullopt;
}
return it->target;
}
Stack Stack::Remove(Token token) const {
const auto reverse_entries{entries | std::views::reverse};
const auto it{std::ranges::find(reverse_entries, token, &StackEntry::token)};
const auto pos{std::distance(reverse_entries.begin(), it)};
const auto it{std::find_if(entries.rbegin(), entries.rend(),
[token](const auto& entry) { return entry.token == token; })};
const auto pos{std::distance(entries.rbegin(), it)};
Stack result;
result.entries.insert(result.entries.end(), entries.begin(), entries.end() - pos - 1);
return result;

1
src/shader_recompiler/frontend/maxwell/control_flow.h
View File

@@ -161,7 +161,6 @@ private:
Environment& env;
ObjectPool<Block>& block_pool;
boost::container::small_vector<Function, 1> functions;
FunctionId current_function_id{0};
Location program_start;
bool exits_to_dispatcher{};
Block* dispatch_block{};

17
src/shader_recompiler/frontend/maxwell/structured_control_flow.cpp
View File

@@ -4,7 +4,6 @@
#include <algorithm>
#include <memory>
#include <ranges>
#include <string>
#include <unordered_map>
#include <utility>
@@ -167,7 +166,7 @@ std::string DumpExpr(const Statement* stmt) {
}
}
std::string DumpTree(const Tree& tree, u32 indentation = 0) {
[[maybe_unused]] std::string DumpTree(const Tree& tree, u32 indentation = 0) {
std::string ret;
std::string indent(indentation, ' ');
for (auto stmt = tree.begin(); stmt != tree.end(); ++stmt) {
@@ -313,12 +312,11 @@ bool NeedsLift(Node goto_stmt, Node label_stmt) noexcept {
class GotoPass {
public:
explicit GotoPass(Flow::CFG& cfg, ObjectPool<IR::Inst>& inst_pool_,
ObjectPool<IR::Block>& block_pool_, ObjectPool<Statement>& stmt_pool)
: inst_pool{inst_pool_}, block_pool{block_pool_}, pool{stmt_pool} {
explicit GotoPass(Flow::CFG& cfg, ObjectPool<Statement>& stmt_pool) : pool{stmt_pool} {
std::vector gotos{BuildTree(cfg)};
for (const Node& goto_stmt : gotos | std::views::reverse) {
RemoveGoto(goto_stmt);
const auto end{gotos.rend()};
for (auto goto_stmt = gotos.rbegin(); goto_stmt != end; ++goto_stmt) {
RemoveGoto(*goto_stmt);
}
}
@@ -616,8 +614,6 @@ private:
return parent_tree.insert(std::next(loop), *new_goto);
}
ObjectPool<IR::Inst>& inst_pool;
ObjectPool<IR::Block>& block_pool;
ObjectPool<Statement>& pool;
Statement root_stmt{FunctionTag{}};
};
@@ -864,7 +860,6 @@ private:
ObjectPool<IR::Block>& block_pool;
Environment& env;
IR::AbstractSyntaxList& syntax_list;
u32 loop_id{};
// TODO: C++20 Remove this when all compilers support constexpr std::vector
#if __cpp_lib_constexpr_vector >= 201907
@@ -878,7 +873,7 @@ private:
IR::AbstractSyntaxList BuildASL(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Block>& block_pool,
Environment& env, Flow::CFG& cfg) {
ObjectPool<Statement> stmt_pool{64};
GotoPass goto_pass{cfg, inst_pool, block_pool, stmt_pool};
GotoPass goto_pass{cfg, stmt_pool};
Statement& root{goto_pass.RootStatement()};
IR::AbstractSyntaxList syntax_list;
TranslatePass{inst_pool, block_pool, stmt_pool, env, root, syntax_list};

12
src/shader_recompiler/frontend/maxwell/translate/impl/atomic_operations_global_memory.cpp
View File

@@ -59,14 +59,14 @@ IR::U32U64 ApplyIntegerAtomOp(IR::IREmitter& ir, const IR::U32U64& offset, const
IR::Value ApplyFpAtomOp(IR::IREmitter& ir, const IR::U64& offset, const IR::Value& op_b, AtomOp op,
AtomSize size) {
static constexpr IR::FpControl f16_control{
.no_contraction{false},
.rounding{IR::FpRounding::RN},
.fmz_mode{IR::FmzMode::DontCare},
.no_contraction = false,
.rounding = IR::FpRounding::RN,
.fmz_mode = IR::FmzMode::DontCare,
};
static constexpr IR::FpControl f32_control{
.no_contraction{false},
.rounding{IR::FpRounding::RN},
.fmz_mode{IR::FmzMode::FTZ},
.no_contraction = false,
.rounding = IR::FpRounding::RN,
.fmz_mode = IR::FmzMode::FTZ,
};
switch (op) {
case AtomOp::ADD:

4
src/shader_recompiler/frontend/maxwell/translate/impl/integer_floating_point_conversion.cpp
View File

@@ -104,7 +104,9 @@ void I2F(TranslatorVisitor& v, u64 insn, IR::U32U64 src) {
.rounding = CastFpRounding(i2f.fp_rounding),
.fmz_mode = IR::FmzMode::DontCare,
};
auto value{v.ir.ConvertIToF(dst_bitsize, conversion_src_bitsize, is_signed, src, fp_control)};
auto value{v.ir.ConvertIToF(static_cast<size_t>(dst_bitsize),
static_cast<size_t>(conversion_src_bitsize), is_signed, src,
fp_control)};
if (i2f.neg != 0) {
if (i2f.abs != 0 || !is_signed) {
// We know the value is positive

12
src/shader_recompiler/frontend/maxwell/translate/impl/load_store_attribute.cpp
View File

@@ -80,10 +80,10 @@ void TranslatorVisitor::ALD(u64 insn) {
for (u32 element = 0; element < num_elements; ++element) {
if (ald.patch != 0) {
const IR::Patch patch{offset / 4 + element};
F(ald.dest_reg + element, ir.GetPatch(patch));
F(ald.dest_reg + static_cast<int>(element), ir.GetPatch(patch));
} else {
const IR::Attribute attr{offset / 4 + element};
F(ald.dest_reg + element, ir.GetAttribute(attr, vertex));
F(ald.dest_reg + static_cast<int>(element), ir.GetAttribute(attr, vertex));
}
}
return;
@@ -92,7 +92,7 @@ void TranslatorVisitor::ALD(u64 insn) {
throw NotImplementedException("Indirect patch read");
}
HandleIndexed(*this, ald.index_reg, num_elements, [&](u32 element, IR::U32 final_offset) {
F(ald.dest_reg + element, ir.GetAttributeIndexed(final_offset, vertex));
F(ald.dest_reg + static_cast<int>(element), ir.GetAttributeIndexed(final_offset, vertex));
});
}
@@ -121,10 +121,10 @@ void TranslatorVisitor::AST(u64 insn) {
for (u32 element = 0; element < num_elements; ++element) {
if (ast.patch != 0) {
const IR::Patch patch{offset / 4 + element};
ir.SetPatch(patch, F(ast.src_reg + element));
ir.SetPatch(patch, F(ast.src_reg + static_cast<int>(element)));
} else {
const IR::Attribute attr{offset / 4 + element};
ir.SetAttribute(attr, F(ast.src_reg + element), vertex);
ir.SetAttribute(attr, F(ast.src_reg + static_cast<int>(element)), vertex);
}
}
return;
@@ -133,7 +133,7 @@ void TranslatorVisitor::AST(u64 insn) {
throw NotImplementedException("Indexed tessellation patch store");
}
HandleIndexed(*this, ast.index_reg, num_elements, [&](u32 element, IR::U32 final_offset) {
ir.SetAttributeIndexed(final_offset, F(ast.src_reg + element), vertex);
ir.SetAttributeIndexed(final_offset, F(ast.src_reg + static_cast<int>(element)), vertex);
});
}

3
src/shader_recompiler/frontend/maxwell/translate/impl/surface_atomic_operations.cpp
View File

@@ -69,9 +69,6 @@ TextureType GetType(Type type) {
}
IR::Value MakeCoords(TranslatorVisitor& v, IR::Reg reg, Type type) {
const auto array{[&](int index) {
return v.ir.BitFieldExtract(v.X(reg + index), v.ir.Imm32(0), v.ir.Imm32(16));
}};
switch (type) {
case Type::_1D:
case Type::BUFFER_1D:

8
src/shader_recompiler/frontend/maxwell/translate/impl/surface_load_store.cpp
View File

@@ -160,10 +160,10 @@ unsigned SwizzleMask(u64 swizzle) {
IR::Value MakeColor(IR::IREmitter& ir, IR::Reg reg, int num_regs) {
std::array<IR::U32, 4> colors;
for (int i = 0; i < num_regs; ++i) {
colors[i] = ir.GetReg(reg + i);
colors[static_cast<size_t>(i)] = ir.GetReg(reg + i);
}
for (int i = num_regs; i < 4; ++i) {
colors[i] = ir.Imm32(0);
colors[static_cast<size_t>(i)] = ir.Imm32(0);
}
return ir.CompositeConstruct(colors[0], colors[1], colors[2], colors[3]);
}
@@ -211,12 +211,12 @@ void TranslatorVisitor::SULD(u64 insn) {
if (is_typed) {
const int num_regs{SizeInRegs(suld.size)};
for (int i = 0; i < num_regs; ++i) {
X(dest_reg + i, IR::U32{ir.CompositeExtract(result, i)});
X(dest_reg + i, IR::U32{ir.CompositeExtract(result, static_cast<size_t>(i))});
}
} else {
const unsigned mask{SwizzleMask(suld.swizzle)};
const int bits{std::popcount(mask)};
if (!IR::IsAligned(dest_reg, bits == 3 ? 4 : bits)) {
if (!IR::IsAligned(dest_reg, bits == 3 ? 4 : static_cast<size_t>(bits))) {
throw NotImplementedException("Unaligned destination register");
}
for (unsigned component = 0; component < 4; ++component) {

20
src/shader_recompiler/frontend/maxwell/translate_program.cpp
View File

@@ -4,7 +4,6 @@
#include <algorithm>
#include <memory>
#include <ranges>
#include <vector>
#include "common/settings.h"
@@ -20,12 +19,19 @@
namespace Shader::Maxwell {
namespace {
IR::BlockList GenerateBlocks(const IR::AbstractSyntaxList& syntax_list) {
auto syntax_blocks{syntax_list | std::views::filter([](const auto& node) {
return node.type == IR::AbstractSyntaxNode::Type::Block;
})};
IR::BlockList blocks(std::ranges::distance(syntax_blocks));
std::ranges::transform(syntax_blocks, blocks.begin(),
[](const IR::AbstractSyntaxNode& node) { return node.data.block; });
size_t num_syntax_blocks{};
for (const auto& node : syntax_list) {
if (node.type == IR::AbstractSyntaxNode::Type::Block) {
++num_syntax_blocks;
}
}
IR::BlockList blocks;
blocks.reserve(num_syntax_blocks);
for (const auto& node : syntax_list) {
if (node.type == IR::AbstractSyntaxNode::Type::Block) {
blocks.push_back(node.data.block);
}
}
return blocks;
}

5
src/shader_recompiler/ir_opt/constant_propagation_pass.cpp
View File

@@ -3,7 +3,6 @@
// Refer to the license.txt file included.
#include <algorithm>
#include <ranges>
#include <tuple>
#include <type_traits>
@@ -599,7 +598,9 @@ void ConstantPropagation(IR::Block& block, IR::Inst& inst) {
} // Anonymous namespace
void ConstantPropagationPass(IR::Program& program) {
for (IR::Block* const block : program.post_order_blocks | std::views::reverse) {
const auto end{program.post_order_blocks.rend()};
for (auto it = program.post_order_blocks.rbegin(); it != end; ++it) {
IR::Block* const block{*it};
for (IR::Inst& inst : block->Instructions()) {
ConstantPropagation(*block, inst);
}

2
src/shader_recompiler/ir_opt/dead_code_elimination_pass.cpp
View File

@@ -2,8 +2,6 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <ranges>
#include "shader_recompiler/frontend/ir/basic_block.h"
#include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/ir_opt/passes.h"

6
src/shader_recompiler/ir_opt/dual_vertex_pass.cpp
View File

@@ -2,12 +2,6 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <ranges>
#include "common/bit_cast.h"
#include "common/bit_util.h"
#include "shader_recompiler/exception.h"
#include "shader_recompiler/frontend/ir/ir_emitter.h"
#include "shader_recompiler/ir_opt/passes.h"

7
src/shader_recompiler/ir_opt/global_memory_to_storage_buffer_pass.cpp
View File

@@ -5,7 +5,6 @@
#include <algorithm>
#include <compare>
#include <optional>
#include <ranges>
#include <queue>
#include <boost/container/flat_set.hpp>
@@ -314,8 +313,8 @@ std::optional<StorageBufferAddr> Track(const IR::Value& value, const Bias* bias)
return std::nullopt;
}
const StorageBufferAddr storage_buffer{
.index{index.U32()},
.offset{offset.U32()},
.index = index.U32(),
.offset = offset.U32(),
};
if (!Common::IsAligned(storage_buffer.offset, 16)) {
// The SSBO pointer has to be aligned
@@ -484,7 +483,7 @@ void GlobalMemoryToStorageBufferPass(IR::Program& program) {
.cbuf_index = storage_buffer.index,
.cbuf_offset = storage_buffer.offset,
.count = 1,
.is_written{info.writes.contains(storage_buffer)},
.is_written = info.writes.contains(storage_buffer),
});
}
for (const StorageInst& storage_inst : info.to_replace) {

5
src/shader_recompiler/ir_opt/lower_int64_to_int32.cpp
View File

@@ -2,7 +2,6 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <ranges>
#include <utility>
#include "shader_recompiler/exception.h"
@@ -207,7 +206,9 @@ void Lower(IR::Block& block, IR::Inst& inst) {
} // Anonymous namespace
void LowerInt64ToInt32(IR::Program& program) {
for (IR::Block* const block : program.post_order_blocks | std::views::reverse) {
const auto end{program.post_order_blocks.rend()};
for (auto it = program.post_order_blocks.rbegin(); it != end; ++it) {
IR::Block* const block{*it};
for (IR::Inst& inst : block->Instructions()) {
Lower(*block, inst);
}

10
src/shader_recompiler/ir_opt/ssa_rewrite_pass.cpp
View File

@@ -14,7 +14,6 @@
// https://link.springer.com/chapter/10.1007/978-3-642-37051-9_6
//
#include <ranges>
#include <span>
#include <variant>
#include <vector>
@@ -243,7 +242,9 @@ public:
void SealBlock(IR::Block* block) {
const auto it{incomplete_phis.find(block)};
if (it != incomplete_phis.end()) {
for (auto& [variant, phi] : it->second) {
for (auto& pair : it->second) {
auto& variant{pair.first};
auto& phi{pair.second};
std::visit([&](auto& variable) { AddPhiOperands(variable, *phi, block); }, variant);
}
}
@@ -373,8 +374,9 @@ void VisitBlock(Pass& pass, IR::Block* block) {
void SsaRewritePass(IR::Program& program) {
Pass pass;
for (IR::Block* const block : program.post_order_blocks | std::views::reverse) {
VisitBlock(pass, block);
const auto end{program.post_order_blocks.rend()};
for (auto block = program.post_order_blocks.rbegin(); block != end; ++block) {
VisitBlock(pass, *block);
}
}

2
src/shader_recompiler/profile.h
View File

@@ -67,6 +67,8 @@ struct Profile {
bool has_gl_precise_bug{};
/// Ignores SPIR-V ordered vs unordered using GLSL semantics
bool ignore_nan_fp_comparisons{};
u32 gl_max_compute_smem_size{};
};
} // namespace Shader