summaryrefslogtreecommitdiffstats
path: root/src/video_core/macro
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core/macro')
-rw-r--r--src/video_core/macro/macro.cpp35
-rw-r--r--src/video_core/macro/macro.h19
-rw-r--r--src/video_core/macro/macro_hle.cpp113
-rw-r--r--src/video_core/macro/macro_hle.h44
-rw-r--r--src/video_core/macro/macro_interpreter.cpp3
-rw-r--r--src/video_core/macro/macro_jit_x64.cpp65
-rw-r--r--src/video_core/macro/macro_jit_x64.h1
7 files changed, 245 insertions, 35 deletions
diff --git a/src/video_core/macro/macro.cpp b/src/video_core/macro/macro.cpp
index 89077a2d8..ef7dad349 100644
--- a/src/video_core/macro/macro.cpp
+++ b/src/video_core/macro/macro.cpp
@@ -2,23 +2,37 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
+#include <boost/container_hash/hash.hpp>
#include "common/assert.h"
#include "common/logging/log.h"
#include "core/settings.h"
+#include "video_core/engines/maxwell_3d.h"
#include "video_core/macro/macro.h"
+#include "video_core/macro/macro_hle.h"
#include "video_core/macro/macro_interpreter.h"
#include "video_core/macro/macro_jit_x64.h"
namespace Tegra {
+MacroEngine::MacroEngine(Engines::Maxwell3D& maxwell3d)
+ : hle_macros{std::make_unique<Tegra::HLEMacro>(maxwell3d)} {}
+
+MacroEngine::~MacroEngine() = default;
+
void MacroEngine::AddCode(u32 method, u32 data) {
uploaded_macro_code[method].push_back(data);
}
-void MacroEngine::Execute(u32 method, const std::vector<u32>& parameters) {
+void MacroEngine::Execute(Engines::Maxwell3D& maxwell3d, u32 method,
+ const std::vector<u32>& parameters) {
auto compiled_macro = macro_cache.find(method);
if (compiled_macro != macro_cache.end()) {
- compiled_macro->second->Execute(parameters, method);
+ const auto& cache_info = compiled_macro->second;
+ if (cache_info.has_hle_program) {
+ cache_info.hle_program->Execute(parameters, method);
+ } else {
+ cache_info.lle_program->Execute(parameters, method);
+ }
} else {
// Macro not compiled, check if it's uploaded and if so, compile it
auto macro_code = uploaded_macro_code.find(method);
@@ -26,8 +40,21 @@ void MacroEngine::Execute(u32 method, const std::vector<u32>& parameters) {
UNREACHABLE_MSG("Macro 0x{0:x} was not uploaded", method);
return;
}
- macro_cache[method] = Compile(macro_code->second);
- macro_cache[method]->Execute(parameters, method);
+ auto& cache_info = macro_cache[method];
+ cache_info.hash = boost::hash_value(macro_code->second);
+ cache_info.lle_program = Compile(macro_code->second);
+
+ auto hle_program = hle_macros->GetHLEProgram(cache_info.hash);
+ if (hle_program.has_value()) {
+ cache_info.has_hle_program = true;
+ cache_info.hle_program = std::move(hle_program.value());
+ }
+
+ if (cache_info.has_hle_program) {
+ cache_info.hle_program->Execute(parameters, method);
+ } else {
+ cache_info.lle_program->Execute(parameters, method);
+ }
}
}
diff --git a/src/video_core/macro/macro.h b/src/video_core/macro/macro.h
index b76ed891f..4d00b84b0 100644
--- a/src/video_core/macro/macro.h
+++ b/src/video_core/macro/macro.h
@@ -11,9 +11,11 @@
#include "common/common_types.h"
namespace Tegra {
+
namespace Engines {
class Maxwell3D;
}
+
namespace Macro {
constexpr std::size_t NUM_MACRO_REGISTERS = 8;
enum class Operation : u32 {
@@ -94,6 +96,8 @@ union MethodAddress {
} // namespace Macro
+class HLEMacro;
+
class CachedMacro {
public:
virtual ~CachedMacro() = default;
@@ -107,20 +111,29 @@ public:
class MacroEngine {
public:
- virtual ~MacroEngine() = default;
+ explicit MacroEngine(Engines::Maxwell3D& maxwell3d);
+ virtual ~MacroEngine();
// Store the uploaded macro code to compile them when they're called.
void AddCode(u32 method, u32 data);
// Compiles the macro if its not in the cache, and executes the compiled macro
- void Execute(u32 method, const std::vector<u32>& parameters);
+ void Execute(Engines::Maxwell3D& maxwell3d, u32 method, const std::vector<u32>& parameters);
protected:
virtual std::unique_ptr<CachedMacro> Compile(const std::vector<u32>& code) = 0;
private:
- std::unordered_map<u32, std::unique_ptr<CachedMacro>> macro_cache;
+ struct CacheInfo {
+ std::unique_ptr<CachedMacro> lle_program{};
+ std::unique_ptr<CachedMacro> hle_program{};
+ u64 hash{};
+ bool has_hle_program{};
+ };
+
+ std::unordered_map<u32, CacheInfo> macro_cache;
std::unordered_map<u32, std::vector<u32>> uploaded_macro_code;
+ std::unique_ptr<HLEMacro> hle_macros;
};
std::unique_ptr<MacroEngine> GetMacroEngine(Engines::Maxwell3D& maxwell3d);
diff --git a/src/video_core/macro/macro_hle.cpp b/src/video_core/macro/macro_hle.cpp
new file mode 100644
index 000000000..410f99018
--- /dev/null
+++ b/src/video_core/macro/macro_hle.cpp
@@ -0,0 +1,113 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <array>
+#include <vector>
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/macro/macro_hle.h"
+#include "video_core/rasterizer_interface.h"
+
+namespace Tegra {
+
+namespace {
+// HLE'd functions
+static void HLE_771BB18C62444DA0(Engines::Maxwell3D& maxwell3d,
+ const std::vector<u32>& parameters) {
+ const u32 instance_count = parameters[2] & maxwell3d.GetRegisterValue(0xD1B);
+
+ maxwell3d.regs.draw.topology.Assign(
+ static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[0] &
+ ~(0x3ffffff << 26)));
+ maxwell3d.regs.vb_base_instance = parameters[5];
+ maxwell3d.mme_draw.instance_count = instance_count;
+ maxwell3d.regs.vb_element_base = parameters[3];
+ maxwell3d.regs.index_array.count = parameters[1];
+ maxwell3d.regs.index_array.first = parameters[4];
+
+ if (maxwell3d.ShouldExecute()) {
+ maxwell3d.GetRasterizer().Draw(true, true);
+ }
+ maxwell3d.regs.index_array.count = 0;
+ maxwell3d.mme_draw.instance_count = 0;
+ maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
+}
+
+static void HLE_0D61FC9FAAC9FCAD(Engines::Maxwell3D& maxwell3d,
+ const std::vector<u32>& parameters) {
+ const u32 count = (maxwell3d.GetRegisterValue(0xD1B) & parameters[2]);
+
+ maxwell3d.regs.vertex_buffer.first = parameters[3];
+ maxwell3d.regs.vertex_buffer.count = parameters[1];
+ maxwell3d.regs.vb_base_instance = parameters[4];
+ maxwell3d.regs.draw.topology.Assign(
+ static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[0]));
+ maxwell3d.mme_draw.instance_count = count;
+
+ if (maxwell3d.ShouldExecute()) {
+ maxwell3d.GetRasterizer().Draw(false, true);
+ }
+ maxwell3d.regs.vertex_buffer.count = 0;
+ maxwell3d.mme_draw.instance_count = 0;
+ maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
+}
+
+static void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d,
+ const std::vector<u32>& parameters) {
+ const u32 instance_count = (maxwell3d.GetRegisterValue(0xD1B) & parameters[2]);
+ const u32 element_base = parameters[4];
+ const u32 base_instance = parameters[5];
+ maxwell3d.regs.index_array.first = parameters[3];
+ maxwell3d.regs.reg_array[0x446] = element_base; // vertex id base?
+ maxwell3d.regs.index_array.count = parameters[1];
+ maxwell3d.regs.vb_element_base = element_base;
+ maxwell3d.regs.vb_base_instance = base_instance;
+ maxwell3d.mme_draw.instance_count = instance_count;
+ maxwell3d.CallMethodFromMME(0x8e3, 0x640);
+ maxwell3d.CallMethodFromMME(0x8e4, element_base);
+ maxwell3d.CallMethodFromMME(0x8e5, base_instance);
+ maxwell3d.regs.draw.topology.Assign(
+ static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[0]));
+ if (maxwell3d.ShouldExecute()) {
+ maxwell3d.GetRasterizer().Draw(true, true);
+ }
+ maxwell3d.regs.reg_array[0x446] = 0x0; // vertex id base?
+ maxwell3d.regs.index_array.count = 0;
+ maxwell3d.regs.vb_element_base = 0x0;
+ maxwell3d.regs.vb_base_instance = 0x0;
+ maxwell3d.mme_draw.instance_count = 0;
+ maxwell3d.CallMethodFromMME(0x8e3, 0x640);
+ maxwell3d.CallMethodFromMME(0x8e4, 0x0);
+ maxwell3d.CallMethodFromMME(0x8e5, 0x0);
+ maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
+}
+} // namespace
+
+constexpr std::array<std::pair<u64, HLEFunction>, 3> hle_funcs{{
+ std::make_pair<u64, HLEFunction>(0x771BB18C62444DA0, &HLE_771BB18C62444DA0),
+ std::make_pair<u64, HLEFunction>(0x0D61FC9FAAC9FCAD, &HLE_0D61FC9FAAC9FCAD),
+ std::make_pair<u64, HLEFunction>(0x0217920100488FF7, &HLE_0217920100488FF7),
+}};
+
+HLEMacro::HLEMacro(Engines::Maxwell3D& maxwell3d) : maxwell3d(maxwell3d) {}
+HLEMacro::~HLEMacro() = default;
+
+std::optional<std::unique_ptr<CachedMacro>> HLEMacro::GetHLEProgram(u64 hash) const {
+ const auto it = std::find_if(hle_funcs.cbegin(), hle_funcs.cend(),
+ [hash](const auto& pair) { return pair.first == hash; });
+ if (it == hle_funcs.end()) {
+ return std::nullopt;
+ }
+ return std::make_unique<HLEMacroImpl>(maxwell3d, it->second);
+}
+
+HLEMacroImpl::~HLEMacroImpl() = default;
+
+HLEMacroImpl::HLEMacroImpl(Engines::Maxwell3D& maxwell3d, HLEFunction func)
+ : maxwell3d(maxwell3d), func(func) {}
+
+void HLEMacroImpl::Execute(const std::vector<u32>& parameters, u32 method) {
+ func(maxwell3d, parameters);
+}
+
+} // namespace Tegra
diff --git a/src/video_core/macro/macro_hle.h b/src/video_core/macro/macro_hle.h
new file mode 100644
index 000000000..37af875a0
--- /dev/null
+++ b/src/video_core/macro/macro_hle.h
@@ -0,0 +1,44 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <memory>
+#include <optional>
+#include <vector>
+#include "common/common_types.h"
+#include "video_core/macro/macro.h"
+
+namespace Tegra {
+
+namespace Engines {
+class Maxwell3D;
+}
+
+using HLEFunction = void (*)(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& parameters);
+
+class HLEMacro {
+public:
+ explicit HLEMacro(Engines::Maxwell3D& maxwell3d);
+ ~HLEMacro();
+
+ std::optional<std::unique_ptr<CachedMacro>> GetHLEProgram(u64 hash) const;
+
+private:
+ Engines::Maxwell3D& maxwell3d;
+};
+
+class HLEMacroImpl : public CachedMacro {
+public:
+ explicit HLEMacroImpl(Engines::Maxwell3D& maxwell3d, HLEFunction func);
+ ~HLEMacroImpl();
+
+ void Execute(const std::vector<u32>& parameters, u32 method) override;
+
+private:
+ Engines::Maxwell3D& maxwell3d;
+ HLEFunction func;
+};
+
+} // namespace Tegra
diff --git a/src/video_core/macro/macro_interpreter.cpp b/src/video_core/macro/macro_interpreter.cpp
index 5edff27aa..aa5256419 100644
--- a/src/video_core/macro/macro_interpreter.cpp
+++ b/src/video_core/macro/macro_interpreter.cpp
@@ -11,7 +11,8 @@
MICROPROFILE_DEFINE(MacroInterp, "GPU", "Execute macro interpreter", MP_RGB(128, 128, 192));
namespace Tegra {
-MacroInterpreter::MacroInterpreter(Engines::Maxwell3D& maxwell3d) : maxwell3d(maxwell3d) {}
+MacroInterpreter::MacroInterpreter(Engines::Maxwell3D& maxwell3d)
+ : MacroEngine::MacroEngine(maxwell3d), maxwell3d(maxwell3d) {}
std::unique_ptr<CachedMacro> MacroInterpreter::Compile(const std::vector<u32>& code) {
return std::make_unique<MacroInterpreterImpl>(maxwell3d, code);
diff --git a/src/video_core/macro/macro_jit_x64.cpp b/src/video_core/macro/macro_jit_x64.cpp
index d4a97ec7b..07292702f 100644
--- a/src/video_core/macro/macro_jit_x64.cpp
+++ b/src/video_core/macro/macro_jit_x64.cpp
@@ -28,7 +28,8 @@ static const std::bitset<32> PERSISTENT_REGISTERS = Common::X64::BuildRegSet({
BRANCH_HOLDER,
});
-MacroJITx64::MacroJITx64(Engines::Maxwell3D& maxwell3d) : maxwell3d(maxwell3d) {}
+MacroJITx64::MacroJITx64(Engines::Maxwell3D& maxwell3d)
+ : MacroEngine::MacroEngine(maxwell3d), maxwell3d(maxwell3d) {}
std::unique_ptr<CachedMacro> MacroJITx64::Compile(const std::vector<u32>& code) {
return std::make_unique<MacroJITx64Impl>(maxwell3d, code);
@@ -54,13 +55,15 @@ void MacroJITx64Impl::Compile_ALU(Macro::Opcode opcode) {
const bool is_a_zero = opcode.src_a == 0;
const bool is_b_zero = opcode.src_b == 0;
const bool valid_operation = !is_a_zero && !is_b_zero;
- const bool is_move_operation = !is_a_zero && is_b_zero;
+ [[maybe_unused]] const bool is_move_operation = !is_a_zero && is_b_zero;
const bool has_zero_register = is_a_zero || is_b_zero;
+ const bool no_zero_reg_skip = opcode.alu_operation == Macro::ALUOperation::AddWithCarry ||
+ opcode.alu_operation == Macro::ALUOperation::SubtractWithBorrow;
Xbyak::Reg32 src_a;
Xbyak::Reg32 src_b;
- if (!optimizer.zero_reg_skip) {
+ if (!optimizer.zero_reg_skip || no_zero_reg_skip) {
src_a = Compile_GetRegister(opcode.src_a, RESULT);
src_b = Compile_GetRegister(opcode.src_b, eax);
} else {
@@ -71,7 +74,6 @@ void MacroJITx64Impl::Compile_ALU(Macro::Opcode opcode) {
src_b = Compile_GetRegister(opcode.src_b, eax);
}
}
- Xbyak::Label skip_carry{};
bool has_emitted = false;
@@ -183,7 +185,8 @@ void MacroJITx64Impl::Compile_AddImmediate(Macro::Opcode opcode) {
opcode.result_operation == Macro::ResultOperation::MoveAndSetMethod) {
if (next_opcode.has_value()) {
const auto next = *next_opcode;
- if (next.result_operation == Macro::ResultOperation::MoveAndSetMethod) {
+ if (next.result_operation == Macro::ResultOperation::MoveAndSetMethod &&
+ opcode.dst == next.dst) {
return;
}
}
@@ -237,10 +240,10 @@ void MacroJITx64Impl::Compile_ExtractInsert(Macro::Opcode opcode) {
}
void MacroJITx64Impl::Compile_ExtractShiftLeftImmediate(Macro::Opcode opcode) {
- auto dst = Compile_GetRegister(opcode.src_a, eax);
- auto src = Compile_GetRegister(opcode.src_b, RESULT);
+ const auto dst = Compile_GetRegister(opcode.src_a, ecx);
+ const auto src = Compile_GetRegister(opcode.src_b, RESULT);
- shr(src, al);
+ shr(src, dst.cvt8());
if (opcode.bf_size != 0 && opcode.bf_size != 31) {
and_(src, opcode.GetBitfieldMask());
} else if (opcode.bf_size == 0) {
@@ -256,8 +259,8 @@ void MacroJITx64Impl::Compile_ExtractShiftLeftImmediate(Macro::Opcode opcode) {
}
void MacroJITx64Impl::Compile_ExtractShiftLeftRegister(Macro::Opcode opcode) {
- auto dst = Compile_GetRegister(opcode.src_a, eax);
- auto src = Compile_GetRegister(opcode.src_b, RESULT);
+ const auto dst = Compile_GetRegister(opcode.src_a, ecx);
+ const auto src = Compile_GetRegister(opcode.src_b, RESULT);
if (opcode.bf_src_bit != 0) {
shr(src, opcode.bf_src_bit);
@@ -266,16 +269,9 @@ void MacroJITx64Impl::Compile_ExtractShiftLeftRegister(Macro::Opcode opcode) {
if (opcode.bf_size != 31) {
and_(src, opcode.GetBitfieldMask());
}
- shl(src, al);
- Compile_ProcessResult(opcode.result_operation, opcode.dst);
-}
+ shl(src, dst.cvt8());
-static u32 Read(Engines::Maxwell3D* maxwell3d, u32 method) {
- return maxwell3d->GetRegisterValue(method);
-}
-
-static void Send(Engines::Maxwell3D* maxwell3d, Macro::MethodAddress method_address, u32 value) {
- maxwell3d->CallMethodFromMME(method_address.address, value);
+ Compile_ProcessResult(opcode.result_operation, opcode.dst);
}
void MacroJITx64Impl::Compile_Read(Macro::Opcode opcode) {
@@ -295,15 +291,27 @@ void MacroJITx64Impl::Compile_Read(Macro::Opcode opcode) {
sub(result, opcode.immediate * -1);
}
}
- Common::X64::ABI_PushRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
- mov(Common::X64::ABI_PARAM1, qword[STATE]);
- mov(Common::X64::ABI_PARAM2, RESULT);
- Common::X64::CallFarFunction(*this, &Read);
- Common::X64::ABI_PopRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
- mov(RESULT, Common::X64::ABI_RETURN.cvt32());
+
+ // Equivalent to Engines::Maxwell3D::GetRegisterValue:
+ if (optimizer.enable_asserts) {
+ Xbyak::Label pass_range_check;
+ cmp(RESULT, static_cast<u32>(Engines::Maxwell3D::Regs::NUM_REGS));
+ jb(pass_range_check);
+ int3();
+ L(pass_range_check);
+ }
+ mov(rax, qword[STATE]);
+ mov(RESULT,
+ dword[rax + offsetof(Engines::Maxwell3D, regs) +
+ offsetof(Engines::Maxwell3D::Regs, reg_array) + RESULT.cvt64() * sizeof(u32)]);
+
Compile_ProcessResult(opcode.result_operation, opcode.dst);
}
+static void Send(Engines::Maxwell3D* maxwell3d, Macro::MethodAddress method_address, u32 value) {
+ maxwell3d->CallMethodFromMME(method_address.address, value);
+}
+
void Tegra::MacroJITx64Impl::Compile_Send(Xbyak::Reg32 value) {
Common::X64::ABI_PushRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
mov(Common::X64::ABI_PARAM1, qword[STATE]);
@@ -435,6 +443,9 @@ void MacroJITx64Impl::Compile() {
// one if our register isn't "dirty"
optimizer.optimize_for_method_move = true;
+ // Enable run-time assertions in JITted code
+ optimizer.enable_asserts = false;
+
// Check to see if we can skip emitting certain instructions
Optimizer_ScanFlags();
@@ -543,7 +554,7 @@ Xbyak::Reg32 MacroJITx64Impl::Compile_GetRegister(u32 index, Xbyak::Reg32 dst) {
}
void MacroJITx64Impl::Compile_ProcessResult(Macro::ResultOperation operation, u32 reg) {
- auto SetRegister = [=](u32 reg, Xbyak::Reg32 result) {
+ const auto SetRegister = [this](u32 reg, const Xbyak::Reg32& result) {
// Register 0 is supposed to always return 0. NOP is implemented as a store to the zero
// register.
if (reg == 0) {
@@ -551,7 +562,7 @@ void MacroJITx64Impl::Compile_ProcessResult(Macro::ResultOperation operation, u3
}
mov(dword[STATE + offsetof(JITState, registers) + reg * sizeof(u32)], result);
};
- auto SetMethodAddress = [=](Xbyak::Reg32 reg) { mov(METHOD_ADDRESS, reg); };
+ const auto SetMethodAddress = [this](const Xbyak::Reg32& reg) { mov(METHOD_ADDRESS, reg); };
switch (operation) {
case Macro::ResultOperation::IgnoreAndFetch:
diff --git a/src/video_core/macro/macro_jit_x64.h b/src/video_core/macro/macro_jit_x64.h
index 51ec090b8..a180e7428 100644
--- a/src/video_core/macro/macro_jit_x64.h
+++ b/src/video_core/macro/macro_jit_x64.h
@@ -76,6 +76,7 @@ private:
bool zero_reg_skip{};
bool skip_dummy_addimmediate{};
bool optimize_for_method_move{};
+ bool enable_asserts{};
};
OptimizerState optimizer{};
generated by cgit v1.2.3 (git 2.25.1) at 2024-10-30 02:40:01 +0100