From 811bff009eca0d0fa2ddb1455fc73fdaec4474da Mon Sep 17 00:00:00 2001 From: Lioncash Date: Fri, 19 Jun 2020 21:57:41 -0400 Subject: macro_jit_x64: Eliminate variable shadowing in Compile_ProcessResult() We can reduce the capture scope so that it's not possible for both "reg" variables to clash with one another. While we're at it, we can prevent unnecessary copies while we're at it. --- src/video_core/macro/macro_jit_x64.cpp | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'src/video_core/macro') diff --git a/src/video_core/macro/macro_jit_x64.cpp b/src/video_core/macro/macro_jit_x64.cpp index bee34a7c0..9eface47e 100644 --- a/src/video_core/macro/macro_jit_x64.cpp +++ b/src/video_core/macro/macro_jit_x64.cpp @@ -546,7 +546,7 @@ 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) { @@ -554,7 +554,7 @@ 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: -- cgit v1.2.3 From 6ce5f3120be6a65a798d3abc6fda0fe6171d0296 Mon Sep 17 00:00:00 2001 From: David Marcec Date: Fri, 5 Jun 2020 01:42:19 +1000 Subject: Macro HLE support --- src/video_core/macro/macro.cpp | 35 ++++++++-- src/video_core/macro/macro.h | 19 ++++- src/video_core/macro/macro_hle.cpp | 108 +++++++++++++++++++++++++++++ src/video_core/macro/macro_hle.h | 43 ++++++++++++ src/video_core/macro/macro_interpreter.cpp | 3 +- src/video_core/macro/macro_jit_x64.cpp | 3 +- 6 files changed, 202 insertions(+), 9 deletions(-) create mode 100644 src/video_core/macro/macro_hle.cpp create mode 100644 src/video_core/macro/macro_hle.h (limited to 'src/video_core/macro') diff --git a/src/video_core/macro/macro.cpp b/src/video_core/macro/macro.cpp index 89077a2d8..c8aa2534a 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 #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(maxwell3d)} {} + +MacroEngine::~MacroEngine() {} + void MacroEngine::AddCode(u32 method, u32 data) { uploaded_macro_code[method].push_back(data); } -void MacroEngine::Execute(u32 method, const std::vector& parameters) { +void MacroEngine::Execute(Engines::Maxwell3D& maxwell3d, u32 method, + const std::vector& 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& 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..5fa8023af 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; + 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& parameters); + void Execute(Engines::Maxwell3D& maxwell3d, u32 method, const std::vector& parameters); protected: virtual std::unique_ptr Compile(const std::vector& code) = 0; private: - std::unordered_map> macro_cache; + struct CacheInfo { + std::unique_ptr lle_program{}; + std::unique_ptr hle_program{}; + u64 hash{}; + bool has_hle_program{}; + }; + + std::unordered_map macro_cache; std::unordered_map> uploaded_macro_code; + std::unique_ptr hle_macros; }; std::unique_ptr 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..51827c822 --- /dev/null +++ b/src/video_core/macro/macro_hle.cpp @@ -0,0 +1,108 @@ +// Copyright 2020 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include +#include +#include "video_core/engines/maxwell_3d.h" +#include "video_core/macro/macro_hle.h" +#include "video_core/rasterizer_interface.h" + +namespace Tegra { + +// HLE'd functions +static void HLE_771BB18C62444DA0(Engines::Maxwell3D& maxwell3d, + const std::vector& parameters) { + const u32 instance_count = parameters[2] & maxwell3d.GetRegisterValue(0xD1B); + + maxwell3d.regs.draw.topology.Assign( + static_cast(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; +} + +static void HLE_0D61FC9FAAC9FCAD(Engines::Maxwell3D& maxwell3d, + const std::vector& 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(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; +} + +static void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d, + const std::vector& 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.regs.const_buffer.cb_pos = 0x640; + maxwell3d.mme_draw.instance_count = instance_count; + maxwell3d.regs.const_buffer.cb_data[0] = element_base; + maxwell3d.regs.const_buffer.cb_data[1] = base_instance; + maxwell3d.regs.draw.topology.Assign( + static_cast(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.regs.const_buffer.cb_pos = 0x640; + maxwell3d.regs.const_buffer.cb_data[0] = 0; + maxwell3d.regs.const_buffer.cb_data[1] = 0; + maxwell3d.mme_draw.instance_count = 0; +} + +static const std::unordered_map hle_funcs{ + {0x771BB18C62444DA0, &HLE_771BB18C62444DA0}, + {0x0D61FC9FAAC9FCAD, &HLE_0D61FC9FAAC9FCAD}, + {0x0217920100488FF7, &HLE_0217920100488FF7}, +}; + +HLEMacro::HLEMacro(Engines::Maxwell3D& maxwell3d) : maxwell3d(maxwell3d) {} +HLEMacro::~HLEMacro() = default; + +std::optional> HLEMacro::GetHLEProgram(u64 hash) const { + auto it = hle_funcs.find(hash); + if (it != hle_funcs.end()) { + return std::make_unique(maxwell3d, it->second); + } else { + return {}; + } +} + +HLEMacroImpl::~HLEMacroImpl() = default; + +HLEMacroImpl::HLEMacroImpl(Engines::Maxwell3D& maxwell3d, HLEFunction func) + : maxwell3d(maxwell3d), func(func) {} + +void HLEMacroImpl::Execute(const std::vector& 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..de7f43dc4 --- /dev/null +++ b/src/video_core/macro/macro_hle.h @@ -0,0 +1,43 @@ +// Copyright 2020 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include +#include +#include +#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& parameters); + +class HLEMacro { +public: + HLEMacro(Engines::Maxwell3D& maxwell3d); + ~HLEMacro(); + std::optional> 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& 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 MacroInterpreter::Compile(const std::vector& code) { return std::make_unique(maxwell3d, code); diff --git a/src/video_core/macro/macro_jit_x64.cpp b/src/video_core/macro/macro_jit_x64.cpp index 30abb66e5..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 MacroJITx64::Compile(const std::vector& code) { return std::make_unique(maxwell3d, code); -- cgit v1.2.3 From 74b4334d510b58d96e8305bc3f5a7c8d05e842ba Mon Sep 17 00:00:00 2001 From: David Marcec Date: Fri, 5 Jun 2020 12:59:59 +1000 Subject: Fix constbuffer for 0217920100488FF7 --- src/video_core/macro/macro_hle.cpp | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) (limited to 'src/video_core/macro') diff --git a/src/video_core/macro/macro_hle.cpp b/src/video_core/macro/macro_hle.cpp index 51827c822..887f40310 100644 --- a/src/video_core/macro/macro_hle.cpp +++ b/src/video_core/macro/macro_hle.cpp @@ -59,10 +59,10 @@ static void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d, maxwell3d.regs.index_array.count = parameters[1]; maxwell3d.regs.vb_element_base = element_base; maxwell3d.regs.vb_base_instance = base_instance; - maxwell3d.regs.const_buffer.cb_pos = 0x640; maxwell3d.mme_draw.instance_count = instance_count; - maxwell3d.regs.const_buffer.cb_data[0] = element_base; - maxwell3d.regs.const_buffer.cb_data[1] = base_instance; + maxwell3d.CallMethodFromMME(0x8e3, 0x640); + maxwell3d.CallMethodFromMME(0x8e4, element_base); + maxwell3d.CallMethodFromMME(0x8e5, base_instance); maxwell3d.regs.draw.topology.Assign( static_cast(parameters[0])); if (maxwell3d.ShouldExecute()) { @@ -72,10 +72,10 @@ static void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d, maxwell3d.regs.index_array.count = 0; maxwell3d.regs.vb_element_base = 0x0; maxwell3d.regs.vb_base_instance = 0x0; - maxwell3d.regs.const_buffer.cb_pos = 0x640; - maxwell3d.regs.const_buffer.cb_data[0] = 0; - maxwell3d.regs.const_buffer.cb_data[1] = 0; maxwell3d.mme_draw.instance_count = 0; + maxwell3d.CallMethodFromMME(0x8e3, 0x640); + maxwell3d.CallMethodFromMME(0x8e4, 0x0); + maxwell3d.CallMethodFromMME(0x8e5, 0x0); } static const std::unordered_map hle_funcs{ -- cgit v1.2.3 From fabdf5d3850c078d173653f259845c26a2ce6e7d Mon Sep 17 00:00:00 2001 From: David Marcec Date: Fri, 5 Jun 2020 13:09:52 +1000 Subject: Addressed issues --- src/video_core/macro/macro.cpp | 2 +- src/video_core/macro/macro.h | 2 +- src/video_core/macro/macro_hle.cpp | 20 ++++++++++---------- src/video_core/macro/macro_hle.h | 2 +- 4 files changed, 13 insertions(+), 13 deletions(-) (limited to 'src/video_core/macro') diff --git a/src/video_core/macro/macro.cpp b/src/video_core/macro/macro.cpp index c8aa2534a..ef7dad349 100644 --- a/src/video_core/macro/macro.cpp +++ b/src/video_core/macro/macro.cpp @@ -17,7 +17,7 @@ namespace Tegra { MacroEngine::MacroEngine(Engines::Maxwell3D& maxwell3d) : hle_macros{std::make_unique(maxwell3d)} {} -MacroEngine::~MacroEngine() {} +MacroEngine::~MacroEngine() = default; void MacroEngine::AddCode(u32 method, u32 data) { uploaded_macro_code[method].push_back(data); diff --git a/src/video_core/macro/macro.h b/src/video_core/macro/macro.h index 5fa8023af..4d00b84b0 100644 --- a/src/video_core/macro/macro.h +++ b/src/video_core/macro/macro.h @@ -111,7 +111,7 @@ public: class MacroEngine { public: - MacroEngine(Engines::Maxwell3D& maxwell3d); + explicit MacroEngine(Engines::Maxwell3D& maxwell3d); virtual ~MacroEngine(); // Store the uploaded macro code to compile them when they're called. diff --git a/src/video_core/macro/macro_hle.cpp b/src/video_core/macro/macro_hle.cpp index 887f40310..1f1348df3 100644 --- a/src/video_core/macro/macro_hle.cpp +++ b/src/video_core/macro/macro_hle.cpp @@ -2,7 +2,7 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. -#include +#include #include #include "video_core/engines/maxwell_3d.h" #include "video_core/macro/macro_hle.h" @@ -78,22 +78,22 @@ static void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d, maxwell3d.CallMethodFromMME(0x8e5, 0x0); } -static const std::unordered_map hle_funcs{ - {0x771BB18C62444DA0, &HLE_771BB18C62444DA0}, - {0x0D61FC9FAAC9FCAD, &HLE_0D61FC9FAAC9FCAD}, - {0x0217920100488FF7, &HLE_0217920100488FF7}, +static const std::array, 3> hle_funcs{ + std::make_pair(0x771BB18C62444DA0, &HLE_771BB18C62444DA0), + std::make_pair(0x0D61FC9FAAC9FCAD, &HLE_0D61FC9FAAC9FCAD), + std::make_pair(0x0217920100488FF7, &HLE_0217920100488FF7), }; HLEMacro::HLEMacro(Engines::Maxwell3D& maxwell3d) : maxwell3d(maxwell3d) {} HLEMacro::~HLEMacro() = default; std::optional> HLEMacro::GetHLEProgram(u64 hash) const { - auto it = hle_funcs.find(hash); - if (it != hle_funcs.end()) { - return std::make_unique(maxwell3d, it->second); - } else { - return {}; + const auto it = std::find_if(hle_funcs.begin(), hle_funcs.end(), + [hash](auto& pair) { return pair.first == hash; }); + if (it == hle_funcs.end()) { + return std::nullopt; } + return std::make_unique(maxwell3d, it->second); } HLEMacroImpl::~HLEMacroImpl() = default; diff --git a/src/video_core/macro/macro_hle.h b/src/video_core/macro/macro_hle.h index de7f43dc4..7cd492a8f 100644 --- a/src/video_core/macro/macro_hle.h +++ b/src/video_core/macro/macro_hle.h @@ -20,7 +20,7 @@ using HLEFunction = void (*)(Engines::Maxwell3D& maxwell3d, const std::vector> GetHLEProgram(u64 hash) const; -- cgit v1.2.3 From 52340e94ac5a64572643f01a23316ad492a40f66 Mon Sep 17 00:00:00 2001 From: David Marcec Date: Fri, 5 Jun 2020 14:00:00 +1000 Subject: clear mme draw mode We already draw, so we can clear it --- src/video_core/macro/macro_hle.cpp | 3 +++ 1 file changed, 3 insertions(+) (limited to 'src/video_core/macro') diff --git a/src/video_core/macro/macro_hle.cpp b/src/video_core/macro/macro_hle.cpp index 1f1348df3..689533f6a 100644 --- a/src/video_core/macro/macro_hle.cpp +++ b/src/video_core/macro/macro_hle.cpp @@ -29,6 +29,7 @@ static void HLE_771BB18C62444DA0(Engines::Maxwell3D& maxwell3d, } 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, @@ -47,6 +48,7 @@ static void HLE_0D61FC9FAAC9FCAD(Engines::Maxwell3D& maxwell3d, } 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, @@ -76,6 +78,7 @@ static void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d, maxwell3d.CallMethodFromMME(0x8e3, 0x640); maxwell3d.CallMethodFromMME(0x8e4, 0x0); maxwell3d.CallMethodFromMME(0x8e5, 0x0); + maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined; } static const std::array, 3> hle_funcs{ -- cgit v1.2.3 From f5e2aec4220ee2b72ec2986e0e60625897b2fd44 Mon Sep 17 00:00:00 2001 From: David Marcec Date: Wed, 24 Jun 2020 12:18:33 +1000 Subject: addressed issues --- src/video_core/macro/macro_hle.cpp | 10 ++++++---- src/video_core/macro/macro_hle.h | 1 + 2 files changed, 7 insertions(+), 4 deletions(-) (limited to 'src/video_core/macro') diff --git a/src/video_core/macro/macro_hle.cpp b/src/video_core/macro/macro_hle.cpp index 689533f6a..410f99018 100644 --- a/src/video_core/macro/macro_hle.cpp +++ b/src/video_core/macro/macro_hle.cpp @@ -10,6 +10,7 @@ namespace Tegra { +namespace { // HLE'd functions static void HLE_771BB18C62444DA0(Engines::Maxwell3D& maxwell3d, const std::vector& parameters) { @@ -80,19 +81,20 @@ static void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d, maxwell3d.CallMethodFromMME(0x8e5, 0x0); maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined; } +} // namespace -static const std::array, 3> hle_funcs{ +constexpr std::array, 3> hle_funcs{{ std::make_pair(0x771BB18C62444DA0, &HLE_771BB18C62444DA0), std::make_pair(0x0D61FC9FAAC9FCAD, &HLE_0D61FC9FAAC9FCAD), std::make_pair(0x0217920100488FF7, &HLE_0217920100488FF7), -}; +}}; HLEMacro::HLEMacro(Engines::Maxwell3D& maxwell3d) : maxwell3d(maxwell3d) {} HLEMacro::~HLEMacro() = default; std::optional> HLEMacro::GetHLEProgram(u64 hash) const { - const auto it = std::find_if(hle_funcs.begin(), hle_funcs.end(), - [hash](auto& pair) { return pair.first == hash; }); + 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; } diff --git a/src/video_core/macro/macro_hle.h b/src/video_core/macro/macro_hle.h index 7cd492a8f..37af875a0 100644 --- a/src/video_core/macro/macro_hle.h +++ b/src/video_core/macro/macro_hle.h @@ -22,6 +22,7 @@ class HLEMacro { public: explicit HLEMacro(Engines::Maxwell3D& maxwell3d); ~HLEMacro(); + std::optional> GetHLEProgram(u64 hash) const; private: -- cgit v1.2.3 From 7c970132b5dd6eaa40f114355e0125091ceb8142 Mon Sep 17 00:00:00 2001 From: David Date: Tue, 30 Jun 2020 15:32:24 +1000 Subject: macro: Add support for "middle methods" on the code cache (#4112) Macro code is just uploaded sequentially from a starting address, however that does not mean the entry point for the macro is at that address. This PR adds preliminary support for executing macros in the middle of our cached code.--- src/video_core/macro/macro.cpp | 35 +++++++++++++++++++++++++++-------- 1 file changed, 27 insertions(+), 8 deletions(-) (limited to 'src/video_core/macro') diff --git a/src/video_core/macro/macro.cpp b/src/video_core/macro/macro.cpp index ef7dad349..a50e7b4e0 100644 --- a/src/video_core/macro/macro.cpp +++ b/src/video_core/macro/macro.cpp @@ -2,6 +2,7 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. +#include #include #include "common/assert.h" #include "common/logging/log.h" @@ -35,22 +36,40 @@ void MacroEngine::Execute(Engines::Maxwell3D& maxwell3d, u32 method, } } else { // Macro not compiled, check if it's uploaded and if so, compile it - auto macro_code = uploaded_macro_code.find(method); + std::optional mid_method = std::nullopt; + const auto macro_code = uploaded_macro_code.find(method); if (macro_code == uploaded_macro_code.end()) { - UNREACHABLE_MSG("Macro 0x{0:x} was not uploaded", method); - return; + 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()) { + UNREACHABLE_MSG("Macro 0x{0:x} was not uploaded", method); + return; + } } auto& cache_info = macro_cache[method]; - cache_info.hash = boost::hash_value(macro_code->second); - cache_info.lle_program = Compile(macro_code->second); + + if (!mid_method.has_value()) { + cache_info.lle_program = Compile(macro_code->second); + cache_info.hash = boost::hash_value(macro_code->second); + } else { + 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)); + cache_info.hash = boost::hash_value(code); + cache_info.lle_program = Compile(code); + } 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); -- cgit v1.2.3