diff options
Diffstat (limited to 'src')
54 files changed, 2388 insertions, 765 deletions
diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt index 264dff546..24b7a083c 100644 --- a/src/common/CMakeLists.txt +++ b/src/common/CMakeLists.txt @@ -171,10 +171,12 @@ if(ARCHITECTURE_x86_64) PRIVATE x64/cpu_detect.cpp x64/cpu_detect.h + x64/xbyak_abi.h + x64/xbyak_util.h ) endif() create_target_directory_groups(common) target_link_libraries(common PUBLIC Boost::boost fmt::fmt microprofile) -target_link_libraries(common PRIVATE lz4::lz4 zstd::zstd) +target_link_libraries(common PRIVATE lz4::lz4 zstd::zstd xbyak) diff --git a/src/common/x64/xbyak_abi.h b/src/common/x64/xbyak_abi.h new file mode 100644 index 000000000..794da8a52 --- /dev/null +++ b/src/common/x64/xbyak_abi.h @@ -0,0 +1,266 @@ +// Copyright 2016 Citra Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <bitset> +#include <initializer_list> +#include <xbyak.h> +#include "common/assert.h" + +namespace Common::X64 { + +inline int RegToIndex(const Xbyak::Reg& reg) { + using Kind = Xbyak::Reg::Kind; + ASSERT_MSG((reg.getKind() & (Kind::REG | Kind::XMM)) != 0, + "RegSet only support GPRs and XMM registers."); + ASSERT_MSG(reg.getIdx() < 16, "RegSet only supports XXM0-15."); + return reg.getIdx() + (reg.getKind() == Kind::REG ? 0 : 16); +} + +inline Xbyak::Reg64 IndexToReg64(int reg_index) { + ASSERT(reg_index < 16); + return Xbyak::Reg64(reg_index); +} + +inline Xbyak::Xmm IndexToXmm(int reg_index) { + ASSERT(reg_index >= 16 && reg_index < 32); + return Xbyak::Xmm(reg_index - 16); +} + +inline Xbyak::Reg IndexToReg(int reg_index) { + if (reg_index < 16) { + return IndexToReg64(reg_index); + } else { + return IndexToXmm(reg_index); + } +} + +inline std::bitset<32> BuildRegSet(std::initializer_list<Xbyak::Reg> regs) { + std::bitset<32> bits; + for (const Xbyak::Reg& reg : regs) { + bits[RegToIndex(reg)] = true; + } + return bits; +} + +const std::bitset<32> ABI_ALL_GPRS(0x0000FFFF); +const std::bitset<32> ABI_ALL_XMMS(0xFFFF0000); + +#ifdef _WIN32 + +// Microsoft x64 ABI +const Xbyak::Reg ABI_RETURN = Xbyak::util::rax; +const Xbyak::Reg ABI_PARAM1 = Xbyak::util::rcx; +const Xbyak::Reg ABI_PARAM2 = Xbyak::util::rdx; +const Xbyak::Reg ABI_PARAM3 = Xbyak::util::r8; +const Xbyak::Reg ABI_PARAM4 = Xbyak::util::r9; + +const std::bitset<32> ABI_ALL_CALLER_SAVED = BuildRegSet({ + // GPRs + Xbyak::util::rcx, + Xbyak::util::rdx, + Xbyak::util::r8, + Xbyak::util::r9, + Xbyak::util::r10, + Xbyak::util::r11, + // XMMs + Xbyak::util::xmm0, + Xbyak::util::xmm1, + Xbyak::util::xmm2, + Xbyak::util::xmm3, + Xbyak::util::xmm4, + Xbyak::util::xmm5, +}); + +const std::bitset<32> ABI_ALL_CALLEE_SAVED = BuildRegSet({ + // GPRs + Xbyak::util::rbx, + Xbyak::util::rsi, + Xbyak::util::rdi, + Xbyak::util::rbp, + Xbyak::util::r12, + Xbyak::util::r13, + Xbyak::util::r14, + Xbyak::util::r15, + // XMMs + Xbyak::util::xmm6, + Xbyak::util::xmm7, + Xbyak::util::xmm8, + Xbyak::util::xmm9, + Xbyak::util::xmm10, + Xbyak::util::xmm11, + Xbyak::util::xmm12, + Xbyak::util::xmm13, + Xbyak::util::xmm14, + Xbyak::util::xmm15, +}); + +constexpr size_t ABI_SHADOW_SPACE = 0x20; + +#else + +// System V x86-64 ABI +const Xbyak::Reg ABI_RETURN = Xbyak::util::rax; +const Xbyak::Reg ABI_PARAM1 = Xbyak::util::rdi; +const Xbyak::Reg ABI_PARAM2 = Xbyak::util::rsi; +const Xbyak::Reg ABI_PARAM3 = Xbyak::util::rdx; +const Xbyak::Reg ABI_PARAM4 = Xbyak::util::rcx; + +const std::bitset<32> ABI_ALL_CALLER_SAVED = BuildRegSet({ + // GPRs + Xbyak::util::rcx, + Xbyak::util::rdx, + Xbyak::util::rdi, + Xbyak::util::rsi, + Xbyak::util::r8, + Xbyak::util::r9, + Xbyak::util::r10, + Xbyak::util::r11, + // XMMs + Xbyak::util::xmm0, + Xbyak::util::xmm1, + Xbyak::util::xmm2, + Xbyak::util::xmm3, + Xbyak::util::xmm4, + Xbyak::util::xmm5, + Xbyak::util::xmm6, + Xbyak::util::xmm7, + Xbyak::util::xmm8, + Xbyak::util::xmm9, + Xbyak::util::xmm10, + Xbyak::util::xmm11, + Xbyak::util::xmm12, + Xbyak::util::xmm13, + Xbyak::util::xmm14, + Xbyak::util::xmm15, +}); + +const std::bitset<32> ABI_ALL_CALLEE_SAVED = BuildRegSet({ + // GPRs + Xbyak::util::rbx, + Xbyak::util::rbp, + Xbyak::util::r12, + Xbyak::util::r13, + Xbyak::util::r14, + Xbyak::util::r15, +}); + +constexpr size_t ABI_SHADOW_SPACE = 0; + +#endif + +inline void ABI_CalculateFrameSize(std::bitset<32> regs, size_t rsp_alignment, + size_t needed_frame_size, s32* out_subtraction, + s32* out_xmm_offset) { + const auto count = (regs & ABI_ALL_GPRS).count(); + rsp_alignment -= count * 8; + size_t subtraction = 0; + const auto xmm_count = (regs & ABI_ALL_XMMS).count(); + if (xmm_count) { + // If we have any XMMs to save, we must align the stack here. + subtraction = rsp_alignment & 0xF; + } + subtraction += 0x10 * xmm_count; + size_t xmm_base_subtraction = subtraction; + subtraction += needed_frame_size; + subtraction += ABI_SHADOW_SPACE; + // Final alignment. + rsp_alignment -= subtraction; + subtraction += rsp_alignment & 0xF; + + *out_subtraction = (s32)subtraction; + *out_xmm_offset = (s32)(subtraction - xmm_base_subtraction); +} + +inline size_t ABI_PushRegistersAndAdjustStack(Xbyak::CodeGenerator& code, std::bitset<32> regs, + size_t rsp_alignment, size_t needed_frame_size = 0) { + s32 subtraction, xmm_offset; + ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size, &subtraction, &xmm_offset); + for (std::size_t i = 0; i < regs.size(); ++i) { + if (regs[i] && ABI_ALL_GPRS[i]) { + code.push(IndexToReg64(static_cast<int>(i))); + } + } + if (subtraction != 0) { + code.sub(code.rsp, subtraction); + } + + for (int i = 0; i < regs.count(); i++) { + if (regs.test(i) & ABI_ALL_GPRS.test(i)) { + code.push(IndexToReg64(i)); + } + } + + for (std::size_t i = 0; i < regs.size(); ++i) { + if (regs[i] && ABI_ALL_XMMS[i]) { + code.movaps(code.xword[code.rsp + xmm_offset], IndexToXmm(static_cast<int>(i))); + xmm_offset += 0x10; + } + } + + return ABI_SHADOW_SPACE; +} + +inline void ABI_PopRegistersAndAdjustStack(Xbyak::CodeGenerator& code, std::bitset<32> regs, + size_t rsp_alignment, size_t needed_frame_size = 0) { + s32 subtraction, xmm_offset; + ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size, &subtraction, &xmm_offset); + + for (std::size_t i = 0; i < regs.size(); ++i) { + if (regs[i] && ABI_ALL_XMMS[i]) { + code.movaps(IndexToXmm(static_cast<int>(i)), code.xword[code.rsp + xmm_offset]); + xmm_offset += 0x10; + } + } + + if (subtraction != 0) { + code.add(code.rsp, subtraction); + } + + // GPRs need to be popped in reverse order + for (int i = 15; i >= 0; i--) { + if (regs[i]) { + code.pop(IndexToReg64(i)); + } + } +} + +inline size_t ABI_PushRegistersAndAdjustStackGPS(Xbyak::CodeGenerator& code, std::bitset<32> regs, + size_t rsp_alignment, + size_t needed_frame_size = 0) { + s32 subtraction, xmm_offset; + ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size, &subtraction, &xmm_offset); + + for (std::size_t i = 0; i < regs.size(); ++i) { + if (regs[i] && ABI_ALL_GPRS[i]) { + code.push(IndexToReg64(static_cast<int>(i))); + } + } + + if (subtraction != 0) { + code.sub(code.rsp, subtraction); + } + + return ABI_SHADOW_SPACE; +} + +inline void ABI_PopRegistersAndAdjustStackGPS(Xbyak::CodeGenerator& code, std::bitset<32> regs, + size_t rsp_alignment, size_t needed_frame_size = 0) { + s32 subtraction, xmm_offset; + ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size, &subtraction, &xmm_offset); + + if (subtraction != 0) { + code.add(code.rsp, subtraction); + } + + // GPRs need to be popped in reverse order + for (int i = 15; i >= 0; i--) { + if (regs[i]) { + code.pop(IndexToReg64(i)); + } + } +} + +} // namespace Common::X64 diff --git a/src/common/x64/xbyak_util.h b/src/common/x64/xbyak_util.h new file mode 100644 index 000000000..df17f8cbe --- /dev/null +++ b/src/common/x64/xbyak_util.h @@ -0,0 +1,47 @@ +// Copyright 2016 Citra Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <type_traits> +#include <xbyak.h> +#include "common/x64/xbyak_abi.h" + +namespace Common::X64 { + +// Constants for use with cmpps/cmpss +enum { + CMP_EQ = 0, + CMP_LT = 1, + CMP_LE = 2, + CMP_UNORD = 3, + CMP_NEQ = 4, + CMP_NLT = 5, + CMP_NLE = 6, + CMP_ORD = 7, +}; + +constexpr bool IsWithin2G(uintptr_t ref, uintptr_t target) { + const u64 distance = target - (ref + 5); + return !(distance >= 0x8000'0000ULL && distance <= ~0x8000'0000ULL); +} + +inline bool IsWithin2G(const Xbyak::CodeGenerator& code, uintptr_t target) { + return IsWithin2G(reinterpret_cast<uintptr_t>(code.getCurr()), target); +} + +template <typename T> +inline void CallFarFunction(Xbyak::CodeGenerator& code, const T f) { + static_assert(std::is_pointer_v<T>, "Argument must be a (function) pointer."); + size_t addr = reinterpret_cast<size_t>(f); + if (IsWithin2G(code, addr)) { + code.call(f); + } else { + // ABI_RETURN is a safe temp register to use before a call + code.mov(ABI_RETURN, addr); + code.call(ABI_RETURN); + } +} + +} // namespace Common::X64 diff --git a/src/core/hle/kernel/readable_event.cpp b/src/core/hle/kernel/readable_event.cpp index 00860fcbd..ef5e19e63 100644 --- a/src/core/hle/kernel/readable_event.cpp +++ b/src/core/hle/kernel/readable_event.cpp @@ -38,7 +38,7 @@ void ReadableEvent::Clear() { ResultCode ReadableEvent::Reset() { if (!is_signaled) { - LOG_ERROR(Kernel, "Handle is not signaled! object_id={}, object_type={}, object_name={}", + LOG_TRACE(Kernel, "Handle is not signaled! object_id={}, object_type={}, object_name={}", GetObjectId(), GetTypeName(), GetName()); return ERR_INVALID_STATE; } diff --git a/src/core/hle/service/hid/hid.cpp b/src/core/hle/service/hid/hid.cpp index c84cb1483..72a050de2 100644 --- a/src/core/hle/service/hid/hid.cpp +++ b/src/core/hle/service/hid/hid.cpp @@ -161,7 +161,7 @@ Hid::Hid(Core::System& system) : ServiceFramework("hid"), system(system) { {40, nullptr, "AcquireXpadIdEventHandle"}, {41, nullptr, "ReleaseXpadIdEventHandle"}, {51, &Hid::ActivateXpad, "ActivateXpad"}, - {55, nullptr, "GetXpadIds"}, + {55, &Hid::GetXpadIDs, "GetXpadIds"}, {56, nullptr, "ActivateJoyXpad"}, {58, nullptr, "GetJoyXpadLifoHandle"}, {59, nullptr, "GetJoyXpadIds"}, @@ -319,6 +319,17 @@ void Hid::ActivateXpad(Kernel::HLERequestContext& ctx) { rb.Push(RESULT_SUCCESS); } +void Hid::GetXpadIDs(Kernel::HLERequestContext& ctx) { + IPC::RequestParser rp{ctx}; + const auto applet_resource_user_id{rp.Pop<u64>()}; + + LOG_DEBUG(Service_HID, "(STUBBED) called, applet_resource_user_id={}", applet_resource_user_id); + + IPC::ResponseBuilder rb{ctx, 3}; + rb.Push(RESULT_SUCCESS); + rb.Push(0); +} + void Hid::ActivateDebugPad(Kernel::HLERequestContext& ctx) { IPC::RequestParser rp{ctx}; const auto applet_resource_user_id{rp.Pop<u64>()}; diff --git a/src/core/hle/service/hid/hid.h b/src/core/hle/service/hid/hid.h index c8ed4ad8b..d481a75f8 100644 --- a/src/core/hle/service/hid/hid.h +++ b/src/core/hle/service/hid/hid.h @@ -86,6 +86,7 @@ public: private: void CreateAppletResource(Kernel::HLERequestContext& ctx); void ActivateXpad(Kernel::HLERequestContext& ctx); + void GetXpadIDs(Kernel::HLERequestContext& ctx); void ActivateDebugPad(Kernel::HLERequestContext& ctx); void ActivateTouchScreen(Kernel::HLERequestContext& ctx); void ActivateMouse(Kernel::HLERequestContext& ctx); diff --git a/src/core/settings.h b/src/core/settings.h index 78eb33737..36cd66fd4 100644 --- a/src/core/settings.h +++ b/src/core/settings.h @@ -474,6 +474,7 @@ struct Values { bool reporting_services; bool quest_flag; bool disable_cpu_opt; + bool disable_macro_jit; // BCAT std::string bcat_backend; diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt index f00c71dae..2bf8d68ce 100644 --- a/src/video_core/CMakeLists.txt +++ b/src/video_core/CMakeLists.txt @@ -25,6 +25,12 @@ add_library(video_core STATIC engines/shader_bytecode.h engines/shader_header.h engines/shader_type.h + macro/macro.cpp + macro/macro.h + macro/macro_interpreter.cpp + macro/macro_interpreter.h + macro/macro_jit_x64.cpp + macro/macro_jit_x64.h fence_manager.h gpu.cpp gpu.h @@ -36,8 +42,6 @@ add_library(video_core STATIC gpu_thread.h guest_driver.cpp guest_driver.h - macro_interpreter.cpp - macro_interpreter.h memory_manager.cpp memory_manager.h morton.cpp @@ -229,7 +233,7 @@ endif() create_target_directory_groups(video_core) target_link_libraries(video_core PUBLIC common core) -target_link_libraries(video_core PRIVATE glad) +target_link_libraries(video_core PRIVATE glad xbyak) if (ENABLE_VULKAN) target_include_directories(video_core PRIVATE sirit ../../externals/Vulkan-Headers/include) diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index d9a4a1b4d..b88fce2cd 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -56,24 +56,28 @@ public: if (use_fast_cbuf || size < max_stream_size) { if (!is_written && !IsRegionWritten(cpu_addr, cpu_addr + size - 1)) { auto& memory_manager = system.GPU().MemoryManager(); + const bool is_granular = memory_manager.IsGranularRange(gpu_addr, size); if (use_fast_cbuf) { - if (memory_manager.IsGranularRange(gpu_addr, size)) { - const auto host_ptr = memory_manager.GetPointer(gpu_addr); - return ConstBufferUpload(host_ptr, size); + u8* dest; + if (is_granular) { + dest = memory_manager.GetPointer(gpu_addr); } else { staging_buffer.resize(size); - memory_manager.ReadBlockUnsafe(gpu_addr, staging_buffer.data(), size); - return ConstBufferUpload(staging_buffer.data(), size); + dest = staging_buffer.data(); + memory_manager.ReadBlockUnsafe(gpu_addr, dest, size); } + return ConstBufferUpload(dest, size); + } + if (is_granular) { + u8* const host_ptr = memory_manager.GetPointer(gpu_addr); + return StreamBufferUpload(size, alignment, [host_ptr, size](u8* dest) { + std::memcpy(dest, host_ptr, size); + }); } else { - if (memory_manager.IsGranularRange(gpu_addr, size)) { - const auto host_ptr = memory_manager.GetPointer(gpu_addr); - return StreamBufferUpload(host_ptr, size, alignment); - } else { - staging_buffer.resize(size); - memory_manager.ReadBlockUnsafe(gpu_addr, staging_buffer.data(), size); - return StreamBufferUpload(staging_buffer.data(), size, alignment); - } + return StreamBufferUpload( + size, alignment, [&memory_manager, gpu_addr, size](u8* dest) { + memory_manager.ReadBlockUnsafe(gpu_addr, dest, size); + }); } } } @@ -101,7 +105,9 @@ public: BufferInfo UploadHostMemory(const void* raw_pointer, std::size_t size, std::size_t alignment = 4) { std::lock_guard lock{mutex}; - return StreamBufferUpload(raw_pointer, size, alignment); + return StreamBufferUpload(size, alignment, [raw_pointer, size](u8* dest) { + std::memcpy(dest, raw_pointer, size); + }); } void Map(std::size_t max_size) { @@ -424,11 +430,11 @@ private: map->MarkAsModified(false, 0); } - BufferInfo StreamBufferUpload(const void* raw_pointer, std::size_t size, - std::size_t alignment) { + template <typename Callable> + BufferInfo StreamBufferUpload(std::size_t size, std::size_t alignment, Callable&& callable) { AlignBuffer(alignment); const std::size_t uploaded_offset = buffer_offset; - std::memcpy(buffer_ptr, raw_pointer, size); + callable(buffer_ptr); buffer_ptr += size; buffer_offset += size; diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index 024c9e43b..e46b153f9 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -25,9 +25,8 @@ constexpr u32 MacroRegistersStart = 0xE00; Maxwell3D::Maxwell3D(Core::System& system, VideoCore::RasterizerInterface& rasterizer, MemoryManager& memory_manager) : system{system}, rasterizer{rasterizer}, memory_manager{memory_manager}, - macro_interpreter{*this}, upload_state{memory_manager, regs.upload} { + macro_engine{GetMacroEngine(*this)}, upload_state{memory_manager, regs.upload} { dirty.flags.flip(); - InitializeRegisterDefaults(); } @@ -106,7 +105,11 @@ void Maxwell3D::InitializeRegisterDefaults() { regs.rasterize_enable = 1; regs.rt_separate_frag_data = 1; regs.framebuffer_srgb = 1; + regs.line_width_aliased = 1.0f; + regs.line_width_smooth = 1.0f; regs.front_face = Maxwell3D::Regs::FrontFace::ClockWise; + regs.polygon_mode_back = Maxwell3D::Regs::PolygonMode::Fill; + regs.polygon_mode_front = Maxwell3D::Regs::PolygonMode::Fill; shadow_state = regs; @@ -116,7 +119,7 @@ void Maxwell3D::InitializeRegisterDefaults() { mme_inline[MAXWELL3D_REG_INDEX(index_array.count)] = true; } -void Maxwell3D::CallMacroMethod(u32 method, std::size_t num_parameters, const u32* parameters) { +void Maxwell3D::CallMacroMethod(u32 method, const std::vector<u32>& parameters) { // Reset the current macro. executing_macro = 0; @@ -125,7 +128,7 @@ void Maxwell3D::CallMacroMethod(u32 method, std::size_t num_parameters, const u3 ((method - MacroRegistersStart) >> 1) % static_cast<u32>(macro_positions.size()); // Execute the current macro. - macro_interpreter.Execute(macro_positions[entry], num_parameters, parameters); + macro_engine->Execute(macro_positions[entry], parameters); if (mme_draw.current_mode != MMEDrawMode::Undefined) { FlushMMEInlineDraw(); } @@ -161,7 +164,7 @@ void Maxwell3D::CallMethod(u32 method, u32 method_argument, bool is_last_call) { // Call the macro when there are no more parameters in the command buffer if (is_last_call) { - CallMacroMethod(executing_macro, macro_params.size(), macro_params.data()); + CallMacroMethod(executing_macro, macro_params); macro_params.clear(); } return; @@ -197,7 +200,7 @@ void Maxwell3D::CallMethod(u32 method, u32 method_argument, bool is_last_call) { break; } case MAXWELL3D_REG_INDEX(macros.data): { - ProcessMacroUpload(arg); + macro_engine->AddCode(regs.macros.upload_address, arg); break; } case MAXWELL3D_REG_INDEX(macros.bind): { @@ -306,7 +309,7 @@ void Maxwell3D::CallMultiMethod(u32 method, const u32* base_start, u32 amount, // Call the macro when there are no more parameters in the command buffer if (amount == methods_pending) { - CallMacroMethod(executing_macro, macro_params.size(), macro_params.data()); + CallMacroMethod(executing_macro, macro_params); macro_params.clear(); } return; @@ -420,9 +423,7 @@ void Maxwell3D::FlushMMEInlineDraw() { } void Maxwell3D::ProcessMacroUpload(u32 data) { - ASSERT_MSG(regs.macros.upload_address < macro_memory.size(), - "upload_address exceeded macro_memory size!"); - macro_memory[regs.macros.upload_address++] = data; + macro_engine->AddCode(regs.macros.upload_address++, data); } void Maxwell3D::ProcessMacroBind(u32 data) { @@ -457,8 +458,9 @@ void Maxwell3D::StampQueryResult(u64 payload, bool long_query) { void Maxwell3D::ProcessQueryGet() { // TODO(Subv): Support the other query units. - ASSERT_MSG(regs.query.query_get.unit == Regs::QueryUnit::Crop, - "Units other than CROP are unimplemented"); + if (regs.query.query_get.unit != Regs::QueryUnit::Crop) { + LOG_DEBUG(HW_GPU, "Units other than CROP are unimplemented"); + } switch (regs.query.query_get.operation) { case Regs::QueryOperation::Release: @@ -534,8 +536,8 @@ void Maxwell3D::ProcessCounterReset() { rasterizer.ResetCounter(QueryType::SamplesPassed); break; default: - LOG_WARNING(Render_OpenGL, "Unimplemented counter reset={}", - static_cast<int>(regs.counter_reset)); + LOG_DEBUG(Render_OpenGL, "Unimplemented counter reset={}", + static_cast<int>(regs.counter_reset)); break; } } @@ -592,8 +594,8 @@ std::optional<u64> Maxwell3D::GetQueryResult() { system.GPU().GetTicks()); return {}; default: - UNIMPLEMENTED_MSG("Unimplemented query select type {}", - static_cast<u32>(regs.query.query_get.select.Value())); + LOG_DEBUG(HW_GPU, "Unimplemented query select type {}", + static_cast<u32>(regs.query.query_get.select.Value())); return 1; } } diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h index 05dd6b39b..79fc9bbea 100644 --- a/src/video_core/engines/maxwell_3d.h +++ b/src/video_core/engines/maxwell_3d.h @@ -23,7 +23,7 @@ #include "video_core/engines/engine_upload.h" #include "video_core/engines/shader_type.h" #include "video_core/gpu.h" -#include "video_core/macro_interpreter.h" +#include "video_core/macro/macro.h" #include "video_core/textures/texture.h" namespace Core { @@ -598,6 +598,7 @@ public: BitField<4, 3, u32> block_height; BitField<8, 3, u32> block_depth; BitField<12, 1, InvMemoryLayout> type; + BitField<16, 1, u32> is_3d; } memory_layout; union { BitField<0, 16, u32> layers; @@ -1411,15 +1412,6 @@ public: const VideoCore::GuestDriverProfile& AccessGuestDriverProfile() const override; - /// Memory for macro code - it's undetermined how big this is, however 1MB is much larger than - /// we've seen used. - using MacroMemory = std::array<u32, 0x40000>; - - /// Gets a reference to macro memory. - const MacroMemory& GetMacroMemory() const { - return macro_memory; - } - bool ShouldExecute() const { return execute_on; } @@ -1468,16 +1460,13 @@ private: std::array<bool, Regs::NUM_REGS> mme_inline{}; - /// Memory for macro code - MacroMemory macro_memory; - /// Macro method that is currently being executed / being fed parameters. u32 executing_macro = 0; /// Parameters that have been submitted to the macro call so far. std::vector<u32> macro_params; /// Interpreter for the macro codes uploaded to the GPU. - MacroInterpreter macro_interpreter; + std::unique_ptr<MacroEngine> macro_engine; static constexpr u32 null_cb_data = 0xFFFFFFFF; struct { @@ -1506,7 +1495,7 @@ private: * @param num_parameters Number of arguments * @param parameters Arguments to the method call */ - void CallMacroMethod(u32 method, std::size_t num_parameters, const u32* parameters); + void CallMacroMethod(u32 method, const std::vector<u32>& parameters); /// Handles writes to the macro uploading register. void ProcessMacroUpload(u32 data); diff --git a/src/video_core/macro/macro.cpp b/src/video_core/macro/macro.cpp new file mode 100644 index 000000000..89077a2d8 --- /dev/null +++ b/src/video_core/macro/macro.cpp @@ -0,0 +1,45 @@ +// Copyright 2020 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include "common/assert.h" +#include "common/logging/log.h" +#include "core/settings.h" +#include "video_core/macro/macro.h" +#include "video_core/macro/macro_interpreter.h" +#include "video_core/macro/macro_jit_x64.h" + +namespace Tegra { + +void MacroEngine::AddCode(u32 method, u32 data) { + uploaded_macro_code[method].push_back(data); +} + +void MacroEngine::Execute(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); + } else { + // Macro not compiled, check if it's uploaded and if so, compile it + 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; + } + macro_cache[method] = Compile(macro_code->second); + macro_cache[method]->Execute(parameters, method); + } +} + +std::unique_ptr<MacroEngine> GetMacroEngine(Engines::Maxwell3D& maxwell3d) { + if (Settings::values.disable_macro_jit) { + return std::make_unique<MacroInterpreter>(maxwell3d); + } +#ifdef ARCHITECTURE_x86_64 + return std::make_unique<MacroJITx64>(maxwell3d); +#else + return std::make_unique<MacroInterpreter>(maxwell3d); +#endif +} + +} // namespace Tegra diff --git a/src/video_core/macro/macro.h b/src/video_core/macro/macro.h new file mode 100644 index 000000000..b76ed891f --- /dev/null +++ b/src/video_core/macro/macro.h @@ -0,0 +1,128 @@ +// Copyright 2020 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <memory> +#include <unordered_map> +#include <vector> +#include "common/bit_field.h" +#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 { + ALU = 0, + AddImmediate = 1, + ExtractInsert = 2, + ExtractShiftLeftImmediate = 3, + ExtractShiftLeftRegister = 4, + Read = 5, + Unused = 6, // This operation doesn't seem to be a valid encoding. + Branch = 7, +}; + +enum class ALUOperation : u32 { + Add = 0, + AddWithCarry = 1, + Subtract = 2, + SubtractWithBorrow = 3, + // Operations 4-7 don't seem to be valid encodings. + Xor = 8, + Or = 9, + And = 10, + AndNot = 11, + Nand = 12 +}; + +enum class ResultOperation : u32 { + IgnoreAndFetch = 0, + Move = 1, + MoveAndSetMethod = 2, + FetchAndSend = 3, + MoveAndSend = 4, + FetchAndSetMethod = 5, + MoveAndSetMethodFetchAndSend = 6, + MoveAndSetMethodSend = 7 +}; + +enum class BranchCondition : u32 { + Zero = 0, + NotZero = 1, +}; + +union Opcode { + u32 raw; + BitField<0, 3, Operation> operation; + BitField<4, 3, ResultOperation> result_operation; + BitField<4, 1, BranchCondition> branch_condition; + // If set on a branch, then the branch doesn't have a delay slot. + BitField<5, 1, u32> branch_annul; + BitField<7, 1, u32> is_exit; + BitField<8, 3, u32> dst; + BitField<11, 3, u32> src_a; + BitField<14, 3, u32> src_b; + // The signed immediate overlaps the second source operand and the alu operation. + BitField<14, 18, s32> immediate; + + BitField<17, 5, ALUOperation> alu_operation; + + // Bitfield instructions data + BitField<17, 5, u32> bf_src_bit; + BitField<22, 5, u32> bf_size; + BitField<27, 5, u32> bf_dst_bit; + + u32 GetBitfieldMask() const { + return (1 << bf_size) - 1; + } + + s32 GetBranchTarget() const { + return static_cast<s32>(immediate * sizeof(u32)); + } +}; + +union MethodAddress { + u32 raw; + BitField<0, 12, u32> address; + BitField<12, 6, u32> increment; +}; + +} // namespace Macro + +class CachedMacro { +public: + virtual ~CachedMacro() = default; + /** + * Executes the macro code with the specified input parameters. + * @param code The macro byte code to execute + * @param parameters The parameters of the macro + */ + virtual void Execute(const std::vector<u32>& parameters, u32 method) = 0; +}; + +class MacroEngine { +public: + virtual ~MacroEngine() = default; + + // 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); + +protected: + virtual std::unique_ptr<CachedMacro> Compile(const std::vector<u32>& code) = 0; + +private: + std::unordered_map<u32, std::unique_ptr<CachedMacro>> macro_cache; + std::unordered_map<u32, std::vector<u32>> uploaded_macro_code; +}; + +std::unique_ptr<MacroEngine> GetMacroEngine(Engines::Maxwell3D& maxwell3d); + +} // namespace Tegra diff --git a/src/video_core/macro_interpreter.cpp b/src/video_core/macro/macro_interpreter.cpp index 947364928..5edff27aa 100644 --- a/src/video_core/macro_interpreter.cpp +++ b/src/video_core/macro/macro_interpreter.cpp @@ -1,4 +1,4 @@ -// Copyright 2018 yuzu Emulator Project +// Copyright 2020 yuzu Emulator Project // Licensed under GPLv2 or any later version // Refer to the license.txt file included. @@ -6,109 +6,46 @@ #include "common/logging/log.h" #include "common/microprofile.h" #include "video_core/engines/maxwell_3d.h" -#include "video_core/macro_interpreter.h" +#include "video_core/macro/macro_interpreter.h" MICROPROFILE_DEFINE(MacroInterp, "GPU", "Execute macro interpreter", MP_RGB(128, 128, 192)); namespace Tegra { -namespace { -enum class Operation : u32 { - ALU = 0, - AddImmediate = 1, - ExtractInsert = 2, - ExtractShiftLeftImmediate = 3, - ExtractShiftLeftRegister = 4, - Read = 5, - Unused = 6, // This operation doesn't seem to be a valid encoding. - Branch = 7, -}; -} // Anonymous namespace - -enum class MacroInterpreter::ALUOperation : u32 { - Add = 0, - AddWithCarry = 1, - Subtract = 2, - SubtractWithBorrow = 3, - // Operations 4-7 don't seem to be valid encodings. - Xor = 8, - Or = 9, - And = 10, - AndNot = 11, - Nand = 12 -}; - -enum class MacroInterpreter::ResultOperation : u32 { - IgnoreAndFetch = 0, - Move = 1, - MoveAndSetMethod = 2, - FetchAndSend = 3, - MoveAndSend = 4, - FetchAndSetMethod = 5, - MoveAndSetMethodFetchAndSend = 6, - MoveAndSetMethodSend = 7 -}; - -enum class MacroInterpreter::BranchCondition : u32 { - Zero = 0, - NotZero = 1, -}; - -union MacroInterpreter::Opcode { - u32 raw; - BitField<0, 3, Operation> operation; - BitField<4, 3, ResultOperation> result_operation; - BitField<4, 1, BranchCondition> branch_condition; - // If set on a branch, then the branch doesn't have a delay slot. - BitField<5, 1, u32> branch_annul; - BitField<7, 1, u32> is_exit; - BitField<8, 3, u32> dst; - BitField<11, 3, u32> src_a; - BitField<14, 3, u32> src_b; - // The signed immediate overlaps the second source operand and the alu operation. - BitField<14, 18, s32> immediate; - - BitField<17, 5, ALUOperation> alu_operation; - - // Bitfield instructions data - BitField<17, 5, u32> bf_src_bit; - BitField<22, 5, u32> bf_size; - BitField<27, 5, u32> bf_dst_bit; - - u32 GetBitfieldMask() const { - return (1 << bf_size) - 1; - } +MacroInterpreter::MacroInterpreter(Engines::Maxwell3D& maxwell3d) : maxwell3d(maxwell3d) {} - s32 GetBranchTarget() const { - return static_cast<s32>(immediate * sizeof(u32)); - } -}; +std::unique_ptr<CachedMacro> MacroInterpreter::Compile(const std::vector<u32>& code) { + return std::make_unique<MacroInterpreterImpl>(maxwell3d, code); +} -MacroInterpreter::MacroInterpreter(Engines::Maxwell3D& maxwell3d) : maxwell3d(maxwell3d) {} +MacroInterpreterImpl::MacroInterpreterImpl(Engines::Maxwell3D& maxwell3d, + const std::vector<u32>& code) + : maxwell3d(maxwell3d), code(code) {} -void MacroInterpreter::Execute(u32 offset, std::size_t num_parameters, const u32* parameters) { +void MacroInterpreterImpl::Execute(const std::vector<u32>& parameters, u32 method) { MICROPROFILE_SCOPE(MacroInterp); Reset(); registers[1] = parameters[0]; + num_parameters = parameters.size(); if (num_parameters > parameters_capacity) { parameters_capacity = num_parameters; this->parameters = std::make_unique<u32[]>(num_parameters); } - std::memcpy(this->parameters.get(), parameters, num_parameters * sizeof(u32)); + std::memcpy(this->parameters.get(), parameters.data(), num_parameters * sizeof(u32)); this->num_parameters = num_parameters; // Execute the code until we hit an exit condition. bool keep_executing = true; while (keep_executing) { - keep_executing = Step(offset, false); + keep_executing = Step(false); } // Assert the the macro used all the input parameters ASSERT(next_parameter_index == num_parameters); } -void MacroInterpreter::Reset() { +void MacroInterpreterImpl::Reset() { registers = {}; pc = 0; delayed_pc = {}; @@ -120,10 +57,10 @@ void MacroInterpreter::Reset() { carry_flag = false; } -bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) { +bool MacroInterpreterImpl::Step(bool is_delay_slot) { u32 base_address = pc; - Opcode opcode = GetOpcode(offset); + Macro::Opcode opcode = GetOpcode(); pc += 4; // Update the program counter if we were delayed @@ -134,18 +71,18 @@ bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) { } switch (opcode.operation) { - case Operation::ALU: { + case Macro::Operation::ALU: { u32 result = GetALUResult(opcode.alu_operation, GetRegister(opcode.src_a), GetRegister(opcode.src_b)); ProcessResult(opcode.result_operation, opcode.dst, result); break; } - case Operation::AddImmediate: { + case Macro::Operation::AddImmediate: { ProcessResult(opcode.result_operation, opcode.dst, GetRegister(opcode.src_a) + opcode.immediate); break; } - case Operation::ExtractInsert: { + case Macro::Operation::ExtractInsert: { u32 dst = GetRegister(opcode.src_a); u32 src = GetRegister(opcode.src_b); @@ -155,7 +92,7 @@ bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) { ProcessResult(opcode.result_operation, opcode.dst, dst); break; } - case Operation::ExtractShiftLeftImmediate: { + case Macro::Operation::ExtractShiftLeftImmediate: { u32 dst = GetRegister(opcode.src_a); u32 src = GetRegister(opcode.src_b); @@ -164,7 +101,7 @@ bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) { ProcessResult(opcode.result_operation, opcode.dst, result); break; } - case Operation::ExtractShiftLeftRegister: { + case Macro::Operation::ExtractShiftLeftRegister: { u32 dst = GetRegister(opcode.src_a); u32 src = GetRegister(opcode.src_b); @@ -173,12 +110,12 @@ bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) { ProcessResult(opcode.result_operation, opcode.dst, result); break; } - case Operation::Read: { + case Macro::Operation::Read: { u32 result = Read(GetRegister(opcode.src_a) + opcode.immediate); ProcessResult(opcode.result_operation, opcode.dst, result); break; } - case Operation::Branch: { + case Macro::Operation::Branch: { ASSERT_MSG(!is_delay_slot, "Executing a branch in a delay slot is not valid"); u32 value = GetRegister(opcode.src_a); bool taken = EvaluateBranchCondition(opcode.branch_condition, value); @@ -191,7 +128,7 @@ bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) { delayed_pc = base_address + opcode.GetBranchTarget(); // Execute one more instruction due to the delay slot. - return Step(offset, true); + return Step(true); } break; } @@ -204,51 +141,44 @@ bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) { // cause an exit if it's executed inside a delay slot. if (opcode.is_exit && !is_delay_slot) { // Exit has a delay slot, execute the next instruction - Step(offset, true); + Step(true); return false; } return true; } -MacroInterpreter::Opcode MacroInterpreter::GetOpcode(u32 offset) const { - const auto& macro_memory{maxwell3d.GetMacroMemory()}; - ASSERT((pc % sizeof(u32)) == 0); - ASSERT((pc + offset) < macro_memory.size() * sizeof(u32)); - return {macro_memory[offset + pc / sizeof(u32)]}; -} - -u32 MacroInterpreter::GetALUResult(ALUOperation operation, u32 src_a, u32 src_b) { +u32 MacroInterpreterImpl::GetALUResult(Macro::ALUOperation operation, u32 src_a, u32 src_b) { switch (operation) { - case ALUOperation::Add: { + case Macro::ALUOperation::Add: { const u64 result{static_cast<u64>(src_a) + src_b}; carry_flag = result > 0xffffffff; return static_cast<u32>(result); } - case ALUOperation::AddWithCarry: { + case Macro::ALUOperation::AddWithCarry: { const u64 result{static_cast<u64>(src_a) + src_b + (carry_flag ? 1ULL : 0ULL)}; carry_flag = result > 0xffffffff; return static_cast<u32>(result); } - case ALUOperation::Subtract: { + case Macro::ALUOperation::Subtract: { const u64 result{static_cast<u64>(src_a) - src_b}; carry_flag = result < 0x100000000; return static_cast<u32>(result); } - case ALUOperation::SubtractWithBorrow: { + case Macro::ALUOperation::SubtractWithBorrow: { const u64 result{static_cast<u64>(src_a) - src_b - (carry_flag ? 0ULL : 1ULL)}; carry_flag = result < 0x100000000; return static_cast<u32>(result); } - case ALUOperation::Xor: + case Macro::ALUOperation::Xor: return src_a ^ src_b; - case ALUOperation::Or: + case Macro::ALUOperation::Or: return src_a | src_b; - case ALUOperation::And: + case Macro::ALUOperation::And: return src_a & src_b; - case ALUOperation::AndNot: + case Macro::ALUOperation::AndNot: return src_a & ~src_b; - case ALUOperation::Nand: + case Macro::ALUOperation::Nand: return ~(src_a & src_b); default: @@ -257,43 +187,43 @@ u32 MacroInterpreter::GetALUResult(ALUOperation operation, u32 src_a, u32 src_b) } } -void MacroInterpreter::ProcessResult(ResultOperation operation, u32 reg, u32 result) { +void MacroInterpreterImpl::ProcessResult(Macro::ResultOperation operation, u32 reg, u32 result) { switch (operation) { - case ResultOperation::IgnoreAndFetch: + case Macro::ResultOperation::IgnoreAndFetch: // Fetch parameter and ignore result. SetRegister(reg, FetchParameter()); break; - case ResultOperation::Move: + case Macro::ResultOperation::Move: // Move result. SetRegister(reg, result); break; - case ResultOperation::MoveAndSetMethod: + case Macro::ResultOperation::MoveAndSetMethod: // Move result and use as Method Address. SetRegister(reg, result); SetMethodAddress(result); break; - case ResultOperation::FetchAndSend: + case Macro::ResultOperation::FetchAndSend: // Fetch parameter and send result. SetRegister(reg, FetchParameter()); Send(result); break; - case ResultOperation::MoveAndSend: + case Macro::ResultOperation::MoveAndSend: // Move and send result. SetRegister(reg, result); Send(result); break; - case ResultOperation::FetchAndSetMethod: + case Macro::ResultOperation::FetchAndSetMethod: // Fetch parameter and use result as Method Address. SetRegister(reg, FetchParameter()); SetMethodAddress(result); break; - case ResultOperation::MoveAndSetMethodFetchAndSend: + case Macro::ResultOperation::MoveAndSetMethodFetchAndSend: // Move result and use as Method Address, then fetch and send parameter. SetRegister(reg, result); SetMethodAddress(result); Send(FetchParameter()); break; - case ResultOperation::MoveAndSetMethodSend: + case Macro::ResultOperation::MoveAndSetMethodSend: // Move result and use as Method Address, then send bits 12:17 of result. SetRegister(reg, result); SetMethodAddress(result); @@ -304,16 +234,28 @@ void MacroInterpreter::ProcessResult(ResultOperation operation, u32 reg, u32 res } } -u32 MacroInterpreter::FetchParameter() { - ASSERT(next_parameter_index < num_parameters); - return parameters[next_parameter_index++]; +bool MacroInterpreterImpl::EvaluateBranchCondition(Macro::BranchCondition cond, u32 value) const { + switch (cond) { + case Macro::BranchCondition::Zero: + return value == 0; + case Macro::BranchCondition::NotZero: + return value != 0; + } + UNREACHABLE(); + return true; } -u32 MacroInterpreter::GetRegister(u32 register_id) const { +Macro::Opcode MacroInterpreterImpl::GetOpcode() const { + ASSERT((pc % sizeof(u32)) == 0); + ASSERT(pc < code.size() * sizeof(u32)); + return {code[pc / sizeof(u32)]}; +} + +u32 MacroInterpreterImpl::GetRegister(u32 register_id) const { return registers.at(register_id); } -void MacroInterpreter::SetRegister(u32 register_id, u32 value) { +void MacroInterpreterImpl::SetRegister(u32 register_id, u32 value) { // Register 0 is hardwired as the zero register. // Ensure no writes to it actually occur. if (register_id == 0) { @@ -323,30 +265,24 @@ void MacroInterpreter::SetRegister(u32 register_id, u32 value) { registers.at(register_id) = value; } -void MacroInterpreter::SetMethodAddress(u32 address) { +void MacroInterpreterImpl::SetMethodAddress(u32 address) { method_address.raw = address; } -void MacroInterpreter::Send(u32 value) { +void MacroInterpreterImpl::Send(u32 value) { maxwell3d.CallMethodFromMME(method_address.address, value); // Increment the method address by the method increment. method_address.address.Assign(method_address.address.Value() + method_address.increment.Value()); } -u32 MacroInterpreter::Read(u32 method) const { +u32 MacroInterpreterImpl::Read(u32 method) const { return maxwell3d.GetRegisterValue(method); } -bool MacroInterpreter::EvaluateBranchCondition(BranchCondition cond, u32 value) const { - switch (cond) { - case BranchCondition::Zero: - return value == 0; - case BranchCondition::NotZero: - return value != 0; - } - UNREACHABLE(); - return true; +u32 MacroInterpreterImpl::FetchParameter() { + ASSERT(next_parameter_index < num_parameters); + return parameters[next_parameter_index++]; } } // namespace Tegra diff --git a/src/video_core/macro_interpreter.h b/src/video_core/macro/macro_interpreter.h index 631146d89..90217fc89 100644 --- a/src/video_core/macro_interpreter.h +++ b/src/video_core/macro/macro_interpreter.h @@ -1,44 +1,37 @@ -// Copyright 2018 yuzu Emulator Project +// Copyright 2020 yuzu Emulator Project // Licensed under GPLv2 or any later version // Refer to the license.txt file included. #pragma once - #include <array> #include <optional> - +#include <vector> #include "common/bit_field.h" #include "common/common_types.h" +#include "video_core/macro/macro.h" namespace Tegra { namespace Engines { class Maxwell3D; } -class MacroInterpreter final { +class MacroInterpreter final : public MacroEngine { public: explicit MacroInterpreter(Engines::Maxwell3D& maxwell3d); - /** - * Executes the macro code with the specified input parameters. - * @param offset Offset to start execution at. - * @param parameters The parameters of the macro. - */ - void Execute(u32 offset, std::size_t num_parameters, const u32* parameters); +protected: + std::unique_ptr<CachedMacro> Compile(const std::vector<u32>& code) override; private: - enum class ALUOperation : u32; - enum class BranchCondition : u32; - enum class ResultOperation : u32; - - union Opcode; + Engines::Maxwell3D& maxwell3d; +}; - union MethodAddress { - u32 raw; - BitField<0, 12, u32> address; - BitField<12, 6, u32> increment; - }; +class MacroInterpreterImpl : public CachedMacro { +public: + MacroInterpreterImpl(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& code); + void Execute(const std::vector<u32>& parameters, u32 method) override; +private: /// Resets the execution engine state, zeroing registers, etc. void Reset(); @@ -49,20 +42,20 @@ private: * @param is_delay_slot Whether the current step is being executed due to a delay slot in a * previous instruction. */ - bool Step(u32 offset, bool is_delay_slot); + bool Step(bool is_delay_slot); /// Calculates the result of an ALU operation. src_a OP src_b; - u32 GetALUResult(ALUOperation operation, u32 src_a, u32 src_b); + u32 GetALUResult(Macro::ALUOperation operation, u32 src_a, u32 src_b); /// Performs the result operation on the input result and stores it in the specified register /// (if necessary). - void ProcessResult(ResultOperation operation, u32 reg, u32 result); + void ProcessResult(Macro::ResultOperation operation, u32 reg, u32 result); /// Evaluates the branch condition and returns whether the branch should be taken or not. - bool EvaluateBranchCondition(BranchCondition cond, u32 value) const; + bool EvaluateBranchCondition(Macro::BranchCondition cond, u32 value) const; /// Reads an opcode at the current program counter location. - Opcode GetOpcode(u32 offset) const; + Macro::Opcode GetOpcode() const; /// Returns the specified register's value. Register 0 is hardcoded to always return 0. u32 GetRegister(u32 register_id) const; @@ -89,13 +82,11 @@ private: /// Program counter to execute at after the delay slot is executed. std::optional<u32> delayed_pc; - static constexpr std::size_t NumMacroRegisters = 8; - /// General purpose macro registers. - std::array<u32, NumMacroRegisters> registers = {}; + std::array<u32, Macro::NUM_MACRO_REGISTERS> registers = {}; /// Method address to use for the next Send instruction. - MethodAddress method_address = {}; + Macro::MethodAddress method_address = {}; /// Input parameters of the current macro. std::unique_ptr<u32[]> parameters; @@ -105,5 +96,7 @@ private: u32 next_parameter_index = 0; bool carry_flag = false; + const std::vector<u32>& code; }; + } // namespace Tegra diff --git a/src/video_core/macro/macro_jit_x64.cpp b/src/video_core/macro/macro_jit_x64.cpp new file mode 100644 index 000000000..11c1cc3be --- /dev/null +++ b/src/video_core/macro/macro_jit_x64.cpp @@ -0,0 +1,640 @@ +// Copyright 2020 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include "common/assert.h" +#include "common/logging/log.h" +#include "common/microprofile.h" +#include "common/x64/xbyak_util.h" +#include "video_core/engines/maxwell_3d.h" +#include "video_core/macro/macro_interpreter.h" +#include "video_core/macro/macro_jit_x64.h" + +MICROPROFILE_DEFINE(MacroJitCompile, "GPU", "Compile macro JIT", MP_RGB(173, 255, 47)); +MICROPROFILE_DEFINE(MacroJitExecute, "GPU", "Execute macro JIT", MP_RGB(255, 255, 0)); + +namespace Tegra { +static const Xbyak::Reg64 PARAMETERS = Xbyak::util::r9; +static const Xbyak::Reg64 REGISTERS = Xbyak::util::r10; +static const Xbyak::Reg64 STATE = Xbyak::util::r11; +static const Xbyak::Reg64 NEXT_PARAMETER = Xbyak::util::r12; +static const Xbyak::Reg32 RESULT = Xbyak::util::r13d; +static const Xbyak::Reg64 RESULT_64 = Xbyak::util::r13; +static const Xbyak::Reg32 METHOD_ADDRESS = Xbyak::util::r14d; +static const Xbyak::Reg64 METHOD_ADDRESS_64 = Xbyak::util::r14; +static const Xbyak::Reg64 BRANCH_HOLDER = Xbyak::util::r15; + +static const std::bitset<32> PERSISTENT_REGISTERS = Common::X64::BuildRegSet({ + PARAMETERS, + REGISTERS, + STATE, + NEXT_PARAMETER, + RESULT, + METHOD_ADDRESS, + BRANCH_HOLDER, +}); + +MacroJITx64::MacroJITx64(Engines::Maxwell3D& maxwell3d) : maxwell3d(maxwell3d) {} + +std::unique_ptr<CachedMacro> MacroJITx64::Compile(const std::vector<u32>& code) { + return std::make_unique<MacroJITx64Impl>(maxwell3d, code); +} + +MacroJITx64Impl::MacroJITx64Impl(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& code) + : Xbyak::CodeGenerator(MAX_CODE_SIZE), code(code), maxwell3d(maxwell3d) { + Compile(); +} + +MacroJITx64Impl::~MacroJITx64Impl() = default; + +void MacroJITx64Impl::Execute(const std::vector<u32>& parameters, u32 method) { + MICROPROFILE_SCOPE(MacroJitExecute); + ASSERT_OR_EXECUTE(program != nullptr, { return; }); + JITState state{}; + state.maxwell3d = &maxwell3d; + state.registers = {}; + state.parameters = parameters.data(); + program(&state); +} + +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; + const bool has_zero_register = is_a_zero || is_b_zero; + + Xbyak::Reg64 src_a; + Xbyak::Reg32 src_b; + + if (!optimizer.zero_reg_skip) { + src_a = Compile_GetRegister(opcode.src_a, RESULT_64); + src_b = Compile_GetRegister(opcode.src_b, ebx); + } else { + if (!is_a_zero) { + src_a = Compile_GetRegister(opcode.src_a, RESULT_64); + } + if (!is_b_zero) { + src_b = Compile_GetRegister(opcode.src_b, ebx); + } + } + Xbyak::Label skip_carry{}; + + bool has_emitted = false; + + switch (opcode.alu_operation) { + case Macro::ALUOperation::Add: + if (optimizer.zero_reg_skip) { + if (valid_operation) { + add(src_a, src_b); + } + } else { + add(src_a, src_b); + } + + if (!optimizer.can_skip_carry) { + setc(byte[STATE + offsetof(JITState, carry_flag)]); + } + break; + case Macro::ALUOperation::AddWithCarry: + bt(dword[STATE + offsetof(JITState, carry_flag)], 0); + adc(src_a, src_b); + setc(byte[STATE + offsetof(JITState, carry_flag)]); + break; + case Macro::ALUOperation::Subtract: + if (optimizer.zero_reg_skip) { + if (valid_operation) { + sub(src_a, src_b); + has_emitted = true; + } + } else { + sub(src_a, src_b); + has_emitted = true; + } + if (!optimizer.can_skip_carry && has_emitted) { + setc(byte[STATE + offsetof(JITState, carry_flag)]); + } + break; + case Macro::ALUOperation::SubtractWithBorrow: + bt(dword[STATE + offsetof(JITState, carry_flag)], 0); + sbb(src_a, src_b); + setc(byte[STATE + offsetof(JITState, carry_flag)]); + break; + case Macro::ALUOperation::Xor: + if (optimizer.zero_reg_skip) { + if (valid_operation) { + xor_(src_a, src_b); + } + } else { + xor_(src_a, src_b); + } + break; + case Macro::ALUOperation::Or: + if (optimizer.zero_reg_skip) { + if (valid_operation) { + or_(src_a, src_b); + } + } else { + or_(src_a, src_b); + } + break; + case Macro::ALUOperation::And: + if (optimizer.zero_reg_skip) { + if (!has_zero_register) { + and_(src_a, src_b); + } + } else { + and_(src_a, src_b); + } + break; + case Macro::ALUOperation::AndNot: + if (optimizer.zero_reg_skip) { + if (!is_a_zero) { + not_(src_b); + and_(src_a, src_b); + } + } else { + not_(src_b); + and_(src_a, src_b); + } + break; + case Macro::ALUOperation::Nand: + if (optimizer.zero_reg_skip) { + if (!is_a_zero) { + and_(src_a, src_b); + not_(src_a); + } + } else { + and_(src_a, src_b); + not_(src_a); + } + break; + default: + UNIMPLEMENTED_MSG("Unimplemented ALU operation {}", + static_cast<std::size_t>(opcode.alu_operation.Value())); + break; + } + Compile_ProcessResult(opcode.result_operation, opcode.dst); +} + +void MacroJITx64Impl::Compile_AddImmediate(Macro::Opcode opcode) { + if (optimizer.skip_dummy_addimmediate) { + // Games tend to use this as an exit instruction placeholder. It's to encode an instruction + // without doing anything. In our case we can just not emit anything. + if (opcode.result_operation == Macro::ResultOperation::Move && opcode.dst == 0) { + return; + } + } + // Check for redundant moves + if (optimizer.optimize_for_method_move && + opcode.result_operation == Macro::ResultOperation::MoveAndSetMethod) { + if (next_opcode.has_value()) { + const auto next = *next_opcode; + if (next.result_operation == Macro::ResultOperation::MoveAndSetMethod) { + return; + } + } + } + if (optimizer.zero_reg_skip && opcode.src_a == 0) { + if (opcode.immediate == 0) { + xor_(RESULT, RESULT); + } else { + mov(RESULT, opcode.immediate); + } + } else { + auto result = Compile_GetRegister(opcode.src_a, RESULT); + if (opcode.immediate > 2) { + add(result, opcode.immediate); + } else if (opcode.immediate == 1) { + inc(result); + } else if (opcode.immediate < 0) { + sub(result, opcode.immediate * -1); + } + } + Compile_ProcessResult(opcode.result_operation, opcode.dst); +} + +void MacroJITx64Impl::Compile_ExtractInsert(Macro::Opcode opcode) { + auto dst = Compile_GetRegister(opcode.src_a, RESULT); + auto src = Compile_GetRegister(opcode.src_b, eax); + + if (opcode.bf_src_bit != 0 && opcode.bf_src_bit != 31) { + shr(src, opcode.bf_src_bit); + } else if (opcode.bf_src_bit == 31) { + xor_(src, src); + } + // Don't bother masking the whole register since we're using a 32 bit register + if (opcode.bf_size != 31 && opcode.bf_size != 0) { + and_(src, opcode.GetBitfieldMask()); + } else if (opcode.bf_size == 0) { + xor_(src, src); + } + if (opcode.bf_dst_bit != 31 && opcode.bf_dst_bit != 0) { + shl(src, opcode.bf_dst_bit); + } else if (opcode.bf_dst_bit == 31) { + xor_(src, src); + } + + const u32 mask = ~(opcode.GetBitfieldMask() << opcode.bf_dst_bit); + if (mask != 0xffffffff) { + and_(dst, mask); + } + or_(dst, src); + Compile_ProcessResult(opcode.result_operation, opcode.dst); +} + +void MacroJITx64Impl::Compile_ExtractShiftLeftImmediate(Macro::Opcode opcode) { + auto dst = Compile_GetRegister(opcode.src_a, eax); + auto src = Compile_GetRegister(opcode.src_b, RESULT); + + shr(src, al); + if (opcode.bf_size != 0 && opcode.bf_size != 31) { + and_(src, opcode.GetBitfieldMask()); + } else if (opcode.bf_size == 0) { + xor_(src, src); + } + + if (opcode.bf_dst_bit != 0 && opcode.bf_dst_bit != 31) { + shl(src, opcode.bf_dst_bit); + } else if (opcode.bf_dst_bit == 31) { + xor_(src, src); + } + Compile_ProcessResult(opcode.result_operation, opcode.dst); +} + +void MacroJITx64Impl::Compile_ExtractShiftLeftRegister(Macro::Opcode opcode) { + auto dst = Compile_GetRegister(opcode.src_a, eax); + auto src = Compile_GetRegister(opcode.src_b, RESULT); + + if (opcode.bf_src_bit != 0) { + shr(src, opcode.bf_src_bit); + } + + if (opcode.bf_size != 31) { + and_(src, opcode.GetBitfieldMask()); + } + shl(src, al); + Compile_ProcessResult(opcode.result_operation, opcode.dst); +} + +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); +} + +void MacroJITx64Impl::Compile_Read(Macro::Opcode opcode) { + if (optimizer.zero_reg_skip && opcode.src_a == 0) { + if (opcode.immediate == 0) { + xor_(RESULT, RESULT); + } else { + mov(RESULT, opcode.immediate); + } + } else { + auto result = Compile_GetRegister(opcode.src_a, RESULT); + if (opcode.immediate > 2) { + add(result, opcode.immediate); + } else if (opcode.immediate == 1) { + inc(result); + } else if (opcode.immediate < 0) { + sub(result, opcode.immediate * -1); + } + } + Common::X64::ABI_PushRegistersAndAdjustStackGPS(*this, PersistentCallerSavedRegs(), 0); + mov(Common::X64::ABI_PARAM1, qword[STATE]); + mov(Common::X64::ABI_PARAM2, RESULT); + Common::X64::CallFarFunction(*this, &Read); + Common::X64::ABI_PopRegistersAndAdjustStackGPS(*this, PersistentCallerSavedRegs(), 0); + mov(RESULT, Common::X64::ABI_RETURN.cvt32()); + Compile_ProcessResult(opcode.result_operation, opcode.dst); +} + +void Tegra::MacroJITx64Impl::Compile_Send(Xbyak::Reg32 value) { + Common::X64::ABI_PushRegistersAndAdjustStackGPS(*this, PersistentCallerSavedRegs(), 0); + mov(Common::X64::ABI_PARAM1, qword[STATE]); + mov(Common::X64::ABI_PARAM2, METHOD_ADDRESS); + mov(Common::X64::ABI_PARAM3, value); + Common::X64::CallFarFunction(*this, &Send); + Common::X64::ABI_PopRegistersAndAdjustStackGPS(*this, PersistentCallerSavedRegs(), 0); + + Xbyak::Label dont_process{}; + // Get increment + test(METHOD_ADDRESS, 0x3f000); + // If zero, method address doesn't update + je(dont_process); + + mov(ecx, METHOD_ADDRESS); + and_(METHOD_ADDRESS, 0xfff); + shr(ecx, 12); + and_(ecx, 0x3f); + lea(eax, ptr[rcx + METHOD_ADDRESS_64]); + sal(ecx, 12); + or_(eax, ecx); + + mov(METHOD_ADDRESS, eax); + + L(dont_process); +} + +void Tegra::MacroJITx64Impl::Compile_Branch(Macro::Opcode opcode) { + ASSERT_MSG(!is_delay_slot, "Executing a branch in a delay slot is not valid"); + const s32 jump_address = + static_cast<s32>(pc) + static_cast<s32>(opcode.GetBranchTarget() / sizeof(s32)); + + Xbyak::Label end; + auto value = Compile_GetRegister(opcode.src_a, eax); + test(value, value); + if (optimizer.has_delayed_pc) { + switch (opcode.branch_condition) { + case Macro::BranchCondition::Zero: + jne(end, T_NEAR); + break; + case Macro::BranchCondition::NotZero: + je(end, T_NEAR); + break; + } + + if (opcode.branch_annul) { + xor_(BRANCH_HOLDER, BRANCH_HOLDER); + jmp(labels[jump_address], T_NEAR); + } else { + Xbyak::Label handle_post_exit{}; + Xbyak::Label skip{}; + jmp(skip, T_NEAR); + if (opcode.is_exit) { + L(handle_post_exit); + // Execute 1 instruction + mov(BRANCH_HOLDER, end_of_code); + // Jump to next instruction to skip delay slot check + jmp(labels[jump_address], T_NEAR); + } else { + L(handle_post_exit); + xor_(BRANCH_HOLDER, BRANCH_HOLDER); + jmp(labels[jump_address], T_NEAR); + } + L(skip); + mov(BRANCH_HOLDER, handle_post_exit); + jmp(delay_skip[pc], T_NEAR); + } + } else { + switch (opcode.branch_condition) { + case Macro::BranchCondition::Zero: + je(labels[jump_address], T_NEAR); + break; + case Macro::BranchCondition::NotZero: + jne(labels[jump_address], T_NEAR); + break; + } + } + + L(end); +} + +void Tegra::MacroJITx64Impl::Optimizer_ScanFlags() { + optimizer.can_skip_carry = true; + optimizer.has_delayed_pc = false; + for (auto raw_op : code) { + Macro::Opcode op{}; + op.raw = raw_op; + + if (op.operation == Macro::Operation::ALU) { + // Scan for any ALU operations which actually use the carry flag, if they don't exist in + // our current code we can skip emitting the carry flag handling operations + if (op.alu_operation == Macro::ALUOperation::AddWithCarry || + op.alu_operation == Macro::ALUOperation::SubtractWithBorrow) { + optimizer.can_skip_carry = false; + } + } + + if (op.operation == Macro::Operation::Branch) { + if (!op.branch_annul) { + optimizer.has_delayed_pc = true; + } + } + } +} + +void MacroJITx64Impl::Compile() { + MICROPROFILE_SCOPE(MacroJitCompile); + bool keep_executing = true; + labels.fill(Xbyak::Label()); + + Common::X64::ABI_PushRegistersAndAdjustStackGPS(*this, Common::X64::ABI_ALL_CALLEE_SAVED, 8); + // JIT state + mov(STATE, Common::X64::ABI_PARAM1); + mov(PARAMETERS, qword[Common::X64::ABI_PARAM1 + + static_cast<Xbyak::uint32>(offsetof(JITState, parameters))]); + mov(REGISTERS, Common::X64::ABI_PARAM1); + add(REGISTERS, static_cast<Xbyak::uint32>(offsetof(JITState, registers))); + xor_(RESULT, RESULT); + xor_(METHOD_ADDRESS, METHOD_ADDRESS); + xor_(NEXT_PARAMETER, NEXT_PARAMETER); + xor_(BRANCH_HOLDER, BRANCH_HOLDER); + + mov(dword[REGISTERS + 4], Compile_FetchParameter()); + + // Track get register for zero registers and mark it as no-op + optimizer.zero_reg_skip = true; + + // AddImmediate tends to be used as a NOP instruction, if we detect this we can + // completely skip the entire code path and no emit anything + optimizer.skip_dummy_addimmediate = true; + + // SMO tends to emit a lot of unnecessary method moves, we can mitigate this by only emitting + // one if our register isn't "dirty" + optimizer.optimize_for_method_move = true; + + // Check to see if we can skip emitting certain instructions + Optimizer_ScanFlags(); + + const u32 op_count = static_cast<u32>(code.size()); + for (u32 i = 0; i < op_count; i++) { + if (i < op_count - 1) { + pc = i + 1; + next_opcode = GetOpCode(); + } else { + next_opcode = {}; + } + pc = i; + Compile_NextInstruction(); + } + + L(end_of_code); + + Common::X64::ABI_PopRegistersAndAdjustStackGPS(*this, Common::X64::ABI_ALL_CALLEE_SAVED, 8); + ret(); + ready(); + program = getCode<ProgramType>(); +} + +bool MacroJITx64Impl::Compile_NextInstruction() { + const auto opcode = GetOpCode(); + if (labels[pc].getAddress()) { + return false; + } + + L(labels[pc]); + + switch (opcode.operation) { + case Macro::Operation::ALU: + Compile_ALU(opcode); + break; + case Macro::Operation::AddImmediate: + Compile_AddImmediate(opcode); + break; + case Macro::Operation::ExtractInsert: + Compile_ExtractInsert(opcode); + break; + case Macro::Operation::ExtractShiftLeftImmediate: + Compile_ExtractShiftLeftImmediate(opcode); + break; + case Macro::Operation::ExtractShiftLeftRegister: + Compile_ExtractShiftLeftRegister(opcode); + break; + case Macro::Operation::Read: + Compile_Read(opcode); + break; + case Macro::Operation::Branch: + Compile_Branch(opcode); + break; + default: + UNIMPLEMENTED_MSG("Unimplemented opcode {}", opcode.operation.Value()); + break; + } + + if (optimizer.has_delayed_pc) { + if (opcode.is_exit) { + mov(rax, end_of_code); + test(BRANCH_HOLDER, BRANCH_HOLDER); + cmove(BRANCH_HOLDER, rax); + // Jump to next instruction to skip delay slot check + je(labels[pc + 1], T_NEAR); + } else { + // TODO(ogniK): Optimize delay slot branching + Xbyak::Label no_delay_slot{}; + test(BRANCH_HOLDER, BRANCH_HOLDER); + je(no_delay_slot, T_NEAR); + mov(rax, BRANCH_HOLDER); + xor_(BRANCH_HOLDER, BRANCH_HOLDER); + jmp(rax); + L(no_delay_slot); + } + L(delay_skip[pc]); + if (opcode.is_exit) { + return false; + } + } else { + test(BRANCH_HOLDER, BRANCH_HOLDER); + jne(end_of_code, T_NEAR); + if (opcode.is_exit) { + inc(BRANCH_HOLDER); + return false; + } + } + return true; +} + +Xbyak::Reg32 Tegra::MacroJITx64Impl::Compile_FetchParameter() { + mov(eax, dword[PARAMETERS + NEXT_PARAMETER * sizeof(u32)]); + inc(NEXT_PARAMETER); + return eax; +} + +Xbyak::Reg32 MacroJITx64Impl::Compile_GetRegister(u32 index, Xbyak::Reg32 dst) { + if (index == 0) { + // Register 0 is always zero + xor_(dst, dst); + } else { + mov(dst, dword[REGISTERS + index * sizeof(u32)]); + } + + return dst; +} + +Xbyak::Reg64 Tegra::MacroJITx64Impl::Compile_GetRegister(u32 index, Xbyak::Reg64 dst) { + if (index == 0) { + // Register 0 is always zero + xor_(dst, dst); + } else { + mov(dst, dword[REGISTERS + index * sizeof(u32)]); + } + + return dst; +} + +void Tegra::MacroJITx64Impl::Compile_WriteCarry(Xbyak::Reg64 dst) { + Xbyak::Label zero{}, end{}; + xor_(ecx, ecx); + shr(dst, 32); + setne(cl); + mov(dword[STATE + offsetof(JITState, carry_flag)], ecx); +} + +void MacroJITx64Impl::Compile_ProcessResult(Macro::ResultOperation operation, u32 reg) { + auto SetRegister = [=](u32 reg, Xbyak::Reg32 result) { + // Register 0 is supposed to always return 0. NOP is implemented as a store to the zero + // register. + if (reg == 0) { + return; + } + mov(dword[REGISTERS + reg * sizeof(u32)], result); + }; + auto SetMethodAddress = [=](Xbyak::Reg32 reg) { mov(METHOD_ADDRESS, reg); }; + + switch (operation) { + case Macro::ResultOperation::IgnoreAndFetch: + SetRegister(reg, Compile_FetchParameter()); + break; + case Macro::ResultOperation::Move: + SetRegister(reg, RESULT); + break; + case Macro::ResultOperation::MoveAndSetMethod: + SetRegister(reg, RESULT); + SetMethodAddress(RESULT); + break; + case Macro::ResultOperation::FetchAndSend: + // Fetch parameter and send result. + SetRegister(reg, Compile_FetchParameter()); + Compile_Send(RESULT); + break; + case Macro::ResultOperation::MoveAndSend: + // Move and send result. + SetRegister(reg, RESULT); + Compile_Send(RESULT); + break; + case Macro::ResultOperation::FetchAndSetMethod: + // Fetch parameter and use result as Method Address. + SetRegister(reg, Compile_FetchParameter()); + SetMethodAddress(RESULT); + break; + case Macro::ResultOperation::MoveAndSetMethodFetchAndSend: + // Move result and use as Method Address, then fetch and send parameter. + SetRegister(reg, RESULT); + SetMethodAddress(RESULT); + Compile_Send(Compile_FetchParameter()); + break; + case Macro::ResultOperation::MoveAndSetMethodSend: + // Move result and use as Method Address, then send bits 12:17 of result. + SetRegister(reg, RESULT); + SetMethodAddress(RESULT); + shr(RESULT, 12); + and_(RESULT, 0b111111); + Compile_Send(RESULT); + break; + default: + UNIMPLEMENTED_MSG("Unimplemented macro operation {}", static_cast<std::size_t>(operation)); + } +} + +Macro::Opcode MacroJITx64Impl::GetOpCode() const { + ASSERT(pc < code.size()); + return {code[pc]}; +} + +std::bitset<32> MacroJITx64Impl::PersistentCallerSavedRegs() const { + return PERSISTENT_REGISTERS & Common::X64::ABI_ALL_CALLER_SAVED; +} + +} // namespace Tegra diff --git a/src/video_core/macro/macro_jit_x64.h b/src/video_core/macro/macro_jit_x64.h new file mode 100644 index 000000000..21ee157cf --- /dev/null +++ b/src/video_core/macro/macro_jit_x64.h @@ -0,0 +1,100 @@ +// Copyright 2020 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <array> +#include <bitset> +#include <xbyak.h> +#include "common/bit_field.h" +#include "common/common_types.h" +#include "common/x64/xbyak_abi.h" +#include "video_core/macro/macro.h" + +namespace Tegra { + +namespace Engines { +class Maxwell3D; +} + +/// MAX_CODE_SIZE is arbitrarily chosen based on current booting games +constexpr size_t MAX_CODE_SIZE = 0x10000; + +class MacroJITx64 final : public MacroEngine { +public: + explicit MacroJITx64(Engines::Maxwell3D& maxwell3d); + +protected: + std::unique_ptr<CachedMacro> Compile(const std::vector<u32>& code) override; + +private: + Engines::Maxwell3D& maxwell3d; +}; + +class MacroJITx64Impl : public Xbyak::CodeGenerator, public CachedMacro { +public: + MacroJITx64Impl(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& code); + ~MacroJITx64Impl(); + + void Execute(const std::vector<u32>& parameters, u32 method) override; + + void Compile_ALU(Macro::Opcode opcode); + void Compile_AddImmediate(Macro::Opcode opcode); + void Compile_ExtractInsert(Macro::Opcode opcode); + void Compile_ExtractShiftLeftImmediate(Macro::Opcode opcode); + void Compile_ExtractShiftLeftRegister(Macro::Opcode opcode); + void Compile_Read(Macro::Opcode opcode); + void Compile_Branch(Macro::Opcode opcode); + +private: + void Optimizer_ScanFlags(); + + void Compile(); + bool Compile_NextInstruction(); + + Xbyak::Reg32 Compile_FetchParameter(); + Xbyak::Reg32 Compile_GetRegister(u32 index, Xbyak::Reg32 dst); + Xbyak::Reg64 Compile_GetRegister(u32 index, Xbyak::Reg64 dst); + void Compile_WriteCarry(Xbyak::Reg64 dst); + + void Compile_ProcessResult(Macro::ResultOperation operation, u32 reg); + void Compile_Send(Xbyak::Reg32 value); + + Macro::Opcode GetOpCode() const; + std::bitset<32> PersistentCallerSavedRegs() const; + + struct JITState { + Engines::Maxwell3D* maxwell3d{}; + std::array<u32, Macro::NUM_MACRO_REGISTERS> registers{}; + const u32* parameters{}; + u32 carry_flag{}; + }; + static_assert(offsetof(JITState, maxwell3d) == 0, "Maxwell3D is not at 0x0"); + using ProgramType = void (*)(JITState*); + + struct OptimizerState { + bool can_skip_carry{}; + bool has_delayed_pc{}; + bool zero_reg_skip{}; + bool skip_dummy_addimmediate{}; + bool optimize_for_method_move{}; + }; + OptimizerState optimizer{}; + + std::optional<Macro::Opcode> next_opcode{}; + ProgramType program{nullptr}; + + std::array<Xbyak::Label, MAX_CODE_SIZE> labels{}; + std::array<Xbyak::Label, MAX_CODE_SIZE> delay_skip{}; + Xbyak::Label end_of_code{}; + + bool is_delay_slot{}; + u32 pc{}; + std::optional<u32> delayed_pc; + + const std::vector<u32>& code; + Engines::Maxwell3D& maxwell3d; +}; + +} // namespace Tegra diff --git a/src/video_core/renderer_opengl/gl_device.cpp b/src/video_core/renderer_opengl/gl_device.cpp index 466a911db..890fc6c63 100644 --- a/src/video_core/renderer_opengl/gl_device.cpp +++ b/src/video_core/renderer_opengl/gl_device.cpp @@ -6,6 +6,7 @@ #include <array> #include <cstddef> #include <cstring> +#include <limits> #include <optional> #include <vector> @@ -26,24 +27,27 @@ constexpr u32 ReservedUniformBlocks = 1; constexpr u32 NumStages = 5; -constexpr std::array LimitUBOs = {GL_MAX_VERTEX_UNIFORM_BLOCKS, GL_MAX_TESS_CONTROL_UNIFORM_BLOCKS, - GL_MAX_TESS_EVALUATION_UNIFORM_BLOCKS, - GL_MAX_GEOMETRY_UNIFORM_BLOCKS, GL_MAX_FRAGMENT_UNIFORM_BLOCKS}; +constexpr std::array LimitUBOs = { + GL_MAX_VERTEX_UNIFORM_BLOCKS, GL_MAX_TESS_CONTROL_UNIFORM_BLOCKS, + GL_MAX_TESS_EVALUATION_UNIFORM_BLOCKS, GL_MAX_GEOMETRY_UNIFORM_BLOCKS, + GL_MAX_FRAGMENT_UNIFORM_BLOCKS, GL_MAX_COMPUTE_UNIFORM_BLOCKS}; constexpr std::array LimitSSBOs = { - GL_MAX_VERTEX_SHADER_STORAGE_BLOCKS, GL_MAX_TESS_CONTROL_SHADER_STORAGE_BLOCKS, + GL_MAX_VERTEX_SHADER_STORAGE_BLOCKS, GL_MAX_TESS_CONTROL_SHADER_STORAGE_BLOCKS, GL_MAX_TESS_EVALUATION_SHADER_STORAGE_BLOCKS, GL_MAX_GEOMETRY_SHADER_STORAGE_BLOCKS, - GL_MAX_FRAGMENT_SHADER_STORAGE_BLOCKS}; + GL_MAX_FRAGMENT_SHADER_STORAGE_BLOCKS, GL_MAX_COMPUTE_SHADER_STORAGE_BLOCKS}; -constexpr std::array LimitSamplers = { - GL_MAX_VERTEX_TEXTURE_IMAGE_UNITS, GL_MAX_TESS_CONTROL_TEXTURE_IMAGE_UNITS, - GL_MAX_TESS_EVALUATION_TEXTURE_IMAGE_UNITS, GL_MAX_GEOMETRY_TEXTURE_IMAGE_UNITS, - GL_MAX_TEXTURE_IMAGE_UNITS}; +constexpr std::array LimitSamplers = {GL_MAX_VERTEX_TEXTURE_IMAGE_UNITS, + GL_MAX_TESS_CONTROL_TEXTURE_IMAGE_UNITS, + GL_MAX_TESS_EVALUATION_TEXTURE_IMAGE_UNITS, + GL_MAX_GEOMETRY_TEXTURE_IMAGE_UNITS, + GL_MAX_TEXTURE_IMAGE_UNITS, + GL_MAX_COMPUTE_TEXTURE_IMAGE_UNITS}; -constexpr std::array LimitImages = {GL_MAX_VERTEX_IMAGE_UNIFORMS, - GL_MAX_TESS_CONTROL_IMAGE_UNIFORMS, - GL_MAX_TESS_EVALUATION_IMAGE_UNIFORMS, - GL_MAX_GEOMETRY_IMAGE_UNIFORMS, GL_MAX_FRAGMENT_IMAGE_UNIFORMS}; +constexpr std::array LimitImages = { + GL_MAX_VERTEX_IMAGE_UNIFORMS, GL_MAX_TESS_CONTROL_IMAGE_UNIFORMS, + GL_MAX_TESS_EVALUATION_IMAGE_UNIFORMS, GL_MAX_GEOMETRY_IMAGE_UNIFORMS, + GL_MAX_FRAGMENT_IMAGE_UNIFORMS, GL_MAX_COMPUTE_IMAGE_UNIFORMS}; template <typename T> T GetInteger(GLenum pname) { @@ -85,6 +89,13 @@ u32 Extract(u32& base, u32& num, u32 amount, std::optional<GLenum> limit = {}) { return std::exchange(base, base + amount); } +std::array<u32, Tegra::Engines::MaxShaderTypes> BuildMaxUniformBuffers() noexcept { + std::array<u32, Tegra::Engines::MaxShaderTypes> max; + std::transform(LimitUBOs.begin(), LimitUBOs.end(), max.begin(), + [](GLenum pname) { return GetInteger<u32>(pname); }); + return max; +} + std::array<Device::BaseBindings, Tegra::Engines::MaxShaderTypes> BuildBaseBindings() noexcept { std::array<Device::BaseBindings, Tegra::Engines::MaxShaderTypes> bindings; @@ -133,6 +144,7 @@ std::array<Device::BaseBindings, Tegra::Engines::MaxShaderTypes> BuildBaseBindin } bool IsASTCSupported() { + static constexpr std::array targets = {GL_TEXTURE_2D, GL_TEXTURE_2D_ARRAY}; static constexpr std::array formats = { GL_COMPRESSED_RGBA_ASTC_4x4_KHR, GL_COMPRESSED_RGBA_ASTC_5x4_KHR, GL_COMPRESSED_RGBA_ASTC_5x5_KHR, GL_COMPRESSED_RGBA_ASTC_6x5_KHR, @@ -149,25 +161,43 @@ bool IsASTCSupported() { GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x8_KHR, GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x10_KHR, GL_COMPRESSED_SRGB8_ALPHA8_ASTC_12x10_KHR, GL_COMPRESSED_SRGB8_ALPHA8_ASTC_12x12_KHR, }; - return std::find_if_not(formats.begin(), formats.end(), [](GLenum format) { - GLint supported; - glGetInternalformativ(GL_TEXTURE_2D, format, GL_INTERNALFORMAT_SUPPORTED, 1, - &supported); - return supported == GL_TRUE; - }) == formats.end(); + static constexpr std::array required_support = { + GL_VERTEX_TEXTURE, GL_TESS_CONTROL_TEXTURE, GL_TESS_EVALUATION_TEXTURE, + GL_GEOMETRY_TEXTURE, GL_FRAGMENT_TEXTURE, GL_COMPUTE_TEXTURE, + }; + + for (const GLenum target : targets) { + for (const GLenum format : formats) { + for (const GLenum support : required_support) { + GLint value; + glGetInternalformativ(GL_TEXTURE_2D, format, support, 1, &value); + if (value != GL_FULL_SUPPORT) { + return false; + } + } + } + } + return true; } } // Anonymous namespace -Device::Device() : base_bindings{BuildBaseBindings()} { +Device::Device() + : max_uniform_buffers{BuildMaxUniformBuffers()}, base_bindings{BuildBaseBindings()} { const std::string_view vendor = reinterpret_cast<const char*>(glGetString(GL_VENDOR)); - const auto renderer = reinterpret_cast<const char*>(glGetString(GL_RENDERER)); + const std::string_view version = reinterpret_cast<const char*>(glGetString(GL_VERSION)); const std::vector extensions = GetExtensions(); const bool is_nvidia = vendor == "NVIDIA Corporation"; const bool is_amd = vendor == "ATI Technologies Inc."; - const bool is_intel = vendor == "Intel"; - const bool is_intel_proprietary = is_intel && std::strstr(renderer, "Mesa") == nullptr; + + bool disable_fast_buffer_sub_data = false; + if (is_nvidia && version == "4.6.0 NVIDIA 443.24") { + LOG_WARNING( + Render_OpenGL, + "Beta driver 443.24 is known to have issues. There might be performance issues."); + disable_fast_buffer_sub_data = true; + } uniform_buffer_alignment = GetInteger<std::size_t>(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT); shader_storage_alignment = GetInteger<std::size_t>(GL_SHADER_STORAGE_BUFFER_OFFSET_ALIGNMENT); @@ -182,10 +212,10 @@ Device::Device() : base_bindings{BuildBaseBindings()} { has_variable_aoffi = TestVariableAoffi(); has_component_indexing_bug = is_amd; has_precise_bug = TestPreciseBug(); - has_broken_compute = is_intel_proprietary; - has_fast_buffer_sub_data = is_nvidia; + has_fast_buffer_sub_data = is_nvidia && !disable_fast_buffer_sub_data; use_assembly_shaders = Settings::values.use_assembly_shaders && GLAD_GL_NV_gpu_program5 && - GLAD_GL_NV_compute_program5; + GLAD_GL_NV_compute_program5 && GLAD_GL_NV_transform_feedback && + GLAD_GL_NV_transform_feedback2; LOG_INFO(Render_OpenGL, "Renderer_VariableAOFFI: {}", has_variable_aoffi); LOG_INFO(Render_OpenGL, "Renderer_ComponentIndexingBug: {}", has_component_indexing_bug); @@ -197,7 +227,9 @@ Device::Device() : base_bindings{BuildBaseBindings()} { } Device::Device(std::nullptr_t) { - uniform_buffer_alignment = 0; + max_uniform_buffers.fill(std::numeric_limits<u32>::max()); + uniform_buffer_alignment = 4; + shader_storage_alignment = 4; max_vertex_attributes = 16; max_varyings = 15; has_warp_intrinsics = true; @@ -205,9 +237,6 @@ Device::Device(std::nullptr_t) { has_vertex_viewport_layer = true; has_image_load_formatted = true; has_variable_aoffi = true; - has_component_indexing_bug = false; - has_broken_compute = false; - has_precise_bug = false; } bool Device::TestVariableAoffi() { diff --git a/src/video_core/renderer_opengl/gl_device.h b/src/video_core/renderer_opengl/gl_device.h index e915dbd86..98cca0254 100644 --- a/src/video_core/renderer_opengl/gl_device.h +++ b/src/video_core/renderer_opengl/gl_device.h @@ -24,6 +24,10 @@ public: explicit Device(); explicit Device(std::nullptr_t); + u32 GetMaxUniformBuffers(Tegra::Engines::ShaderType shader_type) const noexcept { + return max_uniform_buffers[static_cast<std::size_t>(shader_type)]; + } + const BaseBindings& GetBaseBindings(std::size_t stage_index) const noexcept { return base_bindings[stage_index]; } @@ -80,10 +84,6 @@ public: return has_precise_bug; } - bool HasBrokenCompute() const { - return has_broken_compute; - } - bool HasFastBufferSubData() const { return has_fast_buffer_sub_data; } @@ -96,7 +96,8 @@ private: static bool TestVariableAoffi(); static bool TestPreciseBug(); - std::array<BaseBindings, Tegra::Engines::MaxShaderTypes> base_bindings; + std::array<u32, Tegra::Engines::MaxShaderTypes> max_uniform_buffers{}; + std::array<BaseBindings, Tegra::Engines::MaxShaderTypes> base_bindings{}; std::size_t uniform_buffer_alignment{}; std::size_t shader_storage_alignment{}; u32 max_vertex_attributes{}; @@ -109,7 +110,6 @@ private: bool has_variable_aoffi{}; bool has_component_indexing_bug{}; bool has_precise_bug{}; - bool has_broken_compute{}; bool has_fast_buffer_sub_data{}; bool use_assembly_shaders{}; }; diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 8116a5daa..f802fd384 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -54,6 +54,12 @@ MICROPROFILE_DEFINE(OpenGL_PrimitiveAssembly, "OpenGL", "Prim Asmbl", MP_RGB(255 namespace { +constexpr std::size_t NUM_CONST_BUFFERS_PER_STAGE = 18; +constexpr std::size_t NUM_CONST_BUFFERS_BYTES_PER_STAGE = + NUM_CONST_BUFFERS_PER_STAGE * Maxwell::MaxConstBufferSize; +constexpr std::size_t TOTAL_CONST_BUFFER_BYTES = + NUM_CONST_BUFFERS_BYTES_PER_STAGE * Maxwell::MaxShaderStage; + constexpr std::size_t NumSupportedVertexAttributes = 16; template <typename Engine, typename Entry> @@ -87,6 +93,34 @@ std::size_t GetConstBufferSize(const Tegra::Engines::ConstBufferInfo& buffer, return buffer.size; } +/// Translates hardware transform feedback indices +/// @param location Hardware location +/// @return Pair of ARB_transform_feedback3 token stream first and third arguments +/// @note Read https://www.khronos.org/registry/OpenGL/extensions/ARB/ARB_transform_feedback3.txt +std::pair<GLint, GLint> TransformFeedbackEnum(u8 location) { + const u8 index = location / 4; + if (index >= 8 && index <= 39) { + return {GL_GENERIC_ATTRIB_NV, index - 8}; + } + if (index >= 48 && index <= 55) { + return {GL_TEXTURE_COORD_NV, index - 48}; + } + switch (index) { + case 7: + return {GL_POSITION, 0}; + case 40: + return {GL_PRIMARY_COLOR_NV, 0}; + case 41: + return {GL_SECONDARY_COLOR_NV, 0}; + case 42: + return {GL_BACK_PRIMARY_COLOR_NV, 0}; + case 43: + return {GL_BACK_SECONDARY_COLOR_NV, 0}; + } + UNIMPLEMENTED_MSG("index={}", static_cast<int>(index)); + return {GL_POSITION, 0}; +} + void oglEnable(GLenum cap, bool state) { (state ? glEnable : glDisable)(cap); } @@ -104,6 +138,9 @@ RasterizerOpenGL::RasterizerOpenGL(Core::System& system, Core::Frontend::EmuWind screen_info{info}, program_manager{program_manager}, state_tracker{state_tracker} { CheckExtensions(); + unified_uniform_buffer.Create(); + glNamedBufferStorage(unified_uniform_buffer.handle, TOTAL_CONST_BUFFER_BYTES, nullptr, 0); + if (device.UseAssemblyShaders()) { glCreateBuffers(static_cast<GLsizei>(staging_cbufs.size()), staging_cbufs.data()); for (const GLuint cbuf : staging_cbufs) { @@ -655,10 +692,6 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { } void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) { - if (device.HasBrokenCompute()) { - return; - } - buffer_cache.Acquire(); current_cbuf = 0; @@ -846,34 +879,56 @@ void RasterizerOpenGL::SetupDrawConstBuffers(std::size_t stage_index, const Shad MICROPROFILE_SCOPE(OpenGL_UBO); const auto& stages = system.GPU().Maxwell3D().state.shader_stages; const auto& shader_stage = stages[stage_index]; + const auto& entries = shader->GetEntries(); + const bool use_unified = entries.use_unified_uniforms; + const std::size_t base_unified_offset = stage_index * NUM_CONST_BUFFERS_BYTES_PER_STAGE; - u32 binding = - device.UseAssemblyShaders() ? 0 : device.GetBaseBindings(stage_index).uniform_buffer; - for (const auto& entry : shader->GetEntries().const_buffers) { - const auto& buffer = shader_stage.const_buffers[entry.GetIndex()]; - SetupConstBuffer(PARAMETER_LUT[stage_index], binding++, buffer, entry); + const auto base_bindings = device.GetBaseBindings(stage_index); + u32 binding = device.UseAssemblyShaders() ? 0 : base_bindings.uniform_buffer; + for (const auto& entry : entries.const_buffers) { + const u32 index = entry.GetIndex(); + const auto& buffer = shader_stage.const_buffers[index]; + SetupConstBuffer(PARAMETER_LUT[stage_index], binding, buffer, entry, use_unified, + base_unified_offset + index * Maxwell::MaxConstBufferSize); + ++binding; + } + if (use_unified) { + const u32 index = static_cast<u32>(base_bindings.shader_storage_buffer + + entries.global_memory_entries.size()); + glBindBufferRange(GL_SHADER_STORAGE_BUFFER, index, unified_uniform_buffer.handle, + base_unified_offset, NUM_CONST_BUFFERS_BYTES_PER_STAGE); } } void RasterizerOpenGL::SetupComputeConstBuffers(const Shader& kernel) { MICROPROFILE_SCOPE(OpenGL_UBO); const auto& launch_desc = system.GPU().KeplerCompute().launch_description; + const auto& entries = kernel->GetEntries(); + const bool use_unified = entries.use_unified_uniforms; u32 binding = 0; - for (const auto& entry : kernel->GetEntries().const_buffers) { + for (const auto& entry : entries.const_buffers) { const auto& config = launch_desc.const_buffer_config[entry.GetIndex()]; const std::bitset<8> mask = launch_desc.const_buffer_enable_mask.Value(); Tegra::Engines::ConstBufferInfo buffer; buffer.address = config.Address(); buffer.size = config.size; buffer.enabled = mask[entry.GetIndex()]; - SetupConstBuffer(GL_COMPUTE_PROGRAM_PARAMETER_BUFFER_NV, binding++, buffer, entry); + SetupConstBuffer(GL_COMPUTE_PROGRAM_PARAMETER_BUFFER_NV, binding, buffer, entry, + use_unified, entry.GetIndex() * Maxwell::MaxConstBufferSize); + ++binding; + } + if (use_unified) { + const GLuint index = static_cast<GLuint>(entries.global_memory_entries.size()); + glBindBufferRange(GL_SHADER_STORAGE_BUFFER, index, unified_uniform_buffer.handle, 0, + NUM_CONST_BUFFERS_BYTES_PER_STAGE); } } void RasterizerOpenGL::SetupConstBuffer(GLenum stage, u32 binding, const Tegra::Engines::ConstBufferInfo& buffer, - const ConstBufferEntry& entry) { + const ConstBufferEntry& entry, bool use_unified, + std::size_t unified_offset) { if (!buffer.enabled) { // Set values to zero to unbind buffers if (device.UseAssemblyShaders()) { @@ -889,20 +944,29 @@ void RasterizerOpenGL::SetupConstBuffer(GLenum stage, u32 binding, // UBO alignment requirements. const std::size_t size = Common::AlignUp(GetConstBufferSize(buffer, entry), sizeof(GLvec4)); - const auto alignment = device.GetUniformBufferAlignment(); - auto [cbuf, offset] = buffer_cache.UploadMemory(buffer.address, size, alignment, false, - device.HasFastBufferSubData()); - if (!device.UseAssemblyShaders()) { - glBindBufferRange(GL_UNIFORM_BUFFER, binding, cbuf, offset, size); + const bool fast_upload = !use_unified && device.HasFastBufferSubData(); + + const std::size_t alignment = use_unified ? 4 : device.GetUniformBufferAlignment(); + const GPUVAddr gpu_addr = buffer.address; + auto [cbuf, offset] = buffer_cache.UploadMemory(gpu_addr, size, alignment, false, fast_upload); + + if (device.UseAssemblyShaders()) { + UNIMPLEMENTED_IF(use_unified); + if (offset != 0) { + const GLuint staging_cbuf = staging_cbufs[current_cbuf++]; + glCopyNamedBufferSubData(cbuf, staging_cbuf, offset, 0, size); + cbuf = staging_cbuf; + offset = 0; + } + glBindBufferRangeNV(stage, binding, cbuf, offset, size); return; } - if (offset != 0) { - const GLuint staging_cbuf = staging_cbufs[current_cbuf++]; - glCopyNamedBufferSubData(cbuf, staging_cbuf, offset, 0, size); - cbuf = staging_cbuf; - offset = 0; + + if (use_unified) { + glCopyNamedBufferSubData(cbuf, unified_uniform_buffer.handle, offset, unified_offset, size); + } else { + glBindBufferRange(GL_UNIFORM_BUFFER, binding, cbuf, offset, size); } - glBindBufferRangeNV(stage, binding, cbuf, offset, size); } void RasterizerOpenGL::SetupDrawGlobalMemory(std::size_t stage_index, const Shader& shader) { @@ -977,16 +1041,12 @@ void RasterizerOpenGL::SetupTexture(u32 binding, const Tegra::Texture::FullTextu glBindTextureUnit(binding, 0); return; } - glBindTextureUnit(binding, view->GetTexture()); - - if (view->GetSurfaceParams().IsBuffer()) { - return; + const GLuint handle = view->GetTexture(texture.tic.x_source, texture.tic.y_source, + texture.tic.z_source, texture.tic.w_source); + glBindTextureUnit(binding, handle); + if (!view->GetSurfaceParams().IsBuffer()) { + glBindSampler(binding, sampler_cache.GetSampler(texture.tsc)); } - // Apply swizzle to textures that are not buffers. - view->ApplySwizzle(texture.tic.x_source, texture.tic.y_source, texture.tic.z_source, - texture.tic.w_source); - - glBindSampler(binding, sampler_cache.GetSampler(texture.tsc)); } void RasterizerOpenGL::SetupDrawImages(std::size_t stage_index, const Shader& shader) { @@ -1015,14 +1075,11 @@ void RasterizerOpenGL::SetupImage(u32 binding, const Tegra::Texture::TICEntry& t glBindImageTexture(binding, 0, 0, GL_FALSE, 0, GL_READ_ONLY, GL_R8); return; } - if (!tic.IsBuffer()) { - view->ApplySwizzle(tic.x_source, tic.y_source, tic.z_source, tic.w_source); - } if (entry.is_written) { view->MarkAsModified(texture_cache.Tick()); } - glBindImageTexture(binding, view->GetTexture(), 0, GL_TRUE, 0, GL_READ_WRITE, - view->GetFormat()); + const GLuint handle = view->GetTexture(tic.x_source, tic.y_source, tic.z_source, tic.w_source); + glBindImageTexture(binding, handle, 0, GL_TRUE, 0, GL_READ_WRITE, view->GetFormat()); } void RasterizerOpenGL::SyncViewport() { @@ -1031,6 +1088,26 @@ void RasterizerOpenGL::SyncViewport() { const auto& regs = gpu.regs; const bool dirty_viewport = flags[Dirty::Viewports]; + const bool dirty_clip_control = flags[Dirty::ClipControl]; + + if (dirty_clip_control || flags[Dirty::FrontFace]) { + flags[Dirty::FrontFace] = false; + + GLenum mode = MaxwellToGL::FrontFace(regs.front_face); + if (regs.screen_y_control.triangle_rast_flip != 0 && + regs.viewport_transform[0].scale_y < 0.0f) { + switch (mode) { + case GL_CW: + mode = GL_CCW; + break; + case GL_CCW: + mode = GL_CW; + break; + } + } + glFrontFace(mode); + } + if (dirty_viewport || flags[Dirty::ClipControl]) { flags[Dirty::ClipControl] = false; @@ -1128,11 +1205,6 @@ void RasterizerOpenGL::SyncCullMode() { glDisable(GL_CULL_FACE); } } - - if (flags[Dirty::FrontFace]) { - flags[Dirty::FrontFace] = false; - glFrontFace(MaxwellToGL::FrontFace(regs.front_face)); - } } void RasterizerOpenGL::SyncPrimitiveRestart() { @@ -1503,12 +1575,70 @@ void RasterizerOpenGL::SyncFramebufferSRGB() { oglEnable(GL_FRAMEBUFFER_SRGB, gpu.regs.framebuffer_srgb); } +void RasterizerOpenGL::SyncTransformFeedback() { + // TODO(Rodrigo): Inject SKIP_COMPONENTS*_NV when required. An unimplemented message will signal + // when this is required. + const auto& regs = system.GPU().Maxwell3D().regs; + + static constexpr std::size_t STRIDE = 3; + std::array<GLint, 128 * STRIDE * Maxwell::NumTransformFeedbackBuffers> attribs; + std::array<GLint, Maxwell::NumTransformFeedbackBuffers> streams; + + GLint* cursor = attribs.data(); + GLint* current_stream = streams.data(); + + for (std::size_t feedback = 0; feedback < Maxwell::NumTransformFeedbackBuffers; ++feedback) { + const auto& layout = regs.tfb_layouts[feedback]; + UNIMPLEMENTED_IF_MSG(layout.stride != layout.varying_count * 4, "Stride padding"); + if (layout.varying_count == 0) { + continue; + } + + *current_stream = static_cast<GLint>(feedback); + if (current_stream != streams.data()) { + // When stepping one stream, push the expected token + cursor[0] = GL_NEXT_BUFFER_NV; + cursor[1] = 0; + cursor[2] = 0; + cursor += STRIDE; + } + ++current_stream; + + const auto& locations = regs.tfb_varying_locs[feedback]; + std::optional<u8> current_index; + for (u32 offset = 0; offset < layout.varying_count; ++offset) { + const u8 location = locations[offset]; + const u8 index = location / 4; + + if (current_index == index) { + // Increase number of components of the previous attachment + ++cursor[-2]; + continue; + } + current_index = index; + + std::tie(cursor[0], cursor[2]) = TransformFeedbackEnum(location); + cursor[1] = 1; + cursor += STRIDE; + } + } + + const GLsizei num_attribs = static_cast<GLsizei>((cursor - attribs.data()) / STRIDE); + const GLsizei num_strides = static_cast<GLsizei>(current_stream - streams.data()); + glTransformFeedbackStreamAttribsNV(num_attribs, attribs.data(), num_strides, streams.data(), + GL_INTERLEAVED_ATTRIBS); +} + void RasterizerOpenGL::BeginTransformFeedback(GLenum primitive_mode) { const auto& regs = system.GPU().Maxwell3D().regs; if (regs.tfb_enabled == 0) { return; } + if (device.UseAssemblyShaders()) { + SyncTransformFeedback(); + } + UNIMPLEMENTED_IF(regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationControl) || regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationEval) || regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::Geometry)); @@ -1535,6 +1665,10 @@ void RasterizerOpenGL::BeginTransformFeedback(GLenum primitive_mode) { static_cast<GLsizeiptr>(size)); } + // We may have to call BeginTransformFeedbackNV here since they seem to call different + // implementations on Nvidia's driver (the pointer is different) but we are using + // ARB_transform_feedback3 features with NV_transform_feedback interactions and the ARB + // extension doesn't define BeginTransformFeedback (without NV) interactions. It just works. glBeginTransformFeedback(GL_POINTS); } diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index 87f7fe159..7abc8fdbd 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -107,7 +107,8 @@ private: /// Configures a constant buffer. void SetupConstBuffer(GLenum stage, u32 binding, const Tegra::Engines::ConstBufferInfo& buffer, - const ConstBufferEntry& entry); + const ConstBufferEntry& entry, bool use_unified, + std::size_t unified_offset); /// Configures the current global memory entries to use for the draw command. void SetupDrawGlobalMemory(std::size_t stage_index, const Shader& shader); @@ -201,6 +202,10 @@ private: /// Syncs the framebuffer sRGB state to match the guest state void SyncFramebufferSRGB(); + /// Syncs transform feedback state to match guest state + /// @note Only valid on assembly shaders + void SyncTransformFeedback(); + /// Begin a transform feedback void BeginTransformFeedback(GLenum primitive_mode); @@ -253,6 +258,7 @@ private: Tegra::Engines::Maxwell3D::Regs::MaxShaderProgram; std::array<GLuint, NUM_CONSTANT_BUFFERS> staging_cbufs{}; std::size_t current_cbuf = 0; + OGLBuffer unified_uniform_buffer; /// Number of commands queued to the OpenGL driver. Reseted on flush. std::size_t num_queued_commands = 0; diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp index 4cd0f36cf..a991ca64a 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp @@ -241,8 +241,9 @@ Shader CachedShader::CreateStageFromMemory(const ShaderParameters& params, entry.bindless_samplers = registry->GetBindlessSamplers(); params.disk_cache.SaveEntry(std::move(entry)); - return std::shared_ptr<CachedShader>(new CachedShader( - params.cpu_addr, size_in_bytes, std::move(registry), MakeEntries(ir), std::move(program))); + return std::shared_ptr<CachedShader>( + new CachedShader(params.cpu_addr, size_in_bytes, std::move(registry), + MakeEntries(params.device, ir, shader_type), std::move(program))); } Shader CachedShader::CreateKernelFromMemory(const ShaderParameters& params, ProgramCode code) { @@ -265,8 +266,9 @@ Shader CachedShader::CreateKernelFromMemory(const ShaderParameters& params, Prog entry.bindless_samplers = registry->GetBindlessSamplers(); params.disk_cache.SaveEntry(std::move(entry)); - return std::shared_ptr<CachedShader>(new CachedShader( - params.cpu_addr, size_in_bytes, std::move(registry), MakeEntries(ir), std::move(program))); + return std::shared_ptr<CachedShader>( + new CachedShader(params.cpu_addr, size_in_bytes, std::move(registry), + MakeEntries(params.device, ir, ShaderType::Compute), std::move(program))); } Shader CachedShader::CreateFromCache(const ShaderParameters& params, @@ -348,7 +350,7 @@ void ShaderCacheOpenGL::LoadDiskCache(const std::atomic_bool& stop_loading, PrecompiledShader shader; shader.program = std::move(program); shader.registry = std::move(registry); - shader.entries = MakeEntries(ir); + shader.entries = MakeEntries(device, ir, entry.type); std::scoped_lock lock{mutex}; if (callback) { diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index 253484968..d6e30b321 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -61,8 +61,8 @@ struct TextureDerivates {}; using TextureArgument = std::pair<Type, Node>; using TextureIR = std::variant<TextureOffset, TextureDerivates, TextureArgument>; -constexpr u32 MAX_CONSTBUFFER_ELEMENTS = - static_cast<u32>(Maxwell::MaxConstBufferSize) / (4 * sizeof(float)); +constexpr u32 MAX_CONSTBUFFER_SCALARS = static_cast<u32>(Maxwell::MaxConstBufferSize) / sizeof(u32); +constexpr u32 MAX_CONSTBUFFER_ELEMENTS = MAX_CONSTBUFFER_SCALARS / sizeof(u32); constexpr std::string_view CommonDeclarations = R"(#define ftoi floatBitsToInt #define ftou floatBitsToUint @@ -402,6 +402,13 @@ std::string FlowStackTopName(MetaStackClass stack) { return fmt::format("{}_flow_stack_top", GetFlowStackPrefix(stack)); } +bool UseUnifiedUniforms(const Device& device, const ShaderIR& ir, ShaderType stage) { + const u32 num_ubos = static_cast<u32>(ir.GetConstantBuffers().size()); + // We waste one UBO for emulation + const u32 num_available_ubos = device.GetMaxUniformBuffers(stage) - 1; + return num_ubos > num_available_ubos; +} + struct GenericVaryingDescription { std::string name; u8 first_element = 0; @@ -412,8 +419,9 @@ class GLSLDecompiler final { public: explicit GLSLDecompiler(const Device& device, const ShaderIR& ir, const Registry& registry, ShaderType stage, std::string_view identifier, std::string_view suffix) - : device{device}, ir{ir}, registry{registry}, stage{stage}, - identifier{identifier}, suffix{suffix}, header{ir.GetHeader()} { + : device{device}, ir{ir}, registry{registry}, stage{stage}, identifier{identifier}, + suffix{suffix}, header{ir.GetHeader()}, use_unified_uniforms{ + UseUnifiedUniforms(device, ir, stage)} { if (stage != ShaderType::Compute) { transform_feedback = BuildTransformFeedback(registry.GetGraphicsInfo()); } @@ -618,7 +626,9 @@ private: break; } } - if (stage != ShaderType::Vertex || device.HasVertexViewportLayer()) { + + if (stage != ShaderType::Geometry && + (stage != ShaderType::Vertex || device.HasVertexViewportLayer())) { if (ir.UsesLayer()) { code.AddLine("int gl_Layer;"); } @@ -647,6 +657,16 @@ private: --code.scope; code.AddLine("}};"); code.AddNewLine(); + + if (stage == ShaderType::Geometry) { + if (ir.UsesLayer()) { + code.AddLine("out int gl_Layer;"); + } + if (ir.UsesViewportIndex()) { + code.AddLine("out int gl_ViewportIndex;"); + } + } + code.AddNewLine(); } void DeclareRegisters() { @@ -834,12 +854,24 @@ private: } void DeclareConstantBuffers() { + if (use_unified_uniforms) { + const u32 binding = device.GetBaseBindings(stage).shader_storage_buffer + + static_cast<u32>(ir.GetGlobalMemory().size()); + code.AddLine("layout (std430, binding = {}) readonly buffer UnifiedUniforms {{", + binding); + code.AddLine(" uint cbufs[];"); + code.AddLine("}};"); + code.AddNewLine(); + return; + } + u32 binding = device.GetBaseBindings(stage).uniform_buffer; - for (const auto& buffers : ir.GetConstantBuffers()) { - const auto index = buffers.first; + for (const auto [index, info] : ir.GetConstantBuffers()) { + const u32 num_elements = Common::AlignUp(info.GetSize(), 4) / 4; + const u32 size = info.IsIndirect() ? MAX_CONSTBUFFER_ELEMENTS : num_elements; code.AddLine("layout (std140, binding = {}) uniform {} {{", binding++, GetConstBufferBlock(index)); - code.AddLine(" uvec4 {}[{}];", GetConstBuffer(index), MAX_CONSTBUFFER_ELEMENTS); + code.AddLine(" uvec4 {}[{}];", GetConstBuffer(index), size); code.AddLine("}};"); code.AddNewLine(); } @@ -1038,42 +1070,51 @@ private: if (const auto cbuf = std::get_if<CbufNode>(&*node)) { const Node offset = cbuf->GetOffset(); + const u32 base_unified_offset = cbuf->GetIndex() * MAX_CONSTBUFFER_SCALARS; + if (const auto immediate = std::get_if<ImmediateNode>(&*offset)) { // Direct access const u32 offset_imm = immediate->GetValue(); ASSERT_MSG(offset_imm % 4 == 0, "Unaligned cbuf direct access"); - return {fmt::format("{}[{}][{}]", GetConstBuffer(cbuf->GetIndex()), - offset_imm / (4 * 4), (offset_imm / 4) % 4), + if (use_unified_uniforms) { + return {fmt::format("cbufs[{}]", base_unified_offset + offset_imm / 4), + Type::Uint}; + } else { + return {fmt::format("{}[{}][{}]", GetConstBuffer(cbuf->GetIndex()), + offset_imm / (4 * 4), (offset_imm / 4) % 4), + Type::Uint}; + } + } + + // Indirect access + if (use_unified_uniforms) { + return {fmt::format("cbufs[{} + ({} >> 2)]", base_unified_offset, + Visit(offset).AsUint()), Type::Uint}; } - if (std::holds_alternative<OperationNode>(*offset)) { - // Indirect access - const std::string final_offset = code.GenerateTemporary(); - code.AddLine("uint {} = {} >> 2;", final_offset, Visit(offset).AsUint()); + const std::string final_offset = code.GenerateTemporary(); + code.AddLine("uint {} = {} >> 2;", final_offset, Visit(offset).AsUint()); - if (!device.HasComponentIndexingBug()) { - return {fmt::format("{}[{} >> 2][{} & 3]", GetConstBuffer(cbuf->GetIndex()), - final_offset, final_offset), - Type::Uint}; - } - - // AMD's proprietary GLSL compiler emits ill code for variable component access. - // To bypass this driver bug generate 4 ifs, one per each component. - const std::string pack = code.GenerateTemporary(); - code.AddLine("uvec4 {} = {}[{} >> 2];", pack, GetConstBuffer(cbuf->GetIndex()), - final_offset); - - const std::string result = code.GenerateTemporary(); - code.AddLine("uint {};", result); - for (u32 swizzle = 0; swizzle < 4; ++swizzle) { - code.AddLine("if (({} & 3) == {}) {} = {}{};", final_offset, swizzle, result, - pack, GetSwizzle(swizzle)); - } - return {result, Type::Uint}; + if (!device.HasComponentIndexingBug()) { + return {fmt::format("{}[{} >> 2][{} & 3]", GetConstBuffer(cbuf->GetIndex()), + final_offset, final_offset), + Type::Uint}; } - UNREACHABLE_MSG("Unmanaged offset node type"); + // AMD's proprietary GLSL compiler emits ill code for variable component access. + // To bypass this driver bug generate 4 ifs, one per each component. + const std::string pack = code.GenerateTemporary(); + code.AddLine("uvec4 {} = {}[{} >> 2];", pack, GetConstBuffer(cbuf->GetIndex()), + final_offset); + + const std::string result = code.GenerateTemporary(); + code.AddLine("uint {};", result); + for (u32 swizzle = 0; swizzle < 4; ++swizzle) { + code.AddLine("if (({} & 3) == {}) {} = {}{};", final_offset, swizzle, result, pack, + GetSwizzle(swizzle)); + } + return {result, Type::Uint}; } if (const auto gmem = std::get_if<GmemNode>(&*node)) { @@ -2344,7 +2385,12 @@ private: return {}; } - Expression MemoryBarrierGL(Operation) { + Expression MemoryBarrierGroup(Operation) { + code.AddLine("groupMemoryBarrier();"); + return {}; + } + + Expression MemoryBarrierGlobal(Operation) { code.AddLine("memoryBarrier();"); return {}; } @@ -2591,7 +2637,8 @@ private: &GLSLDecompiler::ShuffleIndexed, &GLSLDecompiler::Barrier, - &GLSLDecompiler::MemoryBarrierGL, + &GLSLDecompiler::MemoryBarrierGroup, + &GLSLDecompiler::MemoryBarrierGlobal, }; static_assert(operation_decompilers.size() == static_cast<std::size_t>(OperationCode::Amount)); @@ -2704,6 +2751,7 @@ private: const std::string_view identifier; const std::string_view suffix; const Header header; + const bool use_unified_uniforms; std::unordered_map<u8, VaryingTFB> transform_feedback; ShaderWriter code; @@ -2899,7 +2947,7 @@ void GLSLDecompiler::DecompileAST() { } // Anonymous namespace -ShaderEntries MakeEntries(const VideoCommon::Shader::ShaderIR& ir) { +ShaderEntries MakeEntries(const Device& device, const ShaderIR& ir, ShaderType stage) { ShaderEntries entries; for (const auto& cbuf : ir.GetConstantBuffers()) { entries.const_buffers.emplace_back(cbuf.second.GetMaxOffset(), cbuf.second.IsIndirect(), @@ -2920,6 +2968,7 @@ ShaderEntries MakeEntries(const VideoCommon::Shader::ShaderIR& ir) { entries.clip_distances = (clip_distances[i] ? 1U : 0U) << i; } entries.shader_length = ir.GetLength(); + entries.use_unified_uniforms = UseUnifiedUniforms(device, ir, stage); return entries; } diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.h b/src/video_core/renderer_opengl/gl_shader_decompiler.h index e8a178764..451c9689a 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.h +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.h @@ -53,11 +53,13 @@ struct ShaderEntries { std::vector<GlobalMemoryEntry> global_memory_entries; std::vector<SamplerEntry> samplers; std::vector<ImageEntry> images; - u32 clip_distances{}; std::size_t shader_length{}; + u32 clip_distances{}; + bool use_unified_uniforms{}; }; -ShaderEntries MakeEntries(const VideoCommon::Shader::ShaderIR& ir); +ShaderEntries MakeEntries(const Device& device, const VideoCommon::Shader::ShaderIR& ir, + Tegra::Engines::ShaderType stage); std::string DecompileShader(const Device& device, const VideoCommon::Shader::ShaderIR& ir, const VideoCommon::Shader::Registry& registry, diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp index 94fbd2a22..61505879b 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.cpp +++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp @@ -35,7 +35,7 @@ MICROPROFILE_DEFINE(OpenGL_Texture_Buffer_Copy, "OpenGL", "Texture Buffer Copy", namespace { struct FormatTuple { - GLint internal_format; + GLenum internal_format; GLenum format = GL_NONE; GLenum type = GL_NONE; }; @@ -238,6 +238,12 @@ OGLTexture CreateTexture(const SurfaceParams& params, GLenum target, GLenum inte return texture; } +constexpr u32 EncodeSwizzle(SwizzleSource x_source, SwizzleSource y_source, SwizzleSource z_source, + SwizzleSource w_source) { + return (static_cast<u32>(x_source) << 24) | (static_cast<u32>(y_source) << 16) | + (static_cast<u32>(z_source) << 8) | static_cast<u32>(w_source); +} + } // Anonymous namespace CachedSurface::CachedSurface(const GPUVAddr gpu_addr, const SurfaceParams& params, @@ -257,9 +263,14 @@ CachedSurface::CachedSurface(const GPUVAddr gpu_addr, const SurfaceParams& param target = GetTextureTarget(params.target); texture = CreateTexture(params, target, internal_format, texture_buffer); DecorateSurfaceName(); - main_view = CreateViewInner( - ViewParams(params.target, 0, params.is_layered ? params.depth : 1, 0, params.num_levels), - true); + + u32 num_layers = 1; + if (params.is_layered || params.target == SurfaceTarget::Texture3D) { + num_layers = params.depth; + } + + main_view = + CreateViewInner(ViewParams(params.target, 0, num_layers, 0, params.num_levels), true); } CachedSurface::~CachedSurface() = default; @@ -381,7 +392,7 @@ void CachedSurface::DecorateSurfaceName() { } void CachedSurfaceView::DecorateViewName(GPUVAddr gpu_addr, std::string prefix) { - LabelGLObject(GL_TEXTURE, texture_view.handle, gpu_addr, prefix); + LabelGLObject(GL_TEXTURE, main_view.handle, gpu_addr, prefix); } View CachedSurface::CreateView(const ViewParams& view_key) { @@ -397,32 +408,33 @@ View CachedSurface::CreateViewInner(const ViewParams& view_key, const bool is_pr } CachedSurfaceView::CachedSurfaceView(CachedSurface& surface, const ViewParams& params, - const bool is_proxy) - : VideoCommon::ViewBase(params), surface{surface}, is_proxy{is_proxy} { - target = GetTextureTarget(params.target); - format = GetFormatTuple(surface.GetSurfaceParams().pixel_format).internal_format; + bool is_proxy) + : VideoCommon::ViewBase(params), surface{surface}, format{surface.internal_format}, + target{GetTextureTarget(params.target)}, is_proxy{is_proxy} { if (!is_proxy) { - texture_view = CreateTextureView(); + main_view = CreateTextureView(); } - swizzle = EncodeSwizzle(SwizzleSource::R, SwizzleSource::G, SwizzleSource::B, SwizzleSource::A); } CachedSurfaceView::~CachedSurfaceView() = default; -void CachedSurfaceView::Attach(GLenum attachment, GLenum target) const { +void CachedSurfaceView::Attach(GLenum attachment, GLenum fb_target) const { ASSERT(params.num_levels == 1); + if (params.target == SurfaceTarget::Texture3D) { + if (params.num_layers > 1) { + ASSERT(params.base_layer == 0); + glFramebufferTexture(fb_target, attachment, surface.texture.handle, params.base_level); + } else { + glFramebufferTexture3D(fb_target, attachment, target, surface.texture.handle, + params.base_level, params.base_layer); + } + return; + } + if (params.num_layers > 1) { - // Layered framebuffer attachments UNIMPLEMENTED_IF(params.base_layer != 0); - - switch (params.target) { - case SurfaceTarget::Texture2DArray: - glFramebufferTexture(target, attachment, GetTexture(), 0); - break; - default: - UNIMPLEMENTED(); - } + glFramebufferTexture(fb_target, attachment, GetTexture(), 0); return; } @@ -430,16 +442,16 @@ void CachedSurfaceView::Attach(GLenum attachment, GLenum target) const { const GLuint texture = surface.GetTexture(); switch (surface.GetSurfaceParams().target) { case SurfaceTarget::Texture1D: - glFramebufferTexture1D(target, attachment, view_target, texture, params.base_level); + glFramebufferTexture1D(fb_target, attachment, view_target, texture, params.base_level); break; case SurfaceTarget::Texture2D: - glFramebufferTexture2D(target, attachment, view_target, texture, params.base_level); + glFramebufferTexture2D(fb_target, attachment, view_target, texture, params.base_level); break; case SurfaceTarget::Texture1DArray: case SurfaceTarget::Texture2DArray: case SurfaceTarget::TextureCubemap: case SurfaceTarget::TextureCubeArray: - glFramebufferTextureLayer(target, attachment, texture, params.base_level, + glFramebufferTextureLayer(fb_target, attachment, texture, params.base_level, params.base_layer); break; default: @@ -447,35 +459,62 @@ void CachedSurfaceView::Attach(GLenum attachment, GLenum target) const { } } -void CachedSurfaceView::ApplySwizzle(SwizzleSource x_source, SwizzleSource y_source, +GLuint CachedSurfaceView::GetTexture(SwizzleSource x_source, SwizzleSource y_source, SwizzleSource z_source, SwizzleSource w_source) { - u32 new_swizzle = EncodeSwizzle(x_source, y_source, z_source, w_source); - if (new_swizzle == swizzle) - return; - swizzle = new_swizzle; - const std::array gl_swizzle = {GetSwizzleSource(x_source), GetSwizzleSource(y_source), - GetSwizzleSource(z_source), GetSwizzleSource(w_source)}; - const GLuint handle = GetTexture(); - const PixelFormat format = surface.GetSurfaceParams().pixel_format; - switch (format) { + if (GetSurfaceParams().IsBuffer()) { + return GetTexture(); + } + const u32 new_swizzle = EncodeSwizzle(x_source, y_source, z_source, w_source); + if (current_swizzle == new_swizzle) { + return current_view; + } + current_swizzle = new_swizzle; + + const auto [entry, is_cache_miss] = view_cache.try_emplace(new_swizzle); + OGLTextureView& view = entry->second; + if (!is_cache_miss) { + current_view = view.handle; + return view.handle; + } + view = CreateTextureView(); + current_view = view.handle; + + std::array swizzle{x_source, y_source, z_source, w_source}; + + switch (const PixelFormat format = GetSurfaceParams().pixel_format) { case PixelFormat::Z24S8: case PixelFormat::Z32FS8: case PixelFormat::S8Z24: - glTextureParameteri(handle, GL_DEPTH_STENCIL_TEXTURE_MODE, + UNIMPLEMENTED_IF(x_source != SwizzleSource::R && x_source != SwizzleSource::G); + glTextureParameteri(view.handle, GL_DEPTH_STENCIL_TEXTURE_MODE, GetComponent(format, x_source == SwizzleSource::R)); - break; - default: - glTextureParameteriv(handle, GL_TEXTURE_SWIZZLE_RGBA, gl_swizzle.data()); + + // Make sure we sample the first component + std::transform(swizzle.begin(), swizzle.end(), swizzle.begin(), [](SwizzleSource value) { + return value == SwizzleSource::G ? SwizzleSource::R : value; + }); + [[fallthrough]]; + default: { + const std::array gl_swizzle = {GetSwizzleSource(swizzle[0]), GetSwizzleSource(swizzle[1]), + GetSwizzleSource(swizzle[2]), GetSwizzleSource(swizzle[3])}; + glTextureParameteriv(view.handle, GL_TEXTURE_SWIZZLE_RGBA, gl_swizzle.data()); break; } + } + return view.handle; } OGLTextureView CachedSurfaceView::CreateTextureView() const { OGLTextureView texture_view; texture_view.Create(); - glTextureView(texture_view.handle, target, surface.texture.handle, format, params.base_level, - params.num_levels, params.base_layer, params.num_layers); + if (target == GL_TEXTURE_3D) { + glTextureView(texture_view.handle, target, surface.texture.handle, format, + params.base_level, params.num_levels, 0, 1); + } else { + glTextureView(texture_view.handle, target, surface.texture.handle, format, + params.base_level, params.num_levels, params.base_layer, params.num_layers); + } ApplyTextureDefaults(surface.GetSurfaceParams(), texture_view.handle); return texture_view; @@ -518,8 +557,8 @@ void TextureCacheOpenGL::ImageBlit(View& src_view, View& dst_view, const Tegra::Engines::Fermi2D::Config& copy_config) { const auto& src_params{src_view->GetSurfaceParams()}; const auto& dst_params{dst_view->GetSurfaceParams()}; - UNIMPLEMENTED_IF(src_params.target == SurfaceTarget::Texture3D); - UNIMPLEMENTED_IF(dst_params.target == SurfaceTarget::Texture3D); + UNIMPLEMENTED_IF(src_params.depth != 1); + UNIMPLEMENTED_IF(dst_params.depth != 1); state_tracker.NotifyScissor0(); state_tracker.NotifyFramebuffer(); diff --git a/src/video_core/renderer_opengl/gl_texture_cache.h b/src/video_core/renderer_opengl/gl_texture_cache.h index 02d9981a1..bfc4ddf5d 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.h +++ b/src/video_core/renderer_opengl/gl_texture_cache.h @@ -80,10 +80,12 @@ public: explicit CachedSurfaceView(CachedSurface& surface, const ViewParams& params, bool is_proxy); ~CachedSurfaceView(); - /// Attaches this texture view to the current bound GL_DRAW_FRAMEBUFFER - void Attach(GLenum attachment, GLenum target) const; + /// @brief Attaches this texture view to the currently bound fb_target framebuffer + /// @param attachment Attachment to bind textures to + /// @param fb_target Framebuffer target to attach to (e.g. DRAW_FRAMEBUFFER) + void Attach(GLenum attachment, GLenum fb_target) const; - void ApplySwizzle(Tegra::Texture::SwizzleSource x_source, + GLuint GetTexture(Tegra::Texture::SwizzleSource x_source, Tegra::Texture::SwizzleSource y_source, Tegra::Texture::SwizzleSource z_source, Tegra::Texture::SwizzleSource w_source); @@ -98,7 +100,7 @@ public: if (is_proxy) { return surface.GetTexture(); } - return texture_view.handle; + return main_view.handle; } GLenum GetFormat() const { @@ -110,23 +112,19 @@ public: } private: - u32 EncodeSwizzle(Tegra::Texture::SwizzleSource x_source, - Tegra::Texture::SwizzleSource y_source, - Tegra::Texture::SwizzleSource z_source, - Tegra::Texture::SwizzleSource w_source) const { - return (static_cast<u32>(x_source) << 24) | (static_cast<u32>(y_source) << 16) | - (static_cast<u32>(z_source) << 8) | static_cast<u32>(w_source); - } - OGLTextureView CreateTextureView() const; CachedSurface& surface; - GLenum target{}; - GLenum format{}; + const GLenum format; + const GLenum target; + const bool is_proxy; + + std::unordered_map<u32, OGLTextureView> view_cache; + OGLTextureView main_view; - OGLTextureView texture_view; - u32 swizzle{}; - bool is_proxy{}; + // Use an invalid default so it always fails the comparison test + u32 current_swizzle = 0xffffffff; + GLuint current_view = 0; }; class TextureCacheOpenGL final : public TextureCacheBase { diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp index 6b489e6db..6214fcbc3 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.cpp +++ b/src/video_core/renderer_opengl/renderer_opengl.cpp @@ -751,8 +751,9 @@ void RendererOpenGL::RenderScreenshot() { } bool RendererOpenGL::Init() { - if (GLAD_GL_KHR_debug) { + if (Settings::values.renderer_debug && GLAD_GL_KHR_debug) { glEnable(GL_DEBUG_OUTPUT); + glEnable(GL_DEBUG_OUTPUT_SYNCHRONOUS); glDebugMessageCallback(DebugHandler, nullptr); } diff --git a/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp b/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp index 568744e3c..424278816 100644 --- a/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp +++ b/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp @@ -71,8 +71,7 @@ void FixedPipelineState::Rasterizer::Fill(const Maxwell& regs) noexcept { const u32 topology_index = static_cast<u32>(regs.draw.topology.Value()); u32 packed_front_face = PackFrontFace(regs.front_face); - if (regs.screen_y_control.triangle_rast_flip != 0 && - regs.viewport_transform[0].scale_y > 0.0f) { + if (regs.screen_y_control.triangle_rast_flip != 0) { // Flip front face packed_front_face = 1 - packed_front_face; } diff --git a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp index 2871035f5..62e950d31 100644 --- a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp +++ b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp @@ -149,7 +149,7 @@ struct FormatTuple { {VK_FORMAT_R16_SFLOAT, Attachable | Storage}, // R16F {VK_FORMAT_R16_UNORM, Attachable | Storage}, // R16U {VK_FORMAT_UNDEFINED}, // R16S - {VK_FORMAT_UNDEFINED}, // R16UI + {VK_FORMAT_R16_UINT, Attachable | Storage}, // R16UI {VK_FORMAT_UNDEFINED}, // R16I {VK_FORMAT_R16G16_UNORM, Attachable | Storage}, // RG16 {VK_FORMAT_R16G16_SFLOAT, Attachable | Storage}, // RG16F diff --git a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp index 8e1b46277..281bf9ac3 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp @@ -53,8 +53,9 @@ vk::DescriptorSetLayout VKComputePipeline::CreateDescriptorSetLayout() const { }; add_bindings(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, entries.const_buffers.size()); add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, entries.global_buffers.size()); - add_bindings(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, entries.texel_buffers.size()); + add_bindings(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, entries.uniform_texels.size()); add_bindings(VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, entries.samplers.size()); + add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER, entries.storage_texels.size()); add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, entries.images.size()); VkDescriptorSetLayoutCreateInfo ci; diff --git a/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp b/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp index 890fd52cf..9259b618d 100644 --- a/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp +++ b/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp @@ -42,6 +42,7 @@ vk::DescriptorPool* VKDescriptorPool::AllocateNewPool() { {VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, num_sets * 60}, {VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, num_sets * 64}, {VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, num_sets * 64}, + {VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER, num_sets * 64}, {VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, num_sets * 40}}; VkDescriptorPoolCreateInfo ci; diff --git a/src/video_core/renderer_vulkan/vk_device.cpp b/src/video_core/renderer_vulkan/vk_device.cpp index 750e5a0ca..9fd8ac3f6 100644 --- a/src/video_core/renderer_vulkan/vk_device.cpp +++ b/src/video_core/renderer_vulkan/vk_device.cpp @@ -73,76 +73,79 @@ VkFormatFeatureFlags GetFormatFeatures(VkFormatProperties properties, FormatType std::unordered_map<VkFormat, VkFormatProperties> GetFormatProperties( vk::PhysicalDevice physical, const vk::InstanceDispatch& dld) { - static constexpr std::array formats{VK_FORMAT_A8B8G8R8_UNORM_PACK32, - VK_FORMAT_A8B8G8R8_UINT_PACK32, - VK_FORMAT_A8B8G8R8_SNORM_PACK32, - VK_FORMAT_A8B8G8R8_SRGB_PACK32, - VK_FORMAT_B5G6R5_UNORM_PACK16, - VK_FORMAT_A2B10G10R10_UNORM_PACK32, - VK_FORMAT_A1R5G5B5_UNORM_PACK16, - VK_FORMAT_R32G32B32A32_SFLOAT, - VK_FORMAT_R32G32B32A32_UINT, - VK_FORMAT_R32G32_SFLOAT, - VK_FORMAT_R32G32_UINT, - VK_FORMAT_R16G16B16A16_UINT, - VK_FORMAT_R16G16B16A16_SNORM, - VK_FORMAT_R16G16B16A16_UNORM, - VK_FORMAT_R16G16_UNORM, - VK_FORMAT_R16G16_SNORM, - VK_FORMAT_R16G16_SFLOAT, - VK_FORMAT_R16_UNORM, - VK_FORMAT_R8G8B8A8_SRGB, - VK_FORMAT_R8G8_UNORM, - VK_FORMAT_R8G8_SNORM, - VK_FORMAT_R8G8_UINT, - VK_FORMAT_R8_UNORM, - VK_FORMAT_R8_UINT, - VK_FORMAT_B10G11R11_UFLOAT_PACK32, - VK_FORMAT_R32_SFLOAT, - VK_FORMAT_R32_UINT, - VK_FORMAT_R32_SINT, - VK_FORMAT_R16_SFLOAT, - VK_FORMAT_R16G16B16A16_SFLOAT, - VK_FORMAT_B8G8R8A8_UNORM, - VK_FORMAT_B8G8R8A8_SRGB, - VK_FORMAT_R4G4B4A4_UNORM_PACK16, - VK_FORMAT_D32_SFLOAT, - VK_FORMAT_D16_UNORM, - VK_FORMAT_D16_UNORM_S8_UINT, - VK_FORMAT_D24_UNORM_S8_UINT, - VK_FORMAT_D32_SFLOAT_S8_UINT, - VK_FORMAT_BC1_RGBA_UNORM_BLOCK, - VK_FORMAT_BC2_UNORM_BLOCK, - VK_FORMAT_BC3_UNORM_BLOCK, - VK_FORMAT_BC4_UNORM_BLOCK, - VK_FORMAT_BC5_UNORM_BLOCK, - VK_FORMAT_BC5_SNORM_BLOCK, - VK_FORMAT_BC7_UNORM_BLOCK, - VK_FORMAT_BC6H_UFLOAT_BLOCK, - VK_FORMAT_BC6H_SFLOAT_BLOCK, - VK_FORMAT_BC1_RGBA_SRGB_BLOCK, - VK_FORMAT_BC2_SRGB_BLOCK, - VK_FORMAT_BC3_SRGB_BLOCK, - VK_FORMAT_BC7_SRGB_BLOCK, - VK_FORMAT_ASTC_4x4_SRGB_BLOCK, - VK_FORMAT_ASTC_8x8_SRGB_BLOCK, - VK_FORMAT_ASTC_8x5_SRGB_BLOCK, - VK_FORMAT_ASTC_5x4_SRGB_BLOCK, - VK_FORMAT_ASTC_5x5_UNORM_BLOCK, - VK_FORMAT_ASTC_5x5_SRGB_BLOCK, - VK_FORMAT_ASTC_10x8_UNORM_BLOCK, - VK_FORMAT_ASTC_10x8_SRGB_BLOCK, - VK_FORMAT_ASTC_6x6_UNORM_BLOCK, - VK_FORMAT_ASTC_6x6_SRGB_BLOCK, - VK_FORMAT_ASTC_10x10_UNORM_BLOCK, - VK_FORMAT_ASTC_10x10_SRGB_BLOCK, - VK_FORMAT_ASTC_12x12_UNORM_BLOCK, - VK_FORMAT_ASTC_12x12_SRGB_BLOCK, - VK_FORMAT_ASTC_8x6_UNORM_BLOCK, - VK_FORMAT_ASTC_8x6_SRGB_BLOCK, - VK_FORMAT_ASTC_6x5_UNORM_BLOCK, - VK_FORMAT_ASTC_6x5_SRGB_BLOCK, - VK_FORMAT_E5B9G9R9_UFLOAT_PACK32}; + static constexpr std::array formats{ + VK_FORMAT_A8B8G8R8_UNORM_PACK32, + VK_FORMAT_A8B8G8R8_UINT_PACK32, + VK_FORMAT_A8B8G8R8_SNORM_PACK32, + VK_FORMAT_A8B8G8R8_SRGB_PACK32, + VK_FORMAT_B5G6R5_UNORM_PACK16, + VK_FORMAT_A2B10G10R10_UNORM_PACK32, + VK_FORMAT_A1R5G5B5_UNORM_PACK16, + VK_FORMAT_R32G32B32A32_SFLOAT, + VK_FORMAT_R32G32B32A32_UINT, + VK_FORMAT_R32G32_SFLOAT, + VK_FORMAT_R32G32_UINT, + VK_FORMAT_R16G16B16A16_UINT, + VK_FORMAT_R16G16B16A16_SNORM, + VK_FORMAT_R16G16B16A16_UNORM, + VK_FORMAT_R16G16_UNORM, + VK_FORMAT_R16G16_SNORM, + VK_FORMAT_R16G16_SFLOAT, + VK_FORMAT_R16_UNORM, + VK_FORMAT_R16_UINT, + VK_FORMAT_R8G8B8A8_SRGB, + VK_FORMAT_R8G8_UNORM, + VK_FORMAT_R8G8_SNORM, + VK_FORMAT_R8G8_UINT, + VK_FORMAT_R8_UNORM, + VK_FORMAT_R8_UINT, + VK_FORMAT_B10G11R11_UFLOAT_PACK32, + VK_FORMAT_R32_SFLOAT, + VK_FORMAT_R32_UINT, + VK_FORMAT_R32_SINT, + VK_FORMAT_R16_SFLOAT, + VK_FORMAT_R16G16B16A16_SFLOAT, + VK_FORMAT_B8G8R8A8_UNORM, + VK_FORMAT_B8G8R8A8_SRGB, + VK_FORMAT_R4G4B4A4_UNORM_PACK16, + VK_FORMAT_D32_SFLOAT, + VK_FORMAT_D16_UNORM, + VK_FORMAT_D16_UNORM_S8_UINT, + VK_FORMAT_D24_UNORM_S8_UINT, + VK_FORMAT_D32_SFLOAT_S8_UINT, + VK_FORMAT_BC1_RGBA_UNORM_BLOCK, + VK_FORMAT_BC2_UNORM_BLOCK, + VK_FORMAT_BC3_UNORM_BLOCK, + VK_FORMAT_BC4_UNORM_BLOCK, + VK_FORMAT_BC5_UNORM_BLOCK, + VK_FORMAT_BC5_SNORM_BLOCK, + VK_FORMAT_BC7_UNORM_BLOCK, + VK_FORMAT_BC6H_UFLOAT_BLOCK, + VK_FORMAT_BC6H_SFLOAT_BLOCK, + VK_FORMAT_BC1_RGBA_SRGB_BLOCK, + VK_FORMAT_BC2_SRGB_BLOCK, + VK_FORMAT_BC3_SRGB_BLOCK, + VK_FORMAT_BC7_SRGB_BLOCK, + VK_FORMAT_ASTC_4x4_SRGB_BLOCK, + VK_FORMAT_ASTC_8x8_SRGB_BLOCK, + VK_FORMAT_ASTC_8x5_SRGB_BLOCK, + VK_FORMAT_ASTC_5x4_SRGB_BLOCK, + VK_FORMAT_ASTC_5x5_UNORM_BLOCK, + VK_FORMAT_ASTC_5x5_SRGB_BLOCK, + VK_FORMAT_ASTC_10x8_UNORM_BLOCK, + VK_FORMAT_ASTC_10x8_SRGB_BLOCK, + VK_FORMAT_ASTC_6x6_UNORM_BLOCK, + VK_FORMAT_ASTC_6x6_SRGB_BLOCK, + VK_FORMAT_ASTC_10x10_UNORM_BLOCK, + VK_FORMAT_ASTC_10x10_SRGB_BLOCK, + VK_FORMAT_ASTC_12x12_UNORM_BLOCK, + VK_FORMAT_ASTC_12x12_SRGB_BLOCK, + VK_FORMAT_ASTC_8x6_UNORM_BLOCK, + VK_FORMAT_ASTC_8x6_SRGB_BLOCK, + VK_FORMAT_ASTC_6x5_UNORM_BLOCK, + VK_FORMAT_ASTC_6x5_SRGB_BLOCK, + VK_FORMAT_E5B9G9R9_UFLOAT_PACK32, + }; std::unordered_map<VkFormat, VkFormatProperties> format_properties; for (const auto format : formats) { format_properties.emplace(format, physical.GetFormatProperties(format)); diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp index a5c7b7945..b8ccf164f 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp @@ -45,6 +45,7 @@ constexpr VkDescriptorType UNIFORM_BUFFER = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER; constexpr VkDescriptorType STORAGE_BUFFER = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER; constexpr VkDescriptorType UNIFORM_TEXEL_BUFFER = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER; constexpr VkDescriptorType COMBINED_IMAGE_SAMPLER = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER; +constexpr VkDescriptorType STORAGE_TEXEL_BUFFER = VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER; constexpr VkDescriptorType STORAGE_IMAGE = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE; constexpr VideoCommon::Shader::CompilerSettings compiler_settings{ @@ -104,8 +105,9 @@ u32 FillDescriptorLayout(const ShaderEntries& entries, u32 binding = base_binding; AddBindings<UNIFORM_BUFFER>(bindings, binding, flags, entries.const_buffers); AddBindings<STORAGE_BUFFER>(bindings, binding, flags, entries.global_buffers); - AddBindings<UNIFORM_TEXEL_BUFFER>(bindings, binding, flags, entries.texel_buffers); + AddBindings<UNIFORM_TEXEL_BUFFER>(bindings, binding, flags, entries.uniform_texels); AddBindings<COMBINED_IMAGE_SAMPLER>(bindings, binding, flags, entries.samplers); + AddBindings<STORAGE_TEXEL_BUFFER>(bindings, binding, flags, entries.storage_texels); AddBindings<STORAGE_IMAGE>(bindings, binding, flags, entries.images); return binding; } @@ -312,7 +314,9 @@ VKPipelineCache::DecompileShaders(const GraphicsPipelineCacheKey& key) { ASSERT(point_size != 0.0f); } for (std::size_t i = 0; i < Maxwell::NumVertexAttributes; ++i) { - specialization.attribute_types[i] = fixed_state.vertex_input.attributes[i].Type(); + const auto& attribute = fixed_state.vertex_input.attributes[i]; + specialization.enabled_attributes[i] = attribute.enabled.Value() != 0; + specialization.attribute_types[i] = attribute.Type(); } specialization.ndc_minus_one_to_one = fixed_state.rasterizer.ndc_minus_one_to_one; @@ -375,16 +379,17 @@ void AddEntry(std::vector<VkDescriptorUpdateTemplateEntry>& template_entries, u3 return; } - if constexpr (descriptor_type == UNIFORM_TEXEL_BUFFER) { - // Nvidia has a bug where updating multiple uniform texels at once causes the driver to - // crash. + if constexpr (descriptor_type == UNIFORM_TEXEL_BUFFER || + descriptor_type == STORAGE_TEXEL_BUFFER) { + // Nvidia has a bug where updating multiple texels at once causes the driver to crash. + // Note: Fixed in driver Windows 443.24, Linux 440.66.15 for (u32 i = 0; i < count; ++i) { VkDescriptorUpdateTemplateEntry& entry = template_entries.emplace_back(); entry.dstBinding = binding + i; entry.dstArrayElement = 0; entry.descriptorCount = 1; entry.descriptorType = descriptor_type; - entry.offset = offset + i * entry_size; + entry.offset = static_cast<std::size_t>(offset + i * entry_size); entry.stride = entry_size; } } else if (count > 0) { @@ -405,8 +410,9 @@ void FillDescriptorUpdateTemplateEntries( std::vector<VkDescriptorUpdateTemplateEntryKHR>& template_entries) { AddEntry<UNIFORM_BUFFER>(template_entries, offset, binding, entries.const_buffers); AddEntry<STORAGE_BUFFER>(template_entries, offset, binding, entries.global_buffers); - AddEntry<UNIFORM_TEXEL_BUFFER>(template_entries, offset, binding, entries.texel_buffers); + AddEntry<UNIFORM_TEXEL_BUFFER>(template_entries, offset, binding, entries.uniform_texels); AddEntry<COMBINED_IMAGE_SAMPLER>(template_entries, offset, binding, entries.samplers); + AddEntry<STORAGE_TEXEL_BUFFER>(template_entries, offset, binding, entries.storage_texels); AddEntry<STORAGE_IMAGE>(template_entries, offset, binding, entries.images); } diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index be5b77fae..19b8f9da3 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -468,8 +468,9 @@ void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) { const auto& entries = pipeline.GetEntries(); SetupComputeConstBuffers(entries); SetupComputeGlobalBuffers(entries); - SetupComputeTexelBuffers(entries); + SetupComputeUniformTexels(entries); SetupComputeTextures(entries); + SetupComputeStorageTexels(entries); SetupComputeImages(entries); buffer_cache.Unmap(); @@ -715,7 +716,7 @@ std::tuple<VkFramebuffer, VkExtent2D> RasterizerVulkan::ConfigureFramebuffers( if (!view) { return false; } - key.views.push_back(view->GetHandle()); + key.views.push_back(view->GetAttachment()); key.width = std::min(key.width, view->GetWidth()); key.height = std::min(key.height, view->GetHeight()); key.layers = std::min(key.layers, view->GetNumLayers()); @@ -787,8 +788,9 @@ void RasterizerVulkan::SetupShaderDescriptors( const auto& entries = shader->GetEntries(); SetupGraphicsConstBuffers(entries, stage); SetupGraphicsGlobalBuffers(entries, stage); - SetupGraphicsTexelBuffers(entries, stage); + SetupGraphicsUniformTexels(entries, stage); SetupGraphicsTextures(entries, stage); + SetupGraphicsStorageTexels(entries, stage); SetupGraphicsImages(entries, stage); } texture_cache.GuardSamplers(false); @@ -838,6 +840,10 @@ void RasterizerVulkan::BeginTransformFeedback() { if (regs.tfb_enabled == 0) { return; } + if (!device.IsExtTransformFeedbackSupported()) { + LOG_ERROR(Render_Vulkan, "Transform feedbacks used but not supported"); + return; + } UNIMPLEMENTED_IF(regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationControl) || regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationEval) || @@ -866,6 +872,9 @@ void RasterizerVulkan::EndTransformFeedback() { if (regs.tfb_enabled == 0) { return; } + if (!device.IsExtTransformFeedbackSupported()) { + return; + } scheduler.Record( [](vk::CommandBuffer cmdbuf) { cmdbuf.EndTransformFeedbackEXT(0, 0, nullptr, nullptr); }); @@ -877,14 +886,10 @@ void RasterizerVulkan::SetupVertexArrays(FixedPipelineState::VertexInput& vertex for (std::size_t index = 0; index < Maxwell::NumVertexAttributes; ++index) { const auto& attrib = regs.vertex_attrib_format[index]; - if (!attrib.IsValid()) { + if (attrib.IsConstant()) { vertex_input.SetAttribute(index, false, 0, 0, {}, {}); continue; } - - [[maybe_unused]] const auto& buffer = regs.vertex_array[attrib.buffer]; - ASSERT(buffer.IsEnabled()); - vertex_input.SetAttribute(index, true, attrib.buffer, attrib.offset, attrib.type.Value(), attrib.size.Value()); } @@ -980,12 +985,12 @@ void RasterizerVulkan::SetupGraphicsGlobalBuffers(const ShaderEntries& entries, } } -void RasterizerVulkan::SetupGraphicsTexelBuffers(const ShaderEntries& entries, std::size_t stage) { +void RasterizerVulkan::SetupGraphicsUniformTexels(const ShaderEntries& entries, std::size_t stage) { MICROPROFILE_SCOPE(Vulkan_Textures); const auto& gpu = system.GPU().Maxwell3D(); - for (const auto& entry : entries.texel_buffers) { + for (const auto& entry : entries.uniform_texels) { const auto image = GetTextureInfo(gpu, entry, stage).tic; - SetupTexelBuffer(image, entry); + SetupUniformTexels(image, entry); } } @@ -1000,6 +1005,15 @@ void RasterizerVulkan::SetupGraphicsTextures(const ShaderEntries& entries, std:: } } +void RasterizerVulkan::SetupGraphicsStorageTexels(const ShaderEntries& entries, std::size_t stage) { + MICROPROFILE_SCOPE(Vulkan_Textures); + const auto& gpu = system.GPU().Maxwell3D(); + for (const auto& entry : entries.storage_texels) { + const auto image = GetTextureInfo(gpu, entry, stage).tic; + SetupStorageTexel(image, entry); + } +} + void RasterizerVulkan::SetupGraphicsImages(const ShaderEntries& entries, std::size_t stage) { MICROPROFILE_SCOPE(Vulkan_Images); const auto& gpu = system.GPU().Maxwell3D(); @@ -1032,12 +1046,12 @@ void RasterizerVulkan::SetupComputeGlobalBuffers(const ShaderEntries& entries) { } } -void RasterizerVulkan::SetupComputeTexelBuffers(const ShaderEntries& entries) { +void RasterizerVulkan::SetupComputeUniformTexels(const ShaderEntries& entries) { MICROPROFILE_SCOPE(Vulkan_Textures); const auto& gpu = system.GPU().KeplerCompute(); - for (const auto& entry : entries.texel_buffers) { + for (const auto& entry : entries.uniform_texels) { const auto image = GetTextureInfo(gpu, entry, ComputeShaderIndex).tic; - SetupTexelBuffer(image, entry); + SetupUniformTexels(image, entry); } } @@ -1052,6 +1066,15 @@ void RasterizerVulkan::SetupComputeTextures(const ShaderEntries& entries) { } } +void RasterizerVulkan::SetupComputeStorageTexels(const ShaderEntries& entries) { + MICROPROFILE_SCOPE(Vulkan_Textures); + const auto& gpu = system.GPU().KeplerCompute(); + for (const auto& entry : entries.storage_texels) { + const auto image = GetTextureInfo(gpu, entry, ComputeShaderIndex).tic; + SetupStorageTexel(image, entry); + } +} + void RasterizerVulkan::SetupComputeImages(const ShaderEntries& entries) { MICROPROFILE_SCOPE(Vulkan_Images); const auto& gpu = system.GPU().KeplerCompute(); @@ -1101,8 +1124,8 @@ void RasterizerVulkan::SetupGlobalBuffer(const GlobalBufferEntry& entry, GPUVAdd update_descriptor_queue.AddBuffer(buffer, offset, size); } -void RasterizerVulkan::SetupTexelBuffer(const Tegra::Texture::TICEntry& tic, - const TexelBufferEntry& entry) { +void RasterizerVulkan::SetupUniformTexels(const Tegra::Texture::TICEntry& tic, + const UniformTexelEntry& entry) { const auto view = texture_cache.GetTextureSurface(tic, entry); ASSERT(view->IsBufferView()); @@ -1114,8 +1137,8 @@ void RasterizerVulkan::SetupTexture(const Tegra::Texture::FullTextureInfo& textu auto view = texture_cache.GetTextureSurface(texture.tic, entry); ASSERT(!view->IsBufferView()); - const auto image_view = view->GetHandle(texture.tic.x_source, texture.tic.y_source, - texture.tic.z_source, texture.tic.w_source); + const VkImageView image_view = view->GetImageView(texture.tic.x_source, texture.tic.y_source, + texture.tic.z_source, texture.tic.w_source); const auto sampler = sampler_cache.GetSampler(texture.tsc); update_descriptor_queue.AddSampledImage(sampler, image_view); @@ -1124,6 +1147,14 @@ void RasterizerVulkan::SetupTexture(const Tegra::Texture::FullTextureInfo& textu sampled_views.push_back(ImageView{std::move(view), image_layout}); } +void RasterizerVulkan::SetupStorageTexel(const Tegra::Texture::TICEntry& tic, + const StorageTexelEntry& entry) { + const auto view = texture_cache.GetImageSurface(tic, entry); + ASSERT(view->IsBufferView()); + + update_descriptor_queue.AddTexelBuffer(view->GetBufferView()); +} + void RasterizerVulkan::SetupImage(const Tegra::Texture::TICEntry& tic, const ImageEntry& entry) { auto view = texture_cache.GetImageSurface(tic, entry); @@ -1133,7 +1164,8 @@ void RasterizerVulkan::SetupImage(const Tegra::Texture::TICEntry& tic, const Ima UNIMPLEMENTED_IF(tic.IsBuffer()); - const auto image_view = view->GetHandle(tic.x_source, tic.y_source, tic.z_source, tic.w_source); + const VkImageView image_view = + view->GetImageView(tic.x_source, tic.y_source, tic.z_source, tic.w_source); update_descriptor_queue.AddImage(image_view); const auto image_layout = update_descriptor_queue.GetLastImageLayout(); diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h index 0ed0e48c6..04be37a5e 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.h +++ b/src/video_core/renderer_vulkan/vk_rasterizer.h @@ -193,12 +193,15 @@ private: /// Setup global buffers in the graphics pipeline. void SetupGraphicsGlobalBuffers(const ShaderEntries& entries, std::size_t stage); - /// Setup texel buffers in the graphics pipeline. - void SetupGraphicsTexelBuffers(const ShaderEntries& entries, std::size_t stage); + /// Setup uniform texels in the graphics pipeline. + void SetupGraphicsUniformTexels(const ShaderEntries& entries, std::size_t stage); /// Setup textures in the graphics pipeline. void SetupGraphicsTextures(const ShaderEntries& entries, std::size_t stage); + /// Setup storage texels in the graphics pipeline. + void SetupGraphicsStorageTexels(const ShaderEntries& entries, std::size_t stage); + /// Setup images in the graphics pipeline. void SetupGraphicsImages(const ShaderEntries& entries, std::size_t stage); @@ -209,11 +212,14 @@ private: void SetupComputeGlobalBuffers(const ShaderEntries& entries); /// Setup texel buffers in the compute pipeline. - void SetupComputeTexelBuffers(const ShaderEntries& entries); + void SetupComputeUniformTexels(const ShaderEntries& entries); /// Setup textures in the compute pipeline. void SetupComputeTextures(const ShaderEntries& entries); + /// Setup storage texels in the compute pipeline. + void SetupComputeStorageTexels(const ShaderEntries& entries); + /// Setup images in the compute pipeline. void SetupComputeImages(const ShaderEntries& entries); @@ -222,10 +228,12 @@ private: void SetupGlobalBuffer(const GlobalBufferEntry& entry, GPUVAddr address); - void SetupTexelBuffer(const Tegra::Texture::TICEntry& image, const TexelBufferEntry& entry); + void SetupUniformTexels(const Tegra::Texture::TICEntry& image, const UniformTexelEntry& entry); void SetupTexture(const Tegra::Texture::FullTextureInfo& texture, const SamplerEntry& entry); + void SetupStorageTexel(const Tegra::Texture::TICEntry& tic, const StorageTexelEntry& entry); + void SetupImage(const Tegra::Texture::TICEntry& tic, const ImageEntry& entry); void UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& regs); diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp index 890f34a2c..97429cc59 100644 --- a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp +++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp @@ -400,8 +400,9 @@ private: u32 binding = specialization.base_binding; binding = DeclareConstantBuffers(binding); binding = DeclareGlobalBuffers(binding); - binding = DeclareTexelBuffers(binding); + binding = DeclareUniformTexels(binding); binding = DeclareSamplers(binding); + binding = DeclareStorageTexels(binding); binding = DeclareImages(binding); const Id main = OpFunction(t_void, {}, TypeFunction(t_void)); @@ -741,8 +742,10 @@ private: if (!IsGenericAttribute(index)) { continue; } - const u32 location = GetGenericAttributeLocation(index); + if (!IsAttributeEnabled(location)) { + continue; + } const auto type_descriptor = GetAttributeType(location); Id type; if (IsInputAttributeArray()) { @@ -887,7 +890,7 @@ private: return binding; } - u32 DeclareTexelBuffers(u32 binding) { + u32 DeclareUniformTexels(u32 binding) { for (const auto& sampler : ir.GetSamplers()) { if (!sampler.is_buffer) { continue; @@ -908,7 +911,7 @@ private: Decorate(id, spv::Decoration::Binding, binding++); Decorate(id, spv::Decoration::DescriptorSet, DESCRIPTOR_SET); - texel_buffers.emplace(sampler.index, TexelBuffer{image_type, id}); + uniform_texels.emplace(sampler.index, TexelBuffer{image_type, id}); } return binding; } @@ -943,31 +946,48 @@ private: return binding; } - u32 DeclareImages(u32 binding) { + u32 DeclareStorageTexels(u32 binding) { for (const auto& image : ir.GetImages()) { - const auto [dim, arrayed] = GetImageDim(image); - constexpr int depth = 0; - constexpr bool ms = false; - constexpr int sampled = 2; // This won't be accessed with a sampler - constexpr auto format = spv::ImageFormat::Unknown; - const Id image_type = TypeImage(t_uint, dim, depth, arrayed, ms, sampled, format, {}); - const Id pointer_type = TypePointer(spv::StorageClass::UniformConstant, image_type); - const Id id = OpVariable(pointer_type, spv::StorageClass::UniformConstant); - AddGlobalVariable(Name(id, fmt::format("image_{}", image.index))); - - Decorate(id, spv::Decoration::Binding, binding++); - Decorate(id, spv::Decoration::DescriptorSet, DESCRIPTOR_SET); - if (image.is_read && !image.is_written) { - Decorate(id, spv::Decoration::NonWritable); - } else if (image.is_written && !image.is_read) { - Decorate(id, spv::Decoration::NonReadable); + if (image.type != Tegra::Shader::ImageType::TextureBuffer) { + continue; } + DeclareImage(image, binding); + } + return binding; + } - images.emplace(image.index, StorageImage{image_type, id}); + u32 DeclareImages(u32 binding) { + for (const auto& image : ir.GetImages()) { + if (image.type == Tegra::Shader::ImageType::TextureBuffer) { + continue; + } + DeclareImage(image, binding); } return binding; } + void DeclareImage(const Image& image, u32& binding) { + const auto [dim, arrayed] = GetImageDim(image); + constexpr int depth = 0; + constexpr bool ms = false; + constexpr int sampled = 2; // This won't be accessed with a sampler + const auto format = image.is_atomic ? spv::ImageFormat::R32ui : spv::ImageFormat::Unknown; + const Id image_type = TypeImage(t_uint, dim, depth, arrayed, ms, sampled, format, {}); + const Id pointer_type = TypePointer(spv::StorageClass::UniformConstant, image_type); + const Id id = OpVariable(pointer_type, spv::StorageClass::UniformConstant); + AddGlobalVariable(Name(id, fmt::format("image_{}", image.index))); + + Decorate(id, spv::Decoration::Binding, binding++); + Decorate(id, spv::Decoration::DescriptorSet, DESCRIPTOR_SET); + if (image.is_read && !image.is_written) { + Decorate(id, spv::Decoration::NonWritable); + } else if (image.is_written && !image.is_read) { + Decorate(id, spv::Decoration::NonReadable); + } + + images.emplace(image.index, StorageImage{image_type, id}); + } + bool IsRenderTargetEnabled(u32 rt) const { for (u32 component = 0; component < 4; ++component) { if (header.ps.IsColorComponentOutputEnabled(rt, component)) { @@ -986,6 +1006,10 @@ private: return stage == ShaderType::TesselationControl; } + bool IsAttributeEnabled(u32 location) const { + return stage != ShaderType::Vertex || specialization.enabled_attributes[location]; + } + u32 GetNumInputVertices() const { switch (stage) { case ShaderType::Geometry: @@ -1201,16 +1225,20 @@ private: UNIMPLEMENTED_MSG("Unmanaged FrontFacing element={}", element); return {v_float_zero, Type::Float}; default: - if (IsGenericAttribute(attribute)) { - const u32 location = GetGenericAttributeLocation(attribute); - const auto type_descriptor = GetAttributeType(location); - const Type type = type_descriptor.type; - const Id attribute_id = input_attributes.at(attribute); - const std::vector elements = {element}; - const Id pointer = ArrayPass(type_descriptor.scalar, attribute_id, elements); - return {OpLoad(GetTypeDefinition(type), pointer), type}; + if (!IsGenericAttribute(attribute)) { + break; } - break; + const u32 location = GetGenericAttributeLocation(attribute); + if (!IsAttributeEnabled(location)) { + // Disabled attributes (also known as constant attributes) always return zero. + return {v_float_zero, Type::Float}; + } + const auto type_descriptor = GetAttributeType(location); + const Type type = type_descriptor.type; + const Id attribute_id = input_attributes.at(attribute); + const std::vector elements = {element}; + const Id pointer = ArrayPass(type_descriptor.scalar, attribute_id, elements); + return {OpLoad(GetTypeDefinition(type), pointer), type}; } UNIMPLEMENTED_MSG("Unhandled input attribute: {}", static_cast<u32>(attribute)); return {v_float_zero, Type::Float}; @@ -1246,7 +1274,7 @@ private: } else { UNREACHABLE_MSG("Unmanaged offset node type"); } - pointer = OpAccessChain(t_cbuf_float, buffer_id, Constant(t_uint, 0), buffer_index, + pointer = OpAccessChain(t_cbuf_float, buffer_id, v_uint_zero, buffer_index, buffer_element); } return {OpLoad(t_float, pointer), Type::Float}; @@ -1601,7 +1629,7 @@ private: const Id result = OpIAddCarry(TypeStruct({t_uint, t_uint}), op_a, op_b); const Id carry = OpCompositeExtract(t_uint, result, 1); - return {OpINotEqual(t_bool, carry, Constant(t_uint, 0)), Type::Bool}; + return {OpINotEqual(t_bool, carry, v_uint_zero), Type::Bool}; } Expression LogicalAssign(Operation operation) { @@ -1664,7 +1692,7 @@ private: const auto& meta = std::get<MetaTexture>(operation.GetMeta()); const u32 index = meta.sampler.index; if (meta.sampler.is_buffer) { - const auto& entry = texel_buffers.at(index); + const auto& entry = uniform_texels.at(index); return OpLoad(entry.image_type, entry.image); } else { const auto& entry = sampled_images.at(index); @@ -1941,39 +1969,20 @@ private: return {}; } - Expression AtomicImageAdd(Operation operation) { - UNIMPLEMENTED(); - return {}; - } - - Expression AtomicImageMin(Operation operation) { - UNIMPLEMENTED(); - return {}; - } - - Expression AtomicImageMax(Operation operation) { - UNIMPLEMENTED(); - return {}; - } - - Expression AtomicImageAnd(Operation operation) { - UNIMPLEMENTED(); - return {}; - } - - Expression AtomicImageOr(Operation operation) { - UNIMPLEMENTED(); - return {}; - } + template <Id (Module::*func)(Id, Id, Id, Id, Id)> + Expression AtomicImage(Operation operation) { + const auto& meta{std::get<MetaImage>(operation.GetMeta())}; + ASSERT(meta.values.size() == 1); - Expression AtomicImageXor(Operation operation) { - UNIMPLEMENTED(); - return {}; - } + const Id coordinate = GetCoordinates(operation, Type::Int); + const Id image = images.at(meta.image.index).image; + const Id sample = v_uint_zero; + const Id pointer = OpImageTexelPointer(t_image_uint, image, coordinate, sample); - Expression AtomicImageExchange(Operation operation) { - UNIMPLEMENTED(); - return {}; + const Id scope = Constant(t_uint, static_cast<u32>(spv::Scope::Device)); + const Id semantics = v_uint_zero; + const Id value = AsUint(Visit(meta.values[0])); + return {(this->*func)(t_uint, pointer, scope, semantics, value), Type::Uint}; } template <Id (Module::*func)(Id, Id, Id, Id, Id)> @@ -1988,7 +1997,7 @@ private: return {v_float_zero, Type::Float}; } const Id scope = Constant(t_uint, static_cast<u32>(spv::Scope::Device)); - const Id semantics = Constant(t_uint, 0); + const Id semantics = v_uint_zero; const Id value = AsUint(Visit(operation[1])); return {(this->*func)(t_uint, pointer, scope, semantics, value), Type::Uint}; @@ -2215,8 +2224,8 @@ private: return {}; } - Expression MemoryBarrierGL(Operation) { - const auto scope = spv::Scope::Device; + template <spv::Scope scope> + Expression MemoryBarrier(Operation) { const auto semantics = spv::MemorySemanticsMask::AcquireRelease | spv::MemorySemanticsMask::UniformMemory | spv::MemorySemanticsMask::WorkgroupMemory | @@ -2612,11 +2621,11 @@ private: &SPIRVDecompiler::ImageLoad, &SPIRVDecompiler::ImageStore, - &SPIRVDecompiler::AtomicImageAdd, - &SPIRVDecompiler::AtomicImageAnd, - &SPIRVDecompiler::AtomicImageOr, - &SPIRVDecompiler::AtomicImageXor, - &SPIRVDecompiler::AtomicImageExchange, + &SPIRVDecompiler::AtomicImage<&Module::OpAtomicIAdd>, + &SPIRVDecompiler::AtomicImage<&Module::OpAtomicAnd>, + &SPIRVDecompiler::AtomicImage<&Module::OpAtomicOr>, + &SPIRVDecompiler::AtomicImage<&Module::OpAtomicXor>, + &SPIRVDecompiler::AtomicImage<&Module::OpAtomicExchange>, &SPIRVDecompiler::Atomic<&Module::OpAtomicExchange>, &SPIRVDecompiler::Atomic<&Module::OpAtomicIAdd>, @@ -2681,7 +2690,8 @@ private: &SPIRVDecompiler::ShuffleIndexed, &SPIRVDecompiler::Barrier, - &SPIRVDecompiler::MemoryBarrierGL, + &SPIRVDecompiler::MemoryBarrier<spv::Scope::Workgroup>, + &SPIRVDecompiler::MemoryBarrier<spv::Scope::Device>, }; static_assert(operation_decompilers.size() == static_cast<std::size_t>(OperationCode::Amount)); @@ -2757,8 +2767,11 @@ private: Decorate(TypeStruct(t_gmem_array), spv::Decoration::Block), 0, spv::Decoration::Offset, 0); const Id t_gmem_ssbo = TypePointer(spv::StorageClass::StorageBuffer, t_gmem_struct); + const Id t_image_uint = TypePointer(spv::StorageClass::Image, t_uint); + const Id v_float_zero = Constant(t_float, 0.0f); const Id v_float_one = Constant(t_float, 1.0f); + const Id v_uint_zero = Constant(t_uint, 0); // Nvidia uses these defaults for varyings (e.g. position and generic attributes) const Id v_varying_default = @@ -2783,15 +2796,16 @@ private: std::unordered_map<u8, GenericVaryingDescription> output_attributes; std::map<u32, Id> constant_buffers; std::map<GlobalMemoryBase, Id> global_buffers; - std::map<u32, TexelBuffer> texel_buffers; + std::map<u32, TexelBuffer> uniform_texels; std::map<u32, SampledImage> sampled_images; + std::map<u32, TexelBuffer> storage_texels; std::map<u32, StorageImage> images; + std::array<Id, Maxwell::NumRenderTargets> frag_colors{}; Id instance_index{}; Id vertex_index{}; Id base_instance{}; Id base_vertex{}; - std::array<Id, Maxwell::NumRenderTargets> frag_colors{}; Id frag_depth{}; Id frag_coord{}; Id front_facing{}; @@ -3047,13 +3061,17 @@ ShaderEntries GenerateShaderEntries(const VideoCommon::Shader::ShaderIR& ir) { } for (const auto& sampler : ir.GetSamplers()) { if (sampler.is_buffer) { - entries.texel_buffers.emplace_back(sampler); + entries.uniform_texels.emplace_back(sampler); } else { entries.samplers.emplace_back(sampler); } } for (const auto& image : ir.GetImages()) { - entries.images.emplace_back(image); + if (image.type == Tegra::Shader::ImageType::TextureBuffer) { + entries.storage_texels.emplace_back(image); + } else { + entries.images.emplace_back(image); + } } for (const auto& attribute : ir.GetInputAttributes()) { if (IsGenericAttribute(attribute)) { diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.h b/src/video_core/renderer_vulkan/vk_shader_decompiler.h index f4c05ac3c..2b0e90396 100644 --- a/src/video_core/renderer_vulkan/vk_shader_decompiler.h +++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.h @@ -21,8 +21,9 @@ class VKDevice; namespace Vulkan { using Maxwell = Tegra::Engines::Maxwell3D::Regs; -using TexelBufferEntry = VideoCommon::Shader::Sampler; +using UniformTexelEntry = VideoCommon::Shader::Sampler; using SamplerEntry = VideoCommon::Shader::Sampler; +using StorageTexelEntry = VideoCommon::Shader::Image; using ImageEntry = VideoCommon::Shader::Image; constexpr u32 DESCRIPTOR_SET = 0; @@ -66,13 +67,15 @@ private: struct ShaderEntries { u32 NumBindings() const { return static_cast<u32>(const_buffers.size() + global_buffers.size() + - texel_buffers.size() + samplers.size() + images.size()); + uniform_texels.size() + samplers.size() + storage_texels.size() + + images.size()); } std::vector<ConstBufferEntry> const_buffers; std::vector<GlobalBufferEntry> global_buffers; - std::vector<TexelBufferEntry> texel_buffers; + std::vector<UniformTexelEntry> uniform_texels; std::vector<SamplerEntry> samplers; + std::vector<StorageTexelEntry> storage_texels; std::vector<ImageEntry> images; std::set<u32> attributes; std::array<bool, Maxwell::NumClipDistances> clip_distances{}; @@ -88,7 +91,8 @@ struct Specialization final { u32 shared_memory_size{}; // Graphics specific - std::optional<float> point_size{}; + std::optional<float> point_size; + std::bitset<Maxwell::NumVertexAttributes> enabled_attributes; std::array<Maxwell::VertexAttribute::Type, Maxwell::NumVertexAttributes> attribute_types{}; bool ndc_minus_one_to_one{}; }; diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 55f43e61b..430031665 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -100,8 +100,8 @@ vk::Buffer CreateBuffer(const VKDevice& device, const SurfaceParams& params, ci.pNext = nullptr; ci.flags = 0; ci.size = static_cast<VkDeviceSize>(host_memory_size); - ci.usage = VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_SRC_BIT | - VK_BUFFER_USAGE_TRANSFER_DST_BIT; + ci.usage = VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT | + VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT; ci.sharingMode = VK_SHARING_MODE_EXCLUSIVE; ci.queueFamilyIndexCount = 0; ci.pQueueFamilyIndices = nullptr; @@ -167,6 +167,7 @@ VkImageCreateInfo GenerateImageCreateInfo(const VKDevice& device, const SurfaceP ci.extent = {params.width, params.height, 1}; break; case SurfaceTarget::Texture3D: + ci.flags |= VK_IMAGE_CREATE_2D_ARRAY_COMPATIBLE_BIT; ci.extent = {params.width, params.height, params.depth}; break; case SurfaceTarget::TextureBuffer: @@ -176,6 +177,12 @@ VkImageCreateInfo GenerateImageCreateInfo(const VKDevice& device, const SurfaceP return ci; } +u32 EncodeSwizzle(Tegra::Texture::SwizzleSource x_source, Tegra::Texture::SwizzleSource y_source, + Tegra::Texture::SwizzleSource z_source, Tegra::Texture::SwizzleSource w_source) { + return (static_cast<u32>(x_source) << 24) | (static_cast<u32>(y_source) << 16) | + (static_cast<u32>(z_source) << 8) | static_cast<u32>(w_source); +} + } // Anonymous namespace CachedSurface::CachedSurface(Core::System& system, const VKDevice& device, @@ -203,9 +210,11 @@ CachedSurface::CachedSurface(Core::System& system, const VKDevice& device, } // TODO(Rodrigo): Move this to a virtual function. - main_view = CreateViewInner( - ViewParams(params.target, 0, static_cast<u32>(params.GetNumLayers()), 0, params.num_levels), - true); + u32 num_layers = 1; + if (params.is_layered || params.target == SurfaceTarget::Texture3D) { + num_layers = params.depth; + } + main_view = CreateView(ViewParams(params.target, 0, num_layers, 0, params.num_levels)); } CachedSurface::~CachedSurface() = default; @@ -253,12 +262,8 @@ void CachedSurface::DecorateSurfaceName() { } View CachedSurface::CreateView(const ViewParams& params) { - return CreateViewInner(params, false); -} - -View CachedSurface::CreateViewInner(const ViewParams& params, bool is_proxy) { // TODO(Rodrigo): Add name decorations - return views[params] = std::make_shared<CachedSurfaceView>(device, *this, params, is_proxy); + return views[params] = std::make_shared<CachedSurfaceView>(device, *this, params); } void CachedSurface::UploadBuffer(const std::vector<u8>& staging_buffer) { @@ -342,38 +347,44 @@ VkImageSubresourceRange CachedSurface::GetImageSubresourceRange() const { } CachedSurfaceView::CachedSurfaceView(const VKDevice& device, CachedSurface& surface, - const ViewParams& params, bool is_proxy) + const ViewParams& params) : VideoCommon::ViewBase{params}, params{surface.GetSurfaceParams()}, image{surface.GetImageHandle()}, buffer_view{surface.GetBufferViewHandle()}, aspect_mask{surface.GetAspectMask()}, device{device}, surface{surface}, - base_layer{params.base_layer}, num_layers{params.num_layers}, base_level{params.base_level}, - num_levels{params.num_levels}, image_view_type{image ? GetImageViewType(params.target) - : VK_IMAGE_VIEW_TYPE_1D} {} + base_level{params.base_level}, num_levels{params.num_levels}, + image_view_type{image ? GetImageViewType(params.target) : VK_IMAGE_VIEW_TYPE_1D} { + if (image_view_type == VK_IMAGE_VIEW_TYPE_3D) { + base_layer = 0; + num_layers = 1; + base_slice = params.base_layer; + num_slices = params.num_layers; + } else { + base_layer = params.base_layer; + num_layers = params.num_layers; + } +} CachedSurfaceView::~CachedSurfaceView() = default; -VkImageView CachedSurfaceView::GetHandle(SwizzleSource x_source, SwizzleSource y_source, - SwizzleSource z_source, SwizzleSource w_source) { - const u32 swizzle = EncodeSwizzle(x_source, y_source, z_source, w_source); - if (last_image_view && last_swizzle == swizzle) { +VkImageView CachedSurfaceView::GetImageView(SwizzleSource x_source, SwizzleSource y_source, + SwizzleSource z_source, SwizzleSource w_source) { + const u32 new_swizzle = EncodeSwizzle(x_source, y_source, z_source, w_source); + if (last_image_view && last_swizzle == new_swizzle) { return last_image_view; } - last_swizzle = swizzle; + last_swizzle = new_swizzle; - const auto [entry, is_cache_miss] = view_cache.try_emplace(swizzle); + const auto [entry, is_cache_miss] = view_cache.try_emplace(new_swizzle); auto& image_view = entry->second; if (!is_cache_miss) { return last_image_view = *image_view; } - auto swizzle_x = MaxwellToVK::SwizzleSource(x_source); - auto swizzle_y = MaxwellToVK::SwizzleSource(y_source); - auto swizzle_z = MaxwellToVK::SwizzleSource(z_source); - auto swizzle_w = MaxwellToVK::SwizzleSource(w_source); - + std::array swizzle{MaxwellToVK::SwizzleSource(x_source), MaxwellToVK::SwizzleSource(y_source), + MaxwellToVK::SwizzleSource(z_source), MaxwellToVK::SwizzleSource(w_source)}; if (params.pixel_format == VideoCore::Surface::PixelFormat::A1B5G5R5U) { // A1B5G5R5 is implemented as A1R5G5B5, we have to change the swizzle here. - std::swap(swizzle_x, swizzle_z); + std::swap(swizzle[0], swizzle[2]); } // Games can sample depth or stencil values on textures. This is decided by the swizzle value on @@ -395,11 +406,16 @@ VkImageView CachedSurfaceView::GetHandle(SwizzleSource x_source, SwizzleSource y UNIMPLEMENTED(); } - // Vulkan doesn't seem to understand swizzling of a depth stencil image, use identity - swizzle_x = VK_COMPONENT_SWIZZLE_R; - swizzle_y = VK_COMPONENT_SWIZZLE_G; - swizzle_z = VK_COMPONENT_SWIZZLE_B; - swizzle_w = VK_COMPONENT_SWIZZLE_A; + // Make sure we sample the first component + std::transform( + swizzle.begin(), swizzle.end(), swizzle.begin(), [](VkComponentSwizzle component) { + return component == VK_COMPONENT_SWIZZLE_G ? VK_COMPONENT_SWIZZLE_R : component; + }); + } + + if (image_view_type == VK_IMAGE_VIEW_TYPE_3D) { + ASSERT(base_slice == 0); + ASSERT(num_slices == params.depth); } VkImageViewCreateInfo ci; @@ -409,7 +425,7 @@ VkImageView CachedSurfaceView::GetHandle(SwizzleSource x_source, SwizzleSource y ci.image = surface.GetImageHandle(); ci.viewType = image_view_type; ci.format = surface.GetImage().GetFormat(); - ci.components = {swizzle_x, swizzle_y, swizzle_z, swizzle_w}; + ci.components = {swizzle[0], swizzle[1], swizzle[2], swizzle[3]}; ci.subresourceRange.aspectMask = aspect; ci.subresourceRange.baseMipLevel = base_level; ci.subresourceRange.levelCount = num_levels; @@ -420,6 +436,35 @@ VkImageView CachedSurfaceView::GetHandle(SwizzleSource x_source, SwizzleSource y return last_image_view = *image_view; } +VkImageView CachedSurfaceView::GetAttachment() { + if (render_target) { + return *render_target; + } + + VkImageViewCreateInfo ci; + ci.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO; + ci.pNext = nullptr; + ci.flags = 0; + ci.image = surface.GetImageHandle(); + ci.format = surface.GetImage().GetFormat(); + ci.components = {VK_COMPONENT_SWIZZLE_IDENTITY, VK_COMPONENT_SWIZZLE_IDENTITY, + VK_COMPONENT_SWIZZLE_IDENTITY, VK_COMPONENT_SWIZZLE_IDENTITY}; + ci.subresourceRange.aspectMask = aspect_mask; + ci.subresourceRange.baseMipLevel = base_level; + ci.subresourceRange.levelCount = num_levels; + if (image_view_type == VK_IMAGE_VIEW_TYPE_3D) { + ci.viewType = num_slices > 1 ? VK_IMAGE_VIEW_TYPE_2D_ARRAY : VK_IMAGE_VIEW_TYPE_2D; + ci.subresourceRange.baseArrayLayer = base_slice; + ci.subresourceRange.layerCount = num_slices; + } else { + ci.viewType = image_view_type; + ci.subresourceRange.baseArrayLayer = base_layer; + ci.subresourceRange.layerCount = num_layers; + } + render_target = device.GetLogical().CreateImageView(ci); + return *render_target; +} + VKTextureCache::VKTextureCache(Core::System& system, VideoCore::RasterizerInterface& rasterizer, const VKDevice& device, VKResourceManager& resource_manager, VKMemoryManager& memory_manager, VKScheduler& scheduler, diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index f211ccb1e..807e26c8a 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -91,7 +91,6 @@ protected: void DecorateSurfaceName(); View CreateView(const ViewParams& params) override; - View CreateViewInner(const ViewParams& params, bool is_proxy); private: void UploadBuffer(const std::vector<u8>& staging_buffer); @@ -120,23 +119,20 @@ private: class CachedSurfaceView final : public VideoCommon::ViewBase { public: explicit CachedSurfaceView(const VKDevice& device, CachedSurface& surface, - const ViewParams& params, bool is_proxy); + const ViewParams& params); ~CachedSurfaceView(); - VkImageView GetHandle(Tegra::Texture::SwizzleSource x_source, - Tegra::Texture::SwizzleSource y_source, - Tegra::Texture::SwizzleSource z_source, - Tegra::Texture::SwizzleSource w_source); + VkImageView GetImageView(Tegra::Texture::SwizzleSource x_source, + Tegra::Texture::SwizzleSource y_source, + Tegra::Texture::SwizzleSource z_source, + Tegra::Texture::SwizzleSource w_source); + + VkImageView GetAttachment(); bool IsSameSurface(const CachedSurfaceView& rhs) const { return &surface == &rhs.surface; } - VkImageView GetHandle() { - return GetHandle(Tegra::Texture::SwizzleSource::R, Tegra::Texture::SwizzleSource::G, - Tegra::Texture::SwizzleSource::B, Tegra::Texture::SwizzleSource::A); - } - u32 GetWidth() const { return params.GetMipWidth(base_level); } @@ -180,14 +176,6 @@ public: } private: - static u32 EncodeSwizzle(Tegra::Texture::SwizzleSource x_source, - Tegra::Texture::SwizzleSource y_source, - Tegra::Texture::SwizzleSource z_source, - Tegra::Texture::SwizzleSource w_source) { - return (static_cast<u32>(x_source) << 24) | (static_cast<u32>(y_source) << 16) | - (static_cast<u32>(z_source) << 8) | static_cast<u32>(w_source); - } - // Store a copy of these values to avoid double dereference when reading them const SurfaceParams params; const VkImage image; @@ -196,15 +184,18 @@ private: const VKDevice& device; CachedSurface& surface; - const u32 base_layer; - const u32 num_layers; const u32 base_level; const u32 num_levels; const VkImageViewType image_view_type; + u32 base_layer = 0; + u32 num_layers = 0; + u32 base_slice = 0; + u32 num_slices = 0; VkImageView last_image_view = nullptr; u32 last_swizzle = 0; + vk::ImageView render_target; std::unordered_map<u32, vk::ImageView> view_cache; }; diff --git a/src/video_core/shader/decode/other.cpp b/src/video_core/shader/decode/other.cpp index 694b325e1..c0a8f233f 100644 --- a/src/video_core/shader/decode/other.cpp +++ b/src/video_core/shader/decode/other.cpp @@ -83,7 +83,7 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) { return Operation(OperationCode::YNegate); case SystemVariable::InvocationInfo: LOG_WARNING(HW_GPU, "S2R instruction with InvocationInfo is incomplete"); - return Immediate(0U); + return Immediate(0x00ff'0000U); case SystemVariable::WscaleFactorXY: UNIMPLEMENTED_MSG("S2R WscaleFactorXY is not implemented"); return Immediate(0U); @@ -299,9 +299,19 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) { break; } case OpCode::Id::MEMBAR: { - UNIMPLEMENTED_IF(instr.membar.type != Tegra::Shader::MembarType::GL); UNIMPLEMENTED_IF(instr.membar.unknown != Tegra::Shader::MembarUnknown::Default); - bb.push_back(Operation(OperationCode::MemoryBarrierGL)); + const OperationCode type = [instr] { + switch (instr.membar.type) { + case Tegra::Shader::MembarType::CTA: + return OperationCode::MemoryBarrierGroup; + case Tegra::Shader::MembarType::GL: + return OperationCode::MemoryBarrierGlobal; + default: + UNIMPLEMENTED_MSG("MEMBAR type={}", static_cast<int>(instr.membar.type.Value())); + return OperationCode::MemoryBarrierGlobal; + } + }(); + bb.push_back(Operation(type)); break; } case OpCode::Id::DEPBAR: { diff --git a/src/video_core/shader/node.h b/src/video_core/shader/node.h index c06512413..c5e5165ff 100644 --- a/src/video_core/shader/node.h +++ b/src/video_core/shader/node.h @@ -233,8 +233,9 @@ enum class OperationCode { ThreadLtMask, /// () -> uint ShuffleIndexed, /// (uint value, uint index) -> uint - Barrier, /// () -> void - MemoryBarrierGL, /// () -> void + Barrier, /// () -> void + MemoryBarrierGroup, /// () -> void + MemoryBarrierGlobal, /// () -> void Amount, }; diff --git a/src/video_core/texture_cache/format_lookup_table.cpp b/src/video_core/texture_cache/format_lookup_table.cpp index 7032e0059..f476f03b0 100644 --- a/src/video_core/texture_cache/format_lookup_table.cpp +++ b/src/video_core/texture_cache/format_lookup_table.cpp @@ -41,7 +41,7 @@ struct Table { ComponentType alpha_component; bool is_srgb; }; -constexpr std::array<Table, 77> DefinitionTable = {{ +constexpr std::array<Table, 78> DefinitionTable = {{ {TextureFormat::A8R8G8B8, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::ABGR8U}, {TextureFormat::A8R8G8B8, C, SNORM, SNORM, SNORM, SNORM, PixelFormat::ABGR8S}, {TextureFormat::A8R8G8B8, C, UINT, UINT, UINT, UINT, PixelFormat::ABGR8UI}, @@ -98,6 +98,7 @@ constexpr std::array<Table, 77> DefinitionTable = {{ {TextureFormat::ZF32, C, FLOAT, FLOAT, FLOAT, FLOAT, PixelFormat::Z32F}, {TextureFormat::Z16, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::Z16}, {TextureFormat::S8Z24, C, UINT, UNORM, UNORM, UNORM, PixelFormat::S8Z24}, + {TextureFormat::G24R8, C, UINT, UNORM, UNORM, UNORM, PixelFormat::S8Z24}, {TextureFormat::ZF32_X24S8, C, FLOAT, UINT, UNORM, UNORM, PixelFormat::Z32FS8}, {TextureFormat::DXT1, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::DXT1}, diff --git a/src/video_core/texture_cache/surface_base.cpp b/src/video_core/texture_cache/surface_base.cpp index 715f39d0d..94d3a6ae5 100644 --- a/src/video_core/texture_cache/surface_base.cpp +++ b/src/video_core/texture_cache/surface_base.cpp @@ -248,12 +248,11 @@ void SurfaceBaseImpl::FlushBuffer(Tegra::MemoryManager& memory_manager, // Use an extra temporal buffer auto& tmp_buffer = staging_cache.GetBuffer(1); - // Special case for 3D Texture Segments - const bool must_read_current_data = - params.block_depth > 0 && params.target == VideoCore::Surface::SurfaceTarget::Texture2D; tmp_buffer.resize(guest_memory_size); host_ptr = tmp_buffer.data(); - if (must_read_current_data) { + + if (params.target == SurfaceTarget::Texture3D) { + // Special case for 3D texture segments memory_manager.ReadBlockUnsafe(gpu_addr, host_ptr, guest_memory_size); } diff --git a/src/video_core/texture_cache/surface_base.h b/src/video_core/texture_cache/surface_base.h index 79e10ffbb..173f2edba 100644 --- a/src/video_core/texture_cache/surface_base.h +++ b/src/video_core/texture_cache/surface_base.h @@ -217,8 +217,8 @@ public: } bool IsProtected() const { - // Only 3D Slices are to be protected - return is_target && params.block_depth > 0; + // Only 3D slices are to be protected + return is_target && params.target == SurfaceTarget::Texture3D; } bool IsRenderTarget() const { @@ -250,6 +250,11 @@ public: return GetView(ViewParams(overview_params.target, 0, num_layers, 0, params.num_levels)); } + TView Emplace3DView(u32 slice, u32 depth, u32 base_level, u32 num_levels) { + return GetView(ViewParams(VideoCore::Surface::SurfaceTarget::Texture3D, slice, depth, + base_level, num_levels)); + } + std::optional<TView> EmplaceIrregularView(const SurfaceParams& view_params, const GPUVAddr view_addr, const std::size_t candidate_size, const u32 mipmap, @@ -272,8 +277,8 @@ public: std::optional<TView> EmplaceView(const SurfaceParams& view_params, const GPUVAddr view_addr, const std::size_t candidate_size) { if (params.target == SurfaceTarget::Texture3D || - (params.num_levels == 1 && !params.is_layered) || - view_params.target == SurfaceTarget::Texture3D) { + view_params.target == SurfaceTarget::Texture3D || + (params.num_levels == 1 && !params.is_layered)) { return {}; } const auto layer_mipmap{GetLayerMipmap(view_addr)}; diff --git a/src/video_core/texture_cache/surface_params.cpp b/src/video_core/texture_cache/surface_params.cpp index 884fabffe..0b2b2b8c4 100644 --- a/src/video_core/texture_cache/surface_params.cpp +++ b/src/video_core/texture_cache/surface_params.cpp @@ -215,10 +215,19 @@ SurfaceParams SurfaceParams::CreateForFramebuffer(Core::System& system, std::siz params.num_levels = 1; params.emulated_levels = 1; - const bool is_layered = config.layers > 1 && params.block_depth == 0; - params.is_layered = is_layered; - params.depth = is_layered ? config.layers.Value() : 1; - params.target = is_layered ? SurfaceTarget::Texture2DArray : SurfaceTarget::Texture2D; + if (config.memory_layout.is_3d != 0) { + params.depth = config.layers.Value(); + params.is_layered = false; + params.target = SurfaceTarget::Texture3D; + } else if (config.layers > 1) { + params.depth = config.layers.Value(); + params.is_layered = true; + params.target = SurfaceTarget::Texture2DArray; + } else { + params.depth = 1; + params.is_layered = false; + params.target = SurfaceTarget::Texture2D; + } return params; } @@ -237,7 +246,7 @@ SurfaceParams SurfaceParams::CreateForFermiCopySurface( params.width = config.width; params.height = config.height; params.pitch = config.pitch; - // TODO(Rodrigo): Try to guess the surface target from depth and layer parameters + // TODO(Rodrigo): Try to guess texture arrays from parameters params.target = SurfaceTarget::Texture2D; params.depth = 1; params.num_levels = 1; diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index d6efc34b2..b543fc8c0 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -14,6 +14,7 @@ #include <unordered_map> #include <vector> +#include <boost/container/small_vector.hpp> #include <boost/icl/interval_map.hpp> #include <boost/range/iterator_range.hpp> @@ -53,6 +54,7 @@ using RenderTargetConfig = Tegra::Engines::Maxwell3D::Regs::RenderTargetConfig; template <typename TSurface, typename TView> class TextureCache { + using VectorSurface = boost::container::small_vector<TSurface, 1>; public: void InvalidateRegion(VAddr addr, std::size_t size) { @@ -296,30 +298,30 @@ public: const GPUVAddr src_gpu_addr = src_config.Address(); const GPUVAddr dst_gpu_addr = dst_config.Address(); DeduceBestBlit(src_params, dst_params, src_gpu_addr, dst_gpu_addr); - const std::optional<VAddr> dst_cpu_addr = - system.GPU().MemoryManager().GpuToCpuAddress(dst_gpu_addr); - const std::optional<VAddr> src_cpu_addr = - system.GPU().MemoryManager().GpuToCpuAddress(src_gpu_addr); - std::pair<TSurface, TView> dst_surface = - GetSurface(dst_gpu_addr, *dst_cpu_addr, dst_params, true, false); - std::pair<TSurface, TView> src_surface = - GetSurface(src_gpu_addr, *src_cpu_addr, src_params, true, false); - ImageBlit(src_surface.second, dst_surface.second, copy_config); + + const auto& memory_manager = system.GPU().MemoryManager(); + const std::optional<VAddr> dst_cpu_addr = memory_manager.GpuToCpuAddress(dst_gpu_addr); + const std::optional<VAddr> src_cpu_addr = memory_manager.GpuToCpuAddress(src_gpu_addr); + std::pair dst_surface = GetSurface(dst_gpu_addr, *dst_cpu_addr, dst_params, true, false); + TView src_surface = GetSurface(src_gpu_addr, *src_cpu_addr, src_params, true, false).second; + ImageBlit(src_surface, dst_surface.second, copy_config); dst_surface.first->MarkAsModified(true, Tick()); } - TSurface TryFindFramebufferSurface(VAddr addr) { + TSurface TryFindFramebufferSurface(VAddr addr) const { if (!addr) { return nullptr; } const VAddr page = addr >> registry_page_bits; - std::vector<TSurface>& list = registry[page]; - for (auto& surface : list) { - if (surface->GetCpuAddr() == addr) { - return surface; - } + const auto it = registry.find(page); + if (it == registry.end()) { + return nullptr; } - return nullptr; + const auto& list = it->second; + const auto found = std::find_if(list.begin(), list.end(), [addr](const auto& surface) { + return surface->GetCpuAddr() == addr; + }); + return found != list.end() ? *found : nullptr; } u64 Tick() { @@ -498,18 +500,18 @@ private: * @param untopological Indicates to the recycler that the texture has no way * to match the overlaps due to topological reasons. **/ - RecycleStrategy PickStrategy(std::vector<TSurface>& overlaps, const SurfaceParams& params, + RecycleStrategy PickStrategy(VectorSurface& overlaps, const SurfaceParams& params, const GPUVAddr gpu_addr, const MatchTopologyResult untopological) { if (Settings::IsGPULevelExtreme()) { return RecycleStrategy::Flush; } // 3D Textures decision - if (params.block_depth > 1 || params.target == SurfaceTarget::Texture3D) { + if (params.target == SurfaceTarget::Texture3D) { return RecycleStrategy::Flush; } for (const auto& s : overlaps) { const auto& s_params = s->GetSurfaceParams(); - if (s_params.block_depth > 1 || s_params.target == SurfaceTarget::Texture3D) { + if (s_params.target == SurfaceTarget::Texture3D) { return RecycleStrategy::Flush; } } @@ -538,9 +540,8 @@ private: * @param untopological Indicates to the recycler that the texture has no way to match the * overlaps due to topological reasons. **/ - std::pair<TSurface, TView> RecycleSurface(std::vector<TSurface>& overlaps, - const SurfaceParams& params, const GPUVAddr gpu_addr, - const bool preserve_contents, + std::pair<TSurface, TView> RecycleSurface(VectorSurface& overlaps, const SurfaceParams& params, + const GPUVAddr gpu_addr, const bool preserve_contents, const MatchTopologyResult untopological) { const bool do_load = preserve_contents && Settings::IsGPULevelExtreme(); for (auto& surface : overlaps) { @@ -650,47 +651,65 @@ private: * @param params The parameters on the new surface. * @param gpu_addr The starting address of the new surface. **/ - std::optional<std::pair<TSurface, TView>> TryReconstructSurface(std::vector<TSurface>& overlaps, + std::optional<std::pair<TSurface, TView>> TryReconstructSurface(VectorSurface& overlaps, const SurfaceParams& params, - const GPUVAddr gpu_addr) { + GPUVAddr gpu_addr) { if (params.target == SurfaceTarget::Texture3D) { - return {}; + return std::nullopt; } - bool modified = false; + const auto test_modified = [](TSurface& surface) { return surface->IsModified(); }; TSurface new_surface = GetUncachedSurface(gpu_addr, params); - u32 passed_tests = 0; + + if (std::none_of(overlaps.begin(), overlaps.end(), test_modified)) { + LoadSurface(new_surface); + for (const auto& surface : overlaps) { + Unregister(surface); + } + Register(new_surface); + return {{new_surface, new_surface->GetMainView()}}; + } + + std::size_t passed_tests = 0; for (auto& surface : overlaps) { const SurfaceParams& src_params = surface->GetSurfaceParams(); - if (src_params.is_layered || src_params.num_levels > 1) { - // We send this cases to recycle as they are more complex to handle - return {}; - } - const std::size_t candidate_size = surface->GetSizeInBytes(); - auto mipmap_layer{new_surface->GetLayerMipmap(surface->GetGpuAddr())}; + const auto mipmap_layer{new_surface->GetLayerMipmap(surface->GetGpuAddr())}; if (!mipmap_layer) { continue; } - const auto [layer, mipmap] = *mipmap_layer; - if (new_surface->GetMipmapSize(mipmap) != candidate_size) { + const auto [base_layer, base_mipmap] = *mipmap_layer; + if (new_surface->GetMipmapSize(base_mipmap) != surface->GetMipmapSize(0)) { continue; } - modified |= surface->IsModified(); - // Now we got all the data set up - const u32 width = SurfaceParams::IntersectWidth(src_params, params, 0, mipmap); - const u32 height = SurfaceParams::IntersectHeight(src_params, params, 0, mipmap); - const CopyParams copy_params(0, 0, 0, 0, 0, layer, 0, mipmap, width, height, 1); - passed_tests++; - ImageCopy(surface, new_surface, copy_params); + ++passed_tests; + + // Copy all mipmaps and layers + const u32 block_width = params.GetDefaultBlockWidth(); + const u32 block_height = params.GetDefaultBlockHeight(); + for (u32 mipmap = base_mipmap; mipmap < base_mipmap + src_params.num_levels; ++mipmap) { + const u32 width = SurfaceParams::IntersectWidth(src_params, params, 0, mipmap); + const u32 height = SurfaceParams::IntersectHeight(src_params, params, 0, mipmap); + if (width < block_width || height < block_height) { + // Current APIs forbid copying small compressed textures, avoid errors + break; + } + const CopyParams copy_params(0, 0, 0, 0, 0, base_layer, 0, mipmap, width, height, + src_params.depth); + ImageCopy(surface, new_surface, copy_params); + } } if (passed_tests == 0) { - return {}; + return std::nullopt; + } + if (Settings::IsGPULevelExtreme() && passed_tests != overlaps.size()) { // In Accurate GPU all tests should pass, else we recycle - } else if (Settings::IsGPULevelExtreme() && passed_tests != overlaps.size()) { - return {}; + return std::nullopt; } + + const bool modified = std::any_of(overlaps.begin(), overlaps.end(), test_modified); for (const auto& surface : overlaps) { Unregister(surface); } + new_surface->MarkAsModified(modified, Tick()); Register(new_surface); return {{new_surface, new_surface->GetMainView()}}; @@ -708,53 +727,11 @@ private: * @param preserve_contents Indicates that the new surface should be loaded from memory or * left blank. */ - std::optional<std::pair<TSurface, TView>> Manage3DSurfaces(std::vector<TSurface>& overlaps, + std::optional<std::pair<TSurface, TView>> Manage3DSurfaces(VectorSurface& overlaps, const SurfaceParams& params, - const GPUVAddr gpu_addr, - const VAddr cpu_addr, + GPUVAddr gpu_addr, VAddr cpu_addr, bool preserve_contents) { - if (params.target == SurfaceTarget::Texture3D) { - bool failed = false; - if (params.num_levels > 1) { - // We can't handle mipmaps in 3D textures yet, better fallback to LLE approach - return std::nullopt; - } - TSurface new_surface = GetUncachedSurface(gpu_addr, params); - bool modified = false; - for (auto& surface : overlaps) { - const SurfaceParams& src_params = surface->GetSurfaceParams(); - if (src_params.target != SurfaceTarget::Texture2D) { - failed = true; - break; - } - if (src_params.height != params.height) { - failed = true; - break; - } - if (src_params.block_depth != params.block_depth || - src_params.block_height != params.block_height) { - failed = true; - break; - } - const u32 offset = static_cast<u32>(surface->GetCpuAddr() - cpu_addr); - const auto offsets = params.GetBlockOffsetXYZ(offset); - const auto z = std::get<2>(offsets); - modified |= surface->IsModified(); - const CopyParams copy_params(0, 0, 0, 0, 0, z, 0, 0, params.width, params.height, - 1); - ImageCopy(surface, new_surface, copy_params); - } - if (failed) { - return std::nullopt; - } - for (const auto& surface : overlaps) { - Unregister(surface); - } - new_surface->MarkAsModified(modified, Tick()); - Register(new_surface); - auto view = new_surface->GetMainView(); - return {{std::move(new_surface), view}}; - } else { + if (params.target != SurfaceTarget::Texture3D) { for (const auto& surface : overlaps) { if (!surface->MatchTarget(params.target)) { if (overlaps.size() == 1 && surface->GetCpuAddr() == cpu_addr) { @@ -770,11 +747,60 @@ private: continue; } if (surface->MatchesStructure(params) == MatchStructureResult::FullMatch) { - return {{surface, surface->GetMainView()}}; + return std::make_pair(surface, surface->GetMainView()); } } return InitializeSurface(gpu_addr, params, preserve_contents); } + + if (params.num_levels > 1) { + // We can't handle mipmaps in 3D textures yet, better fallback to LLE approach + return std::nullopt; + } + + if (overlaps.size() == 1) { + const auto& surface = overlaps[0]; + const SurfaceParams& overlap_params = surface->GetSurfaceParams(); + // Don't attempt to render to textures with more than one level for now + // The texture has to be to the right or the sample address if we want to render to it + if (overlap_params.num_levels == 1 && cpu_addr >= surface->GetCpuAddr()) { + const u32 offset = static_cast<u32>(cpu_addr - surface->GetCpuAddr()); + const u32 slice = std::get<2>(params.GetBlockOffsetXYZ(offset)); + if (slice < overlap_params.depth) { + auto view = surface->Emplace3DView(slice, params.depth, 0, 1); + return std::make_pair(std::move(surface), std::move(view)); + } + } + } + + TSurface new_surface = GetUncachedSurface(gpu_addr, params); + bool modified = false; + + for (auto& surface : overlaps) { + const SurfaceParams& src_params = surface->GetSurfaceParams(); + if (src_params.target != SurfaceTarget::Texture2D || + src_params.height != params.height || + src_params.block_depth != params.block_depth || + src_params.block_height != params.block_height) { + return std::nullopt; + } + modified |= surface->IsModified(); + + const u32 offset = static_cast<u32>(surface->GetCpuAddr() - cpu_addr); + const u32 slice = std::get<2>(params.GetBlockOffsetXYZ(offset)); + const u32 width = params.width; + const u32 height = params.height; + const CopyParams copy_params(0, 0, 0, 0, 0, slice, 0, 0, width, height, 1); + ImageCopy(surface, new_surface, copy_params); + } + for (const auto& surface : overlaps) { + Unregister(surface); + } + new_surface->MarkAsModified(modified, Tick()); + Register(new_surface); + + TView view = new_surface->GetMainView(); + return std::make_pair(std::move(new_surface), std::move(view)); } /** @@ -810,7 +836,7 @@ private: TSurface& current_surface = iter->second; const auto topological_result = current_surface->MatchesTopology(params); if (topological_result != MatchTopologyResult::FullMatch) { - std::vector<TSurface> overlaps{current_surface}; + VectorSurface overlaps{current_surface}; return RecycleSurface(overlaps, params, gpu_addr, preserve_contents, topological_result); } @@ -852,7 +878,7 @@ private: } } - // Check if it's a 3D texture + // Manage 3D textures if (params.block_depth > 0) { auto surface = Manage3DSurfaces(overlaps, params, gpu_addr, cpu_addr, preserve_contents); @@ -868,12 +894,9 @@ private: // two things either the candidate surface is a supertexture of the overlap // or they don't match in any known way. if (!current_surface->IsInside(gpu_addr, gpu_addr + candidate_size)) { - if (current_surface->GetGpuAddr() == gpu_addr) { - std::optional<std::pair<TSurface, TView>> view = - TryReconstructSurface(overlaps, params, gpu_addr); - if (view) { - return *view; - } + const std::optional view = TryReconstructSurface(overlaps, params, gpu_addr); + if (view) { + return *view; } return RecycleSurface(overlaps, params, gpu_addr, preserve_contents, MatchTopologyResult::FullMatch); @@ -991,7 +1014,9 @@ private: params.target = target; params.is_tiled = false; params.srgb_conversion = false; - params.is_layered = false; + params.is_layered = + target == SurfaceTarget::Texture1DArray || target == SurfaceTarget::Texture2DArray || + target == SurfaceTarget::TextureCubemap || target == SurfaceTarget::TextureCubeArray; params.block_width = 0; params.block_height = 0; params.block_depth = 0; @@ -1124,23 +1149,25 @@ private: } } - std::vector<TSurface> GetSurfacesInRegion(const VAddr cpu_addr, const std::size_t size) { + VectorSurface GetSurfacesInRegion(const VAddr cpu_addr, const std::size_t size) { if (size == 0) { return {}; } const VAddr cpu_addr_end = cpu_addr + size; - VAddr start = cpu_addr >> registry_page_bits; const VAddr end = (cpu_addr_end - 1) >> registry_page_bits; - std::vector<TSurface> surfaces; - while (start <= end) { - std::vector<TSurface>& list = registry[start]; - for (auto& surface : list) { - if (!surface->IsPicked() && surface->Overlaps(cpu_addr, cpu_addr_end)) { - surface->MarkAsPicked(true); - surfaces.push_back(surface); + VectorSurface surfaces; + for (VAddr start = cpu_addr >> registry_page_bits; start <= end; ++start) { + const auto it = registry.find(start); + if (it == registry.end()) { + continue; + } + for (auto& surface : it->second) { + if (surface->IsPicked() || !surface->Overlaps(cpu_addr, cpu_addr_end)) { + continue; } + surface->MarkAsPicked(true); + surfaces.push_back(surface); } - start++; } for (auto& surface : surfaces) { surface->MarkAsPicked(false); diff --git a/src/yuzu/bootmanager.cpp b/src/yuzu/bootmanager.cpp index 1adf8932b..1f5e43043 100644 --- a/src/yuzu/bootmanager.cpp +++ b/src/yuzu/bootmanager.cpp @@ -106,6 +106,9 @@ public: format.setVersion(4, 3); format.setProfile(QSurfaceFormat::CompatibilityProfile); format.setOption(QSurfaceFormat::FormatOption::DeprecatedFunctions); + if (Settings::values.renderer_debug) { + format.setOption(QSurfaceFormat::FormatOption::DebugContext); + } // TODO: expose a setting for buffer value (ie default/single/double/triple) format.setSwapBehavior(QSurfaceFormat::DefaultSwapBehavior); format.setSwapInterval(0); diff --git a/src/yuzu/configuration/config.cpp b/src/yuzu/configuration/config.cpp index b08b87426..7e9073cc3 100644 --- a/src/yuzu/configuration/config.cpp +++ b/src/yuzu/configuration/config.cpp @@ -533,6 +533,8 @@ void Config::ReadDebuggingValues() { Settings::values.quest_flag = ReadSetting(QStringLiteral("quest_flag"), false).toBool(); Settings::values.disable_cpu_opt = ReadSetting(QStringLiteral("disable_cpu_opt"), false).toBool(); + Settings::values.disable_macro_jit = + ReadSetting(QStringLiteral("disable_macro_jit"), false).toBool(); qt_config->endGroup(); } @@ -1011,6 +1013,7 @@ void Config::SaveDebuggingValues() { WriteSetting(QStringLiteral("dump_nso"), Settings::values.dump_nso, false); WriteSetting(QStringLiteral("quest_flag"), Settings::values.quest_flag, false); WriteSetting(QStringLiteral("disable_cpu_opt"), Settings::values.disable_cpu_opt, false); + WriteSetting(QStringLiteral("disable_macro_jit"), Settings::values.disable_macro_jit, false); qt_config->endGroup(); } diff --git a/src/yuzu/configuration/configure_debug.cpp b/src/yuzu/configuration/configure_debug.cpp index c2026763e..2c77441fd 100644 --- a/src/yuzu/configuration/configure_debug.cpp +++ b/src/yuzu/configuration/configure_debug.cpp @@ -39,6 +39,8 @@ void ConfigureDebug::SetConfiguration() { ui->disable_cpu_opt->setChecked(Settings::values.disable_cpu_opt); ui->enable_graphics_debugging->setEnabled(!Core::System::GetInstance().IsPoweredOn()); ui->enable_graphics_debugging->setChecked(Settings::values.renderer_debug); + ui->disable_macro_jit->setEnabled(!Core::System::GetInstance().IsPoweredOn()); + ui->disable_macro_jit->setChecked(Settings::values.disable_macro_jit); } void ConfigureDebug::ApplyConfiguration() { @@ -51,6 +53,7 @@ void ConfigureDebug::ApplyConfiguration() { Settings::values.quest_flag = ui->quest_flag->isChecked(); Settings::values.disable_cpu_opt = ui->disable_cpu_opt->isChecked(); Settings::values.renderer_debug = ui->enable_graphics_debugging->isChecked(); + Settings::values.disable_macro_jit = ui->disable_macro_jit->isChecked(); Debugger::ToggleConsole(); Log::Filter filter; filter.ParseFilterString(Settings::values.log_filter); diff --git a/src/yuzu/configuration/configure_debug.ui b/src/yuzu/configuration/configure_debug.ui index e0d4c4a44..46f0208c6 100644 --- a/src/yuzu/configuration/configure_debug.ui +++ b/src/yuzu/configuration/configure_debug.ui @@ -148,6 +148,19 @@ </property> </widget> </item> + <item> + <widget class="QCheckBox" name="disable_macro_jit"> + <property name="enabled"> + <bool>true</bool> + </property> + <property name="whatsThis"> + <string>When checked, it disables the macro Just In Time compiler. Enabled this makes games run slower</string> + </property> + <property name="text"> + <string>Disable Macro JIT</string> + </property> + </widget> + </item> </layout> </widget> </item> diff --git a/src/yuzu/configuration/configure_input_player.cpp b/src/yuzu/configuration/configure_input_player.cpp index e4eb5594b..a05fa64ba 100644 --- a/src/yuzu/configuration/configure_input_player.cpp +++ b/src/yuzu/configuration/configure_input_player.cpp @@ -480,7 +480,9 @@ void ConfigureInputPlayer::RestoreDefaults() { SetAnalogButton(params, analogs_param[analog_id], analog_sub_buttons[sub_button_id]); } } + UpdateButtonLabels(); + ApplyConfiguration(); } void ConfigureInputPlayer::ClearAll() { @@ -505,6 +507,7 @@ void ConfigureInputPlayer::ClearAll() { } UpdateButtonLabels(); + ApplyConfiguration(); } void ConfigureInputPlayer::UpdateButtonLabels() { diff --git a/src/yuzu_cmd/config.cpp b/src/yuzu_cmd/config.cpp index c20d48c42..7240270f5 100644 --- a/src/yuzu_cmd/config.cpp +++ b/src/yuzu_cmd/config.cpp @@ -432,6 +432,8 @@ void Config::ReadValues() { Settings::values.quest_flag = sdl2_config->GetBoolean("Debugging", "quest_flag", false); Settings::values.disable_cpu_opt = sdl2_config->GetBoolean("Debugging", "disable_cpu_opt", false); + Settings::values.disable_macro_jit = + sdl2_config->GetBoolean("Debugging", "disable_macro_jit", false); const auto title_list = sdl2_config->Get("AddOns", "title_ids", ""); std::stringstream ss(title_list); diff --git a/src/yuzu_cmd/default_ini.h b/src/yuzu_cmd/default_ini.h index abc6e6e65..6f53e9659 100644 --- a/src/yuzu_cmd/default_ini.h +++ b/src/yuzu_cmd/default_ini.h @@ -291,6 +291,8 @@ quest_flag = # Determines whether or not JIT CPU optimizations are enabled # false: Optimizations Enabled, true: Optimizations Disabled disable_cpu_opt = +# Enables/Disables the macro JIT compiler +disable_macro_jit=false [WebService] # Whether or not to enable telemetry diff --git a/src/yuzu_cmd/emu_window/emu_window_sdl2_gl.cpp b/src/yuzu_cmd/emu_window/emu_window_sdl2_gl.cpp index 411e7e647..09cc0a3b5 100644 --- a/src/yuzu_cmd/emu_window/emu_window_sdl2_gl.cpp +++ b/src/yuzu_cmd/emu_window/emu_window_sdl2_gl.cpp @@ -98,6 +98,9 @@ EmuWindow_SDL2_GL::EmuWindow_SDL2_GL(Core::System& system, bool fullscreen) SDL_GL_SetAttribute(SDL_GL_BLUE_SIZE, 8); SDL_GL_SetAttribute(SDL_GL_ALPHA_SIZE, 0); SDL_GL_SetAttribute(SDL_GL_SHARE_WITH_CURRENT_CONTEXT, 1); + if (Settings::values.renderer_debug) { + SDL_GL_SetAttribute(SDL_GL_CONTEXT_FLAGS, SDL_GL_CONTEXT_DEBUG_FLAG); + } SDL_GL_SetSwapInterval(0); std::string window_title = fmt::format("yuzu {} | {}-{}", Common::g_build_fullname, |