diff options
Diffstat (limited to '')
46 files changed, 1723 insertions, 781 deletions
diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt index a780215c1..0406fbcd9 100644 --- a/src/video_core/CMakeLists.txt +++ b/src/video_core/CMakeLists.txt @@ -1,6 +1,6 @@ add_library(video_core STATIC - command_processor.cpp - command_processor.h + dma_pusher.cpp + dma_pusher.h debug_utils/debug_utils.cpp debug_utils/debug_utils.h engines/fermi_2d.cpp @@ -21,6 +21,8 @@ add_library(video_core STATIC macro_interpreter.h memory_manager.cpp memory_manager.h + morton.cpp + morton.h rasterizer_cache.cpp rasterizer_cache.h rasterizer_interface.h @@ -62,7 +64,6 @@ add_library(video_core STATIC textures/decoders.cpp textures/decoders.h textures/texture.h - utils.h video_core.cpp video_core.h ) diff --git a/src/video_core/command_processor.cpp b/src/video_core/command_processor.cpp index 28e8c13aa..8b9c548cc 100644 --- a/src/video_core/command_processor.cpp +++ b/src/video_core/command_processor.cpp @@ -34,6 +34,9 @@ MICROPROFILE_DEFINE(ProcessCommandLists, "GPU", "Execute command buffer", MP_RGB void GPU::ProcessCommandLists(const std::vector<CommandListHeader>& commands) { MICROPROFILE_SCOPE(ProcessCommandLists); + // On entering GPU code, assume all memory may be touched by the ARM core. + maxwell_3d->dirty_flags.OnMemoryWrite(); + auto WriteReg = [this](u32 method, u32 subchannel, u32 value, u32 remaining_params) { LOG_TRACE(HW_GPU, "Processing method {:08X} on subchannel {} value " diff --git a/src/video_core/command_processor.h b/src/video_core/command_processor.h deleted file mode 100644 index bd766e77a..000000000 --- a/src/video_core/command_processor.h +++ /dev/null @@ -1,53 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <type_traits> -#include "common/bit_field.h" -#include "common/common_types.h" -#include "video_core/memory_manager.h" - -namespace Tegra { - -enum class SubmissionMode : u32 { - IncreasingOld = 0, - Increasing = 1, - NonIncreasingOld = 2, - NonIncreasing = 3, - Inline = 4, - IncreaseOnce = 5 -}; - -struct CommandListHeader { - u32 entry0; // gpu_va_lo - union { - u32 entry1; // gpu_va_hi | (unk_0x02 << 0x08) | (size << 0x0A) | (unk_0x01 << 0x1F) - BitField<0, 8, u32> gpu_va_hi; - BitField<8, 2, u32> unk1; - BitField<10, 21, u32> sz; - BitField<31, 1, u32> unk2; - }; - - GPUVAddr Address() const { - return (static_cast<GPUVAddr>(gpu_va_hi) << 32) | entry0; - } -}; -static_assert(sizeof(CommandListHeader) == 8, "CommandListHeader is incorrect size"); - -union CommandHeader { - u32 hex; - - BitField<0, 13, u32> method; - BitField<13, 3, u32> subchannel; - - BitField<16, 13, u32> arg_count; - BitField<16, 13, u32> inline_data; - - BitField<29, 3, SubmissionMode> mode; -}; -static_assert(std::is_standard_layout_v<CommandHeader>, "CommandHeader is not standard layout"); -static_assert(sizeof(CommandHeader) == sizeof(u32), "CommandHeader has incorrect size!"); - -} // namespace Tegra diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp new file mode 100644 index 000000000..63a958f11 --- /dev/null +++ b/src/video_core/dma_pusher.cpp @@ -0,0 +1,123 @@ +// Copyright 2018 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include "common/microprofile.h" +#include "core/core.h" +#include "core/memory.h" +#include "video_core/dma_pusher.h" +#include "video_core/engines/maxwell_3d.h" +#include "video_core/gpu.h" + +namespace Tegra { + +DmaPusher::DmaPusher(GPU& gpu) : gpu(gpu) {} + +DmaPusher::~DmaPusher() = default; + +MICROPROFILE_DEFINE(DispatchCalls, "GPU", "Execute command buffer", MP_RGB(128, 128, 192)); + +void DmaPusher::DispatchCalls() { + MICROPROFILE_SCOPE(DispatchCalls); + + // On entering GPU code, assume all memory may be touched by the ARM core. + gpu.Maxwell3D().dirty_flags.OnMemoryWrite(); + + dma_pushbuffer_subindex = 0; + + while (Core::System::GetInstance().IsPoweredOn()) { + if (!Step()) { + break; + } + } +} + +bool DmaPusher::Step() { + if (dma_get != dma_put) { + // Push buffer non-empty, read a word + const CommandHeader command_header{ + Memory::Read32(*gpu.MemoryManager().GpuToCpuAddress(dma_get))}; + + dma_get += sizeof(u32); + + if (!non_main) { + dma_mget = dma_get; + } + + // now, see if we're in the middle of a command + if (dma_state.length_pending) { + // Second word of long non-inc methods command - method count + dma_state.length_pending = 0; + dma_state.method_count = command_header.method_count_; + } else if (dma_state.method_count) { + // Data word of methods command + CallMethod(command_header.argument); + + if (!dma_state.non_incrementing) { + dma_state.method++; + } + + if (dma_increment_once) { + dma_state.non_incrementing = true; + } + + dma_state.method_count--; + } else { + // No command active - this is the first word of a new one + switch (command_header.mode) { + case SubmissionMode::Increasing: + SetState(command_header); + dma_state.non_incrementing = false; + dma_increment_once = false; + break; + case SubmissionMode::NonIncreasing: + SetState(command_header); + dma_state.non_incrementing = true; + dma_increment_once = false; + break; + case SubmissionMode::Inline: + dma_state.method = command_header.method; + dma_state.subchannel = command_header.subchannel; + CallMethod(command_header.arg_count); + dma_state.non_incrementing = true; + dma_increment_once = false; + break; + case SubmissionMode::IncreaseOnce: + SetState(command_header); + dma_state.non_incrementing = false; + dma_increment_once = true; + break; + } + } + } else if (ib_enable && !dma_pushbuffer.empty()) { + // Current pushbuffer empty, but we have more IB entries to read + const CommandList& command_list{dma_pushbuffer.front()}; + const CommandListHeader& command_list_header{command_list[dma_pushbuffer_subindex++]}; + dma_get = command_list_header.addr; + dma_put = dma_get + command_list_header.size * sizeof(u32); + non_main = command_list_header.is_non_main; + + if (dma_pushbuffer_subindex >= command_list.size()) { + // We've gone through the current list, remove it from the queue + dma_pushbuffer.pop(); + dma_pushbuffer_subindex = 0; + } + } else { + // Otherwise, pushbuffer empty and IB empty or nonexistent - nothing to do + return {}; + } + + return true; +} + +void DmaPusher::SetState(const CommandHeader& command_header) { + dma_state.method = command_header.method; + dma_state.subchannel = command_header.subchannel; + dma_state.method_count = command_header.method_count; +} + +void DmaPusher::CallMethod(u32 argument) const { + gpu.CallMethod({dma_state.method, argument, dma_state.subchannel, dma_state.method_count}); +} + +} // namespace Tegra diff --git a/src/video_core/dma_pusher.h b/src/video_core/dma_pusher.h new file mode 100644 index 000000000..16e0697c4 --- /dev/null +++ b/src/video_core/dma_pusher.h @@ -0,0 +1,99 @@ +// Copyright 2018 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <vector> +#include <queue> + +#include "common/bit_field.h" +#include "common/common_types.h" +#include "video_core/memory_manager.h" + +namespace Tegra { + +enum class SubmissionMode : u32 { + IncreasingOld = 0, + Increasing = 1, + NonIncreasingOld = 2, + NonIncreasing = 3, + Inline = 4, + IncreaseOnce = 5 +}; + +struct CommandListHeader { + union { + u64 raw; + BitField<0, 40, GPUVAddr> addr; + BitField<41, 1, u64> is_non_main; + BitField<42, 21, u64> size; + }; +}; +static_assert(sizeof(CommandListHeader) == sizeof(u64), "CommandListHeader is incorrect size"); + +union CommandHeader { + u32 argument; + BitField<0, 13, u32> method; + BitField<0, 24, u32> method_count_; + BitField<13, 3, u32> subchannel; + BitField<16, 13, u32> arg_count; + BitField<16, 13, u32> method_count; + BitField<29, 3, SubmissionMode> mode; +}; +static_assert(std::is_standard_layout_v<CommandHeader>, "CommandHeader is not standard layout"); +static_assert(sizeof(CommandHeader) == sizeof(u32), "CommandHeader has incorrect size!"); + +class GPU; + +using CommandList = std::vector<Tegra::CommandListHeader>; + +/** + * The DmaPusher class implements DMA submission to FIFOs, providing an area of memory that the + * emulated app fills with commands and tells PFIFO to process. The pushbuffers are then assembled + * into a "command stream" consisting of 32-bit words that make up "commands". + * See https://envytools.readthedocs.io/en/latest/hw/fifo/dma-pusher.html#fifo-dma-pusher for + * details on this implementation. + */ +class DmaPusher { +public: + explicit DmaPusher(GPU& gpu); + ~DmaPusher(); + + void Push(CommandList&& entries) { + dma_pushbuffer.push(std::move(entries)); + } + + void DispatchCalls(); + +private: + bool Step(); + + void SetState(const CommandHeader& command_header); + + void CallMethod(u32 argument) const; + + GPU& gpu; + + std::queue<CommandList> dma_pushbuffer; ///< Queue of command lists to be processed + std::size_t dma_pushbuffer_subindex{}; ///< Index within a command list within the pushbuffer + + struct DmaState { + u32 method; ///< Current method + u32 subchannel; ///< Current subchannel + u32 method_count; ///< Current method count + u32 length_pending; ///< Large NI command length pending + bool non_incrementing; ///< Current command’s NI flag + }; + + DmaState dma_state{}; + bool dma_increment_once{}; + + GPUVAddr dma_put{}; ///< pushbuffer current end address + GPUVAddr dma_get{}; ///< pushbuffer current read address + GPUVAddr dma_mget{}; ///< main pushbuffer last read address + bool ib_enable{true}; ///< IB mode enabled + bool non_main{}; ///< non-main pushbuffer active +}; + +} // namespace Tegra diff --git a/src/video_core/engines/fermi_2d.cpp b/src/video_core/engines/fermi_2d.cpp index 74e44c7fe..80f70e332 100644 --- a/src/video_core/engines/fermi_2d.cpp +++ b/src/video_core/engines/fermi_2d.cpp @@ -2,8 +2,10 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. +#include "core/core.h" #include "core/memory.h" #include "video_core/engines/fermi_2d.h" +#include "video_core/engines/maxwell_3d.h" #include "video_core/rasterizer_interface.h" #include "video_core/textures/decoders.h" @@ -12,13 +14,13 @@ namespace Tegra::Engines { Fermi2D::Fermi2D(VideoCore::RasterizerInterface& rasterizer, MemoryManager& memory_manager) : memory_manager(memory_manager), rasterizer{rasterizer} {} -void Fermi2D::WriteReg(u32 method, u32 value) { - ASSERT_MSG(method < Regs::NUM_REGS, +void Fermi2D::CallMethod(const GPU::MethodCall& method_call) { + ASSERT_MSG(method_call.method < Regs::NUM_REGS, "Invalid Fermi2D register, increase the size of the Regs structure"); - regs.reg_array[method] = value; + regs.reg_array[method_call.method] = method_call.argument; - switch (method) { + switch (method_call.method) { case FERMI2D_REG_INDEX(trigger): { HandleSurfaceCopy(); break; @@ -47,6 +49,9 @@ void Fermi2D::HandleSurfaceCopy() { u32 dst_bytes_per_pixel = RenderTargetBytesPerPixel(regs.dst.format); if (!rasterizer.AccelerateSurfaceCopy(regs.src, regs.dst)) { + // All copies here update the main memory, so mark all rasterizer states as invalid. + Core::System::GetInstance().GPU().Maxwell3D().dirty_flags.OnMemoryWrite(); + rasterizer.FlushRegion(source_cpu, src_bytes_per_pixel * regs.src.width * regs.src.height); // We have to invalidate the destination region to evict any outdated surfaces from the // cache. We do this before actually writing the new data because the destination address @@ -68,13 +73,13 @@ void Fermi2D::HandleSurfaceCopy() { Texture::CopySwizzledData(regs.src.width, regs.src.height, regs.src.depth, src_bytes_per_pixel, dst_bytes_per_pixel, src_buffer, dst_buffer, true, regs.src.BlockHeight(), - regs.src.BlockDepth()); + regs.src.BlockDepth(), 0); } else { // If the input is linear and the output is tiled, swizzle the input and copy it over. Texture::CopySwizzledData(regs.src.width, regs.src.height, regs.src.depth, src_bytes_per_pixel, dst_bytes_per_pixel, dst_buffer, src_buffer, false, regs.dst.BlockHeight(), - regs.dst.BlockDepth()); + regs.dst.BlockDepth(), 0); } } } diff --git a/src/video_core/engines/fermi_2d.h b/src/video_core/engines/fermi_2d.h index 2a6e8bbbb..50009bf75 100644 --- a/src/video_core/engines/fermi_2d.h +++ b/src/video_core/engines/fermi_2d.h @@ -27,7 +27,7 @@ public: ~Fermi2D() = default; /// Write the value to the register identified by method. - void WriteReg(u32 method, u32 value); + void CallMethod(const GPU::MethodCall& method_call); struct Regs { static constexpr std::size_t NUM_REGS = 0x258; diff --git a/src/video_core/engines/kepler_memory.cpp b/src/video_core/engines/kepler_memory.cpp index 585290d9f..4880191fc 100644 --- a/src/video_core/engines/kepler_memory.cpp +++ b/src/video_core/engines/kepler_memory.cpp @@ -3,8 +3,10 @@ // Refer to the license.txt file included. #include "common/logging/log.h" +#include "core/core.h" #include "core/memory.h" #include "video_core/engines/kepler_memory.h" +#include "video_core/engines/maxwell_3d.h" #include "video_core/rasterizer_interface.h" namespace Tegra::Engines { @@ -15,19 +17,19 @@ KeplerMemory::KeplerMemory(VideoCore::RasterizerInterface& rasterizer, KeplerMemory::~KeplerMemory() = default; -void KeplerMemory::WriteReg(u32 method, u32 value) { - ASSERT_MSG(method < Regs::NUM_REGS, +void KeplerMemory::CallMethod(const GPU::MethodCall& method_call) { + ASSERT_MSG(method_call.method < Regs::NUM_REGS, "Invalid KeplerMemory register, increase the size of the Regs structure"); - regs.reg_array[method] = value; + regs.reg_array[method_call.method] = method_call.argument; - switch (method) { + switch (method_call.method) { case KEPLERMEMORY_REG_INDEX(exec): { state.write_offset = 0; break; } case KEPLERMEMORY_REG_INDEX(data): { - ProcessData(value); + ProcessData(method_call.argument); break; } } @@ -47,6 +49,7 @@ void KeplerMemory::ProcessData(u32 data) { rasterizer.InvalidateRegion(dest_address, sizeof(u32)); Memory::Write32(dest_address, data); + Core::System::GetInstance().GPU().Maxwell3D().dirty_flags.OnMemoryWrite(); state.write_offset++; } diff --git a/src/video_core/engines/kepler_memory.h b/src/video_core/engines/kepler_memory.h index bf4a13cff..fe9ebc5b9 100644 --- a/src/video_core/engines/kepler_memory.h +++ b/src/video_core/engines/kepler_memory.h @@ -9,6 +9,7 @@ #include "common/bit_field.h" #include "common/common_funcs.h" #include "common/common_types.h" +#include "video_core/gpu.h" #include "video_core/memory_manager.h" namespace VideoCore { @@ -26,7 +27,7 @@ public: ~KeplerMemory(); /// Write the value to the register identified by method. - void WriteReg(u32 method, u32 value); + void CallMethod(const GPU::MethodCall& method_call); struct Regs { static constexpr size_t NUM_REGS = 0x7F; diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index 2bc534be3..b19b3a75a 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -97,57 +97,74 @@ void Maxwell3D::CallMacroMethod(u32 method, std::vector<u32> parameters) { macro_interpreter.Execute(search->second, std::move(parameters)); } -void Maxwell3D::WriteReg(u32 method, u32 value, u32 remaining_params) { +void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) { auto debug_context = Core::System::GetInstance().GetGPUDebugContext(); // It is an error to write to a register other than the current macro's ARG register before it // has finished execution. if (executing_macro != 0) { - ASSERT(method == executing_macro + 1); + ASSERT(method_call.method == executing_macro + 1); } // Methods after 0xE00 are special, they're actually triggers for some microcode that was // uploaded to the GPU during initialization. - if (method >= MacroRegistersStart) { + if (method_call.method >= MacroRegistersStart) { // We're trying to execute a macro if (executing_macro == 0) { // A macro call must begin by writing the macro method's register, not its argument. - ASSERT_MSG((method % 2) == 0, + ASSERT_MSG((method_call.method % 2) == 0, "Can't start macro execution by writing to the ARGS register"); - executing_macro = method; + executing_macro = method_call.method; } - macro_params.push_back(value); + macro_params.push_back(method_call.argument); // Call the macro when there are no more parameters in the command buffer - if (remaining_params == 0) { + if (method_call.IsLastCall()) { CallMacroMethod(executing_macro, std::move(macro_params)); } return; } - ASSERT_MSG(method < Regs::NUM_REGS, + ASSERT_MSG(method_call.method < Regs::NUM_REGS, "Invalid Maxwell3D register, increase the size of the Regs structure"); if (debug_context) { debug_context->OnEvent(Tegra::DebugContext::Event::MaxwellCommandLoaded, nullptr); } - if (regs.reg_array[method] != value) { - regs.reg_array[method] = value; - if (method >= MAXWELL3D_REG_INDEX(vertex_attrib_format) && - method < MAXWELL3D_REG_INDEX(vertex_attrib_format) + regs.vertex_attrib_format.size()) { + if (regs.reg_array[method_call.method] != method_call.argument) { + regs.reg_array[method_call.method] = method_call.argument; + // Vertex format + if (method_call.method >= MAXWELL3D_REG_INDEX(vertex_attrib_format) && + method_call.method < + MAXWELL3D_REG_INDEX(vertex_attrib_format) + regs.vertex_attrib_format.size()) { dirty_flags.vertex_attrib_format = true; } + + // Vertex buffer + if (method_call.method >= MAXWELL3D_REG_INDEX(vertex_array) && + method_call.method < MAXWELL3D_REG_INDEX(vertex_array) + 4 * 32) { + dirty_flags.vertex_array |= + 1u << ((method_call.method - MAXWELL3D_REG_INDEX(vertex_array)) >> 2); + } else if (method_call.method >= MAXWELL3D_REG_INDEX(vertex_array_limit) && + method_call.method < MAXWELL3D_REG_INDEX(vertex_array_limit) + 2 * 32) { + dirty_flags.vertex_array |= + 1u << ((method_call.method - MAXWELL3D_REG_INDEX(vertex_array_limit)) >> 1); + } else if (method_call.method >= MAXWELL3D_REG_INDEX(instanced_arrays) && + method_call.method < MAXWELL3D_REG_INDEX(instanced_arrays) + 32) { + dirty_flags.vertex_array |= + 1u << (method_call.method - MAXWELL3D_REG_INDEX(instanced_arrays)); + } } - switch (method) { + switch (method_call.method) { case MAXWELL3D_REG_INDEX(macros.data): { - ProcessMacroUpload(value); + ProcessMacroUpload(method_call.argument); break; } case MAXWELL3D_REG_INDEX(macros.bind): { - ProcessMacroBind(value); + ProcessMacroBind(method_call.argument); break; } case MAXWELL3D_REG_INDEX(const_buffer.cb_data[0]): @@ -166,7 +183,7 @@ void Maxwell3D::WriteReg(u32 method, u32 value, u32 remaining_params) { case MAXWELL3D_REG_INDEX(const_buffer.cb_data[13]): case MAXWELL3D_REG_INDEX(const_buffer.cb_data[14]): case MAXWELL3D_REG_INDEX(const_buffer.cb_data[15]): { - ProcessCBData(value); + ProcessCBData(method_call.argument); break; } case MAXWELL3D_REG_INDEX(cb_bind[0].raw_config): { @@ -270,6 +287,7 @@ void Maxwell3D::ProcessQueryGet() { query_result.timestamp = CoreTiming::GetTicks(); Memory::WriteBlock(*address, &query_result, sizeof(query_result)); } + dirty_flags.OnMemoryWrite(); break; } default: @@ -346,6 +364,7 @@ void Maxwell3D::ProcessCBData(u32 value) { memory_manager.GpuToCpuAddress(buffer_address + regs.const_buffer.cb_pos); Memory::Write32(*address, value); + dirty_flags.OnMemoryWrite(); // Increment the current buffer position. regs.const_buffer.cb_pos = regs.const_buffer.cb_pos + 4; diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h index 9e480dc39..d3b3ed1f0 100644 --- a/src/video_core/engines/maxwell_3d.h +++ b/src/video_core/engines/maxwell_3d.h @@ -389,6 +389,13 @@ public: ReverseSubtract = 3, Min = 4, Max = 5, + + // These values are used by Nouveau and some games. + AddGL = 0x8006, + SubtractGL = 0x8007, + ReverseSubtractGL = 0x8008, + MinGL = 0x800a, + MaxGL = 0x800b }; enum class Factor : u32 { @@ -583,10 +590,18 @@ public: float clear_color[4]; float clear_depth; + INSERT_PADDING_WORDS(0x3); + s32 clear_stencil; - INSERT_PADDING_WORDS(0x17); + INSERT_PADDING_WORDS(0x7); + + u32 polygon_offset_point_enable; + u32 polygon_offset_line_enable; + u32 polygon_offset_fill_enable; + + INSERT_PADDING_WORDS(0xD); std::array<ScissorTest, NumViewports> scissor_test; @@ -624,7 +639,16 @@ public: } } zeta; - INSERT_PADDING_WORDS(0x5B); + INSERT_PADDING_WORDS(0x41); + + union { + BitField<0, 4, u32> stencil; + BitField<4, 4, u32> unknown; + BitField<8, 4, u32> scissor; + BitField<12, 4, u32> viewport; + } clear_flags; + + INSERT_PADDING_WORDS(0x19); std::array<VertexAttribute, NumVertexAttributes> vertex_attrib_format; @@ -712,6 +736,7 @@ public: u32 frag_color_clamp; union { + BitField<0, 1, u32> y_negate; BitField<4, 1, u32> triangle_rast_flip; } screen_y_control; @@ -719,7 +744,20 @@ public: u32 vb_element_base; - INSERT_PADDING_WORDS(0x38); + INSERT_PADDING_WORDS(0x36); + + union { + BitField<0, 1, u32> c0; + BitField<1, 1, u32> c1; + BitField<2, 1, u32> c2; + BitField<3, 1, u32> c3; + BitField<4, 1, u32> c4; + BitField<5, 1, u32> c5; + BitField<6, 1, u32> c6; + BitField<7, 1, u32> c7; + } clip_distance_enabled; + + INSERT_PADDING_WORDS(0x1); float point_size; @@ -745,7 +783,11 @@ public: } } tsc; - INSERT_PADDING_WORDS(0x3); + INSERT_PADDING_WORDS(0x1); + + float polygon_offset_factor; + + INSERT_PADDING_WORDS(0x1); struct { u32 tic_address_high; @@ -770,7 +812,9 @@ public: u32 framebuffer_srgb; - INSERT_PADDING_WORDS(0x12); + float polygon_offset_units; + + INSERT_PADDING_WORDS(0x11); union { BitField<2, 1, u32> coord_origin; @@ -847,7 +891,9 @@ public: INSERT_PADDING_WORDS(0x7); - INSERT_PADDING_WORDS(0x20); + INSERT_PADDING_WORDS(0x1F); + + float polygon_offset_clamp; struct { u32 is_instanced[NumVertexArrays]; @@ -863,8 +909,21 @@ public: Cull cull; - INSERT_PADDING_WORDS(0x28); + u32 pixel_center_integer; + + INSERT_PADDING_WORDS(0x1); + + u32 viewport_transform_enabled; + + INSERT_PADDING_WORDS(0x3); + + union { + BitField<0, 1, u32> depth_range_0_1; + BitField<3, 1, u32> depth_clamp_near; + BitField<4, 1, u32> depth_clamp_far; + } view_volume_clip_control; + INSERT_PADDING_WORDS(0x21); struct { u32 enable; LogicOperation operation; @@ -1028,6 +1087,11 @@ public: struct DirtyFlags { bool vertex_attrib_format = true; + u32 vertex_array = 0xFFFFFFFF; + + void OnMemoryWrite() { + vertex_array = 0xFFFFFFFF; + } }; DirtyFlags dirty_flags; @@ -1036,7 +1100,7 @@ public: u32 GetRegisterValue(u32 method) const; /// Write the value to the register identified by method. - void WriteReg(u32 method, u32 value, u32 remaining_params); + void CallMethod(const GPU::MethodCall& method_call); /// Returns a list of enabled textures for the specified shader stage. std::vector<Texture::FullTextureInfo> GetStageTextures(Regs::ShaderStage stage) const; @@ -1120,6 +1184,9 @@ ASSERT_REG_POSITION(vertex_buffer, 0x35D); ASSERT_REG_POSITION(clear_color[0], 0x360); ASSERT_REG_POSITION(clear_depth, 0x364); ASSERT_REG_POSITION(clear_stencil, 0x368); +ASSERT_REG_POSITION(polygon_offset_point_enable, 0x370); +ASSERT_REG_POSITION(polygon_offset_line_enable, 0x371); +ASSERT_REG_POSITION(polygon_offset_fill_enable, 0x372); ASSERT_REG_POSITION(scissor_test, 0x380); ASSERT_REG_POSITION(stencil_back_func_ref, 0x3D5); ASSERT_REG_POSITION(stencil_back_mask, 0x3D6); @@ -1127,6 +1194,7 @@ ASSERT_REG_POSITION(stencil_back_func_mask, 0x3D7); ASSERT_REG_POSITION(color_mask_common, 0x3E4); ASSERT_REG_POSITION(rt_separate_frag_data, 0x3EB); ASSERT_REG_POSITION(zeta, 0x3F8); +ASSERT_REG_POSITION(clear_flags, 0x43E); ASSERT_REG_POSITION(vertex_attrib_format, 0x458); ASSERT_REG_POSITION(rt_control, 0x487); ASSERT_REG_POSITION(zeta_width, 0x48a); @@ -1153,10 +1221,12 @@ ASSERT_REG_POSITION(stencil_front_mask, 0x4E7); ASSERT_REG_POSITION(frag_color_clamp, 0x4EA); ASSERT_REG_POSITION(screen_y_control, 0x4EB); ASSERT_REG_POSITION(vb_element_base, 0x50D); +ASSERT_REG_POSITION(clip_distance_enabled, 0x544); ASSERT_REG_POSITION(point_size, 0x546); ASSERT_REG_POSITION(zeta_enable, 0x54E); ASSERT_REG_POSITION(multisample_control, 0x54F); ASSERT_REG_POSITION(tsc, 0x557); +ASSERT_REG_POSITION(polygon_offset_factor, 0x55b); ASSERT_REG_POSITION(tic, 0x55D); ASSERT_REG_POSITION(stencil_two_side_enable, 0x565); ASSERT_REG_POSITION(stencil_back_op_fail, 0x566); @@ -1164,13 +1234,18 @@ ASSERT_REG_POSITION(stencil_back_op_zfail, 0x567); ASSERT_REG_POSITION(stencil_back_op_zpass, 0x568); ASSERT_REG_POSITION(stencil_back_func_func, 0x569); ASSERT_REG_POSITION(framebuffer_srgb, 0x56E); +ASSERT_REG_POSITION(polygon_offset_units, 0x56F); ASSERT_REG_POSITION(point_coord_replace, 0x581); ASSERT_REG_POSITION(code_address, 0x582); ASSERT_REG_POSITION(draw, 0x585); ASSERT_REG_POSITION(primitive_restart, 0x591); ASSERT_REG_POSITION(index_array, 0x5F2); +ASSERT_REG_POSITION(polygon_offset_clamp, 0x61F); ASSERT_REG_POSITION(instanced_arrays, 0x620); ASSERT_REG_POSITION(cull, 0x646); +ASSERT_REG_POSITION(pixel_center_integer, 0x649); +ASSERT_REG_POSITION(viewport_transform_enabled, 0x64B); +ASSERT_REG_POSITION(view_volume_clip_control, 0x64F); ASSERT_REG_POSITION(logic_op, 0x671); ASSERT_REG_POSITION(clear_buffers, 0x674); ASSERT_REG_POSITION(color_mask, 0x680); diff --git a/src/video_core/engines/maxwell_compute.cpp b/src/video_core/engines/maxwell_compute.cpp index 8b5f08351..656db6a61 100644 --- a/src/video_core/engines/maxwell_compute.cpp +++ b/src/video_core/engines/maxwell_compute.cpp @@ -8,13 +8,13 @@ namespace Tegra::Engines { -void MaxwellCompute::WriteReg(u32 method, u32 value) { - ASSERT_MSG(method < Regs::NUM_REGS, +void MaxwellCompute::CallMethod(const GPU::MethodCall& method_call) { + ASSERT_MSG(method_call.method < Regs::NUM_REGS, "Invalid MaxwellCompute register, increase the size of the Regs structure"); - regs.reg_array[method] = value; + regs.reg_array[method_call.method] = method_call.argument; - switch (method) { + switch (method_call.method) { case MAXWELL_COMPUTE_REG_INDEX(compute): { LOG_CRITICAL(HW_GPU, "Compute shaders are not implemented"); UNREACHABLE(); diff --git a/src/video_core/engines/maxwell_compute.h b/src/video_core/engines/maxwell_compute.h index 6ea934fb9..1d71f11bd 100644 --- a/src/video_core/engines/maxwell_compute.h +++ b/src/video_core/engines/maxwell_compute.h @@ -9,6 +9,7 @@ #include "common/bit_field.h" #include "common/common_funcs.h" #include "common/common_types.h" +#include "video_core/gpu.h" namespace Tegra::Engines { @@ -42,7 +43,7 @@ public: "MaxwellCompute Regs has wrong size"); /// Write the value to the register identified by method. - void WriteReg(u32 method, u32 value); + void CallMethod(const GPU::MethodCall& method_call); }; #define ASSERT_REG_POSITION(field_name, position) \ diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp index b8a78cf82..06462f570 100644 --- a/src/video_core/engines/maxwell_dma.cpp +++ b/src/video_core/engines/maxwell_dma.cpp @@ -2,7 +2,9 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. +#include "core/core.h" #include "core/memory.h" +#include "video_core/engines/maxwell_3d.h" #include "video_core/engines/maxwell_dma.h" #include "video_core/rasterizer_interface.h" #include "video_core/textures/decoders.h" @@ -12,16 +14,16 @@ namespace Tegra::Engines { MaxwellDMA::MaxwellDMA(VideoCore::RasterizerInterface& rasterizer, MemoryManager& memory_manager) : memory_manager(memory_manager), rasterizer{rasterizer} {} -void MaxwellDMA::WriteReg(u32 method, u32 value) { - ASSERT_MSG(method < Regs::NUM_REGS, +void MaxwellDMA::CallMethod(const GPU::MethodCall& method_call) { + ASSERT_MSG(method_call.method < Regs::NUM_REGS, "Invalid MaxwellDMA register, increase the size of the Regs structure"); - regs.reg_array[method] = value; + regs.reg_array[method_call.method] = method_call.argument; #define MAXWELLDMA_REG_INDEX(field_name) \ (offsetof(Tegra::Engines::MaxwellDMA::Regs, field_name) / sizeof(u32)) - switch (method) { + switch (method_call.method) { case MAXWELLDMA_REG_INDEX(exec): { HandleCopy(); break; @@ -54,6 +56,9 @@ void MaxwellDMA::HandleCopy() { return; } + // All copies here update the main memory, so mark all rasterizer states as invalid. + Core::System::GetInstance().GPU().Maxwell3D().dirty_flags.OnMemoryWrite(); + if (regs.exec.is_dst_linear && regs.exec.is_src_linear) { // When the enable_2d bit is disabled, the copy is performed as if we were copying a 1D // buffer of length `x_count`, otherwise we copy a 2D image of dimensions (x_count, diff --git a/src/video_core/engines/maxwell_dma.h b/src/video_core/engines/maxwell_dma.h index 5f3704f05..1f8cd65d2 100644 --- a/src/video_core/engines/maxwell_dma.h +++ b/src/video_core/engines/maxwell_dma.h @@ -24,7 +24,7 @@ public: ~MaxwellDMA() = default; /// Write the value to the register identified by method. - void WriteReg(u32 method, u32 value); + void CallMethod(const GPU::MethodCall& method_call); struct Regs { static constexpr std::size_t NUM_REGS = 0x1D6; diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h index 83a6fd875..b9faaf8e0 100644 --- a/src/video_core/engines/shader_bytecode.h +++ b/src/video_core/engines/shader_bytecode.h @@ -82,6 +82,8 @@ union Attribute { Position = 7, Attribute_0 = 8, Attribute_31 = 39, + ClipDistances0123 = 44, + ClipDistances4567 = 45, PointCoord = 46, // This attribute contains a tuple of (~, ~, InstanceId, VertexId) when inside a vertex // shader, and a tuple of (TessCoord.x, TessCoord.y, TessCoord.z, ~) when inside a Tess Eval @@ -153,6 +155,7 @@ enum class PredCondition : u64 { NotEqual = 5, GreaterEqual = 6, LessThanWithNan = 9, + LessEqualWithNan = 11, GreaterThanWithNan = 12, NotEqualWithNan = 13, GreaterEqualWithNan = 14, @@ -261,7 +264,7 @@ enum class FlowCondition : u64 { Fcsm_Tr = 0x1C, // TODO(bunnei): What is this used for? }; -enum class ControlCode : u64 { +enum class ConditionCode : u64 { F = 0, LT = 1, EQ = 2, @@ -365,6 +368,11 @@ enum class HalfPrecision : u64 { FMZ = 2, }; +enum class R2pMode : u64 { + Pr = 0, + Cc = 1, +}; + enum class IpaInterpMode : u64 { Linear = 0, Perspective = 1, @@ -569,7 +577,6 @@ union Instruction { BitField<39, 2, u64> tab5cb8_2; BitField<41, 3, u64> tab5c68_1; BitField<44, 2, u64> tab5c68_0; - BitField<47, 1, u64> cc; BitField<48, 1, u64> negate_b; } fmul; @@ -831,7 +838,7 @@ union Instruction { union { BitField<0, 3, u64> pred0; BitField<3, 3, u64> pred3; - BitField<8, 5, ControlCode> cc; // flag in cc + BitField<8, 5, ConditionCode> cc; // flag in cc BitField<39, 3, u64> pred39; BitField<42, 1, u64> neg_pred39; BitField<45, 4, PredOperation> op; // op with pred39 @@ -855,6 +862,12 @@ union Instruction { } hsetp2; union { + BitField<40, 1, R2pMode> mode; + BitField<41, 2, u64> byte; + BitField<20, 7, u64> immediate_mask; + } r2p; + + union { BitField<39, 3, u64> pred39; BitField<42, 1, u64> neg_pred; BitField<43, 1, u64> neg_a; @@ -1235,7 +1248,7 @@ union Instruction { BitField<60, 1, u64> is_b_gpr; BitField<59, 1, u64> is_c_gpr; BitField<20, 24, s64> smem_imm; - BitField<0, 5, ControlCode> flow_control_code; + BitField<0, 5, ConditionCode> flow_condition_code; Attribute attribute; Sampler sampler; @@ -1256,6 +1269,7 @@ public: BFE_C, BFE_R, BFE_IMM, + BFI_IMM_R, BRA, PBK, LD_A, @@ -1381,6 +1395,7 @@ public: PSETP, PSET, CSETP, + R2P_IMM, XMAD_IMM, XMAD_CR, XMAD_RC, @@ -1396,6 +1411,7 @@ public: ArithmeticHalf, ArithmeticHalfImmediate, Bfe, + Bfi, Shift, Ffma, Hfma2, @@ -1410,6 +1426,7 @@ public: HalfSetPredicate, PredicateSetPredicate, PredicateSetRegister, + RegisterSetPredicate, Conversion, Xmad, Unknown, @@ -1613,6 +1630,7 @@ private: INST("0100110000000---", Id::BFE_C, Type::Bfe, "BFE_C"), INST("0101110000000---", Id::BFE_R, Type::Bfe, "BFE_R"), INST("0011100-00000---", Id::BFE_IMM, Type::Bfe, "BFE_IMM"), + INST("0011011-11110---", Id::BFI_IMM_R, Type::Bfi, "BFI_IMM_R"), INST("0100110001000---", Id::LOP_C, Type::ArithmeticInteger, "LOP_C"), INST("0101110001000---", Id::LOP_R, Type::ArithmeticInteger, "LOP_R"), INST("0011100001000---", Id::LOP_IMM, Type::ArithmeticInteger, "LOP_IMM"), @@ -1647,6 +1665,7 @@ private: INST("0101000010001---", Id::PSET, Type::PredicateSetRegister, "PSET"), INST("0101000010010---", Id::PSETP, Type::PredicateSetPredicate, "PSETP"), INST("010100001010----", Id::CSETP, Type::PredicateSetPredicate, "CSETP"), + INST("0011100-11110---", Id::R2P_IMM, Type::RegisterSetPredicate, "R2P_IMM"), INST("0011011-00------", Id::XMAD_IMM, Type::Xmad, "XMAD_IMM"), INST("0100111---------", Id::XMAD_CR, Type::Xmad, "XMAD_CR"), INST("010100010-------", Id::XMAD_RC, Type::Xmad, "XMAD_RC"), diff --git a/src/video_core/engines/shader_header.h b/src/video_core/engines/shader_header.h index a0e015c4b..99c34649f 100644 --- a/src/video_core/engines/shader_header.h +++ b/src/video_core/engines/shader_header.h @@ -62,7 +62,16 @@ struct Header { INSERT_PADDING_BYTES(1); // ImapSystemValuesB INSERT_PADDING_BYTES(16); // ImapGenericVector[32] INSERT_PADDING_BYTES(2); // ImapColor - INSERT_PADDING_BYTES(2); // ImapSystemValuesC + union { + BitField<0, 8, u16> clip_distances; + BitField<8, 1, u16> point_sprite_s; + BitField<9, 1, u16> point_sprite_t; + BitField<10, 1, u16> fog_coordinate; + BitField<12, 1, u16> tessellation_eval_point_u; + BitField<13, 1, u16> tessellation_eval_point_v; + BitField<14, 1, u16> instance_id; + BitField<15, 1, u16> vertex_id; + }; INSERT_PADDING_BYTES(5); // ImapFixedFncTexture[10] INSERT_PADDING_BYTES(1); // ImapReserved INSERT_PADDING_BYTES(3); // OmapSystemValuesA diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp index 51b3904f6..6c81dee64 100644 --- a/src/video_core/gpu.cpp +++ b/src/video_core/gpu.cpp @@ -26,6 +26,7 @@ u32 FramebufferConfig::BytesPerPixel(PixelFormat format) { GPU::GPU(VideoCore::RasterizerInterface& rasterizer) { memory_manager = std::make_unique<Tegra::MemoryManager>(); + dma_pusher = std::make_unique<Tegra::DmaPusher>(*this); maxwell_3d = std::make_unique<Engines::Maxwell3D>(rasterizer, *memory_manager); fermi_2d = std::make_unique<Engines::Fermi2D>(rasterizer, *memory_manager); maxwell_compute = std::make_unique<Engines::MaxwellCompute>(); @@ -51,6 +52,14 @@ const MemoryManager& GPU::MemoryManager() const { return *memory_manager; } +DmaPusher& GPU::DmaPusher() { + return *dma_pusher; +} + +const DmaPusher& GPU::DmaPusher() const { + return *dma_pusher; +} + u32 RenderTargetBytesPerPixel(RenderTargetFormat format) { ASSERT(format != RenderTargetFormat::NONE); @@ -113,4 +122,48 @@ u32 DepthFormatBytesPerPixel(DepthFormat format) { } } +enum class BufferMethods { + BindObject = 0, + CountBufferMethods = 0x40, +}; + +void GPU::CallMethod(const MethodCall& method_call) { + LOG_TRACE(HW_GPU, + "Processing method {:08X} on subchannel {} value " + "{:08X} remaining params {}", + MethCall.method, MethCall.subchannel, value, remaining_params); + + ASSERT(method_call.subchannel < bound_engines.size()); + + if (method_call.method == static_cast<u32>(BufferMethods::BindObject)) { + // Bind the current subchannel to the desired engine id. + LOG_DEBUG(HW_GPU, "Binding subchannel {} to engine {}", method_call.subchannel, + method_call.argument); + bound_engines[method_call.subchannel] = static_cast<EngineID>(method_call.argument); + return; + } + + const EngineID engine = bound_engines[method_call.subchannel]; + + switch (engine) { + case EngineID::FERMI_TWOD_A: + fermi_2d->CallMethod(method_call); + break; + case EngineID::MAXWELL_B: + maxwell_3d->CallMethod(method_call); + break; + case EngineID::MAXWELL_COMPUTE_B: + maxwell_compute->CallMethod(method_call); + break; + case EngineID::MAXWELL_DMA_COPY_A: + maxwell_dma->CallMethod(method_call); + break; + case EngineID::KEPLER_INLINE_TO_MEMORY_B: + kepler_memory->CallMethod(method_call); + break; + default: + UNIMPLEMENTED_MSG("Unimplemented engine"); + } +} + } // namespace Tegra diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h index 5cc1e19ca..af5ccd1e9 100644 --- a/src/video_core/gpu.h +++ b/src/video_core/gpu.h @@ -9,6 +9,7 @@ #include <vector> #include "common/common_types.h" #include "core/hle/service/nvflinger/buffer_queue.h" +#include "video_core/dma_pusher.h" #include "video_core/memory_manager.h" namespace VideoCore { @@ -119,8 +120,23 @@ public: explicit GPU(VideoCore::RasterizerInterface& rasterizer); ~GPU(); - /// Processes a command list stored at the specified address in GPU memory. - void ProcessCommandLists(const std::vector<CommandListHeader>& commands); + struct MethodCall { + u32 method{}; + u32 argument{}; + u32 subchannel{}; + u32 method_count{}; + + bool IsLastCall() const { + return method_count <= 1; + } + + MethodCall(u32 method, u32 argument, u32 subchannel = 0, u32 method_count = 0) + : method(method), argument(argument), subchannel(subchannel), + method_count(method_count) {} + }; + + /// Calls a GPU method. + void CallMethod(const MethodCall& method_call); /// Returns a reference to the Maxwell3D GPU engine. Engines::Maxwell3D& Maxwell3D(); @@ -134,7 +150,14 @@ public: /// Returns a const reference to the GPU memory manager. const Tegra::MemoryManager& MemoryManager() const; + /// Returns a reference to the GPU DMA pusher. + Tegra::DmaPusher& DmaPusher(); + + /// Returns a const reference to the GPU DMA pusher. + const Tegra::DmaPusher& DmaPusher() const; + private: + std::unique_ptr<Tegra::DmaPusher> dma_pusher; std::unique_ptr<Tegra::MemoryManager> memory_manager; /// Mapping of command subchannels to their bound engine ids. diff --git a/src/video_core/macro_interpreter.cpp b/src/video_core/macro_interpreter.cpp index 335a8d407..9c55e9f1e 100644 --- a/src/video_core/macro_interpreter.cpp +++ b/src/video_core/macro_interpreter.cpp @@ -35,6 +35,7 @@ void MacroInterpreter::Reset() { // The next parameter index starts at 1, because $r1 already has the value of the first // parameter. next_parameter_index = 1; + carry_flag = false; } bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) { @@ -135,14 +136,28 @@ MacroInterpreter::Opcode MacroInterpreter::GetOpcode(u32 offset) const { return {macro_memory[offset + pc / sizeof(u32)]}; } -u32 MacroInterpreter::GetALUResult(ALUOperation operation, u32 src_a, u32 src_b) const { +u32 MacroInterpreter::GetALUResult(ALUOperation operation, u32 src_a, u32 src_b) { switch (operation) { - case ALUOperation::Add: - return src_a + src_b; - // TODO(Subv): Implement AddWithCarry - case ALUOperation::Subtract: - return src_a - src_b; - // TODO(Subv): Implement SubtractWithBorrow + case ALUOperation::Add: { + const u64 result{static_cast<u64>(src_a) + src_b}; + carry_flag = result > 0xffffffff; + return static_cast<u32>(result); + } + case 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: { + const u64 result{static_cast<u64>(src_a) - src_b}; + carry_flag = result < 0x100000000; + return static_cast<u32>(result); + } + case 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: return src_a ^ src_b; case ALUOperation::Or: @@ -235,7 +250,7 @@ void MacroInterpreter::SetMethodAddress(u32 address) { } void MacroInterpreter::Send(u32 value) { - maxwell3d.WriteReg(method_address.address, value, 0); + maxwell3d.CallMethod({method_address.address, value}); // Increment the method address by the method increment. method_address.address.Assign(method_address.address.Value() + method_address.increment.Value()); diff --git a/src/video_core/macro_interpreter.h b/src/video_core/macro_interpreter.h index 62d1ce289..cde360288 100644 --- a/src/video_core/macro_interpreter.h +++ b/src/video_core/macro_interpreter.h @@ -117,7 +117,7 @@ private: bool Step(u32 offset, 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) const; + u32 GetALUResult(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). @@ -165,5 +165,7 @@ private: std::vector<u32> parameters; /// Index of the next parameter that will be fetched by the 'parm' instruction. u32 next_parameter_index = 0; + + bool carry_flag{}; }; } // namespace Tegra diff --git a/src/video_core/memory_manager.cpp b/src/video_core/memory_manager.cpp index 77a20bb84..47247f097 100644 --- a/src/video_core/memory_manager.cpp +++ b/src/video_core/memory_manager.cpp @@ -9,6 +9,13 @@ namespace Tegra { +MemoryManager::MemoryManager() { + // Mark the first page as reserved, so that 0 is not a valid GPUVAddr. Otherwise, games might + // try to use 0 as a valid address, which is also used to mean nullptr. This fixes a bug with + // Undertale using 0 for a render target. + PageSlot(0) = static_cast<u64>(PageStatus::Reserved); +} + GPUVAddr MemoryManager::AllocateSpace(u64 size, u64 align) { const std::optional<GPUVAddr> gpu_addr{FindFreeBlock(0, size, align, PageStatus::Unmapped)}; diff --git a/src/video_core/memory_manager.h b/src/video_core/memory_manager.h index 4eb338aa2..fb03497ca 100644 --- a/src/video_core/memory_manager.h +++ b/src/video_core/memory_manager.h @@ -18,7 +18,7 @@ using GPUVAddr = u64; class MemoryManager final { public: - MemoryManager() = default; + MemoryManager(); GPUVAddr AllocateSpace(u64 size, u64 align); GPUVAddr AllocateSpace(GPUVAddr gpu_addr, u64 size, u64 align); @@ -37,6 +37,7 @@ private: enum class PageStatus : u64 { Unmapped = 0xFFFFFFFFFFFFFFFFULL, Allocated = 0xFFFFFFFFFFFFFFFEULL, + Reserved = 0xFFFFFFFFFFFFFFFDULL, }; std::optional<GPUVAddr> FindFreeBlock(GPUVAddr region_start, u64 size, u64 align, diff --git a/src/video_core/morton.cpp b/src/video_core/morton.cpp new file mode 100644 index 000000000..a310491a8 --- /dev/null +++ b/src/video_core/morton.cpp @@ -0,0 +1,355 @@ +// Copyright 2018 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <array> +#include <cstring> +#include "common/assert.h" +#include "common/common_types.h" +#include "core/memory.h" +#include "video_core/morton.h" +#include "video_core/surface.h" +#include "video_core/textures/decoders.h" + +namespace VideoCore { + +using Surface::GetBytesPerPixel; +using Surface::PixelFormat; + +using MortonCopyFn = void (*)(u32, u32, u32, u32, u32, u32, u8*, std::size_t, VAddr); +using ConversionArray = std::array<MortonCopyFn, Surface::MaxPixelFormat>; + +template <bool morton_to_linear, PixelFormat format> +static void MortonCopy(u32 stride, u32 block_height, u32 height, u32 block_depth, u32 depth, + u32 tile_width_spacing, u8* buffer, std::size_t buffer_size, VAddr addr) { + constexpr u32 bytes_per_pixel = GetBytesPerPixel(format); + + // With the BCn formats (DXT and DXN), each 4x4 tile is swizzled instead of just individual + // pixel values. + const u32 tile_size_x{GetDefaultBlockWidth(format)}; + const u32 tile_size_y{GetDefaultBlockHeight(format)}; + + if constexpr (morton_to_linear) { + Tegra::Texture::UnswizzleTexture(buffer, addr, tile_size_x, tile_size_y, bytes_per_pixel, + stride, height, depth, block_height, block_depth, + tile_width_spacing); + } else { + Tegra::Texture::CopySwizzledData( + (stride + tile_size_x - 1) / tile_size_x, (height + tile_size_y - 1) / tile_size_y, + depth, bytes_per_pixel, bytes_per_pixel, Memory::GetPointer(addr), buffer, false, + block_height, block_depth, tile_width_spacing); + } +} + +static constexpr ConversionArray morton_to_linear_fns = { + // clang-format off + MortonCopy<true, PixelFormat::ABGR8U>, + MortonCopy<true, PixelFormat::ABGR8S>, + MortonCopy<true, PixelFormat::ABGR8UI>, + MortonCopy<true, PixelFormat::B5G6R5U>, + MortonCopy<true, PixelFormat::A2B10G10R10U>, + MortonCopy<true, PixelFormat::A1B5G5R5U>, + MortonCopy<true, PixelFormat::R8U>, + MortonCopy<true, PixelFormat::R8UI>, + MortonCopy<true, PixelFormat::RGBA16F>, + MortonCopy<true, PixelFormat::RGBA16U>, + MortonCopy<true, PixelFormat::RGBA16UI>, + MortonCopy<true, PixelFormat::R11FG11FB10F>, + MortonCopy<true, PixelFormat::RGBA32UI>, + MortonCopy<true, PixelFormat::DXT1>, + MortonCopy<true, PixelFormat::DXT23>, + MortonCopy<true, PixelFormat::DXT45>, + MortonCopy<true, PixelFormat::DXN1>, + MortonCopy<true, PixelFormat::DXN2UNORM>, + MortonCopy<true, PixelFormat::DXN2SNORM>, + MortonCopy<true, PixelFormat::BC7U>, + MortonCopy<true, PixelFormat::BC6H_UF16>, + MortonCopy<true, PixelFormat::BC6H_SF16>, + MortonCopy<true, PixelFormat::ASTC_2D_4X4>, + MortonCopy<true, PixelFormat::G8R8U>, + MortonCopy<true, PixelFormat::G8R8S>, + MortonCopy<true, PixelFormat::BGRA8>, + MortonCopy<true, PixelFormat::RGBA32F>, + MortonCopy<true, PixelFormat::RG32F>, + MortonCopy<true, PixelFormat::R32F>, + MortonCopy<true, PixelFormat::R16F>, + MortonCopy<true, PixelFormat::R16U>, + MortonCopy<true, PixelFormat::R16S>, + MortonCopy<true, PixelFormat::R16UI>, + MortonCopy<true, PixelFormat::R16I>, + MortonCopy<true, PixelFormat::RG16>, + MortonCopy<true, PixelFormat::RG16F>, + MortonCopy<true, PixelFormat::RG16UI>, + MortonCopy<true, PixelFormat::RG16I>, + MortonCopy<true, PixelFormat::RG16S>, + MortonCopy<true, PixelFormat::RGB32F>, + MortonCopy<true, PixelFormat::RGBA8_SRGB>, + MortonCopy<true, PixelFormat::RG8U>, + MortonCopy<true, PixelFormat::RG8S>, + MortonCopy<true, PixelFormat::RG32UI>, + MortonCopy<true, PixelFormat::R32UI>, + MortonCopy<true, PixelFormat::ASTC_2D_8X8>, + MortonCopy<true, PixelFormat::ASTC_2D_8X5>, + MortonCopy<true, PixelFormat::ASTC_2D_5X4>, + MortonCopy<true, PixelFormat::BGRA8_SRGB>, + MortonCopy<true, PixelFormat::DXT1_SRGB>, + MortonCopy<true, PixelFormat::DXT23_SRGB>, + MortonCopy<true, PixelFormat::DXT45_SRGB>, + MortonCopy<true, PixelFormat::BC7U_SRGB>, + MortonCopy<true, PixelFormat::ASTC_2D_4X4_SRGB>, + MortonCopy<true, PixelFormat::ASTC_2D_8X8_SRGB>, + MortonCopy<true, PixelFormat::ASTC_2D_8X5_SRGB>, + MortonCopy<true, PixelFormat::ASTC_2D_5X4_SRGB>, + MortonCopy<true, PixelFormat::ASTC_2D_5X5>, + MortonCopy<true, PixelFormat::ASTC_2D_5X5_SRGB>, + MortonCopy<true, PixelFormat::ASTC_2D_10X8>, + MortonCopy<true, PixelFormat::ASTC_2D_10X8_SRGB>, + MortonCopy<true, PixelFormat::Z32F>, + MortonCopy<true, PixelFormat::Z16>, + MortonCopy<true, PixelFormat::Z24S8>, + MortonCopy<true, PixelFormat::S8Z24>, + MortonCopy<true, PixelFormat::Z32FS8>, + // clang-format on +}; + +static constexpr ConversionArray linear_to_morton_fns = { + // clang-format off + MortonCopy<false, PixelFormat::ABGR8U>, + MortonCopy<false, PixelFormat::ABGR8S>, + MortonCopy<false, PixelFormat::ABGR8UI>, + MortonCopy<false, PixelFormat::B5G6R5U>, + MortonCopy<false, PixelFormat::A2B10G10R10U>, + MortonCopy<false, PixelFormat::A1B5G5R5U>, + MortonCopy<false, PixelFormat::R8U>, + MortonCopy<false, PixelFormat::R8UI>, + MortonCopy<false, PixelFormat::RGBA16F>, + MortonCopy<false, PixelFormat::RGBA16U>, + MortonCopy<false, PixelFormat::RGBA16UI>, + MortonCopy<false, PixelFormat::R11FG11FB10F>, + MortonCopy<false, PixelFormat::RGBA32UI>, + MortonCopy<false, PixelFormat::DXT1>, + MortonCopy<false, PixelFormat::DXT23>, + MortonCopy<false, PixelFormat::DXT45>, + MortonCopy<false, PixelFormat::DXN1>, + MortonCopy<false, PixelFormat::DXN2UNORM>, + MortonCopy<false, PixelFormat::DXN2SNORM>, + MortonCopy<false, PixelFormat::BC7U>, + MortonCopy<false, PixelFormat::BC6H_UF16>, + MortonCopy<false, PixelFormat::BC6H_SF16>, + // TODO(Subv): Swizzling ASTC formats are not supported + nullptr, + MortonCopy<false, PixelFormat::G8R8U>, + MortonCopy<false, PixelFormat::G8R8S>, + MortonCopy<false, PixelFormat::BGRA8>, + MortonCopy<false, PixelFormat::RGBA32F>, + MortonCopy<false, PixelFormat::RG32F>, + MortonCopy<false, PixelFormat::R32F>, + MortonCopy<false, PixelFormat::R16F>, + MortonCopy<false, PixelFormat::R16U>, + MortonCopy<false, PixelFormat::R16S>, + MortonCopy<false, PixelFormat::R16UI>, + MortonCopy<false, PixelFormat::R16I>, + MortonCopy<false, PixelFormat::RG16>, + MortonCopy<false, PixelFormat::RG16F>, + MortonCopy<false, PixelFormat::RG16UI>, + MortonCopy<false, PixelFormat::RG16I>, + MortonCopy<false, PixelFormat::RG16S>, + MortonCopy<false, PixelFormat::RGB32F>, + MortonCopy<false, PixelFormat::RGBA8_SRGB>, + MortonCopy<false, PixelFormat::RG8U>, + MortonCopy<false, PixelFormat::RG8S>, + MortonCopy<false, PixelFormat::RG32UI>, + MortonCopy<false, PixelFormat::R32UI>, + nullptr, + nullptr, + nullptr, + MortonCopy<false, PixelFormat::BGRA8_SRGB>, + MortonCopy<false, PixelFormat::DXT1_SRGB>, + MortonCopy<false, PixelFormat::DXT23_SRGB>, + MortonCopy<false, PixelFormat::DXT45_SRGB>, + MortonCopy<false, PixelFormat::BC7U_SRGB>, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + MortonCopy<false, PixelFormat::Z32F>, + MortonCopy<false, PixelFormat::Z16>, + MortonCopy<false, PixelFormat::Z24S8>, + MortonCopy<false, PixelFormat::S8Z24>, + MortonCopy<false, PixelFormat::Z32FS8>, + // clang-format on +}; + +static MortonCopyFn GetSwizzleFunction(MortonSwizzleMode mode, Surface::PixelFormat format) { + switch (mode) { + case MortonSwizzleMode::MortonToLinear: + return morton_to_linear_fns[static_cast<std::size_t>(format)]; + case MortonSwizzleMode::LinearToMorton: + return linear_to_morton_fns[static_cast<std::size_t>(format)]; + } + UNREACHABLE(); +} + +/// 8x8 Z-Order coordinate from 2D coordinates +static u32 MortonInterleave(u32 x, u32 y) { + static const u32 xlut[] = {0x00, 0x01, 0x04, 0x05, 0x10, 0x11, 0x14, 0x15}; + static const u32 ylut[] = {0x00, 0x02, 0x08, 0x0a, 0x20, 0x22, 0x28, 0x2a}; + return xlut[x % 8] + ylut[y % 8]; +} + +/// Calculates the offset of the position of the pixel in Morton order +static u32 GetMortonOffset(u32 x, u32 y, u32 bytes_per_pixel) { + // Images are split into 8x8 tiles. Each tile is composed of four 4x4 subtiles each + // of which is composed of four 2x2 subtiles each of which is composed of four texels. + // Each structure is embedded into the next-bigger one in a diagonal pattern, e.g. + // texels are laid out in a 2x2 subtile like this: + // 2 3 + // 0 1 + // + // The full 8x8 tile has the texels arranged like this: + // + // 42 43 46 47 58 59 62 63 + // 40 41 44 45 56 57 60 61 + // 34 35 38 39 50 51 54 55 + // 32 33 36 37 48 49 52 53 + // 10 11 14 15 26 27 30 31 + // 08 09 12 13 24 25 28 29 + // 02 03 06 07 18 19 22 23 + // 00 01 04 05 16 17 20 21 + // + // This pattern is what's called Z-order curve, or Morton order. + + const unsigned int block_height = 8; + const unsigned int coarse_x = x & ~7; + + u32 i = MortonInterleave(x, y); + + const unsigned int offset = coarse_x * block_height; + + return (i + offset) * bytes_per_pixel; +} + +static u32 MortonInterleave128(u32 x, u32 y) { + // 128x128 Z-Order coordinate from 2D coordinates + static constexpr u32 xlut[] = { + 0x0000, 0x0001, 0x0002, 0x0003, 0x0008, 0x0009, 0x000a, 0x000b, 0x0040, 0x0041, 0x0042, + 0x0043, 0x0048, 0x0049, 0x004a, 0x004b, 0x0800, 0x0801, 0x0802, 0x0803, 0x0808, 0x0809, + 0x080a, 0x080b, 0x0840, 0x0841, 0x0842, 0x0843, 0x0848, 0x0849, 0x084a, 0x084b, 0x1000, + 0x1001, 0x1002, 0x1003, 0x1008, 0x1009, 0x100a, 0x100b, 0x1040, 0x1041, 0x1042, 0x1043, + 0x1048, 0x1049, 0x104a, 0x104b, 0x1800, 0x1801, 0x1802, 0x1803, 0x1808, 0x1809, 0x180a, + 0x180b, 0x1840, 0x1841, 0x1842, 0x1843, 0x1848, 0x1849, 0x184a, 0x184b, 0x2000, 0x2001, + 0x2002, 0x2003, 0x2008, 0x2009, 0x200a, 0x200b, 0x2040, 0x2041, 0x2042, 0x2043, 0x2048, + 0x2049, 0x204a, 0x204b, 0x2800, 0x2801, 0x2802, 0x2803, 0x2808, 0x2809, 0x280a, 0x280b, + 0x2840, 0x2841, 0x2842, 0x2843, 0x2848, 0x2849, 0x284a, 0x284b, 0x3000, 0x3001, 0x3002, + 0x3003, 0x3008, 0x3009, 0x300a, 0x300b, 0x3040, 0x3041, 0x3042, 0x3043, 0x3048, 0x3049, + 0x304a, 0x304b, 0x3800, 0x3801, 0x3802, 0x3803, 0x3808, 0x3809, 0x380a, 0x380b, 0x3840, + 0x3841, 0x3842, 0x3843, 0x3848, 0x3849, 0x384a, 0x384b, 0x0000, 0x0001, 0x0002, 0x0003, + 0x0008, 0x0009, 0x000a, 0x000b, 0x0040, 0x0041, 0x0042, 0x0043, 0x0048, 0x0049, 0x004a, + 0x004b, 0x0800, 0x0801, 0x0802, 0x0803, 0x0808, 0x0809, 0x080a, 0x080b, 0x0840, 0x0841, + 0x0842, 0x0843, 0x0848, 0x0849, 0x084a, 0x084b, 0x1000, 0x1001, 0x1002, 0x1003, 0x1008, + 0x1009, 0x100a, 0x100b, 0x1040, 0x1041, 0x1042, 0x1043, 0x1048, 0x1049, 0x104a, 0x104b, + 0x1800, 0x1801, 0x1802, 0x1803, 0x1808, 0x1809, 0x180a, 0x180b, 0x1840, 0x1841, 0x1842, + 0x1843, 0x1848, 0x1849, 0x184a, 0x184b, 0x2000, 0x2001, 0x2002, 0x2003, 0x2008, 0x2009, + 0x200a, 0x200b, 0x2040, 0x2041, 0x2042, 0x2043, 0x2048, 0x2049, 0x204a, 0x204b, 0x2800, + 0x2801, 0x2802, 0x2803, 0x2808, 0x2809, 0x280a, 0x280b, 0x2840, 0x2841, 0x2842, 0x2843, + 0x2848, 0x2849, 0x284a, 0x284b, 0x3000, 0x3001, 0x3002, 0x3003, 0x3008, 0x3009, 0x300a, + 0x300b, 0x3040, 0x3041, 0x3042, 0x3043, 0x3048, 0x3049, 0x304a, 0x304b, 0x3800, 0x3801, + 0x3802, 0x3803, 0x3808, 0x3809, 0x380a, 0x380b, 0x3840, 0x3841, 0x3842, 0x3843, 0x3848, + 0x3849, 0x384a, 0x384b, 0x0000, 0x0001, 0x0002, 0x0003, 0x0008, 0x0009, 0x000a, 0x000b, + 0x0040, 0x0041, 0x0042, 0x0043, 0x0048, 0x0049, 0x004a, 0x004b, 0x0800, 0x0801, 0x0802, + 0x0803, 0x0808, 0x0809, 0x080a, 0x080b, 0x0840, 0x0841, 0x0842, 0x0843, 0x0848, 0x0849, + 0x084a, 0x084b, 0x1000, 0x1001, 0x1002, 0x1003, 0x1008, 0x1009, 0x100a, 0x100b, 0x1040, + 0x1041, 0x1042, 0x1043, 0x1048, 0x1049, 0x104a, 0x104b, 0x1800, 0x1801, 0x1802, 0x1803, + 0x1808, 0x1809, 0x180a, 0x180b, 0x1840, 0x1841, 0x1842, 0x1843, 0x1848, 0x1849, 0x184a, + 0x184b, 0x2000, 0x2001, 0x2002, 0x2003, 0x2008, 0x2009, 0x200a, 0x200b, 0x2040, 0x2041, + 0x2042, 0x2043, 0x2048, 0x2049, 0x204a, 0x204b, 0x2800, 0x2801, 0x2802, 0x2803, 0x2808, + 0x2809, 0x280a, 0x280b, 0x2840, 0x2841, 0x2842, 0x2843, 0x2848, 0x2849, 0x284a, 0x284b, + 0x3000, 0x3001, 0x3002, 0x3003, 0x3008, 0x3009, 0x300a, 0x300b, 0x3040, 0x3041, 0x3042, + 0x3043, 0x3048, 0x3049, 0x304a, 0x304b, 0x3800, 0x3801, 0x3802, 0x3803, 0x3808, 0x3809, + 0x380a, 0x380b, 0x3840, 0x3841, 0x3842, 0x3843, 0x3848, 0x3849, 0x384a, 0x384b, + }; + static constexpr u32 ylut[] = { + 0x0000, 0x0004, 0x0010, 0x0014, 0x0020, 0x0024, 0x0030, 0x0034, 0x0080, 0x0084, 0x0090, + 0x0094, 0x00a0, 0x00a4, 0x00b0, 0x00b4, 0x0100, 0x0104, 0x0110, 0x0114, 0x0120, 0x0124, + 0x0130, 0x0134, 0x0180, 0x0184, 0x0190, 0x0194, 0x01a0, 0x01a4, 0x01b0, 0x01b4, 0x0200, + 0x0204, 0x0210, 0x0214, 0x0220, 0x0224, 0x0230, 0x0234, 0x0280, 0x0284, 0x0290, 0x0294, + 0x02a0, 0x02a4, 0x02b0, 0x02b4, 0x0300, 0x0304, 0x0310, 0x0314, 0x0320, 0x0324, 0x0330, + 0x0334, 0x0380, 0x0384, 0x0390, 0x0394, 0x03a0, 0x03a4, 0x03b0, 0x03b4, 0x0400, 0x0404, + 0x0410, 0x0414, 0x0420, 0x0424, 0x0430, 0x0434, 0x0480, 0x0484, 0x0490, 0x0494, 0x04a0, + 0x04a4, 0x04b0, 0x04b4, 0x0500, 0x0504, 0x0510, 0x0514, 0x0520, 0x0524, 0x0530, 0x0534, + 0x0580, 0x0584, 0x0590, 0x0594, 0x05a0, 0x05a4, 0x05b0, 0x05b4, 0x0600, 0x0604, 0x0610, + 0x0614, 0x0620, 0x0624, 0x0630, 0x0634, 0x0680, 0x0684, 0x0690, 0x0694, 0x06a0, 0x06a4, + 0x06b0, 0x06b4, 0x0700, 0x0704, 0x0710, 0x0714, 0x0720, 0x0724, 0x0730, 0x0734, 0x0780, + 0x0784, 0x0790, 0x0794, 0x07a0, 0x07a4, 0x07b0, 0x07b4, 0x0000, 0x0004, 0x0010, 0x0014, + 0x0020, 0x0024, 0x0030, 0x0034, 0x0080, 0x0084, 0x0090, 0x0094, 0x00a0, 0x00a4, 0x00b0, + 0x00b4, 0x0100, 0x0104, 0x0110, 0x0114, 0x0120, 0x0124, 0x0130, 0x0134, 0x0180, 0x0184, + 0x0190, 0x0194, 0x01a0, 0x01a4, 0x01b0, 0x01b4, 0x0200, 0x0204, 0x0210, 0x0214, 0x0220, + 0x0224, 0x0230, 0x0234, 0x0280, 0x0284, 0x0290, 0x0294, 0x02a0, 0x02a4, 0x02b0, 0x02b4, + 0x0300, 0x0304, 0x0310, 0x0314, 0x0320, 0x0324, 0x0330, 0x0334, 0x0380, 0x0384, 0x0390, + 0x0394, 0x03a0, 0x03a4, 0x03b0, 0x03b4, 0x0400, 0x0404, 0x0410, 0x0414, 0x0420, 0x0424, + 0x0430, 0x0434, 0x0480, 0x0484, 0x0490, 0x0494, 0x04a0, 0x04a4, 0x04b0, 0x04b4, 0x0500, + 0x0504, 0x0510, 0x0514, 0x0520, 0x0524, 0x0530, 0x0534, 0x0580, 0x0584, 0x0590, 0x0594, + 0x05a0, 0x05a4, 0x05b0, 0x05b4, 0x0600, 0x0604, 0x0610, 0x0614, 0x0620, 0x0624, 0x0630, + 0x0634, 0x0680, 0x0684, 0x0690, 0x0694, 0x06a0, 0x06a4, 0x06b0, 0x06b4, 0x0700, 0x0704, + 0x0710, 0x0714, 0x0720, 0x0724, 0x0730, 0x0734, 0x0780, 0x0784, 0x0790, 0x0794, 0x07a0, + 0x07a4, 0x07b0, 0x07b4, 0x0000, 0x0004, 0x0010, 0x0014, 0x0020, 0x0024, 0x0030, 0x0034, + 0x0080, 0x0084, 0x0090, 0x0094, 0x00a0, 0x00a4, 0x00b0, 0x00b4, 0x0100, 0x0104, 0x0110, + 0x0114, 0x0120, 0x0124, 0x0130, 0x0134, 0x0180, 0x0184, 0x0190, 0x0194, 0x01a0, 0x01a4, + 0x01b0, 0x01b4, 0x0200, 0x0204, 0x0210, 0x0214, 0x0220, 0x0224, 0x0230, 0x0234, 0x0280, + 0x0284, 0x0290, 0x0294, 0x02a0, 0x02a4, 0x02b0, 0x02b4, 0x0300, 0x0304, 0x0310, 0x0314, + 0x0320, 0x0324, 0x0330, 0x0334, 0x0380, 0x0384, 0x0390, 0x0394, 0x03a0, 0x03a4, 0x03b0, + 0x03b4, 0x0400, 0x0404, 0x0410, 0x0414, 0x0420, 0x0424, 0x0430, 0x0434, 0x0480, 0x0484, + 0x0490, 0x0494, 0x04a0, 0x04a4, 0x04b0, 0x04b4, 0x0500, 0x0504, 0x0510, 0x0514, 0x0520, + 0x0524, 0x0530, 0x0534, 0x0580, 0x0584, 0x0590, 0x0594, 0x05a0, 0x05a4, 0x05b0, 0x05b4, + 0x0600, 0x0604, 0x0610, 0x0614, 0x0620, 0x0624, 0x0630, 0x0634, 0x0680, 0x0684, 0x0690, + 0x0694, 0x06a0, 0x06a4, 0x06b0, 0x06b4, 0x0700, 0x0704, 0x0710, 0x0714, 0x0720, 0x0724, + 0x0730, 0x0734, 0x0780, 0x0784, 0x0790, 0x0794, 0x07a0, 0x07a4, 0x07b0, 0x07b4, + }; + return xlut[x % 128] + ylut[y % 128]; +} + +static u32 GetMortonOffset128(u32 x, u32 y, u32 bytes_per_pixel) { + // Calculates the offset of the position of the pixel in Morton order + // Framebuffer images are split into 128x128 tiles. + + constexpr u32 block_height = 128; + const u32 coarse_x = x & ~127; + + const u32 i = MortonInterleave128(x, y); + + const u32 offset = coarse_x * block_height; + + return (i + offset) * bytes_per_pixel; +} + +void MortonSwizzle(MortonSwizzleMode mode, Surface::PixelFormat format, u32 stride, + u32 block_height, u32 height, u32 block_depth, u32 depth, u32 tile_width_spacing, + u8* buffer, std::size_t buffer_size, VAddr addr) { + + GetSwizzleFunction(mode, format)(stride, block_height, height, block_depth, depth, + tile_width_spacing, buffer, buffer_size, addr); +} + +void MortonCopyPixels128(u32 width, u32 height, u32 bytes_per_pixel, u32 linear_bytes_per_pixel, + u8* morton_data, u8* linear_data, bool morton_to_linear) { + u8* data_ptrs[2]; + for (u32 y = 0; y < height; ++y) { + for (u32 x = 0; x < width; ++x) { + const u32 coarse_y = y & ~127; + const u32 morton_offset = + GetMortonOffset128(x, y, bytes_per_pixel) + coarse_y * width * bytes_per_pixel; + const u32 linear_pixel_index = (x + y * width) * linear_bytes_per_pixel; + + data_ptrs[morton_to_linear ? 1 : 0] = morton_data + morton_offset; + data_ptrs[morton_to_linear ? 0 : 1] = &linear_data[linear_pixel_index]; + + std::memcpy(data_ptrs[0], data_ptrs[1], bytes_per_pixel); + } + } +} + +} // namespace VideoCore diff --git a/src/video_core/morton.h b/src/video_core/morton.h new file mode 100644 index 000000000..065f59ce3 --- /dev/null +++ b/src/video_core/morton.h @@ -0,0 +1,21 @@ +// Copyright 2018 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include "common/common_types.h" +#include "video_core/surface.h" + +namespace VideoCore { + +enum class MortonSwizzleMode { MortonToLinear, LinearToMorton }; + +void MortonSwizzle(MortonSwizzleMode mode, VideoCore::Surface::PixelFormat format, u32 stride, + u32 block_height, u32 height, u32 block_depth, u32 depth, u32 tile_width_spacing, + u8* buffer, std::size_t buffer_size, VAddr addr); + +void MortonCopyPixels128(u32 width, u32 height, u32 bytes_per_pixel, u32 linear_bytes_per_pixel, + u8* morton_data, u8* linear_data, bool morton_to_linear); + +} // namespace VideoCore diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.cpp b/src/video_core/renderer_opengl/gl_buffer_cache.cpp index 075192c3f..46a6c0308 100644 --- a/src/video_core/renderer_opengl/gl_buffer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_buffer_cache.cpp @@ -76,7 +76,7 @@ std::tuple<u8*, GLintptr> OGLBufferCache::ReserveMemory(std::size_t size, std::s return std::make_tuple(uploaded_ptr, uploaded_offset); } -void OGLBufferCache::Map(std::size_t max_size) { +bool OGLBufferCache::Map(std::size_t max_size) { bool invalidate; std::tie(buffer_ptr, buffer_offset_base, invalidate) = stream_buffer.Map(static_cast<GLsizeiptr>(max_size), 4); @@ -85,6 +85,7 @@ void OGLBufferCache::Map(std::size_t max_size) { if (invalidate) { InvalidateAll(); } + return invalidate; } void OGLBufferCache::Unmap() { diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.h b/src/video_core/renderer_opengl/gl_buffer_cache.h index 91fca3f6c..c11acfb79 100644 --- a/src/video_core/renderer_opengl/gl_buffer_cache.h +++ b/src/video_core/renderer_opengl/gl_buffer_cache.h @@ -50,7 +50,7 @@ public: /// Reserves memory to be used by host's CPU. Returns mapped address and offset. std::tuple<u8*, GLintptr> ReserveMemory(std::size_t size, std::size_t alignment = 4); - void Map(std::size_t max_size); + bool Map(std::size_t max_size); void Unmap(); GLuint GetHandle() const; diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index ae6aaee4c..a44bbfae8 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -98,14 +98,9 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& window, ScreenInfo has_ARB_direct_state_access = true; } else if (extension == "GL_ARB_multi_bind") { has_ARB_multi_bind = true; - } else if (extension == "GL_ARB_separate_shader_objects") { - has_ARB_separate_shader_objects = true; - } else if (extension == "GL_ARB_vertex_attrib_binding") { - has_ARB_vertex_attrib_binding = true; } } - ASSERT_MSG(has_ARB_separate_shader_objects, "has_ARB_separate_shader_objects is unsupported"); OpenGLState::ApplyDefaultState(); // Create render framebuffer @@ -118,10 +113,24 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& window, ScreenInfo glGetIntegerv(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT, &uniform_buffer_alignment); LOG_CRITICAL(Render_OpenGL, "Sync fixed function OpenGL state here!"); + CheckExtensions(); } RasterizerOpenGL::~RasterizerOpenGL() {} +void RasterizerOpenGL::CheckExtensions() { + if (!GLAD_GL_ARB_texture_filter_anisotropic && !GLAD_GL_EXT_texture_filter_anisotropic) { + LOG_WARNING( + Render_OpenGL, + "Anisotropic filter is not supported! This can cause graphical issues in some games."); + } + if (!GLAD_GL_ARB_buffer_storage) { + LOG_WARNING( + Render_OpenGL, + "Buffer storage control is not supported! This can cause performance degradation."); + } +} + void RasterizerOpenGL::SetupVertexFormat() { auto& gpu = Core::System::GetInstance().GPU().Maxwell3D(); const auto& regs = gpu.regs; @@ -181,15 +190,25 @@ void RasterizerOpenGL::SetupVertexFormat() { } state.draw.vertex_array = VAO.handle; state.ApplyVertexBufferState(); + + // Rebinding the VAO invalidates the vertex buffer bindings. + gpu.dirty_flags.vertex_array = 0xFFFFFFFF; } void RasterizerOpenGL::SetupVertexBuffer() { - MICROPROFILE_SCOPE(OpenGL_VB); - const auto& gpu = Core::System::GetInstance().GPU().Maxwell3D(); + auto& gpu = Core::System::GetInstance().GPU().Maxwell3D(); const auto& regs = gpu.regs; + if (!gpu.dirty_flags.vertex_array) + return; + + MICROPROFILE_SCOPE(OpenGL_VB); + // Upload all guest vertex arrays sequentially to our buffer for (u32 index = 0; index < Maxwell::NumVertexArrays; ++index) { + if (~gpu.dirty_flags.vertex_array & (1u << index)) + continue; + const auto& vertex_array = regs.vertex_array[index]; if (!vertex_array.IsEnabled()) continue; @@ -216,6 +235,8 @@ void RasterizerOpenGL::SetupVertexBuffer() { // Implicit set by glBindVertexBuffer. Stupid glstate handling... state.draw.vertex_buffer = buffer_cache.GetHandle(); + + gpu.dirty_flags.vertex_array = 0; } DrawParameters RasterizerOpenGL::SetupDraw() { @@ -542,6 +563,30 @@ void RasterizerOpenGL::Clear() { ASSERT_MSG(regs.zeta_enable != 0, "Tried to clear stencil but buffer is not enabled!"); use_stencil = true; clear_state.stencil.test_enabled = true; + if (regs.clear_flags.stencil) { + // Stencil affects the clear so fill it with the used masks + clear_state.stencil.front.test_func = GL_ALWAYS; + clear_state.stencil.front.test_mask = regs.stencil_front_func_mask; + clear_state.stencil.front.action_stencil_fail = GL_KEEP; + clear_state.stencil.front.action_depth_fail = GL_KEEP; + clear_state.stencil.front.action_depth_pass = GL_KEEP; + clear_state.stencil.front.write_mask = regs.stencil_front_mask; + if (regs.stencil_two_side_enable) { + clear_state.stencil.back.test_func = GL_ALWAYS; + clear_state.stencil.back.test_mask = regs.stencil_back_func_mask; + clear_state.stencil.back.action_stencil_fail = GL_KEEP; + clear_state.stencil.back.action_depth_fail = GL_KEEP; + clear_state.stencil.back.action_depth_pass = GL_KEEP; + clear_state.stencil.back.write_mask = regs.stencil_back_mask; + } else { + clear_state.stencil.back.test_func = GL_ALWAYS; + clear_state.stencil.back.test_mask = 0xFFFFFFFF; + clear_state.stencil.back.write_mask = 0xFFFFFFFF; + clear_state.stencil.back.action_stencil_fail = GL_KEEP; + clear_state.stencil.back.action_depth_fail = GL_KEEP; + clear_state.stencil.back.action_depth_pass = GL_KEEP; + } + } } if (!use_color && !use_depth && !use_stencil) { @@ -553,6 +598,14 @@ void RasterizerOpenGL::Clear() { ConfigureFramebuffers(clear_state, use_color, use_depth || use_stencil, false, regs.clear_buffers.RT.Value()); + if (regs.clear_flags.scissor) { + SyncScissorTest(clear_state); + } + + if (regs.clear_flags.viewport) { + clear_state.EmulateViewportWithScissor(); + } + clear_state.Apply(); if (use_color) { @@ -573,7 +626,7 @@ void RasterizerOpenGL::DrawArrays() { return; MICROPROFILE_SCOPE(OpenGL_Drawing); - const auto& gpu = Core::System::GetInstance().GPU().Maxwell3D(); + auto& gpu = Core::System::GetInstance().GPU().Maxwell3D(); const auto& regs = gpu.regs; ScopeAcquireGLContext acquire_context{emu_window}; @@ -588,12 +641,13 @@ void RasterizerOpenGL::DrawArrays() { SyncLogicOpState(); SyncCullMode(); SyncPrimitiveRestart(); - SyncScissorTest(); + SyncScissorTest(state); + SyncClipEnabled(); // Alpha Testing is synced on shaders. SyncTransformFeedback(); SyncPointState(); CheckAlphaTests(); - + SyncPolygonOffset(); // TODO(bunnei): Sync framebuffer_scale uniform here // TODO(bunnei): Sync scissorbox uniform(s) here @@ -626,7 +680,11 @@ void RasterizerOpenGL::DrawArrays() { // Add space for at least 18 constant buffers buffer_size += Maxwell::MaxConstBuffers * (MaxConstbufferSize + uniform_buffer_alignment); - buffer_cache.Map(buffer_size); + bool invalidate = buffer_cache.Map(buffer_size); + if (invalidate) { + // As all cached buffers are invalidated, we need to recheck their state. + gpu.dirty_flags.vertex_array = 0xFFFFFFFF; + } SetupVertexFormat(); SetupVertexBuffer(); @@ -815,7 +873,7 @@ void RasterizerOpenGL::SamplerInfo::SyncWithConfig(const Tegra::Texture::TSCEntr } const u32 bias = config.mip_lod_bias.Value(); // Sign extend the 13-bit value. - const u32 mask = 1U << (13 - 1); + constexpr u32 mask = 1U << (13 - 1); const float bias_lod = static_cast<s32>((bias ^ mask) - mask) / 256.f; if (lod_bias != bias_lod) { lod_bias = bias_lod; @@ -942,20 +1000,35 @@ u32 RasterizerOpenGL::SetupTextures(Maxwell::ShaderStage stage, Shader& shader, void RasterizerOpenGL::SyncViewport(OpenGLState& current_state) { const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs; - for (std::size_t i = 0; i < Tegra::Engines::Maxwell3D::Regs::NumViewports; i++) { - const MathUtil::Rectangle<s32> viewport_rect{regs.viewport_transform[i].GetRect()}; + const bool geometry_shaders_enabled = + regs.IsShaderConfigEnabled(static_cast<size_t>(Maxwell::ShaderProgram::Geometry)); + const std::size_t viewport_count = + geometry_shaders_enabled ? Tegra::Engines::Maxwell3D::Regs::NumViewports : 1; + for (std::size_t i = 0; i < viewport_count; i++) { auto& viewport = current_state.viewports[i]; + const auto& src = regs.viewports[i]; + const MathUtil::Rectangle<s32> viewport_rect{regs.viewport_transform[i].GetRect()}; viewport.x = viewport_rect.left; viewport.y = viewport_rect.bottom; - viewport.width = static_cast<GLfloat>(viewport_rect.GetWidth()); - viewport.height = static_cast<GLfloat>(viewport_rect.GetHeight()); + viewport.width = viewport_rect.GetWidth(); + viewport.height = viewport_rect.GetHeight(); viewport.depth_range_far = regs.viewports[i].depth_range_far; viewport.depth_range_near = regs.viewports[i].depth_range_near; } + state.depth_clamp.far_plane = regs.view_volume_clip_control.depth_clamp_far != 0; + state.depth_clamp.near_plane = regs.view_volume_clip_control.depth_clamp_near != 0; } void RasterizerOpenGL::SyncClipEnabled() { - UNREACHABLE(); + const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs; + state.clip_distance[0] = regs.clip_distance_enabled.c0 != 0; + state.clip_distance[1] = regs.clip_distance_enabled.c1 != 0; + state.clip_distance[2] = regs.clip_distance_enabled.c2 != 0; + state.clip_distance[3] = regs.clip_distance_enabled.c3 != 0; + state.clip_distance[4] = regs.clip_distance_enabled.c4 != 0; + state.clip_distance[5] = regs.clip_distance_enabled.c5 != 0; + state.clip_distance[6] = regs.clip_distance_enabled.c6 != 0; + state.clip_distance[7] = regs.clip_distance_enabled.c7 != 0; } void RasterizerOpenGL::SyncClipCoef() { @@ -1120,11 +1193,15 @@ void RasterizerOpenGL::SyncLogicOpState() { state.logic_op.operation = MaxwellToGL::LogicOp(regs.logic_op.operation); } -void RasterizerOpenGL::SyncScissorTest() { +void RasterizerOpenGL::SyncScissorTest(OpenGLState& current_state) { const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs; - for (std::size_t i = 0; i < Tegra::Engines::Maxwell3D::Regs::NumViewports; i++) { + const bool geometry_shaders_enabled = + regs.IsShaderConfigEnabled(static_cast<size_t>(Maxwell::ShaderProgram::Geometry)); + const std::size_t viewport_count = + geometry_shaders_enabled ? Tegra::Engines::Maxwell3D::Regs::NumViewports : 1; + for (std::size_t i = 0; i < viewport_count; i++) { const auto& src = regs.scissor_test[i]; - auto& dst = state.viewports[i].scissor; + auto& dst = current_state.viewports[i].scissor; dst.enabled = (src.enable != 0); if (dst.enabled == 0) { return; @@ -1152,6 +1229,16 @@ void RasterizerOpenGL::SyncPointState() { state.point.size = regs.point_size; } +void RasterizerOpenGL::SyncPolygonOffset() { + const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs; + state.polygon_offset.fill_enable = regs.polygon_offset_fill_enable != 0; + state.polygon_offset.line_enable = regs.polygon_offset_line_enable != 0; + state.polygon_offset.point_enable = regs.polygon_offset_point_enable != 0; + state.polygon_offset.units = regs.polygon_offset_units; + state.polygon_offset.factor = regs.polygon_offset_factor; + state.polygon_offset.clamp = regs.polygon_offset_clamp; +} + void RasterizerOpenGL::CheckAlphaTests() { const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs; diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index 6e78ab4cd..7ec9746b1 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -91,19 +91,20 @@ private: void SyncWithConfig(const Tegra::Texture::TSCEntry& info); private: - Tegra::Texture::TextureFilter mag_filter; - Tegra::Texture::TextureFilter min_filter; - Tegra::Texture::TextureMipmapFilter mip_filter; - Tegra::Texture::WrapMode wrap_u; - Tegra::Texture::WrapMode wrap_v; - Tegra::Texture::WrapMode wrap_p; - bool uses_depth_compare; - Tegra::Texture::DepthCompareFunc depth_compare_func; - GLvec4 border_color; - float min_lod; - float max_lod; - float lod_bias; - float max_anisotropic; + Tegra::Texture::TextureFilter mag_filter = Tegra::Texture::TextureFilter::Nearest; + Tegra::Texture::TextureFilter min_filter = Tegra::Texture::TextureFilter::Nearest; + Tegra::Texture::TextureMipmapFilter mip_filter = Tegra::Texture::TextureMipmapFilter::None; + Tegra::Texture::WrapMode wrap_u = Tegra::Texture::WrapMode::ClampToEdge; + Tegra::Texture::WrapMode wrap_v = Tegra::Texture::WrapMode::ClampToEdge; + Tegra::Texture::WrapMode wrap_p = Tegra::Texture::WrapMode::ClampToEdge; + bool uses_depth_compare = false; + Tegra::Texture::DepthCompareFunc depth_compare_func = + Tegra::Texture::DepthCompareFunc::Always; + GLvec4 border_color = {}; + float min_lod = 0.0f; + float max_lod = 16.0f; + float lod_bias = 0.0f; + float max_anisotropic = 1.0f; }; /** @@ -171,7 +172,7 @@ private: void SyncMultiSampleState(); /// Syncs the scissor test state to match the guest state - void SyncScissorTest(); + void SyncScissorTest(OpenGLState& current_state); /// Syncs the transform feedback state to match the guest state void SyncTransformFeedback(); @@ -182,13 +183,18 @@ private: /// Syncs Color Mask void SyncColorMask(); + /// Syncs the polygon offsets + void SyncPolygonOffset(); + /// Check asserts for alpha testing. void CheckAlphaTests(); + /// Check for extension that are not strictly required + /// but are needed for correct emulation + void CheckExtensions(); + bool has_ARB_direct_state_access = false; bool has_ARB_multi_bind = false; - bool has_ARB_separate_shader_objects = false; - bool has_ARB_vertex_attrib_binding = false; OpenGLState state; diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp index 9ca82c06c..dde2f468d 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp @@ -15,6 +15,7 @@ #include "core/memory.h" #include "core/settings.h" #include "video_core/engines/maxwell_3d.h" +#include "video_core/morton.h" #include "video_core/renderer_opengl/gl_rasterizer.h" #include "video_core/renderer_opengl/gl_rasterizer_cache.h" #include "video_core/renderer_opengl/gl_state.h" @@ -22,10 +23,11 @@ #include "video_core/surface.h" #include "video_core/textures/astc.h" #include "video_core/textures/decoders.h" -#include "video_core/utils.h" namespace OpenGL { +using VideoCore::MortonSwizzle; +using VideoCore::MortonSwizzleMode; using VideoCore::Surface::ComponentTypeFromDepthFormat; using VideoCore::Surface::ComponentTypeFromRenderTarget; using VideoCore::Surface::ComponentTypeFromTexture; @@ -95,6 +97,7 @@ std::size_t SurfaceParams::InnerMemorySize(bool force_gl, bool layer_only, params.block_width = params.is_tiled ? config.tic.BlockWidth() : 0, params.block_height = params.is_tiled ? config.tic.BlockHeight() : 0, params.block_depth = params.is_tiled ? config.tic.BlockDepth() : 0, + params.tile_width_spacing = params.is_tiled ? (1 << config.tic.tile_width_spacing.Value()) : 1; params.srgb_conversion = config.tic.IsSrgbConversionEnabled(); params.pixel_format = PixelFormatFromTextureFormat(config.tic.format, config.tic.r_type.Value(), params.srgb_conversion); @@ -160,6 +163,7 @@ std::size_t SurfaceParams::InnerMemorySize(bool force_gl, bool layer_only, params.block_width = 1 << config.memory_layout.block_width; params.block_height = 1 << config.memory_layout.block_height; params.block_depth = 1 << config.memory_layout.block_depth; + params.tile_width_spacing = 1; params.pixel_format = PixelFormatFromRenderTargetFormat(config.format); params.srgb_conversion = config.format == Tegra::RenderTargetFormat::BGRA8_SRGB || config.format == Tegra::RenderTargetFormat::RGBA8_SRGB; @@ -195,6 +199,7 @@ std::size_t SurfaceParams::InnerMemorySize(bool force_gl, bool layer_only, params.block_width = 1 << std::min(block_width, 5U); params.block_height = 1 << std::min(block_height, 5U); params.block_depth = 1 << std::min(block_depth, 5U); + params.tile_width_spacing = 1; params.pixel_format = PixelFormatFromDepthFormat(format); params.component_type = ComponentTypeFromDepthFormat(format); params.type = GetFormatType(params.pixel_format); @@ -221,6 +226,7 @@ std::size_t SurfaceParams::InnerMemorySize(bool force_gl, bool layer_only, params.block_width = params.is_tiled ? std::min(config.BlockWidth(), 32U) : 0, params.block_height = params.is_tiled ? std::min(config.BlockHeight(), 32U) : 0, params.block_depth = params.is_tiled ? std::min(config.BlockDepth(), 32U) : 0, + params.tile_width_spacing = 1; params.pixel_format = PixelFormatFromRenderTargetFormat(config.format); params.srgb_conversion = config.format == Tegra::RenderTargetFormat::BGRA8_SRGB || config.format == Tegra::RenderTargetFormat::RGBA8_SRGB; @@ -265,11 +271,11 @@ static constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> tex {GL_COMPRESSED_RG_RGTC2, GL_RG, GL_UNSIGNED_INT_8_8_8_8, ComponentType::UNorm, true}, // DXN2UNORM {GL_COMPRESSED_SIGNED_RG_RGTC2, GL_RG, GL_INT, ComponentType::SNorm, true}, // DXN2SNORM - {GL_COMPRESSED_RGBA_BPTC_UNORM_ARB, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8, ComponentType::UNorm, + {GL_COMPRESSED_RGBA_BPTC_UNORM, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8, ComponentType::UNorm, true}, // BC7U - {GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT_ARB, GL_RGB, GL_UNSIGNED_INT_8_8_8_8, - ComponentType::Float, true}, // BC6H_UF16 - {GL_COMPRESSED_RGB_BPTC_SIGNED_FLOAT_ARB, GL_RGB, GL_UNSIGNED_INT_8_8_8_8, ComponentType::Float, + {GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT, GL_RGB, GL_UNSIGNED_INT_8_8_8_8, ComponentType::Float, + true}, // BC6H_UF16 + {GL_COMPRESSED_RGB_BPTC_SIGNED_FLOAT, GL_RGB, GL_UNSIGNED_INT_8_8_8_8, ComponentType::Float, true}, // BC6H_SF16 {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_4X4 {GL_RG8, GL_RG, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // G8R8U @@ -306,8 +312,8 @@ static constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> tex true}, // DXT23_SRGB {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8, ComponentType::UNorm, true}, // DXT45_SRGB - {GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM_ARB, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8, - ComponentType::UNorm, true}, // BC7U_SRGB + {GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8, ComponentType::UNorm, + true}, // BC7U_SRGB {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_4X4_SRGB {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_8X8_SRGB {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_8X5_SRGB @@ -346,7 +352,7 @@ static GLenum SurfaceTargetToGL(SurfaceTarget target) { case SurfaceTarget::TextureCubemap: return GL_TEXTURE_CUBE_MAP; case SurfaceTarget::TextureCubeArray: - return GL_TEXTURE_CUBE_MAP_ARRAY_ARB; + return GL_TEXTURE_CUBE_MAP_ARRAY; } LOG_CRITICAL(Render_OpenGL, "Unimplemented texture target={}", static_cast<u32>(target)); UNREACHABLE(); @@ -370,174 +376,7 @@ MathUtil::Rectangle<u32> SurfaceParams::GetRect(u32 mip_level) const { return {0, actual_height, MipWidth(mip_level), 0}; } -template <bool morton_to_gl, PixelFormat format> -void MortonCopy(u32 stride, u32 block_height, u32 height, u32 block_depth, u32 depth, u8* gl_buffer, - std::size_t gl_buffer_size, VAddr addr) { - constexpr u32 bytes_per_pixel = GetBytesPerPixel(format); - - // With the BCn formats (DXT and DXN), each 4x4 tile is swizzled instead of just individual - // pixel values. - const u32 tile_size_x{GetDefaultBlockWidth(format)}; - const u32 tile_size_y{GetDefaultBlockHeight(format)}; - - if (morton_to_gl) { - Tegra::Texture::UnswizzleTexture(gl_buffer, addr, tile_size_x, tile_size_y, bytes_per_pixel, - stride, height, depth, block_height, block_depth); - } else { - Tegra::Texture::CopySwizzledData((stride + tile_size_x - 1) / tile_size_x, - (height + tile_size_y - 1) / tile_size_y, depth, - bytes_per_pixel, bytes_per_pixel, Memory::GetPointer(addr), - gl_buffer, false, block_height, block_depth); - } -} - -using GLConversionArray = std::array<void (*)(u32, u32, u32, u32, u32, u8*, std::size_t, VAddr), - VideoCore::Surface::MaxPixelFormat>; - -static constexpr GLConversionArray morton_to_gl_fns = { - // clang-format off - MortonCopy<true, PixelFormat::ABGR8U>, - MortonCopy<true, PixelFormat::ABGR8S>, - MortonCopy<true, PixelFormat::ABGR8UI>, - MortonCopy<true, PixelFormat::B5G6R5U>, - MortonCopy<true, PixelFormat::A2B10G10R10U>, - MortonCopy<true, PixelFormat::A1B5G5R5U>, - MortonCopy<true, PixelFormat::R8U>, - MortonCopy<true, PixelFormat::R8UI>, - MortonCopy<true, PixelFormat::RGBA16F>, - MortonCopy<true, PixelFormat::RGBA16U>, - MortonCopy<true, PixelFormat::RGBA16UI>, - MortonCopy<true, PixelFormat::R11FG11FB10F>, - MortonCopy<true, PixelFormat::RGBA32UI>, - MortonCopy<true, PixelFormat::DXT1>, - MortonCopy<true, PixelFormat::DXT23>, - MortonCopy<true, PixelFormat::DXT45>, - MortonCopy<true, PixelFormat::DXN1>, - MortonCopy<true, PixelFormat::DXN2UNORM>, - MortonCopy<true, PixelFormat::DXN2SNORM>, - MortonCopy<true, PixelFormat::BC7U>, - MortonCopy<true, PixelFormat::BC6H_UF16>, - MortonCopy<true, PixelFormat::BC6H_SF16>, - MortonCopy<true, PixelFormat::ASTC_2D_4X4>, - MortonCopy<true, PixelFormat::G8R8U>, - MortonCopy<true, PixelFormat::G8R8S>, - MortonCopy<true, PixelFormat::BGRA8>, - MortonCopy<true, PixelFormat::RGBA32F>, - MortonCopy<true, PixelFormat::RG32F>, - MortonCopy<true, PixelFormat::R32F>, - MortonCopy<true, PixelFormat::R16F>, - MortonCopy<true, PixelFormat::R16U>, - MortonCopy<true, PixelFormat::R16S>, - MortonCopy<true, PixelFormat::R16UI>, - MortonCopy<true, PixelFormat::R16I>, - MortonCopy<true, PixelFormat::RG16>, - MortonCopy<true, PixelFormat::RG16F>, - MortonCopy<true, PixelFormat::RG16UI>, - MortonCopy<true, PixelFormat::RG16I>, - MortonCopy<true, PixelFormat::RG16S>, - MortonCopy<true, PixelFormat::RGB32F>, - MortonCopy<true, PixelFormat::RGBA8_SRGB>, - MortonCopy<true, PixelFormat::RG8U>, - MortonCopy<true, PixelFormat::RG8S>, - MortonCopy<true, PixelFormat::RG32UI>, - MortonCopy<true, PixelFormat::R32UI>, - MortonCopy<true, PixelFormat::ASTC_2D_8X8>, - MortonCopy<true, PixelFormat::ASTC_2D_8X5>, - MortonCopy<true, PixelFormat::ASTC_2D_5X4>, - MortonCopy<true, PixelFormat::BGRA8_SRGB>, - MortonCopy<true, PixelFormat::DXT1_SRGB>, - MortonCopy<true, PixelFormat::DXT23_SRGB>, - MortonCopy<true, PixelFormat::DXT45_SRGB>, - MortonCopy<true, PixelFormat::BC7U_SRGB>, - MortonCopy<true, PixelFormat::ASTC_2D_4X4_SRGB>, - MortonCopy<true, PixelFormat::ASTC_2D_8X8_SRGB>, - MortonCopy<true, PixelFormat::ASTC_2D_8X5_SRGB>, - MortonCopy<true, PixelFormat::ASTC_2D_5X4_SRGB>, - MortonCopy<true, PixelFormat::ASTC_2D_5X5>, - MortonCopy<true, PixelFormat::ASTC_2D_5X5_SRGB>, - MortonCopy<true, PixelFormat::ASTC_2D_10X8>, - MortonCopy<true, PixelFormat::ASTC_2D_10X8_SRGB>, - MortonCopy<true, PixelFormat::Z32F>, - MortonCopy<true, PixelFormat::Z16>, - MortonCopy<true, PixelFormat::Z24S8>, - MortonCopy<true, PixelFormat::S8Z24>, - MortonCopy<true, PixelFormat::Z32FS8>, - // clang-format on -}; - -static constexpr GLConversionArray gl_to_morton_fns = { - // clang-format off - MortonCopy<false, PixelFormat::ABGR8U>, - MortonCopy<false, PixelFormat::ABGR8S>, - MortonCopy<false, PixelFormat::ABGR8UI>, - MortonCopy<false, PixelFormat::B5G6R5U>, - MortonCopy<false, PixelFormat::A2B10G10R10U>, - MortonCopy<false, PixelFormat::A1B5G5R5U>, - MortonCopy<false, PixelFormat::R8U>, - MortonCopy<false, PixelFormat::R8UI>, - MortonCopy<false, PixelFormat::RGBA16F>, - MortonCopy<false, PixelFormat::RGBA16U>, - MortonCopy<false, PixelFormat::RGBA16UI>, - MortonCopy<false, PixelFormat::R11FG11FB10F>, - MortonCopy<false, PixelFormat::RGBA32UI>, - MortonCopy<false, PixelFormat::DXT1>, - MortonCopy<false, PixelFormat::DXT23>, - MortonCopy<false, PixelFormat::DXT45>, - MortonCopy<false, PixelFormat::DXN1>, - MortonCopy<false, PixelFormat::DXN2UNORM>, - MortonCopy<false, PixelFormat::DXN2SNORM>, - MortonCopy<false, PixelFormat::BC7U>, - MortonCopy<false, PixelFormat::BC6H_UF16>, - MortonCopy<false, PixelFormat::BC6H_SF16>, - // TODO(Subv): Swizzling ASTC formats are not supported - nullptr, - MortonCopy<false, PixelFormat::G8R8U>, - MortonCopy<false, PixelFormat::G8R8S>, - MortonCopy<false, PixelFormat::BGRA8>, - MortonCopy<false, PixelFormat::RGBA32F>, - MortonCopy<false, PixelFormat::RG32F>, - MortonCopy<false, PixelFormat::R32F>, - MortonCopy<false, PixelFormat::R16F>, - MortonCopy<false, PixelFormat::R16U>, - MortonCopy<false, PixelFormat::R16S>, - MortonCopy<false, PixelFormat::R16UI>, - MortonCopy<false, PixelFormat::R16I>, - MortonCopy<false, PixelFormat::RG16>, - MortonCopy<false, PixelFormat::RG16F>, - MortonCopy<false, PixelFormat::RG16UI>, - MortonCopy<false, PixelFormat::RG16I>, - MortonCopy<false, PixelFormat::RG16S>, - MortonCopy<false, PixelFormat::RGB32F>, - MortonCopy<false, PixelFormat::RGBA8_SRGB>, - MortonCopy<false, PixelFormat::RG8U>, - MortonCopy<false, PixelFormat::RG8S>, - MortonCopy<false, PixelFormat::RG32UI>, - MortonCopy<false, PixelFormat::R32UI>, - nullptr, - nullptr, - nullptr, - MortonCopy<false, PixelFormat::BGRA8_SRGB>, - MortonCopy<false, PixelFormat::DXT1_SRGB>, - MortonCopy<false, PixelFormat::DXT23_SRGB>, - MortonCopy<false, PixelFormat::DXT45_SRGB>, - MortonCopy<false, PixelFormat::BC7U_SRGB>, - nullptr, - nullptr, - nullptr, - nullptr, - nullptr, - nullptr, - nullptr, - nullptr, - MortonCopy<false, PixelFormat::Z32F>, - MortonCopy<false, PixelFormat::Z16>, - MortonCopy<false, PixelFormat::Z24S8>, - MortonCopy<false, PixelFormat::S8Z24>, - MortonCopy<false, PixelFormat::Z32FS8>, - // clang-format on -}; - -void SwizzleFunc(const GLConversionArray& functions, const SurfaceParams& params, +void SwizzleFunc(const MortonSwizzleMode& mode, const SurfaceParams& params, std::vector<u8>& gl_buffer, u32 mip_level) { u32 depth = params.MipDepth(mip_level); if (params.target == SurfaceTarget::Texture2D) { @@ -550,19 +389,19 @@ void SwizzleFunc(const GLConversionArray& functions, const SurfaceParams& params const u64 layer_size = params.LayerMemorySize(); const u64 gl_size = params.LayerSizeGL(mip_level); for (u32 i = 0; i < params.depth; i++) { - functions[static_cast<std::size_t>(params.pixel_format)]( - params.MipWidth(mip_level), params.MipBlockHeight(mip_level), - params.MipHeight(mip_level), params.MipBlockDepth(mip_level), 1, - gl_buffer.data() + offset_gl, gl_size, params.addr + offset); + MortonSwizzle(mode, params.pixel_format, params.MipWidth(mip_level), + params.MipBlockHeight(mip_level), params.MipHeight(mip_level), + params.MipBlockDepth(mip_level), params.tile_width_spacing, 1, + gl_buffer.data() + offset_gl, gl_size, params.addr + offset); offset += layer_size; offset_gl += gl_size; } } else { const u64 offset = params.GetMipmapLevelOffset(mip_level); - functions[static_cast<std::size_t>(params.pixel_format)]( - params.MipWidth(mip_level), params.MipBlockHeight(mip_level), - params.MipHeight(mip_level), params.MipBlockDepth(mip_level), depth, gl_buffer.data(), - gl_buffer.size(), params.addr + offset); + MortonSwizzle(mode, params.pixel_format, params.MipWidth(mip_level), + params.MipBlockHeight(mip_level), params.MipHeight(mip_level), + params.MipBlockDepth(mip_level), depth, params.tile_width_spacing, + gl_buffer.data(), gl_buffer.size(), params.addr + offset); } } @@ -726,7 +565,7 @@ static void CopySurface(const Surface& src_surface, const Surface& dst_surface, const std::size_t buffer_size = std::max(src_params.size_in_bytes, dst_params.size_in_bytes); glBindBuffer(GL_PIXEL_PACK_BUFFER, copy_pbo_handle); - glBufferData(GL_PIXEL_PACK_BUFFER, buffer_size, nullptr, GL_STREAM_DRAW_ARB); + glBufferData(GL_PIXEL_PACK_BUFFER, buffer_size, nullptr, GL_STREAM_DRAW); if (source_format.compressed) { glGetCompressedTextureImage(src_surface->Texture().handle, src_attachment, static_cast<GLsizei>(src_params.size_in_bytes), nullptr); @@ -996,7 +835,7 @@ void CachedSurface::LoadGLBuffer() { ASSERT_MSG(params.block_width == 1, "Block width is defined as {} on texture type {}", params.block_width, static_cast<u32>(params.target)); for (u32 i = 0; i < params.max_mip_level; i++) - SwizzleFunc(morton_to_gl_fns, params, gl_buffer[i], i); + SwizzleFunc(MortonSwizzleMode::MortonToLinear, params, gl_buffer[i], i); } else { const auto texture_src_data{Memory::GetPointer(params.addr)}; const auto texture_src_data_end{texture_src_data + params.size_in_bytes_gl}; @@ -1035,7 +874,7 @@ void CachedSurface::FlushGLBuffer() { ASSERT_MSG(params.block_width == 1, "Block width is defined as {} on texture type {}", params.block_width, static_cast<u32>(params.target)); - SwizzleFunc(gl_to_morton_fns, params, gl_buffer[0], 0); + SwizzleFunc(MortonSwizzleMode::LinearToMorton, params, gl_buffer[0], 0); } else { std::memcpy(Memory::GetPointer(GetAddr()), gl_buffer[0].data(), GetSizeInBytes()); } @@ -1275,6 +1114,31 @@ Surface RasterizerCacheOpenGL::GetUncachedSurface(const SurfaceParams& params) { return surface; } +void RasterizerCacheOpenGL::FastLayeredCopySurface(const Surface& src_surface, + const Surface& dst_surface) { + const auto& init_params{src_surface->GetSurfaceParams()}; + const auto& dst_params{dst_surface->GetSurfaceParams()}; + VAddr address = init_params.addr; + const std::size_t layer_size = dst_params.LayerMemorySize(); + for (u32 layer = 0; layer < dst_params.depth; layer++) { + for (u32 mipmap = 0; mipmap < dst_params.max_mip_level; mipmap++) { + const VAddr sub_address = address + dst_params.GetMipmapLevelOffset(mipmap); + const Surface& copy = TryGet(sub_address); + if (!copy) + continue; + const auto& src_params{copy->GetSurfaceParams()}; + const u32 width{std::min(src_params.width, dst_params.MipWidth(mipmap))}; + const u32 height{std::min(src_params.height, dst_params.MipHeight(mipmap))}; + + glCopyImageSubData(copy->Texture().handle, SurfaceTargetToGL(src_params.target), 0, 0, + 0, 0, dst_surface->Texture().handle, + SurfaceTargetToGL(dst_params.target), mipmap, 0, 0, layer, width, + height, 1); + } + address += layer_size; + } +} + void RasterizerCacheOpenGL::FermiCopySurface( const Tegra::Engines::Fermi2D::Regs::Surface& src_config, const Tegra::Engines::Fermi2D::Regs::Surface& dst_config) { @@ -1340,11 +1204,13 @@ Surface RasterizerCacheOpenGL::RecreateSurface(const Surface& old_surface, CopySurface(old_surface, new_surface, copy_pbo.handle); } break; - case SurfaceTarget::TextureCubemap: case SurfaceTarget::Texture3D: + AccurateCopySurface(old_surface, new_surface); + break; + case SurfaceTarget::TextureCubemap: case SurfaceTarget::Texture2DArray: case SurfaceTarget::TextureCubeArray: - AccurateCopySurface(old_surface, new_surface); + FastLayeredCopySurface(old_surface, new_surface); break; default: LOG_CRITICAL(Render_OpenGL, "Unimplemented surface target={}", diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.h b/src/video_core/renderer_opengl/gl_rasterizer_cache.h index 494f6b903..c710aa245 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.h +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.h @@ -196,9 +196,15 @@ struct SurfaceParams { /// Checks if surfaces are compatible for caching bool IsCompatibleSurface(const SurfaceParams& other) const { - return std::tie(pixel_format, type, width, height, target, depth) == - std::tie(other.pixel_format, other.type, other.width, other.height, other.target, - other.depth); + if (std::tie(pixel_format, type, width, height, target, depth, is_tiled) == + std::tie(other.pixel_format, other.type, other.width, other.height, other.target, + other.depth, other.is_tiled)) { + if (!is_tiled) + return true; + return std::tie(block_height, block_depth, tile_width_spacing) == + std::tie(other.block_height, other.block_depth, other.tile_width_spacing); + } + return false; } /// Initializes parameters for caching, should be called after everything has been initialized @@ -208,6 +214,7 @@ struct SurfaceParams { u32 block_width; u32 block_height; u32 block_depth; + u32 tile_width_spacing; PixelFormat pixel_format; ComponentType component_type; SurfaceType type; @@ -350,6 +357,7 @@ private: /// Performs a slow but accurate surface copy, flushing to RAM and reinterpreting the data void AccurateCopySurface(const Surface& src_surface, const Surface& dst_surface); + void FastLayeredCopySurface(const Surface& src_surface, const Surface& dst_surface); /// The surface reserve is a "backup" cache, this is where we put unique surfaces that have /// previously been used. This is to prevent surfaces from being constantly created and diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp index a85a7c0c5..038b25c75 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp @@ -84,6 +84,7 @@ CachedShader::CachedShader(VAddr addr, Maxwell::ShaderProgram program_type) } entries = program_result.second; + shader_length = entries.shader_length; if (program_type != Maxwell::ShaderProgram::Geometry) { OGLShader shader; diff --git a/src/video_core/renderer_opengl/gl_shader_cache.h b/src/video_core/renderer_opengl/gl_shader_cache.h index ffbf21831..08f470de3 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.h +++ b/src/video_core/renderer_opengl/gl_shader_cache.h @@ -30,7 +30,7 @@ public: } std::size_t GetSizeInBytes() const override { - return GLShader::MAX_PROGRAM_CODE_LENGTH * sizeof(u64); + return shader_length; } // We do not have to flush this cache as things in it are never modified by us. @@ -82,6 +82,7 @@ private: u32 max_vertices, const std::string& debug_name); VAddr addr; + std::size_t shader_length; Maxwell::ShaderProgram program_type; GLShader::ShaderSetup setup; GLShader::ShaderEntries entries; diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index 90a88b91a..0c4524d5c 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -34,6 +34,17 @@ constexpr u32 PROGRAM_HEADER_SIZE = sizeof(Tegra::Shader::Header); constexpr u32 MAX_GEOMETRY_BUFFERS = 6; constexpr u32 MAX_ATTRIBUTES = 0x100; // Size in vec4s, this value is untested +static const char* INTERNAL_FLAG_NAMES[] = {"zero_flag", "sign_flag", "carry_flag", + "overflow_flag"}; + +enum class InternalFlag : u64 { + ZeroFlag = 0, + SignFlag = 1, + CarryFlag = 2, + OverflowFlag = 3, + Amount +}; + class DecompileFail : public std::runtime_error { public: using std::runtime_error::runtime_error; @@ -84,7 +95,8 @@ struct Subroutine { class ControlFlowAnalyzer { public: ControlFlowAnalyzer(const ProgramCode& program_code, u32 main_offset, const std::string& suffix) - : program_code(program_code) { + : program_code(program_code), shader_coverage_begin(main_offset), + shader_coverage_end(main_offset + 1) { // Recursively finds all subroutines. const Subroutine& program_main = AddSubroutine(main_offset, PROGRAM_END, suffix); @@ -96,10 +108,16 @@ public: return std::move(subroutines); } + std::size_t GetShaderLength() const { + return shader_coverage_end * sizeof(u64); + } + private: const ProgramCode& program_code; std::set<Subroutine> subroutines; std::map<std::pair<u32, u32>, ExitMethod> exit_method_map; + u32 shader_coverage_begin; + u32 shader_coverage_end; /// Adds and analyzes a new subroutine if it is not added yet. const Subroutine& AddSubroutine(u32 begin, u32 end, const std::string& suffix) { @@ -141,6 +159,9 @@ private: return exit_method; for (u32 offset = begin; offset != end && offset != PROGRAM_END; ++offset) { + shader_coverage_begin = std::min(shader_coverage_begin, offset); + shader_coverage_end = std::max(shader_coverage_end, offset + 1); + const Instruction instr = {program_code[offset]}; if (const auto opcode = OpCode::Decode(instr)) { switch (opcode->get().GetId()) { @@ -257,14 +278,6 @@ private: const std::string& suffix; }; -enum class InternalFlag : u64 { - ZeroFlag = 0, - CarryFlag = 1, - OverflowFlag = 2, - NaNFlag = 3, - Amount -}; - /** * Used to manage shader registers that are emulated with GLSL. This class keeps track of the state * of all registers (e.g. whether they are currently being used as Floats or Integers), and @@ -371,7 +384,7 @@ public: if (sets_cc) { const std::string zero_condition = "( " + ConvertIntegerSize(value, size) + " == 0 )"; SetInternalFlag(InternalFlag::ZeroFlag, zero_condition); - LOG_WARNING(HW_GPU, "Control Codes Imcomplete."); + LOG_WARNING(HW_GPU, "Condition codes implementation is incomplete."); } } @@ -454,23 +467,25 @@ public: shader.AddLine("lmem[" + index + "] = " + func + '(' + value + ");"); } - std::string GetControlCode(const Tegra::Shader::ControlCode cc) const { + std::string GetConditionCode(const Tegra::Shader::ConditionCode cc) const { switch (cc) { - case Tegra::Shader::ControlCode::NEU: + case Tegra::Shader::ConditionCode::NEU: return "!(" + GetInternalFlag(InternalFlag::ZeroFlag) + ')'; default: - UNIMPLEMENTED_MSG("Unimplemented Control Code: {}", static_cast<u32>(cc)); + UNIMPLEMENTED_MSG("Unimplemented condition code: {}", static_cast<u32>(cc)); return "false"; } } - std::string GetInternalFlag(const InternalFlag ii) const { - const u32 code = static_cast<u32>(ii); - return "internalFlag_" + std::to_string(code) + suffix; + std::string GetInternalFlag(const InternalFlag flag) const { + const auto index = static_cast<u32>(flag); + ASSERT(index < static_cast<u32>(InternalFlag::Amount)); + + return std::string(INTERNAL_FLAG_NAMES[index]) + '_' + suffix; } - void SetInternalFlag(const InternalFlag ii, const std::string& value) const { - shader.AddLine(GetInternalFlag(ii) + " = " + value + ';'); + void SetInternalFlag(const InternalFlag flag, const std::string& value) const { + shader.AddLine(GetInternalFlag(flag) + " = " + value + ';'); } /** @@ -485,27 +500,42 @@ public: const Register& buf_reg) { const std::string dest = GetOutputAttribute(attribute); const std::string src = GetRegisterAsFloat(val_reg); + if (dest.empty()) + return; - if (!dest.empty()) { - // Can happen with unknown/unimplemented output attributes, in which case we ignore the - // instruction for now. - if (stage == Maxwell3D::Regs::ShaderStage::Geometry) { - // TODO(Rodrigo): nouveau sets some attributes after setting emitting a geometry - // shader. These instructions use a dirty register as buffer index, to avoid some - // drivers from complaining about out of boundary writes, guard them. - const std::string buf_index{"((" + GetRegisterAsInteger(buf_reg) + ") % " + - std::to_string(MAX_GEOMETRY_BUFFERS) + ')'}; - shader.AddLine("amem[" + buf_index + "][" + - std::to_string(static_cast<u32>(attribute)) + ']' + - GetSwizzle(elem) + " = " + src + ';'); - } else { - if (attribute == Attribute::Index::PointSize) { - fixed_pipeline_output_attributes_used.insert(attribute); - shader.AddLine(dest + " = " + src + ';'); - } else { - shader.AddLine(dest + GetSwizzle(elem) + " = " + src + ';'); - } - } + // Can happen with unknown/unimplemented output attributes, in which case we ignore the + // instruction for now. + if (stage == Maxwell3D::Regs::ShaderStage::Geometry) { + // TODO(Rodrigo): nouveau sets some attributes after setting emitting a geometry + // shader. These instructions use a dirty register as buffer index, to avoid some + // drivers from complaining about out of boundary writes, guard them. + const std::string buf_index{"((" + GetRegisterAsInteger(buf_reg) + ") % " + + std::to_string(MAX_GEOMETRY_BUFFERS) + ')'}; + shader.AddLine("amem[" + buf_index + "][" + + std::to_string(static_cast<u32>(attribute)) + ']' + GetSwizzle(elem) + + " = " + src + ';'); + return; + } + + switch (attribute) { + case Attribute::Index::ClipDistances0123: + case Attribute::Index::ClipDistances4567: { + const u64 index = (attribute == Attribute::Index::ClipDistances4567 ? 4 : 0) + elem; + UNIMPLEMENTED_IF_MSG( + ((header.vtg.clip_distances >> index) & 1) == 0, + "Shader is setting gl_ClipDistance{} without enabling it in the header", index); + + fixed_pipeline_output_attributes_used.insert(attribute); + shader.AddLine(dest + '[' + std::to_string(index) + "] = " + src + ';'); + break; + } + case Attribute::Index::PointSize: + fixed_pipeline_output_attributes_used.insert(attribute); + shader.AddLine(dest + " = " + src + ';'); + break; + default: + shader.AddLine(dest + GetSwizzle(elem) + " = " + src + ';'); + break; } } @@ -621,8 +651,8 @@ private: /// Generates declarations for internal flags. void GenerateInternalFlags() { - for (u32 ii = 0; ii < static_cast<u64>(InternalFlag::Amount); ii++) { - const InternalFlag code = static_cast<InternalFlag>(ii); + for (u32 flag = 0; flag < static_cast<u32>(InternalFlag::Amount); flag++) { + const InternalFlag code = static_cast<InternalFlag>(flag); declarations.AddLine("bool " + GetInternalFlag(code) + " = false;"); } declarations.AddNewLine(); @@ -725,12 +755,19 @@ private: void GenerateVertex() { if (stage != Maxwell3D::Regs::ShaderStage::Vertex) return; + bool clip_distances_declared = false; + declarations.AddLine("out gl_PerVertex {"); ++declarations.scope; declarations.AddLine("vec4 gl_Position;"); for (auto& o : fixed_pipeline_output_attributes_used) { if (o == Attribute::Index::PointSize) declarations.AddLine("float gl_PointSize;"); + if (!clip_distances_declared && (o == Attribute::Index::ClipDistances0123 || + o == Attribute::Index::ClipDistances4567)) { + declarations.AddLine("float gl_ClipDistance[];"); + clip_distances_declared = true; + } } --declarations.scope; declarations.AddLine("};"); @@ -830,7 +867,8 @@ private: // vertex shader, and what's the value of the fourth element when inside a Tess Eval // shader. ASSERT(stage == Maxwell3D::Regs::ShaderStage::Vertex); - return "vec4(0, 0, uintBitsToFloat(instance_id.x), uintBitsToFloat(gl_VertexID))"; + // Config pack's first value is instance_id. + return "vec4(0, 0, uintBitsToFloat(config_pack[0]), uintBitsToFloat(gl_VertexID))"; case Attribute::Index::FrontFacing: // TODO(Subv): Find out what the values are for the other elements. ASSERT(stage == Maxwell3D::Regs::ShaderStage::Fragment); @@ -901,6 +939,10 @@ private: return "gl_PointSize"; case Attribute::Index::Position: return "position"; + case Attribute::Index::ClipDistances0123: + case Attribute::Index::ClipDistances4567: { + return "gl_ClipDistance"; + } default: const u32 index{static_cast<u32>(attribute) - static_cast<u32>(Attribute::Index::Attribute_0)}; @@ -939,9 +981,10 @@ private: class GLSLGenerator { public: GLSLGenerator(const std::set<Subroutine>& subroutines, const ProgramCode& program_code, - u32 main_offset, Maxwell3D::Regs::ShaderStage stage, const std::string& suffix) + u32 main_offset, Maxwell3D::Regs::ShaderStage stage, const std::string& suffix, + std::size_t shader_length) : subroutines(subroutines), program_code(program_code), main_offset(main_offset), - stage(stage), suffix(suffix) { + stage(stage), suffix(suffix), shader_length(shader_length) { std::memcpy(&header, program_code.data(), sizeof(Tegra::Shader::Header)); local_memory_size = header.GetLocalMemorySize(); regs.SetLocalMemory(local_memory_size); @@ -954,7 +997,7 @@ public: /// Returns entries in the shader that are useful for external functions ShaderEntries GetEntries() const { - return {regs.GetConstBuffersDeclarations(), regs.GetSamplers()}; + return {regs.GetConstBuffersDeclarations(), regs.GetSamplers(), shader_length}; } private: @@ -1059,11 +1102,17 @@ private: const std::string& op_a, const std::string& op_b) const { using Tegra::Shader::PredCondition; static const std::unordered_map<PredCondition, const char*> PredicateComparisonStrings = { - {PredCondition::LessThan, "<"}, {PredCondition::Equal, "=="}, - {PredCondition::LessEqual, "<="}, {PredCondition::GreaterThan, ">"}, - {PredCondition::NotEqual, "!="}, {PredCondition::GreaterEqual, ">="}, - {PredCondition::LessThanWithNan, "<"}, {PredCondition::NotEqualWithNan, "!="}, - {PredCondition::GreaterThanWithNan, ">"}, {PredCondition::GreaterEqualWithNan, ">="}}; + {PredCondition::LessThan, "<"}, + {PredCondition::Equal, "=="}, + {PredCondition::LessEqual, "<="}, + {PredCondition::GreaterThan, ">"}, + {PredCondition::NotEqual, "!="}, + {PredCondition::GreaterEqual, ">="}, + {PredCondition::LessThanWithNan, "<"}, + {PredCondition::NotEqualWithNan, "!="}, + {PredCondition::LessEqualWithNan, "<="}, + {PredCondition::GreaterThanWithNan, ">"}, + {PredCondition::GreaterEqualWithNan, ">="}}; const auto& comparison{PredicateComparisonStrings.find(condition)}; UNIMPLEMENTED_IF_MSG(comparison == PredicateComparisonStrings.end(), @@ -1072,6 +1121,7 @@ private: std::string predicate{'(' + op_a + ") " + comparison->second + " (" + op_b + ')'}; if (condition == PredCondition::LessThanWithNan || condition == PredCondition::NotEqualWithNan || + condition == PredCondition::LessEqualWithNan || condition == PredCondition::GreaterThanWithNan || condition == PredCondition::GreaterEqualWithNan) { predicate += " || isnan(" + op_a + ") || isnan(" + op_b + ')'; @@ -1250,6 +1300,7 @@ private: shader.AddLine('{'); ++shader.scope; shader.AddLine(coord); + shader.AddLine("vec4 texture_tmp = " + texture + ';'); // TEXS has two destination registers and a swizzle. The first two elements in the swizzle // go into gpr0+0 and gpr0+1, and the rest goes into gpr28+0 and gpr28+1 @@ -1262,18 +1313,17 @@ private: if (written_components < 2) { // Write the first two swizzle components to gpr0 and gpr0+1 - regs.SetRegisterToFloat(instr.gpr0, component, texture, 1, 4, false, + regs.SetRegisterToFloat(instr.gpr0, component, "texture_tmp", 1, 4, false, written_components % 2); } else { ASSERT(instr.texs.HasTwoDestinations()); // Write the rest of the swizzle components to gpr28 and gpr28+1 - regs.SetRegisterToFloat(instr.gpr28, component, texture, 1, 4, false, + regs.SetRegisterToFloat(instr.gpr28, component, "texture_tmp", 1, 4, false, written_components % 2); } ++written_components; } - --shader.scope; shader.AddLine('}'); } @@ -1508,9 +1558,8 @@ private: instr.fmul.tab5c68_0 != 1, "FMUL tab5cb8_0({}) is not implemented", instr.fmul.tab5c68_0 .Value()); // SMO typical sends 1 here which seems to be the default - UNIMPLEMENTED_IF_MSG(instr.fmul.cc != 0, "FMUL cc is not implemented"); UNIMPLEMENTED_IF_MSG(instr.generates_cc, - "FMUL Generates an unhandled Control Code"); + "Condition codes generation in FMUL is not implemented"); op_b = GetOperandAbsNeg(op_b, false, instr.fmul.negate_b); @@ -1522,7 +1571,7 @@ private: case OpCode::Id::FADD_R: case OpCode::Id::FADD_IMM: { UNIMPLEMENTED_IF_MSG(instr.generates_cc, - "FADD Generates an unhandled Control Code"); + "Condition codes generation in FADD is not implemented"); op_a = GetOperandAbsNeg(op_a, instr.alu.abs_a, instr.alu.negate_a); op_b = GetOperandAbsNeg(op_b, instr.alu.abs_b, instr.alu.negate_b); @@ -1572,7 +1621,7 @@ private: case OpCode::Id::FMNMX_R: case OpCode::Id::FMNMX_IMM: { UNIMPLEMENTED_IF_MSG(instr.generates_cc, - "FMNMX Generates an unhandled Control Code"); + "Condition codes generation in FMNMX is not implemented"); op_a = GetOperandAbsNeg(op_a, instr.alu.abs_a, instr.alu.negate_a); op_b = GetOperandAbsNeg(op_b, instr.alu.abs_b, instr.alu.negate_b); @@ -1609,7 +1658,7 @@ private: } case OpCode::Id::FMUL32_IMM: { UNIMPLEMENTED_IF_MSG(instr.op_32.generates_cc, - "FMUL32 Generates an unhandled Control Code"); + "Condition codes generation in FMUL32 is not implemented"); regs.SetRegisterToFloat(instr.gpr0, 0, regs.GetRegisterAsFloat(instr.gpr8) + " * " + @@ -1619,7 +1668,7 @@ private: } case OpCode::Id::FADD32I: { UNIMPLEMENTED_IF_MSG(instr.op_32.generates_cc, - "FADD32 Generates an unhandled Control Code"); + "Condition codes generation in FADD32I is not implemented"); std::string op_a = regs.GetRegisterAsFloat(instr.gpr8); std::string op_b = GetImmediate32(instr); @@ -1654,7 +1703,8 @@ private: switch (opcode->get().GetId()) { case OpCode::Id::BFE_IMM: { - UNIMPLEMENTED_IF_MSG(instr.generates_cc, "BFE Generates an unhandled Control Code"); + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "Condition codes generation in BFE is not implemented"); std::string inner_shift = '(' + op_a + " << " + std::to_string(instr.bfe.GetLeftShiftValue()) + ')'; @@ -1672,6 +1722,26 @@ private: break; } + case OpCode::Type::Bfi: { + UNIMPLEMENTED_IF(instr.generates_cc); + + const auto [base, packed_shift] = [&]() -> std::tuple<std::string, std::string> { + switch (opcode->get().GetId()) { + case OpCode::Id::BFI_IMM_R: + return {regs.GetRegisterAsInteger(instr.gpr39, 0, false), + std::to_string(instr.alu.GetSignedImm20_20())}; + default: + UNREACHABLE(); + } + }(); + const std::string offset = '(' + packed_shift + " & 0xff)"; + const std::string bits = "((" + packed_shift + " >> 8) & 0xff)"; + const std::string insert = regs.GetRegisterAsInteger(instr.gpr8, 0, false); + regs.SetRegisterToInteger( + instr.gpr0, false, 0, + "bitfieldInsert(" + base + ", " + insert + ", " + offset + ", " + bits + ')', 1, 1); + break; + } case OpCode::Type::Shift: { std::string op_a = regs.GetRegisterAsInteger(instr.gpr8, 0, true); std::string op_b; @@ -1691,7 +1761,8 @@ private: case OpCode::Id::SHR_C: case OpCode::Id::SHR_R: case OpCode::Id::SHR_IMM: { - UNIMPLEMENTED_IF_MSG(instr.generates_cc, "SHR Generates an unhandled Control Code"); + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "Condition codes generation in SHR is not implemented"); if (!instr.shift.is_signed) { // Logical shift right @@ -1706,8 +1777,8 @@ private: case OpCode::Id::SHL_C: case OpCode::Id::SHL_R: case OpCode::Id::SHL_IMM: - UNIMPLEMENTED_IF_MSG(instr.generates_cc, "SHL Generates an unhandled Control Code"); - + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "Condition codes generation in SHL is not implemented"); regs.SetRegisterToInteger(instr.gpr0, true, 0, op_a + " << " + op_b, 1, 1); break; default: { @@ -1723,7 +1794,7 @@ private: switch (opcode->get().GetId()) { case OpCode::Id::IADD32I: UNIMPLEMENTED_IF_MSG(instr.op_32.generates_cc, - "IADD32 Generates an unhandled Control Code"); + "Condition codes generation in IADD32I is not implemented"); if (instr.iadd32i.negate_a) op_a = "-(" + op_a + ')'; @@ -1733,7 +1804,7 @@ private: break; case OpCode::Id::LOP32I: { UNIMPLEMENTED_IF_MSG(instr.op_32.generates_cc, - "LOP32I Generates an unhandled Control Code"); + "Condition codes generation in LOP32I is not implemented"); if (instr.alu.lop32i.invert_a) op_a = "~(" + op_a + ')'; @@ -1772,7 +1843,7 @@ private: case OpCode::Id::IADD_R: case OpCode::Id::IADD_IMM: { UNIMPLEMENTED_IF_MSG(instr.generates_cc, - "IADD Generates an unhandled Control Code"); + "Condition codes generation in IADD is not implemented"); if (instr.alu_integer.negate_a) op_a = "-(" + op_a + ')'; @@ -1788,7 +1859,7 @@ private: case OpCode::Id::IADD3_R: case OpCode::Id::IADD3_IMM: { UNIMPLEMENTED_IF_MSG(instr.generates_cc, - "IADD3 Generates an unhandled Control Code"); + "Condition codes generation in IADD3 is not implemented"); std::string op_c = regs.GetRegisterAsInteger(instr.gpr39); @@ -1851,7 +1922,7 @@ private: case OpCode::Id::ISCADD_R: case OpCode::Id::ISCADD_IMM: { UNIMPLEMENTED_IF_MSG(instr.generates_cc, - "ISCADD Generates an unhandled Control Code"); + "Condition codes generation in ISCADD is not implemented"); if (instr.alu_integer.negate_a) op_a = "-(" + op_a + ')'; @@ -1886,7 +1957,8 @@ private: case OpCode::Id::LOP_C: case OpCode::Id::LOP_R: case OpCode::Id::LOP_IMM: { - UNIMPLEMENTED_IF_MSG(instr.generates_cc, "LOP Generates an unhandled Control Code"); + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "Condition codes generation in LOP is not implemented"); if (instr.alu.lop.invert_a) op_a = "~(" + op_a + ')'; @@ -1902,7 +1974,7 @@ private: case OpCode::Id::LOP3_R: case OpCode::Id::LOP3_IMM: { UNIMPLEMENTED_IF_MSG(instr.generates_cc, - "LOP3 Generates an unhandled Control Code"); + "Condition codes generation in LOP3 is not implemented"); const std::string op_c = regs.GetRegisterAsInteger(instr.gpr39); std::string lut; @@ -1921,7 +1993,7 @@ private: case OpCode::Id::IMNMX_IMM: { UNIMPLEMENTED_IF(instr.imnmx.exchange != Tegra::Shader::IMinMaxExchange::None); UNIMPLEMENTED_IF_MSG(instr.generates_cc, - "IMNMX Generates an unhandled Control Code"); + "Condition codes generation in IMNMX is not implemented"); const std::string condition = GetPredicateCondition(instr.imnmx.pred, instr.imnmx.negate_pred != 0); @@ -2094,7 +2166,8 @@ private: instr.ffma.tab5980_0.Value()); // Seems to be 1 by default based on SMO UNIMPLEMENTED_IF_MSG(instr.ffma.tab5980_1 != 0, "FFMA tab5980_1({}) not implemented", instr.ffma.tab5980_1.Value()); - UNIMPLEMENTED_IF_MSG(instr.generates_cc, "FFMA Generates an unhandled Control Code"); + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "Condition codes generation in FFMA is not implemented"); switch (opcode->get().GetId()) { case OpCode::Id::FFMA_CR: { @@ -2204,7 +2277,8 @@ private: case OpCode::Id::I2F_C: { UNIMPLEMENTED_IF(instr.conversion.dest_size != Register::Size::Word); UNIMPLEMENTED_IF(instr.conversion.selector); - UNIMPLEMENTED_IF_MSG(instr.generates_cc, "I2F Generates an unhandled Control Code"); + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "Condition codes generation in I2F is not implemented"); std::string op_a{}; @@ -2234,7 +2308,8 @@ private: case OpCode::Id::F2F_R: { UNIMPLEMENTED_IF(instr.conversion.dest_size != Register::Size::Word); UNIMPLEMENTED_IF(instr.conversion.src_size != Register::Size::Word); - UNIMPLEMENTED_IF_MSG(instr.generates_cc, "F2F Generates an unhandled Control Code"); + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "Condition codes generation in F2F is not implemented"); std::string op_a = regs.GetRegisterAsFloat(instr.gpr20); if (instr.conversion.abs_a) { @@ -2272,7 +2347,8 @@ private: case OpCode::Id::F2I_R: case OpCode::Id::F2I_C: { UNIMPLEMENTED_IF(instr.conversion.src_size != Register::Size::Word); - UNIMPLEMENTED_IF_MSG(instr.generates_cc, "F2I Generates an unhandled Control Code"); + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "Condition codes generation in F2I is not implemented"); std::string op_a{}; if (instr.is_b_gpr) { @@ -2491,61 +2567,83 @@ private: const bool depth_compare = instr.tex.UsesMiscMode(Tegra::Shader::TextureMiscMode::DC); u32 num_coordinates = TextureCoordinates(texture_type); - if (depth_compare) - num_coordinates += 1; + u32 start_index = 0; + std::string array_elem; + if (is_array) { + array_elem = regs.GetRegisterAsInteger(instr.gpr8); + start_index = 1; + } + const auto process_mode = instr.tex.GetTextureProcessMode(); + u32 start_index_b = 0; + std::string lod_value; + if (process_mode != Tegra::Shader::TextureProcessMode::LZ && + process_mode != Tegra::Shader::TextureProcessMode::None) { + start_index_b = 1; + lod_value = regs.GetRegisterAsFloat(instr.gpr20); + } + + std::string depth_value; + if (depth_compare) { + depth_value = regs.GetRegisterAsFloat(instr.gpr20.Value() + start_index_b); + } + + bool depth_compare_extra = false; switch (num_coordinates) { case 1: { + const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + start_index); if (is_array) { - const std::string index = regs.GetRegisterAsInteger(instr.gpr8); - const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - coord = "vec2 coords = vec2(" + x + ", " + index + ");"; + if (depth_compare) { + coord = "vec3 coords = vec3(" + x + ", " + depth_value + ", " + + array_elem + ");"; + } else { + coord = "vec2 coords = vec2(" + x + ", " + array_elem + ");"; + } } else { - const std::string x = regs.GetRegisterAsFloat(instr.gpr8); - coord = "float coords = " + x + ';'; + if (depth_compare) { + coord = "vec2 coords = vec2(" + x + ", " + depth_value + ");"; + } else { + coord = "float coords = " + x + ';'; + } } break; } case 2: { + const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + start_index); + const std::string y = + regs.GetRegisterAsFloat(instr.gpr8.Value() + start_index + 1); if (is_array) { - const std::string index = regs.GetRegisterAsInteger(instr.gpr8); - const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 2); - coord = "vec3 coords = vec3(" + x + ", " + y + ", " + index + ");"; + if (depth_compare) { + coord = "vec4 coords = vec4(" + x + ", " + y + ", " + depth_value + + ", " + array_elem + ");"; + } else { + coord = "vec3 coords = vec3(" + x + ", " + y + ", " + array_elem + ");"; + } } else { - const std::string x = regs.GetRegisterAsFloat(instr.gpr8); - const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - coord = "vec2 coords = vec2(" + x + ", " + y + ");"; + if (depth_compare) { + coord = + "vec3 coords = vec3(" + x + ", " + y + ", " + depth_value + ");"; + } else { + coord = "vec2 coords = vec2(" + x + ", " + y + ");"; + } } break; } case 3: { - if (depth_compare) { - if (is_array) { - const std::string index = regs.GetRegisterAsInteger(instr.gpr8); - const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - const std::string y = regs.GetRegisterAsFloat(instr.gpr20); - const std::string z = regs.GetRegisterAsFloat(instr.gpr20.Value() + 1); - coord = "vec4 coords = vec4(" + x + ", " + y + ", " + z + ", " + index + - ");"; - } else { - const std::string x = regs.GetRegisterAsFloat(instr.gpr8); - const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - const std::string z = regs.GetRegisterAsFloat(instr.gpr20); - coord = "vec3 coords = vec3(" + x + ", " + y + ", " + z + ");"; - } + const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + start_index); + const std::string y = + regs.GetRegisterAsFloat(instr.gpr8.Value() + start_index + 1); + const std::string z = + regs.GetRegisterAsFloat(instr.gpr8.Value() + start_index + 2); + if (is_array) { + depth_compare_extra = depth_compare; + coord = "vec4 coords = vec4(" + x + ", " + y + ", " + z + ", " + + array_elem + ");"; } else { - if (is_array) { - const std::string index = regs.GetRegisterAsInteger(instr.gpr8); - const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 2); - const std::string z = regs.GetRegisterAsFloat(instr.gpr8.Value() + 3); - coord = "vec4 coords = vec4(" + x + ", " + y + ", " + z + ", " + index + - ");"; + if (depth_compare) { + coord = "vec4 coords = vec4(" + x + ", " + y + ", " + z + ", " + + depth_value + ");"; } else { - const std::string x = regs.GetRegisterAsFloat(instr.gpr8); - const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - const std::string z = regs.GetRegisterAsFloat(instr.gpr8.Value() + 2); coord = "vec3 coords = vec3(" + x + ", " + y + ", " + z + ");"; } } @@ -2561,79 +2659,88 @@ private: coord = "vec2 coords = vec2(" + x + ", " + y + ");"; texture_type = Tegra::Shader::TextureType::Texture2D; } - // TODO: make sure coordinates are always indexed to gpr8 and gpr20 is always bias - // or lod. - std::string op_c; const std::string sampler = GetSampler(instr.sampler, texture_type, is_array, depth_compare); // Add an extra scope and declare the texture coords inside to prevent // overwriting them in case they are used as outputs of the texs instruction. - shader.AddLine("{"); + shader.AddLine('{'); ++shader.scope; shader.AddLine(coord); std::string texture; switch (instr.tex.GetTextureProcessMode()) { case Tegra::Shader::TextureProcessMode::None: { - texture = "texture(" + sampler + ", coords)"; + if (!depth_compare_extra) { + texture = "texture(" + sampler + ", coords)"; + } else { + texture = "texture(" + sampler + ", coords, " + depth_value + ')'; + } break; } case Tegra::Shader::TextureProcessMode::LZ: { - texture = "textureLod(" + sampler + ", coords, 0.0)"; + if (!depth_compare_extra) { + texture = "textureLod(" + sampler + ", coords, 0.0)"; + } else { + texture = "texture(" + sampler + ", coords, " + depth_value + ')'; + } break; } case Tegra::Shader::TextureProcessMode::LB: case Tegra::Shader::TextureProcessMode::LBA: { - if (depth_compare) { - if (is_array) - op_c = regs.GetRegisterAsFloat(instr.gpr20.Value() + 2); - else - op_c = regs.GetRegisterAsFloat(instr.gpr20.Value() + 1); + // TODO: Figure if A suffix changes the equation at all. + if (!depth_compare_extra) { + texture = "texture(" + sampler + ", coords, " + lod_value + ')'; } else { - op_c = regs.GetRegisterAsFloat(instr.gpr20); + texture = "texture(" + sampler + ", coords, " + depth_value + ')'; + LOG_WARNING(HW_GPU, + "OpenGL Limitation: can't set bias value along depth compare"); } - // TODO: Figure if A suffix changes the equation at all. - texture = "texture(" + sampler + ", coords, " + op_c + ')'; break; } case Tegra::Shader::TextureProcessMode::LL: case Tegra::Shader::TextureProcessMode::LLA: { - if (num_coordinates <= 2) { - op_c = regs.GetRegisterAsFloat(instr.gpr20); + // TODO: Figure if A suffix changes the equation at all. + if (!depth_compare_extra) { + texture = "textureLod(" + sampler + ", coords, " + lod_value + ')'; } else { - op_c = regs.GetRegisterAsFloat(instr.gpr20.Value() + 1); + texture = "texture(" + sampler + ", coords, " + depth_value + ')'; + LOG_WARNING(HW_GPU, + "OpenGL Limitation: can't set lod value along depth compare"); } - // TODO: Figure if A suffix changes the equation at all. - texture = "textureLod(" + sampler + ", coords, " + op_c + ')'; break; } default: { - texture = "texture(" + sampler + ", coords)"; + if (!depth_compare_extra) { + texture = "texture(" + sampler + ", coords)"; + } else { + texture = "texture(" + sampler + ", coords, " + depth_value + ')'; + } UNIMPLEMENTED_MSG("Unhandled texture process mode {}", static_cast<u32>(instr.tex.GetTextureProcessMode())); } } if (!depth_compare) { + shader.AddLine("vec4 texture_tmp = " + texture + ';'); std::size_t dest_elem{}; for (std::size_t elem = 0; elem < 4; ++elem) { if (!instr.tex.IsComponentEnabled(elem)) { // Skip disabled components continue; } - regs.SetRegisterToFloat(instr.gpr0, elem, texture, 1, 4, false, dest_elem); + regs.SetRegisterToFloat(instr.gpr0, elem, "texture_tmp", 1, 4, false, + dest_elem); ++dest_elem; } } else { regs.SetRegisterToFloat(instr.gpr0, 0, texture, 1, 1, false); } --shader.scope; - shader.AddLine("}"); + shader.AddLine('}'); break; } case OpCode::Id::TEXS: { - std::string coord; Tegra::Shader::TextureType texture_type{instr.texs.GetTextureType()}; bool is_array{instr.texs.IsArrayTexture()}; @@ -2643,37 +2750,76 @@ private: const bool depth_compare = instr.texs.UsesMiscMode(Tegra::Shader::TextureMiscMode::DC); u32 num_coordinates = TextureCoordinates(texture_type); - if (depth_compare) - num_coordinates += 1; + const auto process_mode = instr.texs.GetTextureProcessMode(); + std::string lod_value; + std::string coord; + u32 lod_offset = 0; + if (process_mode == Tegra::Shader::TextureProcessMode::LL) { + if (num_coordinates > 2) { + lod_value = regs.GetRegisterAsFloat(instr.gpr20.Value() + 1); + lod_offset = 2; + } else { + lod_value = regs.GetRegisterAsFloat(instr.gpr20); + lod_offset = 1; + } + } switch (num_coordinates) { + case 1: { + coord = "float coords = " + regs.GetRegisterAsFloat(instr.gpr8) + ';'; + break; + } case 2: { if (is_array) { - const std::string index = regs.GetRegisterAsInteger(instr.gpr8); - const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - const std::string y = regs.GetRegisterAsFloat(instr.gpr20); - coord = "vec3 coords = vec3(" + x + ", " + y + ", " + index + ");"; + if (depth_compare) { + const std::string index = regs.GetRegisterAsInteger(instr.gpr8); + const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); + const std::string y = regs.GetRegisterAsFloat(instr.gpr20); + const std::string z = regs.GetRegisterAsFloat(instr.gpr20.Value() + 1); + coord = "vec4 coords = vec4(" + x + ", " + y + ", " + z + ", " + index + + ");"; + } else { + const std::string index = regs.GetRegisterAsInteger(instr.gpr8); + const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); + const std::string y = regs.GetRegisterAsFloat(instr.gpr20); + coord = "vec3 coords = vec3(" + x + ", " + y + ", " + index + ");"; + } } else { - const std::string x = regs.GetRegisterAsFloat(instr.gpr8); - const std::string y = regs.GetRegisterAsFloat(instr.gpr20); - coord = "vec2 coords = vec2(" + x + ", " + y + ");"; + if (lod_offset != 0) { + if (depth_compare) { + const std::string x = regs.GetRegisterAsFloat(instr.gpr8); + const std::string y = + regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); + const std::string z = + regs.GetRegisterAsFloat(instr.gpr20.Value() + lod_offset); + coord = "vec3 coords = vec3(" + x + ", " + y + ", " + z + ");"; + } else { + const std::string x = regs.GetRegisterAsFloat(instr.gpr8); + const std::string y = + regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); + coord = "vec2 coords = vec2(" + x + ", " + y + ");"; + } + } else { + if (depth_compare) { + const std::string x = regs.GetRegisterAsFloat(instr.gpr8); + const std::string y = + regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); + const std::string z = regs.GetRegisterAsFloat(instr.gpr20); + coord = "vec3 coords = vec3(" + x + ", " + y + ", " + z + ");"; + } else { + const std::string x = regs.GetRegisterAsFloat(instr.gpr8); + const std::string y = regs.GetRegisterAsFloat(instr.gpr20); + coord = "vec2 coords = vec2(" + x + ", " + y + ");"; + } + } } break; } case 3: { - if (is_array) { - const std::string index = regs.GetRegisterAsInteger(instr.gpr8); - const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 2); - const std::string z = regs.GetRegisterAsFloat(instr.gpr20); - coord = - "vec4 coords = vec4(" + x + ", " + y + ", " + z + ", " + index + ");"; - } else { - const std::string x = regs.GetRegisterAsFloat(instr.gpr8); - const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - const std::string z = regs.GetRegisterAsFloat(instr.gpr20); - coord = "vec3 coords = vec3(" + x + ", " + y + ", " + z + ");"; - } + const std::string x = regs.GetRegisterAsFloat(instr.gpr8); + const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); + const std::string z = regs.GetRegisterAsFloat(instr.gpr20); + coord = "vec3 coords = vec3(" + x + ", " + y + ", " + z + ");"; break; } default: @@ -2690,7 +2836,7 @@ private: const std::string sampler = GetSampler(instr.sampler, texture_type, is_array, depth_compare); std::string texture; - switch (instr.texs.GetTextureProcessMode()) { + switch (process_mode) { case Tegra::Shader::TextureProcessMode::None: { texture = "texture(" + sampler + ", coords)"; break; @@ -2704,8 +2850,7 @@ private: break; } case Tegra::Shader::TextureProcessMode::LL: { - const std::string op_c = regs.GetRegisterAsFloat(instr.gpr20.Value() + 1); - texture = "textureLod(" + sampler + ", coords, " + op_c + ')'; + texture = "textureLod(" + sampler + ", coords, " + lod_value + ')'; break; } default: { @@ -2719,10 +2864,10 @@ private: } else { WriteTexsInstruction(instr, coord, "vec4(" + texture + ')'); } + break; } case OpCode::Id::TLDS: { - std::string coord; const Tegra::Shader::TextureType texture_type{instr.tlds.GetTextureType()}; const bool is_array{instr.tlds.IsArrayTexture()}; @@ -2736,12 +2881,17 @@ private: UNIMPLEMENTED_IF_MSG(instr.tlds.UsesMiscMode(Tegra::Shader::TextureMiscMode::MZ), "MZ is not implemented"); - u32 op_c_offset = 0; + u32 extra_op_offset = 0; + + // Scope to avoid variable name overlaps. + shader.AddLine('{'); + ++shader.scope; + std::string coords; switch (texture_type) { case Tegra::Shader::TextureType::Texture1D: { const std::string x = regs.GetRegisterAsInteger(instr.gpr8); - coord = "int coords = " + x + ';'; + coords = "float coords = " + x + ';'; break; } case Tegra::Shader::TextureType::Texture2D: { @@ -2749,8 +2899,9 @@ private: const std::string x = regs.GetRegisterAsInteger(instr.gpr8); const std::string y = regs.GetRegisterAsInteger(instr.gpr20); - coord = "ivec2 coords = ivec2(" + x + ", " + y + ");"; - op_c_offset = 1; + // shader.AddLine("ivec2 coords = ivec2(" + x + ", " + y + ");"); + coords = "ivec2 coords = ivec2(" + x + ", " + y + ");"; + extra_op_offset = 1; break; } default: @@ -2765,9 +2916,10 @@ private: break; } case Tegra::Shader::TextureProcessMode::LL: { - const std::string op_c = - regs.GetRegisterAsInteger(instr.gpr20.Value() + op_c_offset); - texture = "texelFetch(" + sampler + ", coords, " + op_c + ')'; + shader.AddLine( + "float lod = " + + regs.GetRegisterAsInteger(instr.gpr20.Value() + extra_op_offset) + ';'); + texture = "texelFetch(" + sampler + ", coords, lod)"; break; } default: { @@ -2776,7 +2928,10 @@ private: static_cast<u32>(instr.tlds.GetTextureProcessMode())); } } - WriteTexsInstruction(instr, coord, texture); + WriteTexsInstruction(instr, coords, texture); + + --shader.scope; + shader.AddLine('}'); break; } case OpCode::Id::TLD4: { @@ -2799,18 +2954,23 @@ private: if (depth_compare) num_coordinates += 1; + // Add an extra scope and declare the texture coords inside to prevent + // overwriting them in case they are used as outputs of the texs instruction. + shader.AddLine('{'); + ++shader.scope; + switch (num_coordinates) { case 2: { const std::string x = regs.GetRegisterAsFloat(instr.gpr8); const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - coord = "vec2 coords = vec2(" + x + ", " + y + ");"; + shader.AddLine("vec2 coords = vec2(" + x + ", " + y + ");"); break; } case 3: { const std::string x = regs.GetRegisterAsFloat(instr.gpr8); const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); const std::string z = regs.GetRegisterAsFloat(instr.gpr8.Value() + 2); - coord = "vec3 coords = vec3(" + x + ", " + y + ", " + z + ");"; + shader.AddLine("vec3 coords = vec3(" + x + ", " + y + ", " + z + ");"); break; } default: @@ -2818,34 +2978,33 @@ private: static_cast<u32>(num_coordinates)); const std::string x = regs.GetRegisterAsFloat(instr.gpr8); const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - coord = "vec2 coords = vec2(" + x + ", " + y + ");"; + shader.AddLine("vec2 coords = vec2(" + x + ", " + y + ");"); texture_type = Tegra::Shader::TextureType::Texture2D; } const std::string sampler = GetSampler(instr.sampler, texture_type, false, depth_compare); - // Add an extra scope and declare the texture coords inside to prevent - // overwriting them in case they are used as outputs of the texs instruction. - shader.AddLine("{"); - ++shader.scope; - shader.AddLine(coord); + const std::string texture = "textureGather(" + sampler + ", coords, " + std::to_string(instr.tld4.component) + ')'; + if (!depth_compare) { + shader.AddLine("vec4 texture_tmp = " + texture + ';'); std::size_t dest_elem{}; for (std::size_t elem = 0; elem < 4; ++elem) { if (!instr.tex.IsComponentEnabled(elem)) { // Skip disabled components continue; } - regs.SetRegisterToFloat(instr.gpr0, elem, texture, 1, 4, false, dest_elem); + regs.SetRegisterToFloat(instr.gpr0, elem, "texture_tmp", 1, 4, false, + dest_elem); ++dest_elem; } } else { regs.SetRegisterToFloat(instr.gpr0, 0, texture, 1, 1, false); } --shader.scope; - shader.AddLine("}"); + shader.AddLine('}'); break; } case OpCode::Id::TLD4S: { @@ -2856,6 +3015,11 @@ private: instr.tld4s.UsesMiscMode(Tegra::Shader::TextureMiscMode::AOFFI), "AOFFI is not implemented"); + // Scope to avoid variable name overlaps. + shader.AddLine('{'); + ++shader.scope; + std::string coords; + const bool depth_compare = instr.tld4s.UsesMiscMode(Tegra::Shader::TextureMiscMode::DC); const std::string op_a = regs.GetRegisterAsFloat(instr.gpr8); @@ -2863,28 +3027,32 @@ private: // TODO(Subv): Figure out how the sampler type is encoded in the TLD4S instruction. const std::string sampler = GetSampler( instr.sampler, Tegra::Shader::TextureType::Texture2D, false, depth_compare); - std::string coord; if (!depth_compare) { - coord = "vec2 coords = vec2(" + op_a + ", " + op_b + ");"; + coords = "vec2 coords = vec2(" + op_a + ", " + op_b + ");"; } else { // Note: TLD4S coordinate encoding works just like TEXS's - const std::string op_c = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - coord = "vec3 coords = vec3(" + op_a + ", " + op_c + ", " + op_b + ");"; + const std::string op_y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); + coords = "vec3 coords = vec3(" + op_a + ", " + op_y + ", " + op_b + ");"; } const std::string texture = "textureGather(" + sampler + ", coords, " + std::to_string(instr.tld4s.component) + ')'; if (!depth_compare) { - WriteTexsInstruction(instr, coord, texture); + WriteTexsInstruction(instr, coords, texture); } else { - WriteTexsInstruction(instr, coord, "vec4(" + texture + ')'); + WriteTexsInstruction(instr, coords, "vec4(" + texture + ')'); } + + --shader.scope; + shader.AddLine('}'); break; } case OpCode::Id::TXQ: { UNIMPLEMENTED_IF_MSG(instr.txq.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP), "NODEP is not implemented"); + ++shader.scope; + shader.AddLine('{'); // TODO: the new commits on the texture refactor, change the way samplers work. // Sadly, not all texture instructions specify the type of texture their sampler // uses. This must be fixed at a later instance. @@ -2892,8 +3060,14 @@ private: GetSampler(instr.sampler, Tegra::Shader::TextureType::Texture2D, false, false); switch (instr.txq.query_type) { case Tegra::Shader::TextureQueryType::Dimension: { - const std::string texture = "textureQueryLevels(" + sampler + ')'; - regs.SetRegisterToInteger(instr.gpr0, true, 0, texture, 1, 1); + const std::string texture = "textureSize(" + sampler + ", " + + regs.GetRegisterAsInteger(instr.gpr8) + ')'; + const std::string mip_level = "textureQueryLevels(" + sampler + ')'; + shader.AddLine("ivec2 sizes = " + texture + ';'); + regs.SetRegisterToInteger(instr.gpr0, true, 0, "sizes.x", 1, 1); + regs.SetRegisterToInteger(instr.gpr0.Value() + 1, true, 0, "sizes.y", 1, 1); + regs.SetRegisterToInteger(instr.gpr0.Value() + 2, true, 0, "0", 1, 1); + regs.SetRegisterToInteger(instr.gpr0.Value() + 3, true, 0, mip_level, 1, 1); break; } default: { @@ -2901,6 +3075,8 @@ private: static_cast<u32>(instr.txq.query_type.Value())); } } + --shader.scope; + shader.AddLine('}'); break; } case OpCode::Id::TMML: { @@ -3083,7 +3259,8 @@ private: break; } case OpCode::Type::PredicateSetRegister: { - UNIMPLEMENTED_IF_MSG(instr.generates_cc, "PSET Generates an unhandled Control Code"); + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "Condition codes generation in PSET is not implemented"); const std::string op_a = GetPredicateCondition(instr.pset.pred12, instr.pset.neg_pred12 != 0); @@ -3142,14 +3319,14 @@ private: const std::string pred = GetPredicateCondition(instr.csetp.pred39, instr.csetp.neg_pred39 != 0); const std::string combiner = GetPredicateCombiner(instr.csetp.op); - const std::string control_code = regs.GetControlCode(instr.csetp.cc); + const std::string condition_code = regs.GetConditionCode(instr.csetp.cc); if (instr.csetp.pred3 != static_cast<u64>(Pred::UnusedIndex)) { SetPredicate(instr.csetp.pred3, - '(' + control_code + ") " + combiner + " (" + pred + ')'); + '(' + condition_code + ") " + combiner + " (" + pred + ')'); } if (instr.csetp.pred0 != static_cast<u64>(Pred::UnusedIndex)) { SetPredicate(instr.csetp.pred0, - "!(" + control_code + ") " + combiner + " (" + pred + ')'); + "!(" + condition_code + ") " + combiner + " (" + pred + ')'); } break; } @@ -3159,6 +3336,34 @@ private: } break; } + case OpCode::Type::RegisterSetPredicate: { + UNIMPLEMENTED_IF(instr.r2p.mode != Tegra::Shader::R2pMode::Pr); + + const std::string apply_mask = [&]() { + switch (opcode->get().GetId()) { + case OpCode::Id::R2P_IMM: + return std::to_string(instr.r2p.immediate_mask); + default: + UNREACHABLE(); + } + }(); + const std::string mask = '(' + regs.GetRegisterAsInteger(instr.gpr8, 0, false) + + " >> " + std::to_string(instr.r2p.byte) + ')'; + + constexpr u64 programmable_preds = 7; + for (u64 pred = 0; pred < programmable_preds; ++pred) { + const auto shift = std::to_string(1 << pred); + + shader.AddLine("if ((" + apply_mask + " & " + shift + ") != 0) {"); + ++shader.scope; + + SetPredicate(pred, '(' + mask + " & " + shift + ") != 0"); + + --shader.scope; + shader.AddLine('}'); + } + break; + } case OpCode::Type::FloatSet: { const std::string op_a = GetOperandAbsNeg(regs.GetRegisterAsFloat(instr.gpr8), instr.fset.abs_a != 0, instr.fset.neg_a != 0); @@ -3196,6 +3401,10 @@ private: regs.SetRegisterToInteger(instr.gpr0, false, 0, predicate + " ? 0xFFFFFFFF : 0", 1, 1); } + if (instr.generates_cc.Value() != 0) { + regs.SetInternalFlag(InternalFlag::ZeroFlag, predicate); + LOG_WARNING(HW_GPU, "FSET Condition Code is incomplete"); + } break; } case OpCode::Type::IntegerSet: { @@ -3280,7 +3489,8 @@ private: case OpCode::Type::Xmad: { UNIMPLEMENTED_IF(instr.xmad.sign_a); UNIMPLEMENTED_IF(instr.xmad.sign_b); - UNIMPLEMENTED_IF_MSG(instr.generates_cc, "XMAD Generates an unhandled Control Code"); + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "Condition codes generation in XMAD is not implemented"); std::string op_a{regs.GetRegisterAsInteger(instr.gpr8, 0, instr.xmad.sign_a)}; std::string op_b; @@ -3372,9 +3582,9 @@ private: default: { switch (opcode->get().GetId()) { case OpCode::Id::EXIT: { - const Tegra::Shader::ControlCode cc = instr.flow_control_code; - UNIMPLEMENTED_IF_MSG(cc != Tegra::Shader::ControlCode::T, - "EXIT Control Code used: {}", static_cast<u32>(cc)); + const Tegra::Shader::ConditionCode cc = instr.flow_condition_code; + UNIMPLEMENTED_IF_MSG(cc != Tegra::Shader::ConditionCode::T, + "EXIT condition code used: {}", static_cast<u32>(cc)); if (stage == Maxwell3D::Regs::ShaderStage::Fragment) { EmitFragmentOutputsWrite(); @@ -3406,9 +3616,9 @@ private: case OpCode::Id::KIL: { UNIMPLEMENTED_IF(instr.flow.cond != Tegra::Shader::FlowCondition::Always); - const Tegra::Shader::ControlCode cc = instr.flow_control_code; - UNIMPLEMENTED_IF_MSG(cc != Tegra::Shader::ControlCode::T, - "KIL Control Code used: {}", static_cast<u32>(cc)); + const Tegra::Shader::ConditionCode cc = instr.flow_condition_code; + UNIMPLEMENTED_IF_MSG(cc != Tegra::Shader::ConditionCode::T, + "KIL condition code used: {}", static_cast<u32>(cc)); // Enclose "discard" in a conditional, so that GLSL compilation does not complain // about unexecuted instructions that may follow this. @@ -3448,6 +3658,11 @@ private: regs.SetRegisterToInteger(instr.gpr0, false, 0, "0u", 1, 1); break; } + case Tegra::Shader::SystemVariable::Ydirection: { + // Config pack's third value is Y_NEGATE's state. + regs.SetRegisterToFloat(instr.gpr0, 0, "uintBitsToFloat(config_pack[2])", 1, 1); + break; + } default: { UNIMPLEMENTED_MSG("Unhandled system move: {}", static_cast<u32>(instr.sys20.Value())); @@ -3470,12 +3685,18 @@ private: UNIMPLEMENTED_IF_MSG(instr.bra.constant_buffer != 0, "BRA with constant buffers are not implemented"); - const Tegra::Shader::ControlCode cc = instr.flow_control_code; - UNIMPLEMENTED_IF_MSG(cc != Tegra::Shader::ControlCode::T, - "BRA Control Code used: {}", static_cast<u32>(cc)); - + const Tegra::Shader::ConditionCode cc = instr.flow_condition_code; const u32 target = offset + instr.bra.GetBranchTarget(); - shader.AddLine("{ jmp_to = " + std::to_string(target) + "u; break; }"); + if (cc != Tegra::Shader::ConditionCode::T) { + const std::string condition_code = regs.GetConditionCode(cc); + shader.AddLine("if (" + condition_code + "){"); + shader.scope++; + shader.AddLine("{ jmp_to = " + std::to_string(target) + "u; break; }"); + shader.scope--; + shader.AddLine('}'); + } else { + shader.AddLine("{ jmp_to = " + std::to_string(target) + "u; break; }"); + } break; } case OpCode::Id::IPA: { @@ -3515,9 +3736,9 @@ private: break; } case OpCode::Id::SYNC: { - const Tegra::Shader::ControlCode cc = instr.flow_control_code; - UNIMPLEMENTED_IF_MSG(cc != Tegra::Shader::ControlCode::T, - "SYNC Control Code used: {}", static_cast<u32>(cc)); + const Tegra::Shader::ConditionCode cc = instr.flow_condition_code; + UNIMPLEMENTED_IF_MSG(cc != Tegra::Shader::ConditionCode::T, + "SYNC condition code used: {}", static_cast<u32>(cc)); // The SYNC opcode jumps to the address previously set by the SSY opcode EmitPopFromFlowStack(); @@ -3525,10 +3746,10 @@ private: } case OpCode::Id::BRK: { // The BRK opcode jumps to the address previously set by the PBK opcode - const Tegra::Shader::ControlCode cc = instr.flow_control_code; - if (cc != Tegra::Shader::ControlCode::T) { - UNIMPLEMENTED_MSG("BRK Control Code used: {}", static_cast<u32>(cc)); - } + const Tegra::Shader::ConditionCode cc = instr.flow_condition_code; + UNIMPLEMENTED_IF_MSG(cc != Tegra::Shader::ConditionCode::T, + "BRK condition code used: {}", static_cast<u32>(cc)); + EmitPopFromFlowStack(); break; } @@ -3539,6 +3760,9 @@ private: break; } case OpCode::Id::VMAD: { + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "Condition codes generation in VMAD is not implemented"); + const bool result_signed = instr.video.signed_a == 1 || instr.video.signed_b == 1; const std::string op_a = GetVideoOperandA(instr); const std::string op_b = GetVideoOperandB(instr); @@ -3558,10 +3782,6 @@ private: regs.SetRegisterToInteger(instr.gpr0, result_signed, 1, result, 1, 1, instr.vmad.saturate == 1, 0, Register::Size::Word, instr.vmad.cc); - if (instr.generates_cc) { - UNIMPLEMENTED_MSG("VMAD Generates an unhandled Control Code"); - } - break; } case OpCode::Id::VSETP: { @@ -3713,6 +3933,7 @@ private: Maxwell3D::Regs::ShaderStage stage; const std::string& suffix; u64 local_memory_size; + std::size_t shader_length; ShaderWriter shader; ShaderWriter declarations; @@ -3731,9 +3952,10 @@ std::optional<ProgramResult> DecompileProgram(const ProgramCode& program_code, u Maxwell3D::Regs::ShaderStage stage, const std::string& suffix) { try { - const auto subroutines = - ControlFlowAnalyzer(program_code, main_offset, suffix).GetSubroutines(); - GLSLGenerator generator(subroutines, program_code, main_offset, stage, suffix); + ControlFlowAnalyzer analyzer(program_code, main_offset, suffix); + const auto subroutines = analyzer.GetSubroutines(); + GLSLGenerator generator(subroutines, program_code, main_offset, stage, suffix, + analyzer.GetShaderLength()); return ProgramResult{generator.GetShaderCode(), generator.GetEntries()}; } catch (const DecompileFail& exception) { LOG_ERROR(HW_GPU, "Shader decompilation failed: {}", exception.what()); diff --git a/src/video_core/renderer_opengl/gl_shader_gen.cpp b/src/video_core/renderer_opengl/gl_shader_gen.cpp index eea090e52..23ed91e27 100644 --- a/src/video_core/renderer_opengl/gl_shader_gen.cpp +++ b/src/video_core/renderer_opengl/gl_shader_gen.cpp @@ -24,8 +24,7 @@ layout (location = 0) out vec4 position; layout(std140) uniform vs_config { vec4 viewport_flip; - uvec4 instance_id; - uvec4 flip_stage; + uvec4 config_pack; // instance_id, flip_stage, y_direction, padding uvec4 alpha_test; }; )"; @@ -63,7 +62,8 @@ void main() { out += R"( // Check if the flip stage is VertexB - if (flip_stage[0] == 1) { + // Config pack's second value is flip_stage + if (config_pack[1] == 1) { // Viewport can be flipped, which is unsupported by glViewport position.xy *= viewport_flip.xy; } @@ -71,7 +71,7 @@ void main() { // TODO(bunnei): This is likely a hack, position.w should be interpolated as 1.0 // For now, this is here to bring order in lieu of proper emulation - if (flip_stage[0] == 1) { + if (config_pack[1] == 1) { position.w = 1.0; } } @@ -101,8 +101,7 @@ layout (location = 0) out vec4 position; layout (std140) uniform gs_config { vec4 viewport_flip; - uvec4 instance_id; - uvec4 flip_stage; + uvec4 config_pack; // instance_id, flip_stage, y_direction, padding uvec4 alpha_test; }; @@ -139,8 +138,7 @@ layout (location = 0) in vec4 position; layout (std140) uniform fs_config { vec4 viewport_flip; - uvec4 instance_id; - uvec4 flip_stage; + uvec4 config_pack; // instance_id, flip_stage, y_direction, padding uvec4 alpha_test; }; diff --git a/src/video_core/renderer_opengl/gl_shader_gen.h b/src/video_core/renderer_opengl/gl_shader_gen.h index 520b9d4e3..b425d98ae 100644 --- a/src/video_core/renderer_opengl/gl_shader_gen.h +++ b/src/video_core/renderer_opengl/gl_shader_gen.h @@ -163,6 +163,7 @@ private: struct ShaderEntries { std::vector<ConstBufferEntry> const_buffer_entries; std::vector<SamplerEntry> texture_samplers; + std::size_t shader_length; }; using ProgramResult = std::pair<std::string, ShaderEntries>; diff --git a/src/video_core/renderer_opengl/gl_shader_manager.cpp b/src/video_core/renderer_opengl/gl_shader_manager.cpp index 8b8869ecb..6a30c28d2 100644 --- a/src/video_core/renderer_opengl/gl_shader_manager.cpp +++ b/src/video_core/renderer_opengl/gl_shader_manager.cpp @@ -27,16 +27,18 @@ void MaxwellUniformData::SetFromRegs(const Maxwell3D::State::ShaderStageInfo& sh alpha_test.func = func; alpha_test.ref = regs.alpha_test_ref; - // We only assign the instance to the first component of the vector, the rest is just padding. - instance_id[0] = state.current_instance; + instance_id = state.current_instance; // Assign in which stage the position has to be flipped // (the last stage before the fragment shader). if (gpu.regs.shader_config[static_cast<u32>(Maxwell3D::Regs::ShaderProgram::Geometry)].enable) { - flip_stage[0] = static_cast<u32>(Maxwell3D::Regs::ShaderProgram::Geometry); + flip_stage = static_cast<u32>(Maxwell3D::Regs::ShaderProgram::Geometry); } else { - flip_stage[0] = static_cast<u32>(Maxwell3D::Regs::ShaderProgram::VertexB); + flip_stage = static_cast<u32>(Maxwell3D::Regs::ShaderProgram::VertexB); } + + // Y_NEGATE controls what value S2R returns for the Y_DIRECTION system value. + y_direction = regs.screen_y_control.y_negate == 0 ? 1.f : -1.f; } } // namespace OpenGL::GLShader diff --git a/src/video_core/renderer_opengl/gl_shader_manager.h b/src/video_core/renderer_opengl/gl_shader_manager.h index 9a5d7e289..b757f5f44 100644 --- a/src/video_core/renderer_opengl/gl_shader_manager.h +++ b/src/video_core/renderer_opengl/gl_shader_manager.h @@ -21,8 +21,11 @@ using Tegra::Engines::Maxwell3D; struct MaxwellUniformData { void SetFromRegs(const Maxwell3D::State::ShaderStageInfo& shader_stage); alignas(16) GLvec4 viewport_flip; - alignas(16) GLuvec4 instance_id; - alignas(16) GLuvec4 flip_stage; + struct alignas(16) { + GLuint instance_id; + GLuint flip_stage; + GLfloat y_direction; + }; struct alignas(16) { GLuint enabled; GLuint func; @@ -30,7 +33,7 @@ struct MaxwellUniformData { GLuint padding; } alpha_test; }; -static_assert(sizeof(MaxwellUniformData) == 64, "MaxwellUniformData structure size is incorrect"); +static_assert(sizeof(MaxwellUniformData) == 48, "MaxwellUniformData structure size is incorrect"); static_assert(sizeof(MaxwellUniformData) < 16384, "MaxwellUniformData structure must be less than 16kb as per the OpenGL spec"); diff --git a/src/video_core/renderer_opengl/gl_state.cpp b/src/video_core/renderer_opengl/gl_state.cpp index d9910c6e8..dc0a5ed5e 100644 --- a/src/video_core/renderer_opengl/gl_state.cpp +++ b/src/video_core/renderer_opengl/gl_state.cpp @@ -92,6 +92,14 @@ OpenGLState::OpenGLState() { point.size = 1; fragment_color_clamp.enabled = false; + depth_clamp.far_plane = false; + depth_clamp.near_plane = false; + polygon_offset.fill_enable = false; + polygon_offset.line_enable = false; + polygon_offset.point_enable = false; + polygon_offset.factor = 0.0f; + polygon_offset.units = 0.0f; + polygon_offset.clamp = 0.0f; } void OpenGLState::ApplyDefaultState() { @@ -140,7 +148,7 @@ void OpenGLState::ApplyCulling() const { } void OpenGLState::ApplyColorMask() const { - if (GLAD_GL_ARB_viewport_array && independant_blend.enabled) { + if (independant_blend.enabled) { for (size_t i = 0; i < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets; i++) { const auto& updated = color_mask[i]; const auto& current = cur_state.color_mask[i]; @@ -233,16 +241,40 @@ void OpenGLState::ApplyStencilTest() const { config_stencil(GL_BACK, stencil.back, cur_state.stencil.back); } } +// Viewport does not affects glClearBuffer so emulate viewport using scissor test +void OpenGLState::EmulateViewportWithScissor() { + auto& current = viewports[0]; + if (current.scissor.enabled) { + const GLint left = std::max(current.x, current.scissor.x); + const GLint right = + std::max(current.x + current.width, current.scissor.x + current.scissor.width); + const GLint bottom = std::max(current.y, current.scissor.y); + const GLint top = + std::max(current.y + current.height, current.scissor.y + current.scissor.height); + current.scissor.x = std::max(left, 0); + current.scissor.y = std::max(bottom, 0); + current.scissor.width = std::max(right - left, 0); + current.scissor.height = std::max(top - bottom, 0); + } else { + current.scissor.enabled = true; + current.scissor.x = current.x; + current.scissor.y = current.y; + current.scissor.width = current.width; + current.scissor.height = current.height; + } +} void OpenGLState::ApplyViewport() const { - if (GLAD_GL_ARB_viewport_array && geometry_shaders.enabled) { + if (geometry_shaders.enabled) { for (GLuint i = 0; i < static_cast<GLuint>(Tegra::Engines::Maxwell3D::Regs::NumViewports); i++) { const auto& current = cur_state.viewports[i]; const auto& updated = viewports[i]; if (updated.x != current.x || updated.y != current.y || updated.width != current.width || updated.height != current.height) { - glViewportIndexedf(i, updated.x, updated.y, updated.width, updated.height); + glViewportIndexedf( + i, static_cast<GLfloat>(updated.x), static_cast<GLfloat>(updated.y), + static_cast<GLfloat>(updated.width), static_cast<GLfloat>(updated.height)); } if (updated.depth_range_near != current.depth_range_near || updated.depth_range_far != current.depth_range_far) { @@ -270,8 +302,7 @@ void OpenGLState::ApplyViewport() const { const auto& updated = viewports[0]; if (updated.x != current.x || updated.y != current.y || updated.width != current.width || updated.height != current.height) { - glViewport(static_cast<GLint>(updated.x), static_cast<GLint>(updated.y), - static_cast<GLsizei>(updated.width), static_cast<GLsizei>(updated.height)); + glViewport(updated.x, updated.y, updated.width, updated.height); } if (updated.depth_range_near != current.depth_range_near || updated.depth_range_far != current.depth_range_far) { @@ -339,14 +370,14 @@ void OpenGLState::ApplyTargetBlending(std::size_t target, bool force) const { if (blend_changed || updated.src_rgb_func != current.src_rgb_func || updated.dst_rgb_func != current.dst_rgb_func || updated.src_a_func != current.src_a_func || updated.dst_a_func != current.dst_a_func) { - glBlendFuncSeparateiARB(static_cast<GLuint>(target), updated.src_rgb_func, - updated.dst_rgb_func, updated.src_a_func, updated.dst_a_func); + glBlendFuncSeparatei(static_cast<GLuint>(target), updated.src_rgb_func, + updated.dst_rgb_func, updated.src_a_func, updated.dst_a_func); } if (blend_changed || updated.rgb_equation != current.rgb_equation || updated.a_equation != current.a_equation) { - glBlendEquationSeparateiARB(static_cast<GLuint>(target), updated.rgb_equation, - updated.a_equation); + glBlendEquationSeparatei(static_cast<GLuint>(target), updated.rgb_equation, + updated.a_equation); } } @@ -383,6 +414,55 @@ void OpenGLState::ApplyLogicOp() const { } } +void OpenGLState::ApplyPolygonOffset() const { + + const bool fill_enable_changed = + polygon_offset.fill_enable != cur_state.polygon_offset.fill_enable; + const bool line_enable_changed = + polygon_offset.line_enable != cur_state.polygon_offset.line_enable; + const bool point_enable_changed = + polygon_offset.point_enable != cur_state.polygon_offset.point_enable; + const bool factor_changed = polygon_offset.factor != cur_state.polygon_offset.factor; + const bool units_changed = polygon_offset.units != cur_state.polygon_offset.units; + const bool clamp_changed = polygon_offset.clamp != cur_state.polygon_offset.clamp; + + if (fill_enable_changed) { + if (polygon_offset.fill_enable) { + glEnable(GL_POLYGON_OFFSET_FILL); + } else { + glDisable(GL_POLYGON_OFFSET_FILL); + } + } + + if (line_enable_changed) { + if (polygon_offset.line_enable) { + glEnable(GL_POLYGON_OFFSET_LINE); + } else { + glDisable(GL_POLYGON_OFFSET_LINE); + } + } + + if (point_enable_changed) { + if (polygon_offset.point_enable) { + glEnable(GL_POLYGON_OFFSET_POINT); + } else { + glDisable(GL_POLYGON_OFFSET_POINT); + } + } + + if ((polygon_offset.fill_enable || polygon_offset.line_enable || polygon_offset.point_enable) && + (factor_changed || units_changed || clamp_changed)) { + + if (GLAD_GL_EXT_polygon_offset_clamp && polygon_offset.clamp != 0) { + glPolygonOffsetClamp(polygon_offset.factor, polygon_offset.units, polygon_offset.clamp); + } else { + glPolygonOffset(polygon_offset.factor, polygon_offset.units); + UNIMPLEMENTED_IF_MSG(polygon_offset.clamp != 0, + "Unimplemented Depth polygon offset clamp."); + } + } +} + void OpenGLState::ApplyTextures() const { for (std::size_t i = 0; i < std::size(texture_units); ++i) { const auto& texture_unit = texture_units[i]; @@ -446,6 +526,21 @@ void OpenGLState::ApplyVertexBufferState() const { } } +void OpenGLState::ApplyDepthClamp() const { + if (depth_clamp.far_plane == cur_state.depth_clamp.far_plane && + depth_clamp.near_plane == cur_state.depth_clamp.near_plane) { + return; + } + if (depth_clamp.far_plane != depth_clamp.near_plane) { + UNIMPLEMENTED_MSG("Unimplemented Depth Clamp Separation!"); + } + if (depth_clamp.far_plane || depth_clamp.near_plane) { + glEnable(GL_DEPTH_CLAMP); + } else { + glDisable(GL_DEPTH_CLAMP); + } +} + void OpenGLState::Apply() const { ApplyFramebufferState(); ApplyVertexBufferState(); @@ -477,11 +572,9 @@ void OpenGLState::Apply() const { if (point.size != cur_state.point.size) { glPointSize(point.size); } - if (GLAD_GL_ARB_color_buffer_float) { - if (fragment_color_clamp.enabled != cur_state.fragment_color_clamp.enabled) { - glClampColor(GL_CLAMP_FRAGMENT_COLOR_ARB, - fragment_color_clamp.enabled ? GL_TRUE : GL_FALSE); - } + if (fragment_color_clamp.enabled != cur_state.fragment_color_clamp.enabled) { + glClampColor(GL_CLAMP_FRAGMENT_COLOR_ARB, + fragment_color_clamp.enabled ? GL_TRUE : GL_FALSE); } if (multisample_control.alpha_to_coverage != cur_state.multisample_control.alpha_to_coverage) { if (multisample_control.alpha_to_coverage) { @@ -497,7 +590,7 @@ void OpenGLState::Apply() const { glDisable(GL_SAMPLE_ALPHA_TO_ONE); } } - + ApplyDepthClamp(); ApplyColorMask(); ApplyViewport(); ApplyStencilTest(); @@ -509,6 +602,7 @@ void OpenGLState::Apply() const { ApplyLogicOp(); ApplyTextures(); ApplySamplers(); + ApplyPolygonOffset(); cur_state = *this; } diff --git a/src/video_core/renderer_opengl/gl_state.h b/src/video_core/renderer_opengl/gl_state.h index bdc743b0f..439bfbc98 100644 --- a/src/video_core/renderer_opengl/gl_state.h +++ b/src/video_core/renderer_opengl/gl_state.h @@ -49,6 +49,11 @@ public: } fragment_color_clamp; struct { + bool far_plane; + bool near_plane; + } depth_clamp; // GL_DEPTH_CLAMP + + struct { bool enabled; // viewports arrays are only supported when geometry shaders are enabled. } geometry_shaders; @@ -156,10 +161,10 @@ public: } draw; struct viewport { - GLfloat x; - GLfloat y; - GLfloat width; - GLfloat height; + GLint x; + GLint y; + GLint width; + GLint height; GLfloat depth_range_near; // GL_DEPTH_RANGE GLfloat depth_range_far; // GL_DEPTH_RANGE struct { @@ -176,7 +181,16 @@ public: float size; // GL_POINT_SIZE } point; - std::array<bool, 2> clip_distance; // GL_CLIP_DISTANCE + struct { + bool point_enable; + bool line_enable; + bool fill_enable; + GLfloat units; + GLfloat factor; + GLfloat clamp; + } polygon_offset; + + std::array<bool, 8> clip_distance; // GL_CLIP_DISTANCE OpenGLState(); @@ -206,6 +220,7 @@ public: OpenGLState& ResetBuffer(GLuint handle); OpenGLState& ResetVertexArray(GLuint handle); OpenGLState& ResetFramebuffer(GLuint handle); + void EmulateViewportWithScissor(); private: static OpenGLState cur_state; @@ -225,6 +240,8 @@ private: void ApplyLogicOp() const; void ApplyTextures() const; void ApplySamplers() const; + void ApplyDepthClamp() const; + void ApplyPolygonOffset() const; }; } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/maxwell_to_gl.h b/src/video_core/renderer_opengl/maxwell_to_gl.h index 065b3929c..a8833c06e 100644 --- a/src/video_core/renderer_opengl/maxwell_to_gl.h +++ b/src/video_core/renderer_opengl/maxwell_to_gl.h @@ -218,14 +218,19 @@ inline GLenum DepthCompareFunc(Tegra::Texture::DepthCompareFunc func) { inline GLenum BlendEquation(Maxwell::Blend::Equation equation) { switch (equation) { case Maxwell::Blend::Equation::Add: + case Maxwell::Blend::Equation::AddGL: return GL_FUNC_ADD; case Maxwell::Blend::Equation::Subtract: + case Maxwell::Blend::Equation::SubtractGL: return GL_FUNC_SUBTRACT; case Maxwell::Blend::Equation::ReverseSubtract: + case Maxwell::Blend::Equation::ReverseSubtractGL: return GL_FUNC_REVERSE_SUBTRACT; case Maxwell::Blend::Equation::Min: + case Maxwell::Blend::Equation::MinGL: return GL_MIN; case Maxwell::Blend::Equation::Max: + case Maxwell::Blend::Equation::MaxGL: return GL_MAX; } LOG_ERROR(Render_OpenGL, "Unimplemented blend equation={}", static_cast<u32>(equation)); diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp index 27b5b8960..4fd0d66c5 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.cpp +++ b/src/video_core/renderer_opengl/renderer_opengl.cpp @@ -19,9 +19,9 @@ #include "core/settings.h" #include "core/telemetry_session.h" #include "core/tracer/recorder.h" +#include "video_core/morton.h" #include "video_core/renderer_opengl/gl_rasterizer.h" #include "video_core/renderer_opengl/renderer_opengl.h" -#include "video_core/utils.h" namespace OpenGL { @@ -490,7 +490,7 @@ bool RendererOpenGL::Init() { Core::Telemetry().AddField(Telemetry::FieldType::UserSystem, "GPU_Model", gpu_model); Core::Telemetry().AddField(Telemetry::FieldType::UserSystem, "GPU_OpenGL_Version", gl_version); - if (!GLAD_GL_VERSION_3_3) { + if (!GLAD_GL_VERSION_4_3) { return false; } diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp index 7eabd34f1..bbae9285f 100644 --- a/src/video_core/textures/decoders.cpp +++ b/src/video_core/textures/decoders.cpp @@ -127,7 +127,8 @@ void FastProcessBlock(u8* const swizzled_data, u8* const unswizzled_data, const template <bool fast> void SwizzledData(u8* const swizzled_data, u8* const unswizzled_data, const bool unswizzle, const u32 width, const u32 height, const u32 depth, const u32 bytes_per_pixel, - const u32 out_bytes_per_pixel, const u32 block_height, const u32 block_depth) { + const u32 out_bytes_per_pixel, const u32 block_height, const u32 block_depth, + const u32 width_spacing) { auto div_ceil = [](const u32 x, const u32 y) { return ((x + y - 1) / y); }; const u32 stride_x = width * out_bytes_per_pixel; const u32 layer_z = height * stride_x; @@ -137,7 +138,8 @@ void SwizzledData(u8* const swizzled_data, u8* const unswizzled_data, const bool const u32 block_x_elements = gob_elements_x; const u32 block_y_elements = gob_elements_y * block_height; const u32 block_z_elements = gob_elements_z * block_depth; - const u32 blocks_on_x = div_ceil(width, block_x_elements); + const u32 aligned_width = Common::AlignUp(width, gob_elements_x * width_spacing); + const u32 blocks_on_x = div_ceil(aligned_width, block_x_elements); const u32 blocks_on_y = div_ceil(height, block_y_elements); const u32 blocks_on_z = div_ceil(depth, block_z_elements); const u32 xy_block_size = gob_size * block_height; @@ -169,13 +171,15 @@ void SwizzledData(u8* const swizzled_data, u8* const unswizzled_data, const bool void CopySwizzledData(u32 width, u32 height, u32 depth, u32 bytes_per_pixel, u32 out_bytes_per_pixel, u8* const swizzled_data, u8* const unswizzled_data, - bool unswizzle, u32 block_height, u32 block_depth) { + bool unswizzle, u32 block_height, u32 block_depth, u32 width_spacing) { if (bytes_per_pixel % 3 != 0 && (width * bytes_per_pixel) % fast_swizzle_align == 0) { SwizzledData<true>(swizzled_data, unswizzled_data, unswizzle, width, height, depth, - bytes_per_pixel, out_bytes_per_pixel, block_height, block_depth); + bytes_per_pixel, out_bytes_per_pixel, block_height, block_depth, + width_spacing); } else { SwizzledData<false>(swizzled_data, unswizzled_data, unswizzle, width, height, depth, - bytes_per_pixel, out_bytes_per_pixel, block_height, block_depth); + bytes_per_pixel, out_bytes_per_pixel, block_height, block_depth, + width_spacing); } } @@ -228,19 +232,19 @@ u32 BytesPerPixel(TextureFormat format) { void UnswizzleTexture(u8* const unswizzled_data, VAddr address, u32 tile_size_x, u32 tile_size_y, u32 bytes_per_pixel, u32 width, u32 height, u32 depth, u32 block_height, - u32 block_depth) { + u32 block_depth, u32 width_spacing) { CopySwizzledData((width + tile_size_x - 1) / tile_size_x, (height + tile_size_y - 1) / tile_size_y, depth, bytes_per_pixel, bytes_per_pixel, Memory::GetPointer(address), unswizzled_data, true, - block_height, block_depth); + block_height, block_depth, width_spacing); } std::vector<u8> UnswizzleTexture(VAddr address, u32 tile_size_x, u32 tile_size_y, u32 bytes_per_pixel, u32 width, u32 height, u32 depth, - u32 block_height, u32 block_depth) { + u32 block_height, u32 block_depth, u32 width_spacing) { std::vector<u8> unswizzled_data(width * height * depth * bytes_per_pixel); UnswizzleTexture(unswizzled_data.data(), address, tile_size_x, tile_size_y, bytes_per_pixel, - width, height, depth, block_height, block_depth); + width, height, depth, block_height, block_depth, width_spacing); return unswizzled_data; } diff --git a/src/video_core/textures/decoders.h b/src/video_core/textures/decoders.h index f4ef7c73e..85b7e9f7b 100644 --- a/src/video_core/textures/decoders.h +++ b/src/video_core/textures/decoders.h @@ -22,19 +22,20 @@ inline std::size_t GetGOBSize() { void UnswizzleTexture(u8* unswizzled_data, VAddr address, u32 tile_size_x, u32 tile_size_y, u32 bytes_per_pixel, u32 width, u32 height, u32 depth, u32 block_height = TICEntry::DefaultBlockHeight, - u32 block_depth = TICEntry::DefaultBlockHeight); + u32 block_depth = TICEntry::DefaultBlockHeight, u32 width_spacing = 0); /** * Unswizzles a swizzled texture without changing its format. */ std::vector<u8> UnswizzleTexture(VAddr address, u32 tile_size_x, u32 tile_size_y, u32 bytes_per_pixel, u32 width, u32 height, u32 depth, u32 block_height = TICEntry::DefaultBlockHeight, - u32 block_depth = TICEntry::DefaultBlockHeight); + u32 block_depth = TICEntry::DefaultBlockHeight, + u32 width_spacing = 0); /// Copies texture data from a buffer and performs swizzling/unswizzling as necessary. void CopySwizzledData(u32 width, u32 height, u32 depth, u32 bytes_per_pixel, u32 out_bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, - bool unswizzle, u32 block_height, u32 block_depth); + bool unswizzle, u32 block_height, u32 block_depth, u32 width_spacing); /** * Decodes an unswizzled texture into a A8R8G8B8 texture. diff --git a/src/video_core/textures/texture.h b/src/video_core/textures/texture.h index ffa08f5c1..e7c78bee2 100644 --- a/src/video_core/textures/texture.h +++ b/src/video_core/textures/texture.h @@ -166,6 +166,8 @@ struct TICEntry { BitField<3, 3, u32> block_height; BitField<6, 3, u32> block_depth; + BitField<10, 3, u32> tile_width_spacing; + // High 16 bits of the pitch value BitField<0, 16, u32> pitch_high; BitField<26, 1, u32> use_header_opt_control; diff --git a/src/video_core/utils.h b/src/video_core/utils.h deleted file mode 100644 index e0a14d48f..000000000 --- a/src/video_core/utils.h +++ /dev/null @@ -1,164 +0,0 @@ -// Copyright 2014 Citra Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include "common/common_types.h" - -namespace VideoCore { - -// 8x8 Z-Order coordinate from 2D coordinates -static inline u32 MortonInterleave(u32 x, u32 y) { - static const u32 xlut[] = {0x00, 0x01, 0x04, 0x05, 0x10, 0x11, 0x14, 0x15}; - static const u32 ylut[] = {0x00, 0x02, 0x08, 0x0a, 0x20, 0x22, 0x28, 0x2a}; - return xlut[x % 8] + ylut[y % 8]; -} - -/** - * Calculates the offset of the position of the pixel in Morton order - */ -static inline u32 GetMortonOffset(u32 x, u32 y, u32 bytes_per_pixel) { - // Images are split into 8x8 tiles. Each tile is composed of four 4x4 subtiles each - // of which is composed of four 2x2 subtiles each of which is composed of four texels. - // Each structure is embedded into the next-bigger one in a diagonal pattern, e.g. - // texels are laid out in a 2x2 subtile like this: - // 2 3 - // 0 1 - // - // The full 8x8 tile has the texels arranged like this: - // - // 42 43 46 47 58 59 62 63 - // 40 41 44 45 56 57 60 61 - // 34 35 38 39 50 51 54 55 - // 32 33 36 37 48 49 52 53 - // 10 11 14 15 26 27 30 31 - // 08 09 12 13 24 25 28 29 - // 02 03 06 07 18 19 22 23 - // 00 01 04 05 16 17 20 21 - // - // This pattern is what's called Z-order curve, or Morton order. - - const unsigned int block_height = 8; - const unsigned int coarse_x = x & ~7; - - u32 i = VideoCore::MortonInterleave(x, y); - - const unsigned int offset = coarse_x * block_height; - - return (i + offset) * bytes_per_pixel; -} - -static inline u32 MortonInterleave128(u32 x, u32 y) { - // 128x128 Z-Order coordinate from 2D coordinates - static constexpr u32 xlut[] = { - 0x0000, 0x0001, 0x0002, 0x0003, 0x0008, 0x0009, 0x000a, 0x000b, 0x0040, 0x0041, 0x0042, - 0x0043, 0x0048, 0x0049, 0x004a, 0x004b, 0x0800, 0x0801, 0x0802, 0x0803, 0x0808, 0x0809, - 0x080a, 0x080b, 0x0840, 0x0841, 0x0842, 0x0843, 0x0848, 0x0849, 0x084a, 0x084b, 0x1000, - 0x1001, 0x1002, 0x1003, 0x1008, 0x1009, 0x100a, 0x100b, 0x1040, 0x1041, 0x1042, 0x1043, - 0x1048, 0x1049, 0x104a, 0x104b, 0x1800, 0x1801, 0x1802, 0x1803, 0x1808, 0x1809, 0x180a, - 0x180b, 0x1840, 0x1841, 0x1842, 0x1843, 0x1848, 0x1849, 0x184a, 0x184b, 0x2000, 0x2001, - 0x2002, 0x2003, 0x2008, 0x2009, 0x200a, 0x200b, 0x2040, 0x2041, 0x2042, 0x2043, 0x2048, - 0x2049, 0x204a, 0x204b, 0x2800, 0x2801, 0x2802, 0x2803, 0x2808, 0x2809, 0x280a, 0x280b, - 0x2840, 0x2841, 0x2842, 0x2843, 0x2848, 0x2849, 0x284a, 0x284b, 0x3000, 0x3001, 0x3002, - 0x3003, 0x3008, 0x3009, 0x300a, 0x300b, 0x3040, 0x3041, 0x3042, 0x3043, 0x3048, 0x3049, - 0x304a, 0x304b, 0x3800, 0x3801, 0x3802, 0x3803, 0x3808, 0x3809, 0x380a, 0x380b, 0x3840, - 0x3841, 0x3842, 0x3843, 0x3848, 0x3849, 0x384a, 0x384b, 0x0000, 0x0001, 0x0002, 0x0003, - 0x0008, 0x0009, 0x000a, 0x000b, 0x0040, 0x0041, 0x0042, 0x0043, 0x0048, 0x0049, 0x004a, - 0x004b, 0x0800, 0x0801, 0x0802, 0x0803, 0x0808, 0x0809, 0x080a, 0x080b, 0x0840, 0x0841, - 0x0842, 0x0843, 0x0848, 0x0849, 0x084a, 0x084b, 0x1000, 0x1001, 0x1002, 0x1003, 0x1008, - 0x1009, 0x100a, 0x100b, 0x1040, 0x1041, 0x1042, 0x1043, 0x1048, 0x1049, 0x104a, 0x104b, - 0x1800, 0x1801, 0x1802, 0x1803, 0x1808, 0x1809, 0x180a, 0x180b, 0x1840, 0x1841, 0x1842, - 0x1843, 0x1848, 0x1849, 0x184a, 0x184b, 0x2000, 0x2001, 0x2002, 0x2003, 0x2008, 0x2009, - 0x200a, 0x200b, 0x2040, 0x2041, 0x2042, 0x2043, 0x2048, 0x2049, 0x204a, 0x204b, 0x2800, - 0x2801, 0x2802, 0x2803, 0x2808, 0x2809, 0x280a, 0x280b, 0x2840, 0x2841, 0x2842, 0x2843, - 0x2848, 0x2849, 0x284a, 0x284b, 0x3000, 0x3001, 0x3002, 0x3003, 0x3008, 0x3009, 0x300a, - 0x300b, 0x3040, 0x3041, 0x3042, 0x3043, 0x3048, 0x3049, 0x304a, 0x304b, 0x3800, 0x3801, - 0x3802, 0x3803, 0x3808, 0x3809, 0x380a, 0x380b, 0x3840, 0x3841, 0x3842, 0x3843, 0x3848, - 0x3849, 0x384a, 0x384b, 0x0000, 0x0001, 0x0002, 0x0003, 0x0008, 0x0009, 0x000a, 0x000b, - 0x0040, 0x0041, 0x0042, 0x0043, 0x0048, 0x0049, 0x004a, 0x004b, 0x0800, 0x0801, 0x0802, - 0x0803, 0x0808, 0x0809, 0x080a, 0x080b, 0x0840, 0x0841, 0x0842, 0x0843, 0x0848, 0x0849, - 0x084a, 0x084b, 0x1000, 0x1001, 0x1002, 0x1003, 0x1008, 0x1009, 0x100a, 0x100b, 0x1040, - 0x1041, 0x1042, 0x1043, 0x1048, 0x1049, 0x104a, 0x104b, 0x1800, 0x1801, 0x1802, 0x1803, - 0x1808, 0x1809, 0x180a, 0x180b, 0x1840, 0x1841, 0x1842, 0x1843, 0x1848, 0x1849, 0x184a, - 0x184b, 0x2000, 0x2001, 0x2002, 0x2003, 0x2008, 0x2009, 0x200a, 0x200b, 0x2040, 0x2041, - 0x2042, 0x2043, 0x2048, 0x2049, 0x204a, 0x204b, 0x2800, 0x2801, 0x2802, 0x2803, 0x2808, - 0x2809, 0x280a, 0x280b, 0x2840, 0x2841, 0x2842, 0x2843, 0x2848, 0x2849, 0x284a, 0x284b, - 0x3000, 0x3001, 0x3002, 0x3003, 0x3008, 0x3009, 0x300a, 0x300b, 0x3040, 0x3041, 0x3042, - 0x3043, 0x3048, 0x3049, 0x304a, 0x304b, 0x3800, 0x3801, 0x3802, 0x3803, 0x3808, 0x3809, - 0x380a, 0x380b, 0x3840, 0x3841, 0x3842, 0x3843, 0x3848, 0x3849, 0x384a, 0x384b, - }; - static constexpr u32 ylut[] = { - 0x0000, 0x0004, 0x0010, 0x0014, 0x0020, 0x0024, 0x0030, 0x0034, 0x0080, 0x0084, 0x0090, - 0x0094, 0x00a0, 0x00a4, 0x00b0, 0x00b4, 0x0100, 0x0104, 0x0110, 0x0114, 0x0120, 0x0124, - 0x0130, 0x0134, 0x0180, 0x0184, 0x0190, 0x0194, 0x01a0, 0x01a4, 0x01b0, 0x01b4, 0x0200, - 0x0204, 0x0210, 0x0214, 0x0220, 0x0224, 0x0230, 0x0234, 0x0280, 0x0284, 0x0290, 0x0294, - 0x02a0, 0x02a4, 0x02b0, 0x02b4, 0x0300, 0x0304, 0x0310, 0x0314, 0x0320, 0x0324, 0x0330, - 0x0334, 0x0380, 0x0384, 0x0390, 0x0394, 0x03a0, 0x03a4, 0x03b0, 0x03b4, 0x0400, 0x0404, - 0x0410, 0x0414, 0x0420, 0x0424, 0x0430, 0x0434, 0x0480, 0x0484, 0x0490, 0x0494, 0x04a0, - 0x04a4, 0x04b0, 0x04b4, 0x0500, 0x0504, 0x0510, 0x0514, 0x0520, 0x0524, 0x0530, 0x0534, - 0x0580, 0x0584, 0x0590, 0x0594, 0x05a0, 0x05a4, 0x05b0, 0x05b4, 0x0600, 0x0604, 0x0610, - 0x0614, 0x0620, 0x0624, 0x0630, 0x0634, 0x0680, 0x0684, 0x0690, 0x0694, 0x06a0, 0x06a4, - 0x06b0, 0x06b4, 0x0700, 0x0704, 0x0710, 0x0714, 0x0720, 0x0724, 0x0730, 0x0734, 0x0780, - 0x0784, 0x0790, 0x0794, 0x07a0, 0x07a4, 0x07b0, 0x07b4, 0x0000, 0x0004, 0x0010, 0x0014, - 0x0020, 0x0024, 0x0030, 0x0034, 0x0080, 0x0084, 0x0090, 0x0094, 0x00a0, 0x00a4, 0x00b0, - 0x00b4, 0x0100, 0x0104, 0x0110, 0x0114, 0x0120, 0x0124, 0x0130, 0x0134, 0x0180, 0x0184, - 0x0190, 0x0194, 0x01a0, 0x01a4, 0x01b0, 0x01b4, 0x0200, 0x0204, 0x0210, 0x0214, 0x0220, - 0x0224, 0x0230, 0x0234, 0x0280, 0x0284, 0x0290, 0x0294, 0x02a0, 0x02a4, 0x02b0, 0x02b4, - 0x0300, 0x0304, 0x0310, 0x0314, 0x0320, 0x0324, 0x0330, 0x0334, 0x0380, 0x0384, 0x0390, - 0x0394, 0x03a0, 0x03a4, 0x03b0, 0x03b4, 0x0400, 0x0404, 0x0410, 0x0414, 0x0420, 0x0424, - 0x0430, 0x0434, 0x0480, 0x0484, 0x0490, 0x0494, 0x04a0, 0x04a4, 0x04b0, 0x04b4, 0x0500, - 0x0504, 0x0510, 0x0514, 0x0520, 0x0524, 0x0530, 0x0534, 0x0580, 0x0584, 0x0590, 0x0594, - 0x05a0, 0x05a4, 0x05b0, 0x05b4, 0x0600, 0x0604, 0x0610, 0x0614, 0x0620, 0x0624, 0x0630, - 0x0634, 0x0680, 0x0684, 0x0690, 0x0694, 0x06a0, 0x06a4, 0x06b0, 0x06b4, 0x0700, 0x0704, - 0x0710, 0x0714, 0x0720, 0x0724, 0x0730, 0x0734, 0x0780, 0x0784, 0x0790, 0x0794, 0x07a0, - 0x07a4, 0x07b0, 0x07b4, 0x0000, 0x0004, 0x0010, 0x0014, 0x0020, 0x0024, 0x0030, 0x0034, - 0x0080, 0x0084, 0x0090, 0x0094, 0x00a0, 0x00a4, 0x00b0, 0x00b4, 0x0100, 0x0104, 0x0110, - 0x0114, 0x0120, 0x0124, 0x0130, 0x0134, 0x0180, 0x0184, 0x0190, 0x0194, 0x01a0, 0x01a4, - 0x01b0, 0x01b4, 0x0200, 0x0204, 0x0210, 0x0214, 0x0220, 0x0224, 0x0230, 0x0234, 0x0280, - 0x0284, 0x0290, 0x0294, 0x02a0, 0x02a4, 0x02b0, 0x02b4, 0x0300, 0x0304, 0x0310, 0x0314, - 0x0320, 0x0324, 0x0330, 0x0334, 0x0380, 0x0384, 0x0390, 0x0394, 0x03a0, 0x03a4, 0x03b0, - 0x03b4, 0x0400, 0x0404, 0x0410, 0x0414, 0x0420, 0x0424, 0x0430, 0x0434, 0x0480, 0x0484, - 0x0490, 0x0494, 0x04a0, 0x04a4, 0x04b0, 0x04b4, 0x0500, 0x0504, 0x0510, 0x0514, 0x0520, - 0x0524, 0x0530, 0x0534, 0x0580, 0x0584, 0x0590, 0x0594, 0x05a0, 0x05a4, 0x05b0, 0x05b4, - 0x0600, 0x0604, 0x0610, 0x0614, 0x0620, 0x0624, 0x0630, 0x0634, 0x0680, 0x0684, 0x0690, - 0x0694, 0x06a0, 0x06a4, 0x06b0, 0x06b4, 0x0700, 0x0704, 0x0710, 0x0714, 0x0720, 0x0724, - 0x0730, 0x0734, 0x0780, 0x0784, 0x0790, 0x0794, 0x07a0, 0x07a4, 0x07b0, 0x07b4, - }; - return xlut[x % 128] + ylut[y % 128]; -} - -static inline u32 GetMortonOffset128(u32 x, u32 y, u32 bytes_per_pixel) { - // Calculates the offset of the position of the pixel in Morton order - // Framebuffer images are split into 128x128 tiles. - - const unsigned int block_height = 128; - const unsigned int coarse_x = x & ~127; - - u32 i = MortonInterleave128(x, y); - - const unsigned int offset = coarse_x * block_height; - - return (i + offset) * bytes_per_pixel; -} - -static inline void MortonCopyPixels128(u32 width, u32 height, u32 bytes_per_pixel, - u32 gl_bytes_per_pixel, u8* morton_data, u8* gl_data, - bool morton_to_gl) { - u8* data_ptrs[2]; - for (unsigned y = 0; y < height; ++y) { - for (unsigned x = 0; x < width; ++x) { - const u32 coarse_y = y & ~127; - u32 morton_offset = - GetMortonOffset128(x, y, bytes_per_pixel) + coarse_y * width * bytes_per_pixel; - u32 gl_pixel_index = (x + y * width) * gl_bytes_per_pixel; - - data_ptrs[morton_to_gl] = morton_data + morton_offset; - data_ptrs[!morton_to_gl] = &gl_data[gl_pixel_index]; - - memcpy(data_ptrs[0], data_ptrs[1], bytes_per_pixel); - } - } -} - -} // namespace VideoCore |