diff options
Diffstat (limited to 'src/video_core')
39 files changed, 1921 insertions, 1179 deletions
diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt index ddb1a1d69..a780215c1 100644 --- a/src/video_core/CMakeLists.txt +++ b/src/video_core/CMakeLists.txt @@ -21,6 +21,7 @@ add_library(video_core STATIC macro_interpreter.h memory_manager.cpp memory_manager.h + rasterizer_cache.cpp rasterizer_cache.h rasterizer_interface.h renderer_base.cpp @@ -33,6 +34,7 @@ add_library(video_core STATIC renderer_opengl/gl_rasterizer.h renderer_opengl/gl_rasterizer_cache.cpp renderer_opengl/gl_rasterizer_cache.h + renderer_opengl/gl_resource_manager.cpp renderer_opengl/gl_resource_manager.h renderer_opengl/gl_shader_cache.cpp renderer_opengl/gl_shader_cache.h diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index d79c50919..2bc534be3 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -34,8 +34,49 @@ void Maxwell3D::InitializeRegisterDefaults() { // Depth range near/far is not always set, but is expected to be the default 0.0f, 1.0f. This is // needed for ARMS. for (std::size_t viewport{}; viewport < Regs::NumViewports; ++viewport) { - regs.viewport[viewport].depth_range_near = 0.0f; - regs.viewport[viewport].depth_range_far = 1.0f; + regs.viewports[viewport].depth_range_near = 0.0f; + regs.viewports[viewport].depth_range_far = 1.0f; + } + // Doom and Bomberman seems to use the uninitialized registers and just enable blend + // so initialize blend registers with sane values + regs.blend.equation_rgb = Regs::Blend::Equation::Add; + regs.blend.factor_source_rgb = Regs::Blend::Factor::One; + regs.blend.factor_dest_rgb = Regs::Blend::Factor::Zero; + regs.blend.equation_a = Regs::Blend::Equation::Add; + regs.blend.factor_source_a = Regs::Blend::Factor::One; + regs.blend.factor_dest_a = Regs::Blend::Factor::Zero; + for (std::size_t blend_index = 0; blend_index < Regs::NumRenderTargets; blend_index++) { + regs.independent_blend[blend_index].equation_rgb = Regs::Blend::Equation::Add; + regs.independent_blend[blend_index].factor_source_rgb = Regs::Blend::Factor::One; + regs.independent_blend[blend_index].factor_dest_rgb = Regs::Blend::Factor::Zero; + regs.independent_blend[blend_index].equation_a = Regs::Blend::Equation::Add; + regs.independent_blend[blend_index].factor_source_a = Regs::Blend::Factor::One; + regs.independent_blend[blend_index].factor_dest_a = Regs::Blend::Factor::Zero; + } + regs.stencil_front_op_fail = Regs::StencilOp::Keep; + regs.stencil_front_op_zfail = Regs::StencilOp::Keep; + regs.stencil_front_op_zpass = Regs::StencilOp::Keep; + regs.stencil_front_func_func = Regs::ComparisonOp::Always; + regs.stencil_front_func_mask = 0xFFFFFFFF; + regs.stencil_front_mask = 0xFFFFFFFF; + regs.stencil_two_side_enable = 1; + regs.stencil_back_op_fail = Regs::StencilOp::Keep; + regs.stencil_back_op_zfail = Regs::StencilOp::Keep; + regs.stencil_back_op_zpass = Regs::StencilOp::Keep; + regs.stencil_back_func_func = Regs::ComparisonOp::Always; + regs.stencil_back_func_mask = 0xFFFFFFFF; + regs.stencil_back_mask = 0xFFFFFFFF; + // TODO(Rodrigo): Most games do not set a point size. I think this is a case of a + // register carrying a default value. Assume it's OpenGL's default (1). + regs.point_size = 1.0f; + + // TODO(bunnei): Some games do not initialize the color masks (e.g. Sonic Mania). Assuming a + // default of enabled fixes rendering here. + for (std::size_t color_mask = 0; color_mask < Regs::NumRenderTargets; color_mask++) { + regs.color_mask[color_mask].R.Assign(1); + regs.color_mask[color_mask].G.Assign(1); + regs.color_mask[color_mask].B.Assign(1); + regs.color_mask[color_mask].A.Assign(1); } } @@ -92,7 +133,13 @@ void Maxwell3D::WriteReg(u32 method, u32 value, u32 remaining_params) { debug_context->OnEvent(Tegra::DebugContext::Event::MaxwellCommandLoaded, nullptr); } - regs.reg_array[method] = value; + 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()) { + dirty_flags.vertex_attrib_format = true; + } + } switch (method) { case MAXWELL3D_REG_INDEX(macros.data): { diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h index 50873813e..4f137e693 100644 --- a/src/video_core/engines/maxwell_3d.h +++ b/src/video_core/engines/maxwell_3d.h @@ -345,6 +345,14 @@ public: Invert = 6, IncrWrap = 7, DecrWrap = 8, + KeepOGL = 0x1E00, + ZeroOGL = 0, + ReplaceOGL = 0x1E01, + IncrOGL = 0x1E02, + DecrOGL = 0x1E03, + InvertOGL = 0x150A, + IncrWrapOGL = 0x8507, + DecrWrapOGL = 0x8508, }; enum class MemoryLayout : u32 { @@ -381,6 +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 { @@ -462,6 +477,77 @@ public: } }; + struct ColorMask { + union { + u32 raw; + BitField<0, 4, u32> R; + BitField<4, 4, u32> G; + BitField<8, 4, u32> B; + BitField<12, 4, u32> A; + }; + }; + + struct ViewportTransform { + f32 scale_x; + f32 scale_y; + f32 scale_z; + f32 translate_x; + f32 translate_y; + f32 translate_z; + INSERT_PADDING_WORDS(2); + + MathUtil::Rectangle<s32> GetRect() const { + return { + GetX(), // left + GetY() + GetHeight(), // top + GetX() + GetWidth(), // right + GetY() // bottom + }; + }; + + s32 GetX() const { + return static_cast<s32>(std::max(0.0f, translate_x - std::fabs(scale_x))); + } + + s32 GetY() const { + return static_cast<s32>(std::max(0.0f, translate_y - std::fabs(scale_y))); + } + + s32 GetWidth() const { + return static_cast<s32>(translate_x + std::fabs(scale_x)) - GetX(); + } + + s32 GetHeight() const { + return static_cast<s32>(translate_y + std::fabs(scale_y)) - GetY(); + } + }; + + struct ScissorTest { + u32 enable; + union { + BitField<0, 16, u32> min_x; + BitField<16, 16, u32> max_x; + }; + union { + BitField<0, 16, u32> min_y; + BitField<16, 16, u32> max_y; + }; + u32 fill; + }; + + struct ViewPort { + union { + BitField<0, 16, u32> x; + BitField<16, 16, u32> width; + }; + union { + BitField<0, 16, u32> y; + BitField<16, 16, u32> height; + }; + float depth_range_near; + float depth_range_far; + }; + bool IsShaderConfigEnabled(std::size_t index) const { // The VertexB is always enabled. if (index == static_cast<std::size_t>(Regs::ShaderProgram::VertexB)) { @@ -487,55 +573,11 @@ public: INSERT_PADDING_WORDS(0x2E); - RenderTargetConfig rt[NumRenderTargets]; + std::array<RenderTargetConfig, NumRenderTargets> rt; - struct { - f32 scale_x; - f32 scale_y; - f32 scale_z; - f32 translate_x; - f32 translate_y; - f32 translate_z; - INSERT_PADDING_WORDS(2); - - MathUtil::Rectangle<s32> GetRect() const { - return { - GetX(), // left - GetY() + GetHeight(), // top - GetX() + GetWidth(), // right - GetY() // bottom - }; - }; + std::array<ViewportTransform, NumViewports> viewport_transform; - s32 GetX() const { - return static_cast<s32>(std::max(0.0f, translate_x - std::fabs(scale_x))); - } - - s32 GetY() const { - return static_cast<s32>(std::max(0.0f, translate_y - std::fabs(scale_y))); - } - - s32 GetWidth() const { - return static_cast<s32>(translate_x + std::fabs(scale_x)) - GetX(); - } - - s32 GetHeight() const { - return static_cast<s32>(translate_y + std::fabs(scale_y)) - GetY(); - } - } viewport_transform[NumViewports]; - - struct { - union { - BitField<0, 16, u32> x; - BitField<16, 16, u32> width; - }; - union { - BitField<0, 16, u32> y; - BitField<16, 16, u32> height; - }; - float depth_range_near; - float depth_range_far; - } viewport[NumViewports]; + std::array<ViewPort, NumViewports> viewports; INSERT_PADDING_WORDS(0x1D); @@ -553,25 +595,19 @@ public: INSERT_PADDING_WORDS(0x17); - struct { - u32 enable; - union { - BitField<0, 16, u32> min_x; - BitField<16, 16, u32> max_x; - }; - union { - BitField<0, 16, u32> min_y; - BitField<16, 16, u32> max_y; - }; - } scissor_test; + std::array<ScissorTest, NumViewports> scissor_test; - INSERT_PADDING_WORDS(0x52); + INSERT_PADDING_WORDS(0x15); s32 stencil_back_func_ref; u32 stencil_back_mask; u32 stencil_back_func_mask; - INSERT_PADDING_WORDS(0x13); + INSERT_PADDING_WORDS(0xC); + + u32 color_mask_common; + + INSERT_PADDING_WORDS(0x6); u32 rt_separate_frag_data; @@ -595,7 +631,16 @@ 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; @@ -646,8 +691,14 @@ public: ComparisonOp depth_test_func; float alpha_test_ref; ComparisonOp alpha_test_func; - - INSERT_PADDING_WORDS(0x9); + u32 draw_tfb_stride; + struct { + float r; + float g; + float b; + float a; + } blend_color; + INSERT_PADDING_WORDS(0x4); struct { u32 separate_alpha; @@ -672,7 +723,9 @@ public: u32 stencil_front_func_mask; u32 stencil_front_mask; - INSERT_PADDING_WORDS(0x3); + INSERT_PADDING_WORDS(0x2); + + u32 frag_color_clamp; union { BitField<4, 1, u32> triangle_rast_flip; @@ -690,7 +743,12 @@ public: u32 zeta_enable; - INSERT_PADDING_WORDS(0x8); + union { + BitField<0, 1, u32> alpha_to_coverage; + BitField<4, 1, u32> alpha_to_one; + } multisample_control; + + INSERT_PADDING_WORDS(0x7); struct { u32 tsc_address_high; @@ -841,8 +899,9 @@ public: BitField<6, 4, u32> RT; BitField<10, 11, u32> layer; } clear_buffers; - - INSERT_PADDING_WORDS(0x4B); + INSERT_PADDING_WORDS(0xB); + std::array<ColorMask, NumRenderTargets> color_mask; + INSERT_PADDING_WORDS(0x38); struct { u32 query_address_high; @@ -983,6 +1042,12 @@ public: State state{}; MemoryManager& memory_manager; + struct DirtyFlags { + bool vertex_attrib_format = true; + }; + + DirtyFlags dirty_flags; + /// Reads a register value located at the input method address u32 GetRegisterValue(u32 method) const; @@ -1065,8 +1130,8 @@ private: ASSERT_REG_POSITION(macros, 0x45); ASSERT_REG_POSITION(tfb_enabled, 0x1D1); ASSERT_REG_POSITION(rt, 0x200); -ASSERT_REG_POSITION(viewport_transform[0], 0x280); -ASSERT_REG_POSITION(viewport, 0x300); +ASSERT_REG_POSITION(viewport_transform, 0x280); +ASSERT_REG_POSITION(viewports, 0x300); ASSERT_REG_POSITION(vertex_buffer, 0x35D); ASSERT_REG_POSITION(clear_color[0], 0x360); ASSERT_REG_POSITION(clear_depth, 0x364); @@ -1075,8 +1140,10 @@ ASSERT_REG_POSITION(scissor_test, 0x380); ASSERT_REG_POSITION(stencil_back_func_ref, 0x3D5); ASSERT_REG_POSITION(stencil_back_mask, 0x3D6); 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); @@ -1087,6 +1154,10 @@ ASSERT_REG_POSITION(depth_write_enabled, 0x4BA); ASSERT_REG_POSITION(alpha_test_enabled, 0x4BB); ASSERT_REG_POSITION(d3d_cull_mode, 0x4C2); ASSERT_REG_POSITION(depth_test_func, 0x4C3); +ASSERT_REG_POSITION(alpha_test_ref, 0x4C4); +ASSERT_REG_POSITION(alpha_test_func, 0x4C5); +ASSERT_REG_POSITION(draw_tfb_stride, 0x4C6); +ASSERT_REG_POSITION(blend_color, 0x4C7); ASSERT_REG_POSITION(blend, 0x4CF); ASSERT_REG_POSITION(stencil_enable, 0x4E0); ASSERT_REG_POSITION(stencil_front_op_fail, 0x4E1); @@ -1096,10 +1167,12 @@ ASSERT_REG_POSITION(stencil_front_func_func, 0x4E4); ASSERT_REG_POSITION(stencil_front_func_ref, 0x4E5); ASSERT_REG_POSITION(stencil_front_func_mask, 0x4E6); 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(point_size, 0x546); ASSERT_REG_POSITION(zeta_enable, 0x54E); +ASSERT_REG_POSITION(multisample_control, 0x54F); ASSERT_REG_POSITION(tsc, 0x557); ASSERT_REG_POSITION(tic, 0x55D); ASSERT_REG_POSITION(stencil_two_side_enable, 0x565); @@ -1117,6 +1190,7 @@ ASSERT_REG_POSITION(instanced_arrays, 0x620); ASSERT_REG_POSITION(cull, 0x646); ASSERT_REG_POSITION(logic_op, 0x671); ASSERT_REG_POSITION(clear_buffers, 0x674); +ASSERT_REG_POSITION(color_mask, 0x680); ASSERT_REG_POSITION(query, 0x6C0); ASSERT_REG_POSITION(vertex_array[0], 0x700); ASSERT_REG_POSITION(independent_blend, 0x780); diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h index 83a6fd875..7e8449bc4 100644 --- a/src/video_core/engines/shader_bytecode.h +++ b/src/video_core/engines/shader_bytecode.h @@ -153,6 +153,7 @@ enum class PredCondition : u64 { NotEqual = 5, GreaterEqual = 6, LessThanWithNan = 9, + LessEqualWithNan = 11, GreaterThanWithNan = 12, NotEqualWithNan = 13, GreaterEqualWithNan = 14, @@ -261,7 +262,7 @@ 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, @@ -569,7 +570,6 @@ 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 +831,7 @@ 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 @@ -1235,7 +1235,7 @@ 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; diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp index 83c7e5b0b..51b3904f6 100644 --- a/src/video_core/gpu.cpp +++ b/src/video_core/gpu.cpp @@ -17,6 +17,8 @@ u32 FramebufferConfig::BytesPerPixel(PixelFormat format) { switch (format) { case PixelFormat::ABGR8: return 4; + default: + return 4; } UNREACHABLE(); diff --git a/src/video_core/macro_interpreter.cpp b/src/video_core/macro_interpreter.cpp index 335a8d407..2b0dea5cd 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: 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 90a8e825d..77a20bb84 100644 --- a/src/video_core/memory_manager.cpp +++ b/src/video_core/memory_manager.cpp @@ -4,18 +4,21 @@ #include "common/alignment.h" #include "common/assert.h" +#include "common/logging/log.h" #include "video_core/memory_manager.h" namespace Tegra { GPUVAddr MemoryManager::AllocateSpace(u64 size, u64 align) { - std::optional<GPUVAddr> gpu_addr = FindFreeBlock(size, align); - ASSERT(gpu_addr); + const std::optional<GPUVAddr> gpu_addr{FindFreeBlock(0, size, align, PageStatus::Unmapped)}; - for (u64 offset = 0; offset < size; offset += PAGE_SIZE) { - VAddr& slot = PageSlot(*gpu_addr + offset); + ASSERT_MSG(gpu_addr, "unable to find available GPU memory"); + + for (u64 offset{}; offset < size; offset += PAGE_SIZE) { + VAddr& slot{PageSlot(*gpu_addr + offset)}; ASSERT(slot == static_cast<u64>(PageStatus::Unmapped)); + slot = static_cast<u64>(PageStatus::Allocated); } @@ -23,10 +26,11 @@ GPUVAddr MemoryManager::AllocateSpace(u64 size, u64 align) { } GPUVAddr MemoryManager::AllocateSpace(GPUVAddr gpu_addr, u64 size, u64 align) { - for (u64 offset = 0; offset < size; offset += PAGE_SIZE) { - VAddr& slot = PageSlot(gpu_addr + offset); + for (u64 offset{}; offset < size; offset += PAGE_SIZE) { + VAddr& slot{PageSlot(gpu_addr + offset)}; ASSERT(slot == static_cast<u64>(PageStatus::Unmapped)); + slot = static_cast<u64>(PageStatus::Allocated); } @@ -34,17 +38,19 @@ GPUVAddr MemoryManager::AllocateSpace(GPUVAddr gpu_addr, u64 size, u64 align) { } GPUVAddr MemoryManager::MapBufferEx(VAddr cpu_addr, u64 size) { - std::optional<GPUVAddr> gpu_addr = FindFreeBlock(size, PAGE_SIZE); - ASSERT(gpu_addr); + const std::optional<GPUVAddr> gpu_addr{FindFreeBlock(0, size, PAGE_SIZE, PageStatus::Unmapped)}; + + ASSERT_MSG(gpu_addr, "unable to find available GPU memory"); - for (u64 offset = 0; offset < size; offset += PAGE_SIZE) { - VAddr& slot = PageSlot(*gpu_addr + offset); + for (u64 offset{}; offset < size; offset += PAGE_SIZE) { + VAddr& slot{PageSlot(*gpu_addr + offset)}; ASSERT(slot == static_cast<u64>(PageStatus::Unmapped)); + slot = cpu_addr + offset; } - MappedRegion region{cpu_addr, *gpu_addr, size}; + const MappedRegion region{cpu_addr, *gpu_addr, size}; mapped_regions.push_back(region); return *gpu_addr; @@ -53,14 +59,31 @@ GPUVAddr MemoryManager::MapBufferEx(VAddr cpu_addr, u64 size) { GPUVAddr MemoryManager::MapBufferEx(VAddr cpu_addr, GPUVAddr gpu_addr, u64 size) { ASSERT((gpu_addr & PAGE_MASK) == 0); - for (u64 offset = 0; offset < size; offset += PAGE_SIZE) { - VAddr& slot = PageSlot(gpu_addr + offset); + if (PageSlot(gpu_addr) != static_cast<u64>(PageStatus::Allocated)) { + // Page has been already mapped. In this case, we must find a new area of memory to use that + // is different than the specified one. Super Mario Odyssey hits this scenario when changing + // areas, but we do not want to overwrite the old pages. + // TODO(bunnei): We need to write a hardware test to confirm this behavior. + + LOG_ERROR(HW_GPU, "attempting to map addr 0x{:016X}, which is not available!", gpu_addr); + + const std::optional<GPUVAddr> new_gpu_addr{ + FindFreeBlock(gpu_addr, size, PAGE_SIZE, PageStatus::Allocated)}; + + ASSERT_MSG(new_gpu_addr, "unable to find available GPU memory"); + + gpu_addr = *new_gpu_addr; + } + + for (u64 offset{}; offset < size; offset += PAGE_SIZE) { + VAddr& slot{PageSlot(gpu_addr + offset)}; ASSERT(slot == static_cast<u64>(PageStatus::Allocated)); + slot = cpu_addr + offset; } - MappedRegion region{cpu_addr, gpu_addr, size}; + const MappedRegion region{cpu_addr, gpu_addr, size}; mapped_regions.push_back(region); return gpu_addr; @@ -69,11 +92,12 @@ GPUVAddr MemoryManager::MapBufferEx(VAddr cpu_addr, GPUVAddr gpu_addr, u64 size) GPUVAddr MemoryManager::UnmapBuffer(GPUVAddr gpu_addr, u64 size) { ASSERT((gpu_addr & PAGE_MASK) == 0); - for (u64 offset = 0; offset < size; offset += PAGE_SIZE) { - VAddr& slot = PageSlot(gpu_addr + offset); + for (u64 offset{}; offset < size; offset += PAGE_SIZE) { + VAddr& slot{PageSlot(gpu_addr + offset)}; ASSERT(slot != static_cast<u64>(PageStatus::Allocated) && slot != static_cast<u64>(PageStatus::Unmapped)); + slot = static_cast<u64>(PageStatus::Unmapped); } @@ -97,13 +121,14 @@ GPUVAddr MemoryManager::GetRegionEnd(GPUVAddr region_start) const { return {}; } -std::optional<GPUVAddr> MemoryManager::FindFreeBlock(u64 size, u64 align) { - GPUVAddr gpu_addr = 0; - u64 free_space = 0; +std::optional<GPUVAddr> MemoryManager::FindFreeBlock(GPUVAddr region_start, u64 size, u64 align, + PageStatus status) { + GPUVAddr gpu_addr{region_start}; + u64 free_space{}; align = (align + PAGE_MASK) & ~PAGE_MASK; while (gpu_addr + free_space < MAX_ADDRESS) { - if (!IsPageMapped(gpu_addr + free_space)) { + if (PageSlot(gpu_addr + free_space) == static_cast<u64>(status)) { free_space += PAGE_SIZE; if (free_space >= size) { return gpu_addr; @@ -119,7 +144,7 @@ std::optional<GPUVAddr> MemoryManager::FindFreeBlock(u64 size, u64 align) { } std::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr gpu_addr) { - VAddr base_addr = PageSlot(gpu_addr); + const VAddr base_addr{PageSlot(gpu_addr)}; if (base_addr == static_cast<u64>(PageStatus::Allocated) || base_addr == static_cast<u64>(PageStatus::Unmapped)) { @@ -133,19 +158,15 @@ std::vector<GPUVAddr> MemoryManager::CpuToGpuAddress(VAddr cpu_addr) const { std::vector<GPUVAddr> results; for (const auto& region : mapped_regions) { if (cpu_addr >= region.cpu_addr && cpu_addr < (region.cpu_addr + region.size)) { - u64 offset = cpu_addr - region.cpu_addr; + const u64 offset{cpu_addr - region.cpu_addr}; results.push_back(region.gpu_addr + offset); } } return results; } -bool MemoryManager::IsPageMapped(GPUVAddr gpu_addr) { - return PageSlot(gpu_addr) != static_cast<u64>(PageStatus::Unmapped); -} - VAddr& MemoryManager::PageSlot(GPUVAddr gpu_addr) { - auto& block = page_table[(gpu_addr >> (PAGE_BITS + PAGE_TABLE_BITS)) & PAGE_TABLE_MASK]; + auto& block{page_table[(gpu_addr >> (PAGE_BITS + PAGE_TABLE_BITS)) & PAGE_TABLE_MASK]}; if (!block) { block = std::make_unique<PageBlock>(); block->fill(static_cast<VAddr>(PageStatus::Unmapped)); diff --git a/src/video_core/memory_manager.h b/src/video_core/memory_manager.h index b1255fd56..4eb338aa2 100644 --- a/src/video_core/memory_manager.h +++ b/src/video_core/memory_manager.h @@ -34,15 +34,15 @@ public: static constexpr u64 PAGE_MASK = PAGE_SIZE - 1; private: - std::optional<GPUVAddr> FindFreeBlock(u64 size, u64 align = 1); - bool IsPageMapped(GPUVAddr gpu_addr); - VAddr& PageSlot(GPUVAddr gpu_addr); - enum class PageStatus : u64 { Unmapped = 0xFFFFFFFFFFFFFFFFULL, Allocated = 0xFFFFFFFFFFFFFFFEULL, }; + std::optional<GPUVAddr> FindFreeBlock(GPUVAddr region_start, u64 size, u64 align, + PageStatus status); + VAddr& PageSlot(GPUVAddr gpu_addr); + static constexpr u64 MAX_ADDRESS{0x10000000000ULL}; static constexpr u64 PAGE_TABLE_BITS{10}; static constexpr u64 PAGE_TABLE_SIZE{1 << PAGE_TABLE_BITS}; diff --git a/src/video_core/rasterizer_cache.cpp b/src/video_core/rasterizer_cache.cpp new file mode 100644 index 000000000..093b2cdf4 --- /dev/null +++ b/src/video_core/rasterizer_cache.cpp @@ -0,0 +1,7 @@ +// Copyright 2018 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include "video_core/rasterizer_cache.h" + +RasterizerCacheObject::~RasterizerCacheObject() = default; diff --git a/src/video_core/rasterizer_cache.h b/src/video_core/rasterizer_cache.h index 0a3b3951e..bcf0c15a4 100644 --- a/src/video_core/rasterizer_cache.h +++ b/src/video_core/rasterizer_cache.h @@ -5,18 +5,19 @@ #pragma once #include <set> +#include <unordered_map> #include <boost/icl/interval_map.hpp> #include <boost/range/iterator_range_core.hpp> #include "common/common_types.h" -#include "core/core.h" #include "core/settings.h" #include "video_core/rasterizer_interface.h" -#include "video_core/renderer_base.h" class RasterizerCacheObject { public: + virtual ~RasterizerCacheObject(); + /// Gets the address of the shader in guest memory, required for cache management virtual VAddr GetAddr() const = 0; @@ -64,6 +65,8 @@ class RasterizerCache : NonCopyable { friend class RasterizerCacheObject; public: + explicit RasterizerCache(VideoCore::RasterizerInterface& rasterizer) : rasterizer{rasterizer} {} + /// Write any cached resources overlapping the specified region back to memory void FlushRegion(Tegra::GPUVAddr addr, size_t size) { const auto& objects{GetSortedObjectsFromRegion(addr, size)}; @@ -86,45 +89,39 @@ public: /// Invalidates everything in the cache void InvalidateAll() { - while (object_cache.begin() != object_cache.end()) { - Unregister(*object_cache.begin()->second.begin()); + while (interval_cache.begin() != interval_cache.end()) { + Unregister(*interval_cache.begin()->second.begin()); } } protected: /// Tries to get an object from the cache with the specified address T TryGet(VAddr addr) const { - const ObjectInterval interval{addr}; - for (auto& pair : boost::make_iterator_range(object_cache.equal_range(interval))) { - for (auto& cached_object : pair.second) { - if (cached_object->GetAddr() == addr) { - return cached_object; - } - } - } + const auto iter = map_cache.find(addr); + if (iter != map_cache.end()) + return iter->second; return nullptr; } /// Register an object into the cache void Register(const T& object) { object->SetIsRegistered(true); - object_cache.add({GetInterval(object), ObjectSet{object}}); - auto& rasterizer = Core::System::GetInstance().Renderer().Rasterizer(); + interval_cache.add({GetInterval(object), ObjectSet{object}}); + map_cache.insert({object->GetAddr(), object}); rasterizer.UpdatePagesCachedCount(object->GetAddr(), object->GetSizeInBytes(), 1); } /// Unregisters an object from the cache void Unregister(const T& object) { object->SetIsRegistered(false); - auto& rasterizer = Core::System::GetInstance().Renderer().Rasterizer(); rasterizer.UpdatePagesCachedCount(object->GetAddr(), object->GetSizeInBytes(), -1); - // Only flush if use_accurate_gpu_emulation is enabled, as it incurs a performance hit if (Settings::values.use_accurate_gpu_emulation) { FlushObject(object); } - object_cache.subtract({GetInterval(object), ObjectSet{object}}); + interval_cache.subtract({GetInterval(object), ObjectSet{object}}); + map_cache.erase(object->GetAddr()); } /// Returns a ticks counter used for tracking when cached objects were last modified @@ -141,7 +138,7 @@ private: std::vector<T> objects; const ObjectInterval interval{addr, addr + size}; - for (auto& pair : boost::make_iterator_range(object_cache.equal_range(interval))) { + for (auto& pair : boost::make_iterator_range(interval_cache.equal_range(interval))) { for (auto& cached_object : pair.second) { if (!cached_object) { continue; @@ -167,14 +164,17 @@ private: } using ObjectSet = std::set<T>; - using ObjectCache = boost::icl::interval_map<VAddr, ObjectSet>; - using ObjectInterval = typename ObjectCache::interval_type; + using ObjectCache = std::unordered_map<VAddr, T>; + using IntervalCache = boost::icl::interval_map<VAddr, ObjectSet>; + using ObjectInterval = typename IntervalCache::interval_type; static auto GetInterval(const T& object) { return ObjectInterval::right_open(object->GetAddr(), object->GetAddr() + object->GetSizeInBytes()); } - ObjectCache object_cache; ///< Cache of objects - u64 modified_ticks{}; ///< Counter of cache state ticks, used for in-order flushing + ObjectCache map_cache; + IntervalCache interval_cache; ///< Cache of objects + u64 modified_ticks{}; ///< Counter of cache state ticks, used for in-order flushing + VideoCore::RasterizerInterface& rasterizer; }; diff --git a/src/video_core/renderer_base.cpp b/src/video_core/renderer_base.cpp index 0df3725c2..1482cdb40 100644 --- a/src/video_core/renderer_base.cpp +++ b/src/video_core/renderer_base.cpp @@ -5,7 +5,6 @@ #include "core/frontend/emu_window.h" #include "core/settings.h" #include "video_core/renderer_base.h" -#include "video_core/renderer_opengl/gl_rasterizer.h" 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 41a54b3e7..075192c3f 100644 --- a/src/video_core/renderer_opengl/gl_buffer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_buffer_cache.cpp @@ -9,10 +9,12 @@ #include "core/core.h" #include "core/memory.h" #include "video_core/renderer_opengl/gl_buffer_cache.h" +#include "video_core/renderer_opengl/gl_rasterizer.h" namespace OpenGL { -OGLBufferCache::OGLBufferCache(std::size_t size) : stream_buffer(GL_ARRAY_BUFFER, size) {} +OGLBufferCache::OGLBufferCache(RasterizerOpenGL& rasterizer, std::size_t size) + : RasterizerCache{rasterizer}, stream_buffer(GL_ARRAY_BUFFER, size) {} GLintptr OGLBufferCache::UploadMemory(Tegra::GPUVAddr gpu_addr, std::size_t size, std::size_t alignment, bool cache) { diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.h b/src/video_core/renderer_opengl/gl_buffer_cache.h index be29dc8be..91fca3f6c 100644 --- a/src/video_core/renderer_opengl/gl_buffer_cache.h +++ b/src/video_core/renderer_opengl/gl_buffer_cache.h @@ -15,6 +15,8 @@ namespace OpenGL { +class RasterizerOpenGL; + struct CachedBufferEntry final : public RasterizerCacheObject { VAddr GetAddr() const override { return addr; @@ -35,7 +37,7 @@ struct CachedBufferEntry final : public RasterizerCacheObject { class OGLBufferCache final : public RasterizerCache<std::shared_ptr<CachedBufferEntry>> { public: - explicit OGLBufferCache(std::size_t size); + explicit OGLBufferCache(RasterizerOpenGL& rasterizer, std::size_t size); /// Uploads data from a guest GPU address. Returns host's buffer offset where it's been /// allocated. diff --git a/src/video_core/renderer_opengl/gl_primitive_assembler.cpp b/src/video_core/renderer_opengl/gl_primitive_assembler.cpp index 741f14bc3..d9ed08437 100644 --- a/src/video_core/renderer_opengl/gl_primitive_assembler.cpp +++ b/src/video_core/renderer_opengl/gl_primitive_assembler.cpp @@ -6,6 +6,7 @@ #include <array> #include "common/assert.h" #include "common/common_types.h" +#include "core/core.h" #include "core/memory.h" #include "video_core/renderer_opengl/gl_buffer_cache.h" #include "video_core/renderer_opengl/gl_primitive_assembler.h" diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index a0527fe57..630a58e49 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -33,7 +33,8 @@ using Maxwell = Tegra::Engines::Maxwell3D::Regs; using PixelFormat = VideoCore::Surface::PixelFormat; using SurfaceType = VideoCore::Surface::SurfaceType; -MICROPROFILE_DEFINE(OpenGL_VAO, "OpenGL", "Vertex Array Setup", MP_RGB(128, 128, 192)); +MICROPROFILE_DEFINE(OpenGL_VAO, "OpenGL", "Vertex Format Setup", MP_RGB(128, 128, 192)); +MICROPROFILE_DEFINE(OpenGL_VB, "OpenGL", "Vertex Buffer Setup", MP_RGB(128, 128, 192)); MICROPROFILE_DEFINE(OpenGL_Shader, "OpenGL", "Shader Setup", MP_RGB(128, 128, 192)); MICROPROFILE_DEFINE(OpenGL_UBO, "OpenGL", "Const Buffer Setup", MP_RGB(128, 128, 192)); MICROPROFILE_DEFINE(OpenGL_Index, "OpenGL", "Index Buffer Setup", MP_RGB(128, 128, 192)); @@ -79,7 +80,8 @@ struct DrawParameters { }; RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& window, ScreenInfo& info) - : emu_window{window}, screen_info{info}, buffer_cache(STREAM_BUFFER_SIZE) { + : res_cache{*this}, shader_cache{*this}, emu_window{window}, screen_info{info}, + buffer_cache(*this, STREAM_BUFFER_SIZE) { // Create sampler objects for (std::size_t i = 0; i < texture_samplers.size(); ++i) { texture_samplers[i].Create(); @@ -96,17 +98,10 @@ 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(); - // Clipping plane 0 is always enabled for PICA fixed clip plane z <= 0 - state.clip_distance[0] = true; // Create render framebuffer framebuffer.Create(); @@ -122,18 +117,23 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& window, ScreenInfo RasterizerOpenGL::~RasterizerOpenGL() {} -void RasterizerOpenGL::SetupVertexArrays() { - MICROPROFILE_SCOPE(OpenGL_VAO); - const auto& gpu = Core::System::GetInstance().GPU().Maxwell3D(); +void RasterizerOpenGL::SetupVertexFormat() { + auto& gpu = Core::System::GetInstance().GPU().Maxwell3D(); const auto& regs = gpu.regs; + if (!gpu.dirty_flags.vertex_attrib_format) + return; + gpu.dirty_flags.vertex_attrib_format = false; + + MICROPROFILE_SCOPE(OpenGL_VAO); + auto [iter, is_cache_miss] = vertex_array_cache.try_emplace(regs.vertex_attrib_format); auto& VAO = iter->second; if (is_cache_miss) { VAO.Create(); state.draw.vertex_array = VAO.handle; - state.Apply(); + state.ApplyVertexBufferState(); // The index buffer binding is stored within the VAO. Stupid OpenGL, but easy to work // around. @@ -175,8 +175,13 @@ void RasterizerOpenGL::SetupVertexArrays() { } } state.draw.vertex_array = VAO.handle; - state.draw.vertex_buffer = buffer_cache.GetHandle(); - state.Apply(); + state.ApplyVertexBufferState(); +} + +void RasterizerOpenGL::SetupVertexBuffer() { + MICROPROFILE_SCOPE(OpenGL_VB); + const auto& gpu = Core::System::GetInstance().GPU().Maxwell3D(); + const auto& regs = gpu.regs; // Upload all guest vertex arrays sequentially to our buffer for (u32 index = 0; index < Maxwell::NumVertexArrays; ++index) { @@ -203,6 +208,9 @@ void RasterizerOpenGL::SetupVertexArrays() { glVertexBindingDivisor(index, 0); } } + + // Implicit set by glBindVertexBuffer. Stupid glstate handling... + state.draw.vertex_buffer = buffer_cache.GetHandle(); } DrawParameters RasterizerOpenGL::SetupDraw() { @@ -327,8 +335,6 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) { index++; } } - - state.Apply(); } std::size_t RasterizerOpenGL::CalculateVertexArraysSize() const { @@ -397,8 +403,8 @@ void RasterizerOpenGL::UpdatePagesCachedCount(VAddr addr, u64 size, int delta) { cached_pages.add({pages_interval, delta}); } -void RasterizerOpenGL::ConfigureFramebuffers(bool using_color_fb, bool using_depth_fb, - bool preserve_contents, +void RasterizerOpenGL::ConfigureFramebuffers(OpenGLState& current_state, bool using_color_fb, + bool using_depth_fb, bool preserve_contents, std::optional<std::size_t> single_color_target) { MICROPROFILE_SCOPE(OpenGL_Framebuffer); const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs; @@ -414,9 +420,9 @@ void RasterizerOpenGL::ConfigureFramebuffers(bool using_color_fb, bool using_dep ASSERT_MSG(regs.rt_separate_frag_data == 0, "Unimplemented"); // Bind the framebuffer surfaces - state.draw.draw_framebuffer = framebuffer.handle; - state.Apply(); - state.framebuffer_srgb.enabled = regs.framebuffer_srgb != 0; + current_state.draw.draw_framebuffer = framebuffer.handle; + current_state.ApplyFramebufferState(); + current_state.framebuffer_srgb.enabled = regs.framebuffer_srgb != 0; if (using_color_fb) { if (single_color_target) { @@ -494,10 +500,7 @@ void RasterizerOpenGL::ConfigureFramebuffers(bool using_color_fb, bool using_dep glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0); } - - SyncViewport(); - - state.Apply(); + SyncViewport(current_state); } void RasterizerOpenGL::Clear() { @@ -510,22 +513,23 @@ void RasterizerOpenGL::Clear() { bool use_stencil{}; OpenGLState clear_state; - clear_state.draw.draw_framebuffer = framebuffer.handle; - clear_state.color_mask.red_enabled = regs.clear_buffers.R ? GL_TRUE : GL_FALSE; - clear_state.color_mask.green_enabled = regs.clear_buffers.G ? GL_TRUE : GL_FALSE; - clear_state.color_mask.blue_enabled = regs.clear_buffers.B ? GL_TRUE : GL_FALSE; - clear_state.color_mask.alpha_enabled = regs.clear_buffers.A ? GL_TRUE : GL_FALSE; - if (regs.clear_buffers.R || regs.clear_buffers.G || regs.clear_buffers.B || regs.clear_buffers.A) { use_color = true; } + if (use_color) { + clear_state.color_mask[0].red_enabled = regs.clear_buffers.R ? GL_TRUE : GL_FALSE; + clear_state.color_mask[0].green_enabled = regs.clear_buffers.G ? GL_TRUE : GL_FALSE; + clear_state.color_mask[0].blue_enabled = regs.clear_buffers.B ? GL_TRUE : GL_FALSE; + clear_state.color_mask[0].alpha_enabled = regs.clear_buffers.A ? GL_TRUE : GL_FALSE; + } if (regs.clear_buffers.Z) { ASSERT_MSG(regs.zeta_enable != 0, "Tried to clear Z but buffer is not enabled!"); use_depth = true; // Always enable the depth write when clearing the depth buffer. The depth write mask is - // ignored when clearing the buffer in the Switch, but OpenGL obeys it so we set it to true. + // ignored when clearing the buffer in the Switch, but OpenGL obeys it so we set it to + // true. clear_state.depth.test_enabled = true; clear_state.depth.test_func = GL_ALWAYS; } @@ -533,6 +537,30 @@ 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) { @@ -542,11 +570,16 @@ void RasterizerOpenGL::Clear() { ScopeAcquireGLContext acquire_context{emu_window}; - ConfigureFramebuffers(use_color, use_depth || use_stencil, false, + ConfigureFramebuffers(clear_state, use_color, use_depth || use_stencil, false, regs.clear_buffers.RT.Value()); - // Copy the sRGB setting to the clear state to avoid problem with - // specific driver implementations - clear_state.framebuffer_srgb.enabled = state.framebuffer_srgb.enabled; + if (regs.clear_flags.scissor) { + SyncScissorTest(clear_state); + } + + if (regs.clear_flags.viewport) { + clear_state.EmulateViewportWithScissor(); + } + clear_state.Apply(); if (use_color) { @@ -572,16 +605,17 @@ void RasterizerOpenGL::DrawArrays() { ScopeAcquireGLContext acquire_context{emu_window}; - ConfigureFramebuffers(); - + ConfigureFramebuffers(state); + SyncColorMask(); + SyncFragmentColorClampState(); + SyncMultiSampleState(); SyncDepthTestState(); SyncStencilTestState(); SyncBlendState(); SyncLogicOpState(); SyncCullMode(); SyncPrimitiveRestart(); - SyncDepthRange(); - SyncScissorTest(); + SyncScissorTest(state); // Alpha Testing is synced on shaders. SyncTransformFeedback(); SyncPointState(); @@ -594,7 +628,7 @@ void RasterizerOpenGL::DrawArrays() { const bool is_indexed = accelerate_draw == AccelDraw::Indexed; state.draw.vertex_buffer = buffer_cache.GetHandle(); - state.Apply(); + state.ApplyVertexBufferState(); std::size_t buffer_size = CalculateVertexArraysSize(); @@ -621,7 +655,8 @@ void RasterizerOpenGL::DrawArrays() { buffer_cache.Map(buffer_size); - SetupVertexArrays(); + SetupVertexFormat(); + SetupVertexBuffer(); DrawParameters params = SetupDraw(); SetupShaders(params.primitive_mode); @@ -634,7 +669,7 @@ void RasterizerOpenGL::DrawArrays() { params.DispatchDraw(); // Disable scissor test - state.scissor.enabled = false; + state.viewports[0].scissor.enabled = false; accelerate_draw = AccelDraw::Disabled; @@ -727,7 +762,6 @@ void RasterizerOpenGL::SamplerInfo::Create() { void RasterizerOpenGL::SamplerInfo::SyncWithConfig(const Tegra::Texture::TSCEntry& config) { const GLuint s = sampler.handle; - if (mag_filter != config.mag_filter) { mag_filter = config.mag_filter; glSamplerParameteri( @@ -769,15 +803,51 @@ void RasterizerOpenGL::SamplerInfo::SyncWithConfig(const Tegra::Texture::TSCEntr MaxwellToGL::DepthCompareFunc(depth_compare_func)); } - if (wrap_u == Tegra::Texture::WrapMode::Border || wrap_v == Tegra::Texture::WrapMode::Border || - wrap_p == Tegra::Texture::WrapMode::Border) { - const GLvec4 new_border_color = {{config.border_color_r, config.border_color_g, - config.border_color_b, config.border_color_a}}; - if (border_color != new_border_color) { - border_color = new_border_color; - glSamplerParameterfv(s, GL_TEXTURE_BORDER_COLOR, border_color.data()); + GLvec4 new_border_color; + if (config.srgb_conversion) { + new_border_color[0] = config.srgb_border_color_r / 255.0f; + new_border_color[1] = config.srgb_border_color_g / 255.0f; + new_border_color[2] = config.srgb_border_color_g / 255.0f; + } else { + new_border_color[0] = config.border_color_r; + new_border_color[1] = config.border_color_g; + new_border_color[2] = config.border_color_b; + } + new_border_color[3] = config.border_color_a; + + if (border_color != new_border_color) { + border_color = new_border_color; + glSamplerParameterfv(s, GL_TEXTURE_BORDER_COLOR, border_color.data()); + } + + const float anisotropic_max = static_cast<float>(1 << config.max_anisotropy.Value()); + if (anisotropic_max != max_anisotropic) { + max_anisotropic = anisotropic_max; + if (GLAD_GL_ARB_texture_filter_anisotropic) { + glSamplerParameterf(s, GL_TEXTURE_MAX_ANISOTROPY, max_anisotropic); + } else if (GLAD_GL_EXT_texture_filter_anisotropic) { + glSamplerParameterf(s, GL_TEXTURE_MAX_ANISOTROPY_EXT, max_anisotropic); } } + const float lod_min = static_cast<float>(config.min_lod_clamp.Value()) / 256.0f; + if (lod_min != min_lod) { + min_lod = lod_min; + glSamplerParameterf(s, GL_TEXTURE_MIN_LOD, min_lod); + } + + const float lod_max = static_cast<float>(config.max_lod_clamp.Value()) / 256.0f; + if (lod_max != max_lod) { + max_lod = lod_max; + glSamplerParameterf(s, GL_TEXTURE_MAX_LOD, max_lod); + } + const u32 bias = config.mip_lod_bias.Value(); + // Sign extend the 13-bit value. + 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; + glSamplerParameterf(s, GL_TEXTURE_LOD_BIAS, lod_bias); + } } u32 RasterizerOpenGL::SetupConstBuffers(Maxwell::ShaderStage stage, Shader& shader, @@ -897,14 +967,18 @@ u32 RasterizerOpenGL::SetupTextures(Maxwell::ShaderStage stage, Shader& shader, return current_unit + static_cast<u32>(entries.size()); } -void RasterizerOpenGL::SyncViewport() { +void RasterizerOpenGL::SyncViewport(OpenGLState& current_state) { const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs; - const MathUtil::Rectangle<s32> viewport_rect{regs.viewport_transform[0].GetRect()}; - - state.viewport.x = viewport_rect.left; - state.viewport.y = viewport_rect.bottom; - state.viewport.width = static_cast<GLsizei>(viewport_rect.GetWidth()); - state.viewport.height = static_cast<GLsizei>(viewport_rect.GetHeight()); + for (std::size_t i = 0; i < Tegra::Engines::Maxwell3D::Regs::NumViewports; i++) { + const MathUtil::Rectangle<s32> viewport_rect{regs.viewport_transform[i].GetRect()}; + auto& viewport = current_state.viewports[i]; + viewport.x = viewport_rect.left; + viewport.y = viewport_rect.bottom; + 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; + } } void RasterizerOpenGL::SyncClipEnabled() { @@ -946,13 +1020,6 @@ void RasterizerOpenGL::SyncPrimitiveRestart() { state.primitive_restart.index = regs.primitive_restart.index; } -void RasterizerOpenGL::SyncDepthRange() { - const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs; - - state.depth.depth_range_near = regs.viewport->depth_range_near; - state.depth.depth_range_far = regs.viewport->depth_range_far; -} - void RasterizerOpenGL::SyncDepthTestState() { const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs; @@ -973,9 +1040,6 @@ void RasterizerOpenGL::SyncStencilTestState() { return; } - // TODO(bunnei): Verify behavior when this is not set - ASSERT(regs.stencil_two_side_enable); - state.stencil.front.test_func = MaxwellToGL::ComparisonOp(regs.stencil_front_func_func); state.stencil.front.test_ref = regs.stencil_front_func_ref; state.stencil.front.test_mask = regs.stencil_front_func_mask; @@ -983,42 +1047,95 @@ void RasterizerOpenGL::SyncStencilTestState() { state.stencil.front.action_depth_fail = MaxwellToGL::StencilOp(regs.stencil_front_op_zfail); state.stencil.front.action_depth_pass = MaxwellToGL::StencilOp(regs.stencil_front_op_zpass); state.stencil.front.write_mask = regs.stencil_front_mask; + if (regs.stencil_two_side_enable) { + state.stencil.back.test_func = MaxwellToGL::ComparisonOp(regs.stencil_back_func_func); + state.stencil.back.test_ref = regs.stencil_back_func_ref; + state.stencil.back.test_mask = regs.stencil_back_func_mask; + state.stencil.back.action_stencil_fail = MaxwellToGL::StencilOp(regs.stencil_back_op_fail); + state.stencil.back.action_depth_fail = MaxwellToGL::StencilOp(regs.stencil_back_op_zfail); + state.stencil.back.action_depth_pass = MaxwellToGL::StencilOp(regs.stencil_back_op_zpass); + state.stencil.back.write_mask = regs.stencil_back_mask; + } else { + state.stencil.back.test_func = GL_ALWAYS; + state.stencil.back.test_ref = 0; + state.stencil.back.test_mask = 0xFFFFFFFF; + state.stencil.back.write_mask = 0xFFFFFFFF; + state.stencil.back.action_stencil_fail = GL_KEEP; + state.stencil.back.action_depth_fail = GL_KEEP; + state.stencil.back.action_depth_pass = GL_KEEP; + } +} - state.stencil.back.test_func = MaxwellToGL::ComparisonOp(regs.stencil_back_func_func); - state.stencil.back.test_ref = regs.stencil_back_func_ref; - state.stencil.back.test_mask = regs.stencil_back_func_mask; - state.stencil.back.action_stencil_fail = MaxwellToGL::StencilOp(regs.stencil_back_op_fail); - state.stencil.back.action_depth_fail = MaxwellToGL::StencilOp(regs.stencil_back_op_zfail); - state.stencil.back.action_depth_pass = MaxwellToGL::StencilOp(regs.stencil_back_op_zpass); - state.stencil.back.write_mask = regs.stencil_back_mask; +void RasterizerOpenGL::SyncColorMask() { + const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs; + const std::size_t count = + regs.independent_blend_enable ? Tegra::Engines::Maxwell3D::Regs::NumRenderTargets : 1; + for (std::size_t i = 0; i < count; i++) { + const auto& source = regs.color_mask[regs.color_mask_common ? 0 : i]; + auto& dest = state.color_mask[i]; + dest.red_enabled = (source.R == 0) ? GL_FALSE : GL_TRUE; + dest.green_enabled = (source.G == 0) ? GL_FALSE : GL_TRUE; + dest.blue_enabled = (source.B == 0) ? GL_FALSE : GL_TRUE; + dest.alpha_enabled = (source.A == 0) ? GL_FALSE : GL_TRUE; + } } -void RasterizerOpenGL::SyncBlendState() { +void RasterizerOpenGL::SyncMultiSampleState() { const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs; + state.multisample_control.alpha_to_coverage = regs.multisample_control.alpha_to_coverage != 0; + state.multisample_control.alpha_to_one = regs.multisample_control.alpha_to_one != 0; +} - // TODO(Subv): Support more than just render target 0. - state.blend.enabled = regs.blend.enable[0] != 0; +void RasterizerOpenGL::SyncFragmentColorClampState() { + const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs; + state.fragment_color_clamp.enabled = regs.frag_color_clamp != 0; +} - if (!state.blend.enabled) - return; +void RasterizerOpenGL::SyncBlendState() { + const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs; - ASSERT_MSG(regs.logic_op.enable == 0, - "Blending and logic op can't be enabled at the same time."); + state.blend_color.red = regs.blend_color.r; + state.blend_color.green = regs.blend_color.g; + state.blend_color.blue = regs.blend_color.b; + state.blend_color.alpha = regs.blend_color.a; + + state.independant_blend.enabled = regs.independent_blend_enable; + if (!state.independant_blend.enabled) { + auto& blend = state.blend[0]; + const auto& src = regs.blend; + blend.enabled = src.enable[0] != 0; + if (blend.enabled) { + blend.rgb_equation = MaxwellToGL::BlendEquation(src.equation_rgb); + blend.src_rgb_func = MaxwellToGL::BlendFunc(src.factor_source_rgb); + blend.dst_rgb_func = MaxwellToGL::BlendFunc(src.factor_dest_rgb); + blend.a_equation = MaxwellToGL::BlendEquation(src.equation_a); + blend.src_a_func = MaxwellToGL::BlendFunc(src.factor_source_a); + blend.dst_a_func = MaxwellToGL::BlendFunc(src.factor_dest_a); + } + for (std::size_t i = 1; i < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets; i++) { + state.blend[i].enabled = false; + } + return; + } - ASSERT_MSG(regs.independent_blend_enable == 1, "Only independent blending is implemented"); - ASSERT_MSG(!regs.independent_blend[0].separate_alpha, "Unimplemented"); - state.blend.rgb_equation = MaxwellToGL::BlendEquation(regs.independent_blend[0].equation_rgb); - state.blend.src_rgb_func = MaxwellToGL::BlendFunc(regs.independent_blend[0].factor_source_rgb); - state.blend.dst_rgb_func = MaxwellToGL::BlendFunc(regs.independent_blend[0].factor_dest_rgb); - state.blend.a_equation = MaxwellToGL::BlendEquation(regs.independent_blend[0].equation_a); - state.blend.src_a_func = MaxwellToGL::BlendFunc(regs.independent_blend[0].factor_source_a); - state.blend.dst_a_func = MaxwellToGL::BlendFunc(regs.independent_blend[0].factor_dest_a); + for (std::size_t i = 0; i < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets; i++) { + auto& blend = state.blend[i]; + const auto& src = regs.independent_blend[i]; + blend.enabled = regs.blend.enable[i] != 0; + if (!blend.enabled) + continue; + blend.rgb_equation = MaxwellToGL::BlendEquation(src.equation_rgb); + blend.src_rgb_func = MaxwellToGL::BlendFunc(src.factor_source_rgb); + blend.dst_rgb_func = MaxwellToGL::BlendFunc(src.factor_dest_rgb); + blend.a_equation = MaxwellToGL::BlendEquation(src.equation_a); + blend.src_a_func = MaxwellToGL::BlendFunc(src.factor_source_a); + blend.dst_a_func = MaxwellToGL::BlendFunc(src.factor_dest_a); + } } void RasterizerOpenGL::SyncLogicOpState() { const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs; - // TODO(Subv): Support more than just render target 0. state.logic_op.enabled = regs.logic_op.enable != 0; if (!state.logic_op.enabled) @@ -1030,19 +1147,21 @@ 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; - - state.scissor.enabled = (regs.scissor_test.enable != 0); - // TODO(Blinkhawk): Figure if the hardware supports scissor testing per viewport and how it's - // implemented. - if (regs.scissor_test.enable != 0) { - const u32 width = regs.scissor_test.max_x - regs.scissor_test.min_x; - const u32 height = regs.scissor_test.max_y - regs.scissor_test.min_y; - state.scissor.x = regs.scissor_test.min_x; - state.scissor.y = regs.scissor_test.min_y; - state.scissor.width = width; - state.scissor.height = height; + for (std::size_t i = 0; i < Tegra::Engines::Maxwell3D::Regs::NumViewports; i++) { + const auto& src = regs.scissor_test[i]; + auto& dst = current_state.viewports[i].scissor; + dst.enabled = (src.enable != 0); + if (dst.enabled == 0) { + return; + } + const u32 width = src.max_x - src.min_x; + const u32 height = src.max_y - src.min_y; + dst.x = src.min_x; + dst.y = src.min_y; + dst.width = width; + dst.height = height; } } @@ -1057,20 +1176,15 @@ void RasterizerOpenGL::SyncTransformFeedback() { void RasterizerOpenGL::SyncPointState() { const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs; - - // TODO(Rodrigo): Most games do not set a point size. I think this is a case of a - // register carrying a default value. For now, if the point size is zero, assume it's - // OpenGL's default (1). - state.point.size = regs.point_size == 0 ? 1 : regs.point_size; + state.point.size = regs.point_size; } void RasterizerOpenGL::CheckAlphaTests() { const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs; if (regs.alpha_test_enabled != 0 && regs.rt_control.count > 1) { - LOG_CRITICAL( - Render_OpenGL, - "Alpha Testing is enabled with Multiple Render Targets, this behavior is undefined."); + LOG_CRITICAL(Render_OpenGL, "Alpha Testing is enabled with Multiple Render Targets, " + "this behavior is undefined."); UNREACHABLE(); } } diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index 47097c569..f4354289c 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -88,18 +88,23 @@ private: /// SamplerInfo struct. void Create(); /// Syncs the sampler object with the config, updating any necessary state. - void SyncWithConfig(const Tegra::Texture::TSCEntry& config); + 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; + 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; }; /** @@ -109,8 +114,8 @@ private: * @param preserve_contents If true, tries to preserve data from a previously used framebuffer. * @param single_color_target Specifies if a single color buffer target should be used. */ - void ConfigureFramebuffers(bool use_color_fb = true, bool using_depth_fb = true, - bool preserve_contents = true, + void ConfigureFramebuffers(OpenGLState& current_state, bool use_color_fb = true, + bool using_depth_fb = true, bool preserve_contents = true, std::optional<std::size_t> single_color_target = {}); /* @@ -133,8 +138,8 @@ private: u32 SetupTextures(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage, Shader& shader, GLenum primitive_mode, u32 current_unit); - /// Syncs the viewport to match the guest state - void SyncViewport(); + /// Syncs the viewport and depth range to match the guest state + void SyncViewport(OpenGLState& current_state); /// Syncs the clip enabled status to match the guest state void SyncClipEnabled(); @@ -148,9 +153,6 @@ private: /// Syncs the primitve restart to match the guest state void SyncPrimitiveRestart(); - /// Syncs the depth range to match the guest state - void SyncDepthRange(); - /// Syncs the depth test state to match the guest state void SyncDepthTestState(); @@ -163,8 +165,14 @@ private: /// Syncs the LogicOp state to match the guest state void SyncLogicOpState(); + /// Syncs the the color clamp state + void SyncFragmentColorClampState(); + + /// Syncs the alpha coverage and alpha to one + 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(); @@ -172,13 +180,14 @@ private: /// Syncs the point state to match the guest state void SyncPointState(); + /// Syncs Color Mask + void SyncColorMask(); + /// Check asserts for alpha testing. void CheckAlphaTests(); 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; @@ -207,7 +216,8 @@ private: std::size_t CalculateIndexBufferSize() const; - void SetupVertexArrays(); + void SetupVertexFormat(); + void SetupVertexBuffer(); DrawParameters SetupDraw(); diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp index f194a7687..4f434fc31 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp @@ -15,7 +15,9 @@ #include "core/memory.h" #include "core/settings.h" #include "video_core/engines/maxwell_3d.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" #include "video_core/renderer_opengl/utils.h" #include "video_core/surface.h" #include "video_core/textures/astc.h" @@ -58,16 +60,14 @@ void SurfaceParams::InitCacheParameters(Tegra::GPUVAddr gpu_addr_) { std::size_t SurfaceParams::InnerMipmapMemorySize(u32 mip_level, bool force_gl, bool layer_only, bool uncompressed) const { - const u32 compression_factor{GetCompressionFactor(pixel_format)}; + const u32 tile_x{GetDefaultBlockWidth(pixel_format)}; + const u32 tile_y{GetDefaultBlockHeight(pixel_format)}; const u32 bytes_per_pixel{GetBytesPerPixel(pixel_format)}; u32 m_depth = (layer_only ? 1U : depth); u32 m_width = MipWidth(mip_level); u32 m_height = MipHeight(mip_level); - m_width = uncompressed ? m_width - : std::max(1U, (m_width + compression_factor - 1) / compression_factor); - m_height = uncompressed - ? m_height - : std::max(1U, (m_height + compression_factor - 1) / compression_factor); + m_width = uncompressed ? m_width : std::max(1U, (m_width + tile_x - 1) / tile_x); + m_height = uncompressed ? m_height : std::max(1U, (m_height + tile_y - 1) / tile_y); m_depth = std::max(1U, m_depth >> mip_level); u32 m_block_height = MipBlockHeight(mip_level); u32 m_block_depth = MipBlockDepth(mip_level); @@ -128,6 +128,13 @@ std::size_t SurfaceParams::InnerMemorySize(bool force_gl, bool layer_only, params.target = SurfaceTarget::Texture2D; } break; + case SurfaceTarget::TextureCubeArray: + params.depth = config.tic.Depth() * 6; + if (!entry.IsArray()) { + ASSERT(params.depth == 6); + params.target = SurfaceTarget::TextureCubemap; + } + break; default: LOG_CRITICAL(HW_GPU, "Unknown depth for target={}", static_cast<u32>(params.target)); UNREACHABLE(); @@ -258,11 +265,11 @@ 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 @@ -299,12 +306,16 @@ 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 {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_5X4_SRGB + {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_5X5 + {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_5X5_SRGB + {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_10X8 + {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_10X8_SRGB // Depth formats {GL_DEPTH_COMPONENT32F, GL_DEPTH_COMPONENT, GL_FLOAT, ComponentType::Float, false}, // Z32F @@ -334,6 +345,8 @@ static GLenum SurfaceTargetToGL(SurfaceTarget target) { return GL_TEXTURE_2D_ARRAY; case SurfaceTarget::TextureCubemap: return GL_TEXTURE_CUBE_MAP; + case SurfaceTarget::TextureCubeArray: + return GL_TEXTURE_CUBE_MAP_ARRAY; } LOG_CRITICAL(Render_OpenGL, "Unimplemented texture target={}", static_cast<u32>(target)); UNREACHABLE(); @@ -364,15 +377,15 @@ void MortonCopy(u32 stride, u32 block_height, u32 height, u32 block_depth, u32 d // With the BCn formats (DXT and DXN), each 4x4 tile is swizzled instead of just individual // pixel values. - const u32 tile_size{IsFormatBCn(format) ? 4U : 1U}; + const u32 tile_size_x{GetDefaultBlockWidth(format)}; + const u32 tile_size_y{GetDefaultBlockHeight(format)}; if (morton_to_gl) { - const std::vector<u8> data = Tegra::Texture::UnswizzleTexture( - addr, tile_size, bytes_per_pixel, stride, height, depth, block_height, block_depth); - const std::size_t size_to_copy{std::min(gl_buffer_size, data.size())}; - memcpy(gl_buffer, data.data(), size_to_copy); + 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, height / tile_size, depth, + 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); } @@ -440,6 +453,10 @@ static constexpr GLConversionArray morton_to_gl_fns = { 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>, @@ -508,6 +525,10 @@ static constexpr GLConversionArray gl_to_morton_fns = { nullptr, nullptr, nullptr, + nullptr, + nullptr, + nullptr, + nullptr, MortonCopy<false, PixelFormat::Z32F>, MortonCopy<false, PixelFormat::Z16>, MortonCopy<false, PixelFormat::Z24S8>, @@ -526,8 +547,8 @@ void SwizzleFunc(const GLConversionArray& functions, const SurfaceParams& params if (params.is_layered) { u64 offset = params.GetMipmapLevelOffset(mip_level); u64 offset_gl = 0; - u64 layer_size = params.LayerMemorySize(); - u64 gl_size = params.LayerSizeGL(mip_level); + 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), @@ -537,7 +558,7 @@ void SwizzleFunc(const GLConversionArray& functions, const SurfaceParams& params offset_gl += gl_size; } } else { - u64 offset = params.GetMipmapLevelOffset(mip_level); + 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(), @@ -545,9 +566,11 @@ void SwizzleFunc(const GLConversionArray& functions, const SurfaceParams& params } } +MICROPROFILE_DEFINE(OpenGL_BlitSurface, "OpenGL", "BlitSurface", MP_RGB(128, 192, 64)); static bool BlitSurface(const Surface& src_surface, const Surface& dst_surface, GLuint read_fb_handle, GLuint draw_fb_handle, GLenum src_attachment = 0, GLenum dst_attachment = 0, std::size_t cubemap_face = 0) { + MICROPROFILE_SCOPE(OpenGL_BlitSurface); const auto& src_params{src_surface->GetSurfaceParams()}; const auto& dst_params{dst_surface->GetSurfaceParams()}; @@ -560,7 +583,7 @@ static bool BlitSurface(const Surface& src_surface, const Surface& dst_surface, state.draw.draw_framebuffer = draw_fb_handle; // Set sRGB enabled if the destination surfaces need it state.framebuffer_srgb.enabled = dst_params.srgb_conversion; - state.Apply(); + state.ApplyFramebufferState(); u32 buffers{}; @@ -687,21 +710,23 @@ static void FastCopySurface(const Surface& src_surface, const Surface& dst_surfa 0, 0, width, height, 1); } +MICROPROFILE_DEFINE(OpenGL_CopySurface, "OpenGL", "CopySurface", MP_RGB(128, 192, 64)); static void CopySurface(const Surface& src_surface, const Surface& dst_surface, - GLuint copy_pbo_handle, GLenum src_attachment = 0, - GLenum dst_attachment = 0, std::size_t cubemap_face = 0) { + const GLuint copy_pbo_handle, const GLenum src_attachment = 0, + const GLenum dst_attachment = 0, const std::size_t cubemap_face = 0) { + MICROPROFILE_SCOPE(OpenGL_CopySurface); ASSERT_MSG(dst_attachment == 0, "Unimplemented"); const auto& src_params{src_surface->GetSurfaceParams()}; const auto& dst_params{dst_surface->GetSurfaceParams()}; - auto source_format = GetFormatTuple(src_params.pixel_format, src_params.component_type); - auto dest_format = GetFormatTuple(dst_params.pixel_format, dst_params.component_type); + const auto source_format = GetFormatTuple(src_params.pixel_format, src_params.component_type); + const auto dest_format = GetFormatTuple(dst_params.pixel_format, dst_params.component_type); - std::size_t buffer_size = std::max(src_params.size_in_bytes, dst_params.size_in_bytes); + 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); @@ -722,13 +747,10 @@ static void CopySurface(const Surface& src_surface, const Surface& dst_surface, LOG_DEBUG(HW_GPU, "Trying to upload extra texture data from the CPU during " "reinterpretation but the texture is tiled."); } - std::size_t remaining_size = dst_params.size_in_bytes - src_params.size_in_bytes; - std::vector<u8> data(remaining_size); - std::memcpy(data.data(), Memory::GetPointer(dst_params.addr + src_params.size_in_bytes), - data.size()); + const std::size_t remaining_size = dst_params.size_in_bytes - src_params.size_in_bytes; glBufferSubData(GL_PIXEL_PACK_BUFFER, src_params.size_in_bytes, remaining_size, - data.data()); + Memory::GetPointer(dst_params.addr + src_params.size_in_bytes)); } glBindBuffer(GL_PIXEL_PACK_BUFFER, 0); @@ -754,6 +776,7 @@ static void CopySurface(const Surface& src_surface, const Surface& dst_surface, break; case SurfaceTarget::Texture3D: case SurfaceTarget::Texture2DArray: + case SurfaceTarget::TextureCubeArray: glTextureSubImage3D(dst_surface->Texture().handle, 0, 0, 0, 0, width, height, static_cast<GLsizei>(dst_params.depth), dest_format.format, dest_format.type, nullptr); @@ -806,6 +829,7 @@ CachedSurface::CachedSurface(const SurfaceParams& params) break; case SurfaceTarget::Texture3D: case SurfaceTarget::Texture2DArray: + case SurfaceTarget::TextureCubeArray: glTexStorage3D(SurfaceTargetToGL(params.target), params.max_mip_level, format_tuple.internal_format, rect.GetWidth(), rect.GetHeight(), params.depth); @@ -897,21 +921,26 @@ static void ConvertG8R8ToR8G8(std::vector<u8>& data, u32 width, u32 height) { * typical desktop GPUs. */ static void ConvertFormatAsNeeded_LoadGLBuffer(std::vector<u8>& data, PixelFormat pixel_format, - u32 width, u32 height) { + u32 width, u32 height, u32 depth) { switch (pixel_format) { case PixelFormat::ASTC_2D_4X4: case PixelFormat::ASTC_2D_8X8: case PixelFormat::ASTC_2D_8X5: case PixelFormat::ASTC_2D_5X4: + case PixelFormat::ASTC_2D_5X5: case PixelFormat::ASTC_2D_4X4_SRGB: case PixelFormat::ASTC_2D_8X8_SRGB: case PixelFormat::ASTC_2D_8X5_SRGB: - case PixelFormat::ASTC_2D_5X4_SRGB: { + case PixelFormat::ASTC_2D_5X4_SRGB: + case PixelFormat::ASTC_2D_5X5_SRGB: + case PixelFormat::ASTC_2D_10X8: + case PixelFormat::ASTC_2D_10X8_SRGB: { // Convert ASTC pixel formats to RGBA8, as most desktop GPUs do not support ASTC. u32 block_width{}; u32 block_height{}; std::tie(block_width, block_height) = GetASTCBlockSize(pixel_format); - data = Tegra::Texture::ASTC::Decompress(data, width, height, block_width, block_height); + data = + Tegra::Texture::ASTC::Decompress(data, width, height, depth, block_width, block_height); break; } case PixelFormat::S8Z24: @@ -940,7 +969,11 @@ static void ConvertFormatAsNeeded_FlushGLBuffer(std::vector<u8>& data, PixelForm case PixelFormat::ASTC_2D_4X4: case PixelFormat::ASTC_2D_8X8: case PixelFormat::ASTC_2D_4X4_SRGB: - case PixelFormat::ASTC_2D_8X8_SRGB: { + case PixelFormat::ASTC_2D_8X8_SRGB: + case PixelFormat::ASTC_2D_5X5: + case PixelFormat::ASTC_2D_5X5_SRGB: + case PixelFormat::ASTC_2D_10X8: + case PixelFormat::ASTC_2D_10X8_SRGB: { LOG_CRITICAL(HW_GPU, "Conversion of format {} after texture flushing is not implemented", static_cast<u32>(pixel_format)); UNREACHABLE(); @@ -953,7 +986,7 @@ static void ConvertFormatAsNeeded_FlushGLBuffer(std::vector<u8>& data, PixelForm } } -MICROPROFILE_DEFINE(OpenGL_SurfaceLoad, "OpenGL", "Surface Load", MP_RGB(128, 64, 192)); +MICROPROFILE_DEFINE(OpenGL_SurfaceLoad, "OpenGL", "Surface Load", MP_RGB(128, 192, 64)); void CachedSurface::LoadGLBuffer() { MICROPROFILE_SCOPE(OpenGL_SurfaceLoad); gl_buffer.resize(params.max_mip_level); @@ -971,7 +1004,7 @@ void CachedSurface::LoadGLBuffer() { } for (u32 i = 0; i < params.max_mip_level; i++) ConvertFormatAsNeeded_LoadGLBuffer(gl_buffer[i], params.pixel_format, params.MipWidth(i), - params.MipHeight(i)); + params.MipHeight(i), params.MipDepth(i)); } MICROPROFILE_DEFINE(OpenGL_SurfaceFlush, "OpenGL", "Surface Flush", MP_RGB(128, 192, 64)); @@ -1055,6 +1088,7 @@ void CachedSurface::UploadGLMipmapTexture(u32 mip_map, GLuint read_fb_handle, &gl_buffer[mip_map][buffer_offset]); break; case SurfaceTarget::Texture2DArray: + case SurfaceTarget::TextureCubeArray: glCompressedTexImage3D(SurfaceTargetToGL(params.target), mip_map, tuple.internal_format, static_cast<GLsizei>(params.MipWidth(mip_map)), static_cast<GLsizei>(params.MipHeight(mip_map)), @@ -1104,6 +1138,7 @@ void CachedSurface::UploadGLMipmapTexture(u32 mip_map, GLuint read_fb_handle, tuple.format, tuple.type, &gl_buffer[mip_map][buffer_offset]); break; case SurfaceTarget::Texture2DArray: + case SurfaceTarget::TextureCubeArray: glTexSubImage3D(SurfaceTargetToGL(params.target), mip_map, x0, y0, 0, static_cast<GLsizei>(rect.GetWidth()), static_cast<GLsizei>(rect.GetHeight()), params.depth, tuple.format, @@ -1133,7 +1168,7 @@ void CachedSurface::UploadGLMipmapTexture(u32 mip_map, GLuint read_fb_handle, glPixelStorei(GL_UNPACK_ROW_LENGTH, 0); } -MICROPROFILE_DEFINE(OpenGL_TextureUL, "OpenGL", "Texture Upload", MP_RGB(128, 64, 192)); +MICROPROFILE_DEFINE(OpenGL_TextureUL, "OpenGL", "Texture Upload", MP_RGB(128, 192, 64)); void CachedSurface::UploadGLTexture(GLuint read_fb_handle, GLuint draw_fb_handle) { if (params.type == SurfaceType::Fill) return; @@ -1144,7 +1179,8 @@ void CachedSurface::UploadGLTexture(GLuint read_fb_handle, GLuint draw_fb_handle UploadGLMipmapTexture(i, read_fb_handle, draw_fb_handle); } -RasterizerCacheOpenGL::RasterizerCacheOpenGL() { +RasterizerCacheOpenGL::RasterizerCacheOpenGL(RasterizerOpenGL& rasterizer) + : RasterizerCache{rasterizer} { read_framebuffer.Create(); draw_framebuffer.Create(); copy_pbo.Create(); @@ -1239,6 +1275,31 @@ 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) { @@ -1304,10 +1365,14 @@ 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: + FastLayeredCopySurface(old_surface, new_surface); + break; default: LOG_CRITICAL(Render_OpenGL, "Unimplemented surface target={}", static_cast<u32>(new_params.target)); diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.h b/src/video_core/renderer_opengl/gl_rasterizer_cache.h index f255f4419..9ac79c5a4 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.h +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.h @@ -49,6 +49,8 @@ struct SurfaceParams { return "Texture2DArray"; case SurfaceTarget::TextureCubemap: return "TextureCubemap"; + case SurfaceTarget::TextureCubeArray: + return "TextureCubeArray"; default: LOG_CRITICAL(HW_GPU, "Unimplemented surface_target={}", static_cast<u32>(target)); UNREACHABLE(); @@ -139,7 +141,7 @@ struct SurfaceParams { } u32 MipDepth(u32 mip_level) const { - return std::max(1U, depth >> mip_level); + return is_layered ? depth : std::max(1U, depth >> mip_level); } // Auto block resizing algorithm from: @@ -262,6 +264,8 @@ struct hash<SurfaceReserveKey> { namespace OpenGL { +class RasterizerOpenGL; + class CachedSurface final : public RasterizerCacheObject { public: CachedSurface(const SurfaceParams& params); @@ -309,7 +313,7 @@ private: class RasterizerCacheOpenGL final : public RasterizerCache<Surface> { public: - RasterizerCacheOpenGL(); + explicit RasterizerCacheOpenGL(RasterizerOpenGL& rasterizer); /// Get a surface based on the texture configuration Surface GetTextureSurface(const Tegra::Texture::FullTextureInfo& config, @@ -346,6 +350,7 @@ 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_resource_manager.cpp b/src/video_core/renderer_opengl/gl_resource_manager.cpp new file mode 100644 index 000000000..c17d5ac00 --- /dev/null +++ b/src/video_core/renderer_opengl/gl_resource_manager.cpp @@ -0,0 +1,186 @@ +// Copyright 2015 Citra Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <utility> +#include <glad/glad.h> +#include "common/common_types.h" +#include "common/microprofile.h" +#include "video_core/renderer_opengl/gl_resource_manager.h" +#include "video_core/renderer_opengl/gl_shader_util.h" +#include "video_core/renderer_opengl/gl_state.h" + +MICROPROFILE_DEFINE(OpenGL_ResourceCreation, "OpenGL", "Resource Creation", MP_RGB(128, 128, 192)); +MICROPROFILE_DEFINE(OpenGL_ResourceDeletion, "OpenGL", "Resource Deletion", MP_RGB(128, 128, 192)); + +namespace OpenGL { + +void OGLTexture::Create() { + if (handle != 0) + return; + + MICROPROFILE_SCOPE(OpenGL_ResourceCreation); + glGenTextures(1, &handle); +} + +void OGLTexture::Release() { + if (handle == 0) + return; + + MICROPROFILE_SCOPE(OpenGL_ResourceDeletion); + glDeleteTextures(1, &handle); + OpenGLState::GetCurState().UnbindTexture(handle).Apply(); + handle = 0; +} + +void OGLSampler::Create() { + if (handle != 0) + return; + + MICROPROFILE_SCOPE(OpenGL_ResourceCreation); + glGenSamplers(1, &handle); +} + +void OGLSampler::Release() { + if (handle == 0) + return; + + MICROPROFILE_SCOPE(OpenGL_ResourceDeletion); + glDeleteSamplers(1, &handle); + OpenGLState::GetCurState().ResetSampler(handle).Apply(); + handle = 0; +} + +void OGLShader::Create(const char* source, GLenum type) { + if (handle != 0) + return; + if (source == nullptr) + return; + + MICROPROFILE_SCOPE(OpenGL_ResourceCreation); + handle = GLShader::LoadShader(source, type); +} + +void OGLShader::Release() { + if (handle == 0) + return; + + MICROPROFILE_SCOPE(OpenGL_ResourceDeletion); + glDeleteShader(handle); + handle = 0; +} + +void OGLProgram::CreateFromSource(const char* vert_shader, const char* geo_shader, + const char* frag_shader, bool separable_program) { + OGLShader vert, geo, frag; + if (vert_shader) + vert.Create(vert_shader, GL_VERTEX_SHADER); + if (geo_shader) + geo.Create(geo_shader, GL_GEOMETRY_SHADER); + if (frag_shader) + frag.Create(frag_shader, GL_FRAGMENT_SHADER); + + MICROPROFILE_SCOPE(OpenGL_ResourceCreation); + Create(separable_program, vert.handle, geo.handle, frag.handle); +} + +void OGLProgram::Release() { + if (handle == 0) + return; + + MICROPROFILE_SCOPE(OpenGL_ResourceDeletion); + glDeleteProgram(handle); + OpenGLState::GetCurState().ResetProgram(handle).Apply(); + handle = 0; +} + +void OGLPipeline::Create() { + if (handle != 0) + return; + + MICROPROFILE_SCOPE(OpenGL_ResourceCreation); + glGenProgramPipelines(1, &handle); +} + +void OGLPipeline::Release() { + if (handle == 0) + return; + + MICROPROFILE_SCOPE(OpenGL_ResourceDeletion); + glDeleteProgramPipelines(1, &handle); + OpenGLState::GetCurState().ResetPipeline(handle).Apply(); + handle = 0; +} + +void OGLBuffer::Create() { + if (handle != 0) + return; + + MICROPROFILE_SCOPE(OpenGL_ResourceCreation); + glGenBuffers(1, &handle); +} + +void OGLBuffer::Release() { + if (handle == 0) + return; + + MICROPROFILE_SCOPE(OpenGL_ResourceDeletion); + glDeleteBuffers(1, &handle); + OpenGLState::GetCurState().ResetBuffer(handle).Apply(); + handle = 0; +} + +void OGLSync::Create() { + if (handle != 0) + return; + + // Don't profile here, this one is expected to happen ingame. + handle = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0); +} + +void OGLSync::Release() { + if (handle == 0) + return; + + // Don't profile here, this one is expected to happen ingame. + glDeleteSync(handle); + handle = 0; +} + +void OGLVertexArray::Create() { + if (handle != 0) + return; + + MICROPROFILE_SCOPE(OpenGL_ResourceCreation); + glGenVertexArrays(1, &handle); +} + +void OGLVertexArray::Release() { + if (handle == 0) + return; + + MICROPROFILE_SCOPE(OpenGL_ResourceDeletion); + glDeleteVertexArrays(1, &handle); + OpenGLState::GetCurState().ResetVertexArray(handle).Apply(); + handle = 0; +} + +void OGLFramebuffer::Create() { + if (handle != 0) + return; + + MICROPROFILE_SCOPE(OpenGL_ResourceCreation); + glGenFramebuffers(1, &handle); +} + +void OGLFramebuffer::Release() { + if (handle == 0) + return; + + MICROPROFILE_SCOPE(OpenGL_ResourceDeletion); + glDeleteFramebuffers(1, &handle); + OpenGLState::GetCurState().ResetFramebuffer(handle).Apply(); + handle = 0; +} + +} // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_resource_manager.h b/src/video_core/renderer_opengl/gl_resource_manager.h index 3bc1b83b5..e33f1e973 100644 --- a/src/video_core/renderer_opengl/gl_resource_manager.h +++ b/src/video_core/renderer_opengl/gl_resource_manager.h @@ -8,7 +8,6 @@ #include <glad/glad.h> #include "common/common_types.h" #include "video_core/renderer_opengl/gl_shader_util.h" -#include "video_core/renderer_opengl/gl_state.h" namespace OpenGL { @@ -29,20 +28,10 @@ public: } /// Creates a new internal OpenGL resource and stores the handle - void Create() { - if (handle != 0) - return; - glGenTextures(1, &handle); - } + void Create(); /// Deletes the internal OpenGL resource - void Release() { - if (handle == 0) - return; - glDeleteTextures(1, &handle); - OpenGLState::GetCurState().UnbindTexture(handle).Apply(); - handle = 0; - } + void Release(); GLuint handle = 0; }; @@ -64,20 +53,10 @@ public: } /// Creates a new internal OpenGL resource and stores the handle - void Create() { - if (handle != 0) - return; - glGenSamplers(1, &handle); - } + void Create(); /// Deletes the internal OpenGL resource - void Release() { - if (handle == 0) - return; - glDeleteSamplers(1, &handle); - OpenGLState::GetCurState().ResetSampler(handle).Apply(); - handle = 0; - } + void Release(); GLuint handle = 0; }; @@ -98,20 +77,9 @@ public: return *this; } - void Create(const char* source, GLenum type) { - if (handle != 0) - return; - if (source == nullptr) - return; - handle = GLShader::LoadShader(source, type); - } + void Create(const char* source, GLenum type); - void Release() { - if (handle == 0) - return; - glDeleteShader(handle); - handle = 0; - } + void Release(); GLuint handle = 0; }; @@ -141,25 +109,10 @@ public: /// Creates a new internal OpenGL resource and stores the handle void CreateFromSource(const char* vert_shader, const char* geo_shader, const char* frag_shader, - bool separable_program = false) { - OGLShader vert, geo, frag; - if (vert_shader) - vert.Create(vert_shader, GL_VERTEX_SHADER); - if (geo_shader) - geo.Create(geo_shader, GL_GEOMETRY_SHADER); - if (frag_shader) - frag.Create(frag_shader, GL_FRAGMENT_SHADER); - Create(separable_program, vert.handle, geo.handle, frag.handle); - } + bool separable_program = false); /// Deletes the internal OpenGL resource - void Release() { - if (handle == 0) - return; - glDeleteProgram(handle); - OpenGLState::GetCurState().ResetProgram(handle).Apply(); - handle = 0; - } + void Release(); GLuint handle = 0; }; @@ -178,20 +131,10 @@ public: } /// Creates a new internal OpenGL resource and stores the handle - void Create() { - if (handle != 0) - return; - glGenProgramPipelines(1, &handle); - } + void Create(); /// Deletes the internal OpenGL resource - void Release() { - if (handle == 0) - return; - glDeleteProgramPipelines(1, &handle); - OpenGLState::GetCurState().ResetPipeline(handle).Apply(); - handle = 0; - } + void Release(); GLuint handle = 0; }; @@ -213,20 +156,10 @@ public: } /// Creates a new internal OpenGL resource and stores the handle - void Create() { - if (handle != 0) - return; - glGenBuffers(1, &handle); - } + void Create(); /// Deletes the internal OpenGL resource - void Release() { - if (handle == 0) - return; - glDeleteBuffers(1, &handle); - OpenGLState::GetCurState().ResetBuffer(handle).Apply(); - handle = 0; - } + void Release(); GLuint handle = 0; }; @@ -247,19 +180,10 @@ public: } /// Creates a new internal OpenGL resource and stores the handle - void Create() { - if (handle != 0) - return; - handle = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0); - } + void Create(); /// Deletes the internal OpenGL resource - void Release() { - if (handle == 0) - return; - glDeleteSync(handle); - handle = 0; - } + void Release(); GLsync handle = 0; }; @@ -281,20 +205,10 @@ public: } /// Creates a new internal OpenGL resource and stores the handle - void Create() { - if (handle != 0) - return; - glGenVertexArrays(1, &handle); - } + void Create(); /// Deletes the internal OpenGL resource - void Release() { - if (handle == 0) - return; - glDeleteVertexArrays(1, &handle); - OpenGLState::GetCurState().ResetVertexArray(handle).Apply(); - handle = 0; - } + void Release(); GLuint handle = 0; }; @@ -316,20 +230,10 @@ public: } /// Creates a new internal OpenGL resource and stores the handle - void Create() { - if (handle != 0) - return; - glGenFramebuffers(1, &handle); - } + void Create(); /// Deletes the internal OpenGL resource - void Release() { - if (handle == 0) - return; - glDeleteFramebuffers(1, &handle); - OpenGLState::GetCurState().ResetFramebuffer(handle).Apply(); - handle = 0; - } + void Release(); GLuint handle = 0; }; diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp index 9522fd344..038b25c75 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp @@ -6,10 +6,10 @@ #include "core/core.h" #include "core/memory.h" #include "video_core/engines/maxwell_3d.h" +#include "video_core/renderer_opengl/gl_rasterizer.h" #include "video_core/renderer_opengl/gl_shader_cache.h" #include "video_core/renderer_opengl/gl_shader_manager.h" #include "video_core/renderer_opengl/utils.h" -#include "video_core/utils.h" namespace OpenGL { @@ -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; @@ -121,12 +122,16 @@ GLint CachedShader::GetUniformLocation(const GLShader::SamplerEntry& sampler) { } GLuint CachedShader::LazyGeometryProgram(OGLProgram& target_program, - const std::string& glsl_topology, + const std::string& glsl_topology, u32 max_vertices, const std::string& debug_name) { if (target_program.handle != 0) { return target_program.handle; } - const std::string source{geometry_programs.code + "layout (" + glsl_topology + ") in;\n"}; + std::string source = "#version 430 core\n"; + source += "layout (" + glsl_topology + ") in;\n"; + source += "#define MAX_VERTEX_INPUT " + std::to_string(max_vertices) + '\n'; + source += geometry_programs.code; + OGLShader shader; shader.Create(source.c_str(), GL_GEOMETRY_SHADER); target_program.Create(true, shader.handle); @@ -135,6 +140,8 @@ GLuint CachedShader::LazyGeometryProgram(OGLProgram& target_program, return target_program.handle; }; +ShaderCacheOpenGL::ShaderCacheOpenGL(RasterizerOpenGL& rasterizer) : RasterizerCache{rasterizer} {} + Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) { const VAddr program_addr{GetShaderAddress(program)}; diff --git a/src/video_core/renderer_opengl/gl_shader_cache.h b/src/video_core/renderer_opengl/gl_shader_cache.h index a210f1731..08f470de3 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.h +++ b/src/video_core/renderer_opengl/gl_shader_cache.h @@ -16,6 +16,8 @@ namespace OpenGL { class CachedShader; +class RasterizerOpenGL; + using Shader = std::shared_ptr<CachedShader>; using Maxwell = Tegra::Engines::Maxwell3D::Regs; @@ -28,7 +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. @@ -46,22 +48,23 @@ public: } switch (primitive_mode) { case GL_POINTS: - return LazyGeometryProgram(geometry_programs.points, "points", "ShaderPoints"); + return LazyGeometryProgram(geometry_programs.points, "points", 1, "ShaderPoints"); case GL_LINES: case GL_LINE_STRIP: - return LazyGeometryProgram(geometry_programs.lines, "lines", "ShaderLines"); + return LazyGeometryProgram(geometry_programs.lines, "lines", 2, "ShaderLines"); case GL_LINES_ADJACENCY: case GL_LINE_STRIP_ADJACENCY: - return LazyGeometryProgram(geometry_programs.lines_adjacency, "lines_adjacency", + return LazyGeometryProgram(geometry_programs.lines_adjacency, "lines_adjacency", 4, "ShaderLinesAdjacency"); case GL_TRIANGLES: case GL_TRIANGLE_STRIP: case GL_TRIANGLE_FAN: - return LazyGeometryProgram(geometry_programs.triangles, "triangles", "ShaderTriangles"); + return LazyGeometryProgram(geometry_programs.triangles, "triangles", 3, + "ShaderTriangles"); case GL_TRIANGLES_ADJACENCY: case GL_TRIANGLE_STRIP_ADJACENCY: return LazyGeometryProgram(geometry_programs.triangles_adjacency, "triangles_adjacency", - "ShaderLines"); + 6, "ShaderTrianglesAdjacency"); default: UNREACHABLE_MSG("Unknown primitive mode."); } @@ -76,9 +79,10 @@ public: private: /// Generates a geometry shader or returns one that already exists. GLuint LazyGeometryProgram(OGLProgram& target_program, const std::string& glsl_topology, - const std::string& debug_name); + 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; @@ -104,6 +108,8 @@ private: class ShaderCacheOpenGL final : public RasterizerCache<Shader> { public: + explicit ShaderCacheOpenGL(RasterizerOpenGL& rasterizer); + /// Gets the current specified shader stage program Shader GetStageProgram(Maxwell::ShaderProgram program); }; diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index 09b003c59..97b9028c5 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; @@ -49,8 +60,7 @@ static std::string GetTopologyName(Tegra::Shader::OutputTopology topology) { case Tegra::Shader::OutputTopology::TriangleStrip: return "triangle_strip"; default: - LOG_CRITICAL(Render_OpenGL, "Unknown output topology {}", static_cast<u32>(topology)); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unknown output topology: {}", static_cast<u32>(topology)); return "points"; } } @@ -85,7 +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); @@ -97,10 +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) { @@ -142,6 +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()) { @@ -167,8 +187,8 @@ private: case OpCode::Id::SSY: case OpCode::Id::PBK: { // The SSY and PBK use a similar encoding as the BRA instruction. - ASSERT_MSG(instr.bra.constant_buffer == 0, - "Constant buffer branching is not supported"); + UNIMPLEMENTED_IF_MSG(instr.bra.constant_buffer != 0, + "Constant buffer branching is not supported"); const u32 target = offset + instr.bra.GetBranchTarget(); labels.insert(target); // Continue scanning for an exit method. @@ -258,14 +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 @@ -299,8 +311,7 @@ public: // Default - do nothing return value; default: - LOG_CRITICAL(HW_GPU, "Unimplemented conversion size {}", static_cast<u32>(size)); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unimplemented conversion size: {}", static_cast<u32>(size)); } } @@ -363,7 +374,7 @@ public: u64 value_num_components, bool is_saturated = false, u64 dest_elem = 0, Register::Size size = Register::Size::Word, bool sets_cc = false) { - ASSERT_MSG(!is_saturated, "Unimplemented"); + UNIMPLEMENTED_IF(is_saturated); const std::string func{is_signed ? "intBitsToFloat" : "uintBitsToFloat"}; @@ -373,7 +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."); } } @@ -392,7 +403,7 @@ public: Tegra::Shader::HalfMerge merge, u64 dest_num_components, u64 value_num_components, bool is_saturated = false, u64 dest_elem = 0) { - ASSERT_MSG(!is_saturated, "Unimplemented"); + UNIMPLEMENTED_IF(is_saturated); const std::string result = [&]() { switch (merge) { @@ -456,24 +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: - LOG_CRITICAL(HW_GPU, "Unimplemented Control Code {}", static_cast<u32>(cc)); - UNREACHABLE(); + 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 + ';'); } /** @@ -494,10 +506,10 @@ public: // 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 for the out of boundary writes, guard them. - const std::string buf_index{"min(" + GetRegisterAsInteger(buf_reg) + ", " + - std::to_string(MAX_GEOMETRY_BUFFERS - 1) + ')'}; + // 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 + ';'); @@ -624,8 +636,8 @@ 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(); @@ -761,8 +773,7 @@ private: u64 dest_num_components, u64 value_num_components, u64 dest_elem, bool precise) { if (reg == Register::ZeroIndex) { - LOG_CRITICAL(HW_GPU, "Cannot set Register::ZeroIndex"); - UNREACHABLE(); + // Setting RZ is a nop in hardware. return; } @@ -811,7 +822,11 @@ private: std::optional<Register> vertex = {}) { auto GeometryPass = [&](const std::string& name) { if (stage == Maxwell3D::Regs::ShaderStage::Geometry && vertex) { - return "gs_" + name + '[' + GetRegisterAsInteger(*vertex, 0, false) + ']'; + // TODO(Rodrigo): Guard geometry inputs against out of bound reads. Some games set + // an 0x80000000 index for those and the shader fails to build. Find out why this + // happens and what's its intent. + return "gs_" + name + '[' + GetRegisterAsInteger(*vertex, 0, false) + + " % MAX_VERTEX_INPUT]"; } return name; }; @@ -843,16 +858,13 @@ private: if (declr_input_attribute.count(attribute) == 0) { declr_input_attribute[attribute] = input_mode; } else { - if (declr_input_attribute[attribute] != input_mode) { - LOG_CRITICAL(HW_GPU, "Same Input multiple input modes"); - UNREACHABLE(); - } + UNIMPLEMENTED_IF_MSG(declr_input_attribute[attribute] != input_mode, + "Multiple input modes for the same attribute"); } return GeometryPass("input_attribute_" + std::to_string(index)); } - LOG_CRITICAL(HW_GPU, "Unhandled input attribute: {}", static_cast<u32>(attribute)); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled input attribute: {}", static_cast<u32>(attribute)); } return "vec4(0, 0, 0, 0)"; @@ -878,24 +890,20 @@ private: break; } default: { - LOG_CRITICAL(HW_GPU, "Unhandled Ipa InterpMode: {}", static_cast<u32>(interp_mode)); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled IPA interp mode: {}", static_cast<u32>(interp_mode)); } } switch (sample_mode) { - case Tegra::Shader::IpaSampleMode::Centroid: { - // Note not implemented, it can be implemented with the "centroid " keyword in glsl; - LOG_CRITICAL(HW_GPU, "Ipa Sampler Mode: centroid, not implemented"); - UNREACHABLE(); + case Tegra::Shader::IpaSampleMode::Centroid: + // It can be implemented with the "centroid " keyword in glsl + UNIMPLEMENTED_MSG("Unimplemented IPA sampler mode centroid"); break; - } - case Tegra::Shader::IpaSampleMode::Default: { + case Tegra::Shader::IpaSampleMode::Default: // Default, n/a break; - } default: { - LOG_CRITICAL(HW_GPU, "Unhandled Ipa SampleMode: {}", static_cast<u32>(sample_mode)); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unimplemented IPA sampler mode: {}", static_cast<u32>(sample_mode)); + break; } } return out; @@ -916,8 +924,7 @@ private: return "output_attribute_" + std::to_string(index); } - LOG_CRITICAL(HW_GPU, "Unhandled output attribute: {}", index); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled output attribute={}", index); return {}; } } @@ -947,9 +954,10 @@ 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); @@ -962,7 +970,7 @@ 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: @@ -1067,19 +1075,26 @@ 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)}; - ASSERT_MSG(comparison != PredicateComparisonStrings.end(), - "Unknown predicate comparison operation"); + UNIMPLEMENTED_IF_MSG(comparison == PredicateComparisonStrings.end(), + "Unknown predicate comparison operation"); 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 + ')'; @@ -1103,7 +1118,7 @@ private: }; auto op = PredicateOperationStrings.find(operation); - ASSERT_MSG(op != PredicateOperationStrings.end(), "Unknown predicate operation"); + UNIMPLEMENTED_IF_MSG(op == PredicateOperationStrings.end(), "Unknown predicate operation"); return op->second; } @@ -1201,8 +1216,7 @@ private: break; } default: - LOG_CRITICAL(HW_GPU, "Unimplemented logic operation: {}", static_cast<u32>(logic_op)); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unimplemented logic operation={}", static_cast<u32>(logic_op)); } if (dest != Tegra::Shader::Register::ZeroIndex) { @@ -1220,9 +1234,8 @@ private: SetPredicate(static_cast<u64>(predicate), '(' + result + ") != 0"); break; default: - LOG_CRITICAL(HW_GPU, "Unimplemented predicate result mode: {}", - static_cast<u32>(predicate_mode)); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unimplemented predicate result mode: {}", + static_cast<u32>(predicate_mode)); } } @@ -1253,14 +1266,7 @@ private: regs.SetRegisterToInteger(dest, true, 0, result, 1, 1); } - void WriteTexsInstruction(const Instruction& instr, const std::string& coord, - const std::string& texture) { - // 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); - + void WriteTexsInstruction(const Instruction& instr, const std::string& 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 @@ -1283,26 +1289,19 @@ private: ++written_components; } - - --shader.scope; - shader.AddLine('}'); } static u32 TextureCoordinates(Tegra::Shader::TextureType texture_type) { switch (texture_type) { - case Tegra::Shader::TextureType::Texture1D: { + case Tegra::Shader::TextureType::Texture1D: return 1; - } - case Tegra::Shader::TextureType::Texture2D: { + case Tegra::Shader::TextureType::Texture2D: return 2; - } case Tegra::Shader::TextureType::Texture3D: - case Tegra::Shader::TextureType::TextureCube: { + case Tegra::Shader::TextureType::TextureCube: return 3; - } default: - LOG_CRITICAL(HW_GPU, "Unhandled texture type {}", static_cast<u32>(texture_type)); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled texture type: {}", static_cast<u32>(texture_type)); return 0; } } @@ -1338,7 +1337,7 @@ private: void EmitFragmentOutputsWrite() { ASSERT(stage == Maxwell3D::Regs::ShaderStage::Fragment); - ASSERT_MSG(header.ps.omap.sample_mask == 0, "Samplemask write is unimplemented"); + UNIMPLEMENTED_IF_MSG(header.ps.omap.sample_mask != 0, "Samplemask write is unimplemented"); shader.AddLine("if (alpha_test[0] != 0) {"); ++shader.scope; @@ -1404,7 +1403,7 @@ private: case Tegra::Shader::VideoType::Size32: // TODO(Rodrigo): From my hardware tests it becomes a bit "mad" when // this type is used (1 * 1 + 0 == 0x5b800000). Until a better - // explanation is found: assert. + // explanation is found: abort. UNIMPLEMENTED(); return zero; case Tegra::Shader::VideoType::Invalid: @@ -1460,8 +1459,7 @@ private: // Decoding failure if (!opcode) { - LOG_CRITICAL(HW_GPU, "Unhandled instruction: {0:x}", instr.value); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled instruction: {0:x}", instr.value); return offset + 1; } @@ -1469,8 +1467,8 @@ private: fmt::format("// {}: {} (0x{:016x})", offset, opcode->get().GetName(), instr.value)); using Tegra::Shader::Pred; - ASSERT_MSG(instr.pred.full_pred != Pred::NeverExecute, - "NeverExecute predicate not implemented"); + UNIMPLEMENTED_IF_MSG(instr.pred.full_pred == Pred::NeverExecute, + "NeverExecute predicate not implemented"); // Some instructions (like SSY) don't have a predicate field, they are always // unconditionally executed. @@ -1513,37 +1511,36 @@ private: case OpCode::Id::FMUL_R: case OpCode::Id::FMUL_IMM: { // FMUL does not have 'abs' bits and only the second operand has a 'neg' bit. - ASSERT_MSG(instr.fmul.tab5cb8_2 == 0, "FMUL tab5cb8_2({}) is not implemented", - instr.fmul.tab5cb8_2.Value()); - ASSERT_MSG(instr.fmul.tab5c68_1 == 0, "FMUL tab5cb8_1({}) is not implemented", - instr.fmul.tab5c68_1.Value()); - ASSERT_MSG(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 - ASSERT_MSG(instr.fmul.cc == 0, "FMUL cc is not implemented"); + UNIMPLEMENTED_IF_MSG(instr.fmul.tab5cb8_2 != 0, + "FMUL tab5cb8_2({}) is not implemented", + instr.fmul.tab5cb8_2.Value()); + UNIMPLEMENTED_IF_MSG(instr.fmul.tab5c68_1 != 0, + "FMUL tab5cb8_1({}) is not implemented", + instr.fmul.tab5c68_1.Value()); + UNIMPLEMENTED_IF_MSG( + 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.generates_cc, + "Condition codes generation in FMUL is not implemented"); op_b = GetOperandAbsNeg(op_b, false, instr.fmul.negate_b); regs.SetRegisterToFloat(instr.gpr0, 0, op_a + " * " + op_b, 1, 1, instr.alu.saturate_d, 0, true); - if (instr.generates_cc) { - LOG_CRITICAL(HW_GPU, "FMUL Generates an unhandled Control Code"); - UNREACHABLE(); - } break; } case OpCode::Id::FADD_C: case OpCode::Id::FADD_R: case OpCode::Id::FADD_IMM: { + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "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); regs.SetRegisterToFloat(instr.gpr0, 0, op_a + " + " + op_b, 1, 1, instr.alu.saturate_d, 0, true); - if (instr.generates_cc) { - LOG_CRITICAL(HW_GPU, "FADD Generates an unhandled Control Code"); - UNREACHABLE(); - } break; } case OpCode::Id::MUFU: { @@ -1578,15 +1575,17 @@ private: instr.alu.saturate_d, 0, true); break; default: - LOG_CRITICAL(HW_GPU, "Unhandled MUFU sub op: {0:x}", - static_cast<unsigned>(instr.sub_op.Value())); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled MUFU sub op={0:x}", + static_cast<unsigned>(instr.sub_op.Value())); } break; } case OpCode::Id::FMNMX_C: case OpCode::Id::FMNMX_R: case OpCode::Id::FMNMX_IMM: { + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "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); @@ -1597,10 +1596,6 @@ private: '(' + condition + ") ? min(" + parameters + ") : max(" + parameters + ')', 1, 1, false, 0, true); - if (instr.generates_cc) { - LOG_CRITICAL(HW_GPU, "FMNMX Generates an unhandled Control Code"); - UNREACHABLE(); - } break; } case OpCode::Id::RRO_C: @@ -1613,9 +1608,7 @@ private: break; } default: { - LOG_CRITICAL(HW_GPU, "Unhandled arithmetic instruction: {}", - opcode->get().GetName()); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled arithmetic instruction: {}", opcode->get().GetName()); } } break; @@ -1627,17 +1620,19 @@ private: break; } case OpCode::Id::FMUL32_IMM: { + UNIMPLEMENTED_IF_MSG(instr.op_32.generates_cc, + "Condition codes generation in FMUL32 is not implemented"); + regs.SetRegisterToFloat(instr.gpr0, 0, regs.GetRegisterAsFloat(instr.gpr8) + " * " + GetImmediate32(instr), 1, 1, instr.fmul32.saturate, 0, true); - if (instr.op_32.generates_cc) { - LOG_CRITICAL(HW_GPU, "FMUL32 Generates an unhandled Control Code"); - UNREACHABLE(); - } break; } case OpCode::Id::FADD32I: { + UNIMPLEMENTED_IF_MSG(instr.op_32.generates_cc, + "Condition codes generation in FADD32I is not implemented"); + std::string op_a = regs.GetRegisterAsFloat(instr.gpr8); std::string op_b = GetImmediate32(instr); @@ -1658,23 +1653,22 @@ private: } regs.SetRegisterToFloat(instr.gpr0, 0, op_a + " + " + op_b, 1, 1, false, 0, true); - if (instr.op_32.generates_cc) { - LOG_CRITICAL(HW_GPU, "FADD32 Generates an unhandled Control Code"); - UNREACHABLE(); - } break; } } break; } case OpCode::Type::Bfe: { - ASSERT_MSG(!instr.bfe.negate_b, "Unimplemented"); + UNIMPLEMENTED_IF(instr.bfe.negate_b); std::string op_a = instr.bfe.negate_a ? "-" : ""; op_a += regs.GetRegisterAsInteger(instr.gpr8); switch (opcode->get().GetId()) { case OpCode::Id::BFE_IMM: { + 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()) + ')'; std::string outer_shift = @@ -1682,15 +1676,10 @@ private: std::to_string(instr.bfe.GetLeftShiftValue() + instr.bfe.shift_position) + ')'; regs.SetRegisterToInteger(instr.gpr0, true, 0, outer_shift, 1, 1); - if (instr.generates_cc) { - LOG_CRITICAL(HW_GPU, "BFE Generates an unhandled Control Code"); - UNREACHABLE(); - } break; } default: { - LOG_CRITICAL(HW_GPU, "Unhandled BFE instruction: {}", opcode->get().GetName()); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled BFE instruction: {}", opcode->get().GetName()); } } @@ -1715,6 +1704,9 @@ private: case OpCode::Id::SHR_C: case OpCode::Id::SHR_R: case OpCode::Id::SHR_IMM: { + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "Condition codes generation in SHR is not implemented"); + if (!instr.shift.is_signed) { // Logical shift right op_a = "uint(" + op_a + ')'; @@ -1723,24 +1715,17 @@ private: // Cast to int is superfluous for arithmetic shift, it's only for a logical shift regs.SetRegisterToInteger(instr.gpr0, true, 0, "int(" + op_a + " >> " + op_b + ')', 1, 1); - if (instr.generates_cc) { - LOG_CRITICAL(HW_GPU, "SHR Generates an unhandled Control Code"); - UNREACHABLE(); - } break; } case OpCode::Id::SHL_C: case OpCode::Id::SHL_R: case OpCode::Id::SHL_IMM: + 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); - if (instr.generates_cc) { - LOG_CRITICAL(HW_GPU, "SHL Generates an unhandled Control Code"); - UNREACHABLE(); - } break; default: { - LOG_CRITICAL(HW_GPU, "Unhandled shift instruction: {}", opcode->get().GetName()); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled shift instruction: {}", opcode->get().GetName()); } } break; @@ -1751,17 +1736,19 @@ private: switch (opcode->get().GetId()) { case OpCode::Id::IADD32I: + UNIMPLEMENTED_IF_MSG(instr.op_32.generates_cc, + "Condition codes generation in IADD32I is not implemented"); + if (instr.iadd32i.negate_a) op_a = "-(" + op_a + ')'; regs.SetRegisterToInteger(instr.gpr0, true, 0, op_a + " + " + op_b, 1, 1, instr.iadd32i.saturate != 0); - if (instr.op_32.generates_cc) { - LOG_CRITICAL(HW_GPU, "IADD32 Generates an unhandled Control Code"); - UNREACHABLE(); - } break; case OpCode::Id::LOP32I: { + UNIMPLEMENTED_IF_MSG(instr.op_32.generates_cc, + "Condition codes generation in LOP32I is not implemented"); + if (instr.alu.lop32i.invert_a) op_a = "~(" + op_a + ')'; @@ -1771,16 +1758,11 @@ private: WriteLogicOperation(instr.gpr0, instr.alu.lop32i.operation, op_a, op_b, Tegra::Shader::PredicateResultMode::None, Tegra::Shader::Pred::UnusedIndex); - if (instr.op_32.generates_cc) { - LOG_CRITICAL(HW_GPU, "LOP32I Generates an unhandled Control Code"); - UNREACHABLE(); - } break; } default: { - LOG_CRITICAL(HW_GPU, "Unhandled ArithmeticIntegerImmediate instruction: {}", - opcode->get().GetName()); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled ArithmeticIntegerImmediate instruction: {}", + opcode->get().GetName()); } } break; @@ -1803,6 +1785,9 @@ private: case OpCode::Id::IADD_C: case OpCode::Id::IADD_R: case OpCode::Id::IADD_IMM: { + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "Condition codes generation in IADD is not implemented"); + if (instr.alu_integer.negate_a) op_a = "-(" + op_a + ')'; @@ -1811,15 +1796,14 @@ private: regs.SetRegisterToInteger(instr.gpr0, true, 0, op_a + " + " + op_b, 1, 1, instr.alu.saturate_d); - if (instr.generates_cc) { - LOG_CRITICAL(HW_GPU, "IADD Generates an unhandled Control Code"); - UNREACHABLE(); - } break; } case OpCode::Id::IADD3_C: case OpCode::Id::IADD3_R: case OpCode::Id::IADD3_IMM: { + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "Condition codes generation in IADD3 is not implemented"); + std::string op_c = regs.GetRegisterAsInteger(instr.gpr39); auto apply_height = [](auto height, auto& oprand) { @@ -1833,9 +1817,8 @@ private: oprand = "((" + oprand + ") >> 16)"; break; default: - LOG_CRITICAL(HW_GPU, "Unhandled IADD3 height: {}", - static_cast<u32>(height.Value())); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled IADD3 height: {}", + static_cast<u32>(height.Value())); } }; @@ -1876,16 +1859,14 @@ private: } regs.SetRegisterToInteger(instr.gpr0, true, 0, result, 1, 1); - - if (instr.generates_cc) { - LOG_CRITICAL(HW_GPU, "IADD3 Generates an unhandled Control Code"); - UNREACHABLE(); - } break; } case OpCode::Id::ISCADD_C: case OpCode::Id::ISCADD_R: case OpCode::Id::ISCADD_IMM: { + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "Condition codes generation in ISCADD is not implemented"); + if (instr.alu_integer.negate_a) op_a = "-(" + op_a + ')'; @@ -1896,10 +1877,6 @@ private: regs.SetRegisterToInteger(instr.gpr0, true, 0, "((" + op_a + " << " + shift + ") + " + op_b + ')', 1, 1); - if (instr.generates_cc) { - LOG_CRITICAL(HW_GPU, "ISCADD Generates an unhandled Control Code"); - UNREACHABLE(); - } break; } case OpCode::Id::POPC_C: @@ -1923,6 +1900,9 @@ private: case OpCode::Id::LOP_C: case OpCode::Id::LOP_R: case OpCode::Id::LOP_IMM: { + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "Condition codes generation in LOP is not implemented"); + if (instr.alu.lop.invert_a) op_a = "~(" + op_a + ')'; @@ -1931,15 +1911,14 @@ private: WriteLogicOperation(instr.gpr0, instr.alu.lop.operation, op_a, op_b, instr.alu.lop.pred_result_mode, instr.alu.lop.pred48); - if (instr.generates_cc) { - LOG_CRITICAL(HW_GPU, "LOP Generates an unhandled Control Code"); - UNREACHABLE(); - } break; } case OpCode::Id::LOP3_C: case OpCode::Id::LOP3_R: case OpCode::Id::LOP3_IMM: { + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "Condition codes generation in LOP3 is not implemented"); + const std::string op_c = regs.GetRegisterAsInteger(instr.gpr39); std::string lut; @@ -1950,17 +1929,15 @@ private: } WriteLop3Instruction(instr.gpr0, op_a, op_b, op_c, lut); - if (instr.generates_cc) { - LOG_CRITICAL(HW_GPU, "LOP3 Generates an unhandled Control Code"); - UNREACHABLE(); - } break; } case OpCode::Id::IMNMX_C: case OpCode::Id::IMNMX_R: case OpCode::Id::IMNMX_IMM: { - ASSERT_MSG(instr.imnmx.exchange == Tegra::Shader::IMinMaxExchange::None, - "Unimplemented"); + UNIMPLEMENTED_IF(instr.imnmx.exchange != Tegra::Shader::IMinMaxExchange::None); + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "Condition codes generation in IMNMX is not implemented"); + const std::string condition = GetPredicateCondition(instr.imnmx.pred, instr.imnmx.negate_pred != 0); const std::string parameters = op_a + ',' + op_b; @@ -1968,10 +1945,6 @@ private: '(' + condition + ") ? min(" + parameters + ") : max(" + parameters + ')', 1, 1); - if (instr.generates_cc) { - LOG_CRITICAL(HW_GPU, "IMNMX Generates an unhandled Control Code"); - UNREACHABLE(); - } break; } case OpCode::Id::LEA_R2: @@ -2026,24 +1999,19 @@ private: op_b = regs.GetRegisterAsInteger(instr.gpr8); op_a = std::to_string(instr.lea.imm.entry_a); op_c = std::to_string(instr.lea.imm.entry_b); - LOG_CRITICAL(HW_GPU, "Unhandled LEA subinstruction: {}", - opcode->get().GetName()); - UNREACHABLE(); - } + UNIMPLEMENTED_MSG("Unhandled LEA subinstruction: {}", opcode->get().GetName()); } - if (instr.lea.pred48 != static_cast<u64>(Pred::UnusedIndex)) { - LOG_ERROR(HW_GPU, "Unhandled LEA Predicate"); - UNREACHABLE(); } + UNIMPLEMENTED_IF_MSG(instr.lea.pred48 != static_cast<u64>(Pred::UnusedIndex), + "Unhandled LEA Predicate"); const std::string value = '(' + op_a + " + (" + op_b + "*(1 << " + op_c + ")))"; regs.SetRegisterToInteger(instr.gpr0, true, 0, value, 1, 1); break; } default: { - LOG_CRITICAL(HW_GPU, "Unhandled ArithmeticInteger instruction: {}", - opcode->get().GetName()); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled ArithmeticInteger instruction: {}", + opcode->get().GetName()); } } @@ -2052,7 +2020,7 @@ private: case OpCode::Type::ArithmeticHalf: { if (opcode->get().GetId() == OpCode::Id::HADD2_C || opcode->get().GetId() == OpCode::Id::HADD2_R) { - ASSERT_MSG(instr.alu_half.ftz == 0, "Unimplemented"); + UNIMPLEMENTED_IF(instr.alu_half.ftz != 0); } const bool negate_a = opcode->get().GetId() != OpCode::Id::HMUL2_R && instr.alu_half.negate_a != 0; @@ -2090,9 +2058,8 @@ private: case OpCode::Id::HMUL2_R: return '(' + op_a + " * " + op_b + ')'; default: - LOG_CRITICAL(HW_GPU, "Unhandled half float instruction: {}", - opcode->get().GetName()); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled half float instruction: {}", + opcode->get().GetName()); return std::string("0"); } }(); @@ -2103,10 +2070,10 @@ private: } case OpCode::Type::ArithmeticHalfImmediate: { if (opcode->get().GetId() == OpCode::Id::HADD2_IMM) { - ASSERT_MSG(instr.alu_half_imm.ftz == 0, "Unimplemented"); + UNIMPLEMENTED_IF(instr.alu_half_imm.ftz != 0); } else { - ASSERT_MSG(instr.alu_half_imm.precision == Tegra::Shader::HalfPrecision::None, - "Unimplemented"); + UNIMPLEMENTED_IF(instr.alu_half_imm.precision != + Tegra::Shader::HalfPrecision::None); } const std::string op_a = GetHalfFloat( @@ -2136,11 +2103,14 @@ private: std::string op_b = instr.ffma.negate_b ? "-" : ""; std::string op_c = instr.ffma.negate_c ? "-" : ""; - ASSERT_MSG(instr.ffma.cc == 0, "FFMA cc not implemented"); - ASSERT_MSG(instr.ffma.tab5980_0 == 1, "FFMA tab5980_0({}) not implemented", - instr.ffma.tab5980_0.Value()); // Seems to be 1 by default based on SMO - ASSERT_MSG(instr.ffma.tab5980_1 == 0, "FFMA tab5980_1({}) not implemented", - instr.ffma.tab5980_1.Value()); + UNIMPLEMENTED_IF_MSG(instr.ffma.cc != 0, "FFMA cc not implemented"); + UNIMPLEMENTED_IF_MSG( + instr.ffma.tab5980_0 != 1, "FFMA tab5980_0({}) not implemented", + 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, + "Condition codes generation in FFMA is not implemented"); switch (opcode->get().GetId()) { case OpCode::Id::FFMA_CR: { @@ -2166,27 +2136,19 @@ private: break; } default: { - LOG_CRITICAL(HW_GPU, "Unhandled FFMA instruction: {}", opcode->get().GetName()); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled FFMA instruction: {}", opcode->get().GetName()); } } regs.SetRegisterToFloat(instr.gpr0, 0, "fma(" + op_a + ", " + op_b + ", " + op_c + ')', 1, 1, instr.alu.saturate_d, 0, true); - if (instr.generates_cc) { - LOG_CRITICAL(HW_GPU, "FFMA Generates an unhandled Control Code"); - UNREACHABLE(); - } - break; } case OpCode::Type::Hfma2: { if (opcode->get().GetId() == OpCode::Id::HFMA2_RR) { - ASSERT_MSG(instr.hfma2.rr.precision == Tegra::Shader::HalfPrecision::None, - "Unimplemented"); + UNIMPLEMENTED_IF(instr.hfma2.rr.precision != Tegra::Shader::HalfPrecision::None); } else { - ASSERT_MSG(instr.hfma2.precision == Tegra::Shader::HalfPrecision::None, - "Unimplemented"); + UNIMPLEMENTED_IF(instr.hfma2.precision != Tegra::Shader::HalfPrecision::None); } const bool saturate = opcode->get().GetId() == OpCode::Id::HFMA2_RR ? instr.hfma2.rr.saturate != 0 @@ -2236,7 +2198,7 @@ private: case OpCode::Type::Conversion: { switch (opcode->get().GetId()) { case OpCode::Id::I2I_R: { - ASSERT_MSG(!instr.conversion.selector, "Unimplemented"); + UNIMPLEMENTED_IF(instr.conversion.selector); std::string op_a = regs.GetRegisterAsInteger( instr.gpr20, 0, instr.conversion.is_input_signed, instr.conversion.src_size); @@ -2256,8 +2218,10 @@ private: } case OpCode::Id::I2F_R: case OpCode::Id::I2F_C: { - ASSERT_MSG(instr.conversion.dest_size == Register::Size::Word, "Unimplemented"); - ASSERT_MSG(!instr.conversion.selector, "Unimplemented"); + UNIMPLEMENTED_IF(instr.conversion.dest_size != Register::Size::Word); + UNIMPLEMENTED_IF(instr.conversion.selector); + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "Condition codes generation in I2F is not implemented"); std::string op_a{}; @@ -2282,16 +2246,13 @@ private: } regs.SetRegisterToFloat(instr.gpr0, 0, op_a, 1, 1); - - if (instr.generates_cc) { - LOG_CRITICAL(HW_GPU, "I2F Generates an unhandled Control Code"); - UNREACHABLE(); - } break; } case OpCode::Id::F2F_R: { - ASSERT_MSG(instr.conversion.dest_size == Register::Size::Word, "Unimplemented"); - ASSERT_MSG(instr.conversion.src_size == Register::Size::Word, "Unimplemented"); + UNIMPLEMENTED_IF(instr.conversion.dest_size != Register::Size::Word); + UNIMPLEMENTED_IF(instr.conversion.src_size != Register::Size::Word); + 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) { @@ -2318,23 +2279,19 @@ private: op_a = "trunc(" + op_a + ')'; break; default: - LOG_CRITICAL(HW_GPU, "Unimplemented f2f rounding mode {}", - static_cast<u32>(instr.conversion.f2f.rounding.Value())); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unimplemented F2F rounding mode {}", + static_cast<u32>(instr.conversion.f2f.rounding.Value())); break; } regs.SetRegisterToFloat(instr.gpr0, 0, op_a, 1, 1, instr.alu.saturate_d); - - if (instr.generates_cc) { - LOG_CRITICAL(HW_GPU, "F2F Generates an unhandled Control Code"); - UNREACHABLE(); - } break; } case OpCode::Id::F2I_R: case OpCode::Id::F2I_C: { - ASSERT_MSG(instr.conversion.src_size == Register::Size::Word, "Unimplemented"); + UNIMPLEMENTED_IF(instr.conversion.src_size != Register::Size::Word); + UNIMPLEMENTED_IF_MSG(instr.generates_cc, + "Condition codes generation in F2I is not implemented"); std::string op_a{}; if (instr.is_b_gpr) { @@ -2365,9 +2322,8 @@ private: op_a = "trunc(" + op_a + ')'; break; default: - LOG_CRITICAL(HW_GPU, "Unimplemented f2i rounding mode {}", - static_cast<u32>(instr.conversion.f2i.rounding.Value())); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unimplemented F2I rounding mode {}", + static_cast<u32>(instr.conversion.f2i.rounding.Value())); break; } @@ -2379,16 +2335,10 @@ private: regs.SetRegisterToInteger(instr.gpr0, instr.conversion.is_output_signed, 0, op_a, 1, 1, false, 0, instr.conversion.dest_size); - if (instr.generates_cc) { - LOG_CRITICAL(HW_GPU, "F2I Generates an unhandled Control Code"); - UNREACHABLE(); - } break; } default: { - LOG_CRITICAL(HW_GPU, "Unhandled conversion instruction: {}", - opcode->get().GetName()); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled conversion instruction: {}", opcode->get().GetName()); } } break; @@ -2397,10 +2347,10 @@ private: switch (opcode->get().GetId()) { case OpCode::Id::LD_A: { // Note: Shouldn't this be interp mode flat? As in no interpolation made. - ASSERT_MSG(instr.gpr8.Value() == Register::ZeroIndex, - "Indirect attribute loads are not supported"); - ASSERT_MSG((instr.attribute.fmt20.immediate.Value() % sizeof(u32)) == 0, - "Unaligned attribute loads are not supported"); + UNIMPLEMENTED_IF_MSG(instr.gpr8.Value() != Register::ZeroIndex, + "Indirect attribute loads are not supported"); + UNIMPLEMENTED_IF_MSG((instr.attribute.fmt20.immediate.Value() % sizeof(u32)) != 0, + "Unaligned attribute loads are not supported"); Tegra::Shader::IpaMode input_mode{Tegra::Shader::IpaInterpMode::Perspective, Tegra::Shader::IpaSampleMode::Default}; @@ -2427,7 +2377,7 @@ private: break; } case OpCode::Id::LD_C: { - ASSERT_MSG(instr.ld_c.unknown == 0, "Unimplemented"); + UNIMPLEMENTED_IF(instr.ld_c.unknown != 0); // Add an extra scope and declare the index register inside to prevent // overwriting it in case it is used as an output of the LD instruction. @@ -2455,9 +2405,8 @@ private: break; } default: - LOG_CRITICAL(HW_GPU, "Unhandled type: {}", - static_cast<unsigned>(instr.ld_c.type.Value())); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled type: {}", + static_cast<unsigned>(instr.ld_c.type.Value())); } --shader.scope; @@ -2465,6 +2414,9 @@ private: break; } case OpCode::Id::LD_L: { + UNIMPLEMENTED_IF_MSG(instr.ld_l.unknown == 1, "LD_L Unhandled mode: {}", + static_cast<unsigned>(instr.ld_l.unknown.Value())); + // Add an extra scope and declare the index register inside to prevent // overwriting it in case it is used as an output of the LD instruction. shader.AddLine('{'); @@ -2477,20 +2429,13 @@ private: const std::string op_a = regs.GetLocalMemoryAsFloat("index"); - if (instr.ld_l.unknown != 1) { - LOG_CRITICAL(HW_GPU, "LD_L Unhandled mode: {}", - static_cast<unsigned>(instr.ld_l.unknown.Value())); - UNREACHABLE(); - } - switch (instr.ldst_sl.type.Value()) { case Tegra::Shader::StoreType::Bytes32: regs.SetRegisterToFloat(instr.gpr0, 0, op_a, 1, 1); break; default: - LOG_CRITICAL(HW_GPU, "LD_L Unhandled type: {}", - static_cast<unsigned>(instr.ldst_sl.type.Value())); - UNREACHABLE(); + UNIMPLEMENTED_MSG("LD_L Unhandled type: {}", + static_cast<unsigned>(instr.ldst_sl.type.Value())); } --shader.scope; @@ -2498,10 +2443,10 @@ private: break; } case OpCode::Id::ST_A: { - ASSERT_MSG(instr.gpr8.Value() == Register::ZeroIndex, - "Indirect attribute loads are not supported"); - ASSERT_MSG((instr.attribute.fmt20.immediate.Value() % sizeof(u32)) == 0, - "Unaligned attribute loads are not supported"); + UNIMPLEMENTED_IF_MSG(instr.gpr8.Value() != Register::ZeroIndex, + "Indirect attribute loads are not supported"); + UNIMPLEMENTED_IF_MSG((instr.attribute.fmt20.immediate.Value() % sizeof(u32)) != 0, + "Unaligned attribute loads are not supported"); u64 next_element = instr.attribute.fmt20.element; u64 next_index = static_cast<u64>(instr.attribute.fmt20.index.Value()); @@ -2526,6 +2471,9 @@ private: break; } case OpCode::Id::ST_L: { + UNIMPLEMENTED_IF_MSG(instr.st_l.unknown == 0, "ST_L Unhandled mode: {}", + static_cast<unsigned>(instr.st_l.unknown.Value())); + // Add an extra scope and declare the index register inside to prevent // overwriting it in case it is used as an output of the LD instruction. shader.AddLine('{'); @@ -2536,20 +2484,13 @@ private: shader.AddLine("uint index = (" + op + " / 4);"); - if (instr.st_l.unknown != 0) { - LOG_CRITICAL(HW_GPU, "ST_L Unhandled mode: {}", - static_cast<unsigned>(instr.st_l.unknown.Value())); - UNREACHABLE(); - } - switch (instr.ldst_sl.type.Value()) { case Tegra::Shader::StoreType::Bytes32: regs.SetLocalMemoryAsFloat("index", regs.GetRegisterAsFloat(instr.gpr0)); break; default: - LOG_CRITICAL(HW_GPU, "ST_L Unhandled type: {}", - static_cast<unsigned>(instr.ldst_sl.type.Value())); - UNREACHABLE(); + UNIMPLEMENTED_MSG("ST_L Unhandled type: {}", + static_cast<unsigned>(instr.ldst_sl.type.Value())); } --shader.scope; @@ -2561,10 +2502,10 @@ private: std::string coord; const bool is_array = instr.tex.array != 0; - ASSERT_MSG(!instr.tex.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP), - "NODEP is not implemented"); - ASSERT_MSG(!instr.tex.UsesMiscMode(Tegra::Shader::TextureMiscMode::AOFFI), - "AOFFI is not implemented"); + UNIMPLEMENTED_IF_MSG(instr.tex.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP), + "NODEP is not implemented"); + UNIMPLEMENTED_IF_MSG(instr.tex.UsesMiscMode(Tegra::Shader::TextureMiscMode::AOFFI), + "AOFFI is not implemented"); const bool depth_compare = instr.tex.UsesMiscMode(Tegra::Shader::TextureMiscMode::DC); @@ -2630,9 +2571,8 @@ private: break; } default: - LOG_CRITICAL(HW_GPU, "Unhandled coordinates number {}", - static_cast<u32>(num_coordinates)); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled coordinates number {}", + static_cast<u32>(num_coordinates)); // Fallback to interpreting as a 2D texture for now const std::string x = regs.GetRegisterAsFloat(instr.gpr8); @@ -2642,7 +2582,6 @@ private: } // 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); @@ -2665,34 +2604,41 @@ private: } 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); - } else { - op_c = regs.GetRegisterAsFloat(instr.gpr20); - } + const std::string bias = [&]() { + if (depth_compare) { + if (is_array) + return regs.GetRegisterAsFloat(instr.gpr20.Value() + 2); + else + return regs.GetRegisterAsFloat(instr.gpr20.Value() + 1); + } else { + return regs.GetRegisterAsFloat(instr.gpr20); + } + }(); + shader.AddLine("float bias = " + bias + ';'); + // TODO: Figure if A suffix changes the equation at all. - texture = "texture(" + sampler + ", coords, " + op_c + ')'; + texture = "texture(" + sampler + ", coords, bias)"; break; } case Tegra::Shader::TextureProcessMode::LL: case Tegra::Shader::TextureProcessMode::LLA: { - if (num_coordinates <= 2) { - op_c = regs.GetRegisterAsFloat(instr.gpr20); - } else { - op_c = regs.GetRegisterAsFloat(instr.gpr20.Value() + 1); - } + const std::string lod = [&]() { + if (num_coordinates <= 2) { + return regs.GetRegisterAsFloat(instr.gpr20); + } else { + return regs.GetRegisterAsFloat(instr.gpr20.Value() + 1); + } + }(); + shader.AddLine("float lod = " + lod + ';'); + // TODO: Figure if A suffix changes the equation at all. - texture = "textureLod(" + sampler + ", coords, " + op_c + ')'; + texture = "textureLod(" + sampler + ", coords, lod)"; break; } default: { texture = "texture(" + sampler + ", coords)"; - LOG_CRITICAL(HW_GPU, "Unhandled texture process mode {}", - static_cast<u32>(instr.tex.GetTextureProcessMode())); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled texture process mode {}", + static_cast<u32>(instr.tex.GetTextureProcessMode())); } } if (!depth_compare) { @@ -2713,12 +2659,11 @@ private: break; } case OpCode::Id::TEXS: { - std::string coord; Tegra::Shader::TextureType texture_type{instr.texs.GetTextureType()}; bool is_array{instr.texs.IsArrayTexture()}; - ASSERT_MSG(!instr.texs.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP), - "NODEP is not implemented"); + UNIMPLEMENTED_IF_MSG(instr.texs.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP), + "NODEP is not implemented"); const bool depth_compare = instr.texs.UsesMiscMode(Tegra::Shader::TextureMiscMode::DC); @@ -2726,45 +2671,48 @@ private: if (depth_compare) num_coordinates += 1; + // Scope to avoid variable name overlaps. + shader.AddLine('{'); + ++shader.scope; + switch (num_coordinates) { 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 + ");"; + shader.AddLine("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 + ");"; + shader.AddLine("vec2 coords = vec2(" + x + ", " + y + ");"); } break; } case 3: { if (is_array) { - UNIMPLEMENTED_MSG("3-coordinate arrays not fully implemented"); - const std::string x = regs.GetRegisterAsFloat(instr.gpr8); - const std::string y = regs.GetRegisterAsFloat(instr.gpr20); - coord = "vec2 coords = vec2(" + x + ", " + y + ");"; - texture_type = Tegra::Shader::TextureType::Texture2D; - is_array = false; + 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); + shader.AddLine("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 + ");"; + shader.AddLine("vec3 coords = vec3(" + x + ", " + y + ", " + z + ");"); } break; } default: - LOG_CRITICAL(HW_GPU, "Unhandled coordinates number {}", - static_cast<u32>(num_coordinates)); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled coordinates number {}", + static_cast<u32>(num_coordinates)); // Fallback to interpreting as a 2D texture for now const std::string x = regs.GetRegisterAsFloat(instr.gpr8); const std::string y = regs.GetRegisterAsFloat(instr.gpr20); - coord = "vec2 coords = vec2(" + x + ", " + y + ");"; + shader.AddLine("vec2 coords = vec2(" + x + ", " + y + ");"); texture_type = Tegra::Shader::TextureType::Texture2D; is_array = false; } @@ -2777,7 +2725,11 @@ private: break; } case Tegra::Shader::TextureProcessMode::LZ: { - texture = "textureLod(" + sampler + ", coords, 0.0)"; + if (depth_compare && is_array) { + texture = "texture(" + sampler + ", coords)"; + } else { + texture = "textureLod(" + sampler + ", coords, 0.0)"; + } break; } case Tegra::Shader::TextureProcessMode::LL: { @@ -2787,57 +2739,57 @@ private: } default: { texture = "texture(" + sampler + ", coords)"; - LOG_CRITICAL(HW_GPU, "Unhandled texture process mode {}", - static_cast<u32>(instr.texs.GetTextureProcessMode())); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled texture process mode {}", + static_cast<u32>(instr.texs.GetTextureProcessMode())); } } if (!depth_compare) { - WriteTexsInstruction(instr, coord, texture); + WriteTexsInstruction(instr, texture); } else { - WriteTexsInstruction(instr, coord, "vec4(" + texture + ')'); + WriteTexsInstruction(instr, "vec4(" + texture + ')'); } + + shader.AddLine('}'); + --shader.scope; break; } case OpCode::Id::TLDS: { - std::string coord; const Tegra::Shader::TextureType texture_type{instr.tlds.GetTextureType()}; const bool is_array{instr.tlds.IsArrayTexture()}; ASSERT(texture_type == Tegra::Shader::TextureType::Texture2D); ASSERT(is_array == false); - ASSERT_MSG(!instr.tlds.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP), - "NODEP is not implemented"); - ASSERT_MSG(!instr.tlds.UsesMiscMode(Tegra::Shader::TextureMiscMode::AOFFI), - "AOFFI is not implemented"); - ASSERT_MSG(!instr.tlds.UsesMiscMode(Tegra::Shader::TextureMiscMode::MZ), - "MZ is not implemented"); + UNIMPLEMENTED_IF_MSG(instr.tlds.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP), + "NODEP is not implemented"); + UNIMPLEMENTED_IF_MSG(instr.tlds.UsesMiscMode(Tegra::Shader::TextureMiscMode::AOFFI), + "AOFFI is not implemented"); + 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; switch (texture_type) { case Tegra::Shader::TextureType::Texture1D: { const std::string x = regs.GetRegisterAsInteger(instr.gpr8); - coord = "int coords = " + x + ';'; + shader.AddLine("int coords = " + x + ';'); break; } case Tegra::Shader::TextureType::Texture2D: { - if (is_array) { - LOG_CRITICAL(HW_GPU, "Unhandled 2d array texture"); - UNREACHABLE(); - } else { - 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; - } + UNIMPLEMENTED_IF_MSG(is_array, "Unhandled 2d array texture"); + + const std::string x = regs.GetRegisterAsInteger(instr.gpr8); + const std::string y = regs.GetRegisterAsInteger(instr.gpr20); + shader.AddLine("ivec2 coords = ivec2(" + x + ", " + y + ");"); + extra_op_offset = 1; break; } default: - LOG_CRITICAL(HW_GPU, "Unhandled texture type {}", - static_cast<u32>(texture_type)); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled texture type {}", static_cast<u32>(texture_type)); } const std::string sampler = GetSampler(instr.sampler, texture_type, is_array, false); @@ -2848,19 +2800,22 @@ 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: { texture = "texelFetch(" + sampler + ", coords, 0)"; - LOG_CRITICAL(HW_GPU, "Unhandled texture process mode {}", - static_cast<u32>(instr.tlds.GetTextureProcessMode())); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled texture process mode {}", + static_cast<u32>(instr.tlds.GetTextureProcessMode())); } } - WriteTexsInstruction(instr, coord, texture); + WriteTexsInstruction(instr, texture); + + --shader.scope; + shader.AddLine('}'); break; } case OpCode::Id::TLD4: { @@ -2868,14 +2823,14 @@ private: ASSERT(instr.tld4.array == 0); std::string coord; - ASSERT_MSG(!instr.tld4.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP), - "NODEP is not implemented"); - ASSERT_MSG(!instr.tld4.UsesMiscMode(Tegra::Shader::TextureMiscMode::AOFFI), - "AOFFI is not implemented"); - ASSERT_MSG(!instr.tld4.UsesMiscMode(Tegra::Shader::TextureMiscMode::NDV), - "NDV is not implemented"); - ASSERT_MSG(!instr.tld4.UsesMiscMode(Tegra::Shader::TextureMiscMode::PTP), - "PTP is not implemented"); + UNIMPLEMENTED_IF_MSG(instr.tld4.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP), + "NODEP is not implemented"); + UNIMPLEMENTED_IF_MSG(instr.tld4.UsesMiscMode(Tegra::Shader::TextureMiscMode::AOFFI), + "AOFFI is not implemented"); + UNIMPLEMENTED_IF_MSG(instr.tld4.UsesMiscMode(Tegra::Shader::TextureMiscMode::NDV), + "NDV is not implemented"); + UNIMPLEMENTED_IF_MSG(instr.tld4.UsesMiscMode(Tegra::Shader::TextureMiscMode::PTP), + "PTP is not implemented"); const bool depth_compare = instr.tld4.UsesMiscMode(Tegra::Shader::TextureMiscMode::DC); auto texture_type = instr.tld4.texture_type.Value(); @@ -2883,37 +2838,37 @@ 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: - LOG_CRITICAL(HW_GPU, "Unhandled coordinates number {}", - static_cast<u32>(num_coordinates)); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled coordinates number {}", + 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) { @@ -2930,14 +2885,20 @@ private: regs.SetRegisterToFloat(instr.gpr0, 0, texture, 1, 1, false); } --shader.scope; - shader.AddLine("}"); + shader.AddLine('}'); break; } case OpCode::Id::TLD4S: { - ASSERT_MSG(!instr.tld4s.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP), - "NODEP is not implemented"); - ASSERT_MSG(!instr.tld4s.UsesMiscMode(Tegra::Shader::TextureMiscMode::AOFFI), - "AOFFI is not implemented"); + UNIMPLEMENTED_IF_MSG( + instr.tld4s.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP), + "NODEP is not implemented"); + UNIMPLEMENTED_IF_MSG( + instr.tld4s.UsesMiscMode(Tegra::Shader::TextureMiscMode::AOFFI), + "AOFFI is not implemented"); + + // Scope to avoid variable name overlaps. + shader.AddLine('{'); + ++shader.scope; const bool depth_compare = instr.tld4s.UsesMiscMode(Tegra::Shader::TextureMiscMode::DC); @@ -2946,28 +2907,33 @@ 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 + ");"; + shader.AddLine("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 + ");"; + shader.AddLine( + "float op_y = " + regs.GetRegisterAsFloat(instr.gpr8.Value() + 1) + ';'); + shader.AddLine("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, texture); } else { - WriteTexsInstruction(instr, coord, "vec4(" + texture + ')'); + WriteTexsInstruction(instr, "vec4(" + texture + ')'); } + + --shader.scope; + shader.AddLine('}'); break; } case OpCode::Id::TXQ: { - ASSERT_MSG(!instr.txq.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP), - "NODEP is not implemented"); + 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. @@ -2975,23 +2941,30 @@ 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: { - LOG_CRITICAL(HW_GPU, "Unhandled texture query type: {}", - static_cast<u32>(instr.txq.query_type.Value())); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled texture query type: {}", + static_cast<u32>(instr.txq.query_type.Value())); } } + --shader.scope; + shader.AddLine('}'); break; } case OpCode::Id::TMML: { - ASSERT_MSG(!instr.tmml.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP), - "NODEP is not implemented"); - ASSERT_MSG(!instr.tmml.UsesMiscMode(Tegra::Shader::TextureMiscMode::NDV), - "NDV is not implemented"); + UNIMPLEMENTED_IF_MSG(instr.tmml.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP), + "NODEP is not implemented"); + UNIMPLEMENTED_IF_MSG(instr.tmml.UsesMiscMode(Tegra::Shader::TextureMiscMode::NDV), + "NDV is not implemented"); const std::string x = regs.GetRegisterAsFloat(instr.gpr8); const bool is_array = instr.tmml.array != 0; @@ -3013,9 +2986,7 @@ private: break; } default: - LOG_CRITICAL(HW_GPU, "Unhandled texture type {}", - static_cast<u32>(texture_type)); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled texture type {}", static_cast<u32>(texture_type)); // Fallback to interpreting as a 2D texture for now const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); @@ -3038,8 +3009,7 @@ private: break; } default: { - LOG_CRITICAL(HW_GPU, "Unhandled memory instruction: {}", opcode->get().GetName()); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled memory instruction: {}", opcode->get().GetName()); } } break; @@ -3125,7 +3095,7 @@ private: break; } case OpCode::Type::HalfSetPredicate: { - ASSERT_MSG(instr.hsetp2.ftz == 0, "Unimplemented"); + UNIMPLEMENTED_IF(instr.hsetp2.ftz != 0); const std::string op_a = GetHalfFloat(regs.GetRegisterAsInteger(instr.gpr8, 0, false), instr.hsetp2.type_a, @@ -3170,6 +3140,9 @@ private: break; } case OpCode::Type::PredicateSetRegister: { + 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); const std::string op_b = @@ -3190,12 +3163,6 @@ private: const std::string value = '(' + result + ") ? 1.0 : 0.0"; regs.SetRegisterToFloat(instr.gpr0, 0, value, 1, 1); } - - if (instr.generates_cc) { - LOG_CRITICAL(HW_GPU, "PSET Generates an unhandled Control Code"); - UNREACHABLE(); - } - break; } case OpCode::Type::PredicateSetPredicate: { @@ -3233,21 +3200,19 @@ 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; } default: { - LOG_CRITICAL(HW_GPU, "Unhandled predicate instruction: {}", - opcode->get().GetName()); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled predicate instruction: {}", opcode->get().GetName()); } } break; @@ -3327,7 +3292,7 @@ private: break; } case OpCode::Type::HalfSet: { - ASSERT_MSG(instr.hset2.ftz == 0, "Unimplemented"); + UNIMPLEMENTED_IF(instr.hset2.ftz != 0); const std::string op_a = GetHalfFloat(regs.GetRegisterAsInteger(instr.gpr8, 0, false), instr.hset2.type_a, @@ -3371,15 +3336,17 @@ private: break; } case OpCode::Type::Xmad: { - ASSERT_MSG(!instr.xmad.sign_a, "Unimplemented"); - ASSERT_MSG(!instr.xmad.sign_b, "Unimplemented"); + UNIMPLEMENTED_IF(instr.xmad.sign_a); + UNIMPLEMENTED_IF(instr.xmad.sign_b); + 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; std::string op_c; // TODO(bunnei): Needs to be fixed once op_a or op_b is signed - ASSERT_MSG(instr.xmad.sign_a == instr.xmad.sign_b, "Unimplemented"); + UNIMPLEMENTED_IF(instr.xmad.sign_a != instr.xmad.sign_b); const bool is_signed{instr.xmad.sign_a == 1}; bool is_merge{}; @@ -3412,8 +3379,7 @@ private: break; } default: { - LOG_CRITICAL(HW_GPU, "Unhandled XMAD instruction: {}", opcode->get().GetName()); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled XMAD instruction: {}", opcode->get().GetName()); } } @@ -3449,9 +3415,8 @@ private: op_c = "((" + op_c + ") + (" + src2 + "<< 16))"; break; default: { - LOG_CRITICAL(HW_GPU, "Unhandled XMAD mode: {}", - static_cast<u32>(instr.xmad.mode.Value())); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled XMAD mode: {}", + static_cast<u32>(instr.xmad.mode.Value())); } } @@ -3461,25 +3426,19 @@ private: } regs.SetRegisterToInteger(instr.gpr0, is_signed, 0, sum, 1, 1); - if (instr.generates_cc) { - LOG_CRITICAL(HW_GPU, "XMAD Generates an unhandled Control Code"); - UNREACHABLE(); - } break; } default: { switch (opcode->get().GetId()) { case OpCode::Id::EXIT: { + 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(); } - const Tegra::Shader::ControlCode cc = instr.flow_control_code; - if (cc != Tegra::Shader::ControlCode::T) { - LOG_CRITICAL(HW_GPU, "EXIT Control Code used: {}", static_cast<u32>(cc)); - UNREACHABLE(); - } - switch (instr.flow.cond) { case Tegra::Shader::FlowCondition::Always: shader.AddLine("return true;"); @@ -3494,26 +3453,24 @@ private: case Tegra::Shader::FlowCondition::Fcsm_Tr: // TODO(bunnei): What is this used for? If we assume this conditon is not // satisifed, dual vertex shaders in Farming Simulator make more sense - LOG_CRITICAL(HW_GPU, "Skipping unknown FlowCondition::Fcsm_Tr"); + UNIMPLEMENTED_MSG("Skipping unknown FlowCondition::Fcsm_Tr"); break; default: - LOG_CRITICAL(HW_GPU, "Unhandled flow condition: {}", - static_cast<u32>(instr.flow.cond.Value())); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled flow condition: {}", + static_cast<u32>(instr.flow.cond.Value())); } break; } case OpCode::Id::KIL: { - ASSERT(instr.flow.cond == Tegra::Shader::FlowCondition::Always); + UNIMPLEMENTED_IF(instr.flow.cond != Tegra::Shader::FlowCondition::Always); + + 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. - const Tegra::Shader::ControlCode cc = instr.flow_control_code; - if (cc != Tegra::Shader::ControlCode::T) { - LOG_CRITICAL(HW_GPU, "KIL Control Code used: {}", static_cast<u32>(cc)); - UNREACHABLE(); - } shader.AddLine("if (true) {"); ++shader.scope; shader.AddLine("discard;"); @@ -3523,7 +3480,8 @@ private: break; } case OpCode::Id::OUT_R: { - ASSERT(instr.gpr20.Value() == Register::ZeroIndex); + UNIMPLEMENTED_IF_MSG(instr.gpr20.Value() != Register::ZeroIndex, + "Stream buffer is not supported"); ASSERT_MSG(stage == Maxwell3D::Regs::ShaderStage::Geometry, "OUT is expected to be used in a geometry shader."); @@ -3550,18 +3508,17 @@ private: break; } default: { - LOG_CRITICAL(HW_GPU, "Unhandled system move: {}", - static_cast<u32>(instr.sys20.Value())); - UNREACHABLE(); + UNIMPLEMENTED_MSG("Unhandled system move: {}", + static_cast<u32>(instr.sys20.Value())); } } break; } case OpCode::Id::ISBERD: { - ASSERT(instr.isberd.o == 0); - ASSERT(instr.isberd.skew == 0); - ASSERT(instr.isberd.shift == Tegra::Shader::IsberdShift::None); - ASSERT(instr.isberd.mode == Tegra::Shader::IsberdMode::None); + UNIMPLEMENTED_IF(instr.isberd.o != 0); + UNIMPLEMENTED_IF(instr.isberd.skew != 0); + UNIMPLEMENTED_IF(instr.isberd.shift != Tegra::Shader::IsberdShift::None); + UNIMPLEMENTED_IF(instr.isberd.mode != Tegra::Shader::IsberdMode::None); ASSERT_MSG(stage == Maxwell3D::Regs::ShaderStage::Geometry, "ISBERD is expected to be used in a geometry shader."); LOG_WARNING(HW_GPU, "ISBERD instruction is incomplete"); @@ -3569,13 +3526,13 @@ private: break; } case OpCode::Id::BRA: { - ASSERT_MSG(instr.bra.constant_buffer == 0, - "BRA with constant buffers are not implemented"); - const Tegra::Shader::ControlCode cc = instr.flow_control_code; - if (cc != Tegra::Shader::ControlCode::T) { - LOG_CRITICAL(HW_GPU, "BRA Control Code used: {}", static_cast<u32>(cc)); - UNREACHABLE(); - } + UNIMPLEMENTED_IF_MSG(instr.bra.constant_buffer != 0, + "BRA with constant buffers are not implemented"); + + const Tegra::Shader::ConditionCode cc = instr.flow_condition_code; + UNIMPLEMENTED_IF_MSG(cc != Tegra::Shader::ConditionCode::T, + "BRA condition code used: {}", static_cast<u32>(cc)); + const u32 target = offset + instr.bra.GetBranchTarget(); shader.AddLine("{ jmp_to = " + std::to_string(target) + "u; break; }"); break; @@ -3598,7 +3555,8 @@ private: // The SSY opcode tells the GPU where to re-converge divergent execution paths, it // sets the target of the jump that the SYNC instruction will make. The SSY opcode // has a similar structure to the BRA opcode. - ASSERT_MSG(instr.bra.constant_buffer == 0, "Constant buffer flow is not supported"); + UNIMPLEMENTED_IF_MSG(instr.bra.constant_buffer != 0, + "Constant buffer flow is not supported"); const u32 target = offset + instr.bra.GetBranchTarget(); EmitPushToFlowStack(target); @@ -3608,29 +3566,28 @@ private: // PBK pushes to a stack the address where BRK will jump to. This shares stack with // SSY but using SYNC on a PBK address will kill the shader execution. We don't // emulate this because it's very unlikely a driver will emit such invalid shader. - ASSERT_MSG(instr.bra.constant_buffer == 0, "Constant buffer PBK is not supported"); + UNIMPLEMENTED_IF_MSG(instr.bra.constant_buffer != 0, + "Constant buffer PBK is not supported"); const u32 target = offset + instr.bra.GetBranchTarget(); EmitPushToFlowStack(target); break; } case OpCode::Id::SYNC: { + 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 - const Tegra::Shader::ControlCode cc = instr.flow_control_code; - if (cc != Tegra::Shader::ControlCode::T) { - LOG_CRITICAL(HW_GPU, "SYNC Control Code used: {}", static_cast<u32>(cc)); - UNREACHABLE(); - } EmitPopFromFlowStack(); break; } 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) { - LOG_CRITICAL(HW_GPU, "BRK Control Code used: {}", static_cast<u32>(cc)); - UNREACHABLE(); - } + 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; } @@ -3641,6 +3598,9 @@ 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); @@ -3660,11 +3620,6 @@ 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) { - LOG_CRITICAL(HW_GPU, "VMAD Generates an unhandled Control Code"); - UNREACHABLE(); - } - break; } case OpCode::Id::VSETP: { @@ -3691,10 +3646,7 @@ private: } break; } - default: { - LOG_CRITICAL(HW_GPU, "Unhandled instruction: {}", opcode->get().GetName()); - UNREACHABLE(); - } + default: { UNIMPLEMENTED_MSG("Unhandled instruction: {}", opcode->get().GetName()); } } break; @@ -3819,6 +3771,7 @@ private: Maxwell3D::Regs::ShaderStage stage; const std::string& suffix; u64 local_memory_size; + std::size_t shader_length; ShaderWriter shader; ShaderWriter declarations; @@ -3837,9 +3790,10 @@ 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 9d17edd63..eea090e52 100644 --- a/src/video_core/renderer_opengl/gl_shader_gen.cpp +++ b/src/video_core/renderer_opengl/gl_shader_gen.cpp @@ -82,8 +82,8 @@ void main() { } ProgramResult GenerateGeometryShader(const ShaderSetup& setup) { - std::string out = "#version 430 core\n"; - out += "#extension GL_ARB_separate_shader_objects : enable\n\n"; + // Version is intentionally skipped in shader generation, it's added by the lazy compilation. + std::string out = "#extension GL_ARB_separate_shader_objects : enable\n\n"; out += Decompiler::GetCommonDeclarations(); out += "bool exec_geometry();\n"; 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.h b/src/video_core/renderer_opengl/gl_shader_manager.h index 36fe1f04c..9a5d7e289 100644 --- a/src/video_core/renderer_opengl/gl_shader_manager.h +++ b/src/video_core/renderer_opengl/gl_shader_manager.h @@ -7,6 +7,7 @@ #include <glad/glad.h> #include "video_core/renderer_opengl/gl_resource_manager.h" +#include "video_core/renderer_opengl/gl_state.h" #include "video_core/renderer_opengl/maxwell_to_gl.h" namespace OpenGL::GLShader { @@ -66,6 +67,7 @@ public: glUseProgramStages(pipeline.handle, GL_FRAGMENT_SHADER_BIT, fs); state.draw.shader_program = 0; state.draw.program_pipeline = pipeline.handle; + state.geometry_shaders.enabled = (gs != 0); } private: diff --git a/src/video_core/renderer_opengl/gl_state.cpp b/src/video_core/renderer_opengl/gl_state.cpp index b6b426f34..934f4db78 100644 --- a/src/video_core/renderer_opengl/gl_state.cpp +++ b/src/video_core/renderer_opengl/gl_state.cpp @@ -14,7 +14,10 @@ OpenGLState OpenGLState::cur_state; bool OpenGLState::s_rgb_used; OpenGLState::OpenGLState() { // These all match default OpenGL values + geometry_shaders.enabled = false; framebuffer_srgb.enabled = false; + multisample_control.alpha_to_coverage = false; + multisample_control.alpha_to_one = false; cull.enabled = false; cull.mode = GL_BACK; cull.front_face = GL_CCW; @@ -22,17 +25,15 @@ OpenGLState::OpenGLState() { depth.test_enabled = false; depth.test_func = GL_LESS; depth.write_mask = GL_TRUE; - depth.depth_range_near = 0.0f; - depth.depth_range_far = 1.0f; primitive_restart.enabled = false; primitive_restart.index = 0; - - color_mask.red_enabled = GL_TRUE; - color_mask.green_enabled = GL_TRUE; - color_mask.blue_enabled = GL_TRUE; - color_mask.alpha_enabled = GL_TRUE; - + for (auto& item : color_mask) { + item.red_enabled = GL_TRUE; + item.green_enabled = GL_TRUE; + item.blue_enabled = GL_TRUE; + item.alpha_enabled = GL_TRUE; + } stencil.test_enabled = false; auto reset_stencil = [](auto& config) { config.test_func = GL_ALWAYS; @@ -45,19 +46,33 @@ OpenGLState::OpenGLState() { }; reset_stencil(stencil.front); reset_stencil(stencil.back); - - blend.enabled = true; - blend.rgb_equation = GL_FUNC_ADD; - blend.a_equation = GL_FUNC_ADD; - blend.src_rgb_func = GL_ONE; - blend.dst_rgb_func = GL_ZERO; - blend.src_a_func = GL_ONE; - blend.dst_a_func = GL_ZERO; - blend.color.red = 0.0f; - blend.color.green = 0.0f; - blend.color.blue = 0.0f; - blend.color.alpha = 0.0f; - + for (auto& item : viewports) { + item.x = 0; + item.y = 0; + item.width = 0; + item.height = 0; + item.depth_range_near = 0.0f; + item.depth_range_far = 1.0f; + item.scissor.enabled = false; + item.scissor.x = 0; + item.scissor.y = 0; + item.scissor.width = 0; + item.scissor.height = 0; + } + for (auto& item : blend) { + item.enabled = true; + item.rgb_equation = GL_FUNC_ADD; + item.a_equation = GL_FUNC_ADD; + item.src_rgb_func = GL_ONE; + item.dst_rgb_func = GL_ZERO; + item.src_a_func = GL_ONE; + item.dst_a_func = GL_ZERO; + } + independant_blend.enabled = false; + blend_color.red = 0.0f; + blend_color.green = 0.0f; + blend_color.blue = 0.0f; + blend_color.alpha = 0.0f; logic_op.enabled = false; logic_op.operation = GL_COPY; @@ -73,20 +88,10 @@ OpenGLState::OpenGLState() { draw.shader_program = 0; draw.program_pipeline = 0; - scissor.enabled = false; - scissor.x = 0; - scissor.y = 0; - scissor.width = 0; - scissor.height = 0; - - viewport.x = 0; - viewport.y = 0; - viewport.width = 0; - viewport.height = 0; - clip_distance = {}; point.size = 1; + fragment_color_clamp.enabled = false; } void OpenGLState::ApplyDefaultState() { @@ -134,6 +139,32 @@ void OpenGLState::ApplyCulling() const { } } +void OpenGLState::ApplyColorMask() const { + if (GLAD_GL_ARB_viewport_array && 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]; + if (updated.red_enabled != current.red_enabled || + updated.green_enabled != current.green_enabled || + updated.blue_enabled != current.blue_enabled || + updated.alpha_enabled != current.alpha_enabled) { + glColorMaski(static_cast<GLuint>(i), updated.red_enabled, updated.green_enabled, + updated.blue_enabled, updated.alpha_enabled); + } + } + } else { + const auto& updated = color_mask[0]; + const auto& current = cur_state.color_mask[0]; + if (updated.red_enabled != current.red_enabled || + updated.green_enabled != current.green_enabled || + updated.blue_enabled != current.blue_enabled || + updated.alpha_enabled != current.alpha_enabled) { + glColorMask(updated.red_enabled, updated.green_enabled, updated.blue_enabled, + updated.alpha_enabled); + } + } +} + void OpenGLState::ApplyDepth() const { // Depth test const bool depth_test_changed = depth.test_enabled != cur_state.depth.test_enabled; @@ -152,11 +183,6 @@ void OpenGLState::ApplyDepth() const { if (depth.write_mask != cur_state.depth.write_mask) { glDepthMask(depth.write_mask); } - // Depth range - if (depth.depth_range_near != cur_state.depth.depth_range_near || - depth.depth_range_far != cur_state.depth.depth_range_far) { - glDepthRange(depth.depth_range_near, depth.depth_range_far); - } } void OpenGLState::ApplyPrimitiveRestart() const { @@ -207,53 +233,160 @@ 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::ApplyScissorTest() const { - const bool scissor_changed = scissor.enabled != cur_state.scissor.enabled; - if (scissor_changed) { - if (scissor.enabled) { - glEnable(GL_SCISSOR_TEST); - } else { - glDisable(GL_SCISSOR_TEST); +void OpenGLState::ApplyViewport() const { + if (GLAD_GL_ARB_viewport_array && 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, 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) { + glDepthRangeIndexed(i, updated.depth_range_near, updated.depth_range_far); + } + const bool scissor_changed = updated.scissor.enabled != current.scissor.enabled; + if (scissor_changed) { + if (updated.scissor.enabled) { + glEnablei(GL_SCISSOR_TEST, i); + } else { + glDisablei(GL_SCISSOR_TEST, i); + } + } + if (updated.scissor.enabled && + (scissor_changed || updated.scissor.x != current.scissor.x || + updated.scissor.y != current.scissor.y || + updated.scissor.width != current.scissor.width || + updated.scissor.height != current.scissor.height)) { + glScissorIndexed(i, updated.scissor.x, updated.scissor.y, updated.scissor.width, + updated.scissor.height); + } + } + } else { + const auto& current = cur_state.viewports[0]; + const auto& updated = viewports[0]; + if (updated.x != current.x || updated.y != current.y || updated.width != current.width || + updated.height != current.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) { + glDepthRange(updated.depth_range_near, updated.depth_range_far); + } + const bool scissor_changed = updated.scissor.enabled != current.scissor.enabled; + if (scissor_changed) { + if (updated.scissor.enabled) { + glEnable(GL_SCISSOR_TEST); + } else { + glDisable(GL_SCISSOR_TEST); + } + } + if (updated.scissor.enabled && (scissor_changed || updated.scissor.x != current.scissor.x || + updated.scissor.y != current.scissor.y || + updated.scissor.width != current.scissor.width || + updated.scissor.height != current.scissor.height)) { + glScissor(updated.scissor.x, updated.scissor.y, updated.scissor.width, + updated.scissor.height); } - } - if (scissor_changed || scissor_changed || scissor.x != cur_state.scissor.x || - scissor.y != cur_state.scissor.y || scissor.width != cur_state.scissor.width || - scissor.height != cur_state.scissor.height) { - glScissor(scissor.x, scissor.y, scissor.width, scissor.height); } } -void OpenGLState::ApplyBlending() const { - const bool blend_changed = blend.enabled != cur_state.blend.enabled; +void OpenGLState::ApplyGlobalBlending() const { + const Blend& current = cur_state.blend[0]; + const Blend& updated = blend[0]; + const bool blend_changed = updated.enabled != current.enabled; if (blend_changed) { - if (blend.enabled) { - ASSERT(!logic_op.enabled); + if (updated.enabled) { glEnable(GL_BLEND); } else { glDisable(GL_BLEND); } } - if (blend.enabled) { - if (blend_changed || blend.color.red != cur_state.blend.color.red || - blend.color.green != cur_state.blend.color.green || - blend.color.blue != cur_state.blend.color.blue || - blend.color.alpha != cur_state.blend.color.alpha) { - glBlendColor(blend.color.red, blend.color.green, blend.color.blue, blend.color.alpha); - } + if (!updated.enabled) { + return; + } + 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) { + glBlendFuncSeparate(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) { + glBlendEquationSeparate(updated.rgb_equation, updated.a_equation); + } +} - if (blend_changed || blend.src_rgb_func != cur_state.blend.src_rgb_func || - blend.dst_rgb_func != cur_state.blend.dst_rgb_func || - blend.src_a_func != cur_state.blend.src_a_func || - blend.dst_a_func != cur_state.blend.dst_a_func) { - glBlendFuncSeparate(blend.src_rgb_func, blend.dst_rgb_func, blend.src_a_func, - blend.dst_a_func); +void OpenGLState::ApplyTargetBlending(std::size_t target, bool force) const { + const Blend& updated = blend[target]; + const Blend& current = cur_state.blend[target]; + const bool blend_changed = updated.enabled != current.enabled || force; + if (blend_changed) { + if (updated.enabled) { + glEnablei(GL_BLEND, static_cast<GLuint>(target)); + } else { + glDisablei(GL_BLEND, static_cast<GLuint>(target)); } + } + if (!updated.enabled) { + return; + } + 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) { + glBlendFuncSeparatei(static_cast<GLuint>(target), updated.src_rgb_func, + updated.dst_rgb_func, updated.src_a_func, updated.dst_a_func); + } - if (blend_changed || blend.rgb_equation != cur_state.blend.rgb_equation || - blend.a_equation != cur_state.blend.a_equation) { - glBlendEquationSeparate(blend.rgb_equation, blend.a_equation); + if (blend_changed || updated.rgb_equation != current.rgb_equation || + updated.a_equation != current.a_equation) { + glBlendEquationSeparatei(static_cast<GLuint>(target), updated.rgb_equation, + updated.a_equation); + } +} + +void OpenGLState::ApplyBlending() const { + if (independant_blend.enabled) { + for (size_t i = 0; i < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets; i++) { + ApplyTargetBlending(i, + independant_blend.enabled != cur_state.independant_blend.enabled); } + } else { + ApplyGlobalBlending(); + } + if (blend_color.red != cur_state.blend_color.red || + blend_color.green != cur_state.blend_color.green || + blend_color.blue != cur_state.blend_color.blue || + blend_color.alpha != cur_state.blend_color.alpha) { + glBlendColor(blend_color.red, blend_color.green, blend_color.blue, blend_color.alpha); } } @@ -261,7 +394,6 @@ void OpenGLState::ApplyLogicOp() const { const bool logic_op_changed = logic_op.enabled != cur_state.logic_op.enabled; if (logic_op_changed) { if (logic_op.enabled) { - ASSERT(!blend.enabled); glEnable(GL_COLOR_LOGIC_OP); } else { glDisable(GL_COLOR_LOGIC_OP); @@ -315,7 +447,7 @@ void OpenGLState::ApplySamplers() const { } } -void OpenGLState::Apply() const { +void OpenGLState::ApplyFramebufferState() const { // Framebuffer if (draw.read_framebuffer != cur_state.draw.read_framebuffer) { glBindFramebuffer(GL_READ_FRAMEBUFFER, draw.read_framebuffer); @@ -323,7 +455,9 @@ void OpenGLState::Apply() const { if (draw.draw_framebuffer != cur_state.draw.draw_framebuffer) { glBindFramebuffer(GL_DRAW_FRAMEBUFFER, draw.draw_framebuffer); } +} +void OpenGLState::ApplyVertexBufferState() const { // Vertex array if (draw.vertex_array != cur_state.draw.vertex_array) { glBindVertexArray(draw.vertex_array); @@ -333,7 +467,11 @@ void OpenGLState::Apply() const { if (draw.vertex_buffer != cur_state.draw.vertex_buffer) { glBindBuffer(GL_ARRAY_BUFFER, draw.vertex_buffer); } +} +void OpenGLState::Apply() const { + ApplyFramebufferState(); + ApplyVertexBufferState(); // Uniform buffer if (draw.uniform_buffer != cur_state.draw.uniform_buffer) { glBindBuffer(GL_UNIFORM_BUFFER, draw.uniform_buffer); @@ -348,12 +486,6 @@ void OpenGLState::Apply() const { if (draw.program_pipeline != cur_state.draw.program_pipeline) { glBindProgramPipeline(draw.program_pipeline); } - // Viewport - if (viewport.x != cur_state.viewport.x || viewport.y != cur_state.viewport.y || - viewport.width != cur_state.viewport.width || - viewport.height != cur_state.viewport.height) { - glViewport(viewport.x, viewport.y, viewport.width, viewport.height); - } // Clip distance for (std::size_t i = 0; i < clip_distance.size(); ++i) { if (clip_distance[i] != cur_state.clip_distance[i]) { @@ -364,19 +496,33 @@ void OpenGLState::Apply() const { } } } - // Color mask - if (color_mask.red_enabled != cur_state.color_mask.red_enabled || - color_mask.green_enabled != cur_state.color_mask.green_enabled || - color_mask.blue_enabled != cur_state.color_mask.blue_enabled || - color_mask.alpha_enabled != cur_state.color_mask.alpha_enabled) { - glColorMask(color_mask.red_enabled, color_mask.green_enabled, color_mask.blue_enabled, - color_mask.alpha_enabled); - } // Point if (point.size != cur_state.point.size) { glPointSize(point.size); } - ApplyScissorTest(); + 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 (multisample_control.alpha_to_coverage != cur_state.multisample_control.alpha_to_coverage) { + if (multisample_control.alpha_to_coverage) { + glEnable(GL_SAMPLE_ALPHA_TO_COVERAGE); + } else { + glDisable(GL_SAMPLE_ALPHA_TO_COVERAGE); + } + } + if (multisample_control.alpha_to_one != cur_state.multisample_control.alpha_to_one) { + if (multisample_control.alpha_to_one) { + glEnable(GL_SAMPLE_ALPHA_TO_ONE); + } else { + glDisable(GL_SAMPLE_ALPHA_TO_ONE); + } + } + + ApplyColorMask(); + ApplyViewport(); ApplyStencilTest(); ApplySRgb(); ApplyCulling(); diff --git a/src/video_core/renderer_opengl/gl_state.h b/src/video_core/renderer_opengl/gl_state.h index fe648aff6..032fc43f0 100644 --- a/src/video_core/renderer_opengl/gl_state.h +++ b/src/video_core/renderer_opengl/gl_state.h @@ -40,17 +40,28 @@ public: } framebuffer_srgb; struct { + bool alpha_to_coverage; // GL_ALPHA_TO_COVERAGE + bool alpha_to_one; // GL_ALPHA_TO_ONE + } multisample_control; + + struct { + bool enabled; // GL_CLAMP_FRAGMENT_COLOR_ARB + } fragment_color_clamp; + + struct { + bool enabled; // viewports arrays are only supported when geometry shaders are enabled. + } geometry_shaders; + + struct { bool enabled; // GL_CULL_FACE GLenum mode; // GL_CULL_FACE_MODE GLenum front_face; // GL_FRONT_FACE } cull; struct { - bool test_enabled; // GL_DEPTH_TEST - GLenum test_func; // GL_DEPTH_FUNC - GLboolean write_mask; // GL_DEPTH_WRITEMASK - GLfloat depth_range_near; // GL_DEPTH_RANGE - GLfloat depth_range_far; // GL_DEPTH_RANGE + bool test_enabled; // GL_DEPTH_TEST + GLenum test_func; // GL_DEPTH_FUNC + GLboolean write_mask; // GL_DEPTH_WRITEMASK } depth; struct { @@ -58,13 +69,14 @@ public: GLuint index; } primitive_restart; // GL_PRIMITIVE_RESTART - struct { + struct ColorMask { GLboolean red_enabled; GLboolean green_enabled; GLboolean blue_enabled; GLboolean alpha_enabled; - } color_mask; // GL_COLOR_WRITEMASK - + }; + std::array<ColorMask, Tegra::Engines::Maxwell3D::Regs::NumRenderTargets> + color_mask; // GL_COLOR_WRITEMASK struct { bool test_enabled; // GL_STENCIL_TEST struct { @@ -78,7 +90,7 @@ public: } front, back; } stencil; - struct { + struct Blend { bool enabled; // GL_BLEND GLenum rgb_equation; // GL_BLEND_EQUATION_RGB GLenum a_equation; // GL_BLEND_EQUATION_ALPHA @@ -86,14 +98,19 @@ public: GLenum dst_rgb_func; // GL_BLEND_DST_RGB GLenum src_a_func; // GL_BLEND_SRC_ALPHA GLenum dst_a_func; // GL_BLEND_DST_ALPHA + }; + std::array<Blend, Tegra::Engines::Maxwell3D::Regs::NumRenderTargets> blend; - struct { - GLclampf red; - GLclampf green; - GLclampf blue; - GLclampf alpha; - } color; // GL_BLEND_COLOR - } blend; + struct { + bool enabled; + } independant_blend; + + struct { + GLclampf red; + GLclampf green; + GLclampf blue; + GLclampf alpha; + } blend_color; // GL_BLEND_COLOR struct { bool enabled; // GL_LOGIC_OP_MODE @@ -138,20 +155,22 @@ public: GLuint program_pipeline; // GL_PROGRAM_PIPELINE_BINDING } draw; - struct { - bool enabled; // GL_SCISSOR_TEST + struct viewport { GLint x; GLint y; - GLsizei width; - GLsizei height; - } scissor; - - struct { - GLint x; - GLint y; - GLsizei width; - GLsizei height; - } viewport; + GLint width; + GLint height; + GLfloat depth_range_near; // GL_DEPTH_RANGE + GLfloat depth_range_far; // GL_DEPTH_RANGE + struct { + bool enabled; // GL_SCISSOR_TEST + GLint x; + GLint y; + GLsizei width; + GLsizei height; + } scissor; + }; + std::array<viewport, Tegra::Engines::Maxwell3D::Regs::NumViewports> viewports; struct { float size; // GL_POINT_SIZE @@ -173,6 +192,10 @@ public: } /// Apply this state as the current OpenGL state void Apply() const; + /// Apply only the state afecting the framebuffer + void ApplyFramebufferState() const; + /// Apply only the state afecting the vertex buffer + void ApplyVertexBufferState() const; /// Set the initial OpenGL state static void ApplyDefaultState(); /// Resets any references to the given resource @@ -183,6 +206,7 @@ public: OpenGLState& ResetBuffer(GLuint handle); OpenGLState& ResetVertexArray(GLuint handle); OpenGLState& ResetFramebuffer(GLuint handle); + void EmulateViewportWithScissor(); private: static OpenGLState cur_state; @@ -191,10 +215,13 @@ private: static bool s_rgb_used; void ApplySRgb() const; void ApplyCulling() const; + void ApplyColorMask() const; void ApplyDepth() const; void ApplyPrimitiveRestart() const; void ApplyStencilTest() const; - void ApplyScissorTest() const; + void ApplyViewport() const; + void ApplyTargetBlending(std::size_t target, bool force) const; + void ApplyGlobalBlending() const; void ApplyBlending() const; void ApplyLogicOp() const; void ApplyTextures() const; diff --git a/src/video_core/renderer_opengl/gl_stream_buffer.cpp b/src/video_core/renderer_opengl/gl_stream_buffer.cpp index e409228cc..b97b895a4 100644 --- a/src/video_core/renderer_opengl/gl_stream_buffer.cpp +++ b/src/video_core/renderer_opengl/gl_stream_buffer.cpp @@ -6,9 +6,13 @@ #include <vector> #include "common/alignment.h" #include "common/assert.h" +#include "common/microprofile.h" #include "video_core/renderer_opengl/gl_state.h" #include "video_core/renderer_opengl/gl_stream_buffer.h" +MICROPROFILE_DEFINE(OpenGL_StreamBuffer, "OpenGL", "Stream Buffer Orphaning", + MP_RGB(128, 128, 192)); + namespace OpenGL { OGLStreamBuffer::OGLStreamBuffer(GLenum target, GLsizeiptr size, bool prefer_coherent) @@ -75,6 +79,7 @@ std::tuple<u8*, GLintptr, bool> OGLStreamBuffer::Map(GLsizeiptr size, GLintptr a } if (invalidate || !persistent) { + MICROPROFILE_SCOPE(OpenGL_StreamBuffer); GLbitfield flags = GL_MAP_WRITE_BIT | (persistent ? GL_MAP_PERSISTENT_BIT : 0) | (coherent ? GL_MAP_COHERENT_BIT : GL_MAP_FLUSH_EXPLICIT_BIT) | (invalidate ? GL_MAP_INVALIDATE_BUFFER_BIT : GL_MAP_UNSYNCHRONIZED_BIT); diff --git a/src/video_core/renderer_opengl/maxwell_to_gl.h b/src/video_core/renderer_opengl/maxwell_to_gl.h index 87d511c38..a8833c06e 100644 --- a/src/video_core/renderer_opengl/maxwell_to_gl.h +++ b/src/video_core/renderer_opengl/maxwell_to_gl.h @@ -159,10 +159,8 @@ inline GLenum TextureFilterMode(Tegra::Texture::TextureFilter filter_mode, } } } - LOG_CRITICAL(Render_OpenGL, "Unimplemented texture filter mode={}", - static_cast<u32>(filter_mode)); - UNREACHABLE(); - return {}; + LOG_ERROR(Render_OpenGL, "Unimplemented texture filter mode={}", static_cast<u32>(filter_mode)); + return GL_LINEAR; } inline GLenum WrapMode(Tegra::Texture::WrapMode wrap_mode) { @@ -182,10 +180,15 @@ inline GLenum WrapMode(Tegra::Texture::WrapMode wrap_mode) { return GL_CLAMP_TO_BORDER; case Tegra::Texture::WrapMode::MirrorOnceClampToEdge: return GL_MIRROR_CLAMP_TO_EDGE; + case Tegra::Texture::WrapMode::MirrorOnceBorder: + if (GL_EXT_texture_mirror_clamp) { + return GL_MIRROR_CLAMP_TO_BORDER_EXT; + } else { + return GL_MIRROR_CLAMP_TO_EDGE; + } } - LOG_CRITICAL(Render_OpenGL, "Unimplemented texture wrap mode={}", static_cast<u32>(wrap_mode)); - UNREACHABLE(); - return {}; + LOG_ERROR(Render_OpenGL, "Unimplemented texture wrap mode={}", static_cast<u32>(wrap_mode)); + return GL_REPEAT; } inline GLenum DepthCompareFunc(Tegra::Texture::DepthCompareFunc func) { @@ -207,28 +210,31 @@ inline GLenum DepthCompareFunc(Tegra::Texture::DepthCompareFunc func) { case Tegra::Texture::DepthCompareFunc::Always: return GL_ALWAYS; } - LOG_CRITICAL(Render_OpenGL, "Unimplemented texture depth compare function ={}", - static_cast<u32>(func)); - UNREACHABLE(); - return {}; + LOG_ERROR(Render_OpenGL, "Unimplemented texture depth compare function ={}", + static_cast<u32>(func)); + return GL_GREATER; } 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_CRITICAL(Render_OpenGL, "Unimplemented blend equation={}", static_cast<u32>(equation)); - UNREACHABLE(); - return {}; + LOG_ERROR(Render_OpenGL, "Unimplemented blend equation={}", static_cast<u32>(equation)); + return GL_FUNC_ADD; } inline GLenum BlendFunc(Maxwell::Blend::Factor factor) { @@ -291,9 +297,8 @@ inline GLenum BlendFunc(Maxwell::Blend::Factor factor) { case Maxwell::Blend::Factor::OneMinusConstantAlphaGL: return GL_ONE_MINUS_CONSTANT_ALPHA; } - LOG_CRITICAL(Render_OpenGL, "Unimplemented blend factor={}", static_cast<u32>(factor)); - UNREACHABLE(); - return {}; + LOG_ERROR(Render_OpenGL, "Unimplemented blend factor={}", static_cast<u32>(factor)); + return GL_ZERO; } inline GLenum SwizzleSource(Tegra::Texture::SwizzleSource source) { @@ -312,9 +317,8 @@ inline GLenum SwizzleSource(Tegra::Texture::SwizzleSource source) { case Tegra::Texture::SwizzleSource::OneFloat: return GL_ONE; } - LOG_CRITICAL(Render_OpenGL, "Unimplemented swizzle source={}", static_cast<u32>(source)); - UNREACHABLE(); - return {}; + LOG_ERROR(Render_OpenGL, "Unimplemented swizzle source={}", static_cast<u32>(source)); + return GL_ZERO; } inline GLenum ComparisonOp(Maxwell::ComparisonOp comparison) { @@ -344,33 +348,39 @@ inline GLenum ComparisonOp(Maxwell::ComparisonOp comparison) { case Maxwell::ComparisonOp::AlwaysOld: return GL_ALWAYS; } - LOG_CRITICAL(Render_OpenGL, "Unimplemented comparison op={}", static_cast<u32>(comparison)); - UNREACHABLE(); - return {}; + LOG_ERROR(Render_OpenGL, "Unimplemented comparison op={}", static_cast<u32>(comparison)); + return GL_ALWAYS; } inline GLenum StencilOp(Maxwell::StencilOp stencil) { switch (stencil) { case Maxwell::StencilOp::Keep: + case Maxwell::StencilOp::KeepOGL: return GL_KEEP; case Maxwell::StencilOp::Zero: + case Maxwell::StencilOp::ZeroOGL: return GL_ZERO; case Maxwell::StencilOp::Replace: + case Maxwell::StencilOp::ReplaceOGL: return GL_REPLACE; case Maxwell::StencilOp::Incr: + case Maxwell::StencilOp::IncrOGL: return GL_INCR; case Maxwell::StencilOp::Decr: + case Maxwell::StencilOp::DecrOGL: return GL_DECR; case Maxwell::StencilOp::Invert: + case Maxwell::StencilOp::InvertOGL: return GL_INVERT; case Maxwell::StencilOp::IncrWrap: + case Maxwell::StencilOp::IncrWrapOGL: return GL_INCR_WRAP; case Maxwell::StencilOp::DecrWrap: + case Maxwell::StencilOp::DecrWrapOGL: return GL_DECR_WRAP; } - LOG_CRITICAL(Render_OpenGL, "Unimplemented stencil op={}", static_cast<u32>(stencil)); - UNREACHABLE(); - return {}; + LOG_ERROR(Render_OpenGL, "Unimplemented stencil op={}", static_cast<u32>(stencil)); + return GL_KEEP; } inline GLenum FrontFace(Maxwell::Cull::FrontFace front_face) { @@ -380,9 +390,8 @@ inline GLenum FrontFace(Maxwell::Cull::FrontFace front_face) { case Maxwell::Cull::FrontFace::CounterClockWise: return GL_CCW; } - LOG_CRITICAL(Render_OpenGL, "Unimplemented front face cull={}", static_cast<u32>(front_face)); - UNREACHABLE(); - return {}; + LOG_ERROR(Render_OpenGL, "Unimplemented front face cull={}", static_cast<u32>(front_face)); + return GL_CCW; } inline GLenum CullFace(Maxwell::Cull::CullFace cull_face) { @@ -394,9 +403,8 @@ inline GLenum CullFace(Maxwell::Cull::CullFace cull_face) { case Maxwell::Cull::CullFace::FrontAndBack: return GL_FRONT_AND_BACK; } - LOG_CRITICAL(Render_OpenGL, "Unimplemented cull face={}", static_cast<u32>(cull_face)); - UNREACHABLE(); - return {}; + LOG_ERROR(Render_OpenGL, "Unimplemented cull face={}", static_cast<u32>(cull_face)); + return GL_BACK; } inline GLenum LogicOp(Maxwell::LogicOperation operation) { @@ -434,9 +442,8 @@ inline GLenum LogicOp(Maxwell::LogicOperation operation) { case Maxwell::LogicOperation::Set: return GL_SET; } - LOG_CRITICAL(Render_OpenGL, "Unimplemented logic operation={}", static_cast<u32>(operation)); - UNREACHABLE(); - return {}; + LOG_ERROR(Render_OpenGL, "Unimplemented logic operation={}", static_cast<u32>(operation)); + return GL_COPY; } } // namespace MaxwellToGL diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp index ea38da932..1492e063a 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.cpp +++ b/src/video_core/renderer_opengl/renderer_opengl.cpp @@ -304,6 +304,12 @@ void RendererOpenGL::ConfigureFramebufferTexture(TextureInfo& texture, gl_framebuffer_data.resize(texture.width * texture.height * 4); break; default: + internal_format = GL_RGBA; + texture.gl_format = GL_RGBA; + texture.gl_type = GL_UNSIGNED_INT_8_8_8_8_REV; + gl_framebuffer_data.resize(texture.width * texture.height * 4); + LOG_CRITICAL(Render_OpenGL, "Unknown framebuffer pixel format: {}", + static_cast<u32>(framebuffer.pixel_format)); UNREACHABLE(); } @@ -484,7 +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/surface.cpp b/src/video_core/surface.cpp index d9a97e30b..9582dd2ca 100644 --- a/src/video_core/surface.cpp +++ b/src/video_core/surface.cpp @@ -19,6 +19,8 @@ SurfaceTarget SurfaceTargetFromTextureType(Tegra::Texture::TextureType texture_t return SurfaceTarget::Texture3D; case Tegra::Texture::TextureType::TextureCubemap: return SurfaceTarget::TextureCubemap; + case Tegra::Texture::TextureType::TextureCubeArray: + return SurfaceTarget::TextureCubeArray; case Tegra::Texture::TextureType::Texture1DArray: return SurfaceTarget::Texture1DArray; case Tegra::Texture::TextureType::Texture2DArray: @@ -39,6 +41,7 @@ bool SurfaceTargetIsLayered(SurfaceTarget target) { case SurfaceTarget::Texture1DArray: case SurfaceTarget::Texture2DArray: case SurfaceTarget::TextureCubemap: + case SurfaceTarget::TextureCubeArray: return true; default: LOG_CRITICAL(HW_GPU, "Unimplemented surface_target={}", static_cast<u32>(target)); @@ -297,10 +300,14 @@ PixelFormat PixelFormatFromTextureFormat(Tegra::Texture::TextureFormat format, return is_srgb ? PixelFormat::ASTC_2D_4X4_SRGB : PixelFormat::ASTC_2D_4X4; case Tegra::Texture::TextureFormat::ASTC_2D_5X4: return is_srgb ? PixelFormat::ASTC_2D_5X4_SRGB : PixelFormat::ASTC_2D_5X4; + case Tegra::Texture::TextureFormat::ASTC_2D_5X5: + return is_srgb ? PixelFormat::ASTC_2D_5X5_SRGB : PixelFormat::ASTC_2D_5X5; case Tegra::Texture::TextureFormat::ASTC_2D_8X8: return is_srgb ? PixelFormat::ASTC_2D_8X8_SRGB : PixelFormat::ASTC_2D_8X8; case Tegra::Texture::TextureFormat::ASTC_2D_8X5: return is_srgb ? PixelFormat::ASTC_2D_8X5_SRGB : PixelFormat::ASTC_2D_8X5; + case Tegra::Texture::TextureFormat::ASTC_2D_10X8: + return is_srgb ? PixelFormat::ASTC_2D_10X8_SRGB : PixelFormat::ASTC_2D_10X8; case Tegra::Texture::TextureFormat::R16_G16: switch (component_type) { case Tegra::Texture::ComponentType::FLOAT: @@ -440,12 +447,16 @@ bool IsPixelFormatASTC(PixelFormat format) { switch (format) { case PixelFormat::ASTC_2D_4X4: case PixelFormat::ASTC_2D_5X4: + case PixelFormat::ASTC_2D_5X5: case PixelFormat::ASTC_2D_8X8: case PixelFormat::ASTC_2D_8X5: case PixelFormat::ASTC_2D_4X4_SRGB: case PixelFormat::ASTC_2D_5X4_SRGB: + case PixelFormat::ASTC_2D_5X5_SRGB: case PixelFormat::ASTC_2D_8X8_SRGB: case PixelFormat::ASTC_2D_8X5_SRGB: + case PixelFormat::ASTC_2D_10X8: + case PixelFormat::ASTC_2D_10X8_SRGB: return true; default: return false; @@ -453,27 +464,7 @@ bool IsPixelFormatASTC(PixelFormat format) { } std::pair<u32, u32> GetASTCBlockSize(PixelFormat format) { - switch (format) { - case PixelFormat::ASTC_2D_4X4: - return {4, 4}; - case PixelFormat::ASTC_2D_5X4: - return {5, 4}; - case PixelFormat::ASTC_2D_8X8: - return {8, 8}; - case PixelFormat::ASTC_2D_8X5: - return {8, 5}; - case PixelFormat::ASTC_2D_4X4_SRGB: - return {4, 4}; - case PixelFormat::ASTC_2D_5X4_SRGB: - return {5, 4}; - case PixelFormat::ASTC_2D_8X8_SRGB: - return {8, 8}; - case PixelFormat::ASTC_2D_8X5_SRGB: - return {8, 5}; - default: - LOG_CRITICAL(HW_GPU, "Unhandled format: {}", static_cast<u32>(format)); - UNREACHABLE(); - } + return {GetDefaultBlockWidth(format), GetDefaultBlockHeight(format)}; } bool IsFormatBCn(PixelFormat format) { diff --git a/src/video_core/surface.h b/src/video_core/surface.h index 3232e437f..0dd3eb2e4 100644 --- a/src/video_core/surface.h +++ b/src/video_core/surface.h @@ -72,19 +72,23 @@ enum class PixelFormat { ASTC_2D_8X8_SRGB = 54, ASTC_2D_8X5_SRGB = 55, ASTC_2D_5X4_SRGB = 56, + ASTC_2D_5X5 = 57, + ASTC_2D_5X5_SRGB = 58, + ASTC_2D_10X8 = 59, + ASTC_2D_10X8_SRGB = 60, MaxColorFormat, // Depth formats - Z32F = 57, - Z16 = 58, + Z32F = 61, + Z16 = 62, MaxDepthFormat, // DepthStencil formats - Z24S8 = 59, - S8Z24 = 60, - Z32FS8 = 61, + Z24S8 = 63, + S8Z24 = 64, + Z32FS8 = 65, MaxDepthStencilFormat, @@ -118,6 +122,7 @@ enum class SurfaceTarget { Texture1DArray, Texture2DArray, TextureCubemap, + TextureCubeArray, }; /** @@ -188,6 +193,10 @@ static constexpr u32 GetCompressionFactor(PixelFormat format) { 4, // ASTC_2D_8X8_SRGB 4, // ASTC_2D_8X5_SRGB 4, // ASTC_2D_5X4_SRGB + 4, // ASTC_2D_5X5 + 4, // ASTC_2D_5X5_SRGB + 4, // ASTC_2D_10X8 + 4, // ASTC_2D_10X8_SRGB 1, // Z32F 1, // Z16 1, // Z24S8 @@ -199,6 +208,81 @@ static constexpr u32 GetCompressionFactor(PixelFormat format) { return compression_factor_table[static_cast<std::size_t>(format)]; } +static constexpr u32 GetDefaultBlockWidth(PixelFormat format) { + if (format == PixelFormat::Invalid) + return 0; + constexpr std::array<u32, MaxPixelFormat> block_width_table = {{ + 1, // ABGR8U + 1, // ABGR8S + 1, // ABGR8UI + 1, // B5G6R5U + 1, // A2B10G10R10U + 1, // A1B5G5R5U + 1, // R8U + 1, // R8UI + 1, // RGBA16F + 1, // RGBA16U + 1, // RGBA16UI + 1, // R11FG11FB10F + 1, // RGBA32UI + 4, // DXT1 + 4, // DXT23 + 4, // DXT45 + 4, // DXN1 + 4, // DXN2UNORM + 4, // DXN2SNORM + 4, // BC7U + 4, // BC6H_UF16 + 4, // BC6H_SF16 + 4, // ASTC_2D_4X4 + 1, // G8R8U + 1, // G8R8S + 1, // BGRA8 + 1, // RGBA32F + 1, // RG32F + 1, // R32F + 1, // R16F + 1, // R16U + 1, // R16S + 1, // R16UI + 1, // R16I + 1, // RG16 + 1, // RG16F + 1, // RG16UI + 1, // RG16I + 1, // RG16S + 1, // RGB32F + 1, // RGBA8_SRGB + 1, // RG8U + 1, // RG8S + 1, // RG32UI + 1, // R32UI + 8, // ASTC_2D_8X8 + 8, // ASTC_2D_8X5 + 5, // ASTC_2D_5X4 + 1, // BGRA8_SRGB + 4, // DXT1_SRGB + 4, // DXT23_SRGB + 4, // DXT45_SRGB + 4, // BC7U_SRGB + 4, // ASTC_2D_4X4_SRGB + 8, // ASTC_2D_8X8_SRGB + 8, // ASTC_2D_8X5_SRGB + 5, // ASTC_2D_5X4_SRGB + 5, // ASTC_2D_5X5 + 5, // ASTC_2D_5X5_SRGB + 10, // ASTC_2D_10X8 + 10, // ASTC_2D_10X8_SRGB + 1, // Z32F + 1, // Z16 + 1, // Z24S8 + 1, // S8Z24 + 1, // Z32FS8 + }}; + ASSERT(static_cast<std::size_t>(format) < block_width_table.size()); + return block_width_table[static_cast<std::size_t>(format)]; +} + static constexpr u32 GetDefaultBlockHeight(PixelFormat format) { if (format == PixelFormat::Invalid) return 0; @@ -261,6 +345,10 @@ static constexpr u32 GetDefaultBlockHeight(PixelFormat format) { 8, // ASTC_2D_8X8_SRGB 5, // ASTC_2D_8X5_SRGB 4, // ASTC_2D_5X4_SRGB + 5, // ASTC_2D_5X5 + 5, // ASTC_2D_5X5_SRGB + 8, // ASTC_2D_10X8 + 8, // ASTC_2D_10X8_SRGB 1, // Z32F 1, // Z16 1, // Z24S8 @@ -299,7 +387,7 @@ static constexpr u32 GetFormatBpp(PixelFormat format) { 128, // BC7U 128, // BC6H_UF16 128, // BC6H_SF16 - 32, // ASTC_2D_4X4 + 128, // ASTC_2D_4X4 16, // G8R8U 16, // G8R8S 32, // BGRA8 @@ -322,18 +410,22 @@ static constexpr u32 GetFormatBpp(PixelFormat format) { 16, // RG8S 64, // RG32UI 32, // R32UI - 16, // ASTC_2D_8X8 - 16, // ASTC_2D_8X5 - 32, // ASTC_2D_5X4 + 128, // ASTC_2D_8X8 + 128, // ASTC_2D_8X5 + 128, // ASTC_2D_5X4 32, // BGRA8_SRGB 64, // DXT1_SRGB 128, // DXT23_SRGB 128, // DXT45_SRGB 128, // BC7U - 32, // ASTC_2D_4X4_SRGB - 16, // ASTC_2D_8X8_SRGB - 16, // ASTC_2D_8X5_SRGB - 32, // ASTC_2D_5X4_SRGB + 128, // ASTC_2D_4X4_SRGB + 128, // ASTC_2D_8X8_SRGB + 128, // ASTC_2D_8X5_SRGB + 128, // ASTC_2D_5X4_SRGB + 128, // ASTC_2D_5X5 + 128, // ASTC_2D_5X5_SRGB + 128, // ASTC_2D_10X8 + 128, // ASTC_2D_10X8_SRGB 32, // Z32F 16, // Z16 32, // Z24S8 diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp index b1feacae9..bc50a4876 100644 --- a/src/video_core/textures/astc.cpp +++ b/src/video_core/textures/astc.cpp @@ -1598,27 +1598,29 @@ static void DecompressBlock(uint8_t inBuf[16], const uint32_t blockWidth, namespace Tegra::Texture::ASTC { std::vector<uint8_t> Decompress(std::vector<uint8_t>& data, uint32_t width, uint32_t height, - uint32_t block_width, uint32_t block_height) { + uint32_t depth, uint32_t block_width, uint32_t block_height) { uint32_t blockIdx = 0; - std::vector<uint8_t> outData(height * width * 4); - for (uint32_t j = 0; j < height; j += block_height) { - for (uint32_t i = 0; i < width; i += block_width) { + std::vector<uint8_t> outData(height * width * depth * 4); + for (uint32_t k = 0; k < depth; k++) { + for (uint32_t j = 0; j < height; j += block_height) { + for (uint32_t i = 0; i < width; i += block_width) { - uint8_t* blockPtr = data.data() + blockIdx * 16; + uint8_t* blockPtr = data.data() + blockIdx * 16; - // Blocks can be at most 12x12 - uint32_t uncompData[144]; - ASTCC::DecompressBlock(blockPtr, block_width, block_height, uncompData); + // Blocks can be at most 12x12 + uint32_t uncompData[144]; + ASTCC::DecompressBlock(blockPtr, block_width, block_height, uncompData); - uint32_t decompWidth = std::min(block_width, width - i); - uint32_t decompHeight = std::min(block_height, height - j); + uint32_t decompWidth = std::min(block_width, width - i); + uint32_t decompHeight = std::min(block_height, height - j); - uint8_t* outRow = outData.data() + (j * width + i) * 4; - for (uint32_t jj = 0; jj < decompHeight; jj++) { - memcpy(outRow + jj * width * 4, uncompData + jj * block_width, decompWidth * 4); - } + uint8_t* outRow = outData.data() + (j * width + i) * 4; + for (uint32_t jj = 0; jj < decompHeight; jj++) { + memcpy(outRow + jj * width * 4, uncompData + jj * block_width, decompWidth * 4); + } - blockIdx++; + blockIdx++; + } } } diff --git a/src/video_core/textures/astc.h b/src/video_core/textures/astc.h index f0d7c0e56..d419dd025 100644 --- a/src/video_core/textures/astc.h +++ b/src/video_core/textures/astc.h @@ -10,6 +10,6 @@ namespace Tegra::Texture::ASTC { std::vector<uint8_t> Decompress(std::vector<uint8_t>& data, uint32_t width, uint32_t height, - uint32_t block_width, uint32_t block_height); + uint32_t depth, uint32_t block_width, uint32_t block_height); } // namespace Tegra::Texture::ASTC diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp index 550ca856c..7eabd34f1 100644 --- a/src/video_core/textures/decoders.cpp +++ b/src/video_core/textures/decoders.cpp @@ -37,25 +37,28 @@ struct alignas(64) SwizzleTable { std::array<std::array<u16, M>, N> values{}; }; -constexpr auto legacy_swizzle_table = SwizzleTable<8, 64, 1>(); -constexpr auto fast_swizzle_table = SwizzleTable<8, 4, 16>(); +constexpr u32 gob_size_x = 64; +constexpr u32 gob_size_y = 8; +constexpr u32 gob_size_z = 1; +constexpr u32 gob_size = gob_size_x * gob_size_y * gob_size_z; +constexpr u32 fast_swizzle_align = 16; + +constexpr auto legacy_swizzle_table = SwizzleTable<gob_size_y, gob_size_x, gob_size_z>(); +constexpr auto fast_swizzle_table = SwizzleTable<gob_size_y, 4, fast_swizzle_align>(); /** * This function manages ALL the GOBs(Group of Bytes) Inside a single block. * Instead of going gob by gob, we map the coordinates inside a block and manage from * those. Block_Width is assumed to be 1. */ -void PreciseProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle, +void PreciseProcessBlock(u8* const swizzled_data, u8* const unswizzled_data, const bool unswizzle, const u32 x_start, const u32 y_start, const u32 z_start, const u32 x_end, const u32 y_end, const u32 z_end, const u32 tile_offset, const u32 xy_block_size, const u32 layer_z, const u32 stride_x, const u32 bytes_per_pixel, const u32 out_bytes_per_pixel) { std::array<u8*, 2> data_ptrs; u32 z_address = tile_offset; - const u32 gob_size_x = 64; - const u32 gob_size_y = 8; - const u32 gob_size_z = 1; - const u32 gob_size = gob_size_x * gob_size_y * gob_size_z; + for (u32 z = z_start; z < z_end; z++) { u32 y_address = z_address; u32 pixel_base = layer_z * z + y_start * stride_x; @@ -81,7 +84,7 @@ void PreciseProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unsw * Instead of going gob by gob, we map the coordinates inside a block and manage from * those. Block_Width is assumed to be 1. */ -void FastProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle, +void FastProcessBlock(u8* const swizzled_data, u8* const unswizzled_data, const bool unswizzle, const u32 x_start, const u32 y_start, const u32 z_start, const u32 x_end, const u32 y_end, const u32 z_end, const u32 tile_offset, const u32 xy_block_size, const u32 layer_z, const u32 stride_x, @@ -90,23 +93,19 @@ void FastProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswizz u32 z_address = tile_offset; const u32 x_startb = x_start * bytes_per_pixel; const u32 x_endb = x_end * bytes_per_pixel; - const u32 copy_size = 16; - const u32 gob_size_x = 64; - const u32 gob_size_y = 8; - const u32 gob_size_z = 1; - const u32 gob_size = gob_size_x * gob_size_y * gob_size_z; + for (u32 z = z_start; z < z_end; z++) { u32 y_address = z_address; u32 pixel_base = layer_z * z + y_start * stride_x; for (u32 y = y_start; y < y_end; y++) { const auto& table = fast_swizzle_table[y % gob_size_y]; - for (u32 xb = x_startb; xb < x_endb; xb += copy_size) { - const u32 swizzle_offset{y_address + table[(xb / copy_size) % 4]}; + for (u32 xb = x_startb; xb < x_endb; xb += fast_swizzle_align) { + const u32 swizzle_offset{y_address + table[(xb / fast_swizzle_align) % 4]}; const u32 out_x = xb * out_bytes_per_pixel / bytes_per_pixel; const u32 pixel_index{out_x + pixel_base}; data_ptrs[unswizzle] = swizzled_data + swizzle_offset; data_ptrs[!unswizzle] = unswizzled_data + pixel_index; - std::memcpy(data_ptrs[0], data_ptrs[1], copy_size); + std::memcpy(data_ptrs[0], data_ptrs[1], fast_swizzle_align); } pixel_base += stride_x; if ((y + 1) % gob_size_y == 0) @@ -126,23 +125,21 @@ void FastProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswizz * https://envytools.readthedocs.io/en/latest/hw/memory/g80-surface.html#blocklinear-surfaces */ template <bool fast> -void SwizzledData(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle, const u32 width, - const u32 height, const u32 depth, const u32 bytes_per_pixel, +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) { 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; - const u32 gob_x_bytes = 64; - const u32 gob_elements_x = gob_x_bytes / bytes_per_pixel; - const u32 gob_elements_y = 8; - const u32 gob_elements_z = 1; + const u32 gob_elements_x = gob_size_x / bytes_per_pixel; + constexpr u32 gob_elements_y = gob_size_y; + constexpr u32 gob_elements_z = gob_size_z; 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 blocks_on_y = div_ceil(height, block_y_elements); const u32 blocks_on_z = div_ceil(depth, block_z_elements); - const u32 gob_size = gob_x_bytes * gob_elements_y * gob_elements_z; const u32 xy_block_size = gob_size * block_height; const u32 block_size = xy_block_size * block_depth; u32 tile_offset = 0; @@ -171,9 +168,9 @@ void SwizzledData(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle, } void CopySwizzledData(u32 width, u32 height, u32 depth, u32 bytes_per_pixel, - u32 out_bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, + u32 out_bytes_per_pixel, u8* const swizzled_data, u8* const unswizzled_data, bool unswizzle, u32 block_height, u32 block_depth) { - if (bytes_per_pixel % 3 != 0 && (width * bytes_per_pixel) % 16 == 0) { + 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); } else { @@ -202,6 +199,8 @@ u32 BytesPerPixel(TextureFormat format) { case TextureFormat::ASTC_2D_5X4: case TextureFormat::ASTC_2D_8X8: case TextureFormat::ASTC_2D_8X5: + case TextureFormat::ASTC_2D_10X8: + case TextureFormat::ASTC_2D_5X5: case TextureFormat::A8R8G8B8: case TextureFormat::A2B10G10R10: case TextureFormat::BF10GF11RF11: @@ -227,27 +226,38 @@ u32 BytesPerPixel(TextureFormat format) { } } -std::vector<u8> UnswizzleTexture(VAddr address, u32 tile_size, u32 bytes_per_pixel, u32 width, - u32 height, u32 depth, u32 block_height, u32 block_depth) { +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) { + 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); +} + +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) { std::vector<u8> unswizzled_data(width * height * depth * bytes_per_pixel); - CopySwizzledData(width / tile_size, height / tile_size, depth, bytes_per_pixel, bytes_per_pixel, - Memory::GetPointer(address), unswizzled_data.data(), true, block_height, - block_depth); + UnswizzleTexture(unswizzled_data.data(), address, tile_size_x, tile_size_y, bytes_per_pixel, + width, height, depth, block_height, block_depth); return unswizzled_data; } void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32 swizzled_width, u32 bytes_per_pixel, VAddr swizzled_data, VAddr unswizzled_data, u32 block_height) { - const u32 image_width_in_gobs{(swizzled_width * bytes_per_pixel + 63) / 64}; + const u32 image_width_in_gobs{(swizzled_width * bytes_per_pixel + (gob_size_x - 1)) / + gob_size_x}; for (u32 line = 0; line < subrect_height; ++line) { const u32 gob_address_y = - (line / (8 * block_height)) * 512 * block_height * image_width_in_gobs + - (line % (8 * block_height) / 8) * 512; - const auto& table = legacy_swizzle_table[line % 8]; + (line / (gob_size_y * block_height)) * gob_size * block_height * image_width_in_gobs + + ((line % (gob_size_y * block_height)) / gob_size_y) * gob_size; + const auto& table = legacy_swizzle_table[line % gob_size_y]; for (u32 x = 0; x < subrect_width; ++x) { - const u32 gob_address = gob_address_y + (x * bytes_per_pixel / 64) * 512 * block_height; - const u32 swizzled_offset = gob_address + table[(x * bytes_per_pixel) % 64]; + const u32 gob_address = + gob_address_y + (x * bytes_per_pixel / gob_size_x) * gob_size * block_height; + const u32 swizzled_offset = gob_address + table[(x * bytes_per_pixel) % gob_size_x]; const VAddr source_line = unswizzled_data + line * source_pitch + x * bytes_per_pixel; const VAddr dest_addr = swizzled_data + swizzled_offset; @@ -261,13 +271,13 @@ void UnswizzleSubrect(u32 subrect_width, u32 subrect_height, u32 dest_pitch, u32 u32 block_height, u32 offset_x, u32 offset_y) { for (u32 line = 0; line < subrect_height; ++line) { const u32 y2 = line + offset_y; - const u32 gob_address_y = - (y2 / (8 * block_height)) * 512 * block_height + (y2 % (8 * block_height) / 8) * 512; - const auto& table = legacy_swizzle_table[y2 % 8]; + const u32 gob_address_y = (y2 / (gob_size_y * block_height)) * gob_size * block_height + + ((y2 % (gob_size_y * block_height)) / gob_size_y) * gob_size; + const auto& table = legacy_swizzle_table[y2 % gob_size_y]; for (u32 x = 0; x < subrect_width; ++x) { const u32 x2 = (x + offset_x) * bytes_per_pixel; - const u32 gob_address = gob_address_y + (x2 / 64) * 512 * block_height; - const u32 swizzled_offset = gob_address + table[x2 % 64]; + const u32 gob_address = gob_address_y + (x2 / gob_size_x) * gob_size * block_height; + const u32 swizzled_offset = gob_address + table[x2 % gob_size_x]; const VAddr dest_line = unswizzled_data + line * dest_pitch + x * bytes_per_pixel; const VAddr source_addr = swizzled_data + swizzled_offset; @@ -292,6 +302,8 @@ std::vector<u8> DecodeTexture(const std::vector<u8>& texture_data, TextureFormat case TextureFormat::BC6H_SF16: case TextureFormat::ASTC_2D_4X4: case TextureFormat::ASTC_2D_8X8: + case TextureFormat::ASTC_2D_5X5: + case TextureFormat::ASTC_2D_10X8: case TextureFormat::A8R8G8B8: case TextureFormat::A2B10G10R10: case TextureFormat::A1B5G5R5: @@ -319,12 +331,9 @@ std::vector<u8> DecodeTexture(const std::vector<u8>& texture_data, TextureFormat std::size_t CalculateSize(bool tiled, u32 bytes_per_pixel, u32 width, u32 height, u32 depth, u32 block_height, u32 block_depth) { if (tiled) { - const u32 gobs_in_x = 64; - const u32 gobs_in_y = 8; - const u32 gobs_in_z = 1; - const u32 aligned_width = Common::AlignUp(width * bytes_per_pixel, gobs_in_x); - const u32 aligned_height = Common::AlignUp(height, gobs_in_y * block_height); - const u32 aligned_depth = Common::AlignUp(depth, gobs_in_z * block_depth); + const u32 aligned_width = Common::AlignUp(width * bytes_per_pixel, gob_size_x); + const u32 aligned_height = Common::AlignUp(height, gob_size_y * block_height); + const u32 aligned_depth = Common::AlignUp(depth, gob_size_z * block_depth); return aligned_width * aligned_height * aligned_depth; } else { return width * height * depth * bytes_per_pixel; diff --git a/src/video_core/textures/decoders.h b/src/video_core/textures/decoders.h index b390219e4..f4ef7c73e 100644 --- a/src/video_core/textures/decoders.h +++ b/src/video_core/textures/decoders.h @@ -19,8 +19,15 @@ inline std::size_t GetGOBSize() { /** * Unswizzles a swizzled texture without changing its format. */ -std::vector<u8> UnswizzleTexture(VAddr address, u32 tile_size, u32 bytes_per_pixel, u32 width, - u32 height, u32 depth, +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); +/** + * 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); diff --git a/src/video_core/textures/texture.h b/src/video_core/textures/texture.h index d12d2ecb8..ffa08f5c1 100644 --- a/src/video_core/textures/texture.h +++ b/src/video_core/textures/texture.h @@ -168,20 +168,30 @@ struct TICEntry { // High 16 bits of the pitch value BitField<0, 16, u32> pitch_high; - + BitField<26, 1, u32> use_header_opt_control; + BitField<27, 1, u32> depth_texture; BitField<28, 4, u32> max_mip_level; }; union { BitField<0, 16, u32> width_minus_1; BitField<22, 1, u32> srgb_conversion; BitField<23, 4, TextureType> texture_type; + BitField<29, 3, u32> border_size; }; union { BitField<0, 16, u32> height_minus_1; BitField<16, 15, u32> depth_minus_1; }; + union { + BitField<6, 13, u32> mip_lod_bias; + BitField<27, 3, u32> max_anisotropy; + }; - INSERT_PADDING_BYTES(8); + union { + BitField<0, 4, u32> res_min_mip_level; + BitField<4, 4, u32> res_max_mip_level; + BitField<12, 12, u32> min_lod_clamp; + }; GPUVAddr Address() const { return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) | address_low); @@ -275,13 +285,25 @@ struct TSCEntry { BitField<6, 3, WrapMode> wrap_p; BitField<9, 1, u32> depth_compare_enabled; BitField<10, 3, DepthCompareFunc> depth_compare_func; + BitField<13, 1, u32> srgb_conversion; + BitField<20, 3, u32> max_anisotropy; }; union { BitField<0, 2, TextureFilter> mag_filter; BitField<4, 2, TextureFilter> min_filter; BitField<6, 2, TextureMipmapFilter> mip_filter; + BitField<9, 1, u32> cubemap_interface_filtering; + BitField<12, 13, u32> mip_lod_bias; + }; + union { + BitField<0, 12, u32> min_lod_clamp; + BitField<12, 12, u32> max_lod_clamp; + BitField<24, 8, u32> srgb_border_color_r; + }; + union { + BitField<12, 8, u32> srgb_border_color_g; + BitField<20, 8, u32> srgb_border_color_b; }; - INSERT_PADDING_BYTES(8); float border_color_r; float border_color_g; float border_color_b; |