From 642b9b503040f7da02dcb2c52f3cd4cbf6fee4b2 Mon Sep 17 00:00:00 2001 From: bunnei Date: Tue, 21 Jul 2015 19:04:05 -0400 Subject: GPU: Refactor "VertexShader" namespace to "Shader". - Also renames "vertex_shader.*" to "shader_interpreter.*" --- src/citra_qt/debugger/graphics_vertex_shader.cpp | 2 +- src/video_core/CMakeLists.txt | 4 +- src/video_core/clipper.cpp | 2 +- src/video_core/clipper.h | 4 +- src/video_core/command_processor.cpp | 19 +- src/video_core/hwrasterizer_base.h | 8 +- src/video_core/primitive_assembly.cpp | 4 +- src/video_core/primitive_assembly.h | 2 +- src/video_core/rasterizer.cpp | 14 +- src/video_core/rasterizer.h | 8 +- src/video_core/renderer_opengl/gl_rasterizer.cpp | 6 +- src/video_core/renderer_opengl/gl_rasterizer.h | 10 +- src/video_core/shader_interpreter.cpp | 629 +++++++++++++++++++++++ src/video_core/shader_interpreter.h | 72 +++ src/video_core/vertex_shader.cpp | 629 ----------------------- src/video_core/vertex_shader.h | 73 --- 16 files changed, 742 insertions(+), 744 deletions(-) create mode 100644 src/video_core/shader_interpreter.cpp create mode 100644 src/video_core/shader_interpreter.h delete mode 100644 src/video_core/vertex_shader.cpp delete mode 100644 src/video_core/vertex_shader.h diff --git a/src/citra_qt/debugger/graphics_vertex_shader.cpp b/src/citra_qt/debugger/graphics_vertex_shader.cpp index f42a2f4ce..566a986f5 100644 --- a/src/citra_qt/debugger/graphics_vertex_shader.cpp +++ b/src/citra_qt/debugger/graphics_vertex_shader.cpp @@ -8,7 +8,7 @@ #include #include -#include "video_core/vertex_shader.h" +#include "video_core/shader_interpreter.h" #include "graphics_vertex_shader.h" diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt index 162108301..e06d368e5 100644 --- a/src/video_core/CMakeLists.txt +++ b/src/video_core/CMakeLists.txt @@ -11,8 +11,8 @@ set(SRCS pica.cpp primitive_assembly.cpp rasterizer.cpp + shader_interpreter.cpp utils.cpp - vertex_shader.cpp video_core.cpp ) @@ -35,8 +35,8 @@ set(HEADERS primitive_assembly.h rasterizer.h renderer_base.h + shader_interpreter.h utils.h - vertex_shader.h video_core.h ) diff --git a/src/video_core/clipper.cpp b/src/video_core/clipper.cpp index 558b49d60..e397ca2e8 100644 --- a/src/video_core/clipper.cpp +++ b/src/video_core/clipper.cpp @@ -7,7 +7,7 @@ #include "clipper.h" #include "pica.h" #include "rasterizer.h" -#include "vertex_shader.h" +#include "shader_interpreter.h" namespace Pica { diff --git a/src/video_core/clipper.h b/src/video_core/clipper.h index 19ce8e140..6ed01e877 100644 --- a/src/video_core/clipper.h +++ b/src/video_core/clipper.h @@ -6,13 +6,13 @@ namespace Pica { -namespace VertexShader { +namespace Shader { struct OutputVertex; } namespace Clipper { -using VertexShader::OutputVertex; +using Shader::OutputVertex; void ProcessTriangle(OutputVertex& v0, OutputVertex& v1, OutputVertex& v2); diff --git a/src/video_core/command_processor.cpp b/src/video_core/command_processor.cpp index 243abe842..e199424c3 100644 --- a/src/video_core/command_processor.cpp +++ b/src/video_core/command_processor.cpp @@ -18,7 +18,7 @@ #include "pica.h" #include "primitive_assembly.h" #include "renderer_base.h" -#include "vertex_shader.h" +#include "shader_interpreter.h" #include "video_core.h" namespace Pica { @@ -165,7 +165,7 @@ static inline void WritePicaReg(u32 id, u32 value, u32 mask) { DebugUtils::GeometryDumper geometry_dumper; PrimitiveAssembler dumping_primitive_assembler(regs.triangle_topology.Value()); #endif - PrimitiveAssembler primitive_assembler(regs.triangle_topology.Value()); + PrimitiveAssembler primitive_assembler(regs.triangle_topology.Value()); if (g_debug_context) { for (int i = 0; i < 3; ++i) { @@ -210,7 +210,7 @@ static inline void WritePicaReg(u32 id, u32 value, u32 mask) { // The size has been tuned for optimal balance between hit-rate and the cost of lookup const size_t VERTEX_CACHE_SIZE = 32; std::array vertex_cache_ids; - std::array vertex_cache; + std::array vertex_cache; unsigned int vertex_cache_pos = 0; vertex_cache_ids.fill(-1); @@ -224,7 +224,7 @@ static inline void WritePicaReg(u32 id, u32 value, u32 mask) { ASSERT(vertex != -1); bool vertex_cache_hit = false; - VertexShader::OutputVertex output; + Shader::OutputVertex output; if (is_indexed) { if (g_debug_context && Pica::g_debug_context->recorder) { @@ -243,7 +243,7 @@ static inline void WritePicaReg(u32 id, u32 value, u32 mask) { if (!vertex_cache_hit) { // Initialize data for the current vertex - VertexShader::InputVertex input; + Shader::InputVertex input; for (int i = 0; i < attribute_config.GetNumTotalAttributes(); ++i) { if (vertex_attribute_elements[i] != 0) { @@ -306,9 +306,8 @@ static inline void WritePicaReg(u32 id, u32 value, u32 mask) { std::bind(&DebugUtils::GeometryDumper::AddTriangle, &geometry_dumper, _1, _2, _3)); #endif - // Send to vertex shader - output = VertexShader::RunShader(input, attribute_config.GetNumTotalAttributes(), g_state.regs.vs, g_state.vs); + output = Shader::RunShader(input, attribute_config.GetNumTotalAttributes(), g_state.regs.vs, g_state.vs); if (is_indexed) { vertex_cache[vertex_cache_pos] = output; @@ -319,9 +318,9 @@ static inline void WritePicaReg(u32 id, u32 value, u32 mask) { if (Settings::values.use_hw_renderer) { // Send to hardware renderer - static auto AddHWTriangle = [](const Pica::VertexShader::OutputVertex& v0, - const Pica::VertexShader::OutputVertex& v1, - const Pica::VertexShader::OutputVertex& v2) { + static auto AddHWTriangle = [](const Pica::Shader::OutputVertex& v0, + const Pica::Shader::OutputVertex& v1, + const Pica::Shader::OutputVertex& v2) { VideoCore::g_renderer->hw_rasterizer->AddTriangle(v0, v1, v2); }; diff --git a/src/video_core/hwrasterizer_base.h b/src/video_core/hwrasterizer_base.h index c8746c608..54b8892fb 100644 --- a/src/video_core/hwrasterizer_base.h +++ b/src/video_core/hwrasterizer_base.h @@ -7,7 +7,7 @@ #include "common/common_types.h" namespace Pica { -namespace VertexShader { +namespace Shader { struct OutputVertex; } } @@ -24,9 +24,9 @@ public: virtual void Reset() = 0; /// Queues the primitive formed by the given vertices for rendering - virtual void AddTriangle(const Pica::VertexShader::OutputVertex& v0, - const Pica::VertexShader::OutputVertex& v1, - const Pica::VertexShader::OutputVertex& v2) = 0; + virtual void AddTriangle(const Pica::Shader::OutputVertex& v0, + const Pica::Shader::OutputVertex& v1, + const Pica::Shader::OutputVertex& v2) = 0; /// Draw the current batch of triangles virtual void DrawTriangles() = 0; diff --git a/src/video_core/primitive_assembly.cpp b/src/video_core/primitive_assembly.cpp index 2f22bdcce..e15a1daba 100644 --- a/src/video_core/primitive_assembly.cpp +++ b/src/video_core/primitive_assembly.cpp @@ -4,7 +4,7 @@ #include "pica.h" #include "primitive_assembly.h" -#include "vertex_shader.h" +#include "shader_interpreter.h" #include "common/logging/log.h" #include "video_core/debug_utils/debug_utils.h" @@ -56,7 +56,7 @@ void PrimitiveAssembler::SubmitVertex(VertexType& vtx, TriangleHandl // explicitly instantiate use cases template -struct PrimitiveAssembler; +struct PrimitiveAssembler; template struct PrimitiveAssembler; diff --git a/src/video_core/primitive_assembly.h b/src/video_core/primitive_assembly.h index 52ff4cd89..0de0b8810 100644 --- a/src/video_core/primitive_assembly.h +++ b/src/video_core/primitive_assembly.h @@ -8,7 +8,7 @@ #include "video_core/pica.h" -#include "video_core/vertex_shader.h" +#include "video_core/shader_interpreter.h" namespace Pica { diff --git a/src/video_core/rasterizer.cpp b/src/video_core/rasterizer.cpp index 68b7cc05d..4f94313df 100644 --- a/src/video_core/rasterizer.cpp +++ b/src/video_core/rasterizer.cpp @@ -16,7 +16,7 @@ #include "math.h" #include "pica.h" #include "rasterizer.h" -#include "vertex_shader.h" +#include "shader_interpreter.h" #include "video_core/utils.h" namespace Pica { @@ -272,9 +272,9 @@ static Common::Profiling::TimingCategory rasterization_category("Rasterization") * Helper function for ProcessTriangle with the "reversed" flag to allow for implementing * culling via recursion. */ -static void ProcessTriangleInternal(const VertexShader::OutputVertex& v0, - const VertexShader::OutputVertex& v1, - const VertexShader::OutputVertex& v2, +static void ProcessTriangleInternal(const Shader::OutputVertex& v0, + const Shader::OutputVertex& v1, + const Shader::OutputVertex& v2, bool reversed = false) { const auto& regs = g_state.regs; @@ -1107,9 +1107,9 @@ static void ProcessTriangleInternal(const VertexShader::OutputVertex& v0, } } -void ProcessTriangle(const VertexShader::OutputVertex& v0, - const VertexShader::OutputVertex& v1, - const VertexShader::OutputVertex& v2) { +void ProcessTriangle(const Shader::OutputVertex& v0, + const Shader::OutputVertex& v1, + const Shader::OutputVertex& v2) { ProcessTriangleInternal(v0, v1, v2); } diff --git a/src/video_core/rasterizer.h b/src/video_core/rasterizer.h index 42148f8b1..a6a9634b4 100644 --- a/src/video_core/rasterizer.h +++ b/src/video_core/rasterizer.h @@ -6,15 +6,15 @@ namespace Pica { -namespace VertexShader { +namespace Shader { struct OutputVertex; } namespace Rasterizer { -void ProcessTriangle(const VertexShader::OutputVertex& v0, - const VertexShader::OutputVertex& v1, - const VertexShader::OutputVertex& v2); +void ProcessTriangle(const Shader::OutputVertex& v0, + const Shader::OutputVertex& v1, + const Shader::OutputVertex& v2); } // namespace Rasterizer diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index e7c1cfeb7..9f1552adf 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -202,9 +202,9 @@ void RasterizerOpenGL::Reset() { res_cache.FullFlush(); } -void RasterizerOpenGL::AddTriangle(const Pica::VertexShader::OutputVertex& v0, - const Pica::VertexShader::OutputVertex& v1, - const Pica::VertexShader::OutputVertex& v2) { +void RasterizerOpenGL::AddTriangle(const Pica::Shader::OutputVertex& v0, + const Pica::Shader::OutputVertex& v1, + const Pica::Shader::OutputVertex& v2) { vertex_batch.push_back(HardwareVertex(v0)); vertex_batch.push_back(HardwareVertex(v1)); vertex_batch.push_back(HardwareVertex(v2)); diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index ae7b26fc6..9018b5e88 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -9,7 +9,7 @@ #include "common/common_types.h" #include "video_core/hwrasterizer_base.h" -#include "video_core/vertex_shader.h" +#include "video_core/shader_interpreter.h" #include "gl_state.h" #include "gl_rasterizer_cache.h" @@ -27,9 +27,9 @@ public: void Reset() override; /// Queues the primitive formed by the given vertices for rendering - void AddTriangle(const Pica::VertexShader::OutputVertex& v0, - const Pica::VertexShader::OutputVertex& v1, - const Pica::VertexShader::OutputVertex& v2) override; + void AddTriangle(const Pica::Shader::OutputVertex& v0, + const Pica::Shader::OutputVertex& v1, + const Pica::Shader::OutputVertex& v2) override; /// Draw the current batch of triangles void DrawTriangles() override; @@ -82,7 +82,7 @@ private: /// Structure that the hardware rendered vertices are composed of struct HardwareVertex { - HardwareVertex(const Pica::VertexShader::OutputVertex& v) { + HardwareVertex(const Pica::Shader::OutputVertex& v) { position[0] = v.pos.x.ToFloat32(); position[1] = v.pos.y.ToFloat32(); position[2] = v.pos.z.ToFloat32(); diff --git a/src/video_core/shader_interpreter.cpp b/src/video_core/shader_interpreter.cpp new file mode 100644 index 000000000..3cce26d36 --- /dev/null +++ b/src/video_core/shader_interpreter.cpp @@ -0,0 +1,629 @@ +// Copyright 2014 Citra Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include +#include + +#include + +#include + +#include "common/profiler.h" + +#include "pica.h" +#include "shader_interpreter.h" +#include "debug_utils/debug_utils.h" + +using nihstro::OpCode; +using nihstro::Instruction; +using nihstro::RegisterType; +using nihstro::SourceRegister; +using nihstro::SwizzlePattern; + +namespace Pica { + +namespace Shader { + +struct ShaderState { + u32 program_counter; + + const float24* input_register_table[16]; + Math::Vec4 output_registers[16]; + + Math::Vec4 temporary_registers[16]; + bool conditional_code[2]; + + // Two Address registers and one loop counter + // TODO: How many bits do these actually have? + s32 address_registers[3]; + + enum { + INVALID_ADDRESS = 0xFFFFFFFF + }; + + struct CallStackElement { + u32 final_address; // Address upon which we jump to return_address + u32 return_address; // Where to jump when leaving scope + u8 repeat_counter; // How often to repeat until this call stack element is removed + u8 loop_increment; // Which value to add to the loop counter after an iteration + // TODO: Should this be a signed value? Does it even matter? + u32 loop_address; // The address where we'll return to after each loop iteration + }; + + // TODO: Is there a maximal size for this? + boost::container::static_vector call_stack; + + struct { + u32 max_offset; // maximum program counter ever reached + u32 max_opdesc_id; // maximum swizzle pattern index ever used + } debug; +}; + +static void ProcessShaderCode(ShaderState& state) { + const auto& uniforms = g_state.vs.uniforms; + const auto& swizzle_data = g_state.vs.swizzle_data; + const auto& program_code = g_state.vs.program_code; + + // Placeholder for invalid inputs + static float24 dummy_vec4_float24[4]; + + while (true) { + if (!state.call_stack.empty()) { + auto& top = state.call_stack.back(); + if (state.program_counter == top.final_address) { + state.address_registers[2] += top.loop_increment; + + if (top.repeat_counter-- == 0) { + state.program_counter = top.return_address; + state.call_stack.pop_back(); + } else { + state.program_counter = top.loop_address; + } + + // TODO: Is "trying again" accurate to hardware? + continue; + } + } + + bool exit_loop = false; + const Instruction instr = { program_code[state.program_counter] }; + const SwizzlePattern swizzle = { swizzle_data[instr.common.operand_desc_id] }; + + static auto call = [](ShaderState& state, u32 offset, u32 num_instructions, + u32 return_offset, u8 repeat_count, u8 loop_increment) { + state.program_counter = offset - 1; // -1 to make sure when incrementing the PC we end up at the correct offset + ASSERT(state.call_stack.size() < state.call_stack.capacity()); + state.call_stack.push_back({ offset + num_instructions, return_offset, repeat_count, loop_increment, offset }); + }; + state.debug.max_offset = std::max(state.debug.max_offset, 1 + state.program_counter); + + auto LookupSourceRegister = [&](const SourceRegister& source_reg) -> const float24* { + switch (source_reg.GetRegisterType()) { + case RegisterType::Input: + return state.input_register_table[source_reg.GetIndex()]; + + case RegisterType::Temporary: + return &state.temporary_registers[source_reg.GetIndex()].x; + + case RegisterType::FloatUniform: + return &uniforms.f[source_reg.GetIndex()].x; + + default: + return dummy_vec4_float24; + } + }; + + switch (instr.opcode.Value().GetInfo().type) { + case OpCode::Type::Arithmetic: + { + const bool is_inverted = (0 != (instr.opcode.Value().GetInfo().subtype & OpCode::Info::SrcInversed)); + + const int address_offset = (instr.common.address_register_index == 0) + ? 0 : state.address_registers[instr.common.address_register_index - 1]; + + const float24* src1_ = LookupSourceRegister(instr.common.GetSrc1(is_inverted) + (!is_inverted * address_offset)); + const float24* src2_ = LookupSourceRegister(instr.common.GetSrc2(is_inverted) + ( is_inverted * address_offset)); + + const bool negate_src1 = ((bool)swizzle.negate_src1 != false); + const bool negate_src2 = ((bool)swizzle.negate_src2 != false); + + float24 src1[4] = { + src1_[(int)swizzle.GetSelectorSrc1(0)], + src1_[(int)swizzle.GetSelectorSrc1(1)], + src1_[(int)swizzle.GetSelectorSrc1(2)], + src1_[(int)swizzle.GetSelectorSrc1(3)], + }; + if (negate_src1) { + src1[0] = src1[0] * float24::FromFloat32(-1); + src1[1] = src1[1] * float24::FromFloat32(-1); + src1[2] = src1[2] * float24::FromFloat32(-1); + src1[3] = src1[3] * float24::FromFloat32(-1); + } + float24 src2[4] = { + src2_[(int)swizzle.GetSelectorSrc2(0)], + src2_[(int)swizzle.GetSelectorSrc2(1)], + src2_[(int)swizzle.GetSelectorSrc2(2)], + src2_[(int)swizzle.GetSelectorSrc2(3)], + }; + if (negate_src2) { + src2[0] = src2[0] * float24::FromFloat32(-1); + src2[1] = src2[1] * float24::FromFloat32(-1); + src2[2] = src2[2] * float24::FromFloat32(-1); + src2[3] = src2[3] * float24::FromFloat32(-1); + } + + float24* dest = (instr.common.dest.Value() < 0x10) ? &state.output_registers[instr.common.dest.Value().GetIndex()][0] + : (instr.common.dest.Value() < 0x20) ? &state.temporary_registers[instr.common.dest.Value().GetIndex()][0] + : dummy_vec4_float24; + + state.debug.max_opdesc_id = std::max(state.debug.max_opdesc_id, 1+instr.common.operand_desc_id); + + switch (instr.opcode.Value().EffectiveOpCode()) { + case OpCode::Id::ADD: + { + for (int i = 0; i < 4; ++i) { + if (!swizzle.DestComponentEnabled(i)) + continue; + + dest[i] = src1[i] + src2[i]; + } + + break; + } + + case OpCode::Id::MUL: + { + for (int i = 0; i < 4; ++i) { + if (!swizzle.DestComponentEnabled(i)) + continue; + + dest[i] = src1[i] * src2[i]; + } + + break; + } + + case OpCode::Id::FLR: + for (int i = 0; i < 4; ++i) { + if (!swizzle.DestComponentEnabled(i)) + continue; + + dest[i] = float24::FromFloat32(std::floor(src1[i].ToFloat32())); + } + break; + + case OpCode::Id::MAX: + for (int i = 0; i < 4; ++i) { + if (!swizzle.DestComponentEnabled(i)) + continue; + + dest[i] = std::max(src1[i], src2[i]); + } + break; + + case OpCode::Id::MIN: + for (int i = 0; i < 4; ++i) { + if (!swizzle.DestComponentEnabled(i)) + continue; + + dest[i] = std::min(src1[i], src2[i]); + } + break; + + case OpCode::Id::DP3: + case OpCode::Id::DP4: + { + float24 dot = float24::FromFloat32(0.f); + int num_components = (instr.opcode.Value() == OpCode::Id::DP3) ? 3 : 4; + for (int i = 0; i < num_components; ++i) + dot = dot + src1[i] * src2[i]; + + for (int i = 0; i < 4; ++i) { + if (!swizzle.DestComponentEnabled(i)) + continue; + + dest[i] = dot; + } + break; + } + + // Reciprocal + case OpCode::Id::RCP: + { + for (int i = 0; i < 4; ++i) { + if (!swizzle.DestComponentEnabled(i)) + continue; + + // TODO: Be stable against division by zero! + // TODO: I think this might be wrong... we should only use one component here + dest[i] = float24::FromFloat32(1.0f / src1[i].ToFloat32()); + } + + break; + } + + // Reciprocal Square Root + case OpCode::Id::RSQ: + { + for (int i = 0; i < 4; ++i) { + if (!swizzle.DestComponentEnabled(i)) + continue; + + // TODO: Be stable against division by zero! + // TODO: I think this might be wrong... we should only use one component here + dest[i] = float24::FromFloat32(1.0f / sqrt(src1[i].ToFloat32())); + } + + break; + } + + case OpCode::Id::MOVA: + { + for (int i = 0; i < 2; ++i) { + if (!swizzle.DestComponentEnabled(i)) + continue; + + // TODO: Figure out how the rounding is done on hardware + state.address_registers[i] = static_cast(src1[i].ToFloat32()); + } + + break; + } + + case OpCode::Id::MOV: + { + for (int i = 0; i < 4; ++i) { + if (!swizzle.DestComponentEnabled(i)) + continue; + + dest[i] = src1[i]; + } + break; + } + + case OpCode::Id::SLT: + case OpCode::Id::SLTI: + for (int i = 0; i < 4; ++i) { + if (!swizzle.DestComponentEnabled(i)) + continue; + + dest[i] = (src1[i] < src2[i]) ? float24::FromFloat32(1.0f) : float24::FromFloat32(0.0f); + } + break; + + case OpCode::Id::CMP: + for (int i = 0; i < 2; ++i) { + // TODO: Can you restrict to one compare via dest masking? + + auto compare_op = instr.common.compare_op; + auto op = (i == 0) ? compare_op.x.Value() : compare_op.y.Value(); + + switch (op) { + case compare_op.Equal: + state.conditional_code[i] = (src1[i] == src2[i]); + break; + + case compare_op.NotEqual: + state.conditional_code[i] = (src1[i] != src2[i]); + break; + + case compare_op.LessThan: + state.conditional_code[i] = (src1[i] < src2[i]); + break; + + case compare_op.LessEqual: + state.conditional_code[i] = (src1[i] <= src2[i]); + break; + + case compare_op.GreaterThan: + state.conditional_code[i] = (src1[i] > src2[i]); + break; + + case compare_op.GreaterEqual: + state.conditional_code[i] = (src1[i] >= src2[i]); + break; + + default: + LOG_ERROR(HW_GPU, "Unknown compare mode %x", static_cast(op)); + break; + } + } + break; + + default: + LOG_ERROR(HW_GPU, "Unhandled arithmetic instruction: 0x%02x (%s): 0x%08x", + (int)instr.opcode.Value().EffectiveOpCode(), instr.opcode.Value().GetInfo().name, instr.hex); + DEBUG_ASSERT(false); + break; + } + + break; + } + + case OpCode::Type::MultiplyAdd: + { + if ((instr.opcode.Value().EffectiveOpCode() == OpCode::Id::MAD) || + (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::MADI)) { + const SwizzlePattern& swizzle = *(SwizzlePattern*)&swizzle_data[instr.mad.operand_desc_id]; + + bool is_inverted = (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::MADI); + + const float24* src1_ = LookupSourceRegister(instr.mad.GetSrc1(is_inverted)); + const float24* src2_ = LookupSourceRegister(instr.mad.GetSrc2(is_inverted)); + const float24* src3_ = LookupSourceRegister(instr.mad.GetSrc3(is_inverted)); + + const bool negate_src1 = ((bool)swizzle.negate_src1 != false); + const bool negate_src2 = ((bool)swizzle.negate_src2 != false); + const bool negate_src3 = ((bool)swizzle.negate_src3 != false); + + float24 src1[4] = { + src1_[(int)swizzle.GetSelectorSrc1(0)], + src1_[(int)swizzle.GetSelectorSrc1(1)], + src1_[(int)swizzle.GetSelectorSrc1(2)], + src1_[(int)swizzle.GetSelectorSrc1(3)], + }; + if (negate_src1) { + src1[0] = src1[0] * float24::FromFloat32(-1); + src1[1] = src1[1] * float24::FromFloat32(-1); + src1[2] = src1[2] * float24::FromFloat32(-1); + src1[3] = src1[3] * float24::FromFloat32(-1); + } + float24 src2[4] = { + src2_[(int)swizzle.GetSelectorSrc2(0)], + src2_[(int)swizzle.GetSelectorSrc2(1)], + src2_[(int)swizzle.GetSelectorSrc2(2)], + src2_[(int)swizzle.GetSelectorSrc2(3)], + }; + if (negate_src2) { + src2[0] = src2[0] * float24::FromFloat32(-1); + src2[1] = src2[1] * float24::FromFloat32(-1); + src2[2] = src2[2] * float24::FromFloat32(-1); + src2[3] = src2[3] * float24::FromFloat32(-1); + } + float24 src3[4] = { + src3_[(int)swizzle.GetSelectorSrc3(0)], + src3_[(int)swizzle.GetSelectorSrc3(1)], + src3_[(int)swizzle.GetSelectorSrc3(2)], + src3_[(int)swizzle.GetSelectorSrc3(3)], + }; + if (negate_src3) { + src3[0] = src3[0] * float24::FromFloat32(-1); + src3[1] = src3[1] * float24::FromFloat32(-1); + src3[2] = src3[2] * float24::FromFloat32(-1); + src3[3] = src3[3] * float24::FromFloat32(-1); + } + + float24* dest = (instr.mad.dest.Value() < 0x10) ? &state.output_registers[instr.mad.dest.Value().GetIndex()][0] + : (instr.mad.dest.Value() < 0x20) ? &state.temporary_registers[instr.mad.dest.Value().GetIndex()][0] + : dummy_vec4_float24; + + for (int i = 0; i < 4; ++i) { + if (!swizzle.DestComponentEnabled(i)) + continue; + + dest[i] = src1[i] * src2[i] + src3[i]; + } + } else { + LOG_ERROR(HW_GPU, "Unhandled multiply-add instruction: 0x%02x (%s): 0x%08x", + (int)instr.opcode.Value().EffectiveOpCode(), instr.opcode.Value().GetInfo().name, instr.hex); + } + break; + } + + default: + { + static auto evaluate_condition = [](const ShaderState& state, bool refx, bool refy, Instruction::FlowControlType flow_control) { + bool results[2] = { refx == state.conditional_code[0], + refy == state.conditional_code[1] }; + + switch (flow_control.op) { + case flow_control.Or: + return results[0] || results[1]; + + case flow_control.And: + return results[0] && results[1]; + + case flow_control.JustX: + return results[0]; + + case flow_control.JustY: + return results[1]; + } + }; + + // Handle each instruction on its own + switch (instr.opcode.Value()) { + case OpCode::Id::END: + exit_loop = true; + break; + + case OpCode::Id::JMPC: + if (evaluate_condition(state, instr.flow_control.refx, instr.flow_control.refy, instr.flow_control)) { + state.program_counter = instr.flow_control.dest_offset - 1; + } + break; + + case OpCode::Id::JMPU: + if (uniforms.b[instr.flow_control.bool_uniform_id]) { + state.program_counter = instr.flow_control.dest_offset - 1; + } + break; + + case OpCode::Id::CALL: + call(state, + instr.flow_control.dest_offset, + instr.flow_control.num_instructions, + state.program_counter + 1, 0, 0); + break; + + case OpCode::Id::CALLU: + if (uniforms.b[instr.flow_control.bool_uniform_id]) { + call(state, + instr.flow_control.dest_offset, + instr.flow_control.num_instructions, + state.program_counter + 1, 0, 0); + } + break; + + case OpCode::Id::CALLC: + if (evaluate_condition(state, instr.flow_control.refx, instr.flow_control.refy, instr.flow_control)) { + call(state, + instr.flow_control.dest_offset, + instr.flow_control.num_instructions, + state.program_counter + 1, 0, 0); + } + break; + + case OpCode::Id::NOP: + break; + + case OpCode::Id::IFU: + if (uniforms.b[instr.flow_control.bool_uniform_id]) { + call(state, + state.program_counter + 1, + instr.flow_control.dest_offset - state.program_counter - 1, + instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0); + } else { + call(state, + instr.flow_control.dest_offset, + instr.flow_control.num_instructions, + instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0); + } + + break; + + case OpCode::Id::IFC: + { + // TODO: Do we need to consider swizzlers here? + + if (evaluate_condition(state, instr.flow_control.refx, instr.flow_control.refy, instr.flow_control)) { + call(state, + state.program_counter + 1, + instr.flow_control.dest_offset - state.program_counter - 1, + instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0); + } else { + call(state, + instr.flow_control.dest_offset, + instr.flow_control.num_instructions, + instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0); + } + + break; + } + + case OpCode::Id::LOOP: + { + state.address_registers[2] = uniforms.i[instr.flow_control.int_uniform_id].y; + + call(state, + state.program_counter + 1, + instr.flow_control.dest_offset - state.program_counter + 1, + instr.flow_control.dest_offset + 1, + uniforms.i[instr.flow_control.int_uniform_id].x, + uniforms.i[instr.flow_control.int_uniform_id].z); + break; + } + + default: + LOG_ERROR(HW_GPU, "Unhandled instruction: 0x%02x (%s): 0x%08x", + (int)instr.opcode.Value().EffectiveOpCode(), instr.opcode.Value().GetInfo().name, instr.hex); + break; + } + + break; + } + } + + ++state.program_counter; + + if (exit_loop) + break; + } +} + +static Common::Profiling::TimingCategory shader_category("Vertex Shader"); + +OutputVertex RunShader(const InputVertex& input, int num_attributes, const Regs::ShaderConfig& config, const State::ShaderSetup& setup) { + Common::Profiling::ScopeTimer timer(shader_category); + + ShaderState state; + + state.program_counter = config.main_offset; + state.debug.max_offset = 0; + state.debug.max_opdesc_id = 0; + + // Setup input register table + const auto& attribute_register_map = config.input_register_map; + float24 dummy_register; + boost::fill(state.input_register_table, &dummy_register); + + if (num_attributes > 0) state.input_register_table[attribute_register_map.attribute0_register] = &input.attr[0].x; + if (num_attributes > 1) state.input_register_table[attribute_register_map.attribute1_register] = &input.attr[1].x; + if (num_attributes > 2) state.input_register_table[attribute_register_map.attribute2_register] = &input.attr[2].x; + if (num_attributes > 3) state.input_register_table[attribute_register_map.attribute3_register] = &input.attr[3].x; + if (num_attributes > 4) state.input_register_table[attribute_register_map.attribute4_register] = &input.attr[4].x; + if (num_attributes > 5) state.input_register_table[attribute_register_map.attribute5_register] = &input.attr[5].x; + if (num_attributes > 6) state.input_register_table[attribute_register_map.attribute6_register] = &input.attr[6].x; + if (num_attributes > 7) state.input_register_table[attribute_register_map.attribute7_register] = &input.attr[7].x; + if (num_attributes > 8) state.input_register_table[attribute_register_map.attribute8_register] = &input.attr[8].x; + if (num_attributes > 9) state.input_register_table[attribute_register_map.attribute9_register] = &input.attr[9].x; + if (num_attributes > 10) state.input_register_table[attribute_register_map.attribute10_register] = &input.attr[10].x; + if (num_attributes > 11) state.input_register_table[attribute_register_map.attribute11_register] = &input.attr[11].x; + if (num_attributes > 12) state.input_register_table[attribute_register_map.attribute12_register] = &input.attr[12].x; + if (num_attributes > 13) state.input_register_table[attribute_register_map.attribute13_register] = &input.attr[13].x; + if (num_attributes > 14) state.input_register_table[attribute_register_map.attribute14_register] = &input.attr[14].x; + if (num_attributes > 15) state.input_register_table[attribute_register_map.attribute15_register] = &input.attr[15].x; + + state.conditional_code[0] = false; + state.conditional_code[1] = false; + + ProcessShaderCode(state); +#if PICA_DUMP_SHADERS + DebugUtils::DumpShader(setup.program_code.data(), state.debug.max_offset, setup.swizzle_data.data(), + state.debug.max_opdesc_id, config.main_offset, + g_state.regs.vs_output_attributes); // TODO: Don't hardcode VS here +#endif + + // Setup output data + OutputVertex ret; + // TODO(neobrain): Under some circumstances, up to 16 attributes may be output. We need to + // figure out what those circumstances are and enable the remaining outputs then. + for (int i = 0; i < 7; ++i) { + const auto& output_register_map = g_state.regs.vs_output_attributes[i]; // TODO: Don't hardcode VS here + + u32 semantics[4] = { + output_register_map.map_x, output_register_map.map_y, + output_register_map.map_z, output_register_map.map_w + }; + + for (int comp = 0; comp < 4; ++comp) { + float24* out = ((float24*)&ret) + semantics[comp]; + if (semantics[comp] != Regs::VSOutputAttributes::INVALID) { + *out = state.output_registers[i][comp]; + } else { + // Zero output so that attributes which aren't output won't have denormals in them, + // which would slow us down later. + memset(out, 0, sizeof(*out)); + } + } + } + + // The hardware takes the absolute and saturates vertex colors like this, *before* doing interpolation + for (int i = 0; i < 4; ++i) { + ret.color[i] = float24::FromFloat32( + std::fmin(std::fabs(ret.color[i].ToFloat32()), 1.0f)); + } + + LOG_TRACE(Render_Software, "Output vertex: pos (%.2f, %.2f, %.2f, %.2f), col(%.2f, %.2f, %.2f, %.2f), tc0(%.2f, %.2f)", + ret.pos.x.ToFloat32(), ret.pos.y.ToFloat32(), ret.pos.z.ToFloat32(), ret.pos.w.ToFloat32(), + ret.color.x.ToFloat32(), ret.color.y.ToFloat32(), ret.color.z.ToFloat32(), ret.color.w.ToFloat32(), + ret.tc0.u().ToFloat32(), ret.tc0.v().ToFloat32()); + + return ret; +} + + +} // namespace + +} // namespace diff --git a/src/video_core/shader_interpreter.h b/src/video_core/shader_interpreter.h new file mode 100644 index 000000000..942a30841 --- /dev/null +++ b/src/video_core/shader_interpreter.h @@ -0,0 +1,72 @@ +// Copyright 2014 Citra Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include + +#include "common/vector_math.h" + +#include "pica.h" + +namespace Pica { + +namespace Shader { + +struct InputVertex { + Math::Vec4 attr[16]; +}; + +struct OutputVertex { + OutputVertex() = default; + + // VS output attributes + Math::Vec4 pos; + Math::Vec4 dummy; // quaternions (not implemented, yet) + Math::Vec4 color; + Math::Vec2 tc0; + Math::Vec2 tc1; + float24 pad[6]; + Math::Vec2 tc2; + + // Padding for optimal alignment + float24 pad2[4]; + + // Attributes used to store intermediate results + + // position after perspective divide + Math::Vec3 screenpos; + float24 pad3; + + // Linear interpolation + // factor: 0=this, 1=vtx + void Lerp(float24 factor, const OutputVertex& vtx) { + pos = pos * factor + vtx.pos * (float24::FromFloat32(1) - factor); + + // TODO: Should perform perspective correct interpolation here... + tc0 = tc0 * factor + vtx.tc0 * (float24::FromFloat32(1) - factor); + tc1 = tc1 * factor + vtx.tc1 * (float24::FromFloat32(1) - factor); + tc2 = tc2 * factor + vtx.tc2 * (float24::FromFloat32(1) - factor); + + screenpos = screenpos * factor + vtx.screenpos * (float24::FromFloat32(1) - factor); + + color = color * factor + vtx.color * (float24::FromFloat32(1) - factor); + } + + // Linear interpolation + // factor: 0=v0, 1=v1 + static OutputVertex Lerp(float24 factor, const OutputVertex& v0, const OutputVertex& v1) { + OutputVertex ret = v0; + ret.Lerp(factor, v1); + return ret; + } +}; +static_assert(std::is_pod::value, "Structure is not POD"); +static_assert(sizeof(OutputVertex) == 32 * sizeof(float), "OutputVertex has invalid size"); + +OutputVertex RunShader(const InputVertex& input, int num_attributes, const Regs::ShaderConfig& config, const State::ShaderSetup& setup); + +} // namespace + +} // namespace diff --git a/src/video_core/vertex_shader.cpp b/src/video_core/vertex_shader.cpp deleted file mode 100644 index 5f66f3455..000000000 --- a/src/video_core/vertex_shader.cpp +++ /dev/null @@ -1,629 +0,0 @@ -// Copyright 2014 Citra Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include -#include - -#include - -#include - -#include "common/profiler.h" - -#include "pica.h" -#include "vertex_shader.h" -#include "debug_utils/debug_utils.h" - -using nihstro::OpCode; -using nihstro::Instruction; -using nihstro::RegisterType; -using nihstro::SourceRegister; -using nihstro::SwizzlePattern; - -namespace Pica { - -namespace VertexShader { - -struct VertexShaderState { - u32 program_counter; - - const float24* input_register_table[16]; - Math::Vec4 output_registers[16]; - - Math::Vec4 temporary_registers[16]; - bool conditional_code[2]; - - // Two Address registers and one loop counter - // TODO: How many bits do these actually have? - s32 address_registers[3]; - - enum { - INVALID_ADDRESS = 0xFFFFFFFF - }; - - struct CallStackElement { - u32 final_address; // Address upon which we jump to return_address - u32 return_address; // Where to jump when leaving scope - u8 repeat_counter; // How often to repeat until this call stack element is removed - u8 loop_increment; // Which value to add to the loop counter after an iteration - // TODO: Should this be a signed value? Does it even matter? - u32 loop_address; // The address where we'll return to after each loop iteration - }; - - // TODO: Is there a maximal size for this? - boost::container::static_vector call_stack; - - struct { - u32 max_offset; // maximum program counter ever reached - u32 max_opdesc_id; // maximum swizzle pattern index ever used - } debug; -}; - -static void ProcessShaderCode(VertexShaderState& state) { - const auto& uniforms = g_state.vs.uniforms; - const auto& swizzle_data = g_state.vs.swizzle_data; - const auto& program_code = g_state.vs.program_code; - - // Placeholder for invalid inputs - static float24 dummy_vec4_float24[4]; - - while (true) { - if (!state.call_stack.empty()) { - auto& top = state.call_stack.back(); - if (state.program_counter == top.final_address) { - state.address_registers[2] += top.loop_increment; - - if (top.repeat_counter-- == 0) { - state.program_counter = top.return_address; - state.call_stack.pop_back(); - } else { - state.program_counter = top.loop_address; - } - - // TODO: Is "trying again" accurate to hardware? - continue; - } - } - - bool exit_loop = false; - const Instruction instr = { program_code[state.program_counter] }; - const SwizzlePattern swizzle = { swizzle_data[instr.common.operand_desc_id] }; - - static auto call = [](VertexShaderState& state, u32 offset, u32 num_instructions, - u32 return_offset, u8 repeat_count, u8 loop_increment) { - state.program_counter = offset - 1; // -1 to make sure when incrementing the PC we end up at the correct offset - ASSERT(state.call_stack.size() < state.call_stack.capacity()); - state.call_stack.push_back({ offset + num_instructions, return_offset, repeat_count, loop_increment, offset }); - }; - state.debug.max_offset = std::max(state.debug.max_offset, 1 + state.program_counter); - - auto LookupSourceRegister = [&](const SourceRegister& source_reg) -> const float24* { - switch (source_reg.GetRegisterType()) { - case RegisterType::Input: - return state.input_register_table[source_reg.GetIndex()]; - - case RegisterType::Temporary: - return &state.temporary_registers[source_reg.GetIndex()].x; - - case RegisterType::FloatUniform: - return &uniforms.f[source_reg.GetIndex()].x; - - default: - return dummy_vec4_float24; - } - }; - - switch (instr.opcode.Value().GetInfo().type) { - case OpCode::Type::Arithmetic: - { - const bool is_inverted = (0 != (instr.opcode.Value().GetInfo().subtype & OpCode::Info::SrcInversed)); - - const int address_offset = (instr.common.address_register_index == 0) - ? 0 : state.address_registers[instr.common.address_register_index - 1]; - - const float24* src1_ = LookupSourceRegister(instr.common.GetSrc1(is_inverted) + (!is_inverted * address_offset)); - const float24* src2_ = LookupSourceRegister(instr.common.GetSrc2(is_inverted) + ( is_inverted * address_offset)); - - const bool negate_src1 = ((bool)swizzle.negate_src1 != false); - const bool negate_src2 = ((bool)swizzle.negate_src2 != false); - - float24 src1[4] = { - src1_[(int)swizzle.GetSelectorSrc1(0)], - src1_[(int)swizzle.GetSelectorSrc1(1)], - src1_[(int)swizzle.GetSelectorSrc1(2)], - src1_[(int)swizzle.GetSelectorSrc1(3)], - }; - if (negate_src1) { - src1[0] = src1[0] * float24::FromFloat32(-1); - src1[1] = src1[1] * float24::FromFloat32(-1); - src1[2] = src1[2] * float24::FromFloat32(-1); - src1[3] = src1[3] * float24::FromFloat32(-1); - } - float24 src2[4] = { - src2_[(int)swizzle.GetSelectorSrc2(0)], - src2_[(int)swizzle.GetSelectorSrc2(1)], - src2_[(int)swizzle.GetSelectorSrc2(2)], - src2_[(int)swizzle.GetSelectorSrc2(3)], - }; - if (negate_src2) { - src2[0] = src2[0] * float24::FromFloat32(-1); - src2[1] = src2[1] * float24::FromFloat32(-1); - src2[2] = src2[2] * float24::FromFloat32(-1); - src2[3] = src2[3] * float24::FromFloat32(-1); - } - - float24* dest = (instr.common.dest.Value() < 0x10) ? &state.output_registers[instr.common.dest.Value().GetIndex()][0] - : (instr.common.dest.Value() < 0x20) ? &state.temporary_registers[instr.common.dest.Value().GetIndex()][0] - : dummy_vec4_float24; - - state.debug.max_opdesc_id = std::max(state.debug.max_opdesc_id, 1+instr.common.operand_desc_id); - - switch (instr.opcode.Value().EffectiveOpCode()) { - case OpCode::Id::ADD: - { - for (int i = 0; i < 4; ++i) { - if (!swizzle.DestComponentEnabled(i)) - continue; - - dest[i] = src1[i] + src2[i]; - } - - break; - } - - case OpCode::Id::MUL: - { - for (int i = 0; i < 4; ++i) { - if (!swizzle.DestComponentEnabled(i)) - continue; - - dest[i] = src1[i] * src2[i]; - } - - break; - } - - case OpCode::Id::FLR: - for (int i = 0; i < 4; ++i) { - if (!swizzle.DestComponentEnabled(i)) - continue; - - dest[i] = float24::FromFloat32(std::floor(src1[i].ToFloat32())); - } - break; - - case OpCode::Id::MAX: - for (int i = 0; i < 4; ++i) { - if (!swizzle.DestComponentEnabled(i)) - continue; - - dest[i] = std::max(src1[i], src2[i]); - } - break; - - case OpCode::Id::MIN: - for (int i = 0; i < 4; ++i) { - if (!swizzle.DestComponentEnabled(i)) - continue; - - dest[i] = std::min(src1[i], src2[i]); - } - break; - - case OpCode::Id::DP3: - case OpCode::Id::DP4: - { - float24 dot = float24::FromFloat32(0.f); - int num_components = (instr.opcode.Value() == OpCode::Id::DP3) ? 3 : 4; - for (int i = 0; i < num_components; ++i) - dot = dot + src1[i] * src2[i]; - - for (int i = 0; i < 4; ++i) { - if (!swizzle.DestComponentEnabled(i)) - continue; - - dest[i] = dot; - } - break; - } - - // Reciprocal - case OpCode::Id::RCP: - { - for (int i = 0; i < 4; ++i) { - if (!swizzle.DestComponentEnabled(i)) - continue; - - // TODO: Be stable against division by zero! - // TODO: I think this might be wrong... we should only use one component here - dest[i] = float24::FromFloat32(1.0f / src1[i].ToFloat32()); - } - - break; - } - - // Reciprocal Square Root - case OpCode::Id::RSQ: - { - for (int i = 0; i < 4; ++i) { - if (!swizzle.DestComponentEnabled(i)) - continue; - - // TODO: Be stable against division by zero! - // TODO: I think this might be wrong... we should only use one component here - dest[i] = float24::FromFloat32(1.0f / sqrt(src1[i].ToFloat32())); - } - - break; - } - - case OpCode::Id::MOVA: - { - for (int i = 0; i < 2; ++i) { - if (!swizzle.DestComponentEnabled(i)) - continue; - - // TODO: Figure out how the rounding is done on hardware - state.address_registers[i] = static_cast(src1[i].ToFloat32()); - } - - break; - } - - case OpCode::Id::MOV: - { - for (int i = 0; i < 4; ++i) { - if (!swizzle.DestComponentEnabled(i)) - continue; - - dest[i] = src1[i]; - } - break; - } - - case OpCode::Id::SLT: - case OpCode::Id::SLTI: - for (int i = 0; i < 4; ++i) { - if (!swizzle.DestComponentEnabled(i)) - continue; - - dest[i] = (src1[i] < src2[i]) ? float24::FromFloat32(1.0f) : float24::FromFloat32(0.0f); - } - break; - - case OpCode::Id::CMP: - for (int i = 0; i < 2; ++i) { - // TODO: Can you restrict to one compare via dest masking? - - auto compare_op = instr.common.compare_op; - auto op = (i == 0) ? compare_op.x.Value() : compare_op.y.Value(); - - switch (op) { - case compare_op.Equal: - state.conditional_code[i] = (src1[i] == src2[i]); - break; - - case compare_op.NotEqual: - state.conditional_code[i] = (src1[i] != src2[i]); - break; - - case compare_op.LessThan: - state.conditional_code[i] = (src1[i] < src2[i]); - break; - - case compare_op.LessEqual: - state.conditional_code[i] = (src1[i] <= src2[i]); - break; - - case compare_op.GreaterThan: - state.conditional_code[i] = (src1[i] > src2[i]); - break; - - case compare_op.GreaterEqual: - state.conditional_code[i] = (src1[i] >= src2[i]); - break; - - default: - LOG_ERROR(HW_GPU, "Unknown compare mode %x", static_cast(op)); - break; - } - } - break; - - default: - LOG_ERROR(HW_GPU, "Unhandled arithmetic instruction: 0x%02x (%s): 0x%08x", - (int)instr.opcode.Value().EffectiveOpCode(), instr.opcode.Value().GetInfo().name, instr.hex); - DEBUG_ASSERT(false); - break; - } - - break; - } - - case OpCode::Type::MultiplyAdd: - { - if ((instr.opcode.Value().EffectiveOpCode() == OpCode::Id::MAD) || - (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::MADI)) { - const SwizzlePattern& swizzle = *(SwizzlePattern*)&swizzle_data[instr.mad.operand_desc_id]; - - bool is_inverted = (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::MADI); - - const float24* src1_ = LookupSourceRegister(instr.mad.GetSrc1(is_inverted)); - const float24* src2_ = LookupSourceRegister(instr.mad.GetSrc2(is_inverted)); - const float24* src3_ = LookupSourceRegister(instr.mad.GetSrc3(is_inverted)); - - const bool negate_src1 = ((bool)swizzle.negate_src1 != false); - const bool negate_src2 = ((bool)swizzle.negate_src2 != false); - const bool negate_src3 = ((bool)swizzle.negate_src3 != false); - - float24 src1[4] = { - src1_[(int)swizzle.GetSelectorSrc1(0)], - src1_[(int)swizzle.GetSelectorSrc1(1)], - src1_[(int)swizzle.GetSelectorSrc1(2)], - src1_[(int)swizzle.GetSelectorSrc1(3)], - }; - if (negate_src1) { - src1[0] = src1[0] * float24::FromFloat32(-1); - src1[1] = src1[1] * float24::FromFloat32(-1); - src1[2] = src1[2] * float24::FromFloat32(-1); - src1[3] = src1[3] * float24::FromFloat32(-1); - } - float24 src2[4] = { - src2_[(int)swizzle.GetSelectorSrc2(0)], - src2_[(int)swizzle.GetSelectorSrc2(1)], - src2_[(int)swizzle.GetSelectorSrc2(2)], - src2_[(int)swizzle.GetSelectorSrc2(3)], - }; - if (negate_src2) { - src2[0] = src2[0] * float24::FromFloat32(-1); - src2[1] = src2[1] * float24::FromFloat32(-1); - src2[2] = src2[2] * float24::FromFloat32(-1); - src2[3] = src2[3] * float24::FromFloat32(-1); - } - float24 src3[4] = { - src3_[(int)swizzle.GetSelectorSrc3(0)], - src3_[(int)swizzle.GetSelectorSrc3(1)], - src3_[(int)swizzle.GetSelectorSrc3(2)], - src3_[(int)swizzle.GetSelectorSrc3(3)], - }; - if (negate_src3) { - src3[0] = src3[0] * float24::FromFloat32(-1); - src3[1] = src3[1] * float24::FromFloat32(-1); - src3[2] = src3[2] * float24::FromFloat32(-1); - src3[3] = src3[3] * float24::FromFloat32(-1); - } - - float24* dest = (instr.mad.dest.Value() < 0x10) ? &state.output_registers[instr.mad.dest.Value().GetIndex()][0] - : (instr.mad.dest.Value() < 0x20) ? &state.temporary_registers[instr.mad.dest.Value().GetIndex()][0] - : dummy_vec4_float24; - - for (int i = 0; i < 4; ++i) { - if (!swizzle.DestComponentEnabled(i)) - continue; - - dest[i] = src1[i] * src2[i] + src3[i]; - } - } else { - LOG_ERROR(HW_GPU, "Unhandled multiply-add instruction: 0x%02x (%s): 0x%08x", - (int)instr.opcode.Value().EffectiveOpCode(), instr.opcode.Value().GetInfo().name, instr.hex); - } - break; - } - - default: - { - static auto evaluate_condition = [](const VertexShaderState& state, bool refx, bool refy, Instruction::FlowControlType flow_control) { - bool results[2] = { refx == state.conditional_code[0], - refy == state.conditional_code[1] }; - - switch (flow_control.op) { - case flow_control.Or: - return results[0] || results[1]; - - case flow_control.And: - return results[0] && results[1]; - - case flow_control.JustX: - return results[0]; - - case flow_control.JustY: - return results[1]; - } - }; - - // Handle each instruction on its own - switch (instr.opcode.Value()) { - case OpCode::Id::END: - exit_loop = true; - break; - - case OpCode::Id::JMPC: - if (evaluate_condition(state, instr.flow_control.refx, instr.flow_control.refy, instr.flow_control)) { - state.program_counter = instr.flow_control.dest_offset - 1; - } - break; - - case OpCode::Id::JMPU: - if (uniforms.b[instr.flow_control.bool_uniform_id]) { - state.program_counter = instr.flow_control.dest_offset - 1; - } - break; - - case OpCode::Id::CALL: - call(state, - instr.flow_control.dest_offset, - instr.flow_control.num_instructions, - state.program_counter + 1, 0, 0); - break; - - case OpCode::Id::CALLU: - if (uniforms.b[instr.flow_control.bool_uniform_id]) { - call(state, - instr.flow_control.dest_offset, - instr.flow_control.num_instructions, - state.program_counter + 1, 0, 0); - } - break; - - case OpCode::Id::CALLC: - if (evaluate_condition(state, instr.flow_control.refx, instr.flow_control.refy, instr.flow_control)) { - call(state, - instr.flow_control.dest_offset, - instr.flow_control.num_instructions, - state.program_counter + 1, 0, 0); - } - break; - - case OpCode::Id::NOP: - break; - - case OpCode::Id::IFU: - if (uniforms.b[instr.flow_control.bool_uniform_id]) { - call(state, - state.program_counter + 1, - instr.flow_control.dest_offset - state.program_counter - 1, - instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0); - } else { - call(state, - instr.flow_control.dest_offset, - instr.flow_control.num_instructions, - instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0); - } - - break; - - case OpCode::Id::IFC: - { - // TODO: Do we need to consider swizzlers here? - - if (evaluate_condition(state, instr.flow_control.refx, instr.flow_control.refy, instr.flow_control)) { - call(state, - state.program_counter + 1, - instr.flow_control.dest_offset - state.program_counter - 1, - instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0); - } else { - call(state, - instr.flow_control.dest_offset, - instr.flow_control.num_instructions, - instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0); - } - - break; - } - - case OpCode::Id::LOOP: - { - state.address_registers[2] = uniforms.i[instr.flow_control.int_uniform_id].y; - - call(state, - state.program_counter + 1, - instr.flow_control.dest_offset - state.program_counter + 1, - instr.flow_control.dest_offset + 1, - uniforms.i[instr.flow_control.int_uniform_id].x, - uniforms.i[instr.flow_control.int_uniform_id].z); - break; - } - - default: - LOG_ERROR(HW_GPU, "Unhandled instruction: 0x%02x (%s): 0x%08x", - (int)instr.opcode.Value().EffectiveOpCode(), instr.opcode.Value().GetInfo().name, instr.hex); - break; - } - - break; - } - } - - ++state.program_counter; - - if (exit_loop) - break; - } -} - -static Common::Profiling::TimingCategory shader_category("Vertex Shader"); - -OutputVertex RunShader(const InputVertex& input, int num_attributes, const Regs::ShaderConfig& config, const State::ShaderSetup& setup) { - Common::Profiling::ScopeTimer timer(shader_category); - - VertexShaderState state; - - state.program_counter = config.main_offset; - state.debug.max_offset = 0; - state.debug.max_opdesc_id = 0; - - // Setup input register table - const auto& attribute_register_map = config.input_register_map; - float24 dummy_register; - boost::fill(state.input_register_table, &dummy_register); - - if (num_attributes > 0) state.input_register_table[attribute_register_map.attribute0_register] = &input.attr[0].x; - if (num_attributes > 1) state.input_register_table[attribute_register_map.attribute1_register] = &input.attr[1].x; - if (num_attributes > 2) state.input_register_table[attribute_register_map.attribute2_register] = &input.attr[2].x; - if (num_attributes > 3) state.input_register_table[attribute_register_map.attribute3_register] = &input.attr[3].x; - if (num_attributes > 4) state.input_register_table[attribute_register_map.attribute4_register] = &input.attr[4].x; - if (num_attributes > 5) state.input_register_table[attribute_register_map.attribute5_register] = &input.attr[5].x; - if (num_attributes > 6) state.input_register_table[attribute_register_map.attribute6_register] = &input.attr[6].x; - if (num_attributes > 7) state.input_register_table[attribute_register_map.attribute7_register] = &input.attr[7].x; - if (num_attributes > 8) state.input_register_table[attribute_register_map.attribute8_register] = &input.attr[8].x; - if (num_attributes > 9) state.input_register_table[attribute_register_map.attribute9_register] = &input.attr[9].x; - if (num_attributes > 10) state.input_register_table[attribute_register_map.attribute10_register] = &input.attr[10].x; - if (num_attributes > 11) state.input_register_table[attribute_register_map.attribute11_register] = &input.attr[11].x; - if (num_attributes > 12) state.input_register_table[attribute_register_map.attribute12_register] = &input.attr[12].x; - if (num_attributes > 13) state.input_register_table[attribute_register_map.attribute13_register] = &input.attr[13].x; - if (num_attributes > 14) state.input_register_table[attribute_register_map.attribute14_register] = &input.attr[14].x; - if (num_attributes > 15) state.input_register_table[attribute_register_map.attribute15_register] = &input.attr[15].x; - - state.conditional_code[0] = false; - state.conditional_code[1] = false; - - ProcessShaderCode(state); -#if PICA_DUMP_SHADERS - DebugUtils::DumpShader(setup.program_code.data(), state.debug.max_offset, setup.swizzle_data.data(), - state.debug.max_opdesc_id, config.main_offset, - g_state.regs.vs_output_attributes); // TODO: Don't hardcode VS here -#endif - - // Setup output data - OutputVertex ret; - // TODO(neobrain): Under some circumstances, up to 16 attributes may be output. We need to - // figure out what those circumstances are and enable the remaining outputs then. - for (int i = 0; i < 7; ++i) { - const auto& output_register_map = g_state.regs.vs_output_attributes[i]; // TODO: Don't hardcode VS here - - u32 semantics[4] = { - output_register_map.map_x, output_register_map.map_y, - output_register_map.map_z, output_register_map.map_w - }; - - for (int comp = 0; comp < 4; ++comp) { - float24* out = ((float24*)&ret) + semantics[comp]; - if (semantics[comp] != Regs::VSOutputAttributes::INVALID) { - *out = state.output_registers[i][comp]; - } else { - // Zero output so that attributes which aren't output won't have denormals in them, - // which would slow us down later. - memset(out, 0, sizeof(*out)); - } - } - } - - // The hardware takes the absolute and saturates vertex colors like this, *before* doing interpolation - for (int i = 0; i < 4; ++i) { - ret.color[i] = float24::FromFloat32( - std::fmin(std::fabs(ret.color[i].ToFloat32()), 1.0f)); - } - - LOG_TRACE(Render_Software, "Output vertex: pos (%.2f, %.2f, %.2f, %.2f), col(%.2f, %.2f, %.2f, %.2f), tc0(%.2f, %.2f)", - ret.pos.x.ToFloat32(), ret.pos.y.ToFloat32(), ret.pos.z.ToFloat32(), ret.pos.w.ToFloat32(), - ret.color.x.ToFloat32(), ret.color.y.ToFloat32(), ret.color.z.ToFloat32(), ret.color.w.ToFloat32(), - ret.tc0.u().ToFloat32(), ret.tc0.v().ToFloat32()); - - return ret; -} - - -} // namespace - -} // namespace diff --git a/src/video_core/vertex_shader.h b/src/video_core/vertex_shader.h deleted file mode 100644 index 97f9250dd..000000000 --- a/src/video_core/vertex_shader.h +++ /dev/null @@ -1,73 +0,0 @@ -// Copyright 2014 Citra Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include - -#include "common/vector_math.h" - -#include "pica.h" - -namespace Pica { - -namespace VertexShader { - -struct InputVertex { - Math::Vec4 attr[16]; -}; - -struct OutputVertex { - OutputVertex() = default; - - // VS output attributes - Math::Vec4 pos; - Math::Vec4 dummy; // quaternions (not implemented, yet) - Math::Vec4 color; - Math::Vec2 tc0; - Math::Vec2 tc1; - float24 pad[6]; - Math::Vec2 tc2; - - // Padding for optimal alignment - float24 pad2[4]; - - // Attributes used to store intermediate results - - // position after perspective divide - Math::Vec3 screenpos; - float24 pad3; - - // Linear interpolation - // factor: 0=this, 1=vtx - void Lerp(float24 factor, const OutputVertex& vtx) { - pos = pos * factor + vtx.pos * (float24::FromFloat32(1) - factor); - - // TODO: Should perform perspective correct interpolation here... - tc0 = tc0 * factor + vtx.tc0 * (float24::FromFloat32(1) - factor); - tc1 = tc1 * factor + vtx.tc1 * (float24::FromFloat32(1) - factor); - tc2 = tc2 * factor + vtx.tc2 * (float24::FromFloat32(1) - factor); - - screenpos = screenpos * factor + vtx.screenpos * (float24::FromFloat32(1) - factor); - - color = color * factor + vtx.color * (float24::FromFloat32(1) - factor); - } - - // Linear interpolation - // factor: 0=v0, 1=v1 - static OutputVertex Lerp(float24 factor, const OutputVertex& v0, const OutputVertex& v1) { - OutputVertex ret = v0; - ret.Lerp(factor, v1); - return ret; - } -}; -static_assert(std::is_pod::value, "Structure is not POD"); -static_assert(sizeof(OutputVertex) == 32 * sizeof(float), "OutputVertex has invalid size"); - -OutputVertex RunShader(const InputVertex& input, int num_attributes, const Regs::ShaderConfig& config, const State::ShaderSetup& setup); - -} // namespace - -} // namespace - -- cgit v1.2.3