Pica: Add vertex shader implementation.

src/video_core/CMakeLists.txt

@@ -1,5 +1,6 @@
 set(SRCS    command_processor.cpp
             utils.cpp
+            vertex_shader.cpp
             video_core.cpp
             renderer_opengl/renderer_opengl.cpp)
 
@@ -8,6 +9,7 @@ set(HEADERS command_processor.h
             utils.h
             video_core.h
             renderer_base.h
+            vertex_shader.h
             video_core.h
             renderer_opengl/renderer_opengl.h)
 

src/video_core/command_processor.cpp

@@ -2,9 +2,10 @@
 // Licensed under GPLv2
 // Refer to the license.txt file included.
 
-#include "pica.h"
 #include "command_processor.h"
 #include "math.h"
+#include "pica.h"
+#include "vertex_shader.h"
 
 
 namespace Pica {
@@ -13,6 +14,14 @@ Regs registers;
 
 namespace CommandProcessor {
 
+static int float_regs_counter = 0;
+
+static u32 uniform_write_buffer[4];
+
+// Used for VSLoadProgramData and VSLoadSwizzleData
+static u32 vs_binary_write_offset = 0;
+static u32 vs_swizzle_write_offset = 0;
+
 static inline void WritePicaReg(u32 id, u32 value) {
     u32 old_value = registers[id];
     registers[id] = value;
@@ -67,9 +76,7 @@ static inline void WritePicaReg(u32 id, u32 value) {
                 }
 
                 // Initialize data for the current vertex
-                struct {
-                    Math::Vec4<float24> attr[16];
-                } input;
+                VertexShader::InputVertex input;
 
                 for (int i = 0; i < attribute_config.GetNumTotalAttributes(); ++i) {
                     for (int comp = 0; comp < vertex_attribute_elements[i]; ++comp) {
@@ -87,7 +94,7 @@ static inline void WritePicaReg(u32 id, u32 value) {
                                   input.attr[i][comp].ToFloat32());
                     }
                 }
-                // TODO: Run vertex data through vertex shader
+                VertexShader::OutputVertex output = VertexShader::RunShader(input, attribute_config.GetNumTotalAttributes());
 
                 if (is_indexed) {
                     // TODO: Add processed vertex to vertex cache!
@@ -98,6 +105,97 @@ static inline void WritePicaReg(u32 id, u32 value) {
             break;
         }
 
+        case PICA_REG_INDEX_WORKAROUND(vs_uniform_setup.set_value[0], 0x2c1):
+        case PICA_REG_INDEX_WORKAROUND(vs_uniform_setup.set_value[1], 0x2c2):
+        case PICA_REG_INDEX_WORKAROUND(vs_uniform_setup.set_value[2], 0x2c3):
+        case PICA_REG_INDEX_WORKAROUND(vs_uniform_setup.set_value[3], 0x2c4):
+        case PICA_REG_INDEX_WORKAROUND(vs_uniform_setup.set_value[4], 0x2c5):
+        case PICA_REG_INDEX_WORKAROUND(vs_uniform_setup.set_value[5], 0x2c6):
+        case PICA_REG_INDEX_WORKAROUND(vs_uniform_setup.set_value[6], 0x2c7):
+        case PICA_REG_INDEX_WORKAROUND(vs_uniform_setup.set_value[7], 0x2c8):
+        {
+            auto& uniform_setup = registers.vs_uniform_setup;
+
+            // TODO: Does actual hardware indeed keep an intermediate buffer or does
+            //       it directly write the values?
+            uniform_write_buffer[float_regs_counter++] = value;
+
+            // Uniforms are written in a packed format such that 4 float24 values are encoded in
+            // three 32-bit numbers. We write to internal memory once a full such vector is
+            // written.
+            if ((float_regs_counter >= 4 && uniform_setup.IsFloat32()) ||
+                (float_regs_counter >= 3 && !uniform_setup.IsFloat32())) {
+                float_regs_counter = 0;
+
+                auto& uniform = VertexShader::GetFloatUniform(uniform_setup.index);
+
+                if (uniform_setup.index > 95) {
+                    ERROR_LOG(GPU, "Invalid VS uniform index %d", (int)uniform_setup.index);
+                    break;
+                }
+
+                // NOTE: The destination component order indeed is "backwards"
+                if (uniform_setup.IsFloat32()) {
+                    for (auto i : {0,1,2,3})
+                        uniform[3 - i] = float24::FromFloat32(*(float*)(&uniform_write_buffer[i]));
+                } else {
+                    // TODO: Untested
+                    uniform.w = float24::FromRawFloat24(uniform_write_buffer[0] >> 8);
+                    uniform.z = float24::FromRawFloat24(((uniform_write_buffer[0] & 0xFF)<<16) | ((uniform_write_buffer[1] >> 16) & 0xFFFF));
+                    uniform.y = float24::FromRawFloat24(((uniform_write_buffer[1] & 0xFFFF)<<8) | ((uniform_write_buffer[2] >> 24) & 0xFF));
+                    uniform.x = float24::FromRawFloat24(uniform_write_buffer[2] & 0xFFFFFF);
+                }
+
+                DEBUG_LOG(GPU, "Set uniform %x to (%f %f %f %f)", (int)uniform_setup.index,
+                          uniform.x.ToFloat32(), uniform.y.ToFloat32(), uniform.z.ToFloat32(),
+                          uniform.w.ToFloat32());
+
+                // TODO: Verify that this actually modifies the register!
+                uniform_setup.index = uniform_setup.index + 1;
+            }
+            break;
+        }
+
+        // Seems to be used to reset the write pointer for VSLoadProgramData
+        case PICA_REG_INDEX(vs_program.begin_load):
+            vs_binary_write_offset = 0;
+            break;
+
+        // Load shader program code
+        case PICA_REG_INDEX_WORKAROUND(vs_program.set_word[0], 0x2cc):
+        case PICA_REG_INDEX_WORKAROUND(vs_program.set_word[1], 0x2cd):
+        case PICA_REG_INDEX_WORKAROUND(vs_program.set_word[2], 0x2ce):
+        case PICA_REG_INDEX_WORKAROUND(vs_program.set_word[3], 0x2cf):
+        case PICA_REG_INDEX_WORKAROUND(vs_program.set_word[4], 0x2d0):
+        case PICA_REG_INDEX_WORKAROUND(vs_program.set_word[5], 0x2d1):
+        case PICA_REG_INDEX_WORKAROUND(vs_program.set_word[6], 0x2d2):
+        case PICA_REG_INDEX_WORKAROUND(vs_program.set_word[7], 0x2d3):
+        {
+            VertexShader::SubmitShaderMemoryChange(vs_binary_write_offset, value);
+            vs_binary_write_offset++;
+            break;
+        }
+
+        // Seems to be used to reset the write pointer for VSLoadSwizzleData
+        case PICA_REG_INDEX(vs_swizzle_patterns.begin_load):
+            vs_swizzle_write_offset = 0;
+            break;
+
+        // Load swizzle pattern data
+        case PICA_REG_INDEX_WORKAROUND(vs_swizzle_patterns.set_word[0], 0x2d6):
+        case PICA_REG_INDEX_WORKAROUND(vs_swizzle_patterns.set_word[1], 0x2d7):
+        case PICA_REG_INDEX_WORKAROUND(vs_swizzle_patterns.set_word[2], 0x2d8):
+        case PICA_REG_INDEX_WORKAROUND(vs_swizzle_patterns.set_word[3], 0x2d9):
+        case PICA_REG_INDEX_WORKAROUND(vs_swizzle_patterns.set_word[4], 0x2da):
+        case PICA_REG_INDEX_WORKAROUND(vs_swizzle_patterns.set_word[5], 0x2db):
+        case PICA_REG_INDEX_WORKAROUND(vs_swizzle_patterns.set_word[6], 0x2dc):
+        case PICA_REG_INDEX_WORKAROUND(vs_swizzle_patterns.set_word[7], 0x2dd):
+        {
+            VertexShader::SubmitSwizzleDataChange(vs_swizzle_write_offset, value);
+            vs_swizzle_write_offset++;
+            break;
+        }
+
         default:
             break;
     }

src/video_core/pica.h

@@ -50,7 +50,39 @@ struct Regs {
     INSERT_PADDING_WORDS(0x1);
     BitField<0, 24, u32> viewport_size_y;
 
-    INSERT_PADDING_WORDS(0x1bc);
+    INSERT_PADDING_WORDS(0xc);
+
+    union {
+        // Maps components of output vertex attributes to semantics
+        enum Semantic : u32
+        {
+            POSITION_X   =  0,
+            POSITION_Y   =  1,
+            POSITION_Z   =  2,
+            POSITION_W   =  3,
+
+            COLOR_R      =  8,
+            COLOR_G      =  9,
+            COLOR_B      = 10,
+            COLOR_A      = 11,
+
+            TEXCOORD0_U  = 12,
+            TEXCOORD0_V  = 13,
+            TEXCOORD1_U  = 14,
+            TEXCOORD1_V  = 15,
+            TEXCOORD2_U  = 22,
+            TEXCOORD2_V  = 23,
+
+            INVALID      = 31,
+        };
+
+        BitField< 0, 5, Semantic> map_x;
+        BitField< 8, 5, Semantic> map_y;
+        BitField<16, 5, Semantic> map_z;
+        BitField<24, 5, Semantic> map_w;
+    } vs_output_attributes[7];
+
+    INSERT_PADDING_WORDS(0x1a9);
 
     struct {
         enum class Format : u64 {
@@ -133,7 +165,7 @@ struct Regs {
 
         // Attribute loaders map the source vertex data to input attributes
         // This e.g. allows to load different attributes from different memory locations
-        struct Loader {
+        struct {
             // Source attribute data offset from the base address
             u32 data_offset;
 
@@ -189,7 +221,90 @@ struct Regs {
     u32 trigger_draw;
     u32 trigger_draw_indexed;
 
-    INSERT_PADDING_WORDS(0xd0);
+    INSERT_PADDING_WORDS(0x8a);
+
+    // Offset to shader program entry point (in words)
+    BitField<0, 16, u32> vs_main_offset;
+
+    union {
+        BitField< 0, 4, u64> attribute0_register;
+        BitField< 4, 4, u64> attribute1_register;
+        BitField< 8, 4, u64> attribute2_register;
+        BitField<12, 4, u64> attribute3_register;
+        BitField<16, 4, u64> attribute4_register;
+        BitField<20, 4, u64> attribute5_register;
+        BitField<24, 4, u64> attribute6_register;
+        BitField<28, 4, u64> attribute7_register;
+        BitField<32, 4, u64> attribute8_register;
+        BitField<36, 4, u64> attribute9_register;
+        BitField<40, 4, u64> attribute10_register;
+        BitField<44, 4, u64> attribute11_register;
+        BitField<48, 4, u64> attribute12_register;
+        BitField<52, 4, u64> attribute13_register;
+        BitField<56, 4, u64> attribute14_register;
+        BitField<60, 4, u64> attribute15_register;
+
+        int GetRegisterForAttribute(int attribute_index) {
+            u64 fields[] = {
+                attribute0_register,  attribute1_register,  attribute2_register,  attribute3_register,
+                attribute4_register,  attribute5_register,  attribute6_register,  attribute7_register,
+                attribute8_register,  attribute9_register,  attribute10_register, attribute11_register,
+                attribute12_register, attribute13_register, attribute14_register, attribute15_register,
+            };
+            return (int)fields[attribute_index];
+        }
+    } vs_input_register_map;
+
+    INSERT_PADDING_WORDS(0x3);
+
+    struct {
+        enum Format : u32
+        {
+            FLOAT24 = 0,
+            FLOAT32 = 1
+        };
+
+        bool IsFloat32() const {
+            return format == FLOAT32;
+        }
+
+        union {
+            // Index of the next uniform to write to
+            // TODO: ctrulib uses 8 bits for this, however that seems to yield lots of invalid indices
+            BitField<0, 7, u32> index;
+
+            BitField<31, 1, Format> format;
+        };
+
+        // Writing to these registers sets the "current" uniform.
+        // TODO: It's not clear how the hardware stores what the "current" uniform is.
+        u32 set_value[8];
+
+    } vs_uniform_setup;
+
+    INSERT_PADDING_WORDS(0x2);
+
+    struct {
+        u32 begin_load;
+
+        // Writing to these registers sets the "current" word in the shader program.
+        // TODO: It's not clear how the hardware stores what the "current" word is.
+        u32 set_word[8];
+    } vs_program;
+
+    INSERT_PADDING_WORDS(0x1);
+
+    // This register group is used to load an internal table of swizzling patterns,
+    // which are indexed by each shader instruction to specify vector component swizzling.
+    struct {
+        u32 begin_load;
+
+        // Writing to these registers sets the "current" swizzle pattern in the table.
+        // TODO: It's not clear how the hardware stores what the "current" swizzle pattern is.
+        u32 set_word[8];
+    } vs_swizzle_patterns;
+
+    INSERT_PADDING_WORDS(0x22);
 
 #undef INSERT_PADDING_WORDS_HELPER1
 #undef INSERT_PADDING_WORDS_HELPER2
@@ -219,6 +334,11 @@ struct Regs {
         ADD_FIELD(num_vertices);
         ADD_FIELD(trigger_draw);
         ADD_FIELD(trigger_draw_indexed);
+        ADD_FIELD(vs_main_offset);
+        ADD_FIELD(vs_input_register_map);
+        ADD_FIELD(vs_uniform_setup);
+        ADD_FIELD(vs_program);
+        ADD_FIELD(vs_swizzle_patterns);
 
         #undef ADD_FIELD
         #endif // _MSC_VER
@@ -259,17 +379,25 @@ private:
 
 ASSERT_REG_POSITION(viewport_size_x, 0x41);
 ASSERT_REG_POSITION(viewport_size_y, 0x43);
+ASSERT_REG_POSITION(vs_output_attributes[0], 0x50);
+ASSERT_REG_POSITION(vs_output_attributes[1], 0x51);
 ASSERT_REG_POSITION(vertex_attributes, 0x200);
 ASSERT_REG_POSITION(index_array, 0x227);
 ASSERT_REG_POSITION(num_vertices, 0x228);
 ASSERT_REG_POSITION(trigger_draw, 0x22e);
 ASSERT_REG_POSITION(trigger_draw_indexed, 0x22f);
+ASSERT_REG_POSITION(vs_main_offset, 0x2ba);
+ASSERT_REG_POSITION(vs_input_register_map, 0x2bb);
+ASSERT_REG_POSITION(vs_uniform_setup, 0x2c0);
+ASSERT_REG_POSITION(vs_program, 0x2cb);
+ASSERT_REG_POSITION(vs_swizzle_patterns, 0x2d5);
 
 #undef ASSERT_REG_POSITION
 #endif // !defined(_MSC_VER)
 
 // The total number of registers is chosen arbitrarily, but let's make sure it's not some odd value anyway.
-static_assert(sizeof(Regs) == 0x300 * sizeof(u32), "Invalid total size of register set");
+static_assert(sizeof(Regs) <= 0x300 * sizeof(u32), "Register set structure larger than it should be");
+static_assert(sizeof(Regs) >= 0x300 * sizeof(u32), "Register set structure smaller than it should be");
 
 extern Regs registers; // TODO: Not sure if we want to have one global instance for this
 
@@ -347,7 +475,6 @@ private:
     float value;
 };
 
-
 union CommandHeader {
     CommandHeader(u32 h) : hex(h) {}
 

src/video_core/vertex_shader.cpp

@@ -0,0 +1,270 @@
+// Copyright 2014 Citra Emulator Project
+// Licensed under GPLv2
+// Refer to the license.txt file included.
+
+#include "pica.h"
+#include "vertex_shader.h"
+#include <core/mem_map.h>
+#include <common/file_util.h>
+
+namespace Pica {
+
+namespace VertexShader {
+
+static struct {
+    Math::Vec4<float24> f[96];
+} shader_uniforms;
+
+
+// TODO: Not sure where the shader binary and swizzle patterns are supposed to be loaded to!
+// For now, we just keep these local arrays around.
+static u32 shader_memory[1024];
+static u32 swizzle_data[1024];
+
+void SubmitShaderMemoryChange(u32 addr, u32 value)
+{
+    shader_memory[addr] = value;
+}
+
+void SubmitSwizzleDataChange(u32 addr, u32 value)
+{
+    swizzle_data[addr] = value;
+}
+
+Math::Vec4<float24>& GetFloatUniform(u32 index)
+{
+    return shader_uniforms.f[index];
+}
+
+struct VertexShaderState {
+    u32* program_counter;
+
+    const float24* input_register_table[16];
+    float24* output_register_table[7*4];
+
+    Math::Vec4<float24> temporary_registers[16];
+    bool status_registers[2];
+
+    enum {
+        INVALID_ADDRESS = 0xFFFFFFFF
+    };
+    u32 call_stack[8]; // TODO: What is the maximal call stack depth?
+    u32* call_stack_pointer;
+};
+
+static void ProcessShaderCode(VertexShaderState& state) {
+    while (true) {
+        bool increment_pc = true;
+        bool exit_loop = false;
+        const Instruction& instr = *(const Instruction*)state.program_counter;
+
+        const float24* src1_ = (instr.common.src1 < 0x10) ? state.input_register_table[instr.common.src1]
+                             : (instr.common.src1 < 0x20) ? &state.temporary_registers[instr.common.src1-0x10].x
+                             : (instr.common.src1 < 0x80) ? &shader_uniforms.f[instr.common.src1-0x20].x
+                             : nullptr;
+        const float24* src2_ = (instr.common.src2 < 0x10) ? state.input_register_table[instr.common.src2]
+                             : &state.temporary_registers[instr.common.src2-0x10].x;
+        // TODO: Unsure about the limit values
+        float24* dest = (instr.common.dest <= 0x1C) ? state.output_register_table[instr.common.dest]
+                             : (instr.common.dest <= 0x3C) ? nullptr
+                             : (instr.common.dest <= 0x7C) ? &state.temporary_registers[(instr.common.dest-0x40)/4][instr.common.dest%4]
+                             : nullptr;
+
+        const SwizzlePattern& swizzle = *(SwizzlePattern*)&swizzle_data[instr.common.operand_desc_id];
+
+        const float24 src1[4] = {
+            src1_[(int)swizzle.GetSelectorSrc1(0)],
+            src1_[(int)swizzle.GetSelectorSrc1(1)],
+            src1_[(int)swizzle.GetSelectorSrc1(2)],
+            src1_[(int)swizzle.GetSelectorSrc1(3)],
+        };
+        const float24 src2[4] = {
+            src2_[(int)swizzle.GetSelectorSrc2(0)],
+            src2_[(int)swizzle.GetSelectorSrc2(1)],
+            src2_[(int)swizzle.GetSelectorSrc2(2)],
+            src2_[(int)swizzle.GetSelectorSrc2(3)],
+        };
+
+        switch (instr.opcode) {
+            case Instruction::OpCode::ADD:
+            {
+                for (int i = 0; i < 4; ++i) {
+                    if (!swizzle.DestComponentEnabled(i))
+                        continue;
+
+                    dest[i] = src1[i] + src2[i];
+                }
+
+                break;
+            }
+
+            case Instruction::OpCode::MUL:
+            {
+                for (int i = 0; i < 4; ++i) {
+                    if (!swizzle.DestComponentEnabled(i))
+                        continue;
+
+                    dest[i] = src1[i] * src2[i];
+                }
+
+                break;
+            }
+
+            case Instruction::OpCode::DP3:
+            case Instruction::OpCode::DP4:
+            {
+                float24 dot = float24::FromFloat32(0.f);
+                int num_components = (instr.opcode == Instruction::OpCode::DP3) ? 3 : 4;
+                for (int i = 0; i < num_components; ++i)
+                    dot = dot + src1[i] * src2[i];
+
+                for (int i = 0; i < num_components; ++i) {
+                    if (!swizzle.DestComponentEnabled(i))
+                        continue;
+
+                    dest[i] = dot;
+                }
+                break;
+            }
+
+            // Reciprocal
+            case Instruction::OpCode::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(1.0 / src1[i].ToFloat32());
+                }
+
+                break;
+            }
+
+            // Reciprocal Square Root
+            case Instruction::OpCode::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(1.0 / sqrt(src1[i].ToFloat32()));
+                }
+
+                break;
+            }
+
+            case Instruction::OpCode::MOV:
+            {
+                for (int i = 0; i < 4; ++i) {
+                    if (!swizzle.DestComponentEnabled(i))
+                        continue;
+
+                    dest[i] = src1[i];
+                }
+                break;
+            }
+
+            case Instruction::OpCode::RET:
+                if (*state.call_stack_pointer == VertexShaderState::INVALID_ADDRESS) {
+                    exit_loop = true;
+                } else {
+                    state.program_counter = &shader_memory[*state.call_stack_pointer--];
+                    *state.call_stack_pointer = VertexShaderState::INVALID_ADDRESS;
+                }
+
+                break;
+
+            case Instruction::OpCode::CALL:
+                increment_pc = false;
+
+                _dbg_assert_(GPU, state.call_stack_pointer - state.call_stack < sizeof(state.call_stack));
+
+                *++state.call_stack_pointer = state.program_counter - shader_memory;
+                // TODO: Does this offset refer to the beginning of shader memory?
+                state.program_counter = &shader_memory[instr.flow_control.offset_words];
+                break;
+
+            case Instruction::OpCode::FLS:
+                // TODO: Do whatever needs to be done here?
+                break;
+
+            default:
+                ERROR_LOG(GPU, "Unhandled instruction: 0x%02x (%s): 0x%08x",
+                          (int)instr.opcode.Value(), instr.GetOpCodeName().c_str(), instr.hex);
+                break;
+        }
+
+        if (increment_pc)
+            ++state.program_counter;
+
+        if (exit_loop)
+            break;
+    }
+}
+
+OutputVertex RunShader(const InputVertex& input, int num_attributes)
+{
+    VertexShaderState state;
+
+    const u32* main = &shader_memory[registers.vs_main_offset];
+    state.program_counter = (u32*)main;
+
+    // Setup input register table
+    const auto& attribute_register_map = registers.vs_input_register_map;
+    float24 dummy_register;
+    std::fill(&state.input_register_table[0], &state.input_register_table[16], &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;
+
+    // Setup output register table
+    OutputVertex ret;
+    for (int i = 0; i < 7; ++i) {
+        const auto& output_register_map = registers.vs_output_attributes[i];
+
+        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)
+            state.output_register_table[4*i+comp] = ((float24*)&ret) + semantics[comp];
+    }
+
+    state.status_registers[0] = false;
+    state.status_registers[1] = false;
+    std::fill(state.call_stack, state.call_stack + sizeof(state.call_stack) / sizeof(state.call_stack[0]),
+              VertexShaderState::INVALID_ADDRESS);
+    state.call_stack_pointer = &state.call_stack[0];
+
+    ProcessShaderCode(state);
+
+    DEBUG_LOG(GPU, "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

src/video_core/vertex_shader.h

@@ -0,0 +1,211 @@
+// Copyright 2014 Citra Emulator Project
+// Licensed under GPLv2
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <initializer_list>
+
+#include <common/common_types.h>
+
+#include "math.h"
+#include "pica.h"
+
+namespace Pica {
+
+namespace VertexShader {
+
+struct InputVertex {
+    Math::Vec4<float24> attr[16];
+};
+
+struct OutputVertex {
+    OutputVertex() = default;
+
+    // VS output attributes
+    Math::Vec4<float24> pos;
+    Math::Vec4<float24> dummy; // quaternions (not implemented, yet)
+    Math::Vec4<float24> color;
+    Math::Vec2<float24> tc0;
+    float24 tc0_v;
+
+    // Padding for optimal alignment
+    float24 pad[14];
+
+    // Attributes used to store intermediate results
+
+    // position after perspective divide
+    Math::Vec3<float24> screenpos;
+
+    // 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);
+
+        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<OutputVertex>::value, "Structure is not POD");
+
+union Instruction {
+    enum class OpCode : u32 {
+        ADD = 0x0,
+        DP3 = 0x1,
+        DP4 = 0x2,
+
+        MUL = 0x8,
+
+        MAX = 0xC,
+        MIN = 0xD,
+        RCP = 0xE,
+        RSQ = 0xF,
+
+        MOV = 0x13,
+
+        RET = 0x21,
+        FLS = 0x22, // Flush
+        CALL = 0x24,
+    };
+
+    std::string GetOpCodeName() const {
+        std::map<OpCode, std::string> map = {
+            { OpCode::ADD, "ADD" },
+            { OpCode::DP3, "DP3" },
+            { OpCode::DP4, "DP4" },
+            { OpCode::MUL, "MUL" },
+            { OpCode::MAX, "MAX" },
+            { OpCode::MIN, "MIN" },
+            { OpCode::RCP, "RCP" },
+            { OpCode::RSQ, "RSQ" },
+            { OpCode::MOV, "MOV" },
+            { OpCode::RET, "RET" },
+            { OpCode::FLS, "FLS" },
+        };
+        auto it = map.find(opcode);
+        if (it == map.end())
+            return "UNK";
+        else
+            return it->second;
+    }
+
+    u32 hex;
+
+    BitField<0x1a, 0x6, OpCode> opcode;
+
+    // General notes:
+    //
+    // When two input registers are used, one of them uses a 5-bit index while the other
+    // one uses a 7-bit index. This is because at most one floating point uniform may be used
+    // as an input.
+
+
+    // Format used e.g. by arithmetic instructions and comparisons
+    // "src1" and "src2" specify register indices (i.e. indices referring to groups of 4 floats),
+    // while "dest" addresses individual floats.
+    union {
+        BitField<0x00, 0x5, u32> operand_desc_id;
+        BitField<0x07, 0x5, u32> src2;
+        BitField<0x0c, 0x7, u32> src1;
+        BitField<0x13, 0x7, u32> dest;
+    } common;
+
+    // Format used for flow control instructions ("if")
+    union {
+        BitField<0x00, 0x8, u32> num_instructions;
+        BitField<0x0a, 0xc, u32> offset_words;
+    } flow_control;
+};
+
+union SwizzlePattern {
+    u32 hex;
+
+    enum class Selector : u32 {
+        x = 0,
+        y = 1,
+        z = 2,
+        w = 3
+    };
+
+    Selector GetSelectorSrc1(int comp) const {
+        Selector selectors[] = {
+            src1_selector_0, src1_selector_1, src1_selector_2, src1_selector_3
+        };
+        return selectors[comp];
+    }
+
+    Selector GetSelectorSrc2(int comp) const {
+        Selector selectors[] = {
+            src2_selector_0, src2_selector_1, src2_selector_2, src2_selector_3
+        };
+        return selectors[comp];
+    }
+
+    bool DestComponentEnabled(int i) const {
+        return (dest_mask & (0x8 >> i));
+    }
+
+    std::string SelectorToString(bool src2) const {
+        std::map<Selector, std::string> map = {
+            { Selector::x, "x" },
+            { Selector::y, "y" },
+            { Selector::z, "z" },
+            { Selector::w, "w" }
+        };
+        std::string ret;
+        for (int i = 0; i < 4; ++i) {
+            ret += map.at(src2 ? GetSelectorSrc2(i) : GetSelectorSrc1(i));
+        }
+        return ret;
+    }
+
+    std::string DestMaskToString() const {
+        std::string ret;
+        for (int i = 0; i < 4; ++i) {
+            if (!DestComponentEnabled(i))
+                ret += "_";
+            else
+                ret += "xyzw"[i];
+        }
+        return ret;
+    }
+
+    // Components of "dest" that should be written to: LSB=dest.w, MSB=dest.x
+    BitField< 0, 4, u32> dest_mask;
+
+    BitField< 5, 2, Selector> src1_selector_3;
+    BitField< 7, 2, Selector> src1_selector_2;
+    BitField< 9, 2, Selector> src1_selector_1;
+    BitField<11, 2, Selector> src1_selector_0;
+
+    BitField<14, 2, Selector> src2_selector_3;
+    BitField<16, 2, Selector> src2_selector_2;
+    BitField<18, 2, Selector> src2_selector_1;
+    BitField<20, 2, Selector> src2_selector_0;
+
+    BitField<31, 1, u32> flag; // not sure what this means, maybe it's the sign?
+};
+
+void SubmitShaderMemoryChange(u32 addr, u32 value);
+void SubmitSwizzleDataChange(u32 addr, u32 value);
+
+OutputVertex RunShader(const InputVertex& input, int num_attributes);
+
+Math::Vec4<float24>& GetFloatUniform(u32 index);
+
+} // namespace
+
+} // namespace
+

src/video_core/video_core.vcxproj

@@ -22,6 +22,7 @@
     <ClCompile Include="renderer_opengl\renderer_opengl.cpp" />
     <ClCompile Include="command_processor.cpp" />
     <ClCompile Include="utils.cpp" />
+    <ClCompile Include="vertex_shader.cpp" />
     <ClCompile Include="video_core.cpp" />
   </ItemGroup>
   <ItemGroup>
@@ -31,6 +32,7 @@
     <ClInclude Include="pica.h" />
     <ClInclude Include="renderer_base.h" />
     <ClInclude Include="utils.h" />
+    <ClInclude Include="vertex_shader.h" />
     <ClInclude Include="video_core.h" />
     <ClInclude Include="renderer_opengl\renderer_opengl.h" />
   </ItemGroup>

src/video_core/video_core.vcxproj.filters

@@ -11,6 +11,7 @@
     </ClCompile>
     <ClCompile Include="command_processor.cpp" />
     <ClCompile Include="utils.cpp" />
+    <ClCompile Include="vertex_shader.cpp" />
     <ClCompile Include="video_core.cpp" />
   </ItemGroup>
   <ItemGroup>
@@ -23,6 +24,7 @@
     <ClInclude Include="pica.h" />
     <ClInclude Include="renderer_base.h" />
     <ClInclude Include="utils.h" />
+    <ClInclude Include="vertex_shader.h" />
     <ClInclude Include="video_core.h" />
   </ItemGroup>
   <ItemGroup>