Merge pull request #2697 from wwylele/proctex

Implemented Procedural Texture (Texture Unit 3)

src/common/vector_math.h

@@ -652,6 +652,16 @@ static inline decltype((X{} * int{} + X{} * int{}) / base) LerpInt(const X& begi
     return (begin * (base - t) + end * t) / base;
 }
 
+// bilinear interpolation. s is for interpolating x00-x01 and x10-x11, and t is for the second
+// interpolation.
+template <typename X>
+inline auto BilinearInterp(const X& x00, const X& x01, const X& x10, const X& x11, const float s,
+                           const float t) {
+    auto y0 = Lerp(x00, x01, s);
+    auto y1 = Lerp(x10, x11, s);
+    return Lerp(y0, y1, t);
+}
+
 // Utility vector factories
 template <typename T>
 static inline Vec2<T> MakeVec(const T& x, const T& y) {

src/video_core/CMakeLists.txt

@@ -15,6 +15,7 @@ set(SRCS
             shader/shader_interpreter.cpp
             swrasterizer/clipper.cpp
             swrasterizer/framebuffer.cpp
+            swrasterizer/proctex.cpp
             swrasterizer/rasterizer.cpp
             swrasterizer/swrasterizer.cpp
             swrasterizer/texturing.cpp
@@ -54,6 +55,7 @@ set(HEADERS
             shader/shader_interpreter.h
             swrasterizer/clipper.h
             swrasterizer/framebuffer.h
+            swrasterizer/proctex.h
             swrasterizer/rasterizer.h
             swrasterizer/swrasterizer.h
             swrasterizer/texturing.h

src/video_core/command_processor.cpp

@@ -556,6 +556,37 @@ static void WritePicaReg(u32 id, u32 value, u32 mask) {
         break;
     }
 
+    case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[0], 0xb0):
+    case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[1], 0xb1):
+    case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[2], 0xb2):
+    case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[3], 0xb3):
+    case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[4], 0xb4):
+    case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[5], 0xb5):
+    case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[6], 0xb6):
+    case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[7], 0xb7): {
+        auto& index = regs.texturing.proctex_lut_config.index;
+        auto& pt = g_state.proctex;
+
+        switch (regs.texturing.proctex_lut_config.ref_table.Value()) {
+        case TexturingRegs::ProcTexLutTable::Noise:
+            pt.noise_table[index % pt.noise_table.size()].raw = value;
+            break;
+        case TexturingRegs::ProcTexLutTable::ColorMap:
+            pt.color_map_table[index % pt.color_map_table.size()].raw = value;
+            break;
+        case TexturingRegs::ProcTexLutTable::AlphaMap:
+            pt.alpha_map_table[index % pt.alpha_map_table.size()].raw = value;
+            break;
+        case TexturingRegs::ProcTexLutTable::Color:
+            pt.color_table[index % pt.color_table.size()].raw = value;
+            break;
+        case TexturingRegs::ProcTexLutTable::ColorDiff:
+            pt.color_diff_table[index % pt.color_diff_table.size()].raw = value;
+            break;
+        }
+        index.Assign(index + 1);
+        break;
+    }
     default:
         break;
     }

src/video_core/pica_state.h

@@ -7,6 +7,7 @@
 #include <array>
 #include "common/bit_field.h"
 #include "common/common_types.h"
+#include "common/vector_math.h"
 #include "video_core/primitive_assembly.h"
 #include "video_core/regs.h"
 #include "video_core/shader/shader.h"
@@ -25,6 +26,59 @@ struct State {
 
     Shader::AttributeBuffer input_default_attributes;
 
+    struct ProcTex {
+        union ValueEntry {
+            u32 raw;
+
+            // LUT value, encoded as 12-bit fixed point, with 12 fraction bits
+            BitField<0, 12, u32> value; // 0.0.12 fixed point
+
+            // Difference between two entry values. Used for efficient interpolation.
+            // 0.0.12 fixed point with two's complement. The range is [-0.5, 0.5).
+            // Note: the type of this is different from the one of lighting LUT
+            BitField<12, 12, s32> difference;
+
+            float ToFloat() const {
+                return static_cast<float>(value) / 4095.f;
+            }
+
+            float DiffToFloat() const {
+                return static_cast<float>(difference) / 4095.f;
+            }
+        };
+
+        union ColorEntry {
+            u32 raw;
+            BitField<0, 8, u32> r;
+            BitField<8, 8, u32> g;
+            BitField<16, 8, u32> b;
+            BitField<24, 8, u32> a;
+
+            Math::Vec4<u8> ToVector() const {
+                return {static_cast<u8>(r), static_cast<u8>(g), static_cast<u8>(b),
+                        static_cast<u8>(a)};
+            }
+        };
+
+        union ColorDifferenceEntry {
+            u32 raw;
+            BitField<0, 8, s32> r; // half of the difference between two ColorEntry
+            BitField<8, 8, s32> g;
+            BitField<16, 8, s32> b;
+            BitField<24, 8, s32> a;
+
+            Math::Vec4<s32> ToVector() const {
+                return Math::Vec4<s32>{r, g, b, a} * 2;
+            }
+        };
+
+        std::array<ValueEntry, 128> noise_table;
+        std::array<ValueEntry, 128> color_map_table;
+        std::array<ValueEntry, 128> alpha_map_table;
+        std::array<ColorEntry, 256> color_table;
+        std::array<ColorDifferenceEntry, 256> color_diff_table;
+    } proctex;
+
     struct {
         union LutEntry {
             // Used for raw access

src/video_core/regs.h

@@ -101,6 +101,13 @@ ASSERT_REG_POSITION(texturing.texture1, 0x91);
 ASSERT_REG_POSITION(texturing.texture1_format, 0x96);
 ASSERT_REG_POSITION(texturing.texture2, 0x99);
 ASSERT_REG_POSITION(texturing.texture2_format, 0x9e);
+ASSERT_REG_POSITION(texturing.proctex, 0xa8);
+ASSERT_REG_POSITION(texturing.proctex_noise_u, 0xa9);
+ASSERT_REG_POSITION(texturing.proctex_noise_v, 0xaa);
+ASSERT_REG_POSITION(texturing.proctex_noise_frequency, 0xab);
+ASSERT_REG_POSITION(texturing.proctex_lut, 0xac);
+ASSERT_REG_POSITION(texturing.proctex_lut_offset, 0xad);
+ASSERT_REG_POSITION(texturing.proctex_lut_config, 0xaf);
 ASSERT_REG_POSITION(texturing.tev_stage0, 0xc0);
 ASSERT_REG_POSITION(texturing.tev_stage1, 0xc8);
 ASSERT_REG_POSITION(texturing.tev_stage2, 0xd0);

src/video_core/regs_texturing.h

@@ -127,8 +127,8 @@ struct TexturingRegs {
         BitField<0, 1, u32> texture0_enable;
         BitField<1, 1, u32> texture1_enable;
         BitField<2, 1, u32> texture2_enable;
-        BitField<8, 2, u32> texture3_coordinates; // TODO: unimplemented
+        BitField<8, 2, u32> texture3_coordinates;
-        BitField<10, 1, u32> texture3_enable;     // TODO: unimplemented
+        BitField<10, 1, u32> texture3_enable;
         BitField<13, 1, u32> texture2_use_coord1;
         BitField<16, 1, u32> clear_texture_cache; // TODO: unimplemented
     } main_config;
@@ -142,7 +142,7 @@ struct TexturingRegs {
     INSERT_PADDING_WORDS(0x2);
     TextureConfig texture2;
     BitField<0, 4, TextureFormat> texture2_format;
-    INSERT_PADDING_WORDS(0x21);
+    INSERT_PADDING_WORDS(0x9);
 
     struct FullTextureConfig {
         const bool enabled;
@@ -157,6 +157,96 @@ struct TexturingRegs {
         }};
     }
 
+    // 0xa8-0xad: ProcTex Config
+    enum class ProcTexClamp : u32 {
+        ToZero = 0,
+        ToEdge = 1,
+        SymmetricalRepeat = 2,
+        MirroredRepeat = 3,
+        Pulse = 4,
+    };
+
+    enum class ProcTexCombiner : u32 {
+        U = 0,        // u
+        U2 = 1,       // u * u
+        V = 2,        // v
+        V2 = 3,       // v * v
+        Add = 4,      // (u + v) / 2
+        Add2 = 5,     // (u * u + v * v) / 2
+        SqrtAdd2 = 6, // sqrt(u * u + v * v)
+        Min = 7,      // min(u, v)
+        Max = 8,      // max(u, v)
+        RMax = 9,     // Average of Max and SqrtAdd2
+    };
+
+    enum class ProcTexShift : u32 {
+        None = 0,
+        Odd = 1,
+        Even = 2,
+    };
+
+    union {
+        BitField<0, 3, ProcTexClamp> u_clamp;
+        BitField<3, 3, ProcTexClamp> v_clamp;
+        BitField<6, 4, ProcTexCombiner> color_combiner;
+        BitField<10, 4, ProcTexCombiner> alpha_combiner;
+        BitField<14, 1, u32> separate_alpha;
+        BitField<15, 1, u32> noise_enable;
+        BitField<16, 2, ProcTexShift> u_shift;
+        BitField<18, 2, ProcTexShift> v_shift;
+        BitField<20, 8, u32> bias_low; // float16 TODO: unimplemented
+    } proctex;
+
+    union ProcTexNoiseConfig {
+        BitField<0, 16, s32> amplitude; // fixed1.3.12
+        BitField<16, 16, u32> phase;    // float16
+    };
+
+    ProcTexNoiseConfig proctex_noise_u;
+    ProcTexNoiseConfig proctex_noise_v;
+
+    union {
+        BitField<0, 16, u32> u;  // float16
+        BitField<16, 16, u32> v; // float16
+    } proctex_noise_frequency;
+
+    enum class ProcTexFilter : u32 {
+        Nearest = 0,
+        Linear = 1,
+        NearestMipmapNearest = 2,
+        LinearMipmapNearest = 3,
+        NearestMipmapLinear = 4,
+        LinearMipmapLinear = 5,
+    };
+
+    union {
+        BitField<0, 3, ProcTexFilter> filter;
+        BitField<11, 8, u32> width;
+        BitField<19, 8, u32> bias_high; // TODO: unimplemented
+    } proctex_lut;
+
+    BitField<0, 8, u32> proctex_lut_offset;
+
+    INSERT_PADDING_WORDS(0x1);
+
+    // 0xaf-0xb7: ProcTex LUT
+    enum class ProcTexLutTable : u32 {
+        Noise = 0,
+        ColorMap = 2,
+        AlphaMap = 3,
+        Color = 4,
+        ColorDiff = 5,
+    };
+
+    union {
+        BitField<0, 8, u32> index;
+        BitField<8, 4, ProcTexLutTable> ref_table;
+    } proctex_lut_config;
+
+    u32 proctex_lut_data[8];
+
+    INSERT_PADDING_WORDS(0x8);
+
     // 0xc0-0xff: Texture Combiner (akin to glTexEnv)
     struct TevStageConfig {
         enum class Source : u32 {

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -55,6 +55,12 @@ RasterizerOpenGL::RasterizerOpenGL() : shader_dirty(true) {
 
     uniform_block_data.fog_lut_dirty = true;
 
+    uniform_block_data.proctex_noise_lut_dirty = true;
+    uniform_block_data.proctex_color_map_dirty = true;
+    uniform_block_data.proctex_alpha_map_dirty = true;
+    uniform_block_data.proctex_lut_dirty = true;
+    uniform_block_data.proctex_diff_lut_dirty = true;
+
     // Set vertex attributes
     glVertexAttribPointer(GLShader::ATTRIBUTE_POSITION, 4, GL_FLOAT, GL_FALSE,
                           sizeof(HardwareVertex), (GLvoid*)offsetof(HardwareVertex, position));
@@ -115,6 +121,51 @@ RasterizerOpenGL::RasterizerOpenGL() : shader_dirty(true) {
     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
 
+    // Setup the noise LUT for proctex
+    proctex_noise_lut.Create();
+    state.proctex_noise_lut.texture_1d = proctex_noise_lut.handle;
+    state.Apply();
+    glActiveTexture(GL_TEXTURE10);
+    glTexImage1D(GL_TEXTURE_1D, 0, GL_RG32F, 128, 0, GL_RG, GL_FLOAT, nullptr);
+    glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+    glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+
+    // Setup the color map for proctex
+    proctex_color_map.Create();
+    state.proctex_color_map.texture_1d = proctex_color_map.handle;
+    state.Apply();
+    glActiveTexture(GL_TEXTURE11);
+    glTexImage1D(GL_TEXTURE_1D, 0, GL_RG32F, 128, 0, GL_RG, GL_FLOAT, nullptr);
+    glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+    glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+
+    // Setup the alpha map for proctex
+    proctex_alpha_map.Create();
+    state.proctex_alpha_map.texture_1d = proctex_alpha_map.handle;
+    state.Apply();
+    glActiveTexture(GL_TEXTURE12);
+    glTexImage1D(GL_TEXTURE_1D, 0, GL_RG32F, 128, 0, GL_RG, GL_FLOAT, nullptr);
+    glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+    glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+
+    // Setup the LUT for proctex
+    proctex_lut.Create();
+    state.proctex_lut.texture_1d = proctex_lut.handle;
+    state.Apply();
+    glActiveTexture(GL_TEXTURE13);
+    glTexImage1D(GL_TEXTURE_1D, 0, GL_RGBA32F, 256, 0, GL_RGBA, GL_FLOAT, nullptr);
+    glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+    glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+
+    // Setup the difference LUT for proctex
+    proctex_diff_lut.Create();
+    state.proctex_diff_lut.texture_1d = proctex_diff_lut.handle;
+    state.Apply();
+    glActiveTexture(GL_TEXTURE14);
+    glTexImage1D(GL_TEXTURE_1D, 0, GL_RGBA32F, 256, 0, GL_RGBA, GL_FLOAT, nullptr);
+    glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+    glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+
     // Sync fixed function OpenGL state
     SyncCullMode();
     SyncBlendEnabled();
@@ -272,6 +323,36 @@ void RasterizerOpenGL::DrawTriangles() {
         uniform_block_data.fog_lut_dirty = false;
     }
 
+    // Sync the proctex noise lut
+    if (uniform_block_data.proctex_noise_lut_dirty) {
+        SyncProcTexNoiseLUT();
+        uniform_block_data.proctex_noise_lut_dirty = false;
+    }
+
+    // Sync the proctex color map
+    if (uniform_block_data.proctex_color_map_dirty) {
+        SyncProcTexColorMap();
+        uniform_block_data.proctex_color_map_dirty = false;
+    }
+
+    // Sync the proctex alpha map
+    if (uniform_block_data.proctex_alpha_map_dirty) {
+        SyncProcTexAlphaMap();
+        uniform_block_data.proctex_alpha_map_dirty = false;
+    }
+
+    // Sync the proctex lut
+    if (uniform_block_data.proctex_lut_dirty) {
+        SyncProcTexLUT();
+        uniform_block_data.proctex_lut_dirty = false;
+    }
+
+    // Sync the proctex difference lut
+    if (uniform_block_data.proctex_diff_lut_dirty) {
+        SyncProcTexDiffLUT();
+        uniform_block_data.proctex_diff_lut_dirty = false;
+    }
+
     // Sync the uniform data
     if (uniform_block_data.dirty) {
         glBufferData(GL_UNIFORM_BUFFER, sizeof(UniformData), &uniform_block_data.data,
@@ -354,6 +435,47 @@ void RasterizerOpenGL::NotifyPicaRegisterChanged(u32 id) {
         uniform_block_data.fog_lut_dirty = true;
         break;
 
+    // ProcTex state
+    case PICA_REG_INDEX(texturing.proctex):
+    case PICA_REG_INDEX(texturing.proctex_lut):
+    case PICA_REG_INDEX(texturing.proctex_lut_offset):
+        shader_dirty = true;
+        break;
+
+    case PICA_REG_INDEX(texturing.proctex_noise_u):
+    case PICA_REG_INDEX(texturing.proctex_noise_v):
+    case PICA_REG_INDEX(texturing.proctex_noise_frequency):
+        SyncProcTexNoise();
+        break;
+
+    case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[0], 0xb0):
+    case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[1], 0xb1):
+    case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[2], 0xb2):
+    case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[3], 0xb3):
+    case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[4], 0xb4):
+    case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[5], 0xb5):
+    case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[6], 0xb6):
+    case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[7], 0xb7):
+        using Pica::TexturingRegs;
+        switch (regs.texturing.proctex_lut_config.ref_table.Value()) {
+        case TexturingRegs::ProcTexLutTable::Noise:
+            uniform_block_data.proctex_noise_lut_dirty = true;
+            break;
+        case TexturingRegs::ProcTexLutTable::ColorMap:
+            uniform_block_data.proctex_color_map_dirty = true;
+            break;
+        case TexturingRegs::ProcTexLutTable::AlphaMap:
+            uniform_block_data.proctex_alpha_map_dirty = true;
+            break;
+        case TexturingRegs::ProcTexLutTable::Color:
+            uniform_block_data.proctex_lut_dirty = true;
+            break;
+        case TexturingRegs::ProcTexLutTable::ColorDiff:
+            uniform_block_data.proctex_diff_lut_dirty = true;
+            break;
+        }
+        break;
+
     // Alpha test
     case PICA_REG_INDEX(framebuffer.output_merger.alpha_test):
         SyncAlphaTest();
@@ -1072,6 +1194,35 @@ void RasterizerOpenGL::SetShader() {
             glUniform1i(uniform_fog_lut, 9);
         }
 
+        GLuint uniform_proctex_noise_lut =
+            glGetUniformLocation(shader->shader.handle, "proctex_noise_lut");
+        if (uniform_proctex_noise_lut != -1) {
+            glUniform1i(uniform_proctex_noise_lut, 10);
+        }
+
+        GLuint uniform_proctex_color_map =
+            glGetUniformLocation(shader->shader.handle, "proctex_color_map");
+        if (uniform_proctex_color_map != -1) {
+            glUniform1i(uniform_proctex_color_map, 11);
+        }
+
+        GLuint uniform_proctex_alpha_map =
+            glGetUniformLocation(shader->shader.handle, "proctex_alpha_map");
+        if (uniform_proctex_alpha_map != -1) {
+            glUniform1i(uniform_proctex_alpha_map, 12);
+        }
+
+        GLuint uniform_proctex_lut = glGetUniformLocation(shader->shader.handle, "proctex_lut");
+        if (uniform_proctex_lut != -1) {
+            glUniform1i(uniform_proctex_lut, 13);
+        }
+
+        GLuint uniform_proctex_diff_lut =
+            glGetUniformLocation(shader->shader.handle, "proctex_diff_lut");
+        if (uniform_proctex_diff_lut != -1) {
+            glUniform1i(uniform_proctex_diff_lut, 14);
+        }
+
         current_shader = shader_cache.emplace(config, std::move(shader)).first->second.get();
 
         GLuint block_index = glGetUniformBlockIndex(current_shader->shader.handle, "shader_data");
@@ -1105,6 +1256,7 @@ void RasterizerOpenGL::SetShader() {
             }
 
             SyncFogColor();
+            SyncProcTexNoise();
         }
     }
 }
@@ -1204,6 +1356,86 @@ void RasterizerOpenGL::SyncFogLUT() {
     }
 }
 
+void RasterizerOpenGL::SyncProcTexNoise() {
+    const auto& regs = Pica::g_state.regs.texturing;
+    uniform_block_data.data.proctex_noise_f = {
+        Pica::float16::FromRaw(regs.proctex_noise_frequency.u).ToFloat32(),
+        Pica::float16::FromRaw(regs.proctex_noise_frequency.v).ToFloat32(),
+    };
+    uniform_block_data.data.proctex_noise_a = {
+        regs.proctex_noise_u.amplitude / 4095.0f, regs.proctex_noise_v.amplitude / 4095.0f,
+    };
+    uniform_block_data.data.proctex_noise_p = {
+        Pica::float16::FromRaw(regs.proctex_noise_u.phase).ToFloat32(),
+        Pica::float16::FromRaw(regs.proctex_noise_v.phase).ToFloat32(),
+    };
+
+    uniform_block_data.dirty = true;
+}
+
+// helper function for SyncProcTexNoiseLUT/ColorMap/AlphaMap
+static void SyncProcTexValueLUT(const std::array<Pica::State::ProcTex::ValueEntry, 128>& lut,
+                                std::array<GLvec2, 128>& lut_data, GLenum texture) {
+    std::array<GLvec2, 128> new_data;
+    std::transform(lut.begin(), lut.end(), new_data.begin(), [](const auto& entry) {
+        return GLvec2{entry.ToFloat(), entry.DiffToFloat()};
+    });
+
+    if (new_data != lut_data) {
+        lut_data = new_data;
+        glActiveTexture(texture);
+        glTexSubImage1D(GL_TEXTURE_1D, 0, 0, 128, GL_RG, GL_FLOAT, lut_data.data());
+    }
+}
+
+void RasterizerOpenGL::SyncProcTexNoiseLUT() {
+    SyncProcTexValueLUT(Pica::g_state.proctex.noise_table, proctex_noise_lut_data, GL_TEXTURE10);
+}
+
+void RasterizerOpenGL::SyncProcTexColorMap() {
+    SyncProcTexValueLUT(Pica::g_state.proctex.color_map_table, proctex_color_map_data,
+                        GL_TEXTURE11);
+}
+
+void RasterizerOpenGL::SyncProcTexAlphaMap() {
+    SyncProcTexValueLUT(Pica::g_state.proctex.alpha_map_table, proctex_alpha_map_data,
+                        GL_TEXTURE12);
+}
+
+void RasterizerOpenGL::SyncProcTexLUT() {
+    std::array<GLvec4, 256> new_data;
+
+    std::transform(Pica::g_state.proctex.color_table.begin(),
+                   Pica::g_state.proctex.color_table.end(), new_data.begin(),
+                   [](const auto& entry) {
+                       auto rgba = entry.ToVector() / 255.0f;
+                       return GLvec4{rgba.r(), rgba.g(), rgba.b(), rgba.a()};
+                   });
+
+    if (new_data != proctex_lut_data) {
+        proctex_lut_data = new_data;
+        glActiveTexture(GL_TEXTURE13);
+        glTexSubImage1D(GL_TEXTURE_1D, 0, 0, 256, GL_RGBA, GL_FLOAT, proctex_lut_data.data());
+    }
+}
+
+void RasterizerOpenGL::SyncProcTexDiffLUT() {
+    std::array<GLvec4, 256> new_data;
+
+    std::transform(Pica::g_state.proctex.color_diff_table.begin(),
+                   Pica::g_state.proctex.color_diff_table.end(), new_data.begin(),
+                   [](const auto& entry) {
+                       auto rgba = entry.ToVector() / 255.0f;
+                       return GLvec4{rgba.r(), rgba.g(), rgba.b(), rgba.a()};
+                   });
+
+    if (new_data != proctex_diff_lut_data) {
+        proctex_diff_lut_data = new_data;
+        glActiveTexture(GL_TEXTURE14);
+        glTexSubImage1D(GL_TEXTURE_1D, 0, 0, 256, GL_RGBA, GL_FLOAT, proctex_diff_lut_data.data());
+    }
+}
+
 void RasterizerOpenGL::SyncAlphaTest() {
     const auto& regs = Pica::g_state.regs;
     if (regs.framebuffer.output_merger.alpha_test.ref != uniform_block_data.data.alphatest_ref) {

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -143,6 +143,9 @@ private:
         GLint scissor_x2;
         GLint scissor_y2;
         alignas(16) GLvec3 fog_color;
+        alignas(8) GLvec2 proctex_noise_f;
+        alignas(8) GLvec2 proctex_noise_a;
+        alignas(8) GLvec2 proctex_noise_p;
         alignas(16) GLvec3 lighting_global_ambient;
         LightSrc light_src[8];
         alignas(16) GLvec4 const_color[6]; // A vec4 color for each of the six tev stages
@@ -150,7 +153,7 @@ private:
     };
 
     static_assert(
-        sizeof(UniformData) == 0x3C0,
+        sizeof(UniformData) == 0x3E0,
         "The size of the UniformData structure has changed, update the structure in the shader");
     static_assert(sizeof(UniformData) < 16384,
                   "UniformData structure must be less than 16kb as per the OpenGL spec");
@@ -180,6 +183,16 @@ private:
     void SyncFogColor();
     void SyncFogLUT();
 
+    /// Sync the procedural texture noise configuration to match the PICA register
+    void SyncProcTexNoise();
+
+    /// Sync the procedural texture lookup tables
+    void SyncProcTexNoiseLUT();
+    void SyncProcTexColorMap();
+    void SyncProcTexAlphaMap();
+    void SyncProcTexLUT();
+    void SyncProcTexDiffLUT();
+
     /// Syncs the alpha test states to match the PICA register
     void SyncAlphaTest();
 
@@ -248,6 +261,11 @@ private:
         UniformData data;
         bool lut_dirty[6];
         bool fog_lut_dirty;
+        bool proctex_noise_lut_dirty;
+        bool proctex_color_map_dirty;
+        bool proctex_alpha_map_dirty;
+        bool proctex_lut_dirty;
+        bool proctex_diff_lut_dirty;
         bool dirty;
     } uniform_block_data = {};
 
@@ -262,4 +280,19 @@ private:
 
     OGLTexture fog_lut;
     std::array<GLuint, 128> fog_lut_data{};
+
+    OGLTexture proctex_noise_lut;
+    std::array<GLvec2, 128> proctex_noise_lut_data{};
+
+    OGLTexture proctex_color_map;
+    std::array<GLvec2, 128> proctex_color_map_data{};
+
+    OGLTexture proctex_alpha_map;
+    std::array<GLvec2, 128> proctex_alpha_map_data{};
+
+    OGLTexture proctex_lut;
+    std::array<GLvec4, 256> proctex_lut_data{};
+
+    OGLTexture proctex_diff_lut;
+    std::array<GLvec4, 256> proctex_diff_lut_data{};
 };

src/video_core/renderer_opengl/gl_shader_gen.cpp

@@ -114,6 +114,22 @@ PicaShaderConfig PicaShaderConfig::BuildFromRegs(const Pica::Regs& regs) {
     state.lighting.bump_renorm = regs.lighting.config0.disable_bump_renorm == 0;
     state.lighting.clamp_highlights = regs.lighting.config0.clamp_highlights != 0;
 
+    state.proctex.enable = regs.texturing.main_config.texture3_enable;
+    if (state.proctex.enable) {
+        state.proctex.coord = regs.texturing.main_config.texture3_coordinates;
+        state.proctex.u_clamp = regs.texturing.proctex.u_clamp;
+        state.proctex.v_clamp = regs.texturing.proctex.v_clamp;
+        state.proctex.color_combiner = regs.texturing.proctex.color_combiner;
+        state.proctex.alpha_combiner = regs.texturing.proctex.alpha_combiner;
+        state.proctex.separate_alpha = regs.texturing.proctex.separate_alpha;
+        state.proctex.noise_enable = regs.texturing.proctex.noise_enable;
+        state.proctex.u_shift = regs.texturing.proctex.u_shift;
+        state.proctex.v_shift = regs.texturing.proctex.v_shift;
+        state.proctex.lut_width = regs.texturing.proctex_lut.width;
+        state.proctex.lut_offset = regs.texturing.proctex_lut_offset;
+        state.proctex.lut_filter = regs.texturing.proctex_lut.filter;
+    }
+
     return res;
 }
 
@@ -132,8 +148,7 @@ static std::string TexCoord(const PicaShaderConfig& config, int texture_unit) {
     if (texture_unit == 2 && config.state.texture2_use_coord1) {
         return "texcoord[1]";
     }
-    // TODO: if texture unit 3 (procedural texture) implementation also uses this function,
+
-    //       config.state.texture3_coordinates should be repected here.
     return "texcoord[" + std::to_string(texture_unit) + "]";
 }
 
@@ -175,6 +190,14 @@ static void AppendSource(std::string& out, const PicaShaderConfig& config,
     case Source::Texture2:
         out += "texture(tex[2], " + TexCoord(config, 2) + ")";
         break;
+    case Source::Texture3:
+        if (config.state.proctex.enable) {
+            out += "ProcTex()";
+        } else {
+            LOG_ERROR(Render_OpenGL, "Using Texture3 without enabling it");
+            out += "vec4(0.0)";
+        }
+        break;
     case Source::PreviousBuffer:
         out += "combiner_buffer";
         break;
@@ -483,9 +506,18 @@ static void WriteLighting(std::string& out, const PicaShaderConfig& config) {
     if (lighting.bump_mode == LightingRegs::LightingBumpMode::NormalMap) {
         // Bump mapping is enabled using a normal map, read perturbation vector from the selected
         // texture
-        std::string bump_selector = std::to_string(lighting.bump_selector);
+        if (lighting.bump_selector == 3) {
-        out += "vec3 surface_normal = 2.0 * texture(tex[" + bump_selector + "], " +
+            if (config.state.proctex.enable) {
-               TexCoord(config, lighting.bump_selector) + ").rgb - 1.0;\n";
+                out += "vec3 surface_normal = 2.0 * ProcTex().rgb - 1.0;\n";
+            } else {
+                LOG_ERROR(Render_OpenGL, "Using Texture3 without enabling it");
+                out += "vec3 surface_normal = vec3(-1.0);\n";
+            }
+        } else {
+            std::string bump_selector = std::to_string(lighting.bump_selector);
+            out += "vec3 surface_normal = 2.0 * texture(tex[" + bump_selector + "], " +
+                   TexCoord(config, lighting.bump_selector) + ").rgb - 1.0;\n";
+        }
 
         // Recompute Z-component of perturbation if 'renorm' is enabled, this provides a higher
         // precision result
@@ -693,6 +725,221 @@ static void WriteLighting(std::string& out, const PicaShaderConfig& config) {
     out += "secondary_fragment_color = clamp(specular_sum, vec4(0.0), vec4(1.0));\n";
 }
 
+using ProcTexClamp = TexturingRegs::ProcTexClamp;
+using ProcTexShift = TexturingRegs::ProcTexShift;
+using ProcTexCombiner = TexturingRegs::ProcTexCombiner;
+using ProcTexFilter = TexturingRegs::ProcTexFilter;
+
+void AppendProcTexShiftOffset(std::string& out, const std::string& v, ProcTexShift mode,
+                              ProcTexClamp clamp_mode) {
+    std::string offset = (clamp_mode == ProcTexClamp::MirroredRepeat) ? "1.0" : "0.5";
+    switch (mode) {
+    case ProcTexShift::None:
+        out += "0";
+        break;
+    case ProcTexShift::Odd:
+        out += offset + " * ((int(" + v + ") / 2) % 2)";
+        break;
+    case ProcTexShift::Even:
+        out += offset + " * (((int(" + v + ") + 1) / 2) % 2)";
+        break;
+    default:
+        LOG_CRITICAL(HW_GPU, "Unknown shift mode %u", static_cast<u32>(mode));
+        out += "0";
+        break;
+    }
+}
+
+void AppendProcTexClamp(std::string& out, const std::string& var, ProcTexClamp mode) {
+    switch (mode) {
+    case ProcTexClamp::ToZero:
+        out += var + " = " + var + " > 1.0 ? 0 : " + var + ";\n";
+        break;
+    case ProcTexClamp::ToEdge:
+        out += var + " = " + "min(" + var + ", 1.0);\n";
+        break;
+    case ProcTexClamp::SymmetricalRepeat:
+        out += var + " = " + "fract(" + var + ");\n";
+        break;
+    case ProcTexClamp::MirroredRepeat: {
+        out +=
+            var + " = int(" + var + ") % 2 == 0 ? fract(" + var + ") : 1.0 - fract(" + var + ");\n";
+        break;
+    }
+    case ProcTexClamp::Pulse:
+        out += var + " = " + var + " > 0.5 ? 1.0 : 0.0;\n";
+        break;
+    default:
+        LOG_CRITICAL(HW_GPU, "Unknown clamp mode %u", static_cast<u32>(mode));
+        out += var + " = " + "min(" + var + ", 1.0);\n";
+        break;
+    }
+}
+
+void AppendProcTexCombineAndMap(std::string& out, ProcTexCombiner combiner,
+                                const std::string& map_lut) {
+    std::string combined;
+    switch (combiner) {
+    case ProcTexCombiner::U:
+        combined = "u";
+        break;
+    case ProcTexCombiner::U2:
+        combined = "(u * u)";
+        break;
+    case TexturingRegs::ProcTexCombiner::V:
+        combined = "v";
+        break;
+    case TexturingRegs::ProcTexCombiner::V2:
+        combined = "(v * v)";
+        break;
+    case TexturingRegs::ProcTexCombiner::Add:
+        combined = "((u + v) * 0.5)";
+        break;
+    case TexturingRegs::ProcTexCombiner::Add2:
+        combined = "((u * u + v * v) * 0.5)";
+        break;
+    case TexturingRegs::ProcTexCombiner::SqrtAdd2:
+        combined = "min(sqrt(u * u + v * v), 1.0)";
+        break;
+    case TexturingRegs::ProcTexCombiner::Min:
+        combined = "min(u, v)";
+        break;
+    case TexturingRegs::ProcTexCombiner::Max:
+        combined = "max(u, v)";
+        break;
+    case TexturingRegs::ProcTexCombiner::RMax:
+        combined = "min(((u + v) * 0.5 + sqrt(u * u + v * v)) * 0.5, 1.0)";
+        break;
+    default:
+        LOG_CRITICAL(HW_GPU, "Unknown combiner %u", static_cast<u32>(combiner));
+        combined = "0.0";
+        break;
+    }
+    out += "ProcTexLookupLUT(" + map_lut + ", " + combined + ")";
+}
+
+void AppendProcTexSampler(std::string& out, const PicaShaderConfig& config) {
+    // LUT sampling uitlity
+    // For NoiseLUT/ColorMap/AlphaMap, coord=0.0 is lut[0], coord=127.0/128.0 is lut[127] and
+    // coord=1.0 is lut[127]+lut_diff[127]. For other indices, the result is interpolated using
+    // value entries and difference entries.
+    out += R"(
+float ProcTexLookupLUT(sampler1D lut, float coord) {
+    coord *= 128;
+    float index_i = clamp(floor(coord), 0.0, 127.0);
+    float index_f = coord - index_i; // fract() cannot be used here because 128.0 needs to be
+                                     // extracted as index_i = 127.0 and index_f = 1.0
+    vec2 entry = texelFetch(lut, int(index_i), 0).rg;
+    return clamp(entry.r + entry.g * index_f, 0.0, 1.0);
+}
+    )";
+
+    // Noise utility
+    if (config.state.proctex.noise_enable) {
+        // See swrasterizer/proctex.cpp for more information about these functions
+        out += R"(
+int ProcTexNoiseRand1D(int v) {
+    const int table[] = int[](0,4,10,8,4,9,7,12,5,15,13,14,11,15,2,11);
+    return ((v % 9 + 2) * 3 & 0xF) ^ table[(v / 9) & 0xF];
+}
+
+float ProcTexNoiseRand2D(vec2 point) {
+    const int table[] = int[](10,2,15,8,0,7,4,5,5,13,2,6,13,9,3,14);
+    int u2 = ProcTexNoiseRand1D(int(point.x));
+    int v2 = ProcTexNoiseRand1D(int(point.y));
+    v2 += ((u2 & 3) == 1) ? 4 : 0;
+    v2 ^= (u2 & 1) * 6;
+    v2 += 10 + u2;
+    v2 &= 0xF;
+    v2 ^= table[u2];
+    return -1.0 + float(v2) * 2.0/ 15.0;
+}
+
+float ProcTexNoiseCoef(vec2 x) {
+    vec2 grid  = 9.0 * proctex_noise_f * abs(x + proctex_noise_p);
+    vec2 point = floor(grid);
+    vec2 frac  = grid - point;
+
+    float g0 = ProcTexNoiseRand2D(point) * (frac.x + frac.y);
+    float g1 = ProcTexNoiseRand2D(point + vec2(1.0, 0.0)) * (frac.x + frac.y - 1.0);
+    float g2 = ProcTexNoiseRand2D(point + vec2(0.0, 1.0)) * (frac.x + frac.y - 1.0);
+    float g3 = ProcTexNoiseRand2D(point + vec2(1.0, 1.0)) * (frac.x + frac.y - 2.0);
+
+    float x_noise = ProcTexLookupLUT(proctex_noise_lut, frac.x);
+    float y_noise = ProcTexLookupLUT(proctex_noise_lut, frac.y);
+    float x0 = mix(g0, g1, x_noise);
+    float x1 = mix(g2, g3, x_noise);
+    return mix(x0, x1, y_noise);
+}
+        )";
+    }
+
+    out += "vec4 ProcTex() {\n";
+    out += "vec2 uv = abs(texcoord[" + std::to_string(config.state.proctex.coord) + "]);\n";
+
+    // Get shift offset before noise generation
+    out += "float u_shift = ";
+    AppendProcTexShiftOffset(out, "uv.y", config.state.proctex.u_shift,
+                             config.state.proctex.u_clamp);
+    out += ";\n";
+    out += "float v_shift = ";
+    AppendProcTexShiftOffset(out, "uv.x", config.state.proctex.v_shift,
+                             config.state.proctex.v_clamp);
+    out += ";\n";
+
+    // Generate noise
+    if (config.state.proctex.noise_enable) {
+        out += "uv += proctex_noise_a * ProcTexNoiseCoef(uv);\n";
+        out += "uv = abs(uv);\n";
+    }
+
+    // Shift
+    out += "float u = uv.x + u_shift;\n";
+    out += "float v = uv.y + v_shift;\n";
+
+    // Clamp
+    AppendProcTexClamp(out, "u", config.state.proctex.u_clamp);
+    AppendProcTexClamp(out, "v", config.state.proctex.v_clamp);
+
+    // Combine and map
+    out += "float lut_coord = ";
+    AppendProcTexCombineAndMap(out, config.state.proctex.color_combiner, "proctex_color_map");
+    out += ";\n";
+
+    // Look up color
+    // For the color lut, coord=0.0 is lut[offset] and coord=1.0 is lut[offset+width-1]
+    out += "lut_coord *= " + std::to_string(config.state.proctex.lut_width - 1) + ";\n";
+    // TODO(wwylele): implement mipmap
+    switch (config.state.proctex.lut_filter) {
+    case ProcTexFilter::Linear:
+    case ProcTexFilter::LinearMipmapLinear:
+    case ProcTexFilter::LinearMipmapNearest:
+        out += "int lut_index_i = int(lut_coord) + " +
+               std::to_string(config.state.proctex.lut_offset) + ";\n";
+        out += "float lut_index_f = fract(lut_coord);\n";
+        out += "vec4 final_color = texelFetch(proctex_lut, lut_index_i, 0) + lut_index_f * "
+               "texelFetch(proctex_diff_lut, lut_index_i, 0);\n";
+        break;
+    case ProcTexFilter::Nearest:
+    case ProcTexFilter::NearestMipmapLinear:
+    case ProcTexFilter::NearestMipmapNearest:
+        out += "lut_coord += " + std::to_string(config.state.proctex.lut_offset) + ";\n";
+        out += "vec4 final_color = texelFetch(proctex_lut, int(round(lut_coord)), 0);\n";
+        break;
+    }
+
+    if (config.state.proctex.separate_alpha) {
+        // Note: in separate alpha mode, the alpha channel skips the color LUT look up stage. It
+        // uses the output of CombineAndMap directly instead.
+        out += "float final_alpha = ";
+        AppendProcTexCombineAndMap(out, config.state.proctex.alpha_combiner, "proctex_alpha_map");
+        out += ";\n";
+        out += "return vec4(final_color.xyz, final_alpha);\n}\n";
+    } else {
+        out += "return final_color;\n}\n";
+    }
+}
+
 std::string GenerateFragmentShader(const PicaShaderConfig& config) {
     const auto& state = config.state;
 
@@ -735,6 +982,9 @@ layout (std140) uniform shader_data {
     int scissor_x2;
     int scissor_y2;
     vec3 fog_color;
+    vec2 proctex_noise_f;
+    vec2 proctex_noise_a;
+    vec2 proctex_noise_p;
     vec3 lighting_global_ambient;
     LightSrc light_src[NUM_LIGHTS];
     vec4 const_color[NUM_TEV_STAGES];
@@ -744,12 +994,21 @@ layout (std140) uniform shader_data {
 uniform sampler2D tex[3];
 uniform sampler1D lut[6];
 uniform usampler1D fog_lut;
+uniform sampler1D proctex_noise_lut;
+uniform sampler1D proctex_color_map;
+uniform sampler1D proctex_alpha_map;
+uniform sampler1D proctex_lut;
+uniform sampler1D proctex_diff_lut;
 
 // Rotate the vector v by the quaternion q
 vec3 quaternion_rotate(vec4 q, vec3 v) {
     return v + 2.0 * cross(q.xyz, cross(q.xyz, v) + q.w * v);
-}
+})";
 
+    if (config.state.proctex.enable)
+        AppendProcTexSampler(out, config);
+
+    out += R"(
 void main() {
 vec4 primary_fragment_color = vec4(0.0);
 vec4 secondary_fragment_color = vec4(0.0);

src/video_core/renderer_opengl/gl_shader_gen.h

@@ -113,6 +113,19 @@ union PicaShaderConfig {
             } lut_d0, lut_d1, lut_fr, lut_rr, lut_rg, lut_rb;
         } lighting;
 
+        struct {
+            bool enable;
+            u32 coord;
+            Pica::TexturingRegs::ProcTexClamp u_clamp, v_clamp;
+            Pica::TexturingRegs::ProcTexCombiner color_combiner, alpha_combiner;
+            bool separate_alpha;
+            bool noise_enable;
+            Pica::TexturingRegs::ProcTexShift u_shift, v_shift;
+            u32 lut_width;
+            u32 lut_offset;
+            Pica::TexturingRegs::ProcTexFilter lut_filter;
+        } proctex;
+
     } state;
 };
 #if (__GNUC__ >= 5) || defined(__clang__) || defined(_MSC_VER)

src/video_core/renderer_opengl/gl_state.cpp

@@ -58,6 +58,12 @@ OpenGLState::OpenGLState() {
 
     fog_lut.texture_1d = 0;
 
+    proctex_lut.texture_1d = 0;
+    proctex_diff_lut.texture_1d = 0;
+    proctex_color_map.texture_1d = 0;
+    proctex_alpha_map.texture_1d = 0;
+    proctex_noise_lut.texture_1d = 0;
+
     draw.read_framebuffer = 0;
     draw.draw_framebuffer = 0;
     draw.vertex_array = 0;
@@ -201,6 +207,36 @@ void OpenGLState::Apply() const {
         glBindTexture(GL_TEXTURE_1D, fog_lut.texture_1d);
     }
 
+    // ProcTex Noise LUT
+    if (proctex_noise_lut.texture_1d != cur_state.proctex_noise_lut.texture_1d) {
+        glActiveTexture(GL_TEXTURE10);
+        glBindTexture(GL_TEXTURE_1D, proctex_noise_lut.texture_1d);
+    }
+
+    // ProcTex Color Map
+    if (proctex_color_map.texture_1d != cur_state.proctex_color_map.texture_1d) {
+        glActiveTexture(GL_TEXTURE11);
+        glBindTexture(GL_TEXTURE_1D, proctex_color_map.texture_1d);
+    }
+
+    // ProcTex Alpha Map
+    if (proctex_alpha_map.texture_1d != cur_state.proctex_alpha_map.texture_1d) {
+        glActiveTexture(GL_TEXTURE12);
+        glBindTexture(GL_TEXTURE_1D, proctex_alpha_map.texture_1d);
+    }
+
+    // ProcTex LUT
+    if (proctex_lut.texture_1d != cur_state.proctex_lut.texture_1d) {
+        glActiveTexture(GL_TEXTURE13);
+        glBindTexture(GL_TEXTURE_1D, proctex_lut.texture_1d);
+    }
+
+    // ProcTex Diff LUT
+    if (proctex_diff_lut.texture_1d != cur_state.proctex_diff_lut.texture_1d) {
+        glActiveTexture(GL_TEXTURE14);
+        glBindTexture(GL_TEXTURE_1D, proctex_diff_lut.texture_1d);
+    }
+
     // Framebuffer
     if (draw.read_framebuffer != cur_state.draw.read_framebuffer) {
         glBindFramebuffer(GL_READ_FRAMEBUFFER, draw.read_framebuffer);

src/video_core/renderer_opengl/gl_state.h

@@ -71,6 +71,26 @@ public:
         GLuint texture_1d; // GL_TEXTURE_BINDING_1D
     } fog_lut;
 
+    struct {
+        GLuint texture_1d; // GL_TEXTURE_BINDING_1D
+    } proctex_noise_lut;
+
+    struct {
+        GLuint texture_1d; // GL_TEXTURE_BINDING_1D
+    } proctex_color_map;
+
+    struct {
+        GLuint texture_1d; // GL_TEXTURE_BINDING_1D
+    } proctex_alpha_map;
+
+    struct {
+        GLuint texture_1d; // GL_TEXTURE_BINDING_1D
+    } proctex_lut;
+
+    struct {
+        GLuint texture_1d; // GL_TEXTURE_BINDING_1D
+    } proctex_diff_lut;
+
     struct {
         GLuint read_framebuffer; // GL_READ_FRAMEBUFFER_BINDING
         GLuint draw_framebuffer; // GL_DRAW_FRAMEBUFFER_BINDING

src/video_core/swrasterizer/proctex.cpp

@@ -0,0 +1,223 @@
+// Copyright 2017 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <array>
+#include <cmath>
+#include "common/math_util.h"
+#include "video_core/swrasterizer/proctex.h"
+
+namespace Pica {
+namespace Rasterizer {
+
+using ProcTexClamp = TexturingRegs::ProcTexClamp;
+using ProcTexShift = TexturingRegs::ProcTexShift;
+using ProcTexCombiner = TexturingRegs::ProcTexCombiner;
+using ProcTexFilter = TexturingRegs::ProcTexFilter;
+
+static float LookupLUT(const std::array<State::ProcTex::ValueEntry, 128>& lut, float coord) {
+    // For NoiseLUT/ColorMap/AlphaMap, coord=0.0 is lut[0], coord=127.0/128.0 is lut[127] and
+    // coord=1.0 is lut[127]+lut_diff[127]. For other indices, the result is interpolated using
+    // value entries and difference entries.
+    coord *= 128;
+    const int index_int = std::min(static_cast<int>(coord), 127);
+    const float frac = coord - index_int;
+    return lut[index_int].ToFloat() + frac * lut[index_int].DiffToFloat();
+}
+
+// These function are used to generate random noise for procedural texture. Their results are
+// verified against real hardware, but it's not known if the algorithm is the same as hardware.
+static unsigned int NoiseRand1D(unsigned int v) {
+    static constexpr std::array<unsigned int, 16> table{
+        {0, 4, 10, 8, 4, 9, 7, 12, 5, 15, 13, 14, 11, 15, 2, 11}};
+    return ((v % 9 + 2) * 3 & 0xF) ^ table[(v / 9) & 0xF];
+}
+
+static float NoiseRand2D(unsigned int x, unsigned int y) {
+    static constexpr std::array<unsigned int, 16> table{
+        {10, 2, 15, 8, 0, 7, 4, 5, 5, 13, 2, 6, 13, 9, 3, 14}};
+    unsigned int u2 = NoiseRand1D(x);
+    unsigned int v2 = NoiseRand1D(y);
+    v2 += ((u2 & 3) == 1) ? 4 : 0;
+    v2 ^= (u2 & 1) * 6;
+    v2 += 10 + u2;
+    v2 &= 0xF;
+    v2 ^= table[u2];
+    return -1.0f + v2 * 2.0f / 15.0f;
+}
+
+static float NoiseCoef(float u, float v, TexturingRegs regs, State::ProcTex state) {
+    const float freq_u = float16::FromRaw(regs.proctex_noise_frequency.u).ToFloat32();
+    const float freq_v = float16::FromRaw(regs.proctex_noise_frequency.v).ToFloat32();
+    const float phase_u = float16::FromRaw(regs.proctex_noise_u.phase).ToFloat32();
+    const float phase_v = float16::FromRaw(regs.proctex_noise_v.phase).ToFloat32();
+    const float x = 9 * freq_u * std::abs(u + phase_u);
+    const float y = 9 * freq_v * std::abs(v + phase_v);
+    const int x_int = static_cast<int>(x);
+    const int y_int = static_cast<int>(y);
+    const float x_frac = x - x_int;
+    const float y_frac = y - y_int;
+
+    const float g0 = NoiseRand2D(x_int, y_int) * (x_frac + y_frac);
+    const float g1 = NoiseRand2D(x_int + 1, y_int) * (x_frac + y_frac - 1);
+    const float g2 = NoiseRand2D(x_int, y_int + 1) * (x_frac + y_frac - 1);
+    const float g3 = NoiseRand2D(x_int + 1, y_int + 1) * (x_frac + y_frac - 2);
+    const float x_noise = LookupLUT(state.noise_table, x_frac);
+    const float y_noise = LookupLUT(state.noise_table, y_frac);
+    return Math::BilinearInterp(g0, g1, g2, g3, x_noise, y_noise);
+}
+
+static float GetShiftOffset(float v, ProcTexShift mode, ProcTexClamp clamp_mode) {
+    const float offset = (clamp_mode == ProcTexClamp::MirroredRepeat) ? 1 : 0.5f;
+    switch (mode) {
+    case ProcTexShift::None:
+        return 0;
+    case ProcTexShift::Odd:
+        return offset * (((int)v / 2) % 2);
+    case ProcTexShift::Even:
+        return offset * ((((int)v + 1) / 2) % 2);
+    default:
+        LOG_CRITICAL(HW_GPU, "Unknown shift mode %u", static_cast<u32>(mode));
+        return 0;
+    }
+};
+
+static void ClampCoord(float& coord, ProcTexClamp mode) {
+    switch (mode) {
+    case ProcTexClamp::ToZero:
+        if (coord > 1.0f)
+            coord = 0.0f;
+        break;
+    case ProcTexClamp::ToEdge:
+        coord = std::min(coord, 1.0f);
+        break;
+    case ProcTexClamp::SymmetricalRepeat:
+        coord = coord - std::floor(coord);
+        break;
+    case ProcTexClamp::MirroredRepeat: {
+        int integer = static_cast<int>(coord);
+        float frac = coord - integer;
+        coord = (integer % 2) == 0 ? frac : (1.0f - frac);
+        break;
+    }
+    case ProcTexClamp::Pulse:
+        if (coord <= 0.5f)
+            coord = 0.0f;
+        else
+            coord = 1.0f;
+        break;
+    default:
+        LOG_CRITICAL(HW_GPU, "Unknown clamp mode %u", static_cast<u32>(mode));
+        coord = std::min(coord, 1.0f);
+        break;
+    }
+}
+
+float CombineAndMap(float u, float v, ProcTexCombiner combiner,
+                    const std::array<State::ProcTex::ValueEntry, 128>& map_table) {
+    float f;
+    switch (combiner) {
+    case ProcTexCombiner::U:
+        f = u;
+        break;
+    case ProcTexCombiner::U2:
+        f = u * u;
+        break;
+    case TexturingRegs::ProcTexCombiner::V:
+        f = v;
+        break;
+    case TexturingRegs::ProcTexCombiner::V2:
+        f = v * v;
+        break;
+    case TexturingRegs::ProcTexCombiner::Add:
+        f = (u + v) * 0.5f;
+        break;
+    case TexturingRegs::ProcTexCombiner::Add2:
+        f = (u * u + v * v) * 0.5f;
+        break;
+    case TexturingRegs::ProcTexCombiner::SqrtAdd2:
+        f = std::min(std::sqrt(u * u + v * v), 1.0f);
+        break;
+    case TexturingRegs::ProcTexCombiner::Min:
+        f = std::min(u, v);
+        break;
+    case TexturingRegs::ProcTexCombiner::Max:
+        f = std::max(u, v);
+        break;
+    case TexturingRegs::ProcTexCombiner::RMax:
+        f = std::min(((u + v) * 0.5f + std::sqrt(u * u + v * v)) * 0.5f, 1.0f);
+        break;
+    default:
+        LOG_CRITICAL(HW_GPU, "Unknown combiner %u", static_cast<u32>(combiner));
+        f = 0.0f;
+        break;
+    }
+    return LookupLUT(map_table, f);
+}
+
+Math::Vec4<u8> ProcTex(float u, float v, TexturingRegs regs, State::ProcTex state) {
+    u = std::abs(u);
+    v = std::abs(v);
+
+    // Get shift offset before noise generation
+    const float u_shift = GetShiftOffset(v, regs.proctex.u_shift, regs.proctex.u_clamp);
+    const float v_shift = GetShiftOffset(u, regs.proctex.v_shift, regs.proctex.v_clamp);
+
+    // Generate noise
+    if (regs.proctex.noise_enable) {
+        float noise = NoiseCoef(u, v, regs, state);
+        u += noise * regs.proctex_noise_u.amplitude / 4095.0f;
+        v += noise * regs.proctex_noise_v.amplitude / 4095.0f;
+        u = std::abs(u);
+        v = std::abs(v);
+    }
+
+    // Shift
+    u += u_shift;
+    v += v_shift;
+
+    // Clamp
+    ClampCoord(u, regs.proctex.u_clamp);
+    ClampCoord(v, regs.proctex.v_clamp);
+
+    // Combine and map
+    const float lut_coord = CombineAndMap(u, v, regs.proctex.color_combiner, state.color_map_table);
+
+    // Look up the color
+    // For the color lut, coord=0.0 is lut[offset] and coord=1.0 is lut[offset+width-1]
+    const u32 offset = regs.proctex_lut_offset;
+    const u32 width = regs.proctex_lut.width;
+    const float index = offset + (lut_coord * (width - 1));
+    Math::Vec4<u8> final_color;
+    // TODO(wwylele): implement mipmap
+    switch (regs.proctex_lut.filter) {
+    case ProcTexFilter::Linear:
+    case ProcTexFilter::LinearMipmapLinear:
+    case ProcTexFilter::LinearMipmapNearest: {
+        const int index_int = static_cast<int>(index);
+        const float frac = index - index_int;
+        const auto color_value = state.color_table[index_int].ToVector().Cast<float>();
+        const auto color_diff = state.color_diff_table[index_int].ToVector().Cast<float>();
+        final_color = (color_value + frac * color_diff).Cast<u8>();
+        break;
+    }
+    case ProcTexFilter::Nearest:
+    case ProcTexFilter::NearestMipmapLinear:
+    case ProcTexFilter::NearestMipmapNearest:
+        final_color = state.color_table[static_cast<int>(std::round(index))].ToVector();
+        break;
+    }
+
+    if (regs.proctex.separate_alpha) {
+        // Note: in separate alpha mode, the alpha channel skips the color LUT look up stage. It
+        // uses the output of CombineAndMap directly instead.
+        const float final_alpha =
+            CombineAndMap(u, v, regs.proctex.alpha_combiner, state.alpha_map_table);
+        return Math::MakeVec<u8>(final_color.rgb(), static_cast<u8>(final_alpha * 255));
+    } else {
+        return final_color;
+    }
+}
+
+} // namespace Rasterizer
+} // namespace Pica

src/video_core/swrasterizer/proctex.h

@@ -0,0 +1,16 @@
+// Copyright 2017 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/common_types.h"
+#include "common/vector_math.h"
+#include "video_core/pica_state.h"
+
+namespace Pica {
+namespace Rasterizer {
+
+/// Generates procedural texture color for the given coordinates
+Math::Vec4<u8> ProcTex(float u, float v, TexturingRegs regs, State::ProcTex state);
+
+} // namespace Rasterizer
+} // namespace Pica

src/video_core/swrasterizer/rasterizer.cpp

@@ -23,6 +23,7 @@
 #include "video_core/regs_texturing.h"
 #include "video_core/shader/shader.h"
 #include "video_core/swrasterizer/framebuffer.h"
+#include "video_core/swrasterizer/proctex.h"
 #include "video_core/swrasterizer/rasterizer.h"
 #include "video_core/swrasterizer/texturing.h"
 #include "video_core/texture/texture_decode.h"
@@ -268,7 +269,7 @@ static void ProcessTriangleInternal(const Vertex& v0, const Vertex& v1, const Ve
             uv[2].u() = GetInterpolatedAttribute(v0.tc2.u(), v1.tc2.u(), v2.tc2.u());
             uv[2].v() = GetInterpolatedAttribute(v0.tc2.v(), v1.tc2.v(), v2.tc2.v());
 
-            Math::Vec4<u8> texture_color[3]{};
+            Math::Vec4<u8> texture_color[4]{};
             for (int i = 0; i < 3; ++i) {
                 const auto& texture = textures[i];
                 if (!texture.enabled)
@@ -334,6 +335,13 @@ static void ProcessTriangleInternal(const Vertex& v0, const Vertex& v1, const Ve
                 }
             }
 
+            // sample procedural texture
+            if (regs.texturing.main_config.texture3_enable) {
+                const auto& proctex_uv = uv[regs.texturing.main_config.texture3_coordinates];
+                texture_color[3] = ProcTex(proctex_uv.u().ToFloat32(), proctex_uv.v().ToFloat32(),
+                                           g_state.regs.texturing, g_state.proctex);
+            }
+
             // Texture environment - consists of 6 stages of color and alpha combining.
             //
             // Color combiners take three input color values from some source (e.g. interpolated
@@ -376,6 +384,9 @@ static void ProcessTriangleInternal(const Vertex& v0, const Vertex& v1, const Ve
                     case Source::Texture2:
                         return texture_color[2];
 
+                    case Source::Texture3:
+                        return texture_color[3];
+
                     case Source::PreviousBuffer:
                         return combiner_buffer;