pica/swrasterizer: implement procedural texture

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

@@ -458,6 +458,37 @@ 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

@@ -122,8 +122,8 @@ 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<10, 1, u32> texture3_enable;     // TODO: unimplemented
+        BitField<8, 2, u32> texture3_coordinates;
+        BitField<10, 1, u32> texture3_enable;
         BitField<13, 1, u32> texture2_use_coord1;
         BitField<16, 1, u32> clear_texture_cache; // TODO: unimplemented
     } main_config;
@@ -137,7 +137,7 @@ 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;
@@ -152,6 +152,96 @@ 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/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;