Merge pull request #3516 from wwylele/shadow-sw

SwRasterizer: Implement shadow mapping

src/video_core/regs_framebuffer.h

@@ -15,6 +15,12 @@
 namespace Pica {
 
 struct FramebufferRegs {
+    enum class FragmentOperationMode : u32 {
+        Default = 0,
+        Gas = 1,
+        Shadow = 3,
+    };
+
     enum class LogicOp : u32 {
         Clear = 0,
         And = 1,
@@ -84,6 +90,7 @@ struct FramebufferRegs {
 
     struct {
         union {
+            BitField<0, 2, FragmentOperationMode> fragment_operation_mode;
             // If false, logic blending is used
             BitField<8, 1, u32> alphablend_enable;
         };
@@ -274,7 +281,14 @@ struct FramebufferRegs {
         ASSERT_MSG(false, "Unknown depth format %u", static_cast<u32>(format));
     }
 
-    INSERT_PADDING_WORDS(0x20);
+    INSERT_PADDING_WORDS(0x10); // Gas related registers
+
+    union {
+        BitField<0, 16, u32> constant; // float1.5.10
+        BitField<16, 16, u32> linear;  // float1.5.10
+    } shadow;
+
+    INSERT_PADDING_WORDS(0xF);
 };
 
 static_assert(sizeof(FramebufferRegs) == 0x40 * sizeof(u32),

src/video_core/regs_lighting.h

@@ -187,9 +187,15 @@ struct LightingRegs {
     BitField<0, 3, u32> max_light_index; // Number of enabled lights - 1
 
     union {
+        BitField<0, 1, u32> enable_shadow;
         BitField<2, 2, LightingFresnelSelector> fresnel_selector;
         BitField<4, 4, LightingConfig> config;
+        BitField<16, 1, u32> shadow_primary;
+        BitField<17, 1, u32> shadow_secondary;
+        BitField<18, 1, u32> shadow_invert;
+        BitField<19, 1, u32> shadow_alpha;
         BitField<22, 2, u32> bump_selector; // 0: Texture 0, 1: Texture 1, 2: Texture 2
+        BitField<24, 2, u32> shadow_selector;
         BitField<27, 1, u32> clamp_highlights;
         BitField<28, 2, LightingBumpMode> bump_mode;
         BitField<30, 1, u32> disable_bump_renorm;
@@ -198,6 +204,9 @@ struct LightingRegs {
     union {
         u32 raw;
 
+        // Each bit specifies whether shadow should be applied for the corresponding light.
+        BitField<0, 8, u32> disable_shadow;
+
         // Each bit specifies whether spot light attenuation should be applied for the corresponding
         // light.
         BitField<8, 8, u32> disable_spot_atten;
@@ -224,6 +233,10 @@ struct LightingRegs {
         return (config1.disable_spot_atten & (1 << index)) != 0;
     }
 
+    bool IsShadowDisabled(unsigned index) const {
+        return (config1.disable_shadow & (1 << index)) != 0;
+    }
+
     union {
         BitField<0, 8, u32> index; ///< Index at which to set data in the LUT
         BitField<8, 5, u32> type;  ///< Type of LUT for which to set data

src/video_core/regs_texturing.h

@@ -158,7 +158,12 @@ struct TexturingRegs {
         return address * 8;
     }
 
-    INSERT_PADDING_WORDS(0x3);
+    union {
+        BitField<0, 1, u32> orthographic; // 0: enable perspective divide
+        BitField<1, 23, u32> bias;        // 23-bit fraction
+    } shadow;
+
+    INSERT_PADDING_WORDS(0x2);
     BitField<0, 4, TextureFormat> texture0_format;
     BitField<0, 1, u32> fragment_lighting_enable;
     INSERT_PADDING_WORDS(0x1);

src/video_core/swrasterizer/framebuffer.cpp

@@ -359,5 +359,54 @@ u8 LogicOp(u8 src, u8 dest, FramebufferRegs::LogicOp op) {
     UNREACHABLE();
 };
 
+// Decode/Encode for shadow map format. It is similar to D24S8 format, but the depth field is in
+// big-endian
+static const Math::Vec2<u32> DecodeD24S8Shadow(const u8* bytes) {
+    return {static_cast<u32>((bytes[0] << 16) | (bytes[1] << 8) | bytes[2]), bytes[3]};
+}
+
+static void EncodeD24X8Shadow(u32 depth, u8* bytes) {
+    bytes[2] = depth & 0xFF;
+    bytes[1] = (depth >> 8) & 0xFF;
+    bytes[0] = (depth >> 16) & 0xFF;
+}
+
+static void EncodeX24S8Shadow(u8 stencil, u8* bytes) {
+    bytes[3] = stencil;
+}
+
+void DrawShadowMapPixel(int x, int y, u32 depth, u8 stencil) {
+    const auto& framebuffer = g_state.regs.framebuffer.framebuffer;
+    const auto& shadow = g_state.regs.framebuffer.shadow;
+    const PAddr addr = framebuffer.GetColorBufferPhysicalAddress();
+
+    y = framebuffer.height - y;
+
+    const u32 coarse_y = y & ~7;
+    u32 bytes_per_pixel = 4;
+    u32 dst_offset = VideoCore::GetMortonOffset(x, y, bytes_per_pixel) +
+                     coarse_y * framebuffer.width * bytes_per_pixel;
+    u8* dst_pixel = Memory::GetPhysicalPointer(addr) + dst_offset;
+
+    auto ref = DecodeD24S8Shadow(dst_pixel);
+    u32 ref_z = ref.x;
+    u32 ref_s = ref.y;
+
+    if (depth < ref_z) {
+        if (stencil == 0) {
+            EncodeD24X8Shadow(depth, dst_pixel);
+        } else {
+            float16 constant = float16::FromRaw(shadow.constant);
+            float16 linear = float16::FromRaw(shadow.linear);
+            float16 x = float16::FromFloat32(static_cast<float>(depth) / ref_z);
+            float16 stencil_new = float16::FromFloat32(stencil) / (constant + linear * x);
+            stencil = static_cast<u8>(MathUtil::Clamp(stencil_new.ToFloat32(), 0.0f, 255.0f));
+
+            if (stencil < ref_s)
+                EncodeX24S8Shadow(stencil, dst_pixel);
+        }
+    }
+}
+
 } // namespace Rasterizer
 } // namespace Pica

src/video_core/swrasterizer/framebuffer.h

@@ -25,5 +25,7 @@ Math::Vec4<u8> EvaluateBlendEquation(const Math::Vec4<u8>& src, const Math::Vec4
 
 u8 LogicOp(u8 src, u8 dest, FramebufferRegs::LogicOp op);
 
+void DrawShadowMapPixel(int x, int y, u32 depth, u8 stencil);
+
 } // namespace Rasterizer
 } // namespace Pica

src/video_core/swrasterizer/lighting.cpp

@@ -25,6 +25,16 @@ std::tuple<Math::Vec4<u8>, Math::Vec4<u8>> ComputeFragmentsColors(
     const Math::Quaternion<float>& normquat, const Math::Vec3<float>& view,
     const Math::Vec4<u8> (&texture_color)[4]) {
 
+    Math::Vec4<float> shadow;
+    if (lighting.config0.enable_shadow) {
+        shadow = texture_color[lighting.config0.shadow_selector].Cast<float>() / 255.0f;
+        if (lighting.config0.shadow_invert) {
+            shadow = Math::MakeVec(1.0f, 1.0f, 1.0f, 1.0f) - shadow;
+        }
+    } else {
+        shadow = Math::MakeVec(1.0f, 1.0f, 1.0f, 1.0f);
+    }
+
     Math::Vec3<float> surface_normal;
     Math::Vec3<float> surface_tangent;
 
@@ -278,11 +288,38 @@ std::tuple<Math::Vec4<u8>, Math::Vec4<u8>> ComputeFragmentsColors(
         }
 
         auto diffuse =
-            light_config.diffuse.ToVec3f() * dot_product + light_config.ambient.ToVec3f();
-        diffuse_sum += Math::MakeVec(diffuse * dist_atten * spot_atten, 0.0f);
+            (light_config.diffuse.ToVec3f() * dot_product + light_config.ambient.ToVec3f()) *
+            dist_atten * spot_atten;
+        auto specular = (specular_0 + specular_1) * clamp_highlights * dist_atten * spot_atten;
 
-        specular_sum += Math::MakeVec(
-            (specular_0 + specular_1) * clamp_highlights * dist_atten * spot_atten, 0.0f);
+        if (!lighting.IsShadowDisabled(num)) {
+            if (lighting.config0.shadow_primary) {
+                diffuse = diffuse * shadow.xyz();
+            }
+            if (lighting.config0.shadow_secondary) {
+                specular = specular * shadow.xyz();
+            }
+        }
+
+        diffuse_sum += Math::MakeVec(diffuse, 0.0f);
+        specular_sum += Math::MakeVec(specular, 0.0f);
+    }
+
+    if (lighting.config0.shadow_alpha) {
+        // Alpha shadow also uses the Fresnel selecotr to determine which alpha to apply
+        // Enabled for diffuse lighting alpha component
+        if (lighting.config0.fresnel_selector ==
+                LightingRegs::LightingFresnelSelector::PrimaryAlpha ||
+            lighting.config0.fresnel_selector == LightingRegs::LightingFresnelSelector::Both) {
+            diffuse_sum.a() *= shadow.w;
+        }
+
+        // Enabled for the specular lighting alpha component
+        if (lighting.config0.fresnel_selector ==
+                LightingRegs::LightingFresnelSelector::SecondaryAlpha ||
+            lighting.config0.fresnel_selector == LightingRegs::LightingFresnelSelector::Both) {
+            specular_sum.a() *= shadow.w;
+        }
     }
 
     diffuse_sum += Math::MakeVec(lighting.global_ambient.ToVec3f(), 0.0f);

src/video_core/swrasterizer/rasterizer.cpp

@@ -74,8 +74,9 @@ static int SignedArea(const Math::Vec2<Fix12P4>& vtx1, const Math::Vec2<Fix12P4>
 };
 
 /// Convert a 3D vector for cube map coordinates to 2D texture coordinates along with the face name
-static std::tuple<float24, float24, PAddr> ConvertCubeCoord(float24 u, float24 v, float24 w,
-                                                            const TexturingRegs& regs) {
+static std::tuple<float24, float24, float24, PAddr> ConvertCubeCoord(float24 u, float24 v,
+                                                                     float24 w,
+                                                                     const TexturingRegs& regs) {
     const float abs_u = std::abs(u.ToFloat32());
     const float abs_v = std::abs(v.ToFloat32());
     const float abs_w = std::abs(w.ToFloat32());
@@ -112,8 +113,9 @@ static std::tuple<float24, float24, PAddr> ConvertCubeCoord(float24 u, float24 v
         x = u;
         z = w;
     }
+    float24 z_abs = float24::FromFloat32(std::abs(z.ToFloat32()));
     const float24 half = float24::FromFloat32(0.5f);
-    return std::make_tuple(x / z * half + half, y / z * half + half, addr);
+    return std::make_tuple(x / z * half + half, y / z * half + half, z_abs, addr);
 }
 
 MICROPROFILE_DEFINE(GPU_Rasterization, "GPU", "Rasterization", MP_RGB(50, 50, 240));
@@ -331,13 +333,16 @@ static void ProcessTriangleInternal(const Vertex& v0, const Vertex& v1, const Ve
                 // Only unit 0 respects the texturing type (according to 3DBrew)
                 // TODO: Refactor so cubemaps and shadowmaps can be handled
                 PAddr texture_address = texture.config.GetPhysicalAddress();
+                float24 shadow_z;
                 if (i == 0) {
                     switch (texture.config.type) {
                     case TexturingRegs::TextureConfig::Texture2D:
                         break;
+                    case TexturingRegs::TextureConfig::ShadowCube:
                     case TexturingRegs::TextureConfig::TextureCube: {
                         auto w = GetInterpolatedAttribute(v0.tc0_w, v1.tc0_w, v2.tc0_w);
-                        std::tie(u, v, texture_address) = ConvertCubeCoord(u, v, w, regs.texturing);
+                        std::tie(u, v, shadow_z, texture_address) =
+                            ConvertCubeCoord(u, v, w, regs.texturing);
                         break;
                     }
                     case TexturingRegs::TextureConfig::Projection2D: {
@@ -346,6 +351,16 @@ static void ProcessTriangleInternal(const Vertex& v0, const Vertex& v1, const Ve
                         v /= tc0_w;
                         break;
                     }
+                    case TexturingRegs::TextureConfig::Shadow2D: {
+                        auto tc0_w = GetInterpolatedAttribute(v0.tc0_w, v1.tc0_w, v2.tc0_w);
+                        if (!regs.texturing.shadow.orthographic) {
+                            u /= tc0_w;
+                            v /= tc0_w;
+                        }
+
+                        shadow_z = float24::FromFloat32(std::abs(tc0_w.ToFloat32()));
+                        break;
+                    }
                     default:
                         // TODO: Change to LOG_ERROR when more types are handled.
                         LOG_DEBUG(HW_GPU, "Unhandled texture type %x", (int)texture.config.type);
@@ -394,6 +409,22 @@ static void ProcessTriangleInternal(const Vertex& v0, const Vertex& v1, const Ve
                     // TODO: Apply the min and mag filters to the texture
                     texture_color[i] = Texture::LookupTexture(texture_data, s, t, info);
                 }
+
+                if (i == 0 && (texture.config.type == TexturingRegs::TextureConfig::Shadow2D ||
+                               texture.config.type == TexturingRegs::TextureConfig::ShadowCube)) {
+
+                    s32 z_int = static_cast<s32>(std::min(shadow_z.ToFloat32(), 1.0f) * 0xFFFFFF);
+                    z_int -= regs.texturing.shadow.bias << 1;
+                    auto& color = texture_color[i];
+                    s32 z_ref = (color.w << 16) | (color.z << 8) | color.y;
+                    u8 density;
+                    if (z_ref >= z_int) {
+                        density = color.x;
+                    } else {
+                        density = 0;
+                    }
+                    texture_color[i] = {density, density, density, density};
+                }
             }
 
             // sample procedural texture
@@ -541,6 +572,17 @@ static void ProcessTriangleInternal(const Vertex& v0, const Vertex& v1, const Ve
             }
 
             const auto& output_merger = regs.framebuffer.output_merger;
+
+            if (output_merger.fragment_operation_mode ==
+                FramebufferRegs::FragmentOperationMode::Shadow) {
+                u32 depth_int = static_cast<u32>(depth * 0xFFFFFF);
+                // use green color as the shadow intensity
+                u8 stencil = combiner_output.y;
+                DrawShadowMapPixel(x >> 4, y >> 4, depth_int, stencil);
+                // skip the normal output merger pipeline if it is in shadow mode
+                continue;
+            }
+
             // TODO: Does alpha testing happen before or after stencil?
             if (output_merger.alpha_test.enable) {
                 bool pass = false;