renderer: add area sampling scaling method (#57)

Adds Area Sampling to the list of scaling options. Works well to achieve a high-quality, smooth super-sampling effect. Dolphin has had this for a while, and now Ryujinx has recently added it too, so I decided to port it.

Not sure if adding the extra uniform to the OpenGL WindowAdaptPass was a good idea or not, or if using the push constants under Vulkan was either, but I wasn't sure about the best way to get the window size for use in the shader, and other scaling methods still work fine. Implementation seems to work fine under both Vulkan and OpenGL, but might still need some minor tweaks to the shader. Should definitely do some testing before merging, I have tested on an Nvidia RTX 3080 under Windows.

Adapted from these two PRs:
https://github.com/Ryujinx/Ryujinx/pull/7304
https://github.com/dolphin-emu/dolphin/pull/11999

Reviewed-on: http://vub63vv26q6v27xzv2dtcd25xumubshogm67yrpaz2rculqxs7jlfqad.onion/torzu-emu/torzu/pulls/57
Co-authored-by: lui <lui@vub63vv26q6v27xzv2dtcd25xumubshogm67yrpaz2rculqxs7jlfqad.onion>
Co-committed-by: lui <lui@vub63vv26q6v27xzv2dtcd25xumubshogm67yrpaz2rculqxs7jlfqad.onion>

src/common/settings_enums.h

@@ -145,7 +145,7 @@ ENUM(NvdecEmulation, Off, Cpu, Gpu);
 ENUM(ResolutionSetup, Res1_2X, Res3_4X, Res1X, Res3_2X, Res2X, Res3X, Res4X, Res5X, Res6X, Res7X,
      Res8X);
 
-ENUM(ScalingFilter, NearestNeighbor, Bilinear, Bicubic, Gaussian, ScaleForce, Fsr, MaxEnum);
+ENUM(ScalingFilter, NearestNeighbor, Bilinear, Bicubic, Gaussian, ScaleForce, Fsr, Area, MaxEnum);
 
 ENUM(AntiAliasing, None, Fxaa, Smaa, MaxEnum);
 

src/video_core/host_shaders/CMakeLists.txt

@@ -41,6 +41,7 @@ set(SHADER_FILES
     opengl_present_scaleforce.frag
     opengl_smaa.glsl
     pitch_unswizzle.comp
+    present_area.frag
     present_bicubic.frag
     present_gaussian.frag
     queries_prefix_scan_sum.comp

src/video_core/host_shaders/present_area.frag

@@ -0,0 +1,107 @@
+#version 460 core
+
+layout(location = 0) in vec2 frag_tex_coord;
+layout(location = 0) out vec4 color;
+layout(binding = 0) uniform sampler2D color_texture;
+
+#ifdef VULKAN
+
+struct ScreenRectVertex {
+    vec2 position;
+    vec2 tex_coord;
+};
+layout (push_constant) uniform PushConstants {
+    mat4 modelview_matrix;
+    ScreenRectVertex vertices[4];
+};
+
+#else // OpenGL
+
+layout(location = 1) uniform uvec2 screen_size;
+
+#endif
+
+/***** Area Sampling *****/
+
+// By Sam Belliveau and Filippo Tarpini. Public Domain license.
+// Effectively a more accurate sharp bilinear filter when upscaling,
+// that also works as a mathematically perfect downscale filter.
+// https://entropymine.com/imageworsener/pixelmixing/
+// https://github.com/obsproject/obs-studio/pull/1715
+// https://legacy.imagemagick.org/Usage/filter/
+vec4 AreaSampling(sampler2D textureSampler, vec2 texCoords, vec2 source_size, vec2 target_size) {
+    // Determine the sizes of the source and target images.
+    vec2 inverted_target_size = vec2(1.0) / target_size;
+
+    // Determine the range of the source image that the target pixel will cover.
+    vec2 range = source_size * inverted_target_size;
+    vec2 beg = (texCoords.xy * source_size) - (range * 0.5);
+    vec2 end = beg + range;
+
+    // Compute the top-left and bottom-right corners of the pixel box.
+    ivec2 f_beg = ivec2(floor(beg));
+    ivec2 f_end = ivec2(floor(end));
+
+    // Compute how much of the start and end pixels are covered horizontally & vertically.
+    float area_w = 1.0 - fract(beg.x);
+    float area_n = 1.0 - fract(beg.y);
+    float area_e = fract(end.x);
+    float area_s = fract(end.y);
+
+    // Compute the areas of the corner pixels in the pixel box.
+    float area_nw = area_n * area_w;
+    float area_ne = area_n * area_e;
+    float area_sw = area_s * area_w;
+    float area_se = area_s * area_e;
+
+    // Initialize the color accumulator.
+    vec4 avg_color = vec4(0.0, 0.0, 0.0, 0.0);
+
+    // Accumulate corner pixels.
+    avg_color += area_nw * texelFetch(textureSampler, ivec2(f_beg.x, f_beg.y), 0);
+    avg_color += area_ne * texelFetch(textureSampler, ivec2(f_end.x, f_beg.y), 0);
+    avg_color += area_sw * texelFetch(textureSampler, ivec2(f_beg.x, f_end.y), 0);
+    avg_color += area_se * texelFetch(textureSampler, ivec2(f_end.x, f_end.y), 0);
+
+    // Determine the size of the pixel box.
+    int x_range = int(f_end.x - f_beg.x - 0.5);
+    int y_range = int(f_end.y - f_beg.y - 0.5);
+
+    // Accumulate top and bottom edge pixels.
+    for (int x = f_beg.x + 1; x <= f_beg.x + x_range; ++x) {
+        avg_color += area_n * texelFetch(textureSampler, ivec2(x, f_beg.y), 0);
+        avg_color += area_s * texelFetch(textureSampler, ivec2(x, f_end.y), 0);
+    }
+
+    // Accumulate left and right edge pixels and all the pixels in between.
+    for (int y = f_beg.y + 1; y <= f_beg.y + y_range; ++y) {
+        avg_color += area_w * texelFetch(textureSampler, ivec2(f_beg.x, y), 0);
+        avg_color += area_e * texelFetch(textureSampler, ivec2(f_end.x, y), 0);
+
+        for (int x = f_beg.x + 1; x <= f_beg.x + x_range; ++x) {
+            avg_color += texelFetch(textureSampler, ivec2(x, y), 0);
+        }
+    }
+
+    // Compute the area of the pixel box that was sampled.
+    float area_corners = area_nw + area_ne + area_sw + area_se;
+    float area_edges = float(x_range) * (area_n + area_s) + float(y_range) * (area_w + area_e);
+    float area_center = float(x_range) * float(y_range);
+
+    // Return the normalized average color.
+    return avg_color / (area_corners + area_edges + area_center);
+}
+
+void main() {
+    vec2 source_image_size = textureSize(color_texture, 0);
+    vec2 window_size;
+
+    #ifdef VULKAN
+    window_size.x = vertices[1].position.x - vertices[0].position.x;
+    window_size.y = vertices[2].position.y - vertices[0].position.y;
+    #else // OpenGL
+    window_size = screen_size;
+    #endif
+
+    color = AreaSampling(color_texture, frag_tex_coord, source_image_size, window_size);
+}

src/video_core/renderer_opengl/gl_blit_screen.cpp

@@ -86,6 +86,9 @@ void BlitScreen::CreateWindowAdapt() {
     case Settings::ScalingFilter::ScaleForce:
         window_adapt = MakeScaleForce(device);
         break;
+    case Settings::ScalingFilter::Area:
+        window_adapt = MakeArea(device);
+        break;
     case Settings::ScalingFilter::Fsr:
     case Settings::ScalingFilter::Bilinear:
     default:

src/video_core/renderer_opengl/present/filters.cpp

@@ -3,6 +3,7 @@
 
 #include "video_core/host_shaders/opengl_present_frag.h"
 #include "video_core/host_shaders/opengl_present_scaleforce_frag.h"
+#include "video_core/host_shaders/present_area_frag.h"
 #include "video_core/host_shaders/present_bicubic_frag.h"
 #include "video_core/host_shaders/present_gaussian_frag.h"
 #include "video_core/renderer_opengl/present/filters.h"
@@ -36,4 +37,9 @@ std::unique_ptr<WindowAdaptPass> MakeScaleForce(const Device& device) {
         fmt::format("#version 460\n{}", HostShaders::OPENGL_PRESENT_SCALEFORCE_FRAG));
 }
 
+std::unique_ptr<WindowAdaptPass> MakeArea(const Device& device) {
+    return std::make_unique<WindowAdaptPass>(device, CreateBilinearSampler(),
+                                             HostShaders::PRESENT_AREA_FRAG);
+}
+
 } // namespace OpenGL

src/video_core/renderer_opengl/present/filters.h

@@ -13,5 +13,6 @@ std::unique_ptr<WindowAdaptPass> MakeBilinear(const Device& device);
 std::unique_ptr<WindowAdaptPass> MakeBicubic(const Device& device);
 std::unique_ptr<WindowAdaptPass> MakeGaussian(const Device& device);
 std::unique_ptr<WindowAdaptPass> MakeScaleForce(const Device& device);
+std::unique_ptr<WindowAdaptPass> MakeArea(const Device& device);
 
 } // namespace OpenGL

src/video_core/renderer_opengl/present/present_uniforms.h

@@ -10,6 +10,7 @@ namespace OpenGL {
 constexpr GLint PositionLocation = 0;
 constexpr GLint TexCoordLocation = 1;
 constexpr GLint ModelViewMatrixLocation = 0;
+constexpr GLint ScreenSizeLocation = 1;
 
 struct ScreenRectVertex {
     constexpr ScreenRectVertex() = default;

src/video_core/renderer_opengl/present/window_adapt_pass.cpp

@@ -110,6 +110,9 @@ void WindowAdaptPass::DrawToFramebuffer(ProgramManager& program_manager, std::li
         glBindTextureUnit(0, textures[i]);
         glProgramUniformMatrix3x2fv(vert.handle, ModelViewMatrixLocation, 1, GL_FALSE,
                                     matrices[i].data());
+        glProgramUniform2ui(frag.handle, ScreenSizeLocation,
+                            static_cast<GLuint>(layout.screen.GetWidth()),
+                            static_cast<GLuint>(layout.screen.GetHeight()));
         glNamedBufferSubData(vertex_buffer.handle, 0, sizeof(vertices[i]), std::data(vertices[i]));
         glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
     }

src/video_core/renderer_vulkan/present/filters.cpp

@@ -3,6 +3,7 @@
 
 #include "common/common_types.h"
 
+#include "video_core/host_shaders/present_area_frag_spv.h"
 #include "video_core/host_shaders/present_bicubic_frag_spv.h"
 #include "video_core/host_shaders/present_gaussian_frag_spv.h"
 #include "video_core/host_shaders/vulkan_present_frag_spv.h"
@@ -53,4 +54,9 @@ std::unique_ptr<WindowAdaptPass> MakeScaleForce(const Device& device, VkFormat f
                                              SelectScaleForceShader(device));
 }
 
+std::unique_ptr<WindowAdaptPass> MakeArea(const Device& device, VkFormat frame_format) {
+    return std::make_unique<WindowAdaptPass>(device, frame_format, CreateBilinearSampler(device),
+                                             BuildShader(device, PRESENT_AREA_FRAG_SPV));
+}
+
 } // namespace Vulkan

src/video_core/renderer_vulkan/present/filters.h

@@ -14,5 +14,6 @@ std::unique_ptr<WindowAdaptPass> MakeBilinear(const Device& device, VkFormat fra
 std::unique_ptr<WindowAdaptPass> MakeBicubic(const Device& device, VkFormat frame_format);
 std::unique_ptr<WindowAdaptPass> MakeGaussian(const Device& device, VkFormat frame_format);
 std::unique_ptr<WindowAdaptPass> MakeScaleForce(const Device& device, VkFormat frame_format);
+std::unique_ptr<WindowAdaptPass> MakeArea(const Device& device, VkFormat frame_format);
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_blit_screen.cpp

@@ -43,6 +43,9 @@ void BlitScreen::SetWindowAdaptPass() {
     case Settings::ScalingFilter::ScaleForce:
         window_adapt = MakeScaleForce(device, swapchain_view_format);
         break;
+    case Settings::ScalingFilter::Area:
+        window_adapt = MakeArea(device, swapchain_view_format);
+        break;
     case Settings::ScalingFilter::Fsr:
     case Settings::ScalingFilter::Bilinear:
     default:

src/yuzu/configuration/shared_translation.cpp

@@ -411,6 +411,7 @@ std::unique_ptr<ComboboxTranslationMap> ComboboxEnumeration(QWidget* parent) {
                               PAIR(ScalingFilter, Gaussian, tr("Gaussian")),
                               PAIR(ScalingFilter, ScaleForce, tr("ScaleForce")),
                               PAIR(ScalingFilter, Fsr, tr("AMD FidelityFX™️ Super Resolution")),
+                              PAIR(ScalingFilter, Area, tr("Area")),
                           }});
     translations->insert({Settings::EnumMetadata<Settings::AntiAliasing>::Index(),
                           {

src/yuzu/configuration/shared_translation.h

@@ -40,6 +40,7 @@ static const std::map<Settings::ScalingFilter, QString> scaling_filter_texts_map
     {Settings::ScalingFilter::ScaleForce,
      QStringLiteral(QT_TRANSLATE_NOOP("GMainWindow", "ScaleForce"))},
     {Settings::ScalingFilter::Fsr, QStringLiteral(QT_TRANSLATE_NOOP("GMainWindow", "FSR"))},
+    {Settings::ScalingFilter::Area, QStringLiteral(QT_TRANSLATE_NOOP("GMainWindow", "Area"))},
 };
 
 static const std::map<Settings::ConsoleMode, QString> use_docked_mode_texts_map = {