GPU: Add Lanczos scaling option

src/core/gpu_presenter.cpp

@@ -132,6 +132,10 @@ bool GPUPresenter::CompileDisplayPipelines(bool display, bool deinterlace, bool
         fs = shadergen.GenerateDisplayFragmentShader(true, false);
         break;
 
+      case DisplayScalingMode::Lanczos:
+        fs = shadergen.GenerateDisplayLanczosFragmentShader();
+        break;
+
       case DisplayScalingMode::Nearest:
       case DisplayScalingMode::NearestInteger:
       default:
@@ -654,6 +658,8 @@ void GPUPresenter::DrawDisplay(const GSVector2i target_size, const GSVector2i fi
   } uniforms;
   std::memset(uniforms.params, 0, sizeof(uniforms.params));
 
+  const GSVector2 display_texture_size = GSVector2(m_display_texture->GetSizeVec());
+
   switch (g_gpu_settings.display_scaling)
   {
     case DisplayScalingMode::Nearest:
@@ -665,6 +671,14 @@ void GPUPresenter::DrawDisplay(const GSVector2i target_size, const GSVector2i fi
       texture_filter_linear = true;
       break;
 
+    case DisplayScalingMode::Lanczos:
+    {
+      const GSVector2 fdisplay_rect_size = GSVector2(display_rect.rsize());
+      GSVector2::store<true>(&uniforms.params[0], fdisplay_rect_size);
+      GSVector2::store<true>(&uniforms.params[2], GSVector2::cxpr(1.0f) / (fdisplay_rect_size / display_texture_size));
+    }
+    break;
+
     case DisplayScalingMode::BilinearSharp:
     {
       texture_filter_linear = true;
@@ -689,7 +703,6 @@ void GPUPresenter::DrawDisplay(const GSVector2i target_size, const GSVector2i fi
 
   // For bilinear, clamp to 0.5/SIZE-0.5 to avoid bleeding from the adjacent texels in VRAM. This is because
   // 1.0 in UV space is not the bottom-right texel, but a mix of the bottom-right and wrapped/next texel.
-  const GSVector2 display_texture_size = GSVector2(m_display_texture->GetSizeVec());
   const GSVector4 display_texture_size4 = GSVector4::xyxy(display_texture_size);
   const GSVector4 uv_rect = GSVector4(GSVector4i(m_display_texture_view_x, m_display_texture_view_y,
                                                  m_display_texture_view_x + m_display_texture_view_width,

src/core/gpu_shadergen.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: 2019-2024 Connor McLaughlin <stenzek@gmail.com>
+// SPDX-FileCopyrightText: 2019-2025 Connor McLaughlin <stenzek@gmail.com>
 // SPDX-License-Identifier: CC-BY-NC-ND-4.0
 
 #include "gpu_shadergen.h"
@@ -95,6 +95,62 @@ std::string GPUShaderGen::GenerateDisplaySharpBilinearFragmentShader() const
   return ss.str();
 }
 
+std::string GPUShaderGen::GenerateDisplayLanczosFragmentShader() const
+{
+  std::stringstream ss;
+  WriteHeader(ss);
+  WriteDisplayUniformBuffer(ss);
+  DeclareTexture(ss, "samp0", 0, false);
+
+  ss << R"(
+CONSTANT int KERNEL_SIZE = 3;
+CONSTANT float PI = 3.14159265359;
+
+float lanczos(float x)
+ {
+    x = abs(x);
+    if (x < 0.0001)
+      return 1.0;
+
+    if (x > float(KERNEL_SIZE))
+      return 0.0;
+
+    float px = PI * x;
+    return (float(KERNEL_SIZE) * sin(px) * sin(px / float(KERNEL_SIZE))) / (px * px);
+}
+
+)";
+
+  DeclareFragmentEntryPoint(ss, 0, 1);
+  ss << R"(
+{
+  float2 pixel = v_tex0 * u_params.xy;
+  float2 src_pixel = pixel * u_params.zw;
+  float2 src = floor(src_pixel - 0.5) + 0.5;
+    
+  float3 color = float3(0.0, 0.0, 0.0);
+  float total_weight = 0.0;
+    
+  for (int i = -KERNEL_SIZE; i <= KERNEL_SIZE; i++)
+  {
+      for (int j = -KERNEL_SIZE; j <= KERNEL_SIZE; j++)
+      {
+          float2 offset = float2(int2(i, j));
+          float2 sample_pos = (src + offset) * u_src_size.zw;
+          float2 dxdy = src_pixel - (src + offset);
+          float weight = lanczos(dxdy.x) * lanczos(dxdy.y);
+            
+          color += SAMPLE_TEXTURE_LEVEL(samp0, ClampUV(sample_pos), 0.0).rgb * weight;
+          total_weight += weight;
+      }
+  }
+    
+  o_col0 = float4(color / total_weight, 1.0);
+})";
+
+  return std::move(ss).str();
+}
+
 std::string GPUShaderGen::GenerateDeinterlaceWeaveFragmentShader() const
 {
   std::stringstream ss;

src/core/gpu_shadergen.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: 2019-2024 Connor McLaughlin <stenzek@gmail.com>
+// SPDX-FileCopyrightText: 2019-2025 Connor McLaughlin <stenzek@gmail.com>
 // SPDX-License-Identifier: CC-BY-NC-ND-4.0
 
 #pragma once
@@ -14,6 +14,7 @@ public:
   std::string GenerateDisplayVertexShader() const;
   std::string GenerateDisplayFragmentShader(bool clamp_uv, bool nearest) const;
   std::string GenerateDisplaySharpBilinearFragmentShader() const;
+  std::string GenerateDisplayLanczosFragmentShader() const;
 
   std::string GenerateDeinterlaceWeaveFragmentShader() const;
   std::string GenerateDeinterlaceBlendFragmentShader() const;

src/core/settings.cpp

@@ -1900,7 +1900,7 @@ const char* Settings::GetForceVideoTimingDisplayName(ForceVideoTimingMode mode)
 }
 
 static constexpr const std::array s_display_scaling_names = {
-  "Nearest", "NearestInteger", "BilinearSmooth", "BilinearSharp", "BilinearInteger",
+  "Nearest", "NearestInteger", "BilinearSmooth", "BilinearSharp", "BilinearInteger", "Lanczos",
 };
 static constexpr const std::array s_display_scaling_display_names = {
   TRANSLATE_DISAMBIG_NOOP("Settings", "Nearest-Neighbor", "DisplayScalingMode"),
@@ -1908,6 +1908,7 @@ static constexpr const std::array s_display_scaling_display_names = {
   TRANSLATE_DISAMBIG_NOOP("Settings", "Bilinear (Smooth)", "DisplayScalingMode"),
   TRANSLATE_DISAMBIG_NOOP("Settings", "Bilinear (Sharp)", "DisplayScalingMode"),
   TRANSLATE_DISAMBIG_NOOP("Settings", "Bilinear (Integer)", "DisplayScalingMode"),
+  TRANSLATE_DISAMBIG_NOOP("Settings", "Lanczos (Sharp)", "DisplayScalingMode"),
 };
 
 std::optional<DisplayScalingMode> Settings::ParseDisplayScaling(const char* str)

src/core/types.h

@@ -184,6 +184,7 @@ enum class DisplayScalingMode : u8
   BilinearSmooth,
   BilinearSharp,
   BilinearInteger,
+  Lanczos,
   Count
 };