optimize with precomputed kernel

Signed-off-by: lizzie <lizzie@eden-emu.dev>

src/video_core/host_shaders/present_lanczos.frag

@@ -9,32 +9,24 @@ layout (location = 0) in vec2 frag_tex_coord;
 layout (location = 0) out vec4 color;
 layout (binding = 0) uniform sampler2D color_texture;
 
-#define PI 3.1415926535897932384626433
+// precomputed kernel
-
+const float w_kernel[49] = float[] (
-float sinc(float x) {
+    -0.238811f, 0.531959f, 0.961865f, 1.000000f, 0.961865f, 0.531959f, -0.238811f, 0.531959f, 0.957419f, 0.313883f, -0.000000f, 0.313883f, 0.957419f, 0.531959f, 0.961865f, 0.313883f, -0.322602f, 0.000000f, -0.322602f, 0.313883f, 0.961865f, 1.000000f, -0.000000f, 0.000000f, 1.000000f, 0.000000f, -0.000000f, 1.000000f, 0.961865f, 0.313883f, -0.322602f, 0.000000f, -0.322602f, 0.313883f, 0.961865f, 0.531959f, 0.957419f, 0.313883f, -0.000000f, 0.313883f, 0.957419f, 0.531959f, -0.238811f, 0.531959f, 0.961865f, 1.000000f, 0.961865f, 0.531959f, -0.238811f
-    return x == 0.0f ? 1.0f : sin(PI * x) / (PI * x);
+);
-}
+const vec2 w_pos[49] = vec2[] (
-
+    vec2(-0.750000f, -0.750000f), vec2(-0.750000f, -0.500000f), vec2(-0.750000f, -0.250000f), vec2(-0.750000f, 0.000000f), vec2(-0.750000f, 0.250000f), vec2(-0.750000f, 0.500000f), vec2(-0.750000f, 0.750000f), vec2(-0.500000f, -0.750000f), vec2(-0.500000f, -0.500000f), vec2(-0.500000f, -0.250000f), vec2(-0.500000f, 0.000000f), vec2(-0.500000f, 0.250000f), vec2(-0.500000f, 0.500000f), vec2(-0.500000f, 0.750000f), vec2(-0.250000f, -0.750000f), vec2(-0.250000f, -0.500000f), vec2(-0.250000f, -0.250000f), vec2(-0.250000f, 0.000000f), vec2(-0.250000f, 0.250000f), vec2(-0.250000f, 0.500000f), vec2(-0.250000f, 0.750000f), vec2(0.000000f, -0.750000f), vec2(0.000000f, -0.500000f), vec2(0.000000f, -0.250000f), vec2(0.000000f, 0.000000f), vec2(0.000000f, 0.250000f), vec2(0.000000f, 0.500000f), vec2(0.000000f, 0.750000f), vec2(0.250000f, -0.750000f), vec2(0.250000f, -0.500000f), vec2(0.250000f, -0.250000f), vec2(0.250000f, 0.000000f), vec2(0.250000f, 0.250000f), vec2(0.250000f, 0.500000f), vec2(0.250000f, 0.750000f), vec2(0.500000f, -0.750000f), vec2(0.500000f, -0.500000f), vec2(0.500000f, -0.250000f), vec2(0.500000f, 0.000000f), vec2(0.500000f, 0.250000f), vec2(0.500000f, 0.500000f), vec2(0.500000f, 0.750000f), vec2(0.750000f, -0.750000f), vec2(0.750000f, -0.500000f), vec2(0.750000f, -0.250000f), vec2(0.750000f, 0.000000f), vec2(0.750000f, 0.250000f), vec2(0.750000f, 0.500000f), vec2(0.750000f, 0.750000f)
-float lanczos(vec2 v, float a) {
+);
-    float d = sqrt(v.x * v.x + v.y * v.y);
+const float w_sum = 21.045683f;
-    return sinc(d) / sinc(d / a);
-}
-
 vec4 textureLanczos(sampler2D textureSampler, vec2 p) {
-    const int r = 1; //radius (1 = 3 steps)
+    const int r = 3; //radius (1 = 3 steps)
     vec3 c_sum = vec3(0.0f);
-    float w_sum = 0.0f;
     vec2 res = vec2(textureSize(textureSampler, 0));
     vec2 cc = floor(p * res) / res;
-    // kernel size = (r * 2 + 1) * (r * 2 + 1)
+    for (int i = 0; i < 49; i++) { // kernel size = (r * 2 + 1) ^ 2
-    for (int x = -r; x <= r; x++)
+        vec2 kp = w_pos[i] / res;
-        for (int y = -r; y <= r; y++) {
+        vec2 uv = cc + kp;
-            vec2 kp = 0.5f * (vec2(x, y) / res); // 0.5 = half-pixel level resampling
+        c_sum += w_kernel[i] * texture(textureSampler, p + kp).rgb;
-            vec2 uv = cc + kp;
+    }
-            float w = lanczos(kp, float(r));
-            c_sum += w * texture(textureSampler, p + kp).rgb;
-            w_sum += w;
-        }
     return vec4(c_sum / w_sum, 1.0f);
 }
 

tools/lanczos_gen.c

@@ -0,0 +1,48 @@
+// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+// clang -lm tools/lanczos_gen.c -o tools/lanczos_gen && ./tools/lanczos_gen
+#include <stdio.h>
+#include <math.h>
+
+double sinc(double x) {
+    return x == 0.0f ? 1.0f : sin(M_PI * x) / (M_PI * x);
+}
+
+typedef struct vec2 {
+    double x;
+    double y;
+} vec2;
+
+double lanczos(vec2 v, float a) {
+    double d = sqrt(v.x * v.x + v.y * v.y);
+    return sinc(d) / sinc(d / a);
+}
+
+int main(int argc, char* argv[]) {
+    const int r = 3; //radius (1 = 3 steps)
+    const int k_size = (r * 2 + 1) * (r * 2 + 1);
+    double w_sum = 0.0f;
+    // kernel size = (r * 2 + 1) ^ 2
+    printf("const float w_kernel[%i] = float[] (\n    ", k_size);
+    double factor = 1.0f / ((double)r + 1.0f);
+    for (int x = -r; x <= r; x++)
+        for (int y = -r; y <= r; y++) {
+            double w = lanczos((vec2){ .x = x, .y = y }, (double)r);
+            printf("%lff, ", w);
+            w_sum += w;
+        }
+    printf("\n);\n");
+    printf("const vec2 w_pos[%i] = vec2[] (\n    ", k_size);
+    for (int x = -r; x <= r; x++)
+        for (int y = -r; y <= r; y++) {
+            vec2 kp = (vec2){
+                .x = x * factor,
+                .y = y * factor
+            };
+            printf("vec2(%lff, %lff), ", kp.x, kp.y);
+        }
+    printf("\n);\n");
+    printf("const float w_sum = %lff;\n", w_sum);
+    return 0;
+}