mirror of
https://github.com/hrydgard/ppsspp.git
synced 2026-10-01 14:58:14 +00:00
Update fsr_easu.fsh
This commit is contained in:
Benjamin Wegener authored and 
GitHub committed
1 parent
0ff1f6483c
commit
f304232c13
1 file changed
+132
-195
+132
-195
@@ -20,226 +20,163 @@
|
||||
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
||||
// SOFTWARE.
|
||||
|
||||
// Set precision for OpenGL ES
|
||||
#ifdef GL_ES
|
||||
precision mediump float;
|
||||
precision mediump int;
|
||||
#endif
|
||||
|
||||
// Texture sampler for input image
|
||||
uniform sampler2D sampler0;
|
||||
uniform vec2 u_texelDelta; // = 1 / textureSize
|
||||
uniform vec2 u_pixelDelta; // not used
|
||||
|
||||
// Texture coordinate deltas:
|
||||
// u_texelDelta = 1.0 / input texture size (texel size)
|
||||
// u_pixelDelta = 1.0 / output resolution (pixel size)
|
||||
uniform vec2 u_texelDelta;
|
||||
uniform vec2 u_pixelDelta;
|
||||
|
||||
// Interpolated texture coordinates from vertex shader
|
||||
varying vec2 v_texcoord0;
|
||||
|
||||
// Precision definition for high precision calculations when available
|
||||
#ifdef GL_FRAGMENT_PRECISION_HIGH
|
||||
#define HIGHP highp
|
||||
#else
|
||||
#define HIGHP mediump
|
||||
#endif
|
||||
|
||||
// ---------------- Luminance ----------------
|
||||
// Calculate luminance from RGB color using standard weights
|
||||
// Based on Rec. 709 luma coefficients for perceptual brightness
|
||||
HIGHP float Luma(vec3 c) { return dot(c, vec3(0.299, 0.587, 0.114)); }
|
||||
// ---------- FSR-EASU core (unchanged) ----------
|
||||
void FsrEasuCon(
|
||||
out vec4 con0, out vec4 con1, out vec4 con2, out vec4 con3,
|
||||
vec2 inputViewportInPixels,
|
||||
vec2 inputSizeInPixels,
|
||||
vec2 outputSizeInPixels)
|
||||
{
|
||||
con0 = vec4(
|
||||
inputViewportInPixels.x / outputSizeInPixels.x,
|
||||
inputViewportInPixels.y / outputSizeInPixels.y,
|
||||
0.5 * inputViewportInPixels.x / outputSizeInPixels.x - 0.5,
|
||||
0.5 * inputViewportInPixels.y / outputSizeInPixels.y - 0.5);
|
||||
|
||||
// ---------------- Cubic kernel (Catmull-Rom, B=0, C=0.5) ----------------
|
||||
// Separable cubic weights in 1D, evaluated at fractional position t in [0,1)
|
||||
// Implements Catmull-Rom spline with B=0, C=0.5 for smooth interpolation
|
||||
// Returns weights for 4 taps at positions: -1, 0, +1, +2
|
||||
HIGHP vec4 cubicWeights(HIGHP float t) {
|
||||
HIGHP float t2 = t * t;
|
||||
HIGHP float t3 = t2 * t;
|
||||
|
||||
// Calculate weights for each of the 4 taps using Catmull-Rom formula
|
||||
HIGHP float w0 = -0.5 * t + 1.0 * t2 - 0.5 * t3; // Tap at -1
|
||||
HIGHP float w1 = 1.0 - 2.5 * t2 + 1.5 * t3; // Tap at 0
|
||||
HIGHP float w2 = 0.5 * t + 2.0 * t2 - 1.5 * t3; // Tap at +1
|
||||
HIGHP float w3 = -0.5 * t2 + 0.5 * t3; // Tap at +2
|
||||
|
||||
return vec4(w0, w1, w2, w3);
|
||||
con1 = vec4(1.0, 1.0, 1.0, -1.0) / inputSizeInPixels.xyxy;
|
||||
con2 = vec4(-1.0, 2.0, 1.0, 2.0) / inputSizeInPixels.xyxy;
|
||||
con3 = vec4( 0.0, 4.0, 0.0, 0.0) / inputSizeInPixels.xyxy;
|
||||
}
|
||||
|
||||
// ---------------- Neighborhood & gradients ----------------
|
||||
// Structure to hold a 3x3 neighborhood of color values and luminance
|
||||
struct N3x3 {
|
||||
vec3 c[3][3]; // RGB color values for 3x3 grid
|
||||
HIGHP float l[3][3]; // Luminance values for 3x3 grid
|
||||
};
|
||||
|
||||
// Load a 3x3 neighborhood around the given UV coordinate
|
||||
// Samples 9 texels and calculates their luminance values
|
||||
N3x3 load3x3(HIGHP vec2 uv) {
|
||||
N3x3 nb;
|
||||
// Loop through the 3x3 grid centered at uv
|
||||
for (int j = -1; j <= 1; ++j) {
|
||||
for (int i = -1; i <= 1; ++i) {
|
||||
// Calculate texture coordinate for this grid position
|
||||
HIGHP vec2 t = uv + vec2(float(i), float(j)) * u_texelDelta;
|
||||
|
||||
// Sample color and calculate luminance
|
||||
vec3 s = texture2D(sampler0, t).rgb;
|
||||
nb.c[j+1][i+1] = s;
|
||||
nb.l[j+1][i+1] = Luma(s);
|
||||
}
|
||||
}
|
||||
return nb;
|
||||
void FsrEasuTapF(
|
||||
inout vec3 aC, inout float aW,
|
||||
vec2 off,
|
||||
vec2 dir, vec2 len,
|
||||
float lob, float clp,
|
||||
vec3 c)
|
||||
{
|
||||
vec2 v = vec2(dot(off, dir), dot(off, vec2(-dir.y, dir.x))) * len;
|
||||
float d2 = min(dot(v, v), clp);
|
||||
float wB = 0.4 * d2 - 1.0;
|
||||
float wA = lob * d2 - 1.0;
|
||||
wB *= wB; wA *= wA;
|
||||
wB = 1.5625 * wB - 0.5625;
|
||||
float w = wB * wA;
|
||||
aC += c * w;
|
||||
aW += w;
|
||||
}
|
||||
|
||||
// Calculate gradient using Sobel operator on luminance values
|
||||
// Returns gradient vector (gx, gy) representing edge direction and magnitude
|
||||
HIGHP vec2 sobelGrad(in HIGHP float l[3][3]) {
|
||||
// Sobel X kernel (horizontal edge detection)
|
||||
HIGHP float gx = (l[0][2] + 2.0*l[1][2] + l[2][2]) - (l[0][0] + 2.0*l[1][0] + l[2][0]);
|
||||
|
||||
// Sobel Y kernel (vertical edge detection)
|
||||
HIGHP float gy = (l[2][0] + 2.0*l[2][1] + l[2][2]) - (l[0][0] + 2.0*l[0][1] + l[0][2]);
|
||||
|
||||
return vec2(gx, gy);
|
||||
void FsrEasuSetF(
|
||||
inout vec2 dir, inout float len,
|
||||
float w,
|
||||
float lA, float lB, float lC, float lD, float lE)
|
||||
{
|
||||
float lenX = max(abs(lD - lC), abs(lC - lB));
|
||||
float dirX = lD - lB;
|
||||
dir.x += dirX * w;
|
||||
lenX = clamp(abs(dirX) / (lenX + 1e-5), 0.0, 1.0);
|
||||
lenX *= lenX;
|
||||
len += lenX * w;
|
||||
|
||||
float lenY = max(abs(lE - lC), abs(lC - lA));
|
||||
float dirY = lE - lA;
|
||||
dir.y += dirY * w;
|
||||
lenY = clamp(abs(dirY) / (lenY + 1e-5), 0.0, 1.0);
|
||||
lenY *= lenY;
|
||||
len += lenY * w;
|
||||
}
|
||||
|
||||
// ---------------- Anti-ringing clamp vs. base 2x2 ----------------
|
||||
// Clamp the color to prevent ringing artifacts by constraining it
|
||||
// within the range of the base 2x2 pixel block
|
||||
void clampToBase2x2(inout vec3 color, HIGHP vec2 srcPxCenter) {
|
||||
// Find the base pixel coordinates (integer grid)
|
||||
HIGHP vec2 ip = floor(srcPxCenter);
|
||||
HIGHP vec2 baseUV = ip * u_texelDelta;
|
||||
vec3 FsrEasuF(vec2 ip, vec4 con0, vec4 con1, vec4 con2, vec4 con3)
|
||||
{
|
||||
vec2 pp = ip * con0.xy + con0.zw;
|
||||
vec2 fp = floor(pp);
|
||||
pp -= fp;
|
||||
|
||||
// Sample the four corners of the 2x2 base block
|
||||
vec3 c00 = texture2D(sampler0, baseUV).rgb;
|
||||
vec3 c10 = texture2D(sampler0, baseUV + vec2(u_texelDelta.x, 0.0)).rgb;
|
||||
vec3 c01 = texture2D(sampler0, baseUV + vec2(0.0, u_texelDelta.y)).rgb;
|
||||
vec3 c11 = texture2D(sampler0, baseUV + u_texelDelta).rgb;
|
||||
vec2 p0 = fp * con1.xy + con1.zw;
|
||||
vec2 p1 = p0 + con2.xy;
|
||||
vec2 p2 = p0 + con2.zw;
|
||||
vec2 p3 = p0 + con3.xy;
|
||||
|
||||
// Find the minimum and maximum colors in the 2x2 block
|
||||
vec3 lo = min(min(c00, c10), min(c01, c11));
|
||||
vec3 hi = max(max(c00, c10), max(c01, c11));
|
||||
|
||||
// Clamp the color to be within the range of the base block
|
||||
color = clamp(color, lo, hi);
|
||||
vec4 off = vec4(-0.5, 0.5, -0.5, 0.5) * con1.xxyy;
|
||||
|
||||
vec3 bC = texture2D(sampler0, p0 + off.xw).rgb; float bL = bC.g + 0.5*(bC.r + bC.b);
|
||||
vec3 cC = texture2D(sampler0, p0 + off.yw).rgb; float cL = cC.g + 0.5*(cC.r + cC.b);
|
||||
vec3 iC = texture2D(sampler0, p1 + off.xw).rgb; float iL = iC.g + 0.5*(iC.r + iC.b);
|
||||
vec3 jC = texture2D(sampler0, p1 + off.yw).rgb; float jL = jC.g + 0.5*(jC.r + jC.b);
|
||||
vec3 fC = texture2D(sampler0, p1 + off.yz).rgb; float fL = fC.g + 0.5*(fC.r + fC.b);
|
||||
vec3 eC = texture2D(sampler0, p1 + off.xz).rgb; float eL = eC.g + 0.5*(eC.r + eC.b);
|
||||
vec3 kC = texture2D(sampler0, p2 + off.xw).rgb; float kL = kC.g + 0.5*(kC.r + kC.b);
|
||||
vec3 lC = texture2D(sampler0, p2 + off.yw).rgb; float lL = lC.g + 0.5*(lC.r + lC.b);
|
||||
vec3 hC = texture2D(sampler0, p2 + off.yz).rgb; float hL = hC.g + 0.5*(hC.r + hC.b);
|
||||
vec3 gC = texture2D(sampler0, p2 + off.xz).rgb; float gL = gC.g + 0.5*(gC.r + gC.b);
|
||||
vec3 oC = texture2D(sampler0, p3 + off.yz).rgb; float oL = oC.g + 0.5*(oC.r + oC.b);
|
||||
vec3 nC = texture2D(sampler0, p3 + off.xz).rgb; float nL = nC.g + 0.5*(nC.r + nC.b);
|
||||
|
||||
vec2 dir = vec2(0.0);
|
||||
float len = 0.0;
|
||||
|
||||
FsrEasuSetF(dir, len, (1.0 - pp.x)*(1.0 - pp.y), bL, eL, fL, gL, jL);
|
||||
FsrEasuSetF(dir, len, pp.x *(1.0 - pp.y), cL, fL, gL, hL, kL);
|
||||
FsrEasuSetF(dir, len, (1.0 - pp.x)* pp.y , fL, iL, jL, kL, nL);
|
||||
FsrEasuSetF(dir, len, pp.x * pp.y , gL, jL, kL, lL, oL);
|
||||
|
||||
vec2 dir2 = dir * dir;
|
||||
float dirR = dir2.x + dir2.y;
|
||||
bool zro = dirR < 1.0/32768.0;
|
||||
dirR = inversesqrt(dirR);
|
||||
dirR = zro ? 1.0 : dirR;
|
||||
dir = zro ? vec2(1.0, 0.0) : (dir * dirR);
|
||||
|
||||
len = len * 0.5;
|
||||
len *= len;
|
||||
|
||||
float stretch = dot(dir, dir) / max(abs(dir.x), abs(dir.y));
|
||||
vec2 len2 = vec2(1.0 + (stretch - 1.0)*len,
|
||||
1.0 - 0.5*len);
|
||||
|
||||
float lob = 0.5 - 0.29*len;
|
||||
float clp = 1.0/lob;
|
||||
|
||||
vec3 aC = vec3(0.0);
|
||||
float aW = 0.0;
|
||||
|
||||
FsrEasuTapF(aC, aW, vec2( 0,-1)-pp, dir, len2, lob, clp, bC);
|
||||
FsrEasuTapF(aC, aW, vec2( 1,-1)-pp, dir, len2, lob, clp, cC);
|
||||
FsrEasuTapF(aC, aW, vec2(-1, 1)-pp, dir, len2, lob, clp, iC);
|
||||
FsrEasuTapF(aC, aW, vec2( 0, 1)-pp, dir, len2, lob, clp, jC);
|
||||
FsrEasuTapF(aC, aW, vec2( 0, 0)-pp, dir, len2, lob, clp, fC);
|
||||
FsrEasuTapF(aC, aW, vec2(-1, 0)-pp, dir, len2, lob, clp, eC);
|
||||
FsrEasuTapF(aC, aW, vec2( 1, 1)-pp, dir, len2, lob, clp, kC);
|
||||
FsrEasuTapF(aC, aW, vec2( 2, 1)-pp, dir, len2, lob, clp, lC);
|
||||
FsrEasuTapF(aC, aW, vec2( 2, 0)-pp, dir, len2, lob, clp, hC);
|
||||
FsrEasuTapF(aC, aW, vec2( 1, 0)-pp, dir, len2, lob, clp, gC);
|
||||
FsrEasuTapF(aC, aW, vec2( 1, 2)-pp, dir, len2, lob, clp, oC);
|
||||
FsrEasuTapF(aC, aW, vec2( 0, 2)-pp, dir, len2, lob, clp, nC);
|
||||
|
||||
vec3 min4 = min(min(fC, gC), min(jC, kC));
|
||||
vec3 max4 = max(max(fC, gC), max(jC, kC));
|
||||
return min(max4, max(min4, aC / aW));
|
||||
}
|
||||
|
||||
// ---------------- Edge-adaptive anisotropic 4x4 reconstruction ----------------
|
||||
// Main EASU (Edge Adaptive Spatial Upsampling) function
|
||||
// Performs edge-adaptive upscaling using anisotropic filtering
|
||||
vec4 FsrEasu(vec2 uvOut) {
|
||||
// Calculate input and output dimensions
|
||||
HIGHP vec2 inSize = 1.0 / u_texelDelta; // Input texture size
|
||||
HIGHP vec2 outSize = 1.0 / u_pixelDelta; // Output resolution
|
||||
|
||||
// Map output UV coordinate to input pixel space
|
||||
HIGHP vec2 srcPx = uvOut / u_texelDelta;
|
||||
|
||||
// Align to pixel centers in input space
|
||||
HIGHP vec2 center = floor(srcPx - 0.5) + 0.5;
|
||||
HIGHP vec2 phase = fract(srcPx - 0.5);
|
||||
|
||||
// Load 3x3 neighborhood and calculate local gradients
|
||||
N3x3 nb = load3x3(uvOut);
|
||||
HIGHP vec2 g = sobelGrad(nb.l); // Edge gradient
|
||||
HIGHP float gLen = length(g); // Edge strength
|
||||
|
||||
// Calculate edge basis vectors:
|
||||
// nrm: Normal to the edge (points toward edge direction)
|
||||
// tan: Tangent to the edge (along edge direction)
|
||||
HIGHP vec2 nrm = (gLen > 1e-4) ? (g / gLen) : vec2(0.0, 1.0);
|
||||
HIGHP vec2 tan = vec2(-nrm.y, nrm.x);
|
||||
|
||||
// Calculate anisotropy based on edge strength:
|
||||
// Stronger edges = more anisotropy (directional filtering)
|
||||
HIGHP float anis = clamp(gLen * 0.35, 0.0, 1.0);
|
||||
HIGHP float rTan = mix(1.0, 1.6, anis); // Support radius along tangent
|
||||
HIGHP float rNrm = mix(1.0, 0.7, anis); // Support radius along normal
|
||||
|
||||
// Define 4x4 sampling footprint anchored at (center - 1, center - 1)
|
||||
HIGHP vec2 basePx = center - 1.0;
|
||||
HIGHP vec2 baseUV = basePx * u_texelDelta;
|
||||
|
||||
// Precompute separable cubic weights for the current phase
|
||||
HIGHP vec4 wx = cubicWeights(phase.x);
|
||||
HIGHP vec4 wy = cubicWeights(phase.y);
|
||||
|
||||
// Initialize accumulator for weighted samples
|
||||
vec3 accum = vec3(0.0);
|
||||
HIGHP float wsum = 0.0;
|
||||
|
||||
// Process all 16 samples in the 4x4 footprint
|
||||
for (int j = 0; j < 4; ++j) {
|
||||
for (int i = 0; i < 4; ++i) {
|
||||
// Calculate offset from center in pixel units
|
||||
HIGHP vec2 d = vec2(float(i) - 1.0 - (1.0 - phase.x),
|
||||
float(j) - 1.0 - (1.0 - phase.y));
|
||||
|
||||
// Project offset onto tangent and normal directions
|
||||
HIGHP float dt = dot(d, tan) / rTan; // Distance along tangent
|
||||
HIGHP float dn = dot(d, nrm) / rNrm; // Distance along normal
|
||||
|
||||
// Calculate anisotropic weights using Catmull-Rom splines
|
||||
HIGHP float at = abs(dt); // Absolute distance along tangent
|
||||
HIGHP float an = abs(dn); // Absolute distance along normal
|
||||
|
||||
// Calculate cubic weights for tangent direction
|
||||
HIGHP float at2 = at * at; HIGHP float at3 = at2 * at;
|
||||
HIGHP float wt = (at < 1.0)
|
||||
? (1.0 - 2.5*at2 + 1.5*at3) // For |d| < 1
|
||||
: ((at < 2.0) ? (-0.5*at + 2.0*at2 - 1.5*at3) : 0.0); // For 1 ≤ |d| < 2
|
||||
|
||||
// Calculate cubic weights for normal direction
|
||||
HIGHP float an2 = an * an; HIGHP float an3 = an2 * an;
|
||||
HIGHP float wn = (an < 1.0)
|
||||
? (1.0 - 2.5*an2 + 1.5*an3) // For |d| < 1
|
||||
: ((an < 2.0) ? (-0.5*an + 2.0*an2 - 1.5*an3) : 0.0); // For 1 ≤ |d| < 2
|
||||
|
||||
// Combine directional weights
|
||||
HIGHP float wAniso = max(0.0, wt) * max(0.0, wn);
|
||||
|
||||
// Calculate baseline separable grid weight
|
||||
HIGHP float wSep = wx[i] * wy[j];
|
||||
|
||||
// Blend between separable and anisotropic weights based on edge strength
|
||||
HIGHP float kEdge = clamp(gLen * 0.4, 0.0, 1.0);
|
||||
HIGHP float w = mix(wSep, wAniso, kEdge);
|
||||
|
||||
// Sample texture and accumulate weighted color
|
||||
vec2 uv = baseUV + vec2(float(i) * u_texelDelta.x,
|
||||
float(j) * u_texelDelta.y);
|
||||
vec3 s = texture2D(sampler0, uv).rgb;
|
||||
|
||||
accum += s * w;
|
||||
wsum += w;
|
||||
}
|
||||
}
|
||||
|
||||
// Normalize accumulated color by total weight
|
||||
vec3 recon = (wsum > 0.0) ? (accum / wsum) : texture2D(sampler0, uvOut).rgb;
|
||||
|
||||
// Early-out optimization for very flat regions to avoid unnecessary processing
|
||||
HIGHP float lMin = min(min(nb.l[0][0], nb.l[0][1]), min(nb.l[0][2], min(min(nb.l[1][0], nb.l[1][1]), min(nb.l[1][2], min(nb.l[2][0], min(nb.l[2][1], nb.l[2][2]))))));
|
||||
HIGHP float lMax = max(max(nb.l[0][0], nb.l[0][1]), max(nb.l[0][2], max(max(nb.l[1][0], nb.l[1][1]), max(nb.l[1][2], max(nb.l[2][0], max(nb.l[2][1], nb.l[2][2]))))));
|
||||
HIGHP float contrast = lMax - lMin;
|
||||
|
||||
// Skip sharpening in low-contrast regions
|
||||
if (contrast < 0.02) {
|
||||
return vec4(texture2D(sampler0, uvOut).rgb, 1.0);
|
||||
}
|
||||
|
||||
// Apply anti-ringing clamp to prevent overshoots
|
||||
clampToBase2x2(recon, center);
|
||||
|
||||
// Final clamp to valid color range
|
||||
return vec4(clamp(recon, 0.0, 1.0), 1.0);
|
||||
}
|
||||
|
||||
// Main fragment shader entry point
|
||||
// ---------- entry point ----------
|
||||
void main() {
|
||||
// Apply EASU upscaling to the current fragment
|
||||
gl_FragColor = FsrEasu(v_texcoord0);
|
||||
vec2 texSize = vec2(1.0) / u_texelDelta; // texture resolution
|
||||
vec2 outSize = vec2(1.0) / u_texelDelta; // if canvas == texture, same
|
||||
|
||||
vec4 con0, con1, con2, con3;
|
||||
FsrEasuCon(con0, con1, con2, con3,
|
||||
texSize, texSize, outSize);
|
||||
|
||||
// Convert 0-1 texcoord to integer pixel in output space
|
||||
vec2 ip = v_texcoord0 * outSize - vec2(0.5);
|
||||
|
||||
vec3 rgb = FsrEasuF(ip, con0, con1, con2, con3);
|
||||
gl_FragColor = vec4(rgb, 1.0);
|
||||
}
|
||||
Reference in new issue
Block a user