Separate mmpx enhance from mmpx (#3453)

Use conditional weak blending instead of pixel copying to eliminate artifacts on straight lines
Reduce algorithm overhead
use explicitly initialize to fix dx11 fxc uninitialized variable false positive

Update gpu_hw_shadergen.cpp

Update gpu_hw_shadergen.cpp
This commit is contained in:
crashGG 2025-06-18 11:28:19 +03:00 committed by GitHub
parent d83ecb0582
commit 99268f22ec
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194

View File

@ -727,10 +727,9 @@ void FilteredSampleFromVRAM(TEXPAGE_VALUE texpage, float2 coords, float4 uv_limi
)"; )";
} }
else if (texture_filter == GPUTextureFilter::MMPX || texture_filter == GPUTextureFilter::MMPXEnhanced) else if (texture_filter == GPUTextureFilter::MMPX)
{ {
DefineMacro(ss, "TEXTURE_ALPHA_BLENDING", false); DefineMacro(ss, "TEXTURE_ALPHA_BLENDING", false);
DefineMacro(ss, "MMPX_ENHANCED", (texture_filter == GPUTextureFilter::MMPXEnhanced));
ss << "#define src(xoffs, yoffs) packUnorm4x8(SampleFromVRAM(texpage, clamp(bcoords + float2((xoffs), (yoffs)), " ss << "#define src(xoffs, yoffs) packUnorm4x8(SampleFromVRAM(texpage, clamp(bcoords + float2((xoffs), (yoffs)), "
"uv_limits.xy, uv_limits.zw)))\n"; "uv_limits.xy, uv_limits.zw)))\n";
@ -743,11 +742,7 @@ void FilteredSampleFromVRAM(TEXPAGE_VALUE texpage, float2 coords, float4 uv_limi
ss << R"( ss << R"(
uint luma(uint C) { uint luma(uint C) {
uint alpha = (C & 0xFF000000u) >> 24; uint alpha = (C & 0xFF000000u) >> 24;
#if MMPX_ENHANCED
return (((C & 0x00FF0000u) >> 16) + ((C & 0x0000FF00u) >> 8) + (C & 0x000000FFu)) + (255u - alpha) * 3;
#else
return (((C & 0x00FF0000u) >> 16) + ((C & 0x0000FF00u) >> 8) + (C & 0x000000FFu) + 1u) * (256u - alpha); return (((C & 0x00FF0000u) >> 16) + ((C & 0x0000FF00u) >> 8) + (C & 0x000000FFu) + 1u) * (256u - alpha);
#endif
} }
bool all_eq2(uint B, uint A0, uint A1) { bool all_eq2(uint B, uint A0, uint A1) {
@ -774,26 +769,150 @@ bool none_eq4(uint B, uint A0, uint A1, uint A2, uint A3) {
return B != A0 && B != A1 && B != A2 && B != A3; return B != A0 && B != A1 && B != A2 && B != A3;
} }
#if MMPX_ENHANCED void FilteredSampleFromVRAM(TEXPAGE_VALUE texpage, float2 coords, float4 uv_limits, out float4 texcol, out float ialpha)
// Calculate the RGB distance between two ABGR8 colors
float rgb_distance(uint a, uint b)
{ {
vec4 ca = unpackUnorm4x8(a); float2 bcoords = floor(coords);
vec4 cb = unpackUnorm4x8(b);
return distance(ca.rgb, cb.rgb); uint A = src(-1, -1), B = src(+0, -1), C = src(+1, -1);
uint D = src(-1, +0), E = src(+0, +0), F = src(+1, +0);
uint G = src(-1, +1), H = src(+0, +1), I = src(+1, +1);
uint J = E, K = E, L = E, M = E;
if (((A ^ E) | (B ^ E) | (C ^ E) | (D ^ E) | (F ^ E) | (G ^ E) | (H ^ E) | (I ^ E)) != 0u) {
uint P = src(+0, -2), S = src(+0, +2);
uint Q = src(-2, +0), R = src(+2, +0);
uint Bl = luma(B), Dl = luma(D), El = luma(E), Fl = luma(F), Hl = luma(H);
// 1:1 slope rules
if ((D == B && D != H && D != F) && (El >= Dl || E == A) && any_eq3(E, A, C, G) && ((El < Dl) || A != D || E != P || E != Q)) J = D;
if ((B == F && B != D && B != H) && (El >= Bl || E == C) && any_eq3(E, A, C, I) && ((El < Bl) || C != B || E != P || E != R)) K = B;
if ((H == D && H != F && H != B) && (El >= Hl || E == G) && any_eq3(E, A, G, I) && ((El < Hl) || G != H || E != S || E != Q)) L = H;
if ((F == H && F != B && F != D) && (El >= Fl || E == I) && any_eq3(E, C, G, I) && ((El < Fl) || I != H || E != R || E != S)) M = F;
// Intersection rules
if ((E != F && all_eq4(E, C, I, D, Q) && all_eq2(F, B, H)) && (F != src(+3, +0))) K = M = F;
if ((E != D && all_eq4(E, A, G, F, R) && all_eq2(D, B, H)) && (D != src(-3, +0))) J = L = D;
if ((E != H && all_eq4(E, G, I, B, P) && all_eq2(H, D, F)) && (H != src(+0, +3))) L = M = H;
if ((E != B && all_eq4(E, A, C, H, S) && all_eq2(B, D, F)) && (B != src(+0, -3))) J = K = B;
if (Bl < El && all_eq4(E, G, H, I, S) && none_eq4(E, A, D, C, F)) J = K = B;
if (Hl < El && all_eq4(E, A, B, C, P) && none_eq4(E, D, G, I, F)) L = M = H;
if (Fl < El && all_eq4(E, A, D, G, Q) && none_eq4(E, B, C, I, H)) K = M = F;
if (Dl < El && all_eq4(E, C, F, I, R) && none_eq4(E, B, A, G, H)) J = L = D;
// 2:1 slope rules
if (H != B) {
if (H != A && H != E && H != C) {
if (all_eq3(H, G, F, R) && none_eq2(H, D, src(+2, -1))) L = M;
if (all_eq3(H, I, D, Q) && none_eq2(H, F, src(-2, -1))) M = L;
}
if (B != I && B != G && B != E) {
if (all_eq3(B, A, F, R) && none_eq2(B, D, src(+2, +1))) J = K;
if (all_eq3(B, C, D, Q) && none_eq2(B, F, src(-2, +1))) K = J;
}
} // H !== B
if (F != D) {
if (D != I && D != E && D != C) {
if (all_eq3(D, A, H, S) && none_eq2(D, B, src(+1, +2))) J = L;
if (all_eq3(D, G, B, P) && none_eq2(D, H, src(+1, -2))) L = J;
}
if (F != E && F != A && F != G) {
if (all_eq3(F, C, H, S) && none_eq2(F, B, src(-1, +2))) K = M;
if (all_eq3(F, I, B, P) && none_eq2(F, H, src(-1, -2))) M = K;
}
} // F !== D
} // not constant
// select quadrant based on fractional part of texture coordinates
float2 fpart = frac(coords);
uint res = (fpart.x < 0.5f) ? ((fpart.y < 0.5f) ? J : L) : ((fpart.y < 0.5f) ? K : M);
ialpha = float(res != 0u);
texcol = unpackUnorm4x8(res);
} }
// Calculate the luminance difference between two ABGR8 colors and normalize it #undef src
float luma_distance(uint a, uint b) )";
{ }
return abs(int(luma(a)) - int(luma(b))) * 0.0006535948f; // Multiplicative replacement for division by 1530 else if (texture_filter == GPUTextureFilter::MMPXEnhanced)
{
DefineMacro(ss, "TEXTURE_ALPHA_BLENDING", false);
DefineMacro(ss, "MMPX_ENHANCED", (texture_filter == GPUTextureFilter::MMPXEnhanced));
ss << "#define src(xoffs, yoffs) packUnorm4x8(SampleFromVRAM(texpage, clamp(bcoords + float2((xoffs), (yoffs)), "
"uv_limits.xy, uv_limits.zw)))\n";
/*
* This part of the shader is from MMPX.glc from https://casual-effects.com/research/McGuire2021PixelArt/index.html
* Copyright 2020 Morgan McGuire & Mara Gagiu.
* Provided under the Open Source MIT license https://opensource.org/licenses/MIT
*/
ss << R"(
uint luma(uint C) {
uint alpha = (C & 0xFF000000u) >> 24;
return (((C & 0x00FF0000u) >> 16) + ((C & 0x0000FF00u) >> 8) + (C & 0x000000FFu) + 1u) * (256u - alpha);
} }
bool all_eq2(uint B, uint A0, uint A1) {
return ((B ^ A0) | (B ^ A1)) == 0u;
}
bool all_eq3(uint B, uint A0, uint A1, uint A2) {
return ((B ^ A0) | (B ^ A1) | (B ^ A2)) == 0u;
}
bool all_eq4(uint B, uint A0, uint A1, uint A2, uint A3) {
return ((B ^ A0) | (B ^ A1) | (B ^ A2) | (B ^ A3)) == 0u;
}
bool any_eq3(uint B, uint A0, uint A1, uint A2) {
return B == A0 || B == A1 || B == A2;
}
bool none_eq2(uint B, uint A0, uint A1) {
return (B != A0) && (B != A1);
}
bool none_eq4(uint B, uint A0, uint A1, uint A2, uint A3) {
return B != A0 && B != A1 && B != A2 && B != A3;
}
// Two-stage weak blending, mix/none
uint admix2d(uint a, uint b) {
float4 a_float = unpackUnorm4x8(a);
float4 b_float = unpackUnorm4x8(b);
float3 diff_rgb = a_float.rgb - b_float.rgb;
float rgbDist = dot(diff_rgb, diff_rgb);
// Combine conditional judgments (reduce branches)
bool aIsBlack = dot(a_float.rgb, a_float.rgb) < 0.01;
//bool aIsTransparent = a_float.a < 0.01;
//bool bIsTransparent = b_float.a < 0.01;
if (aIsBlack ) return b;
// Determine blending mode based on distance
float4 result;
if (rgbDist < 1.0) {
// Close distance: linearly blend RGB and Alpha
result = (a_float + b_float) * 0.5;
} else {
// Far distance: return b
result = b_float;
}
// Repack as uint
return packUnorm4x8(result);
}
/*============================================================================= /*=============================================================================
Helper function for 4-pixel intersection pattern matching: Scores the number of Auxiliary function for 4-pixel cross determination: scores the number of matches at specific positions of the pattern.
matches for specific morphological positions. Three morphological conditions Three pattern conditions are determined, requiring 6 points to be satisfied.
need to be met with a total score of 6.
A B C A B 1 A B C A B 1
@ -810,23 +929,26 @@ bool countPatternMatches(uint LA, uint LB, uint L1, uint L2, uint L3, uint L4, u
int score3 = 0; // Horizontal/vertical line pattern int score3 = 0; // Horizontal/vertical line pattern
int scoreBonus = 0; int scoreBonus = 0;
// Jointly judge the visual difference between two pixels using RGB distance and luminance distance // Replace Euclidean formula with dot product to save a square root calculation
float pixelDist = rgb_distance(LA, LB)*0.356822 + luma_distance(LA, LB)*0.382; // The ratio is the golden ratio float4 a_float = unpackUnorm4x8(LA);
float4 b_float = unpackUnorm4x8(LB);
// Increase details for very similar colors and reduce details for highly contrasting colors (font edges). float3 diff_rgb = a_float.rgb - b_float.rgb;
if (pixelDist < 0.12) { // Colors are quite similar float rgbDist = dot(diff_rgb, diff_rgb);
// Add details for very close colors, reduce details for highly different colors (font edges)
if (rgbDist < 0.06386) { // Point set after quadratic golden section, colors are quite close
scoreBonus += 1; scoreBonus += 1;
} else if (pixelDist > 0.9) { // Black and white contrast } else if (rgbDist > 2.18847) { // Point set after quadratic golden section, significant difference
scoreBonus -= 1; scoreBonus -= 1;
} }
// Diagonal patterns use a penalty system: intersections deduct points, meeting conditions adds points back // Diagonals use a deduction system: deduct points for crosses, add back if conditions are met
// 1. Diagonal pattern ╲ (Condition: B = 2 or 4) // 1. Diagonal pattern ╲ (Condition: B = 2 or 4)
if (LB == L2 || LB == L4) { if (LB == L2 || LB == L4) {
score1 -= int(LB == L2 && LA == L1) * 1; // A-1, B-2 form an intersection, deduct points score1 -= int(LB == L2 && LA == L1) * 1; // A-1 and B-2 form a cross, deduct points
score1 -= int(LB == L4 && LA == L5) * 1; // A-5, B-4 form an intersection, deduct points score1 -= int(LB == L4 && LA == L5) * 1; // A-5 and B-4 form a cross, deduct points
// If the following triangular patterns are met, they can offset the intersection penalties // If the following triangular pattern is satisfied, offset the above cross deductions
score1 += int(LB == L1 && L1 == L2) * 1; // B-1-2 form a triangular pattern, add points score1 += int(LB == L1 && L1 == L2) * 1; // B-1-2 form a triangular pattern, add points
score1 += int(LB == L4 && L4 == L5) * 1; // B-4-5 form a triangular pattern, add points score1 += int(LB == L4 && L4 == L5) * 1; // B-4-5 form a triangular pattern, add points
score1 += int(L2 == L3 && L3 == L4) * 1; // 2-3-4 form a triangular pattern, add points score1 += int(L2 == L3 && L3 == L4) * 1; // 2-3-4 form a triangular pattern, add points
@ -836,11 +958,11 @@ bool countPatternMatches(uint LA, uint LB, uint L1, uint L2, uint L3, uint L4, u
// 2. Diagonal pattern (Condition: A = 1 or 5) // 2. Diagonal pattern (Condition: A = 1 or 5)
if (LA == L1 || LA == L5) { if (LA == L1 || LA == L5) {
score2 -= int(LB == L2 && LA == L1) * 1; // A-1, B-2 intersection, deduct points score2 -= int(LB == L2 && LA == L1) * 1; // A-1 and B-2 form a cross, deduct points
score2 -= int(LB == L4 && LA == L5) * 1; // A-5, B-4 intersection, deduct points score2 -= int(LB == L4 && LA == L5) * 1; // A-5 and B-4 form a cross, deduct points
score2 -= int(LA == L3) * 1; // A-3 forms an intersection, deduct points score2 -= int(LA == L3) * 1; // A-3 forms a cross, deduct points
// If the following triangular patterns are met, they can offset the intersection penalties // If the following triangular pattern is satisfied, offset the above cross deductions
score2 += int(LB == L1 && L1 == L2) * 1; // B-1-2 form a triangular pattern, add points score2 += int(LB == L1 && L1 == L2) * 1; // B-1-2 form a triangular pattern, add points
score2 += int(LB == L4 && L4 == L5) * 1; // B-4-5 form a triangular pattern, add points score2 += int(LB == L4 && L4 == L5) * 1; // B-4-5 form a triangular pattern, add points
score2 += int(L2 == L3 && L3 == L4) * 1; // 2-3-4 form a triangular pattern, add points score2 += int(L2 == L3 && L3 == L4) * 1; // 2-3-4 form a triangular pattern, add points
@ -848,45 +970,44 @@ bool countPatternMatches(uint LA, uint LB, uint L1, uint L2, uint L3, uint L4, u
score2 += scoreBonus + 6; score2 += scoreBonus + 6;
} }
// 3. Horizontal/vertical line pattern (Condition: horizontal continuity) uses a reward system, only pass if conditions are met // 3. Horizontal/vertical line pattern (Condition: horizontal continuity) uses a point addition system, passes only if conditions are met
if (LA == L2 || LB == L1 || LA == L4 || LB == L5 || (L1 == L2 && L2 == L3) || (L3 == L4 && L4 == L5)) { if (LA == L2 || LB == L1 || LA == L4 || LB == L5 || (L1 == L2 && L2 == L3) || (L3 == L4 && L4 == L5)) {
score3 += int(LA == L2); // A same as 2 +1 score3 += int(LA == L2); // A equals 2, +1
score3 += int(LB == L1); // B same as 1 +1 score3 += int(LB == L1); // B equals 1, +1
score3 += int(L3 == L4); // 3 same as 4 +1 score3 += int(L3 == L4); // 3 equals 4, +1
score3 += int(L4 == L5); // 4 same as 5 +1 score3 += int(L4 == L5); // 4 equals 5, +1
score3 += int(L3 == L4 && L4 == L5); // 3-4-5 continuous score3 += int(L3 == L4 && L4 == L5); // 3-4-5 continuous
score3 += int(LB == L5); // B same as 5 +1 score3 += int(LB == L5); // B equals 5, +1
score3 += int(LA == L4); // A same as 4 +1 score3 += int(LA == L4); // A equals 4, +1
score3 += int(L2 == L3); // 2 same as 3 +1 score3 += int(L2 == L3); // 2 equals 3, +1
score3 += int(L1 == L2); // 1 same as 2 +1 score3 += int(L1 == L2); // 1 equals 2, +1
score3 += int(L1 == L2 && L2 == L3); // 1-2-3 continuous score3 += int(L1 == L2 && L2 == L3); // 1-2-3 continuous
// A x 4 square // A x 4 square
score3 += int(LA == L2 && L2 == L3 && L3 == L4) * 2; score3 += int(LA == L2 && L2 == L3 && L3 == L4) * 2;
// Patch for the previous rule to avoid bubbles in large cross-shaped patterns. // Patch for the previous rule to avoid bubbles in large cross patterns. Some games use single-side patterns,
// Some games utilize single-sided patterns, so it's best to extend to bilateral judgment (WIP) // so it's best to expand for bilateral judgment (Work in Progress)
score3 -= int(LB == L1 && L1 == L5 && LA == L2 && L2 == L4)*3; score3 -= int(LB == L1 && L1 == L5 && LA == L2 && L2 == L4)*3;
score3 -= int(LA == L1 && LA == L5); // Deduct points if L1 and L5 are both A to avoid over-scoring score3 -= int(LA == L1 && LA == L5); // Deduct points if both L1 and L5 are A to avoid excessive scores
// and the pattern turning into a diagonal pattern. // and prevent the pattern from becoming a diagonal pattern.
// Bonus score // Extra points
score3 += scoreBonus; // Experience: Even for very similar colors, don't add too many points here, score3 += scoreBonus; // Experience: Even with very close colors, do not add too many points,
// as some Z-shaped intersections may produce bubbles // as some Z-shaped crosses may produce bubbles.
} }
// Take the maximum of the four scores // Take the maximum of the four scores
int score = max(max(score1, score2), score3); int score = max(max(score1, score2), score3);
return score < 6; // Need to reach 6 points return score < 6; // Requires 6 points to be satisfied
} }
#endif
void FilteredSampleFromVRAM(TEXPAGE_VALUE texpage, float2 coords, float4 uv_limits, out float4 texcol, out float ialpha) void FilteredSampleFromVRAM(TEXPAGE_VALUE texpage, float2 coords, float4 uv_limits, out float4 texcol, out float ialpha)
{ {
float2 bcoords = floor(coords); float2 bcoords = floor(coords);
uint A = src(-1, -1), B = src(+0, -1), C = src(+1, -1); uint A = src(-1, -1), B = src(+0, -1), C = src(+1, -1);
@ -895,79 +1016,74 @@ void FilteredSampleFromVRAM(TEXPAGE_VALUE texpage, float2 coords, float4 uv_limi
uint J = E, K = E, L = E, M = E; uint J = E, K = E, L = E, M = E;
if (((A ^ E) | (B ^ E) | (C ^ E) | (D ^ E) | (F ^ E) | (G ^ E) | (H ^ E) | (I ^ E)) != 0u) { // Explicitly initialize with the central pixel E by default
uint res = E;
ialpha = float(res != 0u);
texcol = unpackUnorm4x8(res);
if (((A ^ E) | (B ^ E) | (C ^ E) | (D ^ E) | (F ^ E) | (G ^ E) | (H ^ E) | (I ^ E)) == 0u) return;
uint P = src(+0, -2), S = src(+0, +2); uint P = src(+0, -2), S = src(+0, +2);
uint Q = src(-2, +0), R = src(+2, +0); uint Q = src(-2, +0), R = src(+2, +0);
uint Bl = luma(B), Dl = luma(D), El = luma(E), Fl = luma(F), Hl = luma(H); uint Bl = luma(B), Dl = luma(D), El = luma(E), Fl = luma(F), Hl = luma(H);
#if MMPX_ENHANCED
if ( // Check the cross state of every 4 pixels in a "field" shape, and pass five surrounding pixels for pattern judgment
// Check for "convex" patterns to avoid single-pixel spurs on long line edges if (A == E && B == D && A != B && countPatternMatches(A, B, C, F, I, H, G)) return;
(A == B && B == C && E == H && A != D && C != F && rgb_distance(D, F) < 0.2 && rgb_distance(B, E) > 0.6) || if (C == E && B == F && C != B && countPatternMatches(C, B, A, D, G, H, I)) return;
(A == D && D == G && E == F && A != B && G != H && rgb_distance(B, H) < 0.2 && rgb_distance(D, E) > 0.6) || if (G == E && D == H && G != H && countPatternMatches(G, H, I, F, C, B, A)) return;
(C == F && F == I && E == D && B != C && H != I && rgb_distance(B, H) < 0.2 && rgb_distance(E, F) > 0.6) || if (I == E && F == H && I != H && countPatternMatches(I, H, G, D, A, B, C)) return;
(G == H && H == I && B == E && D != G && F != I && rgb_distance(D, F) < 0.2 && rgb_distance(E, H) > 0.6) ||
// Check each 4-pixel intersection in a "grid" pattern and pass five surrounding pixels for pattern judgment
(A == E && B == D && A != B && countPatternMatches(A, B, C, F, I, H, G)) ||
(C == E && B == F && C != B && countPatternMatches(C, B, A, D, G, H, I)) ||
(G == E && D == H && G != H && countPatternMatches(G, H, I, F, C, B, A)) ||
(I == E && F == H && I != H && countPatternMatches(I, H, G, D, A, B, C))
) {
// Default to use center pixel E
// Fall through, J/K/L/M are already set to E. Setting the outputs here bugs out fxc.
} else {
#endif
// Original MMPX logic
// 1:1 slope rules // main mmpx logic
if ((D == B && D != H && D != F) && (El >= Dl || E == A) && any_eq3(E, A, C, G) && ((El < Dl) || A != D || E != P || E != Q)) J = D;
if ((B == F && B != D && B != H) && (El >= Bl || E == C) && any_eq3(E, A, C, I) && ((El < Bl) || C != B || E != P || E != R)) K = B;
if ((H == D && H != F && H != B) && (El >= Hl || E == G) && any_eq3(E, A, G, I) && ((El < Hl) || G != H || E != S || E != Q)) L = H;
if ((F == H && F != B && F != D) && (El >= Fl || E == I) && any_eq3(E, C, G, I) && ((El < Fl) || I != H || E != R || E != S)) M = F;
// Intersection rules // 1:1 slope rules
if ((E != F && all_eq4(E, C, I, D, Q) && all_eq2(F, B, H)) && (F != src(+3, +0))) K = M = F; if ((D == B && D != H && D != F) && (El >= Dl || E == A) && any_eq3(E, A, C, G) && ((El < Dl) || A != D || E != P || E != Q)) J = D;
if ((E != D && all_eq4(E, A, G, F, R) && all_eq2(D, B, H)) && (D != src(-3, +0))) J = L = D; if ((B == F && B != D && B != H) && (El >= Bl || E == C) && any_eq3(E, A, C, I) && ((El < Bl) || C != B || E != P || E != R)) K = B;
if ((E != H && all_eq4(E, G, I, B, P) && all_eq2(H, D, F)) && (H != src(+0, +3))) L = M = H; if ((H == D && H != F && H != B) && (El >= Hl || E == G) && any_eq3(E, A, G, I) && ((El < Hl) || G != H || E != S || E != Q)) L = H;
if ((E != B && all_eq4(E, A, C, H, S) && all_eq2(B, D, F)) && (B != src(+0, -3))) J = K = B; if ((F == H && F != B && F != D) && (El >= Fl || E == I) && any_eq3(E, C, G, I) && ((El < Fl) || I != H || E != R || E != S)) M = F;
if (Bl < El && all_eq4(E, G, H, I, S) && none_eq4(E, A, D, C, F)) J = K = B;
if (Hl < El && all_eq4(E, A, B, C, P) && none_eq4(E, D, G, I, F)) L = M = H;
if (Fl < El && all_eq4(E, A, D, G, Q) && none_eq4(E, B, C, I, H)) K = M = F;
if (Dl < El && all_eq4(E, C, F, I, R) && none_eq4(E, B, A, G, H)) J = L = D;
// 2:1 slope rules // Intersection rules
if (H != B) { if ((E != F && all_eq4(E, C, I, D, Q) && all_eq2(F, B, H)) && (F != src(+3, +0))) K = M = F;
if (H != A && H != E && H != C) { if ((E != D && all_eq4(E, A, G, F, R) && all_eq2(D, B, H)) && (D != src(-3, +0))) J = L = D;
if (all_eq3(H, G, F, R) && none_eq2(H, D, src(+2, -1))) L = M; if ((E != H && all_eq4(E, G, I, B, P) && all_eq2(H, D, F)) && (H != src(+0, +3))) L = M = H;
if (all_eq3(H, I, D, Q) && none_eq2(H, F, src(-2, -1))) M = L; if ((E != B && all_eq4(E, A, C, H, S) && all_eq2(B, D, F)) && (B != src(+0, -3))) J = K = B;
}
if (B != I && B != G && B != E) { // Use conditional weak blending instead of pixel copying to eliminate artifacts on straight lines
if (all_eq3(B, A, F, R) && none_eq2(B, D, src(+2, +1))) J = K; if (Bl < El && all_eq4(E, G, H, I, S) && none_eq4(E, A, D, C, F)) {J=admix2d(B,J); K=admix2d(B,K);}
if (all_eq3(B, C, D, Q) && none_eq2(B, F, src(-2, +1))) K = J; if (Hl < El && all_eq4(E, A, B, C, P) && none_eq4(E, D, G, I, F)) {L=admix2d(H,L); M=admix2d(H,M);}
} if (Fl < El && all_eq4(E, A, D, G, Q) && none_eq4(E, B, C, I, H)) {K=admix2d(F,K); M=admix2d(F,M);}
} // H !== B if (Dl < El && all_eq4(E, C, F, I, R) && none_eq4(E, B, A, G, H)) {J=admix2d(D,J); L=admix2d(D,L);}
if (F != D) { // 2:1 slope rules
if (D != I && D != E && D != C) { if (H != B) {
if (all_eq3(D, A, H, S) && none_eq2(D, B, src(+1, +2))) J = L; if (H != A && H != E && H != C) {
if (all_eq3(D, G, B, P) && none_eq2(D, H, src(+1, -2))) L = J; if (all_eq3(H, G, F, R) && none_eq2(H, D, src(+2, -1))) L = M;
} if (all_eq3(H, I, D, Q) && none_eq2(H, F, src(-2, -1))) M = L;
}
if (B != I && B != G && B != E) {
if (all_eq3(B, A, F, R) && none_eq2(B, D, src(+2, +1))) J = K;
if (all_eq3(B, C, D, Q) && none_eq2(B, F, src(-2, +1))) K = J;
}
} // H !== B
if (F != D) {
if (D != I && D != E && D != C) {
if (all_eq3(D, A, H, S) && none_eq2(D, B, src(+1, +2))) J = L;
if (all_eq3(D, G, B, P) && none_eq2(D, H, src(+1, -2))) L = J;
}
if (F != E && F != A && F != G) {
if (all_eq3(F, C, H, S) && none_eq2(F, B, src(-1, +2))) K = M;
if (all_eq3(F, I, B, P) && none_eq2(F, H, src(-1, -2))) M = K;
}
} // F !== D
if (F != E && F != A && F != G) {
if (all_eq3(F, C, H, S) && none_eq2(F, B, src(-1, +2))) K = M;
if (all_eq3(F, I, B, P) && none_eq2(F, H, src(-1, -2))) M = K;
}
} // F !== D
#if MMPX_ENHANCED
}
#endif
} // not constant
// select quadrant based on fractional part of texture coordinates // select quadrant based on fractional part of texture coordinates
float2 fpart = frac(coords); float2 fpart = frac(coords);
uint res = (fpart.x < 0.5f) ? ((fpart.y < 0.5f) ? J : L) : ((fpart.y < 0.5f) ? K : M); res = (fpart.x < 0.5f) ? ((fpart.y < 0.5f) ? J : L) : ((fpart.y < 0.5f) ? K : M);
ialpha = float(res != 0u); ialpha = float(res != 0u);
texcol = unpackUnorm4x8(res); texcol = unpackUnorm4x8(res);