Merge pull request #10398 from liamwhite/bcn

video_core: add ASTC recompression

externals/CMakeLists.txt

@@ -139,3 +139,7 @@ if (NOT TARGET LLVM::Demangle)
     target_sources(demangle PRIVATE demangle/ItaniumDemangle.cpp)
     add_library(LLVM::Demangle ALIAS demangle)
 endif()
+
+add_library(stb STATIC)
+target_include_directories(stb PUBLIC ./stb)
+target_sources(stb PRIVATE stb/stb_dxt.cpp)

externals/stb/stb_dxt.cpp

@@ -0,0 +1,765 @@
+// SPDX-FileCopyrightText: fabian "ryg" giesen
+// SPDX-License-Identifier: MIT
+
+// stb_dxt.h - v1.12 - DXT1/DXT5 compressor
+
+#include <stb_dxt.h>
+
+#include <stdlib.h>
+#include <string.h>
+
+#if !defined(STBD_FABS)
+#include <math.h>
+#endif
+
+#ifndef STBD_FABS
+#define STBD_FABS(x) fabs(x)
+#endif
+
+static const unsigned char stb__OMatch5[256][2] = {
+    {0, 0},   {0, 0},   {0, 1},   {0, 1},   {1, 0},   {1, 0},   {1, 0},   {1, 1},   {1, 1},
+    {1, 1},   {1, 2},   {0, 4},   {2, 1},   {2, 1},   {2, 1},   {2, 2},   {2, 2},   {2, 2},
+    {2, 3},   {1, 5},   {3, 2},   {3, 2},   {4, 0},   {3, 3},   {3, 3},   {3, 3},   {3, 4},
+    {3, 4},   {3, 4},   {3, 5},   {4, 3},   {4, 3},   {5, 2},   {4, 4},   {4, 4},   {4, 5},
+    {4, 5},   {5, 4},   {5, 4},   {5, 4},   {6, 3},   {5, 5},   {5, 5},   {5, 6},   {4, 8},
+    {6, 5},   {6, 5},   {6, 5},   {6, 6},   {6, 6},   {6, 6},   {6, 7},   {5, 9},   {7, 6},
+    {7, 6},   {8, 4},   {7, 7},   {7, 7},   {7, 7},   {7, 8},   {7, 8},   {7, 8},   {7, 9},
+    {8, 7},   {8, 7},   {9, 6},   {8, 8},   {8, 8},   {8, 9},   {8, 9},   {9, 8},   {9, 8},
+    {9, 8},   {10, 7},  {9, 9},   {9, 9},   {9, 10},  {8, 12},  {10, 9},  {10, 9},  {10, 9},
+    {10, 10}, {10, 10}, {10, 10}, {10, 11}, {9, 13},  {11, 10}, {11, 10}, {12, 8},  {11, 11},
+    {11, 11}, {11, 11}, {11, 12}, {11, 12}, {11, 12}, {11, 13}, {12, 11}, {12, 11}, {13, 10},
+    {12, 12}, {12, 12}, {12, 13}, {12, 13}, {13, 12}, {13, 12}, {13, 12}, {14, 11}, {13, 13},
+    {13, 13}, {13, 14}, {12, 16}, {14, 13}, {14, 13}, {14, 13}, {14, 14}, {14, 14}, {14, 14},
+    {14, 15}, {13, 17}, {15, 14}, {15, 14}, {16, 12}, {15, 15}, {15, 15}, {15, 15}, {15, 16},
+    {15, 16}, {15, 16}, {15, 17}, {16, 15}, {16, 15}, {17, 14}, {16, 16}, {16, 16}, {16, 17},
+    {16, 17}, {17, 16}, {17, 16}, {17, 16}, {18, 15}, {17, 17}, {17, 17}, {17, 18}, {16, 20},
+    {18, 17}, {18, 17}, {18, 17}, {18, 18}, {18, 18}, {18, 18}, {18, 19}, {17, 21}, {19, 18},
+    {19, 18}, {20, 16}, {19, 19}, {19, 19}, {19, 19}, {19, 20}, {19, 20}, {19, 20}, {19, 21},
+    {20, 19}, {20, 19}, {21, 18}, {20, 20}, {20, 20}, {20, 21}, {20, 21}, {21, 20}, {21, 20},
+    {21, 20}, {22, 19}, {21, 21}, {21, 21}, {21, 22}, {20, 24}, {22, 21}, {22, 21}, {22, 21},
+    {22, 22}, {22, 22}, {22, 22}, {22, 23}, {21, 25}, {23, 22}, {23, 22}, {24, 20}, {23, 23},
+    {23, 23}, {23, 23}, {23, 24}, {23, 24}, {23, 24}, {23, 25}, {24, 23}, {24, 23}, {25, 22},
+    {24, 24}, {24, 24}, {24, 25}, {24, 25}, {25, 24}, {25, 24}, {25, 24}, {26, 23}, {25, 25},
+    {25, 25}, {25, 26}, {24, 28}, {26, 25}, {26, 25}, {26, 25}, {26, 26}, {26, 26}, {26, 26},
+    {26, 27}, {25, 29}, {27, 26}, {27, 26}, {28, 24}, {27, 27}, {27, 27}, {27, 27}, {27, 28},
+    {27, 28}, {27, 28}, {27, 29}, {28, 27}, {28, 27}, {29, 26}, {28, 28}, {28, 28}, {28, 29},
+    {28, 29}, {29, 28}, {29, 28}, {29, 28}, {30, 27}, {29, 29}, {29, 29}, {29, 30}, {29, 30},
+    {30, 29}, {30, 29}, {30, 29}, {30, 30}, {30, 30}, {30, 30}, {30, 31}, {30, 31}, {31, 30},
+    {31, 30}, {31, 30}, {31, 31}, {31, 31},
+};
+static const unsigned char stb__OMatch6[256][2] = {
+    {0, 0},   {0, 1},   {1, 0},   {1, 1},   {1, 1},   {1, 2},   {2, 1},   {2, 2},   {2, 2},
+    {2, 3},   {3, 2},   {3, 3},   {3, 3},   {3, 4},   {4, 3},   {4, 4},   {4, 4},   {4, 5},
+    {5, 4},   {5, 5},   {5, 5},   {5, 6},   {6, 5},   {6, 6},   {6, 6},   {6, 7},   {7, 6},
+    {7, 7},   {7, 7},   {7, 8},   {8, 7},   {8, 8},   {8, 8},   {8, 9},   {9, 8},   {9, 9},
+    {9, 9},   {9, 10},  {10, 9},  {10, 10}, {10, 10}, {10, 11}, {11, 10}, {8, 16},  {11, 11},
+    {11, 12}, {12, 11}, {9, 17},  {12, 12}, {12, 13}, {13, 12}, {11, 16}, {13, 13}, {13, 14},
+    {14, 13}, {12, 17}, {14, 14}, {14, 15}, {15, 14}, {14, 16}, {15, 15}, {15, 16}, {16, 14},
+    {16, 15}, {17, 14}, {16, 16}, {16, 17}, {17, 16}, {18, 15}, {17, 17}, {17, 18}, {18, 17},
+    {20, 14}, {18, 18}, {18, 19}, {19, 18}, {21, 15}, {19, 19}, {19, 20}, {20, 19}, {20, 20},
+    {20, 20}, {20, 21}, {21, 20}, {21, 21}, {21, 21}, {21, 22}, {22, 21}, {22, 22}, {22, 22},
+    {22, 23}, {23, 22}, {23, 23}, {23, 23}, {23, 24}, {24, 23}, {24, 24}, {24, 24}, {24, 25},
+    {25, 24}, {25, 25}, {25, 25}, {25, 26}, {26, 25}, {26, 26}, {26, 26}, {26, 27}, {27, 26},
+    {24, 32}, {27, 27}, {27, 28}, {28, 27}, {25, 33}, {28, 28}, {28, 29}, {29, 28}, {27, 32},
+    {29, 29}, {29, 30}, {30, 29}, {28, 33}, {30, 30}, {30, 31}, {31, 30}, {30, 32}, {31, 31},
+    {31, 32}, {32, 30}, {32, 31}, {33, 30}, {32, 32}, {32, 33}, {33, 32}, {34, 31}, {33, 33},
+    {33, 34}, {34, 33}, {36, 30}, {34, 34}, {34, 35}, {35, 34}, {37, 31}, {35, 35}, {35, 36},
+    {36, 35}, {36, 36}, {36, 36}, {36, 37}, {37, 36}, {37, 37}, {37, 37}, {37, 38}, {38, 37},
+    {38, 38}, {38, 38}, {38, 39}, {39, 38}, {39, 39}, {39, 39}, {39, 40}, {40, 39}, {40, 40},
+    {40, 40}, {40, 41}, {41, 40}, {41, 41}, {41, 41}, {41, 42}, {42, 41}, {42, 42}, {42, 42},
+    {42, 43}, {43, 42}, {40, 48}, {43, 43}, {43, 44}, {44, 43}, {41, 49}, {44, 44}, {44, 45},
+    {45, 44}, {43, 48}, {45, 45}, {45, 46}, {46, 45}, {44, 49}, {46, 46}, {46, 47}, {47, 46},
+    {46, 48}, {47, 47}, {47, 48}, {48, 46}, {48, 47}, {49, 46}, {48, 48}, {48, 49}, {49, 48},
+    {50, 47}, {49, 49}, {49, 50}, {50, 49}, {52, 46}, {50, 50}, {50, 51}, {51, 50}, {53, 47},
+    {51, 51}, {51, 52}, {52, 51}, {52, 52}, {52, 52}, {52, 53}, {53, 52}, {53, 53}, {53, 53},
+    {53, 54}, {54, 53}, {54, 54}, {54, 54}, {54, 55}, {55, 54}, {55, 55}, {55, 55}, {55, 56},
+    {56, 55}, {56, 56}, {56, 56}, {56, 57}, {57, 56}, {57, 57}, {57, 57}, {57, 58}, {58, 57},
+    {58, 58}, {58, 58}, {58, 59}, {59, 58}, {59, 59}, {59, 59}, {59, 60}, {60, 59}, {60, 60},
+    {60, 60}, {60, 61}, {61, 60}, {61, 61}, {61, 61}, {61, 62}, {62, 61}, {62, 62}, {62, 62},
+    {62, 63}, {63, 62}, {63, 63}, {63, 63},
+};
+
+static int stb__Mul8Bit(int a, int b) {
+    int t = a * b + 128;
+    return (t + (t >> 8)) >> 8;
+}
+
+static void stb__From16Bit(unsigned char* out, unsigned short v) {
+    int rv = (v & 0xf800) >> 11;
+    int gv = (v & 0x07e0) >> 5;
+    int bv = (v & 0x001f) >> 0;
+
+    // expand to 8 bits via bit replication
+    out[0] = static_cast<unsigned char>((rv * 33) >> 2);
+    out[1] = static_cast<unsigned char>((gv * 65) >> 4);
+    out[2] = static_cast<unsigned char>((bv * 33) >> 2);
+    out[3] = 0;
+}
+
+static unsigned short stb__As16Bit(int r, int g, int b) {
+    return (unsigned short)((stb__Mul8Bit(r, 31) << 11) + (stb__Mul8Bit(g, 63) << 5) +
+                            stb__Mul8Bit(b, 31));
+}
+
+// linear interpolation at 1/3 point between a and b, using desired rounding
+// type
+static int stb__Lerp13(int a, int b) {
+#ifdef STB_DXT_USE_ROUNDING_BIAS
+    // with rounding bias
+    return a + stb__Mul8Bit(b - a, 0x55);
+#else
+    // without rounding bias
+    // replace "/ 3" by "* 0xaaab) >> 17" if your compiler sucks or you really
+    // need every ounce of speed.
+    return (2 * a + b) / 3;
+#endif
+}
+
+// linear interpolation at 1/2 point between a and b
+static int stb__Lerp12(int a, int b) {
+    return (a + b) / 2;
+}
+
+// lerp RGB color
+static void stb__Lerp13RGB(unsigned char* out, unsigned char* p1, unsigned char* p2) {
+    out[0] = (unsigned char)stb__Lerp13(p1[0], p2[0]);
+    out[1] = (unsigned char)stb__Lerp13(p1[1], p2[1]);
+    out[2] = (unsigned char)stb__Lerp13(p1[2], p2[2]);
+}
+
+static void stb__Lerp12RGB(unsigned char* out, unsigned char* p1, unsigned char* p2) {
+    out[0] = (unsigned char)stb__Lerp12(p1[0], p2[0]);
+    out[1] = (unsigned char)stb__Lerp12(p1[1], p2[1]);
+    out[2] = (unsigned char)stb__Lerp12(p1[2], p2[2]);
+}
+
+/****************************************************************************/
+
+static void stb__Eval4Colors(unsigned char* color, unsigned short c0, unsigned short c1) {
+    stb__From16Bit(color + 0, c0);
+    stb__From16Bit(color + 4, c1);
+    stb__Lerp13RGB(color + 8, color + 0, color + 4);
+    stb__Lerp13RGB(color + 12, color + 4, color + 0);
+}
+
+static void stb__Eval3Colors(unsigned char* color, unsigned short c0, unsigned short c1) {
+    stb__From16Bit(color + 0, c0);
+    stb__From16Bit(color + 4, c1);
+    stb__Lerp12RGB(color + 8, color + 0, color + 4);
+}
+
+// The color matching function
+static unsigned int stb__MatchColorsBlock(unsigned char* block, unsigned char* color) {
+    unsigned int mask = 0;
+    int dirr = color[0 * 4 + 0] - color[1 * 4 + 0];
+    int dirg = color[0 * 4 + 1] - color[1 * 4 + 1];
+    int dirb = color[0 * 4 + 2] - color[1 * 4 + 2];
+    int dots[16];
+    int stops[4];
+    int i;
+    int c0Point, halfPoint, c3Point;
+
+    for (i = 0; i < 16; i++)
+        dots[i] = block[i * 4 + 0] * dirr + block[i * 4 + 1] * dirg + block[i * 4 + 2] * dirb;
+
+    for (i = 0; i < 4; i++)
+        stops[i] = color[i * 4 + 0] * dirr + color[i * 4 + 1] * dirg + color[i * 4 + 2] * dirb;
+
+    // think of the colors as arranged on a line; project point onto that line,
+    // then choose next color out of available ones. we compute the crossover
+    // points for "best color in top half"/"best in bottom half" and then the same
+    // inside that subinterval.
+    //
+    // relying on this 1d approximation isn't always optimal in terms of euclidean
+    // distance, but it's very close and a lot faster.
+    // http://cbloomrants.blogspot.com/2008/12/12-08-08-dxtc-summary.html
+
+    c0Point = (stops[1] + stops[3]);
+    halfPoint = (stops[3] + stops[2]);
+    c3Point = (stops[2] + stops[0]);
+
+    for (i = 15; i >= 0; i--) {
+        int dot = dots[i] * 2;
+        mask <<= 2;
+
+        if (dot < halfPoint)
+            mask |= (dot < c0Point) ? 1 : 3;
+        else
+            mask |= (dot < c3Point) ? 2 : 0;
+    }
+
+    return mask;
+}
+
+static unsigned int stb__MatchColorsAlphaBlock(unsigned char* block, unsigned char* color) {
+    unsigned int mask = 0;
+    int dirr = color[0 * 4 + 0] - color[1 * 4 + 0];
+    int dirg = color[0 * 4 + 1] - color[1 * 4 + 1];
+    int dirb = color[0 * 4 + 2] - color[1 * 4 + 2];
+    int dots[16];
+    int stops[3];
+    int i;
+    int c0Point, c2Point;
+
+    for (i = 0; i < 16; i++)
+        dots[i] = block[i * 4 + 0] * dirr + block[i * 4 + 1] * dirg + block[i * 4 + 2] * dirb;
+
+    for (i = 0; i < 3; i++)
+        stops[i] = color[i * 4 + 0] * dirr + color[i * 4 + 1] * dirg + color[i * 4 + 2] * dirb;
+
+    c0Point = (stops[1] + stops[2]);
+    c2Point = (stops[2] + stops[0]);
+
+    for (i = 15; i >= 0; i--) {
+        int dot = dots[i] * 2;
+        mask <<= 2;
+
+        if (block[i * 4 + 3] == 0)
+            mask |= 3;
+        else if (dot < c2Point)
+            mask |= (dot < c0Point) ? 0 : 2;
+        else
+            mask |= (dot < c0Point) ? 1 : 0;
+    }
+
+    return mask;
+}
+
+static void stb__ReorderColors(unsigned short* pmax16, unsigned short* pmin16) {
+    if (*pmin16 < *pmax16) {
+        unsigned short t = *pmin16;
+        *pmin16 = *pmax16;
+        *pmax16 = t;
+    }
+}
+
+static void stb__FinalizeColors(unsigned short* pmax16, unsigned short* pmin16,
+                                unsigned int* pmask) {
+    if (*pmax16 < *pmin16) {
+        unsigned short t = *pmin16;
+        *pmin16 = *pmax16;
+        *pmax16 = t;
+        *pmask ^= 0x55555555;
+    }
+}
+
+// The color optimization function. (Clever code, part 1)
+static void stb__OptimizeColorsBlock(unsigned char* block, unsigned short* pmax16,
+                                     unsigned short* pmin16) {
+    int mind, maxd;
+    unsigned char *minp, *maxp;
+    double magn;
+    int v_r, v_g, v_b;
+    static const int nIterPower = 4;
+    float covf[6], vfr, vfg, vfb;
+
+    // determine color distribution
+    int cov[6];
+    int mu[3], min[3], max[3];
+    int ch, i, iter;
+
+    for (ch = 0; ch < 3; ch++) {
+        const unsigned char* bp = ((const unsigned char*)block) + ch;
+        int muv, minv, maxv;
+
+        muv = minv = maxv = bp[0];
+        for (i = 4; i < 64; i += 4) {
+            muv += bp[i];
+            if (bp[i] < minv)
+                minv = bp[i];
+            else if (bp[i] > maxv)
+                maxv = bp[i];
+        }
+
+        mu[ch] = (muv + 8) >> 4;
+        min[ch] = minv;
+        max[ch] = maxv;
+    }
+
+    // determine covariance matrix
+    for (i = 0; i < 6; i++)
+        cov[i] = 0;
+
+    for (i = 0; i < 16; i++) {
+        int r = block[i * 4 + 0] - mu[0];
+        int g = block[i * 4 + 1] - mu[1];
+        int b = block[i * 4 + 2] - mu[2];
+
+        cov[0] += r * r;
+        cov[1] += r * g;
+        cov[2] += r * b;
+        cov[3] += g * g;
+        cov[4] += g * b;
+        cov[5] += b * b;
+    }
+
+    // convert covariance matrix to float, find principal axis via power iter
+    for (i = 0; i < 6; i++)
+        covf[i] = static_cast<float>(cov[i]) / 255.0f;
+
+    vfr = (float)(max[0] - min[0]);
+    vfg = (float)(max[1] - min[1]);
+    vfb = (float)(max[2] - min[2]);
+
+    for (iter = 0; iter < nIterPower; iter++) {
+        float r = vfr * covf[0] + vfg * covf[1] + vfb * covf[2];
+        float g = vfr * covf[1] + vfg * covf[3] + vfb * covf[4];
+        float b = vfr * covf[2] + vfg * covf[4] + vfb * covf[5];
+
+        vfr = r;
+        vfg = g;
+        vfb = b;
+    }
+
+    magn = STBD_FABS(vfr);
+    if (STBD_FABS(vfg) > magn)
+        magn = STBD_FABS(vfg);
+    if (STBD_FABS(vfb) > magn)
+        magn = STBD_FABS(vfb);
+
+    if (magn < 4.0f) { // too small, default to luminance
+        v_r = 299;     // JPEG YCbCr luma coefs, scaled by 1000.
+        v_g = 587;
+        v_b = 114;
+    } else {
+        magn = 512.0 / magn;
+        v_r = (int)(vfr * magn);
+        v_g = (int)(vfg * magn);
+        v_b = (int)(vfb * magn);
+    }
+
+    minp = maxp = block;
+    mind = maxd = block[0] * v_r + block[1] * v_g + block[2] * v_b;
+    // Pick colors at extreme points
+    for (i = 1; i < 16; i++) {
+        int dot = block[i * 4 + 0] * v_r + block[i * 4 + 1] * v_g + block[i * 4 + 2] * v_b;
+
+        if (dot < mind) {
+            mind = dot;
+            minp = block + i * 4;
+        }
+
+        if (dot > maxd) {
+            maxd = dot;
+            maxp = block + i * 4;
+        }
+    }
+
+    *pmax16 = stb__As16Bit(maxp[0], maxp[1], maxp[2]);
+    *pmin16 = stb__As16Bit(minp[0], minp[1], minp[2]);
+    stb__ReorderColors(pmax16, pmin16);
+}
+
+static void stb__OptimizeColorsAlphaBlock(unsigned char* block, unsigned short* pmax16,
+                                          unsigned short* pmin16) {
+    int mind, maxd;
+    unsigned char *minp, *maxp;
+    double magn;
+    int v_r, v_g, v_b;
+    static const int nIterPower = 4;
+    float covf[6], vfr, vfg, vfb;
+
+    // determine color distribution
+    int cov[6];
+    int mu[3], min[3], max[3];
+    int ch, i, iter;
+
+    for (ch = 0; ch < 3; ch++) {
+        const unsigned char* bp = ((const unsigned char*)block) + ch;
+        int muv = 0, minv = 256, maxv = -1;
+        int num = 0;
+
+        for (i = 0; i < 64; i += 4) {
+            if (bp[3 - ch] == 0) {
+                continue;
+            }
+
+            muv += bp[i];
+            if (bp[i] < minv)
+                minv = bp[i];
+            else if (bp[i] > maxv)
+                maxv = bp[i];
+
+            num++;
+        }
+
+        mu[ch] = num > 0 ? (muv + 8) / num : 0;
+        min[ch] = minv;
+        max[ch] = maxv;
+    }
+
+    // determine covariance matrix
+    for (i = 0; i < 6; i++)
+        cov[i] = 0;
+
+    for (i = 0; i < 16; i++) {
+        if (block[i * 4 + 3] == 0) {
+            continue;
+        }
+
+        int r = block[i * 4 + 0] - mu[0];
+        int g = block[i * 4 + 1] - mu[1];
+        int b = block[i * 4 + 2] - mu[2];
+
+        cov[0] += r * r;
+        cov[1] += r * g;
+        cov[2] += r * b;
+        cov[3] += g * g;
+        cov[4] += g * b;
+        cov[5] += b * b;
+    }
+
+    // convert covariance matrix to float, find principal axis via power iter
+    for (i = 0; i < 6; i++)
+        covf[i] = static_cast<float>(cov[i]) / 255.0f;
+
+    vfr = (float)(max[0] - min[0]);
+    vfg = (float)(max[1] - min[1]);
+    vfb = (float)(max[2] - min[2]);
+
+    for (iter = 0; iter < nIterPower; iter++) {
+        float r = vfr * covf[0] + vfg * covf[1] + vfb * covf[2];
+        float g = vfr * covf[1] + vfg * covf[3] + vfb * covf[4];
+        float b = vfr * covf[2] + vfg * covf[4] + vfb * covf[5];
+
+        vfr = r;
+        vfg = g;
+        vfb = b;
+    }
+
+    magn = STBD_FABS(vfr);
+    if (STBD_FABS(vfg) > magn)
+        magn = STBD_FABS(vfg);
+    if (STBD_FABS(vfb) > magn)
+        magn = STBD_FABS(vfb);
+
+    if (magn < 4.0f) { // too small, default to luminance
+        v_r = 299;     // JPEG YCbCr luma coefs, scaled by 1000.
+        v_g = 587;
+        v_b = 114;
+    } else {
+        magn = 512.0 / magn;
+        v_r = (int)(vfr * magn);
+        v_g = (int)(vfg * magn);
+        v_b = (int)(vfb * magn);
+    }
+
+    minp = maxp = NULL;
+    mind = 0x7fffffff;
+    maxd = -0x80000000;
+
+    // Pick colors at extreme points
+    for (i = 0; i < 16; i++) {
+        if (block[i * 4 + 3] == 0) {
+            continue;
+        }
+
+        int dot = block[i * 4 + 0] * v_r + block[i * 4 + 1] * v_g + block[i * 4 + 2] * v_b;
+
+        if (dot < mind) {
+            mind = dot;
+            minp = block + i * 4;
+        }
+
+        if (dot > maxd) {
+            maxd = dot;
+            maxp = block + i * 4;
+        }
+    }
+
+    if (!maxp) {
+        // all alpha, no color
+        *pmin16 = 0xffff;
+        *pmax16 = 0;
+    } else {
+        // endpoint colors found
+        *pmax16 = stb__As16Bit(maxp[0], maxp[1], maxp[2]);
+        *pmin16 = stb__As16Bit(minp[0], minp[1], minp[2]);
+
+        if (*pmax16 == *pmin16) {
+            // modify the endpoints to indicate presence of an alpha block
+            if (*pmax16 > 0) {
+                (*pmax16)--;
+            } else {
+                (*pmin16)++;
+            }
+        }
+
+        stb__ReorderColors(pmax16, pmin16);
+    }
+}
+
+static const float stb__midpoints5[32] = {
+    0.015686f, 0.047059f, 0.078431f, 0.111765f, 0.145098f, 0.176471f, 0.207843f, 0.241176f,
+    0.274510f, 0.305882f, 0.337255f, 0.370588f, 0.403922f, 0.435294f, 0.466667f, 0.5f,
+    0.533333f, 0.564706f, 0.596078f, 0.629412f, 0.662745f, 0.694118f, 0.725490f, 0.758824f,
+    0.792157f, 0.823529f, 0.854902f, 0.888235f, 0.921569f, 0.952941f, 0.984314f, 1.0f};
+
+static const float stb__midpoints6[64] = {
+    0.007843f, 0.023529f, 0.039216f, 0.054902f, 0.070588f, 0.086275f, 0.101961f, 0.117647f,
+    0.133333f, 0.149020f, 0.164706f, 0.180392f, 0.196078f, 0.211765f, 0.227451f, 0.245098f,
+    0.262745f, 0.278431f, 0.294118f, 0.309804f, 0.325490f, 0.341176f, 0.356863f, 0.372549f,
+    0.388235f, 0.403922f, 0.419608f, 0.435294f, 0.450980f, 0.466667f, 0.482353f, 0.500000f,
+    0.517647f, 0.533333f, 0.549020f, 0.564706f, 0.580392f, 0.596078f, 0.611765f, 0.627451f,
+    0.643137f, 0.658824f, 0.674510f, 0.690196f, 0.705882f, 0.721569f, 0.737255f, 0.754902f,
+    0.772549f, 0.788235f, 0.803922f, 0.819608f, 0.835294f, 0.850980f, 0.866667f, 0.882353f,
+    0.898039f, 0.913725f, 0.929412f, 0.945098f, 0.960784f, 0.976471f, 0.992157f, 1.0f};
+
+static unsigned short stb__Quantize5(float x) {
+    unsigned short q;
+    x = x < 0 ? 0 : x > 1 ? 1 : x; // saturate
+    q = (unsigned short)(x * 31);
+    q += (x > stb__midpoints5[q]);
+    return q;
+}
+
+static unsigned short stb__Quantize6(float x) {
+    unsigned short q;
+    x = x < 0 ? 0 : x > 1 ? 1 : x; // saturate
+    q = (unsigned short)(x * 63);
+    q += (x > stb__midpoints6[q]);
+    return q;
+}
+
+// The refinement function. (Clever code, part 2)
+// Tries to optimize colors to suit block contents better.
+// (By solving a least squares system via normal equations+Cramer's rule)
+static int stb__RefineBlock(unsigned char* block, unsigned short* pmax16, unsigned short* pmin16,
+                            unsigned int mask) {
+    static const int w1Tab[4] = {3, 0, 2, 1};
+    static const int prods[4] = {0x090000, 0x000900, 0x040102, 0x010402};
+    // ^some magic to save a lot of multiplies in the accumulating loop...
+    // (precomputed products of weights for least squares system, accumulated
+    // inside one 32-bit register)
+
+    float f;
+    unsigned short oldMin, oldMax, min16, max16;
+    int i, akku = 0, xx, xy, yy;
+    int At1_r, At1_g, At1_b;
+    int At2_r, At2_g, At2_b;
+    unsigned int cm = mask;
+
+    oldMin = *pmin16;
+    oldMax = *pmax16;
+
+    if ((mask ^ (mask << 2)) < 4) // all pixels have the same index?
+    {
+        // yes, linear system would be singular; solve using optimal
+        // single-color match on average color
+        int r = 8, g = 8, b = 8;
+        for (i = 0; i < 16; ++i) {
+            r += block[i * 4 + 0];
+            g += block[i * 4 + 1];
+            b += block[i * 4 + 2];
+        }
+
+        r >>= 4;
+        g >>= 4;
+        b >>= 4;
+
+        max16 = static_cast<unsigned short>((stb__OMatch5[r][0] << 11) | (stb__OMatch6[g][0] << 5) |
+                                            stb__OMatch5[b][0]);
+        min16 = static_cast<unsigned short>((stb__OMatch5[r][1] << 11) | (stb__OMatch6[g][1] << 5) |
+                                            stb__OMatch5[b][1]);
+    } else {
+        At1_r = At1_g = At1_b = 0;
+        At2_r = At2_g = At2_b = 0;
+        for (i = 0; i < 16; ++i, cm >>= 2) {
+            int step = cm & 3;
+            int w1 = w1Tab[step];
+            int r = block[i * 4 + 0];
+            int g = block[i * 4 + 1];
+            int b = block[i * 4 + 2];
+
+            akku += prods[step];
+            At1_r += w1 * r;
+            At1_g += w1 * g;
+            At1_b += w1 * b;
+            At2_r += r;
+            At2_g += g;
+            At2_b += b;
+        }
+
+        At2_r = 3 * At2_r - At1_r;
+        At2_g = 3 * At2_g - At1_g;
+        At2_b = 3 * At2_b - At1_b;
+
+        // extract solutions and decide solvability
+        xx = akku >> 16;
+        yy = (akku >> 8) & 0xff;
+        xy = (akku >> 0) & 0xff;
+
+        f = 3.0f / 255.0f / static_cast<float>(xx * yy - xy * xy);
+
+        max16 = static_cast<unsigned short>(
+            stb__Quantize5(static_cast<float>(At1_r * yy - At2_r * xy) * f) << 11);
+        max16 |= static_cast<unsigned short>(
+            stb__Quantize6(static_cast<float>(At1_g * yy - At2_g * xy) * f) << 5);
+        max16 |= static_cast<unsigned short>(
+            stb__Quantize5(static_cast<float>(At1_b * yy - At2_b * xy) * f) << 0);
+
+        min16 = static_cast<unsigned short>(
+            stb__Quantize5(static_cast<float>(At2_r * xx - At1_r * xy) * f) << 11);
+        min16 |= static_cast<unsigned short>(
+            stb__Quantize6(static_cast<float>(At2_g * xx - At1_g * xy) * f) << 5);
+        min16 |= static_cast<unsigned short>(
+            stb__Quantize5(static_cast<float>(At2_b * xx - At1_b * xy) * f) << 0);
+    }
+
+    *pmin16 = min16;
+    *pmax16 = max16;
+    stb__ReorderColors(pmax16, pmin16);
+
+    return oldMin != min16 || oldMax != max16;
+}
+
+// Color block compression
+static void stb__CompressColorBlock(unsigned char* dest, unsigned char* block, int alpha,
+                                    int mode) {
+    unsigned int mask;
+    int i;
+    int refinecount;
+    unsigned short max16, min16;
+    unsigned char color[4 * 4];
+
+    refinecount = (mode & STB_DXT_HIGHQUAL) ? 2 : 1;
+
+    // check if block is constant
+    for (i = 1; i < 16; i++)
+        if (((unsigned int*)block)[i] != ((unsigned int*)block)[0])
+            break;
+
+    if (i == 16 && block[3] == 0 && alpha) { // constant alpha
+        mask = 0xffffffff;
+        max16 = 0;
+        min16 = 0xffff;
+    } else if (i == 16) { // constant color
+        int r = block[0], g = block[1], b = block[2];
+        mask = 0xaaaaaaaa;
+        max16 = static_cast<unsigned short>((stb__OMatch5[r][0] << 11) | (stb__OMatch6[g][0] << 5) |
+                                            stb__OMatch5[b][0]);
+        min16 = static_cast<unsigned short>((stb__OMatch5[r][1] << 11) | (stb__OMatch6[g][1] << 5) |
+                                            stb__OMatch5[b][1]);
+    } else if (alpha) {
+        stb__OptimizeColorsAlphaBlock(block, &max16, &min16);
+        stb__Eval3Colors(color, max16, min16);
+        mask = stb__MatchColorsAlphaBlock(block, color);
+    } else {
+        // first step: PCA+map along principal axis
+        stb__OptimizeColorsBlock(block, &max16, &min16);
+        if (max16 != min16) {
+            stb__Eval4Colors(color, max16, min16);
+            mask = stb__MatchColorsBlock(block, color);
+        } else
+            mask = 0;
+
+        // third step: refine (multiple times if requested)
+        for (i = 0; i < refinecount; i++) {
+            unsigned int lastmask = mask;
+
+            if (stb__RefineBlock(block, &max16, &min16, mask)) {
+                if (max16 != min16) {
+                    stb__Eval4Colors(color, max16, min16);
+                    mask = stb__MatchColorsBlock(block, color);
+                } else {
+                    mask = 0;
+                    break;
+                }
+            }
+
+            if (mask == lastmask)
+                break;
+        }
+    }
+
+    // write the color block
+    if (!alpha)
+        stb__FinalizeColors(&max16, &min16, &mask);
+
+    dest[0] = (unsigned char)(max16);
+    dest[1] = (unsigned char)(max16 >> 8);
+    dest[2] = (unsigned char)(min16);
+    dest[3] = (unsigned char)(min16 >> 8);
+    dest[4] = (unsigned char)(mask);
+    dest[5] = (unsigned char)(mask >> 8);
+    dest[6] = (unsigned char)(mask >> 16);
+    dest[7] = (unsigned char)(mask >> 24);
+}
+
+// Alpha block compression (this is easy for a change)
+static void stb__CompressAlphaBlock(unsigned char* dest, unsigned char* src, int stride) {
+    int i, dist, bias, dist4, dist2, bits, mask;
+
+    // find min/max color
+    int mn, mx;
+    mn = mx = src[0];
+
+    for (i = 1; i < 16; i++) {
+        if (src[i * stride] < mn)
+            mn = src[i * stride];
+        else if (src[i * stride] > mx)
+            mx = src[i * stride];
+    }
+
+    // encode them
+    dest[0] = (unsigned char)mx;
+    dest[1] = (unsigned char)mn;
+    dest += 2;
+
+    // determine bias and emit color indices
+    // given the choice of mx/mn, these indices are optimal:
+    // http://fgiesen.wordpress.com/2009/12/15/dxt5-alpha-block-index-determination/
+    dist = mx - mn;
+    dist4 = dist * 4;
+    dist2 = dist * 2;
+    bias = (dist < 8) ? (dist - 1) : (dist / 2 + 2);
+    bias -= mn * 7;
+    bits = 0, mask = 0;
+
+    for (i = 0; i < 16; i++) {
+        int a = src[i * stride] * 7 + bias;
+        int ind, t;
+
+        // select index. this is a "linear scale" lerp factor between 0 (val=min)
+        // and 7 (val=max).
+        t = (a >= dist4) ? -1 : 0;
+        ind = t & 4;
+        a -= dist4 & t;
+        t = (a >= dist2) ? -1 : 0;
+        ind += t & 2;
+        a -= dist2 & t;
+        ind += (a >= dist);
+
+        // turn linear scale into DXT index (0/1 are extremal pts)
+        ind = -ind & 7;
+        ind ^= (2 > ind);
+
+        // write index
+        mask |= ind << bits;
+        if ((bits += 3) >= 8) {
+            *dest++ = (unsigned char)mask;
+            mask >>= 8;
+            bits -= 8;
+        }
+    }
+}
+
+void stb_compress_bc1_block(unsigned char* dest, const unsigned char* src, int alpha, int mode) {
+    stb__CompressColorBlock(dest, (unsigned char*)src, alpha, mode);
+}
+
+void stb_compress_bc3_block(unsigned char* dest, const unsigned char* src, int mode) {
+    unsigned char data[16][4];
+    int i;
+
+    stb__CompressAlphaBlock(dest, (unsigned char*)src + 3, 4);
+    dest += 8;
+    // make a new copy of the data in which alpha is opaque,
+    // because code uses a fast test for color constancy
+    memcpy(data, src, 4 * 16);
+    for (i = 0; i < 16; ++i)
+        data[i][3] = 255;
+    src = &data[0][0];
+
+    stb__CompressColorBlock(dest, (unsigned char*)src, 0, mode);
+}

externals/stb/stb_dxt.h

@@ -0,0 +1,36 @@
+// SPDX-FileCopyrightText: fabian "ryg" giesen
+// SPDX-License-Identifier: MIT
+
+// stb_dxt.h - v1.12 - DXT1/DXT5 compressor
+
+#ifndef STB_INCLUDE_STB_DXT_H
+#define STB_INCLUDE_STB_DXT_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef STB_DXT_STATIC
+#define STBDDEF static
+#else
+#define STBDDEF extern
+#endif
+
+// compression mode (bitflags)
+#define STB_DXT_NORMAL 0
+#define STB_DXT_DITHER 1 // use dithering. was always dubious, now deprecated. does nothing!
+#define STB_DXT_HIGHQUAL                                                                           \
+    2 // high quality mode, does two refinement steps instead of 1. ~30-40% slower.
+
+STBDDEF void stb_compress_bc1_block(unsigned char* dest,
+                                    const unsigned char* src_rgba_four_bytes_per_pixel, int alpha,
+                                    int mode);
+
+STBDDEF void stb_compress_bc3_block(unsigned char* dest, const unsigned char* src, int mode);
+
+#define STB_COMPRESS_DXT_BLOCK
+
+#ifdef __cplusplus
+}
+#endif
+#endif // STB_INCLUDE_STB_DXT_H

src/common/settings.cpp

@@ -61,6 +61,7 @@ void LogSettings() {
     log_setting("Renderer_NvdecEmulation", values.nvdec_emulation.GetValue());
     log_setting("Renderer_AccelerateASTC", values.accelerate_astc.GetValue());
     log_setting("Renderer_AsyncASTC", values.async_astc.GetValue());
+    log_setting("Renderer_AstcRecompression", values.astc_recompression.GetValue());
     log_setting("Renderer_UseVsync", values.vsync_mode.GetValue());
     log_setting("Renderer_UseReactiveFlushing", values.use_reactive_flushing.GetValue());
     log_setting("Renderer_ShaderBackend", values.shader_backend.GetValue());
@@ -224,6 +225,7 @@ void RestoreGlobalState(bool is_powered_on) {
     values.nvdec_emulation.SetGlobal(true);
     values.accelerate_astc.SetGlobal(true);
     values.async_astc.SetGlobal(true);
+    values.astc_recompression.SetGlobal(true);
     values.use_reactive_flushing.SetGlobal(true);
     values.shader_backend.SetGlobal(true);
     values.use_asynchronous_shaders.SetGlobal(true);

src/common/settings.h

@@ -90,6 +90,12 @@ enum class AntiAliasing : u32 {
     LastAA = Smaa,
 };
 
+enum class AstcRecompression : u32 {
+    Uncompressed = 0,
+    Bc1 = 1,
+    Bc3 = 2,
+};
+
 struct ResolutionScalingInfo {
     u32 up_scale{1};
     u32 down_shift{0};
@@ -473,6 +479,9 @@ struct Values {
     SwitchableSetting<bool> use_vulkan_driver_pipeline_cache{true,
                                                              "use_vulkan_driver_pipeline_cache"};
     SwitchableSetting<bool> enable_compute_pipelines{false, "enable_compute_pipelines"};
+    SwitchableSetting<AstcRecompression, true> astc_recompression{
+        AstcRecompression::Uncompressed, AstcRecompression::Uncompressed, AstcRecompression::Bc3,
+        "astc_recompression"};
 
     SwitchableSetting<u8> bg_red{0, "bg_red"};
     SwitchableSetting<u8> bg_green{0, "bg_green"};

src/video_core/CMakeLists.txt

@@ -246,10 +246,14 @@ add_library(video_core STATIC
     texture_cache/util.h
     textures/astc.h
     textures/astc.cpp
+    textures/bcn.cpp
+    textures/bcn.h
     textures/decoders.cpp
     textures/decoders.h
     textures/texture.cpp
     textures/texture.h
+    textures/workers.cpp
+    textures/workers.h
     transform_feedback.cpp
     transform_feedback.h
     video_core.cpp
@@ -275,7 +279,7 @@ add_library(video_core STATIC
 create_target_directory_groups(video_core)
 
 target_link_libraries(video_core PUBLIC common core)
-target_link_libraries(video_core PUBLIC glad shader_recompiler)
+target_link_libraries(video_core PUBLIC glad shader_recompiler stb)
 
 if (YUZU_USE_BUNDLED_FFMPEG AND NOT WIN32)
     add_dependencies(video_core ffmpeg-build)

src/video_core/buffer_cache/buffer_cache.h

@@ -1664,7 +1664,7 @@ typename BufferCache<P>::Binding BufferCache<P>::StorageBufferBinding(GPUVAddr s
         // cbufs, which do not store the sizes adjacent to the addresses, so use the fully
         // mapped buffer size for now.
         const u32 memory_layout_size = static_cast<u32>(gpu_memory->GetMemoryLayoutSize(gpu_addr));
-        return memory_layout_size;
+        return std::min(memory_layout_size, static_cast<u32>(8_MiB));
     }();
     const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr);
     if (!cpu_addr || size == 0) {

src/video_core/renderer_opengl/gl_texture_cache.cpp

@@ -233,6 +233,8 @@ void ApplySwizzle(GLuint handle, PixelFormat format, std::array<SwizzleSource, 4
                                     const VideoCommon::ImageInfo& info) {
     if (IsPixelFormatASTC(info.format) && info.size.depth == 1 && !runtime.HasNativeASTC()) {
         return Settings::values.accelerate_astc.GetValue() &&
+               Settings::values.astc_recompression.GetValue() ==
+                   Settings::AstcRecompression::Uncompressed &&
                !Settings::values.async_astc.GetValue();
     }
     // Disable other accelerated uploads for now as they don't implement swizzled uploads
@@ -437,6 +439,19 @@ OGLTexture MakeImage(const VideoCommon::ImageInfo& info, GLenum gl_internal_form
     return GL_R32UI;
 }
 
+[[nodiscard]] GLenum SelectAstcFormat(PixelFormat format, bool is_srgb) {
+    switch (Settings::values.astc_recompression.GetValue()) {
+    case Settings::AstcRecompression::Bc1:
+        return is_srgb ? GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT : GL_COMPRESSED_RGBA_S3TC_DXT1_EXT;
+        break;
+    case Settings::AstcRecompression::Bc3:
+        return is_srgb ? GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT : GL_COMPRESSED_RGBA_S3TC_DXT5_EXT;
+        break;
+    default:
+        return is_srgb ? GL_SRGB8_ALPHA8 : GL_RGBA8;
+    }
+}
+
 } // Anonymous namespace
 
 ImageBufferMap::~ImageBufferMap() {
@@ -739,9 +754,16 @@ Image::Image(TextureCacheRuntime& runtime_, const VideoCommon::ImageInfo& info_,
     if (IsConverted(runtime->device, info.format, info.type)) {
         flags |= ImageFlagBits::Converted;
         flags |= ImageFlagBits::CostlyLoad;
-        gl_internal_format = IsPixelFormatSRGB(info.format) ? GL_SRGB8_ALPHA8 : GL_RGBA8;
+
+        const bool is_srgb = IsPixelFormatSRGB(info.format);
+        gl_internal_format = is_srgb ? GL_SRGB8_ALPHA8 : GL_RGBA8;
         gl_format = GL_RGBA;
         gl_type = GL_UNSIGNED_INT_8_8_8_8_REV;
+
+        if (IsPixelFormatASTC(info.format)) {
+            gl_internal_format = SelectAstcFormat(info.format, is_srgb);
+            gl_format = GL_NONE;
+        }
     } else {
         const auto& tuple = MaxwellToGL::GetFormatTuple(info.format);
         gl_internal_format = tuple.internal_format;
@@ -1130,7 +1152,12 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
       views{runtime.null_image_views} {
     const Device& device = runtime.device;
     if (True(image.flags & ImageFlagBits::Converted)) {
-        internal_format = IsPixelFormatSRGB(info.format) ? GL_SRGB8_ALPHA8 : GL_RGBA8;
+        const bool is_srgb = IsPixelFormatSRGB(info.format);
+        internal_format = is_srgb ? GL_SRGB8_ALPHA8 : GL_RGBA8;
+
+        if (IsPixelFormatASTC(info.format)) {
+            internal_format = SelectAstcFormat(info.format, is_srgb);
+        }
     } else {
         internal_format = MaxwellToGL::GetFormatTuple(format).internal_format;
     }

src/video_core/renderer_vulkan/maxwell_to_vk.cpp

@@ -6,6 +6,7 @@
 #include "common/assert.h"
 #include "common/common_types.h"
 #include "common/logging/log.h"
+#include "common/settings.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/renderer_vulkan/maxwell_to_vk.h"
 #include "video_core/surface.h"
@@ -237,14 +238,25 @@ FormatInfo SurfaceFormat(const Device& device, FormatType format_type, bool with
                          PixelFormat pixel_format) {
     ASSERT(static_cast<size_t>(pixel_format) < std::size(tex_format_tuples));
     FormatTuple tuple = tex_format_tuples[static_cast<size_t>(pixel_format)];
-    // Use A8B8G8R8_UNORM on hardware that doesn't support ASTC natively
+    // Transcode on hardware that doesn't support ASTC natively
     if (!device.IsOptimalAstcSupported() && VideoCore::Surface::IsPixelFormatASTC(pixel_format)) {
         const bool is_srgb = with_srgb && VideoCore::Surface::IsPixelFormatSRGB(pixel_format);
-        if (is_srgb) {
-            tuple.format = VK_FORMAT_A8B8G8R8_SRGB_PACK32;
-        } else {
-            tuple.format = VK_FORMAT_A8B8G8R8_UNORM_PACK32;
-            tuple.usage |= Storage;
+
+        switch (Settings::values.astc_recompression.GetValue()) {
+        case Settings::AstcRecompression::Uncompressed:
+            if (is_srgb) {
+                tuple.format = VK_FORMAT_A8B8G8R8_SRGB_PACK32;
+            } else {
+                tuple.format = VK_FORMAT_A8B8G8R8_UNORM_PACK32;
+                tuple.usage |= Storage;
+            }
+            break;
+        case Settings::AstcRecompression::Bc1:
+            tuple.format = is_srgb ? VK_FORMAT_BC1_RGBA_SRGB_BLOCK : VK_FORMAT_BC1_RGBA_UNORM_BLOCK;
+            break;
+        case Settings::AstcRecompression::Bc3:
+            tuple.format = is_srgb ? VK_FORMAT_BC3_SRGB_BLOCK : VK_FORMAT_BC3_UNORM_BLOCK;
+            break;
         }
     }
     const bool attachable = (tuple.usage & Attachable) != 0;

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -1272,7 +1272,9 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu
     if (IsPixelFormatASTC(info.format) && !runtime->device.IsOptimalAstcSupported()) {
         if (Settings::values.async_astc.GetValue()) {
             flags |= VideoCommon::ImageFlagBits::AsynchronousDecode;
-        } else if (Settings::values.accelerate_astc.GetValue() && info.size.depth == 1) {
+        } else if (Settings::values.astc_recompression.GetValue() ==
+                       Settings::AstcRecompression::Uncompressed &&
+                   Settings::values.accelerate_astc.GetValue() && info.size.depth == 1) {
             flags |= VideoCommon::ImageFlagBits::AcceleratedUpload;
         }
         flags |= VideoCommon::ImageFlagBits::Converted;
@@ -1287,7 +1289,9 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu
         .usage = VK_IMAGE_USAGE_STORAGE_BIT,
     };
     current_image = *original_image;
-    if (IsPixelFormatASTC(info.format) && !runtime->device.IsOptimalAstcSupported()) {
+    if (IsPixelFormatASTC(info.format) && !runtime->device.IsOptimalAstcSupported() &&
+        Settings::values.astc_recompression.GetValue() ==
+            Settings::AstcRecompression::Uncompressed) {
         const auto& device = runtime->device.GetLogical();
         storage_image_views.reserve(info.resources.levels);
         for (s32 level = 0; level < info.resources.levels; ++level) {

src/video_core/texture_cache/util.cpp

@@ -18,6 +18,8 @@
 #include "common/bit_util.h"
 #include "common/common_types.h"
 #include "common/div_ceil.h"
+#include "common/scratch_buffer.h"
+#include "common/settings.h"
 #include "video_core/compatible_formats.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/memory_manager.h"
@@ -28,6 +30,7 @@
 #include "video_core/texture_cache/samples_helper.h"
 #include "video_core/texture_cache/util.h"
 #include "video_core/textures/astc.h"
+#include "video_core/textures/bcn.h"
 #include "video_core/textures/decoders.h"
 
 namespace VideoCommon {
@@ -585,6 +588,21 @@ u32 CalculateConvertedSizeBytes(const ImageInfo& info) noexcept {
         return info.size.width * BytesPerBlock(info.format);
     }
     static constexpr Extent2D TILE_SIZE{1, 1};
+    if (IsPixelFormatASTC(info.format) && Settings::values.astc_recompression.GetValue() !=
+                                              Settings::AstcRecompression::Uncompressed) {
+        const u32 bpp_div =
+            Settings::values.astc_recompression.GetValue() == Settings::AstcRecompression::Bc1 ? 2
+                                                                                               : 1;
+        // NumBlocksPerLayer doesn't account for this correctly, so we have to do it manually.
+        u32 output_size = 0;
+        for (s32 i = 0; i < info.resources.levels; i++) {
+            const auto mip_size = AdjustMipSize(info.size, i);
+            const u32 plane_dim =
+                Common::AlignUp(mip_size.width, 4U) * Common::AlignUp(mip_size.height, 4U);
+            output_size += (plane_dim * info.size.depth * info.resources.layers) / bpp_div;
+        }
+        return output_size;
+    }
     return NumBlocksPerLayer(info, TILE_SIZE) * info.resources.layers * CONVERTED_BYTES_PER_BLOCK;
 }
 
@@ -885,6 +903,7 @@ BufferCopy UploadBufferCopy(Tegra::MemoryManager& gpu_memory, GPUVAddr gpu_addr,
 void ConvertImage(std::span<const u8> input, const ImageInfo& info, std::span<u8> output,
                   std::span<BufferImageCopy> copies) {
     u32 output_offset = 0;
+    Common::ScratchBuffer<u8> decode_scratch;
 
     const Extent2D tile_size = DefaultBlockSize(info.format);
     for (BufferImageCopy& copy : copies) {
@@ -895,22 +914,58 @@ void ConvertImage(std::span<const u8> input, const ImageInfo& info, std::span<u8
         ASSERT(copy.image_extent == mip_size);
         ASSERT(copy.buffer_row_length == Common::AlignUp(mip_size.width, tile_size.width));
         ASSERT(copy.buffer_image_height == Common::AlignUp(mip_size.height, tile_size.height));
-        if (IsPixelFormatASTC(info.format)) {
+
+        const auto input_offset = input.subspan(copy.buffer_offset);
+        copy.buffer_offset = output_offset;
+        copy.buffer_row_length = mip_size.width;
+        copy.buffer_image_height = mip_size.height;
+
+        const auto recompression_setting = Settings::values.astc_recompression.GetValue();
+        const bool astc = IsPixelFormatASTC(info.format);
+
+        if (astc && recompression_setting == Settings::AstcRecompression::Uncompressed) {
             Tegra::Texture::ASTC::Decompress(
-                input.subspan(copy.buffer_offset), copy.image_extent.width,
-                copy.image_extent.height,
+                input_offset, copy.image_extent.width, copy.image_extent.height,
                 copy.image_subresource.num_layers * copy.image_extent.depth, tile_size.width,
                 tile_size.height, output.subspan(output_offset));
+
+            output_offset += copy.image_extent.width * copy.image_extent.height *
+                             copy.image_subresource.num_layers * CONVERTED_BYTES_PER_BLOCK;
+        } else if (astc) {
+            // BC1 uses 0.5 bytes per texel
+            // BC3 uses 1 byte per texel
+            const auto compress = recompression_setting == Settings::AstcRecompression::Bc1
+                                      ? Tegra::Texture::BCN::CompressBC1
+                                      : Tegra::Texture::BCN::CompressBC3;
+            const auto bpp_div = recompression_setting == Settings::AstcRecompression::Bc1 ? 2 : 1;
+
+            const u32 plane_dim = copy.image_extent.width * copy.image_extent.height;
+            const u32 level_size = plane_dim * copy.image_extent.depth *
+                                   copy.image_subresource.num_layers * CONVERTED_BYTES_PER_BLOCK;
+            decode_scratch.resize_destructive(level_size);
+
+            Tegra::Texture::ASTC::Decompress(
+                input_offset, copy.image_extent.width, copy.image_extent.height,
+                copy.image_subresource.num_layers * copy.image_extent.depth, tile_size.width,
+                tile_size.height, decode_scratch);
+
+            compress(decode_scratch, copy.image_extent.width, copy.image_extent.height,
+                     copy.image_subresource.num_layers * copy.image_extent.depth,
+                     output.subspan(output_offset));
+
+            const u32 aligned_plane_dim = Common::AlignUp(copy.image_extent.width, 4) *
+                                          Common::AlignUp(copy.image_extent.height, 4);
+
+            copy.buffer_size =
+                (aligned_plane_dim * copy.image_extent.depth * copy.image_subresource.num_layers) /
+                bpp_div;
+            output_offset += static_cast<u32>(copy.buffer_size);
         } else {
-            DecompressBC4(input.subspan(copy.buffer_offset), copy.image_extent,
-                          output.subspan(output_offset));
-        }
-        copy.buffer_offset = output_offset;
-        copy.buffer_row_length = mip_size.width;
-        copy.buffer_image_height = mip_size.height;
+            DecompressBC4(input_offset, copy.image_extent, output.subspan(output_offset));
 
-        output_offset += copy.image_extent.width * copy.image_extent.height *
-                         copy.image_subresource.num_layers * CONVERTED_BYTES_PER_BLOCK;
+            output_offset += copy.image_extent.width * copy.image_extent.height *
+                             copy.image_subresource.num_layers * CONVERTED_BYTES_PER_BLOCK;
+        }
     }
 }
 

src/video_core/textures/astc.cpp

@@ -16,8 +16,8 @@
 #include "common/alignment.h"
 #include "common/common_types.h"
 #include "common/polyfill_ranges.h"
-#include "common/thread_worker.h"
 #include "video_core/textures/astc.h"
+#include "video_core/textures/workers.h"
 
 class InputBitStream {
 public:
@@ -1656,8 +1656,7 @@ void Decompress(std::span<const uint8_t> data, uint32_t width, uint32_t height,
     const u32 rows = Common::DivideUp(height, block_height);
     const u32 cols = Common::DivideUp(width, block_width);
 
-    static Common::ThreadWorker workers{std::max(std::thread::hardware_concurrency(), 2U) / 2,
-                                        "ASTCDecompress"};
+    Common::ThreadWorker& workers{GetThreadWorkers()};
 
     for (u32 z = 0; z < depth; ++z) {
         const u32 depth_offset = z * height * width * 4;

src/video_core/textures/bcn.cpp

@@ -0,0 +1,87 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include <stb_dxt.h>
+#include <string.h>
+
+#include "common/alignment.h"
+#include "video_core/textures/bcn.h"
+#include "video_core/textures/workers.h"
+
+namespace Tegra::Texture::BCN {
+
+using BCNCompressor = void(u8* block_output, const u8* block_input, bool any_alpha);
+
+template <u32 BytesPerBlock, bool ThresholdAlpha = false>
+void CompressBCN(std::span<const uint8_t> data, uint32_t width, uint32_t height, uint32_t depth,
+                 std::span<uint8_t> output, BCNCompressor f) {
+    constexpr u8 alpha_threshold = 128;
+    constexpr u32 bytes_per_px = 4;
+    const u32 plane_dim = width * height;
+
+    Common::ThreadWorker& workers{GetThreadWorkers()};
+
+    for (u32 z = 0; z < depth; z++) {
+        for (u32 y = 0; y < height; y += 4) {
+            auto compress_row = [z, y, width, height, plane_dim, f, data, output]() {
+                for (u32 x = 0; x < width; x += 4) {
+                    // Gather 4x4 block of RGBA texels
+                    u8 input_colors[4][4][4];
+                    bool any_alpha = false;
+
+                    for (u32 j = 0; j < 4; j++) {
+                        for (u32 i = 0; i < 4; i++) {
+                            const size_t coord =
+                                (z * plane_dim + (y + j) * width + (x + i)) * bytes_per_px;
+
+                            if ((x + i < width) && (y + j < height)) {
+                                if constexpr (ThresholdAlpha) {
+                                    if (data[coord + 3] >= alpha_threshold) {
+                                        input_colors[j][i][0] = data[coord + 0];
+                                        input_colors[j][i][1] = data[coord + 1];
+                                        input_colors[j][i][2] = data[coord + 2];
+                                        input_colors[j][i][3] = 255;
+                                    } else {
+                                        any_alpha = true;
+                                        memset(input_colors[j][i], 0, bytes_per_px);
+                                    }
+                                } else {
+                                    memcpy(input_colors[j][i], &data[coord], bytes_per_px);
+                                }
+                            } else {
+                                memset(input_colors[j][i], 0, bytes_per_px);
+                            }
+                        }
+                    }
+
+                    const u32 bytes_per_row = BytesPerBlock * Common::DivideUp(width, 4U);
+                    const u32 bytes_per_plane = bytes_per_row * Common::DivideUp(height, 4U);
+                    f(output.data() + z * bytes_per_plane + (y / 4) * bytes_per_row +
+                          (x / 4) * BytesPerBlock,
+                      reinterpret_cast<u8*>(input_colors), any_alpha);
+                }
+            };
+            workers.QueueWork(std::move(compress_row));
+        }
+        workers.WaitForRequests();
+    }
+}
+
+void CompressBC1(std::span<const uint8_t> data, uint32_t width, uint32_t height, uint32_t depth,
+                 std::span<uint8_t> output) {
+    CompressBCN<8, true>(data, width, height, depth, output,
+                         [](u8* block_output, const u8* block_input, bool any_alpha) {
+                             stb_compress_bc1_block(block_output, block_input, any_alpha,
+                                                    STB_DXT_NORMAL);
+                         });
+}
+
+void CompressBC3(std::span<const uint8_t> data, uint32_t width, uint32_t height, uint32_t depth,
+                 std::span<uint8_t> output) {
+    CompressBCN<16, false>(data, width, height, depth, output,
+                           [](u8* block_output, const u8* block_input, bool any_alpha) {
+                               stb_compress_bc3_block(block_output, block_input, STB_DXT_NORMAL);
+                           });
+}
+
+} // namespace Tegra::Texture::BCN

src/video_core/textures/bcn.h

@@ -0,0 +1,17 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <span>
+#include <stdint.h>
+
+namespace Tegra::Texture::BCN {
+
+void CompressBC1(std::span<const uint8_t> data, uint32_t width, uint32_t height, uint32_t depth,
+                 std::span<uint8_t> output);
+
+void CompressBC3(std::span<const uint8_t> data, uint32_t width, uint32_t height, uint32_t depth,
+                 std::span<uint8_t> output);
+
+} // namespace Tegra::Texture::BCN

src/video_core/textures/workers.cpp

@@ -0,0 +1,15 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include "video_core/textures/workers.h"
+
+namespace Tegra::Texture {
+
+Common::ThreadWorker& GetThreadWorkers() {
+    static Common::ThreadWorker workers{std::max(std::thread::hardware_concurrency(), 2U) / 2,
+                                        "ImageTranscode"};
+
+    return workers;
+}
+
+} // namespace Tegra::Texture

src/video_core/textures/workers.h

@@ -0,0 +1,12 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include "common/thread_worker.h"
+
+namespace Tegra::Texture {
+
+Common::ThreadWorker& GetThreadWorkers();
+
+}

src/video_core/vulkan_common/vulkan_device.cpp

@@ -1001,6 +1001,11 @@ u64 Device::GetDeviceMemoryUsage() const {
 }
 
 void Device::CollectPhysicalMemoryInfo() {
+    // Account for resolution scaling in memory limits
+    const size_t normal_memory = 6_GiB;
+    const size_t scaler_memory = 1_GiB * Settings::values.resolution_info.ScaleUp(1);
+
+    // Calculate limits using memory budget
     VkPhysicalDeviceMemoryBudgetPropertiesEXT budget{};
     budget.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MEMORY_BUDGET_PROPERTIES_EXT;
     const auto mem_info =
@@ -1030,6 +1035,7 @@ void Device::CollectPhysicalMemoryInfo() {
     if (!is_integrated) {
         const u64 reserve_memory = std::min<u64>(device_access_memory / 8, 1_GiB);
         device_access_memory -= reserve_memory;
+        device_access_memory = std::min<u64>(device_access_memory, normal_memory + scaler_memory);
         return;
     }
     const s64 available_memory = static_cast<s64>(device_access_memory - device_initial_usage);

src/yuzu/configuration/config.cpp

@@ -711,6 +711,7 @@ void Config::ReadRendererValues() {
     ReadGlobalSetting(Settings::values.nvdec_emulation);
     ReadGlobalSetting(Settings::values.accelerate_astc);
     ReadGlobalSetting(Settings::values.async_astc);
+    ReadGlobalSetting(Settings::values.astc_recompression);
     ReadGlobalSetting(Settings::values.use_reactive_flushing);
     ReadGlobalSetting(Settings::values.shader_backend);
     ReadGlobalSetting(Settings::values.use_asynchronous_shaders);
@@ -1359,6 +1360,10 @@ void Config::SaveRendererValues() {
                  Settings::values.nvdec_emulation.UsingGlobal());
     WriteGlobalSetting(Settings::values.accelerate_astc);
     WriteGlobalSetting(Settings::values.async_astc);
+    WriteSetting(QString::fromStdString(Settings::values.astc_recompression.GetLabel()),
+                 static_cast<u32>(Settings::values.astc_recompression.GetValue(global)),
+                 static_cast<u32>(Settings::values.astc_recompression.GetDefault()),
+                 Settings::values.astc_recompression.UsingGlobal());
     WriteGlobalSetting(Settings::values.use_reactive_flushing);
     WriteSetting(QString::fromStdString(Settings::values.shader_backend.GetLabel()),
                  static_cast<u32>(Settings::values.shader_backend.GetValue(global)),

src/yuzu/configuration/config.h

@@ -208,3 +208,4 @@ Q_DECLARE_METATYPE(Settings::ScalingFilter);
 Q_DECLARE_METATYPE(Settings::AntiAliasing);
 Q_DECLARE_METATYPE(Settings::RendererBackend);
 Q_DECLARE_METATYPE(Settings::ShaderBackend);
+Q_DECLARE_METATYPE(Settings::AstcRecompression);

src/yuzu/configuration/configure_graphics_advanced.cpp

@@ -27,6 +27,7 @@ void ConfigureGraphicsAdvanced::SetConfiguration() {
     ui->async_present->setEnabled(runtime_lock);
     ui->renderer_force_max_clock->setEnabled(runtime_lock);
     ui->async_astc->setEnabled(runtime_lock);
+    ui->astc_recompression_combobox->setEnabled(runtime_lock);
     ui->use_asynchronous_shaders->setEnabled(runtime_lock);
     ui->anisotropic_filtering_combobox->setEnabled(runtime_lock);
     ui->enable_compute_pipelines_checkbox->setEnabled(runtime_lock);
@@ -47,14 +48,20 @@ void ConfigureGraphicsAdvanced::SetConfiguration() {
             static_cast<int>(Settings::values.gpu_accuracy.GetValue()));
         ui->anisotropic_filtering_combobox->setCurrentIndex(
             Settings::values.max_anisotropy.GetValue());
+        ui->astc_recompression_combobox->setCurrentIndex(
+            static_cast<int>(Settings::values.astc_recompression.GetValue()));
     } else {
         ConfigurationShared::SetPerGameSetting(ui->gpu_accuracy, &Settings::values.gpu_accuracy);
         ConfigurationShared::SetPerGameSetting(ui->anisotropic_filtering_combobox,
                                                &Settings::values.max_anisotropy);
+        ConfigurationShared::SetPerGameSetting(ui->astc_recompression_combobox,
+                                               &Settings::values.astc_recompression);
         ConfigurationShared::SetHighlight(ui->label_gpu_accuracy,
                                           !Settings::values.gpu_accuracy.UsingGlobal());
         ConfigurationShared::SetHighlight(ui->af_label,
                                           !Settings::values.max_anisotropy.UsingGlobal());
+        ConfigurationShared::SetHighlight(ui->label_astc_recompression,
+                                          !Settings::values.astc_recompression.UsingGlobal());
     }
 }
 
@@ -71,6 +78,8 @@ void ConfigureGraphicsAdvanced::ApplyConfiguration() {
                                              ui->use_reactive_flushing, use_reactive_flushing);
     ConfigurationShared::ApplyPerGameSetting(&Settings::values.async_astc, ui->async_astc,
                                              async_astc);
+    ConfigurationShared::ApplyPerGameSetting(&Settings::values.astc_recompression,
+                                             ui->astc_recompression_combobox);
     ConfigurationShared::ApplyPerGameSetting(&Settings::values.use_asynchronous_shaders,
                                              ui->use_asynchronous_shaders,
                                              use_asynchronous_shaders);
@@ -105,6 +114,8 @@ void ConfigureGraphicsAdvanced::SetupPerGameUI() {
             Settings::values.renderer_force_max_clock.UsingGlobal());
         ui->use_reactive_flushing->setEnabled(Settings::values.use_reactive_flushing.UsingGlobal());
         ui->async_astc->setEnabled(Settings::values.async_astc.UsingGlobal());
+        ui->astc_recompression_combobox->setEnabled(
+            Settings::values.astc_recompression.UsingGlobal());
         ui->use_asynchronous_shaders->setEnabled(
             Settings::values.use_asynchronous_shaders.UsingGlobal());
         ui->use_fast_gpu_time->setEnabled(Settings::values.use_fast_gpu_time.UsingGlobal());
@@ -144,6 +155,9 @@ void ConfigureGraphicsAdvanced::SetupPerGameUI() {
     ConfigurationShared::SetColoredComboBox(
         ui->anisotropic_filtering_combobox, ui->af_label,
         static_cast<int>(Settings::values.max_anisotropy.GetValue(true)));
+    ConfigurationShared::SetColoredComboBox(
+        ui->astc_recompression_combobox, ui->label_astc_recompression,
+        static_cast<int>(Settings::values.astc_recompression.GetValue(true)));
 }
 
 void ConfigureGraphicsAdvanced::ExposeComputeOption() {

src/yuzu/configuration/configure_graphics_advanced.ui

@@ -69,6 +69,50 @@
           </layout>
          </widget>
         </item>
+        <item>
+         <widget class="QWidget" name="astc_recompression_layout" native="true">
+          <layout class="QHBoxLayout" name="horizontalLayout_3">
+            <property name="leftMargin">
+              <number>0</number>
+            </property>
+            <property name="topMargin">
+              <number>0</number>
+            </property>
+            <property name="rightMargin">
+              <number>0</number>
+            </property>
+            <property name="bottomMargin">
+              <number>0</number>
+            </property>
+            <item>
+              <widget class="QLabel" name="label_astc_recompression">
+              <property name="text">
+                <string>ASTC recompression:</string>
+              </property>
+              </widget>
+            </item>
+            <item>
+              <widget class="QComboBox" name="astc_recompression_combobox">
+                <item>
+                <property name="text">
+                  <string>Uncompressed (Best quality)</string>
+                </property>
+                </item>
+                <item>
+                <property name="text">
+                  <string>BC1 (Low quality)</string>
+                </property>
+                </item>
+                <item>
+                <property name="text">
+                  <string>BC3 (Medium quality)</string>
+                </property>
+                </item>
+              </widget>
+            </item>
+          </layout>
+         </widget>
+        </item>
         <item>
          <widget class="QCheckBox" name="async_present">
           <property name="text">

src/yuzu_cmd/config.cpp

@@ -318,6 +318,7 @@ void Config::ReadValues() {
     ReadSetting("Renderer", Settings::values.nvdec_emulation);
     ReadSetting("Renderer", Settings::values.accelerate_astc);
     ReadSetting("Renderer", Settings::values.async_astc);
+    ReadSetting("Renderer", Settings::values.astc_recompression);
     ReadSetting("Renderer", Settings::values.use_fast_gpu_time);
     ReadSetting("Renderer", Settings::values.use_vulkan_driver_pipeline_cache);
 

src/yuzu_cmd/default_ini.h

@@ -360,6 +360,10 @@ accelerate_astc =
 # 0 (default): Off, 1: On
 async_astc =
 
+# Recompress ASTC textures to a different format.
+# 0 (default): Uncompressed, 1: BC1 (Low quality), 2: BC3: (Medium quality)
+async_astc =
+
 # Turns on the speed limiter, which will limit the emulation speed to the desired speed limit value
 # 0: Off, 1: On (default)
 use_speed_limit =