audio_core: interpolate: Improvements to fix audio crackling.

- Fixes audio crackling in Crash Team Racing Nitro-Fueled, Super Mario Odyssey, and others. - Addresses followup issues from #3310.
Merge pull request #3444 from bunnei/linux-audio-fix
2020-02-22 22:26:16 -05:00 · 2020-02-22 03:08:05 -05:00 · 2020-02-22 02:29:41 -05:00 · 2020-02-22 01:54:40 -05:00 · 2020-02-21 22:08:57 -05:00 · 2020-02-21 15:57:27 -03:00
31 changed files with 3085 additions and 898 deletions
@@ -1,4 +1,4 @@
-From https://github.com/yhirose/cpp-httplib/commit/d9479bc0b12e8a1e8bce2d34da4feeef488581f3
+From https://github.com/yhirose/cpp-httplib/tree/fce8e6fefdab4ad48bc5b25c98e5ebfda4f3cf53

 MIT License

@@ -5,21 +5,141 @@
 #define _USE_MATH_DEFINES

 #include <algorithm>
+#include <climits>
 #include <cmath>
 #include <vector>
+
 #include "audio_core/algorithm/interpolate.h"
 #include "common/common_types.h"
 #include "common/logging/log.h"

 namespace AudioCore {

-/// The Lanczos kernel
-static double Lanczos(std::size_t a, double x) {
-    if (x == 0.0)
-        return 1.0;
-    const double px = M_PI * x;
-    return a * std::sin(px) * std::sin(px / a) / (px * px);
-}
+constexpr std::array<s16, 512> curve_lut0{
+    6600,  19426, 6722,  3,     6479,  19424, 6845,  9,     6359,  19419, 6968,  15,    6239,
+    19412, 7093,  22,    6121,  19403, 7219,  28,    6004,  19391, 7345,  34,    5888,  19377,
+    7472,  41,    5773,  19361, 7600,  48,    5659,  19342, 7728,  55,    5546,  19321, 7857,
+    62,    5434,  19298, 7987,  69,    5323,  19273, 8118,  77,    5213,  19245, 8249,  84,
+    5104,  19215, 8381,  92,    4997,  19183, 8513,  101,   4890,  19148, 8646,  109,   4785,
+    19112, 8780,  118,   4681,  19073, 8914,  127,   4579,  19031, 9048,  137,   4477,  18988,
+    9183,  147,   4377,  18942, 9318,  157,   4277,  18895, 9454,  168,   4179,  18845, 9590,
+    179,   4083,  18793, 9726,  190,   3987,  18738, 9863,  202,   3893,  18682, 10000, 215,
+    3800,  18624, 10137, 228,   3709,  18563, 10274, 241,   3618,  18500, 10411, 255,   3529,
+    18436, 10549, 270,   3441,  18369, 10687, 285,   3355,  18300, 10824, 300,   3269,  18230,
+    10962, 317,   3186,  18157, 11100, 334,   3103,  18082, 11238, 351,   3022,  18006, 11375,
+    369,   2942,  17927, 11513, 388,   2863,  17847, 11650, 408,   2785,  17765, 11788, 428,
+    2709,  17681, 11925, 449,   2635,  17595, 12062, 471,   2561,  17507, 12198, 494,   2489,
+    17418, 12334, 517,   2418,  17327, 12470, 541,   2348,  17234, 12606, 566,   2280,  17140,
+    12741, 592,   2213,  17044, 12876, 619,   2147,  16946, 13010, 647,   2083,  16846, 13144,
+    675,   2020,  16745, 13277, 704,   1958,  16643, 13409, 735,   1897,  16539, 13541, 766,
+    1838,  16434, 13673, 798,   1780,  16327, 13803, 832,   1723,  16218, 13933, 866,   1667,
+    16109, 14062, 901,   1613,  15998, 14191, 937,   1560,  15885, 14318, 975,   1508,  15772,
+    14445, 1013,  1457,  15657, 14571, 1052,  1407,  15540, 14695, 1093,  1359,  15423, 14819,
+    1134,  1312,  15304, 14942, 1177,  1266,  15185, 15064, 1221,  1221,  15064, 15185, 1266,
+    1177,  14942, 15304, 1312,  1134,  14819, 15423, 1359,  1093,  14695, 15540, 1407,  1052,
+    14571, 15657, 1457,  1013,  14445, 15772, 1508,  975,   14318, 15885, 1560,  937,   14191,
+    15998, 1613,  901,   14062, 16109, 1667,  866,   13933, 16218, 1723,  832,   13803, 16327,
+    1780,  798,   13673, 16434, 1838,  766,   13541, 16539, 1897,  735,   13409, 16643, 1958,
+    704,   13277, 16745, 2020,  675,   13144, 16846, 2083,  647,   13010, 16946, 2147,  619,
+    12876, 17044, 2213,  592,   12741, 17140, 2280,  566,   12606, 17234, 2348,  541,   12470,
+    17327, 2418,  517,   12334, 17418, 2489,  494,   12198, 17507, 2561,  471,   12062, 17595,
+    2635,  449,   11925, 17681, 2709,  428,   11788, 17765, 2785,  408,   11650, 17847, 2863,
+    388,   11513, 17927, 2942,  369,   11375, 18006, 3022,  351,   11238, 18082, 3103,  334,
+    11100, 18157, 3186,  317,   10962, 18230, 3269,  300,   10824, 18300, 3355,  285,   10687,
+    18369, 3441,  270,   10549, 18436, 3529,  255,   10411, 18500, 3618,  241,   10274, 18563,
+    3709,  228,   10137, 18624, 3800,  215,   10000, 18682, 3893,  202,   9863,  18738, 3987,
+    190,   9726,  18793, 4083,  179,   9590,  18845, 4179,  168,   9454,  18895, 4277,  157,
+    9318,  18942, 4377,  147,   9183,  18988, 4477,  137,   9048,  19031, 4579,  127,   8914,
+    19073, 4681,  118,   8780,  19112, 4785,  109,   8646,  19148, 4890,  101,   8513,  19183,
+    4997,  92,    8381,  19215, 5104,  84,    8249,  19245, 5213,  77,    8118,  19273, 5323,
+    69,    7987,  19298, 5434,  62,    7857,  19321, 5546,  55,    7728,  19342, 5659,  48,
+    7600,  19361, 5773,  41,    7472,  19377, 5888,  34,    7345,  19391, 6004,  28,    7219,
+    19403, 6121,  22,    7093,  19412, 6239,  15,    6968,  19419, 6359,  9,     6845,  19424,
+    6479,  3,     6722,  19426, 6600};
+
+constexpr std::array<s16, 512> curve_lut1{
+    -68,   32639, 69,    -5,    -200,  32630, 212,   -15,   -328,  32613, 359,   -26,   -450,
+    32586, 512,   -36,   -568,  32551, 669,   -47,   -680,  32507, 832,   -58,   -788,  32454,
+    1000,  -69,   -891,  32393, 1174,  -80,   -990,  32323, 1352,  -92,   -1084, 32244, 1536,
+    -103,  -1173, 32157, 1724,  -115,  -1258, 32061, 1919,  -128,  -1338, 31956, 2118,  -140,
+    -1414, 31844, 2322,  -153,  -1486, 31723, 2532,  -167,  -1554, 31593, 2747,  -180,  -1617,
+    31456, 2967,  -194,  -1676, 31310, 3192,  -209,  -1732, 31157, 3422,  -224,  -1783, 30995,
+    3657,  -240,  -1830, 30826, 3897,  -256,  -1874, 30649, 4143,  -272,  -1914, 30464, 4393,
+    -289,  -1951, 30272, 4648,  -307,  -1984, 30072, 4908,  -325,  -2014, 29866, 5172,  -343,
+    -2040, 29652, 5442,  -362,  -2063, 29431, 5716,  -382,  -2083, 29203, 5994,  -403,  -2100,
+    28968, 6277,  -424,  -2114, 28727, 6565,  -445,  -2125, 28480, 6857,  -468,  -2133, 28226,
+    7153,  -490,  -2139, 27966, 7453,  -514,  -2142, 27700, 7758,  -538,  -2142, 27428, 8066,
+    -563,  -2141, 27151, 8378,  -588,  -2136, 26867, 8694,  -614,  -2130, 26579, 9013,  -641,
+    -2121, 26285, 9336,  -668,  -2111, 25987, 9663,  -696,  -2098, 25683, 9993,  -724,  -2084,
+    25375, 10326, -753,  -2067, 25063, 10662, -783,  -2049, 24746, 11000, -813,  -2030, 24425,
+    11342, -844,  -2009, 24100, 11686, -875,  -1986, 23771, 12033, -907,  -1962, 23438, 12382,
+    -939,  -1937, 23103, 12733, -972,  -1911, 22764, 13086, -1005, -1883, 22422, 13441, -1039,
+    -1855, 22077, 13798, -1072, -1825, 21729, 14156, -1107, -1795, 21380, 14516, -1141, -1764,
+    21027, 14877, -1176, -1732, 20673, 15239, -1211, -1700, 20317, 15602, -1246, -1667, 19959,
+    15965, -1282, -1633, 19600, 16329, -1317, -1599, 19239, 16694, -1353, -1564, 18878, 17058,
+    -1388, -1530, 18515, 17423, -1424, -1495, 18151, 17787, -1459, -1459, 17787, 18151, -1495,
+    -1424, 17423, 18515, -1530, -1388, 17058, 18878, -1564, -1353, 16694, 19239, -1599, -1317,
+    16329, 19600, -1633, -1282, 15965, 19959, -1667, -1246, 15602, 20317, -1700, -1211, 15239,
+    20673, -1732, -1176, 14877, 21027, -1764, -1141, 14516, 21380, -1795, -1107, 14156, 21729,
+    -1825, -1072, 13798, 22077, -1855, -1039, 13441, 22422, -1883, -1005, 13086, 22764, -1911,
+    -972,  12733, 23103, -1937, -939,  12382, 23438, -1962, -907,  12033, 23771, -1986, -875,
+    11686, 24100, -2009, -844,  11342, 24425, -2030, -813,  11000, 24746, -2049, -783,  10662,
+    25063, -2067, -753,  10326, 25375, -2084, -724,  9993,  25683, -2098, -696,  9663,  25987,
+    -2111, -668,  9336,  26285, -2121, -641,  9013,  26579, -2130, -614,  8694,  26867, -2136,
+    -588,  8378,  27151, -2141, -563,  8066,  27428, -2142, -538,  7758,  27700, -2142, -514,
+    7453,  27966, -2139, -490,  7153,  28226, -2133, -468,  6857,  28480, -2125, -445,  6565,
+    28727, -2114, -424,  6277,  28968, -2100, -403,  5994,  29203, -2083, -382,  5716,  29431,
+    -2063, -362,  5442,  29652, -2040, -343,  5172,  29866, -2014, -325,  4908,  30072, -1984,
+    -307,  4648,  30272, -1951, -289,  4393,  30464, -1914, -272,  4143,  30649, -1874, -256,
+    3897,  30826, -1830, -240,  3657,  30995, -1783, -224,  3422,  31157, -1732, -209,  3192,
+    31310, -1676, -194,  2967,  31456, -1617, -180,  2747,  31593, -1554, -167,  2532,  31723,
+    -1486, -153,  2322,  31844, -1414, -140,  2118,  31956, -1338, -128,  1919,  32061, -1258,
+    -115,  1724,  32157, -1173, -103,  1536,  32244, -1084, -92,   1352,  32323, -990,  -80,
+    1174,  32393, -891,  -69,   1000,  32454, -788,  -58,   832,   32507, -680,  -47,   669,
+    32551, -568,  -36,   512,   32586, -450,  -26,   359,   32613, -328,  -15,   212,   32630,
+    -200,  -5,    69,    32639, -68};
+
+constexpr std::array<s16, 512> curve_lut2{
+    3195,  26287, 3329,  -32,   3064,  26281, 3467,  -34,   2936,  26270, 3608,  -38,   2811,
+    26253, 3751,  -42,   2688,  26230, 3897,  -46,   2568,  26202, 4046,  -50,   2451,  26169,
+    4199,  -54,   2338,  26130, 4354,  -58,   2227,  26085, 4512,  -63,   2120,  26035, 4673,
+    -67,   2015,  25980, 4837,  -72,   1912,  25919, 5004,  -76,   1813,  25852, 5174,  -81,
+    1716,  25780, 5347,  -87,   1622,  25704, 5522,  -92,   1531,  25621, 5701,  -98,   1442,
+    25533, 5882,  -103,  1357,  25440, 6066,  -109,  1274,  25342, 6253,  -115,  1193,  25239,
+    6442,  -121,  1115,  25131, 6635,  -127,  1040,  25018, 6830,  -133,  967,   24899, 7027,
+    -140,  897,   24776, 7227,  -146,  829,   24648, 7430,  -153,  764,   24516, 7635,  -159,
+    701,   24379, 7842,  -166,  641,   24237, 8052,  -174,  583,   24091, 8264,  -181,  526,
+    23940, 8478,  -187,  472,   23785, 8695,  -194,  420,   23626, 8914,  -202,  371,   23462,
+    9135,  -209,  324,   23295, 9358,  -215,  279,   23123, 9583,  -222,  236,   22948, 9809,
+    -230,  194,   22769, 10038, -237,  154,   22586, 10269, -243,  117,   22399, 10501, -250,
+    81,    22208, 10735, -258,  47,    22015, 10970, -265,  15,    21818, 11206, -271,  -16,
+    21618, 11444, -277,  -44,   21415, 11684, -283,  -71,   21208, 11924, -290,  -97,   20999,
+    12166, -296,  -121,  20786, 12409, -302,  -143,  20571, 12653, -306,  -163,  20354, 12898,
+    -311,  -183,  20134, 13143, -316,  -201,  19911, 13389, -321,  -218,  19686, 13635, -325,
+    -234,  19459, 13882, -328,  -248,  19230, 14130, -332,  -261,  18998, 14377, -335,  -273,
+    18765, 14625, -337,  -284,  18531, 14873, -339,  -294,  18295, 15121, -341,  -302,  18057,
+    15369, -341,  -310,  17817, 15617, -341,  -317,  17577, 15864, -340,  -323,  17335, 16111,
+    -340,  -328,  17092, 16357, -338,  -332,  16848, 16603, -336,  -336,  16603, 16848, -332,
+    -338,  16357, 17092, -328,  -340,  16111, 17335, -323,  -340,  15864, 17577, -317,  -341,
+    15617, 17817, -310,  -341,  15369, 18057, -302,  -341,  15121, 18295, -294,  -339,  14873,
+    18531, -284,  -337,  14625, 18765, -273,  -335,  14377, 18998, -261,  -332,  14130, 19230,
+    -248,  -328,  13882, 19459, -234,  -325,  13635, 19686, -218,  -321,  13389, 19911, -201,
+    -316,  13143, 20134, -183,  -311,  12898, 20354, -163,  -306,  12653, 20571, -143,  -302,
+    12409, 20786, -121,  -296,  12166, 20999, -97,   -290,  11924, 21208, -71,   -283,  11684,
+    21415, -44,   -277,  11444, 21618, -16,   -271,  11206, 21818, 15,    -265,  10970, 22015,
+    47,    -258,  10735, 22208, 81,    -250,  10501, 22399, 117,   -243,  10269, 22586, 154,
+    -237,  10038, 22769, 194,   -230,  9809,  22948, 236,   -222,  9583,  23123, 279,   -215,
+    9358,  23295, 324,   -209,  9135,  23462, 371,   -202,  8914,  23626, 420,   -194,  8695,
+    23785, 472,   -187,  8478,  23940, 526,   -181,  8264,  24091, 583,   -174,  8052,  24237,
+    641,   -166,  7842,  24379, 701,   -159,  7635,  24516, 764,   -153,  7430,  24648, 829,
+    -146,  7227,  24776, 897,   -140,  7027,  24899, 967,   -133,  6830,  25018, 1040,  -127,
+    6635,  25131, 1115,  -121,  6442,  25239, 1193,  -115,  6253,  25342, 1274,  -109,  6066,
+    25440, 1357,  -103,  5882,  25533, 1442,  -98,   5701,  25621, 1531,  -92,   5522,  25704,
+    1622,  -87,   5347,  25780, 1716,  -81,   5174,  25852, 1813,  -76,   5004,  25919, 1912,
+    -72,   4837,  25980, 2015,  -67,   4673,  26035, 2120,  -63,   4512,  26085, 2227,  -58,
+    4354,  26130, 2338,  -54,   4199,  26169, 2451,  -50,   4046,  26202, 2568,  -46,   3897,
+    26230, 2688,  -42,   3751,  26253, 2811,  -38,   3608,  26270, 2936,  -34,   3467,  26281,
+    3064,  -32,   3329,  26287, 3195};

 std::vector<s16> Interpolate(InterpolationState& state, std::vector<s16> input, double ratio) {
    if (input.size() < 2)
@@ -27,43 +147,51 @@ std::vector<s16> Interpolate(InterpolationState& state, std::vector<s16> input,

    if (ratio <= 0) {
        LOG_CRITICAL(Audio, "Nonsensical interpolation ratio {}", ratio);
-        ratio = 1.0;
+        return input;
    }

-    if (ratio != state.current_ratio) {
-        const double cutoff_frequency = std::min(0.5 / ratio, 0.5 * ratio);
-        state.nyquist = CascadingFilter::LowPass(std::clamp(cutoff_frequency, 0.0, 0.4), 3);
-        state.current_ratio = ratio;
-    }
-    state.nyquist.Process(input);
+    const s32 step{static_cast<s32>(ratio * 0x8000)};
+    const std::array<s16, 512>& lut = [step] {
+        if (step > 0xaaaa) {
+            return curve_lut0;
+        }
+        if (step <= 0x8000) {
+            return curve_lut1;
+        }
+        return curve_lut2;
+    }();

-    constexpr std::size_t taps = InterpolationState::lanczos_taps;
-    const std::size_t num_frames = input.size() / 2;
+    const std::size_t num_frames{input.size() / 2};

    std::vector<s16> output;
-    output.reserve(static_cast<std::size_t>(input.size() / ratio + 4));
+    output.reserve(static_cast<std::size_t>(input.size() / ratio + InterpolationState::taps));

-    double& pos = state.position;
-    auto& h = state.history;
-    for (std::size_t i = 0; i < num_frames; ++i) {
-        std::rotate(h.begin(), h.end() - 1, h.end());
-        h[0][0] = input[i * 2 + 0];
-        h[0][1] = input[i * 2 + 1];
+    for (std::size_t frame{}; frame < num_frames; ++frame) {
+        const std::size_t lut_index{(state.fraction >> 8) * InterpolationState::taps};

-        while (pos <= 1.0) {
-            double l = 0.0;
-            double r = 0.0;
-            for (std::size_t j = 0; j < h.size(); j++) {
-                const double lanczos_calc = Lanczos(taps, pos + j - taps + 1);
-                l += lanczos_calc * h[j][0];
-                r += lanczos_calc * h[j][1];
-            }
-            output.emplace_back(static_cast<s16>(std::clamp(l, -32768.0, 32767.0)));
-            output.emplace_back(static_cast<s16>(std::clamp(r, -32768.0, 32767.0)));
+        std::rotate(state.history.begin(), state.history.end() - 1, state.history.end());
+        state.history[0][0] = input[frame * 2 + 0];
+        state.history[0][1] = input[frame * 2 + 1];

-            pos += ratio;
+        while (state.position <= 1.0) {
+            const s32 left{state.history[0][0] * lut[lut_index + 0] +
+                           state.history[1][0] * lut[lut_index + 1] +
+                           state.history[2][0] * lut[lut_index + 2] +
+                           state.history[3][0] * lut[lut_index + 3]};
+            const s32 right{state.history[0][1] * lut[lut_index + 0] +
+                            state.history[1][1] * lut[lut_index + 1] +
+                            state.history[2][1] * lut[lut_index + 2] +
+                            state.history[3][1] * lut[lut_index + 3]};
+            const s32 new_offset{state.fraction + step};
+
+            state.fraction = new_offset & 0x7fff;
+
+            output.emplace_back(static_cast<s16>(std::clamp(left >> 15, SHRT_MIN, SHRT_MAX)));
+            output.emplace_back(static_cast<s16>(std::clamp(right >> 15, SHRT_MIN, SHRT_MAX)));
+
+            state.position += ratio;
        }
-        pos -= 1.0;
+        state.position -= 1.0;
    }

    return output;
@@ -6,19 +6,17 @@

 #include <array>
 #include <vector>
-#include "audio_core/algorithm/filter.h"
+
 #include "common/common_types.h"

 namespace AudioCore {

 struct InterpolationState {
-    static constexpr std::size_t lanczos_taps = 4;
-    static constexpr std::size_t history_size = lanczos_taps * 2 - 1;
-
-    double current_ratio = 0.0;
-    CascadingFilter nyquist;
-    std::array<std::array<s16, 2>, history_size> history = {};
-    double position = 0;
+    static constexpr std::size_t taps{4};
+    static constexpr std::size_t history_size{taps * 2 - 1};
+    std::array<std::array<s16, 2>, history_size> history{};
+    double position{};
+    s32 fraction{};
 };

 /// Interpolates input signal to produce output signal.
@@ -200,7 +200,8 @@ private:
    DownloadResult DownloadInternal(const std::string& resolved_path, u32 timeout_seconds,
                                    const std::string& content_type_name) {
        if (client == nullptr) {
-            client = std::make_unique<httplib::SSLClient>(BOXCAT_HOSTNAME, PORT, timeout_seconds);
+            client = std::make_unique<httplib::SSLClient>(BOXCAT_HOSTNAME, PORT);
+            client->set_timeout_sec(timeout_seconds);
        }

        httplib::Headers headers{
@@ -448,8 +449,8 @@ std::optional<std::vector<u8>> Boxcat::GetLaunchParameter(TitleIDVersion title)

 Boxcat::StatusResult Boxcat::GetStatus(std::optional<std::string>& global,
                                       std::map<std::string, EventStatus>& games) {
-    httplib::SSLClient client{BOXCAT_HOSTNAME, static_cast<int>(PORT),
-                              static_cast<int>(TIMEOUT_SECONDS)};
+    httplib::SSLClient client{BOXCAT_HOSTNAME, static_cast<int>(PORT)};
+    client.set_timeout_sec(static_cast<int>(TIMEOUT_SECONDS));

    httplib::Headers headers{
        {std::string("Game-Assets-API-Version"), std::string(BOXCAT_API_VERSION)},
@@ -37,6 +37,7 @@ add_library(video_core STATIC
    memory_manager.h
    morton.cpp
    morton.h
+    query_cache.h
    rasterizer_accelerated.cpp
    rasterizer_accelerated.h
    rasterizer_cache.cpp
@@ -74,6 +75,8 @@ add_library(video_core STATIC
    renderer_opengl/gl_stream_buffer.h
    renderer_opengl/gl_texture_cache.cpp
    renderer_opengl/gl_texture_cache.h
+    renderer_opengl/gl_query_cache.cpp
+    renderer_opengl/gl_query_cache.h
    renderer_opengl/maxwell_to_gl.h
    renderer_opengl/renderer_opengl.cpp
    renderer_opengl/renderer_opengl.h
@@ -177,6 +180,8 @@ if (ENABLE_VULKAN)
        renderer_vulkan/vk_memory_manager.h
        renderer_vulkan/vk_pipeline_cache.cpp
        renderer_vulkan/vk_pipeline_cache.h
+        renderer_vulkan/vk_query_cache.cpp
+        renderer_vulkan/vk_query_cache.h
        renderer_vulkan/vk_rasterizer.cpp
        renderer_vulkan/vk_rasterizer.h
        renderer_vulkan/vk_renderpass_cache.cpp
@@ -4,6 +4,7 @@

 #include <cinttypes>
 #include <cstring>
+#include <optional>
 #include "common/assert.h"
 #include "core/core.h"
 #include "core/core_timing.h"
@@ -16,6 +17,8 @@

 namespace Tegra::Engines {

+using VideoCore::QueryType;
+
 /// First register id that is actually a Macro call.
 constexpr u32 MacroRegistersStart = 0xE00;

@@ -400,6 +403,10 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
        ProcessQueryCondition();
        break;
    }
+    case MAXWELL3D_REG_INDEX(counter_reset): {
+        ProcessCounterReset();
+        break;
+    }
    case MAXWELL3D_REG_INDEX(sync_info): {
        ProcessSyncPoint();
        break;
@@ -482,7 +489,7 @@ void Maxwell3D::FlushMMEInlineDraw() {

    const bool is_indexed = mme_draw.current_mode == MMEDrawMode::Indexed;
    if (ShouldExecute()) {
-        rasterizer.DrawMultiBatch(is_indexed);
+        rasterizer.Draw(is_indexed, true);
    }

    // TODO(bunnei): Below, we reset vertex count so that we can use these registers to determine if
@@ -544,40 +551,28 @@ void Maxwell3D::ProcessQueryGet() {
               "Units other than CROP are unimplemented");

    switch (regs.query.query_get.operation) {
-    case Regs::QueryOperation::Release: {
-        const u64 result = regs.query.query_sequence;
-        StampQueryResult(result, regs.query.query_get.short_query == 0);
+    case Regs::QueryOperation::Release:
+        StampQueryResult(regs.query.query_sequence, regs.query.query_get.short_query == 0);
        break;
-    }
-    case Regs::QueryOperation::Acquire: {
-        // Todo(Blinkhawk): Under this operation, the GPU waits for the CPU
-        // to write a value that matches the current payload.
+    case Regs::QueryOperation::Acquire:
+        // TODO(Blinkhawk): Under this operation, the GPU waits for the CPU to write a value that
+        // matches the current payload.
        UNIMPLEMENTED_MSG("Unimplemented query operation ACQUIRE");
        break;
-    }
-    case Regs::QueryOperation::Counter: {
-        u64 result{};
-        switch (regs.query.query_get.select) {
-        case Regs::QuerySelect::Zero:
-            result = 0;
-            break;
-        default:
-            result = 1;
-            UNIMPLEMENTED_MSG("Unimplemented query select type {}",
-                              static_cast<u32>(regs.query.query_get.select.Value()));
+    case Regs::QueryOperation::Counter:
+        if (const std::optional<u64> result = GetQueryResult()) {
+            // If the query returns an empty optional it means it's cached and deferred.
+            // In this case we have a non-empty result, so we stamp it immediately.
+            StampQueryResult(*result, regs.query.query_get.short_query == 0);
        }
-        StampQueryResult(result, regs.query.query_get.short_query == 0);
        break;
-    }
-    case Regs::QueryOperation::Trap: {
+    case Regs::QueryOperation::Trap:
        UNIMPLEMENTED_MSG("Unimplemented query operation TRAP");
        break;
-    }
-    default: {
+    default:
        UNIMPLEMENTED_MSG("Unknown query operation");
        break;
    }
-    }
 }

 void Maxwell3D::ProcessQueryCondition() {
@@ -593,20 +588,20 @@ void Maxwell3D::ProcessQueryCondition() {
    }
    case Regs::ConditionMode::ResNonZero: {
        Regs::QueryCompare cmp;
-        memory_manager.ReadBlockUnsafe(condition_address, &cmp, sizeof(cmp));
+        memory_manager.ReadBlock(condition_address, &cmp, sizeof(cmp));
        execute_on = cmp.initial_sequence != 0U && cmp.initial_mode != 0U;
        break;
    }
    case Regs::ConditionMode::Equal: {
        Regs::QueryCompare cmp;
-        memory_manager.ReadBlockUnsafe(condition_address, &cmp, sizeof(cmp));
+        memory_manager.ReadBlock(condition_address, &cmp, sizeof(cmp));
        execute_on =
            cmp.initial_sequence == cmp.current_sequence && cmp.initial_mode == cmp.current_mode;
        break;
    }
    case Regs::ConditionMode::NotEqual: {
        Regs::QueryCompare cmp;
-        memory_manager.ReadBlockUnsafe(condition_address, &cmp, sizeof(cmp));
+        memory_manager.ReadBlock(condition_address, &cmp, sizeof(cmp));
        execute_on =
            cmp.initial_sequence != cmp.current_sequence || cmp.initial_mode != cmp.current_mode;
        break;
@@ -619,6 +614,18 @@ void Maxwell3D::ProcessQueryCondition() {
    }
 }

+void Maxwell3D::ProcessCounterReset() {
+    switch (regs.counter_reset) {
+    case Regs::CounterReset::SampleCnt:
+        rasterizer.ResetCounter(QueryType::SamplesPassed);
+        break;
+    default:
+        LOG_WARNING(Render_OpenGL, "Unimplemented counter reset={}",
+                    static_cast<int>(regs.counter_reset));
+        break;
+    }
+}
+
 void Maxwell3D::ProcessSyncPoint() {
    const u32 sync_point = regs.sync_info.sync_point.Value();
    const u32 increment = regs.sync_info.increment.Value();
@@ -647,7 +654,7 @@ void Maxwell3D::DrawArrays() {

    const bool is_indexed{regs.index_array.count && !regs.vertex_buffer.count};
    if (ShouldExecute()) {
-        rasterizer.DrawBatch(is_indexed);
+        rasterizer.Draw(is_indexed, false);
    }

    // TODO(bunnei): Below, we reset vertex count so that we can use these registers to determine if
@@ -661,6 +668,22 @@ void Maxwell3D::DrawArrays() {
    }
 }

+std::optional<u64> Maxwell3D::GetQueryResult() {
+    switch (regs.query.query_get.select) {
+    case Regs::QuerySelect::Zero:
+        return 0;
+    case Regs::QuerySelect::SamplesPassed:
+        // Deferred.
+        rasterizer.Query(regs.query.QueryAddress(), VideoCore::QueryType::SamplesPassed,
+                         system.GPU().GetTicks());
+        return {};
+    default:
+        UNIMPLEMENTED_MSG("Unimplemented query select type {}",
+                          static_cast<u32>(regs.query.query_get.select.Value()));
+        return 1;
+    }
+}
+
 void Maxwell3D::ProcessCBBind(std::size_t stage_index) {
    // Bind the buffer currently in CB_ADDRESS to the specified index in the desired shader stage.
    auto& shader = state.shader_stages[stage_index];
@@ -6,6 +6,7 @@

 #include <array>
 #include <bitset>
+#include <optional>
 #include <type_traits>
 #include <unordered_map>
 #include <vector>
@@ -409,6 +410,27 @@ public:
            Linear = 1,
        };

+        enum class CounterReset : u32 {
+            SampleCnt = 0x01,
+            Unk02 = 0x02,
+            Unk03 = 0x03,
+            Unk04 = 0x04,
+            EmittedPrimitives = 0x10, // Not tested
+            Unk11 = 0x11,
+            Unk12 = 0x12,
+            Unk13 = 0x13,
+            Unk15 = 0x15,
+            Unk16 = 0x16,
+            Unk17 = 0x17,
+            Unk18 = 0x18,
+            Unk1A = 0x1A,
+            Unk1B = 0x1B,
+            Unk1C = 0x1C,
+            Unk1D = 0x1D,
+            Unk1E = 0x1E,
+            GeneratedPrimitives = 0x1F,
+        };
+
        struct Cull {
            enum class FrontFace : u32 {
                ClockWise = 0x0900,
@@ -857,7 +879,7 @@ public:
                    BitField<7, 1, u32> c7;
                } clip_distance_enabled;

-                INSERT_UNION_PADDING_WORDS(0x1);
+                u32 samplecnt_enable;

                float point_size;

@@ -865,7 +887,11 @@ public:

                u32 point_sprite_enable;

-                INSERT_UNION_PADDING_WORDS(0x5);
+                INSERT_UNION_PADDING_WORDS(0x3);
+
+                CounterReset counter_reset;
+
+                INSERT_UNION_PADDING_WORDS(0x1);

                u32 zeta_enable;

@@ -1412,12 +1438,15 @@ private:
    /// Handles a write to the QUERY_GET register.
    void ProcessQueryGet();

-    // Writes the query result accordingly
+    /// Writes the query result accordingly.
    void StampQueryResult(u64 payload, bool long_query);

-    // Handles Conditional Rendering
+    /// Handles conditional rendering.
    void ProcessQueryCondition();

+    /// Handles counter resets.
+    void ProcessCounterReset();
+
    /// Handles writes to syncing register.
    void ProcessSyncPoint();

@@ -1434,6 +1463,9 @@ private:

    // Handles a instance drawcall from MME
    void StepInstance(MMEDrawMode expected_mode, u32 count);
+
+    /// Returns a query's value or an empty object if the value will be deferred through a cache.
+    std::optional<u64> GetQueryResult();
 };

 #define ASSERT_REG_POSITION(field_name, position)                                                  \
@@ -1499,8 +1531,10 @@ ASSERT_REG_POSITION(screen_y_control, 0x4EB);
 ASSERT_REG_POSITION(vb_element_base, 0x50D);
 ASSERT_REG_POSITION(vb_base_instance, 0x50E);
 ASSERT_REG_POSITION(clip_distance_enabled, 0x544);
+ASSERT_REG_POSITION(samplecnt_enable, 0x545);
 ASSERT_REG_POSITION(point_size, 0x546);
 ASSERT_REG_POSITION(point_sprite_enable, 0x548);
+ASSERT_REG_POSITION(counter_reset, 0x54C);
 ASSERT_REG_POSITION(zeta_enable, 0x54E);
 ASSERT_REG_POSITION(multisample_control, 0x54F);
 ASSERT_REG_POSITION(condition, 0x554);
@@ -9,6 +9,7 @@
 #include "core/hle/kernel/process.h"
 #include "core/hle/kernel/vm_manager.h"
 #include "core/memory.h"
+#include "video_core/gpu.h"
 #include "video_core/memory_manager.h"
 #include "video_core/rasterizer_interface.h"

@@ -84,7 +85,9 @@ GPUVAddr MemoryManager::UnmapBuffer(GPUVAddr gpu_addr, u64 size) {
    const auto cpu_addr = GpuToCpuAddress(gpu_addr);
    ASSERT(cpu_addr);

-    rasterizer.FlushAndInvalidateRegion(cache_addr, aligned_size);
+    // Flush and invalidate through the GPU interface, to be asynchronous if possible.
+    system.GPU().FlushAndInvalidateRegion(cache_addr, aligned_size);
+
    UnmapRange(gpu_addr, aligned_size);
    ASSERT(system.CurrentProcess()
               ->VMManager()
@@ -242,6 +245,8 @@ void MemoryManager::ReadBlock(GPUVAddr src_addr, void* dest_buffer, const std::s
        switch (page_table.attributes[page_index]) {
        case Common::PageType::Memory: {
            const u8* src_ptr{page_table.pointers[page_index] + page_offset};
+            // Flush must happen on the rasterizer interface, such that memory is always synchronous
+            // when it is read (even when in asynchronous GPU mode). Fixes Dead Cells title menu.
            rasterizer.FlushRegion(ToCacheAddr(src_ptr), copy_amount);
            std::memcpy(dest_buffer, src_ptr, copy_amount);
            break;
@@ -292,6 +297,8 @@ void MemoryManager::WriteBlock(GPUVAddr dest_addr, const void* src_buffer, const
        switch (page_table.attributes[page_index]) {
        case Common::PageType::Memory: {
            u8* dest_ptr{page_table.pointers[page_index] + page_offset};
+            // Invalidate must happen on the rasterizer interface, such that memory is always
+            // synchronous when it is written (even when in asynchronous GPU mode).
            rasterizer.InvalidateRegion(ToCacheAddr(dest_ptr), copy_amount);
            std::memcpy(dest_ptr, src_buffer, copy_amount);
            break;
@@ -339,6 +346,8 @@ void MemoryManager::CopyBlock(GPUVAddr dest_addr, GPUVAddr src_addr, const std::

        switch (page_table.attributes[page_index]) {
        case Common::PageType::Memory: {
+            // Flush must happen on the rasterizer interface, such that memory is always synchronous
+            // when it is copied (even when in asynchronous GPU mode).
            const u8* src_ptr{page_table.pointers[page_index] + page_offset};
            rasterizer.FlushRegion(ToCacheAddr(src_ptr), copy_amount);
            WriteBlock(dest_addr, src_ptr, copy_amount);
@@ -0,0 +1,359 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <algorithm>
+#include <array>
+#include <cstring>
+#include <iterator>
+#include <memory>
+#include <mutex>
+#include <optional>
+#include <unordered_map>
+#include <vector>
+
+#include "common/assert.h"
+#include "core/core.h"
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/gpu.h"
+#include "video_core/memory_manager.h"
+#include "video_core/rasterizer_interface.h"
+
+namespace VideoCommon {
+
+template <class QueryCache, class HostCounter>
+class CounterStreamBase {
+public:
+    explicit CounterStreamBase(QueryCache& cache, VideoCore::QueryType type)
+        : cache{cache}, type{type} {}
+
+    /// Updates the state of the stream, enabling or disabling as needed.
+    void Update(bool enabled) {
+        if (enabled) {
+            Enable();
+        } else {
+            Disable();
+        }
+    }
+
+    /// Resets the stream to zero. It doesn't disable the query after resetting.
+    void Reset() {
+        if (current) {
+            current->EndQuery();
+
+            // Immediately start a new query to avoid disabling its state.
+            current = cache.Counter(nullptr, type);
+        }
+        last = nullptr;
+    }
+
+    /// Returns the current counter slicing as needed.
+    std::shared_ptr<HostCounter> Current() {
+        if (!current) {
+            return nullptr;
+        }
+        current->EndQuery();
+        last = std::move(current);
+        current = cache.Counter(last, type);
+        return last;
+    }
+
+    /// Returns true when the counter stream is enabled.
+    bool IsEnabled() const {
+        return current != nullptr;
+    }
+
+private:
+    /// Enables the stream.
+    void Enable() {
+        if (current) {
+            return;
+        }
+        current = cache.Counter(last, type);
+    }
+
+    // Disables the stream.
+    void Disable() {
+        if (current) {
+            current->EndQuery();
+        }
+        last = std::exchange(current, nullptr);
+    }
+
+    QueryCache& cache;
+    const VideoCore::QueryType type;
+
+    std::shared_ptr<HostCounter> current;
+    std::shared_ptr<HostCounter> last;
+};
+
+template <class QueryCache, class CachedQuery, class CounterStream, class HostCounter,
+          class QueryPool>
+class QueryCacheBase {
+public:
+    explicit QueryCacheBase(Core::System& system, VideoCore::RasterizerInterface& rasterizer)
+        : system{system}, rasterizer{rasterizer}, streams{{CounterStream{
+                                                      static_cast<QueryCache&>(*this),
+                                                      VideoCore::QueryType::SamplesPassed}}} {}
+
+    void InvalidateRegion(CacheAddr addr, std::size_t size) {
+        std::unique_lock lock{mutex};
+        FlushAndRemoveRegion(addr, size);
+    }
+
+    void FlushRegion(CacheAddr addr, std::size_t size) {
+        std::unique_lock lock{mutex};
+        FlushAndRemoveRegion(addr, size);
+    }
+
+    /**
+     * Records a query in GPU mapped memory, potentially marked with a timestamp.
+     * @param gpu_addr  GPU address to flush to when the mapped memory is read.
+     * @param type      Query type, e.g. SamplesPassed.
+     * @param timestamp Timestamp, when empty the flushed query is assumed to be short.
+     */
+    void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) {
+        std::unique_lock lock{mutex};
+        auto& memory_manager = system.GPU().MemoryManager();
+        const auto host_ptr = memory_manager.GetPointer(gpu_addr);
+
+        CachedQuery* query = TryGet(ToCacheAddr(host_ptr));
+        if (!query) {
+            const auto cpu_addr = memory_manager.GpuToCpuAddress(gpu_addr);
+            ASSERT_OR_EXECUTE(cpu_addr, return;);
+
+            query = Register(type, *cpu_addr, host_ptr, timestamp.has_value());
+        }
+
+        query->BindCounter(Stream(type).Current(), timestamp);
+    }
+
+    /// Updates counters from GPU state. Expected to be called once per draw, clear or dispatch.
+    void UpdateCounters() {
+        std::unique_lock lock{mutex};
+        const auto& regs = system.GPU().Maxwell3D().regs;
+        Stream(VideoCore::QueryType::SamplesPassed).Update(regs.samplecnt_enable);
+    }
+
+    /// Resets a counter to zero. It doesn't disable the query after resetting.
+    void ResetCounter(VideoCore::QueryType type) {
+        std::unique_lock lock{mutex};
+        Stream(type).Reset();
+    }
+
+    /// Disable all active streams. Expected to be called at the end of a command buffer.
+    void DisableStreams() {
+        std::unique_lock lock{mutex};
+        for (auto& stream : streams) {
+            stream.Update(false);
+        }
+    }
+
+    /// Returns a new host counter.
+    std::shared_ptr<HostCounter> Counter(std::shared_ptr<HostCounter> dependency,
+                                         VideoCore::QueryType type) {
+        return std::make_shared<HostCounter>(static_cast<QueryCache&>(*this), std::move(dependency),
+                                             type);
+    }
+
+    /// Returns the counter stream of the specified type.
+    CounterStream& Stream(VideoCore::QueryType type) {
+        return streams[static_cast<std::size_t>(type)];
+    }
+
+    /// Returns the counter stream of the specified type.
+    const CounterStream& Stream(VideoCore::QueryType type) const {
+        return streams[static_cast<std::size_t>(type)];
+    }
+
+protected:
+    std::array<QueryPool, VideoCore::NumQueryTypes> query_pools;
+
+private:
+    /// Flushes a memory range to guest memory and removes it from the cache.
+    void FlushAndRemoveRegion(CacheAddr addr, std::size_t size) {
+        const u64 addr_begin = static_cast<u64>(addr);
+        const u64 addr_end = addr_begin + static_cast<u64>(size);
+        const auto in_range = [addr_begin, addr_end](CachedQuery& query) {
+            const u64 cache_begin = query.GetCacheAddr();
+            const u64 cache_end = cache_begin + query.SizeInBytes();
+            return cache_begin < addr_end && addr_begin < cache_end;
+        };
+
+        const u64 page_end = addr_end >> PAGE_SHIFT;
+        for (u64 page = addr_begin >> PAGE_SHIFT; page <= page_end; ++page) {
+            const auto& it = cached_queries.find(page);
+            if (it == std::end(cached_queries)) {
+                continue;
+            }
+            auto& contents = it->second;
+            for (auto& query : contents) {
+                if (!in_range(query)) {
+                    continue;
+                }
+                rasterizer.UpdatePagesCachedCount(query.CpuAddr(), query.SizeInBytes(), -1);
+                query.Flush();
+            }
+            contents.erase(std::remove_if(std::begin(contents), std::end(contents), in_range),
+                           std::end(contents));
+        }
+    }
+
+    /// Registers the passed parameters as cached and returns a pointer to the stored cached query.
+    CachedQuery* Register(VideoCore::QueryType type, VAddr cpu_addr, u8* host_ptr, bool timestamp) {
+        rasterizer.UpdatePagesCachedCount(cpu_addr, CachedQuery::SizeInBytes(timestamp), 1);
+        const u64 page = static_cast<u64>(ToCacheAddr(host_ptr)) >> PAGE_SHIFT;
+        return &cached_queries[page].emplace_back(static_cast<QueryCache&>(*this), type, cpu_addr,
+                                                  host_ptr);
+    }
+
+    /// Tries to a get a cached query. Returns nullptr on failure.
+    CachedQuery* TryGet(CacheAddr addr) {
+        const u64 page = static_cast<u64>(addr) >> PAGE_SHIFT;
+        const auto it = cached_queries.find(page);
+        if (it == std::end(cached_queries)) {
+            return nullptr;
+        }
+        auto& contents = it->second;
+        const auto found =
+            std::find_if(std::begin(contents), std::end(contents),
+                         [addr](auto& query) { return query.GetCacheAddr() == addr; });
+        return found != std::end(contents) ? &*found : nullptr;
+    }
+
+    static constexpr std::uintptr_t PAGE_SIZE = 4096;
+    static constexpr unsigned PAGE_SHIFT = 12;
+
+    Core::System& system;
+    VideoCore::RasterizerInterface& rasterizer;
+
+    std::recursive_mutex mutex;
+
+    std::unordered_map<u64, std::vector<CachedQuery>> cached_queries;
+
+    std::array<CounterStream, VideoCore::NumQueryTypes> streams;
+};
+
+template <class QueryCache, class HostCounter>
+class HostCounterBase {
+public:
+    explicit HostCounterBase(std::shared_ptr<HostCounter> dependency_)
+        : dependency{std::move(dependency_)}, depth{dependency ? (dependency->Depth() + 1) : 0} {
+        // Avoid nesting too many dependencies to avoid a stack overflow when these are deleted.
+        constexpr u64 depth_threshold = 96;
+        if (depth > depth_threshold) {
+            depth = 0;
+            base_result = dependency->Query();
+            dependency = nullptr;
+        }
+    }
+    virtual ~HostCounterBase() = default;
+
+    /// Returns the current value of the query.
+    u64 Query() {
+        if (result) {
+            return *result;
+        }
+
+        u64 value = BlockingQuery() + base_result;
+        if (dependency) {
+            value += dependency->Query();
+            dependency = nullptr;
+        }
+
+        result = value;
+        return *result;
+    }
+
+    /// Returns true when flushing this query will potentially wait.
+    bool WaitPending() const noexcept {
+        return result.has_value();
+    }
+
+    u64 Depth() const noexcept {
+        return depth;
+    }
+
+protected:
+    /// Returns the value of query from the backend API blocking as needed.
+    virtual u64 BlockingQuery() const = 0;
+
+private:
+    std::shared_ptr<HostCounter> dependency; ///< Counter to add to this value.
+    std::optional<u64> result;               ///< Filled with the already returned value.
+    u64 depth;                               ///< Number of nested dependencies.
+    u64 base_result = 0;                     ///< Equivalent to nested dependencies value.
+};
+
+template <class HostCounter>
+class CachedQueryBase {
+public:
+    explicit CachedQueryBase(VAddr cpu_addr, u8* host_ptr)
+        : cpu_addr{cpu_addr}, host_ptr{host_ptr} {}
+    virtual ~CachedQueryBase() = default;
+
+    CachedQueryBase(CachedQueryBase&&) noexcept = default;
+    CachedQueryBase(const CachedQueryBase&) = delete;
+
+    CachedQueryBase& operator=(CachedQueryBase&&) noexcept = default;
+    CachedQueryBase& operator=(const CachedQueryBase&) = delete;
+
+    /// Flushes the query to guest memory.
+    virtual void Flush() {
+        // When counter is nullptr it means that it's just been reseted. We are supposed to write a
+        // zero in these cases.
+        const u64 value = counter ? counter->Query() : 0;
+        std::memcpy(host_ptr, &value, sizeof(u64));
+
+        if (timestamp) {
+            std::memcpy(host_ptr + TIMESTAMP_OFFSET, &*timestamp, sizeof(u64));
+        }
+    }
+
+    /// Binds a counter to this query.
+    void BindCounter(std::shared_ptr<HostCounter> counter_, std::optional<u64> timestamp_) {
+        if (counter) {
+            // If there's an old counter set it means the query is being rewritten by the game.
+            // To avoid losing the data forever, flush here.
+            Flush();
+        }
+        counter = std::move(counter_);
+        timestamp = timestamp_;
+    }
+
+    VAddr CpuAddr() const noexcept {
+        return cpu_addr;
+    }
+
+    CacheAddr GetCacheAddr() const noexcept {
+        return ToCacheAddr(host_ptr);
+    }
+
+    u64 SizeInBytes() const noexcept {
+        return SizeInBytes(timestamp.has_value());
+    }
+
+    static constexpr u64 SizeInBytes(bool with_timestamp) noexcept {
+        return with_timestamp ? LARGE_QUERY_SIZE : SMALL_QUERY_SIZE;
+    }
+
+protected:
+    /// Returns true when querying the counter may potentially block.
+    bool WaitPending() const noexcept {
+        return counter && counter->WaitPending();
+    }
+
+private:
+    static constexpr std::size_t SMALL_QUERY_SIZE = 8;   // Query size without timestamp.
+    static constexpr std::size_t LARGE_QUERY_SIZE = 16;  // Query size with timestamp.
+    static constexpr std::intptr_t TIMESTAMP_OFFSET = 8; // Timestamp offset in a large query.
+
+    VAddr cpu_addr;                       ///< Guest CPU address.
+    u8* host_ptr;                         ///< Writable host pointer.
+    std::shared_ptr<HostCounter> counter; ///< Host counter to query, owns the dependency tree.
+    std::optional<u64> timestamp;         ///< Timestamp to flush to guest memory.
+};
+
+} // namespace VideoCommon
@@ -6,6 +6,7 @@

 #include <atomic>
 #include <functional>
+#include <optional>
 #include "common/common_types.h"
 #include "video_core/engines/fermi_2d.h"
 #include "video_core/gpu.h"
@@ -17,6 +18,11 @@ class MemoryManager;

 namespace VideoCore {

+enum class QueryType {
+    SamplesPassed,
+};
+constexpr std::size_t NumQueryTypes = 1;
+
 enum class LoadCallbackStage {
    Prepare,
    Decompile,
@@ -29,11 +35,8 @@ class RasterizerInterface {
 public:
    virtual ~RasterizerInterface() {}

-    /// Draw the current batch of vertex arrays
-    virtual bool DrawBatch(bool is_indexed) = 0;
-
-    /// Draw the current batch of multiple instances of vertex arrays
-    virtual bool DrawMultiBatch(bool is_indexed) = 0;
+    /// Dispatches a draw invocation
+    virtual void Draw(bool is_indexed, bool is_instanced) = 0;

    /// Clear the current framebuffer
    virtual void Clear() = 0;
@@ -41,6 +44,12 @@ public:
    /// Dispatches a compute shader invocation
    virtual void DispatchCompute(GPUVAddr code_addr) = 0;

+    /// Resets the counter of a query
+    virtual void ResetCounter(QueryType type) = 0;
+
+    /// Records a GPU query and caches it
+    virtual void Query(GPUVAddr gpu_addr, QueryType type, std::optional<u64> timestamp) = 0;
+
    /// Notify rasterizer that all caches should be flushed to Switch memory
    virtual void FlushAll() = 0;

@@ -0,0 +1,120 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <algorithm>
+#include <cstring>
+#include <memory>
+#include <unordered_map>
+#include <utility>
+#include <vector>
+
+#include <glad/glad.h>
+
+#include "common/assert.h"
+#include "core/core.h"
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/memory_manager.h"
+#include "video_core/renderer_opengl/gl_query_cache.h"
+#include "video_core/renderer_opengl/gl_rasterizer.h"
+
+namespace OpenGL {
+
+namespace {
+
+constexpr std::array<GLenum, VideoCore::NumQueryTypes> QueryTargets = {GL_SAMPLES_PASSED};
+
+constexpr GLenum GetTarget(VideoCore::QueryType type) {
+    return QueryTargets[static_cast<std::size_t>(type)];
+}
+
+} // Anonymous namespace
+
+QueryCache::QueryCache(Core::System& system, RasterizerOpenGL& gl_rasterizer)
+    : VideoCommon::QueryCacheBase<
+          QueryCache, CachedQuery, CounterStream, HostCounter,
+          std::vector<OGLQuery>>{system,
+                                 static_cast<VideoCore::RasterizerInterface&>(gl_rasterizer)},
+      gl_rasterizer{gl_rasterizer} {}
+
+QueryCache::~QueryCache() = default;
+
+OGLQuery QueryCache::AllocateQuery(VideoCore::QueryType type) {
+    auto& reserve = query_pools[static_cast<std::size_t>(type)];
+    OGLQuery query;
+    if (reserve.empty()) {
+        query.Create(GetTarget(type));
+        return query;
+    }
+
+    query = std::move(reserve.back());
+    reserve.pop_back();
+    return query;
+}
+
+void QueryCache::Reserve(VideoCore::QueryType type, OGLQuery&& query) {
+    query_pools[static_cast<std::size_t>(type)].push_back(std::move(query));
+}
+
+bool QueryCache::AnyCommandQueued() const noexcept {
+    return gl_rasterizer.AnyCommandQueued();
+}
+
+HostCounter::HostCounter(QueryCache& cache, std::shared_ptr<HostCounter> dependency,
+                         VideoCore::QueryType type)
+    : VideoCommon::HostCounterBase<QueryCache, HostCounter>{std::move(dependency)}, cache{cache},
+      type{type}, query{cache.AllocateQuery(type)} {
+    glBeginQuery(GetTarget(type), query.handle);
+}
+
+HostCounter::~HostCounter() {
+    cache.Reserve(type, std::move(query));
+}
+
+void HostCounter::EndQuery() {
+    if (!cache.AnyCommandQueued()) {
+        // There are chances a query waited on without commands (glDraw, glClear, glDispatch). Not
+        // having any of these causes a lock. glFlush is considered a command, so we can safely wait
+        // for this. Insert to the OpenGL command stream a flush.
+        glFlush();
+    }
+    glEndQuery(GetTarget(type));
+}
+
+u64 HostCounter::BlockingQuery() const {
+    GLint64 value;
+    glGetQueryObjecti64v(query.handle, GL_QUERY_RESULT, &value);
+    return static_cast<u64>(value);
+}
+
+CachedQuery::CachedQuery(QueryCache& cache, VideoCore::QueryType type, VAddr cpu_addr, u8* host_ptr)
+    : VideoCommon::CachedQueryBase<HostCounter>{cpu_addr, host_ptr}, cache{&cache}, type{type} {}
+
+CachedQuery::CachedQuery(CachedQuery&& rhs) noexcept
+    : VideoCommon::CachedQueryBase<HostCounter>(std::move(rhs)), cache{rhs.cache}, type{rhs.type} {}
+
+CachedQuery& CachedQuery::operator=(CachedQuery&& rhs) noexcept {
+    VideoCommon::CachedQueryBase<HostCounter>::operator=(std::move(rhs));
+    cache = rhs.cache;
+    type = rhs.type;
+    return *this;
+}
+
+void CachedQuery::Flush() {
+    // Waiting for a query while another query of the same target is enabled locks Nvidia's driver.
+    // To avoid this disable and re-enable keeping the dependency stream.
+    // But we only have to do this if we have pending waits to be done.
+    auto& stream = cache->Stream(type);
+    const bool slice_counter = WaitPending() && stream.IsEnabled();
+    if (slice_counter) {
+        stream.Update(false);
+    }
+
+    VideoCommon::CachedQueryBase<HostCounter>::Flush();
+
+    if (slice_counter) {
+        stream.Update(true);
+    }
+}
+
+} // namespace OpenGL
@@ -0,0 +1,78 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <memory>
+#include <vector>
+
+#include "common/common_types.h"
+#include "video_core/query_cache.h"
+#include "video_core/rasterizer_interface.h"
+#include "video_core/renderer_opengl/gl_resource_manager.h"
+
+namespace Core {
+class System;
+}
+
+namespace OpenGL {
+
+class CachedQuery;
+class HostCounter;
+class QueryCache;
+class RasterizerOpenGL;
+
+using CounterStream = VideoCommon::CounterStreamBase<QueryCache, HostCounter>;
+
+class QueryCache final : public VideoCommon::QueryCacheBase<QueryCache, CachedQuery, CounterStream,
+                                                            HostCounter, std::vector<OGLQuery>> {
+public:
+    explicit QueryCache(Core::System& system, RasterizerOpenGL& rasterizer);
+    ~QueryCache();
+
+    OGLQuery AllocateQuery(VideoCore::QueryType type);
+
+    void Reserve(VideoCore::QueryType type, OGLQuery&& query);
+
+    bool AnyCommandQueued() const noexcept;
+
+private:
+    RasterizerOpenGL& gl_rasterizer;
+};
+
+class HostCounter final : public VideoCommon::HostCounterBase<QueryCache, HostCounter> {
+public:
+    explicit HostCounter(QueryCache& cache, std::shared_ptr<HostCounter> dependency,
+                         VideoCore::QueryType type);
+    ~HostCounter();
+
+    void EndQuery();
+
+private:
+    u64 BlockingQuery() const override;
+
+    QueryCache& cache;
+    const VideoCore::QueryType type;
+    OGLQuery query;
+};
+
+class CachedQuery final : public VideoCommon::CachedQueryBase<HostCounter> {
+public:
+    explicit CachedQuery(QueryCache& cache, VideoCore::QueryType type, VAddr cpu_addr,
+                         u8* host_ptr);
+    CachedQuery(CachedQuery&& rhs) noexcept;
+    CachedQuery(const CachedQuery&) = delete;
+
+    CachedQuery& operator=(CachedQuery&& rhs) noexcept;
+    CachedQuery& operator=(const CachedQuery&) = delete;
+
+    void Flush() override;
+
+private:
+    QueryCache* cache;
+    VideoCore::QueryType type;
+};
+
+} // namespace OpenGL
@@ -25,6 +25,7 @@
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/engines/shader_type.h"
 #include "video_core/memory_manager.h"
+#include "video_core/renderer_opengl/gl_query_cache.h"
 #include "video_core/renderer_opengl/gl_rasterizer.h"
 #include "video_core/renderer_opengl/gl_shader_cache.h"
 #include "video_core/renderer_opengl/gl_shader_gen.h"
@@ -92,8 +93,8 @@ std::size_t GetConstBufferSize(const Tegra::Engines::ConstBufferInfo& buffer,
 RasterizerOpenGL::RasterizerOpenGL(Core::System& system, Core::Frontend::EmuWindow& emu_window,
                                   ScreenInfo& info)
    : RasterizerAccelerated{system.Memory()}, texture_cache{system, *this, device},
-      shader_cache{*this, system, emu_window, device}, system{system}, screen_info{info},
-      buffer_cache{*this, system, device, STREAM_BUFFER_SIZE} {
+      shader_cache{*this, system, emu_window, device}, query_cache{system, *this}, system{system},
+      screen_info{info}, buffer_cache{*this, system, device, STREAM_BUFFER_SIZE} {
    shader_program_manager = std::make_unique<GLShader::ProgramManager>();
    state.draw.shader_program = 0;
    state.Apply();
@@ -541,11 +542,16 @@ void RasterizerOpenGL::Clear() {
    } else if (use_stencil) {
        glClearBufferiv(GL_STENCIL, 0, &regs.clear_stencil);
    }
+
+    ++num_queued_commands;
 }

 void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
    MICROPROFILE_SCOPE(OpenGL_Drawing);
    auto& gpu = system.GPU().Maxwell3D();
+    const auto& regs = gpu.regs;
+
+    query_cache.UpdateCounters();

    SyncRasterizeEnable(state);
    SyncColorMask();
@@ -611,7 +617,7 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {

    // Setup shaders and their used resources.
    texture_cache.GuardSamplers(true);
-    const auto primitive_mode = MaxwellToGL::PrimitiveTopology(gpu.regs.draw.topology);
+    const GLenum primitive_mode = MaxwellToGL::PrimitiveTopology(gpu.regs.draw.topology);
    SetupShaders(primitive_mode);
    texture_cache.GuardSamplers(false);

@@ -638,35 +644,47 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
        glTextureBarrier();
    }

+    ++num_queued_commands;
+
    const GLuint base_instance = static_cast<GLuint>(gpu.regs.vb_base_instance);
    const GLsizei num_instances =
        static_cast<GLsizei>(is_instanced ? gpu.mme_draw.instance_count : 1);
    if (is_indexed) {
-        const GLenum index_format = MaxwellToGL::IndexFormat(gpu.regs.index_array.format);
        const GLint base_vertex = static_cast<GLint>(gpu.regs.vb_element_base);
        const GLsizei num_vertices = static_cast<GLsizei>(gpu.regs.index_array.count);
-        glDrawElementsInstancedBaseVertexBaseInstance(
-            primitive_mode, num_vertices, index_format,
-            reinterpret_cast<const void*>(index_buffer_offset), num_instances, base_vertex,
-            base_instance);
+        const GLvoid* offset = reinterpret_cast<const GLvoid*>(index_buffer_offset);
+        const GLenum format = MaxwellToGL::IndexFormat(gpu.regs.index_array.format);
+        if (num_instances == 1 && base_instance == 0 && base_vertex == 0) {
+            glDrawElements(primitive_mode, num_vertices, format, offset);
+        } else if (num_instances == 1 && base_instance == 0) {
+            glDrawElementsBaseVertex(primitive_mode, num_vertices, format, offset, base_vertex);
+        } else if (base_vertex == 0 && base_instance == 0) {
+            glDrawElementsInstanced(primitive_mode, num_vertices, format, offset, num_instances);
+        } else if (base_vertex == 0) {
+            glDrawElementsInstancedBaseInstance(primitive_mode, num_vertices, format, offset,
+                                                num_instances, base_instance);
+        } else if (base_instance == 0) {
+            glDrawElementsInstancedBaseVertex(primitive_mode, num_vertices, format, offset,
+                                              num_instances, base_vertex);
+        } else {
+            glDrawElementsInstancedBaseVertexBaseInstance(primitive_mode, num_vertices, format,
+                                                          offset, num_instances, base_vertex,
+                                                          base_instance);
+        }
    } else {
        const GLint base_vertex = static_cast<GLint>(gpu.regs.vertex_buffer.first);
        const GLsizei num_vertices = static_cast<GLsizei>(gpu.regs.vertex_buffer.count);
-        glDrawArraysInstancedBaseInstance(primitive_mode, base_vertex, num_vertices, num_instances,
-                                          base_instance);
+        if (num_instances == 1 && base_instance == 0) {
+            glDrawArrays(primitive_mode, base_vertex, num_vertices);
+        } else if (base_instance == 0) {
+            glDrawArraysInstanced(primitive_mode, base_vertex, num_vertices, num_instances);
+        } else {
+            glDrawArraysInstancedBaseInstance(primitive_mode, base_vertex, num_vertices,
+                                              num_instances, base_instance);
+        }
    }
 }

-bool RasterizerOpenGL::DrawBatch(bool is_indexed) {
-    Draw(is_indexed, false);
-    return true;
-}
-
-bool RasterizerOpenGL::DrawMultiBatch(bool is_indexed) {
-    Draw(is_indexed, true);
-    return true;
-}
-
 void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) {
    if (device.HasBrokenCompute()) {
        return;
@@ -707,6 +725,16 @@ void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) {
    state.ApplyProgramPipeline();

    glDispatchCompute(launch_desc.grid_dim_x, launch_desc.grid_dim_y, launch_desc.grid_dim_z);
+    ++num_queued_commands;
+}
+
+void RasterizerOpenGL::ResetCounter(VideoCore::QueryType type) {
+    query_cache.ResetCounter(type);
+}
+
+void RasterizerOpenGL::Query(GPUVAddr gpu_addr, VideoCore::QueryType type,
+                             std::optional<u64> timestamp) {
+    query_cache.Query(gpu_addr, type, timestamp);
 }

 void RasterizerOpenGL::FlushAll() {}
@@ -718,6 +746,7 @@ void RasterizerOpenGL::FlushRegion(CacheAddr addr, u64 size) {
    }
    texture_cache.FlushRegion(addr, size);
    buffer_cache.FlushRegion(addr, size);
+    query_cache.FlushRegion(addr, size);
 }

 void RasterizerOpenGL::InvalidateRegion(CacheAddr addr, u64 size) {
@@ -728,6 +757,7 @@ void RasterizerOpenGL::InvalidateRegion(CacheAddr addr, u64 size) {
    texture_cache.InvalidateRegion(addr, size);
    shader_cache.InvalidateRegion(addr, size);
    buffer_cache.InvalidateRegion(addr, size);
+    query_cache.InvalidateRegion(addr, size);
 }

 void RasterizerOpenGL::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
@@ -738,10 +768,18 @@ void RasterizerOpenGL::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
 }

 void RasterizerOpenGL::FlushCommands() {
+    // Only flush when we have commands queued to OpenGL.
+    if (num_queued_commands == 0) {
+        return;
+    }
+    num_queued_commands = 0;
    glFlush();
 }

 void RasterizerOpenGL::TickFrame() {
+    // Ticking a frame means that buffers will be swapped, calling glFlush implicitly.
+    num_queued_commands = 0;
+
    buffer_cache.TickFrame();
 }

@@ -24,6 +24,7 @@
 #include "video_core/renderer_opengl/gl_buffer_cache.h"
 #include "video_core/renderer_opengl/gl_device.h"
 #include "video_core/renderer_opengl/gl_framebuffer_cache.h"
+#include "video_core/renderer_opengl/gl_query_cache.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
 #include "video_core/renderer_opengl/gl_sampler_cache.h"
 #include "video_core/renderer_opengl/gl_shader_cache.h"
@@ -57,10 +58,11 @@ public:
                              ScreenInfo& info);
    ~RasterizerOpenGL() override;

-    bool DrawBatch(bool is_indexed) override;
-    bool DrawMultiBatch(bool is_indexed) override;
+    void Draw(bool is_indexed, bool is_instanced) override;
    void Clear() override;
    void DispatchCompute(GPUVAddr code_addr) override;
+    void ResetCounter(VideoCore::QueryType type) override;
+    void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) override;
    void FlushAll() override;
    void FlushRegion(CacheAddr addr, u64 size) override;
    void InvalidateRegion(CacheAddr addr, u64 size) override;
@@ -75,6 +77,11 @@ public:
    void LoadDiskResources(const std::atomic_bool& stop_loading,
                           const VideoCore::DiskResourceLoadCallback& callback) override;

+    /// Returns true when there are commands queued to the OpenGL server.
+    bool AnyCommandQueued() const {
+        return num_queued_commands > 0;
+    }
+
 private:
    /// Configures the color and depth framebuffer states.
    void ConfigureFramebuffers();
@@ -102,9 +109,6 @@ private:
    void SetupGlobalMemory(u32 binding, const GLShader::GlobalMemoryEntry& entry, GPUVAddr gpu_addr,
                           std::size_t size);

-    /// Syncs all the state, shaders, render targets and textures setting before a draw call.
-    void Draw(bool is_indexed, bool is_instanced);
-
    /// Configures the current textures to use for the draw command.
    void SetupDrawTextures(std::size_t stage_index, const Shader& shader);

@@ -180,10 +184,23 @@ private:
    /// Syncs the alpha test state to match the guest state
    void SyncAlphaTest();

-    /// Check for extension that are not strictly required
-    /// but are needed for correct emulation
+    /// Check for extension that are not strictly required but are needed for correct emulation
    void CheckExtensions();

+    std::size_t CalculateVertexArraysSize() const;
+
+    std::size_t CalculateIndexBufferSize() const;
+
+    /// Updates and returns a vertex array object representing current vertex format
+    GLuint SetupVertexFormat();
+
+    void SetupVertexBuffer(GLuint vao);
+    void SetupVertexInstances(GLuint vao);
+
+    GLintptr SetupIndexBuffer();
+
+    void SetupShaders(GLenum primitive_mode);
+
    const Device device;
    OpenGLState state;

@@ -191,6 +208,7 @@ private:
    ShaderCacheOpenGL shader_cache;
    SamplerCacheOpenGL sampler_cache;
    FramebufferCacheOpenGL framebuffer_cache;
+    QueryCache query_cache;

    Core::System& system;
    ScreenInfo& screen_info;
@@ -208,19 +226,8 @@ private:
    BindBuffersRangePushBuffer bind_ubo_pushbuffer{GL_UNIFORM_BUFFER};
    BindBuffersRangePushBuffer bind_ssbo_pushbuffer{GL_SHADER_STORAGE_BUFFER};

-    std::size_t CalculateVertexArraysSize() const;
-
-    std::size_t CalculateIndexBufferSize() const;
-
-    /// Updates and returns a vertex array object representing current vertex format
-    GLuint SetupVertexFormat();
-
-    void SetupVertexBuffer(GLuint vao);
-    void SetupVertexInstances(GLuint vao);
-
-    GLintptr SetupIndexBuffer();
-
-    void SetupShaders(GLenum primitive_mode);
+    /// Number of commands queued to the OpenGL driver. Reseted on flush.
+    std::size_t num_queued_commands = 0;
 };

 } // namespace OpenGL
@@ -207,4 +207,21 @@ void OGLFramebuffer::Release() {
    handle = 0;
 }

+void OGLQuery::Create(GLenum target) {
+    if (handle != 0)
+        return;
+
+    MICROPROFILE_SCOPE(OpenGL_ResourceCreation);
+    glCreateQueries(target, 1, &handle);
+}
+
+void OGLQuery::Release() {
+    if (handle == 0)
+        return;
+
+    MICROPROFILE_SCOPE(OpenGL_ResourceDeletion);
+    glDeleteQueries(1, &handle);
+    handle = 0;
+}
+
 } // namespace OpenGL
@@ -266,4 +266,29 @@ public:
    GLuint handle = 0;
 };

+class OGLQuery : private NonCopyable {
+public:
+    OGLQuery() = default;
+
+    OGLQuery(OGLQuery&& o) noexcept : handle(std::exchange(o.handle, 0)) {}
+
+    ~OGLQuery() {
+        Release();
+    }
+
+    OGLQuery& operator=(OGLQuery&& o) noexcept {
+        Release();
+        handle = std::exchange(o.handle, 0);
+        return *this;
+    }
+
+    /// Creates a new internal OpenGL resource and stores the handle
+    void Create(GLenum target);
+
+    /// Deletes the internal OpenGL resource
+    void Release();
+
+    GLuint handle = 0;
+};
+
 } // namespace OpenGL
@@ -164,7 +164,7 @@ struct FormatTuple {
    {vk::Format::eUndefined, {}},                                // ASTC_2D_5X4
    {vk::Format::eUndefined, {}},                                // BGRA8_SRGB
    {vk::Format::eBc1RgbaSrgbBlock, {}},                         // DXT1_SRGB
-    {vk::Format::eUndefined, {}},                                // DXT23_SRGB
+    {vk::Format::eBc2SrgbBlock, {}},                             // DXT23_SRGB
    {vk::Format::eBc3SrgbBlock, {}},                             // DXT45_SRGB
    {vk::Format::eBc7SrgbBlock, {}},                             // BC7U_SRGB
    {vk::Format::eR4G4B4A4UnormPack16, Attachable},              // R4G4B4A4U
@@ -363,6 +363,8 @@ vk::Format VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttr
            return vk::Format::eR8G8B8A8Uint;
        case Maxwell::VertexAttribute::Size::Size_32:
            return vk::Format::eR32Uint;
+        case Maxwell::VertexAttribute::Size::Size_32_32_32_32:
+            return vk::Format::eR32G32B32A32Uint;
        default:
            break;
        }
@@ -104,6 +104,7 @@ bool VKDevice::Create(const vk::DispatchLoaderDynamic& dldi, vk::Instance instan
    features.depthBiasClamp = true;
    features.geometryShader = true;
    features.tessellationShader = true;
+    features.occlusionQueryPrecise = true;
    features.fragmentStoresAndAtomics = true;
    features.shaderImageGatherExtended = true;
    features.shaderStorageImageReadWithoutFormat =
@@ -119,6 +120,10 @@ bool VKDevice::Create(const vk::DispatchLoaderDynamic& dldi, vk::Instance instan
    bit8_storage.uniformAndStorageBuffer8BitAccess = true;
    SetNext(next, bit8_storage);

+    vk::PhysicalDeviceHostQueryResetFeaturesEXT host_query_reset;
+    host_query_reset.hostQueryReset = true;
+    SetNext(next, host_query_reset);
+
    vk::PhysicalDeviceFloat16Int8FeaturesKHR float16_int8;
    if (is_float16_supported) {
        float16_int8.shaderFloat16 = true;
@@ -275,6 +280,7 @@ bool VKDevice::IsSuitable(const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDev
        VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME,
        VK_EXT_SHADER_SUBGROUP_BALLOT_EXTENSION_NAME,
        VK_EXT_SHADER_SUBGROUP_VOTE_EXTENSION_NAME,
+        VK_EXT_HOST_QUERY_RESET_EXTENSION_NAME,
    };
    std::bitset<required_extensions.size()> available_extensions{};

@@ -342,6 +348,7 @@ bool VKDevice::IsSuitable(const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDev
        std::make_pair(features.depthBiasClamp, "depthBiasClamp"),
        std::make_pair(features.geometryShader, "geometryShader"),
        std::make_pair(features.tessellationShader, "tessellationShader"),
+        std::make_pair(features.occlusionQueryPrecise, "occlusionQueryPrecise"),
        std::make_pair(features.fragmentStoresAndAtomics, "fragmentStoresAndAtomics"),
        std::make_pair(features.shaderImageGatherExtended, "shaderImageGatherExtended"),
        std::make_pair(features.shaderStorageImageWriteWithoutFormat,
@@ -378,7 +385,7 @@ std::vector<const char*> VKDevice::LoadExtensions(const vk::DispatchLoaderDynami
        }
    };

-    extensions.reserve(13);
+    extensions.reserve(14);
    extensions.push_back(VK_KHR_SWAPCHAIN_EXTENSION_NAME);
    extensions.push_back(VK_KHR_16BIT_STORAGE_EXTENSION_NAME);
    extensions.push_back(VK_KHR_8BIT_STORAGE_EXTENSION_NAME);
@@ -386,6 +393,7 @@ std::vector<const char*> VKDevice::LoadExtensions(const vk::DispatchLoaderDynami
    extensions.push_back(VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME);
    extensions.push_back(VK_EXT_SHADER_SUBGROUP_BALLOT_EXTENSION_NAME);
    extensions.push_back(VK_EXT_SHADER_SUBGROUP_VOTE_EXTENSION_NAME);
+    extensions.push_back(VK_EXT_HOST_QUERY_RESET_EXTENSION_NAME);

    [[maybe_unused]] const bool nsight =
        std::getenv("NVTX_INJECTION64_PATH") || std::getenv("NSIGHT_LAUNCHED");
@@ -534,6 +542,7 @@ std::unordered_map<vk::Format, vk::FormatProperties> VKDevice::GetFormatProperti
                                        vk::Format::eBc6HUfloatBlock,
                                        vk::Format::eBc6HSfloatBlock,
                                        vk::Format::eBc1RgbaSrgbBlock,
+                                        vk::Format::eBc2SrgbBlock,
                                        vk::Format::eBc3SrgbBlock,
                                        vk::Format::eBc7SrgbBlock,
                                        vk::Format::eAstc4x4SrgbBlock,
@@ -0,0 +1,122 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <algorithm>
+#include <cstddef>
+#include <cstdint>
+#include <utility>
+#include <vector>
+
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/vk_device.h"
+#include "video_core/renderer_vulkan/vk_query_cache.h"
+#include "video_core/renderer_vulkan/vk_resource_manager.h"
+#include "video_core/renderer_vulkan/vk_scheduler.h"
+
+namespace Vulkan {
+
+namespace {
+
+constexpr std::array QUERY_TARGETS = {vk::QueryType::eOcclusion};
+
+constexpr vk::QueryType GetTarget(VideoCore::QueryType type) {
+    return QUERY_TARGETS[static_cast<std::size_t>(type)];
+}
+
+} // Anonymous namespace
+
+QueryPool::QueryPool() : VKFencedPool{GROW_STEP} {}
+
+QueryPool::~QueryPool() = default;
+
+void QueryPool::Initialize(const VKDevice& device_, VideoCore::QueryType type_) {
+    device = &device_;
+    type = type_;
+}
+
+std::pair<vk::QueryPool, std::uint32_t> QueryPool::Commit(VKFence& fence) {
+    std::size_t index;
+    do {
+        index = CommitResource(fence);
+    } while (usage[index]);
+    usage[index] = true;
+
+    return {*pools[index / GROW_STEP], static_cast<std::uint32_t>(index % GROW_STEP)};
+}
+
+void QueryPool::Allocate(std::size_t begin, std::size_t end) {
+    usage.resize(end);
+
+    const auto dev = device->GetLogical();
+    const u32 size = static_cast<u32>(end - begin);
+    const vk::QueryPoolCreateInfo query_pool_ci({}, GetTarget(type), size, {});
+    pools.push_back(dev.createQueryPoolUnique(query_pool_ci, nullptr, device->GetDispatchLoader()));
+}
+
+void QueryPool::Reserve(std::pair<vk::QueryPool, std::uint32_t> query) {
+    const auto it =
+        std::find_if(std::begin(pools), std::end(pools),
+                     [query_pool = query.first](auto& pool) { return query_pool == *pool; });
+    ASSERT(it != std::end(pools));
+
+    const std::ptrdiff_t pool_index = std::distance(std::begin(pools), it);
+    usage[pool_index * GROW_STEP + static_cast<std::ptrdiff_t>(query.second)] = false;
+}
+
+VKQueryCache::VKQueryCache(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
+                           const VKDevice& device, VKScheduler& scheduler)
+    : VideoCommon::QueryCacheBase<VKQueryCache, CachedQuery, CounterStream, HostCounter,
+                                  QueryPool>{system, rasterizer},
+      device{device}, scheduler{scheduler} {
+    for (std::size_t i = 0; i < static_cast<std::size_t>(VideoCore::NumQueryTypes); ++i) {
+        query_pools[i].Initialize(device, static_cast<VideoCore::QueryType>(i));
+    }
+}
+
+VKQueryCache::~VKQueryCache() = default;
+
+std::pair<vk::QueryPool, std::uint32_t> VKQueryCache::AllocateQuery(VideoCore::QueryType type) {
+    return query_pools[static_cast<std::size_t>(type)].Commit(scheduler.GetFence());
+}
+
+void VKQueryCache::Reserve(VideoCore::QueryType type,
+                           std::pair<vk::QueryPool, std::uint32_t> query) {
+    query_pools[static_cast<std::size_t>(type)].Reserve(query);
+}
+
+HostCounter::HostCounter(VKQueryCache& cache, std::shared_ptr<HostCounter> dependency,
+                         VideoCore::QueryType type)
+    : VideoCommon::HostCounterBase<VKQueryCache, HostCounter>{std::move(dependency)}, cache{cache},
+      type{type}, query{cache.AllocateQuery(type)}, ticks{cache.Scheduler().Ticks()} {
+    const auto dev = cache.Device().GetLogical();
+    cache.Scheduler().Record([dev, query = query](vk::CommandBuffer cmdbuf, auto& dld) {
+        dev.resetQueryPoolEXT(query.first, query.second, 1, dld);
+        cmdbuf.beginQuery(query.first, query.second, vk::QueryControlFlagBits::ePrecise, dld);
+    });
+}
+
+HostCounter::~HostCounter() {
+    cache.Reserve(type, query);
+}
+
+void HostCounter::EndQuery() {
+    cache.Scheduler().Record([query = query](auto cmdbuf, auto& dld) {
+        cmdbuf.endQuery(query.first, query.second, dld);
+    });
+}
+
+u64 HostCounter::BlockingQuery() const {
+    if (ticks >= cache.Scheduler().Ticks()) {
+        cache.Scheduler().Flush();
+    }
+
+    const auto dev = cache.Device().GetLogical();
+    const auto& dld = cache.Device().GetDispatchLoader();
+    u64 value;
+    dev.getQueryPoolResults(query.first, query.second, 1, sizeof(value), &value, sizeof(value),
+                            vk::QueryResultFlagBits::e64 | vk::QueryResultFlagBits::eWait, dld);
+    return value;
+}
+
+} // namespace Vulkan
@@ -0,0 +1,104 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <cstddef>
+#include <cstdint>
+#include <memory>
+#include <utility>
+#include <vector>
+
+#include "common/common_types.h"
+#include "video_core/query_cache.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/vk_resource_manager.h"
+
+namespace VideoCore {
+class RasterizerInterface;
+}
+
+namespace Vulkan {
+
+class CachedQuery;
+class HostCounter;
+class VKDevice;
+class VKQueryCache;
+class VKScheduler;
+
+using CounterStream = VideoCommon::CounterStreamBase<VKQueryCache, HostCounter>;
+
+class QueryPool final : public VKFencedPool {
+public:
+    explicit QueryPool();
+    ~QueryPool() override;
+
+    void Initialize(const VKDevice& device, VideoCore::QueryType type);
+
+    std::pair<vk::QueryPool, std::uint32_t> Commit(VKFence& fence);
+
+    void Reserve(std::pair<vk::QueryPool, std::uint32_t> query);
+
+protected:
+    void Allocate(std::size_t begin, std::size_t end) override;
+
+private:
+    static constexpr std::size_t GROW_STEP = 512;
+
+    const VKDevice* device = nullptr;
+    VideoCore::QueryType type = {};
+
+    std::vector<UniqueQueryPool> pools;
+    std::vector<bool> usage;
+};
+
+class VKQueryCache final
+    : public VideoCommon::QueryCacheBase<VKQueryCache, CachedQuery, CounterStream, HostCounter,
+                                         QueryPool> {
+public:
+    explicit VKQueryCache(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
+                          const VKDevice& device, VKScheduler& scheduler);
+    ~VKQueryCache();
+
+    std::pair<vk::QueryPool, std::uint32_t> AllocateQuery(VideoCore::QueryType type);
+
+    void Reserve(VideoCore::QueryType type, std::pair<vk::QueryPool, std::uint32_t> query);
+
+    const VKDevice& Device() const noexcept {
+        return device;
+    }
+
+    VKScheduler& Scheduler() const noexcept {
+        return scheduler;
+    }
+
+private:
+    const VKDevice& device;
+    VKScheduler& scheduler;
+};
+
+class HostCounter final : public VideoCommon::HostCounterBase<VKQueryCache, HostCounter> {
+public:
+    explicit HostCounter(VKQueryCache& cache, std::shared_ptr<HostCounter> dependency,
+                         VideoCore::QueryType type);
+    ~HostCounter();
+
+    void EndQuery();
+
+private:
+    u64 BlockingQuery() const override;
+
+    VKQueryCache& cache;
+    const VideoCore::QueryType type;
+    const std::pair<vk::QueryPool, std::uint32_t> query;
+    const u64 ticks;
+};
+
+class CachedQuery : public VideoCommon::CachedQueryBase<HostCounter> {
+public:
+    explicit CachedQuery(VKQueryCache&, VideoCore::QueryType, VAddr cpu_addr, u8* host_ptr)
+        : VideoCommon::CachedQueryBase<HostCounter>{cpu_addr, host_ptr} {}
+};
+
+} // namespace Vulkan
@@ -289,25 +289,19 @@ RasterizerVulkan::RasterizerVulkan(Core::System& system, Core::Frontend::EmuWind
                    staging_pool),
      pipeline_cache(system, *this, device, scheduler, descriptor_pool, update_descriptor_queue),
      buffer_cache(*this, system, device, memory_manager, scheduler, staging_pool),
-      sampler_cache(device) {}
+      sampler_cache(device), query_cache(system, *this, device, scheduler) {
+    scheduler.SetQueryCache(query_cache);
+}

 RasterizerVulkan::~RasterizerVulkan() = default;

-bool RasterizerVulkan::DrawBatch(bool is_indexed) {
-    Draw(is_indexed, false);
-    return true;
-}
-
-bool RasterizerVulkan::DrawMultiBatch(bool is_indexed) {
-    Draw(is_indexed, true);
-    return true;
-}
-
 void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) {
    MICROPROFILE_SCOPE(Vulkan_Drawing);

    FlushWork();

+    query_cache.UpdateCounters();
+
    const auto& gpu = system.GPU().Maxwell3D();
    GraphicsPipelineCacheKey key{GetFixedPipelineState(gpu.regs)};

@@ -362,6 +356,8 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) {
 void RasterizerVulkan::Clear() {
    MICROPROFILE_SCOPE(Vulkan_Clearing);

+    query_cache.UpdateCounters();
+
    const auto& gpu = system.GPU().Maxwell3D();
    if (!system.GPU().Maxwell3D().ShouldExecute()) {
        return;
@@ -429,6 +425,8 @@ void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) {
    sampled_views.clear();
    image_views.clear();

+    query_cache.UpdateCounters();
+
    const auto& launch_desc = system.GPU().KeplerCompute().launch_description;
    const ComputePipelineCacheKey key{
        code_addr,
@@ -471,17 +469,28 @@ void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) {
    });
 }

+void RasterizerVulkan::ResetCounter(VideoCore::QueryType type) {
+    query_cache.ResetCounter(type);
+}
+
+void RasterizerVulkan::Query(GPUVAddr gpu_addr, VideoCore::QueryType type,
+                             std::optional<u64> timestamp) {
+    query_cache.Query(gpu_addr, type, timestamp);
+}
+
 void RasterizerVulkan::FlushAll() {}

 void RasterizerVulkan::FlushRegion(CacheAddr addr, u64 size) {
    texture_cache.FlushRegion(addr, size);
    buffer_cache.FlushRegion(addr, size);
+    query_cache.FlushRegion(addr, size);
 }

 void RasterizerVulkan::InvalidateRegion(CacheAddr addr, u64 size) {
    texture_cache.InvalidateRegion(addr, size);
    pipeline_cache.InvalidateRegion(addr, size);
    buffer_cache.InvalidateRegion(addr, size);
+    query_cache.InvalidateRegion(addr, size);
 }

 void RasterizerVulkan::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
@@ -24,6 +24,7 @@
 #include "video_core/renderer_vulkan/vk_descriptor_pool.h"
 #include "video_core/renderer_vulkan/vk_memory_manager.h"
 #include "video_core/renderer_vulkan/vk_pipeline_cache.h"
+#include "video_core/renderer_vulkan/vk_query_cache.h"
 #include "video_core/renderer_vulkan/vk_renderpass_cache.h"
 #include "video_core/renderer_vulkan/vk_resource_manager.h"
 #include "video_core/renderer_vulkan/vk_sampler_cache.h"
@@ -96,7 +97,7 @@ struct ImageView {
    vk::ImageLayout* layout = nullptr;
 };

-class RasterizerVulkan : public VideoCore::RasterizerAccelerated {
+class RasterizerVulkan final : public VideoCore::RasterizerAccelerated {
 public:
    explicit RasterizerVulkan(Core::System& system, Core::Frontend::EmuWindow& render_window,
                              VKScreenInfo& screen_info, const VKDevice& device,
@@ -104,10 +105,11 @@ public:
                              VKScheduler& scheduler);
    ~RasterizerVulkan() override;

-    bool DrawBatch(bool is_indexed) override;
-    bool DrawMultiBatch(bool is_indexed) override;
+    void Draw(bool is_indexed, bool is_instanced) override;
    void Clear() override;
    void DispatchCompute(GPUVAddr code_addr) override;
+    void ResetCounter(VideoCore::QueryType type) override;
+    void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) override;
    void FlushAll() override;
    void FlushRegion(CacheAddr addr, u64 size) override;
    void InvalidateRegion(CacheAddr addr, u64 size) override;
@@ -140,8 +142,6 @@ private:

    static constexpr std::size_t ZETA_TEXCEPTION_INDEX = 8;

-    void Draw(bool is_indexed, bool is_instanced);
-
    void FlushWork();

    Texceptions UpdateAttachments();
@@ -247,6 +247,7 @@ private:
    VKPipelineCache pipeline_cache;
    VKBufferCache buffer_cache;
    VKSamplerCache sampler_cache;
+    VKQueryCache query_cache;

    std::array<View, Maxwell::NumRenderTargets> color_attachments;
    View zeta_attachment;
@@ -23,7 +23,14 @@ static std::optional<vk::BorderColor> TryConvertBorderColor(std::array<float, 4>
    } else if (color == std::array<float, 4>{1, 1, 1, 1}) {
        return vk::BorderColor::eFloatOpaqueWhite;
    } else {
-        return {};
+        if (color[0] + color[1] + color[2] > 1.35f) {
+            // If color elements are brighter than roughly 0.5 average, use white border
+            return vk::BorderColor::eFloatOpaqueWhite;
+        }
+        if (color[3] > 0.5f) {
+            return vk::BorderColor::eFloatOpaqueBlack;
+        }
+        return vk::BorderColor::eFloatTransparentBlack;
    }
 }

@@ -37,8 +44,6 @@ UniqueSampler VKSamplerCache::CreateSampler(const Tegra::Texture::TSCEntry& tsc)

    const auto border_color{tsc.GetBorderColor()};
    const auto vk_border_color{TryConvertBorderColor(border_color)};
-    UNIMPLEMENTED_IF_MSG(!vk_border_color, "Unimplemented border color {} {} {} {}",
-                         border_color[0], border_color[1], border_color[2], border_color[3]);

    constexpr bool unnormalized_coords{false};

@@ -6,6 +6,7 @@
 #include "common/microprofile.h"
 #include "video_core/renderer_vulkan/declarations.h"
 #include "video_core/renderer_vulkan/vk_device.h"
+#include "video_core/renderer_vulkan/vk_query_cache.h"
 #include "video_core/renderer_vulkan/vk_resource_manager.h"
 #include "video_core/renderer_vulkan/vk_scheduler.h"

@@ -139,6 +140,8 @@ void VKScheduler::SubmitExecution(vk::Semaphore semaphore) {
 }

 void VKScheduler::AllocateNewContext() {
+    ++ticks;
+
    std::unique_lock lock{mutex};
    current_fence = next_fence;
    next_fence = &resource_manager.CommitFence();
@@ -146,6 +149,10 @@ void VKScheduler::AllocateNewContext() {
    current_cmdbuf = resource_manager.CommitCommandBuffer(*current_fence);
    current_cmdbuf.begin({vk::CommandBufferUsageFlagBits::eOneTimeSubmit},
                         device.GetDispatchLoader());
+    // Enable counters once again. These are disabled when a command buffer is finished.
+    if (query_cache) {
+        query_cache->UpdateCounters();
+    }
 }

 void VKScheduler::InvalidateState() {
@@ -159,6 +166,7 @@ void VKScheduler::InvalidateState() {
 }

 void VKScheduler::EndPendingOperations() {
+    query_cache->DisableStreams();
    EndRenderPass();
 }

@@ -4,6 +4,7 @@

 #pragma once

+#include <atomic>
 #include <condition_variable>
 #include <memory>
 #include <optional>
@@ -18,6 +19,7 @@ namespace Vulkan {

 class VKDevice;
 class VKFence;
+class VKQueryCache;
 class VKResourceManager;

 class VKFenceView {
@@ -67,6 +69,11 @@ public:
    /// Binds a pipeline to the current execution context.
    void BindGraphicsPipeline(vk::Pipeline pipeline);

+    /// Assigns the query cache.
+    void SetQueryCache(VKQueryCache& query_cache_) {
+        query_cache = &query_cache_;
+    }
+
    /// Returns true when viewports have been set in the current command buffer.
    bool TouchViewports() {
        return std::exchange(state.viewports, true);
@@ -112,6 +119,11 @@ public:
        return current_fence;
    }

+    /// Returns the current command buffer tick.
+    u64 Ticks() const {
+        return ticks;
+    }
+
 private:
    class Command {
    public:
@@ -205,6 +217,8 @@ private:

    const VKDevice& device;
    VKResourceManager& resource_manager;
+    VKQueryCache* query_cache = nullptr;
+
    vk::CommandBuffer current_cmdbuf;
    VKFence* current_fence = nullptr;
    VKFence* next_fence = nullptr;
@@ -227,6 +241,7 @@ private:
    Common::SPSCQueue<std::unique_ptr<CommandChunk>> chunk_reserve;
    std::mutex mutex;
    std::condition_variable cv;
+    std::atomic<u64> ticks = 0;
    bool quit = false;
 };

@@ -275,12 +275,14 @@ public:
        AddCapability(spv::Capability::ImageGatherExtended);
        AddCapability(spv::Capability::SampledBuffer);
        AddCapability(spv::Capability::StorageImageWriteWithoutFormat);
+        AddCapability(spv::Capability::DrawParameters);
        AddCapability(spv::Capability::SubgroupBallotKHR);
        AddCapability(spv::Capability::SubgroupVoteKHR);
        AddExtension("SPV_KHR_shader_ballot");
        AddExtension("SPV_KHR_subgroup_vote");
        AddExtension("SPV_KHR_storage_buffer_storage_class");
        AddExtension("SPV_KHR_variable_pointers");
+        AddExtension("SPV_KHR_shader_draw_parameters");

        if (ir.UsesViewportIndex()) {
            AddCapability(spv::Capability::MultiViewport);
@@ -496,9 +498,11 @@ private:
        interfaces.push_back(AddGlobalVariable(Name(out_vertex, "out_vertex")));

        // Declare input attributes
-        vertex_index = DeclareInputBuiltIn(spv::BuiltIn::VertexIndex, t_in_uint, "vertex_index");
+        vertex_index = DeclareInputBuiltIn(spv::BuiltIn::VertexIndex, t_in_int, "vertex_index");
        instance_index =
-            DeclareInputBuiltIn(spv::BuiltIn::InstanceIndex, t_in_uint, "instance_index");
+            DeclareInputBuiltIn(spv::BuiltIn::InstanceIndex, t_in_int, "instance_index");
+        base_vertex = DeclareInputBuiltIn(spv::BuiltIn::BaseVertex, t_in_int, "base_vertex");
+        base_instance = DeclareInputBuiltIn(spv::BuiltIn::BaseInstance, t_in_int, "base_instance");
    }

    void DeclareTessControl() {
@@ -1072,9 +1076,12 @@ private:
                    return {OpLoad(t_float, AccessElement(t_in_float, tess_coord, element)),
                            Type::Float};
                case 2:
-                    return {OpLoad(t_uint, instance_index), Type::Uint};
+                    return {
+                        OpISub(t_int, OpLoad(t_int, instance_index), OpLoad(t_int, base_instance)),
+                        Type::Int};
                case 3:
-                    return {OpLoad(t_uint, vertex_index), Type::Uint};
+                    return {OpISub(t_int, OpLoad(t_int, vertex_index), OpLoad(t_int, base_vertex)),
+                            Type::Int};
                }
                UNIMPLEMENTED_MSG("Unmanaged TessCoordInstanceIDVertexID element={}", element);
                return {Constant(t_uint, 0U), Type::Uint};
@@ -2554,6 +2561,8 @@ private:

    Id instance_index{};
    Id vertex_index{};
+    Id base_instance{};
+    Id base_vertex{};
    std::array<Id, Maxwell::NumRenderTargets> frag_colors{};
    Id frag_depth{};
    Id frag_coord{};
@@ -83,14 +83,14 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {

        const bool input_signed = instr.conversion.is_input_signed;

-        if (instr.conversion.src_size == Register::Size::Byte) {
-            const u32 offset = static_cast<u32>(instr.conversion.int_src.selector) * 8;
-            if (offset > 0) {
-                value = SignedOperation(OperationCode::ILogicalShiftRight, input_signed,
-                                        std::move(value), Immediate(offset));
+        if (const u32 offset = static_cast<u32>(instr.conversion.int_src.selector); offset > 0) {
+            ASSERT(instr.conversion.src_size == Register::Size::Byte ||
+                   instr.conversion.src_size == Register::Size::Short);
+            if (instr.conversion.src_size == Register::Size::Short) {
+                ASSERT(offset == 0 || offset == 2);
            }
-        } else {
-            UNIMPLEMENTED_IF(instr.conversion.int_src.selector != 0);
+            value = SignedOperation(OperationCode::ILogicalShiftRight, input_signed,
+                                    std::move(value), Immediate(offset * 8));
        }

        value = ConvertIntegerSize(value, instr.conversion.src_size, input_signed);
@@ -522,68 +522,53 @@ Node4 ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type,
                               Node array, Node depth_compare, u32 bias_offset,
                               std::vector<Node> aoffi,
                               std::optional<Tegra::Shader::Register> bindless_reg) {
-    const auto is_array = static_cast<bool>(array);
-    const auto is_shadow = static_cast<bool>(depth_compare);
+    const bool is_array = array != nullptr;
+    const bool is_shadow = depth_compare != nullptr;
    const bool is_bindless = bindless_reg.has_value();

-    UNIMPLEMENTED_IF_MSG((texture_type == TextureType::Texture3D && (is_array || is_shadow)) ||
-                             (texture_type == TextureType::TextureCube && is_array && is_shadow),
-                         "This method is not supported.");
+    UNIMPLEMENTED_IF(texture_type == TextureType::TextureCube && is_array && is_shadow);
+    ASSERT_MSG(texture_type != TextureType::Texture3D || !is_array || !is_shadow,
+               "Illegal texture type");

    const SamplerInfo info{texture_type, is_array, is_shadow, false};
-    Node index_var{};
+    Node index_var;
    const Sampler* sampler = is_bindless ? GetBindlessSampler(*bindless_reg, index_var, info)
                                         : GetSampler(instr.sampler, info);
-    Node4 values;
-    if (sampler == nullptr) {
-        for (u32 element = 0; element < values.size(); ++element) {
-            values[element] = Immediate(0);
-        }
-        return values;
+    if (!sampler) {
+        return {Immediate(0), Immediate(0), Immediate(0), Immediate(0)};
    }

    const bool lod_needed = process_mode == TextureProcessMode::LZ ||
                            process_mode == TextureProcessMode::LL ||
                            process_mode == TextureProcessMode::LLA;
-
-    // LOD selection (either via bias or explicit textureLod) not supported in GL for
-    // sampler2DArrayShadow and samplerCubeArrayShadow.
-    const bool gl_lod_supported =
-        !((texture_type == Tegra::Shader::TextureType::Texture2D && is_array && is_shadow) ||
-          (texture_type == Tegra::Shader::TextureType::TextureCube && is_array && is_shadow));
-
-    const OperationCode read_method =
-        (lod_needed && gl_lod_supported) ? OperationCode::TextureLod : OperationCode::Texture;
-
-    UNIMPLEMENTED_IF(process_mode != TextureProcessMode::None && !gl_lod_supported);
+    const OperationCode opcode = lod_needed ? OperationCode::TextureLod : OperationCode::Texture;

    Node bias;
    Node lod;
-    if (process_mode != TextureProcessMode::None && gl_lod_supported) {
-        switch (process_mode) {
-        case TextureProcessMode::LZ:
-            lod = Immediate(0.0f);
-            break;
-        case TextureProcessMode::LB:
-            // If present, lod or bias are always stored in the register
-            // indexed by the gpr20 field with an offset depending on the
-            // usage of the other registers
-            bias = GetRegister(instr.gpr20.Value() + bias_offset);
-            break;
-        case TextureProcessMode::LL:
-            lod = GetRegister(instr.gpr20.Value() + bias_offset);
-            break;
-        default:
-            UNIMPLEMENTED_MSG("Unimplemented process mode={}", static_cast<u32>(process_mode));
-            break;
-        }
+    switch (process_mode) {
+    case TextureProcessMode::None:
+        break;
+    case TextureProcessMode::LZ:
+        lod = Immediate(0.0f);
+        break;
+    case TextureProcessMode::LB:
+        // If present, lod or bias are always stored in the register indexed by the gpr20 field with
+        // an offset depending on the usage of the other registers.
+        bias = GetRegister(instr.gpr20.Value() + bias_offset);
+        break;
+    case TextureProcessMode::LL:
+        lod = GetRegister(instr.gpr20.Value() + bias_offset);
+        break;
+    default:
+        UNIMPLEMENTED_MSG("Unimplemented process mode={}", static_cast<u32>(process_mode));
+        break;
    }

+    Node4 values;
    for (u32 element = 0; element < values.size(); ++element) {
-        auto copy_coords = coords;
        MetaTexture meta{*sampler, array, depth_compare, aoffi,    {}, {}, bias,
                         lod,      {},    element,       index_var};
-        values[element] = Operation(read_method, meta, std::move(copy_coords));
+        values[element] = Operation(opcode, meta, coords);
    }

    return values;
@@ -721,7 +721,6 @@ private:
    std::pair<TSurface, TView> GetSurface(const GPUVAddr gpu_addr, const CacheAddr cache_addr,
                                          const SurfaceParams& params, bool preserve_contents,
                                          bool is_render) {
-
        // Step 1
        // Check Level 1 Cache for a fast structural match. If candidate surface
        // matches at certain level we are pretty much done.
@@ -733,14 +732,18 @@ private:
                return RecycleSurface(overlaps, params, gpu_addr, preserve_contents,
                                      topological_result);
            }
+
            const auto struct_result = current_surface->MatchesStructure(params);
-            if (struct_result != MatchStructureResult::None &&
-                (params.target != SurfaceTarget::Texture3D ||
-                 current_surface->MatchTarget(params.target))) {
-                if (struct_result == MatchStructureResult::FullMatch) {
-                    return ManageStructuralMatch(current_surface, params, is_render);
-                } else {
-                    return RebuildSurface(current_surface, params, is_render);
+            if (struct_result != MatchStructureResult::None) {
+                const auto& old_params = current_surface->GetSurfaceParams();
+                const bool not_3d = params.target != SurfaceTarget::Texture3D &&
+                                    old_params.target != SurfaceTarget::Texture3D;
+                if (not_3d || current_surface->MatchTarget(params.target)) {
+                    if (struct_result == MatchStructureResult::FullMatch) {
+                        return ManageStructuralMatch(current_surface, params, is_render);
+                    } else {
+                        return RebuildSurface(current_surface, params, is_render);
+                    }
                }
            }
        }
@@ -73,14 +73,12 @@ struct Client::Impl {
                if (!parsedUrl.GetPort(&port)) {
                    port = HTTP_PORT;
                }
-                cli = std::make_unique<httplib::Client>(parsedUrl.m_Host.c_str(), port,
-                                                        TIMEOUT_SECONDS);
+                cli = std::make_unique<httplib::Client>(parsedUrl.m_Host.c_str(), port);
            } else if (parsedUrl.m_Scheme == "https") {
                if (!parsedUrl.GetPort(&port)) {
                    port = HTTPS_PORT;
                }
-                cli = std::make_unique<httplib::SSLClient>(parsedUrl.m_Host.c_str(), port,
-                                                           TIMEOUT_SECONDS);
+                cli = std::make_unique<httplib::SSLClient>(parsedUrl.m_Host.c_str(), port);
            } else {
                LOG_ERROR(WebService, "Bad URL scheme {}", parsedUrl.m_Scheme);
                return Common::WebResult{Common::WebResult::Code::InvalidURL, "Bad URL scheme"};
@@ -90,6 +88,7 @@ struct Client::Impl {
            LOG_ERROR(WebService, "Invalid URL {}", host + path);
            return Common::WebResult{Common::WebResult::Code::InvalidURL, "Invalid URL"};
        }
+        cli->set_timeout_sec(TIMEOUT_SECONDS);

        httplib::Headers params;
        if (!jwt.empty()) {
Author	SHA1	Message	Date
bunnei	1989e1b9ac	audio_core: interpolate: Improvements to fix audio crackling. - Fixes audio crackling in Crash Team Racing Nitro-Fueled, Super Mario Odyssey, and others. - Addresses followup issues from #3310.	2020-02-22 22:26:16 -05:00
bunnei	d4da52bbd9	Merge pull request #3444 from bunnei/linux-audio-fix audio_core: interpolate: Fix include for climits (Linux build break).	2020-02-22 03:08:05 -05:00
bunnei	f5cf67140b	audio_core: interpolate: Fix include for climits (Linux build break).	2020-02-22 02:29:41 -05:00
bunnei	19bce3685a	Merge pull request #3310 from FearlessTobi/fast-resample audio_core: Switch to a faster interpolation technique	2020-02-22 01:54:40 -05:00
bunnei	27d57e0c4a	Merge pull request #3442 from ReinUsesLisp/fix-3d-assert shader/texture: Fix illegal 3D texture assert	2020-02-21 22:08:57 -05:00
ReinUsesLisp	7dc488a375	shader/texture: Fix illegal 3D texture assert Fix typo in the illegal 3D texture assert logic. We care about catching arrayed 3D textures or 3D shadow textures, not regular 3D textures.	2020-02-21 15:57:27 -03:00
Rodrigo Locatti	4a6a1aeab4	Merge pull request #3433 from namkazt/patch-1 renderer_vulkan: Add the rest of case for TryConvertBorderColor	2020-02-21 15:56:09 -03:00
Rodrigo Locatti	ef27b4b7b5	Merge pull request #3434 from namkazt/patch-2 vk_shader: Implement ImageLoad	2020-02-21 15:55:05 -03:00
Rodrigo Locatti	6b2719c0bb	Merge pull request #3435 from namkazt/patch-3 vulkan: add DXT23_SRGB	2020-02-21 15:48:19 -03:00
bunnei	dc7ebc2d01	Merge pull request #3423 from ReinUsesLisp/no-match-3d texture_cache: Avoid matches in 3D textures	2020-02-21 12:16:51 -05:00
bunnei	fe8e5d8ae4	Merge pull request #3438 from bunnei/gpu-mem-manager-fix video_core: memory_manager: Flush/invalidate asynchronously when possible.	2020-02-20 20:04:05 -05:00
bunnei	2342c0d50e	Merge pull request #3432 from brianclinkenbeard/update-httplib Update httplib to 0.5.5	2020-02-19 21:15:06 -05:00
bunnei	bf0c929d4c	Merge pull request #3415 from ReinUsesLisp/texture-code shader/texture: Allow 2D shadow arrays and simplify code	2020-02-19 20:06:14 -05:00
bunnei	d65fa7d65c	video_core: memory_manager: Flush/invalidate asynchronously on Unmap. - Minor perf improvement.	2020-02-19 20:03:52 -05:00
Brian Clinkenbeard	d31156931d	fix issue with windows getnameinfo()	2020-02-19 16:16:49 -08:00
bunnei	b2bc7682b4	Merge pull request #3414 from ReinUsesLisp/maxwell-3d-draw maxwell_3d: Unify draw methods	2020-02-19 16:13:50 -05:00
bunnei	c8261a1a57	Merge pull request #3411 from ReinUsesLisp/specific-funcs gl_rasterizer: Use the least generic OpenGL draw function possible	2020-02-19 15:37:41 -05:00
bunnei	fd4c5463e8	Merge pull request #3437 from namkazt/patch-5 shader_conversion: add conversion I2F for Short	2020-02-19 11:27:28 -05:00
Nguyen Dac Nam	1b6308727c	shader_conversion: I2F : add Assert for case src_size is Short	2020-02-19 11:40:35 +07:00
Nguyen Dac Nam	a2c2c5768f	fix warning	2020-02-19 11:10:26 +07:00
Nguyen Dac Nam	a8508f2bc0	clang-format fix	2020-02-19 11:02:59 +07:00
Nguyen Dac Nam	556f3a6e9a	shader_conversion: add conversion I2F for Short	2020-02-19 10:54:37 +07:00
bunnei	e545c2322c	Merge pull request #3410 from ReinUsesLisp/vk-draw-index vk_shader_decompiler: Fix vertex id and instance id	2020-02-18 22:37:33 -05:00
Brian Clinkenbeard	ad4e5c15fb	httplib compatibility	2020-02-18 18:04:33 -08:00
Brian Clinkenbeard	7f6c686d55	update httplib to latest commit	2020-02-18 17:11:40 -08:00
Nguyen Dac Nam	ec206f7f95	fixups mistake auto commit.	2020-02-19 01:24:32 +07:00
Nguyen Dac Nam	eaf60ca5d8	Update code structure Co-Authored-By: Mat M. <mathew1800@gmail.com>	2020-02-19 01:23:08 +07:00
Fernando Sahmkow	93acfbd3a5	Merge pull request #3409 from ReinUsesLisp/host-queries query_cache: Implement a query cache and query 21 (samples passed)	2020-02-18 11:31:06 -04:00
Nguyen Dac Nam	9295966d26	add vertex UnsignedInt size RGBA	2020-02-18 21:52:51 +07:00
Nguyen Dac Nam	9fc42fffd9	add eBc2SrgbBlock to formats	2020-02-18 21:44:09 +07:00
Nguyen Dac Nam	493f0ad904	vulkan: add DXT23_SRGB	2020-02-18 21:39:50 +07:00
Nguyen Dac Nam	ba84f0988f	renderer_vulkan: Add the rest of case for TryConvertBorderColor	2020-02-18 16:52:54 +07:00
Brian Clinkenbeard	9e42025e5b	update httplib README	2020-02-17 22:54:09 -08:00
Brian Clinkenbeard	76b55c3624	0.4.2 works too	2020-02-17 22:53:25 -08:00
Brian Clinkenbeard	293d4d553a	update httplib to 0.2.6	2020-02-17 20:13:24 -08:00
ReinUsesLisp	bfda5ff3f6	texture_cache: Avoid matches in 3D textures Code before this commit was trying to match 3D textures with another target. Fix that.	2020-02-16 04:15:42 -03:00
ReinUsesLisp	6910ade146	shader/texture: Allow 2D shadow arrays and simplify code Shadow sampler 2D arrays are supported on OpenGL, so there's no reason to forbid these. Enable textureLod usage on these. Minor style changes.	2020-02-15 02:36:28 -03:00
ReinUsesLisp	91aa58e410	maxwell_3d: Unify draw methods Pass instanced state of a draw invocation as an argument instead of having two separate virtual methods.	2020-02-14 18:09:40 -03:00
ReinUsesLisp	6d3a046caa	query_cache: Address feedback	2020-02-14 17:38:27 -03:00
ReinUsesLisp	54a00ee4cf	query_cache: Fix ambiguity in CacheAddr getter	2020-02-14 17:38:27 -03:00
ReinUsesLisp	cc0694559f	query_cache: Add a recursive mutex for concurrent usage	2020-02-14 17:38:27 -03:00
ReinUsesLisp	bcd348f238	vk_query_cache: Implement generic query cache on Vulkan	2020-02-14 17:38:27 -03:00
ReinUsesLisp	c31382ced5	query_cache: Abstract OpenGL implementation Abstract the current OpenGL implementation into the VideoCommon namespace and reimplement it on top of that. Doing this avoids repeating code and logic in the Vulkan implementation.	2020-02-14 17:38:27 -03:00
ReinUsesLisp	73d2d3342d	gl_query_cache: Optimize query cache Use a custom cache instead of relying on a ranged cache.	2020-02-14 17:38:27 -03:00
ReinUsesLisp	aae8c180cb	gl_query_cache: Implement host queries using a deferred cache Instead of waiting immediately for executed commands, defer the query until the guest CPU reads it. This way we get closer to what the guest program is doing. To archive this we have to build a dependency queue, because host APIs (like OpenGL and Vulkan) use ranged queries instead of counters like NVN. Waiting for queries implicitly uses fences and this requires a command being queued, otherwise the driver will lock waiting until a timeout. To fix this when there are no commands queued, we explicitly call glFlush.	2020-02-14 17:33:13 -03:00
ReinUsesLisp	ef9920e164	gl_rasterizer: Sort method declarations	2020-02-14 17:27:17 -03:00
ReinUsesLisp	fe1238be7a	gl_rasterizer: Add queued commands counter Keep track of the queued OpenGL commands that can signal a fence if waited on. As a side effect, we avoid calls to glFlush when no commands are queued.	2020-02-14 17:27:17 -03:00
ReinUsesLisp	2b58652f08	maxwell_3d: Slow implementation of passed samples (query 21) Implements GL_SAMPLES_PASSED by waiting immediately for queries.	2020-02-14 17:27:17 -03:00
ReinUsesLisp	3217400dd1	gl_resource_manager: Add managed query class	2020-02-13 22:25:55 -03:00
ReinUsesLisp	336a4f8e99	gl_rasterizer: Use the least generic OpenGL draw function possible This may help some implementations.	2020-02-13 21:55:21 -03:00
ReinUsesLisp	cbea8c74de	vk_shader_decompiler: Fix vertex id and instance id Vulkan's VertexIndex and InstanceIndex don't match with hardware. This is because Nvidia implements gl_VertexID and gl_InstanceID. The math that relates these is: gl_VertexIndex = gl_BaseVertex + gl_VertexID gl_InstanceIndex = gl_InstanceIndex + gl_InstanceID To emulate it using what Vulkan's SPIR-V offers (the Index variants) this commit substracts gl_Base from gl_*Index to obtain the OpenGL and hardware's equivalent.	2020-02-13 20:25:28 -03:00
FearlessTobi	e3cad7d49e	audio_core: Switch to a faster interpolation technique	2020-01-24 00:38:22 +01:00