Merge remote-tracking branch 'origin/main'

REUSE.toml

@@ -37,6 +37,7 @@ path = [
     "src/images/refresh_icon.png",
     "src/images/settings_icon.png",
     "src/images/stop_icon.png",
+    "src/images/utils_icon.png",
     "src/images/shadPS4.icns",
     "src/images/shadps4.ico",
     "src/images/net.shadps4.shadPS4.svg",

src/core/devtools/widget/cmd_list.cpp

@@ -3,6 +3,7 @@
 
 // Credits to https://github.com/psucien/tlg-emu-tools/
 
+#include <cinttypes>
 #include <string>
 #include <gcn/si_ci_vi_merged_offset.h>
 #include <imgui.h>
@@ -1224,12 +1225,12 @@ void CmdListViewer::Draw(bool only_batches_view) {
         }
 
         Text("queue    : %s", queue_name);
-        Text("base addr: %08llX", cmdb_addr);
+        Text("base addr: %08" PRIXPTR, cmdb_addr);
         SameLine();
         if (SmallButton("Memory >")) {
             cmdb_view.Open ^= true;
         }
-        Text("size     : %04llX", cmdb_size);
+        Text("size     : %04zX", cmdb_size);
         Separator();
 
         {
@@ -1292,12 +1293,12 @@ void CmdListViewer::Draw(bool only_batches_view) {
                 if (batch.type == static_cast<AmdGpu::PM4ItOpcode>(0xFF)) {
                     ignore_header = true;
                 } else if (!batch.marker.empty()) {
-                    snprintf(batch_hdr, sizeof(batch_hdr), "%08llX: batch-%03d %s | %s",
+                    snprintf(batch_hdr, sizeof(batch_hdr), "%08" PRIXPTR ": batch-%03d %s | %s",
                              cmdb_addr + batch.start_addr, batch.id,
                              Gcn::GetOpCodeName(static_cast<u32>(batch.type)),
                              batch.marker.c_str());
                 } else {
-                    snprintf(batch_hdr, sizeof(batch_hdr), "%08llX: batch-%03d %s",
+                    snprintf(batch_hdr, sizeof(batch_hdr), "%08" PRIXPTR ": batch-%03d %s",
                              cmdb_addr + batch.start_addr, batch.id,
                              Gcn::GetOpCodeName(static_cast<u32>(batch.type)));
                 }
@@ -1348,7 +1349,7 @@ void CmdListViewer::Draw(bool only_batches_view) {
                 }
 
                 if (show_batch_content) {
-                    auto processed_size = 0ull;
+                    size_t processed_size = 0;
                     auto bb = ctx.LastItemData.Rect;
                     if (group_batches && !ignore_header) {
                         Indent();
@@ -1364,9 +1365,9 @@ void CmdListViewer::Draw(bool only_batches_view) {
                             op = pm4_t3->opcode;
 
                             char header_name[128];
-                            sprintf(header_name, "%08llX: %s",
+                            snprintf(header_name, sizeof(header_name), "%08" PRIXPTR ": %s",
-                                    cmdb_addr + batch.start_addr + processed_size,
+                                     cmdb_addr + batch.start_addr + processed_size,
-                                    Gcn::GetOpCodeName((u32)op));
+                                     Gcn::GetOpCodeName(static_cast<u32>(op)));
 
                             bool open_pm4 = TreeNode(header_name);
                             if (!group_batches) {

src/core/libraries/gnmdriver/gnmdriver.cpp

@@ -703,7 +703,7 @@ s32 PS4_SYSV_ABI sceGnmDrawIndexIndirectCountMulti(u32* cmdbuf, u32 size, u32 da
         cmdbuf[3] = (count_addr != 0 ? 1u : 0u) << 0x1e;
         cmdbuf[4] = max_count;
         *(u64*)(&cmdbuf[5]) = count_addr;
-        cmdbuf[7] = AmdGpu::Liverpool::DrawIndexedIndirectArgsSize;
+        cmdbuf[7] = sizeof(DrawIndexedIndirectArgs);
         cmdbuf[8] = 0;
 
         cmdbuf += 9;

src/core/libraries/kernel/file_system.cpp

@@ -416,7 +416,7 @@ int PS4_SYSV_ABI sceKernelCheckReachability(const char* path) {
     return ORBIS_OK;
 }
 
-s64 PS4_SYSV_ABI sceKernelPread(int d, void* buf, size_t nbytes, s64 offset) {
+s64 PS4_SYSV_ABI sceKernelPreadv(int d, SceKernelIovec* iov, int iovcnt, s64 offset) {
     if (d < 3) {
         return ORBIS_KERNEL_ERROR_EPERM;
     }
@@ -436,10 +436,19 @@ s64 PS4_SYSV_ABI sceKernelPread(int d, void* buf, size_t nbytes, s64 offset) {
         file->f.Seek(pos);
     };
     if (!file->f.Seek(offset)) {
-        LOG_CRITICAL(Kernel_Fs, "sceKernelPread: failed to seek");
+        LOG_CRITICAL(Kernel_Fs, "failed to seek");
         return ORBIS_KERNEL_ERROR_EINVAL;
     }
-    return file->f.ReadRaw<u8>(buf, nbytes);
+    size_t total_read = 0;
+    for (int i = 0; i < iovcnt; i++) {
+        total_read += file->f.ReadRaw<u8>(iov[i].iov_base, iov[i].iov_len);
+    }
+    return total_read;
+}
+
+s64 PS4_SYSV_ABI sceKernelPread(int d, void* buf, size_t nbytes, s64 offset) {
+    SceKernelIovec iovec{buf, nbytes};
+    return sceKernelPreadv(d, &iovec, 1, offset);
 }
 
 int PS4_SYSV_ABI sceKernelFStat(int fd, OrbisKernelStat* sb) {
@@ -537,14 +546,13 @@ static int GetDents(int fd, char* buf, int nbytes, s64* basep) {
     }
     const auto& entry = file->dirents.at(file->dirents_index++);
     auto str = entry.name;
-    auto str_size = str.size() - 1;
     static int fileno = 1000; // random
     OrbisKernelDirent* sce_ent = (OrbisKernelDirent*)buf;
     sce_ent->d_fileno = fileno++; // TODO this should be unique but atm it changes maybe switch to a
     // hash or something?
     sce_ent->d_reclen = sizeof(OrbisKernelDirent);
     sce_ent->d_type = (entry.isFile ? 8 : 4);
-    sce_ent->d_namlen = str_size;
+    sce_ent->d_namlen = str.size();
     strncpy(sce_ent->d_name, str.c_str(), ORBIS_MAX_PATH);
     sce_ent->d_name[ORBIS_MAX_PATH] = '\0';
 
@@ -650,6 +658,7 @@ void RegisterFileSystem(Core::Loader::SymbolsResolver* sym) {
     LIB_FUNCTION("E6ao34wPw+U", "libScePosix", 1, "libkernel", 1, 1, posix_stat);
     LIB_FUNCTION("E6ao34wPw+U", "libkernel", 1, "libkernel", 1, 1, posix_stat);
     LIB_FUNCTION("+r3rMFwItV4", "libkernel", 1, "libkernel", 1, 1, sceKernelPread);
+    LIB_FUNCTION("yTj62I7kw4s", "libkernel", 1, "libkernel", 1, 1, sceKernelPreadv);
     LIB_FUNCTION("uWyW3v98sU4", "libkernel", 1, "libkernel", 1, 1, sceKernelCheckReachability);
     LIB_FUNCTION("fTx66l5iWIA", "libkernel", 1, "libkernel", 1, 1, sceKernelFsync);
     LIB_FUNCTION("juWbTNM+8hw", "libkernel", 1, "libkernel", 1, 1, posix_fsync);

src/core/libraries/kernel/threads/semaphore.cpp

@@ -165,8 +165,7 @@ public:
         while (it != wait_list.end() && (*it)->priority > waiter->priority) {
             ++it;
         }
-        wait_list.insert(it, waiter);
+        return wait_list.insert(it, waiter);
-        return it;
     }
 
     WaitList wait_list;

src/core/libraries/libc_internal/libc_internal.cpp

@@ -3,8 +3,8 @@
 
 #include <cmath>
 
+#include "common/assert.h"
 #include "common/logging/log.h"
-#include "core/libraries/error_codes.h"
 #include "core/libraries/libs.h"
 #include "libc_internal.h"
 
@@ -36,28 +36,184 @@ int PS4_SYSV_ABI internal_strcpy_s(char* dest, size_t dest_size, const char* src
 #endif
 }
 
+int PS4_SYSV_ABI internal_strcat_s(char* dest, size_t dest_size, const char* src) {
+#ifdef _WIN64
+    return strcat_s(dest, dest_size, src);
+#else
+    std::strcat(dest, src);
+    return 0; // ALL OK
+#endif
+}
+
 int PS4_SYSV_ABI internal_memcmp(const void* s1, const void* s2, size_t n) {
     return std::memcmp(s1, s2, n);
 }
 
+int PS4_SYSV_ABI internal_strcmp(const char* str1, const char* str2) {
+    return std::strcmp(str1, str2);
+}
+
 int PS4_SYSV_ABI internal_strncmp(const char* str1, const char* str2, size_t num) {
     return std::strncmp(str1, str2, num);
 }
 
-int PS4_SYSV_ABI internal_strlen(const char* str) {
+size_t PS4_SYSV_ABI internal_strlen(const char* str) {
     return std::strlen(str);
 }
 
+char* PS4_SYSV_ABI internal_strncpy(char* dest, const char* src, std::size_t count) {
+    return std::strncpy(dest, src, count);
+}
+
+char* PS4_SYSV_ABI internal_strcat(char* dest, const char* src) {
+    return std::strcat(dest, src);
+}
+
+const char* PS4_SYSV_ABI internal_strchr(const char* str, int c) {
+    return std::strchr(str, c);
+}
+
+double PS4_SYSV_ABI internal_sin(double x) {
+    return std::sin(x);
+}
+
+float PS4_SYSV_ABI internal_sinf(float x) {
+    return std::sinf(x);
+}
+
+double PS4_SYSV_ABI internal_cos(double x) {
+    return std::cos(x);
+}
+
+float PS4_SYSV_ABI internal_cosf(float x) {
+    return std::cosf(x);
+}
+
+void PS4_SYSV_ABI internal_sincos(double x, double* sinp, double* cosp) {
+    *sinp = std::sin(x);
+    *cosp = std::cos(x);
+}
+
+void PS4_SYSV_ABI internal_sincosf(float x, float* sinp, float* cosp) {
+    *sinp = std::sinf(x);
+    *cosp = std::cosf(x);
+}
+
+double PS4_SYSV_ABI internal_tan(double x) {
+    return std::tan(x);
+}
+
+float PS4_SYSV_ABI internal_tanf(float x) {
+    return std::tanf(x);
+}
+
+double PS4_SYSV_ABI internal_asin(double x) {
+    return std::asin(x);
+}
+
+float PS4_SYSV_ABI internal_asinf(float x) {
+    return std::asinf(x);
+}
+
+double PS4_SYSV_ABI internal_acos(double x) {
+    return std::acos(x);
+}
+
+float PS4_SYSV_ABI internal_acosf(float x) {
+    return std::acosf(x);
+}
+
+double PS4_SYSV_ABI internal_atan(double x) {
+    return std::atan(x);
+}
+
+float PS4_SYSV_ABI internal_atanf(float x) {
+    return std::atanf(x);
+}
+
+double PS4_SYSV_ABI internal_atan2(double y, double x) {
+    return std::atan2(y, x);
+}
+
+float PS4_SYSV_ABI internal_atan2f(float y, float x) {
+    return std::atan2f(y, x);
+}
+
+double PS4_SYSV_ABI internal_exp(double x) {
+    return std::exp(x);
+}
+
 float PS4_SYSV_ABI internal_expf(float x) {
-    return expf(x);
+    return std::expf(x);
+}
+
+double PS4_SYSV_ABI internal_exp2(double x) {
+    return std::exp2(x);
+}
+
+float PS4_SYSV_ABI internal_exp2f(float x) {
+    return std::exp2f(x);
+}
+
+double PS4_SYSV_ABI internal_pow(double x, double y) {
+    return std::pow(x, y);
+}
+
+float PS4_SYSV_ABI internal_powf(float x, float y) {
+    return std::powf(x, y);
+}
+
+double PS4_SYSV_ABI internal_log(double x) {
+    return std::log(x);
+}
+
+float PS4_SYSV_ABI internal_logf(float x) {
+    return std::logf(x);
+}
+
+double PS4_SYSV_ABI internal_log10(double x) {
+    return std::log10(x);
+}
+
+float PS4_SYSV_ABI internal_log10f(float x) {
+    return std::log10f(x);
 }
 
 void* PS4_SYSV_ABI internal_malloc(size_t size) {
     return std::malloc(size);
 }
 
-char* PS4_SYSV_ABI internal_strncpy(char* dest, const char* src, std::size_t count) {
+void PS4_SYSV_ABI internal_free(void* ptr) {
-    return std::strncpy(dest, src, count);
+    std::free(ptr);
+}
+
+void* PS4_SYSV_ABI internal_operator_new(size_t size) {
+    if (size == 0) {
+        // Size of 1 is used if 0 is provided.
+        size = 1;
+    }
+    void* ptr = std::malloc(size);
+    ASSERT_MSG(ptr, "Failed to allocate new object with size {}", size);
+    return ptr;
+}
+
+void PS4_SYSV_ABI internal_operator_delete(void* ptr) {
+    if (ptr) {
+        std::free(ptr);
+    }
+}
+
+int PS4_SYSV_ABI internal_posix_memalign(void** ptr, size_t alignment, size_t size) {
+#ifdef _WIN64
+    void* allocated = _aligned_malloc(size, alignment);
+    if (!allocated) {
+        return errno;
+    }
+    *ptr = allocated;
+    return 0;
+#else
+    return posix_memalign(ptr, alignment, size);
+#endif
 }
 
 void RegisterlibSceLibcInternal(Core::Loader::SymbolsResolver* sym) {
@@ -69,17 +225,71 @@ void RegisterlibSceLibcInternal(Core::Loader::SymbolsResolver* sym) {
                  internal_memset);
     LIB_FUNCTION("5Xa2ACNECdo", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
                  internal_strcpy_s);
+    LIB_FUNCTION("K+gcnFFJKVc", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
+                 internal_strcat_s);
     LIB_FUNCTION("DfivPArhucg", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
                  internal_memcmp);
-    LIB_FUNCTION("8zsu04XNsZ4", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1, internal_expf);
     LIB_FUNCTION("aesyjrHVWy4", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
+                 internal_strcmp);
+    LIB_FUNCTION("Ovb2dSJOAuE", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
                  internal_strncmp);
     LIB_FUNCTION("j4ViWNHEgww", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
                  internal_strlen);
     LIB_FUNCTION("6sJWiWSRuqk", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
                  internal_strncpy);
+    LIB_FUNCTION("Ls4tzzhimqQ", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
+                 internal_strcat);
+    LIB_FUNCTION("ob5xAW4ln-0", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
+                 internal_strchr);
+    LIB_FUNCTION("H8ya2H00jbI", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1, internal_sin);
+    LIB_FUNCTION("Q4rRL34CEeE", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1, internal_sinf);
+    LIB_FUNCTION("2WE3BTYVwKM", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1, internal_cos);
+    LIB_FUNCTION("-P6FNMzk2Kc", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1, internal_cosf);
+    LIB_FUNCTION("jMB7EFyu30Y", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
+                 internal_sincos);
+    LIB_FUNCTION("pztV4AF18iI", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
+                 internal_sincosf);
+    LIB_FUNCTION("T7uyNqP7vQA", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1, internal_tan);
+    LIB_FUNCTION("ZE6RNL+eLbk", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1, internal_tanf);
+    LIB_FUNCTION("7Ly52zaL44Q", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1, internal_asin);
+    LIB_FUNCTION("GZWjF-YIFFk", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
+                 internal_asinf);
+    LIB_FUNCTION("JBcgYuW8lPU", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1, internal_acos);
+    LIB_FUNCTION("QI-x0SL8jhw", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
+                 internal_acosf);
+    LIB_FUNCTION("OXmauLdQ8kY", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1, internal_atan);
+    LIB_FUNCTION("weDug8QD-lE", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
+                 internal_atanf);
+    LIB_FUNCTION("HUbZmOnT-Dg", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
+                 internal_atan2);
+    LIB_FUNCTION("EH-x713A99c", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
+                 internal_atan2f);
+    LIB_FUNCTION("NVadfnzQhHQ", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1, internal_exp);
+    LIB_FUNCTION("8zsu04XNsZ4", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1, internal_expf);
+    LIB_FUNCTION("dnaeGXbjP6E", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1, internal_exp2);
+    LIB_FUNCTION("wuAQt-j+p4o", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
+                 internal_exp2f);
+    LIB_FUNCTION("9LCjpWyQ5Zc", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1, internal_pow);
+    LIB_FUNCTION("1D0H2KNjshE", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1, internal_powf);
+    LIB_FUNCTION("rtV7-jWC6Yg", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1, internal_log);
+    LIB_FUNCTION("RQXLbdT2lc4", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1, internal_logf);
+    LIB_FUNCTION("WuMbPBKN1TU", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
+                 internal_log10);
+    LIB_FUNCTION("lhpd6Wk6ccs", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
+                 internal_log10f);
     LIB_FUNCTION("gQX+4GDQjpM", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
                  internal_malloc);
+    LIB_FUNCTION("tIhsqj0qsFE", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1, internal_free);
+    LIB_FUNCTION("fJnpuVVBbKk", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
+                 internal_operator_new);
+    LIB_FUNCTION("hdm0YfMa7TQ", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
+                 internal_operator_new);
+    LIB_FUNCTION("MLWl90SFWNE", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
+                 internal_operator_delete);
+    LIB_FUNCTION("z+P+xCnWLBk", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
+                 internal_operator_delete);
+    LIB_FUNCTION("cVSk9y8URbc", "libSceLibcInternal", 1, "libSceLibcInternal", 1, 1,
+                 internal_posix_memalign);
 };
 
 } // namespace Libraries::LibcInternal

src/main.cpp

@@ -32,7 +32,7 @@ int main(int argc, char* argv[]) {
                           "  -g, --game <path|ID>          Specify game path to launch\n"
                           "  -p, --patch <patch_file>      Apply specified patch file\n"
                           "  -f, --fullscreen <true|false> Specify window initial fullscreen "
-                          "state. Does not overwrite the config file."
+                          "state. Does not overwrite the config file.\n"
                           "  -h, --help                    Display this help message\n";
              exit(0);
          }},

src/qt_gui/gui_context_menus.h

@@ -451,7 +451,7 @@ private:
             pShellLink->SetWorkingDirectory((LPCWSTR)QFileInfo(exePath).absolutePath().utf16());
 
             // Set arguments, eboot.bin file location
-            QString arguments = QString("\"%1\"").arg(targetPath);
+            QString arguments = QString("-g \"%1\"").arg(targetPath);
             pShellLink->SetArguments((LPCWSTR)arguments.utf16());
 
             // Set the icon for the shortcut

src/qt_gui/main.cpp

@@ -45,7 +45,7 @@ int main(int argc, char* argv[]) {
                           "  -p, --patch <patch_file>      Apply specified patch file\n"
                           "  -s, --show-gui                Show the GUI\n"
                           "  -f, --fullscreen <true|false> Specify window initial fullscreen "
-                          "state. Does not overwrite the config file."
+                          "state. Does not overwrite the config file.\n"
                           "  -h, --help                    Display this help message\n";
              exit(0);
          }},

src/qt_gui/main_window.cpp

@@ -106,6 +106,7 @@ void MainWindow::CreateActions() {
     m_list_mode_act_group = new QActionGroup(this);
     m_list_mode_act_group->addAction(ui->setlistModeListAct);
     m_list_mode_act_group->addAction(ui->setlistModeGridAct);
+    m_list_mode_act_group->addAction(ui->setlistElfAct);
 
     // create action group for themes
     m_theme_act_group = new QActionGroup(this);
@@ -372,7 +373,7 @@ void MainWindow::CreateConnects() {
         ui->sizeSlider->setEnabled(true);
         ui->sizeSlider->setSliderPosition(slider_pos_grid);
     });
-    // Elf
+    // Elf Viewer
     connect(ui->setlistElfAct, &QAction::triggered, m_dock_widget.data(), [this]() {
         BackgroundMusicPlayer::getInstance().stopMusic();
         m_dock_widget->setWidget(m_elf_viewer.data());
@@ -632,10 +633,12 @@ void MainWindow::ConfigureGuiFromSettings() {
                 Config::getMainWindowGeometryW(), Config::getMainWindowGeometryH());
 
     ui->showGameListAct->setChecked(true);
-    if (isTableList) {
+    if (Config::getTableMode() == 0) {
         ui->setlistModeListAct->setChecked(true);
-    } else {
+    } else if (Config::getTableMode() == 1) {
         ui->setlistModeGridAct->setChecked(true);
+    } else if (Config::getTableMode() == 2) {
+        ui->setlistElfAct->setChecked(true);
     }
     BackgroundMusicPlayer::getInstance().setVolume(Config::getBGMvolume());
 }
@@ -967,6 +970,7 @@ void MainWindow::SetUiIcons(bool isWhite) {
     ui->gameInstallPathAct->setIcon(RecolorIcon(ui->gameInstallPathAct->icon(), isWhite));
     ui->menuThemes->setIcon(RecolorIcon(ui->menuThemes->icon(), isWhite));
     ui->menuGame_List_Icons->setIcon(RecolorIcon(ui->menuGame_List_Icons->icon(), isWhite));
+    ui->menuUtils->setIcon(RecolorIcon(ui->menuUtils->icon(), isWhite));
     ui->playButton->setIcon(RecolorIcon(ui->playButton->icon(), isWhite));
     ui->pauseButton->setIcon(RecolorIcon(ui->pauseButton->icon(), isWhite));
     ui->stopButton->setIcon(RecolorIcon(ui->stopButton->icon(), isWhite));

src/qt_gui/main_window_ui.h

@@ -110,15 +110,14 @@ public:
         setIconSizeLargeAct->setCheckable(true);
         setlistModeListAct = new QAction(MainWindow);
         setlistModeListAct->setObjectName("setlistModeListAct");
-        setlistModeListAct->setCheckable(true);
-        setlistModeListAct->setChecked(true);
         setlistModeListAct->setIcon(QIcon(":images/list_icon.png"));
+        setlistModeListAct->setCheckable(true);
         setlistModeGridAct = new QAction(MainWindow);
         setlistModeGridAct->setObjectName("setlistModeGridAct");
-        setlistModeGridAct->setCheckable(true);
         setlistModeGridAct->setIcon(QIcon(":images/grid_icon.png"));
+        setlistModeGridAct->setCheckable(true);
         setlistElfAct = new QAction(MainWindow);
-        setlistElfAct->setObjectName("setlistModeGridAct");
+        setlistElfAct->setObjectName("setlistElfAct");
         setlistElfAct->setCheckable(true);
         gameInstallPathAct = new QAction(MainWindow);
         gameInstallPathAct->setObjectName("gameInstallPathAct");
@@ -250,6 +249,7 @@ public:
         menuSettings->setObjectName("menuSettings");
         menuUtils = new QMenu(menuSettings);
         menuUtils->setObjectName("menuUtils");
+        menuUtils->setIcon(QIcon(":images/utils_icon.png"));
         menuThemes = new QMenu(menuView);
         menuThemes->setObjectName("menuThemes");
         menuThemes->setIcon(QIcon(":images/themes_icon.png"));

src/qt_gui/translations/sq.ts

@@ -118,22 +118,22 @@
 		<message>
 			<location filename="../gui_context_menus.h" line="49"/>
 			<source>Open Folder...</source>
-			<translation>Hapni Dosjen...</translation>
+			<translation>Hap Dosjen...</translation>
 		</message>
 		<message>
 			<location filename="../gui_context_menus.h" line="50"/>
 			<source>Open Game Folder</source>
-			<translation>Hapni Dosjen e Lojës</translation>
+			<translation>Hap Dosjen e Lojës</translation>
 		</message>
 		<message>
 			<location filename="../gui_context_menus.h" line="51"/>
 			<source>Open Save Data Folder</source>
-			<translation>Hapni Dosjen e të Dhënave të Ruajtura</translation>
+			<translation>Hap Dosjen e të Dhënave të Ruajtura</translation>
 		</message>
 		<message>
 			<location filename="../gui_context_menus.h" line="52"/>
 			<source>Open Log Folder</source>
-			<translation>Hapni Dosjen e Regjistrimeve</translation>
+			<translation>Hap Dosjen e Ditarit</translation>
 		</message>
 		<message>
 			<location filename="../gui_context_menus.h" line="59"/>

src/shader_recompiler/info.h

@@ -17,6 +17,7 @@
 #include "shader_recompiler/ir/type.h"
 #include "shader_recompiler/params.h"
 #include "shader_recompiler/runtime_info.h"
+#include "video_core/amdgpu/liverpool.h"
 #include "video_core/amdgpu/resource.h"
 
 namespace Shader {
@@ -251,10 +252,10 @@ struct Info {
         bnd.user_data += ud_mask.NumRegs();
     }
 
-    [[nodiscard]] std::pair<u32, u32> GetDrawOffsets() const {
+    [[nodiscard]] std::pair<u32, u32> GetDrawOffsets(const AmdGpu::Liverpool::Regs& regs) const {
-        u32 vertex_offset = 0;
+        u32 vertex_offset = regs.index_offset;
         u32 instance_offset = 0;
-        if (vertex_offset_sgpr != -1) {
+        if (vertex_offset == 0 && vertex_offset_sgpr != -1) {
             vertex_offset = user_data[vertex_offset_sgpr];
         }
         if (instance_offset_sgpr != -1) {

src/shader_recompiler/ir/passes/flatten_extended_userdata_pass.cpp

@@ -148,17 +148,21 @@ static void GenerateSrtProgram(Info& info, PassInfo& pass_info) {
     // Special case for V# step rate buffers in fetch shader
     for (const auto [sgpr_base, dword_offset, num_dwords] : info.srt_info.srt_reservations) {
         // get pointer to V#
-        c.mov(r10d, ptr[rdi + (sgpr_base << 2)]);
+        if (sgpr_base != IR::NumScalarRegs) {
-
+            PushPtr(c, sgpr_base);
+        }
         u32 src_off = dword_offset << 2;
 
         for (auto j = 0; j < num_dwords; j++) {
-            c.mov(r11d, ptr[r10d + src_off]);
+            c.mov(r11d, ptr[rdi + src_off]);
             c.mov(ptr[rsi + (pass_info.dst_off_dw << 2)], r11d);
 
             src_off += 4;
             ++pass_info.dst_off_dw;
         }
+        if (sgpr_base != IR::NumScalarRegs) {
+            PopPtr(c);
+        }
     }
 
     ASSERT(pass_info.dst_off_dw == info.srt_info.flattened_bufsize_dw);

src/shader_recompiler/runtime_info.h

@@ -20,7 +20,7 @@ enum class Stage : u32 {
     Local,
     Compute,
 };
-constexpr u32 MaxStageTypes = 6;
+constexpr u32 MaxStageTypes = 7;
 
 [[nodiscard]] constexpr Stage StageFromIndex(size_t index) noexcept {
     return static_cast<Stage>(index);

src/shadps4.qrc

@@ -6,6 +6,7 @@
 		<file>images/play_icon.png</file>
 		<file>images/pause_icon.png</file>
 		<file>images/stop_icon.png</file>
+		<file>images/utils_icon.png</file>
 		<file>images/file_icon.png</file>
 		<file>images/folder_icon.png</file>
 		<file>images/themes_icon.png</file>

src/video_core/amdgpu/liverpool.cpp

@@ -410,7 +410,7 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span<const u32> dcb, std::span<c
                 const auto* draw_indirect = reinterpret_cast<const PM4CmdDrawIndirect*>(header);
                 const auto offset = draw_indirect->data_offset;
                 const auto ib_address = mapped_queues[GfxQueueId].indirect_args_addr;
-                const auto size = sizeof(PM4CmdDrawIndirect::DrawInstancedArgs);
+                const auto size = sizeof(DrawIndirectArgs);
                 if (DebugState.DumpingCurrentReg()) {
                     DebugState.PushRegsDump(base_addr, reinterpret_cast<uintptr_t>(header), regs);
                 }
@@ -427,7 +427,7 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span<const u32> dcb, std::span<c
                     reinterpret_cast<const PM4CmdDrawIndexIndirect*>(header);
                 const auto offset = draw_index_indirect->data_offset;
                 const auto ib_address = mapped_queues[GfxQueueId].indirect_args_addr;
-                const auto size = sizeof(PM4CmdDrawIndexIndirect::DrawIndexInstancedArgs);
+                const auto size = sizeof(DrawIndexedIndirectArgs);
                 if (DebugState.DumpingCurrentReg()) {
                     DebugState.PushRegsDump(base_addr, reinterpret_cast<uintptr_t>(header), regs);
                 }
@@ -442,10 +442,9 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span<const u32> dcb, std::span<c
             }
             case PM4ItOpcode::DrawIndexIndirectCountMulti: {
                 const auto* draw_index_indirect =
-                    reinterpret_cast<const PM4CmdDrawIndexIndirect*>(header);
+                    reinterpret_cast<const PM4CmdDrawIndexIndirectMulti*>(header);
                 const auto offset = draw_index_indirect->data_offset;
                 const auto ib_address = mapped_queues[GfxQueueId].indirect_args_addr;
-                const auto size = sizeof(PM4CmdDrawIndexIndirect::DrawIndexInstancedArgs);
                 if (DebugState.DumpingCurrentReg()) {
                     DebugState.PushRegsDump(base_addr, reinterpret_cast<uintptr_t>(header), regs);
                 }
@@ -453,7 +452,7 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span<const u32> dcb, std::span<c
                     const auto cmd_address = reinterpret_cast<const void*>(header);
                     rasterizer->ScopeMarkerBegin(
                         fmt::format("dcb:{}:DrawIndexIndirectCountMulti", cmd_address));
-                    rasterizer->DrawIndirect(true, ib_address, offset, size,
+                    rasterizer->DrawIndirect(true, ib_address, offset, draw_index_indirect->stride,
                                              draw_index_indirect->count,
                                              draw_index_indirect->countAddr);
                     rasterizer->ScopeMarkerEnd();

src/video_core/amdgpu/liverpool.h

@@ -57,8 +57,6 @@ struct Liverpool {
     static constexpr u32 ConfigRegWordOffset = 0x2000;
     static constexpr u32 ShRegWordOffset = 0x2C00;
     static constexpr u32 NumRegs = 0xD000;
-    static constexpr u32 DrawIndirectArgsSize = 0x10u;
-    static constexpr u32 DrawIndexedIndirectArgsSize = 0x14u;
 
     using UserData = std::array<u32, NumShaderUserData>;
 
@@ -1113,7 +1111,8 @@ struct Liverpool {
             INSERT_PADDING_WORDS(2);
             std::array<ViewportScissor, NumViewports> viewport_scissors;
             std::array<ViewportDepth, NumViewports> viewport_depths;
-            INSERT_PADDING_WORDS(0xA103 - 0xA0D4);
+            INSERT_PADDING_WORDS(0xA102 - 0xA0D4);
+            u32 index_offset;
             u32 primitive_restart_index;
             INSERT_PADDING_WORDS(1);
             BlendConstants blend_constants;
@@ -1382,6 +1381,7 @@ static_assert(GFX6_3D_REG_INDEX(color_target_mask) == 0xA08E);
 static_assert(GFX6_3D_REG_INDEX(color_shader_mask) == 0xA08F);
 static_assert(GFX6_3D_REG_INDEX(generic_scissor) == 0xA090);
 static_assert(GFX6_3D_REG_INDEX(viewport_scissors) == 0xA094);
+static_assert(GFX6_3D_REG_INDEX(index_offset) == 0xA102);
 static_assert(GFX6_3D_REG_INDEX(primitive_restart_index) == 0xA103);
 static_assert(GFX6_3D_REG_INDEX(stencil_control) == 0xA10B);
 static_assert(GFX6_3D_REG_INDEX(viewports) == 0xA10F);

src/video_core/amdgpu/pm4_cmds.h

@@ -778,14 +778,15 @@ struct PM4CmdDispatchIndirect {
     u32 dispatch_initiator; ///< Dispatch Initiator Register
 };
 
-struct PM4CmdDrawIndirect {
+struct DrawIndirectArgs {
-    struct DrawInstancedArgs {
+    u32 vertex_count_per_instance;
-        u32 vertex_count_per_instance;
+    u32 instance_count;
-        u32 instance_count;
+    u32 start_vertex_location;
-        u32 start_vertex_location;
+    u32 start_instance_location;
-        u32 start_instance_location;
+};
-    };
+static_assert(sizeof(DrawIndirectArgs) == 0x10u);
 
+struct PM4CmdDrawIndirect {
     PM4Type3Header header; ///< header
     u32 data_offset;       ///< Byte aligned offset where the required data structure starts
     union {
@@ -801,15 +802,16 @@ struct PM4CmdDrawIndirect {
     u32 draw_initiator; ///< Draw Initiator Register
 };
 
-struct PM4CmdDrawIndexIndirect {
+struct DrawIndexedIndirectArgs {
-    struct DrawIndexInstancedArgs {
+    u32 index_count_per_instance;
-        u32 index_count_per_instance;
+    u32 instance_count;
-        u32 instance_count;
+    u32 start_index_location;
-        u32 start_index_location;
+    u32 base_vertex_location;
-        u32 base_vertex_location;
+    u32 start_instance_location;
-        u32 start_instance_location;
+};
-    };
+static_assert(sizeof(DrawIndexedIndirectArgs) == 0x14u);
 
+struct PM4CmdDrawIndexIndirect {
     PM4Type3Header header; ///< header
     u32 data_offset;       ///< Byte aligned offset where the required data structure starts
     union {
@@ -822,16 +824,29 @@ struct PM4CmdDrawIndexIndirect {
         BitField<0, 16, u32> start_inst_loc; ///< Offset where the CP will write the
                                              ///< StartInstanceLocation it fetched from memory
     };
+    u32 draw_initiator; ///< Draw Initiator Register
+};
 
+struct PM4CmdDrawIndexIndirectMulti {
+    PM4Type3Header header; ///< header
+    u32 data_offset;       ///< Byte aligned offset where the required data structure starts
+    union {
+        u32 dw2;
+        BitField<0, 16, u32> base_vtx_loc; ///< Offset where the CP will write the
+                                           ///< BaseVertexLocation it fetched from memory
+    };
+    union {
+        u32 dw3;
+        BitField<0, 16, u32> start_inst_loc; ///< Offset where the CP will write the
+                                             ///< StartInstanceLocation it fetched from memory
+    };
     union {
         u32 dw4;
-        struct {
+        BitField<0, 16, u32> drawIndexLoc; ///< register offset to write the Draw Index count
-            BitField<0, 16, u32> drawIndexLoc; ///< register offset to write the Draw Index count
+        BitField<30, 1, u32>
-            BitField<30, 1, u32>
+            countIndirectEnable; ///< Indicates the data structure count is in memory
-                countIndirectEnable; ///< Indicates the data structure count is in memory
+        BitField<31, 1, u32>
-            BitField<31, 1, u32>
+            drawIndexEnable; ///< Enables writing of Draw Index count to DRAW_INDEX_LOC
-                drawIndexEnable; ///< Enables writing of Draw Index count to DRAW_INDEX_LOC
-        };
     };
     u32 count;          ///< Count of data structures to loop through before going to next packet
     u64 countAddr;      ///< DWord aligned Address[31:2]; Valid if countIndirectEnable is set

src/video_core/buffer_cache/buffer_cache.cpp

@@ -620,10 +620,10 @@ void BufferCache::SynchronizeBuffer(Buffer& buffer, VAddr device_addr, u32 size,
 bool BufferCache::SynchronizeBufferFromImage(Buffer& buffer, VAddr device_addr, u32 size) {
     static constexpr FindFlags find_flags =
         FindFlags::NoCreate | FindFlags::RelaxDim | FindFlags::RelaxFmt | FindFlags::RelaxSize;
-    ImageInfo info{};
+    TextureCache::BaseDesc desc{};
-    info.guest_address = device_addr;
+    desc.info.guest_address = device_addr;
-    info.guest_size_bytes = size;
+    desc.info.guest_size_bytes = size;
-    const ImageId image_id = texture_cache.FindImage(info, find_flags);
+    const ImageId image_id = texture_cache.FindImage(desc, find_flags);
     if (!image_id) {
         return false;
     }
@@ -635,7 +635,7 @@ bool BufferCache::SynchronizeBufferFromImage(Buffer& buffer, VAddr device_addr,
                "Texel buffer aliases image subresources {:x} : {:x}", device_addr,
                image.info.guest_address);
     boost::container::small_vector<vk::BufferImageCopy, 8> copies;
-    u32 offset = buffer.Offset(image.cpu_addr);
+    u32 offset = buffer.Offset(image.info.guest_address);
     const u32 num_layers = image.info.resources.layers;
     const u32 max_offset = offset + size;
     for (u32 m = 0; m < image.info.resources.levels; m++) {

src/video_core/page_manager.cpp

@@ -114,8 +114,8 @@ struct PageManager::Impl {
 
             // Notify rasterizer about the fault.
             const VAddr addr = msg.arg.pagefault.address;
-            const VAddr addr_page = Common::AlignDown(addr, PAGESIZE);
+            const VAddr addr_page = GetPageAddr(addr);
-            rasterizer->InvalidateMemory(addr_page, PAGESIZE);
+            rasterizer->InvalidateMemory(addr, addr_page, PAGESIZE);
         }
     }
 
@@ -157,8 +157,8 @@ struct PageManager::Impl {
         const auto addr = reinterpret_cast<VAddr>(fault_address);
         const bool is_write = Common::IsWriteError(context);
         if (is_write && owned_ranges.find(addr) != owned_ranges.end()) {
-            const VAddr addr_aligned = Common::AlignDown(addr, PAGESIZE);
+            const VAddr addr_aligned = GetPageAddr(addr);
-            rasterizer->InvalidateMemory(addr_aligned, PAGESIZE);
+            rasterizer->InvalidateMemory(addr, addr_aligned, PAGESIZE);
             return true;
         }
         return false;
@@ -174,6 +174,14 @@ PageManager::PageManager(Vulkan::Rasterizer* rasterizer_)
 
 PageManager::~PageManager() = default;
 
+VAddr PageManager::GetPageAddr(VAddr addr) {
+    return Common::AlignDown(addr, PAGESIZE);
+}
+
+VAddr PageManager::GetNextPageAddr(VAddr addr) {
+    return Common::AlignUp(addr + 1, PAGESIZE);
+}
+
 void PageManager::OnGpuMap(VAddr address, size_t size) {
     impl->OnMap(address, size);
 }

src/video_core/page_manager.h

@@ -28,6 +28,9 @@ public:
     /// Increase/decrease the number of surface in pages touching the specified region
     void UpdatePagesCachedCount(VAddr addr, u64 size, s32 delta);
 
+    static VAddr GetPageAddr(VAddr addr);
+    static VAddr GetNextPageAddr(VAddr addr);
+
 private:
     struct Impl;
     std::unique_ptr<Impl> impl;

src/video_core/renderer_vulkan/vk_compute_pipeline.cpp

@@ -15,8 +15,10 @@ ComputePipeline::ComputePipeline(const Instance& instance_, Scheduler& scheduler
                                  DescriptorHeap& desc_heap_, vk::PipelineCache pipeline_cache,
                                  u64 compute_key_, const Shader::Info& info_,
                                  vk::ShaderModule module)
-    : Pipeline{instance_, scheduler_, desc_heap_, pipeline_cache}, compute_key{compute_key_},
+    : Pipeline{instance_, scheduler_, desc_heap_, pipeline_cache, true}, compute_key{compute_key_} {
-      info{&info_} {
+    auto& info = stages[int(Shader::Stage::Compute)];
+    info = &info_;
+
     const vk::PipelineShaderStageCreateInfo shader_ci = {
         .stage = vk::ShaderStageFlagBits::eCompute,
         .module = module,
@@ -118,90 +120,4 @@ ComputePipeline::ComputePipeline(const Instance& instance_, Scheduler& scheduler
 
 ComputePipeline::~ComputePipeline() = default;
 
-bool ComputePipeline::BindResources(VideoCore::BufferCache& buffer_cache,
-                                    VideoCore::TextureCache& texture_cache) const {
-    // Bind resource buffers and textures.
-    boost::container::small_vector<vk::WriteDescriptorSet, 16> set_writes;
-    BufferBarriers buffer_barriers;
-    Shader::PushData push_data{};
-    Shader::Backend::Bindings binding{};
-
-    info->PushUd(binding, push_data);
-
-    buffer_infos.clear();
-    buffer_views.clear();
-    image_infos.clear();
-
-    // Most of the time when a metadata is updated with a shader it gets cleared. It means
-    // we can skip the whole dispatch and update the tracked state instead. Also, it is not
-    // intended to be consumed and in such rare cases (e.g. HTile introspection, CRAA) we
-    // will need its full emulation anyways. For cases of metadata read a warning will be logged.
-    const auto IsMetaUpdate = [&](const auto& desc) {
-        const VAddr address = desc.GetSharp(*info).base_address;
-        if (desc.is_written) {
-            if (texture_cache.TouchMeta(address, true)) {
-                LOG_TRACE(Render_Vulkan, "Metadata update skipped");
-                return true;
-            }
-        } else {
-            if (texture_cache.IsMeta(address)) {
-                LOG_WARNING(Render_Vulkan, "Unexpected metadata read by a CS shader (buffer)");
-            }
-        }
-        return false;
-    };
-
-    for (const auto& desc : info->buffers) {
-        if (desc.is_gds_buffer) {
-            continue;
-        }
-        if (IsMetaUpdate(desc)) {
-            return false;
-        }
-    }
-    for (const auto& desc : info->texture_buffers) {
-        if (IsMetaUpdate(desc)) {
-            return false;
-        }
-    }
-
-    BindBuffers(buffer_cache, texture_cache, *info, binding, push_data, set_writes,
-                buffer_barriers);
-
-    BindTextures(texture_cache, *info, binding, set_writes);
-
-    if (set_writes.empty()) {
-        return false;
-    }
-
-    const auto cmdbuf = scheduler.CommandBuffer();
-    if (!buffer_barriers.empty()) {
-        const auto dependencies = vk::DependencyInfo{
-            .dependencyFlags = vk::DependencyFlagBits::eByRegion,
-            .bufferMemoryBarrierCount = u32(buffer_barriers.size()),
-            .pBufferMemoryBarriers = buffer_barriers.data(),
-        };
-        scheduler.EndRendering();
-        cmdbuf.pipelineBarrier2(dependencies);
-    }
-
-    cmdbuf.pushConstants(*pipeline_layout, vk::ShaderStageFlagBits::eCompute, 0u, sizeof(push_data),
-                         &push_data);
-
-    // Bind descriptor set.
-    if (uses_push_descriptors) {
-        cmdbuf.pushDescriptorSetKHR(vk::PipelineBindPoint::eCompute, *pipeline_layout, 0,
-                                    set_writes);
-        return true;
-    }
-    const auto desc_set = desc_heap.Commit(*desc_layout);
-    for (auto& set_write : set_writes) {
-        set_write.dstSet = desc_set;
-    }
-    instance.GetDevice().updateDescriptorSets(set_writes, {});
-    cmdbuf.bindDescriptorSets(vk::PipelineBindPoint::eCompute, *pipeline_layout, 0, desc_set, {});
-
-    return true;
-}
-
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_compute_pipeline.h

@@ -24,13 +24,8 @@ public:
                     vk::ShaderModule module);
     ~ComputePipeline();
 
-    bool BindResources(VideoCore::BufferCache& buffer_cache,
-                       VideoCore::TextureCache& texture_cache) const;
-
 private:
     u64 compute_key;
-    const Shader::Info* info;
-    bool uses_push_descriptors{};
 };
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -16,10 +16,6 @@
 
 namespace Vulkan {
 
-static constexpr auto gp_stage_flags = vk::ShaderStageFlagBits::eVertex |
-                                       vk::ShaderStageFlagBits::eGeometry |
-                                       vk::ShaderStageFlagBits::eFragment;
-
 GraphicsPipeline::GraphicsPipeline(const Instance& instance_, Scheduler& scheduler_,
                                    DescriptorHeap& desc_heap_, const GraphicsPipelineKey& key_,
                                    vk::PipelineCache pipeline_cache,
@@ -389,67 +385,4 @@ void GraphicsPipeline::BuildDescSetLayout() {
     desc_layout = std::move(layout);
 }
 
-void GraphicsPipeline::BindResources(const Liverpool::Regs& regs,
-                                     VideoCore::BufferCache& buffer_cache,
-                                     VideoCore::TextureCache& texture_cache) const {
-    // Bind resource buffers and textures.
-    boost::container::small_vector<vk::WriteDescriptorSet, 16> set_writes;
-    BufferBarriers buffer_barriers;
-    Shader::PushData push_data{};
-    Shader::Backend::Bindings binding{};
-
-    buffer_infos.clear();
-    buffer_views.clear();
-    image_infos.clear();
-
-    for (const auto* stage : stages) {
-        if (!stage) {
-            continue;
-        }
-        if (stage->uses_step_rates) {
-            push_data.step0 = regs.vgt_instance_step_rate_0;
-            push_data.step1 = regs.vgt_instance_step_rate_1;
-        }
-        stage->PushUd(binding, push_data);
-
-        BindBuffers(buffer_cache, texture_cache, *stage, binding, push_data, set_writes,
-                    buffer_barriers);
-
-        BindTextures(texture_cache, *stage, binding, set_writes);
-    }
-
-    const auto cmdbuf = scheduler.CommandBuffer();
-    SCOPE_EXIT {
-        cmdbuf.pushConstants(*pipeline_layout, gp_stage_flags, 0U, sizeof(push_data), &push_data);
-        cmdbuf.bindPipeline(vk::PipelineBindPoint::eGraphics, Handle());
-    };
-
-    if (set_writes.empty()) {
-        return;
-    }
-
-    if (!buffer_barriers.empty()) {
-        const auto dependencies = vk::DependencyInfo{
-            .dependencyFlags = vk::DependencyFlagBits::eByRegion,
-            .bufferMemoryBarrierCount = u32(buffer_barriers.size()),
-            .pBufferMemoryBarriers = buffer_barriers.data(),
-        };
-        scheduler.EndRendering();
-        cmdbuf.pipelineBarrier2(dependencies);
-    }
-
-    // Bind descriptor set.
-    if (uses_push_descriptors) {
-        cmdbuf.pushDescriptorSetKHR(vk::PipelineBindPoint::eGraphics, *pipeline_layout, 0,
-                                    set_writes);
-        return;
-    }
-    const auto desc_set = desc_heap.Commit(*desc_layout);
-    for (auto& set_write : set_writes) {
-        set_write.dstSet = desc_set;
-    }
-    instance.GetDevice().updateDescriptorSets(set_writes, {});
-    cmdbuf.bindDescriptorSets(vk::PipelineBindPoint::eGraphics, *pipeline_layout, 0, desc_set, {});
-}
-
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_graphics_pipeline.h

@@ -2,6 +2,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <xxhash.h>
+
 #include "common/types.h"
 #include "video_core/renderer_vulkan/liverpool_to_vk.h"
 #include "video_core/renderer_vulkan/vk_common.h"
@@ -14,8 +15,8 @@ class TextureCache;
 
 namespace Vulkan {
 
-static constexpr u32 MaxVertexBufferCount = 32;
 static constexpr u32 MaxShaderStages = 5;
+static constexpr u32 MaxVertexBufferCount = 32;
 
 class Instance;
 class Scheduler;
@@ -61,13 +62,6 @@ public:
                      std::span<const vk::ShaderModule> modules);
     ~GraphicsPipeline();
 
-    void BindResources(const Liverpool::Regs& regs, VideoCore::BufferCache& buffer_cache,
-                       VideoCore::TextureCache& texture_cache) const;
-
-    const Shader::Info& GetStage(Shader::Stage stage) const noexcept {
-        return *stages[u32(stage)];
-    }
-
     bool IsEmbeddedVs() const noexcept {
         static constexpr size_t EmbeddedVsHash = 0x9b2da5cf47f8c29f;
         return key.stage_hashes[u32(Shader::Stage::Vertex)] == EmbeddedVsHash;
@@ -99,9 +93,7 @@ private:
     void BuildDescSetLayout();
 
 private:
-    std::array<const Shader::Info*, MaxShaderStages> stages{};
     GraphicsPipelineKey key;
-    bool uses_push_descriptors{};
 };
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_pipeline_cache.h

@@ -38,8 +38,6 @@ struct Program {
 };
 
 class PipelineCache {
-    static constexpr size_t MaxShaderStages = 5;
-
 public:
     explicit PipelineCache(const Instance& instance, Scheduler& scheduler,
                            AmdGpu::Liverpool* liverpool);

src/video_core/renderer_vulkan/vk_pipeline_common.cpp

@@ -12,230 +12,47 @@
 
 namespace Vulkan {
 
-boost::container::static_vector<vk::DescriptorImageInfo, 32> Pipeline::image_infos;
-boost::container::static_vector<vk::BufferView, 8> Pipeline::buffer_views;
-boost::container::static_vector<vk::DescriptorBufferInfo, 32> Pipeline::buffer_infos;
-
 Pipeline::Pipeline(const Instance& instance_, Scheduler& scheduler_, DescriptorHeap& desc_heap_,
-                   vk::PipelineCache pipeline_cache)
+                   vk::PipelineCache pipeline_cache, bool is_compute_ /*= false*/)
-    : instance{instance_}, scheduler{scheduler_}, desc_heap{desc_heap_} {}
+    : instance{instance_}, scheduler{scheduler_}, desc_heap{desc_heap_}, is_compute{is_compute_} {}
 
 Pipeline::~Pipeline() = default;
 
-void Pipeline::BindBuffers(VideoCore::BufferCache& buffer_cache,
+void Pipeline::BindResources(DescriptorWrites& set_writes, const BufferBarriers& buffer_barriers,
-                           VideoCore::TextureCache& texture_cache, const Shader::Info& stage,
+                             const Shader::PushData& push_data) const {
-                           Shader::Backend::Bindings& binding, Shader::PushData& push_data,
+    const auto cmdbuf = scheduler.CommandBuffer();
-                           DescriptorWrites& set_writes, BufferBarriers& buffer_barriers) const {
+    const auto bind_point =
-    using BufferBindingInfo = std::pair<VideoCore::BufferId, AmdGpu::Buffer>;
+        IsCompute() ? vk::PipelineBindPoint::eCompute : vk::PipelineBindPoint::eGraphics;
-    static boost::container::static_vector<BufferBindingInfo, 32> buffer_bindings;
 
-    buffer_bindings.clear();
+    if (!buffer_barriers.empty()) {
-
+        const auto dependencies = vk::DependencyInfo{
-    for (const auto& desc : stage.buffers) {
+            .dependencyFlags = vk::DependencyFlagBits::eByRegion,
-        const auto vsharp = desc.GetSharp(stage);
+            .bufferMemoryBarrierCount = u32(buffer_barriers.size()),
-        if (!desc.is_gds_buffer && vsharp.base_address != 0 && vsharp.GetSize() > 0) {
+            .pBufferMemoryBarriers = buffer_barriers.data(),
-            const auto buffer_id = buffer_cache.FindBuffer(vsharp.base_address, vsharp.GetSize());
+        };
-            buffer_bindings.emplace_back(buffer_id, vsharp);
+        scheduler.EndRendering();
-        } else {
+        cmdbuf.pipelineBarrier2(dependencies);
-            buffer_bindings.emplace_back(VideoCore::BufferId{}, vsharp);
-        }
     }
 
-    using TexBufferBindingInfo = std::pair<VideoCore::BufferId, AmdGpu::Buffer>;
+    const auto stage_flags = IsCompute() ? vk::ShaderStageFlagBits::eCompute : gp_stage_flags;
-    static boost::container::static_vector<TexBufferBindingInfo, 32> texbuffer_bindings;
+    cmdbuf.pushConstants(*pipeline_layout, stage_flags, 0u, sizeof(push_data), &push_data);
 
-    texbuffer_bindings.clear();
+    // Bind descriptor set.
-
+    if (set_writes.empty()) {
-    for (const auto& desc : stage.texture_buffers) {
+        return;
-        const auto vsharp = desc.GetSharp(stage);
-        if (vsharp.base_address != 0 && vsharp.GetSize() > 0 &&
-            vsharp.GetDataFmt() != AmdGpu::DataFormat::FormatInvalid) {
-            const auto buffer_id = buffer_cache.FindBuffer(vsharp.base_address, vsharp.GetSize());
-            texbuffer_bindings.emplace_back(buffer_id, vsharp);
-        } else {
-            texbuffer_bindings.emplace_back(VideoCore::BufferId{}, vsharp);
-        }
     }
 
-    // Bind the flattened user data buffer as a UBO so it's accessible to the shader
+    if (uses_push_descriptors) {
-    if (stage.has_readconst) {
+        cmdbuf.pushDescriptorSetKHR(bind_point, *pipeline_layout, 0, set_writes);
-        const auto [vk_buffer, offset] = buffer_cache.ObtainHostUBO(stage.flattened_ud_buf);
+        return;
-        buffer_infos.emplace_back(vk_buffer->Handle(), offset,
-                                  stage.flattened_ud_buf.size() * sizeof(u32));
-        set_writes.push_back({
-            .dstSet = VK_NULL_HANDLE,
-            .dstBinding = binding.unified++,
-            .dstArrayElement = 0,
-            .descriptorCount = 1,
-            .descriptorType = vk::DescriptorType::eUniformBuffer,
-            .pBufferInfo = &buffer_infos.back(),
-        });
-        ++binding.buffer;
     }
 
-    // Second pass to re-bind buffers that were updated after binding
+    const auto desc_set = desc_heap.Commit(*desc_layout);
-    for (u32 i = 0; i < buffer_bindings.size(); i++) {
+    for (auto& set_write : set_writes) {
-        const auto& [buffer_id, vsharp] = buffer_bindings[i];
+        set_write.dstSet = desc_set;
-        const auto& desc = stage.buffers[i];
-        const bool is_storage = desc.IsStorage(vsharp);
-        if (!buffer_id) {
-            if (desc.is_gds_buffer) {
-                const auto* gds_buf = buffer_cache.GetGdsBuffer();
-                buffer_infos.emplace_back(gds_buf->Handle(), 0, gds_buf->SizeBytes());
-            } else if (instance.IsNullDescriptorSupported()) {
-                buffer_infos.emplace_back(VK_NULL_HANDLE, 0, VK_WHOLE_SIZE);
-            } else {
-                auto& null_buffer = buffer_cache.GetBuffer(VideoCore::NULL_BUFFER_ID);
-                buffer_infos.emplace_back(null_buffer.Handle(), 0, VK_WHOLE_SIZE);
-            }
-        } else {
-            const auto [vk_buffer, offset] = buffer_cache.ObtainBuffer(
-                vsharp.base_address, vsharp.GetSize(), desc.is_written, false, buffer_id);
-            const u32 alignment =
-                is_storage ? instance.StorageMinAlignment() : instance.UniformMinAlignment();
-            const u32 offset_aligned = Common::AlignDown(offset, alignment);
-            const u32 adjust = offset - offset_aligned;
-            ASSERT(adjust % 4 == 0);
-            push_data.AddOffset(binding.buffer, adjust);
-            buffer_infos.emplace_back(vk_buffer->Handle(), offset_aligned,
-                                      vsharp.GetSize() + adjust);
-        }
-
-        set_writes.push_back({
-            .dstSet = VK_NULL_HANDLE,
-            .dstBinding = binding.unified++,
-            .dstArrayElement = 0,
-            .descriptorCount = 1,
-            .descriptorType = is_storage ? vk::DescriptorType::eStorageBuffer
-                                         : vk::DescriptorType::eUniformBuffer,
-            .pBufferInfo = &buffer_infos.back(),
-        });
-        ++binding.buffer;
-    }
-
-    const auto null_buffer_view =
-        instance.IsNullDescriptorSupported() ? VK_NULL_HANDLE : buffer_cache.NullBufferView();
-    for (u32 i = 0; i < texbuffer_bindings.size(); i++) {
-        const auto& [buffer_id, vsharp] = texbuffer_bindings[i];
-        const auto& desc = stage.texture_buffers[i];
-        vk::BufferView& buffer_view = buffer_views.emplace_back(null_buffer_view);
-        if (buffer_id) {
-            const u32 alignment = instance.TexelBufferMinAlignment();
-            const auto [vk_buffer, offset] = buffer_cache.ObtainBuffer(
-                vsharp.base_address, vsharp.GetSize(), desc.is_written, true, buffer_id);
-            const u32 fmt_stride = AmdGpu::NumBits(vsharp.GetDataFmt()) >> 3;
-            ASSERT_MSG(fmt_stride == vsharp.GetStride(),
-                       "Texel buffer stride must match format stride");
-            const u32 offset_aligned = Common::AlignDown(offset, alignment);
-            const u32 adjust = offset - offset_aligned;
-            ASSERT(adjust % fmt_stride == 0);
-            push_data.AddOffset(binding.buffer, adjust / fmt_stride);
-            buffer_view =
-                vk_buffer->View(offset_aligned, vsharp.GetSize() + adjust, desc.is_written,
-                                vsharp.GetDataFmt(), vsharp.GetNumberFmt());
-            if (auto barrier =
-                    vk_buffer->GetBarrier(desc.is_written ? vk::AccessFlagBits2::eShaderWrite
-                                                          : vk::AccessFlagBits2::eShaderRead,
-                                          vk::PipelineStageFlagBits2::eComputeShader)) {
-                buffer_barriers.emplace_back(*barrier);
-            }
-            if (desc.is_written) {
-                texture_cache.InvalidateMemoryFromGPU(vsharp.base_address, vsharp.GetSize());
-            }
-        }
-
-        set_writes.push_back({
-            .dstSet = VK_NULL_HANDLE,
-            .dstBinding = binding.unified++,
-            .dstArrayElement = 0,
-            .descriptorCount = 1,
-            .descriptorType = desc.is_written ? vk::DescriptorType::eStorageTexelBuffer
-                                              : vk::DescriptorType::eUniformTexelBuffer,
-            .pTexelBufferView = &buffer_view,
-        });
-        ++binding.buffer;
-    }
-}
-
-void Pipeline::BindTextures(VideoCore::TextureCache& texture_cache, const Shader::Info& stage,
-                            Shader::Backend::Bindings& binding,
-                            DescriptorWrites& set_writes) const {
-
-    using ImageBindingInfo = std::tuple<VideoCore::ImageId, AmdGpu::Image, Shader::ImageResource>;
-    static boost::container::static_vector<ImageBindingInfo, 32> image_bindings;
-
-    image_bindings.clear();
-
-    for (const auto& image_desc : stage.images) {
-        const auto tsharp = image_desc.GetSharp(stage);
-        if (tsharp.GetDataFmt() != AmdGpu::DataFormat::FormatInvalid) {
-            VideoCore::ImageInfo image_info{tsharp, image_desc};
-            const auto image_id = texture_cache.FindImage(image_info);
-            auto& image = texture_cache.GetImage(image_id);
-            image.flags |= VideoCore::ImageFlagBits::Bound;
-            image_bindings.emplace_back(image_id, tsharp, image_desc);
-        } else {
-            image_bindings.emplace_back(VideoCore::ImageId{}, tsharp, image_desc);
-        }
-
-        if (texture_cache.IsMeta(tsharp.Address())) {
-            LOG_WARNING(Render_Vulkan, "Unexpected metadata read by a PS shader (texture)");
-        }
-    }
-
-    // Second pass to re-bind images that were updated after binding
-    for (auto [image_id, tsharp, desc] : image_bindings) {
-        if (!image_id) {
-            if (instance.IsNullDescriptorSupported()) {
-                image_infos.emplace_back(VK_NULL_HANDLE, VK_NULL_HANDLE, vk::ImageLayout::eGeneral);
-            } else {
-                auto& null_image = texture_cache.GetImageView(VideoCore::NULL_IMAGE_VIEW_ID);
-                image_infos.emplace_back(VK_NULL_HANDLE, *null_image.image_view,
-                                         vk::ImageLayout::eGeneral);
-            }
-        } else {
-            auto& image = texture_cache.GetImage(image_id);
-            if (True(image.flags & VideoCore::ImageFlagBits::NeedsRebind)) {
-                image_id = texture_cache.FindImage(image.info);
-            }
-            VideoCore::ImageViewInfo view_info{tsharp, desc};
-            auto& image_view = texture_cache.FindTexture(image_id, view_info);
-            image_infos.emplace_back(VK_NULL_HANDLE, *image_view.image_view,
-                                     texture_cache.GetImage(image_id).last_state.layout);
-            image.flags &=
-                ~(VideoCore::ImageFlagBits::NeedsRebind | VideoCore::ImageFlagBits::Bound);
-        }
-
-        set_writes.push_back({
-            .dstSet = VK_NULL_HANDLE,
-            .dstBinding = binding.unified++,
-            .dstArrayElement = 0,
-            .descriptorCount = 1,
-            .descriptorType = desc.is_storage ? vk::DescriptorType::eStorageImage
-                                              : vk::DescriptorType::eSampledImage,
-            .pImageInfo = &image_infos.back(),
-        });
-    }
-
-    for (const auto& sampler : stage.samplers) {
-        auto ssharp = sampler.GetSharp(stage);
-        if (sampler.disable_aniso) {
-            const auto& tsharp = stage.images[sampler.associated_image].GetSharp(stage);
-            if (tsharp.base_level == 0 && tsharp.last_level == 0) {
-                ssharp.max_aniso.Assign(AmdGpu::AnisoRatio::One);
-            }
-        }
-        const auto vk_sampler = texture_cache.GetSampler(ssharp);
-        image_infos.emplace_back(vk_sampler, VK_NULL_HANDLE, vk::ImageLayout::eGeneral);
-        set_writes.push_back({
-            .dstSet = VK_NULL_HANDLE,
-            .dstBinding = binding.unified++,
-            .dstArrayElement = 0,
-            .descriptorCount = 1,
-            .descriptorType = vk::DescriptorType::eSampler,
-            .pImageInfo = &image_infos.back(),
-        });
     }
+    instance.GetDevice().updateDescriptorSets(set_writes, {});
+    cmdbuf.bindDescriptorSets(bind_point, *pipeline_layout, 0, desc_set, {});
 }
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_pipeline_common.h

@@ -6,14 +6,18 @@
 #include "shader_recompiler/backend/bindings.h"
 #include "shader_recompiler/info.h"
 #include "video_core/renderer_vulkan/vk_common.h"
+#include "video_core/texture_cache/texture_cache.h"
 
 namespace VideoCore {
 class BufferCache;
-class TextureCache;
 } // namespace VideoCore
 
 namespace Vulkan {
 
+static constexpr auto gp_stage_flags = vk::ShaderStageFlagBits::eVertex |
+                                       vk::ShaderStageFlagBits::eGeometry |
+                                       vk::ShaderStageFlagBits::eFragment;
+
 class Instance;
 class Scheduler;
 class DescriptorHeap;
@@ -21,7 +25,7 @@ class DescriptorHeap;
 class Pipeline {
 public:
     Pipeline(const Instance& instance, Scheduler& scheduler, DescriptorHeap& desc_heap,
-             vk::PipelineCache pipeline_cache);
+             vk::PipelineCache pipeline_cache, bool is_compute = false);
     virtual ~Pipeline();
 
     vk::Pipeline Handle() const noexcept {
@@ -32,16 +36,27 @@ public:
         return *pipeline_layout;
     }
 
+    auto GetStages() const {
+        if (is_compute) {
+            return std::span{stages.cend() - 1, stages.cend()};
+        } else {
+            return std::span{stages.cbegin(), stages.cend() - 1};
+        }
+    }
+
+    const Shader::Info& GetStage(Shader::Stage stage) const noexcept {
+        return *stages[u32(stage)];
+    }
+
+    bool IsCompute() const {
+        return is_compute;
+    }
+
     using DescriptorWrites = boost::container::small_vector<vk::WriteDescriptorSet, 16>;
     using BufferBarriers = boost::container::small_vector<vk::BufferMemoryBarrier2, 16>;
 
-    void BindBuffers(VideoCore::BufferCache& buffer_cache, VideoCore::TextureCache& texture_cache,
+    void BindResources(DescriptorWrites& set_writes, const BufferBarriers& buffer_barriers,
-                     const Shader::Info& stage, Shader::Backend::Bindings& binding,
+                       const Shader::PushData& push_data) const;
-                     Shader::PushData& push_data, DescriptorWrites& set_writes,
-                     BufferBarriers& buffer_barriers) const;
-
-    void BindTextures(VideoCore::TextureCache& texture_cache, const Shader::Info& stage,
-                      Shader::Backend::Bindings& binding, DescriptorWrites& set_writes) const;
 
 protected:
     const Instance& instance;
@@ -50,9 +65,9 @@ protected:
     vk::UniquePipeline pipeline;
     vk::UniquePipelineLayout pipeline_layout;
     vk::UniqueDescriptorSetLayout desc_layout;
-    static boost::container::static_vector<vk::DescriptorImageInfo, 32> image_infos;
+    std::array<const Shader::Info*, Shader::MaxStageTypes> stages{};
-    static boost::container::static_vector<vk::BufferView, 8> buffer_views;
+    bool uses_push_descriptors{};
-    static boost::container::static_vector<vk::DescriptorBufferInfo, 32> buffer_infos;
+    const bool is_compute;
 };
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_presenter.h

@@ -55,8 +55,8 @@ public:
 
     Frame* PrepareFrame(const Libraries::VideoOut::BufferAttributeGroup& attribute,
                         VAddr cpu_address, bool is_eop) {
-        const auto info = VideoCore::ImageInfo{attribute, cpu_address};
+        auto desc = VideoCore::TextureCache::VideoOutDesc{attribute, cpu_address};
-        const auto image_id = texture_cache.FindImage(info);
+        const auto image_id = texture_cache.FindImage(desc);
         texture_cache.UpdateImage(image_id, is_eop ? nullptr : &flip_scheduler);
         return PrepareFrameInternal(image_id, is_eop);
     }
@@ -68,9 +68,11 @@ public:
     VideoCore::Image& RegisterVideoOutSurface(
         const Libraries::VideoOut::BufferAttributeGroup& attribute, VAddr cpu_address) {
         vo_buffers_addr.emplace_back(cpu_address);
-        const auto info = VideoCore::ImageInfo{attribute, cpu_address};
+        auto desc = VideoCore::TextureCache::VideoOutDesc{attribute, cpu_address};
-        const auto image_id = texture_cache.FindImage(info);
+        const auto image_id = texture_cache.FindImage(desc);
-        return texture_cache.GetImage(image_id);
+        auto& image = texture_cache.GetImage(image_id);
+        image.usage.vo_surface = 1u;
+        return image;
     }
 
     bool IsVideoOutSurface(const AmdGpu::Liverpool::ColorBuffer& color_buffer) {

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -75,6 +75,105 @@ bool Rasterizer::FilterDraw() {
     return true;
 }
 
+RenderState Rasterizer::PrepareRenderState(u32 mrt_mask) {
+    // Prefetch color and depth buffers to let texture cache handle possible overlaps with bound
+    // textures (e.g. mipgen)
+    RenderState state;
+
+    cb_descs.clear();
+    db_desc.reset();
+
+    const auto& regs = liverpool->regs;
+
+    if (regs.color_control.degamma_enable) {
+        LOG_WARNING(Render_Vulkan, "Color buffers require gamma correction");
+    }
+
+    for (auto col_buf_id = 0u; col_buf_id < Liverpool::NumColorBuffers; ++col_buf_id) {
+        const auto& col_buf = regs.color_buffers[col_buf_id];
+        if (!col_buf) {
+            continue;
+        }
+
+        // If the color buffer is still bound but rendering to it is disabled by the target
+        // mask, we need to prevent the render area from being affected by unbound render target
+        // extents.
+        if (!regs.color_target_mask.GetMask(col_buf_id)) {
+            continue;
+        }
+
+        // Skip stale color buffers if shader doesn't output to them. Otherwise it will perform
+        // an unnecessary transition and may result in state conflict if the resource is already
+        // bound for reading.
+        if ((mrt_mask & (1 << col_buf_id)) == 0) {
+            continue;
+        }
+
+        const bool is_clear = texture_cache.IsMetaCleared(col_buf.CmaskAddress());
+        texture_cache.TouchMeta(col_buf.CmaskAddress(), false);
+
+        const auto& hint = liverpool->last_cb_extent[col_buf_id];
+        auto& [image_id, desc] = cb_descs.emplace_back(std::piecewise_construct, std::tuple{},
+                                                       std::tuple{col_buf, hint});
+        const auto& image_view = texture_cache.FindRenderTarget(desc);
+        image_id = bound_images.emplace_back(image_view.image_id);
+        auto& image = texture_cache.GetImage(image_id);
+        image.binding.is_target = 1u;
+
+        const auto mip = image_view.info.range.base.level;
+        state.width = std::min<u32>(state.width, std::max(image.info.size.width >> mip, 1u));
+        state.height = std::min<u32>(state.height, std::max(image.info.size.height >> mip, 1u));
+        state.color_images[state.num_color_attachments] = image.image;
+        state.color_attachments[state.num_color_attachments++] = {
+            .imageView = *image_view.image_view,
+            .imageLayout = vk::ImageLayout::eUndefined,
+            .loadOp = is_clear ? vk::AttachmentLoadOp::eClear : vk::AttachmentLoadOp::eLoad,
+            .storeOp = vk::AttachmentStoreOp::eStore,
+            .clearValue =
+                is_clear ? LiverpoolToVK::ColorBufferClearValue(col_buf) : vk::ClearValue{},
+        };
+    }
+
+    using ZFormat = AmdGpu::Liverpool::DepthBuffer::ZFormat;
+    using StencilFormat = AmdGpu::Liverpool::DepthBuffer::StencilFormat;
+    if (regs.depth_buffer.Address() != 0 &&
+        ((regs.depth_control.depth_enable && regs.depth_buffer.z_info.format != ZFormat::Invalid) ||
+         (regs.depth_control.stencil_enable &&
+          regs.depth_buffer.stencil_info.format != StencilFormat::Invalid))) {
+        const auto htile_address = regs.depth_htile_data_base.GetAddress();
+        const bool is_clear = regs.depth_render_control.depth_clear_enable ||
+                              texture_cache.IsMetaCleared(htile_address);
+        const auto& hint = liverpool->last_db_extent;
+        auto& [image_id, desc] =
+            db_desc.emplace(std::piecewise_construct, std::tuple{},
+                            std::tuple{regs.depth_buffer, regs.depth_view, regs.depth_control,
+                                       htile_address, hint});
+        const auto& image_view = texture_cache.FindDepthTarget(desc);
+        image_id = bound_images.emplace_back(image_view.image_id);
+        auto& image = texture_cache.GetImage(image_id);
+        image.binding.is_target = 1u;
+
+        state.width = std::min<u32>(state.width, image.info.size.width);
+        state.height = std::min<u32>(state.height, image.info.size.height);
+        state.depth_image = image.image;
+        state.depth_attachment = {
+            .imageView = *image_view.image_view,
+            .imageLayout = vk::ImageLayout::eUndefined,
+            .loadOp = is_clear ? vk::AttachmentLoadOp::eClear : vk::AttachmentLoadOp::eLoad,
+            .storeOp = vk::AttachmentStoreOp::eStore,
+            .clearValue = vk::ClearValue{.depthStencil = {.depth = regs.depth_clear,
+                                                          .stencil = regs.stencil_clear}},
+        };
+        texture_cache.TouchMeta(htile_address, false);
+        state.has_depth =
+            regs.depth_buffer.z_info.format != AmdGpu::Liverpool::DepthBuffer::ZFormat::Invalid;
+        state.has_stencil = regs.depth_buffer.stencil_info.format !=
+                            AmdGpu::Liverpool::DepthBuffer::StencilFormat::Invalid;
+    }
+
+    return state;
+}
+
 void Rasterizer::Draw(bool is_indexed, u32 index_offset) {
     RENDERER_TRACE;
 
@@ -82,27 +181,29 @@ void Rasterizer::Draw(bool is_indexed, u32 index_offset) {
         return;
     }
 
-    const auto cmdbuf = scheduler.CommandBuffer();
     const auto& regs = liverpool->regs;
     const GraphicsPipeline* pipeline = pipeline_cache.GetGraphicsPipeline();
     if (!pipeline) {
         return;
     }
 
-    try {
+    auto state = PrepareRenderState(pipeline->GetMrtMask());
-        pipeline->BindResources(regs, buffer_cache, texture_cache);
+
-    } catch (...) {
+    if (!BindResources(pipeline)) {
-        UNREACHABLE();
+        return;
     }
 
     const auto& vs_info = pipeline->GetStage(Shader::Stage::Vertex);
     buffer_cache.BindVertexBuffers(vs_info);
     const u32 num_indices = buffer_cache.BindIndexBuffer(is_indexed, index_offset);
 
-    BeginRendering(*pipeline);
+    BeginRendering(*pipeline, state);
     UpdateDynamicState(*pipeline);
 
-    const auto [vertex_offset, instance_offset] = vs_info.GetDrawOffsets();
+    const auto [vertex_offset, instance_offset] = vs_info.GetDrawOffsets(regs);
+
+    const auto cmdbuf = scheduler.CommandBuffer();
+    cmdbuf.bindPipeline(vk::PipelineBindPoint::eGraphics, pipeline->Handle());
 
     if (is_indexed) {
         cmdbuf.drawIndexed(num_indices, regs.num_instances.NumInstances(), 0, s32(vertex_offset),
@@ -113,9 +214,11 @@ void Rasterizer::Draw(bool is_indexed, u32 index_offset) {
         cmdbuf.draw(num_vertices, regs.num_instances.NumInstances(), vertex_offset,
                     instance_offset);
     }
+
+    ResetBindings();
 }
 
-void Rasterizer::DrawIndirect(bool is_indexed, VAddr arg_address, u32 offset, u32 size,
+void Rasterizer::DrawIndirect(bool is_indexed, VAddr arg_address, u32 offset, u32 stride,
                               u32 max_count, VAddr count_address) {
     RENDERER_TRACE;
 
@@ -124,25 +227,34 @@ void Rasterizer::DrawIndirect(bool is_indexed, VAddr arg_address, u32 offset, u3
     }
 
     const auto& regs = liverpool->regs;
-    const GraphicsPipeline* pipeline = pipeline_cache.GetGraphicsPipeline();
+    if (regs.primitive_type == AmdGpu::PrimitiveType::QuadList) {
-    if (!pipeline) {
+        // For QuadList we use generated index buffer to convert quads to triangles. Since it
+        // changes type of the draw, arguments are not valid for this case. We need to run a
+        // conversion pass to repack the indirect arguments buffer first.
+        LOG_WARNING(Render_Vulkan, "QuadList primitive type is not supported for indirect draw");
         return;
     }
 
     ASSERT_MSG(regs.primitive_type != AmdGpu::PrimitiveType::RectList,
                "Unsupported primitive type for indirect draw");
 
-    try {
+    const GraphicsPipeline* pipeline = pipeline_cache.GetGraphicsPipeline();
-        pipeline->BindResources(regs, buffer_cache, texture_cache);
+    if (!pipeline) {
-    } catch (...) {
+        return;
-        UNREACHABLE();
+    }
+
+    auto state = PrepareRenderState(pipeline->GetMrtMask());
+
+    if (!BindResources(pipeline)) {
+        return;
     }
 
     const auto& vs_info = pipeline->GetStage(Shader::Stage::Vertex);
     buffer_cache.BindVertexBuffers(vs_info);
     buffer_cache.BindIndexBuffer(is_indexed, 0);
 
-    const auto [buffer, base] = buffer_cache.ObtainBuffer(arg_address + offset, size, false);
+    const auto& [buffer, base] =
+        buffer_cache.ObtainBuffer(arg_address + offset, stride * max_count, false);
 
     VideoCore::Buffer* count_buffer{};
     u32 count_base{};
@@ -150,36 +262,36 @@ void Rasterizer::DrawIndirect(bool is_indexed, VAddr arg_address, u32 offset, u3
         std::tie(count_buffer, count_base) = buffer_cache.ObtainBuffer(count_address, 4, false);
     }
 
-    BeginRendering(*pipeline);
+    BeginRendering(*pipeline, state);
     UpdateDynamicState(*pipeline);
 
     // We can safely ignore both SGPR UD indices and results of fetch shader parsing, as vertex and
     // instance offsets will be automatically applied by Vulkan from indirect args buffer.
 
     const auto cmdbuf = scheduler.CommandBuffer();
+    cmdbuf.bindPipeline(vk::PipelineBindPoint::eGraphics, pipeline->Handle());
+
     if (is_indexed) {
-        static_assert(sizeof(VkDrawIndexedIndirectCommand) ==
+        ASSERT(sizeof(VkDrawIndexedIndirectCommand) == stride);
-                      AmdGpu::Liverpool::DrawIndexedIndirectArgsSize);
 
         if (count_address != 0) {
             cmdbuf.drawIndexedIndirectCount(buffer->Handle(), base, count_buffer->Handle(),
-                                            count_base, max_count,
+                                            count_base, max_count, stride);
-                                            AmdGpu::Liverpool::DrawIndexedIndirectArgsSize);
         } else {
-            cmdbuf.drawIndexedIndirect(buffer->Handle(), base, max_count,
+            cmdbuf.drawIndexedIndirect(buffer->Handle(), base, max_count, stride);
-                                       AmdGpu::Liverpool::DrawIndexedIndirectArgsSize);
         }
     } else {
-        static_assert(sizeof(VkDrawIndirectCommand) == AmdGpu::Liverpool::DrawIndirectArgsSize);
+        ASSERT(sizeof(VkDrawIndirectCommand) == stride);
 
         if (count_address != 0) {
             cmdbuf.drawIndirectCount(buffer->Handle(), base, count_buffer->Handle(), count_base,
-                                     max_count, AmdGpu::Liverpool::DrawIndirectArgsSize);
+                                     max_count, stride);
         } else {
-            cmdbuf.drawIndirect(buffer->Handle(), base, max_count,
+            cmdbuf.drawIndirect(buffer->Handle(), base, max_count, stride);
-                                AmdGpu::Liverpool::DrawIndirectArgsSize);
         }
     }
+
+    ResetBindings();
 }
 
 void Rasterizer::DispatchDirect() {
@@ -192,18 +304,15 @@ void Rasterizer::DispatchDirect() {
         return;
     }
 
-    try {
+    if (!BindResources(pipeline)) {
-        const auto has_resources = pipeline->BindResources(buffer_cache, texture_cache);
+        return;
-        if (!has_resources) {
-            return;
-        }
-    } catch (...) {
-        UNREACHABLE();
     }
 
     scheduler.EndRendering();
     cmdbuf.bindPipeline(vk::PipelineBindPoint::eCompute, pipeline->Handle());
     cmdbuf.dispatch(cs_program.dim_x, cs_program.dim_y, cs_program.dim_z);
+
+    ResetBindings();
 }
 
 void Rasterizer::DispatchIndirect(VAddr address, u32 offset, u32 size) {
@@ -216,19 +325,16 @@ void Rasterizer::DispatchIndirect(VAddr address, u32 offset, u32 size) {
         return;
     }
 
-    try {
+    if (!BindResources(pipeline)) {
-        const auto has_resources = pipeline->BindResources(buffer_cache, texture_cache);
+        return;
-        if (!has_resources) {
-            return;
-        }
-    } catch (...) {
-        UNREACHABLE();
     }
 
     scheduler.EndRendering();
     cmdbuf.bindPipeline(vk::PipelineBindPoint::eCompute, pipeline->Handle());
     const auto [buffer, base] = buffer_cache.ObtainBuffer(address + offset, size, false);
     cmdbuf.dispatchIndirect(buffer->Handle(), base);
+
+    ResetBindings();
 }
 
 u64 Rasterizer::Flush() {
@@ -242,86 +348,386 @@ void Rasterizer::Finish() {
     scheduler.Finish();
 }
 
-void Rasterizer::BeginRendering(const GraphicsPipeline& pipeline) {
+bool Rasterizer::BindResources(const Pipeline* pipeline) {
+    buffer_infos.clear();
+    buffer_views.clear();
+    image_infos.clear();
+
     const auto& regs = liverpool->regs;
-    RenderState state;
 
-    if (regs.color_control.degamma_enable) {
+    if (pipeline->IsCompute()) {
-        LOG_WARNING(Render_Vulkan, "Color buffers require gamma correction");
+        const auto& info = pipeline->GetStage(Shader::Stage::Compute);
-    }
 
-    for (auto col_buf_id = 0u; col_buf_id < Liverpool::NumColorBuffers; ++col_buf_id) {
+        // Most of the time when a metadata is updated with a shader it gets cleared. It means
-        const auto& col_buf = regs.color_buffers[col_buf_id];
+        // we can skip the whole dispatch and update the tracked state instead. Also, it is not
-        if (!col_buf) {
+        // intended to be consumed and in such rare cases (e.g. HTile introspection, CRAA) we
-            continue;
+        // will need its full emulation anyways. For cases of metadata read a warning will be
-        }
+        // logged.
-
+        const auto IsMetaUpdate = [&](const auto& desc) {
-        // If the color buffer is still bound but rendering to it is disabled by the target mask,
+            const VAddr address = desc.GetSharp(info).base_address;
-        // we need to prevent the render area from being affected by unbound render target extents.
+            if (desc.is_written) {
-        if (!regs.color_target_mask.GetMask(col_buf_id)) {
+                if (texture_cache.TouchMeta(address, true)) {
-            continue;
+                    LOG_TRACE(Render_Vulkan, "Metadata update skipped");
-        }
+                    return true;
-
+                }
-        // Skip stale color buffers if shader doesn't output to them. Otherwise it will perform
+            } else {
-        // an unnecessary transition and may result in state conflict if the resource is already
+                if (texture_cache.IsMeta(address)) {
-        // bound for reading.
+                    LOG_WARNING(Render_Vulkan, "Unexpected metadata read by a CS shader (buffer)");
-        if ((pipeline.GetMrtMask() & (1 << col_buf_id)) == 0) {
+                }
-            continue;
+            }
-        }
+            return false;
-
-        const auto& hint = liverpool->last_cb_extent[col_buf_id];
-        VideoCore::ImageInfo image_info{col_buf, hint};
-        VideoCore::ImageViewInfo view_info{col_buf};
-        const auto& image_view = texture_cache.FindRenderTarget(image_info, view_info);
-        const auto& image = texture_cache.GetImage(image_view.image_id);
-        state.width = std::min<u32>(state.width, image.info.size.width);
-        state.height = std::min<u32>(state.height, image.info.size.height);
-
-        const bool is_clear = texture_cache.IsMetaCleared(col_buf.CmaskAddress());
-        state.color_images[state.num_color_attachments] = image.image;
-        state.color_attachments[state.num_color_attachments++] = {
-            .imageView = *image_view.image_view,
-            .imageLayout = vk::ImageLayout::eColorAttachmentOptimal,
-            .loadOp = is_clear ? vk::AttachmentLoadOp::eClear : vk::AttachmentLoadOp::eLoad,
-            .storeOp = vk::AttachmentStoreOp::eStore,
-            .clearValue =
-                is_clear ? LiverpoolToVK::ColorBufferClearValue(col_buf) : vk::ClearValue{},
         };
-        texture_cache.TouchMeta(col_buf.CmaskAddress(), false);
+
+        for (const auto& desc : info.buffers) {
+            if (desc.is_gds_buffer) {
+                continue;
+            }
+            if (IsMetaUpdate(desc)) {
+                return false;
+            }
+        }
+        for (const auto& desc : info.texture_buffers) {
+            if (IsMetaUpdate(desc)) {
+                return false;
+            }
+        }
     }
 
-    using ZFormat = AmdGpu::Liverpool::DepthBuffer::ZFormat;
+    set_writes.clear();
-    using StencilFormat = AmdGpu::Liverpool::DepthBuffer::StencilFormat;
+    buffer_barriers.clear();
-    if (regs.depth_buffer.Address() != 0 &&
+
-        ((regs.depth_control.depth_enable && regs.depth_buffer.z_info.format != ZFormat::Invalid) ||
+    // Bind resource buffers and textures.
-         (regs.depth_control.stencil_enable &&
+    Shader::PushData push_data{};
-          regs.depth_buffer.stencil_info.format != StencilFormat::Invalid))) {
+    Shader::Backend::Bindings binding{};
-        const auto htile_address = regs.depth_htile_data_base.GetAddress();
+
-        const bool is_clear = regs.depth_render_control.depth_clear_enable ||
+    for (const auto* stage : pipeline->GetStages()) {
-                              texture_cache.IsMetaCleared(htile_address);
+        if (!stage) {
-        const auto& hint = liverpool->last_db_extent;
+            continue;
-        VideoCore::ImageInfo image_info{regs.depth_buffer, regs.depth_view.NumSlices(),
+        }
-                                        htile_address, hint};
+        if (stage->uses_step_rates) {
-        VideoCore::ImageViewInfo view_info{regs.depth_buffer, regs.depth_view, regs.depth_control};
+            push_data.step0 = regs.vgt_instance_step_rate_0;
-        const auto& image_view = texture_cache.FindDepthTarget(image_info, view_info);
+            push_data.step1 = regs.vgt_instance_step_rate_1;
-        const auto& image = texture_cache.GetImage(image_view.image_id);
+        }
-        state.width = std::min<u32>(state.width, image.info.size.width);
+        stage->PushUd(binding, push_data);
-        state.height = std::min<u32>(state.height, image.info.size.height);
+
-        state.depth_image = image.image;
+        BindBuffers(*stage, binding, push_data, set_writes, buffer_barriers);
-        state.depth_attachment = {
+        BindTextures(*stage, binding, set_writes);
-            .imageView = *image_view.image_view,
-            .imageLayout = image.last_state.layout,
-            .loadOp = is_clear ? vk::AttachmentLoadOp::eClear : vk::AttachmentLoadOp::eLoad,
-            .storeOp = is_clear ? vk::AttachmentStoreOp::eNone : vk::AttachmentStoreOp::eStore,
-            .clearValue = vk::ClearValue{.depthStencil = {.depth = regs.depth_clear,
-                                                          .stencil = regs.stencil_clear}},
-        };
-        texture_cache.TouchMeta(htile_address, false);
-        state.has_depth =
-            regs.depth_buffer.z_info.format != AmdGpu::Liverpool::DepthBuffer::ZFormat::Invalid;
-        state.has_stencil = regs.depth_buffer.stencil_info.format !=
-                            AmdGpu::Liverpool::DepthBuffer::StencilFormat::Invalid;
     }
+
+    pipeline->BindResources(set_writes, buffer_barriers, push_data);
+
+    return true;
+}
+
+void Rasterizer::BindBuffers(const Shader::Info& stage, Shader::Backend::Bindings& binding,
+                             Shader::PushData& push_data, Pipeline::DescriptorWrites& set_writes,
+                             Pipeline::BufferBarriers& buffer_barriers) {
+    buffer_bindings.clear();
+
+    for (const auto& desc : stage.buffers) {
+        const auto vsharp = desc.GetSharp(stage);
+        if (!desc.is_gds_buffer && vsharp.base_address != 0 && vsharp.GetSize() > 0) {
+            const auto buffer_id = buffer_cache.FindBuffer(vsharp.base_address, vsharp.GetSize());
+            buffer_bindings.emplace_back(buffer_id, vsharp);
+        } else {
+            buffer_bindings.emplace_back(VideoCore::BufferId{}, vsharp);
+        }
+    }
+
+    texbuffer_bindings.clear();
+
+    for (const auto& desc : stage.texture_buffers) {
+        const auto vsharp = desc.GetSharp(stage);
+        if (vsharp.base_address != 0 && vsharp.GetSize() > 0 &&
+            vsharp.GetDataFmt() != AmdGpu::DataFormat::FormatInvalid) {
+            const auto buffer_id = buffer_cache.FindBuffer(vsharp.base_address, vsharp.GetSize());
+            texbuffer_bindings.emplace_back(buffer_id, vsharp);
+        } else {
+            texbuffer_bindings.emplace_back(VideoCore::BufferId{}, vsharp);
+        }
+    }
+
+    // Bind the flattened user data buffer as a UBO so it's accessible to the shader
+    if (stage.has_readconst) {
+        const auto [vk_buffer, offset] = buffer_cache.ObtainHostUBO(stage.flattened_ud_buf);
+        buffer_infos.emplace_back(vk_buffer->Handle(), offset,
+                                  stage.flattened_ud_buf.size() * sizeof(u32));
+        set_writes.push_back({
+            .dstSet = VK_NULL_HANDLE,
+            .dstBinding = binding.unified++,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = vk::DescriptorType::eUniformBuffer,
+            .pBufferInfo = &buffer_infos.back(),
+        });
+        ++binding.buffer;
+    }
+
+    // Second pass to re-bind buffers that were updated after binding
+    for (u32 i = 0; i < buffer_bindings.size(); i++) {
+        const auto& [buffer_id, vsharp] = buffer_bindings[i];
+        const auto& desc = stage.buffers[i];
+        const bool is_storage = desc.IsStorage(vsharp);
+        if (!buffer_id) {
+            if (desc.is_gds_buffer) {
+                const auto* gds_buf = buffer_cache.GetGdsBuffer();
+                buffer_infos.emplace_back(gds_buf->Handle(), 0, gds_buf->SizeBytes());
+            } else if (instance.IsNullDescriptorSupported()) {
+                buffer_infos.emplace_back(VK_NULL_HANDLE, 0, VK_WHOLE_SIZE);
+            } else {
+                auto& null_buffer = buffer_cache.GetBuffer(VideoCore::NULL_BUFFER_ID);
+                buffer_infos.emplace_back(null_buffer.Handle(), 0, VK_WHOLE_SIZE);
+            }
+        } else {
+            const auto [vk_buffer, offset] = buffer_cache.ObtainBuffer(
+                vsharp.base_address, vsharp.GetSize(), desc.is_written, false, buffer_id);
+            const u32 alignment =
+                is_storage ? instance.StorageMinAlignment() : instance.UniformMinAlignment();
+            const u32 offset_aligned = Common::AlignDown(offset, alignment);
+            const u32 adjust = offset - offset_aligned;
+            ASSERT(adjust % 4 == 0);
+            push_data.AddOffset(binding.buffer, adjust);
+            buffer_infos.emplace_back(vk_buffer->Handle(), offset_aligned,
+                                      vsharp.GetSize() + adjust);
+        }
+
+        set_writes.push_back({
+            .dstSet = VK_NULL_HANDLE,
+            .dstBinding = binding.unified++,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = is_storage ? vk::DescriptorType::eStorageBuffer
+                                         : vk::DescriptorType::eUniformBuffer,
+            .pBufferInfo = &buffer_infos.back(),
+        });
+        ++binding.buffer;
+    }
+
+    const auto null_buffer_view =
+        instance.IsNullDescriptorSupported() ? VK_NULL_HANDLE : buffer_cache.NullBufferView();
+    for (u32 i = 0; i < texbuffer_bindings.size(); i++) {
+        const auto& [buffer_id, vsharp] = texbuffer_bindings[i];
+        const auto& desc = stage.texture_buffers[i];
+        vk::BufferView& buffer_view = buffer_views.emplace_back(null_buffer_view);
+        if (buffer_id) {
+            const u32 alignment = instance.TexelBufferMinAlignment();
+            const auto [vk_buffer, offset] = buffer_cache.ObtainBuffer(
+                vsharp.base_address, vsharp.GetSize(), desc.is_written, true, buffer_id);
+            const u32 fmt_stride = AmdGpu::NumBits(vsharp.GetDataFmt()) >> 3;
+            ASSERT_MSG(fmt_stride == vsharp.GetStride(),
+                       "Texel buffer stride must match format stride");
+            const u32 offset_aligned = Common::AlignDown(offset, alignment);
+            const u32 adjust = offset - offset_aligned;
+            ASSERT(adjust % fmt_stride == 0);
+            push_data.AddOffset(binding.buffer, adjust / fmt_stride);
+            buffer_view =
+                vk_buffer->View(offset_aligned, vsharp.GetSize() + adjust, desc.is_written,
+                                vsharp.GetDataFmt(), vsharp.GetNumberFmt());
+            if (auto barrier =
+                    vk_buffer->GetBarrier(desc.is_written ? vk::AccessFlagBits2::eShaderWrite
+                                                          : vk::AccessFlagBits2::eShaderRead,
+                                          vk::PipelineStageFlagBits2::eComputeShader)) {
+                buffer_barriers.emplace_back(*barrier);
+            }
+            if (desc.is_written) {
+                texture_cache.InvalidateMemoryFromGPU(vsharp.base_address, vsharp.GetSize());
+            }
+        }
+
+        set_writes.push_back({
+            .dstSet = VK_NULL_HANDLE,
+            .dstBinding = binding.unified++,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = desc.is_written ? vk::DescriptorType::eStorageTexelBuffer
+                                              : vk::DescriptorType::eUniformTexelBuffer,
+            .pTexelBufferView = &buffer_view,
+        });
+        ++binding.buffer;
+    }
+}
+
+void Rasterizer::BindTextures(const Shader::Info& stage, Shader::Backend::Bindings& binding,
+                              Pipeline::DescriptorWrites& set_writes) {
+    image_bindings.clear();
+
+    for (const auto& image_desc : stage.images) {
+        const auto tsharp = image_desc.GetSharp(stage);
+        if (texture_cache.IsMeta(tsharp.Address())) {
+            LOG_WARNING(Render_Vulkan, "Unexpected metadata read by a shader (texture)");
+        }
+
+        if (tsharp.GetDataFmt() == AmdGpu::DataFormat::FormatInvalid) {
+            image_bindings.emplace_back(std::piecewise_construct, std::tuple{}, std::tuple{});
+            continue;
+        }
+
+        auto& [image_id, desc] = image_bindings.emplace_back(std::piecewise_construct, std::tuple{},
+                                                             std::tuple{tsharp, image_desc});
+        image_id = texture_cache.FindImage(desc);
+        auto& image = texture_cache.GetImage(image_id);
+        if (image.binding.is_bound) {
+            // The image is already bound. In case if it is about to be used as storage we need
+            // to force general layout on it.
+            image.binding.force_general |= image_desc.is_storage;
+        }
+        if (image.binding.is_target) {
+            // The image is already bound as target. Since we read and output to it need to force
+            // general layout too.
+            image.binding.force_general = 1u;
+        }
+        image.binding.is_bound = 1u;
+    }
+
+    // Second pass to re-bind images that were updated after binding
+    for (auto& [image_id, desc] : image_bindings) {
+        bool is_storage = desc.type == VideoCore::TextureCache::BindingType::Storage;
+        if (!image_id) {
+            if (instance.IsNullDescriptorSupported()) {
+                image_infos.emplace_back(VK_NULL_HANDLE, VK_NULL_HANDLE, vk::ImageLayout::eGeneral);
+            } else {
+                auto& null_image = texture_cache.GetImageView(VideoCore::NULL_IMAGE_VIEW_ID);
+                image_infos.emplace_back(VK_NULL_HANDLE, *null_image.image_view,
+                                         vk::ImageLayout::eGeneral);
+            }
+        } else {
+            if (auto& old_image = texture_cache.GetImage(image_id);
+                old_image.binding.needs_rebind) {
+                old_image.binding.Reset(); // clean up previous image binding state
+                image_id = texture_cache.FindImage(desc);
+            }
+
+            bound_images.emplace_back(image_id);
+
+            auto& image = texture_cache.GetImage(image_id);
+            auto& image_view = texture_cache.FindTexture(image_id, desc.view_info);
+
+            if (image.binding.force_general || image.binding.is_target) {
+                image.Transit(vk::ImageLayout::eGeneral,
+                              vk::AccessFlagBits2::eShaderRead |
+                                  (image.info.IsDepthStencil()
+                                       ? vk::AccessFlagBits2::eDepthStencilAttachmentWrite
+                                       : vk::AccessFlagBits2::eColorAttachmentWrite),
+                              {});
+            } else {
+                if (is_storage) {
+                    image.Transit(vk::ImageLayout::eGeneral,
+                                  vk::AccessFlagBits2::eShaderRead |
+                                      vk::AccessFlagBits2::eShaderWrite,
+                                  desc.view_info.range);
+                } else {
+                    const auto new_layout = image.info.IsDepthStencil()
+                                                ? vk::ImageLayout::eDepthStencilReadOnlyOptimal
+                                                : vk::ImageLayout::eShaderReadOnlyOptimal;
+                    image.Transit(new_layout, vk::AccessFlagBits2::eShaderRead,
+                                  desc.view_info.range);
+                }
+            }
+            image.usage.storage |= is_storage;
+            image.usage.texture |= !is_storage;
+
+            image_infos.emplace_back(VK_NULL_HANDLE, *image_view.image_view,
+                                     image.last_state.layout);
+        }
+
+        set_writes.push_back({
+            .dstSet = VK_NULL_HANDLE,
+            .dstBinding = binding.unified++,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType =
+                is_storage ? vk::DescriptorType::eStorageImage : vk::DescriptorType::eSampledImage,
+            .pImageInfo = &image_infos.back(),
+        });
+    }
+
+    for (const auto& sampler : stage.samplers) {
+        auto ssharp = sampler.GetSharp(stage);
+        if (sampler.disable_aniso) {
+            const auto& tsharp = stage.images[sampler.associated_image].GetSharp(stage);
+            if (tsharp.base_level == 0 && tsharp.last_level == 0) {
+                ssharp.max_aniso.Assign(AmdGpu::AnisoRatio::One);
+            }
+        }
+        const auto vk_sampler = texture_cache.GetSampler(ssharp);
+        image_infos.emplace_back(vk_sampler, VK_NULL_HANDLE, vk::ImageLayout::eGeneral);
+        set_writes.push_back({
+            .dstSet = VK_NULL_HANDLE,
+            .dstBinding = binding.unified++,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = vk::DescriptorType::eSampler,
+            .pImageInfo = &image_infos.back(),
+        });
+    }
+}
+
+void Rasterizer::BeginRendering(const GraphicsPipeline& pipeline, RenderState& state) {
+    int cb_index = 0;
+    for (auto& [image_id, desc] : cb_descs) {
+        if (auto& old_img = texture_cache.GetImage(image_id); old_img.binding.needs_rebind) {
+            auto& view = texture_cache.FindRenderTarget(desc);
+            ASSERT(view.image_id != image_id);
+            image_id = bound_images.emplace_back(view.image_id);
+            auto& image = texture_cache.GetImage(view.image_id);
+            state.color_attachments[cb_index].imageView = *view.image_view;
+            state.color_attachments[cb_index].imageLayout = image.last_state.layout;
+            state.color_images[cb_index] = image.image;
+
+            const auto mip = view.info.range.base.level;
+            state.width = std::min<u32>(state.width, std::max(image.info.size.width >> mip, 1u));
+            state.height = std::min<u32>(state.height, std::max(image.info.size.height >> mip, 1u));
+            ASSERT(old_img.info.size.width == state.width);
+            ASSERT(old_img.info.size.height == state.height);
+        }
+        auto& image = texture_cache.GetImage(image_id);
+        if (image.binding.force_general) {
+            image.Transit(
+                vk::ImageLayout::eGeneral,
+                vk::AccessFlagBits2::eColorAttachmentWrite | vk::AccessFlagBits2::eShaderRead, {});
+
+        } else {
+            image.Transit(vk::ImageLayout::eColorAttachmentOptimal,
+                          vk::AccessFlagBits2::eColorAttachmentWrite |
+                              vk::AccessFlagBits2::eColorAttachmentRead,
+                          desc.view_info.range);
+        }
+        image.usage.render_target = 1u;
+        state.color_attachments[cb_index].imageLayout = image.last_state.layout;
+        ++cb_index;
+    }
+
+    if (db_desc) {
+        const auto& image_id = std::get<0>(*db_desc);
+        const auto& desc = std::get<1>(*db_desc);
+        auto& image = texture_cache.GetImage(image_id);
+        ASSERT(image.binding.needs_rebind == 0);
+        const bool has_stencil = image.usage.stencil;
+        if (has_stencil) {
+            image.aspect_mask |= vk::ImageAspectFlagBits::eStencil;
+        }
+        if (image.binding.force_general) {
+            image.Transit(vk::ImageLayout::eGeneral,
+                          vk::AccessFlagBits2::eDepthStencilAttachmentWrite |
+                              vk::AccessFlagBits2::eShaderRead,
+                          {});
+        } else {
+            const auto new_layout = desc.view_info.is_storage
+                                        ? has_stencil
+                                              ? vk::ImageLayout::eDepthStencilAttachmentOptimal
+                                              : vk::ImageLayout::eDepthAttachmentOptimal
+                                    : has_stencil ? vk::ImageLayout::eDepthStencilReadOnlyOptimal
+                                                  : vk::ImageLayout::eDepthReadOnlyOptimal;
+            image.Transit(new_layout,
+                          vk::AccessFlagBits2::eDepthStencilAttachmentWrite |
+                              vk::AccessFlagBits2::eDepthStencilAttachmentRead,
+                          desc.view_info.range);
+        }
+        state.depth_attachment.imageLayout = image.last_state.layout;
+        image.usage.depth_target = true;
+        image.usage.stencil = has_stencil;
+    }
+
     scheduler.BeginRendering(state);
 }
 
@@ -331,10 +737,12 @@ void Rasterizer::Resolve() {
     // Read from MRT0, average all samples, and write to MRT1, which is one-sample
     const auto& mrt0_hint = liverpool->last_cb_extent[0];
     const auto& mrt1_hint = liverpool->last_cb_extent[1];
-    VideoCore::ImageInfo mrt0_info{liverpool->regs.color_buffers[0], mrt0_hint};
+    VideoCore::TextureCache::RenderTargetDesc mrt0_desc{liverpool->regs.color_buffers[0],
-    VideoCore::ImageInfo mrt1_info{liverpool->regs.color_buffers[1], mrt1_hint};
+                                                        mrt0_hint};
-    auto& mrt0_image = texture_cache.GetImage(texture_cache.FindImage(mrt0_info));
+    VideoCore::TextureCache::RenderTargetDesc mrt1_desc{liverpool->regs.color_buffers[1],
-    auto& mrt1_image = texture_cache.GetImage(texture_cache.FindImage(mrt1_info));
+                                                        mrt1_hint};
+    auto& mrt0_image = texture_cache.GetImage(texture_cache.FindImage(mrt0_desc));
+    auto& mrt1_image = texture_cache.GetImage(texture_cache.FindImage(mrt1_desc));
 
     VideoCore::SubresourceRange mrt0_range;
     mrt0_range.base.layer = liverpool->regs.color_buffers[0].view.slice_start;
@@ -384,9 +792,9 @@ u32 Rasterizer::ReadDataFromGds(u32 gds_offset) {
     return value;
 }
 
-void Rasterizer::InvalidateMemory(VAddr addr, u64 size) {
+void Rasterizer::InvalidateMemory(VAddr addr, VAddr addr_aligned, u64 size) {
-    buffer_cache.InvalidateMemory(addr, size);
+    buffer_cache.InvalidateMemory(addr_aligned, size);
-    texture_cache.InvalidateMemory(addr, size);
+    texture_cache.InvalidateMemory(addr, addr_aligned, size);
 }
 
 void Rasterizer::MapMemory(VAddr addr, u64 size) {

src/video_core/renderer_vulkan/vk_rasterizer.h

@@ -19,6 +19,7 @@ class MemoryManager;
 namespace Vulkan {
 
 class Scheduler;
+class RenderState;
 class GraphicsPipeline;
 
 class Rasterizer {
@@ -45,7 +46,7 @@ public:
 
     void InlineData(VAddr address, const void* value, u32 num_bytes, bool is_gds);
     u32 ReadDataFromGds(u32 gsd_offset);
-    void InvalidateMemory(VAddr addr, u64 size);
+    void InvalidateMemory(VAddr addr, VAddr addr_aligned, u64 size);
     void MapMemory(VAddr addr, u64 size);
     void UnmapMemory(VAddr addr, u64 size);
 
@@ -54,7 +55,8 @@ public:
     void Finish();
 
 private:
-    void BeginRendering(const GraphicsPipeline& pipeline);
+    RenderState PrepareRenderState(u32 mrt_mask);
+    void BeginRendering(const GraphicsPipeline& pipeline, RenderState& state);
     void Resolve();
 
     void UpdateDynamicState(const GraphicsPipeline& pipeline);
@@ -63,6 +65,21 @@ private:
 
     bool FilterDraw();
 
+    void BindBuffers(const Shader::Info& stage, Shader::Backend::Bindings& binding,
+                     Shader::PushData& push_data, Pipeline::DescriptorWrites& set_writes,
+                     Pipeline::BufferBarriers& buffer_barriers);
+
+    void BindTextures(const Shader::Info& stage, Shader::Backend::Bindings& binding,
+                      Pipeline::DescriptorWrites& set_writes);
+
+    bool BindResources(const Pipeline* pipeline);
+    void ResetBindings() {
+        for (auto& image_id : bound_images) {
+            texture_cache.GetImage(image_id).binding.Reset();
+        }
+        bound_images.clear();
+    }
+
 private:
     const Instance& instance;
     Scheduler& scheduler;
@@ -72,6 +89,25 @@ private:
     AmdGpu::Liverpool* liverpool;
     Core::MemoryManager* memory;
     PipelineCache pipeline_cache;
+
+    boost::container::static_vector<
+        std::pair<VideoCore::ImageId, VideoCore::TextureCache::RenderTargetDesc>, 8>
+        cb_descs;
+    std::optional<std::pair<VideoCore::ImageId, VideoCore::TextureCache::DepthTargetDesc>> db_desc;
+    boost::container::static_vector<vk::DescriptorImageInfo, 32> image_infos;
+    boost::container::static_vector<vk::BufferView, 8> buffer_views;
+    boost::container::static_vector<vk::DescriptorBufferInfo, 32> buffer_infos;
+    boost::container::static_vector<VideoCore::ImageId, 64> bound_images;
+
+    Pipeline::DescriptorWrites set_writes;
+    Pipeline::BufferBarriers buffer_barriers;
+
+    using BufferBindingInfo = std::pair<VideoCore::BufferId, AmdGpu::Buffer>;
+    boost::container::static_vector<BufferBindingInfo, 32> buffer_bindings;
+    using TexBufferBindingInfo = std::pair<VideoCore::BufferId, AmdGpu::Buffer>;
+    boost::container::static_vector<TexBufferBindingInfo, 32> texbuffer_bindings;
+    using ImageBindingInfo = std::pair<VideoCore::ImageId, VideoCore::TextureCache::TextureDesc>;
+    boost::container::static_vector<ImageBindingInfo, 32> image_bindings;
 };
 
 } // namespace Vulkan

src/video_core/texture_cache/image.cpp

@@ -61,6 +61,15 @@ bool ImageInfo::IsDepthStencil() const {
     }
 }
 
+bool ImageInfo::HasStencil() const {
+    if (pixel_format == vk::Format::eD32SfloatS8Uint ||
+        pixel_format == vk::Format::eD24UnormS8Uint ||
+        pixel_format == vk::Format::eD16UnormS8Uint) {
+        return true;
+    }
+    return false;
+}
+
 static vk::ImageUsageFlags ImageUsageFlags(const ImageInfo& info) {
     vk::ImageUsageFlags usage = vk::ImageUsageFlagBits::eTransferSrc |
                                 vk::ImageUsageFlagBits::eTransferDst |
@@ -135,22 +144,24 @@ void UniqueImage::Create(const vk::ImageCreateInfo& image_ci) {
 Image::Image(const Vulkan::Instance& instance_, Vulkan::Scheduler& scheduler_,
              const ImageInfo& info_)
     : instance{&instance_}, scheduler{&scheduler_}, info{info_},
-      image{instance->GetDevice(), instance->GetAllocator()}, cpu_addr{info.guest_address},
+      image{instance->GetDevice(), instance->GetAllocator()} {
-      cpu_addr_end{cpu_addr + info.guest_size_bytes} {
     mip_hashes.resize(info.resources.levels);
     ASSERT(info.pixel_format != vk::Format::eUndefined);
     // Here we force `eExtendedUsage` as don't know all image usage cases beforehand. In normal case
     // the texture cache should re-create the resource with the usage requested
     vk::ImageCreateFlags flags{vk::ImageCreateFlagBits::eMutableFormat |
                                vk::ImageCreateFlagBits::eExtendedUsage};
-    if (info.props.is_cube || (info.type == vk::ImageType::e2D && info.resources.layers >= 6)) {
+    const bool can_be_cube = (info.type == vk::ImageType::e2D) &&
+                             (info.resources.layers % 6 == 0) &&
+                             (info.size.width == info.size.height);
+    if (info.props.is_cube || can_be_cube) {
         flags |= vk::ImageCreateFlagBits::eCubeCompatible;
     } else if (info.props.is_volume) {
         flags |= vk::ImageCreateFlagBits::e2DArrayCompatible;
     }
 
-    usage = ImageUsageFlags(info);
+    usage_flags = ImageUsageFlags(info);
-    format_features = FormatFeatureFlags(usage);
+    format_features = FormatFeatureFlags(usage_flags);
 
     switch (info.pixel_format) {
     case vk::Format::eD16Unorm:
@@ -170,7 +181,7 @@ Image::Image(const Vulkan::Instance& instance_, Vulkan::Scheduler& scheduler_,
     constexpr auto tiling = vk::ImageTiling::eOptimal;
     const auto supported_format = instance->GetSupportedFormat(info.pixel_format, format_features);
     const auto properties = instance->GetPhysicalDevice().getImageFormatProperties(
-        supported_format, info.type, tiling, usage, flags);
+        supported_format, info.type, tiling, usage_flags, flags);
     const auto supported_samples = properties.result == vk::Result::eSuccess
                                        ? properties.value.sampleCounts
                                        : vk::SampleCountFlagBits::e1;
@@ -188,7 +199,7 @@ Image::Image(const Vulkan::Instance& instance_, Vulkan::Scheduler& scheduler_,
         .arrayLayers = static_cast<u32>(info.resources.layers),
         .samples = LiverpoolToVK::NumSamples(info.num_samples, supported_samples),
         .tiling = tiling,
-        .usage = usage,
+        .usage = usage_flags,
         .initialLayout = vk::ImageLayout::eUndefined,
     };
 

src/video_core/texture_cache/image.h

@@ -22,16 +22,15 @@ VK_DEFINE_HANDLE(VmaAllocator)
 namespace VideoCore {
 
 enum ImageFlagBits : u32 {
-    CpuDirty = 1 << 1, ///< Contents have been modified from the CPU
+    Empty = 0,
+    MaybeCpuDirty = 1 << 0, ///< The page this image is in was touched before the image address
+    CpuDirty = 1 << 1,      ///< Contents have been modified from the CPU
     GpuDirty = 1 << 2, ///< Contents have been modified from the GPU (valid data in buffer cache)
-    Dirty = CpuDirty | GpuDirty,
+    Dirty = MaybeCpuDirty | CpuDirty | GpuDirty,
     GpuModified = 1 << 3,    ///< Contents have been modified from the GPU
-    Tracked = 1 << 4,        ///< Writes and reads are being hooked from the CPU
     Registered = 1 << 6,     ///< True when the image is registered
     Picked = 1 << 7,         ///< Temporary flag to mark the image as picked
     MetaRegistered = 1 << 8, ///< True when metadata for this surface is known and registered
-    Bound = 1 << 9,          ///< True when the image is bound to a descriptor set
-    NeedsRebind = 1 << 10,   ///< True when the image needs to be rebound
 };
 DECLARE_ENUM_FLAG_OPERATORS(ImageFlagBits)
 
@@ -80,7 +79,9 @@ struct Image {
 
     [[nodiscard]] bool Overlaps(VAddr overlap_cpu_addr, size_t overlap_size) const noexcept {
         const VAddr overlap_end = overlap_cpu_addr + overlap_size;
-        return cpu_addr < overlap_end && overlap_cpu_addr < cpu_addr_end;
+        const auto image_addr = info.guest_address;
+        const auto image_end = info.guest_address + info.guest_size_bytes;
+        return image_addr < overlap_end && overlap_cpu_addr < image_end;
     }
 
     ImageViewId FindView(const ImageViewInfo& info) const {
@@ -101,19 +102,31 @@ struct Image {
     void CopyImage(const Image& image);
     void CopyMip(const Image& image, u32 mip);
 
+    bool IsTracked() {
+        return track_addr != 0 && track_addr_end != 0;
+    }
+
     const Vulkan::Instance* instance;
     Vulkan::Scheduler* scheduler;
     ImageInfo info;
     UniqueImage image;
     vk::ImageAspectFlags aspect_mask = vk::ImageAspectFlagBits::eColor;
     ImageFlagBits flags = ImageFlagBits::Dirty;
-    VAddr cpu_addr = 0;
+    VAddr track_addr = 0;
-    VAddr cpu_addr_end = 0;
+    VAddr track_addr_end = 0;
     std::vector<ImageViewInfo> image_view_infos;
     std::vector<ImageViewId> image_view_ids;
 
     // Resource state tracking
-    vk::ImageUsageFlags usage;
+    struct {
+        u32 texture : 1;
+        u32 storage : 1;
+        u32 render_target : 1;
+        u32 depth_target : 1;
+        u32 stencil : 1;
+        u32 vo_surface : 1;
+    } usage{};
+    vk::ImageUsageFlags usage_flags;
     vk::FormatFeatureFlags2 format_features;
     struct State {
         vk::Flags<vk::PipelineStageFlagBits2> pl_stage = vk::PipelineStageFlagBits2::eAllCommands;
@@ -124,6 +137,23 @@ struct Image {
     std::vector<State> subresource_states{};
     boost::container::small_vector<u64, 14> mip_hashes{};
     u64 tick_accessed_last{0};
+    u64 hash{0};
+
+    struct {
+        union {
+            struct {
+                u32 is_bound : 1;      // the image is bound to a descriptor set
+                u32 is_target : 1;     // the image is bound as color/depth target
+                u32 needs_rebind : 1;  // the image needs to be rebound
+                u32 force_general : 1; // the image needs to be used in general layout
+            };
+            u32 raw{};
+        };
+
+        void Reset() {
+            raw = 0u;
+        }
+    } binding{};
 };
 
 } // namespace VideoCore

src/video_core/texture_cache/image_info.cpp

@@ -245,7 +245,6 @@ ImageInfo::ImageInfo(const Libraries::VideoOut::BufferAttributeGroup& group,
     size.width = attrib.width;
     size.height = attrib.height;
     pitch = attrib.tiling_mode == TilingMode::Linear ? size.width : (size.width + 127) & (~127);
-    usage.vo_buffer = true;
     num_bits = attrib.pixel_format != VideoOutFormat::A16R16G16B16Float ? 32 : 64;
     ASSERT(num_bits == 32);
 
@@ -277,7 +276,6 @@ ImageInfo::ImageInfo(const AmdGpu::Liverpool::ColorBuffer& buffer,
     resources.layers = buffer.NumSlices();
     meta_info.cmask_addr = buffer.info.fast_clear ? buffer.CmaskAddress() : 0;
     meta_info.fmask_addr = buffer.info.compression ? buffer.FmaskAddress() : 0;
-    usage.render_target = true;
 
     guest_address = buffer.Address();
     const auto color_slice_sz = buffer.GetColorSliceSize();
@@ -299,9 +297,6 @@ ImageInfo::ImageInfo(const AmdGpu::Liverpool::DepthBuffer& buffer, u32 num_slice
     pitch = buffer.Pitch();
     resources.layers = num_slices;
     meta_info.htile_addr = buffer.z_info.tile_surface_en ? htile_address : 0;
-    usage.depth_target = true;
-    usage.stencil =
-        buffer.stencil_info.format != AmdGpu::Liverpool::DepthBuffer::StencilFormat::Invalid;
 
     guest_address = buffer.Address();
     const auto depth_slice_sz = buffer.GetDepthSliceSize();
@@ -330,7 +325,6 @@ ImageInfo::ImageInfo(const AmdGpu::Image& image, const Shader::ImageResource& de
     resources.layers = image.NumLayers(desc.is_array);
     num_samples = image.NumSamples();
     num_bits = NumBits(image.GetDataFmt());
-    usage.texture = true;
 
     guest_address = image.Address();
 
@@ -392,7 +386,6 @@ void ImageInfo::UpdateSize() {
         }
         }
         mip_info.size *= mip_d;
-
         mip_info.offset = guest_size_bytes;
         mips_layout.emplace_back(mip_info);
         guest_size_bytes += mip_info.size;
@@ -400,79 +393,87 @@ void ImageInfo::UpdateSize() {
     guest_size_bytes *= resources.layers;
 }
 
-bool ImageInfo::IsMipOf(const ImageInfo& info) const {
+int ImageInfo::IsMipOf(const ImageInfo& info) const {
     if (!IsCompatible(info)) {
-        return false;
+        return -1;
+    }
+
+    if (IsTilingCompatible(info.tiling_idx, tiling_idx)) {
+        return -1;
     }
 
     // Currently we expect only on level to be copied.
     if (resources.levels != 1) {
-        return false;
+        return -1;
     }
 
-    const int mip = info.resources.levels - resources.levels;
+    if (info.mips_layout.empty()) {
-    if (mip < 1) {
+        UNREACHABLE();
-        return false;
     }
 
+    // Find mip
+    auto mip = -1;
+    for (auto m = 0; m < info.mips_layout.size(); ++m) {
+        if (guest_address == (info.guest_address + info.mips_layout[m].offset)) {
+            mip = m;
+            break;
+        }
+    }
+
+    if (mip < 0) {
+        return -1;
+    }
+    ASSERT(mip != 0);
+
     const auto mip_w = std::max(info.size.width >> mip, 1u);
     const auto mip_h = std::max(info.size.height >> mip, 1u);
     if ((size.width != mip_w) || (size.height != mip_h)) {
-        return false;
+        return -1;
     }
 
     const auto mip_d = std::max(info.size.depth >> mip, 1u);
     if (info.type == vk::ImageType::e3D && type == vk::ImageType::e2D) {
         // In case of 2D array to 3D copy, make sure we have proper number of layers.
         if (resources.layers != mip_d) {
-            return false;
+            return -1;
         }
     } else {
         if (type != info.type) {
-            return false;
+            return -1;
         }
     }
 
-    // Check if the mip has correct size.
+    return mip;
-    if (info.mips_layout.size() <= mip || info.mips_layout[mip].size != guest_size_bytes) {
-        return false;
-    }
-
-    // Ensure that address matches too.
-    if ((info.guest_address + info.mips_layout[mip].offset) != guest_address) {
-        return false;
-    }
-
-    return true;
 }
 
-bool ImageInfo::IsSliceOf(const ImageInfo& info) const {
+int ImageInfo::IsSliceOf(const ImageInfo& info) const {
     if (!IsCompatible(info)) {
-        return false;
+        return -1;
     }
 
     // Array slices should be of the same type.
     if (type != info.type) {
-        return false;
+        return -1;
     }
 
     // 2D dimensions of both images should be the same.
     if ((size.width != info.size.width) || (size.height != info.size.height)) {
-        return false;
+        return -1;
     }
 
     // Check for size alignment.
     const bool slice_size = info.guest_size_bytes / info.resources.layers;
     if (guest_size_bytes % slice_size != 0) {
-        return false;
+        return -1;
     }
 
     // Ensure that address is aligned too.
-    if (((info.guest_address - guest_address) % guest_size_bytes) != 0) {
+    const auto addr_diff = guest_address - info.guest_address;
-        return false;
+    if ((addr_diff % guest_size_bytes) != 0) {
+        return -1;
     }
 
-    return true;
+    return addr_diff / guest_size_bytes;
 }
 
 } // namespace VideoCore

src/video_core/texture_cache/image_info.h

@@ -28,14 +28,28 @@ struct ImageInfo {
     bool IsBlockCoded() const;
     bool IsPacked() const;
     bool IsDepthStencil() const;
+    bool HasStencil() const;
 
-    bool IsMipOf(const ImageInfo& info) const;
+    int IsMipOf(const ImageInfo& info) const;
-    bool IsSliceOf(const ImageInfo& info) const;
+    int IsSliceOf(const ImageInfo& info) const;
 
     /// Verifies if images are compatible for subresource merging.
     bool IsCompatible(const ImageInfo& info) const {
-        return (pixel_format == info.pixel_format && tiling_idx == info.tiling_idx &&
+        return (pixel_format == info.pixel_format && num_samples == info.num_samples &&
-                num_samples == info.num_samples && num_bits == info.num_bits);
+                num_bits == info.num_bits);
+    }
+
+    bool IsTilingCompatible(u32 lhs, u32 rhs) const {
+        if (lhs == rhs) {
+            return true;
+        }
+        if (lhs == 0x0e && rhs == 0x0d) {
+            return true;
+        }
+        if (lhs == 0x0d && rhs == 0x0e) {
+            return true;
+        }
+        return false;
     }
 
     void UpdateSize();
@@ -46,15 +60,6 @@ struct ImageInfo {
         VAddr htile_addr;
     } meta_info{};
 
-    struct {
-        u32 texture : 1;
-        u32 storage : 1;
-        u32 render_target : 1;
-        u32 depth_target : 1;
-        u32 stencil : 1;
-        u32 vo_buffer : 1;
-    } usage{}; // Usage data tracked during image lifetime
-
     struct {
         u32 is_cube : 1;
         u32 is_volume : 1;
@@ -81,6 +86,9 @@ struct ImageInfo {
     VAddr guest_address{0};
     u32 guest_size_bytes{0};
     u32 tiling_idx{0}; // TODO: merge with existing!
+
+    VAddr stencil_addr{0};
+    u32 stencil_size{0};
 };
 
 } // namespace VideoCore

src/video_core/texture_cache/image_view.cpp

@@ -149,7 +149,7 @@ ImageViewInfo::ImageViewInfo(const AmdGpu::Liverpool::DepthBuffer& depth_buffer,
 ImageView::ImageView(const Vulkan::Instance& instance, const ImageViewInfo& info_, Image& image,
                      ImageId image_id_)
     : image_id{image_id_}, info{info_} {
-    vk::ImageViewUsageCreateInfo usage_ci{.usage = image.usage};
+    vk::ImageViewUsageCreateInfo usage_ci{.usage = image.usage_flags};
     if (!info.is_storage) {
         usage_ci.usage &= ~vk::ImageUsageFlagBits::eStorage;
     }

src/video_core/texture_cache/texture_cache.cpp

@@ -29,9 +29,12 @@ TextureCache::TextureCache(const Vulkan::Instance& instance_, Vulkan::Scheduler&
     info.UpdateSize();
     const ImageId null_id = slot_images.insert(instance, scheduler, info);
     ASSERT(null_id.index == NULL_IMAGE_ID.index);
-    const vk::Image& null_image = slot_images[null_id].image;
+    auto& img = slot_images[null_id];
+    const vk::Image& null_image = img.image;
     Vulkan::SetObjectName(instance.GetDevice(), null_image, "Null Image");
-    slot_images[null_id].flags = ImageFlagBits::Tracked;
+    img.flags = ImageFlagBits::Empty;
+    img.track_addr = img.info.guest_address;
+    img.track_addr_end = img.info.guest_address + img.info.guest_size_bytes;
 
     ImageViewInfo view_info;
     const auto null_view_id =
@@ -43,13 +46,43 @@ TextureCache::TextureCache(const Vulkan::Instance& instance_, Vulkan::Scheduler&
 
 TextureCache::~TextureCache() = default;
 
-void TextureCache::InvalidateMemory(VAddr address, size_t size) {
+void TextureCache::MarkAsMaybeDirty(ImageId image_id, Image& image) {
+    if (image.hash == 0) {
+        // Initialize hash
+        const u8* addr = std::bit_cast<u8*>(image.info.guest_address);
+        image.hash = XXH3_64bits(addr, image.info.guest_size_bytes);
+    }
+    image.flags |= ImageFlagBits::MaybeCpuDirty;
+    UntrackImage(image_id);
+}
+
+void TextureCache::InvalidateMemory(VAddr addr, VAddr page_addr, size_t size) {
     std::scoped_lock lock{mutex};
-    ForEachImageInRegion(address, size, [&](ImageId image_id, Image& image) {
+    ForEachImageInRegion(page_addr, size, [&](ImageId image_id, Image& image) {
-        // Ensure image is reuploaded when accessed again.
+        const auto image_begin = image.info.guest_address;
-        image.flags |= ImageFlagBits::CpuDirty;
+        const auto image_end = image.info.guest_address + image.info.guest_size_bytes;
-        // Untrack image, so the range is unprotected and the guest can write freely.
+        const auto page_end = page_addr + size;
-        UntrackImage(image_id);
+        if (image_begin <= addr && addr < image_end) {
+            // This image was definitely accessed by this page fault.
+            // Untrack image, so the range is unprotected and the guest can write freely
+            image.flags |= ImageFlagBits::CpuDirty;
+            UntrackImage(image_id);
+        } else if (page_end < image_end) {
+            // This page access may or may not modify the image.
+            // We should not mark it as dirty now. If it really was modified
+            // it will receive more invalidations on its other pages.
+            // Remove tracking from this page only.
+            UntrackImageHead(image_id);
+        } else if (image_begin < page_addr) {
+            // This page access does not modify the image but the page should be untracked.
+            // We should not mark this image as dirty now. If it really was modified
+            // it will receive more invalidations on its other pages.
+            UntrackImageTail(image_id);
+        } else {
+            // Image begins and ends on this page so it can not receive any more invalidations.
+            // We will check it's hash later to see if it really was modified.
+            MarkAsMaybeDirty(image_id, image);
+        }
     });
 }
 
@@ -77,84 +110,149 @@ void TextureCache::UnmapMemory(VAddr cpu_addr, size_t size) {
     }
 }
 
-ImageId TextureCache::ResolveDepthOverlap(const ImageInfo& requested_info, ImageId cache_image_id) {
+ImageId TextureCache::ResolveDepthOverlap(const ImageInfo& requested_info, BindingType binding,
-    const auto& cache_info = slot_images[cache_image_id].info;
+                                          ImageId cache_image_id) {
+    const auto& cache_image = slot_images[cache_image_id];
 
-    const bool was_bound_as_texture =
+    if (!cache_image.info.IsDepthStencil() && !requested_info.IsDepthStencil()) {
-        !cache_info.usage.depth_target && (cache_info.usage.texture || cache_info.usage.storage);
+        return {};
-    if (requested_info.usage.depth_target && was_bound_as_texture) {
+    }
-        auto new_image_id = slot_images.insert(instance, scheduler, requested_info);
+
+    const bool stencil_match = requested_info.HasStencil() == cache_image.info.HasStencil();
+    const bool bpp_match = requested_info.num_bits == cache_image.info.num_bits;
+
+    // If an image in the cache has less slices we need to expand it
+    bool recreate = cache_image.info.resources < requested_info.resources;
+
+    switch (binding) {
+    case BindingType::Texture:
+        // The guest requires a depth sampled texture, but cache can offer only Rxf. Need to
+        // recreate the image.
+        recreate |= requested_info.IsDepthStencil() && !cache_image.info.IsDepthStencil();
+        break;
+    case BindingType::Storage:
+        // If the guest is going to use previously created depth as storage, the image needs to be
+        // recreated. (TODO: Probably a case with linear rgba8 aliasing is legit)
+        recreate |= cache_image.info.IsDepthStencil();
+        break;
+    case BindingType::RenderTarget:
+        // Render target can have only Rxf format. If the cache contains only Dx[S8] we need to
+        // re-create the image.
+        ASSERT(!requested_info.IsDepthStencil());
+        recreate |= cache_image.info.IsDepthStencil();
+        break;
+    case BindingType::DepthTarget:
+        // The guest has requested previously allocated texture to be bound as a depth target.
+        // In this case we need to convert Rx float to a Dx[S8] as requested
+        recreate |= !cache_image.info.IsDepthStencil();
+
+        // The guest is trying to bind a depth target and cache has it. Need to be sure that aspects
+        // and bpp match
+        recreate |= cache_image.info.IsDepthStencil() && !(stencil_match && bpp_match);
+        break;
+    default:
+        break;
+    }
+
+    if (recreate) {
+        auto new_info{requested_info};
+        new_info.resources = std::max(requested_info.resources, cache_image.info.resources);
+        new_info.UpdateSize();
+        const auto new_image_id = slot_images.insert(instance, scheduler, new_info);
         RegisterImage(new_image_id);
 
+        // Inherit image usage
+        auto& new_image = GetImage(new_image_id);
+        new_image.usage = cache_image.usage;
+
         // TODO: perform a depth copy here
 
         FreeImage(cache_image_id);
         return new_image_id;
     }
 
-    const bool should_bind_as_texture =
+    // Will be handled by view
-        !requested_info.usage.depth_target &&
+    return cache_image_id;
-        (requested_info.usage.texture || requested_info.usage.storage);
-    if (cache_info.usage.depth_target && should_bind_as_texture) {
-        if (cache_info.resources == requested_info.resources) {
-            return cache_image_id;
-        } else {
-            // UNREACHABLE();
-        }
-    }
-
-    return {};
 }
 
-ImageId TextureCache::ResolveOverlap(const ImageInfo& image_info, ImageId cache_image_id,
+std::tuple<ImageId, int, int> TextureCache::ResolveOverlap(const ImageInfo& image_info,
-                                     ImageId merged_image_id) {
+                                                           BindingType binding,
+                                                           ImageId cache_image_id,
+                                                           ImageId merged_image_id) {
     auto& tex_cache_image = slot_images[cache_image_id];
+    // We can assume it is safe to delete the image if it wasn't accessed in some number of frames.
+    const bool safe_to_delete =
+        scheduler.CurrentTick() - tex_cache_image.tick_accessed_last > NumFramesBeforeRemoval;
 
     if (image_info.guest_address == tex_cache_image.info.guest_address) { // Equal address
         if (image_info.size != tex_cache_image.info.size) {
-            // Very likely this kind of overlap is caused by allocation from a pool. We can assume
+            // Very likely this kind of overlap is caused by allocation from a pool.
-            // it is safe to delete the image if it wasn't accessed in some amount of frames.
+            if (safe_to_delete) {
-            if (scheduler.CurrentTick() - tex_cache_image.tick_accessed_last >
-                NumFramesBeforeRemoval) {
-
                 FreeImage(cache_image_id);
             }
-            return merged_image_id;
+            return {merged_image_id, -1, -1};
         }
 
-        if (auto depth_image_id = ResolveDepthOverlap(image_info, cache_image_id)) {
+        if (const auto depth_image_id = ResolveDepthOverlap(image_info, binding, cache_image_id)) {
-            return depth_image_id;
+            return {depth_image_id, -1, -1};
         }
 
         if (image_info.pixel_format != tex_cache_image.info.pixel_format ||
             image_info.guest_size_bytes <= tex_cache_image.info.guest_size_bytes) {
             auto result_id = merged_image_id ? merged_image_id : cache_image_id;
             const auto& result_image = slot_images[result_id];
-            return IsVulkanFormatCompatible(image_info.pixel_format, result_image.info.pixel_format)
+            return {
-                       ? result_id
+                IsVulkanFormatCompatible(image_info.pixel_format, result_image.info.pixel_format)
-                       : ImageId{};
+                    ? result_id
+                    : ImageId{},
+                -1, -1};
         }
 
         ImageId new_image_id{};
         if (image_info.type == tex_cache_image.info.type) {
+            ASSERT(image_info.resources > tex_cache_image.info.resources);
             new_image_id = ExpandImage(image_info, cache_image_id);
         } else {
             UNREACHABLE();
         }
-        return new_image_id;
+        return {new_image_id, -1, -1};
     }
 
     // Right overlap, the image requested is a possible subresource of the image from cache.
     if (image_info.guest_address > tex_cache_image.info.guest_address) {
-        // Should be handled by view. No additional actions needed.
+        if (auto mip = image_info.IsMipOf(tex_cache_image.info); mip >= 0) {
+            return {cache_image_id, mip, -1};
+        }
+
+        if (auto slice = image_info.IsSliceOf(tex_cache_image.info); slice >= 0) {
+            return {cache_image_id, -1, slice};
+        }
+
+        // TODO: slice and mip
+
+        if (safe_to_delete) {
+            FreeImage(cache_image_id);
+        }
+
+        return {{}, -1, -1};
     } else {
         // Left overlap, the image from cache is a possible subresource of the image requested
         if (!merged_image_id) {
             // We need to have a larger, already allocated image to copy this one into
-            return {};
+            return {{}, -1, -1};
         }
 
-        if (tex_cache_image.info.IsMipOf(image_info)) {
+        if (auto mip = tex_cache_image.info.IsMipOf(image_info); mip >= 0) {
+            if (tex_cache_image.binding.is_target) {
+                // We have a larger image created and a separate one, representing a subres of it,
+                // bound as render target. In this case we need to rebind render target.
+                tex_cache_image.binding.needs_rebind = 1u;
+                GetImage(merged_image_id).binding.is_target = 1u;
+
+                FreeImage(cache_image_id);
+                return {merged_image_id, -1, -1};
+            }
+
             tex_cache_image.Transit(vk::ImageLayout::eTransferSrcOptimal,
                                     vk::AccessFlagBits2::eTransferRead, {});
 
@@ -162,13 +260,13 @@ ImageId TextureCache::ResolveOverlap(const ImageInfo& image_info, ImageId cache_
             ASSERT(num_mips_to_copy == 1);
 
             auto& merged_image = slot_images[merged_image_id];
-            merged_image.CopyMip(tex_cache_image, image_info.resources.levels - 1);
+            merged_image.CopyMip(tex_cache_image, mip);
 
             FreeImage(cache_image_id);
         }
     }
 
-    return merged_image_id;
+    return {merged_image_id, -1, -1};
 }
 
 ImageId TextureCache::ExpandImage(const ImageInfo& info, ImageId image_id) {
@@ -181,8 +279,8 @@ ImageId TextureCache::ExpandImage(const ImageInfo& info, ImageId image_id) {
     src_image.Transit(vk::ImageLayout::eTransferSrcOptimal, vk::AccessFlagBits2::eTransferRead, {});
     new_image.CopyImage(src_image);
 
-    if (True(src_image.flags & ImageFlagBits::Bound)) {
+    if (src_image.binding.is_bound || src_image.binding.is_target) {
-        src_image.flags |= ImageFlagBits::NeedsRebind;
+        src_image.binding.needs_rebind = 1u;
     }
 
     FreeImage(image_id);
@@ -192,9 +290,11 @@ ImageId TextureCache::ExpandImage(const ImageInfo& info, ImageId image_id) {
     return new_image_id;
 }
 
-ImageId TextureCache::FindImage(const ImageInfo& info, FindFlags flags) {
+ImageId TextureCache::FindImage(BaseDesc& desc, FindFlags flags) {
+    const auto& info = desc.info;
+
     if (info.guest_address == 0) [[unlikely]] {
-        return NULL_IMAGE_VIEW_ID;
+        return NULL_IMAGE_ID;
     }
 
     std::scoped_lock lock{mutex};
@@ -231,10 +331,16 @@ ImageId TextureCache::FindImage(const ImageInfo& info, FindFlags flags) {
     }
 
     // Try to resolve overlaps (if any)
+    int view_mip{-1};
+    int view_slice{-1};
     if (!image_id) {
         for (const auto& cache_id : image_ids) {
+            view_mip = -1;
+            view_slice = -1;
+
             const auto& merged_info = image_id ? slot_images[image_id].info : info;
-            image_id = ResolveOverlap(merged_info, cache_id, image_id);
+            std::tie(image_id, view_mip, view_slice) =
+                ResolveOverlap(merged_info, desc.type, cache_id, image_id);
         }
     }
 
@@ -254,6 +360,10 @@ ImageId TextureCache::FindImage(const ImageInfo& info, FindFlags flags) {
         RegisterImage(image_id);
     }
 
+    if (view_mip > 0) {
+        desc.view_info.range.base.level = view_mip;
+    }
+
     Image& image = slot_images[image_id];
     image.tick_accessed_last = scheduler.CurrentTick();
 
@@ -275,100 +385,58 @@ ImageView& TextureCache::RegisterImageView(ImageId image_id, const ImageViewInfo
 ImageView& TextureCache::FindTexture(ImageId image_id, const ImageViewInfo& view_info) {
     Image& image = slot_images[image_id];
     UpdateImage(image_id);
-    auto& usage = image.info.usage;
-
-    if (view_info.is_storage) {
-        image.Transit(vk::ImageLayout::eGeneral,
-                      vk::AccessFlagBits2::eShaderRead | vk::AccessFlagBits2::eShaderWrite,
-                      view_info.range);
-        usage.storage = true;
-    } else {
-        const auto new_layout = image.info.IsDepthStencil()
-                                    ? vk::ImageLayout::eDepthStencilReadOnlyOptimal
-                                    : vk::ImageLayout::eShaderReadOnlyOptimal;
-        image.Transit(new_layout, vk::AccessFlagBits2::eShaderRead, view_info.range);
-        usage.texture = true;
-    }
-
     return RegisterImageView(image_id, view_info);
 }
 
-ImageView& TextureCache::FindRenderTarget(const ImageInfo& image_info,
+ImageView& TextureCache::FindRenderTarget(BaseDesc& desc) {
-                                          const ImageViewInfo& view_info) {
+    const ImageId image_id = FindImage(desc);
-    const ImageId image_id = FindImage(image_info);
     Image& image = slot_images[image_id];
     image.flags |= ImageFlagBits::GpuModified;
+    image.usage.render_target = 1u;
     UpdateImage(image_id);
 
-    image.Transit(vk::ImageLayout::eColorAttachmentOptimal,
-                  vk::AccessFlagBits2::eColorAttachmentWrite |
-                      vk::AccessFlagBits2::eColorAttachmentRead,
-                  view_info.range);
-
     // Register meta data for this color buffer
     if (!(image.flags & ImageFlagBits::MetaRegistered)) {
-        if (image_info.meta_info.cmask_addr) {
+        if (desc.info.meta_info.cmask_addr) {
             surface_metas.emplace(
-                image_info.meta_info.cmask_addr,
+                desc.info.meta_info.cmask_addr,
                 MetaDataInfo{.type = MetaDataInfo::Type::CMask, .is_cleared = true});
-            image.info.meta_info.cmask_addr = image_info.meta_info.cmask_addr;
+            image.info.meta_info.cmask_addr = desc.info.meta_info.cmask_addr;
             image.flags |= ImageFlagBits::MetaRegistered;
         }
 
-        if (image_info.meta_info.fmask_addr) {
+        if (desc.info.meta_info.fmask_addr) {
             surface_metas.emplace(
-                image_info.meta_info.fmask_addr,
+                desc.info.meta_info.fmask_addr,
                 MetaDataInfo{.type = MetaDataInfo::Type::FMask, .is_cleared = true});
-            image.info.meta_info.fmask_addr = image_info.meta_info.fmask_addr;
+            image.info.meta_info.fmask_addr = desc.info.meta_info.fmask_addr;
             image.flags |= ImageFlagBits::MetaRegistered;
         }
     }
 
-    // Update tracked image usage
+    return RegisterImageView(image_id, desc.view_info);
-    image.info.usage.render_target = true;
-
-    return RegisterImageView(image_id, view_info);
 }
 
-ImageView& TextureCache::FindDepthTarget(const ImageInfo& image_info,
+ImageView& TextureCache::FindDepthTarget(BaseDesc& desc) {
-                                         const ImageViewInfo& view_info) {
+    const ImageId image_id = FindImage(desc);
-    const ImageId image_id = FindImage(image_info);
     Image& image = slot_images[image_id];
     image.flags |= ImageFlagBits::GpuModified;
     image.flags &= ~ImageFlagBits::Dirty;
-    image.aspect_mask = vk::ImageAspectFlagBits::eDepth;
+    image.usage.depth_target = 1u;
-
+    image.usage.stencil = image.info.HasStencil();
-    const bool has_stencil = image_info.usage.stencil;
-    if (has_stencil) {
-        image.aspect_mask |= vk::ImageAspectFlagBits::eStencil;
-    }
-
-    const auto new_layout = view_info.is_storage
-                                ? has_stencil ? vk::ImageLayout::eDepthStencilAttachmentOptimal
-                                              : vk::ImageLayout::eDepthAttachmentOptimal
-                            : has_stencil ? vk::ImageLayout::eDepthStencilReadOnlyOptimal
-                                          : vk::ImageLayout::eDepthReadOnlyOptimal;
-    image.Transit(new_layout,
-                  vk::AccessFlagBits2::eDepthStencilAttachmentWrite |
-                      vk::AccessFlagBits2::eDepthStencilAttachmentRead,
-                  view_info.range);
 
     // Register meta data for this depth buffer
     if (!(image.flags & ImageFlagBits::MetaRegistered)) {
-        if (image_info.meta_info.htile_addr) {
+        if (desc.info.meta_info.htile_addr) {
             surface_metas.emplace(
-                image_info.meta_info.htile_addr,
+                desc.info.meta_info.htile_addr,
                 MetaDataInfo{.type = MetaDataInfo::Type::HTile, .is_cleared = true});
-            image.info.meta_info.htile_addr = image_info.meta_info.htile_addr;
+            image.info.meta_info.htile_addr = desc.info.meta_info.htile_addr;
             image.flags |= ImageFlagBits::MetaRegistered;
         }
     }
 
-    // Update tracked image usage
+    return RegisterImageView(image_id, desc.view_info);
-    image.info.usage.depth_target = true;
-    image.info.usage.stencil = has_stencil;
-
-    return RegisterImageView(image_id, view_info);
 }
 
 void TextureCache::RefreshImage(Image& image, Vulkan::Scheduler* custom_scheduler /*= nullptr*/) {
@@ -380,6 +448,23 @@ void TextureCache::RefreshImage(Image& image, Vulkan::Scheduler* custom_schedule
         return;
     }
 
+    if (True(image.flags & ImageFlagBits::MaybeCpuDirty) &&
+        False(image.flags & ImageFlagBits::CpuDirty)) {
+        // The image size should be less than page size to be considered MaybeCpuDirty
+        // So this calculation should be very uncommon and reasonably fast
+        // For now we'll just check up to 64 first pixels
+        const auto addr = std::bit_cast<u8*>(image.info.guest_address);
+        const auto w = std::min(image.info.size.width, u32(8));
+        const auto h = std::min(image.info.size.height, u32(8));
+        const auto size = w * h * image.info.num_bits / 8;
+        const u64 hash = XXH3_64bits(addr, size);
+        if (image.hash == hash) {
+            image.flags &= ~ImageFlagBits::MaybeCpuDirty;
+            return;
+        }
+        image.hash = hash;
+    }
+
     const auto& num_layers = image.info.resources.layers;
     const auto& num_mips = image.info.resources.levels;
     ASSERT(num_mips == image.info.mips_layout.size());
@@ -390,14 +475,14 @@ void TextureCache::RefreshImage(Image& image, Vulkan::Scheduler* custom_schedule
         const u32 height = std::max(image.info.size.height >> m, 1u);
         const u32 depth =
             image.info.props.is_volume ? std::max(image.info.size.depth >> m, 1u) : 1u;
-        const auto& [mip_size, mip_pitch, mip_height, mip_ofs] = image.info.mips_layout[m];
+        const auto& mip = image.info.mips_layout[m];
 
         // Protect GPU modified resources from accidental CPU reuploads.
         const bool is_gpu_modified = True(image.flags & ImageFlagBits::GpuModified);
         const bool is_gpu_dirty = True(image.flags & ImageFlagBits::GpuDirty);
         if (is_gpu_modified && !is_gpu_dirty) {
             const u8* addr = std::bit_cast<u8*>(image.info.guest_address);
-            const u64 hash = XXH3_64bits(addr + mip_ofs, mip_size);
+            const u64 hash = XXH3_64bits(addr + mip.offset, mip.size);
             if (image.mip_hashes[m] == hash) {
                 continue;
             }
@@ -405,9 +490,9 @@ void TextureCache::RefreshImage(Image& image, Vulkan::Scheduler* custom_schedule
         }
 
         image_copy.push_back({
-            .bufferOffset = mip_ofs * num_layers,
+            .bufferOffset = mip.offset * num_layers,
-            .bufferRowLength = static_cast<u32>(mip_pitch),
+            .bufferRowLength = static_cast<u32>(mip.pitch),
-            .bufferImageHeight = static_cast<u32>(mip_height),
+            .bufferImageHeight = static_cast<u32>(mip.height),
             .imageSubresource{
                 .aspectMask = image.aspect_mask & ~vk::ImageAspectFlagBits::eStencil,
                 .mipLevel = m,
@@ -420,6 +505,7 @@ void TextureCache::RefreshImage(Image& image, Vulkan::Scheduler* custom_schedule
     }
 
     if (image_copy.empty()) {
+        image.flags &= ~ImageFlagBits::Dirty;
         return;
     }
 
@@ -465,16 +551,16 @@ void TextureCache::RegisterImage(ImageId image_id) {
     ASSERT_MSG(False(image.flags & ImageFlagBits::Registered),
                "Trying to register an already registered image");
     image.flags |= ImageFlagBits::Registered;
-    ForEachPage(image.cpu_addr, image.info.guest_size_bytes,
+    ForEachPage(image.info.guest_address, image.info.guest_size_bytes,
                 [this, image_id](u64 page) { page_table[page].push_back(image_id); });
 }
 
 void TextureCache::UnregisterImage(ImageId image_id) {
     Image& image = slot_images[image_id];
     ASSERT_MSG(True(image.flags & ImageFlagBits::Registered),
-               "Trying to unregister an already registered image");
+               "Trying to unregister an already unregistered image");
     image.flags &= ~ImageFlagBits::Registered;
-    ForEachPage(image.cpu_addr, image.info.guest_size_bytes, [this, image_id](u64 page) {
+    ForEachPage(image.info.guest_address, image.info.guest_size_bytes, [this, image_id](u64 page) {
         const auto page_it = page_table.find(page);
         if (page_it == nullptr) {
             UNREACHABLE_MSG("Unregistering unregistered page=0x{:x}", page << PageShift);
@@ -492,25 +578,106 @@ void TextureCache::UnregisterImage(ImageId image_id) {
 
 void TextureCache::TrackImage(ImageId image_id) {
     auto& image = slot_images[image_id];
-    if (True(image.flags & ImageFlagBits::Tracked)) {
+    const auto image_begin = image.info.guest_address;
+    const auto image_end = image.info.guest_address + image.info.guest_size_bytes;
+    if (image_begin == image.track_addr && image_end == image.track_addr_end) {
         return;
     }
-    image.flags |= ImageFlagBits::Tracked;
+
-    tracker.UpdatePagesCachedCount(image.cpu_addr, image.info.guest_size_bytes, 1);
+    if (!image.IsTracked()) {
+        // Re-track the whole image
+        image.track_addr = image_begin;
+        image.track_addr_end = image_end;
+        tracker.UpdatePagesCachedCount(image_begin, image.info.guest_size_bytes, 1);
+    } else {
+        if (image_begin < image.track_addr) {
+            TrackImageHead(image_id);
+        }
+        if (image.track_addr_end < image_end) {
+            TrackImageTail(image_id);
+        }
+    }
+}
+
+void TextureCache::TrackImageHead(ImageId image_id) {
+    auto& image = slot_images[image_id];
+    const auto image_begin = image.info.guest_address;
+    if (image_begin == image.track_addr) {
+        return;
+    }
+    ASSERT(image.track_addr != 0 && image_begin < image.track_addr);
+    const auto size = image.track_addr - image_begin;
+    image.track_addr = image_begin;
+    tracker.UpdatePagesCachedCount(image_begin, size, 1);
+}
+
+void TextureCache::TrackImageTail(ImageId image_id) {
+    auto& image = slot_images[image_id];
+    const auto image_end = image.info.guest_address + image.info.guest_size_bytes;
+    if (image_end == image.track_addr_end) {
+        return;
+    }
+    ASSERT(image.track_addr_end != 0 && image.track_addr_end < image_end);
+    const auto addr = image.track_addr_end;
+    const auto size = image_end - image.track_addr_end;
+    image.track_addr_end = image_end;
+    tracker.UpdatePagesCachedCount(addr, size, 1);
 }
 
 void TextureCache::UntrackImage(ImageId image_id) {
     auto& image = slot_images[image_id];
-    if (False(image.flags & ImageFlagBits::Tracked)) {
+    if (!image.IsTracked()) {
         return;
     }
-    image.flags &= ~ImageFlagBits::Tracked;
+    const auto addr = image.track_addr;
-    tracker.UpdatePagesCachedCount(image.cpu_addr, image.info.guest_size_bytes, -1);
+    const auto size = image.track_addr_end - image.track_addr;
+    image.track_addr = 0;
+    image.track_addr_end = 0;
+    if (size != 0) {
+        tracker.UpdatePagesCachedCount(addr, size, -1);
+    }
+}
+
+void TextureCache::UntrackImageHead(ImageId image_id) {
+    auto& image = slot_images[image_id];
+    const auto image_begin = image.info.guest_address;
+    if (!image.IsTracked() || image_begin < image.track_addr) {
+        return;
+    }
+    const auto addr = tracker.GetNextPageAddr(image_begin);
+    const auto size = addr - image_begin;
+    image.track_addr = addr;
+    if (image.track_addr == image.track_addr_end) {
+        // This image spans only 2 pages and both are modified,
+        // but the image itself was not directly affected.
+        // Cehck its hash later.
+        MarkAsMaybeDirty(image_id, image);
+    }
+    tracker.UpdatePagesCachedCount(image_begin, size, -1);
+}
+
+void TextureCache::UntrackImageTail(ImageId image_id) {
+    auto& image = slot_images[image_id];
+    const auto image_end = image.info.guest_address + image.info.guest_size_bytes;
+    if (!image.IsTracked() || image.track_addr_end < image_end) {
+        return;
+    }
+    ASSERT(image.track_addr_end != 0);
+    const auto addr = tracker.GetPageAddr(image_end);
+    const auto size = image_end - addr;
+    image.track_addr_end = addr;
+    if (image.track_addr == image.track_addr_end) {
+        // This image spans only 2 pages and both are modified,
+        // but the image itself was not directly affected.
+        // Cehck its hash later.
+        MarkAsMaybeDirty(image_id, image);
+    }
+    tracker.UpdatePagesCachedCount(addr, size, -1);
 }
 
 void TextureCache::DeleteImage(ImageId image_id) {
     Image& image = slot_images[image_id];
-    ASSERT_MSG(False(image.flags & ImageFlagBits::Tracked), "Image was not untracked");
+    ASSERT_MSG(!image.IsTracked(), "Image was not untracked");
     ASSERT_MSG(False(image.flags & ImageFlagBits::Registered), "Image was not unregistered");
 
     // Remove any registered meta areas.

src/video_core/texture_cache/texture_cache.h

@@ -43,12 +43,59 @@ class TextureCache {
     using PageTable = MultiLevelPageTable<Traits>;
 
 public:
-    explicit TextureCache(const Vulkan::Instance& instance, Vulkan::Scheduler& scheduler,
+    enum class BindingType : u32 {
-                          BufferCache& buffer_cache, PageManager& tracker);
+        Texture,
+        Storage,
+        RenderTarget,
+        DepthTarget,
+        VideoOut,
+    };
+
+    struct BaseDesc {
+        ImageInfo info;
+        ImageViewInfo view_info;
+        BindingType type{BindingType::Texture};
+
+        BaseDesc() = default;
+        BaseDesc(BindingType type_, ImageInfo info_, ImageViewInfo view_info_) noexcept
+            : info{std::move(info_)}, view_info{std::move(view_info_)}, type{type_} {}
+    };
+
+    struct TextureDesc : public BaseDesc {
+        TextureDesc() = default;
+        TextureDesc(const AmdGpu::Image& image, const Shader::ImageResource& desc)
+            : BaseDesc{desc.is_storage ? BindingType::Storage : BindingType::Texture,
+                       ImageInfo{image, desc}, ImageViewInfo{image, desc}} {}
+    };
+
+    struct RenderTargetDesc : public BaseDesc {
+        RenderTargetDesc(const AmdGpu::Liverpool::ColorBuffer& buffer,
+                         const AmdGpu::Liverpool::CbDbExtent& hint = {})
+            : BaseDesc{BindingType::RenderTarget, ImageInfo{buffer, hint}, ImageViewInfo{buffer}} {}
+    };
+
+    struct DepthTargetDesc : public BaseDesc {
+        DepthTargetDesc(const AmdGpu::Liverpool::DepthBuffer& buffer,
+                        const AmdGpu::Liverpool::DepthView& view,
+                        const AmdGpu::Liverpool::DepthControl& ctl, VAddr htile_address,
+                        const AmdGpu::Liverpool::CbDbExtent& hint = {})
+            : BaseDesc{BindingType::DepthTarget,
+                       ImageInfo{buffer, view.NumSlices(), htile_address, hint},
+                       ImageViewInfo{buffer, view, ctl}} {}
+    };
+
+    struct VideoOutDesc : public BaseDesc {
+        VideoOutDesc(const Libraries::VideoOut::BufferAttributeGroup& group, VAddr cpu_address)
+            : BaseDesc{BindingType::VideoOut, ImageInfo{group, cpu_address}, ImageViewInfo{}} {}
+    };
+
+public:
+    TextureCache(const Vulkan::Instance& instance, Vulkan::Scheduler& scheduler,
+                 BufferCache& buffer_cache, PageManager& tracker);
     ~TextureCache();
 
     /// Invalidates any image in the logical page range.
-    void InvalidateMemory(VAddr address, size_t size);
+    void InvalidateMemory(VAddr addr, VAddr page_addr, size_t size);
 
     /// Marks an image as dirty if it exists at the provided address.
     void InvalidateMemoryFromGPU(VAddr address, size_t max_size);
@@ -57,18 +104,16 @@ public:
     void UnmapMemory(VAddr cpu_addr, size_t size);
 
     /// Retrieves the image handle of the image with the provided attributes.
-    [[nodiscard]] ImageId FindImage(const ImageInfo& info, FindFlags flags = {});
+    [[nodiscard]] ImageId FindImage(BaseDesc& desc, FindFlags flags = {});
 
     /// Retrieves an image view with the properties of the specified image id.
     [[nodiscard]] ImageView& FindTexture(ImageId image_id, const ImageViewInfo& view_info);
 
     /// Retrieves the render target with specified properties
-    [[nodiscard]] ImageView& FindRenderTarget(const ImageInfo& image_info,
+    [[nodiscard]] ImageView& FindRenderTarget(BaseDesc& desc);
-                                              const ImageViewInfo& view_info);
 
     /// Retrieves the depth target with specified properties
-    [[nodiscard]] ImageView& FindDepthTarget(const ImageInfo& image_info,
+    [[nodiscard]] ImageView& FindDepthTarget(BaseDesc& desc);
-                                             const ImageViewInfo& view_info);
 
     /// Updates image contents if it was modified by CPU.
     void UpdateImage(ImageId image_id, Vulkan::Scheduler* custom_scheduler = nullptr) {
@@ -77,11 +122,13 @@ public:
         RefreshImage(image, custom_scheduler);
     }
 
-    [[nodiscard]] ImageId ResolveOverlap(const ImageInfo& info, ImageId cache_img_id,
+    [[nodiscard]] std::tuple<ImageId, int, int> ResolveOverlap(const ImageInfo& info,
-                                         ImageId merged_image_id);
+                                                               BindingType binding,
+                                                               ImageId cache_img_id,
+                                                               ImageId merged_image_id);
 
     /// Resolves depth overlap and either re-creates the image or returns existing one
-    [[nodiscard]] ImageId ResolveDepthOverlap(const ImageInfo& requested_info,
+    [[nodiscard]] ImageId ResolveDepthOverlap(const ImageInfo& requested_info, BindingType binding,
                                               ImageId cache_img_id);
 
     [[nodiscard]] ImageId ExpandImage(const ImageInfo& info, ImageId image_id);
@@ -195,9 +242,15 @@ private:
 
     /// Track CPU reads and writes for image
     void TrackImage(ImageId image_id);
+    void TrackImageHead(ImageId image_id);
+    void TrackImageTail(ImageId image_id);
 
     /// Stop tracking CPU reads and writes for image
     void UntrackImage(ImageId image_id);
+    void UntrackImageHead(ImageId image_id);
+    void UntrackImageTail(ImageId image_id);
+
+    void MarkAsMaybeDirty(ImageId image_id, Image& image);
 
     /// Removes the image and any views/surface metas that reference it.
     void DeleteImage(ImageId image_id);