Merge branch 'main' into fontlib

CMakeLists.txt

@@ -603,6 +603,8 @@ set(MISC_LIBS  src/core/libraries/screenshot/screenshot.cpp
                src/core/libraries/move/move.h
                src/core/libraries/ulobjmgr/ulobjmgr.cpp
                src/core/libraries/ulobjmgr/ulobjmgr.h
+               src/core/libraries/signin_dialog/signindialog.cpp
+               src/core/libraries/signin_dialog/signindialog.h
 )
 
 set(DEV_TOOLS src/core/devtools/layer.cpp

externals/sirit

@@ -1 +1 @@
-Subproject commit 427a42c9ed99b38204d9107bc3dc14e92458acf1
+Subproject commit 09a1416ab1b59ddfebd2618412f118f2004f3b2c

src/common/io_file.cpp

@@ -131,9 +131,7 @@ namespace {
     case SeekOrigin::End:
         return SEEK_END;
     default:
-        LOG_ERROR(Common_Filesystem, "Unsupported origin {}, defaulting to SEEK_SET",
-                  static_cast<u32>(origin));
-        return SEEK_SET;
+        UNREACHABLE_MSG("Impossible SeekOrigin {}", static_cast<u32>(origin));
     }
 }
 

src/common/io_file.h

@@ -61,8 +61,6 @@ enum class SeekOrigin : u32 {
     SetOrigin,       // Seeks from the start of the file.
     CurrentPosition, // Seeks from the current file pointer position.
     End,             // Seeks from the end of the file.
-    SeekHole,        // Seeks from the start of the next hole in the file.
-    SeekData,        // Seeks from the start of the next non-hole region in the file.
 };
 
 class IOFile final {

src/common/logging/filter.cpp

@@ -139,6 +139,7 @@ bool ParseFilterRule(Filter& instance, Iterator begin, Iterator end) {
     SUB(Lib, Hmd)                                                                                  \
     SUB(Lib, Font)                                                                                 \
     SUB(Lib, FontFt)                                                                               \
+    SUB(Lib, SigninDialog)                                                                         \
     CLS(Frontend)                                                                                  \
     CLS(Render)                                                                                    \
     SUB(Render, Vulkan)                                                                            \

src/common/logging/types.h

@@ -104,8 +104,7 @@ enum class Class : u8 {
     Lib_NpParty,           ///< The LibSceNpParty implementation
     Lib_Zlib,              ///< The LibSceZlib implementation.
     Lib_Hmd,               ///< The LibSceHmd implementation.
-    Lib_Font,              ///< The libSceFont implementation.
-    Lib_FontFt,            ///< The libSceFontFt implementation.
+    Lib_SigninDialog,      ///< The LibSigninDialog implementation.
     Frontend,              ///< Emulator UI
     Render,                ///< Video Core
     Render_Vulkan,         ///< Vulkan backend

src/common/memory_patcher.cpp

@@ -23,7 +23,7 @@
 
 namespace MemoryPatcher {
 
-uintptr_t g_eboot_address;
+EXPORT uintptr_t g_eboot_address;
 uint64_t g_eboot_image_size;
 std::string g_game_serial;
 std::string patchFile;

src/common/memory_patcher.h

@@ -6,9 +6,15 @@
 #include <string>
 #include <vector>
 
+#if defined(WIN32)
+#define EXPORT __declspec(dllexport)
+#else
+#define EXPORT __attribute__((visibility("default")))
+#endif
+
 namespace MemoryPatcher {
 
-extern uintptr_t g_eboot_address;
+extern EXPORT uintptr_t g_eboot_address;
 extern uint64_t g_eboot_image_size;
 extern std::string g_game_serial;
 extern std::string patchFile;

src/core/address_space.h

@@ -19,8 +19,6 @@ enum class MemoryPermission : u32 {
 };
 DECLARE_ENUM_FLAG_OPERATORS(MemoryPermission)
 
-constexpr VAddr CODE_BASE_OFFSET = 0x100000000ULL;
-
 constexpr VAddr SYSTEM_MANAGED_MIN = 0x00000400000ULL;
 constexpr VAddr SYSTEM_MANAGED_MAX = 0x07FFFFBFFFULL;
 constexpr VAddr SYSTEM_RESERVED_MIN = 0x07FFFFC000ULL;

src/core/cpu_patches.cpp

@@ -464,9 +464,8 @@ static std::pair<bool, u64> TryPatch(u8* code, PatchModule* module) {
 
             if (needs_trampoline && instruction.length < 5) {
                 // Trampoline is needed but instruction is too short to patch.
-                // Return false and length to fall back to the illegal instruction handler,
-                // or to signal to AOT compilation that this instruction should be skipped and
-                // handled at runtime.
+                // Return false and length to signal to AOT compilation that this instruction
+                // should be skipped and handled at runtime.
                 return std::make_pair(false, instruction.length);
             }
 
@@ -512,136 +511,137 @@ static std::pair<bool, u64> TryPatch(u8* code, PatchModule* module) {
 
 #if defined(ARCH_X86_64)
 
+static bool Is4ByteExtrqOrInsertq(void* code_address) {
+    u8* bytes = (u8*)code_address;
+    if (bytes[0] == 0x66 && bytes[1] == 0x0F && bytes[2] == 0x79) {
+        return true; // extrq
+    } else if (bytes[0] == 0xF2 && bytes[1] == 0x0F && bytes[2] == 0x79) {
+        return true; // insertq
+    } else {
+        return false;
+    }
+}
+
 static bool TryExecuteIllegalInstruction(void* ctx, void* code_address) {
-    ZydisDecodedInstruction instruction;
-    ZydisDecodedOperand operands[ZYDIS_MAX_OPERAND_COUNT];
-    const auto status =
-        Common::Decoder::Instance()->decodeInstruction(instruction, operands, code_address);
+    // We need to decode the instruction to find out what it is. Normally we'd use a fully fleshed
+    // out decoder like Zydis, however Zydis does a bunch of stuff that impact performance that we
+    // don't care about. We can get information about the instruction a lot faster by writing a mini
+    // decoder here, since we know it is definitely an extrq or an insertq. If for some reason we
+    // need to interpret more instructions in the future (I don't see why we would), we can revert
+    // to using Zydis.
+    ZydisMnemonic mnemonic;
+    u8* bytes = (u8*)code_address;
+    if (bytes[0] == 0x66) {
+        mnemonic = ZYDIS_MNEMONIC_EXTRQ;
+    } else if (bytes[0] == 0xF2) {
+        mnemonic = ZYDIS_MNEMONIC_INSERTQ;
+    } else {
+        ZydisDecodedInstruction instruction;
+        ZydisDecodedOperand operands[ZYDIS_MAX_OPERAND_COUNT];
+        const auto status =
+            Common::Decoder::Instance()->decodeInstruction(instruction, operands, code_address);
+        LOG_ERROR(Core, "Unhandled illegal instruction at code address {}: {}",
+                  fmt::ptr(code_address),
+                  ZYAN_SUCCESS(status) ? ZydisMnemonicGetString(instruction.mnemonic)
+                                       : "Failed to decode");
+        return false;
+    }
 
-    switch (instruction.mnemonic) {
+    ASSERT(bytes[1] == 0x0F && bytes[2] == 0x79);
+
+    // Note: It's guaranteed that there's no REX prefix in these instructions checked by
+    // Is4ByteExtrqOrInsertq
+    u8 modrm = bytes[3];
+    u8 rm = modrm & 0b111;
+    u8 reg = (modrm >> 3) & 0b111;
+    u8 mod = (modrm >> 6) & 0b11;
+
+    ASSERT(mod == 0b11); // Any instruction we interpret here uses reg/reg addressing only
+
+    int dstIndex = reg;
+    int srcIndex = rm;
+
+    switch (mnemonic) {
     case ZYDIS_MNEMONIC_EXTRQ: {
-        bool immediateForm = operands[1].type == ZYDIS_OPERAND_TYPE_IMMEDIATE &&
-                             operands[2].type == ZYDIS_OPERAND_TYPE_IMMEDIATE;
-        if (immediateForm) {
-            LOG_CRITICAL(Core, "EXTRQ immediate form should have been patched at code address: {}",
-                         fmt::ptr(code_address));
-            return false;
+        const auto dst = Common::GetXmmPointer(ctx, dstIndex);
+        const auto src = Common::GetXmmPointer(ctx, srcIndex);
+
+        u64 lowQWordSrc;
+        memcpy(&lowQWordSrc, src, sizeof(lowQWordSrc));
+
+        u64 lowQWordDst;
+        memcpy(&lowQWordDst, dst, sizeof(lowQWordDst));
+
+        u64 length = lowQWordSrc & 0x3F;
+        u64 mask;
+        if (length == 0) {
+            length = 64; // for the check below
+            mask = 0xFFFF'FFFF'FFFF'FFFF;
         } else {
-            ASSERT_MSG(operands[0].type == ZYDIS_OPERAND_TYPE_REGISTER &&
-                           operands[1].type == ZYDIS_OPERAND_TYPE_REGISTER &&
-                           operands[0].reg.value >= ZYDIS_REGISTER_XMM0 &&
-                           operands[0].reg.value <= ZYDIS_REGISTER_XMM15 &&
-                           operands[1].reg.value >= ZYDIS_REGISTER_XMM0 &&
-                           operands[1].reg.value <= ZYDIS_REGISTER_XMM15,
-                       "Unexpected operand types for EXTRQ instruction");
-
-            const auto dstIndex = operands[0].reg.value - ZYDIS_REGISTER_XMM0;
-            const auto srcIndex = operands[1].reg.value - ZYDIS_REGISTER_XMM0;
-
-            const auto dst = Common::GetXmmPointer(ctx, dstIndex);
-            const auto src = Common::GetXmmPointer(ctx, srcIndex);
-
-            u64 lowQWordSrc;
-            memcpy(&lowQWordSrc, src, sizeof(lowQWordSrc));
-
-            u64 lowQWordDst;
-            memcpy(&lowQWordDst, dst, sizeof(lowQWordDst));
-
-            u64 length = lowQWordSrc & 0x3F;
-            u64 mask;
-            if (length == 0) {
-                length = 64; // for the check below
-                mask = 0xFFFF'FFFF'FFFF'FFFF;
-            } else {
-                mask = (1ULL << length) - 1;
-            }
-
-            u64 index = (lowQWordSrc >> 8) & 0x3F;
-            if (length + index > 64) {
-                // Undefined behavior if length + index is bigger than 64 according to the spec,
-                // we'll warn and continue execution.
-                LOG_TRACE(Core,
-                          "extrq at {} with length {} and index {} is bigger than 64, "
-                          "undefined behavior",
-                          fmt::ptr(code_address), length, index);
-            }
-
-            lowQWordDst >>= index;
-            lowQWordDst &= mask;
-
-            memcpy(dst, &lowQWordDst, sizeof(lowQWordDst));
-
-            Common::IncrementRip(ctx, instruction.length);
-
-            return true;
+            mask = (1ULL << length) - 1;
         }
-        break;
+
+        u64 index = (lowQWordSrc >> 8) & 0x3F;
+        if (length + index > 64) {
+            // Undefined behavior if length + index is bigger than 64 according to the spec,
+            // we'll warn and continue execution.
+            LOG_TRACE(Core,
+                      "extrq at {} with length {} and index {} is bigger than 64, "
+                      "undefined behavior",
+                      fmt::ptr(code_address), length, index);
+        }
+
+        lowQWordDst >>= index;
+        lowQWordDst &= mask;
+
+        memcpy(dst, &lowQWordDst, sizeof(lowQWordDst));
+
+        Common::IncrementRip(ctx, 4);
+
+        return true;
     }
     case ZYDIS_MNEMONIC_INSERTQ: {
-        bool immediateForm = operands[2].type == ZYDIS_OPERAND_TYPE_IMMEDIATE &&
-                             operands[3].type == ZYDIS_OPERAND_TYPE_IMMEDIATE;
-        if (immediateForm) {
-            LOG_CRITICAL(Core,
-                         "INSERTQ immediate form should have been patched at code address: {}",
-                         fmt::ptr(code_address));
-            return false;
+        const auto dst = Common::GetXmmPointer(ctx, dstIndex);
+        const auto src = Common::GetXmmPointer(ctx, srcIndex);
+
+        u64 lowQWordSrc, highQWordSrc;
+        memcpy(&lowQWordSrc, src, sizeof(lowQWordSrc));
+        memcpy(&highQWordSrc, (u8*)src + 8, sizeof(highQWordSrc));
+
+        u64 lowQWordDst;
+        memcpy(&lowQWordDst, dst, sizeof(lowQWordDst));
+
+        u64 length = highQWordSrc & 0x3F;
+        u64 mask;
+        if (length == 0) {
+            length = 64; // for the check below
+            mask = 0xFFFF'FFFF'FFFF'FFFF;
         } else {
-            ASSERT_MSG(operands[2].type == ZYDIS_OPERAND_TYPE_UNUSED &&
-                           operands[3].type == ZYDIS_OPERAND_TYPE_UNUSED,
-                       "operands 2 and 3 must be unused for register form.");
-
-            ASSERT_MSG(operands[0].type == ZYDIS_OPERAND_TYPE_REGISTER &&
-                           operands[1].type == ZYDIS_OPERAND_TYPE_REGISTER,
-                       "operands 0 and 1 must be registers.");
-
-            const auto dstIndex = operands[0].reg.value - ZYDIS_REGISTER_XMM0;
-            const auto srcIndex = operands[1].reg.value - ZYDIS_REGISTER_XMM0;
-
-            const auto dst = Common::GetXmmPointer(ctx, dstIndex);
-            const auto src = Common::GetXmmPointer(ctx, srcIndex);
-
-            u64 lowQWordSrc, highQWordSrc;
-            memcpy(&lowQWordSrc, src, sizeof(lowQWordSrc));
-            memcpy(&highQWordSrc, (u8*)src + 8, sizeof(highQWordSrc));
-
-            u64 lowQWordDst;
-            memcpy(&lowQWordDst, dst, sizeof(lowQWordDst));
-
-            u64 length = highQWordSrc & 0x3F;
-            u64 mask;
-            if (length == 0) {
-                length = 64; // for the check below
-                mask = 0xFFFF'FFFF'FFFF'FFFF;
-            } else {
-                mask = (1ULL << length) - 1;
-            }
-
-            u64 index = (highQWordSrc >> 8) & 0x3F;
-            if (length + index > 64) {
-                // Undefined behavior if length + index is bigger than 64 according to the spec,
-                // we'll warn and continue execution.
-                LOG_TRACE(Core,
-                          "insertq at {} with length {} and index {} is bigger than 64, "
-                          "undefined behavior",
-                          fmt::ptr(code_address), length, index);
-            }
-
-            lowQWordSrc &= mask;
-            lowQWordDst &= ~(mask << index);
-            lowQWordDst |= lowQWordSrc << index;
-
-            memcpy(dst, &lowQWordDst, sizeof(lowQWordDst));
-
-            Common::IncrementRip(ctx, instruction.length);
-
-            return true;
+            mask = (1ULL << length) - 1;
         }
-        break;
+
+        u64 index = (highQWordSrc >> 8) & 0x3F;
+        if (length + index > 64) {
+            // Undefined behavior if length + index is bigger than 64 according to the spec,
+            // we'll warn and continue execution.
+            LOG_TRACE(Core,
+                      "insertq at {} with length {} and index {} is bigger than 64, "
+                      "undefined behavior",
+                      fmt::ptr(code_address), length, index);
+        }
+
+        lowQWordSrc &= mask;
+        lowQWordDst &= ~(mask << index);
+        lowQWordDst |= lowQWordSrc << index;
+
+        memcpy(dst, &lowQWordDst, sizeof(lowQWordDst));
+
+        Common::IncrementRip(ctx, 4);
+
+        return true;
     }
     default: {
-        LOG_ERROR(Core, "Unhandled illegal instruction at code address {}: {}",
-                  fmt::ptr(code_address), ZydisMnemonicGetString(instruction.mnemonic));
-        return false;
+        UNREACHABLE();
     }
     }
 
@@ -695,9 +695,22 @@ static bool PatchesAccessViolationHandler(void* context, void* /* fault_address
 
 static bool PatchesIllegalInstructionHandler(void* context) {
     void* code_address = Common::GetRip(context);
-    if (!TryPatchJit(code_address)) {
+    if (Is4ByteExtrqOrInsertq(code_address)) {
+        // The instruction is not big enough for a relative jump, don't try to patch it and pass it
+        // to our illegal instruction interpreter directly
         return TryExecuteIllegalInstruction(context, code_address);
+    } else {
+        if (!TryPatchJit(code_address)) {
+            ZydisDecodedInstruction instruction;
+            ZydisDecodedOperand operands[ZYDIS_MAX_OPERAND_COUNT];
+            const auto status =
+                Common::Decoder::Instance()->decodeInstruction(instruction, operands, code_address);
+            LOG_ERROR(Core, "Failed to patch address {:x} -- mnemonic: {}", (u64)code_address,
+                      ZYAN_SUCCESS(status) ? ZydisMnemonicGetString(instruction.mnemonic)
+                                           : "Failed to decode");
+        }
     }
+
     return true;
 }
 

src/core/devices/random_device.cpp

@@ -2,6 +2,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <cstdlib>
+#include <ctime>
 #include "common/logging/log.h"
 #include "random_device.h"
 

src/core/devices/srandom_device.cpp

@@ -2,6 +2,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <cstdlib>
+#include <ctime>
 #include "common/logging/log.h"
 #include "srandom_device.h"
 

src/core/devices/urandom_device.cpp

@@ -2,6 +2,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <cstdlib>
+#include <ctime>
 #include "common/logging/log.h"
 #include "urandom_device.h"
 

src/core/devtools/widget/frame_dump.cpp

@@ -2,6 +2,7 @@
 //  SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <cstdio>
+#include <ctime>
 #include <fmt/chrono.h>
 #include <imgui.h>
 #include <magic_enum/magic_enum.hpp>

src/core/libraries/gnmdriver/gnmdriver.cpp

@@ -505,9 +505,10 @@ s32 PS4_SYSV_ABI sceGnmDrawIndexIndirectCountMulti(u32* cmdbuf, u32 size, u32 da
                                                    u32 flags) {
     LOG_TRACE(Lib_GnmDriver, "called");
 
-    if ((!sceKernelIsNeoMode() || !UseNeoCompatSequences) && !cmdbuf && (size == 16) &&
-        (shader_stage < ShaderStages::Max) && (vertex_sgpr_offset < 0x10u) &&
-        (instance_sgpr_offset < 0x10u)) {
+    if ((!sceKernelIsNeoMode() || !UseNeoCompatSequences) && cmdbuf && (size == 16) &&
+        (vertex_sgpr_offset < 0x10u) && (instance_sgpr_offset < 0x10u) &&
+        (shader_stage == ShaderStages::Vs || shader_stage == ShaderStages::Es ||
+         shader_stage == ShaderStages::Ls)) {
 
         cmdbuf = WriteHeader<PM4ItOpcode::Nop>(cmdbuf, 2);
         cmdbuf = WriteBody(cmdbuf, 0u);
@@ -535,10 +536,33 @@ s32 PS4_SYSV_ABI sceGnmDrawIndexIndirectCountMulti(u32* cmdbuf, u32 size, u32 da
     return -1;
 }
 
-int PS4_SYSV_ABI sceGnmDrawIndexIndirectMulti() {
-    LOG_ERROR(Lib_GnmDriver, "(STUBBED) called");
-    UNREACHABLE();
-    return ORBIS_OK;
+int PS4_SYSV_ABI sceGnmDrawIndexIndirectMulti(u32* cmdbuf, u32 size, u32 data_offset, u32 max_count,
+                                              u32 shader_stage, u32 vertex_sgpr_offset,
+                                              u32 instance_sgpr_offset, u32 flags) {
+    LOG_TRACE(Lib_GnmDriver, "called");
+
+    if (cmdbuf && (size == 11) && (vertex_sgpr_offset < 0x10u) && (instance_sgpr_offset < 0x10u) &&
+        (shader_stage == ShaderStages::Vs || shader_stage == ShaderStages::Es ||
+         shader_stage == ShaderStages::Ls)) {
+
+        const auto predicate = flags & 1 ? PM4Predicate::PredEnable : PM4Predicate::PredDisable;
+        cmdbuf = WriteHeader<PM4ItOpcode::DrawIndexIndirectMulti>(
+            cmdbuf, 6, PM4ShaderType::ShaderGraphics, predicate);
+
+        const auto sgpr_offset = indirect_sgpr_offsets[shader_stage];
+
+        cmdbuf[0] = data_offset;
+        cmdbuf[1] = vertex_sgpr_offset == 0 ? 0 : (vertex_sgpr_offset & 0xffffu) + sgpr_offset;
+        cmdbuf[2] = instance_sgpr_offset == 0 ? 0 : (instance_sgpr_offset & 0xffffu) + sgpr_offset;
+        cmdbuf[3] = max_count;
+        cmdbuf[4] = sizeof(DrawIndexedIndirectArgs);
+        cmdbuf[5] = sceKernelIsNeoMode() ? flags & 0xe0000000u : 0;
+
+        cmdbuf += 6;
+        WriteTrailingNop<3>(cmdbuf);
+        return ORBIS_OK;
+    }
+    return -1;
 }
 
 int PS4_SYSV_ABI sceGnmDrawIndexMultiInstanced() {

src/core/libraries/gnmdriver/gnmdriver.h

@@ -51,7 +51,9 @@ s32 PS4_SYSV_ABI sceGnmDrawIndexIndirectCountMulti(u32* cmdbuf, u32 size, u32 da
                                                    u32 max_count, u64 count_addr, u32 shader_stage,
                                                    u32 vertex_sgpr_offset, u32 instance_sgpr_offset,
                                                    u32 flags);
-int PS4_SYSV_ABI sceGnmDrawIndexIndirectMulti();
+int PS4_SYSV_ABI sceGnmDrawIndexIndirectMulti(u32* cmdbuf, u32 size, u32 data_offset, u32 max_count,
+                                              u32 shader_stage, u32 vertex_sgpr_offset,
+                                              u32 instance_sgpr_offset, u32 flags);
 int PS4_SYSV_ABI sceGnmDrawIndexMultiInstanced();
 s32 PS4_SYSV_ABI sceGnmDrawIndexOffset(u32* cmdbuf, u32 size, u32 index_offset, u32 index_count,
                                        u32 flags);

src/core/libraries/kernel/file_system.cpp

@@ -67,10 +67,16 @@ s32 PS4_SYSV_ABI open(const char* raw_path, s32 flags, u16 mode) {
     bool write = (flags & 0x3) == ORBIS_KERNEL_O_WRONLY;
     bool rdwr = (flags & 0x3) == ORBIS_KERNEL_O_RDWR;
 
+    if (!read && !write && !rdwr) {
+        // Start by checking for invalid flags.
+        *__Error() = POSIX_EINVAL;
+        return -1;
+    }
+
     bool nonblock = (flags & ORBIS_KERNEL_O_NONBLOCK) != 0;
     bool append = (flags & ORBIS_KERNEL_O_APPEND) != 0;
-    bool fsync = (flags & ORBIS_KERNEL_O_FSYNC) != 0;
-    bool sync = (flags & ORBIS_KERNEL_O_SYNC) != 0;
+    // Flags fsync and sync behave the same
+    bool sync = (flags & ORBIS_KERNEL_O_SYNC) != 0 || (flags & ORBIS_KERNEL_O_FSYNC) != 0;
     bool create = (flags & ORBIS_KERNEL_O_CREAT) != 0;
     bool truncate = (flags & ORBIS_KERNEL_O_TRUNC) != 0;
     bool excl = (flags & ORBIS_KERNEL_O_EXCL) != 0;
@@ -78,6 +84,10 @@ s32 PS4_SYSV_ABI open(const char* raw_path, s32 flags, u16 mode) {
     bool direct = (flags & ORBIS_KERNEL_O_DIRECT) != 0;
     bool directory = (flags & ORBIS_KERNEL_O_DIRECTORY) != 0;
 
+    if (sync || direct || dsync || nonblock) {
+        LOG_WARNING(Kernel_Fs, "flags {:#x} not fully handled", flags);
+    }
+
     std::string_view path{raw_path};
     u32 handle = h->CreateHandle();
     auto* file = h->GetFile(handle);
@@ -94,84 +104,127 @@ s32 PS4_SYSV_ABI open(const char* raw_path, s32 flags, u16 mode) {
         }
     }
 
-    if (directory) {
-        file->type = Core::FileSys::FileType::Directory;
-        file->m_guest_name = path;
-        file->m_host_name = mnt->GetHostPath(file->m_guest_name);
-        if (!std::filesystem::is_directory(file->m_host_name)) { // directory doesn't exist
+    bool read_only = false;
+    file->m_guest_name = path;
+    file->m_host_name = mnt->GetHostPath(file->m_guest_name, &read_only);
+    bool exists = std::filesystem::exists(file->m_host_name);
+    s32 e = 0;
+
+    if (create) {
+        if (excl && exists) {
+            // Error if file exists
             h->DeleteHandle(handle);
-            *__Error() = POSIX_ENOENT;
+            *__Error() = POSIX_EEXIST;
+            return -1;
+        }
+
+        if (read_only) {
+            // Can't create files in a read only directory
+            h->DeleteHandle(handle);
+            *__Error() = POSIX_EROFS;
+            return -1;
+        }
+        // Create a file if it doesn't exist
+        Common::FS::IOFile out(file->m_host_name, Common::FS::FileAccessMode::Write);
+    } else if (!exists) {
+        // If we're not creating a file, and it doesn't exist, return ENOENT
+        h->DeleteHandle(handle);
+        *__Error() = POSIX_ENOENT;
+        return -1;
+    }
+
+    if (std::filesystem::is_directory(file->m_host_name) || directory) {
+        // Directories can be opened even if the directory flag isn't set.
+        // In these cases, error behavior is identical to the directory code path.
+        directory = true;
+    }
+
+    if (directory) {
+        if (!std::filesystem::is_directory(file->m_host_name)) {
+            // If the opened file is not a directory, return ENOTDIR.
+            // This will trigger when create & directory is specified, this is expected.
+            h->DeleteHandle(handle);
+            *__Error() = POSIX_ENOTDIR;
+            return -1;
+        }
+
+        file->type = Core::FileSys::FileType::Directory;
+
+        // Populate directory contents
+        mnt->IterateDirectory(file->m_guest_name,
+                              [&file](const auto& ent_path, const auto ent_is_file) {
+                                  auto& dir_entry = file->dirents.emplace_back();
+                                  dir_entry.name = ent_path.filename().string();
+                                  dir_entry.isFile = ent_is_file;
+                              });
+        file->dirents_index = 0;
+
+        if (read) {
+            e = file->f.Open(file->m_host_name, Common::FS::FileAccessMode::Read);
+        } else if (write || rdwr) {
+            // Cannot open directories with any type of write access
+            h->DeleteHandle(handle);
+            *__Error() = POSIX_EISDIR;
+            return -1;
+        }
+
+        if (e == EACCES) {
+            // Hack to bypass some platform limitations, ignore the error and continue as normal.
+            LOG_WARNING(Kernel_Fs, "Opening directories is not fully supported on this platform");
+            e = 0;
+        }
+
+        if (truncate) {
+            // Cannot open directories with truncate
+            h->DeleteHandle(handle);
+            *__Error() = POSIX_EISDIR;
             return -1;
-        } else {
-            if (create) {
-                return handle; // dir already exists
-            } else {
-                mnt->IterateDirectory(file->m_guest_name,
-                                      [&file](const auto& ent_path, const auto ent_is_file) {
-                                          auto& dir_entry = file->dirents.emplace_back();
-                                          dir_entry.name = ent_path.filename().string();
-                                          dir_entry.isFile = ent_is_file;
-                                      });
-                file->dirents_index = 0;
-            }
         }
     } else {
-        file->m_guest_name = path;
-        file->m_host_name = mnt->GetHostPath(file->m_guest_name);
-        bool exists = std::filesystem::exists(file->m_host_name);
-        int e = 0;
+        file->type = Core::FileSys::FileType::Regular;
 
-        if (create) {
-            if (excl && exists) {
-                // Error if file exists
-                h->DeleteHandle(handle);
-                *__Error() = POSIX_EEXIST;
-                return -1;
-            }
-            // Create file if it doesn't exist
-            Common::FS::IOFile out(file->m_host_name, Common::FS::FileAccessMode::Write);
-        } else if (!exists) {
-            // File to open doesn't exist, return ENOENT
+        if (truncate && read_only) {
+            // Can't open files with truncate flag in a read only directory
             h->DeleteHandle(handle);
-            *__Error() = POSIX_ENOENT;
+            *__Error() = POSIX_EROFS;
             return -1;
+        } else if (truncate) {
+            // Open the file as read-write so we can truncate regardless of flags.
+            // Since open starts by closing the file, this won't interfere with later open calls.
+            e = file->f.Open(file->m_host_name, Common::FS::FileAccessMode::ReadWrite);
+            if (e == 0) {
+                // If the file was opened successfully, reduce size to 0
+                file->f.SetSize(0);
+            }
         }
 
         if (read) {
             // Read only
             e = file->f.Open(file->m_host_name, Common::FS::FileAccessMode::Read);
+        } else if (read_only) {
+            // Can't open files with write/read-write access in a read only directory
+            h->DeleteHandle(handle);
+            *__Error() = POSIX_EROFS;
+            return -1;
+        } else if (append) {
+            // Append can be specified with rdwr or write, but we treat it as a separate mode.
+            e = file->f.Open(file->m_host_name, Common::FS::FileAccessMode::Append);
         } else if (write) {
             // Write only
-            if (append) {
-                e = file->f.Open(file->m_host_name, Common::FS::FileAccessMode::Append);
-            } else {
-                e = file->f.Open(file->m_host_name, Common::FS::FileAccessMode::Write);
-            }
+            e = file->f.Open(file->m_host_name, Common::FS::FileAccessMode::Write);
         } else if (rdwr) {
             // Read and write
-            if (append) {
-                e = file->f.Open(file->m_host_name, Common::FS::FileAccessMode::Append);
-            } else {
-                e = file->f.Open(file->m_host_name, Common::FS::FileAccessMode::ReadWrite);
-            }
-        } else {
-            // Invalid flags
-            *__Error() = POSIX_EINVAL;
-            return -1;
-        }
-
-        if (truncate && e == 0) {
-            // If the file was opened successfully and truncate was enabled, reduce size to 0
-            file->f.SetSize(0);
-        }
-
-        if (e != 0) {
-            // Open failed in platform-specific code, errno needs to be converted.
-            h->DeleteHandle(handle);
-            SetPosixErrno(e);
-            return -1;
+            e = file->f.Open(file->m_host_name, Common::FS::FileAccessMode::ReadWrite);
         }
     }
+
+    if (e != 0) {
+        // Open failed in platform-specific code, errno needs to be converted.
+        h->DeleteHandle(handle);
+        SetPosixErrno(e);
+        return -1;
+    }
+
     file->is_opened = true;
     return handle;
 }
@@ -365,10 +418,10 @@ s64 PS4_SYSV_ABI posix_lseek(s32 fd, s64 offset, s32 whence) {
         origin = Common::FS::SeekOrigin::CurrentPosition;
     } else if (whence == 2) {
         origin = Common::FS::SeekOrigin::End;
-    } else if (whence == 3) {
-        origin = Common::FS::SeekOrigin::SeekHole;
-    } else if (whence == 4) {
-        origin = Common::FS::SeekOrigin::SeekData;
+    } else if (whence == 3 || whence == 4) {
+        // whence parameter belongs to an unsupported POSIX extension
+        *__Error() = POSIX_ENOTTY;
+        return -1;
     } else {
         // whence parameter is invalid
         *__Error() = POSIX_EINVAL;
@@ -486,13 +539,13 @@ s32 PS4_SYSV_ABI posix_rmdir(const char* path) {
 
     const std::filesystem::path dir_name = mnt->GetHostPath(path, &ro);
 
-    if (dir_name.empty() || !std::filesystem::is_directory(dir_name)) {
-        *__Error() = POSIX_ENOTDIR;
+    if (ro) {
+        *__Error() = POSIX_EROFS;
         return -1;
     }
 
-    if (ro) {
-        *__Error() = POSIX_EROFS;
+    if (dir_name.empty() || !std::filesystem::is_directory(dir_name)) {
+        *__Error() = POSIX_ENOTDIR;
         return -1;
     }
 
@@ -523,8 +576,7 @@ s32 PS4_SYSV_ABI sceKernelRmdir(const char* path) {
 s32 PS4_SYSV_ABI posix_stat(const char* path, OrbisKernelStat* sb) {
     LOG_INFO(Kernel_Fs, "(PARTIAL) path = {}", path);
     auto* mnt = Common::Singleton<Core::FileSys::MntPoints>::Instance();
-    bool ro = false;
-    const auto path_name = mnt->GetHostPath(path, &ro);
+    const auto path_name = mnt->GetHostPath(path);
     std::memset(sb, 0, sizeof(OrbisKernelStat));
     const bool is_dir = std::filesystem::is_directory(path_name);
     const bool is_file = std::filesystem::is_regular_file(path_name);
@@ -545,9 +597,6 @@ s32 PS4_SYSV_ABI posix_stat(const char* path, OrbisKernelStat* sb) {
         sb->st_blocks = (sb->st_size + 511) / 512;
         // TODO incomplete
     }
-    if (ro) {
-        sb->st_mode &= ~0000555u;
-    }
 
     return ORBIS_OK;
 }

src/core/libraries/kernel/memory.cpp

@@ -126,9 +126,6 @@ s32 PS4_SYSV_ABI sceKernelAvailableDirectMemorySize(u64 searchStart, u64 searchE
 s32 PS4_SYSV_ABI sceKernelVirtualQuery(const void* addr, int flags, OrbisVirtualQueryInfo* info,
                                        size_t infoSize) {
     LOG_INFO(Kernel_Vmm, "called addr = {}, flags = {:#x}", fmt::ptr(addr), flags);
-    if (!addr) {
-        return ORBIS_KERNEL_ERROR_EACCES;
-    }
     auto* memory = Core::Memory::Instance();
     return memory->VirtualQuery(std::bit_cast<VAddr>(addr), flags, info);
 }
@@ -136,7 +133,6 @@ s32 PS4_SYSV_ABI sceKernelVirtualQuery(const void* addr, int flags, OrbisVirtual
 s32 PS4_SYSV_ABI sceKernelReserveVirtualRange(void** addr, u64 len, int flags, u64 alignment) {
     LOG_INFO(Kernel_Vmm, "addr = {}, len = {:#x}, flags = {:#x}, alignment = {:#x}",
              fmt::ptr(*addr), len, flags, alignment);
-
     if (addr == nullptr) {
         LOG_ERROR(Kernel_Vmm, "Address is invalid!");
         return ORBIS_KERNEL_ERROR_EINVAL;
@@ -155,9 +151,12 @@ s32 PS4_SYSV_ABI sceKernelReserveVirtualRange(void** addr, u64 len, int flags, u
     auto* memory = Core::Memory::Instance();
     const VAddr in_addr = reinterpret_cast<VAddr>(*addr);
     const auto map_flags = static_cast<Core::MemoryMapFlags>(flags);
-    memory->Reserve(addr, in_addr, len, map_flags, alignment);
 
-    return ORBIS_OK;
+    s32 result = memory->Reserve(addr, in_addr, len, map_flags, alignment);
+    if (result == 0) {
+        LOG_INFO(Kernel_Vmm, "out_addr = {}", fmt::ptr(*addr));
+    }
+    return result;
 }
 
 int PS4_SYSV_ABI sceKernelMapNamedDirectMemory(void** addr, u64 len, int prot, int flags,
@@ -172,10 +171,12 @@ int PS4_SYSV_ABI sceKernelMapNamedDirectMemory(void** addr, u64 len, int prot, i
         LOG_ERROR(Kernel_Vmm, "Map size is either zero or not 16KB aligned!");
         return ORBIS_KERNEL_ERROR_EINVAL;
     }
+
     if (!Common::Is16KBAligned(directMemoryStart)) {
         LOG_ERROR(Kernel_Vmm, "Start address is not 16KB aligned!");
         return ORBIS_KERNEL_ERROR_EINVAL;
     }
+
     if (alignment != 0) {
         if ((!std::has_single_bit(alignment) && !Common::Is16KBAligned(alignment))) {
             LOG_ERROR(Kernel_Vmm, "Alignment value is invalid!");
@@ -183,14 +184,19 @@ int PS4_SYSV_ABI sceKernelMapNamedDirectMemory(void** addr, u64 len, int prot, i
         }
     }
 
+    if (std::strlen(name) >= ORBIS_KERNEL_MAXIMUM_NAME_LENGTH) {
+        LOG_ERROR(Kernel_Vmm, "name exceeds 32 bytes!");
+        return ORBIS_KERNEL_ERROR_ENAMETOOLONG;
+    }
+
     const VAddr in_addr = reinterpret_cast<VAddr>(*addr);
     const auto mem_prot = static_cast<Core::MemoryProt>(prot);
     const auto map_flags = static_cast<Core::MemoryMapFlags>(flags);
 
     auto* memory = Core::Memory::Instance();
     const auto ret =
-        memory->MapMemory(addr, in_addr, len, mem_prot, map_flags, Core::VMAType::Direct, "", false,
-                          directMemoryStart, alignment);
+        memory->MapMemory(addr, in_addr, len, mem_prot, map_flags, Core::VMAType::Direct, name,
+                          false, directMemoryStart, alignment);
 
     LOG_INFO(Kernel_Vmm, "out_addr = {}", fmt::ptr(*addr));
     return ret;
@@ -199,7 +205,8 @@ int PS4_SYSV_ABI sceKernelMapNamedDirectMemory(void** addr, u64 len, int prot, i
 int PS4_SYSV_ABI sceKernelMapDirectMemory(void** addr, u64 len, int prot, int flags,
                                           s64 directMemoryStart, u64 alignment) {
     LOG_INFO(Kernel_Vmm, "called, redirected to sceKernelMapNamedDirectMemory");
-    return sceKernelMapNamedDirectMemory(addr, len, prot, flags, directMemoryStart, alignment, "");
+    return sceKernelMapNamedDirectMemory(addr, len, prot, flags, directMemoryStart, alignment,
+                                         "anon");
 }
 
 s32 PS4_SYSV_ABI sceKernelMapNamedFlexibleMemory(void** addr_in_out, std::size_t len, int prot,
@@ -210,17 +217,16 @@ s32 PS4_SYSV_ABI sceKernelMapNamedFlexibleMemory(void** addr_in_out, std::size_t
         return ORBIS_KERNEL_ERROR_EINVAL;
     }
 
-    static constexpr size_t MaxNameSize = 32;
-    if (std::strlen(name) > MaxNameSize) {
-        LOG_ERROR(Kernel_Vmm, "name exceeds 32 bytes!");
-        return ORBIS_KERNEL_ERROR_ENAMETOOLONG;
-    }
-
     if (name == nullptr) {
         LOG_ERROR(Kernel_Vmm, "name is invalid!");
         return ORBIS_KERNEL_ERROR_EFAULT;
     }
 
+    if (std::strlen(name) >= ORBIS_KERNEL_MAXIMUM_NAME_LENGTH) {
+        LOG_ERROR(Kernel_Vmm, "name exceeds 32 bytes!");
+        return ORBIS_KERNEL_ERROR_ENAMETOOLONG;
+    }
+
     const VAddr in_addr = reinterpret_cast<VAddr>(*addr_in_out);
     const auto mem_prot = static_cast<Core::MemoryProt>(prot);
     const auto map_flags = static_cast<Core::MemoryMapFlags>(flags);
@@ -236,7 +242,7 @@ s32 PS4_SYSV_ABI sceKernelMapNamedFlexibleMemory(void** addr_in_out, std::size_t
 
 s32 PS4_SYSV_ABI sceKernelMapFlexibleMemory(void** addr_in_out, std::size_t len, int prot,
                                             int flags) {
-    return sceKernelMapNamedFlexibleMemory(addr_in_out, len, prot, flags, "");
+    return sceKernelMapNamedFlexibleMemory(addr_in_out, len, prot, flags, "anon");
 }
 
 int PS4_SYSV_ABI sceKernelQueryMemoryProtection(void* addr, void** start, void** end, u32* prot) {
@@ -304,7 +310,7 @@ s32 PS4_SYSV_ABI sceKernelBatchMap2(OrbisKernelBatchMapEntry* entries, int numEn
         case MemoryOpTypes::ORBIS_KERNEL_MAP_OP_MAP_DIRECT: {
             result = sceKernelMapNamedDirectMemory(&entries[i].start, entries[i].length,
                                                    entries[i].protection, flags,
-                                                   static_cast<s64>(entries[i].offset), 0, "");
+                                                   static_cast<s64>(entries[i].offset), 0, "anon");
             LOG_INFO(Kernel_Vmm,
                      "entry = {}, operation = {}, len = {:#x}, offset = {:#x}, type = {}, "
                      "result = {}",
@@ -326,7 +332,7 @@ s32 PS4_SYSV_ABI sceKernelBatchMap2(OrbisKernelBatchMapEntry* entries, int numEn
         }
         case MemoryOpTypes::ORBIS_KERNEL_MAP_OP_MAP_FLEXIBLE: {
             result = sceKernelMapNamedFlexibleMemory(&entries[i].start, entries[i].length,
-                                                     entries[i].protection, flags, "");
+                                                     entries[i].protection, flags, "anon");
             LOG_INFO(Kernel_Vmm,
                      "entry = {}, operation = {}, len = {:#x}, type = {}, "
                      "result = {}",
@@ -356,16 +362,16 @@ s32 PS4_SYSV_ABI sceKernelBatchMap2(OrbisKernelBatchMapEntry* entries, int numEn
 }
 
 s32 PS4_SYSV_ABI sceKernelSetVirtualRangeName(const void* addr, size_t len, const char* name) {
-    static constexpr size_t MaxNameSize = 32;
-    if (std::strlen(name) > MaxNameSize) {
-        LOG_ERROR(Kernel_Vmm, "name exceeds 32 bytes!");
-        return ORBIS_KERNEL_ERROR_ENAMETOOLONG;
-    }
-
     if (name == nullptr) {
         LOG_ERROR(Kernel_Vmm, "name is invalid!");
         return ORBIS_KERNEL_ERROR_EFAULT;
     }
+
+    if (std::strlen(name) >= ORBIS_KERNEL_MAXIMUM_NAME_LENGTH) {
+        LOG_ERROR(Kernel_Vmm, "name exceeds 32 bytes!");
+        return ORBIS_KERNEL_ERROR_ENAMETOOLONG;
+    }
+
     auto* memory = Core::Memory::Instance();
     memory->NameVirtualRange(std::bit_cast<VAddr>(addr), len, name);
     return ORBIS_OK;

src/core/libraries/kernel/memory.h

@@ -47,6 +47,8 @@ enum MemoryOpTypes : u32 {
     ORBIS_KERNEL_MAP_OP_TYPE_PROTECT = 4
 };
 
+constexpr u32 ORBIS_KERNEL_MAXIMUM_NAME_LENGTH = 32;
+
 struct OrbisQueryInfo {
     uintptr_t start;
     uintptr_t end;
@@ -59,14 +61,12 @@ struct OrbisVirtualQueryInfo {
     size_t offset;
     s32 protection;
     s32 memory_type;
-    union {
-        BitField<0, 1, u32> is_flexible;
-        BitField<1, 1, u32> is_direct;
-        BitField<2, 1, u32> is_stack;
-        BitField<3, 1, u32> is_pooled;
-        BitField<4, 1, u32> is_committed;
-    };
-    std::array<char, 32> name;
+    u32 is_flexible : 1;
+    u32 is_direct : 1;
+    u32 is_stack : 1;
+    u32 is_pooled : 1;
+    u32 is_committed : 1;
+    char name[ORBIS_KERNEL_MAXIMUM_NAME_LENGTH];
 };
 
 struct OrbisKernelBatchMapEntry {

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

@@ -289,7 +289,12 @@ int PS4_SYSV_ABI posix_pthread_create_name_np(PthreadT* thread, const PthreadAtt
     /* Create thread */
     new_thread->native_thr = Core::NativeThread();
     int ret = new_thread->native_thr.Create(RunThread, new_thread, &new_thread->attr);
+
     ASSERT_MSG(ret == 0, "Failed to create thread with error {}", ret);
+
+    if (attr != nullptr && *attr != nullptr && (*attr)->cpuset != nullptr) {
+        new_thread->SetAffinity((*attr)->cpuset);
+    }
     if (ret) {
         *thread = nullptr;
     }
@@ -521,6 +526,69 @@ int PS4_SYSV_ABI posix_pthread_setcancelstate(PthreadCancelState state,
     return 0;
 }
 
+int Pthread::SetAffinity(const Cpuset* cpuset) {
+    const auto processor_count = std::thread::hardware_concurrency();
+    if (processor_count < 8) {
+        return 0;
+    }
+    if (cpuset == nullptr) {
+        return POSIX_EINVAL;
+    }
+
+    u64 mask = cpuset->bits;
+
+    uintptr_t handle = native_thr.GetHandle();
+    if (handle == 0) {
+        return POSIX_ESRCH;
+    }
+
+    // We don't use this currently because some games gets performance problems
+    // when applying affinity even on strong hardware
+    /*
+    #ifdef _WIN64
+        DWORD_PTR affinity_mask = static_cast<DWORD_PTR>(mask);
+        if (!SetThreadAffinityMask(reinterpret_cast<HANDLE>(handle), affinity_mask)) {
+            return POSIX_EINVAL;
+        }
+
+    #elif defined(__linux__)
+        cpu_set_t cpu_set;
+        CPU_ZERO(&cpu_set);
+
+        u64 mask = cpuset->bits;
+        for (int cpu = 0; cpu < std::min(64, CPU_SETSIZE); ++cpu) {
+            if (mask & (1ULL << cpu)) {
+                CPU_SET(cpu, &cpu_set);
+            }
+        }
+
+        int result =
+            pthread_setaffinity_np(static_cast<pthread_t>(handle), sizeof(cpu_set_t), &cpu_set);
+        if (result != 0) {
+            return POSIX_EINVAL;
+        }
+    #endif
+    */
+    return 0;
+}
+
+int PS4_SYSV_ABI posix_pthread_setaffinity_np(PthreadT thread, size_t cpusetsize,
+                                              const Cpuset* cpusetp) {
+    if (thread == nullptr || cpusetp == nullptr) {
+        return POSIX_EINVAL;
+    }
+    thread->attr.cpusetsize = cpusetsize;
+    return thread->SetAffinity(cpusetp);
+}
+
+int PS4_SYSV_ABI scePthreadSetaffinity(PthreadT thread, const Cpuset mask) {
+    int result = posix_pthread_setaffinity_np(thread, 0x10, &mask);
+    if (result != 0) {
+        return ErrnoToSceKernelError(result);
+    }
+    return 0;
+}
+
 void RegisterThread(Core::Loader::SymbolsResolver* sym) {
     // Posix
     LIB_FUNCTION("Z4QosVuAsA0", "libScePosix", 1, "libkernel", 1, 1, posix_pthread_once);
@@ -544,6 +612,7 @@ void RegisterThread(Core::Loader::SymbolsResolver* sym) {
     LIB_FUNCTION("Z4QosVuAsA0", "libkernel", 1, "libkernel", 1, 1, posix_pthread_once);
     LIB_FUNCTION("EotR8a3ASf4", "libkernel", 1, "libkernel", 1, 1, posix_pthread_self);
     LIB_FUNCTION("OxhIB8LB-PQ", "libkernel", 1, "libkernel", 1, 1, posix_pthread_create);
+    LIB_FUNCTION("5KWrg7-ZqvE", "libkernel", 1, "libkernel", 1, 1, posix_pthread_setaffinity_np);
 
     // Orbis
     LIB_FUNCTION("14bOACANTBo", "libkernel", 1, "libkernel", 1, 1, ORBIS(posix_pthread_once));
@@ -566,6 +635,7 @@ void RegisterThread(Core::Loader::SymbolsResolver* sym) {
     LIB_FUNCTION("W0Hpm2X0uPE", "libkernel", 1, "libkernel", 1, 1, ORBIS(posix_pthread_setprio));
     LIB_FUNCTION("rNhWz+lvOMU", "libkernel", 1, "libkernel", 1, 1, _sceKernelSetThreadDtors);
     LIB_FUNCTION("6XG4B33N09g", "libkernel", 1, "libkernel", 1, 1, sched_yield);
+    LIB_FUNCTION("bt3CTBKmGyI", "libkernel", 1, "libkernel", 1, 1, scePthreadSetaffinity)
 }
 
 } // namespace Libraries::Kernel

src/core/libraries/kernel/threads/pthread.h

@@ -332,6 +332,8 @@ struct Pthread {
             return true;
         }
     }
+
+    int SetAffinity(const Cpuset* cpuset);
 };
 using PthreadT = Pthread*;
 

src/core/libraries/libc_internal/printf.h

@@ -56,7 +56,6 @@
 #include <stdio.h>
 
 #include <cstdarg>
-#include <cstdbool>
 #include <cstddef>
 #include <cstdint>
 #include <cstring>

src/core/libraries/libs.cpp

@@ -45,6 +45,7 @@
 #include "core/libraries/save_data/savedata.h"
 #include "core/libraries/screenshot/screenshot.h"
 #include "core/libraries/share_play/shareplay.h"
+#include "core/libraries/signin_dialog/signindialog.h"
 #include "core/libraries/system/commondialog.h"
 #include "core/libraries/system/msgdialog.h"
 #include "core/libraries/system/posix.h"
@@ -120,6 +121,7 @@ void InitHLELibs(Core::Loader::SymbolsResolver* sym) {
     Libraries::Hmd::RegisterlibSceHmd(sym);
     Libraries::DiscMap::RegisterlibSceDiscMap(sym);
     Libraries::Ulobjmgr::RegisterlibSceUlobjmgr(sym);
+    Libraries::SigninDialog::RegisterlibSceSigninDialog(sym);
 }
 
 } // namespace Libraries

src/core/libraries/save_data/dialog/savedatadialog_ui.cpp

@@ -49,13 +49,11 @@ void SaveDialogResult::CopyTo(OrbisSaveDataDialogResult& result) const {
     result.mode = this->mode;
     result.result = this->result;
     result.buttonId = this->button_id;
-    if (mode == SaveDataDialogMode::LIST || ElfInfo::Instance().FirmwareVer() >= ElfInfo::FW_45) {
-        if (result.dirName != nullptr) {
-            result.dirName->data.FromString(this->dir_name);
-        }
-        if (result.param != nullptr && this->param.GetString(SaveParams::MAINTITLE).has_value()) {
-            result.param->FromSFO(this->param);
-        }
+    if (result.dirName != nullptr) {
+        result.dirName->data.FromString(this->dir_name);
+    }
+    if (result.param != nullptr && this->param.GetString(SaveParams::MAINTITLE).has_value()) {
+        result.param->FromSFO(this->param);
     }
     result.userData = this->user_data;
 }
@@ -345,12 +343,15 @@ SaveDialogUi::SaveDialogUi(SaveDialogUi&& other) noexcept
     }
 }
 
-SaveDialogUi& SaveDialogUi::operator=(SaveDialogUi other) {
+SaveDialogUi& SaveDialogUi::operator=(SaveDialogUi&& other) noexcept {
     std::scoped_lock lock(draw_mutex, other.draw_mutex);
     using std::swap;
-    swap(state, other.state);
-    swap(status, other.status);
-    swap(result, other.result);
+    state = other.state;
+    other.state = nullptr;
+    status = other.status;
+    other.status = nullptr;
+    result = other.result;
+    other.result = nullptr;
     if (status && *status == Status::RUNNING) {
         first_render = true;
         AddLayer(this);

src/core/libraries/save_data/dialog/savedatadialog_ui.h

@@ -300,7 +300,8 @@ public:
     ~SaveDialogUi() override;
     SaveDialogUi(const SaveDialogUi& other) = delete;
     SaveDialogUi(SaveDialogUi&& other) noexcept;
-    SaveDialogUi& operator=(SaveDialogUi other);
+    SaveDialogUi& operator=(SaveDialogUi& other) = delete;
+    SaveDialogUi& operator=(SaveDialogUi&& other) noexcept;
 
     void Finish(ButtonId buttonId, CommonDialog::Result r = CommonDialog::Result::OK);
 

src/core/libraries/signin_dialog/signindialog.cpp

@@ -0,0 +1,64 @@
+// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+// Generated By moduleGenerator
+#include "common/logging/log.h"
+#include "core/libraries/error_codes.h"
+#include "core/libraries/libs.h"
+#include "signindialog.h"
+
+namespace Libraries::SigninDialog {
+
+s32 PS4_SYSV_ABI sceSigninDialogInitialize() {
+    LOG_ERROR(Lib_SigninDialog, "(STUBBED) called");
+    return ORBIS_OK;
+}
+
+s32 PS4_SYSV_ABI sceSigninDialogOpen() {
+    LOG_ERROR(Lib_SigninDialog, "(STUBBED) called");
+    return ORBIS_OK;
+}
+
+Status PS4_SYSV_ABI sceSigninDialogGetStatus() {
+    LOG_ERROR(Lib_SigninDialog, "(STUBBED) called, return 'finished' status");
+    return Status::FINISHED;
+}
+
+Status PS4_SYSV_ABI sceSigninDialogUpdateStatus() {
+    LOG_ERROR(Lib_SigninDialog, "(STUBBED) called, return 'finished' status");
+    return Status::FINISHED;
+}
+
+s32 PS4_SYSV_ABI sceSigninDialogGetResult() {
+    LOG_ERROR(Lib_SigninDialog, "(STUBBED) called");
+    return ORBIS_OK;
+}
+
+s32 PS4_SYSV_ABI sceSigninDialogClose() {
+    LOG_ERROR(Lib_SigninDialog, "(STUBBED) called");
+    return ORBIS_OK;
+}
+
+s32 PS4_SYSV_ABI sceSigninDialogTerminate() {
+    LOG_ERROR(Lib_SigninDialog, "(STUBBED) called");
+    return ORBIS_OK;
+}
+
+void RegisterlibSceSigninDialog(Core::Loader::SymbolsResolver* sym) {
+    LIB_FUNCTION("mlYGfmqE3fQ", "libSceSigninDialog", 1, "libSceSigninDialog", 1, 1,
+                 sceSigninDialogInitialize);
+    LIB_FUNCTION("JlpJVoRWv7U", "libSceSigninDialog", 1, "libSceSigninDialog", 1, 1,
+                 sceSigninDialogOpen);
+    LIB_FUNCTION("2m077aeC+PA", "libSceSigninDialog", 1, "libSceSigninDialog", 1, 1,
+                 sceSigninDialogGetStatus);
+    LIB_FUNCTION("Bw31liTFT3A", "libSceSigninDialog", 1, "libSceSigninDialog", 1, 1,
+                 sceSigninDialogUpdateStatus);
+    LIB_FUNCTION("nqG7rqnYw1U", "libSceSigninDialog", 1, "libSceSigninDialog", 1, 1,
+                 sceSigninDialogGetResult);
+    LIB_FUNCTION("M3OkENHcyiU", "libSceSigninDialog", 1, "libSceSigninDialog", 1, 1,
+                 sceSigninDialogClose);
+    LIB_FUNCTION("LXlmS6PvJdU", "libSceSigninDialog", 1, "libSceSigninDialog", 1, 1,
+                 sceSigninDialogTerminate);
+};
+
+} // namespace Libraries::SigninDialog

src/core/libraries/signin_dialog/signindialog.h

@@ -0,0 +1,29 @@
+// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+#include "common/types.h"
+
+namespace Core::Loader {
+class SymbolsResolver;
+}
+
+enum class Status : u32 {
+    NONE = 0,
+    INITIALIZED = 1,
+    RUNNING = 2,
+    FINISHED = 3,
+};
+
+namespace Libraries::SigninDialog {
+
+s32 PS4_SYSV_ABI sceSigninDialogInitialize();
+s32 PS4_SYSV_ABI sceSigninDialogOpen();
+Status PS4_SYSV_ABI sceSigninDialogGetStatus();
+Status PS4_SYSV_ABI sceSigninDialogUpdateStatus();
+s32 PS4_SYSV_ABI sceSigninDialogGetResult();
+s32 PS4_SYSV_ABI sceSigninDialogClose();
+s32 PS4_SYSV_ABI sceSigninDialogTerminate();
+
+void RegisterlibSceSigninDialog(Core::Loader::SymbolsResolver* sym);
+} // namespace Libraries::SigninDialog

src/core/memory.cpp

@@ -75,7 +75,8 @@ u64 MemoryManager::ClampRangeSize(VAddr virtual_addr, u64 size) {
 
     // Clamp size to the remaining size of the current VMA.
     auto vma = FindVMA(virtual_addr);
-    ASSERT_MSG(vma != vma_map.end(), "Attempted to access invalid GPU address {:#x}", virtual_addr);
+    ASSERT_MSG(vma->second.Contains(virtual_addr, 0),
+               "Attempted to access invalid GPU address {:#x}", virtual_addr);
     u64 clamped_size = vma->second.base + vma->second.size - virtual_addr;
     ++vma;
 
@@ -96,6 +97,8 @@ u64 MemoryManager::ClampRangeSize(VAddr virtual_addr, u64 size) {
 bool MemoryManager::TryWriteBacking(void* address, const void* data, u32 num_bytes) {
     const VAddr virtual_addr = std::bit_cast<VAddr>(address);
     const auto& vma = FindVMA(virtual_addr)->second;
+    ASSERT_MSG(vma.Contains(virtual_addr, 0),
+               "Attempting to access out of bounds memory at address {:#x}", virtual_addr);
     if (vma.type != VMAType::Direct) {
         return false;
     }
@@ -145,10 +148,12 @@ PAddr MemoryManager::Allocate(PAddr search_start, PAddr search_end, size_t size,
     auto mapping_end = mapping_start + size;
 
     // Find the first free, large enough dmem area in the range.
-    while ((!dmem_area->second.is_free || dmem_area->second.GetEnd() < mapping_end) &&
-           dmem_area != dmem_map.end()) {
+    while (!dmem_area->second.is_free || dmem_area->second.GetEnd() < mapping_end) {
         // The current dmem_area isn't suitable, move to the next one.
         dmem_area++;
+        if (dmem_area == dmem_map.end()) {
+            break;
+        }
 
         // Update local variables based on the new dmem_area
         mapping_start = Common::AlignUp(dmem_area->second.base, alignment);
@@ -172,7 +177,6 @@ void MemoryManager::Free(PAddr phys_addr, size_t size) {
     std::scoped_lock lk{mutex};
 
     auto dmem_area = CarveDmemArea(phys_addr, size);
-    ASSERT(dmem_area != dmem_map.end() && dmem_area->second.size >= size);
 
     // Release any dmem mappings that reference this physical block.
     std::vector<std::pair<VAddr, u64>> remove_list;
@@ -216,12 +220,18 @@ int MemoryManager::PoolReserve(void** out_addr, VAddr virtual_addr, size_t size,
             vma = FindVMA(mapped_addr)->second;
         }
         const size_t remaining_size = vma.base + vma.size - mapped_addr;
-        ASSERT_MSG(vma.type == VMAType::Free && remaining_size >= size);
+        ASSERT_MSG(vma.type == VMAType::Free && remaining_size >= size,
+                   "Memory region {:#x} to {:#x} is not large enough to reserve {:#x} to {:#x}",
+                   vma.base, vma.base + vma.size, virtual_addr, virtual_addr + size);
     }
 
     // Find the first free area starting with provided virtual address.
     if (False(flags & MemoryMapFlags::Fixed)) {
         mapped_addr = SearchFree(mapped_addr, size, alignment);
+        if (mapped_addr == -1) {
+            // No suitable memory areas to map to
+            return ORBIS_KERNEL_ERROR_ENOMEM;
+        }
     }
 
     // Add virtual memory area
@@ -229,7 +239,7 @@ int MemoryManager::PoolReserve(void** out_addr, VAddr virtual_addr, size_t size,
     auto& new_vma = new_vma_handle->second;
     new_vma.disallow_merge = True(flags & MemoryMapFlags::NoCoalesce);
     new_vma.prot = MemoryProt::NoAccess;
-    new_vma.name = "";
+    new_vma.name = "anon";
     new_vma.type = VMAType::PoolReserved;
     MergeAdjacent(vma_map, new_vma_handle);
 
@@ -247,19 +257,25 @@ int MemoryManager::Reserve(void** out_addr, VAddr virtual_addr, size_t size, Mem
 
     // Fixed mapping means the virtual address must exactly match the provided one.
     if (True(flags & MemoryMapFlags::Fixed)) {
-        auto& vma = FindVMA(mapped_addr)->second;
+        auto vma = FindVMA(mapped_addr)->second;
         // If the VMA is mapped, unmap the region first.
         if (vma.IsMapped()) {
             UnmapMemoryImpl(mapped_addr, size);
             vma = FindVMA(mapped_addr)->second;
         }
         const size_t remaining_size = vma.base + vma.size - mapped_addr;
-        ASSERT_MSG(vma.type == VMAType::Free && remaining_size >= size);
+        ASSERT_MSG(vma.type == VMAType::Free && remaining_size >= size,
+                   "Memory region {:#x} to {:#x} is not large enough to reserve {:#x} to {:#x}",
+                   vma.base, vma.base + vma.size, virtual_addr, virtual_addr + size);
     }
 
     // Find the first free area starting with provided virtual address.
     if (False(flags & MemoryMapFlags::Fixed)) {
         mapped_addr = SearchFree(mapped_addr, size, alignment);
+        if (mapped_addr == -1) {
+            // No suitable memory areas to map to
+            return ORBIS_KERNEL_ERROR_ENOMEM;
+        }
     }
 
     // Add virtual memory area
@@ -267,7 +283,7 @@ int MemoryManager::Reserve(void** out_addr, VAddr virtual_addr, size_t size, Mem
     auto& new_vma = new_vma_handle->second;
     new_vma.disallow_merge = True(flags & MemoryMapFlags::NoCoalesce);
     new_vma.prot = MemoryProt::NoAccess;
-    new_vma.name = "";
+    new_vma.name = "anon";
     new_vma.type = VMAType::Reserved;
     MergeAdjacent(vma_map, new_vma_handle);
 
@@ -288,7 +304,9 @@ int MemoryManager::PoolCommit(VAddr virtual_addr, size_t size, MemoryProt prot)
     // This should return SCE_KERNEL_ERROR_ENOMEM but shouldn't normally happen.
     const auto& vma = FindVMA(mapped_addr)->second;
     const size_t remaining_size = vma.base + vma.size - mapped_addr;
-    ASSERT_MSG(!vma.IsMapped() && remaining_size >= size);
+    ASSERT_MSG(!vma.IsMapped() && remaining_size >= size,
+               "Memory region {:#x} to {:#x} isn't free enough to map region {:#x} to {:#x}",
+               vma.base, vma.base + vma.size, virtual_addr, virtual_addr + size);
 
     // Perform the mapping.
     void* out_addr = impl.Map(mapped_addr, size, alignment, -1, false);
@@ -302,7 +320,10 @@ int MemoryManager::PoolCommit(VAddr virtual_addr, size_t size, MemoryProt prot)
     new_vma.is_exec = false;
     new_vma.phys_base = 0;
 
-    rasterizer->MapMemory(mapped_addr, size);
+    if (IsValidGpuMapping(mapped_addr, size)) {
+        rasterizer->MapMemory(mapped_addr, size);
+    }
+
     return ORBIS_OK;
 }
 
@@ -325,15 +346,34 @@ int MemoryManager::MapMemory(void** out_addr, VAddr virtual_addr, size_t size, M
 
     // Fixed mapping means the virtual address must exactly match the provided one.
     if (True(flags & MemoryMapFlags::Fixed)) {
-        // This should return SCE_KERNEL_ERROR_ENOMEM but shouldn't normally happen.
-        const auto& vma = FindVMA(mapped_addr)->second;
-        const size_t remaining_size = vma.base + vma.size - mapped_addr;
-        ASSERT_MSG(!vma.IsMapped() && remaining_size >= size);
+        auto vma = FindVMA(mapped_addr)->second;
+        size_t remaining_size = vma.base + vma.size - mapped_addr;
+        // There's a possible edge case where we're mapping to a partially reserved range.
+        // To account for this, unmap any reserved areas within this mapping range first.
+        auto unmap_addr = mapped_addr;
+        auto unmap_size = size;
+        while (!vma.IsMapped() && unmap_addr < mapped_addr + size && remaining_size < size) {
+            auto unmapped = UnmapBytesFromEntry(unmap_addr, vma, unmap_size);
+            unmap_addr += unmapped;
+            unmap_size -= unmapped;
+            vma = FindVMA(unmap_addr)->second;
+        }
+
+        // This should return SCE_KERNEL_ERROR_ENOMEM but rarely happens.
+        vma = FindVMA(mapped_addr)->second;
+        remaining_size = vma.base + vma.size - mapped_addr;
+        ASSERT_MSG(!vma.IsMapped() && remaining_size >= size,
+                   "Memory region {:#x} to {:#x} isn't free enough to map region {:#x} to {:#x}",
+                   vma.base, vma.base + vma.size, virtual_addr, virtual_addr + size);
     }
 
     // Find the first free area starting with provided virtual address.
     if (False(flags & MemoryMapFlags::Fixed)) {
         mapped_addr = SearchFree(mapped_addr, size, alignment);
+        if (mapped_addr == -1) {
+            // No suitable memory areas to map to
+            return ORBIS_KERNEL_ERROR_ENOMEM;
+        }
     }
 
     // Perform the mapping.
@@ -353,7 +393,10 @@ int MemoryManager::MapMemory(void** out_addr, VAddr virtual_addr, size_t size, M
     if (type == VMAType::Flexible) {
         flexible_usage += size;
     }
-    rasterizer->MapMemory(mapped_addr, size);
+
+    if (IsValidGpuMapping(mapped_addr, size)) {
+        rasterizer->MapMemory(mapped_addr, size);
+    }
 
     return ORBIS_OK;
 }
@@ -366,12 +409,18 @@ int MemoryManager::MapFile(void** out_addr, VAddr virtual_addr, size_t size, Mem
     // Find first free area to map the file.
     if (False(flags & MemoryMapFlags::Fixed)) {
         mapped_addr = SearchFree(mapped_addr, size_aligned, 1);
+        if (mapped_addr == -1) {
+            // No suitable memory areas to map to
+            return ORBIS_KERNEL_ERROR_ENOMEM;
+        }
     }
 
     if (True(flags & MemoryMapFlags::Fixed)) {
         const auto& vma = FindVMA(virtual_addr)->second;
         const size_t remaining_size = vma.base + vma.size - virtual_addr;
-        ASSERT_MSG(!vma.IsMapped() && remaining_size >= size);
+        ASSERT_MSG(!vma.IsMapped() && remaining_size >= size,
+                   "Memory region {:#x} to {:#x} isn't free enough to map region {:#x} to {:#x}",
+                   vma.base, vma.base + vma.size, virtual_addr, virtual_addr + size);
     }
 
     // Map the file.
@@ -404,7 +453,9 @@ void MemoryManager::PoolDecommit(VAddr virtual_addr, size_t size) {
     const auto start_in_vma = virtual_addr - vma_base_addr;
     const auto type = vma_base.type;
 
-    rasterizer->UnmapMemory(virtual_addr, size);
+    if (IsValidGpuMapping(virtual_addr, size)) {
+        rasterizer->UnmapMemory(virtual_addr, size);
+    }
 
     // Mark region as free and attempt to coalesce it with neighbours.
     const auto new_it = CarveVMA(virtual_addr, size);
@@ -444,7 +495,10 @@ u64 MemoryManager::UnmapBytesFromEntry(VAddr virtual_addr, VirtualMemoryArea vma
     if (type == VMAType::Flexible) {
         flexible_usage -= adjusted_size;
     }
-    rasterizer->UnmapMemory(virtual_addr, adjusted_size);
+
+    if (IsValidGpuMapping(virtual_addr, adjusted_size)) {
+        rasterizer->UnmapMemory(virtual_addr, adjusted_size);
+    }
 
     // Mark region as free and attempt to coalesce it with neighbours.
     const auto new_it = CarveVMA(virtual_addr, adjusted_size);
@@ -471,6 +525,8 @@ s32 MemoryManager::UnmapMemoryImpl(VAddr virtual_addr, u64 size) {
     do {
         auto it = FindVMA(virtual_addr + unmapped_bytes);
         auto& vma_base = it->second;
+        ASSERT_MSG(vma_base.Contains(virtual_addr + unmapped_bytes, 0),
+                   "Address {:#x} is out of bounds", virtual_addr + unmapped_bytes);
         auto unmapped =
             UnmapBytesFromEntry(virtual_addr + unmapped_bytes, vma_base, size - unmapped_bytes);
         ASSERT_MSG(unmapped > 0, "Failed to unmap memory, progress is impossible");
@@ -485,7 +541,10 @@ int MemoryManager::QueryProtection(VAddr addr, void** start, void** end, u32* pr
 
     const auto it = FindVMA(addr);
     const auto& vma = it->second;
-    ASSERT_MSG(vma.type != VMAType::Free, "Provided address is not mapped");
+    if (!vma.Contains(addr, 0) || vma.IsFree()) {
+        LOG_ERROR(Kernel_Vmm, "Address {:#x} is not mapped", addr);
+        return ORBIS_KERNEL_ERROR_EACCES;
+    }
 
     if (start != nullptr) {
         *start = reinterpret_cast<void*>(vma.base);
@@ -555,6 +614,8 @@ s32 MemoryManager::Protect(VAddr addr, size_t size, MemoryProt prot) {
     do {
         auto it = FindVMA(addr + protected_bytes);
         auto& vma_base = it->second;
+        ASSERT_MSG(vma_base.Contains(addr + protected_bytes, 0), "Address {:#x} is out of bounds",
+                   addr + protected_bytes);
         auto result = 0;
         result = ProtectBytes(addr + protected_bytes, vma_base, size - protected_bytes, prot);
         if (result < 0) {
@@ -571,8 +632,16 @@ int MemoryManager::VirtualQuery(VAddr addr, int flags,
                                 ::Libraries::Kernel::OrbisVirtualQueryInfo* info) {
     std::scoped_lock lk{mutex};
 
-    auto it = FindVMA(addr);
-    if (it->second.type == VMAType::Free && flags == 1) {
+    // FindVMA on addresses before the vma_map return garbage data.
+    auto query_addr =
+        addr < impl.SystemManagedVirtualBase() ? impl.SystemManagedVirtualBase() : addr;
+    if (addr < query_addr && flags == 0) {
+        LOG_WARNING(Kernel_Vmm, "VirtualQuery on free memory region");
+        return ORBIS_KERNEL_ERROR_EACCES;
+    }
+    auto it = FindVMA(query_addr);
+
+    while (it->second.type == VMAType::Free && flags == 1 && it != --vma_map.end()) {
         ++it;
     }
     if (it->second.type == VMAType::Free) {
@@ -585,15 +654,17 @@ int MemoryManager::VirtualQuery(VAddr addr, int flags,
     info->end = vma.base + vma.size;
     info->offset = vma.phys_base;
     info->protection = static_cast<s32>(vma.prot);
-    info->is_flexible.Assign(vma.type == VMAType::Flexible);
-    info->is_direct.Assign(vma.type == VMAType::Direct);
-    info->is_stack.Assign(vma.type == VMAType::Stack);
-    info->is_pooled.Assign(vma.type == VMAType::PoolReserved || vma.type == VMAType::Pooled);
-    info->is_committed.Assign(vma.IsMapped());
-    vma.name.copy(info->name.data(), std::min(info->name.size(), vma.name.size()));
+    info->is_flexible = vma.type == VMAType::Flexible ? 1 : 0;
+    info->is_direct = vma.type == VMAType::Direct ? 1 : 0;
+    info->is_stack = vma.type == VMAType::Stack ? 1 : 0;
+    info->is_pooled = vma.type == VMAType::PoolReserved || vma.type == VMAType::Pooled ? 1 : 0;
+    info->is_committed = vma.IsMapped() ? 1 : 0;
+
+    strncpy(info->name, vma.name.data(), ::Libraries::Kernel::ORBIS_KERNEL_MAXIMUM_NAME_LENGTH);
+
     if (vma.type == VMAType::Direct) {
         const auto dmem_it = FindDmemArea(vma.phys_base);
-        ASSERT(dmem_it != dmem_map.end());
+        ASSERT_MSG(vma.phys_base <= dmem_it->second.GetEnd(), "vma.phys_base is not in dmem_map!");
         info->memory_type = dmem_it->second.memory_type;
     } else {
         info->memory_type = ::Libraries::Kernel::SCE_KERNEL_WB_ONION;
@@ -607,11 +678,11 @@ int MemoryManager::DirectMemoryQuery(PAddr addr, bool find_next,
     std::scoped_lock lk{mutex};
 
     auto dmem_area = FindDmemArea(addr);
-    while (dmem_area != dmem_map.end() && dmem_area->second.is_free && find_next) {
+    while (dmem_area != --dmem_map.end() && dmem_area->second.is_free && find_next) {
         dmem_area++;
     }
 
-    if (dmem_area == dmem_map.end() || dmem_area->second.is_free) {
+    if (dmem_area->second.is_free) {
         LOG_ERROR(Core, "Unable to find allocated direct memory region to query!");
         return ORBIS_KERNEL_ERROR_EACCES;
     }
@@ -691,36 +762,56 @@ VAddr MemoryManager::SearchFree(VAddr virtual_addr, size_t size, u32 alignment)
         virtual_addr = min_search_address;
     }
 
+    // If the requested address is beyond the maximum our code can handle, throw an assert
+    auto max_search_address = impl.UserVirtualBase() + impl.UserVirtualSize();
+    ASSERT_MSG(virtual_addr <= max_search_address, "Input address {:#x} is out of bounds",
+               virtual_addr);
+
     auto it = FindVMA(virtual_addr);
-    ASSERT_MSG(it != vma_map.end(), "Specified mapping address was not found!");
 
     // If the VMA is free and contains the requested mapping we are done.
     if (it->second.IsFree() && it->second.Contains(virtual_addr, size)) {
         return virtual_addr;
     }
+
     // Search for the first free VMA that fits our mapping.
-    const auto is_suitable = [&] {
+    while (it != vma_map.end()) {
         if (!it->second.IsFree()) {
-            return false;
+            it++;
+            continue;
         }
+
         const auto& vma = it->second;
         virtual_addr = Common::AlignUp(vma.base, alignment);
         // Sometimes the alignment itself might be larger than the VMA.
         if (virtual_addr > vma.base + vma.size) {
-            return false;
+            it++;
+            continue;
         }
+
+        // Make sure the address is within our defined bounds
+        if (virtual_addr >= max_search_address) {
+            // There are no free mappings within our safely usable address space.
+            break;
+        }
+
+        // If there's enough space in the VMA, return the address.
         const size_t remaining_size = vma.base + vma.size - virtual_addr;
-        return remaining_size >= size;
-    };
-    while (!is_suitable()) {
-        ++it;
+        if (remaining_size >= size) {
+            return virtual_addr;
+        }
+        it++;
     }
-    return virtual_addr;
+
+    // Couldn't find a suitable VMA, return an error.
+    LOG_ERROR(Kernel_Vmm, "Couldn't find a free mapping for address {:#x}, size {:#x}",
+              virtual_addr, size);
+    return -1;
 }
 
 MemoryManager::VMAHandle MemoryManager::CarveVMA(VAddr virtual_addr, size_t size) {
     auto vma_handle = FindVMA(virtual_addr);
-    ASSERT_MSG(vma_handle != vma_map.end(), "Virtual address not in vm_map");
+    ASSERT_MSG(vma_handle->second.Contains(virtual_addr, 0), "Virtual address not in vm_map");
 
     const VirtualMemoryArea& vma = vma_handle->second;
     ASSERT_MSG(vma.base <= virtual_addr, "Adding a mapping to already mapped region");
@@ -749,7 +840,7 @@ MemoryManager::VMAHandle MemoryManager::CarveVMA(VAddr virtual_addr, size_t size
 
 MemoryManager::DMemHandle MemoryManager::CarveDmemArea(PAddr addr, size_t size) {
     auto dmem_handle = FindDmemArea(addr);
-    ASSERT_MSG(dmem_handle != dmem_map.end(), "Physical address not in dmem_map");
+    ASSERT_MSG(addr <= dmem_handle->second.GetEnd(), "Physical address not in dmem_map");
 
     const DirectMemoryArea& area = dmem_handle->second;
     ASSERT_MSG(area.base <= addr, "Adding an allocation to already allocated region");
@@ -804,7 +895,7 @@ int MemoryManager::GetDirectMemoryType(PAddr addr, int* directMemoryTypeOut,
 
     auto dmem_area = FindDmemArea(addr);
 
-    if (dmem_area == dmem_map.end() || dmem_area->second.is_free) {
+    if (addr > dmem_area->second.GetEnd() || dmem_area->second.is_free) {
         LOG_ERROR(Core, "Unable to find allocated direct memory region to check type!");
         return ORBIS_KERNEL_ERROR_ENOENT;
     }

src/core/memory.h

@@ -157,6 +157,12 @@ public:
         return impl.SystemReservedVirtualBase();
     }
 
+    bool IsValidGpuMapping(VAddr virtual_addr, u64 size) {
+        // The PS4's GPU can only handle 40 bit addresses.
+        const VAddr max_gpu_address{0x10000000000};
+        return virtual_addr + size < max_gpu_address;
+    }
+
     bool IsValidAddress(const void* addr) const noexcept {
         const VAddr virtual_addr = reinterpret_cast<VAddr>(addr);
         const auto end_it = std::prev(vma_map.end());
@@ -186,7 +192,7 @@ public:
     int PoolCommit(VAddr virtual_addr, size_t size, MemoryProt prot);
 
     int MapMemory(void** out_addr, VAddr virtual_addr, size_t size, MemoryProt prot,
-                  MemoryMapFlags flags, VMAType type, std::string_view name = "",
+                  MemoryMapFlags flags, VMAType type, std::string_view name = "anon",
                   bool is_exec = false, PAddr phys_addr = -1, u64 alignment = 0);
 
     int MapFile(void** out_addr, VAddr virtual_addr, size_t size, MemoryProt prot,

src/core/module.cpp

@@ -19,8 +19,7 @@ namespace Core {
 
 using EntryFunc = PS4_SYSV_ABI int (*)(size_t args, const void* argp, void* param);
 
-static u64 LoadOffset = CODE_BASE_OFFSET;
-static constexpr u64 CODE_BASE_INCR = 0x010000000u;
+static constexpr u64 ModuleLoadBase = 0x800000000;
 
 static u64 GetAlignedSize(const elf_program_header& phdr) {
     return (phdr.p_align != 0 ? (phdr.p_memsz + (phdr.p_align - 1)) & ~(phdr.p_align - 1)
@@ -84,7 +83,7 @@ static std::string StringToNid(std::string_view symbol) {
 }
 
 Module::Module(Core::MemoryManager* memory_, const std::filesystem::path& file_, u32& max_tls_index)
-    : memory{memory_}, file{file_}, name{file.stem().string()} {
+    : memory{memory_}, file{file_}, name{file.filename().string()} {
     elf.Open(file);
     if (elf.IsElfFile()) {
         LoadModuleToMemory(max_tls_index);
@@ -113,10 +112,8 @@ void Module::LoadModuleToMemory(u32& max_tls_index) {
 
     // Map module segments (and possible TLS trampolines)
     void** out_addr = reinterpret_cast<void**>(&base_virtual_addr);
-    memory->MapMemory(out_addr, memory->SystemReservedVirtualBase() + LoadOffset,
-                      aligned_base_size + TrampolineSize, MemoryProt::CpuReadWrite,
-                      MemoryMapFlags::Fixed, VMAType::Code, name, true);
-    LoadOffset += CODE_BASE_INCR * (1 + aligned_base_size / CODE_BASE_INCR);
+    memory->MapMemory(out_addr, ModuleLoadBase, aligned_base_size + TrampolineSize,
+                      MemoryProt::CpuReadWrite, MemoryMapFlags::NoFlags, VMAType::Code, name, true);
     LOG_INFO(Core_Linker, "Loading module {} to {}", name, fmt::ptr(*out_addr));
 
 #ifdef ARCH_X86_64
@@ -229,7 +226,7 @@ void Module::LoadModuleToMemory(u32& max_tls_index) {
     LOG_INFO(Core_Linker, "program entry addr ..........: {:#018x}", entry_addr);
 
     if (MemoryPatcher::g_eboot_address == 0) {
-        if (name == "eboot") {
+        if (name == "eboot.bin") {
             MemoryPatcher::g_eboot_address = base_virtual_addr;
             MemoryPatcher::g_eboot_image_size = base_size;
             MemoryPatcher::OnGameLoaded();

src/core/tls.h

@@ -5,6 +5,8 @@
 
 #include "common/types.h"
 
+void* memset(void* ptr, int value, size_t num);
+
 namespace Xbyak {
 class CodeGenerator;
 }
@@ -41,9 +43,18 @@ Tcb* GetTcbBase();
 /// Makes sure TLS is initialized for the thread before entering guest.
 void EnsureThreadInitialized();
 
+template <size_t size>
+__attribute__((optnone)) void ClearStack() {
+    volatile void* buf = alloca(size);
+    memset(const_cast<void*>(buf), 0, size);
+    buf = nullptr;
+}
+
 template <class ReturnType, class... FuncArgs, class... CallArgs>
 ReturnType ExecuteGuest(PS4_SYSV_ABI ReturnType (*func)(FuncArgs...), CallArgs&&... args) {
     EnsureThreadInitialized();
+    // clear stack to avoid trash from EnsureThreadInitialized
+    ClearStack<13_KB>();
     return func(std::forward<CallArgs>(args)...);
 }
 

src/emulator.cpp

@@ -10,7 +10,6 @@
 #include "common/logging/log.h"
 #ifdef ENABLE_QT_GUI
 #include <QtCore>
-#include "common/memory_patcher.h"
 #endif
 #include "common/assert.h"
 #ifdef ENABLE_DISCORD_RPC
@@ -20,6 +19,7 @@
 #include <WinSock2.h>
 #endif
 #include "common/elf_info.h"
+#include "common/memory_patcher.h"
 #include "common/ntapi.h"
 #include "common/path_util.h"
 #include "common/polyfill_thread.h"
@@ -56,27 +56,6 @@ Emulator::Emulator() {
     WSADATA wsaData;
     WSAStartup(versionWanted, &wsaData);
 #endif
-
-    // Create stdin/stdout/stderr
-    Common::Singleton<FileSys::HandleTable>::Instance()->CreateStdHandles();
-
-    // Defer until after logging is initialized.
-    memory = Core::Memory::Instance();
-    controller = Common::Singleton<Input::GameController>::Instance();
-    linker = Common::Singleton<Core::Linker>::Instance();
-
-    // Load renderdoc module.
-    VideoCore::LoadRenderDoc();
-
-    // Start the timer (Play Time)
-#ifdef ENABLE_QT_GUI
-    start_time = std::chrono::steady_clock::now();
-    const auto user_dir = Common::FS::GetUserPath(Common::FS::PathType::UserDir);
-    QString filePath = QString::fromStdString((user_dir / "play_time.txt").string());
-    QFile file(filePath);
-    ASSERT_MSG(file.open(QIODevice::ReadWrite | QIODevice::Text),
-               "Error opening or creating play_time.txt");
-#endif
 }
 
 Emulator::~Emulator() {
@@ -100,58 +79,93 @@ void Emulator::Run(const std::filesystem::path& file, const std::vector<std::str
 
     // Applications expect to be run from /app0 so mount the file's parent path as app0.
     auto* mnt = Common::Singleton<Core::FileSys::MntPoints>::Instance();
-    mnt->Mount(game_folder, "/app0");
+    mnt->Mount(game_folder, "/app0", true);
     // Certain games may use /hostapp as well such as CUSA001100
-    mnt->Mount(game_folder, "/hostapp");
+    mnt->Mount(game_folder, "/hostapp", true);
 
-    auto& game_info = Common::ElfInfo::Instance();
+    const auto param_sfo_path = mnt->GetHostPath("/app0/sce_sys/param.sfo");
+    const auto param_sfo_exists = std::filesystem::exists(param_sfo_path);
 
-    // Loading param.sfo file if exists
+    // Load param.sfo details if it exists
     std::string id;
     std::string title;
     std::string app_version;
     u32 fw_version;
     Common::PSFAttributes psf_attributes{};
-
-    const auto param_sfo_path = mnt->GetHostPath("/app0/sce_sys/param.sfo");
-    if (!std::filesystem::exists(param_sfo_path) || !Config::getSeparateLogFilesEnabled()) {
-        Common::Log::Initialize();
-        Common::Log::Start();
-    }
-
-    if (std::filesystem::exists(param_sfo_path)) {
+    if (param_sfo_exists) {
         auto* param_sfo = Common::Singleton<PSF>::Instance();
-        const bool success = param_sfo->Open(param_sfo_path);
-        ASSERT_MSG(success, "Failed to open param.sfo");
+        ASSERT_MSG(param_sfo->Open(param_sfo_path), "Failed to open param.sfo");
+
         const auto content_id = param_sfo->GetString("CONTENT_ID");
         ASSERT_MSG(content_id.has_value(), "Failed to get CONTENT_ID");
+
         id = std::string(*content_id, 7, 9);
-
-        if (Config::getSeparateLogFilesEnabled()) {
-            Common::Log::Initialize(id + ".log");
-            Common::Log::Start();
+        title = param_sfo->GetString("TITLE").value_or("Unknown title");
+        fw_version = param_sfo->GetInteger("SYSTEM_VER").value_or(0x4700000);
+        app_version = param_sfo->GetString("APP_VER").value_or("Unknown version");
+        if (const auto raw_attributes = param_sfo->GetInteger("ATTRIBUTE")) {
+            psf_attributes.raw = *raw_attributes;
         }
-        LOG_INFO(Loader, "Starting shadps4 emulator v{} ", Common::g_version);
-        LOG_INFO(Loader, "Revision {}", Common::g_scm_rev);
-        LOG_INFO(Loader, "Branch {}", Common::g_scm_branch);
-        LOG_INFO(Loader, "Description {}", Common::g_scm_desc);
-        LOG_INFO(Loader, "Remote {}", Common::g_scm_remote_url);
+    }
 
-        LOG_INFO(Config, "General LogType: {}", Config::getLogType());
-        LOG_INFO(Config, "General isNeo: {}", Config::isNeoModeConsole());
-        LOG_INFO(Config, "GPU isNullGpu: {}", Config::nullGpu());
-        LOG_INFO(Config, "GPU shouldDumpShaders: {}", Config::dumpShaders());
-        LOG_INFO(Config, "GPU vblankDivider: {}", Config::vblankDiv());
-        LOG_INFO(Config, "Vulkan gpuId: {}", Config::getGpuId());
-        LOG_INFO(Config, "Vulkan vkValidation: {}", Config::vkValidationEnabled());
-        LOG_INFO(Config, "Vulkan vkValidationSync: {}", Config::vkValidationSyncEnabled());
-        LOG_INFO(Config, "Vulkan vkValidationGpu: {}", Config::vkValidationGpuEnabled());
-        LOG_INFO(Config, "Vulkan crashDiagnostics: {}", Config::getVkCrashDiagnosticEnabled());
-        LOG_INFO(Config, "Vulkan hostMarkers: {}", Config::getVkHostMarkersEnabled());
-        LOG_INFO(Config, "Vulkan guestMarkers: {}", Config::getVkGuestMarkersEnabled());
-        LOG_INFO(Config, "Vulkan rdocEnable: {}", Config::isRdocEnabled());
+    // Initialize logging as soon as possible
+    if (!id.empty() && Config::getSeparateLogFilesEnabled()) {
+        Common::Log::Initialize(id + ".log");
+    } else {
+        Common::Log::Initialize();
+    }
+    Common::Log::Start();
 
+    LOG_INFO(Loader, "Starting shadps4 emulator v{} ", Common::g_version);
+    LOG_INFO(Loader, "Revision {}", Common::g_scm_rev);
+    LOG_INFO(Loader, "Branch {}", Common::g_scm_branch);
+    LOG_INFO(Loader, "Description {}", Common::g_scm_desc);
+    LOG_INFO(Loader, "Remote {}", Common::g_scm_remote_url);
+
+    LOG_INFO(Config, "General LogType: {}", Config::getLogType());
+    LOG_INFO(Config, "General isNeo: {}", Config::isNeoModeConsole());
+    LOG_INFO(Config, "GPU isNullGpu: {}", Config::nullGpu());
+    LOG_INFO(Config, "GPU shouldDumpShaders: {}", Config::dumpShaders());
+    LOG_INFO(Config, "GPU vblankDivider: {}", Config::vblankDiv());
+    LOG_INFO(Config, "Vulkan gpuId: {}", Config::getGpuId());
+    LOG_INFO(Config, "Vulkan vkValidation: {}", Config::vkValidationEnabled());
+    LOG_INFO(Config, "Vulkan vkValidationSync: {}", Config::vkValidationSyncEnabled());
+    LOG_INFO(Config, "Vulkan vkValidationGpu: {}", Config::vkValidationGpuEnabled());
+    LOG_INFO(Config, "Vulkan crashDiagnostics: {}", Config::getVkCrashDiagnosticEnabled());
+    LOG_INFO(Config, "Vulkan hostMarkers: {}", Config::getVkHostMarkersEnabled());
+    LOG_INFO(Config, "Vulkan guestMarkers: {}", Config::getVkGuestMarkersEnabled());
+    LOG_INFO(Config, "Vulkan rdocEnable: {}", Config::isRdocEnabled());
+
+    if (param_sfo_exists) {
+        LOG_INFO(Loader, "Game id: {} Title: {}", id, title);
+        LOG_INFO(Loader, "Fw: {:#x} App Version: {}", fw_version, app_version);
+    }
+    if (!args.empty()) {
+        const auto argc = std::min<size_t>(args.size(), 32);
+        for (auto i = 0; i < argc; i++) {
+            LOG_INFO(Loader, "Game argument {}: {}", i, args[i]);
+        }
+        if (args.size() > 32) {
+            LOG_ERROR(Loader, "Too many game arguments, only passing the first 32");
+        }
+    }
+
+    // Create stdin/stdout/stderr
+    Common::Singleton<FileSys::HandleTable>::Instance()->CreateStdHandles();
+
+    // Initialize components
+    memory = Core::Memory::Instance();
+    controller = Common::Singleton<Input::GameController>::Instance();
+    linker = Common::Singleton<Core::Linker>::Instance();
+
+    // Load renderdoc module
+    VideoCore::LoadRenderDoc();
+
+    // Initialize patcher and trophies
+    if (!id.empty()) {
+        MemoryPatcher::g_game_serial = id;
         Libraries::NpTrophy::game_serial = id;
+
         const auto trophyDir =
             Common::FS::GetUserPath(Common::FS::PathType::MetaDataDir) / id / "TrophyFiles";
         if (!std::filesystem::exists(trophyDir)) {
@@ -160,41 +174,9 @@ void Emulator::Run(const std::filesystem::path& file, const std::vector<std::str
                 LOG_ERROR(Loader, "Couldn't extract trophies");
             }
         }
-#ifdef ENABLE_QT_GUI
-        MemoryPatcher::g_game_serial = id;
-
-        // Timer for 'Play Time'
-        QTimer* timer = new QTimer();
-        QObject::connect(timer, &QTimer::timeout, [this, id]() {
-            UpdatePlayTime(id);
-            start_time = std::chrono::steady_clock::now();
-        });
-        timer->start(60000); // 60000 ms = 1 minute
-#endif
-        title = param_sfo->GetString("TITLE").value_or("Unknown title");
-        LOG_INFO(Loader, "Game id: {} Title: {}", id, title);
-        fw_version = param_sfo->GetInteger("SYSTEM_VER").value_or(0x4700000);
-        app_version = param_sfo->GetString("APP_VER").value_or("Unknown version");
-        LOG_INFO(Loader, "Fw: {:#x} App Version: {}", fw_version, app_version);
-        if (const auto raw_attributes = param_sfo->GetInteger("ATTRIBUTE")) {
-            psf_attributes.raw = *raw_attributes;
-        }
-        if (!args.empty()) {
-            int argc = std::min<int>(args.size(), 32);
-            for (int i = 0; i < argc; i++) {
-                LOG_INFO(Loader, "Game argument {}: {}", i, args[i]);
-            }
-            if (args.size() > 32) {
-                LOG_ERROR(Loader, "Too many game arguments, only passing the first 32");
-            }
-        }
-    }
-
-    const auto pic1_path = mnt->GetHostPath("/app0/sce_sys/pic1.png");
-    if (std::filesystem::exists(pic1_path)) {
-        game_info.splash_path = pic1_path;
     }
 
+    auto& game_info = Common::ElfInfo::Instance();
     game_info.initialized = true;
     game_info.game_serial = id;
     game_info.title = title;
@@ -203,6 +185,11 @@ void Emulator::Run(const std::filesystem::path& file, const std::vector<std::str
     game_info.raw_firmware_ver = fw_version;
     game_info.psf_attributes = psf_attributes;
 
+    const auto pic1_path = mnt->GetHostPath("/app0/sce_sys/pic1.png");
+    if (std::filesystem::exists(pic1_path)) {
+        game_info.splash_path = pic1_path;
+    }
+
     std::string game_title = fmt::format("{} - {} <{}>", id, title, app_version);
     std::string window_title = "";
     if (Common::g_is_release) {
@@ -233,11 +220,15 @@ void Emulator::Run(const std::filesystem::path& file, const std::vector<std::str
         std::filesystem::create_directory(mount_data_dir);
     }
     mnt->Mount(mount_data_dir, "/data"); // should just exist, manually create with game serial
+
+    // Mounting temp folders
     const auto& mount_temp_dir = Common::FS::GetUserPath(Common::FS::PathType::TempDataDir) / id;
-    if (!std::filesystem::exists(mount_temp_dir)) {
-        std::filesystem::create_directory(mount_temp_dir);
+    if (std::filesystem::exists(mount_temp_dir)) {
+        // Temp folder should be cleared on each boot.
+        std::filesystem::remove_all(mount_temp_dir);
     }
-    mnt->Mount(mount_temp_dir, "/temp0"); // called in app_content ==> stat/mkdir
+    std::filesystem::create_directory(mount_temp_dir);
+    mnt->Mount(mount_temp_dir, "/temp0");
     mnt->Mount(mount_temp_dir, "/temp");
 
     const auto& mount_download_dir =
@@ -282,6 +273,25 @@ void Emulator::Run(const std::filesystem::path& file, const std::vector<std::str
     }
 #endif
 
+    // Start the timer (Play Time)
+#ifdef ENABLE_QT_GUI
+    if (!id.empty()) {
+        auto* timer = new QTimer();
+        QObject::connect(timer, &QTimer::timeout, [this, id]() {
+            UpdatePlayTime(id);
+            start_time = std::chrono::steady_clock::now();
+        });
+        timer->start(60000); // 60000 ms = 1 minute
+
+        start_time = std::chrono::steady_clock::now();
+        const auto user_dir = Common::FS::GetUserPath(Common::FS::PathType::UserDir);
+        QString filePath = QString::fromStdString((user_dir / "play_time.txt").string());
+        QFile file(filePath);
+        ASSERT_MSG(file.open(QIODevice::ReadWrite | QIODevice::Text),
+                   "Error opening or creating play_time.txt");
+    }
+#endif
+
     linker->Execute(args);
 
     window->InitTimers();

src/shader_recompiler/backend/spirv/emit_spirv.cpp

@@ -270,6 +270,10 @@ void SetupCapabilities(const Info& info, const Profile& profile, EmitContext& ct
     if (info.has_image_query) {
         ctx.AddCapability(spv::Capability::ImageQuery);
     }
+    if (info.uses_atomic_float_min_max && profile.supports_image_fp32_atomic_min_max) {
+        ctx.AddExtension("SPV_EXT_shader_atomic_float_min_max");
+        ctx.AddCapability(spv::Capability::AtomicFloat32MinMaxEXT);
+    }
     if (info.uses_lane_id) {
         ctx.AddCapability(spv::Capability::GroupNonUniform);
     }

src/shader_recompiler/backend/spirv/emit_spirv_atomic.cpp

@@ -75,6 +75,14 @@ Id ImageAtomicU32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id va
     const auto [scope, semantics]{AtomicArgs(ctx)};
     return (ctx.*atomic_func)(ctx.U32[1], pointer, scope, semantics, value);
 }
+
+Id ImageAtomicF32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id value,
+                  Id (Sirit::Module::*atomic_func)(Id, Id, Id, Id, Id)) {
+    const auto& texture = ctx.images[handle & 0xFFFF];
+    const Id pointer{ctx.OpImageTexelPointer(ctx.image_f32, texture.id, coords, ctx.ConstU32(0U))};
+    const auto [scope, semantics]{AtomicArgs(ctx)};
+    return (ctx.*atomic_func)(ctx.F32[1], pointer, scope, semantics, value);
+}
 } // Anonymous namespace
 
 Id EmitSharedAtomicIAdd32(EmitContext& ctx, Id offset, Id value) {
@@ -187,6 +195,40 @@ Id EmitImageAtomicUMax32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords
     return ImageAtomicU32(ctx, inst, handle, coords, value, &Sirit::Module::OpAtomicUMax);
 }
 
+Id EmitImageAtomicFMax32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id value) {
+    if (ctx.profile.supports_image_fp32_atomic_min_max) {
+        return ImageAtomicF32(ctx, inst, handle, coords, value, &Sirit::Module::OpAtomicFMax);
+    }
+
+    const auto u32_value = ctx.OpBitcast(ctx.U32[1], value);
+    const auto sign_bit_set =
+        ctx.OpBitFieldUExtract(ctx.U32[1], u32_value, ctx.ConstU32(31u), ctx.ConstU32(1u));
+
+    const auto result = ctx.OpSelect(
+        ctx.F32[1], sign_bit_set,
+        EmitBitCastF32U32(ctx, EmitImageAtomicUMin32(ctx, inst, handle, coords, u32_value)),
+        EmitBitCastF32U32(ctx, EmitImageAtomicSMax32(ctx, inst, handle, coords, u32_value)));
+
+    return result;
+}
+
+Id EmitImageAtomicFMin32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id value) {
+    if (ctx.profile.supports_image_fp32_atomic_min_max) {
+        return ImageAtomicF32(ctx, inst, handle, coords, value, &Sirit::Module::OpAtomicFMin);
+    }
+
+    const auto u32_value = ctx.OpBitcast(ctx.U32[1], value);
+    const auto sign_bit_set =
+        ctx.OpBitFieldUExtract(ctx.U32[1], u32_value, ctx.ConstU32(31u), ctx.ConstU32(1u));
+
+    const auto result = ctx.OpSelect(
+        ctx.F32[1], sign_bit_set,
+        EmitBitCastF32U32(ctx, EmitImageAtomicUMax32(ctx, inst, handle, coords, u32_value)),
+        EmitBitCastF32U32(ctx, EmitImageAtomicSMin32(ctx, inst, handle, coords, u32_value)));
+
+    return result;
+}
+
 Id EmitImageAtomicInc32(EmitContext&, IR::Inst*, u32, Id, Id) {
     // TODO: This is not yet implemented
     throw NotImplementedException("SPIR-V Instruction");

src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp

@@ -529,7 +529,7 @@ void EmitStoreBufferBoundsCheck(EmitContext& ctx, Id index, Id buffer_size, auto
         // Bounds checking enabled, wrap in a conditional branch.
         auto compare_index = index;
         if (N > 1) {
-            index = ctx.OpIAdd(ctx.U32[1], index, ctx.ConstU32(N - 1));
+            compare_index = ctx.OpIAdd(ctx.U32[1], index, ctx.ConstU32(N - 1));
         }
         const Id in_bounds = ctx.OpULessThan(ctx.U1[1], compare_index, buffer_size);
         const Id in_bounds_label = ctx.OpLabel();

src/shader_recompiler/backend/spirv/emit_spirv_instructions.h

@@ -482,6 +482,8 @@ Id EmitImageAtomicSMin32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords
 Id EmitImageAtomicUMin32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id value);
 Id EmitImageAtomicSMax32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id value);
 Id EmitImageAtomicUMax32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id value);
+Id EmitImageAtomicFMax32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id value);
+Id EmitImageAtomicFMin32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id value);
 Id EmitImageAtomicInc32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id value);
 Id EmitImageAtomicDec32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id value);
 Id EmitImageAtomicAnd32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id value);

src/shader_recompiler/backend/spirv/spirv_emit_context.cpp

@@ -869,6 +869,7 @@ void EmitContext::DefineImagesAndSamplers() {
     }
     if (std::ranges::any_of(info.images, &ImageResource::is_atomic)) {
         image_u32 = TypePointer(spv::StorageClass::Image, U32[1]);
+        image_f32 = TypePointer(spv::StorageClass::Image, F32[1]);
     }
     if (info.samplers.empty()) {
         return;

src/shader_recompiler/backend/spirv/spirv_emit_context.h

@@ -207,6 +207,7 @@ public:
     Id invocation_id{};
     Id subgroup_local_invocation_id{};
     Id image_u32{};
+    Id image_f32{};
 
     Id shared_memory_u8{};
     Id shared_memory_u16{};

src/shader_recompiler/frontend/translate/translate.cpp

@@ -517,7 +517,9 @@ void Translator::EmitFetch(const GcnInst& inst) {
         const auto values =
             ir.CompositeConstruct(ir.GetAttribute(attr, 0), ir.GetAttribute(attr, 1),
                                   ir.GetAttribute(attr, 2), ir.GetAttribute(attr, 3));
-        const auto swizzled = ApplySwizzle(ir, values, buffer.DstSelect());
+        const auto converted =
+            IR::ApplyReadNumberConversionVec4(ir, values, buffer.GetNumberConversion());
+        const auto swizzled = ApplySwizzle(ir, converted, buffer.DstSelect());
         for (u32 i = 0; i < 4; i++) {
             ir.SetVectorReg(dst_reg++, IR::F32{ir.CompositeExtract(swizzled, i)});
         }

src/shader_recompiler/frontend/translate/vector_memory.cpp

@@ -115,8 +115,12 @@ void Translator::EmitVectorMemory(const GcnInst& inst) {
         return IMAGE_ATOMIC(AtomicOp::Smin, inst);
     case Opcode::IMAGE_ATOMIC_UMIN:
         return IMAGE_ATOMIC(AtomicOp::Umin, inst);
+    case Opcode::IMAGE_ATOMIC_FMIN:
+        return IMAGE_ATOMIC(AtomicOp::Fmin, inst);
     case Opcode::IMAGE_ATOMIC_SMAX:
         return IMAGE_ATOMIC(AtomicOp::Smax, inst);
+    case Opcode::IMAGE_ATOMIC_FMAX:
+        return IMAGE_ATOMIC(AtomicOp::Fmax, inst);
     case Opcode::IMAGE_ATOMIC_UMAX:
         return IMAGE_ATOMIC(AtomicOp::Umax, inst);
     case Opcode::IMAGE_ATOMIC_AND:
@@ -139,6 +143,7 @@ void Translator::EmitVectorMemory(const GcnInst& inst) {
     case Opcode::IMAGE_SAMPLE_C_LZ:
     case Opcode::IMAGE_SAMPLE_O:
     case Opcode::IMAGE_SAMPLE_L_O:
+    case Opcode::IMAGE_SAMPLE_B_O:
     case Opcode::IMAGE_SAMPLE_LZ_O:
     case Opcode::IMAGE_SAMPLE_C_O:
     case Opcode::IMAGE_SAMPLE_C_LZ_O:
@@ -466,6 +471,10 @@ void Translator::IMAGE_ATOMIC(AtomicOp op, const GcnInst& inst) {
             return ir.ImageAtomicIMax(handle, body, value, true, info);
         case AtomicOp::Umax:
             return ir.ImageAtomicUMax(handle, body, value, info);
+        case AtomicOp::Fmax:
+            return ir.ImageAtomicFMax(handle, body, value, info);
+        case AtomicOp::Fmin:
+            return ir.ImageAtomicFMin(handle, body, value, info);
         case AtomicOp::And:
             return ir.ImageAtomicAnd(handle, body, value, info);
         case AtomicOp::Or:

src/shader_recompiler/info.h

@@ -196,6 +196,7 @@ struct Info {
     bool has_discard{};
     bool has_image_gather{};
     bool has_image_query{};
+    bool uses_atomic_float_min_max{};
     bool uses_lane_id{};
     bool uses_group_quad{};
     bool uses_group_ballot{};
@@ -280,6 +281,11 @@ constexpr AmdGpu::Image ImageResource::GetSharp(const Info& info) const noexcept
         // Fall back to null image if unbound.
         return AmdGpu::Image::Null();
     }
+    const auto data_fmt = image.GetDataFmt();
+    if (is_depth && data_fmt != AmdGpu::DataFormat::Format16 &&
+        data_fmt != AmdGpu::DataFormat::Format32) {
+        return AmdGpu::Image::NullDepth();
+    }
     return image;
 }
 

src/shader_recompiler/ir/ir_emitter.cpp

@@ -1870,6 +1870,16 @@ Value IREmitter::ImageAtomicUMax(const Value& handle, const Value& coords, const
     return Inst(Opcode::ImageAtomicUMax32, Flags{info}, handle, coords, value);
 }
 
+Value IREmitter::ImageAtomicFMax(const Value& handle, const Value& coords, const Value& value,
+                                 TextureInstInfo info) {
+    return Inst(Opcode::ImageAtomicFMax32, Flags{info}, handle, coords, value);
+}
+
+Value IREmitter::ImageAtomicFMin(const Value& handle, const Value& coords, const Value& value,
+                                 TextureInstInfo info) {
+    return Inst(Opcode::ImageAtomicFMin32, Flags{info}, handle, coords, value);
+}
+
 Value IREmitter::ImageAtomicIMax(const Value& handle, const Value& coords, const Value& value,
                                  bool is_signed, TextureInstInfo info) {
     return is_signed ? ImageAtomicSMax(handle, coords, value, info)

src/shader_recompiler/ir/ir_emitter.h

@@ -321,6 +321,10 @@ public:
                                         const Value& value, TextureInstInfo info);
     [[nodiscard]] Value ImageAtomicUMax(const Value& handle, const Value& coords,
                                         const Value& value, TextureInstInfo info);
+    [[nodiscard]] Value ImageAtomicFMax(const Value& handle, const Value& coords,
+                                        const Value& value, TextureInstInfo info);
+    [[nodiscard]] Value ImageAtomicFMin(const Value& handle, const Value& coords,
+                                        const Value& value, TextureInstInfo info);
     [[nodiscard]] Value ImageAtomicIMax(const Value& handle, const Value& coords,
                                         const Value& value, bool is_signed, TextureInstInfo info);
     [[nodiscard]] Value ImageAtomicInc(const Value& handle, const Value& coords, const Value& value,

src/shader_recompiler/ir/microinstruction.cpp

@@ -94,6 +94,8 @@ bool Inst::MayHaveSideEffects() const noexcept {
     case Opcode::ImageAtomicUMin32:
     case Opcode::ImageAtomicSMax32:
     case Opcode::ImageAtomicUMax32:
+    case Opcode::ImageAtomicFMax32:
+    case Opcode::ImageAtomicFMin32:
     case Opcode::ImageAtomicInc32:
     case Opcode::ImageAtomicDec32:
     case Opcode::ImageAtomicAnd32:

src/shader_recompiler/ir/opcodes.inc

@@ -420,6 +420,8 @@ OPCODE(ImageAtomicSMin32,                                   U32,            Opaq
 OPCODE(ImageAtomicUMin32,                                   U32,            Opaque,         Opaque,         U32,                                            )
 OPCODE(ImageAtomicSMax32,                                   U32,            Opaque,         Opaque,         U32,                                            )
 OPCODE(ImageAtomicUMax32,                                   U32,            Opaque,         Opaque,         U32,                                            )
+OPCODE(ImageAtomicFMax32,                                   F32,            Opaque,         Opaque,         F32,                                            )
+OPCODE(ImageAtomicFMin32,                                   F32,            Opaque,         Opaque,         F32,                                            )
 OPCODE(ImageAtomicInc32,                                    U32,            Opaque,         Opaque,         U32,                                            )
 OPCODE(ImageAtomicDec32,                                    U32,            Opaque,         Opaque,         U32,                                            )
 OPCODE(ImageAtomicAnd32,                                    U32,            Opaque,         Opaque,         U32,                                            )

src/shader_recompiler/ir/passes/lower_buffer_format_to_raw.cpp

@@ -206,7 +206,8 @@ static void LowerBufferFormatInst(IR::Block& block, IR::Inst& inst, Info& info)
         .swizzle = is_inst_typed
                        ? AmdGpu::RemapSwizzle(flags.inst_data_fmt.Value(), AmdGpu::IdentityMapping)
                        : buffer.DstSelect(),
-        .num_conversion = is_inst_typed ? AmdGpu::MapNumberConversion(flags.inst_num_fmt.Value())
+        .num_conversion = is_inst_typed ? AmdGpu::MapNumberConversion(flags.inst_num_fmt.Value(),
+                                                                      flags.inst_data_fmt.Value())
                                         : buffer.GetNumberConversion(),
         .num_components = AmdGpu::NumComponents(data_format),
     };

src/shader_recompiler/ir/passes/resource_tracking_pass.cpp

@@ -101,6 +101,8 @@ bool IsImageAtomicInstruction(const IR::Inst& inst) {
     case IR::Opcode::ImageAtomicUMin32:
     case IR::Opcode::ImageAtomicSMax32:
     case IR::Opcode::ImageAtomicUMax32:
+    case IR::Opcode::ImageAtomicFMax32:
+    case IR::Opcode::ImageAtomicFMin32:
     case IR::Opcode::ImageAtomicInc32:
     case IR::Opcode::ImageAtomicDec32:
     case IR::Opcode::ImageAtomicAnd32:
@@ -361,6 +363,12 @@ void PatchImageSharp(IR::Block& block, IR::Inst& inst, Info& info, Descriptors&
         LOG_ERROR(Render_Vulkan, "Shader compiled with unbound image!");
         image = AmdGpu::Image::Null();
     }
+    const auto data_fmt = image.GetDataFmt();
+    if (inst_info.is_depth && data_fmt != AmdGpu::DataFormat::Format16 &&
+        data_fmt != AmdGpu::DataFormat::Format32) {
+        LOG_ERROR(Render_Vulkan, "Shader compiled using non-depth image with depth instruction!");
+        image = AmdGpu::Image::NullDepth();
+    }
     ASSERT(image.GetType() != AmdGpu::ImageType::Invalid);
     const bool is_written = inst.GetOpcode() == IR::Opcode::ImageWrite;
 

src/shader_recompiler/ir/passes/shader_info_collection_pass.cpp

@@ -71,6 +71,10 @@ void Visit(Info& info, const IR::Inst& inst) {
     case IR::Opcode::ImageQueryLod:
         info.has_image_query = true;
         break;
+    case IR::Opcode::ImageAtomicFMax32:
+    case IR::Opcode::ImageAtomicFMin32:
+        info.uses_atomic_float_min_max = true;
+        break;
     case IR::Opcode::LaneId:
         info.uses_lane_id = true;
         break;

src/shader_recompiler/ir/reinterpret.h

@@ -34,6 +34,18 @@ inline F32 ApplyReadNumberConversion(IREmitter& ir, const F32& value,
     case AmdGpu::NumberConversion::UnormToUbnorm:
         // Convert 0...1 to -1...1
         return ir.FPSub(ir.FPMul(value, ir.Imm32(2.f)), ir.Imm32(1.f));
+    case AmdGpu::NumberConversion::Sint8ToSnormNz: {
+        const IR::U32 additon = ir.IAdd(ir.IMul(ir.BitCast<U32>(value), ir.Imm32(2)), ir.Imm32(1));
+        const IR::F32 left = ir.ConvertSToF(32, 32, additon);
+        const IR::F32 max = ir.Imm32(float(std::numeric_limits<u8>::max()));
+        return ir.FPDiv(left, max);
+    }
+    case AmdGpu::NumberConversion::Sint16ToSnormNz: {
+        const IR::U32 additon = ir.IAdd(ir.IMul(ir.BitCast<U32>(value), ir.Imm32(2)), ir.Imm32(1));
+        const IR::F32 left = ir.ConvertSToF(32, 32, additon);
+        const IR::F32 max = ir.Imm32(float(std::numeric_limits<u16>::max()));
+        return ir.FPDiv(left, max);
+    }
     default:
         UNREACHABLE();
     }
@@ -66,6 +78,20 @@ inline F32 ApplyWriteNumberConversion(IREmitter& ir, const F32& value,
     case AmdGpu::NumberConversion::UnormToUbnorm:
         // Convert -1...1 to 0...1
         return ir.FPDiv(ir.FPAdd(value, ir.Imm32(1.f)), ir.Imm32(2.f));
+    case AmdGpu::NumberConversion::Sint8ToSnormNz: {
+        const IR::F32 max = ir.Imm32(float(std::numeric_limits<u8>::max()));
+        const IR::F32 mul = ir.FPMul(ir.FPClamp(value, ir.Imm32(-1.f), ir.Imm32(1.f)), max);
+        const IR::F32 left = ir.FPSub(mul, ir.Imm32(1.f));
+        const IR::U32 raw = ir.ConvertFToS(32, ir.FPDiv(left, ir.Imm32(2.f)));
+        return ir.BitCast<F32>(raw);
+    }
+    case AmdGpu::NumberConversion::Sint16ToSnormNz: {
+        const IR::F32 max = ir.Imm32(float(std::numeric_limits<u16>::max()));
+        const IR::F32 mul = ir.FPMul(ir.FPClamp(value, ir.Imm32(-1.f), ir.Imm32(1.f)), max);
+        const IR::F32 left = ir.FPSub(mul, ir.Imm32(1.f));
+        const IR::U32 raw = ir.ConvertFToS(32, ir.FPDiv(left, ir.Imm32(2.f)));
+        return ir.BitCast<F32>(raw);
+    }
     default:
         UNREACHABLE();
     }

src/shader_recompiler/profile.h

@@ -29,6 +29,7 @@ struct Profile {
     bool supports_native_cube_calc{};
     bool supports_trinary_minmax{};
     bool supports_robust_buffer_access{};
+    bool supports_image_fp32_atomic_min_max{};
     bool has_broken_spirv_clamp{};
     bool lower_left_origin_mode{};
     bool needs_manual_interpolation{};

src/shader_recompiler/runtime_info.h

@@ -169,10 +169,10 @@ static constexpr u32 MaxColorBuffers = 8;
 
 struct PsColorBuffer {
     AmdGpu::NumberFormat num_format : 4;
-    AmdGpu::NumberConversion num_conversion : 2;
+    AmdGpu::NumberConversion num_conversion : 3;
     AmdGpu::Liverpool::ShaderExportFormat export_format : 4;
     u32 needs_unorm_fixup : 1;
-    u32 pad : 21;
+    u32 pad : 20;
     AmdGpu::CompMapping swizzle;
 
     auto operator<=>(const PsColorBuffer&) const noexcept = default;

src/video_core/amdgpu/liverpool.cpp

@@ -455,14 +455,14 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span<const u32> dcb, std::span<c
             case PM4ItOpcode::DrawIndirect: {
                 const auto* draw_indirect = reinterpret_cast<const PM4CmdDrawIndirect*>(header);
                 const auto offset = draw_indirect->data_offset;
-                const auto size = sizeof(DrawIndirectArgs);
+                const auto stride = sizeof(DrawIndirectArgs);
                 if (DebugState.DumpingCurrentReg()) {
                     DebugState.PushRegsDump(base_addr, reinterpret_cast<uintptr_t>(header), regs);
                 }
                 if (rasterizer) {
                     const auto cmd_address = reinterpret_cast<const void*>(header);
                     rasterizer->ScopeMarkerBegin(fmt::format("gfx:{}:DrawIndirect", cmd_address));
-                    rasterizer->DrawIndirect(false, indirect_args_addr, offset, size, 1, 0);
+                    rasterizer->DrawIndirect(false, indirect_args_addr, offset, stride, 1, 0);
                     rasterizer->ScopeMarkerEnd();
                 }
                 break;
@@ -471,7 +471,7 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span<const u32> dcb, std::span<c
                 const auto* draw_index_indirect =
                     reinterpret_cast<const PM4CmdDrawIndexIndirect*>(header);
                 const auto offset = draw_index_indirect->data_offset;
-                const auto size = sizeof(DrawIndexedIndirectArgs);
+                const auto stride = sizeof(DrawIndexedIndirectArgs);
                 if (DebugState.DumpingCurrentReg()) {
                     DebugState.PushRegsDump(base_addr, reinterpret_cast<uintptr_t>(header), regs);
                 }
@@ -479,25 +479,46 @@ 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("gfx:{}:DrawIndexIndirect", cmd_address));
-                    rasterizer->DrawIndirect(true, indirect_args_addr, offset, size, 1, 0);
+                    rasterizer->DrawIndirect(true, indirect_args_addr, offset, stride, 1, 0);
                     rasterizer->ScopeMarkerEnd();
                 }
                 break;
             }
-            case PM4ItOpcode::DrawIndexIndirectCountMulti: {
+            case PM4ItOpcode::DrawIndexIndirectMulti: {
                 const auto* draw_index_indirect =
                     reinterpret_cast<const PM4CmdDrawIndexIndirectMulti*>(header);
                 const auto offset = draw_index_indirect->data_offset;
                 if (DebugState.DumpingCurrentReg()) {
                     DebugState.PushRegsDump(base_addr, reinterpret_cast<uintptr_t>(header), regs);
                 }
+                if (rasterizer) {
+                    const auto cmd_address = reinterpret_cast<const void*>(header);
+                    rasterizer->ScopeMarkerBegin(
+                        fmt::format("gfx:{}:DrawIndexIndirectMulti", cmd_address));
+                    rasterizer->DrawIndirect(true, indirect_args_addr, offset,
+                                             draw_index_indirect->stride,
+                                             draw_index_indirect->count, 0);
+                    rasterizer->ScopeMarkerEnd();
+                }
+                break;
+            }
+            case PM4ItOpcode::DrawIndexIndirectCountMulti: {
+                const auto* draw_index_indirect =
+                    reinterpret_cast<const PM4CmdDrawIndexIndirectCountMulti*>(header);
+                const auto offset = draw_index_indirect->data_offset;
+                if (DebugState.DumpingCurrentReg()) {
+                    DebugState.PushRegsDump(base_addr, reinterpret_cast<uintptr_t>(header), regs);
+                }
                 if (rasterizer) {
                     const auto cmd_address = reinterpret_cast<const void*>(header);
                     rasterizer->ScopeMarkerBegin(
                         fmt::format("gfx:{}:DrawIndexIndirectCountMulti", cmd_address));
-                    rasterizer->DrawIndirect(
-                        true, indirect_args_addr, offset, draw_index_indirect->stride,
-                        draw_index_indirect->count, draw_index_indirect->countAddr);
+                    rasterizer->DrawIndirect(true, indirect_args_addr, offset,
+                                             draw_index_indirect->stride,
+                                             draw_index_indirect->count,
+                                             draw_index_indirect->count_indirect_enable.Value()
+                                                 ? draw_index_indirect->count_addr
+                                                 : 0);
                     rasterizer->ScopeMarkerEnd();
                 }
                 break;

src/video_core/amdgpu/liverpool.h

@@ -924,15 +924,11 @@ struct Liverpool {
         }
 
         [[nodiscard]] NumberFormat GetNumberFmt() const {
-            // There is a small difference between T# and CB number types, account for it.
-            return RemapNumberFormat(info.number_type == NumberFormat::SnormNz
-                                         ? NumberFormat::Srgb
-                                         : info.number_type.Value(),
-                                     info.format);
+            return RemapNumberFormat(GetFixedNumberFormat(), info.format);
         }
 
         [[nodiscard]] NumberConversion GetNumberConversion() const {
-            return MapNumberConversion(info.number_type);
+            return MapNumberConversion(GetFixedNumberFormat(), info.format);
         }
 
         [[nodiscard]] CompMapping Swizzle() const {
@@ -973,6 +969,13 @@ struct Liverpool {
             const auto mrt_swizzle = mrt_swizzles[swap_idx][components_idx];
             return RemapSwizzle(info.format, mrt_swizzle);
         }
+
+    private:
+        [[nodiscard]] NumberFormat GetFixedNumberFormat() const {
+            // There is a small difference between T# and CB number types, account for it.
+            return info.number_type == NumberFormat::SnormNz ? NumberFormat::Srgb
+                                                             : info.number_type.Value();
+        }
     };
 
     enum ContextRegs : u32 {

src/video_core/amdgpu/pm4_cmds.h

@@ -860,6 +860,24 @@ struct PM4CmdDrawIndexIndirect {
 };
 
 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
+    };
+    u32 count;          ///< Count of data structures to loop through before going to next packet
+    u32 stride;         ///< Stride in memory from one data structure to the next
+    u32 draw_initiator; ///< Draw Initiator Register
+};
+
+struct PM4CmdDrawIndexIndirectCountMulti {
     PM4Type3Header header; ///< header
     u32 data_offset;       ///< Byte aligned offset where the required data structure starts
     union {
@@ -874,14 +892,14 @@ struct PM4CmdDrawIndexIndirectMulti {
     };
     union {
         u32 dw4;
-        BitField<0, 16, u32> drawIndexLoc; ///< register offset to write the Draw Index count
+        BitField<0, 16, u32> draw_index_loc; ///< register offset to write the Draw Index count
         BitField<30, 1, u32>
-            countIndirectEnable; ///< Indicates the data structure count is in memory
+            count_indirect_enable; ///< Indicates the data structure count is in memory
         BitField<31, 1, u32>
-            drawIndexEnable; ///< Enables writing of Draw Index count to DRAW_INDEX_LOC
+            draw_index_enable; ///< 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
+    u64 count_addr;     ///< DWord aligned Address[31:2]; Valid if countIndirectEnable is set
     u32 stride;         ///< Stride in memory from one data structure to the next
     u32 draw_initiator; ///< Draw Initiator Register
 };

src/video_core/amdgpu/resource.h

@@ -68,7 +68,7 @@ struct Buffer {
     }
 
     NumberConversion GetNumberConversion() const noexcept {
-        return MapNumberConversion(NumberFormat(num_format));
+        return MapNumberConversion(NumberFormat(num_format), DataFormat(data_format));
     }
 
     u32 GetStride() const noexcept {
@@ -219,6 +219,19 @@ struct Image {
         return image;
     }
 
+    static constexpr Image NullDepth() {
+        Image image{};
+        image.data_format = u64(DataFormat::Format32);
+        image.num_format = u64(NumberFormat::Float);
+        image.dst_sel_x = u64(CompSwizzle::Red);
+        image.dst_sel_y = u64(CompSwizzle::Green);
+        image.dst_sel_z = u64(CompSwizzle::Blue);
+        image.dst_sel_w = u64(CompSwizzle::Alpha);
+        image.tiling_index = u64(TilingMode::Texture_MicroTiled);
+        image.type = u64(ImageType::Color2D);
+        return image;
+    }
+
     bool Valid() const {
         return (type & 0x8u) != 0;
     }
@@ -292,7 +305,7 @@ struct Image {
     }
 
     NumberConversion GetNumberConversion() const noexcept {
-        return MapNumberConversion(NumberFormat(num_format));
+        return MapNumberConversion(NumberFormat(num_format), DataFormat(data_format));
     }
 
     TilingMode GetTilingMode() const {

src/video_core/amdgpu/types.h

@@ -197,6 +197,8 @@ enum class NumberConversion : u32 {
     UintToUscaled = 1,
     SintToSscaled = 2,
     UnormToUbnorm = 3,
+    Sint8ToSnormNz = 5,
+    Sint16ToSnormNz = 6,
 };
 
 struct CompMapping {
@@ -287,6 +289,7 @@ inline NumberFormat RemapNumberFormat(const NumberFormat format, const DataForma
     case NumberFormat::Uscaled:
         return NumberFormat::Uint;
     case NumberFormat::Sscaled:
+    case NumberFormat::SnormNz:
         return NumberFormat::Sint;
     case NumberFormat::Ubnorm:
         return NumberFormat::Unorm;
@@ -336,14 +339,28 @@ inline CompMapping RemapSwizzle(const DataFormat format, const CompMapping swizz
     }
 }
 
-inline NumberConversion MapNumberConversion(const NumberFormat format) {
-    switch (format) {
+inline NumberConversion MapNumberConversion(const NumberFormat num_fmt, const DataFormat data_fmt) {
+    switch (num_fmt) {
     case NumberFormat::Uscaled:
         return NumberConversion::UintToUscaled;
     case NumberFormat::Sscaled:
         return NumberConversion::SintToSscaled;
     case NumberFormat::Ubnorm:
         return NumberConversion::UnormToUbnorm;
+    case NumberFormat::SnormNz: {
+        switch (data_fmt) {
+        case DataFormat::Format8:
+        case DataFormat::Format8_8:
+        case DataFormat::Format8_8_8_8:
+            return NumberConversion::Sint8ToSnormNz;
+        case DataFormat::Format16:
+        case DataFormat::Format16_16:
+        case DataFormat::Format16_16_16_16:
+            return NumberConversion::Sint16ToSnormNz;
+        default:
+            UNREACHABLE_MSG("data_fmt = {}", u32(data_fmt));
+        }
+    }
     default:
         return NumberConversion::None;
     }

src/video_core/buffer_cache/buffer_cache.cpp

@@ -177,8 +177,8 @@ void BufferCache::BindVertexBuffers(const Vulkan::GraphicsPipeline& pipeline) {
     if (instance.IsVertexInputDynamicState()) {
         cmdbuf.bindVertexBuffers(0, num_buffers, host_buffers.data(), host_offsets.data());
     } else {
-        cmdbuf.bindVertexBuffers2EXT(0, num_buffers, host_buffers.data(), host_offsets.data(),
-                                     host_sizes.data(), host_strides.data());
+        cmdbuf.bindVertexBuffers2(0, num_buffers, host_buffers.data(), host_offsets.data(),
+                                  host_sizes.data(), host_strides.data());
     }
 }
 

src/video_core/renderer_vulkan/vk_instance.cpp

@@ -210,7 +210,8 @@ bool Instance::CreateDevice() {
                           vk::PhysicalDeviceRobustness2FeaturesEXT,
                           vk::PhysicalDeviceExtendedDynamicState3FeaturesEXT,
                           vk::PhysicalDevicePrimitiveTopologyListRestartFeaturesEXT,
-                          vk::PhysicalDevicePortabilitySubsetFeaturesKHR>();
+                          vk::PhysicalDevicePortabilitySubsetFeaturesKHR,
+                          vk::PhysicalDeviceShaderAtomicFloat2FeaturesEXT>();
     features = feature_chain.get().features;
 
     const vk::StructureChain properties_chain = physical_device.getProperties2<
@@ -272,6 +273,13 @@ bool Instance::CreateDevice() {
     image_load_store_lod = add_extension(VK_AMD_SHADER_IMAGE_LOAD_STORE_LOD_EXTENSION_NAME);
     amd_gcn_shader = add_extension(VK_AMD_GCN_SHADER_EXTENSION_NAME);
     amd_shader_trinary_minmax = add_extension(VK_AMD_SHADER_TRINARY_MINMAX_EXTENSION_NAME);
+    shader_atomic_float2 = add_extension(VK_EXT_SHADER_ATOMIC_FLOAT_2_EXTENSION_NAME);
+    if (shader_atomic_float2) {
+        shader_atomic_float2_features =
+            feature_chain.get<vk::PhysicalDeviceShaderAtomicFloat2FeaturesEXT>();
+        LOG_INFO(Render_Vulkan, "- shaderImageFloat32AtomicMinMax: {}",
+                 shader_atomic_float2_features.shaderImageFloat32AtomicMinMax);
+    }
     const bool calibrated_timestamps =
         TRACY_GPU_ENABLED ? add_extension(VK_EXT_CALIBRATED_TIMESTAMPS_EXTENSION_NAME) : false;
 
@@ -330,6 +338,7 @@ bool Instance::CreateDevice() {
                 .geometryShader = features.geometryShader,
                 .tessellationShader = features.tessellationShader,
                 .logicOp = features.logicOp,
+                .multiDrawIndirect = features.multiDrawIndirect,
                 .depthBiasClamp = features.depthBiasClamp,
                 .fillModeNonSolid = features.fillModeNonSolid,
                 .depthBounds = features.depthBounds,
@@ -401,6 +410,10 @@ bool Instance::CreateDevice() {
         vk::PhysicalDeviceLegacyVertexAttributesFeaturesEXT{
             .legacyVertexAttributes = true,
         },
+        vk::PhysicalDeviceShaderAtomicFloat2FeaturesEXT{
+            .shaderImageFloat32AtomicMinMax =
+                shader_atomic_float2_features.shaderImageFloat32AtomicMinMax,
+        },
 #ifdef __APPLE__
         portability_features,
 #endif
@@ -430,6 +443,9 @@ bool Instance::CreateDevice() {
     if (!legacy_vertex_attributes) {
         device_chain.unlink<vk::PhysicalDeviceLegacyVertexAttributesFeaturesEXT>();
     }
+    if (!shader_atomic_float2) {
+        device_chain.unlink<vk::PhysicalDeviceShaderAtomicFloat2FeaturesEXT>();
+    }
 
     auto [device_result, dev] = physical_device.createDeviceUnique(device_chain.get());
     if (device_result != vk::Result::eSuccess) {

src/video_core/renderer_vulkan/vk_instance.h

@@ -165,6 +165,12 @@ public:
         return amd_shader_trinary_minmax;
     }
 
+    /// Returns true when the shaderImageFloat32AtomicMinMax feature of
+    /// VK_EXT_shader_atomic_float2 is supported.
+    bool IsShaderAtomicFloatImage32MinMaxSupported() const {
+        return shader_atomic_float2 && shader_atomic_float2_features.shaderImageFloat32AtomicMinMax;
+    }
+
     /// Returns true when geometry shaders are supported by the device
     bool IsGeometryStageSupported() const {
         return features.geometryShader;
@@ -336,6 +342,7 @@ private:
     vk::PhysicalDevicePortabilitySubsetFeaturesKHR portability_features;
     vk::PhysicalDeviceExtendedDynamicState3FeaturesEXT dynamic_state_3_features;
     vk::PhysicalDeviceRobustness2FeaturesEXT robustness2_features;
+    vk::PhysicalDeviceShaderAtomicFloat2FeaturesEXT shader_atomic_float2_features;
     vk::DriverIdKHR driver_id;
     vk::UniqueDebugUtilsMessengerEXT debug_callback{};
     std::string vendor_name;
@@ -360,6 +367,7 @@ private:
     bool image_load_store_lod{};
     bool amd_gcn_shader{};
     bool amd_shader_trinary_minmax{};
+    bool shader_atomic_float2{};
     bool portability_subset{};
 };
 

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -206,6 +206,7 @@ PipelineCache::PipelineCache(const Instance& instance_, Scheduler& scheduler_,
         .supports_native_cube_calc = instance_.IsAmdGcnShaderSupported(),
         .supports_trinary_minmax = instance_.IsAmdShaderTrinaryMinMaxSupported(),
         .supports_robust_buffer_access = instance_.IsRobustBufferAccess2Supported(),
+        .supports_image_fp32_atomic_min_max = instance_.IsShaderAtomicFloatImage32MinMaxSupported(),
         .needs_manual_interpolation = instance.IsFragmentShaderBarycentricSupported() &&
                                       instance.GetDriverID() == vk::DriverId::eNvidiaProprietary,
         .needs_lds_barriers = instance.GetDriverID() == vk::DriverId::eNvidiaProprietary ||

src/video_core/renderer_vulkan/vk_presenter.cpp

@@ -270,7 +270,25 @@ Frame* Presenter::PrepareLastFrame() {
     return frame;
 }
 
-Frame* Presenter::PrepareFrameInternal(VideoCore::ImageId image_id, bool is_eop) {
+static vk::Format GetFrameViewFormat(const Libraries::VideoOut::PixelFormat format) {
+    switch (format) {
+    case Libraries::VideoOut::PixelFormat::A8B8G8R8Srgb:
+        return vk::Format::eR8G8B8A8Srgb;
+    case Libraries::VideoOut::PixelFormat::A8R8G8B8Srgb:
+        return vk::Format::eB8G8R8A8Srgb;
+    case Libraries::VideoOut::PixelFormat::A2R10G10B10:
+    case Libraries::VideoOut::PixelFormat::A2R10G10B10Srgb:
+    case Libraries::VideoOut::PixelFormat::A2R10G10B10Bt2020Pq:
+        return vk::Format::eA2R10G10B10UnormPack32;
+    default:
+        break;
+    }
+    UNREACHABLE_MSG("Unknown format={}", static_cast<u32>(format));
+    return {};
+}
+
+Frame* Presenter::PrepareFrameInternal(VideoCore::ImageId image_id,
+                                       const Libraries::VideoOut::PixelFormat format, bool is_eop) {
     // Request a free presentation frame.
     Frame* frame = GetRenderFrame();
 
@@ -324,7 +342,7 @@ Frame* Presenter::PrepareFrameInternal(VideoCore::ImageId image_id, bool is_eop)
                       cmdbuf);
 
         VideoCore::ImageViewInfo info{};
-        info.format = image.info.pixel_format;
+        info.format = GetFrameViewFormat(format);
         // Exclude alpha from output frame to avoid blending with UI.
         info.mapping = vk::ComponentMapping{
             .r = vk::ComponentSwizzle::eIdentity,

src/video_core/renderer_vulkan/vk_presenter.h

@@ -70,11 +70,12 @@ public:
         auto desc = VideoCore::TextureCache::VideoOutDesc{attribute, cpu_address};
         const auto image_id = texture_cache.FindImage(desc);
         texture_cache.UpdateImage(image_id, is_eop ? nullptr : &flip_scheduler);
-        return PrepareFrameInternal(image_id, is_eop);
+        return PrepareFrameInternal(image_id, attribute.attrib.pixel_format, is_eop);
     }
 
     Frame* PrepareBlankFrame(bool is_eop) {
-        return PrepareFrameInternal(VideoCore::NULL_IMAGE_ID, is_eop);
+        return PrepareFrameInternal(VideoCore::NULL_IMAGE_ID,
+                                    Libraries::VideoOut::PixelFormat::Unknown, is_eop);
     }
 
     VideoCore::Image& RegisterVideoOutSurface(
@@ -119,7 +120,8 @@ public:
     }
 
 private:
-    Frame* PrepareFrameInternal(VideoCore::ImageId image_id, bool is_eop = true);
+    Frame* PrepareFrameInternal(VideoCore::ImageId image_id,
+                                Libraries::VideoOut::PixelFormat format, bool is_eop = true);
     Frame* GetRenderFrame();
 
     void SetExpectedGameSize(s32 width, s32 height);

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -618,8 +618,9 @@ void Rasterizer::BindTextures(const Shader::Info& stage, Shader::Backend::Bindin
             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,
+                auto& null_image_view =
+                    texture_cache.FindTexture(VideoCore::NULL_IMAGE_ID, desc.view_info);
+                image_infos.emplace_back(VK_NULL_HANDLE, *null_image_view.image_view,
                                          vk::ImageLayout::eGeneral);
             }
         } else {

src/video_core/texture_cache/image_info.cpp

@@ -16,14 +16,15 @@ using VideoOutFormat = Libraries::VideoOut::PixelFormat;
 
 static vk::Format ConvertPixelFormat(const VideoOutFormat format) {
     switch (format) {
-    case VideoOutFormat::A8R8G8B8Srgb:
-        return vk::Format::eB8G8R8A8Srgb;
     case VideoOutFormat::A8B8G8R8Srgb:
+    // Remaining formats are mapped to RGBA for internal consistency and changed to BGRA in the
+    // frame image view.
+    case VideoOutFormat::A8R8G8B8Srgb:
         return vk::Format::eR8G8B8A8Srgb;
     case VideoOutFormat::A2R10G10B10:
     case VideoOutFormat::A2R10G10B10Srgb:
     case VideoOutFormat::A2R10G10B10Bt2020Pq:
-        return vk::Format::eA2R10G10B10UnormPack32;
+        return vk::Format::eA2B10G10R10UnormPack32;
     default:
         break;
     }

src/video_core/texture_cache/image_view.h

@@ -34,8 +34,6 @@ struct ImageViewInfo {
 
 struct Image;
 
-constexpr Common::SlotId NULL_IMAGE_VIEW_ID{0};
-
 struct ImageView {
     ImageView(const Vulkan::Instance& instance, const ImageViewInfo& info, Image& image,
               ImageId image_id);

src/video_core/texture_cache/texture_cache.cpp

@@ -8,6 +8,7 @@
 #include "common/debug.h"
 #include "video_core/buffer_cache/buffer_cache.h"
 #include "video_core/page_manager.h"
+#include "video_core/renderer_vulkan/liverpool_to_vk.h"
 #include "video_core/renderer_vulkan/vk_instance.h"
 #include "video_core/renderer_vulkan/vk_scheduler.h"
 #include "video_core/texture_cache/host_compatibility.h"
@@ -23,31 +24,41 @@ TextureCache::TextureCache(const Vulkan::Instance& instance_, Vulkan::Scheduler&
                            BufferCache& buffer_cache_, PageManager& tracker_)
     : instance{instance_}, scheduler{scheduler_}, buffer_cache{buffer_cache_}, tracker{tracker_},
       tile_manager{instance, scheduler} {
+    // Create basic null image at fixed image ID.
+    const auto null_id = GetNullImage(vk::Format::eR8G8B8A8Unorm);
+    ASSERT(null_id.index == NULL_IMAGE_ID.index);
+}
+
+TextureCache::~TextureCache() = default;
+
+ImageId TextureCache::GetNullImage(const vk::Format format) {
+    const auto existing_image = null_images.find(format);
+    if (existing_image != null_images.end()) {
+        return existing_image->second;
+    }
+
     ImageInfo info{};
-    info.pixel_format = vk::Format::eR8G8B8A8Unorm;
+    info.pixel_format = format;
     info.type = vk::ImageType::e2D;
-    info.tiling_idx = u32(AmdGpu::TilingMode::Texture_MicroTiled);
+    info.tiling_idx = static_cast<u32>(AmdGpu::TilingMode::Texture_MicroTiled);
     info.num_bits = 32;
     info.UpdateSize();
+
     const ImageId null_id = slot_images.insert(instance, scheduler, info);
-    ASSERT(null_id.index == NULL_IMAGE_ID.index);
     auto& img = slot_images[null_id];
+
     const vk::Image& null_image = img.image;
-    Vulkan::SetObjectName(instance.GetDevice(), null_image, "Null Image");
+    Vulkan::SetObjectName(instance.GetDevice(), null_image,
+                          fmt::format("Null Image ({})", vk::to_string(format)));
+
     img.flags = ImageFlagBits::Empty;
     img.track_addr = img.info.guest_address;
     img.track_addr_end = img.info.guest_address + img.info.guest_size;
 
-    ImageViewInfo view_info;
-    const auto null_view_id =
-        slot_image_views.insert(instance, view_info, slot_images[null_id], null_id);
-    ASSERT(null_view_id.index == NULL_IMAGE_VIEW_ID.index);
-    const vk::ImageView& null_image_view = slot_image_views[null_view_id].image_view.get();
-    Vulkan::SetObjectName(instance.GetDevice(), null_image_view, "Null Image View");
+    null_images.emplace(format, null_id);
+    return null_id;
 }
 
-TextureCache::~TextureCache() = default;
-
 void TextureCache::MarkAsMaybeDirty(ImageId image_id, Image& image) {
     if (image.hash == 0) {
         // Initialize hash
@@ -296,7 +307,7 @@ ImageId TextureCache::FindImage(BaseDesc& desc, FindFlags flags) {
     const auto& info = desc.info;
 
     if (info.guest_address == 0) [[unlikely]] {
-        return NULL_IMAGE_ID;
+        return GetNullImage(info.pixel_format);
     }
 
     std::scoped_lock lock{mutex};

src/video_core/texture_cache/texture_cache.h

@@ -246,6 +246,9 @@ private:
         }
     }
 
+    /// Gets or creates a null image for a particular format.
+    ImageId GetNullImage(vk::Format format);
+
     /// Create an image from the given parameters
     [[nodiscard]] ImageId InsertImage(const ImageInfo& info, VAddr cpu_addr);
 
@@ -285,6 +288,7 @@ private:
     Common::SlotVector<Image> slot_images;
     Common::SlotVector<ImageView> slot_image_views;
     tsl::robin_map<u64, Sampler> samplers;
+    tsl::robin_map<vk::Format, ImageId> null_images;
     PageTable page_table;
     std::mutex mutex;