Merge pull request #206 from shadps4-emu/shader-again

Add more shader instructions and some memory functions

src/core/libraries/kernel/event_flag/event_flag.cpp

@@ -73,7 +73,7 @@ int PS4_SYSV_ABI sceKernelCloseEventFlag() {
     return ORBIS_OK;
 }
 int PS4_SYSV_ABI sceKernelClearEventFlag(OrbisKernelEventFlag ef, u64 bitPattern) {
-    LOG_ERROR(Kernel_Event, "called");
+    LOG_INFO(Kernel_Event, "called");
     ef->Clear(bitPattern);
     return ORBIS_OK;
 }
@@ -177,10 +177,10 @@ int PS4_SYSV_ABI sceKernelWaitEventFlag(OrbisKernelEventFlag ef, u64 bitPattern,
         UNREACHABLE();
     }
 
-    auto result = ef->Wait(bitPattern, wait, clear, pResultPat, pTimeout);
+    u32 result = ef->Wait(bitPattern, wait, clear, pResultPat, pTimeout);
 
     if (result != ORBIS_OK) {
-        LOG_ERROR(Kernel_Event, "returned {}", result);
+        LOG_ERROR(Kernel_Event, "returned {:#x}", result);
     }
 
     return result;

src/core/libraries/kernel/libkernel.cpp

@@ -177,7 +177,7 @@ s32 PS4_SYSV_ABI sceKernelLoadStartModule(const char* moduleFileName, size_t arg
 
     // Load PRX module and relocate any modules that import it.
     auto* linker = Common::Singleton<Core::Linker>::Instance();
-    u32 handle = linker->LoadModule(path);
+    u32 handle = linker->LoadModule(path, true);
     if (handle == -1) {
         return ORBIS_KERNEL_ERROR_EINVAL;
     }
@@ -298,6 +298,8 @@ void LibKernel_Register(Core::Loader::SymbolsResolver* sym) {
                  sceKernelAllocateMainDirectMemory);
     LIB_FUNCTION("C0f7TJcbfac", "libkernel", 1, "libkernel", 1, 1,
                  sceKernelAvailableDirectMemorySize);
+    LIB_FUNCTION("hwVSPCmp5tM", "libkernel", 1, "libkernel", 1, 1,
+                 sceKernelCheckedReleaseDirectMemory);
     LIB_FUNCTION("rVjRvHJ0X6c", "libkernel", 1, "libkernel", 1, 1, sceKernelVirtualQuery);
     LIB_FUNCTION("pO96TwzOm5E", "libkernel", 1, "libkernel", 1, 1, sceKernelGetDirectMemorySize);
     LIB_FUNCTION("NcaWUxfMNIQ", "libkernel", 1, "libkernel", 1, 1, sceKernelMapNamedDirectMemory);
@@ -307,6 +309,8 @@ void LibKernel_Register(Core::Loader::SymbolsResolver* sym) {
     LIB_FUNCTION("MBuItvba6z8", "libkernel", 1, "libkernel", 1, 1, sceKernelReleaseDirectMemory);
     LIB_FUNCTION("cQke9UuBQOk", "libkernel", 1, "libkernel", 1, 1, sceKernelMunmap);
     LIB_FUNCTION("mL8NDH86iQI", "libkernel", 1, "libkernel", 1, 1, sceKernelMapNamedFlexibleMemory);
+    LIB_FUNCTION("aNz11fnnzi4", "libkernel", 1, "libkernel", 1, 1,
+                 sceKernelAvailableFlexibleMemorySize);
     LIB_FUNCTION("IWIBBdTHit4", "libkernel", 1, "libkernel", 1, 1, sceKernelMapFlexibleMemory);
     LIB_FUNCTION("p5EcQeEeJAE", "libkernel", 1, "libkernel", 1, 1,
                  _sceKernelRtldSetApplicationHeapAPI);

src/core/libraries/kernel/memory_management.cpp

@@ -173,6 +173,13 @@ int PS4_SYSV_ABI sceKernelDirectMemoryQuery(u64 offset, int flags, OrbisQueryInf
     return memory->DirectMemoryQuery(offset, flags == 1, query_info);
 }
 
+s32 PS4_SYSV_ABI sceKernelAvailableFlexibleMemorySize(size_t* out_size) {
+    auto* memory = Core::Memory::Instance();
+    *out_size = memory->GetAvailableFlexibleSize();
+    LOG_INFO(Kernel_Vmm, "called size = {:#x}", *out_size);
+    return ORBIS_OK;
+}
+
 void PS4_SYSV_ABI _sceKernelRtldSetApplicationHeapAPI(void* func) {
     auto* linker = Common::Singleton<Core::Linker>::Instance();
     linker->SetHeapApiFunc(func);

src/core/libraries/kernel/memory_management.h

@@ -78,6 +78,7 @@ int PS4_SYSV_ABI sceKernelQueryMemoryProtection(void* addr, void** start, void**
 
 int PS4_SYSV_ABI sceKernelDirectMemoryQuery(u64 offset, int flags, OrbisQueryInfo* query_info,
                                             size_t infoSize);
+s32 PS4_SYSV_ABI sceKernelAvailableFlexibleMemorySize(size_t* sizeOut);
 void PS4_SYSV_ABI _sceKernelRtldSetApplicationHeapAPI(void* func);
 
 } // namespace Libraries::Kernel

src/core/libraries/kernel/thread_management.cpp

@@ -1276,6 +1276,10 @@ int PS4_SYSV_ABI scePthreadOnce(int* once_control, void (*init_routine)(void)) {
     return pthread_once(reinterpret_cast<pthread_once_t*>(once_control), init_routine);
 }
 
+[[noreturn]] void PS4_SYSV_ABI scePthreadExit(void* value_ptr) {
+    pthread_exit(value_ptr);
+}
+
 void pthreadSymbolsRegister(Core::Loader::SymbolsResolver* sym) {
     LIB_FUNCTION("lZzFeSxPl08", "libScePosix", 1, "libkernel", 1, 1, posix_pthread_setcancelstate);
     LIB_FUNCTION("0TyVk4MSLt0", "libScePosix", 1, "libkernel", 1, 1, posix_pthread_cond_init);
@@ -1293,6 +1297,7 @@ void pthreadSymbolsRegister(Core::Loader::SymbolsResolver* sym) {
     LIB_FUNCTION("onNY9Byn-W8", "libkernel", 1, "libkernel", 1, 1, scePthreadJoin);
     LIB_FUNCTION("4qGrR6eoP9Y", "libkernel", 1, "libkernel", 1, 1, scePthreadDetach);
     LIB_FUNCTION("3PtV6p3QNX4", "libkernel", 1, "libkernel", 1, 1, scePthreadEqual);
+    LIB_FUNCTION("3kg7rT0NQIs", "libkernel", 1, "libkernel", 1, 1, scePthreadExit);
     LIB_FUNCTION("7Xl257M4VNI", "libScePosix", 1, "libkernel", 1, 1, posix_pthread_equal);
     LIB_FUNCTION("h9CcP3J0oVM", "libScePosix", 1, "libkernel", 1, 1, posix_pthread_join);
 

src/core/libraries/kernel/time_management.cpp

@@ -56,10 +56,20 @@ int PS4_SYSV_ABI sceKernelUsleep(u32 microseconds) {
     } else {
         std::this_thread::sleep_for(std::chrono::microseconds(microseconds));
     }
-#else
-    usleep(microseconds);
-#endif
     return 0;
+#else
+    timespec start;
+    timespec remain;
+    start.tv_sec = microseconds / 1000000;
+    start.tv_nsec = (microseconds % 1000000) * 1000;
+    timespec* requested = &start;
+    int ret = 0;
+    do {
+        ret = nanosleep(requested, &remain);
+        requested = &remain;
+    } while (ret != 0);
+    return ret;
+#endif
 }
 
 int PS4_SYSV_ABI posix_usleep(u32 microseconds) {

src/core/linker.cpp

@@ -13,6 +13,7 @@
 #include "core/libraries/kernel/memory_management.h"
 #include "core/libraries/kernel/thread_management.h"
 #include "core/linker.h"
+#include "core/memory.h"
 #include "core/tls.h"
 #include "core/virtual_memory.h"
 
@@ -46,7 +47,7 @@ static void RunMainEntry(VAddr addr, EntryParams* params, ExitFunc exit_func) {
                  : "rax", "rsi", "rdi");
 }
 
-Linker::Linker() = default;
+Linker::Linker() : memory{Memory::Instance()} {}
 
 Linker::~Linker() = default;
 
@@ -66,6 +67,11 @@ void Linker::Execute() {
         Relocate(m.get());
     }
 
+    // Configure used flexible memory size.
+    if (u64* flexible_size = GetProcParam()->mem_param->flexible_memory_size) {
+        memory->SetTotalFlexibleSize(*flexible_size);
+    }
+
     // Init primary thread.
     Common::SetCurrentThreadName("GAME_MainThread");
     Libraries::Kernel::pthreadInitSelfMainThread();
@@ -90,7 +96,7 @@ void Linker::Execute() {
     }
 }
 
-s32 Linker::LoadModule(const std::filesystem::path& elf_name) {
+s32 Linker::LoadModule(const std::filesystem::path& elf_name, bool is_dynamic) {
     std::scoped_lock lk{mutex};
 
     if (!std::filesystem::exists(elf_name)) {
@@ -98,12 +104,13 @@ s32 Linker::LoadModule(const std::filesystem::path& elf_name) {
         return -1;
     }
 
-    auto module = std::make_unique<Module>(elf_name, max_tls_index);
+    auto module = std::make_unique<Module>(memory, elf_name, max_tls_index);
     if (!module->IsValid()) {
         LOG_ERROR(Core_Linker, "Provided file {} is not valid ELF file", elf_name.string());
         return -1;
     }
 
+    num_static_modules += !is_dynamic;
     m_modules.emplace_back(std::move(module));
     return m_modules.size() - 1;
 }
@@ -143,11 +150,13 @@ void Linker::Relocate(Module* module) {
         case R_X86_64_RELATIVE:
             rel_value = rel_base_virtual_addr + addend;
             rel_is_resolved = true;
+            module->SetRelaBit(bit_idx);
             break;
         case R_X86_64_DTPMOD64:
             rel_value = static_cast<u64>(module->tls.modid);
             rel_is_resolved = true;
             rel_sym_type = Loader::SymbolType::Tls;
+            module->SetRelaBit(bit_idx);
             break;
         case R_X86_64_GLOB_DAT:
         case R_X86_64_JUMP_SLOT:
@@ -343,7 +352,8 @@ void Linker::InitTlsForThread(bool is_primary) {
     dtv_table[1].counter = num_dtvs;
 
     // Copy init images to TLS thread blocks and map them to DTV slots.
-    for (const auto& module : m_modules) {
+    for (u32 i = 0; i < num_static_modules; i++) {
+        auto* module = m_modules[i].get();
         if (module->tls.image_size == 0) {
             continue;
         }

src/core/linker.h

@@ -12,6 +12,33 @@ namespace Core {
 
 struct DynamicModuleInfo;
 class Linker;
+class MemoryManager;
+
+struct OrbisKernelMemParam {
+    u64 size;
+    u64* extended_page_table;
+    u64* flexible_memory_size;
+    u8* extended_memory_1;
+    u64* extended_gpu_page_table;
+    u8* extended_memory_2;
+    u64* exnteded_cpu_page_table;
+};
+
+struct OrbisProcParam {
+    u64 size;
+    u32 magic;
+    u32 entry_count;
+    u64 sdk_version;
+    char* process_name;
+    char* main_thread_name;
+    u32* main_thread_prio;
+    u32* main_thread_stack_size;
+    void* libc_param;
+    OrbisKernelMemParam* mem_param;
+    void* fs_param;
+    u32* process_preload_enable;
+    u64 unknown1;
+};
 
 struct EntryParams {
     int argc;
@@ -30,8 +57,8 @@ public:
         return m_hle_symbols;
     }
 
-    VAddr GetProcParam() const {
-        return m_modules[0]->GetProcParam();
+    OrbisProcParam* GetProcParam() const {
+        return m_modules[0]->GetProcParam<OrbisProcParam*>();
     }
 
     Module* GetModule(s32 index) const {
@@ -59,7 +86,7 @@ public:
     void* TlsGetAddr(u64 module_index, u64 offset);
     void InitTlsForThread(bool is_primary = false);
 
-    s32 LoadModule(const std::filesystem::path& elf_name);
+    s32 LoadModule(const std::filesystem::path& elf_name, bool is_dynamic = false);
     Module* FindByAddress(VAddr address);
 
     void Relocate(Module* module);
@@ -71,10 +98,12 @@ public:
 private:
     const Module* FindExportedModule(const ModuleInfo& m, const LibraryInfo& l);
 
+    MemoryManager* memory;
     std::mutex mutex;
     u32 dtv_generation_counter{1};
     size_t static_tls_size{};
     u32 max_tls_index{};
+    u32 num_static_modules{};
     HeapApiFunc heap_api_func{};
     std::vector<std::unique_ptr<Module>> m_modules;
     Loader::SymbolsResolver m_hle_symbols{};

src/core/memory.cpp

@@ -84,7 +84,7 @@ int MemoryManager::MapMemory(void** out_addr, VAddr virtual_addr, size_t size, M
                              MemoryMapFlags flags, VMAType type, std::string_view name,
                              bool is_exec, PAddr phys_addr, u64 alignment) {
     std::scoped_lock lk{mutex};
-    if (type == VMAType::Flexible && total_flexible_usage + size > 448_MB) {
+    if (type == VMAType::Flexible && flexible_usage + size > total_flexible_size) {
         return SCE_KERNEL_ERROR_ENOMEM;
     }
 
@@ -106,7 +106,7 @@ int MemoryManager::MapMemory(void** out_addr, VAddr virtual_addr, size_t size, M
             MapVulkanMemory(mapped_addr, size);
         }
         if (type == VMAType::Flexible) {
-            total_flexible_usage += size;
+            flexible_usage += size;
         }
     };
 
@@ -184,7 +184,7 @@ void MemoryManager::UnmapMemory(VAddr virtual_addr, size_t size) {
         UnmapVulkanMemory(virtual_addr, size);
     }
     if (type == VMAType::Flexible) {
-        total_flexible_usage -= size;
+        flexible_usage -= size;
     }
 
     // Mark region as free and attempt to coalesce it with neighbours.

src/core/memory.h

@@ -124,6 +124,14 @@ public:
         instance = instance_;
     }
 
+    void SetTotalFlexibleSize(u64 size) {
+        total_flexible_size = size;
+    }
+
+    u64 GetAvailableFlexibleSize() const {
+        return total_flexible_size - flexible_usage;
+    }
+
     PAddr Allocate(PAddr search_start, PAddr search_end, size_t size, u64 alignment,
                    int memory_type);
 
@@ -195,7 +203,8 @@ private:
     DMemMap dmem_map;
     VMAMap vma_map;
     std::recursive_mutex mutex;
-    size_t total_flexible_usage{};
+    size_t total_flexible_size = 448_MB;
+    size_t flexible_usage{};
 
     struct MappedMemory {
         vk::UniqueBuffer buffer;

src/core/module.cpp

@@ -55,8 +55,8 @@ static std::string EncodeId(u64 nVal) {
     return enc;
 }
 
-Module::Module(const std::filesystem::path& file_, u32& max_tls_index)
-    : file{file_}, name{file.stem().string()} {
+Module::Module(Core::MemoryManager* memory_, const std::filesystem::path& file_, u32& max_tls_index)
+    : memory{memory_}, file{file_}, name{file.stem().string()} {
     elf.Open(file);
     if (elf.IsElfFile()) {
         LoadModuleToMemory(max_tls_index);
@@ -84,7 +84,6 @@ void Module::LoadModuleToMemory(u32& max_tls_index) {
     aligned_base_size = Common::AlignUp(base_size, BlockAlign);
 
     // Map module segments (and possible TLS trampolines)
-    auto* memory = Core::Memory::Instance();
     void** out_addr = reinterpret_cast<void**>(&base_virtual_addr);
     memory->MapMemory(out_addr, LoadAddress, aligned_base_size + TrampolineSize,
                       MemoryProt::CpuReadWrite, MemoryMapFlags::Fixed, VMAType::Code, name, true);

src/core/module.h

@@ -137,10 +137,12 @@ struct DynamicModuleInfo {
 };
 
 using ModuleFunc = int (*)(size_t, const void*);
+class MemoryManager;
 
 class Module {
 public:
-    explicit Module(const std::filesystem::path& file, u32& max_tls_index);
+    explicit Module(Core::MemoryManager* memory, const std::filesystem::path& file,
+                    u32& max_tls_index);
     ~Module();
 
     VAddr GetBaseAddress() const noexcept {
@@ -220,6 +222,7 @@ public:
     const LibraryInfo* FindLibrary(std::string_view id);
 
 public:
+    Core::MemoryManager* memory;
     std::filesystem::path file;
     std::string name;
     Loader::Elf elf;

src/shader_recompiler/backend/spirv/emit_spirv_image.cpp

@@ -54,7 +54,11 @@ Id EmitImageFetch(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id of
                   Id ms) {
     const auto& texture = ctx.images[handle & 0xFFFF];
     const Id image = ctx.OpLoad(texture.image_type, texture.id);
-    return ctx.OpImageFetch(ctx.F32[4], image, coords, spv::ImageOperandsMask::Lod, lod);
+    if (Sirit::ValidId(lod)) {
+        return ctx.OpImageFetch(ctx.F32[4], image, coords, spv::ImageOperandsMask::Lod, lod);
+    } else {
+        return ctx.OpImageFetch(ctx.F32[4], image, coords);
+    }
 }
 
 Id EmitImageQueryDimensions(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id lod,

src/shader_recompiler/frontend/translate/scalar_alu.cpp

@@ -216,6 +216,14 @@ void Translator::S_AND_B32(const GcnInst& inst) {
     ir.SetScc(ir.INotEqual(result, ir.Imm32(0)));
 }
 
+void Translator::S_OR_B32(const GcnInst& inst) {
+    const IR::U32 src0{GetSrc(inst.src[0])};
+    const IR::U32 src1{GetSrc(inst.src[1])};
+    const IR::U32 result{ir.BitwiseOr(src0, src1)};
+    SetDst(inst.dst[0], result);
+    ir.SetScc(ir.INotEqual(result, ir.Imm32(0)));
+}
+
 void Translator::S_LSHR_B32(const GcnInst& inst) {
     const IR::U32 src0{GetSrc(inst.src[0])};
     const IR::U32 src1{GetSrc(inst.src[1])};
@@ -285,4 +293,36 @@ void Translator::S_BFM_B32(const GcnInst& inst) {
     SetDst(inst.dst[0], ir.ShiftLeftLogical(mask, src1));
 }
 
+void Translator::S_NOT_B64(const GcnInst& inst) {
+    const auto get_src = [&](const InstOperand& operand) {
+        switch (operand.field) {
+        case OperandField::VccLo:
+            return ir.GetVcc();
+        case OperandField::ExecLo:
+            return ir.GetExec();
+        case OperandField::ScalarGPR:
+            return ir.GetThreadBitScalarReg(IR::ScalarReg(operand.code));
+        default:
+            UNREACHABLE();
+        }
+    };
+    const IR::U1 src0{get_src(inst.src[0])};
+    const IR::U1 result = ir.LogicalNot(src0);
+    ir.SetScc(result);
+    switch (inst.dst[0].field) {
+    case OperandField::VccLo:
+        ir.SetVcc(result);
+        break;
+    case OperandField::ScalarGPR:
+        ir.SetThreadBitScalarReg(IR::ScalarReg(inst.dst[0].code), result);
+        break;
+    default:
+        UNREACHABLE();
+    }
+}
+
+void Translator::S_BREV_B32(const GcnInst& inst) {
+    SetDst(inst.dst[0], ir.BitReverse(GetSrc(inst.src[0])));
+}
+
 } // namespace Shader::Gcn

src/shader_recompiler/frontend/translate/translate.cpp

@@ -324,7 +324,10 @@ void Translate(IR::Block* block, std::span<const GcnInst> inst_list, Info& info)
             translator.IMAGE_STORE(inst);
             break;
         case Opcode::IMAGE_LOAD_MIP:
-            translator.IMAGE_LOAD_MIP(inst);
+            translator.IMAGE_LOAD(true, inst);
+            break;
+        case Opcode::IMAGE_LOAD:
+            translator.IMAGE_LOAD(false, inst);
             break;
         case Opcode::V_CMP_GE_I32:
             translator.V_CMP_U32(ConditionOp::GE, true, false, inst);
@@ -335,6 +338,9 @@ void Translate(IR::Block* block, std::span<const GcnInst> inst_list, Info& info)
         case Opcode::V_CMP_LE_I32:
             translator.V_CMP_U32(ConditionOp::LE, true, false, inst);
             break;
+        case Opcode::V_CMP_NE_I32:
+            translator.V_CMP_U32(ConditionOp::LG, true, false, inst);
+            break;
         case Opcode::V_CMP_NE_U32:
             translator.V_CMP_U32(ConditionOp::LG, false, false, inst);
             break;
@@ -386,6 +392,9 @@ void Translate(IR::Block* block, std::span<const GcnInst> inst_list, Info& info)
         case Opcode::V_CMP_NLT_F32:
             translator.V_CMP_F32(ConditionOp::GE, false, inst);
             break;
+        case Opcode::S_CMP_LT_U32:
+            translator.S_CMP(ConditionOp::LT, false, inst);
+            break;
         case Opcode::S_CMP_LG_U32:
             translator.S_CMP(ConditionOp::LG, false, inst);
             break;
@@ -585,6 +594,9 @@ void Translate(IR::Block* block, std::span<const GcnInst> inst_list, Info& info)
         case Opcode::S_AND_B64:
             translator.S_AND_B64(false, inst);
             break;
+        case Opcode::S_NOT_B64:
+            translator.S_NOT_B64(inst);
+            break;
         case Opcode::S_NAND_B64:
             translator.S_AND_B64(true, inst);
             break;
@@ -627,6 +639,9 @@ void Translate(IR::Block* block, std::span<const GcnInst> inst_list, Info& info)
         case Opcode::S_AND_B32:
             translator.S_AND_B32(inst);
             break;
+        case Opcode::S_OR_B32:
+            translator.S_OR_B32(inst);
+            break;
         case Opcode::S_LSHR_B32:
             translator.S_LSHR_B32(inst);
             break;
@@ -657,9 +672,27 @@ void Translate(IR::Block* block, std::span<const GcnInst> inst_list, Info& info)
         case Opcode::S_BFM_B32:
             translator.S_BFM_B32(inst);
             break;
+        case Opcode::V_MIN_U32:
+            translator.V_MIN_U32(inst);
+            break;
+        case Opcode::V_CMP_NE_U64:
+            translator.V_CMP_NE_U64(inst);
+            break;
         case Opcode::V_TRUNC_F32:
             translator.V_TRUNC_F32(inst);
             break;
+        case Opcode::V_CEIL_F32:
+            translator.V_CEIL_F32(inst);
+            break;
+        case Opcode::V_BFI_B32:
+            translator.V_BFI_B32(inst);
+            break;
+        case Opcode::S_BREV_B32:
+            translator.S_BREV_B32(inst);
+            break;
+        case Opcode::S_TTRACEDATA:
+            LOG_WARNING(Render_Vulkan, "S_TTRACEDATA instruction!");
+            break;
         case Opcode::S_NOP:
         case Opcode::S_CBRANCH_EXECZ:
         case Opcode::S_CBRANCH_SCC0:

src/shader_recompiler/frontend/translate/translate.h

@@ -45,12 +45,15 @@ public:
     void S_AND_B64(bool negate, const GcnInst& inst);
     void S_ADD_I32(const GcnInst& inst);
     void S_AND_B32(const GcnInst& inst);
+    void S_OR_B32(const GcnInst& inst);
     void S_LSHR_B32(const GcnInst& inst);
     void S_CSELECT_B32(const GcnInst& inst);
     void S_CSELECT_B64(const GcnInst& inst);
     void S_BFE_U32(const GcnInst& inst);
     void S_LSHL_B32(const GcnInst& inst);
     void S_BFM_B32(const GcnInst& inst);
+    void S_NOT_B64(const GcnInst& inst);
+    void S_BREV_B32(const GcnInst& inst);
 
     // Scalar Memory
     void S_LOAD_DWORD(int num_dwords, const GcnInst& inst);
@@ -115,6 +118,10 @@ public:
     void V_MIN_I32(const GcnInst& inst);
     void V_MUL_LO_U32(const GcnInst& inst);
     void V_TRUNC_F32(const GcnInst& inst);
+    void V_CEIL_F32(const GcnInst& inst);
+    void V_MIN_U32(const GcnInst& inst);
+    void V_CMP_NE_U64(const GcnInst& inst);
+    void V_BFI_B32(const GcnInst& inst);
 
     // Vector Memory
     void BUFFER_LOAD_FORMAT(u32 num_dwords, bool is_typed, const GcnInst& inst);
@@ -132,7 +139,7 @@ public:
     void IMAGE_GET_RESINFO(const GcnInst& inst);
     void IMAGE_SAMPLE(const GcnInst& inst);
     void IMAGE_STORE(const GcnInst& inst);
-    void IMAGE_LOAD_MIP(const GcnInst& inst);
+    void IMAGE_LOAD(bool has_mip, const GcnInst& inst);
 
     // Export
     void EXP(const GcnInst& inst);

src/shader_recompiler/frontend/translate/vector_alu.cpp

@@ -430,4 +430,52 @@ void Translator::V_TRUNC_F32(const GcnInst& inst) {
     SetDst(inst.dst[0], ir.FPTrunc(src0));
 }
 
+void Translator::V_CEIL_F32(const GcnInst& inst) {
+    const IR::F32 src0{GetSrc(inst.src[0], true)};
+    SetDst(inst.dst[0], ir.FPCeil(src0));
+}
+
+void Translator::V_MIN_U32(const GcnInst& inst) {
+    const IR::U32 src0{GetSrc(inst.src[0])};
+    const IR::U32 src1{GetSrc(inst.src[1])};
+    SetDst(inst.dst[0], ir.IMin(src0, src1, false));
+}
+
+void Translator::V_CMP_NE_U64(const GcnInst& inst) {
+    const auto get_src = [&](const InstOperand& operand) {
+        switch (operand.field) {
+        case OperandField::VccLo:
+            return ir.GetVcc();
+        case OperandField::ExecLo:
+            return ir.GetExec();
+        case OperandField::ScalarGPR:
+            return ir.GetThreadBitScalarReg(IR::ScalarReg(operand.code));
+        case OperandField::ConstZero:
+            return ir.Imm1(false);
+        default:
+            UNREACHABLE();
+        }
+    };
+    const IR::U1 src0{get_src(inst.src[0])};
+    ASSERT(inst.src[1].field == OperandField::ConstZero); // src0 != 0
+    switch (inst.dst[1].field) {
+    case OperandField::VccLo:
+        ir.SetVcc(src0);
+        break;
+    case OperandField::ScalarGPR:
+        ir.SetThreadBitScalarReg(IR::ScalarReg(inst.dst[1].code), src0);
+        break;
+    default:
+        UNREACHABLE();
+    }
+}
+
+void Translator::V_BFI_B32(const GcnInst& inst) {
+    const IR::U32 src0{GetSrc(inst.src[0])};
+    const IR::U32 src1{GetSrc(inst.src[1])};
+    const IR::U32 src2{GetSrc(inst.src[2])};
+    SetDst(inst.dst[0],
+           ir.BitwiseOr(ir.BitwiseAnd(src0, src1), ir.BitwiseAnd(ir.BitwiseNot(src0), src2)));
+}
+
 } // namespace Shader::Gcn

src/shader_recompiler/frontend/translate/vector_memory.cpp

@@ -31,7 +31,9 @@ void Translator::IMAGE_GET_RESINFO(const GcnInst& inst) {
 
 void Translator::IMAGE_SAMPLE(const GcnInst& inst) {
     const auto& mimg = inst.control.mimg;
-    ASSERT(!mimg.da);
+    if (mimg.da) {
+        LOG_WARNING(Render_Vulkan, "Image instruction declares an array");
+    }
 
     IR::VectorReg addr_reg{inst.src[0].code};
     IR::VectorReg dest_reg{inst.dst[0].code};
@@ -107,7 +109,7 @@ void Translator::IMAGE_SAMPLE(const GcnInst& inst) {
     }
 }
 
-void Translator::IMAGE_LOAD_MIP(const GcnInst& inst) {
+void Translator::IMAGE_LOAD(bool has_mip, const GcnInst& inst) {
     const auto& mimg = inst.control.mimg;
     IR::VectorReg addr_reg{inst.src[0].code};
     IR::VectorReg dest_reg{inst.dst[0].code};
@@ -119,7 +121,7 @@ void Translator::IMAGE_LOAD_MIP(const GcnInst& inst) {
                               ir.GetVectorReg(addr_reg + 2), ir.GetVectorReg(addr_reg + 3));
 
     IR::TextureInstInfo info{};
-    info.explicit_lod.Assign(1);
+    info.explicit_lod.Assign(has_mip);
     const IR::Value texel = ir.ImageFetch(handle, body, {}, {}, {}, info);
 
     for (u32 i = 0; i < 4; i++) {

src/shader_recompiler/ir/passes/resource_tracking_pass.cpp

@@ -251,7 +251,9 @@ IR::Value PatchCubeCoord(IR::IREmitter& ir, const IR::Value& s, const IR::Value&
 
 void PatchImageInstruction(IR::Block& block, IR::Inst& inst, Info& info, Descriptors& descriptors) {
     IR::Inst* producer = inst.Arg(0).InstRecursive();
-    ASSERT(producer->GetOpcode() == IR::Opcode::CompositeConstructU32x2 ||
+    ASSERT(producer->GetOpcode() ==
+               IR::Opcode::CompositeConstructU32x2 ||        // IMAGE_SAMPLE (image+sampler)
+           producer->GetOpcode() == IR::Opcode::ReadConst || // IMAGE_LOAD (image only)
            producer->GetOpcode() == IR::Opcode::GetUserData);
     const auto [tsharp_handle, ssharp_handle] = [&] -> std::pair<IR::Inst*, IR::Inst*> {
         if (producer->GetOpcode() == IR::Opcode::CompositeConstructU32x2) {

src/video_core/amdgpu/liverpool.h

@@ -827,7 +827,8 @@ struct Liverpool {
             PolygonControl polygon_control;
             ViewportControl viewport_control;
             VsOutputControl vs_output_control;
-            INSERT_PADDING_WORDS(0xA29E - 0xA207 - 1);
+            INSERT_PADDING_WORDS(0xA29E - 0xA207 - 2);
+            u32 index_size;
             u32 max_index_size;
             IndexBufferType index_buffer_type;
             INSERT_PADDING_WORDS(0xA2A1 - 0xA29E - 2);
@@ -993,6 +994,7 @@ static_assert(GFX6_3D_REG_INDEX(depth_control) == 0xA200);
 static_assert(GFX6_3D_REG_INDEX(clipper_control) == 0xA204);
 static_assert(GFX6_3D_REG_INDEX(viewport_control) == 0xA206);
 static_assert(GFX6_3D_REG_INDEX(vs_output_control) == 0xA207);
+static_assert(GFX6_3D_REG_INDEX(index_size) == 0xA29D);
 static_assert(GFX6_3D_REG_INDEX(index_buffer_type) == 0xA29F);
 static_assert(GFX6_3D_REG_INDEX(enable_primitive_id) == 0xA2A1);
 static_assert(GFX6_3D_REG_INDEX(poly_offset) == 0xA2DF);

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -23,7 +23,7 @@ Rasterizer::Rasterizer(const Instance& instance_, Scheduler& scheduler_,
     : instance{instance_}, scheduler{scheduler_}, texture_cache{texture_cache_},
       liverpool{liverpool_}, memory{Core::Memory::Instance()},
       pipeline_cache{instance, scheduler, liverpool},
-      vertex_index_buffer{instance, scheduler, VertexIndexFlags, 32_MB} {
+      vertex_index_buffer{instance, scheduler, VertexIndexFlags, 128_MB} {
     if (!Config::nullGpu()) {
         liverpool->BindRasterizer(this);
     }

src/video_core/texture_cache/tile_manager.cpp

@@ -191,6 +191,7 @@ vk::Format DemoteImageFormatForDetiling(vk::Format format) {
     case vk::Format::eBc3SrgbBlock:
         [[fallthrough]];
     case vk::Format::eBc3UnormBlock:
+    case vk::Format::eBc7SrgbBlock:
         return vk::Format::eR32G32B32A32Uint;
     default:
         break;
@@ -225,7 +226,8 @@ static constexpr vk::BufferUsageFlags StagingFlags = vk::BufferUsageFlagBits::eT
                                                      vk::BufferUsageFlagBits::eStorageBuffer;
 
 TileManager::TileManager(const Vulkan::Instance& instance, Vulkan::Scheduler& scheduler)
-    : instance{instance}, scheduler{scheduler}, staging{instance, scheduler, StagingFlags, 64_MB} {
+    : instance{instance}, scheduler{scheduler},
+      staging{instance, scheduler, StagingFlags, 64_MB, Vulkan::BufferType::Upload} {
 
     static const std::array detiler_shaders{
         HostShaders::DETILE_M8X1_COMP,