kernel: Module loading

src/core/file_sys/fs.cpp

@@ -52,7 +52,6 @@ std::string MntPoints::GetHostFile(const std::string& guest_file) {
         if (find == 0) {
             std::string npath = guest_file.substr(pair.guest_path.size(), guest_file.size() - 1);
             std::replace(pair.host_path.begin(), pair.host_path.end(), '\\', '/');
-            fmt::print("GetHostFile: {}\n", pair.host_path + npath);
             return pair.host_path + npath;
         }
     }

src/core/libraries/kernel/libkernel.cpp

@@ -236,13 +236,42 @@ size_t PS4_SYSV_ABI sceKernelPread(int fd, void *buf, size_t count, uint64_t off
     return file.ReadRaw<u8>(buf, count);
 }
 
-int PS4_SYSV_ABI Unknown1() {
+s32 PS4_SYSV_ABI sceKernelLoadStartModule(const char *moduleFileName, size_t args, const void *argp,
-    return 0x81100004;
+                                          u32 flags, const void *pOpt, int *pRes) {
+    LOG_INFO(Lib_Kernel, "called filename = {}, args = {}", moduleFileName, args);
+
+    if (flags != 0) {
+        return 0x80020016;
+    }
+
+    auto* mnt = Common::Singleton<Core::FileSys::MntPoints>::Instance();
+    const auto path = mnt->GetHostFile(moduleFileName);
+
+    // Load PRX module.
+    auto* linker = Common::Singleton<Core::Linker>::Instance();
+    u32 handle = linker->LoadModule(path);
+    auto* module = linker->GetModule(handle);
+    linker->Relocate(module);
+
+    // Retrieve and verify proc param according to libkernel.
+    const u64* param = module->GetProcParam<u64*>();
+    ASSERT_MSG(!param || param[0] >= 0x18, "Invalid module param size: {}", param[0]);
+    module->Start(args, argp, 0);
+
+    return handle;
+}
+
+s32 PS4_SYSV_ABI sceKernelDlsym(s32 handle, const char *symbol, void **addrp) {
+    auto* linker = Common::Singleton<Core::Linker>::Instance();
+    auto* module = linker->GetModule(handle);
+    *addrp = module->FindByName(symbol);
+    if (*addrp == nullptr) {
+        return 0x80020003;
+    }
+    return ORBIS_OK;
 }
 
 void LibKernel_Register(Core::Loader::SymbolsResolver* sym) {
-    LIB_FUNCTION("fJgP+wqifno", "libSceDiscMap", 1, "libSceDiscMap", 1, 1, Unknown1);
-
     // obj
     LIB_OBJ("f7uOxY9mM1U", "libkernel", 1, "libkernel", 1, 1, &g_stack_chk_guard);
     // memory
@@ -260,6 +289,8 @@ void LibKernel_Register(Core::Loader::SymbolsResolver* sym) {
     LIB_FUNCTION("IWIBBdTHit4", "libkernel", 1, "libkernel", 1, 1, sceKernelMapFlexibleMemory);
     LIB_FUNCTION("rVjRvHJ0X6c", "libkernel", 1, "libkernel", 1, 1, sceKernelVirtualQuery);
     LIB_FUNCTION("p5EcQeEeJAE", "libkernel", 1, "libkernel", 1, 1, _sceKernelRtldSetApplicationHeapAPI);
+    LIB_FUNCTION("wzvqT4UqKX8", "libkernel", 1, "libkernel", 1, 1, sceKernelLoadStartModule);
+    LIB_FUNCTION("LwG8g3niqwA", "libkernel", 1, "libkernel", 1, 1, sceKernelDlsym);
 
     // equeue
     LIB_FUNCTION("D0OdFMjp46I", "libkernel", 1, "libkernel", 1, 1, sceKernelCreateEqueue);

src/core/linker.cpp

@@ -94,23 +94,22 @@ void Linker::Execute() {
     }
 }
 
-Module* Linker::LoadModule(const std::filesystem::path& elf_name) {
+s32 Linker::LoadModule(const std::filesystem::path& elf_name) {
+    std::scoped_lock lk{mutex};
+
     if (!std::filesystem::exists(elf_name)) {
         LOG_ERROR(Core_Linker, "Provided file {} does not exist", elf_name.string());
-        return nullptr;
+        return -1;
     }
 
     auto module = std::make_unique<Module>(elf_name);
     if (!module->IsValid()) {
         LOG_ERROR(Core_Linker, "Provided file {} is not valid ELF file", elf_name.string());
-        return nullptr;
+        return -1;
     }
 
-    if (!module->IsSharedLib()) {
+    m_modules.emplace_back(std::move(module));
-        main_proc_param_addr = module->GetProcParam();
+    return m_modules.size() - 1;
-    }
-
-    return m_modules.emplace_back(std::move(module)).get();
 }
 
 void Linker::Relocate(Module* module) {

src/core/linker.h

@@ -3,6 +3,7 @@
 
 #pragma once
 
+#include <mutex>
 #include <vector>
 #include "core/module.h"
 
@@ -29,7 +30,11 @@ public:
     }
 
     VAddr GetProcParam() const {
-        return main_proc_param_addr;
+        return m_modules[0]->GetProcParam();
+    }
+
+    Module* GetModule(s32 index) const {
+        return m_modules.at(index).get();
     }
 
     void SetHeapApiFunc(void* func) {
@@ -39,7 +44,7 @@ public:
     void* TlsGetAddr(u64 module_index, u64 offset);
     void InitTlsForThread(bool is_primary = false);
 
-    Module* LoadModule(const std::filesystem::path& elf_name);
+    s32 LoadModule(const std::filesystem::path& elf_name);
 
     void Relocate(Module* module);
     void Resolve(const std::string& name, Loader::SymbolType type,
@@ -49,15 +54,14 @@ public:
 
 private:
     const Module* FindExportedModule(const ModuleInfo& m, const LibraryInfo& l);
-    void InitTls();
 
+    std::mutex mutex;
     u32 dtv_generation_counter{1};
     size_t static_tls_size{};
     size_t max_tls_index{};
     HeapApiFunc heap_api_func{};
     std::vector<std::unique_ptr<Module>> m_modules;
     Loader::SymbolsResolver m_hle_symbols{};
-    VAddr main_proc_param_addr{};
 };
 
 } // namespace Core

src/core/loader/symbols_resolver.cpp

@@ -11,7 +11,7 @@
 namespace Core::Loader {
 
 void SymbolsResolver::AddSymbol(const SymbolResolver& s, u64 virtual_addr) {
-    m_symbols.emplace_back(GenerateName(s), virtual_addr);
+    m_symbols.emplace_back(GenerateName(s), s.nidName, virtual_addr);
 }
 
 std::string SymbolsResolver::GenerateName(const SymbolResolver& s) {

src/core/loader/symbols_resolver.h

@@ -6,6 +6,7 @@
 #include <filesystem>
 #include <string>
 #include <vector>
+#include <span>
 #include "common/types.h"
 
 namespace Core::Loader {
@@ -20,6 +21,7 @@ enum class SymbolType {
 
 struct SymbolRecord {
     std::string name;
+    std::string nid_name;
     u64 virtual_address;
 };
 
@@ -42,6 +44,10 @@ public:
     void AddSymbol(const SymbolResolver& s, u64 virtual_addr);
     const SymbolRecord* FindSymbol(const SymbolResolver& s) const;
 
+    std::span<const SymbolRecord> GetSymbols() const {
+        return m_symbols;
+    }
+
     size_t GetSize() const noexcept {
         return m_symbols.size();
     }

src/core/memory.cpp

@@ -240,7 +240,7 @@ MemoryManager::VMAHandle MemoryManager::MergeAdjacent(VMAHandle iter) {
 }
 
 void MemoryManager::MapVulkanMemory(VAddr addr, size_t size) {
-    return;
+    //return;
     const vk::Device device = instance->GetDevice();
     const auto memory_props = instance->GetPhysicalDevice().getMemoryProperties();
     void* host_pointer = reinterpret_cast<void*>(addr);
@@ -312,7 +312,7 @@ void MemoryManager::MapVulkanMemory(VAddr addr, size_t size) {
 }
 
 void MemoryManager::UnmapVulkanMemory(VAddr addr, size_t size) {
-    return;
+    //return;
     const auto it = mapped_memories.find(addr);
     ASSERT(it != mapped_memories.end() && it->second.buffer_size == size);
     mapped_memories.erase(it);

src/core/module.cpp

@@ -349,6 +349,7 @@ void Module::LoadSymbols() {
 
             const auto aeronid = AeroLib::FindByNid(ids.at(0).c_str());
             const auto nid_name = aeronid ? aeronid->name : "UNK";
+            LOG_INFO(Core_Linker, "NidName {}", nid_name);
 
             Loader::SymbolResolver sym_r{};
             sym_r.name = ids.at(0);

src/core/module.h

@@ -119,8 +119,18 @@ public:
         return elf.IsSharedLib();
     }
 
-    VAddr GetProcParam() const noexcept {
+    void* FindByName(std::string_view name) {
-        return proc_param_virtual_addr;
+        const auto symbols = export_sym.GetSymbols();
+        const auto it = std::ranges::find(symbols, name, &Loader::SymbolRecord::nid_name);
+        if (it != symbols.end()) {
+            return reinterpret_cast<void*>(it->virtual_address);
+        }
+        return nullptr;
+    }
+
+    template <typename T = VAddr>
+    const T GetProcParam() const noexcept {
+        return reinterpret_cast<T>(proc_param_virtual_addr);
     }
 
     std::span<const ModuleInfo> GetImportModules() const {

src/shader_recompiler/frontend/translate/scalar_alu.cpp

@@ -247,4 +247,12 @@ void Translator::S_BFE_U32(const GcnInst& inst) {
     ir.SetScc(ir.INotEqual(result, ir.Imm32(0)));
 }
 
+void Translator::S_LSHL_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.ShiftLeftLogical(src0, ir.BitwiseAnd(src1, ir.Imm32(0x1F)));
+    SetDst(inst.dst[0], result);
+    ir.SetScc(ir.INotEqual(result, ir.Imm32(0)));
+}
+
 } // namespace Shader::Gcn

src/shader_recompiler/frontend/translate/translate.cpp

@@ -371,6 +371,9 @@ void Translate(IR::Block* block, std::span<const GcnInst> inst_list, Info& info)
         case Opcode::S_CMP_EQ_U32:
             translator.S_CMP(ConditionOp::EQ, false, inst);
             break;
+        case Opcode::S_LSHL_B32:
+            translator.S_LSHL_B32(inst);
+            break;
         case Opcode::V_CNDMASK_B32:
             translator.V_CNDMASK_B32(inst);
             break;
@@ -527,6 +530,9 @@ void Translate(IR::Block* block, std::span<const GcnInst> inst_list, Info& info)
         case Opcode::S_BFE_U32:
             translator.S_BFE_U32(inst);
             break;
+        case Opcode::V_RNDNE_F32:
+            translator.V_RNDNE_F32(inst);
+            break;
         case Opcode::S_NOP:
         case Opcode::S_CBRANCH_EXECZ:
         case Opcode::S_CBRANCH_SCC0:

src/shader_recompiler/frontend/translate/translate.h

@@ -48,6 +48,7 @@ public:
     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);
 
     // Scalar Memory
     void S_LOAD_DWORD(int num_dwords, const GcnInst& inst);
@@ -102,6 +103,7 @@ public:
     void V_LSHR_B32(const GcnInst& inst);
     void V_ASHRREV_I32(const GcnInst& inst);
     void V_MAD_U32_U24(const GcnInst& inst);
+    void V_RNDNE_F32(const GcnInst& inst);
 
     // Vector Memory
     void BUFFER_LOAD_FORMAT(u32 num_dwords, bool is_typed, const GcnInst& inst);

src/shader_recompiler/frontend/translate/vector_alu.cpp

@@ -367,4 +367,9 @@ void Translator::V_MAD_U32_U24(const GcnInst& inst) {
     V_MAD_I32_I24(inst);
 }
 
+void Translator::V_RNDNE_F32(const GcnInst& inst) {
+    const IR::F32 src0{GetSrc(inst.src[0], true)};
+    SetDst(inst.dst[0], ir.FPRoundEven(src0));
+}
+
 } // namespace Shader::Gcn

src/shader_recompiler/frontend/translate/vector_memory.cpp

@@ -31,7 +31,7 @@ void Translator::IMAGE_GET_RESINFO(const GcnInst& inst) {
 
 void Translator::IMAGE_SAMPLE(const GcnInst& inst) {
     const auto& mimg = inst.control.mimg;
-    ASSERT(!mimg.da);
+    //ASSERT(!mimg.da);
 
     IR::VectorReg addr_reg{inst.src[0].code};
     IR::VectorReg dest_reg{inst.dst[0].code};

src/video_core/amdgpu/liverpool.cpp

@@ -252,6 +252,16 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span<const u32> dcb, std::span<c
             }
             break;
         }
+        case PM4ItOpcode::DrawIndexOffset2: {
+            const auto* draw_index_off = reinterpret_cast<const PM4CmdDrawIndexOffset2*>(header);
+            regs.max_index_size = draw_index_off->max_size;
+            regs.num_indices = draw_index_off->index_count;
+            regs.draw_initiator = draw_index_off->draw_initiator;
+            if (rasterizer) {
+                rasterizer->Draw(true, draw_index_off->index_offset);
+            }
+            break;
+        }
         case PM4ItOpcode::DrawIndexAuto: {
             const auto* draw_index = reinterpret_cast<const PM4CmdDrawIndexAuto*>(header);
             regs.num_indices = draw_index->index_count;
@@ -272,6 +282,17 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span<const u32> dcb, std::span<c
             }
             break;
         }
+        case PM4ItOpcode::NumInstances: {
+            const auto* num_instances = reinterpret_cast<const PM4CmdDrawNumInstances*>(header);
+            regs.num_instances.num_instances = num_instances->num_instances;
+            break;
+        }
+        case PM4ItOpcode::IndexBase: {
+            const auto* index_base = reinterpret_cast<const PM4CmdDrawIndexBase*>(header);
+            regs.index_base_address.base_addr_lo = index_base->addr_lo;
+            regs.index_base_address.base_addr_hi.Assign(index_base->addr_hi);
+            break;
+        }
         case PM4ItOpcode::EventWrite: {
             // const auto* event = reinterpret_cast<const PM4CmdEventWrite*>(header);
             break;

src/video_core/amdgpu/pm4_cmds.h

@@ -548,4 +548,15 @@ struct PM4CmdDispatchDirect {
     u32 dispatch_initiator; ///< Dispatch Initiator Register
 };
 
+struct PM4CmdDrawNumInstances {
+    PM4Type3Header header;
+    u32 num_instances;
+};
+
+struct PM4CmdDrawIndexBase {
+    PM4Type3Header header;
+    u32 addr_lo;
+    u32 addr_hi;
+};
+
 } // namespace AmdGpu

src/video_core/renderer_vulkan/liverpool_to_vk.cpp

@@ -309,6 +309,12 @@ vk::Format SurfaceFormat(AmdGpu::DataFormat data_format, AmdGpu::NumberFormat nu
     if (data_format == AmdGpu::DataFormat::FormatBc3 && num_format == AmdGpu::NumberFormat::Srgb) {
         return vk::Format::eBc3SrgbBlock;
     }
+    if (data_format == AmdGpu::DataFormat::Format16_16_16_16 && num_format == AmdGpu::NumberFormat::Sint) {
+        return vk::Format::eR16G16B16A16Sint;
+    }
+    if (data_format == AmdGpu::DataFormat::FormatBc7 && num_format == AmdGpu::NumberFormat::Srgb) {
+        return vk::Format::eBc7SrgbBlock;
+    }
     UNREACHABLE();
 }
 

src/video_core/renderer_vulkan/vk_compute_pipeline.cpp

@@ -81,8 +81,17 @@ ComputePipeline::ComputePipeline(const Instance& instance_, Scheduler& scheduler
 
 ComputePipeline::~ComputePipeline() = default;
 
-void ComputePipeline::BindResources(Core::MemoryManager* memory,
+void ComputePipeline::BindResources(Core::MemoryManager* memory, StreamBuffer& staging,
                                     VideoCore::TextureCache& texture_cache) const {
+    static constexpr u64 MinUniformAlignment = 64;
+
+    const auto map_staging = [&](auto src, size_t size) {
+        const auto [data, offset, _] = staging.Map(size, MinUniformAlignment);
+        std::memcpy(data, reinterpret_cast<const void*>(src), size);
+        staging.Commit(size);
+        return offset;
+    };
+
     // Bind resource buffers and textures.
     boost::container::static_vector<vk::DescriptorBufferInfo, 4> buffer_infos;
     boost::container::static_vector<vk::DescriptorImageInfo, 8> image_infos;
@@ -94,8 +103,9 @@ void ComputePipeline::BindResources(Core::MemoryManager* memory,
         const u32 size = vsharp.GetSize();
         const VAddr addr = vsharp.base_address.Value();
         texture_cache.OnCpuWrite(addr);
-        const auto [vk_buffer, offset] = memory->GetVulkanBuffer(addr);
+        const u32 offset = map_staging(addr, size);
-        buffer_infos.emplace_back(vk_buffer, offset, size);
+        //const auto [vk_buffer, offset] = memory->GetVulkanBuffer(addr);
+        buffer_infos.emplace_back(staging.Handle(), offset, size);
         set_writes.push_back({
             .dstSet = VK_NULL_HANDLE,
             .dstBinding = binding++,

src/video_core/renderer_vulkan/vk_compute_pipeline.h

@@ -31,7 +31,8 @@ public:
         return *pipeline;
     }
 
-    void BindResources(Core::MemoryManager* memory, VideoCore::TextureCache& texture_cache) const;
+    void BindResources(Core::MemoryManager* memory, StreamBuffer& staging,
+                       VideoCore::TextureCache& texture_cache) const;
 
 private:
     const Instance& instance;

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -32,16 +32,15 @@ Rasterizer::Rasterizer(const Instance& instance_, Scheduler& scheduler_,
 
 Rasterizer::~Rasterizer() = default;
 
-void Rasterizer::Draw(bool is_indexed) {
+void Rasterizer::Draw(bool is_indexed, u32 index_offset) {
     const auto cmdbuf = scheduler.CommandBuffer();
     const auto& regs = liverpool->regs;
-    const u32 num_indices = SetupIndexBuffer(is_indexed);
+    const u32 num_indices = SetupIndexBuffer(is_indexed, index_offset);
     const GraphicsPipeline* pipeline = pipeline_cache.GetGraphicsPipeline();
     pipeline->BindResources(memory, vertex_index_buffer, texture_cache);
 
     boost::container::static_vector<vk::RenderingAttachmentInfo, Liverpool::NumColorBuffers>
         color_attachments{};
-    const std::array color = {0.5f, 0.5f, 0.5f, 1.f};
     for (auto col_buf_id = 0u; col_buf_id < Liverpool::NumColorBuffers; ++col_buf_id) {
         const auto& col_buf = regs.color_buffers[col_buf_id];
         if (!col_buf) {
@@ -54,12 +53,10 @@ void Rasterizer::Draw(bool is_indexed) {
         color_attachments.push_back({
             .imageView = *image_view.image_view,
             .imageLayout = vk::ImageLayout::eGeneral,
-            .loadOp = compute_done ? vk::AttachmentLoadOp::eClear : vk::AttachmentLoadOp::eLoad,
+            .loadOp = vk::AttachmentLoadOp::eLoad,
             .storeOp = vk::AttachmentStoreOp::eStore,
-            .clearValue = vk::ClearValue{color},
         });
     }
-    compute_done = false;
 
     // TODO: Don't restart renderpass every draw
     const auto& scissor = regs.screen_scissor;
@@ -88,18 +85,16 @@ void Rasterizer::Draw(bool is_indexed) {
 }
 
 void Rasterizer::DispatchDirect() {
-    compute_done = true;
-    return;
     const auto cmdbuf = scheduler.CommandBuffer();
     const auto& cs_program = liverpool->regs.cs_program;
     const ComputePipeline* pipeline = pipeline_cache.GetComputePipeline();
-    pipeline->BindResources(memory, texture_cache);
+    pipeline->BindResources(memory, vertex_index_buffer, texture_cache);
 
     cmdbuf.bindPipeline(vk::PipelineBindPoint::eCompute, pipeline->Handle());
     cmdbuf.dispatch(cs_program.dim_x, cs_program.dim_y, cs_program.dim_z);
 }
 
-u32 Rasterizer::SetupIndexBuffer(bool& is_indexed) {
+u32 Rasterizer::SetupIndexBuffer(bool& is_indexed, u32 index_offset) {
     // Emulate QuadList primitive type with CPU made index buffer.
     const auto& regs = liverpool->regs;
     if (liverpool->regs.primitive_type == Liverpool::PrimitiveType::QuadList) {
@@ -135,7 +130,7 @@ u32 Rasterizer::SetupIndexBuffer(bool& is_indexed) {
 
     // Bind index buffer.
     const auto cmdbuf = scheduler.CommandBuffer();
-    cmdbuf.bindIndexBuffer(vertex_index_buffer.Handle(), offset, index_type);
+    cmdbuf.bindIndexBuffer(vertex_index_buffer.Handle(), offset + index_offset * index_size, index_type);
     return regs.num_indices;
 }
 

src/video_core/renderer_vulkan/vk_rasterizer.h

@@ -29,12 +29,12 @@ public:
                         VideoCore::TextureCache& texture_cache, AmdGpu::Liverpool* liverpool);
     ~Rasterizer();
 
-    void Draw(bool is_indexed);
+    void Draw(bool is_indexed, u32 index_offset = 0);
 
     void DispatchDirect();
 
 private:
-    u32 SetupIndexBuffer(bool& is_indexed);
+    u32 SetupIndexBuffer(bool& is_indexed, u32 index_offset);
     void MapMemory(VAddr addr, size_t size);
 
     void UpdateDynamicState(const GraphicsPipeline& pipeline);
@@ -49,7 +49,6 @@ private:
     Core::MemoryManager* memory;
     PipelineCache pipeline_cache;
     StreamBuffer vertex_index_buffer;
-    bool compute_done{};
 };
 
 } // namespace Vulkan