early-access version 1550

README.md

@@ -1,7 +1,7 @@
 yuzu emulator early access
 =============
 
-This is the source code for early-access 1548.
+This is the source code for early-access 1550.
 
 ## Legal Notice
 

externals/dynarmic/include/dynarmic/A32/config.h

@@ -197,6 +197,13 @@ struct UserConfig {
     /// NOTE: Calling Jit::SetCpsr with CPSR.E=1 while this option is enabled may result
     ///       in unusual behavior.
     bool always_little_endian = false;
+
+    // Minimum size is about 8MiB. Maximum size is about 2GiB. Maximum size is limited by
+    // the maximum length of a x64 jump.
+    size_t code_cache_size = 256 * 1024 * 1024; // bytes
+    // Determines the relative size of the near and far code caches. Must be smaller than
+    // code_cache_size.
+    size_t far_code_offset = 200 * 1024 * 1024; // bytes
 };
 
 } // namespace A32

externals/dynarmic/include/dynarmic/A64/config.h

@@ -245,6 +245,13 @@ struct UserConfig {
     // Determines whether AddTicks and GetTicksRemaining are called.
     // If false, execution will continue until soon after Jit::HaltExecution is called.
     // bool enable_ticks = true; // TODO
+
+    // Minimum size is about 8MiB. Maximum size is about 2GiB. Maximum size is limited by
+    // the maximum length of a x64 jump.
+    size_t code_cache_size = 256 * 1024 * 1024; // bytes
+    // Determines the relative size of the near and far code caches. Must be smaller than
+    // code_cache_size.
+    size_t far_code_offset = 200 * 1024 * 1024; // bytes
 };
 
 } // namespace A64

externals/dynarmic/src/backend/x64/a32_interface.cpp

@@ -53,7 +53,7 @@ static std::function<void(BlockOfCode&)> GenRCP(const A32::UserConfig& conf) {
 
 struct Jit::Impl {
     Impl(Jit* jit, A32::UserConfig conf)
-            : block_of_code(GenRunCodeCallbacks(conf.callbacks, &GetCurrentBlockThunk, this), JitStateInfo{jit_state}, GenRCP(conf))
+            : block_of_code(GenRunCodeCallbacks(conf.callbacks, &GetCurrentBlockThunk, this), JitStateInfo{jit_state}, conf.code_cache_size, conf.far_code_offset, GenRCP(conf))
             , emitter(block_of_code, conf, jit)
             , conf(std::move(conf))
             , jit_interface(jit)

externals/dynarmic/src/backend/x64/a64_interface.cpp

@@ -45,7 +45,7 @@ struct Jit::Impl final {
 public:
     Impl(Jit* jit, UserConfig conf)
         : conf(conf)
-        , block_of_code(GenRunCodeCallbacks(conf.callbacks, &GetCurrentBlockThunk, this), JitStateInfo{jit_state}, GenRCP(conf))
+        , block_of_code(GenRunCodeCallbacks(conf.callbacks, &GetCurrentBlockThunk, this), JitStateInfo{jit_state}, conf.code_cache_size, conf.far_code_offset, GenRCP(conf))
         , emitter(block_of_code, conf, jit)
     {
         ASSERT(conf.page_table_address_space_bits >= 12 && conf.page_table_address_space_bits <= 64);

externals/dynarmic/src/backend/x64/block_of_code.cpp

@@ -44,8 +44,6 @@ const std::array<Xbyak::Reg64, 6> BlockOfCode::ABI_PARAMS = {BlockOfCode::ABI_PA
 
 namespace {
 
-constexpr size_t TOTAL_CODE_SIZE = 256 * 1024 * 1024;
-constexpr size_t FAR_CODE_OFFSET = 200 * 1024 * 1024;
 constexpr size_t CONSTANT_POOL_SIZE = 2 * 1024 * 1024;
 
 class CustomXbyakAllocator : public Xbyak::Allocator {
@@ -75,10 +73,11 @@ void ProtectMemory(const void* base, size_t size, bool is_executable) {
 
 } // anonymous namespace
 
-BlockOfCode::BlockOfCode(RunCodeCallbacks cb, JitStateInfo jsi, std::function<void(BlockOfCode&)> rcp)
-        : Xbyak::CodeGenerator(TOTAL_CODE_SIZE, nullptr, &s_allocator)
+BlockOfCode::BlockOfCode(RunCodeCallbacks cb, JitStateInfo jsi, size_t total_code_size, size_t far_code_offset, std::function<void(BlockOfCode&)> rcp)
+        : Xbyak::CodeGenerator(total_code_size, nullptr, &s_allocator)
         , cb(std::move(cb))
         , jsi(jsi)
+        , far_code_offset(far_code_offset)
         , constant_pool(*this, CONSTANT_POOL_SIZE)
 {
     EnableWriting();
@@ -88,7 +87,7 @@ BlockOfCode::BlockOfCode(RunCodeCallbacks cb, JitStateInfo jsi, std::function<vo
 void BlockOfCode::PreludeComplete() {
     prelude_complete = true;
     near_code_begin = getCurr();
-    far_code_begin = getCurr() + FAR_CODE_OFFSET;
+    far_code_begin = getCurr() + far_code_offset;
     ClearCache();
     DisableWriting();
 }
@@ -115,22 +114,13 @@ void BlockOfCode::ClearCache() {
 
 size_t BlockOfCode::SpaceRemaining() const {
     ASSERT(prelude_complete);
-    // This function provides an underestimate of near-code-size but that's okay.
-    // (Why? The maximum size of near code should be measured from near_code_begin, not top_.)
-    // These are offsets from Xbyak::CodeArray::top_.
-    std::size_t far_code_offset, near_code_offset;
-    if (in_far_code) {
-        near_code_offset = static_cast<const u8*>(near_code_ptr) - getCode();
-        far_code_offset = getCurr() - getCode();
-    } else {
-        near_code_offset = getCurr() - getCode();
-        far_code_offset = static_cast<const u8*>(far_code_ptr) - getCode();
-    }
-    if (far_code_offset > TOTAL_CODE_SIZE)
+    const u8* current_near_ptr = in_far_code ? reinterpret_cast<const u8*>(near_code_ptr) : getCode<const u8*>();
+    const u8* current_far_ptr = in_far_code ? getCode<const u8*>() : reinterpret_cast<const u8*>(far_code_ptr);
+    if (current_near_ptr >= far_code_begin)
         return 0;
-    if (near_code_offset > FAR_CODE_OFFSET)
+    if (current_far_ptr >= &top_[maxSize_])
         return 0;
-    return std::min(TOTAL_CODE_SIZE - far_code_offset, FAR_CODE_OFFSET - near_code_offset);
+    return std::min(current_near_ptr - reinterpret_cast<const u8*>(far_code_begin), current_far_ptr - &top_[maxSize_]);
 }
 
 void BlockOfCode::RunCode(void* jit_state, CodePtr code_ptr) const {

externals/dynarmic/src/backend/x64/block_of_code.h

@@ -31,7 +31,7 @@ struct RunCodeCallbacks {
 
 class BlockOfCode final : public Xbyak::CodeGenerator {
 public:
-    BlockOfCode(RunCodeCallbacks cb, JitStateInfo jsi, std::function<void(BlockOfCode&)> rcp);
+    BlockOfCode(RunCodeCallbacks cb, JitStateInfo jsi, size_t total_code_size, size_t far_code_offset, std::function<void(BlockOfCode&)> rcp);
     BlockOfCode(const BlockOfCode&) = delete;
 
     /// Call when external emitters have finished emitting their preludes.
@@ -164,10 +164,11 @@ public:
 private:
     RunCodeCallbacks cb;
     JitStateInfo jsi;
+    size_t far_code_offset;
 
     bool prelude_complete = false;
-    CodePtr near_code_begin;
-    CodePtr far_code_begin;
+    CodePtr near_code_begin = nullptr;
+    CodePtr far_code_begin = nullptr;
 
     ConstantPool constant_pool;
 

src/core/arm/dynarmic/arm_dynarmic_32.cpp

@@ -137,6 +137,10 @@ std::shared_ptr<Dynarmic::A32::Jit> ARM_Dynarmic_32::MakeJit(Common::PageTable*
     // Timing
     config.wall_clock_cntpct = uses_wall_clock;
 
+    // Code cache size
+    config.code_cache_size = 512 * 1024 * 1024;
+    config.far_code_offset = 256 * 1024 * 1024;
+
     // Safe optimizations
     if (Settings::values.cpu_accuracy == Settings::CPUAccuracy::DebugMode) {
         if (!Settings::values.cpuopt_page_tables) {

src/core/arm/dynarmic/arm_dynarmic_64.cpp

@@ -177,6 +177,10 @@ std::shared_ptr<Dynarmic::A64::Jit> ARM_Dynarmic_64::MakeJit(Common::PageTable*
     // Timing
     config.wall_clock_cntpct = uses_wall_clock;
 
+    // Code cache size
+    config.code_cache_size = 512 * 1024 * 1024;
+    config.far_code_offset = 256 * 1024 * 1024;
+
     // Safe optimizations
     if (Settings::values.cpu_accuracy == Settings::CPUAccuracy::DebugMode) {
         if (!Settings::values.cpuopt_page_tables) {