Minor decoder optimizations, kernel fixes, cpu backend fixes

.gitmodules

@@ -36,7 +36,7 @@
 	url = https://github.com/skystrife/cpptoml.git
 [submodule "third_party/cxxopts"]
 	path = third_party/cxxopts
-	url = https://github.com/chrisps/cxxopts.git
+	url = https://github.com/jarro2783/cxxopts.git
 [submodule "third_party/SDL2"]
 	path = third_party/SDL2
 	url = https://github.com/libsdl-org/SDL.git

src/xenia/app/emulator_window.cc

@@ -614,7 +614,7 @@ bool EmulatorWindow::Initialize() {
         MenuItem::Type::kString, "Build commit on GitHub...", "F2",
         std::bind(&EmulatorWindow::ShowBuildCommit, this)));
     help_menu->AddChild(MenuItem::Create(
-        MenuItem::Type::kString, "Recent changes on GitHub...", [this]() {
+        MenuItem::Type::kString, "Recent changes on GitHub...", []() {
           LaunchWebBrowser(
               "https://github.com/xenia-project/xenia/compare/" XE_BUILD_COMMIT
               "..." XE_BUILD_BRANCH);
@@ -622,7 +622,7 @@ bool EmulatorWindow::Initialize() {
     help_menu->AddChild(MenuItem::Create(MenuItem::Type::kSeparator));
     help_menu->AddChild(MenuItem::Create(
         MenuItem::Type::kString, "&About...",
-        [this]() { LaunchWebBrowser("https://xenia.jp/about/"); }));
+        []() { LaunchWebBrowser("https://xenia.jp/about/"); }));
   }
   main_menu->AddChild(std::move(help_menu));
 

src/xenia/apu/conversion.h

@@ -71,8 +71,6 @@ inline void sequential_6_BE_to_interleaved_2_LE(float* output,
                                                 const float* input,
                                                 size_t ch_sample_count) {
   assert_true(ch_sample_count % 4 == 0);
-  const uint32_t* in = reinterpret_cast<const uint32_t*>(input);
-  uint32_t* out = reinterpret_cast<uint32_t*>(output);
   const __m128i byte_swap_shuffle =
       _mm_set_epi8(12, 13, 14, 15, 8, 9, 10, 11, 4, 5, 6, 7, 0, 1, 2, 3);
   const __m128 half = _mm_set1_ps(0.5f);

src/xenia/apu/xaudio2/xaudio2_audio_driver.cc

@@ -28,16 +28,16 @@ class XAudio2AudioDriver::VoiceCallback : public api::IXAudio2VoiceCallback {
       : semaphore_(semaphore) {}
   ~VoiceCallback() {}
 
-  void OnStreamEnd() {}
+  void OnStreamEnd() noexcept {}
-  void OnVoiceProcessingPassEnd() {}
+  void OnVoiceProcessingPassEnd() noexcept {}
-  void OnVoiceProcessingPassStart(uint32_t samples_required) {}
+  void OnVoiceProcessingPassStart(uint32_t samples_required) noexcept {}
-  void OnBufferEnd(void* context) {
+  void OnBufferEnd(void* context) noexcept {
     auto ret = semaphore_->Release(1, nullptr);
     assert_true(ret);
   }
-  void OnBufferStart(void* context) {}
+  void OnBufferStart(void* context) noexcept {}
-  void OnLoopEnd(void* context) {}
+  void OnLoopEnd(void* context) noexcept {}
-  void OnVoiceError(void* context, HRESULT result) {}
+  void OnVoiceError(void* context, HRESULT result) noexcept {}
 
  private:
   xe::threading::Semaphore* semaphore_ = nullptr;

src/xenia/base/assert.h

@@ -21,8 +21,11 @@ namespace xe {
                 "bad definition for " #type ": must be " #size " bytes")
 
 // We rely on assert being compiled out in NDEBUG.
+#if defined(NDEBUG)
+#define xenia_assert static_cast<void>
+#else
 #define xenia_assert assert
-
+#endif
 #define __XENIA_EXPAND(x) x
 #define __XENIA_ARGC(...)                                                     \
   __XENIA_EXPAND(__XENIA_ARGC_IMPL(__VA_ARGS__, 15, 14, 13, 12, 11, 10, 9, 8, \

src/xenia/base/cvar.h

@@ -170,8 +170,10 @@ CommandVar<T>::CommandVar(const char* name, T* default_value,
                           const char* description)
     : name_(name),
       default_value_(*default_value),
-      description_(description),
+      current_value_(default_value),
-      current_value_(default_value) {}
+      commandline_value_(),
+      description_(description)
+      {}
 
 template <class T>
 ConfigVar<T>::ConfigVar(const char* name, T* default_value,

src/xenia/base/filesystem_win.cc

@@ -149,7 +149,7 @@ class Win32FileHandle : public FileHandle {
       return false;
     }
   }
-  bool SetLength(size_t length) {
+  bool SetLength(size_t length) override {
     LARGE_INTEGER position;
     position.QuadPart = length;
     if (!SetFilePointerEx(handle_, position, nullptr, SEEK_SET)) {

src/xenia/base/memory.cc

@@ -59,7 +59,7 @@ static void XeCopy16384StreamingAVX(CacheLine* XE_RESTRICT to,
 
   CacheLine* dest4 = to + (NUM_CACHELINES_IN_PAGE * 3);
   CacheLine* src4 = from + (NUM_CACHELINES_IN_PAGE * 3);
-#pragma loop(no_vector)
+  
   for (uint32_t i = 0; i < num_lines_for_8k; ++i) {
     xe::swcache::CacheLine line0, line1, line2, line3;
 
@@ -92,7 +92,6 @@ static void XeCopy16384Movdir64M(CacheLine* XE_RESTRICT to,
 
   CacheLine* dest4 = to + (NUM_CACHELINES_IN_PAGE * 3);
   CacheLine* src4 = from + (NUM_CACHELINES_IN_PAGE * 3);
-#pragma loop(no_vector)
   for (uint32_t i = 0; i < num_lines_for_8k; ++i) {
     _movdir64b(dest1 + i, src1 + i);
     _movdir64b(dest2 + i, src2 + i);

src/xenia/base/memory.h

@@ -620,23 +620,23 @@ static void Prefetch(const void* addr) {
 }
 
 template <>
-void Prefetch<PrefetchTag::Write>(const void* addr) {
+XE_MAYBE_UNUSED void Prefetch<PrefetchTag::Write>(const void* addr) {
   PrefetchW(addr);
 }
 template <>
-void Prefetch<PrefetchTag::Nontemporal>(const void* addr) {
+XE_MAYBE_UNUSED void Prefetch<PrefetchTag::Nontemporal>(const void* addr) {
   PrefetchNTA(addr);
 }
 template <>
-void Prefetch<PrefetchTag::Level3>(const void* addr) {
+XE_MAYBE_UNUSED void Prefetch<PrefetchTag::Level3>(const void* addr) {
   PrefetchL3(addr);
 }
 template <>
-void Prefetch<PrefetchTag::Level2>(const void* addr) {
+XE_MAYBE_UNUSED void Prefetch<PrefetchTag::Level2>(const void* addr) {
   PrefetchL2(addr);
 }
 template <>
-void Prefetch<PrefetchTag::Level1>(const void* addr) {
+XE_MAYBE_UNUSED void Prefetch<PrefetchTag::Level1>(const void* addr) {
   PrefetchL1(addr);
 }
 // todo: does aarch64 have streaming stores/loads?

src/xenia/base/mutex.h

@@ -25,6 +25,7 @@ namespace xe {
 */
 
 class alignas(4096) xe_global_mutex {
+	XE_MAYBE_UNUSED
   char detail[64];
 
  public:
@@ -38,6 +39,7 @@ class alignas(4096) xe_global_mutex {
 using global_mutex_type = xe_global_mutex;
 
 class alignas(64) xe_fast_mutex {
+	XE_MAYBE_UNUSED
   char detail[64];
 
  public:
@@ -62,8 +64,6 @@ class xe_unlikely_mutex {
   ~xe_unlikely_mutex() { mut = 0; }
 
   void lock() {
-    uint32_t lock_expected = 0;
-
     if (XE_LIKELY(_tryget())) {
       return;
     } else {

src/xenia/base/platform.h

@@ -144,9 +144,11 @@
 #define XE_MSVC_OPTIMIZE_SMALL()
 #define XE_MSVC_OPTIMIZE_REVERT()
 #endif
+
 #if XE_COMPILER_HAS_GNU_EXTENSIONS == 1
 #define XE_LIKELY_IF(...) if (XE_LIKELY(__VA_ARGS__))
 #define XE_UNLIKELY_IF(...) if (XE_UNLIKELY(__VA_ARGS__))
+#define XE_MAYBE_UNUSED		__attribute__((unused))
 #else
 #if __cplusplus >= 202002
 #define XE_LIKELY_IF(...) if (!!(__VA_ARGS__)) [[likely]]
@@ -155,6 +157,7 @@
 #define XE_LIKELY_IF(...) if (!!(__VA_ARGS__))
 #define XE_UNLIKELY_IF(...) if (!!(__VA_ARGS__))
 #endif
+#define XE_MAYBE_UNUSED		
 #endif
 // only use __restrict if MSVC, for clang/gcc we can use -fstrict-aliasing which
 // acts as __restrict across the board todo: __restrict is part of the type

src/xenia/base/ring_buffer.cc

@@ -78,7 +78,9 @@ size_t RingBuffer::Read(uint8_t* buffer, size_t _count) {
   if (read_offset_ < write_offset_) {
     assert_true(read_offset_ + count <= write_offset_);
   } else if (read_offset_ + count >= capacity_) {
+    XE_MAYBE_UNUSED
     ring_size_t left_half = capacity_ - read_offset_;
+
     assert_true(count - left_half <= write_offset_);
   }
 
@@ -107,6 +109,7 @@ size_t RingBuffer::Write(const uint8_t* buffer, size_t _count) {
   if (write_offset_ < read_offset_) {
     assert_true(write_offset_ + count <= read_offset_);
   } else if (write_offset_ + count >= capacity_) {
+    XE_MAYBE_UNUSED
     size_t left_half = capacity_ - write_offset_;
     assert_true(count - left_half <= read_offset_);
   }

src/xenia/base/ring_buffer.h

@@ -68,7 +68,6 @@ class RingBuffer {
     ring_size_t offset_delta = write_offs - read_offs;
     ring_size_t wrap_read_count = (cap - read_offs) + write_offs;
 
-    ring_size_t comparison_value = read_offs <= write_offs;
 
     if (XE_LIKELY(read_offs <= write_offs)) {
       return offset_delta;  // will be 0 if they are equal, semantically

src/xenia/base/split_map.h

@@ -67,8 +67,6 @@ class split_map {
   void InsertAt(TKey k, TValue v, uint32_t idx) {
     uint32_t old_size = size();
 
-    bool needs_shiftup = idx != old_size;
-
     values_.insert(values_.begin() + idx, v);
     keys_.insert(keys_.begin() + idx, k);
   }

src/xenia/base/threading_win.cc

@@ -117,7 +117,7 @@ void set_name(const std::string_view name) {
 
 // checked ntoskrnl, it does not modify delay, so we can place this as a
 // constant and avoid creating a stack variable
-static const LARGE_INTEGER sleepdelay0_for_maybeyield{0LL};
+static const LARGE_INTEGER sleepdelay0_for_maybeyield{{0LL}};
 
 void MaybeYield() {
 #if 0
@@ -314,7 +314,8 @@ class Win32Event : public Win32Handle<Event> {
   }
 #endif
 
-  EventInfo Query() { EventInfo result{};
+  EventInfo Query() override {
+    EventInfo result{};
     NtQueryEventPointer.invoke(handle_, 0, &result, sizeof(EventInfo), nullptr);
     return result;
   }
@@ -429,7 +430,7 @@ class Win32Timer : public Win32Handle<Timer> {
   }
   bool SetRepeatingAt(GClock_::time_point due_time,
                       std::chrono::milliseconds period,
-                      std::function<void()> opt_callback = nullptr) {
+                      std::function<void()> opt_callback = nullptr) override {
     return SetRepeatingAt(date::clock_cast<WClock_>(due_time), period,
                           std::move(opt_callback));
   }

src/xenia/cpu/backend/x64/x64_code_cache.h

@@ -93,7 +93,8 @@ class X64CodeCache : public CodeCache {
   // This is picked to be high enough to cover whatever we can reasonably
   // expect. If we hit issues with this it probably means some corner case
   // in analysis triggering.
-  static const size_t kMaximumFunctionCount = 100000;
+  //chrispy: raised this, some games that were compiled with low optimization levels can exceed this
+  static const size_t kMaximumFunctionCount = 1000000;
 
   struct UnwindReservation {
     size_t data_size = 0;

src/xenia/cpu/backend/x64/x64_code_cache_win.cc

@@ -209,7 +209,16 @@ bool Win32X64CodeCache::Initialize() {
 
 Win32X64CodeCache::UnwindReservation
 Win32X64CodeCache::RequestUnwindReservation(uint8_t* entry_address) {
+#if defined(NDEBUG)
+  if (unwind_table_count_ >= kMaximumFunctionCount) {
+    // we should not just be ignoring this in release if it happens
+    xe::FatalError(
+        "Unwind table count (unwind_table_count_) exceeded maximum! Please report this to "
+        "Xenia/Canary developers");
+  }
+#else
   assert_false(unwind_table_count_ >= kMaximumFunctionCount);
+#endif
   UnwindReservation unwind_reservation;
   unwind_reservation.data_size = xe::round_up(kUnwindInfoSize, 16);
   unwind_reservation.table_slot = unwind_table_count_++;

src/xenia/cpu/backend/x64/x64_emitter.cc

@@ -46,10 +46,6 @@ DEFINE_bool(ignore_undefined_externs, true,
 DEFINE_bool(emit_source_annotations, false,
             "Add extra movs and nops to make disassembly easier to read.",
             "CPU");
-DEFINE_bool(resolve_rel32_guest_calls, true,
-            "Experimental optimization, directly call already resolved "
-            "functions via x86 rel32 call/jmp",
-            "CPU");
 
 DEFINE_bool(enable_incorrect_roundingmode_behavior, false,
             "Disables the FPU/VMX MXCSR sharing workaround, potentially "
@@ -78,7 +74,6 @@ using namespace xe::literals;
 
 static const size_t kMaxCodeSize = 1_MiB;
 
-static const size_t kStashOffset = 32;
 // static const size_t kStashOffsetHigh = 32 + 32;
 
 const uint32_t X64Emitter::gpr_reg_map_[X64Emitter::GPR_COUNT] = {
@@ -141,55 +136,6 @@ bool X64Emitter::Emit(GuestFunction* function, HIRBuilder* builder,
 
   return true;
 }
-#pragma pack(push, 1)
-struct RGCEmitted {
-  uint8_t ff_;
-  uint32_t rgcid_;
-};
-#pragma pack(pop)
-
-#if 0
-void X64Emitter::InjectCallAddresses(void* new_execute_address) {
-  for (auto&& callsite : call_sites_) {
-    RGCEmitted* hunter = (RGCEmitted*)new_execute_address;
-    while (hunter->ff_ != 0xFF || hunter->rgcid_ != callsite.offset_) {
-      hunter =
-          reinterpret_cast<RGCEmitted*>(reinterpret_cast<char*>(hunter) + 1);
-    }
-
-    hunter->ff_ = callsite.is_jump_ ? 0xE9 : 0xE8;
-    hunter->rgcid_ =
-        static_cast<uint32_t>(static_cast<intptr_t>(callsite.destination_) -
-                              reinterpret_cast<intptr_t>(hunter + 1));
-  }
-}
-
-#else
-void X64Emitter::InjectCallAddresses(void* new_execute_address) {
-#if 0
-	RGCEmitted* hunter = (RGCEmitted*)new_execute_address;
-
-  std::map<uint32_t, ResolvableGuestCall*> id_to_rgc{};
-
-  for (auto&& callsite : call_sites_) {
-    id_to_rgc[callsite.offset_] = &callsite;
-  }
-#else
-  RGCEmitted* hunter = (RGCEmitted*)new_execute_address;
-  for (auto&& callsite : call_sites_) {
-    while (hunter->ff_ != 0xFF || hunter->rgcid_ != callsite.offset_) {
-      hunter =
-          reinterpret_cast<RGCEmitted*>(reinterpret_cast<char*>(hunter) + 1);
-    }
-
-    hunter->ff_ = callsite.is_jump_ ? 0xE9 : 0xE8;
-    hunter->rgcid_ =
-        static_cast<uint32_t>(static_cast<intptr_t>(callsite.destination_) -
-                              reinterpret_cast<intptr_t>(hunter + 1));
-  }
-#endif
-}
-#endif
 void* X64Emitter::Emplace(const EmitFunctionInfo& func_info,
                           GuestFunction* function) {
   // To avoid changing xbyak, we do a switcharoo here.
@@ -207,10 +153,6 @@ void* X64Emitter::Emplace(const EmitFunctionInfo& func_info,
   if (function) {
     code_cache_->PlaceGuestCode(function->address(), top_, func_info, function,
                                 new_execute_address, new_write_address);
-
-    if (cvars::resolve_rel32_guest_calls) {
-      InjectCallAddresses(new_execute_address);
-    }
   } else {
     code_cache_->PlaceHostCode(0, top_, func_info, new_execute_address,
                                new_write_address);
@@ -219,7 +161,6 @@ void* X64Emitter::Emplace(const EmitFunctionInfo& func_info,
   ready();
   top_ = old_address;
   reset();
-  call_sites_.clear();
   tail_code_.clear();
   for (auto&& cached_label : label_cache_) {
     delete cached_label;
@@ -336,7 +277,7 @@ bool X64Emitter::Emit(HIRBuilder* builder, EmitFunctionInfo& func_info) {
     // Mark block labels.
     auto label = block->label_head;
     while (label) {
-      L(label->name);
+      L(std::to_string(label->id));
       label = label->next;
     }
 
@@ -418,7 +359,6 @@ void X64Emitter::EmitProfilerEpilogue() {
     //  actually... lets just try without atomics lol
     // lock();
     add(qword[r10], rdx);
-
   }
 #endif
 }
@@ -534,44 +474,23 @@ void X64Emitter::Call(const hir::Instr* instr, GuestFunction* function) {
   auto fn = static_cast<X64Function*>(function);
   // Resolve address to the function to call and store in rax.
 
-  if (cvars::resolve_rel32_guest_calls && fn->machine_code()) {
+  if (fn->machine_code()) {
-    ResolvableGuestCall rgc;
-    rgc.destination_ = uint32_t(uint64_t(fn->machine_code()));
-    rgc.offset_ = current_rgc_id_;
-    current_rgc_id_++;
-
     if (!(instr->flags & hir::CALL_TAIL)) {
       mov(rcx, qword[rsp + StackLayout::GUEST_CALL_RET_ADDR]);
 
-      db(0xFF);
+      call((void*)fn->machine_code());
-      rgc.is_jump_ = false;
-
-      dd(rgc.offset_);
 
     } else {
       // tail call
       EmitTraceUserCallReturn();
-
+      EmitProfilerEpilogue();
-      rgc.is_jump_ = true;
       // Pass the callers return address over.
       mov(rcx, qword[rsp + StackLayout::GUEST_RET_ADDR]);
 
       add(rsp, static_cast<uint32_t>(stack_size()));
-      db(0xFF);
+      jmp((void*)fn->machine_code(), T_NEAR);
-      dd(rgc.offset_);
     }
-    call_sites_.push_back(rgc);
     return;
-  }
-
-  if (fn->machine_code()) {
-    // TODO(benvanik): is it worth it to do this? It removes the need for
-    // a ResolveFunction call, but makes the table less useful.
-    assert_zero(uint64_t(fn->machine_code()) & 0xFFFFFFFF00000000);
-    // todo: this should be changed so that we can actually do a call to
-    // fn->machine_code. the code will be emitted near us, so 32 bit rel jmp
-    // should be possible
-    mov(eax, uint32_t(uint64_t(fn->machine_code())));
   } else if (code_cache_->has_indirection_table()) {
     // Load the pointer to the indirection table maintained in X64CodeCache.
     // The target dword will either contain the address of the generated code
@@ -1017,7 +936,10 @@ static const vec128_t xmm_consts[] = {
     /*XMMSTVLShuffle*/
     v128_setr_bytes(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15),
     /* XMMSTVRSwapMask*/
-    vec128b((uint8_t)0x83)};
+    vec128b((uint8_t)0x83), /*XMMVSRShlByteshuf*/
+    v128_setr_bytes(13, 14, 15, 8, 9, 10, 11, 4, 5, 6, 7, 0, 1, 2, 3, 0x80),
+    // XMMVSRMask
+    vec128b(1)};
 
 void* X64Emitter::FindByteConstantOffset(unsigned bytevalue) {
   for (auto& vec : xmm_consts) {

src/xenia/cpu/backend/x64/x64_emitter.h

@@ -66,7 +66,7 @@ enum class SimdDomain : uint32_t {
 };
 
 enum class MXCSRMode : uint32_t { Unknown, Fpu, Vmx };
-
+XE_MAYBE_UNUSED
 static SimdDomain PickDomain2(SimdDomain dom1, SimdDomain dom2) {
   if (dom1 == dom2) {
     return dom1;
@@ -172,7 +172,9 @@ enum XmmConst {
   XMMLVLShuffle,
   XMMLVRCmp16,
   XMMSTVLShuffle,
-  XMMSTVRSwapMask  // swapwordmask with bit 7 set
+  XMMSTVRSwapMask,  // swapwordmask with bit 7 set
+  XMMVSRShlByteshuf,
+  XMMVSRMask
 
 };
 using amdfx::xopcompare_e;
@@ -190,13 +192,6 @@ class XbyakAllocator : public Xbyak::Allocator {
   virtual bool useProtect() const { return false; }
 };
 
-class ResolvableGuestCall {
- public:
-  bool is_jump_;
-  uintptr_t destination_;
-  // rgcid
-  unsigned offset_;
-};
 class X64Emitter;
 using TailEmitCallback = std::function<void(X64Emitter& e, Xbyak::Label& lbl)>;
 struct TailEmitter {
@@ -220,7 +215,6 @@ class X64Emitter : public Xbyak::CodeGenerator {
             uint32_t debug_info_flags, FunctionDebugInfo* debug_info,
             void** out_code_address, size_t* out_code_size,
             std::vector<SourceMapEntry>* out_source_map);
-  void InjectCallAddresses(void* new_execute_addr);
 
  public:
   // Reserved:  rsp, rsi, rdi
@@ -230,7 +224,7 @@ class X64Emitter : public Xbyak::CodeGenerator {
   //            xmm4-xmm15 (save to get xmm3)
   static const int GPR_COUNT = 7;
   static const int XMM_COUNT = 12;
-
+  static constexpr size_t kStashOffset = 32;
   static void SetupReg(const hir::Value* v, Xbyak::Reg8& r) {
     auto idx = gpr_reg_map_[v->reg.index];
     r = Xbyak::Reg8(idx);
@@ -410,8 +404,6 @@ class X64Emitter : public Xbyak::CodeGenerator {
 
   static const uint32_t gpr_reg_map_[GPR_COUNT];
   static const uint32_t xmm_reg_map_[XMM_COUNT];
-  uint32_t current_rgc_id_ = 0xEEDDF00F;
-  std::vector<ResolvableGuestCall> call_sites_;
   /*
     set to true if the low 32 bits of membase == 0.
     only really advantageous if you are storing 32 bit 0 to a displaced address,

src/xenia/cpu/backend/x64/x64_op.h

@@ -398,21 +398,22 @@ struct I<OPCODE, DEST, SRC1, SRC2, SRC3> : DestField<DEST> {
 };
 
 template <typename T>
+XE_MAYBE_UNUSED
 static const T GetTempReg(X64Emitter& e);
 template <>
-const Reg8 GetTempReg<Reg8>(X64Emitter& e) {
+XE_MAYBE_UNUSED const Reg8 GetTempReg<Reg8>(X64Emitter& e) {
   return e.al;
 }
 template <>
-const Reg16 GetTempReg<Reg16>(X64Emitter& e) {
+XE_MAYBE_UNUSED const Reg16 GetTempReg<Reg16>(X64Emitter& e) {
   return e.ax;
 }
 template <>
-const Reg32 GetTempReg<Reg32>(X64Emitter& e) {
+XE_MAYBE_UNUSED const Reg32 GetTempReg<Reg32>(X64Emitter& e) {
   return e.eax;
 }
 template <>
-const Reg64 GetTempReg<Reg64>(X64Emitter& e) {
+XE_MAYBE_UNUSED const Reg64 GetTempReg<Reg64>(X64Emitter& e) {
   return e.rax;
 }
 

src/xenia/cpu/backend/x64/x64_seq_control.cc

@@ -25,46 +25,46 @@ static void EmitFusedBranch(X64Emitter& e, const T& i) {
   bool valid = i.instr->prev && i.instr->prev->dest == i.src1.value;
   auto opcode = valid ? i.instr->prev->opcode->num : -1;
   if (valid) {
-    auto name = i.src2.value->name;
+    std::string name = i.src2.value->GetIdString();
     switch (opcode) {
       case OPCODE_COMPARE_EQ:
-        e.je(name, e.T_NEAR);
+        e.je(std::move(name), e.T_NEAR);
         break;
       case OPCODE_COMPARE_NE:
-        e.jne(name, e.T_NEAR);
+        e.jne(std::move(name), e.T_NEAR);
         break;
       case OPCODE_COMPARE_SLT:
-        e.jl(name, e.T_NEAR);
+        e.jl(std::move(name), e.T_NEAR);
         break;
       case OPCODE_COMPARE_SLE:
-        e.jle(name, e.T_NEAR);
+        e.jle(std::move(name), e.T_NEAR);
         break;
       case OPCODE_COMPARE_SGT:
-        e.jg(name, e.T_NEAR);
+        e.jg(std::move(name), e.T_NEAR);
         break;
       case OPCODE_COMPARE_SGE:
-        e.jge(name, e.T_NEAR);
+        e.jge(std::move(name), e.T_NEAR);
         break;
       case OPCODE_COMPARE_ULT:
-        e.jb(name, e.T_NEAR);
+        e.jb(std::move(name), e.T_NEAR);
         break;
       case OPCODE_COMPARE_ULE:
-        e.jbe(name, e.T_NEAR);
+        e.jbe(std::move(name), e.T_NEAR);
         break;
       case OPCODE_COMPARE_UGT:
-        e.ja(name, e.T_NEAR);
+        e.ja(std::move(name), e.T_NEAR);
         break;
       case OPCODE_COMPARE_UGE:
-        e.jae(name, e.T_NEAR);
+        e.jae(std::move(name), e.T_NEAR);
         break;
       default:
         e.test(i.src1, i.src1);
-        e.jnz(name, e.T_NEAR);
+        e.jnz(std::move(name), e.T_NEAR);
         break;
     }
   } else {
     e.test(i.src1, i.src1);
-    e.jnz(i.src2.value->name, e.T_NEAR);
+    e.jnz(i.src2.value->GetIdString(), e.T_NEAR);
   }
 }
 // ============================================================================
@@ -490,7 +490,7 @@ EMITTER_OPCODE_TABLE(OPCODE_SET_RETURN_ADDRESS, SET_RETURN_ADDRESS);
 // ============================================================================
 struct BRANCH : Sequence<BRANCH, I<OPCODE_BRANCH, VoidOp, LabelOp>> {
   static void Emit(X64Emitter& e, const EmitArgType& i) {
-    e.jmp(i.src1.value->name, e.T_NEAR);
+    e.jmp(i.src1.value->GetIdString(), e.T_NEAR);
   }
 };
 EMITTER_OPCODE_TABLE(OPCODE_BRANCH, BRANCH);
@@ -534,7 +534,7 @@ struct BRANCH_TRUE_F32
     Xmm input = GetInputRegOrConstant(e, i.src1, e.xmm0);
     e.vmovd(e.eax, input);
     e.test(e.eax, e.eax);
-    e.jnz(i.src2.value->name, e.T_NEAR);
+    e.jnz(i.src2.value->GetIdString(), e.T_NEAR);
   }
 };
 struct BRANCH_TRUE_F64
@@ -543,7 +543,7 @@ struct BRANCH_TRUE_F64
     Xmm input = GetInputRegOrConstant(e, i.src1, e.xmm0);
     e.vmovq(e.rax, input);
     e.test(e.rax, e.rax);
-    e.jnz(i.src2.value->name, e.T_NEAR);
+    e.jnz(i.src2.value->GetIdString(), e.T_NEAR);
   }
 };
 EMITTER_OPCODE_TABLE(OPCODE_BRANCH_TRUE, BRANCH_TRUE_I8, BRANCH_TRUE_I16,
@@ -557,7 +557,7 @@ struct BRANCH_FALSE_I8
     : Sequence<BRANCH_FALSE_I8, I<OPCODE_BRANCH_FALSE, VoidOp, I8Op, LabelOp>> {
   static void Emit(X64Emitter& e, const EmitArgType& i) {
     e.test(i.src1, i.src1);
-    e.jz(i.src2.value->name, e.T_NEAR);
+    e.jz(i.src2.value->GetIdString(), e.T_NEAR);
   }
 };
 struct BRANCH_FALSE_I16
@@ -565,7 +565,7 @@ struct BRANCH_FALSE_I16
                I<OPCODE_BRANCH_FALSE, VoidOp, I16Op, LabelOp>> {
   static void Emit(X64Emitter& e, const EmitArgType& i) {
     e.test(i.src1, i.src1);
-    e.jz(i.src2.value->name, e.T_NEAR);
+    e.jz(i.src2.value->GetIdString(), e.T_NEAR);
   }
 };
 struct BRANCH_FALSE_I32
@@ -573,7 +573,7 @@ struct BRANCH_FALSE_I32
                I<OPCODE_BRANCH_FALSE, VoidOp, I32Op, LabelOp>> {
   static void Emit(X64Emitter& e, const EmitArgType& i) {
     e.test(i.src1, i.src1);
-    e.jz(i.src2.value->name, e.T_NEAR);
+    e.jz(i.src2.value->GetIdString(), e.T_NEAR);
   }
 };
 struct BRANCH_FALSE_I64
@@ -581,7 +581,7 @@ struct BRANCH_FALSE_I64
                I<OPCODE_BRANCH_FALSE, VoidOp, I64Op, LabelOp>> {
   static void Emit(X64Emitter& e, const EmitArgType& i) {
     e.test(i.src1, i.src1);
-    e.jz(i.src2.value->name, e.T_NEAR);
+    e.jz(i.src2.value->GetIdString(), e.T_NEAR);
   }
 };
 struct BRANCH_FALSE_F32
@@ -591,7 +591,7 @@ struct BRANCH_FALSE_F32
     Xmm input = GetInputRegOrConstant(e, i.src1, e.xmm0);
     e.vmovd(e.eax, input);
     e.test(e.eax, e.eax);
-    e.jz(i.src2.value->name, e.T_NEAR);
+    e.jz(i.src2.value->GetIdString(), e.T_NEAR);
   }
 };
 struct BRANCH_FALSE_F64
@@ -601,7 +601,7 @@ struct BRANCH_FALSE_F64
     Xmm input = GetInputRegOrConstant(e, i.src1, e.xmm0);
     e.vmovq(e.rax, input);
     e.test(e.rax, e.rax);
-    e.jz(i.src2.value->name, e.T_NEAR);
+    e.jz(i.src2.value->GetIdString(), e.T_NEAR);
   }
 };
 EMITTER_OPCODE_TABLE(OPCODE_BRANCH_FALSE, BRANCH_FALSE_I8, BRANCH_FALSE_I16,

src/xenia/cpu/backend/x64/x64_seq_vector.cc

@@ -805,22 +805,7 @@ EMITTER_OPCODE_TABLE(OPCODE_VECTOR_SUB, VECTOR_SUB);
 // ============================================================================
 // OPCODE_VECTOR_SHL
 // ============================================================================
-template <typename T, std::enable_if_t<std::is_integral<T>::value, int> = 0>
-static __m128i EmulateVectorShl(void*, __m128i src1, __m128i src2) {
-  alignas(16) T value[16 / sizeof(T)];
-  alignas(16) T shamt[16 / sizeof(T)];
 
-  // Load SSE registers into a C array.
-  _mm_store_si128(reinterpret_cast<__m128i*>(value), src1);
-  _mm_store_si128(reinterpret_cast<__m128i*>(shamt), src2);
-
-  for (size_t i = 0; i < (16 / sizeof(T)); ++i) {
-    value[i] = value[i] << (shamt[i] & ((sizeof(T) * 8) - 1));
-  }
-
-  // Store result and return it.
-  return _mm_load_si128(reinterpret_cast<__m128i*>(value));
-}
 static XmmConst GetShiftmaskForType(unsigned typ) {
   if (typ == INT8_TYPE) {
     return XMMXOPByteShiftMask;
@@ -914,28 +899,14 @@ struct VECTOR_SHL_V128
           }
         }
         if (all_same) {
-          // mul by two
+          e.vpmovzxbw(e.ymm0, i.src1);
-          /*if (seenvalue == 1) {
+          e.vpsllw(e.ymm0, e.ymm0, seenvalue);
-            e.vpaddb(i.dest, i.src1, i.src1);
+          e.vextracti128(e.xmm1, e.ymm0, 1);
-          } else if (seenvalue == 2) {
-            e.vpaddb(i.dest, i.src1, i.src1);
-            e.vpaddb(i.dest, i.dest, i.dest);
-          } else if (seenvalue == 3) {
-            // mul by 8
-            e.vpaddb(i.dest, i.src1, i.src1);
-            e.vpaddb(i.dest, i.dest, i.dest);
-            e.vpaddb(i.dest, i.dest, i.dest);
-          } else*/
-          {
-            e.vpmovzxbw(e.ymm0, i.src1);
-            e.vpsllw(e.ymm0, e.ymm0, seenvalue);
-            e.vextracti128(e.xmm1, e.ymm0, 1);
 
-            e.vpshufb(e.xmm0, e.xmm0, e.GetXmmConstPtr(XMMShortsToBytes));
+          e.vpshufb(e.xmm0, e.xmm0, e.GetXmmConstPtr(XMMShortsToBytes));
-            e.vpshufb(e.xmm1, e.xmm1, e.GetXmmConstPtr(XMMShortsToBytes));
+          e.vpshufb(e.xmm1, e.xmm1, e.GetXmmConstPtr(XMMShortsToBytes));
-            e.vpunpcklqdq(i.dest, e.xmm0, e.xmm1);
+          e.vpunpcklqdq(i.dest, e.xmm0, e.xmm1);
-            return;
+          return;
-          }
 
         } else {
           e.LoadConstantXmm(e.xmm2, constmask);
@@ -966,14 +937,41 @@ struct VECTOR_SHL_V128
         }
       }
     }
-    if (i.src2.is_constant) {
+
-      e.lea(e.GetNativeParam(1), e.StashConstantXmm(1, i.src2.constant()));
+    unsigned stack_offset_src1 = StackLayout::GUEST_SCRATCH;
+    unsigned stack_offset_src2 = StackLayout::GUEST_SCRATCH + 16;
+
+    if (i.src1.is_constant) {
+      e.StashConstantXmm(0, i.src1.constant());
+      stack_offset_src1 = X64Emitter::kStashOffset;
     } else {
-      e.lea(e.GetNativeParam(1), e.StashXmm(1, i.src2));
+      e.vmovdqa(e.ptr[e.rsp + stack_offset_src1], i.src1);
     }
-    e.lea(e.GetNativeParam(0), e.StashXmm(0, i.src1));
+    if (i.src2.is_constant) {
-    e.CallNativeSafe(reinterpret_cast<void*>(EmulateVectorShl<uint8_t>));
+      e.StashConstantXmm(1, i.src2.constant());
-    e.vmovaps(i.dest, e.xmm0);
+      stack_offset_src2 = X64Emitter::kStashOffset + 16;
+    } else {
+      e.vmovdqa(e.ptr[e.rsp + stack_offset_src2], i.src2);
+    }
+
+    Xbyak::Label looper;
+
+    e.xor_(e.edx, e.edx);
+
+    e.L(looper);
+    e.movzx(e.ecx, e.byte[e.rsp + stack_offset_src2 + e.rdx]);
+
+    e.shl(e.byte[e.rsp + stack_offset_src1 + e.rdx], e.cl);
+
+    if (e.IsFeatureEnabled(kX64FlagsIndependentVars)) {
+      e.inc(e.edx);
+    } else {
+      e.add(e.edx, 1);
+    }
+
+    e.cmp(e.edx, 16);
+    e.jnz(looper);
+    e.vmovdqa(i.dest, e.byte[e.rsp + stack_offset_src1]);
   }
   static void EmitInt16(X64Emitter& e, const EmitArgType& i) {
     Xmm src1;
@@ -1022,14 +1020,32 @@ struct VECTOR_SHL_V128
 
     // TODO(benvanik): native version (with shift magic).
     e.L(emu);
+
+    unsigned stack_offset_src1 = StackLayout::GUEST_SCRATCH;
+    unsigned stack_offset_src2 = StackLayout::GUEST_SCRATCH + 16;
+
+    e.vmovdqa(e.ptr[e.rsp + stack_offset_src1], src1);
     if (i.src2.is_constant) {
-      e.lea(e.GetNativeParam(1), e.StashConstantXmm(1, i.src2.constant()));
+      e.StashConstantXmm(1, i.src2.constant());
+      stack_offset_src2 = X64Emitter::kStashOffset + 16;
     } else {
-      e.lea(e.GetNativeParam(1), e.StashXmm(1, i.src2));
+      e.vmovdqa(e.ptr[e.rsp + stack_offset_src2], i.src2);
     }
-    e.lea(e.GetNativeParam(0), e.StashXmm(0, src1));
+
-    e.CallNativeSafe(reinterpret_cast<void*>(EmulateVectorShl<uint16_t>));
+    Xbyak::Label looper;
-    e.vmovaps(i.dest, e.xmm0);
+
+    e.xor_(e.edx, e.edx);
+
+    e.L(looper);
+    e.movzx(e.ecx, e.word[e.rsp + stack_offset_src2 + e.rdx]);
+
+    e.shl(e.word[e.rsp + stack_offset_src1 + e.rdx], e.cl);
+
+    e.add(e.edx, 2);
+
+    e.cmp(e.edx, 16);
+    e.jnz(looper);
+    e.vmovdqa(i.dest, e.byte[e.rsp + stack_offset_src1]);
 
     e.L(end);
   }
@@ -1098,14 +1114,32 @@ struct VECTOR_SHL_V128
 
       // TODO(benvanik): native version (with shift magic).
       e.L(emu);
+
+      unsigned stack_offset_src1 = StackLayout::GUEST_SCRATCH;
+      unsigned stack_offset_src2 = StackLayout::GUEST_SCRATCH + 16;
+
+      e.vmovdqa(e.ptr[e.rsp + stack_offset_src1], src1);
       if (i.src2.is_constant) {
-        e.lea(e.GetNativeParam(1), e.StashConstantXmm(1, i.src2.constant()));
+        e.StashConstantXmm(1, i.src2.constant());
+        stack_offset_src2 = X64Emitter::kStashOffset + 16;
       } else {
-        e.lea(e.GetNativeParam(1), e.StashXmm(1, i.src2));
+        e.vmovdqa(e.ptr[e.rsp + stack_offset_src2], i.src2);
       }
-      e.lea(e.GetNativeParam(0), e.StashXmm(0, src1));
+
-      e.CallNativeSafe(reinterpret_cast<void*>(EmulateVectorShl<uint32_t>));
+      Xbyak::Label looper;
-      e.vmovaps(i.dest, e.xmm0);
+
+      e.xor_(e.edx, e.edx);
+
+      e.L(looper);
+      e.mov(e.ecx, e.dword[e.rsp + stack_offset_src2 + e.rdx]);
+
+      e.shl(e.dword[e.rsp + stack_offset_src1 + e.rdx], e.cl);
+
+      e.add(e.edx, 4);
+
+      e.cmp(e.edx, 16);
+      e.jnz(looper);
+      e.vmovdqa(i.dest, e.byte[e.rsp + stack_offset_src1]);
 
       e.L(end);
     }
@@ -1116,22 +1150,6 @@ EMITTER_OPCODE_TABLE(OPCODE_VECTOR_SHL, VECTOR_SHL_V128);
 // ============================================================================
 // OPCODE_VECTOR_SHR
 // ============================================================================
-template <typename T, std::enable_if_t<std::is_integral<T>::value, int> = 0>
-static __m128i EmulateVectorShr(void*, __m128i src1, __m128i src2) {
-  alignas(16) T value[16 / sizeof(T)];
-  alignas(16) T shamt[16 / sizeof(T)];
-
-  // Load SSE registers into a C array.
-  _mm_store_si128(reinterpret_cast<__m128i*>(value), src1);
-  _mm_store_si128(reinterpret_cast<__m128i*>(shamt), src2);
-
-  for (size_t i = 0; i < (16 / sizeof(T)); ++i) {
-    value[i] = value[i] >> (shamt[i] & ((sizeof(T) * 8) - 1));
-  }
-
-  // Store result and return it.
-  return _mm_load_si128(reinterpret_cast<__m128i*>(value));
-}
 
 struct VECTOR_SHR_V128
     : Sequence<VECTOR_SHR_V128, I<OPCODE_VECTOR_SHR, V128Op, V128Op, V128Op>> {
@@ -1179,34 +1197,63 @@ struct VECTOR_SHR_V128
   }
 
   static void EmitInt8(X64Emitter& e, const EmitArgType& i) {
-    // TODO(benvanik): native version (with shift magic).
+    if (i.src2.is_constant && e.IsFeatureEnabled(kX64EmitGFNI)) {
-    if (i.src2.is_constant) {
+      const auto& shamt = i.src2.constant();
-      if (e.IsFeatureEnabled(kX64EmitGFNI)) {
+      bool all_same = true;
-        const auto& shamt = i.src2.constant();
+      for (size_t n = 0; n < 16 - n; ++n) {
-        bool all_same = true;
+        if (shamt.u8[n] != shamt.u8[n + 1]) {
-        for (size_t n = 0; n < 16 - n; ++n) {
+          all_same = false;
-          if (shamt.u8[n] != shamt.u8[n + 1]) {
+          break;
-            all_same = false;
-            break;
-          }
-        }
-        if (all_same) {
-          // Every count is the same, so we can use gf2p8affineqb.
-          const uint8_t shift_amount = shamt.u8[0] & 0b111;
-          const uint64_t shift_matrix = UINT64_C(0x0102040810204080)
-                                        << (shift_amount * 8);
-          e.vgf2p8affineqb(i.dest, i.src1,
-                           e.StashConstantXmm(0, vec128q(shift_matrix)), 0);
-          return;
         }
       }
-      e.lea(e.GetNativeParam(1), e.StashConstantXmm(1, i.src2.constant()));
+      if (all_same) {
-    } else {
+        // Every count is the same, so we can use gf2p8affineqb.
-      e.lea(e.GetNativeParam(1), e.StashXmm(1, i.src2));
+        const uint8_t shift_amount = shamt.u8[0] & 0b111;
+        const uint64_t shift_matrix = UINT64_C(0x0102040810204080)
+                                      << (shift_amount * 8);
+        e.vgf2p8affineqb(i.dest, i.src1,
+                         e.StashConstantXmm(0, vec128q(shift_matrix)), 0);
+        return;
+      }
     }
-    e.lea(e.GetNativeParam(0), e.StashXmm(0, i.src1));
+    unsigned stack_offset_src1 = StackLayout::GUEST_SCRATCH;
-    e.CallNativeSafe(reinterpret_cast<void*>(EmulateVectorShr<uint8_t>));
+    unsigned stack_offset_src2 = StackLayout::GUEST_SCRATCH + 16;
-    e.vmovaps(i.dest, e.xmm0);
+
+    if (i.src1.is_constant) {
+      e.StashConstantXmm(0, i.src1.constant());
+      stack_offset_src1 = X64Emitter::kStashOffset;
+    } else {
+      e.vmovdqa(e.ptr[e.rsp + stack_offset_src1], i.src1);
+    }
+    if (i.src2.is_constant) {
+      e.StashConstantXmm(1, i.src2.constant());
+      stack_offset_src2 = X64Emitter::kStashOffset + 16;
+    } else {
+      e.vmovdqa(e.ptr[e.rsp + stack_offset_src2], i.src2);
+    }
+
+    Xbyak::Label looper;
+
+    e.xor_(e.edx, e.edx);
+
+    e.L(looper);
+    // movzx is to eliminate any possible dep on previous value of rcx at start
+    // of loop
+    e.movzx(e.ecx, e.byte[e.rsp + stack_offset_src2 + e.rdx]);
+    // maybe using a memory operand as the left side isn't the best idea lol,
+    // still better than callnativesafe though agners docs have no timing info
+    // on shx [m], cl so shrug
+    e.shr(e.byte[e.rsp + stack_offset_src1 + e.rdx], e.cl);
+
+    if (e.IsFeatureEnabled(kX64FlagsIndependentVars)) {
+      e.inc(e.edx);
+    } else {
+      e.add(e.edx, 1);
+    }
+
+    e.cmp(e.edx, 16);
+    e.jnz(looper);
+    e.vmovdqa(i.dest, e.byte[e.rsp + stack_offset_src1]);
   }
 
   static void EmitInt16(X64Emitter& e, const EmitArgType& i) {
@@ -1248,14 +1295,38 @@ struct VECTOR_SHR_V128
 
     // TODO(benvanik): native version (with shift magic).
     e.L(emu);
-    if (i.src2.is_constant) {
+
-      e.lea(e.GetNativeParam(1), e.StashConstantXmm(1, i.src2.constant()));
+    unsigned stack_offset_src1 = StackLayout::GUEST_SCRATCH;
+    unsigned stack_offset_src2 = StackLayout::GUEST_SCRATCH + 16;
+    if (i.src1.is_constant) {
+      e.StashConstantXmm(0, i.src1.constant());
+      stack_offset_src1 = X64Emitter::kStashOffset;
+
     } else {
-      e.lea(e.GetNativeParam(1), e.StashXmm(1, i.src2));
+      e.vmovdqa(e.ptr[e.rsp + stack_offset_src1], i.src1);
     }
-    e.lea(e.GetNativeParam(0), e.StashXmm(0, i.src1));
+
-    e.CallNativeSafe(reinterpret_cast<void*>(EmulateVectorShr<uint16_t>));
+    if (i.src2.is_constant) {
-    e.vmovaps(i.dest, e.xmm0);
+      e.StashConstantXmm(1, i.src2.constant());
+      stack_offset_src2 = X64Emitter::kStashOffset + 16;
+    } else {
+      e.vmovdqa(e.ptr[e.rsp + stack_offset_src2], i.src2);
+    }
+
+    Xbyak::Label looper;
+
+    e.xor_(e.edx, e.edx);
+
+    e.L(looper);
+    e.movzx(e.ecx, e.word[e.rsp + stack_offset_src2 + e.rdx]);
+
+    e.shr(e.word[e.rsp + stack_offset_src1 + e.rdx], e.cl);
+
+    e.add(e.edx, 2);
+
+    e.cmp(e.edx, 16);
+    e.jnz(looper);
+    e.vmovdqa(i.dest, e.byte[e.rsp + stack_offset_src1]);
 
     e.L(end);
   }
@@ -1324,14 +1395,37 @@ struct VECTOR_SHR_V128
 
       // TODO(benvanik): native version.
       e.L(emu);
-      if (i.src2.is_constant) {
+
-        e.lea(e.GetNativeParam(1), e.StashConstantXmm(1, i.src2.constant()));
+      unsigned stack_offset_src1 = StackLayout::GUEST_SCRATCH;
+      unsigned stack_offset_src2 = StackLayout::GUEST_SCRATCH + 16;
+      if (i.src1.is_constant) {
+        e.StashConstantXmm(0, i.src1.constant());
+        stack_offset_src1 = X64Emitter::kStashOffset;
+
       } else {
-        e.lea(e.GetNativeParam(1), e.StashXmm(1, i.src2));
+        e.vmovdqa(e.ptr[e.rsp + stack_offset_src1], i.src1);
       }
-      e.lea(e.GetNativeParam(0), e.StashXmm(0, src1));
+
-      e.CallNativeSafe(reinterpret_cast<void*>(EmulateVectorShr<uint32_t>));
+      if (i.src2.is_constant) {
-      e.vmovaps(i.dest, e.xmm0);
+        e.StashConstantXmm(1, i.src2.constant());
+        stack_offset_src2 = X64Emitter::kStashOffset + 16;
+      } else {
+        e.vmovdqa(e.ptr[e.rsp + stack_offset_src2], i.src2);
+      }
+
+      Xbyak::Label looper;
+
+      e.xor_(e.edx, e.edx);
+
+      e.L(looper);
+      e.mov(e.ecx, e.dword[e.rsp + stack_offset_src2 + e.rdx]);
+      e.shr(e.dword[e.rsp + stack_offset_src1 + e.rdx], e.cl);
+
+      e.add(e.edx, 4);
+
+      e.cmp(e.edx, 16);
+      e.jnz(looper);
+      e.vmovdqa(i.dest, e.byte[e.rsp + stack_offset_src1]);
 
       e.L(end);
     }
@@ -1388,7 +1482,8 @@ struct VECTOR_SHA_V128
   }
 
   static void EmitInt8(X64Emitter& e, const EmitArgType& i) {
-    // TODO(benvanik): native version (with shift magic).
+    unsigned stack_offset_src1 = StackLayout::GUEST_SCRATCH;
+    unsigned stack_offset_src2 = StackLayout::GUEST_SCRATCH + 16;
     if (i.src2.is_constant) {
       const auto& shamt = i.src2.constant();
       bool all_same = true;
@@ -1399,7 +1494,6 @@ struct VECTOR_SHA_V128
         }
       }
 
-	  
       if (e.IsFeatureEnabled(kX64EmitGFNI)) {
         if (all_same) {
           // Every count is the same, so we can use gf2p8affineqb.
@@ -1412,8 +1506,7 @@ struct VECTOR_SHA_V128
                            e.StashConstantXmm(0, vec128q(shift_matrix)), 0);
           return;
         }
-      }
+      } else if (all_same) {
-      else if (all_same) {
         Xmm to_be_shifted = GetInputRegOrConstant(e, i.src1, e.xmm1);
 
         e.vpmovsxbw(e.xmm0, to_be_shifted);  //_mm_srai_epi16 / psraw
@@ -1425,14 +1518,41 @@ struct VECTOR_SHA_V128
         return;
       }
 
-
+      e.StashConstantXmm(1, i.src2.constant());
-      e.lea(e.GetNativeParam(1), e.StashConstantXmm(1, i.src2.constant()));
+      stack_offset_src2 = X64Emitter::kStashOffset + 16;
     } else {
-      e.lea(e.GetNativeParam(1), e.StashXmm(1, i.src2));
+      e.vmovdqa(e.ptr[e.rsp + stack_offset_src2], i.src2);
     }
-    e.lea(e.GetNativeParam(0), e.StashXmm(0, i.src1));
+
-    e.CallNativeSafe(reinterpret_cast<void*>(EmulateVectorShr<int8_t>));
+    if (i.src1.is_constant) {
-    e.vmovaps(i.dest, e.xmm0);
+      e.StashConstantXmm(0, i.src1.constant());
+      stack_offset_src1 = X64Emitter::kStashOffset;
+    } else {
+      e.vmovdqa(e.ptr[e.rsp + stack_offset_src1], i.src1);
+    }
+
+    Xbyak::Label looper;
+
+    e.xor_(e.edx, e.edx);
+
+    e.L(looper);
+    // movzx is to eliminate any possible dep on previous value of rcx at start
+    // of loop
+    e.movzx(e.ecx, e.byte[e.rsp + stack_offset_src2 + e.rdx]);
+    // maybe using a memory operand as the left side isn't the best idea lol,
+    // still better than callnativesafe though agners docs have no timing info
+    // on shx [m], cl so shrug
+    e.sar(e.byte[e.rsp + stack_offset_src1 + e.rdx], e.cl);
+
+    if (e.IsFeatureEnabled(kX64FlagsIndependentVars)) {
+      e.inc(e.edx);
+    } else {
+      e.add(e.edx, 1);
+    }
+
+    e.cmp(e.edx, 16);
+    e.jnz(looper);
+    e.vmovdqa(i.dest, e.byte[e.rsp + stack_offset_src1]);
   }
 
   static void EmitInt16(X64Emitter& e, const EmitArgType& i) {
@@ -1474,14 +1594,38 @@ struct VECTOR_SHA_V128
 
     // TODO(benvanik): native version (with shift magic).
     e.L(emu);
-    if (i.src2.is_constant) {
+
-      e.lea(e.GetNativeParam(1), e.StashConstantXmm(1, i.src2.constant()));
+    unsigned stack_offset_src1 = StackLayout::GUEST_SCRATCH;
+    unsigned stack_offset_src2 = StackLayout::GUEST_SCRATCH + 16;
+    if (i.src1.is_constant) {
+      e.StashConstantXmm(0, i.src1.constant());
+      stack_offset_src1 = X64Emitter::kStashOffset;
+
     } else {
-      e.lea(e.GetNativeParam(1), e.StashXmm(1, i.src2));
+      e.vmovdqa(e.ptr[e.rsp + stack_offset_src1], i.src1);
     }
-    e.lea(e.GetNativeParam(0), e.StashXmm(0, i.src1));
+
-    e.CallNativeSafe(reinterpret_cast<void*>(EmulateVectorShr<int16_t>));
+    if (i.src2.is_constant) {
-    e.vmovaps(i.dest, e.xmm0);
+      e.StashConstantXmm(1, i.src2.constant());
+      stack_offset_src2 = X64Emitter::kStashOffset + 16;
+    } else {
+      e.vmovdqa(e.ptr[e.rsp + stack_offset_src2], i.src2);
+    }
+
+    Xbyak::Label looper;
+
+    e.xor_(e.edx, e.edx);
+
+    e.L(looper);
+    e.movzx(e.ecx, e.word[e.rsp + stack_offset_src2 + e.rdx]);
+
+    e.sar(e.word[e.rsp + stack_offset_src1 + e.rdx], e.cl);
+
+    e.add(e.edx, 2);
+
+    e.cmp(e.edx, 16);
+    e.jnz(looper);
+    e.vmovdqa(i.dest, e.byte[e.rsp + stack_offset_src1]);
 
     e.L(end);
   }
@@ -1508,9 +1652,9 @@ struct VECTOR_SHA_V128
       // that happens so we mask.
       if (i.src2.is_constant) {
         e.LoadConstantXmm(e.xmm0, i.src2.constant());
-        e.vandps(e.xmm0, e.GetXmmConstPtr(XMMShiftMaskPS));
+        e.vpand(e.xmm0, e.GetXmmConstPtr(XMMShiftMaskPS));
       } else {
-        e.vandps(e.xmm0, i.src2, e.GetXmmConstPtr(XMMShiftMaskPS));
+        e.vpand(e.xmm0, i.src2, e.GetXmmConstPtr(XMMShiftMaskPS));
       }
       e.vpsravd(i.dest, i.src1, e.xmm0);
     } else {
@@ -1535,14 +1679,36 @@ struct VECTOR_SHA_V128
 
       // TODO(benvanik): native version.
       e.L(emu);
-      if (i.src2.is_constant) {
+      unsigned stack_offset_src1 = StackLayout::GUEST_SCRATCH;
-        e.lea(e.GetNativeParam(1), e.StashConstantXmm(1, i.src2.constant()));
+      unsigned stack_offset_src2 = StackLayout::GUEST_SCRATCH + 16;
+      if (i.src1.is_constant) {
+        e.StashConstantXmm(0, i.src1.constant());
+        stack_offset_src1 = X64Emitter::kStashOffset;
+
       } else {
-        e.lea(e.GetNativeParam(1), e.StashXmm(1, i.src2));
+        e.vmovdqa(e.ptr[e.rsp + stack_offset_src1], i.src1);
       }
-      e.lea(e.GetNativeParam(0), e.StashXmm(0, i.src1));
+
-      e.CallNativeSafe(reinterpret_cast<void*>(EmulateVectorShr<int32_t>));
+      if (i.src2.is_constant) {
-      e.vmovaps(i.dest, e.xmm0);
+        e.StashConstantXmm(1, i.src2.constant());
+        stack_offset_src2 = X64Emitter::kStashOffset + 16;
+      } else {
+        e.vmovdqa(e.ptr[e.rsp + stack_offset_src2], i.src2);
+      }
+
+      Xbyak::Label looper;
+
+      e.xor_(e.edx, e.edx);
+
+      e.L(looper);
+      e.mov(e.ecx, e.dword[e.rsp + stack_offset_src2 + e.rdx]);
+      e.sar(e.dword[e.rsp + stack_offset_src1 + e.rdx], e.cl);
+
+      e.add(e.edx, 4);
+
+      e.cmp(e.edx, 16);
+      e.jnz(looper);
+      e.vmovdqa(i.dest, e.byte[e.rsp + stack_offset_src1]);
 
       e.L(end);
     }
@@ -1550,26 +1716,6 @@ struct VECTOR_SHA_V128
 };
 EMITTER_OPCODE_TABLE(OPCODE_VECTOR_SHA, VECTOR_SHA_V128);
 
-// ============================================================================
-// OPCODE_VECTOR_ROTATE_LEFT
-// ============================================================================
-template <typename T, std::enable_if_t<std::is_integral<T>::value, int> = 0>
-static __m128i EmulateVectorRotateLeft(void*, __m128i src1, __m128i src2) {
-  alignas(16) T value[16 / sizeof(T)];
-  alignas(16) T shamt[16 / sizeof(T)];
-
-  // Load SSE registers into a C array.
-  _mm_store_si128(reinterpret_cast<__m128i*>(value), src1);
-  _mm_store_si128(reinterpret_cast<__m128i*>(shamt), src2);
-
-  for (size_t i = 0; i < (16 / sizeof(T)); ++i) {
-    value[i] = xe::rotate_left<T>(value[i], shamt[i] & ((sizeof(T) * 8) - 1));
-  }
-
-  // Store result and return it.
-  return _mm_load_si128(reinterpret_cast<__m128i*>(value));
-}
-
 struct VECTOR_ROTATE_LEFT_V128
     : Sequence<VECTOR_ROTATE_LEFT_V128,
                I<OPCODE_VECTOR_ROTATE_LEFT, V128Op, V128Op, V128Op>> {
@@ -1594,33 +1740,72 @@ struct VECTOR_ROTATE_LEFT_V128
       }
 
     } else {
+      unsigned stack_offset_src1 = StackLayout::GUEST_SCRATCH;
+      unsigned stack_offset_src2 = StackLayout::GUEST_SCRATCH + 16;
       switch (i.instr->flags) {
-        case INT8_TYPE:
+        case INT8_TYPE: {
-          // TODO(benvanik): native version (with shift magic).
+          if (i.src1.is_constant) {
-          if (i.src2.is_constant) {
+            e.StashConstantXmm(0, i.src1.constant());
-            e.lea(e.GetNativeParam(1),
+            stack_offset_src1 = X64Emitter::kStashOffset;
-                  e.StashConstantXmm(1, i.src2.constant()));
+
           } else {
-            e.lea(e.GetNativeParam(1), e.StashXmm(1, i.src2));
+            e.vmovdqa(e.ptr[e.rsp + stack_offset_src1], i.src1);
           }
-          e.lea(e.GetNativeParam(0), e.StashXmm(0, i.src1));
+
-          e.CallNativeSafe(
-              reinterpret_cast<void*>(EmulateVectorRotateLeft<uint8_t>));
-          e.vmovaps(i.dest, e.xmm0);
-          break;
-        case INT16_TYPE:
-          // TODO(benvanik): native version (with shift magic).
           if (i.src2.is_constant) {
-            e.lea(e.GetNativeParam(1),
+            e.StashConstantXmm(1, i.src2.constant());
-                  e.StashConstantXmm(1, i.src2.constant()));
+            stack_offset_src2 = X64Emitter::kStashOffset + 16;
           } else {
-            e.lea(e.GetNativeParam(1), e.StashXmm(1, i.src2));
+            e.vmovdqa(e.ptr[e.rsp + stack_offset_src2], i.src2);
           }
-          e.lea(e.GetNativeParam(0), e.StashXmm(0, i.src1));
+
-          e.CallNativeSafe(
+          Xbyak::Label rotate_iter;
-              reinterpret_cast<void*>(EmulateVectorRotateLeft<uint16_t>));
+
-          e.vmovaps(i.dest, e.xmm0);
+          e.xor_(e.edx, e.edx);
-          break;
+
+          e.L(rotate_iter);
+          e.movzx(e.ecx, e.byte[e.rsp + stack_offset_src2 + e.rdx]);
+
+          e.rol(e.byte[e.rsp + stack_offset_src1 + e.rdx], e.cl);
+
+          e.add(e.edx, 1);
+
+          e.cmp(e.edx, 16);
+          e.jnz(rotate_iter);
+          e.vmovdqa(i.dest, e.byte[e.rsp + stack_offset_src1]);
+
+        } break;
+        case INT16_TYPE: {
+          if (i.src1.is_constant) {
+            e.StashConstantXmm(0, i.src1.constant());
+            stack_offset_src1 = X64Emitter::kStashOffset;
+
+          } else {
+            e.vmovdqa(e.ptr[e.rsp + stack_offset_src1], i.src1);
+          }
+
+          if (i.src2.is_constant) {
+            e.StashConstantXmm(1, i.src2.constant());
+            stack_offset_src2 = X64Emitter::kStashOffset + 16;
+          } else {
+            e.vmovdqa(e.ptr[e.rsp + stack_offset_src2], i.src2);
+          }
+
+          Xbyak::Label rotate_iter;
+
+          e.xor_(e.edx, e.edx);
+
+          e.L(rotate_iter);
+          e.movzx(e.ecx, e.word[e.rsp + stack_offset_src2 + e.rdx]);
+          e.rol(e.word[e.rsp + stack_offset_src1 + e.rdx], e.cl);
+
+          e.add(e.edx, 2);
+
+          e.cmp(e.edx, 16);
+          e.jnz(rotate_iter);
+          e.vmovdqa(i.dest, e.byte[e.rsp + stack_offset_src1]);
+
+        } break;
         case INT32_TYPE: {
           if (e.IsFeatureEnabled(kX64EmitAVX512Ortho)) {
             e.vprolvd(i.dest, i.src1, i.src2);
@@ -1638,23 +1823,40 @@ struct VECTOR_ROTATE_LEFT_V128
             }
             e.vpsllvd(e.xmm1, i.src1, e.xmm0);
             // Shift right (to get low bits):
-            e.vmovaps(temp, e.GetXmmConstPtr(XMMPI32));
+            e.vmovdqa(temp, e.GetXmmConstPtr(XMMPI32));
             e.vpsubd(temp, e.xmm0);
             e.vpsrlvd(i.dest, i.src1, temp);
             // Merge:
             e.vpor(i.dest, e.xmm1);
           } else {
-            // TODO(benvanik): non-AVX2 native version.
+            if (i.src1.is_constant) {
-            if (i.src2.is_constant) {
+              e.StashConstantXmm(0, i.src1.constant());
-              e.lea(e.GetNativeParam(1),
+              stack_offset_src1 = X64Emitter::kStashOffset;
-                    e.StashConstantXmm(1, i.src2.constant()));
+
             } else {
-              e.lea(e.GetNativeParam(1), e.StashXmm(1, i.src2));
+              e.vmovdqa(e.ptr[e.rsp + stack_offset_src1], i.src1);
             }
-            e.lea(e.GetNativeParam(0), e.StashXmm(0, i.src1));
+
-            e.CallNativeSafe(
+            if (i.src2.is_constant) {
-                reinterpret_cast<void*>(EmulateVectorRotateLeft<uint32_t>));
+              e.StashConstantXmm(1, i.src2.constant());
-            e.vmovaps(i.dest, e.xmm0);
+              stack_offset_src2 = X64Emitter::kStashOffset + 16;
+            } else {
+              e.vmovdqa(e.ptr[e.rsp + stack_offset_src2], i.src2);
+            }
+
+            Xbyak::Label rotate_iter;
+
+            e.xor_(e.edx, e.edx);
+
+            e.L(rotate_iter);
+            e.mov(e.ecx, e.dword[e.rsp + stack_offset_src2 + e.rdx]);
+            e.rol(e.dword[e.rsp + stack_offset_src1 + e.rdx], e.cl);
+
+            e.add(e.edx, 4);
+
+            e.cmp(e.edx, 16);
+            e.jnz(rotate_iter);
+            e.vmovdqa(i.dest, e.byte[e.rsp + stack_offset_src1]);
           }
           break;
         }
@@ -1667,80 +1869,120 @@ struct VECTOR_ROTATE_LEFT_V128
 };
 EMITTER_OPCODE_TABLE(OPCODE_VECTOR_ROTATE_LEFT, VECTOR_ROTATE_LEFT_V128);
 
-// ============================================================================
-// OPCODE_VECTOR_AVERAGE
-// ============================================================================
-template <typename T, std::enable_if_t<std::is_integral<T>::value, int> = 0>
-static __m128i EmulateVectorAverage(void*, __m128i src1, __m128i src2) {
-  alignas(16) T src1v[16 / sizeof(T)];
-  alignas(16) T src2v[16 / sizeof(T)];
-  alignas(16) T value[16 / sizeof(T)];
-
-  // Load SSE registers into a C array.
-  _mm_store_si128(reinterpret_cast<__m128i*>(src1v), src1);
-  _mm_store_si128(reinterpret_cast<__m128i*>(src2v), src2);
-
-  for (size_t i = 0; i < (16 / sizeof(T)); ++i) {
-    auto t = (uint64_t(src1v[i]) + uint64_t(src2v[i]) + 1) / 2;
-    value[i] = T(t);
-  }
-
-  // Store result and return it.
-  return _mm_load_si128(reinterpret_cast<__m128i*>(value));
-}
-
 struct VECTOR_AVERAGE
     : Sequence<VECTOR_AVERAGE,
                I<OPCODE_VECTOR_AVERAGE, V128Op, V128Op, V128Op>> {
   static void Emit(X64Emitter& e, const EmitArgType& i) {
+    auto i_flags = i.instr->flags;
     EmitCommutativeBinaryXmmOp(
         e, i,
-        [&i](X64Emitter& e, const Xmm& dest, const Xmm& src1, const Xmm& src2) {
+        [i_flags](X64Emitter& e, const Xmm& dest, const Xmm& src1,
-          const TypeName part_type =
+                  const Xmm& src2) {
-              static_cast<TypeName>(i.instr->flags & 0xFF);
+          const TypeName part_type = static_cast<TypeName>(i_flags & 0xFF);
-          const uint32_t arithmetic_flags = i.instr->flags >> 8;
+          const uint32_t arithmetic_flags = i_flags >> 8;
           bool is_unsigned = !!(arithmetic_flags & ARITHMETIC_UNSIGNED);
+          unsigned stack_offset_src1 = StackLayout::GUEST_SCRATCH;
+          unsigned stack_offset_src2 = StackLayout::GUEST_SCRATCH + 16;
           switch (part_type) {
             case INT8_TYPE:
               if (is_unsigned) {
                 e.vpavgb(dest, src1, src2);
               } else {
-                assert_always();
+                // todo: avx2 version or version that sign extends to two __m128
+
+                e.vmovdqa(e.ptr[e.rsp + stack_offset_src1], src1);
+                e.vmovdqa(e.ptr[e.rsp + stack_offset_src2], src2);
+
+                Xbyak::Label looper;
+
+                e.xor_(e.edx, e.edx);
+
+                e.L(looper);
+
+                e.movsx(e.ecx, e.byte[e.rsp + stack_offset_src2 + e.rdx]);
+                e.movsx(e.eax, e.byte[e.rsp + stack_offset_src1 + e.rdx]);
+
+                e.lea(e.ecx, e.ptr[e.ecx + e.eax + 1]);
+                e.sar(e.ecx, 1);
+                e.mov(e.byte[e.rsp + stack_offset_src1 + e.rdx], e.cl);
+
+                if (e.IsFeatureEnabled(kX64FlagsIndependentVars)) {
+                  e.inc(e.edx);
+                } else {
+                  e.add(e.edx, 1);
+                }
+
+                e.cmp(e.edx, 16);
+                e.jnz(looper);
+                e.vmovdqa(dest, e.ptr[e.rsp + stack_offset_src1]);
               }
               break;
             case INT16_TYPE:
               if (is_unsigned) {
                 e.vpavgw(dest, src1, src2);
               } else {
-                assert_always();
+                e.vmovdqa(e.ptr[e.rsp + stack_offset_src1], src1);
+                e.vmovdqa(e.ptr[e.rsp + stack_offset_src2], src2);
+
+                Xbyak::Label looper;
+
+                e.xor_(e.edx, e.edx);
+
+                e.L(looper);
+
+                e.movsx(e.ecx, e.word[e.rsp + stack_offset_src2 + e.rdx]);
+                e.movsx(e.eax, e.word[e.rsp + stack_offset_src1 + e.rdx]);
+
+                e.lea(e.ecx, e.ptr[e.ecx + e.eax + 1]);
+                e.sar(e.ecx, 1);
+                e.mov(e.word[e.rsp + stack_offset_src1 + e.rdx], e.cx);
+
+                e.add(e.edx, 2);
+
+                e.cmp(e.edx, 16);
+                e.jnz(looper);
+                e.vmovdqa(dest, e.ptr[e.rsp + stack_offset_src1]);
               }
               break;
-            case INT32_TYPE:
+            case INT32_TYPE: {
               // No 32bit averages in AVX.
+              e.vmovdqa(e.ptr[e.rsp + stack_offset_src1], src1);
+              e.vmovdqa(e.ptr[e.rsp + stack_offset_src2], src2);
+
+              Xbyak::Label looper;
+
+              e.xor_(e.edx, e.edx);
+
+              e.L(looper);
+              auto src2_current_ptr =
+                  e.dword[e.rsp + stack_offset_src2 + e.rdx];
+              auto src1_current_ptr =
+                  e.dword[e.rsp + stack_offset_src1 + e.rdx];
+
               if (is_unsigned) {
-                if (i.src2.is_constant) {
+                // implicit zero-ext
-                  e.lea(e.GetNativeParam(1),
+                e.mov(e.ecx, src2_current_ptr);
-                        e.StashConstantXmm(1, i.src2.constant()));
+                e.mov(e.eax, src1_current_ptr);
-                } else {
-                  e.lea(e.GetNativeParam(1), e.StashXmm(1, i.src2));
-                }
-                e.lea(e.GetNativeParam(0), e.StashXmm(0, i.src1));
-                e.CallNativeSafe(
-                    reinterpret_cast<void*>(EmulateVectorAverage<uint32_t>));
-                e.vmovaps(i.dest, e.xmm0);
               } else {
-                if (i.src2.is_constant) {
+                e.movsxd(e.rcx, src2_current_ptr);
-                  e.lea(e.GetNativeParam(1),
+                e.movsxd(e.rax, src1_current_ptr);
-                        e.StashConstantXmm(1, i.src2.constant()));
-                } else {
-                  e.lea(e.GetNativeParam(1), e.StashXmm(1, i.src2));
-                }
-                e.lea(e.GetNativeParam(0), e.StashXmm(0, i.src1));
-                e.CallNativeSafe(
-                    reinterpret_cast<void*>(EmulateVectorAverage<int32_t>));
-                e.vmovaps(i.dest, e.xmm0);
               }
-              break;
+
+              e.lea(e.rcx, e.ptr[e.rcx + e.rax + 1]);
+              if (is_unsigned) {
+                e.shr(e.rcx, 1);
+              } else {
+                e.sar(e.rcx, 1);
+              }
+              e.mov(e.dword[e.rsp + stack_offset_src1 + e.rdx], e.ecx);
+
+              e.add(e.edx, 4);
+
+              e.cmp(e.edx, 16);
+              e.jnz(looper);
+              e.vmovdqa(dest, e.ptr[e.rsp + stack_offset_src1]);
+            } break;
+
             default:
               assert_unhandled_case(part_type);
               break;
@@ -2163,82 +2405,6 @@ struct PERMUTE_V128
 };
 EMITTER_OPCODE_TABLE(OPCODE_PERMUTE, PERMUTE_I32, PERMUTE_V128);
 
-#define LCPI(name, quad1) const __m128i name = _mm_set1_epi32(quad1)
-// xmm0 is precasted to int, but contains float
-// chrispy: todo: make available to gpu code
-static __m128i xenos_float4_to_float16_x4(__m128i xmm0) {
-  LCPI(LCPI0_0, 2147483647);
-  LCPI(LCPI0_1, 1207951360);
-  LCPI(LCPI0_2, 134217728);
-  LCPI(LCPI0_3, 3347054592);
-  LCPI(LCPI0_4, 260038655);
-  LCPI(LCPI0_5, 32767);
-  LCPI(LCPI0_6, 4294934528);
-
-  __m128i xmm1 = _mm_and_si128(xmm0, LCPI0_0);
-
-  __m128i xmm2 = LCPI0_1;
-
-  __m128i xmm3 = _mm_add_epi32(xmm0, LCPI0_2);
-  xmm2 = _mm_cmpgt_epi32(xmm2, xmm1);
-  xmm3 = _mm_srli_epi32(xmm3, 13);
-  xmm1 = _mm_add_epi32(xmm1, LCPI0_3);
-  __m128i xmm4 = _mm_min_epu32(xmm1, LCPI0_4);
-  xmm1 = _mm_cmpeq_epi32(xmm1, xmm4);
-  xmm4 = LCPI0_5;
-  xmm3 = _mm_and_si128(xmm3, xmm4);
-  xmm1 = _mm_and_si128(xmm1, xmm3);
-
-  xmm1 = _mm_castps_si128(_mm_blendv_ps(
-      _mm_castsi128_ps(xmm4), _mm_castsi128_ps(xmm1), _mm_castsi128_ps(xmm2)));
-  xmm0 = _mm_srli_epi32(xmm0, 16);
-  xmm0 = _mm_and_si128(xmm0, LCPI0_6);
-  xmm0 = _mm_or_si128(xmm1, xmm0);
-  xmm0 = _mm_packus_epi32(xmm0, _mm_setzero_si128());
-  return xmm0;
-}
-// returns floats, uncasted
-// chrispy: todo, make this available to gpu code?
-static __m128i xenos_halves_to_floats(__m128i xmm0) {
-  LCPI(LCPI3_0, 0x1f);
-  LCPI(LCPI3_1, 0x80000000);
-  LCPI(LCPI3_2, 0x38000000);
-  LCPI(LCPI3_3, 0x7fe000);
-
-  __m128i xmm1, xmm2, xmm3, xmm4;
-
-  xmm1 = _mm_cvtepu16_epi32(xmm0);
-
-  xmm2 = _mm_srli_epi32(xmm1, 10);
-
-  xmm2 = _mm_and_si128(xmm2, LCPI3_0);
-
-  xmm0 = _mm_cvtepi16_epi32(xmm0);
-
-  xmm0 = _mm_and_si128(xmm0, LCPI3_1);
-
-  xmm3 = _mm_setzero_si128();
-
-  xmm4 = _mm_slli_epi32(xmm2, 23);
-
-  xmm4 = _mm_add_epi32(xmm4, LCPI3_2);
-
-  xmm2 = _mm_cmpeq_epi32(xmm2, xmm3);
-
-  xmm1 = _mm_slli_epi32(xmm1, 13);
-
-  xmm1 = _mm_and_si128(xmm1, LCPI3_3);
-
-  xmm3 = _mm_andnot_si128(xmm2, xmm4);
-
-  xmm1 = _mm_andnot_si128(xmm2, xmm1);
-
-  xmm0 = _mm_or_si128(xmm1, xmm0);
-  xmm0 = _mm_or_si128(xmm0, xmm3);
-  return xmm0;
-}
-
-#undef LCPI
 template <typename Inst>
 static void emit_fast_f16_unpack(X64Emitter& e, const Inst& i,
                                  XmmConst initial_shuffle) {

src/xenia/cpu/compiler/passes/conditional_group_pass.cc

@@ -48,9 +48,7 @@ bool ConditionalGroupPass::Initialize(Compiler* compiler) {
 
 bool ConditionalGroupPass::Run(HIRBuilder* builder) {
   bool dirty;
-  int loops = 0;
   do {
-    assert_true(loops < 20);  // arbitrary number
     dirty = false;
     for (size_t i = 0; i < passes_.size(); ++i) {
       scratch_arena()->Reset();
@@ -68,7 +66,6 @@ bool ConditionalGroupPass::Run(HIRBuilder* builder) {
         dirty |= result;
       }
     }
-    loops++;
   } while (dirty);
   return true;
 }

src/xenia/cpu/compiler/passes/finalization_pass.cc

@@ -41,18 +41,6 @@ bool FinalizationPass::Run(HIRBuilder* builder) {
     block->ordinal = block_ordinal++;
 
     // Ensure all labels have names.
-    auto label = block->label_head;
-    while (label) {
-      if (!label->name) {
-        const size_t label_len = 6 + 4;
-        char* name = reinterpret_cast<char*>(arena->Alloc(label_len + 1, 1));
-        assert_true(label->id <= 9999);
-        auto end = fmt::format_to_n(name, label_len, "_label{}", label->id);
-        name[end.size] = '\0';
-        label->name = name;
-      }
-      label = label->next;
-    }
 
     // Remove unneeded jumps.
     auto tail = block->instr_tail;

src/xenia/cpu/compiler/passes/simplification_pass.cc

@@ -23,52 +23,6 @@ using namespace xe::cpu::hir;
 using xe::cpu::hir::HIRBuilder;
 using xe::cpu::hir::Instr;
 using xe::cpu::hir::Value;
-using vmask_portion_t = uint64_t;
-template <uint32_t Ndwords>
-struct Valuemask_t {
-  vmask_portion_t bits[Ndwords];
-
-  static Valuemask_t create_empty(vmask_portion_t fill = 0) {
-    Valuemask_t result;
-    for (uint32_t i = 0; i < Ndwords; ++i) {
-      result.bits[i] = fill;
-    }
-    return result;
-  }
-  template <typename TCallable>
-  Valuemask_t operate(TCallable&& oper) const {
-    Valuemask_t result = create_empty();
-
-    for (uint32_t i = 0; i < Ndwords; ++i) {
-      result.bits[i] = oper(bits[i]);
-    }
-    return result;
-  }
-  template <typename TCallable>
-  Valuemask_t operate(TCallable&& oper, Valuemask_t other) const {
-    Valuemask_t result = create_empty();
-
-    for (uint32_t i = 0; i < Ndwords; ++i) {
-      result.bits[i] = oper(bits[i], other.bits[i]);
-    }
-    return result;
-  }
-  Valuemask_t operator&(ValueMask other) const {
-    return operate([](vmask_portion_t x, vmask_portion_t y) { return x & y; },
-                   other);
-  }
-  Valuemask_t operator|(ValueMask other) const {
-    return operate([](vmask_portion_t x, vmask_portion_t y) { return x | y; },
-                   other);
-  }
-  Valuemask_t operator^(ValueMask other) const {
-    return operate([](vmask_portion_t x, vmask_portion_t y) { return x ^ y; },
-                   other);
-  }
-  Valuemask_t operator~() const {
-    return operate([](vmask_portion_t x) { return ~x; }, other);
-  }
-};
 
 SimplificationPass::SimplificationPass() : ConditionalGroupSubpass() {}
 
@@ -76,17 +30,13 @@ SimplificationPass::~SimplificationPass() {}
 
 bool SimplificationPass::Run(HIRBuilder* builder, bool& result) {
   result = false;
-  bool iter_result = false;
 
-  do {
+  result |= SimplifyBitArith(builder);
-    iter_result = false;
+  result |= EliminateConversions(builder);
-    iter_result |= SimplifyBitArith(builder);
+  result |= SimplifyAssignments(builder);
-    iter_result |= EliminateConversions(builder);
+  result |= SimplifyBasicArith(builder);
-    iter_result |= SimplifyAssignments(builder);
+  result |= SimplifyVectorOps(builder);
-    iter_result |= SimplifyBasicArith(builder);
+
-    iter_result |= SimplifyVectorOps(builder);
-    result |= iter_result;
-  } while (iter_result);
   return true;
 }
 // simplifications that apply to both or and xor
@@ -735,7 +685,9 @@ bool SimplificationPass::CheckAdd(hir::Instr* i, hir::HIRBuilder* builder) {
     auto [added_constant_neg, added_var_neg] =
         i->BinaryValueArrangeAsConstAndVar();
 
-    if (!added_constant_neg) return false;
+    if (!added_constant_neg) {
+      return false;
+    }
     if (added_constant_neg->AsUint64() &
         GetScalarSignbitMask(added_constant_neg->type)) {
       // adding a value that has its signbit set!
@@ -882,11 +834,6 @@ bool SimplificationPass::CheckScalarConstCmp(hir::Instr* i,
 
     } else if (cmpop == OPCODE_COMPARE_UGT) {
       // impossible, cannot be greater than mask
-
-      /* i->Replace(&OPCODE_ASSIGN_info, 0);
-      i->set_src1(builder->LoadZeroInt8());
-      return true;
-      */
       constant_replacement = builder->LoadZeroInt8();
 
     } else if (cmpop == OPCODE_COMPARE_ULE) {  // less than or equal to mask =
@@ -914,9 +861,9 @@ bool SimplificationPass::CheckIsTrueIsFalse(hir::Instr* i,
   bool istrue = i->opcode == &OPCODE_COMPARE_NE_info;
   bool isfalse = i->opcode == &OPCODE_COMPARE_EQ_info;
 
-  auto [input_cosntant, input] = i->BinaryValueArrangeAsConstAndVar();
+  auto [input_constant, input] = i->BinaryValueArrangeAsConstAndVar();
 
-  if (!input_cosntant || input_cosntant->AsUint64() != 0) {
+  if (!input_constant || input_constant->AsUint64() != 0) {
     return false;
   }
 
@@ -957,12 +904,6 @@ bool SimplificationPass::CheckIsTrueIsFalse(hir::Instr* i,
     }
   }
 
-  /* Instr* input_def = input->def;
-   if (!input_def) {
-     return false;
-   }
-
-   input_def = input_def->GetDestDefSkipAssigns();*/
   return false;
 }
 bool SimplificationPass::CheckSHRByConst(hir::Instr* i,

src/xenia/cpu/hir/label.h

@@ -26,6 +26,13 @@ class Label {
   char* name;
 
   void* tag;
+  // just use stringification of label id
+  // this will later be used as an input to xbyak. xbyak only accepts
+  // std::string as a value, not passed by reference, so precomputing the
+  // stringification does not help
+  std::string GetIdString() {
+	  return std::to_string(id);
+  }
 };
 
 }  // namespace hir

src/xenia/cpu/hir/opcodes.h

@@ -11,7 +11,7 @@
 #define XENIA_CPU_HIR_OPCODES_H_
 
 #include <cstdint>
-
+#include "xenia/base/platform.h"
 namespace xe {
 namespace cpu {
 namespace hir {
@@ -361,13 +361,16 @@ enum OpcodeSignature {
 #define GET_OPCODE_SIG_TYPE_SRC1(sig) (OpcodeSignatureType)((sig >> 3) & 0x7)
 #define GET_OPCODE_SIG_TYPE_SRC2(sig) (OpcodeSignatureType)((sig >> 6) & 0x7)
 #define GET_OPCODE_SIG_TYPE_SRC3(sig) (OpcodeSignatureType)((sig >> 9) & 0x7)
+XE_MAYBE_UNUSED
 static bool IsOpcodeBinaryValue(uint32_t signature) {
   return (signature & ~(0x7)) ==
          ((OPCODE_SIG_TYPE_V << 3) | (OPCODE_SIG_TYPE_V << 6));
 }
+XE_MAYBE_UNUSED
 static bool IsOpcodeUnaryValue(uint32_t signature) {
   return (signature & ~(0x7)) == ((OPCODE_SIG_TYPE_V << 3));
 }
+XE_MAYBE_UNUSED
 static void UnpackOpcodeSig(uint32_t sig, OpcodeSignatureType& dest,
                             OpcodeSignatureType& src1,
                             OpcodeSignatureType& src2,

src/xenia/cpu/mmio_handler.cc

@@ -185,7 +185,7 @@ bool MMIOHandler::TryDecodeLoadStore(const uint8_t* p,
   uint8_t rex_b = rex & 0b0001;
   uint8_t rex_x = rex & 0b0010;
   uint8_t rex_r = rex & 0b0100;
-  uint8_t rex_w = rex & 0b1000;
+  //uint8_t rex_w = rex & 0b1000;
 
   // http://www.sandpile.org/x86/opc_rm.htm
   // http://www.sandpile.org/x86/opc_sib.htm
@@ -418,7 +418,6 @@ bool MMIOHandler::ExceptionCallback(Exception* ex) {
     // Quick kill anything outside our mapping.
     return false;
   }
-  uint64_t hostip = ex->pc();
 
   void* fault_host_address = reinterpret_cast<void*>(ex->fault_address());
 

src/xenia/cpu/module.h

@@ -54,6 +54,7 @@ class Module {
 
   bool ReadMap(const char* file_name);
 
+  virtual void Precompile() {}
  protected:
   virtual std::unique_ptr<Function> CreateFunction(uint32_t address) = 0;
 

src/xenia/cpu/ppc/ppc_context.h

@@ -425,6 +425,27 @@ typedef struct alignas(64) PPCContext_s {
   uint64_t reserved_val;
   ThreadState* thread_state;
   uint8_t* virtual_membase;
+
+  template <typename T = uint8_t*>
+  inline T TranslateVirtual(uint32_t guest_address) XE_RESTRICT const {
+#if XE_PLATFORM_WIN32 == 1
+    uint8_t* host_address = virtual_membase + guest_address;
+    if (guest_address >= static_cast<uint32_t>(reinterpret_cast<uintptr_t>(this))) {
+      host_address += 0x1000;
+    }
+    return reinterpret_cast<T>(host_address);
+#else
+    return processor->memory()->TranslateVirtual<T>(guest_address);
+
+#endif
+  }
+  //for convenience in kernel functions, version that auto narrows to uint32
+  template <typename T = uint8_t*>
+  inline T TranslateVirtualGPR(uint64_t guest_address) XE_RESTRICT const {
+    return TranslateVirtual<T>(static_cast<uint32_t>(guest_address));
+  
+  }
+
   static std::string GetRegisterName(PPCRegister reg);
   std::string GetStringFromValue(PPCRegister reg) const;
   void SetValueFromString(PPCRegister reg, std::string value);

src/xenia/cpu/ppc/ppc_decode_data.h

@@ -46,6 +46,7 @@ struct PPCDecodeData {
     uint32_t LEV() const { return bits_.LEV; }
 
    private:
+	   XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -74,6 +75,7 @@ struct PPCDecodeData {
     uint32_t L() const { return bits_.RT & 0x1; }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -95,6 +97,7 @@ struct PPCDecodeData {
     int32_t ds() const { return static_cast<int32_t>(XEEXTS16(DS() << 2)); }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -174,6 +177,7 @@ struct PPCDecodeData {
     uint32_t CRFS() const { return bits_.RA >> 2; }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -200,6 +204,7 @@ struct PPCDecodeData {
     uint32_t CRFS() const { return CRBA() >> 2; }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -223,6 +228,7 @@ struct PPCDecodeData {
     }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -244,6 +250,7 @@ struct PPCDecodeData {
     bool Rc() const { return bits_.Rc ? true : false; }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -266,6 +273,7 @@ struct PPCDecodeData {
     bool Rc() const { return bits_.Rc ? true : false; }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -289,6 +297,7 @@ struct PPCDecodeData {
     bool Rc() const { return bits_.Rc ? true : false; }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -314,6 +323,7 @@ struct PPCDecodeData {
     bool Rc() const { return bits_.Rc ? true : false; }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -339,6 +349,7 @@ struct PPCDecodeData {
     bool Rc() const { return bits_.Rc ? true : false; }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -363,6 +374,7 @@ struct PPCDecodeData {
     bool Rc() const { return bits_.Rc ? true : false; }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -389,6 +401,7 @@ struct PPCDecodeData {
     bool Rc() const { return bits_.Rc ? true : false; }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -412,6 +425,7 @@ struct PPCDecodeData {
     int32_t SIMM() const { return static_cast<int32_t>(XEEXTS16(VA())); }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -431,6 +445,7 @@ struct PPCDecodeData {
     bool Rc() const { return bits_.Rc ? true : false; }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -452,6 +467,7 @@ struct PPCDecodeData {
     uint32_t SHB() const { return VC() & 0xF; }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -473,6 +489,7 @@ struct PPCDecodeData {
     uint32_t VB() const { return bits_.VB128l | (bits_.VB128h << 5); }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -498,6 +515,7 @@ struct PPCDecodeData {
     uint32_t RB() const { return bits_.RB; }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -521,6 +539,7 @@ struct PPCDecodeData {
     uint32_t VC() const { return bits_.VC; }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -546,6 +565,7 @@ struct PPCDecodeData {
     int32_t SIMM() const { return static_cast<int32_t>(XEEXTS16(bits_.UIMM)); }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -567,6 +587,7 @@ struct PPCDecodeData {
     uint32_t z() const { return bits_.z; }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -592,6 +613,7 @@ struct PPCDecodeData {
     uint32_t SH() const { return bits_.SH; }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -618,6 +640,7 @@ struct PPCDecodeData {
     bool Rc() const { return bits_.Rc ? true : false; }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;
@@ -642,6 +665,7 @@ struct PPCDecodeData {
     uint32_t UIMM() const { return bits_.PERMl | (bits_.PERMh << 5); }
 
    private:
+    XE_MAYBE_UNUSED
     uint32_t address_;
     union {
       uint32_t value_;

src/xenia/cpu/ppc/ppc_emit_altivec.cc

@@ -2014,8 +2014,7 @@ int InstrEmit_vupkhsh(PPCHIRBuilder& f, const InstrData& i) {
   return InstrEmit_vupkhsh_(f, i.VX.VD, i.VX.VB);
 }
 int InstrEmit_vupkhsh128(PPCHIRBuilder& f, const InstrData& i) {
-  uint32_t va = VX128_VA128;
+  assert_zero(VX128_VA128);
-  assert_zero(va);
   return InstrEmit_vupkhsh_(f, VX128_VD128, VX128_VB128);
 }
 
@@ -2032,8 +2031,7 @@ int InstrEmit_vupklsh(PPCHIRBuilder& f, const InstrData& i) {
   return InstrEmit_vupklsh_(f, i.VX.VD, i.VX.VB);
 }
 int InstrEmit_vupklsh128(PPCHIRBuilder& f, const InstrData& i) {
-  uint32_t va = VX128_VA128;
+  assert_zero(VX128_VA128);
-  assert_zero(va);
   return InstrEmit_vupklsh_(f, VX128_VD128, VX128_VB128);
 }
 

src/xenia/cpu/ppc/ppc_emit_memory.cc

@@ -16,7 +16,7 @@
 #include "xenia/cpu/ppc/ppc_hir_builder.h"
 
 DEFINE_bool(
-    disable_prefetch_and_cachecontrol, false,
+    disable_prefetch_and_cachecontrol, true,
     "Disables translating ppc prefetch/cache flush instructions to host "
     "prefetch/cacheflush instructions. This may improve performance as these "
     "instructions were written with the Xbox 360's cache in mind, and modern "

src/xenia/cpu/ppc/ppc_frontend.cc

@@ -105,6 +105,11 @@ bool PPCFrontend::Initialize() {
 }
 
 bool PPCFrontend::DeclareFunction(GuestFunction* function) {
+
+	//chrispy: make sure we aren't declaring a function that is actually padding data, this will mess up PPCScanner and is hard to debug
+	//wow, this halo reach actually has branches into 0 opcodes, look into further
+  //xenia_assert(*reinterpret_cast<const uint32_t*>(
+  //                 this->memory()->TranslateVirtual(function->address())) != 0);
   // Could scan or something here.
   // Could also check to see if it's a well-known function type and classify
   // for later.

src/xenia/cpu/ppc/ppc_hir_builder.cc

@@ -34,6 +34,11 @@ DEFINE_bool(
     "unimplemented PowerPC instruction is encountered.",
     "CPU");
 
+DEFINE_bool(
+    emit_useless_fpscr_updates, false,
+    "Emit useless fpscr update instructions (pre-10/30/2022 behavior). ",
+    "CPU");
+
 namespace xe {
 namespace cpu {
 namespace ppc {
@@ -89,6 +94,9 @@ bool PPCHIRBuilder::Emit(GuestFunction* function, uint32_t flags) {
 
   function_ = function;
   start_address_ = function_->address();
+  //chrispy: i've seen this one happen, not sure why but i think from trying to precompile twice
+  //i've also seen ones with a start and end address that are the same...
+  assert_true(function_->address() <= function_->end_address());
   instr_count_ = (function_->end_address() - function_->address()) / 4 + 1;
 
   with_debug_info_ = (flags & EMIT_DEBUG_COMMENTS) == EMIT_DEBUG_COMMENTS;
@@ -242,6 +250,7 @@ void PPCHIRBuilder::MaybeBreakOnInstruction(uint32_t address) {
 }
 
 void PPCHIRBuilder::AnnotateLabel(uint32_t address, Label* label) {
+	//chrispy: label->name is unused, it would be nice to be able to remove the field and this code
   char name_buffer[13];
   auto format_result = fmt::format_to_n(name_buffer, 12, "loc_{:08X}", address);
   name_buffer[format_result.size] = '\0';
@@ -447,31 +456,38 @@ void PPCHIRBuilder::StoreFPSCR(Value* value) {
 void PPCHIRBuilder::UpdateFPSCR(Value* result, bool update_cr1) {
   // TODO(benvanik): detect overflow and nan cases.
   // fx and vx are the most important.
-  Value* fx = LoadConstantInt8(0);
+  /*
-  Value* fex = LoadConstantInt8(0);
+          chrispy: stubbed this out because right now all it does is waste
-  Value* vx = LoadConstantInt8(0);
+     memory and CPU time
-  Value* ox = LoadConstantInt8(0);
+  */
+  if (cvars::emit_useless_fpscr_updates) {
+    Value* fx = LoadConstantInt8(0);
+    Value* fex = LoadConstantInt8(0);
+    Value* vx = LoadConstantInt8(0);
+    Value* ox = LoadConstantInt8(0);
 
-  if (update_cr1) {
+    if (update_cr1) {
-    // Store into the CR1 field.
+      // Store into the CR1 field.
-    // We do this instead of just calling CopyFPSCRToCR1 so that we don't
+      // We do this instead of just calling CopyFPSCRToCR1 so that we don't
-    // have to read back the bits and do shifting work.
+      // have to read back the bits and do shifting work.
-    StoreContext(offsetof(PPCContext, cr1.cr1_fx), fx);
+      StoreContext(offsetof(PPCContext, cr1.cr1_fx), fx);
-    StoreContext(offsetof(PPCContext, cr1.cr1_fex), fex);
+      StoreContext(offsetof(PPCContext, cr1.cr1_fex), fex);
-    StoreContext(offsetof(PPCContext, cr1.cr1_vx), vx);
+      StoreContext(offsetof(PPCContext, cr1.cr1_vx), vx);
-    StoreContext(offsetof(PPCContext, cr1.cr1_ox), ox);
+      StoreContext(offsetof(PPCContext, cr1.cr1_ox), ox);
+    }
+
+    // Generate our new bits.
+    Value* new_bits = Shl(ZeroExtend(fx, INT32_TYPE), 31);
+    new_bits = Or(new_bits, Shl(ZeroExtend(fex, INT32_TYPE), 30));
+    new_bits = Or(new_bits, Shl(ZeroExtend(vx, INT32_TYPE), 29));
+    new_bits = Or(new_bits, Shl(ZeroExtend(ox, INT32_TYPE), 28));
+
+    // Mix into fpscr while preserving sticky bits (FX and OX).
+    Value* bits = LoadFPSCR();
+    bits = Or(And(bits, LoadConstantUint32(0x9FFFFFFF)), new_bits);
+    StoreFPSCR(bits);
   }
 
-  // Generate our new bits.
-  Value* new_bits = Shl(ZeroExtend(fx, INT32_TYPE), 31);
-  new_bits = Or(new_bits, Shl(ZeroExtend(fex, INT32_TYPE), 30));
-  new_bits = Or(new_bits, Shl(ZeroExtend(vx, INT32_TYPE), 29));
-  new_bits = Or(new_bits, Shl(ZeroExtend(ox, INT32_TYPE), 28));
-
-  // Mix into fpscr while preserving sticky bits (FX and OX).
-  Value* bits = LoadFPSCR();
-  bits = Or(And(bits, LoadConstantUint32(0x9FFFFFFF)), new_bits);
-  StoreFPSCR(bits);
 }
 
 void PPCHIRBuilder::CopyFPSCRToCR1() {

src/xenia/cpu/ppc/ppc_instr.h

@@ -21,13 +21,7 @@ namespace xe {
 namespace cpu {
 namespace ppc {
 
-// DEPRECATED
+struct PPCOpcodeBits {
-// TODO(benvanik): move code to PPCDecodeData.
-struct InstrData {
-  PPCOpcode opcode;
-  const PPCOpcodeInfo* opcode_info;
-  uint32_t address;
-
   union {
     uint32_t code;
 
@@ -329,6 +323,14 @@ struct InstrData {
   };
 };
 
+// DEPRECATED
+// TODO(benvanik): move code to PPCDecodeData.
+struct InstrData : public PPCOpcodeBits {
+  PPCOpcode opcode;
+  const PPCOpcodeInfo* opcode_info;
+  uint32_t address;
+};
+
 }  // namespace ppc
 }  // namespace cpu
 }  // namespace xe

src/xenia/cpu/xex_module.cc

@@ -31,14 +31,17 @@
 #include "third_party/crypto/rijndael-alg-fst.c"
 #include "third_party/crypto/rijndael-alg-fst.h"
 #include "third_party/pe/pe_image.h"
-
+#include "xenia/cpu/ppc/ppc_decode_data.h"
+#include "xenia/cpu/ppc/ppc_instr.h"
 DEFINE_bool(disable_instruction_infocache, false,
             "Disables caching records of called instructions/mmio accesses.",
             "CPU");
-DEFINE_bool(disable_function_precompilation, true,
+
-            "Disables pre-compiling guest functions that we know we've called "
+DEFINE_bool(
-            "on previous runs",
+    disable_early_precompilation, false,
-            "CPU");
+    "Disables pre-compiling guest functions that we know we've called/that "
+    "we've recognized as being functions via simple heuristics.",
+    "CPU");
 
 static const uint8_t xe_xex2_retail_key[16] = {
     0x20, 0xB1, 0x85, 0xA5, 0x9D, 0x28, 0xFD, 0xC3,
@@ -1057,29 +1060,6 @@ bool XexModule::LoadContinue() {
       library_offset += library->size;
     }
   }
-  sha1::SHA1 final_image_sha_;
-
-  final_image_sha_.reset();
-
-  unsigned high_code = this->high_address_ - this->low_address_;
-
-  final_image_sha_.processBytes(memory()->TranslateVirtual(this->low_address_),
-                                high_code);
-  final_image_sha_.finalize(image_sha_bytes_);
-
-  char fmtbuf[16];
-
-  for (unsigned i = 0; i < 16; ++i) {
-    sprintf_s(fmtbuf, "%X", image_sha_bytes_[i]);
-    image_sha_str_ += &fmtbuf[0];
-  }
-
-  info_cache_.Init(this);
-  // Find __savegprlr_* and __restgprlr_* and the others.
-  // We can flag these for special handling (inlining/etc).
-  if (!FindSaveRest()) {
-    return false;
-  }
 
   // Load a specified module map and diff.
   if (cvars::load_module_map.size()) {
@@ -1112,6 +1092,32 @@ bool XexModule::LoadContinue() {
   return true;
 }
 
+void XexModule::Precompile() {
+  sha1::SHA1 final_image_sha_;
+
+  final_image_sha_.reset();
+
+  unsigned high_code = this->high_address_ - this->low_address_;
+
+  final_image_sha_.processBytes(memory()->TranslateVirtual(this->low_address_),
+                                high_code);
+  final_image_sha_.finalize(image_sha_bytes_);
+
+  char fmtbuf[16];
+
+  for (unsigned i = 0; i < 16; ++i) {
+    sprintf_s(fmtbuf, "%X", image_sha_bytes_[i]);
+    image_sha_str_ += &fmtbuf[0];
+  }
+
+  // Find __savegprlr_* and __restgprlr_* and the others.
+  // We can flag these for special handling (inlining/etc).
+  if (!FindSaveRest()) {
+    return;
+  }
+  info_cache_.Init(this);
+  PrecompileDiscoveredFunctions();
+}
 bool XexModule::Unload() {
   if (!loaded_) {
     return true;
@@ -1363,9 +1369,25 @@ InfoCacheFlags* XexModule::GetInstructionAddressFlags(uint32_t guest_addr) {
 
   return info_cache_.LookupFlags(guest_addr);
 }
+void XexModule::PrecompileDiscoveredFunctions() {
+  if (cvars::disable_early_precompilation) {
+	return;
+  }
+  auto others = PreanalyzeCode();
 
+  for (auto&& other : others) {
+    if (other < low_address_ || other >= high_address_) {
+      continue;
+    }
+    auto sym = processor_->LookupFunction(other);
+
+    if (!sym || sym->status() != Symbol::Status::kDefined) {
+      processor_->ResolveFunction(other);
+    }
+  }
+}
 void XexModule::PrecompileKnownFunctions() {
-  if (cvars::disable_function_precompilation) {
+  if (cvars::disable_early_precompilation) {
     return;
   }
   uint32_t start = 0;
@@ -1374,12 +1396,160 @@ void XexModule::PrecompileKnownFunctions() {
   if (!flags) {
     return;
   }
+  //maybe should pre-acquire global crit?
   for (uint32_t i = 0; i < end; i++) {
     if (flags[i].was_resolved) {
-      processor_->ResolveFunction(low_address_ + (i * 4));
+      uint32_t addr = low_address_ + (i * 4);
+      auto sym = processor_->LookupFunction(addr);
+
+      if (!sym || sym->status() != Symbol::Status::kDefined) {
+        processor_->ResolveFunction(addr);
+      }
     }
   }
 }
+
+static uint32_t GetBLCalledFunction(XexModule* xexmod, uint32_t current_base,
+                                    ppc::PPCOpcodeBits wrd) {
+  int32_t displ = static_cast<int32_t>(ppc::XEEXTS26(wrd.I.LI << 2));
+
+  if (wrd.I.AA) {
+    return static_cast<uint32_t>(displ);
+  } else {
+    return static_cast<uint32_t>(static_cast<int32_t>(current_base) + displ);
+  }
+}
+static bool IsOpcodeBL(unsigned w) {
+  return (w >> (32 - 6)) == 18 && ppc::PPCOpcodeBits{w}.I.LK;
+}
+
+std::vector<uint32_t> XexModule::PreanalyzeCode() {
+  uint32_t low_8_aligned = xe::align<uint32_t>(low_address_, 8);
+  uint32_t high_8_aligned = high_address_ & ~(8U - 1);
+
+  uint32_t n_possible_8byte_addresses = (high_8_aligned - low_8_aligned) / 8;
+  uint32_t* funcstart_candidate_stack =
+      new uint32_t[n_possible_8byte_addresses];
+  uint32_t* funcstart_candstack2 = new uint32_t[n_possible_8byte_addresses];
+
+  uint32_t stack_pos = 0;
+  {
+    // all functions seem to start on 8 byte boundaries, except for obvious ones
+    // like the save/rest funcs
+    uint32_t* range_start =
+        (uint32_t*)memory()->TranslateVirtual(low_8_aligned);
+    uint32_t* range_end = (uint32_t*)memory()->TranslateVirtual(
+        high_8_aligned);  // align down to multiple of 8
+
+    const uint8_t mfspr_r12_lr[4] = {0x7D, 0x88, 0x02, 0xA6};
+
+    // a blr instruction, with 4 zero bytes afterwards to pad the next address
+    // to 8 byte alignment
+    // if we see this prior to our address, we can assume we are a function
+    // start
+    const uint8_t blr[4] = {0x4E, 0x80, 0x0, 0x20};
+
+    uint32_t blr32 = *reinterpret_cast<const uint32_t*>(&blr[0]);
+
+    uint32_t mfspr_r12_lr32 =
+        *reinterpret_cast<const uint32_t*>(&mfspr_r12_lr[0]);
+    /*
+                First pass: detect save of the link register at an eight byte
+       aligned address
+        */
+    for (uint32_t* first_pass = range_start; first_pass < range_end;
+         first_pass += 2) {
+      if (*first_pass == mfspr_r12_lr32) {
+        // Push our newly discovered function start into our list
+        // All addresses in the list are sorted until the second pass
+        funcstart_candidate_stack[stack_pos++] =
+            static_cast<uint32_t>(reinterpret_cast<uintptr_t>(first_pass) -
+                                  reinterpret_cast<uintptr_t>(range_start)) +
+            low_8_aligned;
+      } else if (first_pass[-1] == 0 && *first_pass != 0) {
+        // originally i checked for blr followed by 0, but some functions are
+        // actually aligned to greater boundaries. something that appears to be
+        // longjmp (it occurs in most games, so standard library, and loads ctx,
+        // so longjmp) is aligned to 16 bytes in most games
+        uint32_t* check_iter = &first_pass[-2];
+
+        while (!*check_iter) {
+          --check_iter;
+        }
+
+        XE_LIKELY_IF(*check_iter == blr32) {
+          funcstart_candidate_stack[stack_pos++] =
+              static_cast<uint32_t>(reinterpret_cast<uintptr_t>(first_pass) -
+                                    reinterpret_cast<uintptr_t>(range_start)) +
+              low_8_aligned;
+        }
+      }
+    }
+    uint32_t current_guestaddr = low_8_aligned;
+    // Second pass: detect branch with link instructions and decode the target
+    // address. We can safely assume that if bl is to address, that address is
+    // the start of the function
+    for (uint32_t* second_pass = range_start; second_pass < range_end;
+         second_pass++, current_guestaddr += 4) {
+      uint32_t current_call = xe::byte_swap(*second_pass);
+
+      if (IsOpcodeBL(current_call)) {
+        funcstart_candidate_stack[stack_pos++] = GetBLCalledFunction(
+            this, current_guestaddr, ppc::PPCOpcodeBits{current_call});
+      }
+    }
+
+    auto pdata = this->GetPESection(".pdata");
+
+    if (pdata) {
+      uint32_t* pdata_base =
+          (uint32_t*)this->memory()->TranslateVirtual(pdata->address);
+
+      uint32_t n_pdata_entries = pdata->raw_size / 8;
+
+      for (uint32_t i = 0; i < n_pdata_entries; ++i) {
+        uint32_t funcaddr = xe::load_and_swap<uint32_t>(&pdata_base[i * 2]);
+        if (funcaddr >= low_address_ && funcaddr <= high_address_) {
+          funcstart_candidate_stack[stack_pos++] = funcaddr;
+        } else {
+          // we hit 0 for func addr, that means we're done
+          break;
+        }
+      }
+    }
+  }
+
+  // Sort the list of function starts and then ensure that all addresses are
+  // unique
+  uint32_t n_known_funcaddrs = 0;
+  {
+    // make addresses unique
+
+    std::sort(funcstart_candidate_stack, funcstart_candidate_stack + stack_pos);
+
+    uint32_t read_pos = 0;
+    uint32_t write_pos = 0;
+    uint32_t previous_addr = ~0u;
+    while (read_pos < stack_pos) {
+      uint32_t current_addr = funcstart_candidate_stack[read_pos++];
+
+      if (current_addr != previous_addr) {
+        previous_addr = current_addr;
+        funcstart_candstack2[write_pos++] = current_addr;
+      }
+    }
+    n_known_funcaddrs = write_pos;
+  }
+
+  delete[] funcstart_candidate_stack;
+
+  std::vector<uint32_t> result;
+  result.resize(n_known_funcaddrs);
+  memcpy(&result[0], funcstart_candstack2,
+         sizeof(uint32_t) * n_known_funcaddrs);
+  delete[] funcstart_candstack2;
+  return result;
+}
 bool XexModule::FindSaveRest() {
   // Special stack save/restore functions.
   // http://research.microsoft.com/en-us/um/redmond/projects/invisible/src/crt/md/ppc/xxx.s.htm
@@ -1552,6 +1722,8 @@ bool XexModule::FindSaveRest() {
 
   auto page_size = base_address_ <= 0x90000000 ? 64 * 1024 : 4 * 1024;
   auto sec_header = xex_security_info();
+  std::vector<uint32_t> resolve_on_exit{};
+  resolve_on_exit.reserve(256);
   for (uint32_t i = 0, page = 0; i < sec_header->page_descriptor_count; i++) {
     // Byteswap the bitfield manually.
     xex2_page_descriptor desc;
@@ -1586,13 +1758,20 @@ bool XexModule::FindSaveRest() {
 
   // Add function stubs.
   char name[32];
+
+  auto AddXexFunction = [this, &resolve_on_exit](uint32_t address,
+                                                 Function** function) {
+    DeclareFunction(address, function);
+    resolve_on_exit.push_back(address);
+  };
   if (gplr_start) {
     uint32_t address = gplr_start;
     for (int n = 14; n <= 31; n++) {
       auto format_result =
           fmt::format_to_n(name, xe::countof(name), "__savegprlr_{}", n);
       Function* function;
-      DeclareFunction(address, &function);
+
+      AddXexFunction(address, &function);
       function->set_end_address(address + (31 - n) * 4 + 2 * 4);
       function->set_name(std::string_view(name, format_result.size));
       // TODO(benvanik): set type  fn->type = FunctionSymbol::User;
@@ -1608,7 +1787,7 @@ bool XexModule::FindSaveRest() {
       auto format_result =
           fmt::format_to_n(name, xe::countof(name), "__restgprlr_{}", n);
       Function* function;
-      DeclareFunction(address, &function);
+      AddXexFunction(address, &function);
       function->set_end_address(address + (31 - n) * 4 + 3 * 4);
       function->set_name(std::string_view(name, format_result.size));
       // TODO(benvanik): set type  fn->type = FunctionSymbol::User;
@@ -1625,7 +1804,7 @@ bool XexModule::FindSaveRest() {
       auto format_result =
           fmt::format_to_n(name, xe::countof(name), "__savefpr_{}", n);
       Function* function;
-      DeclareFunction(address, &function);
+      AddXexFunction(address, &function);
       function->set_end_address(address + (31 - n) * 4 + 1 * 4);
       function->set_name(std::string_view(name, format_result.size));
       // TODO(benvanik): set type  fn->type = FunctionSymbol::User;
@@ -1641,7 +1820,7 @@ bool XexModule::FindSaveRest() {
       auto format_result =
           fmt::format_to_n(name, xe::countof(name), "__restfpr_{}", n);
       Function* function;
-      DeclareFunction(address, &function);
+      AddXexFunction(address, &function);
       function->set_end_address(address + (31 - n) * 4 + 1 * 4);
       function->set_name(std::string_view(name, format_result.size));
       // TODO(benvanik): set type  fn->type = FunctionSymbol::User;
@@ -1663,7 +1842,7 @@ bool XexModule::FindSaveRest() {
       auto format_result =
           fmt::format_to_n(name, xe::countof(name), "__savevmx_{}", n);
       Function* function;
-      DeclareFunction(address, &function);
+      AddXexFunction(address, &function);
       function->set_name(std::string_view(name, format_result.size));
       // TODO(benvanik): set type  fn->type = FunctionSymbol::User;
       // TODO(benvanik): set flags fn->flags |= FunctionSymbol::kFlagSaveVmx;
@@ -1677,7 +1856,7 @@ bool XexModule::FindSaveRest() {
       auto format_result =
           fmt::format_to_n(name, xe::countof(name), "__savevmx_{}", n);
       Function* function;
-      DeclareFunction(address, &function);
+      AddXexFunction(address, &function);
       function->set_name(std::string_view(name, format_result.size));
       // TODO(benvanik): set type  fn->type = FunctionSymbol::User;
       // TODO(benvanik): set flags fn->flags |= FunctionSymbol::kFlagSaveVmx;
@@ -1691,7 +1870,7 @@ bool XexModule::FindSaveRest() {
       auto format_result =
           fmt::format_to_n(name, xe::countof(name), "__restvmx_{}", n);
       Function* function;
-      DeclareFunction(address, &function);
+      AddXexFunction(address, &function);
       function->set_name(std::string_view(name, format_result.size));
       // TODO(benvanik): set type  fn->type = FunctionSymbol::User;
       // TODO(benvanik): set flags fn->flags |= FunctionSymbol::kFlagRestVmx;
@@ -1705,7 +1884,7 @@ bool XexModule::FindSaveRest() {
       auto format_result =
           fmt::format_to_n(name, xe::countof(name), "__restvmx_{}", n);
       Function* function;
-      DeclareFunction(address, &function);
+      AddXexFunction(address, &function);
       function->set_name(std::string_view(name, format_result.size));
       // TODO(benvanik): set type  fn->type = FunctionSymbol::User;
       // TODO(benvanik): set flags fn->flags |= FunctionSymbol::kFlagRestVmx;
@@ -1715,7 +1894,15 @@ bool XexModule::FindSaveRest() {
       address += 2 * 4;
     }
   }
-
+  if (!cvars::disable_early_precompilation) {
+    for (auto&& to_ensure_precompiled : resolve_on_exit) {
+      // we want to make sure an address for these functions is available before
+      // any other functions are compiled for code generation purposes but we do
+      // it outside of our loops, because we also want to make sure we've marked
+      // up the symbol with info about it being save/rest and whatnot
+      processor_->ResolveFunction(to_ensure_precompiled);
+    }
+  }
   return true;
 }
 

src/xenia/cpu/xex_module.h

@@ -34,7 +34,8 @@ struct InfoCacheFlags {
   uint32_t was_resolved : 1;  // has this address ever been called/requested
                               // via resolvefunction?
   uint32_t accessed_mmio : 1;
-  uint32_t reserved : 30;
+  uint32_t is_syscall_func : 1;
+  uint32_t reserved : 29;
 };
 struct XexInfoCache {
   struct InfoCacheFlagsHeader {
@@ -208,12 +209,15 @@ class XexModule : public xe::cpu::Module {
   }
 
   InfoCacheFlags* GetInstructionAddressFlags(uint32_t guest_addr);
-  void PrecompileKnownFunctions();
 
+  virtual void Precompile() override; 
  protected:
   std::unique_ptr<Function> CreateFunction(uint32_t address) override;
 
  private:
+  void PrecompileKnownFunctions();
+  void PrecompileDiscoveredFunctions();
+  std::vector<uint32_t> PreanalyzeCode();
   friend struct XexInfoCache;
   void ReadSecurityInfo();
 

src/xenia/debug/ui/debug_window.h

@@ -33,7 +33,7 @@ namespace ui {
 
 class DebugWindow : public cpu::DebugListener {
  public:
-  ~DebugWindow();
+ virtual ~DebugWindow();
 
   static std::unique_ptr<DebugWindow> Create(
       Emulator* emulator, xe::ui::WindowedAppContext& app_context);

src/xenia/gpu/d3d12/d3d12_command_processor.cc

@@ -2042,10 +2042,6 @@ D3D12CommandProcessor::WriteRegisterRangeFromRing_WithKnownBound(
   RingBuffer::ReadRange range =
       ring->BeginRead(num_registers * sizeof(uint32_t));
 
-  constexpr auto bounds_has_reg =
-      bounds_may_have_reg<register_lower_bound, register_upper_bound>;
-  constexpr auto bounds_has_bounds =
-      bounds_may_have_bounds<register_lower_bound, register_upper_bound>;
 
   XE_LIKELY_IF(!range.second) {
     WriteRegisterRangeFromMem_WithKnownBound<register_lower_bound,
@@ -5152,6 +5148,7 @@ void D3D12CommandProcessor::WriteGammaRampSRV(
 #define COMMAND_PROCESSOR D3D12CommandProcessor
 
 #include "../pm4_command_processor_implement.h"
+#undef COMMAND_PROCESSOR
 }  // namespace d3d12
 }  // namespace gpu
 }  // namespace xe

src/xenia/gpu/d3d12/d3d12_command_processor.h

@@ -1,5 +1,3 @@
-/**
-/**
 /**
  ******************************************************************************
  * Xenia : Xbox 360 Emulator Research Project                                 *
@@ -50,8 +48,9 @@ struct MemExportRange {
 };
 class D3D12CommandProcessor final : public CommandProcessor {
  protected:
+#define OVERRIDING_BASE_CMDPROCESSOR
 #include "../pm4_command_processor_declare.h"
-
+#undef OVERRIDING_BASE_CMDPROCESSOR
  public:
   explicit D3D12CommandProcessor(D3D12GraphicsSystem* graphics_system,
                                  kernel::KernelState* kernel_state);

src/xenia/gpu/d3d12/d3d12_nvapi.hpp

@@ -108,15 +108,4 @@ inline nvapi_state_t::~nvapi_state_t() {
     call_deinit_interface();
   }
 }
-inline void init_nvapi() {
-  /// HMODULE moddy = LoadLibraryA("nvapi64.dll");
-
-  // FARPROC quif = GetProcAddress(moddy, "nvapi_QueryInterface");
-
-  nvapi_state_t nvapi{};
-
-  auto queryvisible = nvapi.query_interface<void>(0x26322BC3);
-  return;
-}
-
 }  // namespace lightweight_nvapi

src/xenia/gpu/d3d12/d3d12_texture_cache.h

@@ -87,7 +87,7 @@ class D3D12TextureCache final : public TextureCache {
 
   ~D3D12TextureCache();
 
-  void ClearCache();
+  void ClearCache() override;
 
   void BeginSubmission(uint64_t new_submission_index) override;
   void BeginFrame() override;

src/xenia/gpu/pm4_command_processor_declare.h

@@ -1,8 +1,15 @@
 
 
-void ExecuteIndirectBuffer(uint32_t ptr, uint32_t count) XE_RESTRICT;
+#if defined(OVERRIDING_BASE_CMDPROCESSOR)
-virtual uint32_t ExecutePrimaryBuffer(uint32_t start_index, uint32_t end_index) XE_RESTRICT;
+#define	PM4_OVERRIDE		override
-virtual bool ExecutePacket();
+#else
+#define PM4_OVERRIDE
+#endif
+void ExecuteIndirectBuffer(uint32_t ptr,
+                           uint32_t count) XE_RESTRICT;
+virtual uint32_t ExecutePrimaryBuffer(uint32_t start_index, uint32_t end_index)
+    XE_RESTRICT PM4_OVERRIDE;
+virtual bool ExecutePacket() PM4_OVERRIDE;
 
 public:
 void ExecutePacket(uint32_t ptr, uint32_t count);
@@ -111,4 +118,6 @@ XE_COLD
 bool ExecutePacketType3_CountOverflow(uint32_t count);
 XE_NOINLINE
 XE_COLD
-bool ExecutePacketType0_CountOverflow(uint32_t count);
+bool ExecutePacketType0_CountOverflow(uint32_t count);
+
+#undef PM4_OVERRIDE

src/xenia/gpu/pm4_command_processor_implement.h

@@ -4,32 +4,38 @@ void COMMAND_PROCESSOR::ExecuteIndirectBuffer(uint32_t ptr,
                                               uint32_t count) XE_RESTRICT {
   SCOPE_profile_cpu_f("gpu");
 
+
   trace_writer_.WriteIndirectBufferStart(ptr, count * sizeof(uint32_t));
+  if (count != 0) {
+    RingBuffer old_reader = reader_;
 
-  RingBuffer old_reader = reader_;
+    // Execute commands!
+    new (&reader_)
+        RingBuffer(memory_->TranslatePhysical(ptr), count * sizeof(uint32_t));
+    reader_.set_write_offset(count * sizeof(uint32_t));
+    // prefetch the wraparound range
+    // it likely is already in L3 cache, but in a zen system it may be another
+    // chiplets l3
+    reader_.BeginPrefetchedRead<swcache::PrefetchTag::Level2>(
+        COMMAND_PROCESSOR::GetCurrentRingReadCount());
+    do {
+      if (COMMAND_PROCESSOR::ExecutePacket()) {
+        continue;
+      } else {
+        // Return up a level if we encounter a bad packet.
+        XELOGE("**** INDIRECT RINGBUFFER: Failed to execute packet.");
+        assert_always();
+        // break;
+      }
+    } while (reader_.read_count());
 
-  // Execute commands!
+    trace_writer_.WriteIndirectBufferEnd();
-  new (&reader_)
+    reader_ = old_reader;
-      RingBuffer(memory_->TranslatePhysical(ptr), count * sizeof(uint32_t));
+  } else {
-  reader_.set_write_offset(count * sizeof(uint32_t));
+	  //rare, but i've seen it happen! (and then a division by 0 occurs)
-  // prefetch the wraparound range
+    return;
-  // it likely is already in L3 cache, but in a zen system it may be another
+  }
-  // chiplets l3
-  reader_.BeginPrefetchedRead<swcache::PrefetchTag::Level2>(
-      COMMAND_PROCESSOR::GetCurrentRingReadCount());
-  do {
-    if (COMMAND_PROCESSOR::ExecutePacket()) {
-      continue;
-    } else {
-      // Return up a level if we encounter a bad packet.
-      XELOGE("**** INDIRECT RINGBUFFER: Failed to execute packet.");
-      assert_always();
-      // break;
-    }
-  } while (reader_.read_count());
 
-  trace_writer_.WriteIndirectBufferEnd();
-  reader_ = old_reader;
 }
 
 bool COMMAND_PROCESSOR::ExecutePacket() {

src/xenia/gpu/render_target_cache.h

@@ -229,9 +229,10 @@ class RenderTargetCache {
                     TraceWriter* trace_writer, uint32_t draw_resolution_scale_x,
                     uint32_t draw_resolution_scale_y)
       : register_file_(register_file),
-        draw_extent_estimator_(register_file, memory, trace_writer),
         draw_resolution_scale_x_(draw_resolution_scale_x),
-        draw_resolution_scale_y_(draw_resolution_scale_y) {
+        draw_resolution_scale_y_(draw_resolution_scale_y),
+        draw_extent_estimator_(register_file, memory, trace_writer)
+  {
     assert_not_zero(draw_resolution_scale_x);
     assert_not_zero(draw_resolution_scale_y);
   }

src/xenia/gpu/trace_viewer.cc

@@ -941,7 +941,6 @@ void TraceViewer::DrawVertexFetcher(Shader* shader,
           ImGui::NextColumn();
           break;
         case xenos::VertexFormat::k_2_10_10_10: {
-          auto e0 = LOADEL(uint32_t, 0);
           ImGui::Text("??");
           ImGui::NextColumn();
           ImGui::Text("??");
@@ -1066,8 +1065,6 @@ void ProgressBar(float frac, float width, float height = 0,
   }
   frac = xe::saturate_unsigned(frac);
 
-  const auto fontAtlas = ImGui::GetIO().Fonts;
-
   auto pos = ImGui::GetCursorScreenPos();
   auto col = ImGui::ColorConvertFloat4ToU32(color);
   auto border_col = ImGui::ColorConvertFloat4ToU32(border_color);
@@ -1137,7 +1134,6 @@ void TraceViewer::DrawStateUI() {
   std::memset(&draw_info, 0, sizeof(draw_info));
   switch (opcode) {
     case PM4_DRAW_INDX: {
-      uint32_t dword0 = xe::load_and_swap<uint32_t>(packet_head + 4);
       uint32_t dword1 = xe::load_and_swap<uint32_t>(packet_head + 8);
       draw_info.index_count = dword1 >> 16;
       draw_info.prim_type = static_cast<xenos::PrimitiveType>(dword1 & 0x3F);
@@ -1187,7 +1183,6 @@ void TraceViewer::DrawStateUI() {
   auto enable_mode =
       static_cast<ModeControl>(regs[XE_GPU_REG_RB_MODECONTROL].u32 & 0x7);
 
-  const char* mode_name = "Unknown";
   switch (enable_mode) {
     case ModeControl::kIgnore:
       ImGui::Text("Ignored Command %d", player_->current_command_index());

src/xenia/gpu/vulkan/vulkan_command_processor.h

@@ -48,8 +48,9 @@ namespace vulkan {
 
 class VulkanCommandProcessor final : public CommandProcessor {
  protected:
+#define OVERRIDING_BASE_CMDPROCESSOR
 #include "../pm4_command_processor_declare.h"
-
+#undef OVERRIDING_BASE_CMDPROCESSOR
  public:
   // Single-descriptor layouts for use within a single frame.
   enum class SingleTransientDescriptorLayout {

src/xenia/kernel/kernel_state.cc

@@ -421,6 +421,9 @@ X_RESULT KernelState::FinishLoadingUserModule(
   emulator_->patcher()->ApplyPatchesForTitle(memory_, module->title_id(),
                                              module->hash());
   emulator_->on_patch_apply();
+  if (module->xex_module()) {
+    module->xex_module()->Precompile();
+  }
 
   if (module->is_dll_module() && module->entry_point() && call_entry) {
     // Call DllMain(DLL_PROCESS_ATTACH):

src/xenia/kernel/util/shim_utils.h

@@ -36,7 +36,7 @@ using PPCContext = xe::cpu::ppc::PPCContext;
       (xe::cpu::xe_kernel_export_shim_fn)export_name##_entry);
 
 #define SHIM_MEM_ADDR(a) \
-  ((a) ? ppc_context->kernel_state->memory()->TranslateVirtual(a) : nullptr)
+  ((a) ? ppc_context->TranslateVirtual(a) : nullptr)
 
 #define SHIM_MEM_8(a) xe::load_and_swap<uint8_t>(SHIM_MEM_ADDR(a))
 #define SHIM_MEM_16(a) xe::load_and_swap<uint16_t>(SHIM_MEM_ADDR(a))
@@ -159,7 +159,7 @@ class Param {
       uint32_t stack_ptr =
           uint32_t(init.ppc_context->r[1]) + 0x54 + (ordinal_ - 8) * 8;
       *out_value = xe::load_and_swap<V>(
-          init.ppc_context->kernel_state->memory()->TranslateVirtual(
+          init.ppc_context->TranslateVirtual(
               stack_ptr));
     }
   }
@@ -212,7 +212,7 @@ class PointerParam : public ParamBase<uint32_t> {
   PointerParam(Init& init) : ParamBase(init) {
     host_ptr_ =
         value_
-            ? init.ppc_context->kernel_state->memory()->TranslateVirtual(value_)
+            ? init.ppc_context->TranslateVirtual(value_)
             : nullptr;
   }
   PointerParam(void* host_ptr) : ParamBase(), host_ptr_(host_ptr) {}
@@ -251,8 +251,7 @@ template <typename T>
 class PrimitivePointerParam : public ParamBase<uint32_t> {
  public:
   PrimitivePointerParam(Init& init) : ParamBase(init) {
-    host_ptr_ = value_ ? init.ppc_context->kernel_state->memory()
+    host_ptr_ = value_ ? init.ppc_context->TranslateVirtual<xe::be<T>*>(value_)
-                             ->TranslateVirtual<xe::be<T>*>(value_)
                        : nullptr;
   }
   PrimitivePointerParam(T* host_ptr) : ParamBase() {
@@ -285,7 +284,7 @@ class StringPointerParam : public ParamBase<uint32_t> {
   StringPointerParam(Init& init) : ParamBase(init) {
     host_ptr_ =
         value_
-            ? init.ppc_context->kernel_state->memory()->TranslateVirtual<CHAR*>(
+            ? init.ppc_context->TranslateVirtual<CHAR*>(
                   value_)
             : nullptr;
   }
@@ -311,7 +310,7 @@ class TypedPointerParam : public ParamBase<uint32_t> {
  public:
   TypedPointerParam(Init& init) : ParamBase(init) {
     host_ptr_ =
-        value_ ? init.ppc_context->kernel_state->memory()->TranslateVirtual<T*>(
+        value_ ? init.ppc_context->TranslateVirtual<T*>(
                      value_)
                : nullptr;
   }

src/xenia/kernel/xam/xam_info.cc

@@ -195,7 +195,8 @@ void XCustomRegisterDynamicActions_entry() {
 DECLARE_XAM_EXPORT1(XCustomRegisterDynamicActions, kNone, kStub);
 
 dword_result_t XGetAVPack_entry() {
-  // Value from https://github.com/Free60Project/libxenon/blob/920146f/libxenon/drivers/xenos/xenos_videomodes.h
+  // Value from
+  // https://github.com/Free60Project/libxenon/blob/920146f/libxenon/drivers/xenos/xenos_videomodes.h
   // DWORD
   // Not sure what the values are for this, but 6 is VGA.
   // Other likely values are 3/4/8 for HDMI or something.
@@ -321,11 +322,16 @@ void XamLoaderTerminateTitle_entry() {
 }
 DECLARE_XAM_EXPORT1(XamLoaderTerminateTitle, kNone, kSketchy);
 
-dword_result_t XamAlloc_entry(dword_t unk, dword_t size, lpdword_t out_ptr) {
+dword_result_t XamAlloc_entry(dword_t flags, dword_t size, lpdword_t out_ptr) {
-  assert_true(unk == 0);
+  if (flags & 0x00100000) {  // HEAP_ZERO_memory used unless this flag
+    // do nothing!
+    // maybe we ought to fill it with nonzero garbage, but otherwise this is a
+    // flag we can safely ignore
+  }
 
   // Allocate from the heap. Not sure why XAM does this specially, perhaps
   // it keeps stuff in a separate heap?
+  //chrispy: there is a set of different heaps it uses, an array of them. the top 4 bits of the 32 bit flags seems to select the heap
   uint32_t ptr = kernel_state()->memory()->SystemHeapAlloc(size);
   *out_ptr = ptr;
 

src/xenia/kernel/xam/xam_ui.cc

@@ -55,6 +55,7 @@ class XamDialog : public xe::ui::ImGuiDialog {
   XamDialog(xe::ui::ImGuiDrawer* imgui_drawer)
       : xe::ui::ImGuiDialog(imgui_drawer) {}
 
+  virtual ~XamDialog() {}
   void OnClose() override {
     if (close_callback_) {
       close_callback_();
@@ -254,6 +255,7 @@ class MessageBoxDialog : public XamDialog {
       Close();
     }
   }
+  virtual ~MessageBoxDialog() {}
 
  private:
   bool has_opened_ = false;
@@ -264,8 +266,7 @@ class MessageBoxDialog : public XamDialog {
   uint32_t chosen_button_ = 0;
 };
 
-// https://www.se7ensins.com/forums/threads/working-xshowmessageboxui.844116/
+static dword_result_t XamShowMessageBoxUi(
-dword_result_t XamShowMessageBoxUI_entry(
     dword_t user_index, lpu16string_t title_ptr, lpu16string_t text_ptr,
     dword_t button_count, lpdword_t button_ptrs, dword_t active_button,
     dword_t flags, lpdword_t result_ptr, pointer_t<XAM_OVERLAPPED> overlapped) {
@@ -321,8 +322,28 @@ dword_result_t XamShowMessageBoxUI_entry(
   }
   return result;
 }
+
+// https://www.se7ensins.com/forums/threads/working-xshowmessageboxui.844116/
+dword_result_t XamShowMessageBoxUI_entry(
+    dword_t user_index, lpu16string_t title_ptr, lpu16string_t text_ptr,
+    dword_t button_count, lpdword_t button_ptrs, dword_t active_button,
+    dword_t flags, lpdword_t result_ptr, pointer_t<XAM_OVERLAPPED> overlapped) {
+  return XamShowMessageBoxUi(user_index, title_ptr, text_ptr, button_count,
+                             button_ptrs, active_button, flags, result_ptr,
+                             overlapped);
+}
 DECLARE_XAM_EXPORT1(XamShowMessageBoxUI, kUI, kImplemented);
 
+dword_result_t XamShowMessageBoxUIEx_entry(
+    dword_t user_index, lpu16string_t title_ptr, lpu16string_t text_ptr,
+    dword_t button_count, lpdword_t button_ptrs, dword_t active_button,
+    dword_t flags, dword_t unknown_unused, lpdword_t result_ptr,
+    pointer_t<XAM_OVERLAPPED> overlapped) {
+  return XamShowMessageBoxUi(user_index, title_ptr, text_ptr, button_count,
+                             button_ptrs, active_button, flags, result_ptr,
+                             overlapped);
+}
+DECLARE_XAM_EXPORT1(XamShowMessageBoxUIEx, kUI, kImplemented);
 class KeyboardInputDialog : public XamDialog {
  public:
   KeyboardInputDialog(xe::ui::ImGuiDrawer* imgui_drawer, std::string title,
@@ -347,6 +368,7 @@ class KeyboardInputDialog : public XamDialog {
     xe::string_util::copy_truncating(text_buffer_.data(), default_text_,
                                      text_buffer_.size());
   }
+  virtual ~KeyboardInputDialog() {}
 
   const std::string& text() const { return text_; }
   bool cancelled() const { return cancelled_; }

src/xenia/kernel/xbdm/xbdm_misc.cc

@@ -13,11 +13,12 @@
 #include "xenia/kernel/xbdm/xbdm_private.h"
 #include "xenia/kernel/xthread.h"
 #include "xenia/xbox.h"
-
+//chrispy: no idea what a real valid value is for this
+static constexpr const char DmXboxName[] = "Xbox360Name";
 namespace xe {
-namespace kernel {
+  namespace kernel {
 namespace xbdm {
-
+#define XBDM_SUCCESSFULL 0x02DA0000
 #define MAKE_DUMMY_STUB_PTR(x)             \
   dword_result_t x##_entry() { return 0; } \
   DECLARE_XBDM_EXPORT1(x, kDebug, kStub)
@@ -36,11 +37,27 @@ MAKE_DUMMY_STUB_STATUS(DmFreePool);
 dword_result_t DmGetXbeInfo_entry() {
   // TODO(gibbed): 4D5307DC appears to expect this as success?
   // Unknown arguments -- let's hope things don't explode.
-  return 0x02DA0000;
+  return XBDM_SUCCESSFULL;
 }
 DECLARE_XBDM_EXPORT1(DmGetXbeInfo, kDebug, kStub);
 
-MAKE_DUMMY_STUB_STATUS(DmGetXboxName);
+dword_result_t DmGetXboxName_entry(const ppc_context_t& ctx) {
+  uint64_t arg1 = ctx->r[3];
+  uint64_t arg2 = ctx->r[4];
+  if (!arg1 || !arg2) {
+    return 0x80070057;
+  }
+  char* name_out = ctx->TranslateVirtualGPR<char*>(arg1);
+
+  uint32_t* max_name_chars_ptr = ctx->TranslateVirtualGPR<uint32_t*>(arg2);
+
+  uint32_t max_name_chars = xe::load_and_swap<uint32_t>(max_name_chars_ptr);
+  strncpy(name_out, DmXboxName, sizeof(DmXboxName));
+
+
+  return XBDM_SUCCESSFULL;
+}
+DECLARE_XBDM_EXPORT1(DmGetXboxName, kDebug, kImplemented)
 
 dword_result_t DmIsDebuggerPresent_entry() { return 0; }
 DECLARE_XBDM_EXPORT1(DmIsDebuggerPresent, kDebug, kStub);
@@ -49,15 +66,15 @@ void DmSendNotificationString_entry(lpdword_t unk0_ptr) {}
 DECLARE_XBDM_EXPORT1(DmSendNotificationString, kDebug, kStub);
 
 dword_result_t DmRegisterCommandProcessor_entry(lpdword_t name_ptr,
-                                          lpdword_t handler_fn) {
+                                                lpdword_t handler_fn) {
   // Return success to prevent some games from crashing
   return X_STATUS_SUCCESS;
 }
 DECLARE_XBDM_EXPORT1(DmRegisterCommandProcessor, kDebug, kStub);
 
 dword_result_t DmRegisterCommandProcessorEx_entry(lpdword_t name_ptr,
-                                            lpdword_t handler_fn,
+                                                  lpdword_t handler_fn,
-                                            dword_t unk3) {
+                                                  dword_t unk3) {
   // Return success to prevent some games from stalling
   return X_STATUS_SUCCESS;
 }
@@ -65,9 +82,12 @@ DECLARE_XBDM_EXPORT1(DmRegisterCommandProcessorEx, kDebug, kStub);
 
 MAKE_DUMMY_STUB_STATUS(DmStartProfiling);
 MAKE_DUMMY_STUB_STATUS(DmStopProfiling);
-
+// two arguments, first is num frames i think, second is some kind of pointer to
-dword_result_t DmCaptureStackBackTrace_entry(lpdword_t unk0_ptr,
+// where to capture
-                                             lpdword_t unk1_ptr) {
+dword_result_t DmCaptureStackBackTrace_entry(const ppc_context_t& ctx) {
+  uint32_t nframes = static_cast<uint32_t>(ctx->r[3]);
+  uint8_t* unknown_addr =
+      ctx->TranslateVirtual(static_cast<uint32_t>(ctx->r[4]));
   return X_STATUS_INVALID_PARAMETER;
 }
 DECLARE_XBDM_EXPORT1(DmCaptureStackBackTrace, kDebug, kStub);
@@ -82,7 +102,10 @@ dword_result_t DmWalkLoadedModules_entry(lpdword_t unk0_ptr,
 }
 DECLARE_XBDM_EXPORT1(DmWalkLoadedModules, kDebug, kStub);
 
-void DmMapDevkitDrive_entry() {}
+void DmMapDevkitDrive_entry(const ppc_context_t& ctx) {
+  // games check for nonzero result, failure if nz
+  ctx->r[3] = 0ULL;
+}
 DECLARE_XBDM_EXPORT1(DmMapDevkitDrive, kDebug, kStub);
 
 dword_result_t DmFindPdbSignature_entry(lpdword_t unk0_ptr,

src/xenia/kernel/xboxkrnl/xboxkrnl_rtl.cc

@@ -28,7 +28,11 @@
 namespace xe {
 namespace kernel {
 namespace xboxkrnl {
-
+struct X_STRING {
+  unsigned short length;
+  unsigned short pad;
+  uint32_t ptr;
+};
 // https://msdn.microsoft.com/en-us/library/ff561778
 dword_result_t RtlCompareMemory_entry(lpvoid_t source1, lpvoid_t source2,
                                       dword_t length) {
@@ -142,38 +146,80 @@ dword_result_t RtlLowerChar_entry(dword_t in) {
 }
 DECLARE_XBOXKRNL_EXPORT1(RtlLowerChar, kNone, kImplemented);
 
-dword_result_t RtlCompareString_entry(lpstring_t string_1, lpstring_t string_2,
+static int RtlCompareStringN_impl(uint8_t* string_1, unsigned int string_1_len,
-                                      dword_t case_insensitive) {
+                                  uint8_t* string_2, unsigned int string_2_len,
-  int ret = case_insensitive ? xe_strcasecmp(string_1, string_2)
+                                  int case_insensitive) {
-                             : std::strcmp(string_1, string_2);
+  if (string_1_len == 0xFFFFFFFF) {
-
+    uint8_t* string1_strlen_iter = string_1;
-  return ret;
+    while (*string1_strlen_iter++)
+      ;
+    string_1_len =
+        static_cast<unsigned int>(string1_strlen_iter - string_1 - 1);
+  }
+  if (string_2_len == 0xFFFFFFFF) {
+    uint8_t* string2_strlen_iter = string_2;
+    while (*string2_strlen_iter++)
+      ;
+    string_2_len =
+        static_cast<unsigned int>(string2_strlen_iter - string_2 - 1);
+  }
+  uint8_t* string1_end = &string_1[std::min(string_2_len, string_1_len)];
+  if (case_insensitive) {
+    while (string_1 < string1_end) {
+      unsigned c1 = *string_1++;
+      unsigned c2 = *string_2++;
+      if (c1 != c2) {
+        unsigned cu1 = rtl_upper_table[c1];
+        unsigned cu2 = rtl_upper_table[c2];
+        if (cu1 != cu2) {
+          return cu1 - cu2;
+        }
+      }
+    }
+  } else {
+    while (string_1 < string1_end) {
+      unsigned c1 = *string_1++;
+      unsigned c2 = *string_2++;
+      if (c1 != c2) {
+        return c1 - c2;
+      }
+    }
+  }
+  // why? not sure, but its the original logic
+  return string_1_len - string_2_len;
 }
-DECLARE_XBOXKRNL_EXPORT1(RtlCompareString, kNone, kImplemented);
-
 dword_result_t RtlCompareStringN_entry(lpstring_t string_1,
                                        dword_t string_1_len,
                                        lpstring_t string_2,
                                        dword_t string_2_len,
                                        dword_t case_insensitive) {
-  uint32_t len1 = string_1_len;
+  return RtlCompareStringN_impl(
-  uint32_t len2 = string_2_len;
+      reinterpret_cast<uint8_t*>(string_1.host_address()), string_1_len,
-
+      reinterpret_cast<uint8_t*>(string_2.host_address()), string_2_len,
-  if (string_1_len == 0xFFFF) {
+      case_insensitive);
-    len1 = uint32_t(std::strlen(string_1));
-  }
-  if (string_2_len == 0xFFFF) {
-    len2 = uint32_t(std::strlen(string_2));
-  }
-  auto len = std::min(string_1_len, string_2_len);
-
-  int ret = case_insensitive ? xe_strncasecmp(string_1, string_2, len)
-                             : std::strncmp(string_1, string_2, len);
-
-  return ret;
 }
+
 DECLARE_XBOXKRNL_EXPORT1(RtlCompareStringN, kNone, kImplemented);
 
+dword_result_t RtlCompareString_entry(lpvoid_t string_1, lpvoid_t string_2,
+                                      dword_t case_insensitive) {
+  X_STRING* xs1 = string_1.as<X_STRING*>();
+  X_STRING* xs2 = string_2.as<X_STRING*>();
+
+  unsigned length_1 = xe::load_and_swap<uint16_t>(&xs1->length);
+  unsigned length_2 = xe::load_and_swap<uint16_t>(&xs2->length);
+
+  uint32_t ptr_1 = xe::load_and_swap<uint32_t>(&xs1->ptr);
+
+  uint32_t ptr_2 = xe::load_and_swap<uint32_t>(&xs2->ptr);
+
+  auto kmem = kernel_memory();
+
+  return RtlCompareStringN_impl(
+      kmem->TranslateVirtual<uint8_t*>(ptr_1), length_1,
+      kmem->TranslateVirtual<uint8_t*>(ptr_2), length_2, case_insensitive);
+}
+DECLARE_XBOXKRNL_EXPORT1(RtlCompareString, kNone, kImplemented);
 // https://msdn.microsoft.com/en-us/library/ff561918
 void RtlInitAnsiString_entry(pointer_t<X_ANSI_STRING> destination,
                              lpstring_t source) {
@@ -188,13 +234,13 @@ void RtlInitAnsiString_entry(pointer_t<X_ANSI_STRING> destination,
   destination->pointer = source.guest_address();
 }
 DECLARE_XBOXKRNL_EXPORT1(RtlInitAnsiString, kNone, kImplemented);
-//https://learn.microsoft.com/en-us/windows-hardware/drivers/ddi/wdm/nf-wdm-rtlupcaseunicodechar
+// https://learn.microsoft.com/en-us/windows-hardware/drivers/ddi/wdm/nf-wdm-rtlupcaseunicodechar
 dword_result_t RtlUpcaseUnicodeChar_entry(dword_t SourceCharacter) {
-  return std::use_facet<std::ctype<char16_t>>(std::locale()).toupper(SourceCharacter);
+  return std::use_facet<std::ctype<char16_t>>(std::locale())
+      .toupper(SourceCharacter);
 }
 DECLARE_XBOXKRNL_EXPORT1(RtlUpcaseUnicodeChar, kNone, kImplemented);
 
-
 // https://msdn.microsoft.com/en-us/library/ff561899
 void RtlFreeAnsiString_entry(pointer_t<X_ANSI_STRING> string) {
   if (string->pointer) {
@@ -206,8 +252,8 @@ void RtlFreeAnsiString_entry(pointer_t<X_ANSI_STRING> string) {
 DECLARE_XBOXKRNL_EXPORT1(RtlFreeAnsiString, kNone, kImplemented);
 
 // https://msdn.microsoft.com/en-us/library/ff561934
-void RtlInitUnicodeString_entry(pointer_t<X_UNICODE_STRING> destination,
+pointer_result_t RtlInitUnicodeString_entry(
-                                lpu16string_t source) {
+    pointer_t<X_UNICODE_STRING> destination, lpu16string_t source) {
   if (source) {
     destination->length = (uint16_t)source.value().size() * 2;
     destination->maximum_length = (uint16_t)(source.value().size() + 1) * 2;
@@ -215,6 +261,7 @@ void RtlInitUnicodeString_entry(pointer_t<X_UNICODE_STRING> destination,
   } else {
     destination->reset();
   }
+  return destination.guest_address();
 }
 DECLARE_XBOXKRNL_EXPORT1(RtlInitUnicodeString, kNone, kImplemented);
 
@@ -671,6 +718,26 @@ dword_result_t RtlComputeCrc32_entry(dword_t seed, lpvoid_t buffer,
 }
 DECLARE_XBOXKRNL_EXPORT1(RtlComputeCrc32, kNone, kImplemented);
 
+static void RtlRip_entry(const ppc_context_t& ctx) {
+  uint32_t arg1 = static_cast<uint32_t>(ctx->r[3]);
+  uint32_t arg2 = static_cast<uint32_t>(ctx->r[4]);
+  const char* msg_str1 = "";
+
+  const char* msg_str2 = "";
+
+  if (arg1) {
+    msg_str1 = ctx->TranslateVirtual<const char*>(arg1);
+  }
+
+  if (arg2) {
+    msg_str2 = ctx->TranslateVirtual<const char*>(arg2);
+  }
+
+  XELOGE("RtlRip called, arg1 = {}, arg2 = {}\n", msg_str1, msg_str2);
+
+  //we should break here... not sure what to do exactly
+}
+DECLARE_XBOXKRNL_EXPORT1(RtlRip, kNone, kImportant);
 }  // namespace xboxkrnl
 }  // namespace kernel
 }  // namespace xe

src/xenia/kernel/xboxkrnl/xboxkrnl_threading.cc

@@ -9,7 +9,6 @@
 
 #include <algorithm>
 #include <vector>
-
 #include "xenia/base/atomic.h"
 #include "xenia/base/clock.h"
 #include "xenia/base/logging.h"
@@ -913,7 +912,7 @@ dword_result_t NtWaitForMultipleObjectsEx_entry(
     dword_t count, lpdword_t handles, dword_t wait_type, dword_t wait_mode,
     dword_t alertable, lpqword_t timeout_ptr) {
   uint64_t timeout = timeout_ptr ? static_cast<uint64_t>(*timeout_ptr) : 0u;
-  if (!count || count > 64 || wait_type != 1 && wait_type) {
+  if (!count || count > 64 || (wait_type != 1 && wait_type)) {
     return X_STATUS_INVALID_PARAMETER;
   }
   return xeNtWaitForMultipleObjectsEx(count, handles, wait_type, wait_mode,
@@ -964,7 +963,7 @@ uint32_t xeKeKfAcquireSpinLock(uint32_t* lock, uint64_t r13 = 1) {
   PrefetchForCAS(lock);
   assert_true(*lock != static_cast<uint32_t>(r13));
   // Lock.
-  while (!xe::atomic_cas(0, static_cast<uint32_t>(r13), lock)) {
+  while (!xe::atomic_cas(0, xe::byte_swap(static_cast<uint32_t>(r13)), lock)) {
     // Spin!
     // TODO(benvanik): error on deadlock?
     xe::threading::MaybeYield();
@@ -978,7 +977,7 @@ uint32_t xeKeKfAcquireSpinLock(uint32_t* lock, uint64_t r13 = 1) {
 }
 
 dword_result_t KfAcquireSpinLock_entry(lpdword_t lock_ptr,
-                                       ppc_context_t& ppc_context) {
+                                       const ppc_context_t& ppc_context) {
   auto lock = reinterpret_cast<uint32_t*>(lock_ptr.host_address());
   return xeKeKfAcquireSpinLock(lock, ppc_context->r[13]);
 }
@@ -997,9 +996,7 @@ void xeKeKfReleaseSpinLock(uint32_t* lock, dword_t old_irql) {
 }
 
 void KfReleaseSpinLock_entry(lpdword_t lock_ptr, dword_t old_irql,
-                             ppc_context_t& ppc_ctx) {
+                             const ppc_context_t& ppc_ctx) {
-  auto lock = reinterpret_cast<uint32_t*>(lock_ptr.host_address());
-
   assert_true(*lock_ptr == static_cast<uint32_t>(ppc_ctx->r[13]));
 
   *lock_ptr = 0;
@@ -1014,14 +1011,14 @@ DECLARE_XBOXKRNL_EXPORT2(KfReleaseSpinLock, kThreading, kImplemented,
                          kHighFrequency);
 // todo: this is not accurate
 void KeAcquireSpinLockAtRaisedIrql_entry(lpdword_t lock_ptr,
-                                         ppc_context_t& ppc_ctx) {
+                                         const ppc_context_t& ppc_ctx) {
   // Lock.
   auto lock = reinterpret_cast<uint32_t*>(lock_ptr.host_address());
   // must not be our own thread
   assert_true(*lock_ptr != static_cast<uint32_t>(ppc_ctx->r[13]));
 
   PrefetchForCAS(lock);
-  while (!xe::atomic_cas(0, static_cast<uint32_t>(ppc_ctx->r[13]), lock)) {
+  while (!xe::atomic_cas(0, xe::byte_swap(static_cast<uint32_t>(ppc_ctx->r[13])), lock)) {
 #if XE_ARCH_AMD64 == 1
     // todo: this is just a nop if they don't have SMT, which is not great
     // either...
@@ -1036,12 +1033,12 @@ DECLARE_XBOXKRNL_EXPORT3(KeAcquireSpinLockAtRaisedIrql, kThreading,
                          kImplemented, kBlocking, kHighFrequency);
 
 dword_result_t KeTryToAcquireSpinLockAtRaisedIrql_entry(
-    lpdword_t lock_ptr, ppc_context_t& ppc_ctx) {
+    lpdword_t lock_ptr, const ppc_context_t& ppc_ctx) {
   // Lock.
   auto lock = reinterpret_cast<uint32_t*>(lock_ptr.host_address());
   assert_true(*lock_ptr != static_cast<uint32_t>(ppc_ctx->r[13]));
   PrefetchForCAS(lock);
-  if (!xe::atomic_cas(0, static_cast<uint32_t>(ppc_ctx->r[13]), lock)) {
+  if (!xe::atomic_cas(0, xe::byte_swap(static_cast<uint32_t>(ppc_ctx->r[13])), lock)) {
     return 0;
   }
   return 1;
@@ -1050,10 +1047,9 @@ DECLARE_XBOXKRNL_EXPORT4(KeTryToAcquireSpinLockAtRaisedIrql, kThreading,
                          kImplemented, kBlocking, kHighFrequency, kSketchy);
 
 void KeReleaseSpinLockFromRaisedIrql_entry(lpdword_t lock_ptr,
-                                           ppc_context_t& ppc_ctx) {
+                                           const ppc_context_t& ppc_ctx) {
   // Unlock.
   assert_true(*lock_ptr == static_cast<uint32_t>(ppc_ctx->r[13]));
-  auto lock = reinterpret_cast<uint32_t*>(lock_ptr.host_address());
   *lock_ptr = 0;
 }
 DECLARE_XBOXKRNL_EXPORT2(KeReleaseSpinLockFromRaisedIrql, kThreading,
@@ -1283,7 +1279,8 @@ void ExInitializeReadWriteLock_entry(pointer_t<X_ERWLOCK> lock_ptr) {
 }
 DECLARE_XBOXKRNL_EXPORT1(ExInitializeReadWriteLock, kThreading, kImplemented);
 
-void ExAcquireReadWriteLockExclusive_entry(pointer_t<X_ERWLOCK> lock_ptr, ppc_context_t& ppc_context) {
+void ExAcquireReadWriteLockExclusive_entry(pointer_t<X_ERWLOCK> lock_ptr,
+                                           const ppc_context_t& ppc_context) {
   auto old_irql = xeKeKfAcquireSpinLock(&lock_ptr->spin_lock, ppc_context->r[13]);
 
   int32_t lock_count = ++lock_ptr->lock_count;
@@ -1301,7 +1298,7 @@ DECLARE_XBOXKRNL_EXPORT2(ExAcquireReadWriteLockExclusive, kThreading,
                          kImplemented, kBlocking);
 
 dword_result_t ExTryToAcquireReadWriteLockExclusive_entry(
-    pointer_t<X_ERWLOCK> lock_ptr, ppc_context_t& ppc_context) {
+    pointer_t<X_ERWLOCK> lock_ptr, const ppc_context_t& ppc_context) {
   auto old_irql =
       xeKeKfAcquireSpinLock(&lock_ptr->spin_lock, ppc_context->r[13]);
 
@@ -1320,7 +1317,7 @@ DECLARE_XBOXKRNL_EXPORT1(ExTryToAcquireReadWriteLockExclusive, kThreading,
                          kImplemented);
 
 void ExAcquireReadWriteLockShared_entry(pointer_t<X_ERWLOCK> lock_ptr,
-                                        ppc_context_t& ppc_context) {
+                                        const ppc_context_t& ppc_context) {
   auto old_irql = xeKeKfAcquireSpinLock(&lock_ptr->spin_lock, ppc_context->r[13]);
 
   int32_t lock_count = ++lock_ptr->lock_count;
@@ -1340,7 +1337,7 @@ DECLARE_XBOXKRNL_EXPORT2(ExAcquireReadWriteLockShared, kThreading, kImplemented,
                          kBlocking);
 
 dword_result_t ExTryToAcquireReadWriteLockShared_entry(
-    pointer_t<X_ERWLOCK> lock_ptr, ppc_context_t& ppc_context) {
+    pointer_t<X_ERWLOCK> lock_ptr, const ppc_context_t& ppc_context) {
   auto old_irql =
       xeKeKfAcquireSpinLock(&lock_ptr->spin_lock, ppc_context->r[13]);
 
@@ -1361,7 +1358,7 @@ DECLARE_XBOXKRNL_EXPORT1(ExTryToAcquireReadWriteLockShared, kThreading,
                          kImplemented);
 
 void ExReleaseReadWriteLock_entry(pointer_t<X_ERWLOCK> lock_ptr,
-                                  ppc_context_t& ppc_context) {
+                                  const ppc_context_t& ppc_context) {
   auto old_irql =
       xeKeKfAcquireSpinLock(&lock_ptr->spin_lock, ppc_context->r[13]);
 
@@ -1404,7 +1401,7 @@ pointer_result_t InterlockedPushEntrySList_entry(
   assert_not_null(entry);
 
   alignas(8) X_SLIST_HEADER old_hdr = *plist_ptr;
-  alignas(8) X_SLIST_HEADER new_hdr = {0};
+  alignas(8) X_SLIST_HEADER new_hdr = {{0}, 0, 0};
   uint32_t old_head = 0;
   do {
     old_hdr = *plist_ptr;
@@ -1428,8 +1425,8 @@ pointer_result_t InterlockedPopEntrySList_entry(
   assert_not_null(plist_ptr);
 
   uint32_t popped = 0;
-  alignas(8) X_SLIST_HEADER old_hdr = {0};
+  alignas(8) X_SLIST_HEADER old_hdr = {{0}, 0, 0};
-  alignas(8) X_SLIST_HEADER new_hdr = {0};
+  alignas(8) X_SLIST_HEADER new_hdr = {{0}, 0, 0};
   do {
     old_hdr = *plist_ptr;
     auto next = kernel_memory()->TranslateVirtual<X_SINGLE_LIST_ENTRY*>(
@@ -1456,7 +1453,7 @@ pointer_result_t InterlockedFlushSList_entry(
   assert_not_null(plist_ptr);
 
   alignas(8) X_SLIST_HEADER old_hdr = *plist_ptr;
-  alignas(8) X_SLIST_HEADER new_hdr = {0};
+  alignas(8) X_SLIST_HEADER new_hdr = {{0}, 0, 0};
   uint32_t first = 0;
   do {
     old_hdr = *plist_ptr;

src/xenia/kernel/xboxkrnl/xboxkrnl_video.cc

@@ -433,7 +433,7 @@ void VdSwap_entry(
     return;
   }
   gpu_fetch.base_address = frontbuffer_physical_address >> 12;
-
+  XE_MAYBE_UNUSED
   auto texture_format = gpu::xenos::TextureFormat(texture_format_ptr.value());
   auto color_space = *color_space_ptr;
   assert_true(texture_format == gpu::xenos::TextureFormat::k_8_8_8_8 ||

src/xenia/kernel/xthread.h

@@ -41,8 +41,7 @@ struct XAPC {
   // KAPC is 0x28(40) bytes? (what's passed to ExAllocatePoolWithTag)
   // This is 4b shorter than NT - looks like the reserved dword at +4 is gone.
   // NOTE: stored in guest memory.
-  uint8_t type;                      // +0
+  uint16_t type;                      // +0
-  uint8_t unk1;                      // +1
   uint8_t processor_mode;            // +2
   uint8_t enqueued;                  // +3
   xe::be<uint32_t> thread_ptr;       // +4
@@ -57,7 +56,6 @@ struct XAPC {
 
   void Initialize() {
     type = 18;  // ApcObject
-    unk1 = 0;
     processor_mode = 0;
     enqueued = 0;
     thread_ptr = 0;

src/xenia/memory.cc

@@ -316,9 +316,10 @@ void Memory::Reset() {
   heaps_.v90000000.Reset();
   heaps_.physical.Reset();
 }
+//clang does not like non-standard layout offsetof
+#if XE_COMPILER_MSVC == 1 && XE_COMPILER_CLANG_CL==0
 XE_NOALIAS
 const BaseHeap* Memory::LookupHeap(uint32_t address) const {
-#if 1
 #define HEAP_INDEX(name) \
   offsetof(Memory, heaps_.name) - offsetof(Memory, heaps_)
 
@@ -354,8 +355,11 @@ const BaseHeap* Memory::LookupHeap(uint32_t address) const {
     heap_select = nullptr;
   }
   return reinterpret_cast<const BaseHeap*>(selected_heap_offset + heap_select);
-
+}
 #else
+XE_NOALIAS
+const BaseHeap* Memory::LookupHeap(uint32_t address) const {
+
   if (address < 0x40000000) {
     return &heaps_.v00000000;
   } else if (address < 0x7F000000) {
@@ -375,9 +379,8 @@ const BaseHeap* Memory::LookupHeap(uint32_t address) const {
   } else {
     return nullptr;
   }
-#endif
 }
-
+#endif
 BaseHeap* Memory::LookupHeapByType(bool physical, uint32_t page_size) {
   if (physical) {
     if (page_size <= 4096) {
@@ -1069,7 +1072,7 @@ bool BaseHeap::AllocRange(uint32_t low_address, uint32_t high_address,
   if (start_page_number == UINT_MAX || end_page_number == UINT_MAX) {
     // Out of memory.
     XELOGE("BaseHeap::Alloc failed to find contiguous range");
-    assert_always("Heap exhausted!");
+    //assert_always("Heap exhausted!");
     return false;
   }
 

src/xenia/memory.h

@@ -285,7 +285,8 @@ class PhysicalHeap : public BaseHeap {
   uint32_t GetPhysicalAddress(uint32_t address) const;
 
   uint32_t SystemPagenumToGuestPagenum(uint32_t num) const {
-    return ((num << system_page_shift_) - host_address_offset()) >> page_size_shift_;
+    return ((num << system_page_shift_) - host_address_offset()) >>
+           page_size_shift_;
   }
 
   uint32_t GuestPagenumToSystemPagenum(uint32_t num) {
@@ -294,6 +295,7 @@ class PhysicalHeap : public BaseHeap {
     num >>= system_page_shift_;
     return num;
   }
+
  protected:
   VirtualHeap* parent_heap_;
 
@@ -351,12 +353,21 @@ class Memory {
   // Note that the contents at the specified host address are big-endian.
   template <typename T = uint8_t*>
   inline T TranslateVirtual(uint32_t guest_address) const {
+#if XE_PLATFORM_WIN32 == 1
+    uint8_t* host_address = virtual_membase_ + guest_address;
+    if (guest_address >= 0xE0000000) {
+      host_address += 0x1000;
+    }
+    return reinterpret_cast<T>(host_address);
+#else
     uint8_t* host_address = virtual_membase_ + guest_address;
     const auto heap = LookupHeap(guest_address);
     if (heap) {
       host_address += heap->host_address_offset();
     }
     return reinterpret_cast<T>(host_address);
+
+#endif
   }
 
   // Base address of physical memory in the host address space.

src/xenia/vfs/devices/null_device.cc

@@ -21,7 +21,7 @@ namespace vfs {
 
 NullDevice::NullDevice(const std::string& mount_path,
                        const std::initializer_list<std::string>& null_paths)
-    : Device(mount_path), null_paths_(null_paths), name_("NullDevice") {}
+    : Device(mount_path), name_("NullDevice"), null_paths_(null_paths) {}
 
 NullDevice::~NullDevice() = default;
 

third_party/FFmpeg

@@ -1 +1 @@
-Subproject commit a437fe6d8efef17c8ad33d39f5815032e7adf5d7
+Subproject commit a14f5c03834a79fc401626a4dad7a58a2da0c445