diff --git a/.appveyor.yml b/.appveyor.yml
index 483b57e61..47e28b5ef 100644
--- a/.appveyor.yml
+++ b/.appveyor.yml
@@ -13,6 +13,10 @@ skip_commits:
     - docs/*/*
     - .github/*
     - .github/*/*
+    - src/**/*_posix.*
+    - src/**/*_linux.*
+    - src/**/*_x11.*
+    - src/**/*_gtk.*    
     - LICENSE
     - README.md
     - .azure-pipelines.yml
diff --git a/.clang-format b/.clang-format
index 9f8b391e3..f9aa6536d 100644
--- a/.clang-format
+++ b/.clang-format
@@ -3,3 +3,6 @@ BasedOnStyle: Google
 DerivePointerAlignment: false
 PointerAlignment: Left
 SortIncludes: true
+
+# Regroup causes unnecessary noise due to clang-format bug.
+IncludeBlocks: Preserve
diff --git a/.gitmodules b/.gitmodules
index 9da0b4148..b7fb5ee6e 100644
--- a/.gitmodules
+++ b/.gitmodules
@@ -25,9 +25,6 @@
 [submodule "third_party/premake-export-compile-commands"]
 	path = third_party/premake-export-compile-commands
 	url = https://github.com/xenia-project/premake-export-compile-commands.git
-[submodule "third_party/yaml-cpp"]
-	path = third_party/yaml-cpp
-	url = https://github.com/jbeder/yaml-cpp.git
 [submodule "third_party/spirv-headers"]
 	path = third_party/spirv-headers
 	url = https://github.com/KhronosGroup/SPIRV-Headers.git
diff --git a/.travis.yml b/.travis.yml
index 96ebf723e..0135e71c4 100644
--- a/.travis.yml
+++ b/.travis.yml
@@ -10,10 +10,11 @@ addons:
   apt:
     sources:
       - ubuntu-toolchain-r-test
-      - llvm-toolchain-bionic-7
+      - llvm-toolchain-9
     packages:
-      - clang-7
-      - llvm-7-dev
+      - clang-9
+      - clang-format-9
+      - llvm-9-dev
       - g++-8
       - python3
       - libc++-dev
@@ -28,10 +29,10 @@ addons:
 
 matrix:
   include:
-    - env: C_COMPILER=clang-7 CXX_COMPILER=clang++-7 LINT=true
+    - env: C_COMPILER=clang-9 CXX_COMPILER=clang++-9 LINT=true
       sudo: false
-    - env: C_COMPILER=clang-7 CXX_COMPILER=clang++-7 BUILD=true CONFIG=Debug
-    - env: C_COMPILER=clang-7 CXX_COMPILER=clang++-7 BUILD=true CONFIG=Release
+    - env: C_COMPILER=clang-9 CXX_COMPILER=clang++-9 BUILD=true CONFIG=Debug
+    - env: C_COMPILER=clang-9 CXX_COMPILER=clang++-9 BUILD=true CONFIG=Release
 
 git:
   # We handle submodules ourselves in xenia-build setup.
@@ -44,6 +45,8 @@ before_script:
   # Dump useful info.
   - $CXX --version
   - python3 --version
+  - clang-format-9 --version
+  - clang-format-9 -style=file -dump-config
   # Add Vulkan dependencies.
   - travis_retry wget http://mirrors.kernel.org/ubuntu/pool/universe/v/vulkan/libvulkan1_$LIBVULKAN_VERSION+dfsg1-1_amd64.deb
   - travis_retry wget http://mirrors.kernel.org/ubuntu/pool/universe/v/vulkan/libvulkan-dev_$LIBVULKAN_VERSION+dfsg1-1_amd64.deb
diff --git a/premake5.lua b/premake5.lua
index 20a5cd454..e4a858601 100644
--- a/premake5.lua
+++ b/premake5.lua
@@ -229,7 +229,6 @@ solution("xenia")
   include("third_party/spirv-tools.lua")
   include("third_party/volk.lua")
   include("third_party/xxhash.lua")
-  include("third_party/yaml-cpp.lua")
 
   include("src/xenia")
   include("src/xenia/app")
diff --git a/src/xenia/app/emulator_window.cc b/src/xenia/app/emulator_window.cc
index 7a7bc52f8..c1bc733fb 100644
--- a/src/xenia/app/emulator_window.cc
+++ b/src/xenia/app/emulator_window.cc
@@ -2,7 +2,7 @@
  ******************************************************************************
  * Xenia : Xbox 360 Emulator Research Project                                 *
  ******************************************************************************
- * Copyright 2015 Ben Vanik. All rights reserved.                             *
+ * Copyright 2020 Ben Vanik. All rights reserved.                             *
  * Released under the BSD license - see LICENSE in the root for more details. *
  ******************************************************************************
  */
@@ -172,7 +172,9 @@ bool EmulatorWindow::Initialize() {
         ShowCommitID();
       } break;
 
-      default: { handled = false; } break;
+      default: {
+        handled = false;
+      } break;
     }
     e->set_handled(handled);
   });
diff --git a/src/xenia/app/xenia_main.cc b/src/xenia/app/xenia_main.cc
index 06c47fefc..25f2e1f64 100644
--- a/src/xenia/app/xenia_main.cc
+++ b/src/xenia/app/xenia_main.cc
@@ -89,7 +89,8 @@ class Factory {
 
   void Add(const std::string& name,
            std::function<std::unique_ptr<T>(Args...)> instantiate) {
-    Add(name, []() { return true; }, instantiate);
+    auto always_available = []() { return true; };
+    Add(name, always_available, instantiate);
   }
 
   template <typename DT>
diff --git a/src/xenia/cpu/backend/x64/x64_sequences.cc b/src/xenia/cpu/backend/x64/x64_sequences.cc
index cdd1a6fe5..515862659 100644
--- a/src/xenia/cpu/backend/x64/x64_sequences.cc
+++ b/src/xenia/cpu/backend/x64/x64_sequences.cc
@@ -2,7 +2,7 @@
  ******************************************************************************
  * Xenia : Xbox 360 Emulator Research Project                                 *
  ******************************************************************************
- * Copyright 2019 Ben Vanik. All rights reserved.                             *
+ * Copyright 2020 Ben Vanik. All rights reserved.                             *
  * Released under the BSD license - see LICENSE in the root for more details. *
  ******************************************************************************
  */
@@ -2668,6 +2668,48 @@ struct LOG2_V128 : Sequence<LOG2_V128, I<OPCODE_LOG2, V128Op, V128Op>> {
 };
 EMITTER_OPCODE_TABLE(OPCODE_LOG2, LOG2_F32, LOG2_F64, LOG2_V128);
 
+struct DOT_PRODUCT_V128 {
+  static void Emit(X64Emitter& e, Xmm dest, Xmm src1, Xmm src2, uint8_t imm) {
+    // TODO(benvanik): apparently this is very slow
+    // - find alternative?
+    Xbyak::Label end;
+    e.inLocalLabel();
+
+    // Grab space to put MXCSR.
+    // TODO(gibbed): stick this in TLS or
+    // something?
+    e.sub(e.rsp, 8);
+
+    // Grab MXCSR and mask off the overflow flag,
+    // because it's sticky.
+    e.vstmxcsr(e.dword[e.rsp]);
+    e.mov(e.eax, e.dword[e.rsp]);
+    e.and_(e.eax, uint32_t(~8));
+    e.mov(e.dword[e.rsp], e.eax);
+    e.vldmxcsr(e.dword[e.rsp]);
+
+    // Hey we can do the dot product now.
+    e.vdpps(dest, src1, src2, imm);
+
+    // Load MXCSR...
+    e.vstmxcsr(e.dword[e.rsp]);
+
+    // ..free our temporary space and get MXCSR at
+    // the same time
+    e.pop(e.rax);
+
+    // Did we overflow?
+    e.test(e.al, 8);
+    e.jz(end);
+
+    // Infinity? HA! Give NAN.
+    e.vmovdqa(dest, e.GetXmmConstPtr(XMMQNaN));
+
+    e.L(end);
+    e.outLocalLabel();
+  }
+};
+
 // ============================================================================
 // OPCODE_DOT_PRODUCT_3
 // ============================================================================
@@ -2676,12 +2718,10 @@ struct DOT_PRODUCT_3_V128
              I<OPCODE_DOT_PRODUCT_3, F32Op, V128Op, V128Op>> {
   static void Emit(X64Emitter& e, const EmitArgType& i) {
     // https://msdn.microsoft.com/en-us/library/bb514054(v=vs.90).aspx
-    EmitCommutativeBinaryXmmOp(e, i,
-                               [](X64Emitter& e, Xmm dest, Xmm src1, Xmm src2) {
-                                 // TODO(benvanik): apparently this is very slow
-                                 // - find alternative?
-                                 e.vdpps(dest, src1, src2, 0b01110001);
-                               });
+    EmitCommutativeBinaryXmmOp(
+        e, i, [](X64Emitter& e, Xmm dest, Xmm src1, Xmm src2) {
+          DOT_PRODUCT_V128::Emit(e, dest, src1, src2, 0b01110001);
+        });
   }
 };
 EMITTER_OPCODE_TABLE(OPCODE_DOT_PRODUCT_3, DOT_PRODUCT_3_V128);
@@ -2694,12 +2734,10 @@ struct DOT_PRODUCT_4_V128
              I<OPCODE_DOT_PRODUCT_4, F32Op, V128Op, V128Op>> {
   static void Emit(X64Emitter& e, const EmitArgType& i) {
     // https://msdn.microsoft.com/en-us/library/bb514054(v=vs.90).aspx
-    EmitCommutativeBinaryXmmOp(e, i,
-                               [](X64Emitter& e, Xmm dest, Xmm src1, Xmm src2) {
-                                 // TODO(benvanik): apparently this is very slow
-                                 // - find alternative?
-                                 e.vdpps(dest, src1, src2, 0b11110001);
-                               });
+    EmitCommutativeBinaryXmmOp(
+        e, i, [](X64Emitter& e, Xmm dest, Xmm src1, Xmm src2) {
+          DOT_PRODUCT_V128::Emit(e, dest, src1, src2, 0b11110001);
+        });
   }
 };
 EMITTER_OPCODE_TABLE(OPCODE_DOT_PRODUCT_4, DOT_PRODUCT_4_V128);
diff --git a/src/xenia/cpu/compiler/passes/register_allocation_pass.cc b/src/xenia/cpu/compiler/passes/register_allocation_pass.cc
index 1b1f4acd3..bd7380184 100644
--- a/src/xenia/cpu/compiler/passes/register_allocation_pass.cc
+++ b/src/xenia/cpu/compiler/passes/register_allocation_pass.cc
@@ -2,7 +2,7 @@
  ******************************************************************************
  * Xenia : Xbox 360 Emulator Research Project                                 *
  ******************************************************************************
- * Copyright 2014 Ben Vanik. All rights reserved.                             *
+ * Copyright 2020 Ben Vanik. All rights reserved.                             *
  * Released under the BSD license - see LICENSE in the root for more details. *
  ******************************************************************************
  */
@@ -347,9 +347,9 @@ bool RegisterAllocationPass::SpillOneRegister(HIRBuilder* builder, Block* block,
   DumpUsage("SpillOneRegister (pre)");
   // Pick the one with the furthest next use.
   assert_true(!usage_set->upcoming_uses.empty());
-  auto furthest_usage = std::max_element(usage_set->upcoming_uses.begin(),
-                                         usage_set->upcoming_uses.end(),
-                                         RegisterUsage::Comparer());
+  auto furthest_usage =
+      std::max_element(usage_set->upcoming_uses.begin(),
+                       usage_set->upcoming_uses.end(), &RegisterUsage::Compare);
   assert_true(furthest_usage->value->def->block == block);
   assert_true(furthest_usage->use->instr->block == block);
   auto spill_value = furthest_usage->value;
diff --git a/src/xenia/cpu/compiler/passes/register_allocation_pass.h b/src/xenia/cpu/compiler/passes/register_allocation_pass.h
index f14c41dc6..27f7e6560 100644
--- a/src/xenia/cpu/compiler/passes/register_allocation_pass.h
+++ b/src/xenia/cpu/compiler/passes/register_allocation_pass.h
@@ -2,7 +2,7 @@
  ******************************************************************************
  * Xenia : Xbox 360 Emulator Research Project                                 *
  ******************************************************************************
- * Copyright 2014 Ben Vanik. All rights reserved.                             *
+ * Copyright 2020 Ben Vanik. All rights reserved.                             *
  * Released under the BSD license - see LICENSE in the root for more details. *
  ******************************************************************************
  */
@@ -39,11 +39,9 @@ class RegisterAllocationPass : public CompilerPass {
     RegisterUsage() : value(nullptr), use(nullptr) {}
     RegisterUsage(hir::Value* value_, hir::Value::Use* use_)
         : value(value_), use(use_) {}
-    struct Comparer : std::binary_function<RegisterUsage, RegisterUsage, bool> {
-      bool operator()(const RegisterUsage& a, const RegisterUsage& b) const {
-        return a.use->instr->ordinal < b.use->instr->ordinal;
-      }
-    };
+    static bool Compare(const RegisterUsage& a, const RegisterUsage& b) {
+      return a.use->instr->ordinal < b.use->instr->ordinal;
+    }
   };
   struct RegisterSetUsage {
     const backend::MachineInfo::RegisterSet* set = nullptr;
diff --git a/src/xenia/debug/ui/debug_window.cc b/src/xenia/debug/ui/debug_window.cc
index 5d9472359..2fd758d24 100644
--- a/src/xenia/debug/ui/debug_window.cc
+++ b/src/xenia/debug/ui/debug_window.cc
@@ -2,7 +2,7 @@
  ******************************************************************************
  * Xenia : Xbox 360 Emulator Research Project                                 *
  ******************************************************************************
- * Copyright 2015 Ben Vanik. All rights reserved.                             *
+ * Copyright 2020 Ben Vanik. All rights reserved.                             *
  * Released under the BSD license - see LICENSE in the root for more details. *
  ******************************************************************************
  */
@@ -17,7 +17,6 @@
 #include "third_party/capstone/include/capstone/x86.h"
 #include "third_party/imgui/imgui.h"
 #include "third_party/imgui/imgui_internal.h"
-#include "third_party/yaml-cpp/include/yaml-cpp/yaml.h"
 #include "xenia/base/clock.h"
 #include "xenia/base/logging.h"
 #include "xenia/base/math.h"
diff --git a/src/xenia/debug/ui/premake5.lua b/src/xenia/debug/ui/premake5.lua
index 84ca9bbd8..7503528b3 100644
--- a/src/xenia/debug/ui/premake5.lua
+++ b/src/xenia/debug/ui/premake5.lua
@@ -11,11 +11,9 @@ project("xenia-debug-ui")
     "xenia-base",
     "xenia-cpu",
     "xenia-ui",
-    "yaml-cpp",
   })
   defines({
   })
   includedirs({
-    project_root.."/third_party/yaml-cpp/include/",
   })
   local_platform_files()
diff --git a/src/xenia/gpu/command_processor.cc b/src/xenia/gpu/command_processor.cc
index 98566a35a..658e7f707 100644
--- a/src/xenia/gpu/command_processor.cc
+++ b/src/xenia/gpu/command_processor.cc
@@ -1157,44 +1157,51 @@ bool CommandProcessor::ExecutePacketType3_DRAW_INDX(RingBuffer* reader,
   // ID = dword0 & 0x3F;
   // use = dword0 & 0x40;
   uint32_t dword0 = reader->ReadAndSwap<uint32_t>();  // viz query info
-  uint32_t dword1 = reader->ReadAndSwap<uint32_t>();
-  uint32_t index_count = dword1 >> 16;
-  auto prim_type = static_cast<PrimitiveType>(dword1 & 0x3F);
+  reg::VGT_DRAW_INITIATOR vgt_draw_initiator;
+  vgt_draw_initiator.value = reader->ReadAndSwap<uint32_t>();
+  WriteRegister(XE_GPU_REG_VGT_DRAW_INITIATOR, vgt_draw_initiator.value);
+
   bool is_indexed = false;
   IndexBufferInfo index_buffer_info;
-  uint32_t src_sel = (dword1 >> 6) & 0x3;
-  if (src_sel == 0x0) {
-    // DI_SRC_SEL_DMA
-    // Indexed draw.
-    is_indexed = true;
-    index_buffer_info.guest_base = reader->ReadAndSwap<uint32_t>();
-    uint32_t index_size = reader->ReadAndSwap<uint32_t>();
-    index_buffer_info.endianness = static_cast<Endian>(index_size >> 30);
-    index_size &= 0x00FFFFFF;
-    bool index_32bit = (dword1 >> 11) & 0x1;
-    index_buffer_info.format =
-        index_32bit ? IndexFormat::kInt32 : IndexFormat::kInt16;
-    index_size *= index_32bit ? 4 : 2;
-    index_buffer_info.length = index_size;
-    index_buffer_info.count = index_count;
-  } else if (src_sel == 0x1) {
-    // DI_SRC_SEL_IMMEDIATE
-    assert_always();
-  } else if (src_sel == 0x2) {
-    // DI_SRC_SEL_AUTO_INDEX
-    // Auto draw.
-    index_buffer_info.guest_base = 0;
-    index_buffer_info.length = 0;
-  } else {
-    // Invalid source select.
-    assert_always();
+  switch (vgt_draw_initiator.source_select) {
+    case xenos::SourceSelect::kDMA: {
+      // Indexed draw.
+      is_indexed = true;
+      index_buffer_info.guest_base = reader->ReadAndSwap<uint32_t>();
+      uint32_t index_size = reader->ReadAndSwap<uint32_t>();
+      index_buffer_info.endianness = static_cast<Endian>(index_size >> 30);
+      index_size &= 0x00FFFFFF;
+      index_buffer_info.format = vgt_draw_initiator.index_size;
+      index_size *=
+          (vgt_draw_initiator.index_size == IndexFormat::kInt32) ? 4 : 2;
+      index_buffer_info.length = index_size;
+      index_buffer_info.count = vgt_draw_initiator.num_indices;
+    } break;
+    case xenos::SourceSelect::kImmediate: {
+      // TODO(Triang3l): VGT_IMMED_DATA.
+      assert_always();
+    } break;
+    case xenos::SourceSelect::kAutoIndex: {
+      // Auto draw.
+      index_buffer_info.guest_base = 0;
+      index_buffer_info.length = 0;
+    } break;
+    default: {
+      // Invalid source select.
+      assert_always();
+    } break;
   }
 
-  bool success = IssueDraw(prim_type, index_count,
-                           is_indexed ? &index_buffer_info : nullptr);
+  bool success =
+      IssueDraw(vgt_draw_initiator.prim_type, vgt_draw_initiator.num_indices,
+                is_indexed ? &index_buffer_info : nullptr,
+                xenos::IsMajorModeExplicit(vgt_draw_initiator.major_mode,
+                                           vgt_draw_initiator.prim_type));
   if (!success) {
-    XELOGE("PM4_DRAW_INDX(%d, %d, %d): Failed in backend", index_count,
-           prim_type, src_sel);
+    XELOGE("PM4_DRAW_INDX(%d, %d, %d): Failed in backend",
+           vgt_draw_initiator.num_indices,
+           uint32_t(vgt_draw_initiator.prim_type),
+           uint32_t(vgt_draw_initiator.source_select));
   }
 
   return true;
@@ -1204,21 +1211,27 @@ bool CommandProcessor::ExecutePacketType3_DRAW_INDX_2(RingBuffer* reader,
                                                       uint32_t packet,
                                                       uint32_t count) {
   // draw using supplied indices in packet
-  uint32_t dword0 = reader->ReadAndSwap<uint32_t>();
-  uint32_t index_count = dword0 >> 16;
-  auto prim_type = static_cast<PrimitiveType>(dword0 & 0x3F);
-  uint32_t src_sel = (dword0 >> 6) & 0x3;
-  assert_true(src_sel == 0x2);  // 'SrcSel=AutoIndex'
+  reg::VGT_DRAW_INITIATOR vgt_draw_initiator;
+  vgt_draw_initiator.value = reader->ReadAndSwap<uint32_t>();
+  WriteRegister(XE_GPU_REG_VGT_DRAW_INITIATOR, vgt_draw_initiator.value);
+  assert_true(vgt_draw_initiator.source_select ==
+              xenos::SourceSelect::kAutoIndex);
   // Index buffer unused as automatic.
-  // bool index_32bit = (dword0 >> 11) & 0x1;
-  // uint32_t indices_size = index_count * (index_32bit ? 4 : 2);
+  // uint32_t indices_size =
+  //     vgt_draw_initiator.num_indices *
+  //         (vgt_draw_initiator.index_size == IndexFormat::kInt32 ? 4 : 2);
   // uint32_t index_ptr = reader->ptr();
+  // TODO(Triang3l): VGT_IMMED_DATA.
   reader->AdvanceRead((count - 1) * sizeof(uint32_t));
 
-  bool success = IssueDraw(prim_type, index_count, nullptr);
+  bool success = IssueDraw(
+      vgt_draw_initiator.prim_type, vgt_draw_initiator.num_indices, nullptr,
+      xenos::IsMajorModeExplicit(vgt_draw_initiator.major_mode,
+                                 vgt_draw_initiator.prim_type));
   if (!success) {
-    XELOGE("PM4_DRAW_INDX_IMM(%d, %d): Failed in backend", index_count,
-           prim_type);
+    XELOGE("PM4_DRAW_INDX_IMM(%d, %d): Failed in backend",
+           vgt_draw_initiator.num_indices,
+           uint32_t(vgt_draw_initiator.prim_type));
   }
 
   return true;
diff --git a/src/xenia/gpu/command_processor.h b/src/xenia/gpu/command_processor.h
index c224e4504..caa53b33a 100644
--- a/src/xenia/gpu/command_processor.h
+++ b/src/xenia/gpu/command_processor.h
@@ -239,7 +239,8 @@ class CommandProcessor {
                              uint32_t dword_count) = 0;
 
   virtual bool IssueDraw(PrimitiveType prim_type, uint32_t index_count,
-                         IndexBufferInfo* index_buffer_info) = 0;
+                         IndexBufferInfo* index_buffer_info,
+                         bool major_mode_explicit) = 0;
   virtual bool IssueCopy() = 0;
 
   virtual void InitializeTrace() = 0;
diff --git a/src/xenia/gpu/d3d12/d3d12_command_processor.cc b/src/xenia/gpu/d3d12/d3d12_command_processor.cc
index a1088f427..ae05946f6 100644
--- a/src/xenia/gpu/d3d12/d3d12_command_processor.cc
+++ b/src/xenia/gpu/d3d12/d3d12_command_processor.cc
@@ -1245,7 +1245,8 @@ Shader* D3D12CommandProcessor::LoadShader(ShaderType shader_type,
 
 bool D3D12CommandProcessor::IssueDraw(PrimitiveType primitive_type,
                                       uint32_t index_count,
-                                      IndexBufferInfo* index_buffer_info) {
+                                      IndexBufferInfo* index_buffer_info,
+                                      bool major_mode_explicit) {
   auto device = GetD3D12Context()->GetD3D12Provider()->GetDevice();
   auto& regs = *register_file_;
 
@@ -1272,8 +1273,7 @@ bool D3D12CommandProcessor::IssueDraw(PrimitiveType primitive_type,
 
   // Check if using tessellation to get the correct primitive type.
   bool tessellated;
-  if (uint32_t(primitive_type) >=
-      uint32_t(PrimitiveType::kExplicitMajorModeForceStart)) {
+  if (major_mode_explicit) {
     tessellated = regs.Get<reg::VGT_OUTPUT_PATH_CNTL>().path_select ==
                   xenos::VGTOutputPath::kTessellationEnable;
   } else {
@@ -2115,8 +2115,7 @@ bool D3D12CommandProcessor::EndSubmission(bool is_swap) {
       }
       root_signatures_.clear();
 
-      // TODO(Triang3l): Shared memory cache clear.
-      // shared_memory_->ClearCache();
+      shared_memory_->ClearCache();
     }
   }
 
diff --git a/src/xenia/gpu/d3d12/d3d12_command_processor.h b/src/xenia/gpu/d3d12/d3d12_command_processor.h
index b0ed394cc..c6bfa8c2e 100644
--- a/src/xenia/gpu/d3d12/d3d12_command_processor.h
+++ b/src/xenia/gpu/d3d12/d3d12_command_processor.h
@@ -175,7 +175,8 @@ class D3D12CommandProcessor : public CommandProcessor {
                      uint32_t dword_count) override;
 
   bool IssueDraw(PrimitiveType primitive_type, uint32_t index_count,
-                 IndexBufferInfo* index_buffer_info) override;
+                 IndexBufferInfo* index_buffer_info,
+                 bool major_mode_explicit) override;
   bool IssueCopy() override;
 
   void InitializeTrace() override;
diff --git a/src/xenia/gpu/d3d12/shared_memory.cc b/src/xenia/gpu/d3d12/shared_memory.cc
index a88537672..e87567d01 100644
--- a/src/xenia/gpu/d3d12/shared_memory.cc
+++ b/src/xenia/gpu/d3d12/shared_memory.cc
@@ -140,7 +140,22 @@ bool SharedMemory::Initialize() {
 void SharedMemory::Shutdown() {
   ResetTraceGPUWrittenBuffer();
 
-  // TODO(Triang3l): Do something in case any watches are still registered.
+  FireWatches(0, (kBufferSize - 1) >> page_size_log2_, false);
+  assert_true(global_watches_.empty());
+  // No watches now, so no references to the pools accessible by guest threads -
+  // safe not to enter the global critical region.
+  watch_node_first_free_ = nullptr;
+  watch_node_current_pool_allocated_ = 0;
+  for (WatchNode* pool : watch_node_pools_) {
+    delete[] pool;
+  }
+  watch_node_pools_.clear();
+  watch_range_first_free_ = nullptr;
+  watch_range_current_pool_allocated_ = 0;
+  for (WatchRange* pool : watch_range_pools_) {
+    delete[] pool;
+  }
+  watch_range_pools_.clear();
 
   if (memory_invalidation_callback_handle_ != nullptr) {
     memory_->UnregisterPhysicalMemoryInvalidationCallback(
@@ -164,6 +179,36 @@ void SharedMemory::Shutdown() {
   }
 }
 
+void SharedMemory::ClearCache() {
+  upload_buffer_pool_->ClearCache();
+
+  // Keeping GPU-written data, so "invalidated by GPU".
+  FireWatches(0, (kBufferSize - 1) >> page_size_log2_, true);
+  // No watches now, so no references to the pools accessible by guest threads -
+  // safe not to enter the global critical region.
+  watch_node_first_free_ = nullptr;
+  watch_node_current_pool_allocated_ = 0;
+  for (WatchNode* pool : watch_node_pools_) {
+    delete[] pool;
+  }
+  watch_node_pools_.clear();
+  watch_range_first_free_ = nullptr;
+  watch_range_current_pool_allocated_ = 0;
+  for (WatchRange* pool : watch_range_pools_) {
+    delete[] pool;
+  }
+  watch_range_pools_.clear();
+
+  {
+    auto global_lock = global_critical_region_.Acquire();
+    for (SystemPageFlagsBlock& block : system_page_flags_) {
+      block.valid = block.valid_and_gpu_written;
+    }
+  }
+
+  // TODO(Triang3l): Unmap and destroy heaps.
+}
+
 void SharedMemory::CompletedSubmissionUpdated() {
   upload_buffer_pool_->Reclaim(command_processor_->GetCompletedSubmission());
 }
diff --git a/src/xenia/gpu/d3d12/shared_memory.h b/src/xenia/gpu/d3d12/shared_memory.h
index af99fa15b..d0eec5093 100644
--- a/src/xenia/gpu/d3d12/shared_memory.h
+++ b/src/xenia/gpu/d3d12/shared_memory.h
@@ -37,6 +37,7 @@ class SharedMemory {
 
   bool Initialize();
   void Shutdown();
+  void ClearCache();
 
   ID3D12Resource* GetBuffer() const { return buffer_; }
   D3D12_GPU_VIRTUAL_ADDRESS GetGPUAddress() const {
diff --git a/src/xenia/gpu/null/null_command_processor.cc b/src/xenia/gpu/null/null_command_processor.cc
index ba5500acd..b19e7fcb0 100644
--- a/src/xenia/gpu/null/null_command_processor.cc
+++ b/src/xenia/gpu/null/null_command_processor.cc
@@ -44,7 +44,8 @@ Shader* NullCommandProcessor::LoadShader(ShaderType shader_type,
 
 bool NullCommandProcessor::IssueDraw(PrimitiveType prim_type,
                                      uint32_t index_count,
-                                     IndexBufferInfo* index_buffer_info) {
+                                     IndexBufferInfo* index_buffer_info,
+                                     bool major_mode_explicit) {
   return true;
 }
 
diff --git a/src/xenia/gpu/null/null_command_processor.h b/src/xenia/gpu/null/null_command_processor.h
index 916668269..f33c1b126 100644
--- a/src/xenia/gpu/null/null_command_processor.h
+++ b/src/xenia/gpu/null/null_command_processor.h
@@ -41,7 +41,8 @@ class NullCommandProcessor : public CommandProcessor {
                      uint32_t dword_count) override;
 
   bool IssueDraw(PrimitiveType prim_type, uint32_t index_count,
-                 IndexBufferInfo* index_buffer_info) override;
+                 IndexBufferInfo* index_buffer_info,
+                 bool major_mode_explicit) override;
   bool IssueCopy() override;
 
   void InitializeTrace() override;
diff --git a/src/xenia/gpu/register_table.inc b/src/xenia/gpu/register_table.inc
index c305b5a03..8ac7ec5d2 100644
--- a/src/xenia/gpu/register_table.inc
+++ b/src/xenia/gpu/register_table.inc
@@ -131,7 +131,10 @@ XE_GPU_REGISTER(0x2182, kDword, SQ_INTERPOLATOR_CNTL)
 XE_GPU_REGISTER(0x2183, kDword, SQ_WRAPPING_0)
 XE_GPU_REGISTER(0x2184, kDword, SQ_WRAPPING_1)
 
+// These three registers are set by the command processor.
 XE_GPU_REGISTER(0x21F9, kDword, VGT_EVENT_INITIATOR)
+XE_GPU_REGISTER(0x21FC, kDword, VGT_DRAW_INITIATOR)
+XE_GPU_REGISTER(0x21FD, kDword, VGT_IMMED_DATA)
 
 XE_GPU_REGISTER(0x2200, kDword, RB_DEPTHCONTROL)
 XE_GPU_REGISTER(0x2201, kDword, RB_BLENDCONTROL0)
diff --git a/src/xenia/gpu/registers.h b/src/xenia/gpu/registers.h
index 5b6fdc54b..af5ee74d9 100644
--- a/src/xenia/gpu/registers.h
+++ b/src/xenia/gpu/registers.h
@@ -145,6 +145,22 @@ union SQ_CONTEXT_MISC {
 
 *******************************************************************************/
 
+union VGT_DRAW_INITIATOR {
+  // Different than on A2xx and R6xx/R7xx.
+  struct {
+    PrimitiveType prim_type : 6;            // +0
+    xenos::SourceSelect source_select : 2;  // +6
+    xenos::MajorMode major_mode : 2;        // +8
+    uint32_t : 1;                           // +10
+    IndexFormat index_size : 1;             // +11
+    uint32_t not_eop : 1;                   // +12
+    uint32_t : 3;                           // +13
+    uint32_t num_indices : 16;              // +16
+  };
+  uint32_t value;
+  static constexpr Register register_index = XE_GPU_REG_VGT_DRAW_INITIATOR;
+};
+
 union VGT_OUTPUT_PATH_CNTL {
   struct {
     xenos::VGTOutputPath path_select : 2;  // +0
diff --git a/src/xenia/gpu/trace_viewer.cc b/src/xenia/gpu/trace_viewer.cc
index 77ab7c022..d11faaecc 100644
--- a/src/xenia/gpu/trace_viewer.cc
+++ b/src/xenia/gpu/trace_viewer.cc
@@ -2,7 +2,7 @@
  ******************************************************************************
  * Xenia : Xbox 360 Emulator Research Project                                 *
  ******************************************************************************
- * Copyright 2015 Ben Vanik. All rights reserved.                             *
+ * Copyright 2020 Ben Vanik. All rights reserved.                             *
  * Released under the BSD license - see LICENSE in the root for more details. *
  ******************************************************************************
  */
@@ -123,9 +123,9 @@ bool TraceViewer::Setup() {
 
   // Create the emulator but don't initialize so we can setup the window.
   emulator_ = std::make_unique<Emulator>(L"", L"");
-  X_STATUS result =
-      emulator_->Setup(window_.get(), nullptr,
-                       [this]() { return CreateGraphicsSystem(); }, nullptr);
+  X_STATUS result = emulator_->Setup(
+      window_.get(), nullptr, [this]() { return CreateGraphicsSystem(); },
+      nullptr);
   if (XFAILED(result)) {
     XELOGE("Failed to setup emulator: %.8X", result);
     return false;
diff --git a/src/xenia/gpu/vk/vulkan_command_processor.cc b/src/xenia/gpu/vk/vulkan_command_processor.cc
index 62bdb5677..d7c46090f 100644
--- a/src/xenia/gpu/vk/vulkan_command_processor.cc
+++ b/src/xenia/gpu/vk/vulkan_command_processor.cc
@@ -40,7 +40,8 @@ Shader* VulkanCommandProcessor::LoadShader(ShaderType shader_type,
 
 bool VulkanCommandProcessor::IssueDraw(PrimitiveType primitive_type,
                                        uint32_t index_count,
-                                       IndexBufferInfo* index_buffer_info) {
+                                       IndexBufferInfo* index_buffer_info,
+                                       bool major_mode_explicit) {
   return true;
 }
 
diff --git a/src/xenia/gpu/vk/vulkan_command_processor.h b/src/xenia/gpu/vk/vulkan_command_processor.h
index 8157c3590..43e621814 100644
--- a/src/xenia/gpu/vk/vulkan_command_processor.h
+++ b/src/xenia/gpu/vk/vulkan_command_processor.h
@@ -40,7 +40,8 @@ class VulkanCommandProcessor : public CommandProcessor {
                      uint32_t dword_count) override;
 
   bool IssueDraw(PrimitiveType primitive_type, uint32_t index_count,
-                 IndexBufferInfo* index_buffer_info) override;
+                 IndexBufferInfo* index_buffer_info,
+                 bool major_mode_explicit) override;
   bool IssueCopy() override;
 
   void InitializeTrace() override;
diff --git a/src/xenia/gpu/vulkan/vulkan_command_processor.cc b/src/xenia/gpu/vulkan/vulkan_command_processor.cc
index 46244fbb4..43f734a62 100644
--- a/src/xenia/gpu/vulkan/vulkan_command_processor.cc
+++ b/src/xenia/gpu/vulkan/vulkan_command_processor.cc
@@ -598,7 +598,8 @@ Shader* VulkanCommandProcessor::LoadShader(ShaderType shader_type,
 
 bool VulkanCommandProcessor::IssueDraw(PrimitiveType primitive_type,
                                        uint32_t index_count,
-                                       IndexBufferInfo* index_buffer_info) {
+                                       IndexBufferInfo* index_buffer_info,
+                                       bool major_mode_explicit) {
   auto& regs = *register_file_;
 
 #if FINE_GRAINED_DRAW_SCOPES
diff --git a/src/xenia/gpu/vulkan/vulkan_command_processor.h b/src/xenia/gpu/vulkan/vulkan_command_processor.h
index 94f7ae401..c4959fd0a 100644
--- a/src/xenia/gpu/vulkan/vulkan_command_processor.h
+++ b/src/xenia/gpu/vulkan/vulkan_command_processor.h
@@ -81,7 +81,8 @@ class VulkanCommandProcessor : public CommandProcessor {
                      uint32_t dword_count) override;
 
   bool IssueDraw(PrimitiveType primitive_type, uint32_t index_count,
-                 IndexBufferInfo* index_buffer_info) override;
+                 IndexBufferInfo* index_buffer_info,
+                 bool major_mode_explicit) override;
   bool PopulateConstants(VkCommandBuffer command_buffer,
                          VulkanShader* vertex_shader,
                          VulkanShader* pixel_shader);
diff --git a/src/xenia/gpu/xenos.h b/src/xenia/gpu/xenos.h
index 0cda866ba..aa8174ecc 100644
--- a/src/xenia/gpu/xenos.h
+++ b/src/xenia/gpu/xenos.h
@@ -37,12 +37,11 @@ enum class PrimitiveType : uint32_t {
   kQuadStrip = 0x0E,
   kPolygon = 0x0F,
 
-  // Starting with this primitive mode, registers like VGT_OUTPUT_PATH_CNTL have
-  // effect (deduced from R6xx/R7xx registers, and Halo 3 also doesn't reset
-  // VGT_OUTPUT_PATH_CNTL after the first draw with tessellation).
-  // TODO(Triang3l): Find out if VGT_DRAW_INITIATOR (0x21FC on Adreno 2xx, but
-  // not seen being used in games) specifies the major mode (or if it's set
-  // somewhere else).
+  // Starting with this primitive type, explicit major mode is assumed (in the
+  // R6xx/R7xx registers, k2DCopyRectListV0 is 22, and implicit major mode is
+  // only used for primitive types 0 through 21) - and tessellation patches use
+  // the range that starts from k2DCopyRectListV0.
+  // TODO(Triang3l): Verify if this is also true for the Xenos.
   kExplicitMajorModeForceStart = 0x10,
 
   k2DCopyRectListV0 = 0x10,
@@ -460,6 +459,25 @@ typedef enum {
   XE_GPU_INVALIDATE_MASK_ALL = 0x7FFF,
 } XE_GPU_INVALIDATE_MASK;
 
+// VGT_DRAW_INITIATOR::DI_SRC_SEL_*
+enum class SourceSelect : uint32_t {
+  kDMA,
+  kImmediate,
+  kAutoIndex,
+};
+
+// VGT_DRAW_INITIATOR::DI_MAJOR_MODE_*
+enum class MajorMode : uint32_t {
+  kImplicit,
+  kExplicit,
+};
+
+inline bool IsMajorModeExplicit(MajorMode major_mode,
+                                PrimitiveType primitive_type) {
+  return major_mode != MajorMode::kImplicit ||
+         primitive_type >= PrimitiveType::kExplicitMajorModeForceStart;
+}
+
 // instr_arbitrary_filter_t
 enum class ArbitraryFilter : uint32_t {
   k2x4Sym = 0,
diff --git a/src/xenia/kernel/xboxkrnl/xboxkrnl_io.cc b/src/xenia/kernel/xboxkrnl/xboxkrnl_io.cc
index 26b67ff94..fcc2bb588 100644
--- a/src/xenia/kernel/xboxkrnl/xboxkrnl_io.cc
+++ b/src/xenia/kernel/xboxkrnl/xboxkrnl_io.cc
@@ -170,23 +170,15 @@ dword_result_t NtReadFile(dword_t file_handle, dword_t event_handle,
 
   if (XSUCCEEDED(result)) {
     if (true || file->is_synchronous()) {
-      // some games NtReadFile() directly into texture memory
-      auto heap = kernel_memory()->LookupHeap(buffer.guest_address());
-      if (heap && heap->IsGuestPhysicalHeap()) {
-        kernel_memory()->TriggerPhysicalMemoryCallbacks(
-            xe::global_critical_region::AcquireDirect(), buffer.guest_address(),
-            buffer_length, true, true);
-      }
-
       // Synchronous.
-      size_t bytes_read = 0;
+      uint32_t bytes_read = 0;
       result = file->Read(
-          buffer, buffer_length,
+          buffer.guest_address(), buffer_length,
           byte_offset_ptr ? static_cast<uint64_t>(*byte_offset_ptr) : -1,
           &bytes_read, apc_context);
       if (io_status_block) {
         io_status_block->status = result;
-        io_status_block->information = static_cast<uint32_t>(bytes_read);
+        io_status_block->information = bytes_read;
       }
 
       // Queue the APC callback. It must be delivered via the APC mechanism even
@@ -218,7 +210,8 @@ dword_result_t NtReadFile(dword_t file_handle, dword_t event_handle,
       state, file,
       (XAsyncRequest::CompletionCallback)xeNtReadFileCompleted,
       call_state);*/
-      // result = file->Read(buffer, buffer_length, byte_offset, request);
+      // result = file->Read(buffer.guest_address(), buffer_length, byte_offset,
+      //                     request);
       if (io_status_block) {
         io_status_block->status = X_STATUS_PENDING;
         io_status_block->information = 0;
@@ -266,9 +259,9 @@ dword_result_t NtWriteFile(dword_t file_handle, dword_t event_handle,
     // TODO(benvanik): async path.
     if (true || file->is_synchronous()) {
       // Synchronous request.
-      size_t bytes_written = 0;
+      uint32_t bytes_written = 0;
       result = file->Write(
-          buffer, buffer_length,
+          buffer.guest_address(), buffer_length,
           byte_offset_ptr ? static_cast<uint64_t>(*byte_offset_ptr) : -1,
           &bytes_written, apc_context);
       if (io_status_block) {
@@ -517,10 +510,12 @@ dword_result_t NtQueryInformationFile(
         // XctdDecompression.
         /*
         uint32_t magic;
-        size_t bytes_read;
-        size_t cur_pos = file->position();
+        uint32_t bytes_read;
+        uint64_t cur_pos = file->position();
 
         file->set_position(0);
+        // FIXME(Triang3l): For now, XFile can be read only to guest buffers -
+        // this line won't work, implement reading to host buffers if needed.
         result = file->Read(&magic, sizeof(magic), 0, &bytes_read);
         if (XSUCCEEDED(result)) {
           if (bytes_read == sizeof(magic)) {
diff --git a/src/xenia/kernel/xboxkrnl/xboxkrnl_memory.cc b/src/xenia/kernel/xboxkrnl/xboxkrnl_memory.cc
index 7a384bcc2..62179f6ed 100644
--- a/src/xenia/kernel/xboxkrnl/xboxkrnl_memory.cc
+++ b/src/xenia/kernel/xboxkrnl/xboxkrnl_memory.cc
@@ -347,8 +347,8 @@ dword_result_t MmAllocatePhysicalMemoryEx(dword_t flags, dword_t region_size,
   uint32_t heap_max_addr =
       xe::sat_sub(max_addr_range.value(), heap_physical_address_offset);
   uint32_t heap_size = heap->heap_size();
-  heap_min_addr = heap_base + std::min(heap_min_addr, heap_size);
-  heap_max_addr = heap_base + std::min(heap_max_addr, heap_size);
+  heap_min_addr = heap_base + std::min(heap_min_addr, heap_size - 1);
+  heap_max_addr = heap_base + std::min(heap_max_addr, heap_size - 1);
   uint32_t base_address;
   if (!heap->AllocRange(heap_min_addr, heap_max_addr, adjusted_size,
                         adjusted_alignment, allocation_type, protect, top_down,
diff --git a/src/xenia/kernel/xboxkrnl/xboxkrnl_strings.cc b/src/xenia/kernel/xboxkrnl/xboxkrnl_strings.cc
index aa1bbf245..88b59c74e 100644
--- a/src/xenia/kernel/xboxkrnl/xboxkrnl_strings.cc
+++ b/src/xenia/kernel/xboxkrnl/xboxkrnl_strings.cc
@@ -2,7 +2,7 @@
  ******************************************************************************
  * Xenia : Xbox 360 Emulator Research Project                                 *
  ******************************************************************************
- * Copyright 2013 Ben Vanik. All rights reserved.                             *
+ * Copyright 2020 Ben Vanik. All rights reserved.                             *
  * Released under the BSD license - see LICENSE in the root for more details. *
  ******************************************************************************
  */
@@ -49,6 +49,7 @@ enum FormatFlags {
   FF_IsWide = 1 << 9,
   FF_IsSigned = 1 << 10,
   FF_ForceLeadingZero = 1 << 11,
+  FF_InvertWide = 1 << 12,
 };
 
 enum ArgumentSize {
@@ -174,7 +175,9 @@ int32_t format_core(PPCContext* ppc_context, FormatData& data, ArgList& args,
       case FS_Invalid:
       case FS_Unknown:
       case FS_End:
-      default: { assert_always(); }
+      default: {
+        assert_always();
+      }
 
       case FS_Start: {
         if (c == '%') {
@@ -220,7 +223,6 @@ int32_t format_core(PPCContext* ppc_context, FormatData& data, ArgList& args,
           flags |= FF_AddPrefix;
           continue;
         }
-
         state = FS_Width;
         // fall through, don't need to goto restart
       }
@@ -240,7 +242,6 @@ int32_t format_core(PPCContext* ppc_context, FormatData& data, ArgList& args,
           width += c - '0';
           continue;
         }
-
         state = FS_PrecisionStart;
         // fall through, don't need to goto restart
       }
@@ -252,7 +253,6 @@ int32_t format_core(PPCContext* ppc_context, FormatData& data, ArgList& args,
           precision = 0;
           continue;
         }
-
         state = FS_Size;
         goto restart;
       }
@@ -271,7 +271,6 @@ int32_t format_core(PPCContext* ppc_context, FormatData& data, ArgList& args,
           precision += c - '0';
           continue;
         }
-
         state = FS_Size;
         // fall through
       }
@@ -310,7 +309,6 @@ int32_t format_core(PPCContext* ppc_context, FormatData& data, ArgList& args,
             continue;
           }
         }
-
         // fall through
       }
 
@@ -319,16 +317,14 @@ int32_t format_core(PPCContext* ppc_context, FormatData& data, ArgList& args,
         // wide character
         switch (c) {
           case 'C': {
-            if (!(flags & (FF_IsShort | FF_IsLong | FF_IsWide))) {
-              flags |= FF_IsWide;
-            }
+            flags |= FF_InvertWide;
             // fall through
           }
 
           // character
           case 'c': {
             bool is_wide;
-            if (flags & FF_IsLong) {
+            if (flags & (FF_IsLong | FF_IsWide)) {
               // "An lc, lC, wc or wC type specifier is synonymous with C in
               // printf functions and with c in wprintf functions."
               is_wide = true;
@@ -337,7 +333,7 @@ int32_t format_core(PPCContext* ppc_context, FormatData& data, ArgList& args,
               // functions and with C in wprintf functions."
               is_wide = false;
             } else {
-              is_wide = ((flags & FF_IsWide) != 0) ^ wide;
+              is_wide = ((flags & FF_InvertWide) != 0) ^ wide;
             }
 
             auto value = args.get32();
@@ -524,9 +520,7 @@ int32_t format_core(PPCContext* ppc_context, FormatData& data, ArgList& args,
 
           // wide string
           case 'S': {
-            if (!(flags & (FF_IsShort | FF_IsLong | FF_IsWide))) {
-              flags |= FF_IsWide;
-            }
+            flags |= FF_InvertWide;
             // fall through
           }
 
@@ -543,7 +537,7 @@ int32_t format_core(PPCContext* ppc_context, FormatData& data, ArgList& args,
             } else {
               void* str = SHIM_MEM_ADDR(pointer);
               bool is_wide;
-              if (flags & FF_IsLong) {
+              if (flags & (FF_IsLong | FF_IsWide)) {
                 // "An ls, lS, ws or wS type specifier is synonymous with S in
                 // printf functions and with s in wprintf functions."
                 is_wide = true;
@@ -552,7 +546,7 @@ int32_t format_core(PPCContext* ppc_context, FormatData& data, ArgList& args,
                 // functions and with S in wprintf functions."
                 is_wide = false;
               } else {
-                is_wide = ((flags & (FF_IsWide)) != 0) ^ wide;
+                is_wide = ((flags & FF_InvertWide) != 0) ^ wide;
               }
               int32_t length;
 
@@ -581,7 +575,9 @@ int32_t format_core(PPCContext* ppc_context, FormatData& data, ArgList& args,
             break;
           }
 
-          default: { assert_always(); }
+          default: {
+            assert_always();
+          }
         }
       }
     }
@@ -728,7 +724,7 @@ class StringFormatData : public FormatData {
 
   void skip(int32_t count) {
     while (count-- > 0) {
-      if (!*input_) {
+      if (!get()) {
         break;
       }
     }
@@ -761,11 +757,11 @@ class WideStringFormatData : public FormatData {
     return xe::byte_swap(result);
   }
 
-  uint16_t peek(int32_t offset) { return input_[offset]; }
+  uint16_t peek(int32_t offset) { return xe::byte_swap(input_[offset]); }
 
   void skip(int32_t count) {
     while (count-- > 0) {
-      if (!*input_) {
+      if (!get()) {
         break;
       }
     }
@@ -795,11 +791,11 @@ class WideCountFormatData : public FormatData {
     return xe::byte_swap(result);
   }
 
-  uint16_t peek(int32_t offset) { return input_[offset]; }
+  uint16_t peek(int32_t offset) { return xe::byte_swap(input_[offset]); }
 
   void skip(int32_t count) {
     while (count-- > 0) {
-      if (!*input_) {
+      if (!get()) {
         break;
       }
     }
diff --git a/src/xenia/kernel/xfile.cc b/src/xenia/kernel/xfile.cc
index 8a38b207a..857b7444a 100644
--- a/src/xenia/kernel/xfile.cc
+++ b/src/xenia/kernel/xfile.cc
@@ -13,8 +13,10 @@
 #include "xenia/base/byte_stream.h"
 #include "xenia/base/logging.h"
 #include "xenia/base/math.h"
+#include "xenia/base/mutex.h"
 #include "xenia/kernel/kernel_state.h"
 #include "xenia/kernel/xevent.h"
+#include "xenia/memory.h"
 
 namespace xe {
 namespace kernel {
@@ -87,18 +89,73 @@ X_STATUS XFile::QueryDirectory(X_FILE_DIRECTORY_INFORMATION* out_info,
   return X_STATUS_SUCCESS;
 }
 
-X_STATUS XFile::Read(void* buffer, size_t buffer_length, size_t byte_offset,
-                     size_t* out_bytes_read, uint32_t apc_context) {
-  if (byte_offset == -1) {
+X_STATUS XFile::Read(uint32_t buffer_guest_address, uint32_t buffer_length,
+                     uint64_t byte_offset, uint32_t* out_bytes_read,
+                     uint32_t apc_context) {
+  if (byte_offset == uint64_t(-1)) {
     // Read from current position.
     byte_offset = position_;
   }
 
   size_t bytes_read = 0;
-  X_STATUS result =
-      file_->ReadSync(buffer, buffer_length, byte_offset, &bytes_read);
-  if (XSUCCEEDED(result)) {
-    position_ += bytes_read;
+  X_STATUS result = X_STATUS_SUCCESS;
+  // Zero length means success for a valid file object according to Windows
+  // tests.
+  if (buffer_length) {
+    if (UINT32_MAX - buffer_guest_address < buffer_length) {
+      result = X_STATUS_ACCESS_VIOLATION;
+    } else {
+      // Games often read directly to texture/vertex buffer memory - in this
+      // case, invalidation notifications must be sent. However, having any
+      // memory callbacks in the range will result in STATUS_ACCESS_VIOLATION at
+      // least on Windows, without anything being read or any callbacks being
+      // triggered. So for physical memory, host protection must be bypassed,
+      // and invalidation callbacks must be triggered manually (it's also wrong
+      // to trigger invalidation callbacks before reading in this case, because
+      // during the read, the guest may still access the data around the buffer
+      // that is located in the same host pages as the buffer's start and end,
+      // on the GPU - and that must not trigger a race condition).
+      uint32_t buffer_guest_high_address =
+          buffer_guest_address + buffer_length - 1;
+      xe::BaseHeap* buffer_start_heap =
+          memory()->LookupHeap(buffer_guest_address);
+      const xe::BaseHeap* buffer_end_heap =
+          memory()->LookupHeap(buffer_guest_high_address);
+      if (!buffer_start_heap || !buffer_end_heap ||
+          buffer_start_heap->IsGuestPhysicalHeap() !=
+              buffer_end_heap->IsGuestPhysicalHeap() ||
+          (buffer_start_heap->IsGuestPhysicalHeap() &&
+           buffer_start_heap != buffer_end_heap)) {
+        result = X_STATUS_ACCESS_VIOLATION;
+      } else {
+        xe::PhysicalHeap* buffer_physical_heap =
+            buffer_start_heap->IsGuestPhysicalHeap()
+                ? static_cast<xe::PhysicalHeap*>(buffer_start_heap)
+                : nullptr;
+        if (buffer_physical_heap &&
+            buffer_physical_heap->QueryRangeAccess(buffer_guest_address,
+                                                   buffer_guest_high_address) !=
+                memory::PageAccess::kReadWrite) {
+          result = X_STATUS_ACCESS_VIOLATION;
+        } else {
+          result = file_->ReadSync(
+              buffer_physical_heap
+                  ? memory()->TranslatePhysical(
+                        buffer_physical_heap->GetPhysicalAddress(
+                            buffer_guest_address))
+                  : memory()->TranslateVirtual(buffer_guest_address),
+              buffer_length, size_t(byte_offset), &bytes_read);
+          if (XSUCCEEDED(result)) {
+            if (buffer_physical_heap) {
+              buffer_physical_heap->TriggerCallbacks(
+                  xe::global_critical_region::AcquireDirect(),
+                  buffer_guest_address, buffer_length, true, true);
+            }
+            position_ += bytes_read;
+          }
+        }
+      }
+    }
   }
 
   XIOCompletion::IONotification notify;
@@ -109,24 +166,25 @@ X_STATUS XFile::Read(void* buffer, size_t buffer_length, size_t byte_offset,
   NotifyIOCompletionPorts(notify);
 
   if (out_bytes_read) {
-    *out_bytes_read = bytes_read;
+    *out_bytes_read = uint32_t(bytes_read);
   }
 
   async_event_->Set();
   return result;
 }
 
-X_STATUS XFile::Write(const void* buffer, size_t buffer_length,
-                      size_t byte_offset, size_t* out_bytes_written,
+X_STATUS XFile::Write(uint32_t buffer_guest_address, uint32_t buffer_length,
+                      uint64_t byte_offset, uint32_t* out_bytes_written,
                       uint32_t apc_context) {
-  if (byte_offset == -1) {
+  if (byte_offset == uint64_t(-1)) {
     // Write from current position.
     byte_offset = position_;
   }
 
   size_t bytes_written = 0;
   X_STATUS result =
-      file_->WriteSync(buffer, buffer_length, byte_offset, &bytes_written);
+      file_->WriteSync(memory()->TranslateVirtual(buffer_guest_address),
+                       buffer_length, size_t(byte_offset), &bytes_written);
   if (XSUCCEEDED(result)) {
     position_ += bytes_written;
   }
@@ -139,7 +197,7 @@ X_STATUS XFile::Write(const void* buffer, size_t buffer_length,
   NotifyIOCompletionPorts(notify);
 
   if (out_bytes_written) {
-    *out_bytes_written = bytes_written;
+    *out_bytes_written = uint32_t(bytes_written);
   }
 
   async_event_->Set();
diff --git a/src/xenia/kernel/xfile.h b/src/xenia/kernel/xfile.h
index fda7abb47..2e4feb87a 100644
--- a/src/xenia/kernel/xfile.h
+++ b/src/xenia/kernel/xfile.h
@@ -92,17 +92,22 @@ class XFile : public XObject {
   const std::string& path() const { return file_->entry()->path(); }
   const std::string& name() const { return file_->entry()->name(); }
 
-  size_t position() const { return position_; }
-  void set_position(size_t value) { position_ = value; }
+  uint64_t position() const { return position_; }
+  void set_position(uint64_t value) { position_ = value; }
 
   X_STATUS QueryDirectory(X_FILE_DIRECTORY_INFORMATION* out_info, size_t length,
                           const char* file_name, bool restart);
 
-  X_STATUS Read(void* buffer, size_t buffer_length, size_t byte_offset,
-                size_t* out_bytes_read, uint32_t apc_context);
+  // Don't do within the global critical region because invalidation callbacks
+  // may be triggered (as per the usual rule of not doing I/O within the global
+  // critical region).
+  X_STATUS Read(uint32_t buffer_guess_address, uint32_t buffer_length,
+                uint64_t byte_offset, uint32_t* out_bytes_read,
+                uint32_t apc_context);
 
-  X_STATUS Write(const void* buffer, size_t buffer_length, size_t byte_offset,
-                 size_t* out_bytes_written, uint32_t apc_context);
+  X_STATUS Write(uint32_t buffer_guess_address, uint32_t buffer_length,
+                 uint64_t byte_offset, uint32_t* out_bytes_written,
+                 uint32_t apc_context);
 
   X_STATUS SetLength(size_t length);
 
@@ -133,7 +138,7 @@ class XFile : public XObject {
 
   // TODO(benvanik): create flags, open state, etc.
 
-  size_t position_ = 0;
+  uint64_t position_ = 0;
 
   xe::filesystem::WildcardEngine find_engine_;
   size_t find_index_ = 0;
diff --git a/src/xenia/memory.cc b/src/xenia/memory.cc
index 026cd21ba..3083554cc 100644
--- a/src/xenia/memory.cc
+++ b/src/xenia/memory.cc
@@ -365,7 +365,7 @@ uint32_t Memory::HostToGuestVirtual(const void* host_address) const {
       size_t(heaps_.vE0000000.heap_base()) + vE0000000_host_offset;
   if (virtual_address >= vE0000000_host_base &&
       virtual_address <=
-          (vE0000000_host_base + heaps_.vE0000000.heap_size() - 1)) {
+          (vE0000000_host_base + (heaps_.vE0000000.heap_size() - 1))) {
     virtual_address -= vE0000000_host_offset;
   }
   return uint32_t(virtual_address);
@@ -646,7 +646,7 @@ void BaseHeap::Initialize(Memory* memory, uint8_t* membase, uint32_t heap_base,
   memory_ = memory;
   membase_ = membase;
   heap_base_ = heap_base;
-  heap_size_ = heap_size - 1;
+  heap_size_ = heap_size;
   page_size_ = page_size;
   host_address_offset_ = host_address_offset;
   page_table_.resize(heap_size / page_size);
@@ -668,7 +668,7 @@ void BaseHeap::Dispose() {
 void BaseHeap::DumpMap() {
   auto global_lock = global_critical_region_.Acquire();
   XELOGE("------------------------------------------------------------------");
-  XELOGE("Heap: %.8X-%.8X", heap_base_, heap_base_ + heap_size_);
+  XELOGE("Heap: %.8X-%.8X", heap_base_, heap_base_ + (heap_size_ - 1));
   XELOGE("------------------------------------------------------------------");
   XELOGE("            Heap Base: %.8X", heap_base_);
   XELOGE("            Heap Size: %d (%.8X)", heap_size_, heap_size_);
@@ -710,7 +710,8 @@ void BaseHeap::DumpMap() {
   }
   if (is_empty_span) {
     XELOGE("  %.8X-%.8X - %d unreserved pages)",
-           heap_base_ + empty_span_start * page_size_, heap_base_ + heap_size_,
+           heap_base_ + empty_span_start * page_size_,
+           heap_base_ + (heap_size_ - 1),
            page_table_.size() - empty_span_start);
   }
 }
@@ -745,7 +746,7 @@ uint32_t BaseHeap::GetUnreservedPageCount() {
 }
 
 bool BaseHeap::Save(ByteStream* stream) {
-  XELOGD("Heap %.8X-%.8X", heap_base_, heap_base_ + heap_size_);
+  XELOGD("Heap %.8X-%.8X", heap_base_, heap_base_ + (heap_size_ - 1));
 
   for (size_t i = 0; i < page_table_.size(); i++) {
     auto& page = page_table_[i];
@@ -773,7 +774,7 @@ bool BaseHeap::Save(ByteStream* stream) {
 }
 
 bool BaseHeap::Restore(ByteStream* stream) {
-  XELOGD("Heap %.8X-%.8X", heap_base_, heap_base_ + heap_size_);
+  XELOGD("Heap %.8X-%.8X", heap_base_, heap_base_ + (heap_size_ - 1));
 
   for (size_t i = 0; i < page_table_.size(); i++) {
     auto& page = page_table_[i];
@@ -830,7 +831,7 @@ bool BaseHeap::Alloc(uint32_t size, uint32_t alignment,
   size = xe::round_up(size, page_size_);
   alignment = xe::round_up(alignment, page_size_);
   uint32_t low_address = heap_base_;
-  uint32_t high_address = heap_base_ + heap_size_;
+  uint32_t high_address = heap_base_ + (heap_size_ - 1);
   return AllocRange(low_address, high_address, size, alignment, allocation_type,
                     protect, top_down, out_address);
 }
@@ -922,8 +923,8 @@ bool BaseHeap::AllocRange(uint32_t low_address, uint32_t high_address,
   alignment = xe::round_up(alignment, page_size_);
   uint32_t page_count = get_page_count(size, page_size_);
   low_address = std::max(heap_base_, xe::align(low_address, alignment));
-  high_address =
-      std::min(heap_base_ + heap_size_, xe::align(high_address, alignment));
+  high_address = std::min(heap_base_ + (heap_size_ - 1),
+                          xe::align(high_address, alignment));
   uint32_t low_page_number = (low_address - heap_base_) / page_size_;
   uint32_t high_page_number = (high_address - heap_base_) / page_size_;
   low_page_number = std::min(uint32_t(page_table_.size()) - 1, low_page_number);
@@ -1302,6 +1303,24 @@ bool BaseHeap::QueryProtect(uint32_t address, uint32_t* out_protect) {
   return true;
 }
 
+xe::memory::PageAccess BaseHeap::QueryRangeAccess(uint32_t low_address,
+                                                  uint32_t high_address) {
+  if (low_address > high_address || low_address < heap_base_ ||
+      (high_address - heap_base_) >= heap_size_) {
+    return xe::memory::PageAccess::kNoAccess;
+  }
+  uint32_t low_page_number = (low_address - heap_base_) / page_size_;
+  uint32_t high_page_number = (high_address - heap_base_) / page_size_;
+  uint32_t protect = kMemoryProtectRead | kMemoryProtectWrite;
+  {
+    auto global_lock = global_critical_region_.Acquire();
+    for (uint32_t i = low_page_number; protect && i <= high_page_number; ++i) {
+      protect &= page_table_[i].current_protect;
+    }
+  }
+  return ToPageAccess(protect);
+}
+
 VirtualHeap::VirtualHeap() = default;
 
 VirtualHeap::~VirtualHeap() = default;
@@ -1333,7 +1352,7 @@ void PhysicalHeap::Initialize(Memory* memory, uint8_t* membase,
   system_page_size_ = uint32_t(xe::memory::page_size());
 
   system_page_count_ =
-      (heap_size_ /* already - 1 */ + host_address_offset + system_page_size_) /
+      (size_t(heap_size_) + host_address_offset + (system_page_size_ - 1)) /
       system_page_size_;
   system_page_flags_.resize((system_page_count_ + 63) / 64);
 }
@@ -1355,7 +1374,7 @@ bool PhysicalHeap::Alloc(uint32_t size, uint32_t alignment,
 
   // Allocate from parent heap (gets our physical address in 0-512mb).
   uint32_t parent_heap_start = GetPhysicalAddress(heap_base_);
-  uint32_t parent_heap_end = GetPhysicalAddress(heap_base_ + heap_size_);
+  uint32_t parent_heap_end = GetPhysicalAddress(heap_base_ + (heap_size_ - 1));
   uint32_t parent_address;
   if (!parent_heap_->AllocRange(parent_heap_start, parent_heap_end, size,
                                 alignment, allocation_type, protect, top_down,
@@ -1375,11 +1394,6 @@ bool PhysicalHeap::Alloc(uint32_t size, uint32_t alignment,
     // TODO(benvanik): don't leak parent memory.
     return false;
   }
-
-  if (protect & kMemoryProtectWrite) {
-    TriggerCallbacks(std::move(global_lock), address, size, true, true, false);
-  }
-
   *out_address = address;
   return true;
 }
@@ -1417,10 +1431,6 @@ bool PhysicalHeap::AllocFixed(uint32_t base_address, uint32_t size,
     return false;
   }
 
-  if (protect & kMemoryProtectWrite) {
-    TriggerCallbacks(std::move(global_lock), address, size, true, true, false);
-  }
-
   return true;
 }
 
@@ -1438,7 +1448,7 @@ bool PhysicalHeap::AllocRange(uint32_t low_address, uint32_t high_address,
 
   // Allocate from parent heap (gets our physical address in 0-512mb).
   low_address = std::max(heap_base_, low_address);
-  high_address = std::min(heap_base_ + heap_size_, high_address);
+  high_address = std::min(heap_base_ + (heap_size_ - 1), high_address);
   uint32_t parent_low_address = GetPhysicalAddress(low_address);
   uint32_t parent_high_address = GetPhysicalAddress(high_address);
   uint32_t parent_address;
@@ -1461,32 +1471,49 @@ bool PhysicalHeap::AllocRange(uint32_t low_address, uint32_t high_address,
     // TODO(benvanik): don't leak parent memory.
     return false;
   }
-
-  if (protect & kMemoryProtectWrite) {
-    TriggerCallbacks(std::move(global_lock), address, size, true, true, false);
-  }
-
   *out_address = address;
   return true;
 }
 
 bool PhysicalHeap::Decommit(uint32_t address, uint32_t size) {
   auto global_lock = global_critical_region_.Acquire();
+
   uint32_t parent_address = GetPhysicalAddress(address);
   if (!parent_heap_->Decommit(parent_address, size)) {
     XELOGE("PhysicalHeap::Decommit failed due to parent heap failure");
     return false;
   }
+
+  // Not caring about the contents anymore.
+  TriggerCallbacks(std::move(global_lock), address, size, true, true);
+
   return BaseHeap::Decommit(address, size);
 }
 
 bool PhysicalHeap::Release(uint32_t base_address, uint32_t* out_region_size) {
   auto global_lock = global_critical_region_.Acquire();
+
   uint32_t parent_base_address = GetPhysicalAddress(base_address);
   if (!parent_heap_->Release(parent_base_address, out_region_size)) {
     XELOGE("PhysicalHeap::Release failed due to parent heap failure");
     return false;
   }
+
+  // Must invalidate here because the range being released may be reused in
+  // another mapping of physical memory - but callback flags are set in each
+  // heap separately (https://github.com/xenia-project/xenia/issues/1559 -
+  // dynamic vertices in Viva Pinata start screen and menu allocated in
+  // 0xA0000000 at addresses that overlap intro video textures in 0xE0000000,
+  // with the state of the allocator as of February 24th, 2020). If memory is
+  // invalidated in Alloc instead, Alloc won't be aware of callbacks enabled in
+  // other heaps, thus callback handlers will keep considering this range valid
+  // forever.
+  uint32_t region_size;
+  if (QuerySize(base_address, &region_size)) {
+    TriggerCallbacks(std::move(global_lock), base_address, region_size, true,
+                     true);
+  }
+
   return BaseHeap::Release(base_address, out_region_size);
 }
 
@@ -1527,10 +1554,10 @@ void PhysicalHeap::EnableAccessCallbacks(uint32_t physical_address,
     physical_address = physical_address_offset;
   }
   uint32_t heap_relative_address = physical_address - physical_address_offset;
-  if (heap_relative_address >= heap_size_ + 1) {
+  if (heap_relative_address >= heap_size_) {
     return;
   }
-  length = std::min(length, heap_size_ + 1 - heap_relative_address);
+  length = std::min(length, heap_size_ - heap_relative_address);
   if (length == 0) {
     return;
   }
@@ -1637,10 +1664,10 @@ bool PhysicalHeap::TriggerCallbacks(
     virtual_address = heap_base_;
   }
   uint32_t heap_relative_address = virtual_address - heap_base_;
-  if (heap_relative_address >= heap_size_ + 1) {
+  if (heap_relative_address >= heap_size_) {
     return false;
   }
-  length = std::min(length, heap_size_ + 1 - heap_relative_address);
+  length = std::min(length, heap_size_ - heap_relative_address);
   if (length == 0) {
     return false;
   }
@@ -1689,7 +1716,7 @@ bool PhysicalHeap::TriggerCallbacks(
       xe::sat_sub(system_page_last * system_page_size_ + system_page_size_,
                   host_address_offset()) +
           physical_address_offset - physical_address_start,
-      heap_size_ + 1 - (physical_address_start - physical_address_offset));
+      heap_size_ - (physical_address_start - physical_address_offset));
   uint32_t unwatch_first = 0;
   uint32_t unwatch_last = UINT32_MAX;
   for (auto invalidation_callback :
@@ -1724,8 +1751,8 @@ bool PhysicalHeap::TriggerCallbacks(
     unwatch_first = xe::sat_sub(unwatch_first, physical_address_offset);
     unwatch_last = xe::sat_sub(unwatch_last, physical_address_offset);
     // Clamp to the heap upper bound.
-    unwatch_first = std::min(unwatch_first, heap_size_);
-    unwatch_last = std::min(unwatch_last, heap_size_);
+    unwatch_first = std::min(unwatch_first, heap_size_ - 1);
+    unwatch_last = std::min(unwatch_last, heap_size_ - 1);
     // Convert to system pages and update the range.
     unwatch_first += host_address_offset();
     unwatch_last += host_address_offset();
@@ -1794,7 +1821,7 @@ bool PhysicalHeap::TriggerCallbacks(
 uint32_t PhysicalHeap::GetPhysicalAddress(uint32_t address) const {
   assert_true(address >= heap_base_);
   address -= heap_base_;
-  assert_true(address <= heap_size_);
+  assert_true(address < heap_size_);
   if (heap_base_ >= 0xE0000000) {
     address += 0x1000;
   }
diff --git a/src/xenia/memory.h b/src/xenia/memory.h
index 3c7d0c93b..266b8acd8 100644
--- a/src/xenia/memory.h
+++ b/src/xenia/memory.h
@@ -172,6 +172,11 @@ class BaseHeap {
   // address.
   bool QueryProtect(uint32_t address, uint32_t* out_protect);
 
+  // Queries the currently strictest readability and writability for the entire
+  // range.
+  xe::memory::PageAccess QueryRangeAccess(uint32_t low_address,
+                                          uint32_t high_address);
+
   // Whether the heap is a guest virtual memory mapping of the physical memory.
   virtual bool IsGuestPhysicalHeap() const { return false; }
 
@@ -387,7 +392,9 @@ class Memory {
   //
   // May be triggered for a single page (in case of a write access violation or
   // when need to synchronize data given by data providers) or for multiple
-  // pages (like when memory is allocated).
+  // pages (like when memory is released, or explicitly to trigger callbacks
+  // when host-side code can't rely on regular access violations, like when
+  // accessing a file).
   //
   // Since granularity of callbacks is one single page, an invalidation
   // notification handler must invalidate the all the data stored in the touched
diff --git a/third_party/yaml-cpp b/third_party/yaml-cpp
deleted file mode 160000
index e92321aee..000000000
--- a/third_party/yaml-cpp
+++ /dev/null
@@ -1 +0,0 @@
-Subproject commit e92321aee52fd27566601f9cca53ba90b41e15c1
diff --git a/third_party/yaml-cpp.lua b/third_party/yaml-cpp.lua
deleted file mode 100644
index 138e99a46..000000000
--- a/third_party/yaml-cpp.lua
+++ /dev/null
@@ -1,17 +0,0 @@
-group("third_party")
-project("yaml-cpp")
-  uuid("47bfe853-a3f8-4902-921d-d564608ff355")
-  kind("StaticLib")
-  language("C++")
-
-  defines({
-    "_LIB",
-  })
-  includedirs({
-    "yaml-cpp/include/",
-  })
-  recursive_platform_files("yaml-cpp/include/yaml-cpp")
-  recursive_platform_files("yaml-cpp/src")
-
-  filter("platforms:Windows")
-    warnings("Off")  -- Too many warnings.
diff --git a/tools/diff.py b/tools/diff.py
index 29bc1ff33..1ec660e68 100644
--- a/tools/diff.py
+++ b/tools/diff.py
@@ -6,7 +6,7 @@ import difflib
 import sys
 
 diff = difflib.unified_diff(
-    open(sys.argv[1]).readlines(),
-    open(sys.argv[2]).readlines())
-with open(sys.argv[3], 'w') as f:
+    open(sys.argv[1], encoding='utf-8').readlines(),
+    open(sys.argv[2], encoding='utf-8').readlines())
+with open(sys.argv[3], 'w', encoding='utf-8') as f:
     f.write(''.join(diff))
diff --git a/xenia-build b/xenia-build
index cff6dfe5c..aa6859317 100755
--- a/xenia-build
+++ b/xenia-build
@@ -96,7 +96,7 @@ def import_vs_environment():
     install_path = None
     env_tool_args = None
 
-    vswhere = subprocess.check_output('third_party/vswhere/vswhere.exe -version "[15,)" -latest -prerelease -format json -utf8', shell=False, universal_newlines=True)
+    vswhere = subprocess.check_output('third_party/vswhere/vswhere.exe -version "[15,)" -latest -prerelease -format json -utf8', shell=False, universal_newlines=True, encoding="utf-8")
     if vswhere:
         vswhere = json.loads(vswhere)
     if vswhere and len(vswhere) > 0:
@@ -185,7 +185,7 @@ def get_bin(binary):
     return None
 
 
-def shell_call(command, throw_on_error=True, stdout_path=None):
+def shell_call(command, throw_on_error=True, stdout_path=None, shell=False):
     """Executes a shell command.
 
     Args:
@@ -203,10 +203,10 @@ def shell_call(command, throw_on_error=True, stdout_path=None):
     try:
         if throw_on_error:
             result = 1
-            subprocess.check_call(command, shell=False, stdout=stdout_file)
+            subprocess.check_call(command, shell=shell, stdout=stdout_file)
             result = 0
         else:
-            result = subprocess.call(command, shell=False, stdout=stdout_file)
+            result = subprocess.call(command, shell=shell, stdout=stdout_file)
     finally:
         if stdout_file:
             stdout_file.close()
@@ -319,7 +319,7 @@ def get_clang_format_binary():
     attempts = [
         'C:\\Program Files\\LLVM\\bin\\clang-format.exe',
         'C:\\Program Files (x86)\\LLVM\\bin\\clang-format.exe',
-        'clang-format-3.8',
+        'clang-format-9',
         'clang-format',
         ]
     for binary in attempts:
@@ -327,7 +327,7 @@ def get_clang_format_binary():
             return binary
     print('ERROR: clang-format is not on PATH')
     print('LLVM is available from https://llvm.org/releases/download.html')
-    print('At least version 3.8 is required.')
+    print('At least version 9 is required.')
     print('See docs/style_guide.md for instructions on how to get it.')
     sys.exit(1)
 
@@ -1215,6 +1215,7 @@ class LintCommand(Command):
 
         if args['all']:
             all_files = find_all_source_files()
+            all_files.sort()
             print('- linting %d files' % (len(all_files)))
             any_errors = False
             for file_path in all_files:
@@ -1247,7 +1248,7 @@ class LintCommand(Command):
                     shell_call([
                         'type' if sys.platform == 'win32' else 'cat',
                         difftemp,
-                        ])
+                        ], shell=True if sys.platform == 'win32' else False)
                     if os.path.exists(difftemp): os.remove(difftemp)
                     print('')
             print('')
@@ -1265,6 +1266,7 @@ class LintCommand(Command):
                 'third_party/clang-format/git-clang-format',
                 '--binary=%s' % (clang_format_binary),
                 '--commit=%s' % ('origin/master' if args['origin'] else 'HEAD'),
+                '--style=file',
                 '--diff',
                 ], throw_on_error=False, stdout_path=difftemp)
             with open(difftemp) as f:
@@ -1281,6 +1283,7 @@ class LintCommand(Command):
                     'third_party/clang-format/git-clang-format',
                     '--binary=%s' % (clang_format_binary),
                     '--commit=%s' % ('origin/master' if args['origin'] else 'HEAD'),
+                    '--style=file',
                     '--diff',
                     ])
                 print('ERROR: 1+ diffs. Stage changes and run \'xb format\' to fix.')
@@ -1311,6 +1314,7 @@ class FormatCommand(Command):
 
         if args['all']:
             all_files = find_all_source_files()
+            all_files.sort()
             print('- clang-format [%d files]' % (len(all_files)))
             any_errors = False
             for file_path in all_files: