From 6cb8f0aab4a647ebab9ab98c051d40cb3c2fbc90 Mon Sep 17 00:00:00 2001
From: Triang3l <triang3l@yandex.ru>
Date: Sat, 3 Oct 2020 14:12:30 +0300
Subject: [PATCH] [D3D12] Nicer DeferredCommandList uintmax_t alignment

---
 src/xenia/gpu/d3d12/deferred_command_list.cc | 41 ++++++++++----------
 src/xenia/gpu/d3d12/deferred_command_list.h  | 17 +++++---
 2 files changed, 33 insertions(+), 25 deletions(-)

diff --git a/src/xenia/gpu/d3d12/deferred_command_list.cc b/src/xenia/gpu/d3d12/deferred_command_list.cc
index f6eede18b..8bbd08224 100644
--- a/src/xenia/gpu/d3d12/deferred_command_list.cc
+++ b/src/xenia/gpu/d3d12/deferred_command_list.cc
@@ -20,22 +20,22 @@ namespace d3d12 {
 DeferredCommandList::DeferredCommandList(
     D3D12CommandProcessor& command_processor, size_t initial_size)
     : command_processor_(command_processor) {
-  command_stream_.reserve(initial_size);
+  command_stream_.reserve(initial_size / sizeof(uintmax_t));
 }
 
 void DeferredCommandList::Reset() { command_stream_.clear(); }
 
 void DeferredCommandList::Execute(ID3D12GraphicsCommandList* command_list,
                                   ID3D12GraphicsCommandList1* command_list_1) {
-  const uint8_t* stream = command_stream_.data();
+  const uintmax_t* stream = command_stream_.data();
   size_t stream_remaining = command_stream_.size();
   ID3D12PipelineState* current_pipeline_state = nullptr;
   while (stream_remaining != 0) {
-    const uint32_t* header = reinterpret_cast<const uint32_t*>(stream);
-    const size_t header_size = xe::align(2 * sizeof(uint32_t), kAlignment);
-    stream += header_size;
-    stream_remaining -= header_size;
-    switch (Command(header[0])) {
+    const CommandHeader& header =
+        *reinterpret_cast<const CommandHeader*>(stream);
+    stream += kCommandHeaderSizeElements;
+    stream_remaining -= kCommandHeaderSizeElements;
+    switch (header.command) {
       case Command::kD3DClearUnorderedAccessViewUint: {
         auto& args =
             *reinterpret_cast<const ClearUnorderedAccessViewHeader*>(stream);
@@ -121,7 +121,7 @@ void DeferredCommandList::Execute(ID3D12GraphicsCommandList* command_list,
         command_list->ResourceBarrier(
             *reinterpret_cast<const UINT*>(stream),
             reinterpret_cast<const D3D12_RESOURCE_BARRIER*>(
-                stream +
+                reinterpret_cast<const uint8_t*>(stream) +
                 xe::align(sizeof(UINT), alignof(D3D12_RESOURCE_BARRIER))));
       } break;
       case Command::kRSSetScissorRect: {
@@ -221,25 +221,26 @@ void DeferredCommandList::Execute(ID3D12GraphicsCommandList* command_list,
         }
       } break;
       default:
-        assert_unhandled_case(Command(header[0]));
+        assert_unhandled_case(header.command);
         break;
     }
-    stream += header[1];
-    stream_remaining -= header[1];
+    stream += header.arguments_size_elements;
+    stream_remaining -= header.arguments_size_elements;
   }
 }
 
 void* DeferredCommandList::WriteCommand(Command command,
-                                        size_t arguments_size) {
-  arguments_size = xe::align(arguments_size, kAlignment);
-  const size_t header_size = xe::align(2 * sizeof(uint32_t), kAlignment);
+                                        size_t arguments_size_bytes) {
+  size_t arguments_size_elements =
+      (arguments_size_bytes + sizeof(uintmax_t) - 1) / sizeof(uintmax_t);
   size_t offset = command_stream_.size();
-  command_stream_.resize(offset + header_size + arguments_size);
-  uint32_t* header =
-      reinterpret_cast<uint32_t*>(command_stream_.data() + offset);
-  header[0] = uint32_t(command);
-  header[1] = uint32_t(arguments_size);
-  return command_stream_.data() + (offset + header_size);
+  command_stream_.resize(offset + kCommandHeaderSizeElements +
+                         arguments_size_elements);
+  CommandHeader& header =
+      *reinterpret_cast<CommandHeader*>(command_stream_.data() + offset);
+  header.command = command;
+  header.arguments_size_elements = uint32_t(arguments_size_elements);
+  return command_stream_.data() + (offset + kCommandHeaderSizeElements);
 }
 
 }  // namespace d3d12
diff --git a/src/xenia/gpu/d3d12/deferred_command_list.h b/src/xenia/gpu/d3d12/deferred_command_list.h
index 78f1b605b..64118d651 100644
--- a/src/xenia/gpu/d3d12/deferred_command_list.h
+++ b/src/xenia/gpu/d3d12/deferred_command_list.h
@@ -11,6 +11,7 @@
 #define XENIA_GPU_D3D12_DEFERRED_COMMAND_LIST_H_
 
 #include <algorithm>
+#include <cstdint>
 #include <cstring>
 #include <vector>
 
@@ -26,7 +27,7 @@ class D3D12CommandProcessor;
 class DeferredCommandList {
  public:
   DeferredCommandList(D3D12CommandProcessor& command_processor,
-                      size_t initial_size = 256 * 1024);
+                      size_t initial_size_bytes = 1024 * 1024);
 
   void Reset();
   void Execute(ID3D12GraphicsCommandList* command_list,
@@ -332,8 +333,6 @@ class DeferredCommandList {
   }
 
  private:
-  static constexpr size_t kAlignment = std::max(sizeof(void*), sizeof(UINT64));
-
   enum class Command : uint32_t {
     kD3DClearUnorderedAccessViewUint,
     kD3DCopyBufferRegion,
@@ -365,6 +364,13 @@ class DeferredCommandList {
     kD3DSetSamplePositions,
   };
 
+  struct CommandHeader {
+    Command command;
+    uint32_t arguments_size_elements;
+  };
+  static constexpr size_t kCommandHeaderSizeElements =
+      (sizeof(CommandHeader) + sizeof(uintmax_t) - 1) / sizeof(uintmax_t);
+
   struct ClearUnorderedAccessViewHeader {
     D3D12_GPU_DESCRIPTOR_HANDLE view_gpu_handle_in_current_heap;
     D3D12_CPU_DESCRIPTOR_HANDLE view_cpu_handle;
@@ -460,11 +466,12 @@ class DeferredCommandList {
     D3D12_SAMPLE_POSITION sample_positions[16];
   };
 
-  void* WriteCommand(Command command, size_t arguments_size);
+  void* WriteCommand(Command command, size_t arguments_size_bytes);
 
   D3D12CommandProcessor& command_processor_;
 
-  std::vector<uint8_t> command_stream_;
+  // uintmax_t to ensure uint64_t and pointer alignment of all structures.
+  std::vector<uintmax_t> command_stream_;
 };
 
 }  // namespace d3d12