[D3D12] DXBC RDEF header

src/xenia/gpu/dxbc_shader_translator.cc

@@ -14,6 +14,7 @@
 #include "third_party/dxbc/DXBCChecksum.h"
 #include "third_party/dxbc/d3d12TokenizedProgramFormat.hpp"
 
+#include "xenia/base/assert.h"
 #include "xenia/base/math.h"
 
 namespace xe {
@@ -34,15 +35,31 @@ std::vector<uint8_t> DxbcShaderTranslator::CompleteTranslation() {
 
   // Write the shader object header.
   shader_object_.push_back('CBXD');
-  // Reserve space for the checksum.
+  // Checksum (set later).
   for (uint32_t i = 0; i < 4; ++i) {
     shader_object_.push_back(0);
   }
   shader_object_.push_back(1);
-  // Reserve space for the size.
+  // Size (set later).
   shader_object_.push_back(0);
   // 5 chunks - RDEF, ISGN, OSGN, SHEX, STAT.
   shader_object_.push_back(5);
+  // Chunk offsets (set later).
+  for (uint32_t i = 0; i < shader_object_[7]; ++i) {
+    shader_object_.push_back(0);
+  }
+
+  uint32_t chunk_position_dwords;
+
+  // Write Resource DEFinitions.
+  chunk_position_dwords = uint32_t(shader_object_.size());
+  shader_object_[8] = chunk_position_dwords * sizeof(uint32_t);
+  shader_object_.push_back('FEDR');
+  shader_object_.push_back(0);
+  WriteResourceDefinitions();
+  shader_object_[chunk_position_dwords + 1] =
+      (uint32_t(shader_object_.size()) - chunk_position_dwords - 2) *
+      sizeof(uint32_t);
 
   // Fill the remaining fields of the header and copy bytes out.
   uint32_t shader_object_size =
@@ -64,13 +81,61 @@ uint32_t DxbcShaderTranslator::AppendString(std::vector<uint32_t>& dest,
                                             const char* source) {
   size_t size = std::strlen(source) + 1;
   size_t size_aligned = xe::align(size_aligned, sizeof(uint32_t));
-  size_t dest_pos = dest.size();
+  size_t dest_position = dest.size();
-  dest.resize(dest_pos + size_aligned / sizeof(uint32_t));
+  dest.resize(dest_position + size_aligned / sizeof(uint32_t));
-  std::memcpy(&dest[dest_pos], source, size);
+  std::memcpy(&dest[dest_position], source, size);
-  std::memset(reinterpret_cast<uint8_t*>(&dest[dest_pos]) + size, 0xAB,
+  std::memset(reinterpret_cast<uint8_t*>(&dest[dest_position]) + size, 0xAB,
               size_aligned - size);
   return uint32_t(size_aligned);
 }
 
+void DxbcShaderTranslator::WriteResourceDefinitions() {
+  uint32_t chunk_position_dwords = uint32_t(shader_object_.size());
+
+  // Write the header.
+  // Constant buffer count.
+  // b0 - Xenia system constants.
+  // b1 - bool and loop constants.
+  // b2 - fetch constants.
+  // b3-b10 - float constants.
+  shader_object_.push_back(4);
+  // Constant buffer offset (set later).
+  shader_object_.push_back(0);
+  // TODO(Triang3l): Bound resource count (CBV + other views) (set later).
+  shader_object_.push_back(4);
+  // TODO(Triang3l): Bound resource buffer offset (set later).
+  shader_object_.push_back(0);
+  if (is_vertex_shader()) {
+    // vs_5_1
+    shader_object_.push_back(0xFFFE0501u);
+  } else {
+    assert_true(is_pixel_shader());
+    // ps_5_1
+    shader_object_.push_back(0xFFFF0501u);
+  }
+  // Compiler flags - default for SM 5.1 (no preshader, prefer flow control).
+  shader_object_.push_back(0x500);
+  // Generator offset (directly after the RDEF header in our case).
+  shader_object_.push_back(60);
+  // RD11, but with nibbles inverted (unlike in SM 5.0).
+  shader_object_.push_back(0x25441313);
+  // Unknown fields.
+  shader_object_.push_back(60);
+  shader_object_.push_back(24);
+  // Was 32 in SM 5.0.
+  shader_object_.push_back(40);
+  shader_object_.push_back(40);
+  shader_object_.push_back(36);
+  shader_object_.push_back(12);
+  shader_object_.push_back(0);
+  // Generator name.
+  AppendString(shader_object_, "Xenia");
+
+  // Constant buffers.
+  uint32_t cbuffer_position_dwords = uint32_t(shader_object_.size());
+  shader_object_[chunk_position_dwords + 1] =
+      cbuffer_position_dwords * sizeof(uint32_t);
+}
+
 }  // namespace gpu
 }  // namespace xe

src/xenia/gpu/dxbc_shader_translator.h

@@ -32,6 +32,8 @@ class DxbcShaderTranslator : public ShaderTranslator {
   // Appends a string to a DWORD stream, returns the DWORD-aligned length.
   static uint32_t AppendString(std::vector<uint32_t>& dest, const char* source);
 
+  void WriteResourceDefinitions();
+
   // Executable instructions - generated during translation.
   std::vector<uint32_t> shader_code_;
   // Complete shader object, with all the needed chunks and dcl_ instructions -