GPU: Fix incorrectly calculating the primary ringbuffer's size

src/xenia/gpu/command_processor.cc

@@ -129,7 +129,7 @@ void CommandProcessor::WorkerThreadMain() {
 
     uint32_t write_ptr_index = write_ptr_index_.load();
     if (write_ptr_index == 0xBAADF00D || read_ptr_index_ == write_ptr_index) {
-      SCOPE_profile_cpu_i("gpu", "xe::gpu::gl4::CommandProcessor::Stall");
+      SCOPE_profile_cpu_i("gpu", "xe::gpu::CommandProcessor::Stall");
       // We've run out of commands to execute.
       // We spin here waiting for new ones, as the overhead of waiting on our
       // event is too high.
@@ -223,12 +223,9 @@ bool CommandProcessor::SetupContext() { return true; }
 
 void CommandProcessor::ShutdownContext() { context_.reset(); }
 
-void CommandProcessor::InitializeRingBuffer(uint32_t ptr, uint32_t page_count) {
+void CommandProcessor::InitializeRingBuffer(uint32_t ptr, uint32_t log2_size) {
   primary_buffer_ptr_ = ptr;
-  // Not sure this is correct, but it's a way to take the page_count back to
-  // the number of bytes allocated by the physical alloc.
-  uint32_t original_size = 1 << (0x1C - page_count - 1);
-  primary_buffer_size_ = original_size;
+  primary_buffer_size_ = uint32_t(std::pow(2u, log2_size));
 }
 
 void CommandProcessor::EnableReadPointerWriteBack(uint32_t ptr,
@@ -1280,6 +1277,8 @@ bool CommandProcessor::ExecutePacketType3_VIZ_QUERY(RingBuffer* reader,
   // Some sort of ID?
   // This appears to reset a viz query context.
   // This ID matches the ID in conditional draw commands.
+  // Patent says the driver sets the viz_query register with info about the
+  // context ID.
   uint32_t dword0 = reader->Read<uint32_t>(true);
 
   return true;

src/xenia/gpu/graphics_system.cc

@@ -181,8 +181,8 @@ void GraphicsSystem::WriteRegister(uint32_t addr, uint32_t value) {
   register_file_.values[r].u32 = value;
 }
 
-void GraphicsSystem::InitializeRingBuffer(uint32_t ptr, uint32_t page_count) {
-  command_processor_->InitializeRingBuffer(ptr, page_count);
+void GraphicsSystem::InitializeRingBuffer(uint32_t ptr, uint32_t log2_size) {
+  command_processor_->InitializeRingBuffer(ptr, (log2_size | 0x2) + 1);
 }
 
 void GraphicsSystem::EnableReadPointerWriteBack(uint32_t ptr,

src/xenia/gpu/graphics_system.h

@@ -49,7 +49,7 @@ class GraphicsSystem {
     return command_processor_.get();
   }
 
-  void InitializeRingBuffer(uint32_t ptr, uint32_t page_count);
+  void InitializeRingBuffer(uint32_t ptr, uint32_t log2_size);
   void EnableReadPointerWriteBack(uint32_t ptr, uint32_t block_size);
 
   void SetInterruptCallback(uint32_t callback, uint32_t user_data);

src/xenia/kernel/xboxkrnl/xboxkrnl_video.cc

@@ -200,15 +200,18 @@ void VdSetGraphicsInterruptCallback(function_t callback, lpvoid_t user_data) {
 }
 DECLARE_XBOXKRNL_EXPORT(VdSetGraphicsInterruptCallback, ExportTag::kVideo);
 
-void VdInitializeRingBuffer(lpvoid_t ptr, int_t page_count) {
+void VdInitializeRingBuffer(lpvoid_t ptr, int_t log2_size) {
   // r3 = result of MmGetPhysicalAddress
-  // r4 = number of pages? page size?
-  //      0x8000 -> cntlzw=16 -> 0x1C - 16 = 12
+  // r4 = log2(size)
+  // r4 is or'd with 0x802 and then stuffed into CP_RB_CNTL
+  // according to AMD docs, this corresponds with RB_BUFSZ, which is log2
+  // actual size.
+  // 0x8 is RB_BLKSZ, or number of words gpu will read before updating the
+  // host read pointer.
+  // So being or'd with 0x2 makes the ring buffer size always a multiple of 4.
   // Buffer pointers are from MmAllocatePhysicalMemory with WRITE_COMBINE.
-  // Sizes could be zero? XBLA games seem to do this. Default sizes?
-  // D3D does size / region_count - must be > 1024
   auto graphics_system = kernel_state()->emulator()->graphics_system();
-  graphics_system->InitializeRingBuffer(ptr, page_count);
+  graphics_system->InitializeRingBuffer(ptr, log2_size);
 }
 DECLARE_XBOXKRNL_EXPORT(VdInitializeRingBuffer, ExportTag::kVideo);