Batch flushing buffer.

src/xenia/gpu/gl4/circular_buffer.cc

@@ -24,6 +24,8 @@ CircularBuffer::CircularBuffer(size_t capacity, size_t alignment)
     : capacity_(capacity),
       alignment_(alignment),
       write_head_(0),
+      dirty_start_(UINT64_MAX),
+      dirty_end_(0),
       buffer_(0),
       gpu_base_(0),
       host_base_(nullptr) {}
@@ -63,12 +65,15 @@ void CircularBuffer::Shutdown() {
   buffer_ = 0;
 }
 
+bool CircularBuffer::CanAcquire(size_t length) {
+  size_t aligned_length = poly::round_up(length, alignment_);
+  return write_head_ + aligned_length <= capacity_;
+}
+
 CircularBuffer::Allocation CircularBuffer::Acquire(size_t length) {
   // Addresses must always be % 256.
   size_t aligned_length = poly::round_up(length, alignment_);
-
   assert_true(aligned_length <= capacity_, "Request too large");
-
   if (write_head_ + aligned_length > capacity_) {
     // Flush and wait.
     WaitUntilClean();
@@ -79,16 +84,30 @@ CircularBuffer::Allocation CircularBuffer::Acquire(size_t length) {
   allocation.gpu_ptr = gpu_base_ + write_head_;
   allocation.offset = write_head_;
   allocation.length = length;
+  allocation.aligned_length = aligned_length;
   write_head_ += aligned_length;
   return allocation;
 }
 
 void CircularBuffer::Commit(Allocation allocation) {
-  glFlushMappedNamedBufferRange(buffer_, allocation.gpu_ptr - gpu_base_,
-                                allocation.length);
+  uintptr_t start = allocation.gpu_ptr - gpu_base_;
+  uintptr_t end = start + allocation.aligned_length;
+  dirty_start_ = std::min(dirty_start_, start);
+  dirty_end_ = std::max(dirty_end_, end);
+}
+
+void CircularBuffer::Flush() {
+  if (dirty_start_ == dirty_end_ || dirty_start_ == UINT64_MAX) {
+    return;
+  }
+  glFlushMappedNamedBufferRange(buffer_, dirty_start_,
+                                dirty_end_ - dirty_start_);
+  dirty_start_ = UINT64_MAX;
+  dirty_end_ = 0;
 }
 
 void CircularBuffer::WaitUntilClean() {
+  Flush();
   glFinish();
   write_head_ = 0;
 }

src/xenia/gpu/gl4/circular_buffer.h

@@ -28,6 +28,7 @@ class CircularBuffer {
     GLuint64 gpu_ptr;
     size_t offset;
     size_t length;
+    size_t aligned_length;
   };
 
   bool Initialize();
@@ -35,8 +36,10 @@ class CircularBuffer {
 
   GLuint handle() const { return buffer_; }
 
+  bool CanAcquire(size_t length);
   Allocation Acquire(size_t length);
   void Commit(Allocation allocation);
+  void Flush();
 
   void WaitUntilClean();
 
@@ -44,6 +47,8 @@ class CircularBuffer {
   size_t capacity_;
   size_t alignment_;
   uintptr_t write_head_;
+  uintptr_t dirty_start_;
+  uintptr_t dirty_end_;
   GLuint buffer_;
   GLuint64 gpu_base_;
   uint8_t* host_base_;

src/xenia/gpu/gl4/command_processor.cc

@@ -1428,6 +1428,7 @@ bool CommandProcessor::IssueDraw(DrawCommand* draw_command) {
   glBindBufferRange(GL_SHADER_STORAGE_BUFFER, 0, scratch_buffer_.handle(),
                     allocation.offset, allocation.length);
   scratch_buffer_.Commit(std::move(allocation));
+  scratch_buffer_.Flush();
 
   if (cmd.index_buffer.address) {
     // Indexed draw.

src/xenia/gpu/gl4/gl4_profiler_display.cc

@@ -382,7 +382,11 @@ GL4ProfilerDisplay::Vertex* GL4ProfilerDisplay::BeginVertices(size_t count) {
   if (draw_command_count_ + 1 > kMaxCommands) {
     Flush();
   }
-  current_allocation_ = vertex_buffer_.Acquire(sizeof(Vertex) * count);
+  size_t total_length = sizeof(Vertex) * count;
+  if (!vertex_buffer_.CanAcquire(total_length)) {
+    Flush();
+  }
+  current_allocation_ = vertex_buffer_.Acquire(total_length);
   return reinterpret_cast<Vertex*>(current_allocation_.host_ptr);
 }
 
@@ -408,12 +412,14 @@ void GL4ProfilerDisplay::Flush() {
   if (!draw_command_count_) {
     return;
   }
+  vertex_buffer_.Flush();
   for (size_t i = 0; i < draw_command_count_; ++i) {
     glDrawArrays(draw_commands_[i].prim_type,
                  GLint(draw_commands_[i].vertex_offset),
                  GLsizei(draw_commands_[i].vertex_count));
   }
   draw_command_count_ = 0;
+  // TODO(benvanik): don't finish here.
   vertex_buffer_.WaitUntilClean();
 }