Use mapped memory for uploading large textures.

gpu/command_buffer/client/transfer_buffer.cc

Index: gpu/command_buffer/client/transfer_buffer.cc
diff --git a/gpu/command_buffer/client/transfer_buffer.cc b/gpu/command_buffer/client/transfer_buffer.cc
index cb02f15144e0eda9327edb867a9c2516670da09a..3ff257da5ad7c09aa1f448b1194783910a48acdf 100644
--- a/gpu/command_buffer/client/transfer_buffer.cc
+++ b/gpu/command_buffer/client/transfer_buffer.cc
@@ -73,6 +73,10 @@ RingBuffer::Offset TransferBuffer::GetOffset(void* pointer) const {
   return ring_buffer_->GetOffset(pointer);
 }
 
+void TransferBuffer::DiscardBlock(void* p) {
+  ring_buffer_->DiscardBlock(p);
+}
+
 void TransferBuffer::FreePendingToken(void* p, unsigned int token) {
   ring_buffer_->FreePendingToken(p, token);
   if (bytes_since_last_flush_ >= size_to_flush_ && size_to_flush_ > 0) {
@@ -188,6 +192,14 @@ void ScopedTransferBufferPtr::Release() {
   }
 }
 
+void ScopedTransferBufferPtr::Discard() {
+  if (buffer_) {
+    transfer_buffer_->DiscardBlock(buffer_);
+    buffer_ = NULL;
+    size_ = 0;
+  }
+}
+
 void ScopedTransferBufferPtr::Reset(unsigned int new_size) {
   Release();
   // NOTE: we allocate buffers of size 0 so that HaveBuffer will be true, so