Lunderberg commented on a change in pull request #8157:
URL: https://github.com/apache/tvm/pull/8157#discussion_r641561663
##########
File path: src/runtime/vulkan/vulkan_stream.h
##########
@@ -43,135 +44,64 @@ struct VulkanStreamToken {
std::vector<VkBuffer> buffers_;
};
+/*!
+ * \brief Wrapper around a vulkan command buffer
+ *
+ * The VulkanStream collects commands into a VkCommandBuffer. When a
+ * newly submitted command requires resources reserved by an
+ * already-submitted command, all of the queued commands are
+ * submitted to the GPU, and the CPU waits for all queued commands to
+ * finish. The queued commands can also be explicitly pushed/waited
+ * on by calling VulkanStream::Synchronize.
+ *
+ * If the device is
Review comment:
Thank you for that catch, and next time I should probably separate out
the documentation from the refactoring to make those items be easier to find
ahead of time. I have no memory of what I was going to finish that paragraph
with, so I have deleted it for now.
Also, fixed up the linting errors.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
