perf: reduce copies and skip zeroing memory we read into #1238
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Benchmark results (using benchmarks in #1235): |
changhiskhan
reviewed
Sep 6, 2023
| @@ -230,9 +231,22 @@ impl<'a, T: ByteArrayType> BinaryDecoder<'a, T> { | |||
| .null_bit_buffer(Some(null_buf.into())); | |||
| } | |||
|
|
|||
| // TODO: replace this with safe method once arrow-rs 47.0.0 comes out. | |||
There was a problem hiding this comment.
when is 47.0.0 coming out?
There was a problem hiding this comment.
Maybe a week or two? And then we have to wait another two weeks for the next Datafusion release cycle. So we can do this in like 4 weeks I think.
| // Zero-copy conversion from bytes | ||
| // Safety: the bytes are owned by the `data` value, so the pointer | ||
| // will be valid for the lifetime of the Arc we are passing in. | ||
| let buf = unsafe { |
There was a problem hiding this comment.
we don't need the alignment check because this is already byte_width aligned?
also, technically bytes can be empty right? Would that cause any issues with NonNull::new(bytes.as_ptr() as _).unwrap()?
There was a problem hiding this comment.
Yeah for bytes I think we just require need it to be aligned with u8, which is true of any allocation.
There was a problem hiding this comment.
Our simd code uses unaligned load now, so it is unnecessary to be word-aligned (i.e., 8 bytes). About 10% overhead due to misaligned memory loading in SIMD computation, much less in overall e2e latency.
There was a problem hiding this comment.
Yeah I align the primitive buffers because arrow-rs errors otherwise
eddyxu
approved these changes
Sep 6, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
When reading primitive and binary data, we are copying data. We can skip the copies if the buffer is already correctly aligned.
Also, for local filesystem we are spending a lot of time zeroing memory that we immediately write into. This PR change to just initialize without zeroing and write into the data. A check is added to make sure that we don't expose any uninitialized data in the output buffer.