Reuse the ZSTD context and skip compressing all-zero samples in CompressionAnalyzer. - #999
Draft
copybara-service[bot] wants to merge 1 commit into
Draft
copybara-service[bot] wants to merge 1 commit into
copybara-service[bot] wants to merge 1 commit into
Conversation
…essionAnalyzer. ZSTD_compress() builds and tears down a full compression context on every call: at level 1 that is a ~1 MiB workspace malloc'd and freed through the page heap once per sampled allocation while building a heap profile. Create one ZSTD_CCtx per CompressionAnalyzer and drive it with ZSTD_compressCCtx(), which mirrors ZSTD_compress() semantics (explicit level, no sticky parameters) so compressed byte counts are unchanged. An instance was already single-use scratch state (local_copy_, compress_buf_) and profile_builder walks samples serially with one instance; the class is now non-copyable. Count zero bytes a uint64_t at a time (SWAR zero-lane mask + popcount) instead of absl::c_count, which clang did not vectorize (~2.4 GiB/s). An all-zero word costs one load+compare, so all-zero samples scan at ~15 GiB/s and mixed data at 4-7 GiB/s. The count doubles as the all-zero test: such samples skip ZSTD entirely and report compressed_size = 0. Previously they reported only ZSTD framing overhead (~11 bytes for 8 KiB, ~73 bytes for 2 MiB, scaled), so the estimate differs by at most 0.004% of the sample; zero_bytes is identical. PiperOrigin-RevId: 983583135
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Reuse the ZSTD context and skip compressing all-zero samples in CompressionAnalyzer.
ZSTD_compress() builds and tears down a full compression context on every
call: at level 1 that is a ~1 MiB workspace malloc'd and freed through the page
heap once per sampled allocation while building a heap profile. Create one
ZSTD_CCtx per CompressionAnalyzer and drive it with ZSTD_compressCCtx(), which
mirrors ZSTD_compress() semantics (explicit level, no sticky parameters) so
compressed byte counts are unchanged. An instance was already single-use
scratch state (local_copy_, compress_buf_) and profile_builder walks samples
serially with one instance; the class is now non-copyable.
Count zero bytes a uint64_t at a time (SWAR zero-lane mask + popcount) instead
of absl::c_count, which clang did not vectorize (~2.4 GiB/s). An all-zero
word costs one load+compare, so all-zero samples scan at ~15 GiB/s and mixed
data at 4-7 GiB/s. The count doubles as the all-zero test: such samples skip
ZSTD entirely and report compressed_size = 0. Previously they reported only
ZSTD framing overhead (~11 bytes for 8 KiB, ~73 bytes for 2 MiB, scaled), so
the estimate differs by at most 0.004% of the sample; zero_bytes is identical.