[Felix-Gong]

Summary

• Add RISC-V 64-bit prefetch optimization to zstd doubleFast block compressor
• Extend existing aarch64 PREFETCH_L1(ip+256) to RISC-V via macro guard
• 2 lines changed in zstd_double_fast.c

Changes

Extended the architecture guard from #if defined(__aarch64__) to #if defined(__aarch64__) || (defined(__riscv) && (__riscv_xlen == 64)) in two locations:

1. ZSTD_compressBlock_doubleFast_noDict_generic (line 236)
2. ZSTD_compressBlock_doubleFast_dictMatchState_generic (line 451)

Performance Results (RISC-V 64-bit, RV64GC)

Data Type	Compress	Decompress
Random (100MB)	+9.0% (337.3 → 367.6 MB/s)	+8.0% (2763.0 → 2983.9 MB/s)
Zero (100MB)	-0.3% (within noise)	-0.05% (within noise)

Testing

• Platform: RISC-V 64-bit server
• Compiler: GCC 15.1
• Test data: Random and zero-filled data (100MB each)
• Benchmark: zstd -b

Notes

• The prefetch targets sequential data access patterns in the compression loop
• No additional computation added to the hot loop
• Follows the same pattern as the existing aarch64 optimization

[meta-cla]

Hi @Felix-Gong!

Thank you for your pull request and welcome to our community.

Action Required

In order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you.

Process

In order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.

Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.

If you have received this in error or have any questions, please contact us at cla@meta.com. Thanks!

[meta-cla]

Thank you for signing our Contributor License Agreement. We can now accept your code for this (and any) Meta Open Source project. Thanks!

[luhenry]

@gctony hey 👋 would you know who would be a good person to review this? Thank you!

[Cyan4973]

This is a fairly simple and risk-free modification.
It should not impact any other architecture beyond risc-v.
And even on risc-v, it only impacts speed, not correctness.

This repository is only equipped to validate correctness, via qemu.
For speed, we are entirely reliant on external measurements,
keeping in mind that results can vary a lot depending on exact silicon.

In this case, my only gripe is that speed has been measured on 2 opposite corner cases,
impossible to compress, or a bunch of zeroes,
none of which being representative of any real-world scenario.

For something a bit more representative,
I would recommend using the lorem ipsum generator,
which is provided as a stand-alone tool tests/datagen,
or the synthetic generator directly integrated within zstd -b.
Both will provide "reasonably compressible" data, without falling into either extreme.

Also: this should really only impact compression speed, given the modification only mutates the source file for compression levels 3 & 4. Speed impact on the decompression side should be regarded as fishy, or at least unintentional.