Add 1-way SSE4 SHA256 implementation using intrinsics for MSVC builds #28526

1. 
   hebasto commented at 2:18 pm on September 24, 2023: member

   This PR reintroduces the 1-way SSE4 SHA256 implementation using intrinsics, as suggested in #13442, specifically for MSVC builds, where a 50% performance gain has been achieved.

   Here are benchmarks on my machine with Intel Core i5-8350U CPU (no sha_ni flag) + Windows 11 Pro 22H2:

   • before this PR (8a9e37fb95cbb0bf7f6e06fa05d8381db04d61e2):

    0>.\src\bench_bitcoin.exe -filter=SHA256_.*
    1
    2|             ns/byte |              byte/s |    err% |     total | benchmark
    3|--------------------:|--------------------:|--------:|----------:|:----------
    4|                9.92 |      100,826,852.23 |    0.1% |      0.01 | SHA256_32b_AVX2 using the 'standard,sse41(4way),avx2(8way)' SHA256 implementation
    5|                9.90 |      101,038,141.67 |    0.3% |      0.01 | SHA256_32b_SHANI using the 'standard,sse41(4way)' SHA256 implementation
    6|               10.02 |       99,788,852.31 |    0.9% |      0.01 | SHA256_32b_SSE4 using the 'standard,sse41(4way)' SHA256 implementation
    7|               10.01 |       99,883,509.98 |    0.8% |      0.01 | SHA256_32b_STANDARD using the 'standard' SHA256 implementation
    8|                4.48 |      223,348,893.31 |    1.1% |      0.05 | SHA256_AVX2 using the 'standard,sse41(4way),avx2(8way)' SHA256 implementation
    9|                4.47 |      223,668,612.58 |    1.2% |      0.05 | SHA256_SHANI using the 'standard,sse41(4way)' SHA256 implementation
   10|                4.45 |      224,638,332.29 |    0.7% |      0.05 | SHA256_SSE4 using the 'standard,sse41(4way)' SHA256 implementation
   11|                4.45 |      224,542,494.67 |    0.6% |      0.05 | SHA256_STANDARD using the 'standard' SHA256 implementation
   

   • with this PR:

    0>.\src\bench_bitcoin.exe -filter=SHA256_.*
    1
    2|             ns/byte |              byte/s |    err% |     total | benchmark
    3|--------------------:|--------------------:|--------:|----------:|:----------
    4|                7.04 |      142,024,691.36 |    0.2% |      0.01 | SHA256_32b_AVX2 using the 'sse41(1way),sse41(4way),avx2(8way)' SHA256 implementation
    5|                7.03 |      142,222,222.22 |    0.2% |      0.01 | SHA256_32b_SHANI using the 'sse41(1way),sse41(4way)' SHA256 implementation
    6|                7.08 |      141,231,323.51 |    0.8% |      0.01 | SHA256_32b_SSE4 using the 'sse41(1way),sse41(4way)' SHA256 implementation
    7|                9.88 |      101,196,866.84 |    0.4% |      0.01 | SHA256_32b_STANDARD using the 'standard' SHA256 implementation
    8|                3.01 |      332,270,069.11 |    1.3% |      0.03 | SHA256_AVX2 using the 'sse41(1way),sse41(4way),avx2(8way)' SHA256 implementation
    9|                3.00 |      332,989,244.45 |    0.3% |      0.03 | SHA256_SHANI using the 'sse41(1way),sse41(4way)' SHA256 implementation
   10|                3.04 |      328,612,270.38 |    2.0% |      0.03 | SHA256_SSE4 using the 'sse41(1way),sse41(4way)' SHA256 implementation
   11|                4.45 |      224,678,709.45 |    0.4% |      0.05 | SHA256_STANDARD using the 'standard' SHA256 implementation
   

   Based on #24773.

2. 
   DrahtBot commented at 2:18 pm on September 24, 2023: contributor

   The following sections might be updated with supplementary metadata relevant to reviewers and maintainers.

   Code Coverage

   For detailed information about the code coverage, see the test coverage report.

   Reviews

   See the guideline for information on the review process. A summary of reviews will appear here.

   Conflicts

   Reviewers, this pull request conflicts with the following ones:

   • #29774 (build: Enable fuzz binary in MSVC by hebasto)
   • #29625 (Several randomness improvements by sipa)

   If you consider this pull request important, please also help to review the conflicting pull requests. Ideally, start with the one that should be merged first.

3. hebasto marked this as a draft on Sep 24, 2023
4. hebasto force-pushed on Oct 4, 2023
5. 
   hebasto commented at 3:19 pm on October 4, 2023: member

   Rebased on top of the merged #27598.
6. DrahtBot added the label CI failed on Jan 17, 2024
7. DrahtBot added the label Needs rebase on Jan 27, 2024
8. Add MSVC implementation of GetCPUID() cf96097de3
9. Add MSVC implementation of AVXEnabled() 1dc8641dad
10. msvc: Enable AVX2 implementation of SHA256 2df3619a97
11. msvc: Enable SSE4.1 implementation of SHA256 422d813081
12. msvc: Enable x86 SHA-NI implementation of SHA256 6e2513f03e
13. Add 1-way SSE4 SHA256 implementation using intrinsics 95bafb0007
14. msvc: Use 1-way SSE4 SHA256 intrinsics-based implementation 0e9c80507b
15. hebasto force-pushed on Feb 12, 2024
16. DrahtBot removed the label Needs rebase on Feb 12, 2024
17. DrahtBot removed the label CI failed on Feb 12, 2024
18. DrahtBot added the label Needs rebase on Apr 28, 2024
19. 
    DrahtBot commented at 3:29 am on April 28, 2024: contributor

    🐙 This pull request conflicts with the target branch and needs rebase.

20. 
    hebasto commented at 5:25 am on April 28, 2024: member

    Based on #24773.

    Deferring to after cmake.

21. hebasto closed this on Apr 28, 2024

    ---

Contributors
hebasto DrahtBot

Labels
Needs rebase