Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[Enhancement] Optimize code in arm (backport #55072) #55488

Open
wants to merge 1 commit into
base: branch-3.3
Choose a base branch
from

Conversation

mergify[bot]
Copy link
Contributor

@mergify mergify bot commented Jan 27, 2025

Why I'm doing:

arm is slower than x86 in some cases

What I'm doing:

  1. vectorize rf's insert_hash using Neon intrinsics
  2. streamvbyte's cmakelist is wrong, which cause performance downgrade in arm because vectorization cannot work properly
  3. arm's int128_mul_overflow is super slow becase of divide operation, __builtin_mul_overflow(int128_t a, int128_t b, int128_t* c) is fast enough when compile with gcc. But gcc's __builtin_mul_overflow is at least 5 times faster then clang in arm, we already reported it to the community: [clang++][aarch64] help optimize __builtin_mul_overflow performance llvm/llvm-project#123262. So we still use gcc as default compiler and use __builtin_mul_overflow to replace original int128_mul_overflow implementation
  4. arm's cast int128 to double is super slow in arm with gcc because the bad implementation of __floattidf, clang runtime-rt's implementation is 20 times faster then gcc, so I used clang compiler-rt's implementation to replace gcc's version

after this pr, arm is faster then gcc in the most of cases.

| Query   | arm-opt | x86 |
|---------|--------|--------|
| QUERY01 | 36     | 61     |
| QUERY02 | 39     | 62     |
| QUERY14 | 1510   | 1514   |
| QUERY15 | 1407   | 1496   |
| QUERY17 | 21     | 88     |
| QUERY20 | 151    | 279    |
| QUERY21 | 1526   | 1529   |
| QUERY24 | 1399   | 1504   |
| QUERY26 | 32     | 122    |
| QUERY27 | 1493   | 1519   |
| QUERY90 | 3399   | 4030   |
| QUERY97 | 3859   | 4776   |
| QUERY98 | 2763   | 3208   |
| QUERY99 | 868    | 1259   |

What type of PR is this:

  • BugFix
  • Feature
  • Enhancement
  • Refactor
  • UT
  • Doc
  • Tool

Does this PR entail a change in behavior?

  • Yes, this PR will result in a change in behavior.
  • No, this PR will not result in a change in behavior.

If yes, please specify the type of change:

  • Interface/UI changes: syntax, type conversion, expression evaluation, display information
  • Parameter changes: default values, similar parameters but with different default values
  • Policy changes: use new policy to replace old one, functionality automatically enabled
  • Feature removed
  • Miscellaneous: upgrade & downgrade compatibility, etc.

Checklist:

  • I have added test cases for my bug fix or my new feature
  • This pr needs user documentation (for new or modified features or behaviors)
  • I have added documentation for my new feature or new function
  • This is a backport pr

## Why I'm doing:
arm is slower than x86 in some cases

## What I'm doing:
1. vectorize rf's insert_hash using Neon intrinsics
2. streamvbyte's cmakelist is wrong, which cause performance downgrade in arm because vectorization cannot work properly
3. arm's int128_mul_overflow is super slow becase of divide operation,  __builtin_mul_overflow(int128_t a, int128_t b, int128_t* c) is fast enough when compile with gcc. But gcc's __builtin_mul_overflow is at least 5 times faster then clang in arm, we already reported it to the community: llvm/llvm-project#123262. So we still use gcc as default compiler and use __builtin_mul_overflow to replace original int128_mul_overflow implementation
4. arm's cast int128 to double  is super slow in arm with gcc because the bad implementation of __floattidf, clang runtime-rt's implementation is 20 times faster then gcc, so I used clang compiler-rt's implementation to replace gcc's version

after this pr, arm is faster then gcc in  the most of cases.

```
| Query   | arm-opt | x86 |
|---------|--------|--------|
| QUERY01 | 36     | 61     |
| QUERY02 | 39     | 62     |
| QUERY14 | 1510   | 1514   |
| QUERY15 | 1407   | 1496   |
| QUERY17 | 21     | 88     |
| QUERY20 | 151    | 279    |
| QUERY21 | 1526   | 1529   |
| QUERY24 | 1399   | 1504   |
| QUERY26 | 32     | 122    |
| QUERY27 | 1493   | 1519   |
| QUERY90 | 3399   | 4030   |
| QUERY97 | 3859   | 4776   |
| QUERY98 | 2763   | 3208   |
| QUERY99 | 868    | 1259   |
```

Signed-off-by: before-Sunrise <[email protected]>
(cherry picked from commit e88bb85)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant