[Enhancement] Optimize code in arm #55072

before-Sunrise · 2025-01-14T10:59:38Z

Why I'm doing:

arm is slower than x86 in some cases

What I'm doing:

vectorize rf's insert_hash using Neon intrinsics
streamvbyte's cmakelist is wrong, which cause performance downgrade in arm because vectorization cannot work properly
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
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:

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

Bugfix cherry-pick branch check:

Signed-off-by: before-Sunrise <[email protected]>

github-actions · 2025-01-26T14:51:37Z

[Java-Extensions Incremental Coverage Report]

✅ pass : 0 / 0 (0%)

github-actions · 2025-01-26T14:51:42Z

[FE Incremental Coverage Report]

✅ pass : 0 / 0 (0%)

github-actions · 2025-01-26T15:02:51Z

[BE Incremental Coverage Report]

✅ pass : 35 / 35 (100.00%)

file detail

	path	covered_line	new_line	coverage	not_covered_line_detail
🔵	be/src/runtime/int128_to_double.cpp	35	35	100.00%	[]

github-actions · 2025-01-27T03:18:21Z

@Mergifyio backport branch-3.4

github-actions · 2025-01-27T03:18:22Z

@Mergifyio backport branch-3.3

mergify · 2025-01-27T03:18:47Z

backport branch-3.4

✅ Backports have been created

#55487 [Enhancement] Optimize code in arm (backport #55072) has been created for branch branch-3.4

mergify · 2025-01-27T03:18:48Z

backport branch-3.3

✅ Backports have been created

#55488 [Enhancement] Optimize code in arm (backport #55072) has been created for branch branch-3.3

## 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)

## 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]>

Signed-off-by: before-Sunrise <[email protected]>

before-Sunrise added 3 commits January 14, 2025 18:19

support int128 check

97d1fcc

Signed-off-by: before-Sunrise <[email protected]>

optimize insert_hash

5503db6

Signed-off-by: before-Sunrise <[email protected]>

patch

e7e28e9

Signed-off-by: before-Sunrise <[email protected]>

before-Sunrise requested review from a team as code owners January 14, 2025 10:59

mergify bot assigned before-Sunrise Jan 14, 2025

github-actions bot added 3.4 3.3 labels Jan 14, 2025

stdpain previously approved these changes Jan 15, 2025

View reviewed changes

kangkaisen previously approved these changes Jan 22, 2025

View reviewed changes

before-Sunrise dismissed stale reviews from kangkaisen and stdpain via 4741726 January 26, 2025 08:50

before-Sunrise changed the title ~~[Enhancement] Optimize code in arm~~ [WIP] Optimize code in arm Jan 26, 2025

github-actions bot added the title needs [type] label Jan 26, 2025

wrap __floattidf

5c2976b

Signed-off-by: before-Sunrise <[email protected]>

before-Sunrise force-pushed the optimize_arm branch from 4741726 to 5c2976b Compare January 26, 2025 09:00

before-Sunrise changed the title ~~[WIP] Optimize code in arm~~ [Enhancement] Optimize code in arm Jan 26, 2025

github-actions bot removed the title needs [type] label Jan 26, 2025

before-Sunrise added 2 commits January 26, 2025 17:25

add dcheck

188ba50

Signed-off-by: before-Sunrise <[email protected]>

fix compile

5529ca0

Signed-off-by: before-Sunrise <[email protected]>

satanson approved these changes Jan 27, 2025

View reviewed changes

stdpain approved these changes Jan 27, 2025

View reviewed changes

alvin-celerdata approved these changes Jan 27, 2025

View reviewed changes

alvin-celerdata merged commit e88bb85 into StarRocks:main Jan 27, 2025
62 checks passed

github-actions bot removed the 3.4 label Jan 27, 2025

github-actions bot removed the 3.3 label Jan 27, 2025

mergify bot mentioned this pull request Jan 27, 2025

[Enhancement] Optimize code in arm (backport #55072) #55487

Open

18 tasks

mergify bot mentioned this pull request Jan 27, 2025

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

Open

18 tasks

before-Sunrise mentioned this pull request Jan 29, 2025

[BugFix] fix dcheck error for __floattidf #55510

Merged

24 tasks

before-Sunrise mentioned this pull request Feb 6, 2025

[Enhancement] Optimize code in arm (backport #55072) #55607

Merged

24 tasks

before-Sunrise mentioned this pull request Feb 6, 2025

[Enhancement] Optimize code in arm (backport #55072) #55608

Merged

24 tasks

silverbullet233 pushed a commit that referenced this pull request Feb 6, 2025

[Enhancement] Optimize code in arm (backport #55072) (#55607)

f7492cd

Signed-off-by: before-Sunrise <[email protected]>

github-actions bot added the 3.4-merged label Feb 6, 2025

kangkaisen pushed a commit that referenced this pull request Feb 6, 2025

[Enhancement] Optimize code in arm (backport #55072) (#55608)

127e1cf

Signed-off-by: before-Sunrise <[email protected]>

github-actions bot added the 3.3-merged label Feb 6, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Enhancement] Optimize code in arm #55072

[Enhancement] Optimize code in arm #55072

before-Sunrise commented Jan 14, 2025 •

edited

Loading

github-actions bot commented Jan 26, 2025

github-actions bot commented Jan 26, 2025

github-actions bot commented Jan 26, 2025

github-actions bot commented Jan 27, 2025

github-actions bot commented Jan 27, 2025

mergify bot commented Jan 27, 2025 •

edited

Loading

mergify bot commented Jan 27, 2025 •

edited

Loading

[Enhancement] Optimize code in arm #55072

[Enhancement] Optimize code in arm #55072

Conversation

before-Sunrise commented Jan 14, 2025 • edited Loading

Why I'm doing:

What I'm doing:

What type of PR is this:

Checklist:

Bugfix cherry-pick branch check:

github-actions bot commented Jan 26, 2025

[Java-Extensions Incremental Coverage Report]

github-actions bot commented Jan 26, 2025

[FE Incremental Coverage Report]

github-actions bot commented Jan 26, 2025

[BE Incremental Coverage Report]

file detail

github-actions bot commented Jan 27, 2025

github-actions bot commented Jan 27, 2025

mergify bot commented Jan 27, 2025 • edited Loading

✅ Backports have been created

mergify bot commented Jan 27, 2025 • edited Loading

✅ Backports have been created

before-Sunrise commented Jan 14, 2025 •

edited

Loading

mergify bot commented Jan 27, 2025 •

edited

Loading

mergify bot commented Jan 27, 2025 •

edited

Loading