Skip to content

Optimize call result push to the stack#602

Open
chfast wants to merge 3 commits intomasterfrom
optimize_call_result_push
Open

Optimize call result push to the stack#602
chfast wants to merge 3 commits intomasterfrom
optimize_call_result_push

Conversation

@chfast
Copy link
Collaborator

@chfast chfast commented Oct 13, 2020

No description provided.

@chfast chfast requested review from axic and gumb0 October 13, 2020 17:47
@codecov
Copy link

codecov bot commented Oct 13, 2020 •
edited
Loading

Codecov Report

Merging #602 (dcd8e2f) into master (8482896) will decrease coverage by 0.00%.
The diff coverage is n/a.

❗ Current head dcd8e2f differs from pull request most recent head 90bd063. Consider uploading reports for the commit 90bd063 to get more accurate results

@@            Coverage Diff             @@
##           master     #602      +/-   ##
==========================================
- Coverage   99.28%   99.27%   -0.01%     
==========================================
  Files          86       86              
  Lines       13221    13221              
==========================================
- Hits        13126    13125       -1     
- Misses         95       96       +1
Flag Coverage Δ
rust 98.58% <ø> (-0.10%) ⬇️
spectests 89.98% <ø> (ø)
unittests 99.21% <ø> (ø)
unittests-32 99.31% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files Coverage Δ
bindings/rust/src/lib.rs 98.85% <0.00%> (-0.10%) ⬇️

@chfast
Copy link
Collaborator Author

chfast commented Oct 13, 2020

Clang 11

fizzy/execute/blake2b/512_bytes_rounds_1_mean                     -0.0471         -0.0471            86            82            86            82
fizzy/execute/blake2b/512_bytes_rounds_16_mean                    -0.0561         -0.0561          1306          1232          1306          1232
fizzy/execute/ecpairing/onepoint_mean                             +0.0807         +0.0807        401835        434259        401839        434263
fizzy/execute/keccak256/512_bytes_rounds_1_mean                   -0.0354         -0.0354           101            98           101            98
fizzy/execute/keccak256/512_bytes_rounds_16_mean                  -0.0459         -0.0459          1488          1420          1488          1420
fizzy/execute/memset/256_bytes_mean                               -0.0019         -0.0019             7             7             7             7
fizzy/execute/memset/60000_bytes_mean                             +0.0040         +0.0040          1580          1586          1580          1586
fizzy/execute/mul256_opt0/input1_mean                             -0.0336         -0.0336            29            28            29            28
fizzy/execute/ramanujan_pi/33_runs_mean                           +0.0016         +0.0016           115           115           115           115
fizzy/execute/sha1/512_bytes_rounds_1_mean                        -0.0011         -0.0011            90            90            90            90
fizzy/execute/sha1/512_bytes_rounds_16_mean                       -0.0034         -0.0034          1253          1248          1253          1248
fizzy/execute/sha256/512_bytes_rounds_1_mean                      +0.0524         +0.0523            84            88            84            88
fizzy/execute/sha256/512_bytes_rounds_16_mean                     +0.0542         +0.0542          1149          1211          1149          1211
fizzy/execute/taylor_pi/pi_1000000_runs_mean                      -0.0017         -0.0017         42370         42300         42371         42301
fizzy/execute/micro/eli_interpreter/exec105_mean                  -0.0080         -0.0080             5             5             5             5
fizzy/execute/micro/factorial/20_mean                             +0.0012         +0.0012             1             1             1             1
fizzy/execute/micro/fibonacci/24_mean                             -0.0055         -0.0055          5425          5395          5425          5395
fizzy/execute/micro/host_adler32/1_mean                           -0.0024         -0.0024             0             0             0             0
fizzy/execute/micro/host_adler32/1000_mean                        +0.0091         +0.0091            35            36            35            36
fizzy/execute/micro/spinner/1_mean                                +0.0239         +0.0239             0             0             0             0
fizzy/execute/micro/spinner/1000_mean                             +0.0002         +0.0002            10            10            10            10

GCC 10

fizzy/execute/blake2b/512_bytes_rounds_1_mean                     -0.0468         -0.0468            87            83            87            83
fizzy/execute/blake2b/512_bytes_rounds_16_mean                    -0.0500         -0.0500          1316          1250          1316          1250
fizzy/execute/ecpairing/onepoint_mean                             -0.0422         -0.0422        410735        393403        410738        393405
fizzy/execute/keccak256/512_bytes_rounds_1_mean                   -0.0474         -0.0474           101            97           101            97
fizzy/execute/keccak256/512_bytes_rounds_16_mean                  -0.0620         -0.0620          1503          1410          1503          1410
fizzy/execute/memset/256_bytes_mean                               -0.0594         -0.0594             7             7             7             7
fizzy/execute/memset/60000_bytes_mean                             -0.0638         -0.0638          1599          1497          1599          1497
fizzy/execute/mul256_opt0/input1_mean                             -0.0595         -0.0595            29            27            29            27
fizzy/execute/ramanujan_pi/33_runs_mean                           -0.0391         -0.0391           131           126           131           126
fizzy/execute/sha1/512_bytes_rounds_1_mean                        -0.0450         -0.0450            94            90            94            90
fizzy/execute/sha1/512_bytes_rounds_16_mean                       -0.0465         -0.0465          1314          1253          1314          1253
fizzy/execute/sha256/512_bytes_rounds_1_mean                      -0.1120         -0.1120           100            88           100            88
fizzy/execute/sha256/512_bytes_rounds_16_mean                     -0.1107         -0.1107          1367          1216          1367          1216
fizzy/execute/taylor_pi/pi_1000000_runs_mean                      -0.0119         -0.0119         40540         40059         40541         40059
fizzy/execute/micro/eli_interpreter/exec105_mean                  -0.0232         -0.0232             5             5             5             5
fizzy/execute/micro/factorial/20_mean                             -0.0319         -0.0319             1             1             1             1
fizzy/execute/micro/fibonacci/24_mean                             -0.0041         -0.0041          5230          5208          5230          5208
fizzy/execute/micro/host_adler32/1_mean                           +0.0009         +0.0009             0             0             0             0
fizzy/execute/micro/host_adler32/1000_mean                        +0.0113         +0.0113            29            30            29            30
fizzy/execute/micro/spinner/1_mean                                -0.0549         -0.0549             0             0             0             0
fizzy/execute/micro/spinner/1000_mean                             -0.0260         -0.0260            10            10            10            10

@chfast chfast added the optimization Performance optimization label Oct 13, 2020
@axic
Copy link
Member

axic commented Oct 13, 2020

I'd merge #554 first / instead of this.

@chfast
Copy link
Collaborator Author

chfast commented Oct 13, 2020

I'd merge #554 first / instead of this.

This is the only effective change in #554. Rest is done by the compiler optimization.

@chfast chfast mentioned this pull request Oct 20, 2020
Closed
axic
axic reviewed Oct 20, 2020
View reviewed changes
axic
axic reviewed Oct 20, 2020
View reviewed changes
@axic
Copy link
Member

axic commented Oct 20, 2020

I'd merge #554 first / instead of this.

This is the only effective change in #554. Rest is done by the compiler optimization.

The reason I like #554 because that makes it apparent those are two different code paths and we should optimise separately. We can wait for the std::function work first though.

@chfast
Copy link
Collaborator Author

chfast commented Oct 20, 2020

The reason I like #554 because that makes it apparent those are two different code paths and we should optimise separately. We can wait for the std::function work first though.

Inlining gives you the same effect. The commit 11a3ee4 does not change the assembly at all.

@axic
Copy link
Member

axic commented Oct 20, 2020

The reason I like #554 because that makes it apparent those are two different code paths and we should optimise separately. We can wait for the std::function work first though.

Inlining gives you the same effect. The commit 11a3ee4 does not change the assembly at all.

I totally understand that and didn't mean it will change anything, right now, but makes it apparent for us to deviate that code when looking at it.

@axic axic force-pushed the optimize_call_result_push branch from d5b745e to d30f8c6 Compare November 6, 2020 17:19
@axic
Copy link
Member

axic commented Nov 6, 2020

The question about #554 is moot now, since #616 was merged.

@axic axic mentioned this pull request Nov 6, 2020
Draft
axic
axic approved these changes Nov 6, 2020
View reviewed changes
Copy link
Member

@axic axic left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Would need to squash prior to merging.

@chfast
Copy link
Collaborator Author

chfast commented Nov 6, 2020 

fizzy/execute/blake2b/512_bytes_rounds_1_mean                     -0.0009         -0.0009           221           221           221           221
fizzy/execute/blake2b/512_bytes_rounds_16_mean                    +0.0059         +0.0058          3357          3377          3356          3376
fizzy/execute/ecpairing/onepoint_mean                             +0.0274         +0.0276       1172130       1204208       1171467       1203788
fizzy/execute/keccak256/512_bytes_rounds_1_mean                   +0.0330         +0.0327           265           273           265           273
fizzy/execute/keccak256/512_bytes_rounds_16_mean                  +0.0341         +0.0340          3901          4034          3901          4033
fizzy/execute/memset/256_bytes_mean                               +0.0473         +0.0470            20            21            20            21
fizzy/execute/memset/60000_bytes_mean                             +0.0501         +0.0501          4407          4628          4407          4627
fizzy/execute/mul256_opt0/input1_mean                             -0.0678         -0.0679            92            86            92            86
fizzy/execute/ramanujan_pi/33_runs_mean                           +0.0012         +0.0012           438           439           438           439
fizzy/execute/sha1/512_bytes_rounds_1_mean                        +0.0697         +0.0700           237           254           237           254
fizzy/execute/sha1/512_bytes_rounds_16_mean                       +0.0700         +0.0700          3308          3540          3308          3539
fizzy/execute/sha256/512_bytes_rounds_1_mean                      +0.1390         +0.1388           244           278           244           278
fizzy/execute/sha256/512_bytes_rounds_16_mean                     +0.1457         +0.1457          3369          3859          3368          3859
fizzy/execute/taylor_pi/pi_1000000_runs_mean                      -0.1457         -0.1457        118667        101382        118653        101369
fizzy/execute/micro/eli_interpreter/exec105_mean                  -0.0354         -0.0355            13            12            13            12
fizzy/execute/micro/factorial/20_mean                             -0.0095         -0.0098             1             1             1             1
fizzy/execute/micro/fibonacci/24_mean                             -0.1910         -0.1911         13832         11191         13831         11189
fizzy/execute/micro/host_adler32/1_mean                           +0.3003         +0.3002             0             0             0             0
fizzy/execute/micro/host_adler32/1000_mean                        +0.0337         +0.0335            58            60            58            60
fizzy/execute/micro/icall_hash/1000_steps_mean                    +0.0122         +0.0122           124           125           124           125
fizzy/execute/micro/spinner/1_mean                                +0.0286         +0.0283             0             0             0             0
fizzy/execute/micro/spinner/1000_mean                             +0.0343         +0.0347            17            18            17            18

axic
axic reviewed May 21, 2022
View reviewed changes
@axic axic force-pushed the optimize_call_result_push branch from c0abd7b to 71bcab2 Compare May 21, 2022 10:52
@axic
Copy link
Member

axic commented May 21, 2022

Rebased.

@axic axic requested a review from gumb0 May 21, 2022 10:52
gumb0
gumb0 approved these changes May 21, 2022
View reviewed changes
@axic axic force-pushed the optimize_call_result_push branch from 71bcab2 to c7d1c71 Compare May 21, 2022 12:16
@axic
Copy link
Member

axic commented May 21, 2022

@chfast merge or not merge?

@axic axic force-pushed the optimize_call_result_push branch from c7d1c71 to 62311e8 Compare May 23, 2022 08:22
@chfast chfast force-pushed the optimize_call_result_push branch from 62311e8 to 7dd0274 Compare May 23, 2022 21:35
@chfast
Copy link
Collaborator Author

chfast commented May 23, 2022

This is no good.

GCC 12

fizzy/execute/blake2b/512_bytes_rounds_1_mean                      +0.2228         +0.2228            67            82            67            82
fizzy/execute/blake2b/512_bytes_rounds_16_mean                     +0.2179         +0.2179          1011          1231          1011          1231
fizzy/execute/ecpairing/onepoint_mean                              +0.1895         +0.1895        327477        389546        327477        389545
fizzy/execute/keccak256/512_bytes_rounds_1_mean                    +0.2082         +0.2082            76            92            76            92
fizzy/execute/keccak256/512_bytes_rounds_16_mean                   +0.1316         +0.1316          1105          1250          1105          1250
fizzy/execute/memset/256_bytes_mean                                +0.0068         +0.0068             6             6             6             6
fizzy/execute/memset/60000_bytes_mean                              +0.0027         +0.0027          1279          1282          1279          1282
fizzy/execute/mul256_opt0/input1_mean                              -0.0010         -0.0010            25            25            25            25
fizzy/execute/ramanujan_pi/33_runs_mean                            +0.0076         +0.0076            98            99            98            99
fizzy/execute/sha1/512_bytes_rounds_1_mean                         +0.0285         +0.0286            74            76            74            76
fizzy/execute/sha1/512_bytes_rounds_16_mean                        +0.0314         +0.0314          1033          1065          1033          1065
fizzy/execute/sha256/512_bytes_rounds_1_mean                       +0.0059         +0.0059            73            73            73            73
fizzy/execute/sha256/512_bytes_rounds_16_mean                      +0.0126         +0.0126          1000          1012          1000          1012
fizzy/execute/taylor_pi/pi_1000000_runs_mean                       -0.0002         -0.0002         38223         38214         38223         38215
fizzy/execute/micro/eli_interpreter/exec105_mean                   +0.1523         +0.1523             4             5             4             5
fizzy/execute/micro/factorial/20_mean                              +0.0109         +0.0109             1             1             1             1
fizzy/execute/micro/fibonacci/24_mean                              +0.0437         +0.0437          4378          4570          4378          4570
fizzy/execute/micro/host_adler32/1_mean                            -0.0127         -0.0127             0             0             0             0
fizzy/execute/micro/host_adler32/1000_mean                         -0.0102         -0.0102            26            26            26            26
fizzy/execute/micro/icall_hash/1000_steps_mean                     +0.0098         +0.0098            60            61            60            61
fizzy/execute/micro/spinner/1_mean                                 +0.0789         +0.0789             0             0             0             0
fizzy/execute/micro/spinner/1000_mean                              +0.0441         +0.0441             9             9             9             9

clang 14

fizzy/execute/blake2b/512_bytes_rounds_1_mean                     +0.0447         +0.0447            68            72            68            72
fizzy/execute/blake2b/512_bytes_rounds_16_mean                    +0.0405         +0.0405          1035          1077          1035          1077
fizzy/execute/ecpairing/onepoint_mean                             +0.0598         +0.0598        337338        357516        337340        357517
fizzy/execute/keccak256/512_bytes_rounds_1_mean                   +0.0868         +0.0869            82            89            82            89
fizzy/execute/keccak256/512_bytes_rounds_16_mean                  +0.0960         +0.0960          1191          1305          1191          1305
fizzy/execute/memset/256_bytes_mean                               +0.0337         +0.0337             6             6             6             6
fizzy/execute/memset/60000_bytes_mean                             +0.0349         +0.0349          1258          1302          1258          1302
fizzy/execute/mul256_opt0/input1_mean                             +0.0013         +0.0013            23            23            23            23
fizzy/execute/ramanujan_pi/33_runs_mean                           +0.0665         +0.0665            95           102            95           102
fizzy/execute/sha1/512_bytes_rounds_1_mean                        +0.0486         +0.0486            74            78            74            78
fizzy/execute/sha1/512_bytes_rounds_16_mean                       +0.0449         +0.0449          1033          1079          1033          1079
fizzy/execute/sha256/512_bytes_rounds_1_mean                      +0.3413         +0.3413            71            95            71            95
fizzy/execute/sha256/512_bytes_rounds_16_mean                     +0.4002         +0.4002           976          1366           976          1366
fizzy/execute/taylor_pi/pi_1000000_runs_mean                      +0.0045         +0.0045         37441         37610         37442         37611
fizzy/execute/micro/eli_interpreter/exec105_mean                  +0.0661         +0.0661             4             4             4             4
fizzy/execute/micro/factorial/20_mean                             -0.0245         -0.0245             1             1             1             1
fizzy/execute/micro/fibonacci/24_mean                             +0.0001         +0.0001          4939          4940          4939          4940
fizzy/execute/micro/host_adler32/1_mean                           +0.0037         +0.0037             0             0             0             0
fizzy/execute/micro/host_adler32/1000_mean                        +0.0241         +0.0241            28            29            28            29
fizzy/execute/micro/icall_hash/1000_steps_mean                    +0.0531         +0.0531            64            67            64            67
fizzy/execute/micro/spinner/1_mean                                +0.0531         +0.0531             0             0             0             0
fizzy/execute/micro/spinner/1000_mean                             +0.0676         +0.0676             8             8             8             8
OVERALL_GEOMEAN                                                   +0.0661         +0.0661             0             0             0             0

@chfast chfast force-pushed the optimize_call_result_push branch from 7dd0274 to 90bd063 Compare May 24, 2022 12:37
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@axic axic axic approved these changes

@gumb0 gumb0 gumb0 approved these changes

Assignees

No one assigned

Labels

optimization Performance optimization

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@chfast @axic @gumb0