[CompilerPerf] Using reflection to enhance #5112 and #5278 #5307

manofstick · 2018-07-08T03:12:51Z

Third in a series of PRs, each building on the other, sequentially merge-able without the subsequent ones - although ideally all of them would be merged.

This is realizing the discussion begun here.

It's moving the relevant parts of reflection.fs down into prim-types.fs, and then utilizing those functions to enhance the ability of the functions canUseDefaultComparer and canUseDefaultEqualityComparer to determine if PER = ER on the requested type - hence providing a more efficient comparer, especially in the value type case where it means that boxing can be avoided.

manofstick · 2018-07-09T09:04:15Z

Some information about this PR:

Minimal subset of reflect.fs moved to prim-types.fs for querying fsharp types (records, unions)
Function stubs left in reflect.fs call implementations in prim-types.fs
Removed any stubs in reflect.fs which were no longer required
Copied required reshapedreflection.fs functions to prim-types.fs (I'm hoping this will all go away as .netstandard 2.0 becomes standard?)
Used Reflection functions to query objects to determine if ER comparer could be used.
All project internal

Some information about moving the functions

No access to option<>, list<>, Seq, etc.
reflect.fs has what appeared to be optimizations for option<> and list<>. This code was left in the stub.
tryXXX functions whose signature was ...->option<X> where changed to ...->byref<X>->bool
(Seq|Array).filter was converted to Array.FindAll
(Seq|Array).select was converted to Array.ConvertAll
A reshaped Array.ConvertAll was required as that only was introduced in .netstandard 2.0
sortFreshArray didn't originally create a fresh array, and still doesn't, but I didn't rename it

manofstick · 2018-07-14T00:40:31Z

I've run the tests from #5112 & #5278 again after this PR. The following results are compared against those already optimized results reported in those PRs, rather than back to master.

I haven't rerun the old results, just used what was in the original PR.

Note that +/- 5% probably is just a fluctuation of the runs, I'm not doing this fully scientifically. The results should be significant enough that such "minor" performance changes are negligible.

Test 1

#5112 (comment)
https://gist.github.com/manofstick/2b5f11c4574f206ad27b

64-bit

test	#5112	#5307	Percent
custom dynamic	120.33	102	85%
custom structural	136.83	110.83	81%
custom default	123.67	106.83	86%
value dynamic	933.83	382.83	41%
value structural	501	438.5	88%
value default	465.67	410.33	88%
gen value dynamic	1273.67	415.33	33%
gen value structural	838	578	69%
gen value default	572	399.33	70%
ref dynamic	1002.67	664.33	66%
ref structural	740	658.33	89%
ref default	783.5	712.5	91%
gen ref dynamic	1469.83	733.67	50%
gen ref structural	1135.83	878.33	77%
gen ref default	1011.67	764.67	76%
tuple dynamic	593.17	528.5	89%
tuple structural	251.33	241.33	96%
tuple default	706.83	664.5	94%
value tuple dynamic	149.5	126.5	85%
value tuple structural	150.67	127	84%
value tuple default	149.33	128.17	86%

32-bit

test	#5112	#5307	Percent
custom dynamic	113.17	94.67	84%
custom structural	277.17	185.17	67%
custom default	112.67	100.17	89%
value dynamic	2365.5	400.17	17%
value structural	577.67	502.83	87%
value default	481.67	413.67	86%
gen value dynamic	3186.83	511.67	16%
gen value structural	1702	1275	75%
gen value default	603.33	521.67	86%
ref dynamic	2289	654.67	29%
ref structural	722	656.17	91%
ref default	774.5	688.33	89%
gen ref dynamic	3055.83	849.33	28%
gen ref structural	2231.5	1798.33	81%
gen ref default	1079.33	887	82%
tuple dynamic	1208.67	949.83	79%
tuple structural	261	240.67	92%
tuple default	792.33	743.83	94%
value tuple dynamic	152.67	134.17	88%
value tuple structural	153.17	132.83	87%
value tuple default	154.33	134.5	87%

manofstick · 2018-07-14T01:01:22Z

Test 2

#5112 (comment)
https://gist.github.com/manofstick/847d965aff5c1de360cf63d66e8a53ed

32-bit

test	#5112	#5307	Percent
seqGroupBy	674	510	76%
seqCountBy	676	456	67%
listGroupBy	716	252	35%
listCountBy	457	211	46%
arrayCountBy	304	151	50%
arrayGroupBy	217	126	58%
arrayCountBy	384	304	79%

64-bit

test	#5112	#5307	Percent
seqGroupBy	986	569	58%
seqCountBy	1296	559	43%
listGroupBy	758	214	28%
listCountBy	949	212	22%
arrayCountBy	818	162	20%
arrayGroupBy	432	121	28%
arrayCountBy	525	301	57%

forki · 2018-07-14T05:53:50Z

Wow. So it's ready?

manofstick · 2018-07-14T06:17:54Z

Test 3

#5112 (comment)
https://gist.github.com/manofstick/9b06efefe1396808611cf55e812c0ea9

test	#5112	#5307	Percent
32-bit	5512	5628	102%
64-bit	4822	3970	82%

manofstick · 2018-07-14T06:22:50Z

Test 4

#5112 (comment)
https://gist.github.com/manofstick/42bfe5f1a888480af3d83190fe364e85

test	#5112	#5307	Percent
32-bit non-generic	2664	2488	93%
32-bit generic	3083	4098	133%
64-bit non-generic	2639	2449	93%
64-bit generic	2891	2658	92%

manofstick · 2018-07-14T06:26:14Z

Test 5

#5112 (comment)
https://gist.github.com/manofstick/d54cf0e11265e10fa0480f18c3dea8a5

test	#5112	#5307	Percent
32-bit	288	341	118%
64-bit	301	279	93%

manofstick · 2018-07-14T06:30:05Z

Test 6

#5112 (comment)
https://gist.github.com/manofstick/b7544e94ebae75ff0c29290056512558

test	#5112	#5307	Percent
32-bit	915	365	40%
64-bit	710	291	41%

manofstick · 2018-07-14T06:40:18Z

@forki

Yeah, I'm happy with it, so any code reviews now would be appreciated. I'll just finish posting my test times and then remove the WIP in the name...

Oh, and just as I got to test 6 I realized that I was comparing my checks to times from the first cut I did which was removing tailcalls. But it's not too far off - but does explain a couple of anomalies... I should have been referring back to #5112 (comment) - but doesn't totally invalidate I dont' think... just painful posting as I haven't automated this... Grrr.. But just slog through.

manofstick · 2018-07-14T06:54:12Z

Test 7

#5112 (comment)
https://gist.github.com/manofstick/46ae43f5969cc1bb4c125509703299d5

32-bit

test	#5112	#5307	Percent
int	2439	2296	94%
float	3795	3482	92%
struct int*int	3079	2788	91%
struct int*int64	3200	2883	90%
struct int*float	8216	7517	91%
struct int64int64int64int64int64int64int64*int64	5498	5272	96%
struct floatfloatfloatfloatfloatfloatfloat*float	22137	20456	92%
int*int	8348	7742	93%
int*float	8042	7454	93%
int64int64int64int64int64int64int64*int64	21023	19362	92%
floatfloatfloatfloatfloatfloatfloat*float	18985	17860	94%
TestRecordData	4542	4188	92%
TestGenericRecordData	6482	6000	93%
TestUnionData	5094	4788	94%
TestGenericUnionData	5739	5203	91%

64-bit

test	#5112	#5307	Percent
int	671	615	92%
float	926	917	99%
struct int*int	1230	1040	85%
struct int*int64	18486	16990	92%
struct int*float	21521	19824	92%
struct int64int64int64int64int64int64int64*int64	19626	18585	95%
struct floatfloatfloatfloatfloatfloatfloat*float	28992	25770	89%
int*int	3021	2914	96%
int*float	3008	2988	99%
int64int64int64int64int64int64int64*int64	8928	8322	93%
floatfloatfloatfloatfloatfloatfloat*float	8600	10502	122%
TestRecordData	1317	1203	91%
TestGenericRecordData	1495	1379	92%
TestUnionData	1907	1557	82%
TestGenericUnionData	1991	1693	85%

manofstick · 2018-07-14T20:49:41Z

Test 8

#5278 (comment)
https://gist.github.com/manofstick/33fefb8ac6b60d401375be738b521ac3

32-bit

test	#5278	#5307	Percent
custom dynamic	1192.33	1239.67	104%
custom structural	3798.5	3866.00	102%
custom default	1208.83	1244.50	103%
value dynamic	3648.83	1301.00	36%
value structural	1251.33	1288.83	103%
value default	1259.83	1303.83	103%
gen value dynamic	6444.83	3659.67	57%
gen value structural	3522.83	3697.17	105%
gen value default	3560.67	3714.67	104%
ref dynamic	3766.83	1403.00	37%
ref structural	1281.67	1302.67	102%
ref default	1438.83	1469.17	102%
gen ref dynamic	6484.83	3991.83	62%
gen ref structural	4804.17	4934.33	103%
gen ref default	3961.5	4091.17	103%
tuple dynamic	11176.33	11145.33	100%
tuple structural	1256.67	1297.33	103%
tuple default	8640.83	9091.50	105%
value tuple dynamic	1348	1346.83	100%
value tuple structural	1336.83	1342.17	100%
value tuple default	1349.5	1350.33	100%

64-bit

test	#5112	#5307	Percent
custom dynamic	1193.17	1204.50	101%
custom structural	1380.33	1410.67	102%
custom default	1197.33	1224.50	102%
value dynamic	2187.17	1269.67	58%
value structural	1240	1279.50	103%
value default	1249	1270.33	102%
gen value dynamic	2466.5	1521.50	62%
gen value structural	1503.33	1551.00	103%
gen value default	1524.83	1570.33	103%
ref dynamic	1749.67	1369.17	78%
ref structural	1262.83	1262.00	100%
ref default	1361.67	1410.33	104%
gen ref dynamic	2026.17	1665.33	82%
gen ref structural	1572.33	1511.00	96%
gen ref default	1673.5	1710.00	102%
tuple dynamic	4430	4650.33	105%
tuple structural	1070.33	1118.83	105%
tuple default	7844.33	8283.17	106%
value tuple dynamic	1570.17	1637.33	104%
value tuple structural	1558.83	1636.83	105%
value tuple default	1535.17	1633.17	106%

manofstick · 2018-07-14T20:52:37Z

Test 9

#5278 (comment)
https://gist.github.com/manofstick/04ba0c70c398bf5edeaba9b17d0b17c5

test	#5278	#5307	Percent
mapTest 32-bit	3356	3248	97%
mapTest 64-bit	4441	2815	63%

manofstick · 2018-07-14T20:54:52Z

Test 10

#5278 (comment)
https://gist.github.com/manofstick/bd374436a6d40218b2d836db34c494ca

test	#5278	#5307	Percent
32-bit	74615	70751	95%
64-bit	68278	65026	95%

manofstick · 2018-07-14T20:57:20Z

Test 11

#5278 (comment)
https://gist.github.com/manofstick/258edf7e7d39a76e9d1cdf0316ae2b47

test	#5278	#5307	Percent
non-generic 32-bit	24545	11420	47%
generic 32-bit	46421	26119	56%
non-generic 64-bit	18227	12071	66%
generic 64-bit	22051	14137	64%

manofstick · 2018-07-14T22:08:43Z

Test 12

#5278 (comment)
https://gist.github.com/manofstick/847922bcce2e2f47d3eca033ed9dc068

test	#5278	#5307	Percent
32-bit	363	350	96%
64-bit	306.8	287	94%

manofstick · 2018-07-14T22:11:16Z

Test 13

#5278 (comment)
https://gist.github.com/manofstick/b4fa5acc1abe3ca73a77a7eea12d205a

test	#5278	#5307	Percent
32-bit	575.8	558	97%
64-bit	498.8	509	102%

manofstick · 2018-07-14T22:14:38Z

Done.

Remember that the times presented here are relative to already optimized #5278 and #5112 !

This PR is good to go! (well code review, etc.)

manofstick · 2018-07-14T22:42:00Z

... and one final test ...

Five years ago I posted a response on stackoverflow where I compared various key types for Dictionary<_,_>, dict and Map. I have slightly modified the code in this gist (added ValueTuple and inlined some functions) but quite interesting to see how far FSharp.Core has come! (I have done some of this by removing layers in value types in dict and now this PR, but others have improved Map - maybe it's just general compiler improvements? I'm not sure...)

So below are the results! From the original posting:

Using .net's System.Collections.Generic.Dictionary
checksum -55.339450 elapsed 874/562 (KeyRecord)
checksum -55.339450 elapsed 1251/898 (KeyGenericRecord`1)
checksum -55.339450 elapsed 569/1024 (KeyStruct)
checksum -55.339450 elapsed 740/1427 (KeyGenericStruct`1)
checksum -55.339450 elapsed 2497/2218 (Tuple`3)
Using f# 'dict'
checksum -55.339450 elapsed 979/628 (KeyRecord)
checksum -55.339450 elapsed 1614/1206 (KeyGenericRecord`1)
checksum -55.339450 elapsed 3237/5625 (KeyStruct)
checksum -55.339450 elapsed 3290/5626 (KeyGenericStruct`1)
checksum -55.339450 elapsed 2448/1914 (Tuple`3)
Using f# 'Map'
checksum -55.339450 elapsed 8453/2638 (KeyRecord)
checksum -55.339450 elapsed 31301/25441 (KeyGenericRecord`1)
checksum -55.339450 elapsed 30956/26931 (KeyStruct)
checksum -55.339450 elapsed 53699/49274 (KeyGenericStruct`1)
checksum -55.339450 elapsed 32203/25274 (Tuple`3)
Using custom array
checksum -55.339450 elapsed 484/160 (Tuple`3)

(read the results as n/m where n is the time to create the container, m is the time to access the container)

Unfortunately I didn't record if this was the 32-bit or 64-bit build

The less important thing here is not so much the absolute times (as I don't remember what the hardware was) but rather using the Dictionary with KeyRecord has a baseline from which the other numbers can be interpreted.

The results from a current run (prior to this PR) are:

32-bit

Using .net's System.Collections.Generic.Dictionary
checksum -55.339450 elapsed 634/444 (KeyRecord)
checksum -55.339450 elapsed 1001/829 (KeyGenericRecord`1)
checksum -55.339450 elapsed 380/283 (KeyStruct)
checksum -55.339450 elapsed 543/669 (KeyGenericStruct`1)
checksum -55.339450 elapsed 1700/1529 (Tuple`3)
checksum -55.339450 elapsed 597/539 (ValueTuple`3)
Using f# 'dict'
checksum -55.339450 elapsed 750/608 (KeyRecord)
checksum -55.339450 elapsed 1095/1185 (KeyGenericRecord`1)
checksum -55.339450 elapsed 526/565 (KeyStruct)
checksum -55.339450 elapsed 760/1121 (KeyGenericStruct`1)
checksum -55.339450 elapsed 1987/1922 (Tuple`3)
checksum -55.339450 elapsed 1500/1743 (ValueTuple`3)
Using f# 'Map'
checksum -55.339450 elapsed 6720/3540 (KeyRecord)
checksum -55.339450 elapsed 11355/7586 (KeyGenericRecord`1)
checksum -55.339450 elapsed 9682/3243 (KeyStruct)
checksum -55.339450 elapsed 10804/7364 (KeyGenericStruct`1)
checksum -55.339450 elapsed 10695/6700 (Tuple`3)
checksum -55.339450 elapsed 13516/6492 (ValueTuple`3)
Using custom array
checksum -55.339450 elapsed 272/3857 (Tuple`3)

64-bit

Using .net's System.Collections.Generic.Dictionary
checksum -55.339450 elapsed 591/434 (KeyRecord)
checksum -55.339450 elapsed 864/629 (KeyGenericRecord`1)
checksum -55.339450 elapsed 706/568 (KeyStruct)
checksum -55.339450 elapsed 799/811 (KeyGenericStruct`1)
checksum -55.339450 elapsed 1619/1585 (Tuple`3)
checksum -55.339450 elapsed 941/857 (ValueTuple`3)
Using f# 'dict'
checksum -55.339450 elapsed 575/526 (KeyRecord)
checksum -55.339450 elapsed 784/695 (KeyGenericRecord`1)
checksum -55.339450 elapsed 1381/1679 (KeyStruct)
checksum -55.339450 elapsed 1515/1912 (KeyGenericStruct`1)
checksum -55.339450 elapsed 1450/1338 (Tuple`3)
checksum -55.339450 elapsed 3129/3510 (ValueTuple`3)
Using f# 'Map'
checksum -55.339450 elapsed 5180/1748 (KeyRecord)
checksum -55.339450 elapsed 7364/3298 (KeyGenericRecord`1)
checksum -55.339450 elapsed 17353/13944 (KeyStruct)
checksum -55.339450 elapsed 19098/15394 (KeyGenericStruct`1)
checksum -55.339450 elapsed 7422/3325 (Tuple`3)
checksum -55.339450 elapsed 19207/15323 (ValueTuple`3)
Using custom array
checksum -55.339450 elapsed 274/99 (Tuple`3)

And post this PR are:

32-bit

Using .net's System.Collections.Generic.Dictionary
checksum -55.339450 elapsed 601/430 (KeyRecord)
checksum -55.339450 elapsed 741/553 (KeyGenericRecord`1)
checksum -55.339450 elapsed 359/284 (KeyStruct)
checksum -55.339450 elapsed 437/389 (KeyGenericStruct`1)
checksum -55.339450 elapsed 1772/1526 (Tuple`3)
checksum -55.339450 elapsed 590/527 (ValueTuple`3)
Using f# 'dict'
checksum -55.339450 elapsed 610/451 (KeyRecord)
checksum -55.339450 elapsed 711/570 (KeyGenericRecord`1)
checksum -55.339450 elapsed 358/327 (KeyStruct)
checksum -55.339450 elapsed 436/431 (KeyGenericStruct`1)
checksum -55.339450 elapsed 1866/1808 (Tuple`3)
checksum -55.339450 elapsed 556/540 (ValueTuple`3)
Using f# 'Map'
checksum -55.339450 elapsed 4299/1155 (KeyRecord)
checksum -55.339450 elapsed 5758/2476 (KeyGenericRecord`1)
checksum -55.339450 elapsed 6775/794 (KeyStruct)
checksum -55.339450 elapsed 4690/2081 (KeyGenericStruct`1)
checksum -55.339450 elapsed 12744/5136 (Tuple`3)
checksum -55.339450 elapsed 7207/943 (ValueTuple`3)
Using custom array
checksum -55.339450 elapsed 265/121 (Tuple`3)

64-bit

Using .net's System.Collections.Generic.Dictionary
checksum -55.339450 elapsed 569/425 (KeyRecord)
checksum -55.339450 elapsed 601/467 (KeyGenericRecord`1)
checksum -55.339450 elapsed 672/544 (KeyStruct)
checksum -55.339450 elapsed 698/592 (KeyGenericStruct`1)
checksum -55.339450 elapsed 1578/1538 (Tuple`3)
checksum -55.339450 elapsed 911/840 (ValueTuple`3)
Using f# 'dict'
checksum -55.339450 elapsed 533/447 (KeyRecord)
checksum -55.339450 elapsed 556/482 (KeyGenericRecord`1)
checksum -55.339450 elapsed 669/575 (KeyStruct)
checksum -55.339450 elapsed 699/624 (KeyGenericStruct`1)
checksum -55.339450 elapsed 1399/1274 (Tuple`3)
checksum -55.339450 elapsed 911/862 (ValueTuple`3)
Using f# 'Map'
checksum -55.339450 elapsed 4603/1203 (KeyRecord)
checksum -55.339450 elapsed 5409/1887 (KeyGenericRecord`1)
checksum -55.339450 elapsed 4323/1111 (KeyStruct)
checksum -55.339450 elapsed 4950/1585 (KeyGenericStruct`1)
checksum -55.339450 elapsed 7014/3007 (Tuple`3)
checksum -55.339450 elapsed 5078/1281 (ValueTuple`3)
Using custom array
checksum -55.339450 elapsed 263/88 (Tuple`3)

So finally we see that dict almost add no overhead, and Map is now quite competitive!

Time for me to enjoy my Sunday.

dsyme · 2018-07-16T20:15:52Z

Amazing work!

…ized GetHashCode)

…(+ comments)

…or EqualityComparer

… unnecessary

…sed where a stable one was required

KevinRansom · 2020-05-28T07:02:27Z

@dsyme , @manofstick --- guys, is this orphaned, should I close it? I am on a mission to reduce the PR's in the repo, and an amazing number haven't been worked on in over a year. I think the techniques in this look useful?

But if it's not going to be productized I would like to close it, what do you think?

Thanks

Kevin

manofstick · 2020-05-28T07:36:13Z

@KevinRansom,

I will pick these up again if anyone was going to do anything with them. They were complete, but obviously code base has moved on.

So I'll just go and close 'em all, and if anyone decides that they are worth anything then they can poke me with a stick.

manofstick changed the title ~~Using reflection to enhance #5112 and #5278~~ [WIP] Using reflection to enhance #5112 and #5278 Jul 8, 2018

This was referenced Jul 9, 2018

[RFC Needed] Allow creation of Map using custom comparer #1770

Closed

Equality operator causes boxing on value types #526

Closed

Make typedefof more efficient #5019

Open

manofstick mentioned this pull request Jul 9, 2018

[CompilerPerf] Faster equality in generic contexts #5112

Closed

manofstick force-pushed the nobox_reflection branch from df382a4 to 69f8092 Compare July 11, 2018 07:09

manofstick mentioned this pull request Jul 13, 2018

[CompilerPerf] Extends #5112 ideas to Comparers (i.e >, >=, <, <= and LanguagePrimitives.FastGenericComparer) #5278

Closed

manofstick changed the title ~~[WIP] Using reflection to enhance #5112 and #5278~~ [CompilerPerf] Using reflection to enhance #5112 and #5278 Jul 14, 2018

dsyme added the Candidate for F# vNext label Jul 16, 2018

manofstick force-pushed the nobox_reflection branch 2 times, most recently from 080801c to 9e75253 Compare July 19, 2018 05:52

manofstick added 18 commits August 7, 2018 19:27

Replaced duplicated GetHashCode code with inline function

26ed397

Common array EqualityComparers (for types that previously had special…

ad78ec0

…ized GetHashCode)

Fix regression in regards to hash code on fast-path covariant arrays …

2cc05f0

…(+ comments)

Similar treatment within prim-types for Comparer as dotnet#5112 was f…

858cdf3

…or EqualityComparer

Moved throwing exceptions closer

fb5427c

Added comparison usage, so could remove exception catching logic when…

b8f5ac8

… unnecessary

Consolidated functions

780f303

Consolidated ArrayComparisons and addressed issue dotnet#5263

e3c1869

Fixed bug introduced in dotnet#5278 where an unstable sort could be u…

508111f

…sed where a stable one was required

Removed now unnecessary optimization from stableSortWithKeysAndComparer

4090b2b

Moved isRecordType down to prim-types.fs (internal only)

291dd75

Moved enough of reflect.fs to get Record fields

ca7e290

Moved reflection functions for Union types into prim-types.fs

f3f9aa3

Removed methods from prim-types.fsi that are now just used internally

e38a1e3

Used the reflection functions to handle records and unions

0492980

Added check for fsharp value types

bf784e5

Fixed struct check where a [<DefaultValue>] field is used.

ccd2156

Minor shift in logic order for less comparisons

3a4f0b7

manofstick force-pushed the nobox_reflection branch from 9e75253 to 3a4f0b7 Compare August 7, 2018 09:29

TIHan mentioned this pull request Dec 21, 2018

Ranges allocate 670 MB in 94 seconds of normal IDE usage in service.fs only #6047

Closed

manofstick mentioned this pull request Feb 10, 2019

Boxing occurs when hashing a range - 200+ MB over a 90 seconds of normal IDE usage #6052

Open

cartermp modified the milestones: 16.0, Unknown Feb 21, 2019

manofstick closed this May 28, 2020

manofstick mentioned this pull request May 31, 2020

The performance of Comparing and Ordering things #9348

Open

manofstick mentioned this pull request Jun 8, 2020

Avoid boxing on equality and comparisons #9404

Closed

cartermp removed this from the Unknown / not bug milestone Aug 27, 2020

vzarytovskii mentioned this pull request Oct 16, 2023

Resurrect compiler perf improvements PRs #16125

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[CompilerPerf] Using reflection to enhance #5112 and #5278 #5307

[CompilerPerf] Using reflection to enhance #5112 and #5278 #5307

manofstick commented Jul 8, 2018

manofstick commented Jul 9, 2018

manofstick commented Jul 14, 2018

manofstick commented Jul 14, 2018 •

edited

Loading

forki commented Jul 14, 2018

manofstick commented Jul 14, 2018

manofstick commented Jul 14, 2018

manofstick commented Jul 14, 2018

manofstick commented Jul 14, 2018

manofstick commented Jul 14, 2018

manofstick commented Jul 14, 2018

manofstick commented Jul 14, 2018 •

edited

Loading

manofstick commented Jul 14, 2018 •

edited

Loading

manofstick commented Jul 14, 2018 •

edited

Loading

manofstick commented Jul 14, 2018 •

edited

Loading

manofstick commented Jul 14, 2018 •

edited

Loading

manofstick commented Jul 14, 2018 •

edited

Loading

manofstick commented Jul 14, 2018

manofstick commented Jul 14, 2018 •

edited

Loading

dsyme commented Jul 16, 2018

KevinRansom commented May 28, 2020 •

edited

Loading

manofstick commented May 28, 2020

[CompilerPerf] Using reflection to enhance #5112 and #5278 #5307

[CompilerPerf] Using reflection to enhance #5112 and #5278 #5307

Conversation

manofstick commented Jul 8, 2018

manofstick commented Jul 9, 2018

manofstick commented Jul 14, 2018

Test 1

manofstick commented Jul 14, 2018 • edited Loading

Test 2

forki commented Jul 14, 2018

manofstick commented Jul 14, 2018

Test 3

manofstick commented Jul 14, 2018

Test 4

manofstick commented Jul 14, 2018

Test 5

manofstick commented Jul 14, 2018

Test 6

manofstick commented Jul 14, 2018

manofstick commented Jul 14, 2018

Test 7

manofstick commented Jul 14, 2018 • edited Loading

Test 8

manofstick commented Jul 14, 2018 • edited Loading

Test 9

manofstick commented Jul 14, 2018 • edited Loading

Test 10

manofstick commented Jul 14, 2018 • edited Loading

Test 11

manofstick commented Jul 14, 2018 • edited Loading

Test 12

manofstick commented Jul 14, 2018 • edited Loading

Test 13

manofstick commented Jul 14, 2018

manofstick commented Jul 14, 2018 • edited Loading

dsyme commented Jul 16, 2018

KevinRansom commented May 28, 2020 • edited Loading

manofstick commented May 28, 2020

manofstick commented Jul 14, 2018 •

edited

Loading

manofstick commented Jul 14, 2018 •

edited

Loading

manofstick commented Jul 14, 2018 •

edited

Loading

manofstick commented Jul 14, 2018 •

edited

Loading

manofstick commented Jul 14, 2018 •

edited

Loading

manofstick commented Jul 14, 2018 •

edited

Loading

manofstick commented Jul 14, 2018 •

edited

Loading

manofstick commented Jul 14, 2018 •

edited

Loading

KevinRansom commented May 28, 2020 •

edited

Loading