Commit Graph

11 Commits

Author SHA1 Message Date
Hugh Bellamy
f001b7562b Use relatively new LLVM_FALLLTHROUGH instead of our own SWIFT_FALLTHROUGH 2017-02-12 10:47:03 +07:00
practicalswift
6d1ae2a39c [gardening] 2016 → 2017 2017-01-06 16:41:22 +01:00
practicalswift
797b80765f [gardening] Use the correct base URL (https://swift.org) in references to the Swift website
Remove all references to the old non-TLS enabled base URL (http://swift.org)
2016-11-20 17:36:03 +01:00
Xin Tong
6d6a82f44b Remove code motion in ASO. This code has been disabled for sometime.
As promised, we separate the duty of moving retain release pairs with the
task of removing them. Now the task of moving retains and releases are in
Retain Release Code Motion committed in 51b1c0bc68.
2016-05-04 17:23:02 -07:00
Xin Tong
7d26a47200 Fix a non-deterministic retain/release insertion in RRCodeMotion 2016-04-19 22:14:38 -07:00
Xin Tong
c2e8c81227 Disable retain release code motion. There is some nondeterminism in
how we insert the new retain release instructions
2016-04-19 21:42:59 -07:00
Xin Tong
51b1c0bc68 Implement retain, release code motion.
Iterative data flow retain sinking and release hoisting.

This allows us to sink retains and hoist releases across harmless loops. which is
an improvement on the SILCodeMotion retain sinking and release hoisting.

It also separates the duty of moving retain and release with the duty of eliminating them
in ASO.

This should eventually replace RR code motion in SILcodemotion and insertion point
in ARCsequence opts (ASO).

This is the performance difference i get with retain sinking and release hoisting.
After disabling retain release code motion in ASO and SILCodeMotion. we can start to take
those code out once this lands.

I see that we go from 24.5% of time spent in SILOptimizations w.r.t. the whole stdlib compilation
to 25.1%.

Improvement is better (i.e. retain sinking and hoisting releases result in performance gain).

<details open>
  <summary>Regression (7)</summary>

TEST                                                    | OLD_MIN | NEW_MIN | DELTA (%) | SPEEDUP
---                                                     | ---     | ---     | ---       | ---
SetIsSubsetOf                                           | 441     | 510     | +15.7%    | **0.86x**
SetIntersect                                            | 1041    | 1197    | +15.0%    | **0.87x**
BenchLangCallingCFunction                               | 184     | 211     | +14.7%    | **0.87x**
Sim2DArray                                              | 326     | 372     | +14.1%    | **0.88x**
SetIsSubsetOf_OfObjects                                 | 498     | 567     | +13.9%    | **0.88x**
GeekbenchGEMM                                           | 945     | 1022    | +8.2%     | **0.92x**
COWTree                                                 | 3839    | 4181    | +8.9%     | **0.92x(?)**

</details>

<details >
  <summary>Improvement (31)</summary>

TEST                                                    | OLD_MIN | NEW_MIN | DELTA (%) | SPEEDUP
---                                                     | ---     | ---     | ---       | ---
ObjectiveCBridgeFromNSDictionaryAnyObjectToString       | 174526  | 165392  | -5.2%     | **1.06x**
RGBHistogram                                            | 3128    | 2957    | -5.5%     | **1.06x**
ObjectiveCBridgeToNSDictionary                          | 16510   | 15494   | -6.2%     | **1.07x**
LuhnAlgoLazy                                            | 2294    | 2120    | -7.6%     | **1.08x**
DictionarySwapOfObjects                                 | 6477    | 5994    | -7.5%     | **1.08x**
StringRemoveDupes                                       | 1610    | 1485    | -7.8%     | **1.08x**
ObjectiveCBridgeFromNSSetAnyObjectToString              | 159358  | 147824  | -7.2%     | **1.08x**
ObjectiveCBridgeToNSSet                                 | 16191   | 14924   | -7.8%     | **1.08x**
DictionaryHashableClass                                 | 1839    | 1704    | -7.3%     | **1.08x**
DictionaryLiteral                                       | 2906    | 2678    | -7.8%     | **1.09x(?)**
StringUtilsUnderscoreCase                               | 10031   | 9187    | -8.4%     | **1.09x**
LuhnAlgoEager                                           | 2320    | 2113    | -8.9%     | **1.10x**
ObjectiveCBridgeFromNSSetAnyObjectToStringForced        | 99553   | 90348   | -9.2%     | **1.10x**
RIPEMD                                                  | 3327    | 3009    | -9.6%     | **1.11x**
Combos                                                  | 595     | 538     | -9.6%     | **1.11x**
Roman                                                   | 10      | 9       | -10.0%    | **1.11x**
StringUtilsCamelCase                                    | 10783   | 9646    | -10.5%    | **1.12x**
SetIntersect_OfObjects                                  | 2511    | 2182    | -13.1%    | **1.15x**
SwiftStructuresTrie                                     | 28331   | 24339   | -14.1%    | **1.16x**
Dictionary2OfObjects                                    | 3748    | 3115    | -16.9%    | **1.20x**
DictionaryOfObjects                                     | 2473    | 2050    | -17.1%    | **1.21x**
Dictionary                                              | 894     | 737     | -17.6%    | **1.21x**
Dictionary2                                             | 2268    | 1859    | -18.0%    | **1.22x**
StringIteration                                         | 8027    | 6344    | -21.0%    | **1.27x**
Phonebook                                               | 8207    | 6436    | -21.6%    | **1.28x**
BenchLangArray                                          | 119     | 91      | -23.5%    | **1.31x**
LinkedList                                              | 8267    | 6297    | -23.8%    | **1.31x**
StrToInt                                                | 5585    | 4180    | -25.2%    | **1.34x**
Dictionary3OfObjects                                    | 1122    | 831     | -25.9%    | **1.35x**
Dictionary3                                             | 731     | 515     | -29.6%    | **1.42x**
SuperChars                                              | 513353  | 258735  | -49.6%    | **1.98x**
2016-04-18 15:39:17 -07:00
Xin Tong
d7dad7884f Add an option to disable code motion in ARC 2016-03-30 22:14:17 -07:00
Xin Tong
e24ce07eed Remove some unreachable code. Also add an assertion. NFC 2016-02-23 16:52:35 -08:00
practicalswift
2f547e9356 [gardening] Use consistent header structure in newly introduced files 2016-02-15 16:47:48 +01:00
Michael Gottesman
e152746277 [arc] Split GlobalARCPairingAnalysis into the matching set builder part and the
top level driver . Move the top level driver of the pairing analysis into
ARCSequenceOpts and have ARCSequenceOpts use ARCMatchingSetBuilder directly.

This patch is the first in a series of patches that improve ARC compile
time performance by ensuring that ARC only visits the full CFG at most
one time.

Previously when ARC was split into an analysis and a pass, the split in
the codebase occurred at the boundary in between ARCSequenceOpts and
ARCPairingAnalysis. I used a callback to allow ARCSequenceOpts to inject
code into ARCPairingAnalysis.

Now that the analysis has been moved together with the pass this
unnecessarily complicates the code. More importantly though it creates
obstacles towards reducing compile time by visiting the CFG only once.

Specifically, we need to visit the full cfg once to gather interesting
instructions. Then when performing the actual dataflow analysis, we only
visit the interesting instructions. This causes an interesting problem
since retains/releases can have dependencies on each other implying that
I need to be able to update where various "interesting instructions" are
located after ARC moves it. The "interesting instruction" information is
stored at the pairing analysis level, but the moving/removal of
instructions is injected in via the callback.

By moving the top level driver part of ARCPairingAnalysis into
ARCSequenceOpts, we simplify the code by eliminating the dependency
injection callback and also make it easier to manage the cached CFG
state in the face of the ARC optimizer moving/removing retains/releases.
2016-02-14 14:56:13 -08:00