[llvm-commits] [llvm] r154816 - in /llvm/trunk: lib/CodeGen/ test/CodeGen/ARM/ test/CodeGen/CellSPU/ test/CodeGen/Mips/ test/CodeGen/Thumb2/ test/CodeGen/X86/ test/CodeGen/XCore/ test/Transforms/LoopStrengthReduce/X86/

[Tobias Grosser]

On 04/18/2012 03:30 AM, Evan Cheng wrote:
>
> On Apr 16, 2012, at 6:49 AM, Chandler Carruth wrote:
>
>> Author: chandlerc
>> Date: Mon Apr 16 08:49:17 2012
>> New Revision: 154816
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=154816&view=rev
>> Log:
>> Flip the new block-placement pass to be on by default.
>>
>> This is mostly to test the waters. I'd like to get results from FNT
>> build bots and other bots running on non-x86 platforms.
>
> And how do the results look?

Here are some LNT results that only include 154812 and 154816 (both 
related commits of Chandler) as relevant commits:

http://llvm.org/perf/db_default/v4/nts/287

Some execution time changes up to 6.5% slower and 20% faster. In 
general, there is more code that became faster than slower. (Also the 
test base is still not too large).

Cheers
Tobi

P.S: Here the detailed changes inline

===========================
Run-Over-Run Changes Detail
===========================
Performance Regressions - Compile Time
--------------------------------------
   MultiSource/Applications/kimwitu++/kc: 1.03% (16.6838 => 16.8562, 
std. dev.: 0.0190)
   MultiSource/Applications/siod/siod: 1.02% (2.1895 => 2.2119, std. 
dev.: 0.0028)

Performance Regressions - Execution Time
----------------------------------------
   SingleSource/Benchmarks/Adobe-C++/stepanov_vector: 6.50% (2.5630 => 
2.7297, std. dev.: 0.0001) t
   SingleSource/Benchmarks/Adobe-C++/stepanov_abstraction: 4.78% (4.5946 
=> 4.8144, std. dev.: 0.0001)
   MultiSource/Benchmarks/Fhourstones-3_1/fhourstones3_1: 4.28% (1.6064 
=> 1.6752, std. dev.: 0.0002)
   MultiSource/Benchmarks/Fhourstones/fhourstones: 4.28% (1.7592 => 
1.8345, std. dev.: 0.0060)
   MultiSource/Benchmarks/ASC_Sequoia/CrystalMk/CrystalMk: 2.63% (7.9170 
=> 8.1250, std. dev.: 0.0075)
   MultiSource/Applications/hexxagon/hexxagon: 2.55% (12.1443 => 
12.4543, std. dev.: 0.0004)
   SingleSource/Benchmarks/Misc/ffbench: 2.25% (0.8272 => 0.8458, std. 
dev.: 0.0001)
   SingleSource/Benchmarks/BenchmarkGame/recursive: 1.33% (2.0187 => 
2.0456, std. dev.: 0.0003)
   SingleSource/Benchmarks/Shootout/sieve: 1.18% (5.1531 => 5.2137, std. 
dev.: 0.0004)

Performance Improvements - Execution Time
-----------------------------------------
   SingleSource/Benchmarks/Misc/richards_benchmark: -19.02% (1.2491 => 
1.0115, std. dev.: 0.0042)
   SingleSource/Benchmarks/McGill/chomp: -10.69% (1.2417 => 1.1090, std. 
dev.: 0.0063)
   MultiSource/Benchmarks/Ptrdist/yacr2/yacr2: -6.97% (1.0105 => 0.9401, 
std. dev.: 0.0013)
   MultiSource/Benchmarks/Olden/perimeter/perimeter: -6.36% (0.3131 => 
0.2932, std. dev.: 0.0001)
   SingleSource/Benchmarks/BenchmarkGame/nsieve-bits: -5.14% (1.3784 => 
1.3076, std. dev.: 0.0059)
   MultiSource/Benchmarks/BitBench/drop3/drop3: -4.67% (0.4366 => 
0.4162, std. dev.: 0.0009)
   MultiSource/Benchmarks/Olden/health/health: -3.54% (0.4599 => 0.4436, 
std. dev.: 0.0011)
   MultiSource/Benchmarks/Olden/power/power: -2.62% (1.5099 => 1.4703, 
std. dev.: 0.0000)
   MultiSource/Applications/minisat/minisat: -1.95% (9.3446 => 9.1626, 
std. dev.: 0.0060)
   SingleSource/Benchmarks/Shootout/lists: -1.83% (6.3426 => 6.2263, 
std. dev.: 0.0026)
   MultiSource/Applications/aha/aha: -1.68% (3.1727 => 3.1193, std. 
dev.: 0.0003)
   MultiSource/Benchmarks/VersaBench/8b10b/8b10b: -1.60% (6.6071 => 
6.5014, std. dev.: 0.0053)
   SingleSource/Benchmarks/Shootout/heapsort: -1.56% (3.9200 => 3.8587, 
std. dev.: 0.0038)
   SingleSource/Benchmarks/CoyoteBench/huffbench: -1.54% (20.3661 => 
20.0528, std. dev.: 0.0378)
   MultiSource/Benchmarks/MallocBench/cfrac/cfrac: -1.24% (2.3292 => 
2.3003, std. dev.: 0.0000)