[PATCH] D20401: [Lexer] Don't merge macro args from different macro files

[Vedant Kumar via cfe-commits]

Yes, here are the results from the unpatched compiler:

Avg. wall time (s): 0.73241
Std. deviation: 0.05850

And here are the results from the patched compiler:

Avg. wall time (s): 0.75554
Std. deviation: 0.07492

The testing methodology was the same (100 trials), except I used `clang -x objective-c -Xclang -emit-pch`. There is a slight difference, but it still basically looks noisy to me.

vedant


> On May 19, 2016, at 2:48 PM, Argyrios Kyrtzidis <akyrtzi at gmail.com> wrote:
> 
> Could you also check performance of creating a PCH file out of Cocoa.h ?
> 
>> On May 19, 2016, at 1:47 PM, Vedant Kumar <vsk at apple.com> wrote:
>> 
>> vsk added a comment.
>> 
>> I discussed this bug with Argyrios off-list, who lgtm'd on the condition that it doesn't introduce a performance regression. He suggested preprocessing Cocoa.h to stress the patch. After running a stabilization script, I used this command to stress RelNoAsserts builds of clang both with and without this patch.
>> 
>> for I in $(seq 1 100); do time $CC -F /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX10.12.sdk/System/Library/Frameworks -E Cocoa.h -o /dev/null; done
>> 
>> The results are basically in the noise (link to raw data: https://ghostbin.com/paste/r6cyh):
>> 
>> | Compiler           | **Unpatched** TOT | **Patched** TOT |
>> | Avg. wall time (s) | 0.21709           | 0.21608         |
>> | Std. deviation     | 0.02101           | 0.02219         |
>> 
>> I also made sure that the preprocessed sources emitted by the two compilers are the same.
>> 
>> 
>> http://reviews.llvm.org/D20401
>> 
>> 
>> 
>