[PATCH] D16095: [Polly] Prepare unit tests for update to ISL 0.16 (WIP)

[Michael Kruse via llvm-commits]

Meinersbur added a comment.

In http://reviews.llvm.org/D16095#324467, @grosser wrote:

> the patch looks to me very sensible. I would be generally comfortable to commit something like this. What are the remaining changes you had in mind?


Removing additional lines in reduction_simple_iv_debug_wrapped_dependences.ll and sequential_loops.ll, re-add possibly interesting CHECK-NOT lines that are not covered by CHECK-NEXTs.

> Regarding the script, I think it would be great if the --check-include options could eventually (not necessary for this commit) be auto-derived.


Can you share your idea how this autodetection could work?

> Also, does --check-position=autodetect work after this initial formatting?


Yes, it looks for the position of the first CHECK, and the reformatted one with --check-position=autodetect will still be in the same position.

> If both would, the majority of the options in https://raw.githubusercontent.com/Meinersbur/polly/update_test/test/CMakeLists.txt would not be necessary any more, right?


For the moment I do not plan commit the CMakeLists.txt (appart from the configure_file) in its current form as it is too preliminary, would again destroy any manual edits and from our last phone call discussion it seemed there wouldn't be that much interest.

>   I think this is would be desirable.


Why?


================
Comment at: test/ScopInfo/remarks.ll:3
@@ -2,3 +2,1 @@
 ;
-; CHECK: remark: test/ScopInfo/remarks.c:4:7: SCoP begins here.
-; CHECK: remark: test/ScopInfo/remarks.c:8:5: Finite loop assumption: [M, N, Debug] -> {  : M >= 0 or (M <= -2 and N <= 0) or (M = -1 and N <= 0) }
----------------
grosser wrote:
> Why was this line dropped?
Good catch

The RUN line here emits binary data to FileCheck, so I added -analyze (alternatively -S would be possible, but there is another unit test also checking the remark which uses -analyze)

The -analyze output is printed to stdout while the remarks are printed to stderr. How they are combined depends on the receiving program (Terminal, FileCheck, or the python script). For some reason (buffering?) python combines the stdline without its \n, so it writes:
```
Printing analysis 'Polly - Create polyhedral description of Scops' for region: 'entry.split => if.remark: test/ScopInfo/remarks.c:4:7: SCoP begins here
```
This is not recognized as a "Remark"-line by the script. Does opt maybe have an option to suppress its output?


http://reviews.llvm.org/D16095