Strange bug with the -O2-for-C option

[Simon Peyton-Jones]

The consensus here seems to be

* O2-for-C has the sole effect of passing O2 to gcc
* maybe that exposes a bug in gcc
* or maybe it makes gcc produce output that the "evil mangler"
	doesn't recognise

Either way, we don't plan to pursue this, unless you (our trusty user
community) report that
	a) O2-for-C really is significant performance win
	b) preferably some tighter characterisation of the error

Thanks for reporting it though.  We'll put this on 'known-bugs' list.

Simon

| -----Original Message-----
| From: Alain Cremieux [mailto:alcremi at pobox.com] 
| Sent: 15 August 2001 18:48
| To: glasgow-haskell-bugs at haskell.org
| Subject: Strange bug with the -O2-for-C option
| 
| 
| Hi,
| I encountered a strange and potentially dangerous bug trying 
| to test optimization. I was trying to optimize the execution 
| time of a canny edge detection program coming from the nofib 
| suite. What is happening is not what I expected :
| 1) compiled with -O2, the image (edges) is correctly created
| 2) compiled with -O2 -O2-for-C, the image is completely 
| black. My environment : Linux with RedHat 7.1, GHC 5.00.2 
| compiled with itself I have a test set for anyone interested 
| in reproducing the bug, Alain
| 
| 
| _______________________________________________
| Glasgow-haskell-bugs mailing list 
| Glasgow-haskell-bugs at haskell.org 
| http://www.haskell.org/mailman/listinfo/glasgow-| haskell-bugs
|