[PATCH, rs6000] Undo part of October 23rd patch for PR target/41787

[Michael Meissner <meissner at linux dot vnet dot ibm dot com>]

On October 23rd, I commited two separate changes for PR target/41787.  The
patch to rs6000.h does still need to be kept, but I would like to undo the
patch to rs6000.c and the test that I added for this.  When I wrote the patch,
I was under the impression that we needed to set VRSAVE to non-zero whenever we
were using vector registers so that the kernel context swap code would save
these registers.  However, that isn't the case, and always setting VRSAVE will
slow down leaf functions that don't preserve vector registers (and need to set
VRSAVE), since the move to/from vrsave register instructions are serializing.

I have checked this with dual bootstraps and make checks.  There are no
regressions with this patch, and test gcc.c-torture/execute/pr33669.c now works
again.  Is this ok to check in?

-- 
Michael Meissner, IBM
4 Technology Place Drive, MS 2203A, Westford, MA, 01886, USA
meissner@linux.vnet.ibm.com

Attachment: gcc-power7.patch91
Description: Text document