This is the mail archive of the
binutils@sourceware.org
mailing list for the binutils project.
Re: [PATCH] .bundle_align_mode
- From: nick clifton <nickc at redhat dot com>
- To: Roland McGrath <mcgrathr at google dot com>
- Cc: binutils at sourceware dot org, David Sehr <sehr at google dot com>
- Date: Thu, 08 Mar 2012 17:39:46 +0000
- Subject: Re: [PATCH] .bundle_align_mode
- References: <x57jfwdtno8d.fsf@frobland.mtv.corp.google.com>
Hi Roland,
Ok for trunk?
Sorry for the delay in reviewing this patch.
The patch is fine apart from two problems:
1. You are modifying the style of the copyright notices at the start
of the files that you are patching. Whilst I have no issue with the
change itself, I feel that it would be better to have a completely
separate patch to update all of the copyright notices in one go, rather
than changing them piecemeal over time. Plus the change to the
copyright notices has nothing to do with the implementation of the
.bundle_align feature, so it really should not be included in the same
patch.
2. When I tested the patch locally I encountered a new failure in
the gas testsuite for an arm-eabi toolchain:
FAIL: ARM .bundle_lock
Looking in the gas.log the failure is this:
extra regexps in gas/testsuite/gas/arm/bundle-lock.d starting with "^
*2e0:\s+(be00\s+bkpt|1840\s+adds)\s+.+$"
EOF from dump.out
And the dump.out file for that region looks like this:
000002d0 <thumb_sequence_3_offset_12>:
2d0: be00 bkpt 0x0000
2d2: be00 bkpt 0x0000
2d4: be00 bkpt 0x0000
2d6: be00 bkpt 0x0000
2d8: be00 bkpt 0x0000
2da: be00 bkpt 0x0000
2dc: 1840 adds r0, r0, r1
2de: 1a40 subs r0, r0, r1
2e0: 1a40 subs r0, r0, r1
2e2: bf00 nop
2e4: f3af 8000 nop.w
2e8: f3af 8000 nop.w
2ec: f3af 8000 nop.w
Similarly an i686-pc-linux-gnu toolchain failed with:
extra regexps in gas/testsuite/gas/i386/bundle.d starting with "^
*22e0:\s+([0-9a-f][0-9a-f] )+\s+[a-z].+$"
EOF from dump.out
FAIL: i386 .bundle_align_mode
extra regexps in gas/testsuite/gas/i386/bundle-lock.d starting with
"^ *22e0:\s+(f4\s+hlt|f8\s+clc)\s*$"
EOF from dump.out
FAIL: i386 .bundle_lock
[If it matters this is on a 32-bit x86 host running Fedora 16]
I would prefer not to introduce new gas testsuite failures. Plus it
looks to me as if the failures are indicating the the .bundle_align
feature is not working for long sequences of instructions. Please could
you check this and determine if this is just my test environment being
funny or if something needs to be fixed.
Cheers
Nick