[Patch ARM] Fix ICE in neon shift patterns with constant ints.
Ramana Radhakrishnan
ramana.radhakrishnan@arm.com
Mon Aug 24 09:34:00 GMT 2009
Hi,
The vshl<vmode>, vlshr<vmode> and vashr<vmode> patterns have not been
defined for Neon and this causes an ICE in a well known embedded
benchmark using the options -O3 -mcpu=cortex-a8 -mfpu=neon
-mfloat-abi=softfp . The neon backend has ashl<vmode>,lshr<vmode>* and
ashr<vmode> patterns already defined. I've renamed them here to the
corresponding vector shift patterns and that fixes the ICE.
I've built and tested a cross toolchain for arm-none-eabi with cpu =
cortex-a8, fpu=neon and float-abi=softfp on qemu.
Ok for trunk ?
cheers
Ramana
2009-08-23 Ramana Radhakrishnan <ramana.radhakrishnan@arm.com>
* config/arm/neon.md (vashl<mode>3): Rename from ashl<mode>3.
(vashr<mode>3): Rename from ashr<mode>3.
(vlshr<mode>3): Rename from lshr<mode>3.
--
Ramana Radhakrishnan
GNU Tools
ARM Ltd.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: neon-ice-fix.patch
Type: text/x-patch
Size: 1104 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20090824/45449e81/attachment.bin>
More information about the Gcc-patches
mailing list