On 22 September 2012 04:29, Christophe Lyon christophe.lyon@linaro.org wrote:
== Progress ==
Discussed big-endian patches for vext tests: careful review of the specification is required and this patch might actually expose GCC bugs in big-endian/Neon.
builtin_bswap16:
- Posted 2 implementations of the generic patch for (x<<8)|(x>>8). The 2nd one looks OK, but I was asked to make some checks on powerPC, which are on-going.
Matthias and Ulrich are my go-to guys for PowerPC if you need help.
backporting it to Linaro/4.7. On-going: merged obvious patches but one test does not produce the same code as on trunk.
validations in cbuild in with GCC configured for thumb and hard FP showed that some bswap16 tests fail because they force armv6, which leads to unsupported thumb1 configuration. Proposed testsuite modification to make this easier, under discussion with Matt & Richard.
== Next ==
- builtin_bswap16:
- check powerPC results and have the patch accepted upstream.
- measure probable regression in libav when using the builtin instead of an asm().
I've added this to today's agenda.
-- Michael