20090908 * Fix 64-bit shifts by a constant, limiting it to 31 bits left or 32 right * Disable int->float instructions to work round a GCC bug 20090809 * Patches ported to gcc-4.2.4. * Update 4.3 to 4.3.4 release * Extra fix to hardware bug "ldr rN, foo; cfldrd mvdX, [rN, #bar]" when ldr is conditional. It used to bite vorbis/lib/vorbisenc.c 20090327 * Correct some Cirrus "type" attributes * Remove code to handle reversed PLUS in cirrus_reorg() (it "can't happen") 20090322 * Remove the optimization of omitting a single nop after "branch;non_cirrus" when length(non_cirrus)>4 because length attribute is a maximum: some instructions declare a length of 8 but can then generate 4 (e.g. movdf). 20090319 * Update from gcc-4.3.2 to gcc-4.3.3 * change package name from armv4tl-crunch-linux-gnueabi-gcc to gcc-4.3-crunch (installing the new-style package will replace the old one properly) 20090310 * Always enable revision D1-E2 workarounds, drop -mcirrus-fix-invalid-insns flag and remove bogus D0 workarounds. * Added code to avoid a new hardware bug found by FFTW testsuite (ldr rN, x; cfstrd mvdX,[rN]) and rewrite the workaround code. * Disable all 64-bit instructions to avoid bugs: 64-bit shifts of more than 31 places give junk results and unidentified hardware bugs break the OpenSSL testsuite. They can be re-enabled with the new -mcirrus-di flag. 20090216 * Fixed workarounds for buggy instruction sequences and always enable workarounds for bugs present in all silicon. -mcirrus-fix-invalid-insns now just adds the extra workarounds for bugs in revision d0 silicon, as the manual always suggested it did. 20081211 * Added -mieee flag to give full IEEE precision at about half speed * Applied patch to fix PR target/37668 20081202 * Enabled cfnegs instruction ((float)a = -a)