On 2012-12-14 17:12:05 -0800, Ronald S. Bultje wrote: > Hi, > > On Fri, Dec 14, 2012 at 8:56 AM, Janne Grunau <[email protected]> wrote: > > On 2012-12-14 08:32:14 -0800, Ronald S. Bultje wrote: > >> Hi, > >> > >> On Dec 14, 2012 3:30 AM, "Janne Grunau" <[email protected]> wrote: > >> > > >> > Arguments on the stack are handled properly and functions can use more > >> > than arguments than the 7 registers available on x86_32. > >> > Fixes nasm build, yasm would fail too if it would error out on %error > >> > instead of just emitting a warning. > >> > --- > >> > libavutil/x86/x86inc.asm | 1 - > >> > 1 file changed, 1 deletion(-) > >> > > >> > diff --git a/libavutil/x86/x86inc.asm b/libavutil/x86/x86inc.asm > >> > index 60d05f4..40705e0 100644 > >> > --- a/libavutil/x86/x86inc.asm > >> > +++ b/libavutil/x86/x86inc.asm > >> > @@ -554,7 +554,6 @@ DECLARE_ARG 7, 8, 9, 10, 11, 12, 13, 14 > >> > %endif > >> > SETUP_STACK_POINTER %4 > >> > ASSERT regs_used <= 7 > >> > - ASSERT regs_used >= num_args > >> > PUSH_IF_USED 3, 4, 5, 6 > >> > ALLOC_STACK %4 > >> > LOAD_IF_USED 0, 1, 2, 3, 4, 5, 6 > >> > >> Please submit to x264 first. > >> > >> Where does this trigger? This shouldn't happen, an assert is the right > >> thing to do. > > > > It is? make fate passes on x86_32 instances and yasm is affected too > > except that it emits just a warning on %error. > > > > It triggers on h264_loop_filter_strength_mmxext in h264_deblock.asm > > with 9 arguments. > > But we have code above that clips both num_regs and regs_used to 7, so > both should be 7?
if you mean num_args with num_regs then you removed the code to clamp it to 7 in 6f40e9f070f7 'x86inc: support stack mem allocation and re-alignment in PROLOGUE'. Re-adding that breaks several fate-h264 tests. Nasm also fails on '%1 %+ SUFFIX' with empty SUFFIX in cpuid.asm. It was previously working since it was always followed by ', %2'. Easiest solution I've found was adding parameters to cpu_cpuid_test in cpuid.asm. Janne _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
