Herman ten Brugge via Tinycc-devel wrote:
I did test this on arm32 openbsd. This is where the problem was.
However the above patch fails. I will provide a new patch.
I need to add support float/double/long double/pointers for this to work.
I would suggest just to forget it. Nobody will ever need this.
If you want to do something then maybe rather try to get rid of
the ugly "sec->relocplt". Instead you could use plt->reloc which
will get the correct section name "rel(a).plt" automatically and
hopefully will allow to get rid of some other related extra hacks too.
Btw, do we really need all these gen_increment_tcov ()'s?
Can't we just use the inc(0, TOK_INC) for all targets?
I did some measurements.
Compile tcc with itself 100 times with test coverage. (times in seconds):
function TOK_INC slowdown
x86_64 11.28 14.09 25%
i386 14.80 18.08 22%
arm32 233.53 415.94 78%
arm64 83.53 104.47 25%
riscv64 26.70 28.48 7% (qemu 10 times)
I did this because I saw a large slowdown.
The risc target is measured in qemu so not that reliable.
But why needs it to be fast? How can one make better conclusions
from studying a tcov data file if one conld open it 0.04 seconds
earlier?
-- gr
Herman
_______________________________________________
Tinycc-devel mailing list
Tinycc-devel@nongnu.org
https://lists.nongnu.org/mailman/listinfo/tinycc-devel
