Qi Wang
Qi Wang
Can you share the perf report output? High futex cost isn't expected. Also malloc_stats would help too. You can ask jemalloc to output the stats at the end of the...
Thanks for reporting this @fxcoudert . The assert msg is saying that, we expect an invalid alignment to result in failed allocation, which didn't seem to happen. Could you help...
To enable logging, you will need to configure with `--enable-log`, I'd suggesting adding `--enable-debug` as well. Then invoke that unit test with the MALLOC_CONF env var: `MALLOC_CONF="log:." test/integration/aligned_alloc` the beginning...
@fxcoudert it looks all the invalid ones are not going through jemalloc; I wonder if the new compiler is more aware of the syntax and optimized things away? Can you...
> : Number of CPUs detected is not deterministic. This implies the number of available / accessible cores don't match our expectation. Usually it's because of explicit affinity binding such...
> Changed /sys/kernel/mm/transparent_hugepage/enabled to madvise, and now they are used. > (n_thp 1766) > Does it mean with always setting it's used anyway or do I have a misunderstanding here?...
@nullptr0-0 can you help to take a look here?
@roblabla Thank you for the PRs and pushing for the improvements on Windows!
@nullptr0-0 can you help to take another look?
I'll try taking a more detailed look at this soon. This size class (the first in the `large` category) looks concerning on the ARM side: ``` large: size ind allocated...