anderspapitto

"performance has regressed" when throughput went up

1

I changed a benchmark to do twice as much work per iteration, and added a multiplier of 2 in the `.throughput()` to account for this. Time taken per iteration went...

the caffe2 operators backing the RNN/LSTM/GRU conversions require that sequence_lengths be int32. If we are provided a LongTensor in pytorch for sequence lengths, we should automatically cast to IntTensor

/dev/pts/0 is owned by root user, which breaks emacsclient

1

``` [nixos@nixos:~]$ emacsclient -a '' foo.txt -t /home/nixos/.nix-profile/bin/emacsclient: can't find socket; have you started the server? /home/nixos/.nix-profile/bin/emacsclient: To start the server in Emacs, type "M-x server-start". Warning: due to a...

Zero time per iteration is possible with criterion-perf-events

2

I've just run into this error https://github.com/bheisler/criterion.rs/blob/2f5360737807cbe90d149db6199783236f0ef634/src/analysis/mod.rs#L116 on a benchmark using criterion-perf-events (https://github.com/jbreitbart/criterion-perf-events) to count cache misses. On small inputs it's completely possible to have zero cache misses per iteration.

I've made a wrapper around `libc::mmap` that does literally nothing except apply the #[no_panic] attribute, and I get the "detected panic in function `my_mmap`" error. ``` #[no_panic] pub unsafe extern...

when running tests with RUST_BACKTRACE set, a test which panics produces output like ``` thread '...' panicked at '...', src/foo.rs:60:7 stack backtrace: 0: rust_begin_unwind at /rustc/2f2c438dce75d8cc532c3baa849eeddc0901802c/library/std/src/panicking.rs:578:5 1: core::panicking::panic_fmt at /rustc/2f2c438dce75d8cc532c3baa849eeddc0901802c/library/core/src/panicking.rs:67:14...

I regularly kick off some tests (with `rustic-cargo-test`), and then, since they take a while, edit the code and want to recompile/restart the test/etc, before the original tests have completed....