Fix rounding for subnormal floats, closes #8685 #8687
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
josevalim
commented
Jan 24, 2019
lib/elixir/lib/float.ex
Outdated
Contributor
There was a problem hiding this comment.
I have found that this improves performance by about 20%. I am looking for other optimisations now.
Member
Author
There was a problem hiding this comment.
That's great. I wonder if we can implement shift_right, scale_up and scale_down using log2 and if that would be more efficient too.
Contributor
There was a problem hiding this comment.
It would be the best if there would be popcnt (population count aka return number of set bits) and lzcnt (leading zeroes count) support. But this probably would need to be implemented as NIF, so not possible for Elixir, maybe as functions in Erlang itself.
Contributor
There was a problem hiding this comment.
@josevalim BTW for greater speed improvement you need to inline that, as currently the improvement is negligible.
Contributor
There was a problem hiding this comment.
Function call:
Compiling 1 file (.ex)
Operating System: macOS"
CPU Information: Intel(R) Core(TM) i5-5250U CPU @ 1.60GHz
Number of Available Cores: 4
Available memory: 8 GB
Elixir 1.8.0
Erlang 21.1
Benchmark suite executing with the following configuration:
warmup: 2 s
time: 5 s
memory time: 0 μs
parallel: 1
inputs: Large | large precision, Medium | large precision, Trivial | large precision
Estimated total run time: 42 s
Benchmarking built in with input Large | large precision...
Benchmarking built in with input Medium | large precision...
Benchmarking built in with input Trivial | large precision...
Benchmarking built in fix with input Large | large precision...
Benchmarking built in fix with input Medium | large precision...
Benchmarking built in fix with input Trivial | large precision...
##### With input Large | large precision #####
Name ips average deviation median 99th %
built in 172.81 5.79 ms ±9.42% 5.61 ms 8.27 ms
built in fix 170.80 5.85 ms ±19.30% 5.41 ms 10.51 ms
Comparison:
built in 172.81
built in fix 170.80 - 1.01x slower
##### With input Medium | large precision #####
Name ips average deviation median 99th %
built in fix 154.15 6.49 ms ±15.14% 6.13 ms 10.06 ms
built in 151.96 6.58 ms ±26.71% 6.14 ms 13.16 ms
Comparison:
built in fix 154.15
built in 151.96 - 1.01x slower
##### With input Trivial | large precision #####
Name ips average deviation median 99th %
built in fix 160.36 6.24 ms ±21.60% 5.80 ms 11.78 ms
built in 156.33 6.40 ms ±18.28% 6.03 ms 10.99 ms
Comparison:
built in fix 160.36
built in 156.33 - 1.03x slower
Inlined:
Compiling 1 file (.ex)
Operating System: macOS"
CPU Information: Intel(R) Core(TM) i5-5250U CPU @ 1.60GHz
Number of Available Cores: 4
Available memory: 8 GB
Elixir 1.8.0
Erlang 21.1
Benchmark suite executing with the following configuration:
warmup: 2 s
time: 5 s
memory time: 0 μs
parallel: 1
inputs: Large | large precision, Medium | large precision, Trivial | large precision
Estimated total run time: 42 s
Benchmarking built in with input Large | large precision...
Benchmarking built in with input Medium | large precision...
Benchmarking built in with input Trivial | large precision...
Benchmarking built in fix with input Large | large precision...
Benchmarking built in fix with input Medium | large precision...
Benchmarking built in fix with input Trivial | large precision...
##### With input Large | large precision #####
Name ips average deviation median 99th %
built in fix 186.26 5.37 ms ±29.93% 4.78 ms 11.67 ms
built in 170.14 5.88 ms ±11.70% 5.62 ms 9.29 ms
Comparison:
built in fix 186.26
built in 170.14 - 1.09x slower
##### With input Medium | large precision #####
Name ips average deviation median 99th %
built in fix 175.94 5.68 ms ±28.99% 5.18 ms 11.74 ms
built in 159.96 6.25 ms ±10.12% 6.00 ms 9.41 ms
Comparison:
built in fix 175.94
built in 159.96 - 1.10x slower
##### With input Trivial | large precision #####
Name ips average deviation median 99th %
built in fix 201.76 4.96 ms ±12.30% 4.76 ms 7.44 ms
built in 168.05 5.95 ms ±13.54% 5.65 ms 9.01 ms
Comparison:
built in fix 201.76
built in 168.05 - 1.20x slower
Member
Author
There was a problem hiding this comment.
Yeah, I inlined it after your comment about performance. Inlined sign as well.
josevalim
force-pushed
the
jv-subnormal-float-round
branch
from
6397c88 to
2104890
Compare
Contributor
|
@josevalim about performance I am also think that we could check if the number is less than |
Member
Author
|
❤️ 💚 💙 💛 💜 |
josevalim
added a commit
that referenced
this pull request
Jan 25, 2019
Closes #8685 Signed-off-by: José Valim <jose.valim@plataformatec.com.br>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.