-
Notifications
You must be signed in to change notification settings - Fork 66
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
if statements & difficulties using matrices vs vectors #22
Comments
Attaching file with tests (just pull off the .txt required by Github to upload) |
I'm working on these. As a note # FIXME - would prefer to replace 2nd loop w/ if stmt
# k <= maxk && (qq[i,k] = tmp) Once it works, it'll probably be slower than the 2 loops, because it will evaluate the store on every iteration (but it'll apply a mask). |
Thanks, @chriselrod! Good to know about the if statement & two loops -- I'll keep them separate then. |
I'm now testing basic conditionals and your softmax3 functions, based on the second (previously broken) version. However, that FIXME wont work yet, because the library doesn't have much support for using the iterating variable for anything other than indexing. |
Oh, fantastic! Thanks, @chriselrod! |
Closed because adding the store to the same loop has now been fixed and added to the testsuite. |
Thanks, @chriselrod! |
Hi @chriselrod, I've upgraded from
@vectorize
to@avx
and am excited to use an officially registered package.Looks like a change (maybe in SIMDPirates?) broke something for me -- the code below works unvectorized, but is broken when I add the
@avx
. Seems like there are 2 issues. First, it looks like if statements in an@avx
loop don't work. Second, it appears that something goes wrong when I iterate over a matrix or a view, not a vector.@code_warntype
suggests there are problems with type inference. See unvectorized, broken vectorized, and fixed vectorized code below.I'm using julia 1.3.1 and
breakage is of 2 forms
The vector/matrix issue:
and
The text was updated successfully, but these errors were encountered: