| 1 |
Andrea Arteaga vas escriure el dia dl, 19 set 2011: |
| 2 |
|
| 3 |
> Dear Xavier, |
| 4 |
> |
| 5 |
> thank you for your email and sorry for the late reply. |
| 6 |
> In theory it should be useless to provide optimization flags when |
| 7 |
> compiling ATLAS, because it automatically uses and tests many flags, |
| 8 |
> but I could be wrong. |
| 9 |
|
| 10 |
Yes, Atlas tries to find the best options. However, I was worried because |
| 11 |
I had -Ofast in CFLAGS, and thought that maybe some GCC optimizations were |
| 12 |
harming the binary. |
| 13 |
|
| 14 |
> |
| 15 |
> How did you install ATLAS? From the science overlay? I suggest that |
| 16 |
> you update it. Now in the overlay we have sci-libs/atlas-3.9.49 |
| 17 |
> (~keyworded) [BTW, I will probably bump the version to 3.9.51 today]. |
| 18 |
> Could you install it and redo the test? Otherwise, could you shortly |
| 19 |
> explain how to reproduce the problem, so that I can try? |
| 20 |
|
| 21 |
No, I simply use the atlas provided in the official gentoo branch. |
| 22 |
|
| 23 |
François sent an email pointing that there was a problem with 3.9-xx, and |
| 24 |
suggesting going bach to 3.8. This is what I did and it is working fine, |
| 25 |
without errors. So I assume that, effectively, it was something related |
| 26 |
to the 3.9 branch, rather than compiling options. |
| 27 |
|
| 28 |
A simplification of the code that was giving trouble is: |
| 29 |
|
| 30 |
-----8<--------------- |
| 31 |
B <- c(0, 4.9039) |
| 32 |
for (l in 10:100) { |
| 33 |
X <- cbind(1, matrix(rnorm(l, 0, 3), ncol=1)) |
| 34 |
X <- ics |
| 35 |
for (i in 1:1000) { |
| 36 |
M <- X %*% B |
| 37 |
w <- which(is.nan(M)) |
| 38 |
l <- length(w) |
| 39 |
if (l!=0) { |
| 40 |
print(paste("Length", l, "error(s) on", X[w,2], sep=" ")) |
| 41 |
} |
| 42 |
} |
| 43 |
} |
| 44 |
-----8<--------------- |
| 45 |
|
| 46 |
Thank you, |
| 47 |
-- |
| 48 |
- Xavier - |
Archives