grmat / opencl-amd

OpenCL on amdgpu for Arch
15 stars 4 forks source link

Installation of AUR package fails due to sha256sums validation error on amdgpu-pro-17.50-511655.tar.xz #3

Closed enzomich closed 6 years ago

enzomich commented 6 years ago

Here's the final part of the installation log:

[...]
Loaded CA certificate '/etc/ssl/certs/ca-certificates.crt'
Resolving www2.ati.com (www2.ati.com)... 23.56.181.23
Connecting to www2.ati.com (www2.ati.com)|23.56.181.23|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 318509792 (304M) [application/x-tar]
Saving to: ‘amdgpu-pro-17.50-511655.tar.xz’

amdgpu-pro-17.50-511655.tar.x 100%[================================================>] 303.75M   919KB/s    in 5m 43s  

2018-03-06 16:46:04 (906 KB/s) - ‘amdgpu-pro-17.50-511655.tar.xz’ saved [318509792/318509792]

==> Validating source files with sha256sums...
    amdgpu-pro-17.50-511655.tar.xz ... FAILED
==> ERROR: One or more files did not pass the validity check!

EDIT: it appears that the downloaded file amdgpu-pro-17.50-511655.tar.xz is corrupted: not only does it fail the sha256sums integrity check, but it can't be recognised by xz or tar tJf . See also the thread at https://community.amd.com/thread/223507 .

On the other hand, the downloads for the version 17.40 are still available at https://support.amd.com/en-us/kb-articles/Pages/AMDGPU-PRO-Driver-for-Linux-Release-Notes.aspx . However, the filenames of the .deb files inside the download for the AMDGPU-Pro Driver Version 17.40 for Ubuntu 16.04.3 are different (as well as its sha256sums hash, of course), and the PKGBUILD file in the AUR package should be amended accordingly.

grmat commented 6 years ago

You can go back the version history for a 17.40 version of the PKGBUILD, there is also a branch for the 16.40 version which some people seemingly wanted to stick with. You could choose those and call makepkg.

I can't reproduce the errors on downloading the current .deb, and obviously complaining about a corrupted file is not a bug but the intended handling for this case. What AUR-helper are you using? It seems like it didn't even pick up the correct file name (...tar.x instead of ...tar.xz).

enzomich commented 6 years ago

The problem is not with your PKGBUILD but with the corrupted file on the AMD website (as I mentioned, this is also noted at https://community.amd.com/thread/223507). But until (or unless) they fix it, it's better to stick with the version 17.40.

To access AUR I used Manjaro's Pamac, which, by itself, works fine (where do you see "...tar.x instead of ...tar.xz"?)

grmat commented 6 years ago

As I already said, I can't reproduce problems downloading the file. Each time I checked (last time right now), I get the valid file. It's strange because others complained about the same issue on the AUR. Have you tried without pamac, using makepkg on the PKGBUILD directly or downloading the file manually?

The thread at community.amd.com mentions 0 byte file size, you've got 300M, the actual size is >500M. I don't think they are serving a corrupt file but instead have problems with individual downloads.

As I don't think I can do anything about the issue and you already have a workaround, I'll keep this one closed.

enzomich commented 6 years ago

I had indeed run makepkg on the PKGBUILD directly, and the result was the same. But, lo and behold, today everything worked flawlessly from AUR using Pamac! Maybe they fixed the file on www2.ati.com...

(Unfortunately I've found that opencl-amd is unsuitable for me because my GPU (HD 8870M) doesn't seem to be supported by the amdgpu driver, having architecture GCN 1.0 rather than >=1.1, so I'm stuck with the radeon driver and the OpenCL 1.1 supplied by opencl-catalyst, but that's another story).