Re: ROCm ITP
On 12/20/21 18:29, Cordell Bloor wrote:
You're welcome to do so in Debian, but I would be very surprised
if those patches would be accepted upstream. Is it somehow causing
problems?
No, but it is quite a significant work that you have to maintain on your
side.
Hopefully this is not needed anymore soon!
The original hipcc script in the HIP repo was written in perl.
This new hipcc program is written in C++. I wouldn't worry about this
until it's actually released.
Thanks, I had not realized that the hipcc one was a script :)
Can you please try to explain, why there is openCL in the process of
building hip,
as if all I understood was that I am putting legos together?
The targets for the "hip module", look to be :
1. rocm-clr:
librocclr.a (Builds currently static by default)
2. rocm-opencl:
khronos/icd/libopencl.so 47K
khronos/icd/libcltrace.so 186K
tools/clinfo/clinfo 289K
amdocl/libamdocl64.so 3.8M
3. rocm-hipamd:
For hip, after having compiled thanks to your script:
lib/libamdhip64.so 14M
lib/libhiprtc-builtins.so 327K => What is the purpose of this library?
The "rtc" acronym?
bin/hipdemangleatp 1.9K
bin/hipconfig 8.1K
bin/hipcc 27K
Am I missing something in the noise of executable files?
This is notwithstanding the hip testsuite, examples, hipify... that
could have separate packages.
BR, Maxime
Reply to: