add code example and new slides #371
Conversation
| // wait the queued submissions to complete | ||
| Q.wait(); | ||
|
|
||
| return 0; |
There was a problem hiding this comment.
Nit: we don't need this since C++11
There was a problem hiding this comment.
Oh. Nice, forgot about it
IMO It's kind of ugly that an int function returns nothing... But I can remove it if you have a strong preference.
| } catch (const std::runtime_error &err) { | ||
| std::cout << err.what() << std::endl; | ||
| std::cout << program; | ||
| exit(0); |
There was a problem hiding this comment.
std::abort is maybe better
There was a problem hiding this comment.
It print Aborted in my terminal pretty ugly... Can do std::exit(1) as a middle ground? :)
| @@ -0,0 +1,22 @@ | |||
| CXX ?= icpx | |||
There was a problem hiding this comment.
Why do we need a Makefile as well as a CMakeLists.txt?
There was a problem hiding this comment.
Because I hate CMakeList and have no clue how do debug it.
More seriously because this CMakeList will not work for adaptive CPP and like the simplificy of Makefile.
| global_range, | ||
| [=](sycl::id<1> idx) { | ||
| // Explicit cast because of printf shenaningan. | ||
| sycl::ext::oneapi::experimental::printf("Hello, World! World rank %d\n", |
There was a problem hiding this comment.
I don't know what you've planned exactly here, but from the outside it doesn't really feel appropriate to me to use implementation-specific extensions in a generic SYCL tutorial.
There was a problem hiding this comment.
Yeah, I agree... You support printf aren't you? We can IFDEF and write a "portable" printf and
Without the "print," I need to use the horrible sycl::stream, but then I need to teach command group handler, and it has always been a pain...
So I just took the "less horrible" thingy, but I don't like it either.
There was a problem hiding this comment.
Ah, in the context of moving away from handler I can see the issue.
We don't universally have a variadic printf (and we most certainly will not implement oneAPI interfaces after Intel has set the precedent of implementing AdaptiveCpp extensions but changing the extension name without technical reason so that the API becomes incompatible).
But if it's acceptable to just print "Hello World" without rank, I can give you some internal detail function for printing at least strings, which would happen to work for now ;)
That would be sycl::detail::print(const char* s).
There was a problem hiding this comment.
Yes, good enough! I can always cast my int into string for the
Q.parallel_for(global_range, [=](sycl::id<1> idx) {
// Explicit cast because of printf shenaningan.
sycl::ext::oneapi::experimental::printf("Hello, World! World rank %d\n", static_cast<int>(idx));
});
:)
Yeah, nested lambda tripped off a lot of people in the previous tutorial... Even more the Fortran dev who saw c++ for the "first" time. (but maybe I was just particularly bad at explaining cgh...)
There was a problem hiding this comment.
Great :) Main reason there's no int supported directly is because I didn't want to spend a lot of time developing a generic device string formatting library :P But just for int it's reasonably simple.
Yeah, I'm all for killing buffer and command group handler ;)
These are just additional unnecessary complexities in my opinion.
| # Polaris | ||
| module use /soft/modulefiles | ||
| module load oneapi/upstream | ||
| CXX=clang++ CXXFLAGS="-fsycl -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend --cuda-gpu-arch=sm_80" make -j |
There was a problem hiding this comment.
I had set up an AdaptiveCpp installation on Polaris for the IWOCL tutorial. If there's interest I could see if that still works.
There was a problem hiding this comment.
Oh yes! With pleasure. Can update the cmake too if you want :P
There was a problem hiding this comment.
Okay, will try Polaris tomorrow.
We always had working AdaptiveCpp in SYCLAcademy cmake. I think the only thing that would need to change in your CMakeLists.txt is a find_package(AdaptiveCpp).
Not sure if it's an issue if add_sycl_to_target is invoked on every source file (usually it's invoked once per target with a list of source files), but probably it will work.
AdaptiveCpp already ships this cmake interface.
The corresponding changes in the Makefile would be CXX=acpp and no -fsycl (empty CXXFLAGS, except standard arguments like -O3 etc)
There was a problem hiding this comment.
Not sure if it's an issue if add_sycl_to_target is invoked on every source file (usually it's invoked once per target with a list of source files), but probably it will work.
Each .cpp are independent mains, so can use the foreach hack :)
We always had working AdaptiveCpp in SYCLAcademy cmake. I think the only thing that would need to change in your CMakeLists.txt is a find_package(AdaptiveCpp).
Oh then perfect. Can try on Polaris with Adaptive cpp (when you give me the green light) and move to the SYCLAcamedy CMake away
There was a problem hiding this comment.
Ah, I missed that these are multiple targets, one for each file. Then all should be fine and you just need find_package(AdaptiveCpp).
Will let you know once I've tested on Polaris.
368 removed the Right now we just have clang format in |
We have "hands-on" on the name, so lot of of example.
Copy of https://github.com/argonne-lcf/sycltrain/tree/master/9_sycl_of_hell
Will not go over nd_range as it covered later in the day.