Thomas,
These two examples work in our in-house systems with the current drop. Let me test in JLSE to see if I can find what is going on.
Carlos
From: "Peterka, Thomas" <tpeterka@mcs.anl.gov>
Date: Friday, November 15, 2019 at 8:34 AM
To: "Rosales-fernandez, Carlos" <carlos.rosales-fernandez@intel.com>
Cc: Ecp-aurora-sdk-discuss <ecp-aurora-sdk-discuss@lists.jlse.anl.gov>
Subject: Re: [Ecp-aurora-sdk-discuss] New A21 SDK coming to JLSE later today
Hi Carlos,
I think there may be something wrong with this latest drop, because SYCL code that used to work now crashes, and reverting to the old version version of dpcpp/2019.9.30.013b continues to work.
Specifically, I see two problems:
1. My test of USM generates the following run time error:
symbol lookup error: ./fixed: undefined symbol: _ZN2cl4sycl6detail3usm12alignedAllocEmmRKNS0_7contextENS0_3usm5allocE
The code is this:
auto dev = Queue.get_device();
auto ctxt = Queue.get_context();
sycl::usm_allocator<int, sycl::usm::alloc::host> alloc(ctxt, dev);
std::vector<int, decltype(alloc)> n(alloc);
n.resize(1);
The last line, resize, is where the error occurs. The same problem happens if I give the size in the constructor and don’t call the resize.
2. The following trivial example generates a segmentation fault:
sycl::queue Queue;
sycl::buffer<sycl::cl_int, 1> Buffer(4);
sycl::range<1> NumOfWorkItems{Buffer.get_count()};
Queue.submit([&](sycl::handler &cgh)
{
auto Accessor = Buffer.get_access<sycl::access::mode::write>(cgh);
cgh.parallel_for<class FillBuffer>(NumOfWorkItems, [=](sycl::id<1> WIid)
{
Accessor[WIid] = (cl::sycl::cl_int)WIid.get(0);
});
});
The seg fault happens in the parallel_for loop.
Tom
-----------------
Tom Peterka
Mathematics and Computer Science Division
Argonne National Laboratory
9700 S. Cass Ave.
Lemont, IL 60439
Bldg 240, Rm 2148
e-mail: tpeterka@mcs.anl.gov
web: www.mcs.anl.gov/~tpeterka
phone: 630-252-7198