diff --git a/SYCL/Plugin/retain_events.cpp b/SYCL/Plugin/retain_events.cpp new file mode 100644 index 0000000000..7611a54ad9 --- /dev/null +++ b/SYCL/Plugin/retain_events.cpp @@ -0,0 +1,80 @@ +// REQUIRES: gpu +// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out +// RUN: %GPU_RUN_PLACEHOLDER %t.out + +#include +#include +#include +#include + +namespace { + +void print_device_properties(sycl::device const &dev) { + auto plat_name = + dev.get_platform().template get_info(); + auto dev_name = dev.template get_info(); + auto driver_version = + dev.template get_info(); + + fprintf(stdout, " platform name: %s\n", plat_name.c_str()); + fprintf(stdout, " device name: %s\n", dev_name.c_str()); + fprintf(stdout, "driver version: %s\n", driver_version.c_str()); +} + +void async_sycl_error(cl::sycl::exception_list el) { + fprintf(stderr, "async exceptions caught:\n"); + for (auto l = el.begin(); l != el.end(); ++l) { + try { + std::rethrow_exception(*l); + } catch (const cl::sycl::exception &e) { + fprintf(stderr, "what: %s code: %d\n", e.what(), e.get_cl_code()); + std::exit(-1); + } + } +} + +} // namespace + +int test(sycl::device &D) { + sycl::context C(D); + constexpr size_t N = 1024 * 1000; + + int *src = sycl::malloc_shared(N, D, C); + int *dst = sycl::malloc_shared(N, D, C); + + std::iota(src, src + N, 0); + std::fill(dst, dst + N, 0); + + for (int i = 0; i < 256; ++i) { + sycl::queue Q{ + C, D, + async_sycl_error}; //, sycl::property::queue::enable_profiling() }; + + Q.submit([&](sycl::handler &SH) { + sycl::range<1> r(N); + SH.parallel_for(r, [=](sycl::id<1> id) { + size_t i = id.get(0); + dst[i] = src[i] + 1; + }); + }); + } + + sycl::free(dst, C); + sycl::free(src, C); + + return 0; +} + +int main() { + try { + sycl::device D1{sycl::gpu_selector{}}; + print_device_properties(D1); + for (int i = 0; i < 10; ++i) { + test(D1); + } + } catch (sycl::exception &e) { + fprintf(stderr, "...sycl failed to entertain with \"%s\" (%d) \n", e.what(), + e.get_cl_code()); + } + return 0; +}