SYCLomatic
SYCLomatic copied to clipboard
oneapi-src
I tried to convert an cuda program that calls cublas functions with the Intel DPCT tool at Intel DevCloud. https://github.com/zhangzibin/cu-nmf/blob/master/NMF_pgd.cu Please see the following message Migration of cusparseXcoo2csr is not...
Incorrect migration of CUB's run length encoding: https://github.com/zjin-lcf/HeCBench/blob/master/rle-cuda/main.cu This issue is reported by @jinz2014 from comments of https://github.com/oneapi-src/SYCLomatic/pull/1132
This is the line in a CUDA program: cudnnSetTensor4dDescriptor(srcTensorDesc, CUDNN_TENSOR_NCHW, CUDNN_DATA_DOUBLE, n, c, h, w) Thanks.
Building the generated program from the CUDA application produces the following error: ./kernels.dp.cpp:110:18: error: no viable conversion from 'typename AccessorCommonT::AccessorSubscript' to 'volatile uint *' (aka 'volatile unsigned int *') volatile...
The generated codes seem incomplete. The CUDA program is https://github.com/zjin-lcf/HeCBench/tree/master/wsm5-cuda. Thanks. ``` q_ct1.memcpy).wait() ; dpct::get_default_queue().memcpy).wait() ; dpct::get_default_queue().memcpy; dpct::get_default_queue().memcpyz).wait() ; dpct::get_default_queue().memcpy.wait(); dpct::get_default_queue().memcpy; dpct::get_default_queue().memcpy.wait(); dpct::get_default_queue().memcpy.wait(); dpct::get_default_queue().memcpy.wait(); dpct::get_default_queue().memcpy.wait(); dpct::get_default_queue().memcpyn).wait() ; dpct::get_default_queue().memcpynncv).wait() ; dpct::get_default_queue().memcpy.wait();...
Draft PR: Extra APIs in NO-USM mode to launch algorithm on the device if the incoming input iterator is a pointer. If the input pointers are pointers in the virtual...
The tool maps __expf() to sycl::exp(). Native math function may be desired.
Compiling the migrated program shows a few errors. Thanks. main.dp.cpp:35:25: error: unknown type name 'T' void someKernel(int *x, T cmem) { ^ main.dp.cpp:36:15: error: 'Foo' does not refer to a...
Compiling the migrated program produces a few errors. Thanks. Please see the pasted Cuda program. ``` /* Copyright (c) 2022, NVIDIA CORPORATION. All rights reserved. * * Redistribution and use...