Intel® DPC++ Compatibility Tool Developer Guide and Reference
A newer version of this document is available. Customers should click here to go to the newest version.
Visible to Intel only — GUID: GUID-439F9A3A-2CC8-4254-9E80-E490353709B5
Visible to Intel only — GUID: GUID-439F9A3A-2CC8-4254-9E80-E490353709B5
Query CUDA* to SYCL* API Mapping
Use the --query-api-mapping option to discover which SYCL* API is functionally compatible with a specified CUDA* API. The API query functionality can assist users on manual code migration and help users understand the SYCL API.
The following example queries the functionally compatible SYCL API for the cudaMalloc function:
$ dpct --query-api-mapping=cudaMalloc CUDA API: cudaMalloc(pDev /*void ***/, s /*size_t*/); Is migrated to: *pDev = (void *)sycl::malloc_device(s, dpct::get_in_order_queue());
On Linux, you can use the tab key to auto-complete the CUDA API name specified for the --query-api-mapping option. For example :
Specify the partial CUDA function name “cudaMa” to the --query-api-mapping option:
dpct --query-api-mapping=cudaMaPress tab.
The CUDA API name will auto-complete to cudaMalloc:
dpct --query-api-mapping=cudaMalloc
Intel® DPC++ Compatibility Tool also supports fuzzy matching of CUDA API names. For example:
$ dpct --query-api-mapping=cudamalloc CUDA API: cudaMalloc(pDev /*void ***/, s /*size_t*/); Is migrated to: *pDev = (void *)sycl::malloc_device(s, dpct::get_in_order_queue());