SYCL Sanitizer: Allow unknown host pointers

[al42and]

Is your feature request related to a problem? Please describe

sycl::memcpy and its friends can work fine with non-USM host pointers. This could be sub-optimal, but is fully legal:

void memcpy(void* dest, const void* src, size_t numBytes)

Copies numBytes of data from the pointer src to the pointer dest. The dest and src parameters must each either be a host pointer or a pointer within a USM allocation that is accessible on the handler’s device.

Source: Table 132 in SYCL 2020 spec rev. 9

However, currently, running the code using non-USM pointers with SYCL Sanitizer produces fatal errors.

I suggest adding a way to either instrument and check non-USM memory allocations, or at least add a flag to sycl-sanitizer to make such errors non-fatal.

Describe the solution you would like

A sycl-sanitize flag to allow the calls to SYCL API with unknown pointers. --allow-unknown-pointers?

Describe alternatives you have considered

• Keep things as they are. Not great, valid SYCL programs will fail under sycl-sanitizer.
• Allow unknown pointers by default. Might miss some errors, but those would likely segfault anyway?
• Also instrument non-USM host memory allocations and track stack state, thus making all pointers known. The best solution for users, but perhaps more overhead/complexity?

Additional context

$ cat test-sanitizer-unknown.cpp 
#include <iostream>
#include <sycl/sycl.hpp>

int main() {
  for (const auto &dev :
       sycl::device::get_devices(sycl::info::device_type::gpu)) {
    std::cout << dev.get_info<sycl::info::device::name>() << std::endl;
    sycl::queue q{dev};

    int stack = 1;
    int *heap = static_cast<int *>(malloc(sizeof(int)));
    *heap = 2;
    int *host = sycl::malloc_host<int>(1, q);
    *host = 3;

    int *d = sycl::malloc_device<int>(3, q);
    q.memcpy(d + 0, &stack, sizeof(int));
    q.memcpy(d + 1, heap, sizeof(int));
    q.memcpy(d + 2, host, sizeof(int));
    q.wait();
    sycl::free(d, q);
    free(heap);
    sycl::free(host, q);
  }
  std::cout << "Done!" << std::endl;
}

$ clang++ -g -fsycl test-sanitizer-unknown.cpp && ONEAPI_DEVICE_SELECTOR=level_zero:gpu ./a.out 
ZE_LOADER_DEBUG_TRACE:Using Loader Library Path: 
ZE_LOADER_DEBUG_TRACE:Tracing Layer Library Path: libze_tracing_layer.so.1
Intel(R) Arc(TM) A770 Graphics
Intel(R) UHD Graphics 770
Done!

$ clang++ -g -fsycl test-sanitizer-unknown.cpp && ONEAPI_DEVICE_SELECTOR=level_zero:gpu sycl-sanitize ./a.out 
ZE_LOADER_DEBUG_TRACE:Using Loader Library Path: 
ZE_LOADER_DEBUG_TRACE:Tracing Layer Library Path: libze_tracing_layer.so.1
Intel(R) Arc(TM) A770 Graphics

[USM] Function uses unknown USM pointer (could be already released or not allocated as USM) as source memory block.
      | memcpy location:  function main at test-sanitizer-unknown.cpp:17
terminate called without an active exception
Aborted (core dumped)