CTS math_builtin_float_base fail with CUDA backend

[XunZhangIntel]

Describe the bug
CTS math_builtin_float_base fail with CUDA backend on latest SYCL builds. See logs below:

--- math_builtin_float_base
  . value: -1, reference: 0
  . line: 178
  . tests case: 4000217. Correctness check for ptr failed.
  . value: -1, reference: -3
  . line: 252
  . tests case: 4000218. Correctness check for ptr failed.
  . value: -2, reference: -1
  . line: 212
  . tests case: 4000219. Correctness check for ptr failed.
  . value: -2, reference: 0
  . line: 178
  . tests case: 4000220. Correctness check for ptr failed.
  . value: -2, reference: 0
  . line: 252
  . tests case: 4000221. Correctness check for ptr failed.
  . value: 0, reference: -1
  . line: 212
  . tests case: 4000222. Correctness check for ptr failed.
  . value: 0, reference: -1
  . line: 178
  . tests case: 4000223. Correctness check for ptr failed.
  . value: 0, reference: -1
  . line: 252
  . tests case: 4000224. Correctness check for ptr failed.
  . value: -2, reference: 0
  . line: 212
  . tests case: 4000225. Correctness check for ptr failed.
  - fail

  . Passed 0/13 tests (0%)

To Reproduce
Note: Due to a recent uplift of minimal supported C++ standard, one should modify this line to require C++17.

1. Get SYCL with CUDA support (GSG) and SYCL-CTS
2. Build CTS tests

cd SYCL-CTS
mkdir build && cd build
cmake -GNinja  -DSYCL_IMPLEMENTATION=Intel_SYCL -DINTEL_SYCL_ROOT=<path to built sycl> -Dopencl_platform_name=nvidia -Dopencl_device_name=opencl_gpu -DCMAKE_BUILD_TYPE=Release -DINTEL_SYCL_FLAGS="-Xsycl-target-backend;--cuda-gpu-arch=sm_50" -DCMAKE_EXE_LINKER_FLAGS=" -Wl,-no-relax " -DINTEL_SYCL_TRIPLE=nvptx64-nvidia-cuda-sycldevice -DSYCL_CTS_ENABLE_OPENCL_INTEROP_TESTS=Off -DSYCL_CTS_ENABLE_DOUBLE_TESTS=On -DSYCL_CTS_ENABLE_HALF_TESTS=On ..
ninja test_math_builtin_api

3. Run Test

SYCL_DEVICE_FILTER="cuda:gpu:0" ./bin/test_math_builtin_api -p nvidia -d opencl_gpu  --test math_builtin_float_base

Environment (please complete the following information):
OS version: Ubuntu 18.04.5 LTS (Bionic Beaver)
CUDA version:

nvidia-smi

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 435.21       Driver Version: 435.21       CUDA Version: 10.1     |
|-------------------------------+----------------------+----------------------+

[hdelan]

Bugs resolved for CUDA backend of math functions sycl::frexp (#4788) and sycl::mul_hi (#4802).

Bugs remaining for sycl::nextafter. Investigating source.

sycl::nextafter also returning incorrect result on host for sycl::half. This is likely to be a problem with llvm builtins.

@bader will I create a new issue for this?

[bader]

sycl::nextafter also returning incorrect result on host for sycl::half. This is likely to be a problem with llvm builtins.

@bader will I create a new issue for this?

Yes, please.

[hdelan]

New issue created #4815

[hdelan]

All issues resolved by:
#4939
#4788
#4802

[AerialMantis]

The pull requests for addressing the issues in the math built-in functions have been merged and the test is now passing so we believe this issue to be resolved, so we're closing the ticket, if you continue to see any problems we can re-open it.