Marcel Breyer

Compiling SYCL kernels in different cpp files

1

Lets say I have a simple test program in a single `cpp` file: ```cpp #include #include using namespace sycl; int main() { constexpr std::size_t size = 64; queue myQueue; int...

Correct way to enable Intel GPU support

4

I wanted to test hipSYCL on Intel's devcloud using the experimental Level Zero backend to target Intel GPUs. Therefore I installed the sycl branch from Intel's LLVM fork as stated...

atomic function load() on GPU

5

Calling `load()` on an atomic sycl accessor targeting the GPU (NVIDIA) result in a linker error: ``` $ syclcc-clang -O3 --hipsycl-platform=cuda main.cpp ptxas fatal : Unresolved extern function '_ZNVK7hipsycl4sycl6atomicIiLNS0_6access13address_spaceE0EE4loadENS0_12memory_orderE' clang-10:...

--hipsycl-platform=cpu problems with CMake

9

I've installed hipSYCL stable using the spack package. Setting following options before compiling: ``` source /home/marcel/Programs/spack/share/spack/setup-env.sh spack load hipsycl@master spack load [email protected] export PATH=/home/marcel/Programs/spack/opt/spack/linux-linuxmint19-zen/gcc-8.4.0/hipsycl-master-44bjhhamjo3w4k74hsanb73ulhc27now/bin/syclcc-clang:$PATH ``` After that syclcc-clang outputs the...

Segmentation fault with sycl::buffer(InputIt, InputIt) constructor

5

```cpp #include #include namespace sycl = cl::sycl; struct A { sycl::buffer buffer; A(std::vector& vec) : buffer(vec.begin(), vec.end()) { } // A(std::vector& vec) : buffer(vec.size()) { // auto acc = buffer.get_access();...

More useful vendor string on Intel GPUs

6

**Describe the motivation for the feature request** Executing `opensycl-info` gives `pci:32902` as vendor string for an integrated Intel Gen11 GPU (which is obviously also the case if you programmatically query...

Godbolt links for code snippets

3

It would be a nice improvement if you could add an additional button (besides the copy button) that directly opens the current code snippet in a new [https://godbolt.org/](https://godbolt.org/) tab.

Improve SSCP std builtins remapping for -ffast-math

1

**Bug summary** If fast-math is enabled, the JIT compilation (on an NVIDIA GPU) fails when using `std::pow` inside `std::for_each` with: ``` fatal error: error in backend: Undefined external symbol "__powidf2"...

**Bug summary** When using GoogleTest's `EXPECT_DEATH`/`ASSERT_DEATH` functions, it should be possible to specify the error message printed to stderr that should be emitted if a specific assertion fails. However, when...

### Problem Description The ROCm SMI function `rsmi_dev_energy_count_get` should return the total energy consumed since the last driver restart. However, the output is constant over time during one program execution....