Janis

@AdamJMiles please if you get it working on CUDA, tell me how. I tried to build in VS 2017 and bump against `Error MSB4019 The imported project "C:\BuildCustomizations\IntelOpenCL.props" was not...

@yiifanLu no. :/

I am also dealing with this problem. @OmegaJak , @emilk any temporal workarounds?

```c++ runBuldKernel > > (ng, object_size_ui); printf("1\n"); //stdgpu::unordered_map map = stdgpu::unordered_map::createDeviceObject(8); Pointer* p; cudaMalloc(&p, 1 * sizeof(Pointer)); printf("2\n") ``` This in contrast works in async. Allocation happens without waiting for...

1 is not possible. And I am not sure what 2 does, need to read about it, so it doesn't brake something else.

@stotko doesn't seem like default stream is the issue. This below works in async.. ```c++ runBuldKernel > > (ng, object_size_ui); printf("1\n"); //stdgpu::unordered_map map = stdgpu::unordered_map::createDeviceObject(8); Pointer* p; cudaMalloc(&p, 1 *...

```c++ runBuldKernel > > (ng, object_size_ui); printf("1\n"); k_2 > (); printf("2\n"); stdgpu::unordered_map map = stdgpu::unordered_map::createDeviceObject(8); printf("3\n"); ``` k_2 is executed without waiting. An then it blocks in map creation, i.e....

So there is currently no solution to make this happen in async?

> can you try NIS_CLAMP_OUTPUT = 1 in your defines? Also try to use a bit less sharpness Are you using NIS with HDR? `NIS_CLAMP_OUTPUT` did not make a difference....

 `Sharpness 0 + NIS_CLAMP_OUTPUT `