[Bug 52689] New: NVCUDA: Daz Studio 4.20 and CUDA SDK11 implementation

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=52689

Bug ID: 52689
           Summary: NVCUDA: Daz Studio 4.20 and CUDA SDK11 implementation
           Product: Wine-staging
           Version: 7.4
          Hardware: x86-64
                OS: Linux
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: -unknown
          Assignee: [e-mail]
          Reporter: [e-mail]
                CC: [e-mail], [e-mail]
      Distribution: ---

Created attachment 72031
  --> https://bugs.winehq.org/attachment.cgi?id=72031
Series of patches to implement basic SDK11 functionality

Recently DAZ Studio4 updated to 4.20 and the Iray engine also got updated to
the point it requires support for CUDA 11.4

I have worked through a great deal of the issues regarding support for SDK
11.4, aside from the D3D9/10/11 elements that have no equivalent Linux support.
The result is adding a few functions to the nvcuda implementation in
wine-staging. (Attachment with a series of 4 patches).

I have not yet been able to get SDK 11.6 samples to run, as this will (for now)
fail the CUDA runtime validation.. but DAZ Studio 4.20.1 is running fine.

2022-03-17 22:24:11.731 Iray [INFO] - IRAY:RENDER ::   1.1   IRAY   rend info :
NVIDIA display driver version: 510.54
2022-03-17 22:24:11.732 Iray [INFO] - IRAY:RENDER ::   1.1   IRAY   rend info :
Your NVIDIA driver supports CUDA version up to 11.6; iray requires CUDA version
11.4; all is good.
2022-03-17 22:24:11.732 Iray [INFO] - IRAY:RENDER ::   1.1   IRAY   rend info :
Using iray plugin version 5.3, build 349500.7063 n, 18 Jan 2022,
nt-x86-64-vc142.
2022-03-17 22:24:11.768 Iray [INFO] - IRAY:RENDER ::   1.1   IRAY   rend info :
CUDA device 0 (NVIDIA GeForce RTX 2070): compute capability 7.5, 7.773 GiB
total, 7.229 GiB available, display attached

I have built and run the main demo samples from the cuda-samples source github,
aswell as 30 some samples in addition to that, compiled with VS2019 and using
the SDK11.4 runtime library.

[Bug 52689] NVCUDA: Daz Studio 4.20 and CUDA SDK11 implementation

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=52689

Sveinar Søpler  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #72031|0                           |1
        is obsolete|                            |

--- Comment #1 from Sveinar Søpler  ---
Created attachment 72223
  --> https://bugs.winehq.org/attachment.cgi?id=72223
Series of patches to implement basic SDK11 functionality

Updated series of patches that will run a few more OptiX SDK samples aswell as
running most of the CUDA SDK 11.4 samples. (The D3D samples does not work).

I chose to use a bit more opaque voids in the cuda.h header instead of
including all the structs and enums.

[Bug 52689] NVCUDA: Daz Studio 4.20 and CUDA SDK11 implementation

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=52689

[e-mail] changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #2 from [e-mail] ---
Thanks for this patch. PyTorch calls the Unknown7 function introduced in the
patch, probably via cudaGetDeviceCount (a cudart function, cudart being the
higher-level api on top of the cuda driver api).

As a side note, the game I'm trying to run, AI Roguelite (which uses pytorch),
also depends on nvml.dll, for which I needed
https://github.com/Saancreed/wine-nvml .