January 2024 - Wine-GitLab - list.winehq.org

Re: [PATCH v3 0/3] MR4857: Draft: gdi32,win32u: Implement D3DKMTEnumAdapters2.
by Krzysztof Bogacki (＠Saancreed) Jan. 19, 2024

Jan. 19, 2024

Thanks for reviewing, I'll probably need a few days to figure this out. I'll mark this MR as draft while it's work-in-progress and I'll unmark it when the next version is ready. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/4857#note_58259

1 0

Re: [PATCH v3 0/3] MR4857: Draft: gdi32,win32u: Implement D3DKMTEnumAdapters2.
by Krzysztof Bogacki (＠Saancreed) Jan. 19, 2024

Jan. 19, 2024

On Fri Jan 19 04:03:14 2024 +0000, Zhiyi Zhang wrote: > Why enum_adapters_2_desc.NumAdapters + 1 and not > enum_adapters_2_desc.NumAdapters? I don't think it's necessary to check > buffer that is out of range. Sure, will remove. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/4857#note_58258

1 0

Re: [PATCH v3 0/3] MR4857: Draft: gdi32,win32u: Implement D3DKMTEnumAdapters2.
by Krzysztof Bogacki (＠Saancreed) Jan. 19, 2024

Jan. 19, 2024

> This is wrong. adapters in sysparams.c are GDI adapters, they are not the same adapter in D3DKMT terms. Adapters in D3DKMT terms mean GPUs. So you need to add a list of GPUs to the cache and then do enumeration from that. Ah, okay, I see. I was unaware of GDI having a different concept of adapters than D3D(KMT), my bad there :sweat_smile: > You should do the initialization from the SetupAPI registry data. For example, basically with the following code. However, since you're in win32u. You can't use SetupAPI directly and you need to convert these to NT registry calls instead. See prepare_devices() for some example code for how to do it. Thanks, I'll see if I can figure this out. > Then `NumOfSources` is the number of GDI adapters with the same GPU LUID. Okay. From what I've observed, on Windows (checking 2 separate machines), it seems to be constant 4 for adapters that represent physical GPUs and 0 for adapter that seems to represent Microsoft's Basic Render thingy. > As for bPrecisePresentRegionsPreferred, I guess you can just return TRUE. For what it's worth, I haven't seen any adapter to report `TRUE` here on Windows. So it might be safer to default to `FALSE` instead. > The GPU list construction can be done in update_display_cache_from_registry(). Got it, thanks. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/4857#note_58257

1 0

Re: [PATCH v3 0/3] MR4857: Draft: gdi32,win32u: Implement D3DKMTEnumAdapters2.
by Krzysztof Bogacki (＠Saancreed) Jan. 19, 2024

Jan. 19, 2024

On Fri Jan 19 04:03:17 2024 +0000, Zhiyi Zhang wrote: > D3DKMT_ENUMADAPTERS2 already has an array for this. I think you can > avoid this memory allocation. Indeed, but I thought it would be safer to allocate some memory for a temporary working area, try enumerating adapters and only copy the result if the entire operation did not fail halfway through rather than using the array I was passed by the caller directly and risk writing some partially misleading garbage to it. I also wasn't sure how to proceed in such case: if I already wrote some LUIDs and handles, should I try to `memset(0)` them just in case or leave the data there and trust that all callers of this API will ignore whatever was written, treating the content as unspecified? It's also somewhat tricky to cause Windows to fail here so I don't know how exactly does it behave when it encounters some error internally. Copying the full result only on full success seemed like an easy way out that allowed me to not bother figuring this out, even if it costs an extra allocation. But I can revert to writing to `pAdapters` directly if you'd prefer it that way. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/4857#note_58256

1 0

Re: [PATCH v3 0/3] MR4857: Draft: gdi32,win32u: Implement D3DKMTEnumAdapters2.
by Krzysztof Bogacki (＠Saancreed) Jan. 19, 2024

Jan. 19, 2024

On Fri Jan 19 04:03:15 2024 +0000, Zhiyi Zhang wrote: > You can remove this line. Together with the +1 above, this was meant to verify that array entries past number returned by the call that actually fills them are untouched (and +1 was for the case where the implementation returns exactly as many adapters as the length of the array that was passed). If you believe it's not something worth testing, I'll just remove this and +1 above. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/4857#note_58254

1 0

Re: [PATCH v3 0/3] MR4857: Draft: gdi32,win32u: Implement D3DKMTEnumAdapters2.
by Krzysztof Bogacki (＠Saancreed) Jan. 19, 2024

Jan. 19, 2024

On Fri Jan 19 04:03:16 2024 +0000, Zhiyi Zhang wrote: > Please add a test for this. While two separate Windows machines on which I queried the size of array to allocate did return 32, this number isn't something that's documented anywhere and probably shouldn't be relied on directly. I can add a test that checks that `NumAdapters` is always set to `32` if `pAdapters` is `NULL` and it might even pass on most if not all Windows versions currently out there but Microsoft could one day decide that in such cases `64` should now be returned and there would be nothing wrong with that… or so I believe. Otherwise, this is already tested by first `ok(enum_adapters_2_desc.NumAdapters, "Expect not zero.\n");`, although not very strictly. I'm also not sure what would Windows return on systems with more than 32 adapters. Can Windows even support that many GPUs at the same time? -- https://gitlab.winehq.org/wine/wine/-/merge_requests/4857#note_58255

1 0

[PATCH v2 0/2] MR569: tests/shader_runer: Support SRV buffers.
by Francisco Casas (＠fcasas) Jan. 19, 2024

Jan. 19, 2024

This allows to declare SRV buffers in the shader tests using `[buffer srv n]` blocks, and passing them to the different backends. Also, 1/2 includes a missing bit required to parse `Buffer<>` types in the HLSL compiler. This is a step towards supporting StructuredBuffer types in further patches. -- v2: tests/shader_runner: Support SRV buffers. https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/569

2 2

[PATCH 0/2] MR569: tests/shader_runer: Support SRV buffers.
by Francisco Casas (＠fcasas) Jan. 19, 2024

Jan. 19, 2024

This allows to declare SRV buffers in the shader tests using `[buffer srv n]` blocks, and passing them to the different backends. Also, 1/2 includes a missing bit required to parse `Buffer<>` types in the HLSL compiler. This is a step towards supporting StructuredBuffer types in further patches. -- https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/569

3 4

Re: [PATCH v7 0/1] MR4343: setupapi: Add stub for DriverStoreFindDriverPackageW
by Fabian Maurer (＠DarkShadow44) Jan. 19, 2024

Jan. 19, 2024

On Mon Nov 20 18:35:34 2023 +0000, Alex Henrie wrote: > Sounds reasonable. Does that mean you have no objections to merging this > patch as it currently stands? any news on this? -- https://gitlab.winehq.org/wine/wine/-/merge_requests/4343#note_58246

1 0

Re: [PATCH v3 0/5] MR4877: include: Add some ncrypt definitions. - approved
by Alexandre Julliard (＠julliard) Jan. 19, 2024

Jan. 19, 2024

This merge request was approved by Alexandre Julliard. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/4877

1 0