Support for NVIDIA Image Scaling #488
Conversation
|
Maybe it would make sense to use a submodule or a Meson wrap instead of vendoring the SDK? Or maybe not, it's not that large. |
|
why is it removing some fsr support? |
I wanted to avoid pulling in the NIS examples directory which would just take up extra space and be unnecessary. Does Meson have a way to exclude some directories when it pulls in a dependency? |
This change is not removing support for FSR. We are adding support for NIS as an upscaling option. This change will not impact any users who are currently using FSR in Gamescope. |
esullivan-nvidia
changed the title
How would this work from the Meson side? I assume you would download https://github.com/NVIDIAGameWorks/NVIDIAImageScaling/archive/refs/tags/v1.0.2.tar.gz and then extract only the files you need... except that you have already gone and downloaded all 27 mb of that release tarball, which is exactly what you did not want. I don't think there is anything that any application can do about that. Possible options include:
I would advise at a minimum doing option 4, option 3 would be the best if it can be arranged. |
|
@DadSchoorse maybe you want to take a look too? |
|
(also needs rebasing, sorry :( ) |
src/shaders/descriptor_set.h
Outdated
| #if defined(NIS_SCALER) | ||
| layout(binding = 4) uniform const_buffer | ||
| { | ||
| float kDetectRatio; | ||
| float kDetectThres; | ||
| float kMinContrastRatio; | ||
| float kRatioNorm; | ||
|
|
||
| float kContrastBoost; | ||
| float kEps; | ||
| float kSharpStartY; | ||
| float kSharpScaleY; | ||
|
|
||
| float kSharpStrengthMin; | ||
| float kSharpStrengthScale; | ||
| float kSharpLimitMin; | ||
| float kSharpLimitScale; | ||
|
|
||
| float kScaleX; | ||
| float kScaleY; | ||
|
|
||
| float kDstNormX; | ||
| float kDstNormY; | ||
| float kSrcNormX; | ||
| float kSrcNormY; | ||
|
|
||
| uint kInputViewportOriginX; | ||
| uint kInputViewportOriginY; | ||
| uint kInputViewportWidth; | ||
| uint kInputViewportHeight; | ||
|
|
||
| uint kOutputViewportOriginX; | ||
| uint kOutputViewportOriginY; | ||
| uint kOutputViewportWidth; | ||
| uint kOutputViewportHeight; | ||
|
|
||
| float reserved0; | ||
| float reserved1; | ||
| }; |
There was a problem hiding this comment.
This is < 128 bytes, so just use push constants.
There was a problem hiding this comment.
I think we would want to keep this as a uniform buffer to allow Gamescope to keep sharing a single layout between all of its pipelines without adding extra space to the push constant range size.
There was a problem hiding this comment.
Pretty sure we're using like 100 or so bytes of push constant anyway...
There was a problem hiding this comment.
Yes, our pipeline layout uses the full 128 byte range.
There was a problem hiding this comment.
My bad, I should have double checked that. I can move this over to using push constants.
src/shaders/descriptor_set.h
Outdated
| layout(binding = 5) uniform sampler2D coef_scaler; | ||
| layout(binding = 6) uniform sampler2D coef_usm; | ||
| #endif |
There was a problem hiding this comment.
You could use s_sampler_extra, and just add a second one.
src/rendervulkan.cpp
Outdated
| vkCmdCopyBufferToImage(commandBuffer, stagingBuffer, g_output.nisScalerImage, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &imageCopy); | ||
|
|
||
| imageCopy.bufferOffset = coef_scale_size; | ||
|
|
||
| vkCmdCopyBufferToImage(commandBuffer, stagingBuffer, g_output.nisUsmImage, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &imageCopy); |
There was a problem hiding this comment.
I guess I would really like to reuse vulkan_create_texture_from_bits here, but I'll rework most of this in future anyway so it's fine for now.
The docs, and samples directories have been removed along with the third_party_licenses.txt file. Source downloaded from: https://github.com/NVIDIAGameWorks/NVIDIAImageScaling/archive/refs/tags/v1.0.2.tar.gz
esullivan-nvidia
force-pushed
the
master
branch
5 times, most recently
from
0ddb357
to
da1177d
Compare
This PR adds support for selecting NVIDIA Image Scaling as the Gamescope upscaler implementation. It can be enabled by passing the -Y or --nis-upscaling option to Gamescope.