(H)ybrid high-(Q)uality (A)nti-(A)liasing (HQAA)

I think I've managed to fix it, if the error you're getting is a NaN/Infinity literals not allowed by shader model. I managed to reproduce the behavior on Borderlands TPS (oddly, it failed to compile in setup mode and compiled properly in performance mode...). It was one particular offending term tied to the edge threshold causing the issue, and the fix was super janky - I flipped the numerator and denominator in each one causing an error and either added or removed rcp (based on whether rcp was already being used).

that was indeed the error ill post back later today sometime if its solved the issue on my end. As far as the fix goes would this cause any bugs in terms of how it is meant to function ?

No. Rcp just calculates the reciprocal (1/x) - when used on an explicit division it's identical to reversing the positions of the numerator and denominator. The reason this is janky is because the compiler chokes on it one way, but not the other, yet both are functionally identical.

v29.2 is now available, and although the changes are functionally minor, they make a big difference to the output quality in most cases. HQAA's optional CAS sharpening has some new tricks up its sleeve that, combined with improved calibration of the defaults and global presets, leave scenes looking nice and sharp without really over- or under-sharpening anything.

Hello, where can I download the shader? Copying the code from here does not work, gives a compilation error

Direct link: github.com/lordbean-git/HQAA

Also can be found on the 1st page in the beginning of the 1st message by lordbean, hyperlinked as # HQAA.

v29.3 is now available, featuring a major optimization of the FXAA code that significantly improves both performance and output quality and a new "Lossless Mode" global preset, which is designed to retain the maximum possible amount of detail in the scene after processing. This version is worth updating to for the performance improvements alone as it runs significantly faster than any prior version of HQAA.

HQAA v30.0 final is now available.

The output quality and performance of v30 are both better than v29.3 thanks to improvements in the accuracy of both SMAA and FXAA and a general fine-tuning pass to squeeze as much out of the shader as possible. This is the final major revision of HQAA as I do not believe I can push the output quality and flexibility of the shader any further than I already have at this point. I cannot rule out that there may be smaller updates in the future if bugs or compatibility problems are found, but apart from that, version 30 can be considered the official "the project is done" release of the shader.

WoW .. Very Thanks for all your work lordbean, this is a VERY useful shader since AA is a primary aspect to improve on games and your shader does it WELL .. so GooD !
[^.^)

This makes me wonder: is it possible to write a shader code that would use RTX Tensor Cores or RX 7000 series AI cores to process? I know ReShade can't do (true) temporal stuff because it requires access to complicated  functions of the render, but addressing via tensor cores / AI acceleration should be achievable, right? 

v30.1 update now available. It fixes a bug that would cause a compile failure when TAA was configured in certain ways.

HQAA v30.2 is now available, which is a collection of (mostly) minor improvements compared to 30.1. The most notable change is an improvement in the accuracy of the FXAA output in 30.2, which tends to produce sharper final images than previously.

HQAA v30.3 is now available.

30.3 includes a lot of small improvements, fixes, and optimizations, and a new major feature - when using TAA Persistence, HQAA's TAA will now perform next-frame prediction and use this to help reduce shimmering in the scene.

Find it here: github.com/lordbean-git/HQAA