'vImageSepConvolve_Planar8' is only available on macOS 11.0 or newer about melatonin_blur HOT 22 CLOSED

just not sure how to properly test...

I will need to confirm, but I think I have 10.9 or 10.10 as lowest versions of macOS I still keep on my old Mac Pro just for this kind of scenario. If I can be of any assistance hit me up.

from melatonin_blur.

Those seems to be macOS 14.0+

from melatonin_blur.

Still getting these warnings on ee269d3 it compiles tho, so it might be correctly not using these in reality, but maybe they should be protected by macros / availability guards

from melatonin_blur.

Thanks for reporting.

I couldn't easily find versioning information on Accelerate functions. Did you see any Apple information about this, or is it just the experience you made compiling?

from melatonin_blur.

Ah right, I always forget they are specified this way. Also the header specifies

API_AVAILABLE(macos(11.0), ios(14.0), watchos(7.0), tvos(14.0));

I've was always confused about if that was when the function was added/became available to the SDK or where it's currently available.

Assuming the latter, I'll pick an alternate implementation for < 11.0, just not sure how to properly test...

from melatonin_blur.

I'm targeting 10.13 (High Sierra) and it seems to compile fine. I'll look into adding a runtime check.

Did you run into problems compiling?

from melatonin_blur.

Thanks! I'll ping the issue when I get something to try!

from melatonin_blur.

I'm getting other compilation issues as well, and yes, i've added Accelerate.framework

from melatonin_blur.

Looks like I have some work to do here! Sorry about that, was just working hard on finding the best performing solutions, now I need to work on compatibility, it seems!

from melatonin_blur.

Probably the route to go is have things fall back to the included Gin implementation at runtime where needed (until i find time to benchmark alternatives).

from melatonin_blur.

I've tried to add (JUCE_MAC && MAC_OS_X_VERSION_MAX_ALLOWED > MAC_OS_VERSION_13_3) for those methods but it seems like this doesn't work:

    static void argb (const juce::Image& srcImage, juce::Image& dstImage, size_t radius)
    {
        jassert (srcImage.getFormat() == juce::Image::PixelFormat::ARGB);
#if (JUCE_MAC && MAC_OS_X_VERSION_MAX_ALLOWED > MAC_OS_VERSION_13_3) || JUCE_IOS
        auto kernel = createFloatKernel (radius);

        const auto w = (unsigned int) srcImage.getWidth();
        const auto h = (unsigned int) srcImage.getHeight();
        juce::Image::BitmapData srcData (srcImage, juce::Image::BitmapData::readWrite);
        juce::Image::BitmapData dstData (dstImage, juce::Image::BitmapData::readWrite);

        // vImageSepConvolve isn't happy operating in-place
        vImage_Buffer src = { srcData.getLinePointer (0), h, w, (size_t) srcData.lineStride };
        vImage_Buffer dst = { dstData.getLinePointer (0), h, w, (size_t) dstData.lineStride };
        vImageSepConvolve_ARGB8888 (&src, &dst, nullptr, 0, 0, kernel.data(), (unsigned int) kernel.size(), kernel.data(), (unsigned int) kernel.size(), 0, Pixel_8888 { 0, 0, 0, 0 }, kvImageEdgeExtend);
#else
        stackBlur::ginRGBA (img, radius); // There seems to be missing a non in-place blur in gin
#endif

from melatonin_blur.

Probably the route to go is have things fall back to the included Gin implementation at runtime where needed (until i find time to benchmark alternatives).

Or provide a good simd convolution alternative.

from melatonin_blur.

Or provide a good simd convolution alternative.

There are couple people interested in building SIMD variations of the vector implementation, so I'm betting this will happen!

from melatonin_blur.

Ok, I have a draft implementation to make all the vImage stuff conditional at both compile and run time #16

I'm going to have to get #6 done first before merging, to ensure it's happy on at least macOS 11/12/13 and Windows. Don't want any more breakage!

from melatonin_blur.

@kunitoki You should be good to go, I merged the compatibility changes in #16.

I still would like to see some evidence that the runtime checks are needed (i.e. if i can compile vImageSepConvolve on one machine, will it work on earlier targets before the API was available?) but I went the conservative route for now.

#19 and #20 are both followups for the compatibility work. Off to ADC now though!

from melatonin_blur.

Hello, I'm trying to compile on 10.14.6 mac Intel Xcode 11.3.1 and got the following compile error:
vImage.h:44:9: error: use of undeclared identifier 'vImageSepConvolve_Planar8'

Just to be sure, can you confirm you only want to target macOS 11+?

from melatonin_blur.

@kmolcard I opened a new issue for this, it should run happily on 10.14 (when compiled with something newer) but it looks like we'd need to put in a few more conditionals for it to actually compile there. I'll track the issue here: #37

from melatonin_blur.

thanks for the reply, definitely willing to help you on this if you are interesed

from melatonin_blur.

Sure! You'll just have to navigate the mess that is implementations.h and do some preprocessor juggling so MELATONIN_BLUR_VIMAGE is only set for > 11.0.

Maybe one note we should make somewhere is.. by compiling on older OS, I believe this will exclude newer macs from taking advantage of the newer/faster APIs since the code paths that run those APIs won't be available in the binary.

Also related: I'm working on a better/faster fallback implementation for older OSes/machines.

from melatonin_blur.

Thx — I meant to ask, is this on a Projucer project, exported to xcode?

from melatonin_blur.

Yes

from melatonin_blur.

#53 should handle those warnings still showing up (despite being guarded at runtime)

from melatonin_blur.