Support double precision processing #150
Conversation
If `supprotsDoublePrecision` in the processor is true, advertise FLOAT64 audio ports and adjust the processing loop. Tested by running Airwindows Consolidated as CLAP in reaper (which supports double) and Bitwig (which does not) and seeing appropriate behavior.
|
@jatinchowdhury18 if you wouldn't mind a peek I would appreciate it! |
src/wrapper/clap-juce-wrapper.cpp
Outdated
| @@ -1559,40 +1583,84 @@ class ClapJuceWrapper : public clap::helpers::Plugin< | |||
| { | |||
| for (uint32_t ch = 0; ch < process->audio_inputs[idx].channel_count; ++ch) | |||
| { | |||
| auto *ic = process->audio_inputs[idx].data32[ch] + n; | |||
| if (inputChannels < outputChannels) | |||
| if (supportsDouble && process->audio_outputs[idx].data64 != nullptr) | |||
There was a problem hiding this comment.
I don't think it makes a difference logically, but it seems a little bit odd from the "reader's" perspective to be checking the process->audio_outputs pointers, when the body of this if statement is mostly concerned with process->audio_inputs.
There was a problem hiding this comment.
Oh no that’s a mistake for sure! I’ll fix and merge thanks!
|
|
||
| if (supportsDouble) | ||
| { | ||
| doubleBusses.fill(nullptr); |
There was a problem hiding this comment.
Neither this fill nor busses.fill(nullptr) should be necessary since both busses and doubleBusses are already zero-initialized with {}.
There was a problem hiding this comment.
So you are absolutely correct we init busses with {}
I thought we had not used the {} form, hence the fill, so conditionalized the double busses fill so it is unused.
I think our best bet is to remove the {} on double busses and keep the fill but it's not that much assembly difference. Basically I wanted case three here: https://godbolt.org/z/ev7TzrKGj but I accidentally wrote case 4!
There was a problem hiding this comment.
Sounds like a good idea.
For busses, zero-init with {} and no fill, and for doubleBusses, don't use zero-init and do a conditional fill. That would probably generate the best assembly.
While it isn't a huge assembly difference, it's inside the process method and calling fill is O(N), so it would be good to fix this.
There was a problem hiding this comment.
Yup that’s why I put the fill in an if. Just I didn’t see the {} above!
If
supprotsDoublePrecisionin the processor is true, advertise FLOAT64 audio ports and adjust the processing loop.Tested by running Airwindows Consolidated as CLAP in reaper (which supports double) and Bitwig (which does not) and seeing appropriate behavior.