Grain FX by alter-alter · Pull Request #363 · SynthstromAudible/DelugeFirmware

alter-alter · 2023-08-18T05:34:44Z

Added Grain FX to Mod FX
https://youtu.be/9W70SZYMKXo

Parameters:
Mod Depth: Controls Grain Volume / Dry Wet Mix
Mod Offset: Adjusts Grain Size
Mod Rate: Sets Grain Rate
Mod Feedback: Selects Grain Type

Note:
This effect is somewhat resource-intensive, so please use only one instance per song.

sichtbeton · 2023-08-18T10:51:45Z

Tested it with a simple field recording. Instant blade runner.

sichtbeton · 2023-08-18T10:52:26Z

Some options can be modulated, some not. Why is that? Cpu?

alter-alter · 2023-08-18T11:11:32Z

@sichtbeton

Some options can be modulated, some not. Why is that? Cpu?

Currently, it is following the standard behavior of Deluge. Changing that might be a bit challenging at the moment.

m-m-adams

So first off this sounds great! Awesome job and thanks for contributing

Testing this it looks like there's 2 performance issues. Using a test song that can sustain 50 voices, I lose 5 for each clip I add a grain fx too and they stay lost until the modfx is turned off (not just away from grain). E.g. going from off->stereochorus will get 48 voices, going from grain to stereochorus will get 45. This is just caused by the memory for the grain buffer. A first fix is to deallocate that buffer when it's not needed. A potential further fix could be to allocate a single global grain buffer and treat it as a pseudo send fx. Each voice could pull from it and add to it with it's own depth and rate. This is more work to implement so I haven't tested it but I think it would work and still sound cool, if a bit less 'traditional' granular.

Secondly the DSP code is more intensive than the other modfx. I think this is improvable, I've left some comments on easy wins but the bigger one is probably vectorizing the grain calculations. They're fully parallelizable so should be able to get on NEON for a 2-3x speedup

Third I think it is nicer to have the feedback scale with the mix level. There is a pulsing effect that becomes audible on sustained notes and it is alleviated with lower feedback settings. I've left a comment with the calculation I used for testing and for me it produces a more usable range of depth settings

m-m-adams · 2023-08-18T12:52:53Z

+			grainRate = std::max<int32_t>(1, grainRate);
+			grainRate = (kSampleRate << 1) / grainRate;
+			// Preset 0=default
+			grainPitchType = (int8_t)(multiply_32x32_rshift32_rounded(


The goal here is to quantize to 5 levels? It could use a comment

m-m-adams · 2023-08-18T12:58:36Z

+			// Preset 0=default
+			grainPitchType = (int8_t)(multiply_32x32_rshift32_rounded(
+			    unpatchedParams->getValue(Param::Unpatched::MOD_FX_FEEDBACK), 5));
+			grainPitchType = std::max<int8_t>(-2, (std::min<int8_t>(2, grainPitchType)));


Clamp instead

m-m-adams · 2023-08-18T13:06:36Z

+						int32_t pos =
+						    (grains[i].startPoint + delta + kModFXGrainBufferSize) & kModFXGrainBufferIndexMask;
+
+						grains_l = add_saturation(


Multiply and accumulate to save a cycle

multiply_accumulate_32x32_rshift32_rounded (smmlar) does not seem to take saturation operations into account.

It doesn't, but if this is overflowing then that should be fixed instead. The saturation in saturating arithmetic is hard clipping in audio terms and not really desirable

m-m-adams · 2023-08-18T13:17:02Z

+							    & kModFXGrainBufferIndexMask;
+							grains[i].counter = 0;
+							grains[i].rev = (getRandom255() < 76);
+							if (grainPitchType == -2) {


Can you use a switch for readability?

m-m-adams · 2023-08-18T13:18:37Z

+							}
+							else if (grainPitchType == 2) { // temopo sync
+								grains[i].pitch = (getRandom255() < 25)    ? 512
+								                  : (getRandom255() < 153) ? 2048


You can skip a call by storing the previous random value and comparing to 176 instead

m-m-adams · 2023-08-18T14:49:19Z

+
+				int32_t grains_l = 0;
+				int32_t grains_r = 0;
+				for (int32_t i = 0; i < 8; i++) {


You should be able to render this in a vector to speed it up. Not critical to merge but look at the oscillator render function for an example

m-m-adams · 2023-08-18T14:53:23Z

+									}
+								}
+							}
+							grains[i].volScale = (2147483647 / (grains[i].length >> 1));


Stick this in the previous if statement, division is expensive and there's no point in calculating it if grain length is 0

m-m-adams · 2023-08-18T15:11:59Z

+			grainShift = 44 * 300; //(kSampleRate / 1000) * 300;
+			// Size
+			grainSize = 44 * ((((unpatchedParams->getValue(Param::Unpatched::MOD_FX_OFFSET) >> 1) + 1073741824) >> 21));
+			grainSize = std::max<int32_t>(440, std::min<int32_t>(35280, grainSize)); //10ms - 800ms


std::Clamp instead (applies in a bunch of places)

m-m-adams · 2023-08-18T15:13:50Z

+			if (!grainInitialized && modFXGrainBufferWriteIndex >= 65536) {
+				grainInitialized = true;
+			}
+			*postFXVolume = multiply_32x32_rshift32(*postFXVolume, 1518500250) << 1; // Divide by sqrt(2)


define a constant for root 2 in fixedpoint.h (ONE_OVER_SQRT2_Q31) and use that

m-m-adams · 2023-08-18T15:25:36Z

+			    (multiply_32x32_rshift32_rounded(multiply_32x32_rshift32_rounded(grainVol, grainVol), grainVol) << 2)
+			    + 2147483648; //Cubic
+			grainVol = std::max<int32_t>(0, std::min<int32_t>(2147483647, grainVol));
+			grainDryVol =


It would be nice to have feedback scale with the mix level - grainFeedbackVol = grainVol >> 3; works well in my experimentation. The current feedback makes a pulsing sound with sustained notes and I'd like to be able to have lower mixes without it

m-m-adams · 2023-08-18T17:46:19Z

It also might be good to swap the feedback and depth pads

- Fixing buffer dealloc timing - Using MAC - std::clamp - grainFeedbackVol = grainVol >> 3

alter-alter · 2023-08-19T09:11:59Z

@m-m-adams

It also might be good to swap the feedback and depth pads

Since the modulatable parameters are only Depth and Rate, I believe Depth would be suitable for controlling the volume of the grain.

m-m-adams · 2023-08-19T18:15:56Z

You can just make it modulatable by changing it from a Param::Unpatched to a Param::Global. Automating it works fine with the current modfx so modulating should be fine as well

alter-alter · 2023-08-20T01:52:28Z

@m-m-adams

You can just make it modulatable by changing it from a Param::Unpatched to a Param::Global. Automating it works fine with the current modfx so modulating should be fine as well

Thank you! However, the change seems to be static and permanent, and I don't want it to affect the entire system. Is there a way to dynamically change the patch type only when ModFxType is set to Grain?

m-m-adams

Good to go, the lshift by 0 should be removed but it's optimized out anyway

m-m-adams · 2023-08-20T18:22:50Z

-						        << 2,
-						    grains_r);
+						grains_l = multiply_accumulate_32x32_rshift32_rounded(
+						    grains_l, multiply_32x32_rshift32(modFXGrainBuffer[pos].l, vol) << 0, grains[i].panVolL);


<< 0 is a nop but it's optimized out anyway

m-m-adams · 2023-08-20T18:28:39Z

No problem - it can always be made patched in the future

Grain FX

41c1fa1

m-m-adams requested changes Aug 18, 2023

View reviewed changes

Grain FX (fixing buffer dealloc timing, using MAC)

76a59ae

- Fixing buffer dealloc timing - Using MAC - std::clamp - grainFeedbackVol = grainVol >> 3

m-m-adams approved these changes Aug 20, 2023

View reviewed changes

jamiefaye added this pull request to the merge queue Aug 20, 2023

Merged via the queue into SynthstromAudible:community with commit 246804a Aug 20, 2023

m-m-adams mentioned this pull request Aug 21, 2023

Standardize retrieval of quantized values from the param manager #390

Closed

m-m-adams mentioned this pull request Sep 12, 2023

Grainfx uses significant amounts of DSP even when muted #538

Closed

Uh oh!

Conversation

alter-alter commented Aug 18, 2023

Uh oh!

sichtbeton commented Aug 18, 2023

Uh oh!

sichtbeton commented Aug 18, 2023

Uh oh!

alter-alter commented Aug 18, 2023

Uh oh!

m-m-adams left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

m-m-adams commented Aug 18, 2023

Uh oh!

alter-alter commented Aug 19, 2023

Uh oh!

m-m-adams commented Aug 19, 2023

Uh oh!

alter-alter commented Aug 20, 2023

Uh oh!

m-m-adams left a comment

Choose a reason for hiding this comment

Uh oh!

m-m-adams Aug 20, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

m-m-adams commented Aug 20, 2023

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

m-m-adams Aug 20, 2023 •

edited

Loading