Sound overhaul for TKG - Page 7

Karlos · 05 July 2024, 19:57

Can you run again with USE060 to force it down the multiplication path, but skip the lookup?

abu_the_monkey · 05 July 2024, 20:01

two ticks

abu_the_monkey · 05 July 2024, 20:10

as requested.

Karlos · 05 July 2024, 20:12

It feels like this might be too slow already:

1549660/709379 = 2.18453s to mix 189 packets, or 11.56ms/packet.

Under the setup configuration, that's 11.56ms every 20ms, using a 50Hz update rate. Having said that, it's also a worst case, all 16 channels running. And I am still using multiplication based normalisation, so that's up to 640 multiplications per stereo packet.

abu_the_monkey · 05 July 2024, 20:14

it's also the worst hardware setup.
all original commodore, no local fast ram on the cpu card, etc

Karlos · 05 July 2024, 20:14

I will have to do what @paraj suggested and make this a set of tests that run automatically.

Each path - including a linear (non-delta) version for 040, tested with 0-16 channels each.

We can always limit the channel count on slower machines.

Will you have access to the machine for long?

abu_the_monkey · 05 July 2024, 20:16

yes, it's just a pain to work on and my time is limited at best

Karlos · 05 July 2024, 20:17

Quote:

Originally Posted by abu_the_monkey

yes, it's just a pain to work on and my time is limited at best

I better hurry then!

abu_the_monkey · 05 July 2024, 20:18

Quote:

Originally Posted by Karlos

I better hurry then!

abu_the_monkey · 05 July 2024, 20:21

send me a PM when you have something to test and I will see what I can do

paraj · 05 July 2024, 20:28

Great with some 040 numbers finally, thanks abu! But am I reading it right that it's about as slow as 060 w/o D$? Much more cleverness and/or cheating will be needed then..

Karlos · 05 July 2024, 20:32

Quote:

Originally Posted by paraj

Great with some 040 numbers finally, thanks abu! But am I reading it right that it's about as slow as 060 w/o D$? Much more cleverness and/or cheating will be needed then..

We don't have the worst of it yet. I am working on it...

Karlos · 05 July 2024, 20:42

@abu & @paraj - sorry to be a git, can you get the latest binary from here: https://github.com/0xABADCAFE/tkg-mixer and run 3 tests?

mixer linear
mixer
mixer use060

Each one will do 1-16 channels. I will do a more proper structured test when I have time, but this will give us some useful numbers.

Lol, I have an off by one bug it only does up to 15 channels. One sec.

Karlos · 05 July 2024, 20:47

Fixed. Go!

Karlos · 05 July 2024, 20:53

Quote:

Originally Posted by paraj

Great with some 040 numbers finally, thanks abu! But am I reading it right that it's about as slow as 060 w/o D$? Much more cleverness and/or cheating will be needed then..

The whole assumption that delta lookups are faster needs to be tested. The latest binary has a linear version - i.e. the original as it was before adding delta. I have my own numbers for 1-16 channels but they are totally unscientific, coming from UAE as cycle exact doesn't mean much for 68040 emulation.

paraj · 05 July 2024, 20:57

Results (2nd run w/o D$ included):

Code:

cpu >> test.txt
mixer linear >> test.txt
mixer >> test.txt
mixer use060 >> test.txt



System: 68040 68882 (INST: Cache Burst) (DATA: Cache CopyBack)
Using 68040 linear lookup code path
Got Timer, frequency is 709379 Hz
Loaded sounds/airstrike.raw [60460 bytes] at 0x68bf8a10
Testing with 1 channel(s)
Mixed 189 Packets in 51021 EClockVal ticks (709379/s)
Testing with 2 channel(s)
Mixed 189 Packets in 123965 EClockVal ticks (709379/s)
Testing with 3 channel(s)
Mixed 189 Packets in 215264 EClockVal ticks (709379/s)
Testing with 4 channel(s)
Mixed 189 Packets in 324937 EClockVal ticks (709379/s)
Testing with 5 channel(s)
Mixed 189 Packets in 452998 EClockVal ticks (709379/s)
Testing with 6 channel(s)
Mixed 189 Packets in 600165 EClockVal ticks (709379/s)
Testing with 7 channel(s)
Mixed 189 Packets in 768720 EClockVal ticks (709379/s)
Testing with 8 channel(s)
Mixed 189 Packets in 957028 EClockVal ticks (709379/s)
Testing with 9 channel(s)
Mixed 189 Packets in 1164564 EClockVal ticks (709379/s)
Testing with 10 channel(s)
Mixed 189 Packets in 1389403 EClockVal ticks (709379/s)
Testing with 11 channel(s)
Mixed 189 Packets in 1633406 EClockVal ticks (709379/s)
Testing with 12 channel(s)
Mixed 189 Packets in 1895738 EClockVal ticks (709379/s)
Testing with 13 channel(s)
Mixed 189 Packets in 2176737 EClockVal ticks (709379/s)
Testing with 14 channel(s)
Mixed 189 Packets in 2475481 EClockVal ticks (709379/s)
Testing with 15 channel(s)
Mixed 189 Packets in 2793057 EClockVal ticks (709379/s)
Testing with 16 channel(s)
Mixed 189 Packets in 3122699 EClockVal ticks (709379/s)
Using 68040 delta lookup code path
Got Timer, frequency is 709379 Hz
Loaded sounds/airstrike.raw [60460 bytes] at 0x68bf8a10
Testing with 1 channel(s)
Mixed 189 Packets in 51772 EClockVal ticks (709379/s)
Testing with 2 channel(s)
Mixed 189 Packets in 126929 EClockVal ticks (709379/s)
Testing with 3 channel(s)
Mixed 189 Packets in 221924 EClockVal ticks (709379/s)
Testing with 4 channel(s)
Mixed 189 Packets in 336889 EClockVal ticks (709379/s)
Testing with 5 channel(s)
Mixed 189 Packets in 471496 EClockVal ticks (709379/s)
Testing with 6 channel(s)
Mixed 189 Packets in 627027 EClockVal ticks (709379/s)
Testing with 7 channel(s)
Mixed 189 Packets in 803925 EClockVal ticks (709379/s)
Testing with 8 channel(s)
Mixed 189 Packets in 1003656 EClockVal ticks (709379/s)
Testing with 9 channel(s)
Mixed 189 Packets in 1222200 EClockVal ticks (709379/s)
Testing with 10 channel(s)
Mixed 189 Packets in 1462542 EClockVal ticks (709379/s)
Testing with 11 channel(s)
Mixed 189 Packets in 1722122 EClockVal ticks (709379/s)
Testing with 12 channel(s)
Mixed 189 Packets in 2001524 EClockVal ticks (709379/s)
Testing with 13 channel(s)
Mixed 189 Packets in 2300132 EClockVal ticks (709379/s)
Testing with 14 channel(s)
Mixed 189 Packets in 2619335 EClockVal ticks (709379/s)
Testing with 15 channel(s)
Mixed 189 Packets in 2957387 EClockVal ticks (709379/s)
Testing with 16 channel(s)
Mixed 189 Packets in 3308640 EClockVal ticks (709379/s)
Using 68060 code path
Got Timer, frequency is 709379 Hz
Loaded sounds/airstrike.raw [60460 bytes] at 0x68bf8a10
Testing with 1 channel(s)
Mixed 189 Packets in 47705 EClockVal ticks (709379/s)
Testing with 2 channel(s)
Mixed 189 Packets in 113150 EClockVal ticks (709379/s)
Testing with 3 channel(s)
Mixed 189 Packets in 192903 EClockVal ticks (709379/s)
Testing with 4 channel(s)
Mixed 189 Packets in 286925 EClockVal ticks (709379/s)
Testing with 5 channel(s)
Mixed 189 Packets in 395263 EClockVal ticks (709379/s)
Testing with 6 channel(s)
Mixed 189 Packets in 517948 EClockVal ticks (709379/s)
Testing with 7 channel(s)
Mixed 189 Packets in 654865 EClockVal ticks (709379/s)
Testing with 8 channel(s)
Mixed 189 Packets in 806486 EClockVal ticks (709379/s)
Testing with 9 channel(s)
Mixed 189 Packets in 971725 EClockVal ticks (709379/s)
Testing with 10 channel(s)
Mixed 189 Packets in 1151325 EClockVal ticks (709379/s)
Testing with 11 channel(s)
Mixed 189 Packets in 1345978 EClockVal ticks (709379/s)
Testing with 12 channel(s)
Mixed 189 Packets in 1556008 EClockVal ticks (709379/s)
Testing with 13 channel(s)
Mixed 189 Packets in 1779039 EClockVal ticks (709379/s)
Testing with 14 channel(s)
Mixed 189 Packets in 2015664 EClockVal ticks (709379/s)
Testing with 15 channel(s)
Mixed 189 Packets in 2266452 EClockVal ticks (709379/s)
Testing with 16 channel(s)
Mixed 189 Packets in 2527587 EClockVal ticks (709379/s)
System: 68040 68882 (INST: Cache Burst) (DATA: NoCache NoBurst)
Using 68040 linear lookup code path
Got Timer, frequency is 709379 Hz
Loaded sounds/airstrike.raw [60460 bytes] at 0x68bf8a10
Testing with 1 channel(s)
Mixed 189 Packets in 130134 EClockVal ticks (709379/s)
Testing with 2 channel(s)
Mixed 189 Packets in 343012 EClockVal ticks (709379/s)
Testing with 3 channel(s)
Mixed 189 Packets in 633512 EClockVal ticks (709379/s)
Testing with 4 channel(s)
Mixed 189 Packets in 999706 EClockVal ticks (709379/s)
Testing with 5 channel(s)
Mixed 189 Packets in 1444709 EClockVal ticks (709379/s)
Testing with 6 channel(s)
Mixed 189 Packets in 1963594 EClockVal ticks (709379/s)
Testing with 7 channel(s)
Mixed 189 Packets in 2558842 EClockVal ticks (709379/s)
Testing with 8 channel(s)
Mixed 189 Packets in 3230893 EClockVal ticks (709379/s)
Testing with 9 channel(s)
Mixed 189 Packets in 3978930 EClockVal ticks (709379/s)
Testing with 10 channel(s)
Mixed 189 Packets in 4802723 EClockVal ticks (709379/s)
Testing with 11 channel(s)
Mixed 189 Packets in 5704471 EClockVal ticks (709379/s)
Testing with 12 channel(s)
Mixed 189 Packets in 6680380 EClockVal ticks (709379/s)
Testing with 13 channel(s)
Mixed 189 Packets in 7732476 EClockVal ticks (709379/s)
Testing with 14 channel(s)
Mixed 189 Packets in 8862389 EClockVal ticks (709379/s)
Testing with 15 channel(s)
Mixed 189 Packets in 10064878 EClockVal ticks (709379/s)
Testing with 16 channel(s)
Mixed 189 Packets in 11308520 EClockVal ticks (709379/s)
Using 68040 delta lookup code path
Got Timer, frequency is 709379 Hz
Loaded sounds/airstrike.raw [60460 bytes] at 0x68bf8a10
Testing with 1 channel(s)
Mixed 189 Packets in 130742 EClockVal ticks (709379/s)
Testing with 2 channel(s)
Mixed 189 Packets in 346249 EClockVal ticks (709379/s)
Testing with 3 channel(s)
Mixed 189 Packets in 642202 EClockVal ticks (709379/s)
Testing with 4 channel(s)
Mixed 189 Packets in 1014234 EClockVal ticks (709379/s)
Testing with 5 channel(s)
Mixed 189 Packets in 1463917 EClockVal ticks (709379/s)
Testing with 6 channel(s)
Mixed 189 Packets in 1991654 EClockVal ticks (709379/s)
Testing with 7 channel(s)
Mixed 189 Packets in 2598109 EClockVal ticks (709379/s)
Testing with 8 channel(s)
Mixed 189 Packets in 3281854 EClockVal ticks (709379/s)
Testing with 9 channel(s)
Mixed 189 Packets in 4043272 EClockVal ticks (709379/s)
Testing with 10 channel(s)
Mixed 189 Packets in 4884476 EClockVal ticks (709379/s)
Testing with 11 channel(s)
Mixed 189 Packets in 5800665 EClockVal ticks (709379/s)
Testing with 12 channel(s)
Mixed 189 Packets in 6794944 EClockVal ticks (709379/s)
Testing with 13 channel(s)
Mixed 189 Packets in 7869235 EClockVal ticks (709379/s)
Testing with 14 channel(s)
Mixed 189 Packets in 9019214 EClockVal ticks (709379/s)
Testing with 15 channel(s)
Mixed 189 Packets in 10245139 EClockVal ticks (709379/s)
Testing with 16 channel(s)
Mixed 189 Packets in 11514676 EClockVal ticks (709379/s)
Using 68060 code path
Got Timer, frequency is 709379 Hz
Loaded sounds/airstrike.raw [60460 bytes] at 0x68bf8a10
Testing with 1 channel(s)
Mixed 189 Packets in 119500 EClockVal ticks (709379/s)
Testing with 2 channel(s)
Mixed 189 Packets in 304367 EClockVal ticks (709379/s)
Testing with 3 channel(s)
Mixed 189 Packets in 548008 EClockVal ticks (709379/s)
Testing with 4 channel(s)
Mixed 189 Packets in 850057 EClockVal ticks (709379/s)
Testing with 5 channel(s)
Mixed 189 Packets in 1211362 EClockVal ticks (709379/s)
Testing with 6 channel(s)
Mixed 189 Packets in 1631729 EClockVal ticks (709379/s)
Testing with 7 channel(s)
Mixed 189 Packets in 2109590 EClockVal ticks (709379/s)
Testing with 8 channel(s)
Mixed 189 Packets in 2646495 EClockVal ticks (709379/s)
Testing with 9 channel(s)
Mixed 189 Packets in 3243626 EClockVal ticks (709379/s)
Testing with 10 channel(s)
Mixed 189 Packets in 3896928 EClockVal ticks (709379/s)
Testing with 11 channel(s)
Mixed 189 Packets in 4608026 EClockVal ticks (709379/s)
Testing with 12 channel(s)
Mixed 189 Packets in 5378123 EClockVal ticks (709379/s)
Testing with 13 channel(s)
Mixed 189 Packets in 6206602 EClockVal ticks (709379/s)
Testing with 14 channel(s)
Mixed 189 Packets in 7094838 EClockVal ticks (709379/s)
Testing with 15 channel(s)
Mixed 189 Packets in 8038441 EClockVal ticks (709379/s)
Testing with 16 channel(s)
Mixed 189 Packets in 9014083 EClockVal ticks (709379/s)

abu_the_monkey · 05 July 2024, 21:16

not as neat as @paraj presentation, but, as requested

Karlos · 05 July 2024, 21:29

Thanks, mentlegen!

I will extract the numbers and graph later.

As a reward for your cooperation you are now required to playtest level C lol

It looks like the delta model is a tiny bit better on this 040 but not at all by much. I think the cost of the extra instructions is quite high relative to the (presumably) increased cache hit ratio for the lookup.

Would be nice to see for a better 040 as the problems could be in the basic data fetches in the first place.

abu_the_monkey · 05 July 2024, 21:47

Quote:

Originally Posted by Karlos

Would be nice to see for a better 040 as the problems could be in the basic data fetches in the first place.

this.

and playing through the new and improved level C is on my radar

Karlos · 05 July 2024, 22:07

Quote:

Originally Posted by abu_the_monkey

this.

and playing through the new and improved level C is on my radar

You should play the original t17 one first, to appreciate the blood sweat and tears lol

05 July 2024, 20:10	#123
abu_the_monkey Registered User Join Date: Oct 2020 Location: Bicester Posts: 2,039	as requested. Attached Thumbnails

05 July 2024, 21:16	#137
abu_the_monkey Registered User Join Date: Oct 2020 Location: Bicester Posts: 2,039	not as neat as @paraj presentation, but, as requested Attached Thumbnails

05 July 2024, 21:29	#138
Karlos Alien Bleed Join Date: Aug 2022 Location: UK Posts: 4,585	Thanks, mentlegen! I will extract the numbers and graph later. As a reward for your cooperation you are now required to playtest level C lol It looks like the delta model is a tiny bit better on this 040 but not at all by much. I think the cost of the extra instructions is quite high relative to the (presumably) increased cache hit ratio for the lookup. Would be nice to see for a better 040 as the problems could be in the basic data fetches in the first place. Last edited by Karlos; 05 July 2024 at 21:34.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Slow A4000 after overhaul	Screechstar	support.Hardware	57	11 July 2023 23:02
Amiga Font Editor overhaul	buggs	Coders. Releases	19	09 March 2021 17:39
Escom A1200 overhaul	Ox.	Amiga scene	8	26 August 2014 08:54
Will Bridge Practice series needs an overhaul	mk1	HOL data problems	1	02 April 2009 21:55

05 July 2024, 19:57	#121
Karlos Alien Bleed Join Date: Aug 2022 Location: UK Posts: 4,585	Can you run again with USE060 to force it down the multiplication path, but skip the lookup?

05 July 2024, 20:01	#122
abu_the_monkey Registered User Join Date: Oct 2020 Location: Bicester Posts: 2,039	two ticks

05 July 2024, 20:12	#124
Karlos Alien Bleed Join Date: Aug 2022 Location: UK Posts: 4,585	It feels like this might be too slow already: 1549660/709379 = 2.18453s to mix 189 packets, or 11.56ms/packet. Under the setup configuration, that's 11.56ms every 20ms, using a 50Hz update rate. Having said that, it's also a worst case, all 16 channels running. And I am still using multiplication based normalisation, so that's up to 640 multiplications per stereo packet.

05 July 2024, 20:14	#125
abu_the_monkey Registered User Join Date: Oct 2020 Location: Bicester Posts: 2,039	it's also the worst hardware setup. all original commodore, no local fast ram on the cpu card, etc

05 July 2024, 20:14	#126
Karlos Alien Bleed Join Date: Aug 2022 Location: UK Posts: 4,585	I will have to do what @paraj suggested and make this a set of tests that run automatically. Each path - including a linear (non-delta) version for 040, tested with 0-16 channels each. We can always limit the channel count on slower machines. Will you have access to the machine for long?

05 July 2024, 20:16	#127
abu_the_monkey Registered User Join Date: Oct 2020 Location: Bicester Posts: 2,039	yes, it's just a pain to work on and my time is limited at best

05 July 2024, 20:21	#130
abu_the_monkey Registered User Join Date: Oct 2020 Location: Bicester Posts: 2,039	send me a PM when you have something to test and I will see what I can do

05 July 2024, 20:28	#131
paraj Registered User Join Date: Feb 2017 Location: Denmark Posts: 1,245	Great with some 040 numbers finally, thanks abu! But am I reading it right that it's about as slow as 060 w/o D$? Much more cleverness and/or cheating will be needed then..

05 July 2024, 20:42	#133
Karlos Alien Bleed Join Date: Aug 2022 Location: UK Posts: 4,585	@abu & @paraj - sorry to be a git, can you get the latest binary from here: https://github.com/0xABADCAFE/tkg-mixer and run 3 tests? mixer linear mixer mixer use060 Each one will do 1-16 channels. I will do a more proper structured test when I have time, but this will give us some useful numbers. Lol, I have an off by one bug it only does up to 15 channels. One sec.

05 July 2024, 20:47	#134
Karlos Alien Bleed Join Date: Aug 2022 Location: UK Posts: 4,585	Fixed. Go!

Currently Active Users Viewing This Thread: 2 (0 members and 2 guests)