Re: [Csnd-dev] [EXTERNAL] [Csnd-dev] sysdep.h and MAXLEN for ftables

For multichannel, you can use one table per channel instead of interleaved and there's no loss in length.

It looks like 2^31 is possible without having to edit the full codebase.

Prof. Victor Lazzarini
Maynooth University
Ireland

> On 2 Oct 2024, at 01:14, joachim heintz <[email protected]> wrote:
> 
> but what giovanni mentioned as multi-channel import should be considered.  if we import interleaved, we can hit the limit more easy. so if it is safe to extend to 2^32 as maximum length, it might be good.
> 
>> On 02/10/2024 00:46, Victor Lazzarini wrote:
>> I figured it out:
>> 1) Starting with 6.18.1 on Mac I released universal binaries
>> 2) The check for B64BIT was only picking up the x86_64 nor arm64
>> The result was that for  i1 ftgen 1, 0, 2^28, 2, 0, 2^28, 0
>> 1) on arm64 (Csound 6.18.1)
>> /usr/local/bin/csound test.csd -n
>> INIT ERROR in instr 0 (opcode ftgen) line 8: ftgen error
>>  from file test.csd (1),i1 ftgen 1 0 268435456 2 0 268435456 0
>> 2) on x86_64 (Csound 6.18.1)
>> arch -x86_64 /usr/local/bin/csound test.csd -n
>> ftable 1: 268435456 points, scalemax 1.000
>> So for 6.18.1, only the x86_64 allows up to 2^28-length tables.
>> With Steven’s fix, Csound 7 is building with B64BIT on arm64. But I think this changes does not need to be limited to 64bit platforms.
>> We probably can roll it out to all platforms.
>>>> On 1 Oct 2024, at 23:12, vlz <[email protected]> wrote:
>>> 
>>> I think the check was broken, Steven seems to have fixed it.
>>> 
>>>> On 1 Oct 2024, at 23:06, vlz <[email protected]> wrote:
>>>> 
>>>> Then for some reason in the build B64BIT is not defined. If that’s defined you can do
>>>> 
>>>> i1 ftgen 1, 0, 2^28, 2, 0, 2^28, 0
>>>> 
>>>> We need to check that.
>>>> 
>>>>> On 1 Oct 2024, at 22:47, Giovanni Bedetti <[email protected]> wrote:
>>>>> 
>>>>> I tried to create a table (using the API, compileOrc and CopyTableIn) after reading samples from a stereo file (so the table will hold both channels, interleaved): this works: audioClip samples 4549938x2 = 9099876, length (seconds): 103.1732, sr: 44100
>>>>> this doesn't: audioClip samples 10612538x2 = 21225076, length (seconds): 240.6471, sr: 44100
>>>>> with this log:
>>>>> illegal table length
>>>>> INIT ERROR in instr 0 (opcode ftgen) line 11: ftgen error gisampletable100 ftgen 100 0 -21225076 -7 0 0
>>>>> 
>>>>> I'm still using Csound6 (6.18) on CsoundUnity on macOS
>>>>> 
>>>>> Il giorno mar 1 ott 2024 alle ore 21:58 Victor Lazzarini <[email protected]> ha scritto:
>>>>> That’s if B64BIT is not defined. Actually I don’t think we need to check for this.
>>>>> ========================
>>>>> Prof. Victor Lazzarini
>>>>> Maynooth University
>>>>> Ireland
>>>>> 
>>>>>> On 1 Oct 2024, at 20:20, Steven Yi <[email protected]> wrote:
>>>>>> 
>>>>>> I was going off what's documented in the manual (https://csound.com/docs/manual/f.html)
>>>>>> 
>>>>>> "p3 -- Size of function table (i.e. number of points) Must be a power of 2, or a power-of-2 plus 1 if this number is positive. Maximum table size is 16777216 (224) points."
>>>>>> 
>>>>>> I think Giovanni ran into issues with longer sized tables but I don't recall how long.
>>>>>> 
>>>>>> On Tue, Oct 1, 2024 at 3:00 PM Victor Lazzarini <[email protected]> wrote:
>>>>>> MAXLEN is 0x10000000 for most modern archs, that's 2^28 is it not? That translates into
>>>>>> about 101 minutes of audio at 44100.
>>>>>> 
>>>>>> It does not strike me as too short. I think we could get away with 2^32 without having to do a major overhaul, but I need to check if my idea works correctly. That would allows 1603 minutes or 27 hours of audio.
>>>>>> 
>>>>>> Prof. Victor Lazzarini
>>>>>> Maynooth University
>>>>>> Ireland
>>>>>> 
>>>>>>> On 1 Oct 2024, at 13:36, Steven Yi <[email protected]> wrote:
>>>>>>> 
>>>>>>>  *Warning*
>>>>>>> This email originated from outside of Maynooth University's Mail System. Do not reply, click links or open attachments unless you recognise the sender and know the content is safe.
>>>>>>> Hi All,
>>>>>>> 
>>>>>>> I was chatting with Giovanni Bedetti on Discord about max table length. It seems awfully small at 2^24 (~16MB). Any reason not to bump that up to something larger?
>>>>>>> 
>>>>>>> Also, while looking at that, I noticed B64BIT detection in sysdep.h. I saw that it was missing aarch64 amongst other things, so I've submitted a PR to update this.
>>>>>>> 
>>>>>>> steven
>> ========================
>> Prof. Victor Lazzarini
>> Maynooth University
>> Ireland