I2s experimental #2501

Palatis · 2016-09-11T00:57:55Z

various fixes to the i2s subsystem.

add missing convenient function (i2s_write_lr_nb())
fix memory leak in i2s_slc_begin()
sample rate handling refactored

we `malloc()` memories in `i2s_slc_begin()`, but never `free()` them. so the cpu runs out of memory after several `i2s_begin(); i2s_end();` so we add a `i2s_init()` to do the `malloc()` and `i2s_deinit()` to `free()` those allocated buffers.

- do a search to find the closest divider for optimal performance. - use the current cpu clock to get the correct sample rate. - only set the sample rate if it's changed

codecov-io · 2016-09-11T01:16:31Z

Current coverage is 27.80% (diff: 100%)

Merging #2501 into master will not change coverage

@@             master      #2501   diff @@
==========================================
  Files            20         20          
  Lines          3625       3625          
  Methods         335        335          
  Messages          0          0          
  Branches        656        656          
==========================================
  Hits           1008       1008          
  Misses         2441       2441          
  Partials        176        176

Powered by Codecov. Last update 4897e00...6a3a77c

return the underlying buffer used by I2S SLC DMA subsystem, so client code can write data directly into it, instead of calling i2s_write_sample() / i2s_write_sample_nb(). greatly boost the performance, and reduce memory footprint when using an audio codec. because the codec can now decode the stream directly into the buffer, instead of cache them and doing i2s_write_sample{,_nb}().

me-no-dev · 2016-09-13T05:34:06Z

cores/esp8266/core_esp8266_i2s.c

@@ -194,46 +219,94 @@ bool ICACHE_FLASH_ATTR i2s_write_lr(int16_t left, int16_t right){
  return i2s_write_sample(sample);
 }

+bool ICACHE_FLASH_ATTR i2s_write_lr_nb(int16_t left, int16_t right){
+  int sample = right & 0xFFFF;


why signed integer here and not uint32_t?

i2s_write_lr() does use int32_t, I copied the code and only changed that i2s_write_sample(sample); to i2s_write_sample_nb(sample);

Palatis · 2016-09-13T07:54:39Z

the clock divider guessing code is copied from mp3 example.
it use 160Mhz statically.

the modifications I made are reading runtime clock, and do not allow bit mode (like sending with 19 bit or 20 bit per sample, and let the DAC output at a lower volume.)

I'm still waiting for my hw college to provide me with a PCB with a working i2s DAC and amp, tho.

i2s_write_lr() takes two int16_t, so I think the non-blocking version should take signed as well.

Palatis · 2016-09-13T08:01:48Z

as for the i2s base clock, I confirmed that the samples are pushed out at almost correct rate, such as RATE_HZ samples pushed out around ~1015ms.

I think some rounding error and dma buffer consume speed and maybe some random weirdness are acceptable......

me-no-dev · 2016-09-13T21:07:24Z

I have also started writing this lib from the mp3 example ;) did many calculations and came up that the formula I used was the fastest and easiest option, but please prove me wrong :) we are people and make mistakes.
your left and right channels are signed 16 bit values, but both combined, it's unsigned 32 bit value ;)
do not remember if I run the tests at both 80 and 160 MHz, but I would think that I have and that is why I asked about the base clock in your calculations

Palatis · 2016-09-14T10:03:32Z

I will provide a easy sketch to test the sample rate sometime later.
Not at home right now.

the i2s buffer was fixed to 8*64 uint32_t, this may be inefficient for some codecs. by allowing the user to adjust the buffer size, the codec can now decode the whole window directly into the acquired buffer, thus avoid a lot of memory copy.

earlephilhower · 2018-01-15T21:41:46Z

Hi @Palatis I see this has been sitting in the backlog for two years, ouch. Are you still interested in getting this patch into 2.5 or 2.6 of the Arduino SDK? There was a massive cleanup this past release and we're trying to be more methodical in getting these PRs reviewed and accepted in a reasonable process and time.

If you're still interested, please update the conflicts and I'll do a review ASAP.

Thx
-EFP3

Palatis · 2018-01-16T12:27:51Z

i'm still willing to help, but i've shifted onto some other things since then.
I'm currently working in Shenzhen, but all my esp8266 (I'm referring to both hw and sw...) are at home in Taipei...

earlephilhower · 2018-01-17T04:21:07Z

No worries, don't want to put you out. We can shelve this for now and if you ever want to get back to it we can re-open it. Sound OK?

Crypter · 2018-01-21T01:32:12Z

I'll take a look in the following days, there sure must be something we can use and clean up this pull request too :)

earlephilhower · 2018-10-04T17:14:08Z

@Crypter and @Palatis, since the I2S subsystem has been rewritten to support bidirectional transfer and the method for choosing the proper clock setup for a given frequency has also been redone, I think this PR should be closed.

The PR is pretty much unmergeable now, and the only thing I'm seeing is missing is way of writing multiple samples-worth of data to the buffer (which is basically what the get/put buffer calls that were added let you do). That's a much smaller PR and should be started against the current head to preserve everyone's sanity. If I ever get time to get back to working on my ESP8266Audio library (which obviously uses the I2S extensively), adding that capability (uint32_t i2s_write_block/non(uint16_t samples[], int count));) is something I'll definitely look into as the overhead of calling the I2S add-sample call 44,000 times per second can't be small.

Palatis added 4 commits September 11, 2016 08:54

fix whitespace in i2s stuff

a01ccd9

i2s: add non-blocking version of i2s_write_lr()

ff67578

i2s: properly configure sample rate

da2ea6f

- do a search to find the closest divider for optimal performance. - use the current cpu clock to get the correct sample rate. - only set the sample rate if it's changed

me-no-dev reviewed Sep 13, 2016
View reviewed changes

Palatis added 2 commits September 20, 2016 23:52

allow user to adjust the size of i2s buffer

6b3dca9

the i2s buffer was fixed to 8*64 uint32_t, this may be inefficient for some codecs. by allowing the user to adjust the buffer size, the codec can now decode the whole window directly into the acquired buffer, thus avoid a lot of memory copy.

make sample variable in i2s_write_lr*() uint32_t

6a3a77c

devyte assigned earlephilhower Jan 10, 2018

devyte added component: core type: enhancement labels Jan 10, 2018

earlephilhower added this to the 2.5.0 milestone Mar 9, 2018

earlephilhower closed this Oct 4, 2018

earlephilhower mentioned this pull request Oct 4, 2018

Enhancement - Add support for writing/reading multiple I2S samples #5206

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

I2s experimental #2501

I2s experimental #2501

Palatis commented Sep 11, 2016 •

edited

Loading

codecov-io commented Sep 11, 2016 •

edited

Loading

me-no-dev Sep 13, 2016

Palatis Sep 20, 2016

Palatis commented Sep 13, 2016

Palatis commented Sep 13, 2016 •

edited

Loading

me-no-dev commented Sep 13, 2016

Palatis commented Sep 14, 2016

earlephilhower commented Jan 15, 2018

Palatis commented Jan 16, 2018

earlephilhower commented Jan 17, 2018

Crypter commented Jan 21, 2018

earlephilhower commented Oct 4, 2018

I2s experimental #2501

I2s experimental #2501

Conversation

Palatis commented Sep 11, 2016 • edited Loading

codecov-io commented Sep 11, 2016 • edited Loading

Current coverage is 27.80% (diff: 100%)

me-no-dev Sep 13, 2016

Choose a reason for hiding this comment

Palatis Sep 20, 2016

Choose a reason for hiding this comment

Palatis commented Sep 13, 2016

Palatis commented Sep 13, 2016 • edited Loading

me-no-dev commented Sep 13, 2016

Palatis commented Sep 14, 2016

earlephilhower commented Jan 15, 2018

Palatis commented Jan 16, 2018

earlephilhower commented Jan 17, 2018

Crypter commented Jan 21, 2018

earlephilhower commented Oct 4, 2018

Palatis commented Sep 11, 2016 •

edited

Loading

codecov-io commented Sep 11, 2016 •

edited

Loading

Palatis commented Sep 13, 2016 •

edited

Loading