SPI transmit only

[robzed]

I'm working with an LCD that you can only transmit to:

1 lcd_clk spi-pin \ SPI1 SCK
1 lcd_din spi-pin \ SPI1 TX
1 master-spi
10000000 1 spi-baud!
8 1 spi-data-size!

false false 1 motorola-spi
1 enable-spi

When transmitting a byte can I just do:
( n -- ) 1 >spi

Or do I need to drop the incoming buyte?
( n -- ) 1 >spi 1 spi> drop

[robzed]

Also I write 25600 bytes to >SPI and it takes almost 0.5s ...

lcd_buf_size . 25600 ok

\ Write a set of bytes to the SPI port
: spi_write ( address n -- )
0 do
dup C@ 1 >spi 1+
loop drop
;

It appears to take 458ms?
(mecrisp using bit bashing took 83ms - haven't timed zeptforth with bitbashing)

: lcd.display ( -- )
0 lcd_height 1- 0 lcd_width 1- lcd.SetWindows
1 lcd_dc pin!
0 lcd_cs pin!
us-counter-lsb
lcd_buffer lcd_buf_size spi_write
us-counter-lsb swap - u. cr

1 lcd_cs pin!
;
: test_fill
lcd.init
BLUE lcd.fill lcd.display
1000 ms
RED lcd.fill lcd.display
1000 ms
GREEN lcd.fill lcd.display
1000 ms
BLACK lcd.fill lcd.display
;

test_fill 458064
458147
457919
457802
ok

[robzed]

Software SPI (bitbash) version in Zeptoforth

test_fill 411605
411624
411907
409925

[robzed]

Maybe I should try buffer>spi

[tabemann]

I would definitely say try buffer>spi - just using >spi and spi> (they need to pair up) has a lot of extra overhead.

[tabemann]

BTW, have you used spi-baud! to set the baud for SPI?

[tabemann]

Forget that, I missed your earlier post.

[tabemann]

Just deleted my post, thinking I got something wrong in my math, but at second thought it looks right so I'm reposting this:

I wrote the following code:

spi import
timer import

\ 1 lcd_clk spi-pin \ SPI1 SCK
\ 1 lcd_din spi-pin \ SPI1 TX
1 master-spi
10000000 1 spi-baud!
8 1 spi-data-size!

false false 1 motorola-spi
1 enable-spi

25600 constant lcd_buf_size
lcd_buf_size buffer: lcd_buf

lcd_buf lcd_buf_size 0 fill

: spi_write ( address n -- )
0 do
dup C@ 1 >spi 1+ 1 spi> drop
loop drop
;

: time-it ( -- bitrate )
  us-counter-lsb { start }
  lcd_buf lcd_buf_size spi_write
  us-counter-lsb { end }
  end start - 1000000 swap / lcd_buf_size * 8 *
;

: time-it-again ( -- bitrate )
  us-counter-lsb { start }
  lcd_buf lcd_buf_size 1 buffer>spi
  us-counter-lsb { end }
  end start - 1000000 swap / lcd_buf_size * 8 *
;

and got the following result:

time-it-again . 1024000  ok
time-it . 614400  ok

i.e. when using >spi paired with spi> I got 614400 b/s and when I used buffer>spi I got 1024000 b/s.

[robzed]

Ok, so I must have a spi> and drop go every >spi

[tabemann]

I wonder if there's an issue with the clock scalers in use, i.e. why your bitbashed version wasn't much faster than the version using >spi.

[tabemann]

Okay, I did some calculations and got the following:

10000000 find-spi-prescale . 2  ok
10000000 2 find-spi-postdiv . 7  ok
125000000,0 2,0 7,0 1,0 d+ f* f/ f. 7812500,0  ok

i.e. it is really trying to do 7812500 baud, yet the software cannot keep up with the full speed and is instead limited to 1024000 baud. I could potentially rewrite this code to use DMA or PIO or like.

[tabemann]

I managed to optimize buffer>spi and spi>buffer quite a bit, speeding up buffer>spi by approximately a factor of 4.5:

time-it-again . 4505600  ok

Specifically, it takes 43.977 ms to now send the data.

[tabemann]

The code for this is in both the build and devel branches but is not currently in any builds.

[robzed]

Wow you have been busy!!! I will try this out later

[robzed]

Unmodified buffer>spi I get around 216915 for my text (216ms) - which is about twice as fast as my >spi one.

Haven't tried the updated one yet

[robzed]

Times in microseconds for buffer copy: 43877 43788 43788 43762
Faster than mecrisp bitbash!

Buffer fill and write to screen (with coords) 46813 46722 46690 46722.
That's about 21 frames per seconds.

[tabemann]

And that's without using DMA too - I tried using DMA in hopes of pushing the bandwidth to the maximum of what the SPI peripheral theoretically supports but ran into issues with it, so I had to fall back on a highly optimized tight loop instead.