Optimize graphical display with selective rendering #5288

thinkyhead · 2016-11-24T00:33:43Z

The draw loop wastes a lot of cycles because it re-calculates the whole display 2 or more times per screen render. This PR employs two strategies to reduce rendering time:

Add bounds-checking so that only items within the current page being rendered are drawn.
For XYZ —overlapping both halves of the screen— the value strings are cached.
Use a frame with normal text instead of a solid box with inverse text.
Reduce the delay between bytes sent over SPI by 5-6µs.

Added up, these optimizations save at least ~19ms (380,000 cycles) on screens with 2 stripes.

Additionally:

Use uint8_t for u8g coordinates. (u8g also supplies a u8g_uint_t type)
Use const for function arguments and locals, where appropriate.

I look forward to your test results!

Sebastianv650 · 2016-11-24T15:53:26Z

This one is better than #5279: It saves 2ms (82.5 to 80.5ms) on my printer.

thinkyhead · 2016-11-24T21:38:05Z

2ms is significant! That's 32,000 cycles.

@olikraus suggests we can do even better with u8g2. But it is still a work-in-progress…

Please read olikraus/u8glib#447 (comment)

thinkyhead · 2016-11-24T21:42:38Z

Also @yhfudev take note! U8g2 apparently has some helpful features that may make your work easier. Specifically:

"UTF-8 support, Chinese fonts are already included, limit of 256 glyphs per font removed."

Proper UTF-8 support means we don't have to do so much remapping.

Very fast, LiquidCrystal like char based API (U8x8) if graphics output is not required.

Only the Info screen needs graphics. Other screens can take advantage of this optimization.

olikraus/u8glib#447 (comment)

AnHardt · 2016-11-24T22:19:33Z

Very good!

Todays tests:

Todays test of the GLCD enhacements

RCBugFix Status screen
2 Stripes 77ms
4 Stripes 86ms
8 Stripes 110ms

Selective rendering
2 Stripes 75ms (-2ms)
4 Stripes 76ms (-10ms)
8 Stripes 85ms (-25ms)

Selective + unroll
2 Stripes 78ms (+1ms)

Selective + unroll + decrease delay
2 Stripes 76ms (-1ms)

Selective + decrease delay
2 Stripes 73ms (-4ms)

Detailed timings in the commits comments. (AnHardt#67)

Conclusion:
Selective rendering is a small win for the two stripe ST7920 but makes a huge difference for displays with more stripes. Go for it.
Unrolling the loops seems to make the updates a tiny bit slower. Don't merge.
Decreasing the 10µs-delay to 5Ms speeds up ~2ms per complete update. Today i could not go below 5µs. If we should do that - with the original 10 µseconds and a big fat warning - decrease on your own risk.

ghost · 2016-11-24T23:52:52Z

I also tested.

Selective + 10μs delay	Selective + 5μs delay	Selective + 4μs delay	Selective + 3μs delay
2 Stripes	2 Stripes	2 Stripes	2 Stripes
74 ~ 75ms	73 ~ 74ms	72 ~ 73ms	72 ~ 73ms
~~Displaying is broken~~ It's displayed properly	It's displayed properly ~~, but bit noisy when encoder is rotated~~	It's displayed properly ~~, but bit noisy when encoder is rotated~~	Displaying is broken

~~A branch that used for testing: https://github.com/esenapaj/Marlin/tree/glcd-testes~~

~~And, I created a branch that it doesn't include benchmark and 5μs patch and tested. https://github.com/esenapaj/Marlin/tree/glcd-testes2~~
~~Displaying is broken.~~ It's displayed properly.
~~Video clip:~~

AnHardt · 2016-11-25T00:30:56Z

@thinkyhead
Would you pleas check AnHardt#67 (comment).

AnHardt · 2016-11-25T00:34:18Z

@esenapaj
Please reset the display (shut down power completely) and retest Selective with the normal 10µs delay. Simply rebooting the Arduino sometimes does not work when the delays previously have been too short..

ghost · 2016-11-25T01:04:10Z

@AnHardt
Sorry and thanks for the advice, now I've re-tested and confirmed that both this PR and this PR + 5μs patch works without any problems.

ghost · 2016-11-25T01:36:32Z

@AnHardt
I've tested your "frame" solution.

Selective + 5μs delay + "frame"
2 Stripes
58 ~59ms (-19ms)
It's displayed properly.

Selective + 4μs delay + "frame"
2 Stripes
58 ~59ms (-19ms)
It's displayed properly.

Selective + 3μs delay + "frame"
2 Stripes
57 ~58ms (-20ms)
Displaying is broken.

~~A branch that used for testing: https://github.com/esenapaj/Marlin/tree/glcd-testes3~~

AnHardt · 2016-11-25T01:59:35Z

Tomorrow i'll make some experiments with 'hollow' extruder and sd-card symbols.

thinkyhead · 2016-11-25T03:51:17Z

These are really encouraging results! A savings of 19ms frees up 304,000 CPU cycles. Wow.

Can you also test with the options USE_BIG_EDIT_FONT and USE_SMALL_INFOFONT to make sure the display still works correctly, including menus and edit items?

Also, try the display with and without LCD_SCREEN_ROT_180. Oliver suggested that the page_y0 and page_y1 values might be reversed in that case.

I've added an option DOGM_SPI_DELAY_US for tuning that 10µs delay. It's in Configuration_adv.h.

Drawing the frame instead of a solid box is also very smart. As another alternative I was also thinking of trying pre-clipping the solid box. That is, instead of drawing the whole box, starting at 30 with a height of INFO_FONT_HEIGHT + 2, make sure to only draw the portion that's within the current stripe.

Or… place that part of the display at the very top of the screen instead of in the middle, so it's only ever drawn once (including the XYZ strings).

ghost · 2016-11-25T04:30:44Z

I got a compilation error.

error message

In file included from \AppData\Local\Temp\arduino_build_683885\sketch\ultralcd_impl_DOGM.h:46:0,

                 from \AppData\Local\Temp\arduino_build_683885\sketch\ultralcd.cpp:56:

\AppData\Local\Temp\arduino_build_683885\sketch\ultralcd_st7920_u8glib_rrd.h: In function 'void marlin_SPI_delay()':

ultralcd_st7920_u8glib_rrd.h:112: error: '_delay_system_ticks_sub' was not declared in this scope

       _delay_system_ticks_sub((CYCLES_PER_MICROSECOND) * (DOGM_SPI_DELAY_US));

                                                                             ^

ghost · 2016-11-25T08:37:35Z

Still got a compilation error

error message

In file included from \AppData\Local\Temp\arduino_build_498533\sketch\ultralcd_impl_DOGM.h:46:0,

                 from \AppData\Local\Temp\arduino_build_498533\sketch\ultralcd.cpp:56:

\AppData\Local\Temp\arduino_build_498533\sketch\ultralcd_st7920_u8glib_rrd.h: In function 'void marlin_SPI_delay()':

ultralcd_st7920_u8glib_rrd.h:112: error: 'delay_system_ticks' was not declared in this scope

       delay_system_ticks((CYCLES_PER_MICROSECOND) * (DOGM_SPI_DELAY_US));

                                                                        ^

, thus I avoid it tentatively.

Selective + 5μs delay + "frame"	Selective + 5μs delay + "frame"	Selective + 4μs delay + "frame"	Selective + 3μs delay + "frame"
without both `USE_BIG_EDIT_FONT` and `USE_SMALL_INFOFONT`	`USE_BIG_EDIT_FONT` and `USE_SMALL_INFOFONT`	`USE_BIG_EDIT_FONT` and `USE_SMALL_INFOFONT`	`USE_BIG_EDIT_FONT` and `USE_SMALL_INFOFONT`
without `LCD_SCREEN_ROT_180`	without `LCD_SCREEN_ROT_180`	without `LCD_SCREEN_ROT_180`
2 Stripes	2 Stripes	2 Stripes
58 ~ 59ms	55 ~ 56ms	54 ~ 55ms	54 ~ 55ms
It's displayed properly	It's displayed properly but adjustment is needed	It's displayed properly but adjustment is needed	Displaying is broken

without this PR	Selective + 10μs delay + "frame"	Selective + 10μs delay + "frame"	Selective + 5μs delay + "frame"
`USE_BIG_EDIT_FONT` and `USE_SMALL_INFOFONT`	without both `USE_BIG_EDIT_FONT` and `USE_SMALL_INFOFONT`	`USE_BIG_EDIT_FONT` and `USE_SMALL_INFOFONT`	`USE_BIG_EDIT_FONT` and `USE_SMALL_INFOFONT`
with `LCD_SCREEN_ROT_180`	with `LCD_SCREEN_ROT_180`	with `LCD_SCREEN_ROT_180`	with `LCD_SCREEN_ROT_180`
2 Stripes	2 Stripes	2 Stripes	2 Stripes
-	45 ~ 46ms	39 ~ 40ms	37 ~ 38ms
It's displayed properly	It isn't displayed properly	It isn't displayed properly	It isn't displayed properly

~~A branch that used for testing: https://github.com/esenapaj/Marlin/tree/glcd-testes4~~

jbrazio · 2016-11-25T13:20:15Z

@thinkyhead

U8g2 apparently has some helpful features

Last time I checked they were still missing PROGMEM support.

thinkyhead · 2016-11-25T18:08:35Z

Still got a compilation error

Oops, forgot to push. Should be better now.

thinkyhead · 2016-11-25T18:29:05Z

@esenapaj Helpful results. So, the rotated screen needs to be accounted for. Apparently there's a way to access the page bounds that will work with any rotation…

u8g.getU8g()->current_page.y0
u8g.getU8g()->current_page.y1

…so I'll try adapting the code to use these.

I can also try to fix the positioning of the smaller font.

Are the glyphs for X Y Z broken with the small font?

thinkyhead · 2016-11-25T18:33:44Z

they were still missing PROGMEM support

Oliver is being very responsive! He has already submitted this for us:

Required display support for Marlin olikraus/u8g2#94 Required display support for Marlin

The AVR-specific optimizations that make u8g faster than u8g2 are still extant, however.

thinkyhead · 2016-11-26T11:16:31Z

Maybe this?

#if DISABLED(DISPLAY_CHARSET_ISO10646_1)
  #undef USE_BIG_EDIT_FONT
  #undef USE_SMALL_INFOFONT
#endif

That would correspond with:

#if DISABLED(SIMULATE_ROMFONT) \
 && DISABLED(DISPLAY_CHARSET_ISO10646_1) \
 && DISABLED(DISPLAY_CHARSET_ISO10646_5) \
 && DISABLED(DISPLAY_CHARSET_ISO10646_KANA) \
 && DISABLED(DISPLAY_CHARSET_ISO10646_GREEK) \
 && DISABLED(DISPLAY_CHARSET_ISO10646_CN) \
 && DISABLED(DISPLAY_CHARSET_ISO10646_TR)
  #define DISPLAY_CHARSET_ISO10646_1 // use the better font on full graphic displays.
#endif

…and…

// Western only. Not available for Cyrillic, Kana, Turkish, Greek, or Chinese.

ghost · 2016-11-26T11:32:59Z

Maybe this?

Maybe so. At least, I've confirmed that USE_BIG_EDIT_FONT works properly, and "top line missing" problem is solved in English just now.
Also LCD_SCREEN_ROT_180 is ok.

About "g", I test now.

ghost · 2016-11-26T11:46:49Z

How about?

`USE_BIG_EDIT_FONT`	without `USE_BIG_EDIT_FONT`
with `LCD_SCREEN_ROT_180`	with `LCD_SCREEN_ROT_180`

thinkyhead · 2016-11-26T11:51:24Z

Looks good. I count 13 pixels tall for the big font. I'll see about getting it better-centered based on that info, and perhaps we can get one less render on 8-stripe displays too.

ghost · 2016-11-26T12:09:50Z

It looks like that

#if DISABLED(DISPLAY_CHARSET_ISO10646_1)
  #undef USE_BIG_EDIT_FONT
  #undef USE_SMALL_INFOFONT
#endif

is forgotten.

thinkyhead · 2016-11-26T12:16:05Z

It looks like that … is forgotten

I'm suggesting that you try using those lines in place of these:

#if DISABLED(MAPPER_C2C3) && DISABLED(MAPPER_NON) && ENABLED(USE_BIG_EDIT_FONT)
  #undef USE_BIG_EDIT_FONT
#endif

ghost · 2016-11-26T12:33:06Z

Well... I used it already at test in #5288 (comment)

Perhaps I may be misunderstanding the interaction with you cause by my bad English.

Blue-Marlin · 2016-11-26T12:34:41Z

@@ -52,12 +52,13 @@
 
 #if ENABLED(SHOW_BOOTSCREEN) && ENABLED(SHOW_CUSTOM_BOOTSCREEN)
   #include "_Bootscreen.h"
 #endif
 
-#if DISABLED(MAPPER_C2C3) && DISABLED(MAPPER_NON) && ENABLED(USE_BIG_EDIT_FONT)
+#if DISABLED(DISPLAY_CHARSET_ISO10646_1)
   #undef USE_BIG_EDIT_FONT
+  #undef USE_SMALL_INFOFONT
 #endif
 
 #if ENABLED(USE_SMALL_INFOFONT)
   #include "dogm_font_data_6x9_marlin.h"
   #define FONT_STATUSMENU_NAME u8g_font_6x9

is enough to bring back USE_BIG_EDIT_FONT to RCBugFix. Tested with en and de.
It's the most direct condition we can test. Small and big font are both coded in ISO10646_1 so only when ISO10646_1 is used for the 'main' font it makes sense to use the other.

ghost · 2016-11-26T12:38:05Z

Also, I confirmed no problem with kana and kana_utf8.

thinkyhead · 2016-11-26T12:40:28Z

Perhaps I may be misunderstanding the interaction with you cause by my bad English.

From your statement "It looks like that (code) is forgotten" I thought you meant something else. The words "is forgotten" are quite poetic in this context.

I'm glad to hear that fixes it. I will add that to the PR and merge this thing. We can continue to patch this up in the coming days, but I think it's showing great promise.

AnHardt · 2016-11-26T12:41:56Z

I'll make a separate PR to fix 'USE_BIG_EDIT_FONT' only.

ghost · 2016-11-26T12:42:38Z

The words "is forgotten" are quite poetic in this context.

Oh sorry...

thinkyhead · 2016-11-26T12:48:25Z

http://sourcecodepoetry.com/

AnHardt · 2016-11-26T20:56:39Z

And we should check the edit items that need 2 lines to display (because the label is long).

Works.

AnHardt · 2016-11-27T19:14:31Z

About u8g2:

Last time I checked they were still missing PROGMEM support.

the solution for this is looking like:

#define printP(s) print(F((s)))
#define TEST_STR "Test"

u8g2.printP(TEST_STR);
// or
u8g2.printP("An other String");

Fonts are already in PROGMEM.

Something else missing? - besides of speed?

Last night's patch for u8g2 is a big step forward - but not enough. :-(
https://www.youtube.com/watch?v=c6XYxvCf2Fc

landodragon141 · 2016-11-27T23:31:53Z

It looks like he just posted another performance update for SW SPI.
I don't think there is a video up yet.

AnHardt · 2016-11-28T00:23:16Z

u8g vs u8g2 - now maybe a bit too fast - somewhere
https://www.youtube.com/watch?v=ZChlUhMJ3fQ
Downloaded u8g2 ~0:00UTC

@olikraus

if ( b == 0 )
{
  *arduino_data_port &= arduino_data_n_mask;
  for( i = 0; i < 4; i++ )
  {
    *arduino_clock_port |= arduino_clock_mask;      
    *arduino_clock_port &= arduino_clock_n_mask;
          __asm__("nop\n\t");                             ///////////////////////////
    *arduino_clock_port |= arduino_clock_mask;      
    *arduino_clock_port &= arduino_clock_n_mask;
  }
}
else

A nop in both of the 0-loops seems to be enough to have no pixel errors on my display.

thinkyhead added PR: Improvement C: LCD & Controllers labels Nov 24, 2016

thinkyhead mentioned this pull request Nov 24, 2016

Unroll loops in ultralcd_st7920_u8glib_rrd.h for performance #5279

Closed

thinkyhead force-pushed the rc_selective_rendering branch 2 times, most recently from b13ab85 to 0de53b4 Compare November 24, 2016 01:15

thinkyhead mentioned this pull request Nov 24, 2016

Q: How to get the current page bounds olikraus/u8glib#447

Closed

thinkyhead force-pushed the rc_selective_rendering branch 3 times, most recently from 9f05af9 to 6700374 Compare November 24, 2016 07:00

thinkyhead force-pushed the rc_selective_rendering branch from a503e49 to 23c4949 Compare November 25, 2016 06:51

thinkyhead force-pushed the rc_selective_rendering branch from 23c4949 to c2b827f Compare November 25, 2016 18:08

thinkyhead force-pushed the rc_selective_rendering branch 2 times, most recently from 03a316a to c4532c1 Compare November 25, 2016 18:45

thinkyhead force-pushed the rc_selective_rendering branch from 831c3d7 to 2296827 Compare November 26, 2016 11:58

thinkyhead force-pushed the rc_selective_rendering branch from 2296827 to c35b500 Compare November 26, 2016 12:42

AnHardt mentioned this pull request Nov 26, 2016

Fix USE_BIG_EDIT_FONT #5301

Merged

thinkyhead added 5 commits November 26, 2016 06:53

Optimize graphical display with selective rendering

b1abd7e

Apply const to LCD arguments and locals

71842b6

Add option DOGM_SPI_DELAY_US

5460a52

Adjust spacing in utf_mapper.h

05d8d18

Minor formatting in language.h

aee71c5

thinkyhead force-pushed the rc_selective_rendering branch from c35b500 to aee71c5 Compare November 26, 2016 12:54

thinkyhead merged commit 668e737 into MarlinFirmware:RCBugFix Nov 26, 2016

thinkyhead deleted the rc_selective_rendering branch November 26, 2016 13:01

thinkyhead mentioned this pull request Nov 28, 2016

Release 1.1.0-RC8 #5077

Closed

fiveangle mentioned this pull request Sep 19, 2017

[FR] Support U8g2 #7668

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Optimize graphical display with selective rendering #5288

Optimize graphical display with selective rendering #5288

thinkyhead commented Nov 24, 2016 •

edited

Loading

Sebastianv650 commented Nov 24, 2016

thinkyhead commented Nov 24, 2016 •

edited

Loading

thinkyhead commented Nov 24, 2016 •

edited

Loading

AnHardt commented Nov 24, 2016 •

edited by thinkyhead

Loading

ghost commented Nov 24, 2016 •

edited by ghost

Loading

AnHardt commented Nov 25, 2016

AnHardt commented Nov 25, 2016 •

edited by thinkyhead

Loading

ghost commented Nov 25, 2016

ghost commented Nov 25, 2016 •

edited by ghost

Loading

AnHardt commented Nov 25, 2016

thinkyhead commented Nov 25, 2016 •

edited

Loading

ghost commented Nov 25, 2016

ghost commented Nov 25, 2016 •

edited by ghost

Loading

jbrazio commented Nov 25, 2016

thinkyhead commented Nov 25, 2016

thinkyhead commented Nov 25, 2016 •

edited

Loading

thinkyhead commented Nov 25, 2016 •

edited

Loading

thinkyhead commented Nov 26, 2016 •

edited

Loading

ghost commented Nov 26, 2016 •

edited by ghost

Loading

ghost commented Nov 26, 2016

thinkyhead commented Nov 26, 2016

ghost commented Nov 26, 2016

thinkyhead commented Nov 26, 2016 •

edited

Loading

ghost commented Nov 26, 2016 •

edited by ghost

Loading

Blue-Marlin commented Nov 26, 2016 •

edited by thinkyhead

Loading

ghost commented Nov 26, 2016

thinkyhead commented Nov 26, 2016

AnHardt commented Nov 26, 2016

ghost commented Nov 26, 2016

thinkyhead commented Nov 26, 2016

AnHardt commented Nov 26, 2016

AnHardt commented Nov 27, 2016 •

edited by thinkyhead

Loading

landodragon141 commented Nov 27, 2016 •

edited by thinkyhead

Loading

AnHardt commented Nov 28, 2016 •

edited by thinkyhead

Loading

Optimize graphical display with selective rendering #5288

Optimize graphical display with selective rendering #5288

Conversation

thinkyhead commented Nov 24, 2016 • edited Loading

Sebastianv650 commented Nov 24, 2016

thinkyhead commented Nov 24, 2016 • edited Loading

thinkyhead commented Nov 24, 2016 • edited Loading

AnHardt commented Nov 24, 2016 • edited by thinkyhead Loading

ghost commented Nov 24, 2016 • edited by ghost Loading

AnHardt commented Nov 25, 2016

AnHardt commented Nov 25, 2016 • edited by thinkyhead Loading

ghost commented Nov 25, 2016

ghost commented Nov 25, 2016 • edited by ghost Loading

AnHardt commented Nov 25, 2016

thinkyhead commented Nov 25, 2016 • edited Loading

ghost commented Nov 25, 2016

ghost commented Nov 25, 2016 • edited by ghost Loading

jbrazio commented Nov 25, 2016

thinkyhead commented Nov 25, 2016

thinkyhead commented Nov 25, 2016 • edited Loading

thinkyhead commented Nov 25, 2016 • edited Loading

thinkyhead commented Nov 26, 2016 • edited Loading

ghost commented Nov 26, 2016 • edited by ghost Loading

ghost commented Nov 26, 2016

thinkyhead commented Nov 26, 2016

ghost commented Nov 26, 2016

thinkyhead commented Nov 26, 2016 • edited Loading

ghost commented Nov 26, 2016 • edited by ghost Loading

Blue-Marlin commented Nov 26, 2016 • edited by thinkyhead Loading

ghost commented Nov 26, 2016

thinkyhead commented Nov 26, 2016

AnHardt commented Nov 26, 2016

ghost commented Nov 26, 2016

thinkyhead commented Nov 26, 2016

AnHardt commented Nov 26, 2016

AnHardt commented Nov 27, 2016 • edited by thinkyhead Loading

landodragon141 commented Nov 27, 2016 • edited by thinkyhead Loading

AnHardt commented Nov 28, 2016 • edited by thinkyhead Loading

thinkyhead commented Nov 24, 2016 •

edited

Loading

thinkyhead commented Nov 24, 2016 •

edited

Loading

thinkyhead commented Nov 24, 2016 •

edited

Loading

AnHardt commented Nov 24, 2016 •

edited by thinkyhead

Loading

ghost commented Nov 24, 2016 •

edited by ghost

Loading

AnHardt commented Nov 25, 2016 •

edited by thinkyhead

Loading

ghost commented Nov 25, 2016 •

edited by ghost

Loading

thinkyhead commented Nov 25, 2016 •

edited

Loading

ghost commented Nov 25, 2016 •

edited by ghost

Loading

thinkyhead commented Nov 25, 2016 •

edited

Loading

thinkyhead commented Nov 25, 2016 •

edited

Loading

thinkyhead commented Nov 26, 2016 •

edited

Loading

ghost commented Nov 26, 2016 •

edited by ghost

Loading

thinkyhead commented Nov 26, 2016 •

edited

Loading

ghost commented Nov 26, 2016 •

edited by ghost

Loading

Blue-Marlin commented Nov 26, 2016 •

edited by thinkyhead

Loading

AnHardt commented Nov 27, 2016 •

edited by thinkyhead

Loading

landodragon141 commented Nov 27, 2016 •

edited by thinkyhead

Loading

AnHardt commented Nov 28, 2016 •

edited by thinkyhead

Loading