Gray timings on fx-9750GII
Posté le 15/06/2021 14:46
I found the "correct" gray timings for my calculator fx-9750GII
The screen runs at 64Hz (maybe slightly less 63.99998 or TMU is slightly off??)
so the timings for gray are just flickering the screen at some ratio of 64Hz
tho sadly there are only 4 "usable" shades of gray, with only 2 being "good"
No tuning adjustment required
66:33 produces ~98% gray (almost black)
50:50 produces ~55% gray (good)
33:66 produces ~20% gray (good)
25:75 produces ~10% gray (very light, flickers)
I'm using the TMU, to control the frequency
Pϕ/1024 to Pϕ/4 works. Pϕ/4 seems to flicker a bit tho
Can use the RTC, tho it flickers every few sec (skips ticks every so often for some reason?)
(Looks better in real life)
!!Warning!!
The addin attached here is very unstable
It can and will freeze your calculator (only requires reset button on back)
If you leave it running for too long. TMU overflows (it runs in the background too)
Use at your own risk
It is only a prototype
It does take Overclocking into account to maintain 64fps
It should work:tm: on Graph35EII+ and Graph 75+E(fx-9860GII)
Does NOT work on SH3, including the SDK
gray9750.g1a
Controls:
[MENU] (twice) to exit program
[REPLAY] to modify target fps. Number is fixedpointed 5 to the left
[EXE] Fullscreen
[EXP] change Pϕ (Pϕ/4, Pϕ/16, Pϕ/256)
[TAN] increase gray level
[->] decrease gray level
On screen numbers from the top
current fps
number of ticks to wait (Pϕ/X)
active spinning while loop. if value goes to 0, calc is lagging
gray level (does go negative)
Pϕ clock timing
target fps
Fichier joint
Citer : Posté le 15/06/2021 22:13 | #
This is nice, thank you!
Since you mention that the display runs at 64 Hz, I checked the documentation and it actually specifies the frequency (page 32), which makes me feel really stupid for not checking before. xD
As you can see there are various possibilities for the frame frequency f_FR, it would be nice if we could compute it to properly synchronize the driver and screen on all models (since it is known that not all models look good at the same frequencies).
Can use the RTC, tho it flickers every few sec (skips ticks every so often for some reason?)
Pϕ/4 is more precise, I don't see why it would be different - maybe with a slighly different delay (in the order of ±128 units)? The RTC is exactly 64 Hz, there is a chance that the extra processing cost of interrupts delays some frames and causes irregularities. I assume this is a standalone executable and you actively wait on the timers to expire without using interrupts?
It can and will freeze your calculator (only requires reset button on back)
Can you share the source code? Usually freezing would happen if interrupts are requested in shorter intervals than the CPU can handle, but of course this wouldn't happen if you don't use interrupts.
I have tested it on my Graph 75+E (fx-9860G II) and the results are already better than gint's default. This is a very nice progress! Thanks a lot
As for the 63.99998 this is a 5 nano-second difference. Even without counting the Pϕ divider, variations in the clock generation, overhead CPU time, variations in frame generation or access to VRAM due to cache, variations in the power supply, or really any low-level effect, 5 nanoseconds is still 10 times smaller than a Pϕ cycle. I'm pretty sure there's no way the timer can capture that small a difference
Citer : Posté le 15/06/2021 22:26 | # |
Fichier joint
Here is the attached code GRAY9750.zip
I shall warn you, the code is very very messy
I've copy and pasted various pieces of code from old projects and online libs
A lot of variables don't do what they say they are
And a lot of functions don't even get used
Ajouté le 16/06/2021 à 08:28 :
Should try overclocking, seting the fps to 768 and looking at the fullscreen screen. All shades (except 1) are static. Happens at every frame setting that is a multiple of 64. Also notice that everything slowly moves down and setting the fps slightly higher causes it to move upwards
Citer : Posté le 16/06/2021 09:11 | #
Your figures are definitely accurate. It tried with 768 Hz, and got the same result as you (alternating black/white stripes due to VRAM changing while rendering is performed).
I had an idea. Some shades are much cleaner at higher refresh rates, while others are only visible at smaller ones (typically: just 64, anything else gives static stripes). This suggests that a different encodingof gray values on the VRAM might give better uniform results. (I haven't looked at your VRAM encoding yet, but I will sure do soon enough!)
Maybe 64 Hz is the "unifying" factor that works everywhere, and previous engine delays were affected by overclock or other independent sources. I hope so, it would make things much easier! I will try on a variety of models when I can.
This seems to make sense. In this case, let's use whatever works best!
Citer : Posté le 16/06/2021 16:38 | # |
Fichier joint
[TAN] and [->] range is now from 13 to 127
70 is halfway, 26 and 115 are smooth
ManyGray9750.zip
Should work on SH3 now
Citer : Posté le 16/06/2021 16:50 | #
Now that does look good for sure! Unfortunately, it's unstable on my Graph 75+E and it doesn't work at all on my Graph 35+E II (freezes at startup). Are there specific overclock needs or other settings that this might depend on?
I wouldn't be surprised if differences between models still existed ultimately, so as long as we can find parameters for all models, this would end up working.
Anyway, this is starting to look less like a cool improvement and more like a complete breakthrough! I love it!
Citer : Posté le 18/06/2021 18:54 | #
In short, we could get 64 steps from white to black, instead of 4 currently ?
Sounds great
Citer : Posté le 19/06/2021 14:15 | #
64 steps ? It doesn't seem to be headed that way (to me at least). 64 Hz doesn't mean you can get 64 steps. I think the first obvious outcome is increase the quality of 4-color gray, then look for 8 or 16 if that is possible. But that would still need quite a bit of work in software to convert and display images.
Citer : Posté le 24/03/2022 10:15 | # |
Fichier joint
Sadly flickering the screen with a different sequence other than 10, 110 or 100 causes this bad 'pulsing' effect
where the screen will 'pulse' off for quite a long time every few seconds
sequence 1110 and 1000 do flicker a small amount. more than what Gint currently does
The problem is that the 'ghosting' time of the pixels is not enough to flash a sequence more than 4 steps long
You need to have a regular cycle that is shorter than the "ghost time", and in this case the smallest period is far too long.
Its almost impossible to see its flickering and there are very very few visual artifacts
110 and 100 is about on par with Gint's flickering
Next step to making more better grays
is to use another bad flaw with the display; which is its terrible 'bleed through' effect
when a pixel is on, it will 'bleed through' to all other pixels in the same column
this effect compounds. so as more pixels are turned on in the column the darker the 'off' pixels become
untill when you have 126/127 pixels on, the last 1/2 pixels will look almost 90% on
currently all gray engines turn every single pixel on the first frame, then every single pixel off the next frame
what if we instead delay every 2nd pixel by one frame
so only half the pixels are on the first frame, next frame the other half is on
this will create a interlacing effect (which you cant see) and because only half the pixels are on at any one time. (01010101)
the shade of gray will be around almost ~50% lighter!
(Left is normal, right is interlaced (still running at the same frequency and sequence))
one property of this is that the more pixels that are on together, the greater the effect of the bleed through
so instead of every 2nd pixel, why not do every 2nd group of 2 pixels. 01010101 -> 00110011
this creates a gray that is half way between the single interlaced and the original
the doubling can be done 4 more times til we alternative between the bottom and top half of the screen
giving us 7 shades of gray out of one single sequence
which can be applied to the other 4 sequences. allowing us to have 35 different grays! (tho some are very close to others)
so can safely say 8-16 shades
only problem with interlacing is that it requires all other pixels in the same coloumn to either be the same gray or off
every pixel that is black will change the shade of gray
but this is true for every other gray engine, so isn't too bad
I've recorded my screen at almost 1000 frames per second!
showing us the refresh rate of the display
Video was recorded at 960fps
Played back at 30fps
Display runs at 64fps
64Hz * 30fps / 2 screen refreshes = 960FPS
The screen refreshes every 15 frames or twice per second in the video
I've also recorded a single pixel and ploted the shades as it turns on and off
https://docs.google.com/spreadsheets/d/1s0ZnQ1j9Ve8PYBb7UETDX3ZCqb7Q8-O18C2aUyKzcU0
Every line is 1/64 of a second apart
Notice how quickly the pixel starts to fade after every screen refresh while its turning on
and then how slowly it is to turn off. only because its still getting refreshed!
as you can see in the graph, the pixel darkens extremely quickly, but the display driver only powers it for an extremely small amount of time
maybe increasing this time will give much better pixel response times, decreasing ghosting
tho the pixel is also still powered a small amount when it is off.
which can be very easily seen when increasing the contrast
Attached file: GrayGIII.G1A
Add-in is compatible with fx-9750GII, Graph 75+E (fx-9860GII) and Graph 35+E II (tho the Graph 35+E II seems to run at a different display frequency, giving bad grays) and the fx-9860GII emulator
Hold [EXE] while starting the add-in to see some stats
[VARS] changes the interlaced value on the 2nd half of the example gray
[EXP] switches between Pϕ/4 (0), Pϕ/16 (1), Pϕ/256 (2) and Pϕ/1024 (3) (Pϕ/4 seems to flicker a bit?)
[EXE] swtich to fullscreen mode, showing all 128 different sequences
[tan] increase sequence
[→] decrease sequence
[MENU] press twice to exit
[REPLAY] to change the frequency to which the display runs at (default 64fps)
List of sequences:
1 10000000
2 1000000
3 1000001000000
4 100000
5 10000100000
6 10000
7 10001000010000
8 100010000
9 1000100010000
10 10001000100010000
11 1000
12 1001000100010001000
13 100100010001000
14 10010001000
15 100100010010001000
16 1001000
17 10010010001001000
18 1001001000
19 10010010010001001001000
20 1001001001000
21 1001001001001000
22 1001001001001001000
23 1001001001001001001000
24 1001001001001001001001000
25 100
26 10100100100100100100100100
27 10100100100100100100100
28 10100100100100100100
29 10100100100100100
30 10100100100100
31 1010010010010100100100100
32 10100100100
33 1010010010100100100
34 101001001010010010100100100
35 10100100
36 101001010010010100100
37 1010010100100
38 101001010010100100
39 10100101001010010100100
40 10100
41 101010010100101001010010100
42 1010100101001010010100
43 10101001010010100
44 101010010100
45 1010100101010010100
46 10101001010100101010010100
47 1010100
48 10101010010101001010100
49 1010101001010100
50 1010101001010101001010100
51 101010100
52 10101010100101010100
53 10101010100
54 101010101010010101010100
55 1010101010100
56 101010101010100
57 10101010101010100
58 1010101010101010100
59 101010101010101010100
60 10101010101010101010100
61 1010101010101010101010100
62 101010101010101010101010100
63 10
64 110101010101010101010101010
65 1101010101010101010101010
66 11010101010101010101010
67 110101010101010101010
68 1101010101010101010
69 11010101010101010
70 110101010101010
71 1101010101010
72 110101010101101010101010
73 11010101010
74 11010101011010101010
75 110101010
76 1101010110101010110101010
77 1101010110101010
78 11010101101010110101010
79 1101010
80 11010110101011010101101010
81 1101011010101101010
82 110101101010
83 11010110101101010
84 1101011010110101101010
85 110101101011010110101101010
86 11010
87 11011010110101101011010
88 110110101101011010
89 1101101011010
90 110110101101101011010
91 11011010
92 110110110101101101011011010
93 1101101101011011010
94 11011011010
95 1101101101101011011011010
96 11011011011010
97 11011011011011010
98 11011011011011011010
99 11011011011011011011010
100 11011011011011011011011010
101 110
102 1110110110110110110110110
103 1110110110110110110110
104 1110110110110110110
105 1110110110110110
106 1110110110110
107 11101101101110110110110
108 1110110110
109 11101101110110110
110 1110110
111 111011101101110110
112 11101110110
113 111011101110110
114 1110111011101110110
115 1110
116 11110111011101110
117 1111011101110
118 111101110
119 11110111101110
120 11110
121 11111011110
122 111110
123 1111110111110
124 1111110
125 11111110
126 111111110
127 1