Mandelbrot Generator
Posté le 15/06/2019 10:38
Hello allô
Default 1x zoom takes 7sec
Max zoom takes around 5-10min
It has a max zoom of 2^50: over one Quadrillion!
Going over 2^48 can be rather buggy
This is because numbers are limited to the 8 byte double variables
Attached file is both SH4 and SH3 compatible:
MANDEL.G1A
This does need the 'MonochromeLib' libs the code comes with it now
Controls
[-] Zoom out
[+] Zoom in
[F1] Hide/show HUD which contains Cords, Zoom level and Max Iterations. (Heads Up Display)
[F2] Changes colours of camera rectangle: Black, White & Inverted
[AC] Resets screen back to default state
[EXE] Draw set
[EXIT] Stop drawing the Mandelbrot (If it's taking too long)
[MENU] Return to the menu screen
[REPLAY] Move camera rectangle around (Arrow Keys: [LEFT], [RIGHT], [UP], [DOWN])
How can I optimize this code to run faster or zoom in further?
#include "fxlib.h"
#include "stdio.h"
#define TRUE 1
#define FALSE 0
#define ML_vram_adress (*(sc_cpv)sc0135)
typedef enum { ML_TRANSPARENT = -1, ML_WHITE, ML_BLACK, ML_XOR, ML_CHECKER } ML_Color;
typedef char* (*sc_cpv)(void);
const unsigned int sc0135[] = { 0xD201D002, 0x422B0009, 0x80010070, 0x0135 };
unsigned int key; //pause until key press
int kcode1, kcode2; //row & col keycode for Bkey_GetKeyWait()
char unused; //unused (cause CASIO dumb dumb)
unsigned short dispX, dispY; //cords on display when drawing mandelbrot
void ML_clear_vram() {
int i, end, * pointer_long, vram;
char* pointer_byte;
vram = (int)ML_vram_adress();
end = 4 - vram & 3;
pointer_byte = (char*)vram;
for (i = 0; i < end; i++) pointer_byte[i] = 0;
pointer_long = (int*)(vram + end);
for (i = 0; i < 255; i++) pointer_long[i] = 0;
pointer_byte += 1020 + end;
end = vram & 3;
for (i = 0; i < end; i++) pointer_byte[i] = 0;
}
void ML_display_vram() {
char* LCD_register_selector = (char*)0xB4000000, * LCD_data_register = (char*)0xB4010000, * vram;
int i, j;
vram = ML_vram_adress();
for (i = 0; i < 64; i++) {
*LCD_register_selector = 4;
*LCD_data_register = i | 192;
*LCD_register_selector = 4;
*LCD_data_register = 0;
*LCD_register_selector = 7;
for (j = 0; j < 16; j++)
*LCD_data_register = *vram++;
}
}
void ML_display_vram_row(int row) { //faster than ML_display_vram() which displays the entire screen instead of a single row
unsigned char i;
char* LCD_register_selector = (char*)0xB4000000, *LCD_data_register = (char*)0xB4010000, *vram;
vram = (row << 4) + ML_vram_adress();
*LCD_register_selector = 4;
*LCD_data_register = row | 192;
*LCD_register_selector = 4;
*LCD_data_register = 0;
*LCD_register_selector = 7;
for (i = 0; i < 16; i++)
* LCD_data_register = *vram++;
}
void ML_horizontal_line(int y, int x1, int x2, ML_Color color) {
int i;
char checker;
char* vram = ML_vram_adress();
if (y & ~63 || (x1 < 0 && x2 < 0) || (x1 > 127 && x2 > 127))
return;
if (x1 > x2) {
i = x1;
x1 = x2;
x2 = i;
}
if (x1 < 0)
x1 = 0;
if (x2 > 127)
x2 = 127;
switch (color) {
case ML_BLACK:
if (x1 >> 3 != x2 >> 3) {
vram[(y << 4) + (x1 >> 3)] |= 255 >> (x1 & 7);
vram[(y << 4) + (x2 >> 3)] |= 255 << 7 - (x2 & 7);
for (i = (x1 >> 3) + 1; i < x2 >> 3; i++)
vram[(y << 4) + i] = 255;
} else
vram[(y << 4) + (x1 >> 3)] |= (255 >> (x1 % 8 + 7 - x2 % 8)) << (7 - (x2 & 7));
break;
case ML_WHITE:
if (x1 >> 3 != x2 >> 3) {
vram[(y << 4) + (x1 >> 3)] &= 255 << 8 - (x1 & 7);
vram[(y << 4) + (x2 >> 3)] &= 255 >> 1 + (x2 & 7);
for (i = (x1 >> 3) + 1; i < x2 >> 3; i++)
vram[(y << 4) + i] = 0;
} else
vram[(y << 4) + (x1 >> 3)] &= (255 << 8 - (x1 & 7)) | (255 >> 1 + (x2 & 7));
break;
case ML_XOR:
if (x1 >> 3 != x2 >> 3) {
vram[(y << 4) + (x1 >> 3)] ^= 255 >> (x1 & 7);
vram[(y << 4) + (x2 >> 3)] ^= 255 << 7 - (x2 & 7);
for (i = (x1 >> 3) + 1; i < (x2 >> 3); i++)
vram[(y << 4) + i] ^= 255;
} else
vram[(y << 4) + (x1 >> 3)] ^= (255 >> ((x1 & 7) + 7 - (x2 & 7))) << (7 - (x2 & 7));
break;
case ML_CHECKER:
checker = (y & 1 ? 85 : 170);
if (x1 >> 3 != x2 >> 3) {
vram[(y << 4) + (x1 >> 3)] &= 255 << 8 - (x1 & 7);
vram[(y << 4) + (x2 >> 3)] &= 255 >> 1 + (x2 & 7);
vram[(y << 4) + (x1 >> 3)] |= checker & 255 >> (x1 & 7);
vram[(y << 4) + (x2 >> 3)] |= checker & 255 << 7 - (x2 & 7);
for (i = (x1 >> 3) + 1; i < x2 >> 3; i++)
vram[(y << 4) + i] = checker;
} else {
vram[(y << 4) + (x1 >> 3)] &= (255 << 8 - (x1 & 7)) | (255 >> 1 + (x2 & 7));
vram[(y << 4) + (x1 >> 3)] |= checker & (255 >> (x1 % 8 + 7 - x2 % 8)) << (7 - (x2 & 7));
}
break;
}
}
void ML_vertical_line(int x, int y1, int y2, ML_Color color) {
int i, j;
char checker, byte, * vram = ML_vram_adress();
if (x & ~127 || (y1 < 0 && y2 < 0) || (y1 > 63 && y2 > 63)) return;
if (y1 > y2) {
int tmp = y1;
y1 = y2;
y2 = tmp;
}
if (y1 < 0) y1 = 0;
if (y2 > 63) y2 = 63;
i = (y1 << 4) + (x >> 3);
j = (y2 << 4) + (x >> 3);
switch (color) {
case ML_BLACK:
byte = 128 >> (x & 7);
for (; i <= j; i += 16)
vram[i] |= byte;
break;
case ML_WHITE:
byte = ~(128 >> (x & 7));
for (; i <= j; i += 16)
vram[i] &= byte;
break;
case ML_XOR:
byte = 128 >> (x & 7);
for (; i <= j; i += 16)
vram[i] ^= byte;
break;
case ML_CHECKER:
byte = 128 >> (x & 7);
checker = y1 & 1 ^ x & 1;
for (; i <= j; i += 16) {
if (checker) vram[i] &= ~byte;
else vram[i] |= byte;
checker = !checker;
}
break;
}
}
void ML_pixel(int x, int y, ML_Color color) {
char* vram = ML_vram_adress();
if (x & ~127 || y & ~63) return;
switch (color) {
case ML_BLACK:
vram[(y << 4) + (x >> 3)] |= 128 >> (x & 7);
break;
case ML_WHITE:
vram[(y << 4) + (x >> 3)] &= ~(128 >> (x & 7));
break;
case ML_XOR:
vram[(y << 4) + (x >> 3)] ^= 128 >> (x & 7);
break;
case ML_CHECKER:
if (y & 1 ^ x & 1) vram[(y << 4) + (x >> 3)] &= ~(128 >> (x & 7));
else vram[(y << 4) + (x >> 3)] |= 128 >> (x & 7);
break;
}
}
double divByPow(double n, double x, int p) { //Divide OR Times n by x, p times (n / x^p): used for numbers bigger than 2^32 (int limit)
if (p < 0)
for (; p < 0; p++)
n *= x;
else
for (; p > 0; p--)
n /= x;
return n;
}
void stop(void) { //stops drawing set if user presses [EXIT] or [MENU]
if (Bkey_GetKeyWait(&kcode1, &kcode2, 1, 0, 1, &unused))
if (kcode1 == 4 && (kcode2 == 8 || kcode2 == 9)) {
dispX = 128; //Very hacky stop function
dispY = 64;
}
}
int AddIn_main(int isAppli, unsigned short OptionNum) { //Main function
unsigned int graphZoom = 1; //zoom level for graph
char screenZoom; //zoom level on screen (rectangle)
int screenX1, screenX2; //corner X cords for drawing rectangle to screen
int screenY1, screenY2; //corner Y cords for drawing rectangle to screen
unsigned char string[1]; //Used in converting int/double to char
char HUD = TRUE; //Heads Up Display: Cords, Zoom level & Max iteration: toggle with [F1]
char colour = ML_XOR; //Colour of rectangle: Black, White or Inverted
int screenX, screenY; //offset cords on screen from 0,0 for rectangle
double graphX = 0, graphY = 0; //cords on graph - where to center mandelbrot
double graphMove; //amount graphX & Y changes by when moving rectangle around
int screenMove; //amount screenX & Y changes by when moving rectangle around with arrow keys
short tempPixel = 0; //Write pixels to temp variable then write the entire 2bytes to VRAM all at once
register double zr, zi; //zr is real, zi imaginary
register double zr2, zi2; //zr2 = zr^2, zi2 = zi^2
register double x1 = -2.0; //bounding box cords on graph
register double x2 = 2.0; //bounding box cords on graph
register double y1 = -1.0; //bounding box cords on graph
register double y2 = 1.0; //bounding box cords on graph
register double x, y; //pixel cords on graph tested if in set
register double xIsz, yIsz; //amount x/y increases by when ploting graph
register unsigned short iMax = 32; //max iterations
register unsigned short i; //iterations
while (TRUE) {
register char* vram = ML_vram_adress();
SetTimer(1, 200, stop);
ML_clear_vram();
ML_display_vram();
xIsz = (x2 - x1) / 128;
yIsz = (y2 - y1) / 64;
y = y1;
for (dispY = 0; dispY < 64; dispY++) {
x = x1;
y += yIsz;
for (dispX = 0; dispX < 128; dispX++) {
zr = x;
zi = y;
for (i = 0; i < iMax; i++) {
zr2 = zr * zr;
zi2 = zi * zi;
if (zr2 + zi2 > 4)
break;
zi = zr * zi;
zi += zi + y;
zr = zr2 - zi2 + x;
}
tempPixel = (tempPixel << 1) | (i == iMax);
if ((dispX & 7) == 7)
*vram++ = tempPixel;
x += xIsz;
}
ML_display_vram_row(dispY);
}
SaveDisp(1);
KillTimer(1);
screenX = 0;
screenY = 0;
screenZoom = 1;
Bkey_GetKeyWait(&kcode1, &kcode2, 2, 1, 1, &unused);
do {
GetKey(&key);
screenMove = screenZoom > 4 ? 1 : divByPow(16, 2, screenZoom);
graphMove = screenZoom > 4 ? divByPow(1, 2, graphZoom - (double)screenZoom) : divByPow(16, 2, graphZoom);
switch (key) {
case KEY_CHAR_PLUS:
if (graphZoom < 51) {
graphZoom++;
screenZoom++;
}
break;
case KEY_CHAR_MINUS:
if (graphZoom) {
graphZoom--;
screenZoom--;
}
break;
case KEY_CTRL_UP:
screenY -= screenMove;
graphY -= graphMove;
break;
case KEY_CTRL_DOWN:
screenY += screenMove;
graphY += graphMove;
break;
case KEY_CTRL_LEFT:
screenX -= screenMove;
graphX -= graphMove;
break;
case KEY_CTRL_RIGHT:
screenX += screenMove;
graphX += graphMove;
break;
case KEY_CTRL_F1:
HUD = !HUD;
break;
case KEY_CTRL_F2:
if (colour)
colour--;
else
colour = ML_XOR;
break;
case KEY_CTRL_F3:
//Gray scale, by refreshing screen multiple times per sec at different max iterations (iMax)
break;
case KEY_CTRL_AC:
graphZoom = 1;
graphX = 0;
graphY = 0;
screenZoom = 1;
screenX = 0;
screenY = 0;
key = KEY_CTRL_EXE;
break;
}
RestoreDisp(1);
iMax = 8 * (graphZoom + 3);
if (screenZoom < 8) {
screenX1 = 65 - divByPow(128, 2, screenZoom) + screenX;
screenX2 = 62 + divByPow(128, 2, screenZoom) + screenX;
screenY1 = 32 - (screenZoom > 6 ? 1 : divByPow(64, 2, screenZoom)) + screenY;
screenY2 = 31 + (screenZoom > 6 ? 0 : divByPow(64, 2, screenZoom)) + screenY;
ML_horizontal_line(screenY1, screenX1, screenX2, colour);
ML_horizontal_line(screenY2, screenX1, screenX2, colour);
ML_vertical_line(screenX1 - 1, screenY1, screenY2, colour);
ML_vertical_line(screenX2 + 1, screenY1, screenY2, colour);
} else
ML_pixel(screenX + 64, screenY + 31, colour);
x1 = divByPow(-4, 2, graphZoom) + (0.03125 * graphX);
x2 = divByPow(4, 2, graphZoom) + (0.03125 * graphX);
y1 = divByPow(-2, 2, graphZoom) + (0.03125 * graphY);
y2 = divByPow(2, 2, graphZoom) + (0.03125 * graphY);
if (HUD == TRUE) {
sprintf(&string, "X1:%f", x1);
PrintMini(0, 0, string, 0);
sprintf(&string, "Y1:%f", y1);
PrintMini(0, 6, string, 0);
sprintf(&string, "X2:%f", x2);
PrintMini(81, 53, string, 0);
sprintf(&string, "Y2:%f", y2);
PrintMini(81, 59, string, 0);
sprintf(&string, "MaxI:%u", iMax);
PrintMini(0, 53, string, 0);
if (graphZoom > 32)
sprintf(&string, "Zoom:2^%ux", graphZoom - 1);
else
sprintf(&string, "Zoom:%ux", (int)divByPow(1, 2, -graphZoom + 1));
PrintMini(0, 59, string, 0);
}
ML_display_vram();
} while (key != KEY_CTRL_EXE);
}
return 0;
}
#pragma section _BR_Size
unsigned long BR_Size;
#pragma section
#pragma section _TOP
int InitializeSystem(int isAppli, unsigned short OptionNum) {
return INIT_ADDIN_APPLICATION(isAppli, OptionNum);
}
#pragma section
Fichier joint
Citer : Posté le 16/02/2020 05:49 | #
What is this 1/32768 timer?
I think I did try RTC_GetTicks(), but couldn't get it working (errors)
I thought it was just cause of my calculator, but you say C.Basic uses it
If the operating speed was fixed, I would have to revert it back to the overclocked speed when the mandelbrot is computed
And it would be different across the different models?
How to create an interrupt (without breaking all others) with a smaller time than 25ms?
Maybe I don't use any interrupts, or timers and just make it with "clock cycle" based delays only
Citer : Posté le 16/02/2020 06:47 | # |
Fichier joint
What is this 1/32768 timer?
1/32768 timer is only for SH4A model,
You can use a count-up timer by default.
*(unsigned int*)P7305_EXTRA_TMU5_COUNT
I think I did try RTC_GetTicks(), but couldn't get it working (errors)
I thought it was just cause of my calculator, but you say C.Basic uses it
The following is an example of using in C.Basic that mimics the microseconds waiting SysCall in Prizm.
int i,j;
volatile int t=0;
for ( i=0; i<n; i++ ) for ( j=0; j<1; j++ ) t+=RTC_GetTicks();
}
If the operating speed was fixed, I would have to revert it back to the overclocked speed when the mandelbrot is computed
And it would be different across the different models?
Dynamic clock switching is easy if the operation target calculator is limited to SH4A.
This allows you to set Ftune2 default settings directly.
[Fichier joint]:Ftune2_simple.zip
How to create an interrupt (without breaking all others) with a smaller time than 25ms?
Sorry,I don't know how to change the minimum interrupt time.
Overclocking utilitaire Ftune/Ptune2/Ptune3 est également disponible.
Si vous avez des questions ou un rapport de bogue, n'hésitez pas à me le faire savoir.
Citer : Posté le 16/02/2020 07:20 | #
CASIO really has made it hard for me to do this
There's no 'perfect', always works in every case function
I want this to be compatible with Graph 35+EII, SH3 and SH4 (and more if possible)
My current setup, SetTimer() works fine on my calc SH4, but breaks badly on the Graph 35+EII
For some reason?
Citer : Posté le 16/02/2020 08:08 | #
I want this to be compatible with Graph 35+EII, SH3 and SH4 (and more if possible)
My current setup, SetTimer() works fine on my calc SH4, but breaks badly on the Graph 35+EII
For some reason?
I think it is best to create a dedicated routine for each model.
Looping to create a delay of about 1 microsecond.
volatile is to prevent deletion by optimization.
Overclocking utilitaire Ftune/Ptune2/Ptune3 est également disponible.
Si vous avez des questions ou un rapport de bogue, n'hésitez pas à me le faire savoir.
Citer : Posté le 16/02/2020 09:01 | #
Good catch, Sentaro. Indeed you can try and use the 32768 Hz hardware timer. However, as I explained before, you cannot get an interrupt out of it. What I missed is that Redmcd is already doing active waiting here, so doing it with a hardware timer is not going to change much.
Edit: There is one 32768 Hz timer on SH3 but it's used to power SetTimer().
Redcmd, would you mind if I made a few tests with gint? I have a few ideas I'd like to try, but I don't want to make it sound like I'm competing with you.
Citer : Posté le 16/02/2020 09:59 | #
Go ahead :ThumbsUp:
One single person can’t come with every idea possible
Citer : Posté le 16/02/2020 10:46 | #
@Lephenixnoir
Thanks!
Is it possible to read the value of 32768 Hz timer in SH3?
Overclocking utilitaire Ftune/Ptune2/Ptune3 est également disponible.
Si vous avez des questions ou un rapport de bogue, n'hésitez pas à me le faire savoir.
Citer : Posté le 16/02/2020 10:51 | #
Yes. You can read from 0xa44c0030. First is TSTR (1 byte), then 3 bytes gap, then TCOR (4 bytes), TCNT (4 bytes) and TCR (1 byte). The structure is the same as on SH4, but there is only one 32768 Hz, while there are six on SH4.
If you want to use this you will probably need to stop all timers of SetTimer() first, or observe the counter without reconfiguring the timer. The default constant should be about 820. Note that you can also use the normal hardware timers that are connected to the CPG and count at Pϕ/4. To the best of my knowledge they are not used too much, you can probably find an unused one automatically.
More information about the 32768 Hz timers is available in the Planète Casio bible.
Citer : Posté le 16/02/2020 11:54 | #
Thank you very much!
Overclocking utilitaire Ftune/Ptune2/Ptune3 est également disponible.
Si vous avez des questions ou un rapport de bogue, n'hésitez pas à me le faire savoir.
Citer : Posté le 17/02/2020 10:20 | # |
Fichier joint
What's the movua.l instruction?
It doesn't show up in https://bible.planet-casio.com/common/hardware/mpu/sh3_manual.pdf
7.1.1 Data Transfer Instructions
With doing this
mov.b r0, @r2 ;*LCD_data_register = (byte)long
shlr8 r0, ;long >>= 8;
mov.b r0, @r2 ;*LCD_data_register = (byte)long
shlr8 r0, ;long >>= 8;
mov.b r0, @r2 ;*LCD_data_register = (byte)long
shlr8 r0, ;long >>= 8;
mov.b r0, @r2 ;*LCD_data_register = (byte)long
VRAM being 1 aligned doesn't matter cause I'm not using it for Gray
But only problem is that the bytes are loaded to the display backwards
so instead of 1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16 it loads them like 4,3,2,1,8,7,6,5,12,11,10,9,16,15,14,13
Maybe I can just mix up the buffer so it gets unmixed when loaded to the display
byte = Gray.VRAM.Dark[i + 0];
Gray.VRAM.Dark[i + 0] = Gray.VRAM.Dark[i + 3];
Gray.VRAM.Dark[i + 3] = byte;
byte = Gray.VRAM.Dark[i + 1];
Gray.VRAM.Dark[i + 1] = Gray.VRAM.Dark[i + 2];
Gray.VRAM.Dark[i + 2] = byte;
byte = Gray.VRAM.Light[i + 0];
Gray.VRAM.Light[i + 0] = Gray.VRAM.Light[i + 3];
Gray.VRAM.Light[i + 3] = byte;
byte = Gray.VRAM.Light[i + 1];
Gray.VRAM.Light[i + 1] = Gray.VRAM.Light[i + 2];
Gray.VRAM.Light[i + 2] = byte;
}
Did a small GUI update and added that code above
Addin is slowly getting bigger: 34KB
MANDEL.G1A
Citer : Posté le 17/02/2020 13:25 | #
It's a unaligned move. It has been added in SH4 models so you need to look into the SH4 documentation.
It's basically movual.l @rm, rn (or movua.l @rm+, rn) and it loads four consecutive bytes from address rm into rn. When rm is a multiple of 4 it is equivalent to mov.l @rm, rn (or mov.l @rm+, rn). But if rm is not a multiple of 4, then it still loads four bytes from address rm while mov.l would fail with an address error.
There is no corresponding unaligned-write instruction.
Maybe I can just mix up the buffer so it gets unmixed when loaded to the display
Or you can sacrifice a few cycles and keep the order. You should compare performance if this matters to you, although all of this is about 0.25% of the total screen update time so it's not going to change much I believe...
swap.w r0, r0
swap.b r0, r0
mov.b r0, @r2
swap.b r0, r0
mov.b r0, @r2
swap.w r0, r0
swap.b r0, r0
mov.b r0, @r2
swap.b r0, r0
mob.b r0, @r2
Citer : Posté le 12/09/2020 06:52 | #
Is it possible to test if the power cable (usb cable) is connected? and/or not running on battery?
When I connect my calc (9750) to power, the RTC clock seems to run faster
Citer : Posté le 12/09/2020 06:59 | #
Take the batteries out whilst connected via USB.
Citer : Posté le 12/09/2020 07:01 | #
I mean from within software
Like a SYSCALL
But syscall Battery_GetStatus() returns 0 when on battery and power
Citer : Posté le 12/09/2020 07:05 | #
Simon's battery syscall documentation does not list anything that would be able to check if the calc is on USB power - I suspect that's a circuitboard-level feature disconnected from the kernel.
Citer : Posté le 12/09/2020 10:21 | #
The A/DC will likely output a different recognizable pattern when running over USB.
Citer : Posté le 12/09/2020 10:54 | #
I swaped out the Delay() function I was using in the gray engine
For the timer value at 0xA44D0038 (SH4)
Once the gray is tuned (no vsync lines etc) I can OC the calc, without having to retune it again
But when pluging into power, I get heaps of vsync lines and have to retune again
If I use the 0xA44D00D8 32768Hz timer value.
Pluging it in or running on battery doesn't desync the gray
But OC now does
Using the old Gint Demo program
The Timer/RTC comparison test stays at 100% Accuracy on battery
But slowly drops to 99.99% accurate on power once the timer hits roughly 0x0900
And 99.98% at around 0x1500
99.97% at 0x1b00 etc...
Here are some timer addresses (SH4) and some of the values I've seen on them:
0xA413FEC0; 0x00000000
0xA449000C; 0xFFFFFFFF
0xA4490018; 0xFFFFFFFF and 0x000167DD
0xA4490024; 0xFFFFFE33 // last digits flicker
0xA44D0038; 819 // Used by SetTimer and keyboard // 819 ticks = 25ms
0xA44D0058; 0 // set to 819 after restarting calc
0xA44D0078; 0 // set to 819 after restarting calc
0xA44D0098; 0 // set to 819 after restarting calc
0xA44D00B8; 0 - 9000 // Used when screen sharing to PC
0xA44D00D8; 32bit down counter // 32768Hz
https://bible.planet-casio.com/simlo/chm/v20/fx_7305_Registers.htm
https://bible.planet-casio.com/simlo/chm/v20/fxCG20_timer.htm
https://bible.planet-casio.com/lephenixnoir/etmu.html
Is it possible to make my own interrupt timer?
Like the SDK's SetTimer(). But with better timings other than just 25ms
Citer : Posté le 12/09/2020 11:14 | #
This is interesting. Unfortunately I don't know how this could be handled propertly. If it's the RTC, using a TMU might help. TMU-based versions of gint's gray engine do not produce different visual results when running on battery and USB power.
No. Controlling interrupts is the basis for gint. There are three stages, probably none of which is a satisfying answer for you.
• You can take control of interrupts and handle only the timer (Revolution-FX method). Breaks everything else, mandatory reboot when add-in finishes.
• You can take control of interrupts, handle only the timer, and redirect the rest (Kristaba's auto-debugger method). Expect pain debugging the OS since the OS' interrupt handler will take back VBR control from you.
• You can take control of interrupts and handle everything (the gint method), but there's no point in reimplementing everything in your project.
However, there is a more viable way. You can start a TMU with interrupts disabled and just watch the counter run as you do with the ETMU. It will give similar results without you needing to care about interrupts at all.
TMU work almost exactly like ETMU, they just have a couple more settings. See the SH7724 documentation for details, or this tmu.h for a bit of reusable code. Basically you want to set eg. TMU0 to run on Pphi/64 or faster (most precise is Pphi/4), set the delay in TCOR and TCNT, and start watching. gint uses Pphi/64 with the following delays. If I remember correctly you only use one shade of gray but hopefully this can help you find good-looking settings.