Cumulative optimizations #336
Conversation
ps2.c
Outdated
| void | ||
| mouse_button_down(int num) | ||
| { | ||
| buttons |= 1 << num; | ||
| mouse_send_state(); |
There was a problem hiding this comment.
Pull request #321 removed those function calls, for a good reason. Why did you add them back?
There was a problem hiding this comment.
That's essentially a merge error since I completely rewrote ps2.c/ps2.h on my fork. Will fix.
There was a problem hiding this comment.
Should be cleaned up now. Sorry about that. And the emulator should be even faster still as a result. 😃
|
Changes resolved with latest submits to master. |
|
With the latest i2c stuff checked in, I'm seeing 76% -> barely eeking out 100% on an RPi4. I'll be looking at the i2c stuff eventually, I'm sure, to see if I can't give it the same treatment as the ps2 emulation. But not for this PR, and probably not for at least a week. |
| void mouse_move(int x, int y); | ||
| void mouse_send_state(); | ||
| // fake mouse | ||
| void mouse_button_down(int num); |
There was a problem hiding this comment.
please no indentation to line up function names
There was a problem hiding this comment.
Will try to keep this in mind in the future. I blame clang-format.
| @@ -18,12 +19,19 @@ | |||
| // VIA#1 | |||
| // | |||
| // PA0: PS2KDAT PS/2 DATA keyboard | |||
| // PB0-2 ROM bank | |||
There was a problem hiding this comment.
where do these outdated lines come from? :)
| @@ -37,59 +45,55 @@ | |||
|
|
|||
| static uint8_t via1registers[16]; | |||
|
|
|||
| void | |||
| via1_init() | |||
| void via1_init() | |||
There was a problem hiding this comment.
please keep the type in a separate line, like in all other code
| { | ||
| uint8_t out_mode = reg_composer[0] & 3; | ||
|
|
||
| bool new_frame = false; | ||
| float advance = ((out_mode & 2) ? NTSC_PIXEL_FREQ : VGA_PIXEL_FREQ) / mhz; | ||
| float advance = ((out_mode & 2) ? NTSC_PIXEL_FREQ : VGA_PIXEL_FREQ) * steps / mhz; |
There was a problem hiding this comment.
no identation of this kind, please. it will keep diffs minimal.
|
Never mind, I merged it, and I'll change the nits myself :) |
* Removing branches and needless loops from ps2, joystick, video_step, vera_spi_step. * byte -> value: Variable name change: Windows SDK conflict. * Documentation updates
|
amended by changes the commit, new commit: abfe032 |
* Removing branches and needless loops from ps2, joystick, video_step, vera_spi_step. * byte -> value: Variable name change: Windows SDK conflict. * Documentation updates
|
Awesome, thanks! |
A collection of optimizations ported over from my C++-based fork of the X16 emulator:
Altogether, warp speed on a Raspberri Pi4 has gone from 101% to 157%.
Non-warp speed on a Raspberri Pi4 has gone from 80% to a rock-steady 100%.
I'm certain my previous experiments with RPi4 pegged performance in the mid-30% range, and I cannot explain the discrepancy in baseline performance between then and now. Go team? Compiler optimizations added at some point? RPi4 update made more resources available? Whatever the case, this still represents a significant jump in performance by removing the most significant bottlenecks outside of the hot paths of the VERA and base CPU emulation, and all platforms will benefit from this.