The JS implementation does content hashing to not blit unchanged
framebuffer contents (see mihaip/dingusppc@171ff2d407).
However, that is not necessary for the ATI adapters that already track
this and only set draw_fb if the framebuffer has actually changed.
Pass through a fb_known_to_be_changed for these cases, and also add an
optional update_skipped method (since the JS still wants to know when
the last logical screen update was).
Switching from Watch to Arrow cursor at vertical position zero (top of the screen) would sometimes cut off the top half of the Arrow cursor. The following changes fix this:
Use the new cursor_dirty flag to signal when the cursor should be updated. This reduces the number of cursor updates and doesn't depend on registers being accessed in a specific order.
Set the cursor_dirty flag for any change that should cause the hardware cursor to be updated:
- CUR_CLR0, CUR_CLR1: Cursor color changes. We don't check if the colors actually changed (all cursors are usually black and white). Rather, writes to these registers usually means the cursor bytes have changed or will change.
- CUR_OFFSET: Offset to cursor bytes.
- CUR_VERT_OFF: First vertical line of cursor to draw.
Other changes that don't require the cursor to be updated:
- CUR_HORZ_OFF: Horizontal offset of cursor position. The cursor is unchanged - just need to adjust the drawing position.
- CUR_HORZ_VERT_POSN: The cursor is unchanged - only the drawing position is changed.
The only thing that could change the cursor that we don't check is a change to the cursor bytes.
- Read 8 bytes at a time instead of just 1.
- Remove multiply operations from loop. We just need increments or additions.
- Change compares with int to compares with zero.
CUR_HORZ_OFF becomes non-zero when the cursor needs to be drawn to the left of the left edge of the frame buffer.
CUR_VERT_OFF is handled differently. When CUR_VERT_OFF is non-zero, CUR_OFFSET is changed to point to the first line of the cursor that will be drawn, so CUR_VERT_OFF is the number of lines to remove from the total height of the cursor.
Alternatively, we could handle CUR_VERT_OFF the same way as CUR_HORZ_OFF by leaving the cursor height constant, drawing the cursor starting from the CUR_VERT_OFF line, and adjusting cursor Y position by negative CUR_VERT_OFF.
Add pixel format and pixel clock to the list of fields that will initiate a recalculation.
If frame rate is less than 24 or greater than 120 then assume 60Hz.
Consider write-only bits: ATI_CLOCK_STROBE can't be read so clear it.
8 bits at Offset 2 is PLL_DATA. If we don't modify PLL_DATA, then insert the current value of PLL_DATA into the value that will be read from ATI_CLOCK_CNTL.
When checking if a particular byte of a register is accessed, check both the starting position (offset) and ending position (offset + size) of the bytes being access.
- Add BAR 2 decode. This BAR isn't actually used by Mac OS X, but decode it anyway just in case.
- Support updating of BARs (using change_one_bar method).
Cherrypicks a small piece of joevt/dingusppc@117ca1e449
so that booting from the 10.2 CD gets past it trying to change the video
mode to 15bpp.
Co-authored-by: joevt <joevt@shaw.ca>
Result of running IWYU (https://include-what-you-use.org/) and
applying most of the suggestions about unncessary includes and
forward declarations.
Was motivated by observing that <thread> was being included in
ppcopcodes.cpp even though it was unused (found while researching
the use of threads), but seems generally good to help with build
times and correctness.
The simplest solution is to cut the aperture size by the amount
of video RAM installed. This way, accesses to the big-endian
aperture located above the installed VRAM will be catched and
reported by the MMU.
- Don't log anything if the I/O access is not for this device. A different device might handle it.
- Don't return true for I/O access if an I/O access is not performed. Otherwise the I/O access won't be passed to other devices.
