Skip to content

perf(kernel): the NV2A ack thread backs off when busy, off Windows #113

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
phobos665 merged 1 commit into from
Nov 2, 2026
Merged

perf(kernel): the NV2A ack thread backs off when busy, off Windows #113

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 17 additions & 5 deletions src/kernel/xbox_memory_layout.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2631,19 +2631,31 @@ static void framebuffer_probe_tick(void)
* seconds, this thread's CPU time and how many changes a pass found right
* after a sleep (each of those may have waited for the sleep to end).
*
* Why the busy backoff is off by default. TimeSplitters 2 in its level on an
* M4, quiet machine, back-to-back runs (Oct 2026): with the old spin this
* Why the busy backoff is on by default off Windows. TimeSplitters 2 in its
* level on an M4, quiet machine, back-to-back runs (Oct 2026): with the old spin this
* thread used 21% of a core, with RECOMP_NV2A_ACK_BUSY_US=50 1.7%, and
* D3D_BlockOnTime -- the fence wait -- never waited in either (63 calls a
* second, ~7 us a second in total, none over 20 us); frame rate, "title's
* Swap" and "rest of frame" stayed within 0.1 ms. But ~170 changes a second
* were found right after a sleep, and a title that waits on the FIFO's GET
* or on a fence more often than TimeSplitters 2 does (Burnout 2 and OutRun 2
* pace on the fence) would wait out the sleep each time. Turn it on after
* those have been measured. */
* pace on the fence) would wait out the sleep each time.
*
* Measured on an M4 (Nov 2026), 0 against 50, alternated runs: OutRun 2, which
* paces on the fence, held 60 fps with the same frame times capped and
* uncapped while this thread went from 8% of a core to 1.7%; TimeSplitters 2
* went from 20-28% to 2-3% and Constantine from 25-35% to 2%, at a cost of
* 1.5-3% of their uncapped frame rate (117 against 119 fps, 92 against 94)
* and nothing measurable at 60. A fifth of a core is the larger cost on a
* laptop or a phone, so it is 50 off Windows. Windows keeps 0 until it has
* been measured there; Burnout 2 has not been measured anywhere. */
#define ACK_BUSY_YIELDS 8
#ifndef ACK_BUSY_US_DEFAULT
#define ACK_BUSY_US_DEFAULT 0
# ifdef _WIN32
# define ACK_BUSY_US_DEFAULT 0
# else
# define ACK_BUSY_US_DEFAULT 50
# endif
#endif

/* The idle wait, RECOMP_NV2A_ACK_IDLE_US. On POSIX it is a plain sleep
Expand Down
Loading