mirror of
https://github.com/tangerineturtle1/NX2navi.git
synced 2026-11-15 19:19:58 -04:00
39fa0be78c
Only the large compute shaders' background builder looked at the system's commit before it started a build, and it saw nothing of what the builds beside it were about to take. The boot's recorded pipelines (up to 64 builders), the pool's background builds during play, the boot's compute pipelines and stage libraries and the library warm asked nobody. Each of them now asks the gate before a build. A build that starts reserves an estimate of the driver compiler's working memory until it ends (build_reserve_mb, 64 MB; 4096 MB for a large compute shader), and the gate admits a build only while the commit headroom holds the part kept free (build_free_mb, 4096 MB), the reservations and its own. A refused boot builder ends and leaves its pipelines to the pool; the pool leaves a background build queued and asks again every two seconds; the stage library prebuild and the library warm wait up to 30 seconds and then end, and the warm goes on at the next start. A pipeline a draw or a dispatch waits for is never refused. The preparation of the pipeline list runs without the gate: no game runs beside it. tests/build_gate_check.py checks the gate's arithmetic and, through --probe-buildgate, its answers with the machine's own commit. What the builders do when refused is not exercised by a test. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
38 lines
2.2 KiB
C
38 lines
2.2 KiB
C
// The memory gate of the background pipeline builds (vk_pipeline.c gate_enter): whether a build may start, given the
|
|
// system's commit and what the builds already under way were promised. Own header so tests/build_gate_check.c checks
|
|
// it without Vulkan.
|
|
//
|
|
// A pipeline build runs the driver's shader compiler in this process, and what the compiler takes while it works is
|
|
// committed memory like any other. A gate that only reads the commit in use sees none of it for the builds that have
|
|
// just started: sixteen builders that each find "enough room" start together and take sixteen shares of the same
|
|
// room. So every build that starts reserves an estimate of what its compile takes, the gate counts the reservations
|
|
// as if they were in use already, and a build that does not fit is refused: it does not start, and its caller leaves
|
|
// it to a later moment or to the draw that first needs the pipeline. A machine at its commit limit fails allocations
|
|
// (Windows) or pages until the driver gives the device up; a pipeline built later costs one hitch.
|
|
#pragma once
|
|
|
|
typedef struct {
|
|
double free_mb; // commit headroom in MB that stays free of background builds; 0 or less: nothing is refused
|
|
double reserved_mb; // the estimates of the builds under way
|
|
} BuildGate;
|
|
|
|
// 1 when a build estimated at est_mb may start with used_mb of limit_mb of the system's commit in use. A limit below
|
|
// 0 is an unknown one (platform.h plat_commit_info): nothing is refused.
|
|
static inline int build_gate_admits(const BuildGate *g, double used_mb, double limit_mb, double est_mb) {
|
|
if (g->free_mb <= 0 || limit_mb < 0) return 1;
|
|
return limit_mb - used_mb - g->reserved_mb >= g->free_mb + est_mb;
|
|
}
|
|
|
|
// Reserves est_mb for a build that starts now; 0 when it is refused, and then nothing is reserved.
|
|
static inline int build_gate_enter(BuildGate *g, double used_mb, double limit_mb, double est_mb) {
|
|
if (!build_gate_admits(g, used_mb, limit_mb, est_mb)) return 0;
|
|
g->reserved_mb += est_mb;
|
|
return 1;
|
|
}
|
|
|
|
// The build that reserved est_mb is over.
|
|
static inline void build_gate_leave(BuildGate *g, double est_mb) {
|
|
g->reserved_mb -= est_mb;
|
|
if (g->reserved_mb < 0) g->reserved_mb = 0;
|
|
}
|