mirror of
https://github.com/tangerineturtle1/NX2navi.git
synced 2026-11-15 19:19:58 -04:00
a1fcbb0b07
The driver keeps the prepared shaders' libraries in its shader cache, and under Windows that cache has a size limit set in NVIDIA's settings. A limit smaller than the set drops the oldest libraries as new ones come in (a 2048 MB cache could not hold them on the reporter's RTX 5070 Ti), so every start builds them again in the background and play stutters until that is done. The cold lines of the library warm now say so on NVIDIA under Windows, most plainly when the cache is cold right after a warm that finished on the same driver; the build kit's notes and the example file say it too. tests/warm_hint_check.py checks which hint each case gets. From pull request #403 by tarballz; two conflicts resolved (an include and the build kit's notes). Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
56 lines
3.2 KiB
C
56 lines
3.2 KiB
C
// What the library warm's log adds when the driver's cache of stage libraries is cold (vk_pipeline.c lw_main), by GPU
|
|
// vendor, by platform and by why the warm found it cold. Own header so tests/warm_hint_check.c checks it without
|
|
// Vulkan.
|
|
//
|
|
// On NVIDIA under Windows the cache has a size limit, Shader Cache Size in the NVIDIA app. A limit smaller than the
|
|
// prepared libraries empties the cache as the warm fills it: the driver drops the oldest libraries to make room for
|
|
// the newest, and the next start's sample finds them gone. Measured on an RTX 5070 Ti (driver 617.42): the game's cache
|
|
// file stood at exactly 2048 MB, a full warm of 42,122 variants was cold again at the next start (a sample's median
|
|
// 241 ms), and with Shader Cache Size set to Unlimited the warm finished, the file grew to 2711 MB and later starts
|
|
// found the libraries in the cache (median 1.1 ms). Elsewhere there is no NVIDIA app and the driver's cache is set
|
|
// otherwise, so the hints that name the setting are for Windows only.
|
|
#pragma once
|
|
|
|
#include <stdint.h>
|
|
|
|
enum {
|
|
LW_COLD_FIRST, // cold with no finished warm before it on this driver: a driver update, a change of what
|
|
// the libraries are built with, or the build's preparation, which writes no shader-warm.txt
|
|
LW_COLD_AFTER_WARM, // cold again right after a finished warm on the same driver
|
|
LW_COLD_GAVE_UP, // cold after two finished warms in a row: warming is no longer tried on this driver
|
|
LW_WARM_FORCED, // lib_warm = force on a cache the sample found warm: nothing lost, no cause to name
|
|
};
|
|
|
|
#define LW_VENDOR_NVIDIA 0x10de
|
|
#ifdef _WIN32
|
|
#define LW_ON_WINDOWS 1
|
|
#else
|
|
#define LW_ON_WINDOWS 0
|
|
#endif
|
|
|
|
// A sentence or two for the end of the log line ("" when there is nothing to add), each starting with a space.
|
|
// windows: the game runs on Windows (LW_ON_WINDOWS), where NVIDIA's limit is the NVIDIA app's Shader Cache Size.
|
|
static inline const char *lw_cold_hint(uint32_t vendor, int why, int windows) {
|
|
int nvidia = vendor == LW_VENDOR_NVIDIA && windows;
|
|
switch (why) {
|
|
case LW_COLD_FIRST:
|
|
return nvidia ? " On NVIDIA the driver's Shader Cache Size limit empties it too, when it is smaller than the "
|
|
"prepared libraries (a 2048 MB cache dropped the oldest as the warm built the newest): "
|
|
"Shader Cache Size Unlimited in the NVIDIA app keeps them."
|
|
: "";
|
|
case LW_COLD_AFTER_WARM:
|
|
return nvidia ? " A finished warm did not stay in the driver's cache: on NVIDIA that is the Shader Cache Size "
|
|
"limit, smaller than the prepared libraries; set Shader Cache Size to Unlimited in the NVIDIA "
|
|
"app, or this warm is lost again."
|
|
: "";
|
|
case LW_COLD_GAVE_UP:
|
|
return nvidia ? " On NVIDIA that is the Shader Cache Size limit: after setting it to Unlimited in the NVIDIA "
|
|
"app, start once with lib_warm = force (or delete shader-warm.txt) to warm again."
|
|
: " Once the driver's cache can hold them, start once with lib_warm = force (or delete "
|
|
"shader-warm.txt) to warm again.";
|
|
case LW_WARM_FORCED:
|
|
return "";
|
|
}
|
|
return "";
|
|
}
|