Files
NX2navi/docs/releases/v0.8.1.md
epicmeme546 4c5855ce43 Hold ASTC textures whose every block decodes opaque as BC1 instead of BC3: the same texels in half the video memory.
The BC3 encoder's colour half is a four-colour BC1 block (c0 above c1, or every index 0 when they are equal), so a BC1 image of an opaque texture decodes to exactly the texels the BC3 one did. The encoder writes that half alone with its specialization constant BC1 set; its push constants are as before, so the BC3 pipeline is unchanged.

Every ASTC texture the CPU provides is made BC3, and its first upload checks its blocks while it copies them into staging: when all decode with alpha 255 (tex_astc.c astc_blocks_opaque, from the block header for all but the endpoint modes with alpha, which are decoded), the entry gets a BC1 image in place of the BC3 one before the decode. Later uploads into a BC1 image check again and go back to BC3 for content with alpha. A large texture's copy and check run on the ASTC decode workers, so its first upload copies faster than before. A check of its own where the texture is made cost 1.7 s of the render thread over a seven-minute teleport tour.

On that tour (Kakariko Village, the Graveyard, Death Mountain Trail, Kokiri Forest, Hyrule Field, RX 7900 XT), against BC3 alone in the same batch: the ASTC textures in Hyrule Field at the end 2,568 instead of 3,104 MiB, video memory at its peak 11,814 instead of 12,374 MiB, process commit at its peak 25,539 instead of 26,139 MiB; the first 7,200 frames in 93.7 instead of 101.3 s and the run's slowest 1 % at 24.1 instead of 21.1 fps. --probe-astc checks the header test against a decode of every generated block, the loop, its copy and the worker pool against it, and each BC1 block against the colour half of its BC3 block. NX2NAVI_ASTC_BC1=0 (astc_bc1 = 0) keeps every ASTC texture BC3; the sampling probe NX2NAVI_ASTC_BC_PROBE it replaces is gone.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-12 02:11:46 -05:00

6.1 KiB

Fixed

  • Stutter while running through open areas. To skip memory the game has not written for a while when it looks for changes each frame, the runtime write-protects it in 2 MiB pieces, and the game's first write takes the protection off again. The buffers the game streams the next stretch of ground into were protected again as soon as they had been idle a moment and written again moments later: running from the Kakariko Village gate across Hyrule Field did that some 30,000 times in nine minutes, and while the game streamed, each time took Windows 15 to 20 ms, with the game waiting. A piece the game writes again soon after it was protected is now left alone for a while, longer each time. On that run: no frame over 100 ms while running where there had been 29 to 52 a minute, the slowest 1 % of frames at about 20 fps instead of 7 to 9, and standing at the edge of Lake Hylia afterwards 60 fps instead of 38.
  • Ground that vanished in open areas, leaving a view into the void below. The game streams the terrain's height maps through a fixed set of memory and sets up the next piece of ground there before it lets go of the one before; the runtime took that letting go for the new piece too, and the piece was drawn at height zero, far below the ground. Seen after some minutes of running in Hyrule Field, and in Kokiri Forest, Kakariko Village and the Graveyard. It is no rare case: a Hyrule Field save does it 173 times while it loads, and running through the field 40 to 365 times a second. Found and fixed by MartinIsla (pull request #535).
  • Ground flickering white on PCs with 16 GB of memory. Past a bound on the game's own memory, textures gave way one after the other, although dropping them could never bring the game back under that bound: the bound leaves the video memory out, and a dropped texture lowers both sides alike. Textures now give way only when the system's memory is really running low (tex_proc_commit_mb only writes a line into the log now).
  • Stutter and textures missing for a moment on cards with 8 to 12 GB. The game lets go of textures and takes them back within frames, hundreds a frame in towns. Above 80 % of the card's memory those were dropped, only to be made and uploaded again a moment later, and the card never got under its budget: with an 8 GB card's budget, seven minutes of teleports between Hyrule Field, Kakariko Village, the Graveyard, Death Mountain Trail and Kokiri Forest dropped 82,288 of them, and with a 12 GB card's budget a texture was drawn while it was gone 178 times (9 times with the card's full 20 GB). Only textures that stayed let go of for two seconds, and that nothing looked up since, are dropped now: on the same tour with the 8 GB budget, 34,699 drops, a fifth fewer textures made again, a third less time the game waited on textures, 4 draws of a missing texture instead of 11, and in Hyrule Field afterwards the slowest 1 % of frames at 32 fps instead of 22; with the 12 GB budget 60 % less waiting on textures, 10 draws of a missing texture instead of 178, and 43 fps instead of 39 in the field afterwards.
  • AMD FSR 4 turning into FSR 3.1 by itself. Each upscaling mode made an FSR 4 context of its own (about 220 MB of video memory each, up to six of them beside FSR 3.1's), and three creations that failed in a row, as they can while a save loads and the area's textures fill the card, turned FSR 4 off until the game was restarted; the menu then showed FSR 3.1 and no longer listed FSR 4. Now one context per output size serves every mode, contexts nothing used for 90 seconds go, a creation that fails frees the others and tries again two seconds later, and should FSR 4 still give up, choosing it in the menu again tries it anew.
  • Debug menu: a teleport that left Link falling for good. A teleport to a start position the area does not have put Link where nothing held him, and the game refuses every teleport while he falls. A teleport now always goes to one of the area's own start positions, and when Link falls through the ground after arriving it takes him to the area's next one.
  • Debug menu: the free camera no longer goes under the map. Flown while looking down it sank into the ground, and flown level it went through hills, and below the ground the picture showed the void under the map, which looks like missing ground. The left stick now moves it level, only ZR and ZL change its height, and with DLSS, DLAA or FSR 3.1 or 4 on it rises where the ground ahead or below comes within a few metres (it reads the distances the upscaler's motion pass works out for the scene).

Changed

  • One Resolution row instead of Window size and Resolution. Each choice is a size in your game's shape (with 21:9: 2560 x 1080, 3440 x 1440, 4300 x 1800 and up) and sets both the picture's size and the window outside fullscreen, which takes it at once, within your monitor; the picture changes after a restart, as before. Native draws at the window's size and leaves the window as it is.
  • FSR 4 is named by the version your driver runs: AMD FSR 4.1 Native AA and so on with AMD's current drivers.
  • Less video memory for the game's textures, with the same picture. On cards that get the game's ASTC textures as BC3 (all with less than 20 GB, and larger ones on PCs with less than 48 GB of memory), a texture none of whose texels is see-through is now BC1, half the size: BC1 holds exactly the colour half of the BC3 blocks it replaces, so every texel is the same as before. About a third of the game's ASTC textures are, and in Hyrule Field after a seven-minute teleport tour through five areas the ASTC textures took 2.5 instead of 3.0 GB, the game's video memory at its peak 11.5 instead of 12.1 GB and its commit at its peak 24.9 instead of 25.5 GB (Windows counts video memory in the commit too). A texture's first upload checks its blocks while it copies them, and a large texture's copy now runs on several threads: the tour's first 7,200 frames, its area loads, took 94 instead of 101 seconds, and its slowest 1 % of frames ran at 24 instead of 21 fps. astc_bc1 = 0 in nx2navi.ini keeps every ASTC texture BC3.