mirror of
https://github.com/tangerineturtle1/NX2navi.git
synced 2026-11-17 04:59:57 -04:00
Let a texture slot follow another texture at its storage only when the two are the same size.
When a slot's own entry was released, slot_resolve took any live entry at the same address, and Ocarina's per-frame render targets share storage. The half-size ambient occlusion term (640x360 at a 1280x720 scene) followed the bloom pass's 480x270 target there every frame: the blur that writes the term kept its top-left 480x270, the upsample stretched that over the screen, and every contact shadow sat 1.33 times further from the top-left corner than its object (doubled shadows below and right of Link, shadows of terrain out of view). During a hit effect the same slots followed its 1130x86 strip instead, the term read zero over most of the screen, and the right 44% of the scene went black. Block-compressed formats keep the old rule, as a GPU encoder renders their blocks through a smaller view of the same storage. The first eight refusals are logged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
@@ -500,6 +500,28 @@ static int same_but_swizzle(const TexKey *a, const TexKey *b) {
|
||||
return !memcmp(&x, &y, sizeof x);
|
||||
}
|
||||
|
||||
// Whether a slot (key k) may follow entry p, another descriptor at the same storage, when the slot's own entry is
|
||||
// released (slot_resolve). Ocarina's per-frame render targets share storage, so the live entry at an address is often
|
||||
// a target of another size that the frame laid out there: an effect's 1130x86 strip over the 480x270 half-size
|
||||
// lighting term at a 1280x720 scene. Followed, the compute pass that writes the term and the full-screen pass that
|
||||
// reads it both got the strip's image, the term read back as zero over most of the screen and a jump attack's hit
|
||||
// effect drew the right 44% of the scene black. A slot follows only an entry of its own size; block-compressed
|
||||
// formats are left out of the test, as a GPU encoder renders their blocks through a smaller uncompressed view of the
|
||||
// same storage. Counted in n_follow_refused.
|
||||
static uint64_t n_follow_refused;
|
||||
static int follow_size_ok(const Tex *p, const TexKey *k) {
|
||||
int bw1 = 1, bh1 = 1, bw2 = 1, bh2 = 1;
|
||||
nvn_fmt_bpb(p->key.d.format, &bw1, &bh1);
|
||||
nvn_fmt_bpb(k->d.format, &bw2, &bh2);
|
||||
if (bw1 > 1 || bh1 > 1 || bw2 > 1 || bh2 > 1) return 1;
|
||||
if (p->key.d.width == k->d.width && p->key.d.height == k->d.height && p->key.d.depth == k->d.depth) return 1;
|
||||
if (n_follow_refused++ < 8)
|
||||
rt_log("vk: a texture slot at %#llx (%ux%u format %#x) did not follow a %ux%u texture at the same storage: "
|
||||
"its own image instead", (unsigned long long)k->addr, k->d.width, k->d.height, k->d.format,
|
||||
p->key.d.width, p->key.d.height);
|
||||
return 0;
|
||||
}
|
||||
|
||||
// Textures whose content the CPU provides in pool memory. Display, compressible, zcull,
|
||||
// depth, multisample and GPU-only storage stays GPU-owned.
|
||||
// mirror_refused: NULL for such a texture, else the rule that leaves it out, in words for a log line.
|
||||
@@ -4129,7 +4151,7 @@ static Tex *slot_resolve(TexSlot *s, view_t *view, int *has_view) {
|
||||
// With the target kept they get the volume itself.
|
||||
if (p->key.d.target != k.d.target) continue;
|
||||
if (same_but_swizzle(&p->key, &k) && (!cpu_slot || !s->finalized || alias_has_gpu_content(p))) { e = p; break; }
|
||||
if (!any && (!cpu_slot || alias_has_gpu_content(p))) any = p;
|
||||
if (!any && (!cpu_slot || alias_has_gpu_content(p)) && follow_size_ok(p, &k)) any = p;
|
||||
}
|
||||
// A finalized CPU descriptor must not recreate its image from freed storage. The slot keeps this state even
|
||||
// after its released image was evicted; registering the descriptor again makes it current. While its own
|
||||
|
||||
Reference in New Issue
Block a user