Prepare shaders under the code hashes play asks for, from every shader package, with better guessed bind states.

The install step seeded only 63% of the programs a play session met under their run-time code hash. A Switch 2 program with an empty c1 range past its code was hashed with that range, the Switch 2 agl and NDSR packages had no reader, and Layout2Shader/ was not walked. It now seeds 99.3% of the programs play asks for.

A default variant's answers are read again from its first translation (runtime/gfx_shader_guess.c): vertex inputs the shader reads as integers become uint, and a compute shader takes the texture and image types its translation declares. Task and mesh defaults carry the workgroup words play answers with. Against a play session's recorded variants, --prepare-shaders now matches 98.3%; the plain default guess matched 89.5%.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
chrg6
2026-11-07 22:21:28 -05:00
parent d0da862c87
commit fc8a9523a4
5 changed files with 204 additions and 23 deletions

View File

@@ -2118,7 +2118,10 @@ void gfx_shader_record_flush(int force) {
// ---- preparation from record files: the answers met in play (NX2NAVI_TEST_PREPARE_SHADERS=1) or the seed that
// tools/prepare_shaders.py writes from its shader list (nx2navi.exe --prepare-shaders). Both use the record
// layouts above; a seed record has a zero input id (the use mask, hence the id, is the translator's to find) and its
// pad byte holds the list priority (0 title, 1 warm-up, 2 rest).
// pad byte holds the list priority (0 title, 1 warm-up, 2 rest), with SEED_DEFAULT set on a default variant: its
// answers are read again from its first translation (gfx_shader_guess.c) and the variant it resolves to is the
// translation with those, when they differ.
#define SEED_DEFAULT 0x80 // tools/prepare_shaders.py DEFAULT_FLAG
typedef struct { const uint8_t *rec; } PrgIdx;
typedef struct { PrgIdx *idx; size_t n; } PrgList;
typedef struct {
@@ -2126,7 +2129,7 @@ typedef struct {
const uint8_t *ans;
size_t n_ans;
uint8_t (*iid)[32]; // per answers record: the resolved input id, zero on failure
volatile int64_t next, done, missing, failed;
volatile int64_t next, done, missing, failed, refined;
} Prep;
static int prg_cmp(const void *x, const void *y) { return memcmp(((const PrgIdx *)x)->rec + 12, ((const PrgIdx *)y)->rec + 12, 32); }
@@ -2152,9 +2155,12 @@ static void prep_worker(void *arg) {
uint32_t st, len;
memcpy(&st, e->rec, 4), memcpy(&len, e->rec + 4, 4);
const uint8_t *control = e->rec + 44, *code = control + CONTROL_SIZE, *c1 = code + len;
GfxShaderAnswers a;
GfxShaderAnswers a, b;
unpack_answers(r + 68, &a);
const GfxShader *s = gfx_shader_get((int)st, control, code, len, c1, &a);
const GfxShader *s = gfx_shader_get((int)st, control, code, len, c1, &a), *t;
if (s && r[ANS_REC - 1] & SEED_DEFAULT && gfx_shader_guess_refine(s, (int)st, &a, &b) &&
(t = gfx_shader_get((int)st, control, code, len, c1, &b)))
s = t, plat_inc64(&p->refined);
if (!s) plat_inc64(&p->failed);
else if (p->iid) memcpy(p->iid[i], s->iid, 32);
plat_inc64(&p->done);
@@ -2195,8 +2201,9 @@ static long long prepare_files(const char *ans_path, const char *prg_path, int t
(unsigned long long)((uint64_t)n_xlated - xl0), (double)plat_now_ms() - t0);
svc_set_limit(svc_runtime_limit());
rt_log("gfx_shader: prepare: %zu %s of %zu programs: %llu translated, %lld failed, %lld without a program, the rest "
"cached; %.0f ms", p.n_ans, what, p.progs.n, (unsigned long long)((uint64_t)n_xlated - xl0), (long long)p.failed,
(long long)p.missing, (double)plat_now_ms() - t0);
"cached; %lld default variants with refined answers; %.0f ms", p.n_ans, what, p.progs.n,
(unsigned long long)((uint64_t)n_xlated - xl0), (long long)p.failed, (long long)p.missing, (long long)p.refined,
(double)plat_now_ms() - t0);
free(p.progs.idx), free(pb);
if (ans_out) *ans_out = ab;
else free(ab);
@@ -2236,7 +2243,8 @@ int gfx_shader_prepare_seed(const char *dir, int threads, int procs) {
if (!memcmp(iid[i], zero, 32) || !set_add(&seen, hash8(iid[i]))) continue;
uint8_t *r = recs + u * PRP_REC;
const uint8_t *a = ab + (size_t)i * ANS_REC;
r[0] = a[0] < 6 ? a[0] : 0, r[1] = a[ANS_REC - 1] > 2 ? 2 : a[ANS_REC - 1];
uint8_t prio = a[ANS_REC - 1] & ~SEED_DEFAULT;
r[0] = a[0] < 6 ? a[0] : 0, r[1] = prio > 2 ? 2 : prio;
memcpy(r + 4, iid[i], 32);
by[r[1]]++, stages[r[0]]++, u++;
}