// NVN front end, second slice: descriptor pools, textures, samplers, programs,
// windows with present pacing, and the state objects. Ported from MacNativeGames
// games/totk/third_party/d2r_gfx/gfx/nvn/hle_nvn.c @ b75be8cf (MIT, Copyright (c) 2026 jhaizhou; origin d2r-mobile),
// lines 254-535, 1775-1829, 2070-2215 and 2579-2727, and nvn_pace.h; Metal, mach, the translation thread, frame
// interpolation and Apple diagnostics dropped. Calls that reached Metal go to gfx_backend() (gfx.h).
//
// Objects are stored in place in the NVN object memory the game provides (nvn_obj.h); every argument is read from the
// AAPCS64 registers (host.h arg_x / arg_s / arg_p) and every result returned with ret_x. Command-buffer procs are in
// nvn_api_cmd.c, the TOTK extension procs in nvn_ext.c.
#include <stdint.h>
#include <stdlib.h>
#include <string.h>

#include "cfg.h"
#include "gfx.h"
#include "host.h"
#include "nvn_api.h"
#include "nvn_hmap.h"
#include "nvn_tex.h"
#include "platform/platform.h"

// ---------- Descriptor pools
// NVNboolean nvnTexturePoolInitialize(NVNtexturePool *tp, const NVNmemoryPool *p, ptrdiff_t offset, int n)
NVN_API(nvnTexturePoolInitialize) {
    *(dpool_t *)arg_p(c, 0) = (dpool_t){arg_p(c, 1), (int64_t)arg_x(c, 2), (int32_t)arg_x(c, 3)};
    NVN_LOG("TexturePool n=%d", (int)arg_x(c, 3));
    ret_x(c, 1);
}
// void nvnTexturePoolFinalize(NVNtexturePool *tp)
NVN_API(nvnTexturePoolFinalize) { (void)c; }
// void nvnTexturePoolRegisterTexture(const NVNtexturePool *tp, int id, const NVNtexture *t, const NVNtextureView *v)
NVN_API(nvnTexturePoolRegisterTexture) {
    gfx_backend()->texture_register((int)arg_x(c, 1), arg_p(c, 2), arg_p(c, 3));
}
// NVNboolean nvnSamplerPoolInitialize(NVNsamplerPool *sp, const NVNmemoryPool *p, ptrdiff_t offset, int n)
NVN_API(nvnSamplerPoolInitialize) {
    *(dpool_t *)arg_p(c, 0) = (dpool_t){arg_p(c, 1), (int64_t)arg_x(c, 2), (int32_t)arg_x(c, 3)};
    NVN_LOG("SamplerPool n=%d", (int)arg_x(c, 3));
    ret_x(c, 1);
}
// void nvnSamplerPoolFinalize(NVNsamplerPool *sp)
NVN_API(nvnSamplerPoolFinalize) { (void)c; }
// void nvnSamplerPoolRegisterSampler(const NVNsamplerPool *sp, int id, const NVNsampler *s)
NVN_API(nvnSamplerPoolRegisterSampler) {
    gfx_backend()->sampler_register((int)arg_x(c, 1), &((const samp_t *)arg_p(c, 2))->p);
}

// ---------- Texture builder
#define TB(c) ((tex_builder_t *)arg_p(c, 0))
// void nvnTextureBuilderSetDefaults(NVNtextureBuilder *b): keeps the device
NVN_API(nvnTextureBuilderSetDefaults) {
    tex_builder_t *b = TB(c);
    NVNdevice *dev = b->dev;
    memset(b, 0, sizeof *b);
    b->dev = dev;
    b->d.swizzle_r = 2;
    b->d.swizzle_g = 3;
    b->d.swizzle_b = 4;
    b->d.swizzle_a = 5;
    b->d.target = NVN_TEXTURE_TARGET_2D;
    b->d.width = b->d.height = b->d.depth = b->d.levels = 1;
}
NVN_API(nvnTextureBuilderSetDevice) { TB(c)->dev = arg_p(c, 1); }
NVN_API(nvnTextureBuilderSetFlags) { TB(c)->d.flags = (int)arg_x(c, 1); }
NVN_API(nvnTextureBuilderSetTarget) { TB(c)->d.target = (int)arg_x(c, 1); }
NVN_API(nvnTextureBuilderSetWidth) { TB(c)->d.width = (uint32_t)arg_x(c, 1); }
NVN_API(nvnTextureBuilderSetHeight) { TB(c)->d.height = (uint32_t)arg_x(c, 1); }
NVN_API(nvnTextureBuilderSetDepth) { TB(c)->d.depth = (uint32_t)arg_x(c, 1); }
NVN_API(nvnTextureBuilderSetSize1D) {
    tex_builder_t *b = TB(c);
    b->d.width = (uint32_t)arg_x(c, 1);
    b->d.height = b->d.depth = 1;
}
NVN_API(nvnTextureBuilderSetSize2D) {
    tex_builder_t *b = TB(c);
    b->d.width = (uint32_t)arg_x(c, 1);
    b->d.height = (uint32_t)arg_x(c, 2);
    b->d.depth = 1;
}
NVN_API(nvnTextureBuilderSetSize3D) {
    tex_builder_t *b = TB(c);
    b->d.width = (uint32_t)arg_x(c, 1);
    b->d.height = (uint32_t)arg_x(c, 2);
    b->d.depth = (uint32_t)arg_x(c, 3);
}
NVN_API(nvnTextureBuilderSetLevels) { TB(c)->d.levels = (uint32_t)arg_x(c, 1); }
NVN_API(nvnTextureBuilderSetFormat) { TB(c)->d.format = (int)arg_x(c, 1); }
// The back end samples the depth plane; stencil-plane sampling stops.
NVN_API(nvnTextureBuilderSetDepthStencilMode) {
    if ((uint32_t)arg_x(c, 1) != NVN_TEXTURE_DEPTH_STENCIL_MODE_DEPTH)
        rt_fatal("nvn: stencil-plane texture sampling is not implemented");
}
NVN_API(nvnTextureBuilderSetSwizzle) {
    uint32_t s[4] = {(uint32_t)arg_x(c, 1), (uint32_t)arg_x(c, 2), (uint32_t)arg_x(c, 3), (uint32_t)arg_x(c, 4)};
    for (int i = 0; i < 4; i++)
        if (s[i] > 5) rt_fatal("nvn: invalid texture swizzle %u", s[i]);
    nvn_texdesc *d = &TB(c)->d;
    d->swizzle_r = s[0];
    d->swizzle_g = s[1];
    d->swizzle_b = s[2];
    d->swizzle_a = s[3];
}
NVN_API(nvnTextureBuilderSetSamples) { TB(c)->d.samples = (int)arg_x(c, 1); }
NVN_API(nvnTextureBuilderSetStride) { TB(c)->d.stride = (uint32_t)arg_x(c, 1); }
// void nvnTextureBuilderSetStorage(NVNtextureBuilder *b, NVNmemoryPool *p, ptrdiff_t offset)
NVN_API(nvnTextureBuilderSetStorage) {
    tex_builder_t *b = TB(c);
    b->pool = arg_p(c, 1);
    b->off = (int64_t)arg_x(c, 2);
}
// void nvnTextureBuilderSetPackagedTextureLayout(NVNtextureBuilder *b, const NVNpackagedTextureLayout *l)
NVN_API(nvnTextureBuilderSetPackagedTextureLayout) {
    tex_builder_t *b = TB(c);
    const NVNpackagedTextureLayout *l = arg_p(c, 1);
    b->has_layout = l != NULL;
    if (l) memcpy(b->layout, l->layout, 8);
    b->d.pkg_block_h = l ? l->layout[0] + 1 : 0;
}
// void nvnTextureBuilderSetPackagedTextureData(NVNtextureBuilder *b, const void *data): packaged (already GPU-laid-out)
// data, copied into the texture's pool storage at Initialize
NVN_API(nvnTextureBuilderSetPackagedTextureData) { TB(c)->pkg_data = arg_p(c, 1); }
NVN_API(nvnTextureBuilderGetPackagedTextureData) { ret_x(c, (uintptr_t)TB(c)->pkg_data); }
// Storage classes are opaque compatibility identifiers: all texture storage lives in byte-addressed pools and is decoded
// per texture description, so class 1 names that convention; buffer textures use class 0. Not a Tegra memory kind.
static uint64_t storage_class(const nvn_texdesc *d) { return d->target == NVN_TEXTURE_TARGET_BUFFER ? 0 : 1; }
NVN_API(nvnTextureBuilderGetStorageClass) { ret_x(c, storage_class(&TB(c)->d)); }
NVN_API(nvnTextureBuilderGetFlags) { ret_x(c, (uint32_t)TB(c)->d.flags); }
// size_t nvnTextureBuilderGetStorageSize(const NVNtextureBuilder *b)
NVN_API(nvnTextureBuilderGetStorageSize) {
    nvn_texlayout L;
    nvn_tex_layout(&TB(c)->d, &L);
    if (!L.bpb) rt_log("nvn: TextureBuilderGetStorageSize: unknown format %#x", TB(c)->d.format);
    ret_x(c, L.size);
}
// size_t nvnTextureBuilderGetStorageAlignment(const NVNtextureBuilder *b)
NVN_API(nvnTextureBuilderGetStorageAlignment) {
    nvn_texlayout L;
    nvn_tex_layout(&TB(c)->d, &L);
    ret_x(c, L.align);
}

// ---------- Textures
// The game re-initializes transient textures (G-buffer and the like) every frame at the same pool memory without
// registering the descriptor again, as an NVN descriptor refers to memory and stays valid on hardware. The table maps
// storage address -> a copy of the live texture so the back end can redirect a registration whose serial was finalized.
static nvn_hmap live = {.klen = 8};
static PlatMutex live_lock = PLAT_MUTEX_INIT;

static void live_put(const tex_t *o) {
    if (!o->mem) return;
    uint64_t k = (uintptr_t)o->mem;
    tex_t *copy = malloc(sizeof *copy);
    if (!copy) rt_fatal("nvn: out of memory for the live texture table");
    *copy = *o;
    plat_mutex_lock(&live_lock);
    tex_t *old = nvn_hm_put(&live, &k, copy);
    plat_mutex_unlock(&live_lock);
    free(old);
}

static void live_del(const tex_t *o) {
    if (!o->mem) return;
    uint64_t k = (uintptr_t)o->mem;
    plat_mutex_lock(&live_lock);
    tex_t *cur = nvn_hm_get(&live, &k);
    if (cur && cur->serial == o->serial) nvn_hm_del(&live, &k);
    else cur = NULL;
    plat_mutex_unlock(&live_lock);
    free(cur);
}

int nvn_tex_live_at(const void *mem, tex_t *out) {
    if (!mem) return 0;
    uint64_t k = (uintptr_t)mem;
    plat_mutex_lock_shared(&live_lock);
    const tex_t *cur = nvn_hm_get(&live, &k);
    if (cur) *out = *cur;
    plat_mutex_unlock_shared(&live_lock);
    return cur != NULL;
}

// Releases the back end's texture and the live entry of an initialized texture object.
static void tex_release(tex_t *o) {
    live_del(o);
    gfx_backend()->texture_release(o);
}

// NVNboolean nvnTextureInitialize(NVNtexture *t, const NVNtextureBuilder *b)
NVN_API(nvnTextureInitialize) {
    tex_t *o = arg_p(c, 0);
    const tex_builder_t *s = arg_p(c, 1);
    if (o->magic == NVN_TEX_MAGIC && o->serial) {   // initialized again without Finalize: release the old serial
        NVN_LOG("texture %p initialized again without Finalize (serial %llu released)", (void *)o,
                (unsigned long long)o->serial);
        tex_release(o);
    }
    memset(o, 0, sizeof *o);
    o->magic = NVN_TEX_MAGIC;
    o->d = s->d;
    o->pool = s->pool;
    o->pool_flags = s->pool ? (uint16_t)s->pool->flags : 0;
    o->off = s->off;
    o->has_layout = s->has_layout;
    memcpy(o->layout, s->layout, 8);
    nvn_tex_layout(&o->d, &o->L);
    o->mem = s->pool && s->pool->mem ? s->pool->mem + s->off : NULL;
    if (s->pkg_data) {
        if (!o->mem || nvn_gpu_only((uintptr_t)o->mem))
            rt_fatal("nvn: packaged texture data without CPU-visible storage");
        memcpy(o->mem, s->pkg_data, (size_t)o->L.size);
        nvn_guest_written((uintptr_t)o->mem, o->L.size);
    }
    o->serial = nvn_new_serial();
    live_put(o);
    NVN_LOG("Texture %p %ux%ux%u lv%u fmt=%#x tgt=%d flags=%#x size=%#llx", (void *)o, o->d.width, o->d.height,
            o->d.depth, o->d.levels, o->d.format, o->d.target, o->d.flags, (unsigned long long)o->L.size);
    ret_x(c, 1);
}
// void nvnTextureFinalize(NVNtexture *t)
NVN_API(nvnTextureFinalize) {
    tex_t *o = arg_p(c, 0);
    if (o->magic == NVN_TEX_MAGIC && o->serial) tex_release(o);
    o->magic = 0;
}
#define CTX(c) ((const tex_t *)arg_p(c, 0))
NVN_API(nvnTextureGetStorageClass) { ret_x(c, storage_class(&CTX(c)->d)); }
NVN_API(nvnTextureGetFlags) { ret_x(c, (uint32_t)CTX(c)->d.flags); }
NVN_API(nvnTextureGetWidth) { ret_x(c, CTX(c)->d.width); }
NVN_API(nvnTextureGetHeight) { ret_x(c, CTX(c)->d.height); }
NVN_API(nvnTextureGetDepth) { ret_x(c, CTX(c)->d.depth); }
NVN_API(nvnTextureGetStorageSize) { ret_x(c, (uint32_t)(int)CTX(c)->L.size); }
// NVNboolean nvnTextureCompare(const NVNtexture *a, const NVNtexture *b)
NVN_API(nvnTextureCompare) {
    const tex_t *x = arg_p(c, 0), *y = arg_p(c, 1);
    ret_x(c, !memcmp(&x->d, &y->d, sizeof x->d) && x->mem == y->mem);
}
// int nvnTextureGetRawStorageClass(const NVNtexture *t): no prototype in nvn.h. Call sites compare the result with 1
// and set a status bit when it differs; the meaning is unverified, so 1.
NVN_API(nvnTextureGetRawStorageClass) { ret_x(c, 1); }

// ---------- Texture views
#define VW(c) ((view_t *)arg_p(c, 0))
NVN_API(nvnTextureViewSetDefaults) { memset(arg_p(c, 0), 0, sizeof(view_t)); }
NVN_API(nvnTextureViewSetLevels) {
    uint32_t base = (uint32_t)arg_x(c, 1), n = (uint32_t)arg_x(c, 2);
    if (base > 255 || n > 255) rt_fatal("nvn: texture view levels %d/%d", (int)base, (int)n);
    view_t *v = VW(c);
    v->set_levels = 1;
    v->base_level = (uint8_t)base;
    v->nlevels = (uint8_t)n;
}
NVN_API(nvnTextureViewSetLayers) {
    uint32_t first = (uint32_t)arg_x(c, 1), n = (uint32_t)arg_x(c, 2);
    if (first > 65535 || n > 65535) rt_fatal("nvn: texture view layers %d/%d", (int)first, (int)n);
    view_t *v = VW(c);
    v->set_layers = 1;
    v->min_layer = (uint16_t)first;
    v->nlayers = (uint16_t)n;
}
NVN_API(nvnTextureViewSetFormat) {
    uint32_t f = (uint32_t)arg_x(c, 1);
    if (f > 65535) rt_fatal("nvn: texture view format %d", (int)f);
    view_t *v = VW(c);
    v->set_format = 1;
    v->format = (uint16_t)f;
}
NVN_API(nvnTextureViewSetTarget) {
    uint32_t t = (uint32_t)arg_x(c, 1);
    if (t > 255) rt_fatal("nvn: texture view target %d", (int)t);
    view_t *v = VW(c);
    v->set_target = 1;
    v->target = (uint8_t)t;
}
// NVNboolean nvnTextureViewGetLayers(const NVNtextureView *v, int *min, int *n)
NVN_API(nvnTextureViewGetLayers) {
    const view_t *v = arg_p(c, 0);
    *(int *)arg_p(c, 1) = v->min_layer;
    *(int *)arg_p(c, 2) = v->nlayers;
    ret_x(c, v->set_layers);
}

uint64_t nvn_tex_view_offset(const tex_t *t, const view_t *v) {
    if (!v) return 0;
    uint64_t off = v->set_levels ? nvn_tex_level_offset(&t->d, v->base_level) : 0;
    if (v->set_layers) off += (uint64_t)v->min_layer * t->L.layer_size;
    return off;
}
// ptrdiff_t nvnTextureGetViewOffset(const NVNtexture *t, const NVNtextureView *v)
NVN_API(nvnTextureGetViewOffset) { ret_x(c, nvn_tex_view_offset(arg_p(c, 0), arg_p(c, 1))); }

// ---------- Texel access: the back end decodes the view's level and layer
// row / img: 0 means tightly packed
static void texel_strides(const tex_t *o, const NVNcopyRegion *r, uint64_t *row, uint64_t *img, const char *what) {
    int bw, bh, bpb = nvn_fmt_bpb(o->d.format, &bw, &bh);
    if (!bpb) rt_fatal("nvn: %s: unknown format %#x", what, o->d.format);
    uint64_t wb = (uint64_t)(((uint32_t)r->width + bw - 1) / bw) * bpb, hb = ((uint32_t)r->height + bh - 1) / bh;
    if (!*row) *row = wb;
    if (!*img) *img = *row * hb;
    if (*row < wb || *img < *row * hb) rt_fatal("nvn: %s: stride smaller than the region", what);
}
// void nvnTextureReadTexels[Strided](const NVNtexture *t, const NVNtextureView *v, const NVNcopyRegion *r, void *dst
// [, ptrdiff_t row, ptrdiff_t img])
static void read_texels(Cpu *c, int strided) {
    const tex_t *o = arg_p(c, 0);
    const NVNcopyRegion *r = arg_p(c, 2);
    uint64_t row = strided ? arg_x(c, 4) : 0, img = strided ? arg_x(c, 5) : 0;
    texel_strides(o, r, &row, &img, "ReadTexels");
    NVN_BLOCKING();
    gfx_backend()->read_texels(o, arg_p(c, 1), r, arg_p(c, 3), row, img);
}
// void nvnTextureWriteTexels[Strided](const NVNtexture *t, const NVNtextureView *v, const NVNcopyRegion *r, const void
// *src[, ptrdiff_t row, ptrdiff_t img])
static void write_texels(Cpu *c, int strided) {
    const tex_t *o = arg_p(c, 0);
    const view_t *v = arg_p(c, 1);
    const NVNcopyRegion *r = arg_p(c, 2);
    uint64_t row = strided ? arg_x(c, 4) : 0, img = strided ? arg_x(c, 5) : 0;
    texel_strides(o, r, &row, &img, "WriteTexels");
    if (v && v->set_format && v->format != o->d.format) rt_fatal("nvn: WriteTexels: format-reinterpreting view not supported");
    NVN_BLOCKING();
    gfx_backend()->write_texels(o, v, r, arg_p(c, 3), row, img);
}
NVN_API(nvnTextureReadTexels) { read_texels(c, 0); }
NVN_API(nvnTextureReadTexelsStrided) { read_texels(c, 1); }
NVN_API(nvnTextureWriteTexels) { write_texels(c, 0); }
NVN_API(nvnTextureWriteTexelsStrided) { write_texels(c, 1); }
// void nvnTexture{Flush,Invalidate}Texels(const NVNtexture *t, const NVNtextureView *v, const NVNcopyRegion *r): texel
// access is coherent; flush marks the storage as written for re-upload
NVN_API(nvnTextureFlushTexels) {
    const tex_t *t = arg_p(c, 0);
    if (t && t->mem && !nvn_gpu_only((uintptr_t)t->mem)) nvn_guest_written((uintptr_t)t->mem, t->L.size);
}
NVN_API(nvnTextureInvalidateTexels) { (void)c; }

// ---------- Samplers
#define SB(c) (&((samp_builder_t *)arg_p(c, 0))->p)
// void nvnSamplerBuilderSetDefaults(NVNsamplerBuilder *b): keeps the device; GL defaults
NVN_API(nvnSamplerBuilderSetDefaults) {
    samp_builder_t *b = arg_p(c, 0);
    NVNdevice *dev = b->dev;
    memset(b, 0, sizeof *b);
    b->dev = dev;
    samp_par *o = &b->p;
    o->minf = NVN_MIN_FILTER_NEAREST_MIPMAP_LINEAR;
    o->magf = NVN_MAG_FILTER_LINEAR;
    o->ws = o->wt = o->wr = NVN_WRAP_MODE_REPEAT;
    o->cmp_func = NVN_COMPARE_FUNC_LESS;
    o->lod_min = -1000.f;
    o->lod_max = 1000.f;
    o->aniso = 1.f;
}
NVN_API(nvnSamplerBuilderSetDevice) { ((samp_builder_t *)arg_p(c, 0))->dev = arg_p(c, 1); }
NVN_API(nvnSamplerBuilderSetMinMagFilter) {
    samp_par *p = SB(c);
    p->minf = (uint8_t)arg_x(c, 1);
    p->magf = (uint8_t)arg_x(c, 2);
}
NVN_API(nvnSamplerBuilderSetWrapMode) {
    samp_par *p = SB(c);
    p->ws = (uint8_t)arg_x(c, 1);
    p->wt = (uint8_t)arg_x(c, 2);
    p->wr = (uint8_t)arg_x(c, 3);
}
// void nvnSamplerBuilderSetLodClamp(NVNsamplerBuilder *b, float min, float max): the floats arrive in v0, v1
NVN_API(nvnSamplerBuilderSetLodClamp) {
    samp_par *p = SB(c);
    p->lod_min = arg_s(c, 0);
    p->lod_max = arg_s(c, 1);
}
NVN_API(nvnSamplerBuilderSetLodBias) { SB(c)->lod_bias = arg_s(c, 0); }
NVN_API(nvnSamplerBuilderSetCompare) {
    samp_par *p = SB(c);
    p->cmp_mode = (uint8_t)arg_x(c, 1);
    p->cmp_func = (uint8_t)arg_x(c, 2);
}
NVN_API(nvnSamplerBuilderSetBorderColor) { memcpy(SB(c)->border, arg_p(c, 1), 16); }
NVN_API(nvnSamplerBuilderSetMaxAnisotropy) { SB(c)->aniso = arg_s(c, 0); }
NVN_API(nvnSamplerBuilderSetLodSnap) { SB(c)->lod_snap = arg_s(c, 0); }   // stored for queries; no host equivalent
// void nvnSamplerBuilderGetMinMagFilter(const NVNsamplerBuilder *b, NVNminFilter *min, NVNmagFilter *mag)
NVN_API(nvnSamplerBuilderGetMinMagFilter) {
    const samp_par *p = SB(c);
    *(uint32_t *)arg_p(c, 1) = p->minf;
    *(uint32_t *)arg_p(c, 2) = p->magf;
}
// void nvnSamplerBuilderGetWrapMode(const NVNsamplerBuilder *b, NVNwrapMode *s, NVNwrapMode *t, NVNwrapMode *r)
NVN_API(nvnSamplerBuilderGetWrapMode) {
    const samp_par *p = SB(c);
    *(uint32_t *)arg_p(c, 1) = p->ws;
    *(uint32_t *)arg_p(c, 2) = p->wt;
    *(uint32_t *)arg_p(c, 3) = p->wr;
}
NVN_API(nvnSamplerBuilderSetReductionFilter) {
    uint32_t r = (uint32_t)arg_x(c, 1);
    if (r > 2) rt_fatal("nvn: invalid sampler reduction %u", r);
    SB(c)->reduction = (uint8_t)r;
}
NVN_API(nvnSamplerBuilderGetReductionFilter) { ret_x(c, SB(c)->reduction); }
NVN_API(nvnSamplerGetReductionFilter) { ret_x(c, ((const samp_t *)arg_p(c, 0))->p.reduction); }
// NVNboolean nvnSamplerInitialize(NVNsampler *s, const NVNsamplerBuilder *b)
NVN_API(nvnSamplerInitialize) {
    samp_t *o = arg_p(c, 0);
    o->magic = NVN_SAMP_MAGIC;
    o->p = ((const samp_builder_t *)arg_p(c, 1))->p;
    ret_x(c, 1);
}
NVN_API(nvnSamplerFinalize) { ((samp_t *)arg_p(c, 0))->magic = 0; }

// ---------- Programs
// NVNboolean nvnProgramInitialize(NVNprogram *p, NVNdevice *d)
NVN_API(nvnProgramInitialize) {
    prog_t *o = arg_p(c, 0);
    memset(o, 0, sizeof *o);
    o->magic = NVN_PROG_MAGIC;
    ret_x(c, 1);
}
static void prog_release(prog_t *o) {
    if (o->serial) gfx_backend()->program_release(o);
    o->serial = 0;
}
// void nvnProgramFinalize(NVNprogram *p)
NVN_API(nvnProgramFinalize) {
    prog_t *o = arg_p(c, 0);
    prog_release(o);
    o->magic = 0;
}
static int prog_set_shaders(prog_t *o, int n, const NVNshaderData *sd) {
    if (n < 0 || n > 6 || (n && !sd)) {
        rt_log("nvn: ProgramSetShaders with %d stages at %p", n, (const void *)sd);
        return 0;
    }
    prog_release(o);
    o->nstage = (uint32_t)n;
    if (n) memcpy(o->sd, sd, (size_t)n * sizeof *sd);
    o->serial = nvn_new_serial();
    NVN_LOG("ProgramSetShaders %p n=%d data0=%#llx", (void *)o, n, n ? (unsigned long long)sd[0].data : 0ull);
    return 1;
}
// NVNboolean nvnProgramSetShaders(NVNprogram *p, int n, const NVNshaderData *sd)
NVN_API(nvnProgramSetShaders) {
    NVN_BLOCKING();
    ret_x(c, prog_set_shaders(arg_p(c, 0), (int)arg_x(c, 1), arg_p(c, 2)));
}
// NVNboolean nvnProgramSetShadersExt(NVNprogram *p, int n, const NVNshaderDataExt *sd): no prototype in nvn.h; the
// entries are {data, control, reserved} (0x18 bytes, reserved observed 0)
typedef struct { NVNbufferAddress data; const void *control; uint64_t reserved; } shader_data_ext;
_Static_assert(sizeof(shader_data_ext) == 0x18, "NVNshaderDataExt stride");
NVN_API(nvnProgramSetShadersExt) {
    NVN_BLOCKING();
    prog_t *o = arg_p(c, 0);
    int n = (int)arg_x(c, 1);
    const shader_data_ext *sd = arg_p(c, 2);
    if (n < 0 || n > 6 || (n && !sd)) {
        rt_log("nvn: ProgramSetShadersExt with %d stages at %p", n, (const void *)sd);
        ret_x(c, 0);
        return;
    }
    NVNshaderData plain[6];
    for (int i = 0; i < n; i++) {
        if (sd[i].reserved)
            rt_fatal("nvn: ProgramSetShadersExt: unknown non-zero field %#llx in stage %d", (unsigned long long)sd[i].reserved, i);
        plain[i] = (NVNshaderData){sd[i].data, sd[i].control};
    }
    ret_x(c, prog_set_shaders(o, n, plain));
}

// ---------- State objects: packed in place (nvn_obj.h)
#define ST(T, c) ((T *)arg_p(c, 0))
NVN_API(nvnBlendStateSetDefaults) {
    *ST(blend_st, c) = (blend_st){0, NVN_BLEND_FUNC_ONE, NVN_BLEND_FUNC_ZERO, NVN_BLEND_FUNC_ONE, NVN_BLEND_FUNC_ZERO,
                                  NVN_BLEND_EQUATION_ADD, NVN_BLEND_EQUATION_ADD};
}
NVN_API(nvnBlendStateSetBlendTarget) { ST(blend_st, c)->target = (uint32_t)arg_x(c, 1); }
NVN_API(nvnBlendStateSetBlendFunc) {
    blend_st *o = ST(blend_st, c);
    o->src_rgb = (uint32_t)arg_x(c, 1);
    o->dst_rgb = (uint32_t)arg_x(c, 2);
    o->src_a = (uint32_t)arg_x(c, 3);
    o->dst_a = (uint32_t)arg_x(c, 4);
}
NVN_API(nvnBlendStateSetBlendEquation) {
    blend_st *o = ST(blend_st, c);
    o->eq_rgb = (uint32_t)arg_x(c, 1);
    o->eq_a = (uint32_t)arg_x(c, 2);
}
NVN_API(nvnColorStateSetDefaults) { *ST(color_st, c) = (color_st){0, NVN_LOGIC_OP_COPY, NVN_ALPHA_FUNC_ALWAYS}; }
// void nvnColorStateSetBlendEnable(NVNcolorState *s, int index, NVNboolean enable)
NVN_API(nvnColorStateSetBlendEnable) {
    color_st *o = ST(color_st, c);
    int i = (int)arg_x(c, 1);
    if (i < 0 || i > 7) return;
    o->blend_en = nvn_arg_bool(c, 2) ? (o->blend_en | 1u << i) : (o->blend_en & ~(1u << i));
}
NVN_API(nvnChannelMaskStateSetDefaults) { ST(chmask_st, c)->mask = 0xffffffffu; }
// void nvnChannelMaskStateSetChannelMask(NVNchannelMaskState *s, int index, NVNboolean r, g, b, a)
NVN_API(nvnChannelMaskStateSetChannelMask) {
    int i = (int)arg_x(c, 1);
    if (i < 0 || i > 7) return;
    uint32_t m = (nvn_arg_bool(c, 2) ? 1u : 0) | (nvn_arg_bool(c, 3) ? 2u : 0) | (nvn_arg_bool(c, 4) ? 4u : 0) |
                 (nvn_arg_bool(c, 5) ? 8u : 0);
    chmask_st *o = ST(chmask_st, c);
    o->mask = (o->mask & ~(0xfu << (4 * i))) | m << (4 * i);
}
NVN_API(nvnMultisampleStateSetDefaults) {
    memset(arg_p(c, 0), 0, sizeof(ms_st));
    ST(ms_st, c)->enable = 1;
}
NVN_API(nvnMultisampleStateSetMultisampleEnable) { ST(ms_st, c)->enable = nvn_arg_bool(c, 1) != 0; }
NVN_API(nvnMultisampleStateSetSamples) { ST(ms_st, c)->samples = (uint8_t)arg_x(c, 1); }
NVN_API(nvnMultisampleStateSetAlphaToCoverageEnable) { ST(ms_st, c)->a2c = nvn_arg_bool(c, 1) != 0; }
NVN_API(nvnMultisampleStateSetSampleLocationsEnable) { ST(ms_st, c)->sl_en = nvn_arg_bool(c, 1) != 0; }
// void nvnMultisampleStateSetSampleLocations(NVNmultisampleState *s, int start, int n, const float *v): programmable
// sample positions are not replayed
NVN_API(nvnMultisampleStateSetSampleLocations) { (void)c; }
NVN_API(nvnPolygonStateSetDefaults) {
    *ST(poly_st, c) = (poly_st){NVN_FACE_NONE, NVN_FRONT_FACE_CCW, NVN_POLYGON_MODE_FILL, 0};
}
NVN_API(nvnPolygonStateSetCullFace) { ST(poly_st, c)->cull = (uint32_t)arg_x(c, 1); }
NVN_API(nvnPolygonStateSetFrontFace) { ST(poly_st, c)->front = (uint32_t)arg_x(c, 1); }
NVN_API(nvnPolygonStateSetPolygonMode) { ST(poly_st, c)->mode = (uint32_t)arg_x(c, 1); }
NVN_API(nvnPolygonStateSetPolygonOffsetEnables) { ST(poly_st, c)->offset_en = (uint32_t)arg_x(c, 1); }
NVN_API(nvnDepthStencilStateSetDefaults) {
    *ST(ds_st, c) = (ds_st){0, 1, NVN_DEPTH_FUNC_LESS, 0, NVN_STENCIL_FUNC_ALWAYS, NVN_STENCIL_FUNC_ALWAYS,
                            NVN_STENCIL_OP_KEEP, NVN_STENCIL_OP_KEEP, NVN_STENCIL_OP_KEEP,
                            NVN_STENCIL_OP_KEEP, NVN_STENCIL_OP_KEEP, NVN_STENCIL_OP_KEEP};
}
NVN_API(nvnDepthStencilStateSetDepthTestEnable) { ST(ds_st, c)->dt = nvn_arg_bool(c, 1) != 0; }
NVN_API(nvnDepthStencilStateSetDepthWriteEnable) { ST(ds_st, c)->dw = nvn_arg_bool(c, 1) != 0; }
NVN_API(nvnDepthStencilStateSetDepthFunc) { ST(ds_st, c)->df = (uint32_t)arg_x(c, 1); }
NVN_API(nvnDepthStencilStateSetStencilTestEnable) { ST(ds_st, c)->st = nvn_arg_bool(c, 1) != 0; }
// void nvnDepthStencilStateSetStencilFunc(NVNdepthStencilState *s, NVNface face, NVNstencilFunc func)
NVN_API(nvnDepthStencilStateSetStencilFunc) {
    ds_st *o = ST(ds_st, c);
    uint32_t face = (uint32_t)arg_x(c, 1), fn = (uint32_t)arg_x(c, 2);
    if (face & NVN_FACE_FRONT) o->sf_f = fn;
    if (face & NVN_FACE_BACK) o->sf_b = fn;
}
// void nvnDepthStencilStateSetStencilOp(NVNdepthStencilState *s, NVNface face, NVNstencilOp fail, zfail, zpass)
NVN_API(nvnDepthStencilStateSetStencilOp) {
    ds_st *o = ST(ds_st, c);
    uint32_t face = (uint32_t)arg_x(c, 1), fail = (uint32_t)arg_x(c, 2), zfail = (uint32_t)arg_x(c, 3),
             zpass = (uint32_t)arg_x(c, 4);
    if (face & NVN_FACE_FRONT) { o->fail_f = fail; o->zfail_f = zfail; o->zpass_f = zpass; }
    if (face & NVN_FACE_BACK) { o->fail_b = fail; o->zfail_b = zfail; o->zpass_b = zpass; }
}
NVN_API(nvnVertexAttribStateSetDefaults) { *ST(va_st, c) = (va_st){0, 0, 0}; }
// void nvnVertexAttribStateSetFormat(NVNvertexAttribState *s, NVNformat format, ptrdiff_t offset)
NVN_API(nvnVertexAttribStateSetFormat) {
    int64_t off = (int64_t)arg_x(c, 2);
    if (off < 0 || off >= (1 << 19)) rt_fatal("nvn: vertex attribute offset %lld outside the packed range", (long long)off);
    va_st *o = ST(va_st, c);
    o->fmt = (uint32_t)arg_x(c, 1);
    o->offset = (uint32_t)off;
}
NVN_API(nvnVertexAttribStateSetStreamIndex) { ST(va_st, c)->stream = (uint32_t)arg_x(c, 1); }
NVN_API(nvnVertexStreamStateSetDefaults) { *ST(vs_st, c) = (vs_st){0, 0}; }
NVN_API(nvnVertexStreamStateSetStride) { ST(vs_st, c)->stride = (uint32_t)arg_x(c, 1); }
NVN_API(nvnVertexStreamStateSetDivisor) { ST(vs_st, c)->divisor = (uint32_t)arg_x(c, 1); }

// ---------- Windows
#define WB(c) ((win_builder_t *)arg_p(c, 0))
#define WN(c) ((win_t *)arg_p(c, 0))
NVN_API(nvnWindowBuilderSetDefaults) {
    win_builder_t *b = WB(c);
    NVNdevice *dev = b->dev;
    memset(b, 0, sizeof *b);
    b->dev = dev;
    b->interval = 1;
}
NVN_API(nvnWindowBuilderSetDevice) { WB(c)->dev = arg_p(c, 1); }
NVN_API(nvnWindowBuilderSetNativeWindow) { WB(c)->native = arg_p(c, 1); }
// void nvnWindowBuilderSetTextures(NVNwindowBuilder *b, int n, NVNtexture *const *textures)
NVN_API(nvnWindowBuilderSetTextures) {
    win_builder_t *b = WB(c);
    int n = (int)arg_x(c, 1);
    NVNtexture *const *t = arg_p(c, 2);
    if (n < 0 || n > 4 || (n && !t)) rt_fatal("nvn: WindowBuilderSetTextures with %d textures at %p", n, (const void *)t);
    b->ntex = n;
    for (int i = 0; i < n; i++) b->tex[i] = t[i];
}
// TOTK sets two active images; the limit bounds what WindowAcquireTexture hands out. No prototype in nvn.h.
NVN_API(nvnWindowBuilderSetNumActiveTextures) {
    int n = (int)arg_x(c, 1);
    if (n < 2 || n > 4) rt_fatal("nvn: unsupported active texture count %d", n);
    WB(c)->active = n;
}
NVN_API(nvnWindowBuilderGetNumActiveTextures) {
    const win_builder_t *b = WB(c);
    ret_x(c, (uint32_t)(b->active ? b->active : b->ntex));
}
NVN_API(nvnWindowSetNumActiveTextures) {
    win_t *w = WN(c);
    int n = (int)arg_x(c, 1);
    if (n < 2 || n > w->ntex) rt_fatal("nvn: invalid active window texture count %d", n);
    w->active = n;
}
NVN_API(nvnWindowGetNumActiveTextures) {
    const win_t *w = WN(c);
    ret_x(c, (uint32_t)(w->active ? w->active : w->ntex));
}
NVN_API(nvnWindowBuilderSetPresentInterval) { WB(c)->interval = (int)arg_x(c, 1); }
// NVNboolean nvnWindowInitialize(NVNwindow *w, const NVNwindowBuilder *b)
NVN_API(nvnWindowInitialize) {
    win_t *o = arg_p(c, 0);
    const win_builder_t *s = arg_p(c, 1);
    memset(o, 0, sizeof *o);
    o->magic = NVN_WIN_MAGIC;
    o->native = s->native;
    o->ntex = s->ntex;
    o->interval = s->interval;
    o->active = s->active ? s->active : s->ntex;
    if (o->ntex < 2 || o->ntex > 4 || o->active < 2 || o->active > o->ntex) {
        rt_log("nvn: WindowInitialize with %d textures, %d active", o->ntex, o->active);
        ret_x(c, 0);
        return;
    }
    memcpy(o->tex, s->tex, sizeof o->tex);
    const tex_t *t0 = (const tex_t *)o->tex[0];
    NVN_LOG("WindowInitialize %d textures, interval %d, first %ux%u format %#x", o->ntex, o->interval,
            t0 ? t0->d.width : 0, t0 ? t0->d.height : 0, t0 ? t0->d.format : 0);
    ret_x(c, 1);
}
// void nvnWindowFinalize(NVNwindow *w): nothing to release (MNG lists it in g_nops)
NVN_API(nvnWindowFinalize) { (void)c; }
NVN_API(nvnWindowSetPresentInterval) {
    win_t *o = WN(c);
    int v = (int)arg_x(c, 1);
    if (o->interval != v) NVN_LOG("present interval %d -> %d", o->interval, v);
    o->interval = v;
}
// void nvnWindowSetCrop(NVNwindow *w, int x, int y, int w, int h)
NVN_API(nvnWindowSetCrop) {
    win_t *o = WN(c);
    int v[4] = {(int)arg_x(c, 1), (int)arg_x(c, 2), (int)arg_x(c, 3), (int)arg_x(c, 4)};
    if (memcmp(o->crop, v, sizeof v)) NVN_LOG("WindowCrop %d,%d %dx%d", v[0], v[1], v[2], v[3]);
    memcpy(o->crop, v, sizeof v);
}

// ---------- Present pacing
// The vsync thread ticks at a fixed 60 Hz whatever the monitor does, so one game frame lasts `interval` ticks. With
// the real-time frame step (frame_rate.c) game speed does not depend on the interval, so a frame is shown when it is
// ready, at most at the frame limit (TOTK_FPS_LIMIT: 30 to 240 per second or the monitor's rate): holding it to the
// 2nd or 3rd tick would only add latency and 50 ms frames.
#define VSYNC_HZ 60ull
static uint64_t present_period(const win_t *o) {
    if (frame_step_realtime()) return frame_step_cap_ns();
    return (o->interval > 0 ? (uint64_t)o->interval : 1ull) * 1000000000ull / VSYNC_HZ;
}

// MacNativeGames nvn_pace.h: the present slot is the previous slot plus one period. Early: wait for the slot. Late by
// less than a period: present at once and keep the phase. Behind by a period or more: restart from now, no catch-up
// burst. Returns the ns to wait.
static uint64_t pace(uint64_t *target, uint64_t now, uint64_t period) {
    if (!period || !*target) { *target = now; return 0; }
    uint64_t next = *target + period;
    if (now <= next) { *target = next; return next - now; }
    *target = now - next < period ? next : now;
    return 0;
}

static void sleep_until(uint64_t deadline) {
    for (;;) {
        uint64_t now = kern_now_ns();
        if (now >= deadline) return;
        uint64_t d = deadline - now;
        if (d >= 2000000) plat_sleep_ms((uint32_t)(d / 1000000 - 1));
        else plat_yield();
    }
}

// NVNwindowAcquireTextureResult nvnWindowAcquireTexture(NVNwindow *w, NVNsync *s, int *index): blocks here, not in the
// present, until the image is free (the game often presents with a lock held). Without the frame queue that is the
// image's previous present plus one period, when the next frame goes to the display, so a frame gets exactly one period
// from acquire to present. Its present is a copy already on the queue by its display time, so with the frame queue
// (TOTK_FRAME_QUEUE=1) the image is free from that time on: the game draws the next frame while the previous one waits
// for its turn, and a frame that takes up to one more period still makes its slot, for up to one frame more latency.
// Town at 4K with DLAA at 60 on an RTX 5070: frames over 17.5 ms 1-8 per 30 s with the queue against 25-33 without.
NVN_API(nvnWindowAcquireTexture) {
    win_t *o = arg_p(c, 0);
    sync_t *s = arg_p(c, 1);
    int *out = arg_p(c, 2);
    NVN_BLOCKING();
    int idx = o->active ? o->next++ % o->active : 0;
    *out = idx;
    uint64_t at = idx < 4 ? o->avail[idx] : 0;
    if (at) {
        uint64_t p0 = perf_begin();
        sleep_until(at);
        perf_end(PERF_PACE, p0);
    }
    // The sync means "the display no longer reads this image". Presents are copies queued in submit order, so it is
    // already satisfied: fenced = 3 with fence 0.
    if (s) {
        s->fenced = 3;
        s->fence = 0;
    }
    ret_x(c, NVN_WINDOW_ACQUIRE_TEXTURE_RESULT_SUCCESS);
}

// void nvnQueuePresentTexture(NVNqueue *q, NVNwindow *w, int index): queues the copy and returns at once. The display
// time comes from the pacer: the back end's present thread holds the host present until then (back-edge pacing, as the
// Switch's display flips at vblank), so the present cadence does not follow the replay time; one period later (at that
// time with the frame queue) the image is free for the next acquire. Added latency: the slack of the frame (period minus
// its work), as on the Switch, plus up to a period with the frame queue.
NVN_API(nvnQueuePresentTexture) {
    win_t *o = arg_p(c, 1);
    int idx = (int)arg_x(c, 2);
    movie_tick(c);   // ov_movie.c: hand over decoded movie frames that are due; needs the core, so before the wait
    res_present(c);  // resolution.c: a DLSS mode change sets the game's render size again, between frames
    NVN_BLOCKING();
    if (o->magic != NVN_WIN_MAGIC || idx < 0 || idx >= o->ntex || idx >= 4 || !o->tex[idx])
        rt_fatal("nvn: QueuePresentTexture of image %d on window %p", idx, (void *)o);
    o->presents++;
    const tex_t *t = (const tex_t *)o->tex[idx];
    uint64_t period = present_period(o), now = kern_now_ns();
    uint64_t slot = now + pace(&o->pace, now, period);
    nvn_sync_ahead_present();
    gfx_backend()->present(t, o->crop, slot);
    o->ticket[idx] = o->presents;
    o->avail[idx] = cfg_bool("TOTK_FRAME_QUEUE") ? slot : slot + period;
    frame_step_present(slot, o->interval);   // frame_rate.c: the real-time step follows the display times
    rt_frame_presented(t, o->presents);
}