/*
 * Cycle 4 Phase 6 — QPU bench for VP9 wd=8 LPF.
 * Mirrors bench_v3d_lpf.c (cycle 2); changes: calls the wd=8 ref
 * + asserts dst_stride >= 6 (cycle 4 contract).
 */
#define _POSIX_C_SOURCE 200809L
#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
#include <string.h>
#include <stddef.h>
#include <assert.h>
#include <time.h>
#include <getopt.h>
#include <vulkan/vulkan.h>

#include "v3d_runner.h"

extern void daedalus_vp9_loop_filter_h_8_8_ref(
    uint8_t *dst, ptrdiff_t stride, int E, int I, int H);

#define EDGE_STRIDE 8
#define EDGE_BYTES 64

static uint64_t xs_state;
static inline uint64_t xs(void) {
    uint64_t x = xs_state; x ^= x<<13; x ^= x>>7; x ^= x<<17;
    return xs_state = x;
}
static void gen_edge_pixels(uint8_t *buf) {
    int a = (int)(xs() % 200) + 20;
    int b = (int)(xs() % 200) + 20;
    int n = (int)(xs() % 30);
    for (int r = 0; r < 8; r++)
        for (int c = 0; c < 8; c++) {
            int base = (c < 4) ? a : b;
            int noise = ((int)(xs() % (2*n + 1))) - n;
            int v = base + noise;
            buf[r*EDGE_STRIDE + c] = (uint8_t)(v < 0 ? 0 : v > 255 ? 255 : v);
        }
}
static void gen_thresholds(int *E, int *I, int *H) {
    *E = (int)(xs() % 81);
    *I = (int)(xs() % 41);
    *H = (int)(xs() % 11);
}
static double now_seconds(void) {
    struct timespec ts; clock_gettime(CLOCK_MONOTONIC_RAW, &ts);
    return ts.tv_sec + ts.tv_nsec * 1e-9;
}

typedef struct { uint32_t n_edges, blocks_per_row, dst_stride_u8, _pad; } push_consts;

int main(int argc, char **argv)
{
    int n_edges = 65536, iters = 100, verify_only = 0;
    uint64_t seed = 0;
    const char *spv = "v3d_lpf_h_8_8.spv";
    static struct option opts[] = {
        {"edges", required_argument, 0, 'e'},
        {"iters", required_argument, 0, 'i'},
        {"seed",  required_argument, 0, 's'},
        {"spv",   required_argument, 0, 'S'},
        {"verify-only", no_argument, 0, 'V'},
        {0,0,0,0}
    };
    for (int c; (c = getopt_long(argc, argv, "e:i:s:S:V", opts, 0)) != -1;) {
        switch (c) {
        case 'e': n_edges = atoi(optarg); break;
        case 'i': iters = atoi(optarg); break;
        case 's': seed = strtoull(optarg, 0, 0); break;
        case 'S': spv = optarg; break;
        case 'V': verify_only = 1; break;
        default: return 2;
        }
    }

    xs_state = seed ? seed : 0xa57edbeef5717ULL;

    v3d_runner *r = v3d_runner_create();
    if (!r) return 1;
    printf("=== v3d LPF h_8_8 bench ===\n");
    printf("  device: %s\n  n_edges: %d  iters: %d\n",
           v3d_runner_device_name(r), n_edges, iters);

    size_t dst_bytes  = (size_t) n_edges * EDGE_BYTES;
    size_t meta_bytes = (size_t) n_edges * 4 * sizeof(uint32_t);
    v3d_buffer buf_meta = {0}, buf_dst = {0};
    v3d_runner_create_buffer(r, meta_bytes, &buf_meta);
    v3d_runner_create_buffer(r, dst_bytes,  &buf_dst);

    uint8_t *master = malloc(dst_bytes);
    uint8_t *expected = malloc(dst_bytes);
    int *Es = malloc(n_edges*sizeof(int)), *Is = malloc(n_edges*sizeof(int)), *Hs = malloc(n_edges*sizeof(int));
    for (int i = 0; i < n_edges; i++) {
        gen_edge_pixels(master + (size_t)i * EDGE_BYTES);
        gen_thresholds(&Es[i], &Is[i], &Hs[i]);
    }
    memcpy(expected, master, dst_bytes);
    for (int i = 0; i < n_edges; i++)
        daedalus_vp9_loop_filter_h_8_8_ref(expected + (size_t)i * EDGE_BYTES + 4,
                                           EDGE_STRIDE, Es[i], Is[i], Hs[i]);

    uint32_t dst_stride = EDGE_STRIDE;
    assert(dst_stride >= 6 && "cycle 4 §4 contract: dst_stride_u8 >= 6 (flat8in 6-write)");
    uint32_t *meta = buf_meta.mapped;
    for (int i = 0; i < n_edges; i++) {
        uint32_t mx = (uint32_t)((size_t)i * EDGE_BYTES + 4);
        assert(mx >= 4);
        meta[4*i + 0] = mx;
        meta[4*i + 1] = (uint32_t) Es[i];
        meta[4*i + 2] = (uint32_t) Is[i];
        meta[4*i + 3] = (uint32_t) Hs[i];
    }
    memcpy(buf_dst.mapped, master, dst_bytes);

    v3d_pipeline pipe = {0};
    if (v3d_runner_create_pipeline(r, spv, 2, sizeof(push_consts), &pipe)) return 1;
    v3d_buffer bufs[2] = { buf_meta, buf_dst };
    v3d_runner_bind_buffers(r, &pipe, bufs, 2);

    const uint32_t edges_per_wg = 32;
    uint32_t gc = (uint32_t)((n_edges + edges_per_wg - 1) / edges_per_wg);
    push_consts pc = { .n_edges = (uint32_t) n_edges, .dst_stride_u8 = dst_stride };

    VkCommandBuffer cb = v3d_runner_alloc_cmdbuf(r);
    VkCommandBufferBeginInfo cbbi = { .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO };
    vkBeginCommandBuffer(cb, &cbbi);
    vkCmdBindPipeline(cb, VK_PIPELINE_BIND_POINT_COMPUTE, pipe.pipeline);
    vkCmdBindDescriptorSets(cb, VK_PIPELINE_BIND_POINT_COMPUTE,
                            pipe.layout, 0, 1, &pipe.desc_set, 0, NULL);
    vkCmdPushConstants(cb, pipe.layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(pc), &pc);
    vkCmdDispatch(cb, gc, 1, 1);
    vkEndCommandBuffer(cb);

    /* M1'''' */
    printf("\n=== M1'''': QPU vs C bit-exact ===\n");
    memcpy(buf_dst.mapped, master, dst_bytes);
    if (v3d_runner_submit_wait(r, cb)) return 1;
    int mis = 0, bytediffs = 0;
    for (int i = 0; i < n_edges; i++) {
        const uint8_t *q = (uint8_t *) buf_dst.mapped + (size_t)i * EDGE_BYTES;
        const uint8_t *e = expected + (size_t)i * EDGE_BYTES;
        if (memcmp(q, e, EDGE_BYTES) != 0) {
            int d = 0;
            for (int j = 0; j < EDGE_BYTES; j++) if (q[j] != e[j]) d++;
            bytediffs += d;
            if (mis < 3) fprintf(stderr, "MISMATCH edge %d (E=%d I=%d H=%d): %d/64 bytes\n",
                                 i, Es[i], Is[i], Hs[i], d);
            mis++;
        }
    }
    printf("  edges bit-exact: %d / %d (%.4f%%)\n",
           n_edges - mis, n_edges, 100.0 * (n_edges - mis) / n_edges);
    if (mis > 0) { fprintf(stderr, "REFUSING throughput on broken kernel.\n"); return 1; }
    if (verify_only) return 0;

    /* M2'''' */
    printf("\n=== M2'''': QPU throughput ===\n");
    for (int i = 0; i < 10; i++) { memcpy(buf_dst.mapped, master, dst_bytes); v3d_runner_submit_wait(r, cb); }
    double t0 = now_seconds();
    for (int i = 0; i < iters; i++) { memcpy(buf_dst.mapped, master, dst_bytes); v3d_runner_submit_wait(r, cb); }
    double t1 = now_seconds();
    double s0 = now_seconds();
    for (int i = 0; i < iters; i++) memcpy(buf_dst.mapped, master, dst_bytes);
    double s1 = now_seconds();
    double ks = (t1 - t0) - (s1 - s0);
    double total = (double) n_edges * iters;
    double mes = total / ks / 1e6;

    printf("  edges/dispatch: %d, iters: %d, total: %.0f\n", n_edges, iters, total);
    printf("  elapsed (kernel)=%.6f s\n  per-edge       = %.1f ns\n  per-dispatch   = %.1f us\n",
           ks, ks / total * 1e9, ks / iters * 1e6);
    printf("  M2'''' = %.3f Medge/s\n", mes);
    double M3 = 52.382;   /* k4 phase 3 baseline */
    double R = mes / M3;
    printf("\n  Cycle 4 NEON M3'''' = %.3f Medge/s\n", M3);
    printf("  R'''' = M2''''/M3''''  = %.3f\n", R);
    if      (R >= 1.0) printf("  decision band       = GREEN\n");
    else if (R >= 0.5) printf("  decision band       = YELLOW\n");
    else if (R >= 0.1) printf("  decision band       = ORANGE\n");
    else               printf("  decision band       = RED\n");
    double floor30 = 64530.0 * 30 / 1e6;
    printf("  30fps@1080p floor   : %.3f Medge/s — %.1fx margin\n",
           floor30, mes / floor30);

    v3d_runner_destroy_pipeline(r, &pipe);
    v3d_runner_destroy_buffer(r, &buf_dst);
    v3d_runner_destroy_buffer(r, &buf_meta);
    v3d_runner_destroy(r);
    return 0;
}