/* Hash throughput bench — the Phase 2 gate (SHA-256 >= 1.5 GB/s/core). */
#include "naut/hash.h"
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <time.h>

static double now(void) {
    struct timespec t; clock_gettime(CLOCK_MONOTONIC, &t);
    return t.tv_sec + t.tv_nsec * 1e-9;
}

static double bench(const char *name, void (*h)(const void *, size_t, uint8_t *),
                    uint8_t *buf, size_t len, int iters, int outlen) {
    uint8_t out[32];
    /* warm */
    h(buf, len, out);
    double t0 = now();
    for (int i = 0; i < iters; i++) h(buf, len, out);
    double dt = now() - t0;
    double gb = (double)len * iters / 1e9;
    printf("  %-10s %6.2f GB/s   (%d-byte digest)\n", name, gb / dt, outlen);
    return gb / dt;
}

static void s1(const void *d, size_t n, uint8_t *o) { naut_sha1(d, n, o); }
static void s256(const void *d, size_t n, uint8_t *o) { naut_sha256(d, n, o); }

int main(int argc, char **argv) {
    size_t len = (argc > 1) ? (size_t)atoll(argv[1]) * 1024 * 1024 : 256u * 1024 * 1024;
    int iters = (argc > 2) ? atoi(argv[2]) : 8;

    uint8_t *buf = malloc(len);
    if (!buf) { perror("malloc"); return 1; }
    memset(buf, 0xa5, len);

    printf("buffer %zu MiB x %d iters | sha256 backend: %s\n",
           len >> 20, iters, naut_sha256_backend());
    double s256_gbs = bench("sha256", s256, buf, len, iters, 32);
    bench("sha1", s1, buf, len, iters, 20);

    free(buf);
    /* gate: a single core must clear the 1.25 GB/s line rate with margin */
    if (s256_gbs < 1.5) {
        fprintf(stderr, "GATE FAIL: sha256 %.2f GB/s < 1.5 GB/s\n", s256_gbs);
        return 1;
    }
    printf("GATE PASS: sha256 %.2f GB/s >= 1.5 GB/s\n", s256_gbs);
    return 0;
}