wip: spv search + bench abort

tests/bench_h264_primitives.c

@@ -44,12 +44,28 @@ static double now_ms(void) {
 /* Per-1080p-frame counts (8160 MBs at 1920x1088). */
 #define MBS_1080P  8160
 
-/* Standard benchmark loop.  fn() is called n times per iteration. */
-typedef void (*bench_fn)(void);
+/* Standard benchmark loop.  fn() is called n times per iteration.
+ *
+ * fn() now returns the dispatch's int rc.  A single preflight call is
+ * made before the hot loop; if rc != 0 (which on the QPU substrate
+ * almost always means "SPV not found via any search path"), bench_ns
+ * returns -1 and the caller must NOT report the kernel as measured.
+ *
+ * Without this, a missing SPV makes every dispatch fail fast at the
+ * cost of one fprintf+open call (~1-5 µs), and the loop times that
+ * cost as if it were real QPU work — producing absurdly-small ns/op
+ * numbers that look like a QPU speedup.  This is exactly what made
+ * PR #36's bench numbers a measurement artifact. */
+typedef int (*bench_fn)(void);
 
 static double bench_ns(const char *name, int iters, int warmup,
                         int ops_per_iter, bench_fn fn)
 {
+    int rc = fn();
+    if (rc != 0) {
+        printf("  %-32s    DISPATCH FAILED rc=%d — kernel skipped\n", name, rc);
+        return -1;
+    }
     for (int i = 0; i < warmup; i++) fn();
     double t0 = now_ms();
     for (int i = 0; i < iters; i++) fn();
@@ -76,8 +92,8 @@ static int16_t              idct4_coeffs[1024 * 16];
 static daedalus_h264_block_meta idct4_meta[1024];
 static uint8_t              idct_dst[64 * 4 * 16 * 16];   /* 64 MB-rows × ... */
 
-static void bench_idct4(void) {
-    daedalus_dispatch_h264_idct4(ctx, g_sub,
+static int bench_idct4(void) {
+    return daedalus_dispatch_h264_idct4(ctx, g_sub,
                                   idct_dst, 64*16, idct4_coeffs, 1024, idct4_meta);
 }
 
@@ -85,8 +101,8 @@ static void bench_idct4(void) {
 static int16_t              idct8_coeffs[256 * 64];
 static daedalus_h264_block_meta idct8_meta[256];
 
-static void bench_idct8(void) {
-    daedalus_dispatch_h264_idct8(ctx, g_sub,
+static int bench_idct8(void) {
+    return daedalus_dispatch_h264_idct8(ctx, g_sub,
                                   idct_dst, 64*16, idct8_coeffs, 256, idct8_meta);
 }
 
@@ -94,13 +110,13 @@ static void bench_idct8(void) {
 static daedalus_h264_deblock_meta deblock_meta[256];
 static uint8_t deblock_dst[256 * 16 * 16];
 
-static void bench_deblock_v(void) {
-    daedalus_dispatch_h264_deblock_luma_v(ctx, g_sub,
+static int bench_deblock_v(void) {
+    return daedalus_dispatch_h264_deblock_luma_v(ctx, g_sub,
                                            deblock_dst, 16, 256, deblock_meta);
 }
 
-static void bench_deblock_h(void) {
-    daedalus_dispatch_h264_deblock_luma_h(ctx, g_sub,
+static int bench_deblock_h(void) {
+    return daedalus_dispatch_h264_deblock_luma_h(ctx, g_sub,
                                            deblock_dst, 16, 256, deblock_meta);
 }
 
@@ -109,16 +125,16 @@ static uint8_t qpel_src[256 * 16 * 16];
 static uint8_t qpel_dst[256 * 16 * 16];
 static daedalus_h264_qpel_meta qpel_meta[256];
 
-static void bench_qpel_mc20(void) {
-    daedalus_dispatch_h264_qpel_mc20(ctx, g_sub,
+static int bench_qpel_mc20(void) {
+    return daedalus_dispatch_h264_qpel_mc20(ctx, g_sub,
                                       qpel_dst, qpel_src, 16, 256, qpel_meta);
 }
-static void bench_qpel_mc02(void) {
-    daedalus_dispatch_h264_qpel_mc02(ctx, g_sub,
+static int bench_qpel_mc02(void) {
+    return daedalus_dispatch_h264_qpel_mc02(ctx, g_sub,
                                       qpel_dst, qpel_src, 16, 256, qpel_meta);
 }
-static void bench_qpel_mc22(void) {
-    daedalus_dispatch_h264_qpel_mc22(ctx, g_sub,
+static int bench_qpel_mc22(void) {
+    return daedalus_dispatch_h264_qpel_mc22(ctx, g_sub,
                                       qpel_dst, qpel_src, 16, 256, qpel_meta);
 }
 
@@ -197,11 +213,25 @@ int main(int argc, char **argv)
         rows[i].cpu_ns = bench_ns(rows[i].name, iters, warmup, rows[i].n_per_call, rows[i].fn);
 
     /* Pass 2: QPU compute (if available). */
+    int qpu_failures = 0;
     if (has_qpu) {
         g_sub = DAEDALUS_SUBSTRATE_QPU;
         printf("\n== QPU V3D7 compute ==\n");
-        for (int i = 0; i < N_ROWS; i++)
+        for (int i = 0; i < N_ROWS; i++) {
             rows[i].qpu_ns = bench_ns(rows[i].name, iters, warmup, rows[i].n_per_call, rows[i].fn);
+            if (rows[i].qpu_ns < 0) qpu_failures++;
+        }
+        if (qpu_failures) {
+            fprintf(stderr,
+                "\nbench_h264_primitives: %d of %d QPU dispatches failed.\n"
+                "  Almost always means SPV files were not found via any of:\n"
+                "    cwd  /  $DAEDALUS_SHADER_DIR  /  binary-dir  /\n"
+                "    /opt/fourier/share/daedalus-fourier  /  /usr/share/daedalus-fourier\n"
+                "  Set DAEDALUS_SHADER_DIR=<path> or run from a dir where the\n"
+                "  .spv files exist (e.g. the cmake build dir).\n",
+                qpu_failures, N_ROWS);
+            return 2;
+        }
     }
 
     /* Summary table — both substrates side by side. */