checkasm: Improve cdef_filter test

1 files changed, 42 insertions, 38 deletions

diff --git a/tests/checkasm/cdef.c b/tests/checkasm/cdef.c
index b88a0a4..b177de5 100644
--- a/tests/checkasm/cdef.c
+++ b/tests/checkasm/cdef.c
@@ -35,68 +35,71 @@
 #include "src/levels.h"
 #include "src/cdef.h"
 
+static int to_binary(int x) { /* 0-15 -> 0000-1111 */
+    return (x & 1) + 5 * (x & 2) + 25 * (x & 4) + 125 * (x & 8);
+}
+
 static void init_tmp(pixel *buf, int n, const int bitdepth_max) {
     while (n--)
         *buf++ = rnd() & bitdepth_max;
 }
 
-static void check_cdef_filter(const cdef_fn fn, const int w, const int h,
-                              const char *const name)
-{
-    ALIGN_STK_32(pixel, src, 10 * 16 + 8, );
-    ALIGN_STK_32(pixel, c_src, 10 * 16 + 8, ), *const c_src_ptr = c_src + 8;
-    ALIGN_STK_32(pixel, a_src, 10 * 16 + 8, ), *const a_src_ptr = a_src + 8;
-    ALIGN_STK_32(pixel, top, 16 * 2 + 8, ), *const top_ptr = top + 8;
+static void check_cdef_filter(const cdef_fn fn, const int w, const int h) {
+    ALIGN_STK_32(pixel, c_src, 10 * 16 + 8, ), *const c_dst = c_src + 8;
+    ALIGN_STK_32(pixel, a_src, 10 * 16 + 8, ), *const a_dst = a_src + 8;
+    ALIGN_STK_32(pixel, top, 16 * 2 + 8, );
     pixel left[8][2];
+    pixel *const top_ptrs[2] = { top + 8, top + 24 };
+    const ptrdiff_t stride = 16 * sizeof(pixel);
 
     declare_func(void, pixel *dst, ptrdiff_t dst_stride, const pixel (*left)[2],
                  pixel *const top[2], int pri_strength, int sec_strength,
                  int dir, int damping, enum CdefEdgeFlags edges HIGHBD_DECL_SUFFIX);
 
-    if (check_func(fn, "%s_%dbpc", name, BITDEPTH)) {
+    if (check_func(fn, "cdef_filter_%dx%d_%dbpc", w, h, BITDEPTH)) {
         for (int dir = 0; dir < 8; dir++) {
-            for (enum CdefEdgeFlags edges = 0; edges <= 0xf; edges++) {
+            for (enum CdefEdgeFlags edges = 0x0; edges <= 0xf; edges++) {
 #if BITDEPTH == 16
                 const int bitdepth_max = rnd() & 1 ? 0x3ff : 0xfff;
 #else
                 const int bitdepth_max = 0xff;
 #endif
                 const int bitdepth_min_8 = bitdepth_from_max(bitdepth_max) - 8;
-                init_tmp(src, 10 * 16 + 8, bitdepth_max);
-                init_tmp(top, 16 * 2 + 8, bitdepth_max);
-                init_tmp((pixel *) left,8 * 2, bitdepth_max);
 
-                memcpy(a_src, src, (10 * 16 + 8) * sizeof(pixel));
-                memcpy(c_src, src, (10 * 16 + 8) * sizeof(pixel));
+                init_tmp(c_src, 10 * 16 + 8, bitdepth_max);
+                init_tmp(top, 16 * 2 + 8, bitdepth_max);
+                init_tmp((pixel *) left, 8 * 2, bitdepth_max);
+                memcpy(a_src, c_src, (10 * 16 + 8) * sizeof(pixel));
 
                 const int lvl = 1 + (rnd() % 62);
                 const int damping = 3 + (rnd() & 3) + bitdepth_min_8 - (w == 4 || (rnd() & 1));
-                const int pri_strength = (lvl >> 2) << bitdepth_min_8;
+                int pri_strength = (lvl >> 2) << bitdepth_min_8;
                 int sec_strength = lvl & 3;
                 sec_strength += sec_strength == 3;
                 sec_strength <<= bitdepth_min_8;
-                call_ref(c_src_ptr, 16 * sizeof(pixel), left,
-                         (pixel *[2]) { top_ptr, top_ptr + 16 },
-                         pri_strength, sec_strength, dir, damping, edges
-                         HIGHBD_TAIL_SUFFIX);
-                call_new(a_src_ptr, 16 * sizeof(pixel), left,
-                         (pixel *[2]) { top_ptr, top_ptr + 16 },
-                         pri_strength, sec_strength, dir, damping, edges
-                         HIGHBD_TAIL_SUFFIX);
-                checkasm_check_pixel(c_src, 16 * sizeof(pixel),
-                                     a_src, 16 * sizeof(pixel),
-                                     16, 10, "src");
-                checkasm_check_pixel(c_src + 16 * 10, 16 * sizeof(pixel),
-                                     a_src + 16 * 10, 16 * sizeof(pixel),
-                                     8, 1, "src last row");
-                bench_new(a_src_ptr, 16 * sizeof(pixel), left,
-                          (pixel *[2]) { top_ptr, top_ptr + 16 },
-                          pri_strength, sec_strength, dir, damping, edges
-                          HIGHBD_TAIL_SUFFIX);
+                call_ref(c_dst, stride, left, top_ptrs, pri_strength, sec_strength,
+                         dir, damping, edges HIGHBD_TAIL_SUFFIX);
+                call_new(a_dst, stride, left, top_ptrs, pri_strength, sec_strength,
+                         dir, damping, edges HIGHBD_TAIL_SUFFIX);
+                if (checkasm_check_pixel(c_dst, stride, a_dst, stride, w, h, "dst")) {
+                    fprintf(stderr, "strength = %d:%d, dir = %d, damping = %d, edges = %04d\n",
+                            pri_strength, sec_strength, dir, damping, to_binary(edges));
+                    return;
+                }
+                if (dir == 7 && (edges == 0x5 || edges == 0xa || edges == 0xf)) {
+                    /* Benchmark a fixed set of cases to get consistent results:
+                     *  1) top/left edges and pri_strength only
+                     *  2) bottom/right edges and sec_strength only
+                     *  3) all edges and both pri_strength and sec_strength
+                     */
+                    pri_strength = (edges & 1) << bitdepth_min_8;
+                    sec_strength = edges & 2;
+                    bench_new(a_dst, stride, left, top_ptrs, pri_strength, sec_strength,
+                              dir, damping, edges HIGHBD_TAIL_SUFFIX);
+                }
             }
         }
     }
-    report(name);
 }
 
 static void check_cdef_direction(const cdef_dir_fn fn) {
@@ -129,11 +132,12 @@ static void check_cdef_direction(const cdef_dir_fn fn) {
 
 void bitfn(checkasm_check_cdef)(void) {
     Dav1dCdefDSPContext c;
-
     bitfn(dav1d_cdef_dsp_init)(&c);
 
     check_cdef_direction(c.dir);
-    check_cdef_filter(c.fb[0], 8, 8, "cdef_filter_8x8");
-    check_cdef_filter(c.fb[1], 4, 8, "cdef_filter_4x8");
-    check_cdef_filter(c.fb[2], 4, 4, "cdef_filter_4x4");
+
+    check_cdef_filter(c.fb[0], 8, 8);
+    check_cdef_filter(c.fb[1], 4, 8);
+    check_cdef_filter(c.fb[2], 4, 4);
+    report("cdef_filter");
 }