shithub: dav1d

Info  •  Files  •  Log  •  Branches

looprestoration: Add a bpc parameter to the init func

This allows using completely different codepaths for 10 and 12 bpc,
or just adding SIMD functions for either of them.

--- a/src/arm/looprestoration_init_tmpl.c

+++ b/src/arm/looprestoration_init_tmpl.c

@@ -270,7 +270,7 @@

 }

 #endif // BITDEPTH == 8

-COLD void bitfn(dav1d_loop_restoration_dsp_init_arm)(Dav1dLoopRestorationDSPContext *const c) {

+COLD void bitfn(dav1d_loop_restoration_dsp_init_arm)(Dav1dLoopRestorationDSPContext *const c, int bpc) {

     const unsigned flags = dav1d_get_cpu_flags();

     if (!(flags & DAV1D_ARM_CPU_FLAG_NEON)) return;

--- a/src/decode.c

+++ b/src/decode.c

@@ -3242,7 +3242,7 @@

             dav1d_intra_pred_dsp_init_##bd##bpc(&dsp->ipred); \

             dav1d_itx_dsp_init_##bd##bpc(&dsp->itx); \

             dav1d_loop_filter_dsp_init_##bd##bpc(&dsp->lf); \

-            dav1d_loop_restoration_dsp_init_##bd##bpc(&dsp->lr); \

+            dav1d_loop_restoration_dsp_init_##bd##bpc(&dsp->lr, bpc); \

             dav1d_mc_dsp_init_##bd##bpc(&dsp->mc); \

             dav1d_film_grain_dsp_init_##bd##bpc(&dsp->fg); \

             break

--- a/src/looprestoration.h

+++ b/src/looprestoration.h

@@ -72,8 +72,8 @@

     selfguided_fn selfguided;

 } Dav1dLoopRestorationDSPContext;

-bitfn_decls(void dav1d_loop_restoration_dsp_init, Dav1dLoopRestorationDSPContext *c);

-bitfn_decls(void dav1d_loop_restoration_dsp_init_arm, Dav1dLoopRestorationDSPContext *c);

+bitfn_decls(void dav1d_loop_restoration_dsp_init, Dav1dLoopRestorationDSPContext *c, int bpc);

+bitfn_decls(void dav1d_loop_restoration_dsp_init_arm, Dav1dLoopRestorationDSPContext *c, int bpc);

 bitfn_decls(void dav1d_loop_restoration_dsp_init_x86, Dav1dLoopRestorationDSPContext *c);

 bitfn_decls(void dav1d_loop_restoration_dsp_init_ppc, Dav1dLoopRestorationDSPContext *c);

--- a/src/looprestoration_tmpl.c

+++ b/src/looprestoration_tmpl.c

@@ -573,13 +573,13 @@

     }

 }

-COLD void bitfn(dav1d_loop_restoration_dsp_init)(Dav1dLoopRestorationDSPContext *const c) {

+COLD void bitfn(dav1d_loop_restoration_dsp_init)(Dav1dLoopRestorationDSPContext *const c, int bpc) {

     c->wiener = wiener_c;

     c->selfguided = selfguided_c;

 #if HAVE_ASM

 #if ARCH_AARCH64 || ARCH_ARM

-    bitfn(dav1d_loop_restoration_dsp_init_arm)(c);

+    bitfn(dav1d_loop_restoration_dsp_init_arm)(c, bpc);

 #elif ARCH_PPC64LE

     bitfn(dav1d_loop_restoration_dsp_init_ppc)(c);

 #elif ARCH_X86

--- a/tests/checkasm/looprestoration.c

+++ b/tests/checkasm/looprestoration.c

@@ -43,7 +43,7 @@

     }

 }

-static void check_wiener(Dav1dLoopRestorationDSPContext *const c) {

+static void check_wiener(Dav1dLoopRestorationDSPContext *const c, const int bpc) {

     ALIGN_STK_64(pixel, c_dst, 448 * 64,);

     ALIGN_STK_64(pixel, a_dst, 448 * 64,);

     ALIGN_STK_64(pixel, h_edge, 448 * 8,);

@@ -58,7 +58,7 @@

     for (int pl = 0; pl < 2; pl++) {

         if (check_func(c->wiener, "wiener_%s_%dbpc",

-                       pl ? "chroma" : "luma", BITDEPTH))

+                       pl ? "chroma" : "luma", bpc))

         {

             int16_t filter[2][3], filter_v[7], filter_h[7];

@@ -81,11 +81,7 @@

             const int base_w = 1 + (rnd() % 384);

             const int base_h = 1 + (rnd() & 63);

-#if BITDEPTH == 16

-            const int bitdepth_max = rnd() & 1 ? 0x3ff : 0xfff;

-#else

-            const int bitdepth_max = 0xff;

-#endif

+            const int bitdepth_max = (1 << bpc) - 1;

             init_tmp(c_dst, 448 * sizeof(pixel), 448, 64, bitdepth_max);

             init_tmp(h_edge, 448 * sizeof(pixel), 448, 8, bitdepth_max);

@@ -112,10 +108,9 @@

                       256, 64, filter_h, filter_v, 0xf HIGHBD_TAIL_SUFFIX);

         }

     }

-    report("wiener");

 }

-static void check_sgr(Dav1dLoopRestorationDSPContext *const c) {

+static void check_sgr(Dav1dLoopRestorationDSPContext *const c, const int bpc) {

     ALIGN_STK_64(pixel, c_dst, 448 * 64,);

     ALIGN_STK_64(pixel, a_dst, 448 * 64,);

     ALIGN_STK_64(pixel, h_edge, 448 * 8,);

@@ -130,7 +125,7 @@

     for (int sgr_idx = 14; sgr_idx >= 6; sgr_idx -= 4) {

         if (check_func(c->selfguided, "selfguided_%s_%dbpc",

-                       sgr_idx == 6 ? "mix" : sgr_idx == 10 ? "3x3" : "5x5", BITDEPTH))

+                       sgr_idx == 6 ? "mix" : sgr_idx == 10 ? "3x3" : "5x5", bpc))

         {

             int16_t sgr_wt[2];

@@ -140,11 +135,7 @@

             const int base_w = 1 + (rnd() % 384);

             const int base_h = 1 + (rnd() & 63);

-#if BITDEPTH == 16

-            const int bitdepth_max = rnd() & 1 ? 0x3ff : 0xfff;

-#else

-            const int bitdepth_max = 0xff;

-#endif

+            const int bitdepth_max = (1 << bpc) - 1;

             init_tmp(c_dst, 448 * sizeof(pixel), 448, 64, bitdepth_max);

             init_tmp(h_edge, 448 * sizeof(pixel), 448, 8, bitdepth_max);

@@ -171,14 +162,24 @@

                       256, 64, sgr_idx, sgr_wt, 0xf HIGHBD_TAIL_SUFFIX);

         }

     }

-    report("sgr");

 }

 void bitfn(checkasm_check_looprestoration)(void) {

-    Dav1dLoopRestorationDSPContext c;

-

-    bitfn(dav1d_loop_restoration_dsp_init)(&c);

-

-    check_wiener(&c);

-    check_sgr(&c);

+#if BITDEPTH == 16

+    const int bpc_min = 10, bpc_max = 12;

+#else

+    const int bpc_min = 8, bpc_max = 8;

+#endif

+    for (int bpc = bpc_min; bpc <= bpc_max; bpc += 2) {

+        Dav1dLoopRestorationDSPContext c;

+        bitfn(dav1d_loop_restoration_dsp_init)(&c, bpc);

+        check_wiener(&c, bpc);

+    }

+    report("wiener");

+    for (int bpc = bpc_min; bpc <= bpc_max; bpc += 2) {

+        Dav1dLoopRestorationDSPContext c;

+        bitfn(dav1d_loop_restoration_dsp_init)(&c, bpc);

+        check_sgr(&c, bpc);

+    }

+    report("sgr");

 }