#
# Motion search
#
-add_proto qw/int vp9_full_search_sad/, "const struct macroblock *x, const struct mv *ref_mv, int sad_per_bit, int distance, const struct vp9_variance_vtable *fn_ptr, const struct mv *center_mv, struct mv *best_mv";
-specialize qw/vp9_full_search_sad sse3 sse4_1/;
-$vp9_full_search_sad_sse3=vp9_full_search_sadx3;
-$vp9_full_search_sad_sse4_1=vp9_full_search_sadx8;
-
add_proto qw/int vp9_diamond_search_sad/, "const struct macroblock *x, const struct search_site_config *cfg, struct mv *ref_mv, struct mv *best_mv, int search_param, int sad_per_bit, int *num00, const struct vp9_variance_vtable *fn_ptr, const struct mv *center_mv";
specialize qw/vp9_diamond_search_sad avx/;
}
#if CONFIG_VP9_HIGHBITDEPTH
-#define HIGHBD_BFP(BT, SDF, SDAF, VF, SVF, SVAF, SDX3F, SDX8F, SDX4DF) \
- cpi->fn_ptr[BT].sdf = SDF; \
- cpi->fn_ptr[BT].sdaf = SDAF; \
- cpi->fn_ptr[BT].vf = VF; \
- cpi->fn_ptr[BT].svf = SVF; \
- cpi->fn_ptr[BT].svaf = SVAF; \
- cpi->fn_ptr[BT].sdx3f = SDX3F; \
- cpi->fn_ptr[BT].sdx8f = SDX8F; \
+#define HIGHBD_BFP(BT, SDF, SDAF, VF, SVF, SVAF, SDX4DF) \
+ cpi->fn_ptr[BT].sdf = SDF; \
+ cpi->fn_ptr[BT].sdaf = SDAF; \
+ cpi->fn_ptr[BT].vf = VF; \
+ cpi->fn_ptr[BT].svf = SVF; \
+ cpi->fn_ptr[BT].svaf = SVAF; \
cpi->fn_ptr[BT].sdx4df = SDX4DF;
#define MAKE_BFP_SAD_WRAPPER(fnname) \
4; \
}
-#define MAKE_BFP_SAD3_WRAPPER(fnname) \
- static void fnname##_bits8(const uint8_t *src_ptr, int source_stride, \
- const uint8_t *ref_ptr, int ref_stride, \
- unsigned int *sad_array) { \
- fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
- } \
- static void fnname##_bits10(const uint8_t *src_ptr, int source_stride, \
- const uint8_t *ref_ptr, int ref_stride, \
- unsigned int *sad_array) { \
- int i; \
- fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
- for (i = 0; i < 3; i++) sad_array[i] >>= 2; \
- } \
- static void fnname##_bits12(const uint8_t *src_ptr, int source_stride, \
- const uint8_t *ref_ptr, int ref_stride, \
- unsigned int *sad_array) { \
- int i; \
- fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
- for (i = 0; i < 3; i++) sad_array[i] >>= 4; \
- }
-
-#define MAKE_BFP_SAD8_WRAPPER(fnname) \
- static void fnname##_bits8(const uint8_t *src_ptr, int source_stride, \
- const uint8_t *ref_ptr, int ref_stride, \
- unsigned int *sad_array) { \
- fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
- } \
- static void fnname##_bits10(const uint8_t *src_ptr, int source_stride, \
- const uint8_t *ref_ptr, int ref_stride, \
- unsigned int *sad_array) { \
- int i; \
- fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
- for (i = 0; i < 8; i++) sad_array[i] >>= 2; \
- } \
- static void fnname##_bits12(const uint8_t *src_ptr, int source_stride, \
- const uint8_t *ref_ptr, int ref_stride, \
- unsigned int *sad_array) { \
- int i; \
- fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
- for (i = 0; i < 8; i++) sad_array[i] >>= 4; \
- }
#define MAKE_BFP_SAD4D_WRAPPER(fnname) \
static void fnname##_bits8(const uint8_t *src_ptr, int source_stride, \
const uint8_t *const ref_ptr[], int ref_stride, \
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad32x64x4d)
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad32x32)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad32x32_avg)
-MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad32x32x3)
-MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad32x32x8)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad32x32x4d)
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad64x64)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad64x64_avg)
-MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad64x64x3)
-MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad64x64x8)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad64x64x4d)
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad16x16)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad16x16_avg)
-MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad16x16x3)
-MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad16x16x8)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad16x16x4d)
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad16x8)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad16x8_avg)
-MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad16x8x3)
-MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad16x8x8)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad16x8x4d)
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad8x16)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad8x16_avg)
-MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad8x16x3)
-MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad8x16x8)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad8x16x4d)
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad8x8)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad8x8_avg)
-MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad8x8x3)
-MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad8x8x8)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad8x8x4d)
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad8x4)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad8x4_avg)
-MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad8x4x8)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad8x4x4d)
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad4x8)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad4x8_avg)
-MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad4x8x8)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad4x8x4d)
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad4x4)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad4x4_avg)
-MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad4x4x3)
-MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad4x4x8)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad4x4x4d)
static void highbd_set_var_fns(VP9_COMP *const cpi) {
HIGHBD_BFP(BLOCK_32X16, vpx_highbd_sad32x16_bits8,
vpx_highbd_sad32x16_avg_bits8, vpx_highbd_8_variance32x16,
vpx_highbd_8_sub_pixel_variance32x16,
- vpx_highbd_8_sub_pixel_avg_variance32x16, NULL, NULL,
+ vpx_highbd_8_sub_pixel_avg_variance32x16,
vpx_highbd_sad32x16x4d_bits8)
HIGHBD_BFP(BLOCK_16X32, vpx_highbd_sad16x32_bits8,
vpx_highbd_sad16x32_avg_bits8, vpx_highbd_8_variance16x32,
vpx_highbd_8_sub_pixel_variance16x32,
- vpx_highbd_8_sub_pixel_avg_variance16x32, NULL, NULL,
+ vpx_highbd_8_sub_pixel_avg_variance16x32,
vpx_highbd_sad16x32x4d_bits8)
HIGHBD_BFP(BLOCK_64X32, vpx_highbd_sad64x32_bits8,
vpx_highbd_sad64x32_avg_bits8, vpx_highbd_8_variance64x32,
vpx_highbd_8_sub_pixel_variance64x32,
- vpx_highbd_8_sub_pixel_avg_variance64x32, NULL, NULL,
+ vpx_highbd_8_sub_pixel_avg_variance64x32,
vpx_highbd_sad64x32x4d_bits8)
HIGHBD_BFP(BLOCK_32X64, vpx_highbd_sad32x64_bits8,
vpx_highbd_sad32x64_avg_bits8, vpx_highbd_8_variance32x64,
vpx_highbd_8_sub_pixel_variance32x64,
- vpx_highbd_8_sub_pixel_avg_variance32x64, NULL, NULL,
+ vpx_highbd_8_sub_pixel_avg_variance32x64,
vpx_highbd_sad32x64x4d_bits8)
HIGHBD_BFP(BLOCK_32X32, vpx_highbd_sad32x32_bits8,
vpx_highbd_sad32x32_avg_bits8, vpx_highbd_8_variance32x32,
vpx_highbd_8_sub_pixel_variance32x32,
vpx_highbd_8_sub_pixel_avg_variance32x32,
- vpx_highbd_sad32x32x3_bits8, vpx_highbd_sad32x32x8_bits8,
vpx_highbd_sad32x32x4d_bits8)
HIGHBD_BFP(BLOCK_64X64, vpx_highbd_sad64x64_bits8,
vpx_highbd_sad64x64_avg_bits8, vpx_highbd_8_variance64x64,
vpx_highbd_8_sub_pixel_variance64x64,
vpx_highbd_8_sub_pixel_avg_variance64x64,
- vpx_highbd_sad64x64x3_bits8, vpx_highbd_sad64x64x8_bits8,
vpx_highbd_sad64x64x4d_bits8)
HIGHBD_BFP(BLOCK_16X16, vpx_highbd_sad16x16_bits8,
vpx_highbd_sad16x16_avg_bits8, vpx_highbd_8_variance16x16,
vpx_highbd_8_sub_pixel_variance16x16,
vpx_highbd_8_sub_pixel_avg_variance16x16,
- vpx_highbd_sad16x16x3_bits8, vpx_highbd_sad16x16x8_bits8,
vpx_highbd_sad16x16x4d_bits8)
- HIGHBD_BFP(
- BLOCK_16X8, vpx_highbd_sad16x8_bits8, vpx_highbd_sad16x8_avg_bits8,
- vpx_highbd_8_variance16x8, vpx_highbd_8_sub_pixel_variance16x8,
- vpx_highbd_8_sub_pixel_avg_variance16x8, vpx_highbd_sad16x8x3_bits8,
- vpx_highbd_sad16x8x8_bits8, vpx_highbd_sad16x8x4d_bits8)
+ HIGHBD_BFP(BLOCK_16X8, vpx_highbd_sad16x8_bits8,
+ vpx_highbd_sad16x8_avg_bits8, vpx_highbd_8_variance16x8,
+ vpx_highbd_8_sub_pixel_variance16x8,
+ vpx_highbd_8_sub_pixel_avg_variance16x8,
+ vpx_highbd_sad16x8x4d_bits8)
- HIGHBD_BFP(
- BLOCK_8X16, vpx_highbd_sad8x16_bits8, vpx_highbd_sad8x16_avg_bits8,
- vpx_highbd_8_variance8x16, vpx_highbd_8_sub_pixel_variance8x16,
- vpx_highbd_8_sub_pixel_avg_variance8x16, vpx_highbd_sad8x16x3_bits8,
- vpx_highbd_sad8x16x8_bits8, vpx_highbd_sad8x16x4d_bits8)
+ HIGHBD_BFP(BLOCK_8X16, vpx_highbd_sad8x16_bits8,
+ vpx_highbd_sad8x16_avg_bits8, vpx_highbd_8_variance8x16,
+ vpx_highbd_8_sub_pixel_variance8x16,
+ vpx_highbd_8_sub_pixel_avg_variance8x16,
+ vpx_highbd_sad8x16x4d_bits8)
HIGHBD_BFP(
BLOCK_8X8, vpx_highbd_sad8x8_bits8, vpx_highbd_sad8x8_avg_bits8,
vpx_highbd_8_variance8x8, vpx_highbd_8_sub_pixel_variance8x8,
- vpx_highbd_8_sub_pixel_avg_variance8x8, vpx_highbd_sad8x8x3_bits8,
- vpx_highbd_sad8x8x8_bits8, vpx_highbd_sad8x8x4d_bits8)
+ vpx_highbd_8_sub_pixel_avg_variance8x8, vpx_highbd_sad8x8x4d_bits8)
- HIGHBD_BFP(BLOCK_8X4, vpx_highbd_sad8x4_bits8,
- vpx_highbd_sad8x4_avg_bits8, vpx_highbd_8_variance8x4,
- vpx_highbd_8_sub_pixel_variance8x4,
- vpx_highbd_8_sub_pixel_avg_variance8x4, NULL,
- vpx_highbd_sad8x4x8_bits8, vpx_highbd_sad8x4x4d_bits8)
+ HIGHBD_BFP(
+ BLOCK_8X4, vpx_highbd_sad8x4_bits8, vpx_highbd_sad8x4_avg_bits8,
+ vpx_highbd_8_variance8x4, vpx_highbd_8_sub_pixel_variance8x4,
+ vpx_highbd_8_sub_pixel_avg_variance8x4, vpx_highbd_sad8x4x4d_bits8)
- HIGHBD_BFP(BLOCK_4X8, vpx_highbd_sad4x8_bits8,
- vpx_highbd_sad4x8_avg_bits8, vpx_highbd_8_variance4x8,
- vpx_highbd_8_sub_pixel_variance4x8,
- vpx_highbd_8_sub_pixel_avg_variance4x8, NULL,
- vpx_highbd_sad4x8x8_bits8, vpx_highbd_sad4x8x4d_bits8)
+ HIGHBD_BFP(
+ BLOCK_4X8, vpx_highbd_sad4x8_bits8, vpx_highbd_sad4x8_avg_bits8,
+ vpx_highbd_8_variance4x8, vpx_highbd_8_sub_pixel_variance4x8,
+ vpx_highbd_8_sub_pixel_avg_variance4x8, vpx_highbd_sad4x8x4d_bits8)
HIGHBD_BFP(
BLOCK_4X4, vpx_highbd_sad4x4_bits8, vpx_highbd_sad4x4_avg_bits8,
vpx_highbd_8_variance4x4, vpx_highbd_8_sub_pixel_variance4x4,
- vpx_highbd_8_sub_pixel_avg_variance4x4, vpx_highbd_sad4x4x3_bits8,
- vpx_highbd_sad4x4x8_bits8, vpx_highbd_sad4x4x4d_bits8)
+ vpx_highbd_8_sub_pixel_avg_variance4x4, vpx_highbd_sad4x4x4d_bits8)
break;
case VPX_BITS_10:
HIGHBD_BFP(BLOCK_32X16, vpx_highbd_sad32x16_bits10,
vpx_highbd_sad32x16_avg_bits10, vpx_highbd_10_variance32x16,
vpx_highbd_10_sub_pixel_variance32x16,
- vpx_highbd_10_sub_pixel_avg_variance32x16, NULL, NULL,
+ vpx_highbd_10_sub_pixel_avg_variance32x16,
vpx_highbd_sad32x16x4d_bits10)
HIGHBD_BFP(BLOCK_16X32, vpx_highbd_sad16x32_bits10,
vpx_highbd_sad16x32_avg_bits10, vpx_highbd_10_variance16x32,
vpx_highbd_10_sub_pixel_variance16x32,
- vpx_highbd_10_sub_pixel_avg_variance16x32, NULL, NULL,
+ vpx_highbd_10_sub_pixel_avg_variance16x32,
vpx_highbd_sad16x32x4d_bits10)
HIGHBD_BFP(BLOCK_64X32, vpx_highbd_sad64x32_bits10,
vpx_highbd_sad64x32_avg_bits10, vpx_highbd_10_variance64x32,
vpx_highbd_10_sub_pixel_variance64x32,
- vpx_highbd_10_sub_pixel_avg_variance64x32, NULL, NULL,
+ vpx_highbd_10_sub_pixel_avg_variance64x32,
vpx_highbd_sad64x32x4d_bits10)
HIGHBD_BFP(BLOCK_32X64, vpx_highbd_sad32x64_bits10,
vpx_highbd_sad32x64_avg_bits10, vpx_highbd_10_variance32x64,
vpx_highbd_10_sub_pixel_variance32x64,
- vpx_highbd_10_sub_pixel_avg_variance32x64, NULL, NULL,
+ vpx_highbd_10_sub_pixel_avg_variance32x64,
vpx_highbd_sad32x64x4d_bits10)
HIGHBD_BFP(BLOCK_32X32, vpx_highbd_sad32x32_bits10,
vpx_highbd_sad32x32_avg_bits10, vpx_highbd_10_variance32x32,
vpx_highbd_10_sub_pixel_variance32x32,
vpx_highbd_10_sub_pixel_avg_variance32x32,
- vpx_highbd_sad32x32x3_bits10, vpx_highbd_sad32x32x8_bits10,
vpx_highbd_sad32x32x4d_bits10)
HIGHBD_BFP(BLOCK_64X64, vpx_highbd_sad64x64_bits10,
vpx_highbd_sad64x64_avg_bits10, vpx_highbd_10_variance64x64,
vpx_highbd_10_sub_pixel_variance64x64,
vpx_highbd_10_sub_pixel_avg_variance64x64,
- vpx_highbd_sad64x64x3_bits10, vpx_highbd_sad64x64x8_bits10,
vpx_highbd_sad64x64x4d_bits10)
HIGHBD_BFP(BLOCK_16X16, vpx_highbd_sad16x16_bits10,
vpx_highbd_sad16x16_avg_bits10, vpx_highbd_10_variance16x16,
vpx_highbd_10_sub_pixel_variance16x16,
vpx_highbd_10_sub_pixel_avg_variance16x16,
- vpx_highbd_sad16x16x3_bits10, vpx_highbd_sad16x16x8_bits10,
vpx_highbd_sad16x16x4d_bits10)
HIGHBD_BFP(BLOCK_16X8, vpx_highbd_sad16x8_bits10,
vpx_highbd_sad16x8_avg_bits10, vpx_highbd_10_variance16x8,
vpx_highbd_10_sub_pixel_variance16x8,
vpx_highbd_10_sub_pixel_avg_variance16x8,
- vpx_highbd_sad16x8x3_bits10, vpx_highbd_sad16x8x8_bits10,
vpx_highbd_sad16x8x4d_bits10)
HIGHBD_BFP(BLOCK_8X16, vpx_highbd_sad8x16_bits10,
vpx_highbd_sad8x16_avg_bits10, vpx_highbd_10_variance8x16,
vpx_highbd_10_sub_pixel_variance8x16,
vpx_highbd_10_sub_pixel_avg_variance8x16,
- vpx_highbd_sad8x16x3_bits10, vpx_highbd_sad8x16x8_bits10,
vpx_highbd_sad8x16x4d_bits10)
- HIGHBD_BFP(
- BLOCK_8X8, vpx_highbd_sad8x8_bits10, vpx_highbd_sad8x8_avg_bits10,
- vpx_highbd_10_variance8x8, vpx_highbd_10_sub_pixel_variance8x8,
- vpx_highbd_10_sub_pixel_avg_variance8x8, vpx_highbd_sad8x8x3_bits10,
- vpx_highbd_sad8x8x8_bits10, vpx_highbd_sad8x8x4d_bits10)
+ HIGHBD_BFP(BLOCK_8X8, vpx_highbd_sad8x8_bits10,
+ vpx_highbd_sad8x8_avg_bits10, vpx_highbd_10_variance8x8,
+ vpx_highbd_10_sub_pixel_variance8x8,
+ vpx_highbd_10_sub_pixel_avg_variance8x8,
+ vpx_highbd_sad8x8x4d_bits10)
HIGHBD_BFP(BLOCK_8X4, vpx_highbd_sad8x4_bits10,
vpx_highbd_sad8x4_avg_bits10, vpx_highbd_10_variance8x4,
vpx_highbd_10_sub_pixel_variance8x4,
- vpx_highbd_10_sub_pixel_avg_variance8x4, NULL,
- vpx_highbd_sad8x4x8_bits10, vpx_highbd_sad8x4x4d_bits10)
+ vpx_highbd_10_sub_pixel_avg_variance8x4,
+ vpx_highbd_sad8x4x4d_bits10)
HIGHBD_BFP(BLOCK_4X8, vpx_highbd_sad4x8_bits10,
vpx_highbd_sad4x8_avg_bits10, vpx_highbd_10_variance4x8,
vpx_highbd_10_sub_pixel_variance4x8,
- vpx_highbd_10_sub_pixel_avg_variance4x8, NULL,
- vpx_highbd_sad4x8x8_bits10, vpx_highbd_sad4x8x4d_bits10)
-
- HIGHBD_BFP(
- BLOCK_4X4, vpx_highbd_sad4x4_bits10, vpx_highbd_sad4x4_avg_bits10,
- vpx_highbd_10_variance4x4, vpx_highbd_10_sub_pixel_variance4x4,
- vpx_highbd_10_sub_pixel_avg_variance4x4, vpx_highbd_sad4x4x3_bits10,
- vpx_highbd_sad4x4x8_bits10, vpx_highbd_sad4x4x4d_bits10)
+ vpx_highbd_10_sub_pixel_avg_variance4x8,
+ vpx_highbd_sad4x8x4d_bits10)
+
+ HIGHBD_BFP(BLOCK_4X4, vpx_highbd_sad4x4_bits10,
+ vpx_highbd_sad4x4_avg_bits10, vpx_highbd_10_variance4x4,
+ vpx_highbd_10_sub_pixel_variance4x4,
+ vpx_highbd_10_sub_pixel_avg_variance4x4,
+ vpx_highbd_sad4x4x4d_bits10)
break;
case VPX_BITS_12:
HIGHBD_BFP(BLOCK_32X16, vpx_highbd_sad32x16_bits12,
vpx_highbd_sad32x16_avg_bits12, vpx_highbd_12_variance32x16,
vpx_highbd_12_sub_pixel_variance32x16,
- vpx_highbd_12_sub_pixel_avg_variance32x16, NULL, NULL,
+ vpx_highbd_12_sub_pixel_avg_variance32x16,
vpx_highbd_sad32x16x4d_bits12)
HIGHBD_BFP(BLOCK_16X32, vpx_highbd_sad16x32_bits12,
vpx_highbd_sad16x32_avg_bits12, vpx_highbd_12_variance16x32,
vpx_highbd_12_sub_pixel_variance16x32,
- vpx_highbd_12_sub_pixel_avg_variance16x32, NULL, NULL,
+ vpx_highbd_12_sub_pixel_avg_variance16x32,
vpx_highbd_sad16x32x4d_bits12)
HIGHBD_BFP(BLOCK_64X32, vpx_highbd_sad64x32_bits12,
vpx_highbd_sad64x32_avg_bits12, vpx_highbd_12_variance64x32,
vpx_highbd_12_sub_pixel_variance64x32,
- vpx_highbd_12_sub_pixel_avg_variance64x32, NULL, NULL,
+ vpx_highbd_12_sub_pixel_avg_variance64x32,
vpx_highbd_sad64x32x4d_bits12)
HIGHBD_BFP(BLOCK_32X64, vpx_highbd_sad32x64_bits12,
vpx_highbd_sad32x64_avg_bits12, vpx_highbd_12_variance32x64,
vpx_highbd_12_sub_pixel_variance32x64,
- vpx_highbd_12_sub_pixel_avg_variance32x64, NULL, NULL,
+ vpx_highbd_12_sub_pixel_avg_variance32x64,
vpx_highbd_sad32x64x4d_bits12)
HIGHBD_BFP(BLOCK_32X32, vpx_highbd_sad32x32_bits12,
vpx_highbd_sad32x32_avg_bits12, vpx_highbd_12_variance32x32,
vpx_highbd_12_sub_pixel_variance32x32,
vpx_highbd_12_sub_pixel_avg_variance32x32,
- vpx_highbd_sad32x32x3_bits12, vpx_highbd_sad32x32x8_bits12,
vpx_highbd_sad32x32x4d_bits12)
HIGHBD_BFP(BLOCK_64X64, vpx_highbd_sad64x64_bits12,
vpx_highbd_sad64x64_avg_bits12, vpx_highbd_12_variance64x64,
vpx_highbd_12_sub_pixel_variance64x64,
vpx_highbd_12_sub_pixel_avg_variance64x64,
- vpx_highbd_sad64x64x3_bits12, vpx_highbd_sad64x64x8_bits12,
vpx_highbd_sad64x64x4d_bits12)
HIGHBD_BFP(BLOCK_16X16, vpx_highbd_sad16x16_bits12,
vpx_highbd_sad16x16_avg_bits12, vpx_highbd_12_variance16x16,
vpx_highbd_12_sub_pixel_variance16x16,
vpx_highbd_12_sub_pixel_avg_variance16x16,
- vpx_highbd_sad16x16x3_bits12, vpx_highbd_sad16x16x8_bits12,
vpx_highbd_sad16x16x4d_bits12)
HIGHBD_BFP(BLOCK_16X8, vpx_highbd_sad16x8_bits12,
vpx_highbd_sad16x8_avg_bits12, vpx_highbd_12_variance16x8,
vpx_highbd_12_sub_pixel_variance16x8,
vpx_highbd_12_sub_pixel_avg_variance16x8,
- vpx_highbd_sad16x8x3_bits12, vpx_highbd_sad16x8x8_bits12,
vpx_highbd_sad16x8x4d_bits12)
HIGHBD_BFP(BLOCK_8X16, vpx_highbd_sad8x16_bits12,
vpx_highbd_sad8x16_avg_bits12, vpx_highbd_12_variance8x16,
vpx_highbd_12_sub_pixel_variance8x16,
vpx_highbd_12_sub_pixel_avg_variance8x16,
- vpx_highbd_sad8x16x3_bits12, vpx_highbd_sad8x16x8_bits12,
vpx_highbd_sad8x16x4d_bits12)
- HIGHBD_BFP(
- BLOCK_8X8, vpx_highbd_sad8x8_bits12, vpx_highbd_sad8x8_avg_bits12,
- vpx_highbd_12_variance8x8, vpx_highbd_12_sub_pixel_variance8x8,
- vpx_highbd_12_sub_pixel_avg_variance8x8, vpx_highbd_sad8x8x3_bits12,
- vpx_highbd_sad8x8x8_bits12, vpx_highbd_sad8x8x4d_bits12)
+ HIGHBD_BFP(BLOCK_8X8, vpx_highbd_sad8x8_bits12,
+ vpx_highbd_sad8x8_avg_bits12, vpx_highbd_12_variance8x8,
+ vpx_highbd_12_sub_pixel_variance8x8,
+ vpx_highbd_12_sub_pixel_avg_variance8x8,
+ vpx_highbd_sad8x8x4d_bits12)
HIGHBD_BFP(BLOCK_8X4, vpx_highbd_sad8x4_bits12,
vpx_highbd_sad8x4_avg_bits12, vpx_highbd_12_variance8x4,
vpx_highbd_12_sub_pixel_variance8x4,
- vpx_highbd_12_sub_pixel_avg_variance8x4, NULL,
- vpx_highbd_sad8x4x8_bits12, vpx_highbd_sad8x4x4d_bits12)
+ vpx_highbd_12_sub_pixel_avg_variance8x4,
+ vpx_highbd_sad8x4x4d_bits12)
HIGHBD_BFP(BLOCK_4X8, vpx_highbd_sad4x8_bits12,
vpx_highbd_sad4x8_avg_bits12, vpx_highbd_12_variance4x8,
vpx_highbd_12_sub_pixel_variance4x8,
- vpx_highbd_12_sub_pixel_avg_variance4x8, NULL,
- vpx_highbd_sad4x8x8_bits12, vpx_highbd_sad4x8x4d_bits12)
-
- HIGHBD_BFP(
- BLOCK_4X4, vpx_highbd_sad4x4_bits12, vpx_highbd_sad4x4_avg_bits12,
- vpx_highbd_12_variance4x4, vpx_highbd_12_sub_pixel_variance4x4,
- vpx_highbd_12_sub_pixel_avg_variance4x4, vpx_highbd_sad4x4x3_bits12,
- vpx_highbd_sad4x4x8_bits12, vpx_highbd_sad4x4x4d_bits12)
+ vpx_highbd_12_sub_pixel_avg_variance4x8,
+ vpx_highbd_sad4x8x4d_bits12)
+
+ HIGHBD_BFP(BLOCK_4X4, vpx_highbd_sad4x4_bits12,
+ vpx_highbd_sad4x4_avg_bits12, vpx_highbd_12_variance4x4,
+ vpx_highbd_12_sub_pixel_variance4x4,
+ vpx_highbd_12_sub_pixel_avg_variance4x4,
+ vpx_highbd_sad4x4x4d_bits12)
break;
default:
cpi->source_var_thresh = 0;
cpi->frames_till_next_var_check = 0;
-#define BFP(BT, SDF, SDAF, VF, SVF, SVAF, SDX3F, SDX8F, SDX4DF) \
- cpi->fn_ptr[BT].sdf = SDF; \
- cpi->fn_ptr[BT].sdaf = SDAF; \
- cpi->fn_ptr[BT].vf = VF; \
- cpi->fn_ptr[BT].svf = SVF; \
- cpi->fn_ptr[BT].svaf = SVAF; \
- cpi->fn_ptr[BT].sdx3f = SDX3F; \
- cpi->fn_ptr[BT].sdx8f = SDX8F; \
+#define BFP(BT, SDF, SDAF, VF, SVF, SVAF, SDX4DF) \
+ cpi->fn_ptr[BT].sdf = SDF; \
+ cpi->fn_ptr[BT].sdaf = SDAF; \
+ cpi->fn_ptr[BT].vf = VF; \
+ cpi->fn_ptr[BT].svf = SVF; \
+ cpi->fn_ptr[BT].svaf = SVAF; \
cpi->fn_ptr[BT].sdx4df = SDX4DF;
BFP(BLOCK_32X16, vpx_sad32x16, vpx_sad32x16_avg, vpx_variance32x16,
- vpx_sub_pixel_variance32x16, vpx_sub_pixel_avg_variance32x16, NULL, NULL,
+ vpx_sub_pixel_variance32x16, vpx_sub_pixel_avg_variance32x16,
vpx_sad32x16x4d)
BFP(BLOCK_16X32, vpx_sad16x32, vpx_sad16x32_avg, vpx_variance16x32,
- vpx_sub_pixel_variance16x32, vpx_sub_pixel_avg_variance16x32, NULL, NULL,
+ vpx_sub_pixel_variance16x32, vpx_sub_pixel_avg_variance16x32,
vpx_sad16x32x4d)
BFP(BLOCK_64X32, vpx_sad64x32, vpx_sad64x32_avg, vpx_variance64x32,
- vpx_sub_pixel_variance64x32, vpx_sub_pixel_avg_variance64x32, NULL, NULL,
+ vpx_sub_pixel_variance64x32, vpx_sub_pixel_avg_variance64x32,
vpx_sad64x32x4d)
BFP(BLOCK_32X64, vpx_sad32x64, vpx_sad32x64_avg, vpx_variance32x64,
- vpx_sub_pixel_variance32x64, vpx_sub_pixel_avg_variance32x64, NULL, NULL,
+ vpx_sub_pixel_variance32x64, vpx_sub_pixel_avg_variance32x64,
vpx_sad32x64x4d)
BFP(BLOCK_32X32, vpx_sad32x32, vpx_sad32x32_avg, vpx_variance32x32,
vpx_sub_pixel_variance32x32, vpx_sub_pixel_avg_variance32x32,
- vpx_sad32x32x3, vpx_sad32x32x8, vpx_sad32x32x4d)
+ vpx_sad32x32x4d)
BFP(BLOCK_64X64, vpx_sad64x64, vpx_sad64x64_avg, vpx_variance64x64,
vpx_sub_pixel_variance64x64, vpx_sub_pixel_avg_variance64x64,
- vpx_sad64x64x3, vpx_sad64x64x8, vpx_sad64x64x4d)
+ vpx_sad64x64x4d)
BFP(BLOCK_16X16, vpx_sad16x16, vpx_sad16x16_avg, vpx_variance16x16,
vpx_sub_pixel_variance16x16, vpx_sub_pixel_avg_variance16x16,
- vpx_sad16x16x3, vpx_sad16x16x8, vpx_sad16x16x4d)
+ vpx_sad16x16x4d)
BFP(BLOCK_16X8, vpx_sad16x8, vpx_sad16x8_avg, vpx_variance16x8,
- vpx_sub_pixel_variance16x8, vpx_sub_pixel_avg_variance16x8, vpx_sad16x8x3,
- vpx_sad16x8x8, vpx_sad16x8x4d)
+ vpx_sub_pixel_variance16x8, vpx_sub_pixel_avg_variance16x8,
+ vpx_sad16x8x4d)
BFP(BLOCK_8X16, vpx_sad8x16, vpx_sad8x16_avg, vpx_variance8x16,
- vpx_sub_pixel_variance8x16, vpx_sub_pixel_avg_variance8x16, vpx_sad8x16x3,
- vpx_sad8x16x8, vpx_sad8x16x4d)
+ vpx_sub_pixel_variance8x16, vpx_sub_pixel_avg_variance8x16,
+ vpx_sad8x16x4d)
BFP(BLOCK_8X8, vpx_sad8x8, vpx_sad8x8_avg, vpx_variance8x8,
- vpx_sub_pixel_variance8x8, vpx_sub_pixel_avg_variance8x8, vpx_sad8x8x3,
- vpx_sad8x8x8, vpx_sad8x8x4d)
+ vpx_sub_pixel_variance8x8, vpx_sub_pixel_avg_variance8x8, vpx_sad8x8x4d)
BFP(BLOCK_8X4, vpx_sad8x4, vpx_sad8x4_avg, vpx_variance8x4,
- vpx_sub_pixel_variance8x4, vpx_sub_pixel_avg_variance8x4, NULL,
- vpx_sad8x4x8, vpx_sad8x4x4d)
+ vpx_sub_pixel_variance8x4, vpx_sub_pixel_avg_variance8x4, vpx_sad8x4x4d)
BFP(BLOCK_4X8, vpx_sad4x8, vpx_sad4x8_avg, vpx_variance4x8,
- vpx_sub_pixel_variance4x8, vpx_sub_pixel_avg_variance4x8, NULL,
- vpx_sad4x8x8, vpx_sad4x8x4d)
+ vpx_sub_pixel_variance4x8, vpx_sub_pixel_avg_variance4x8, vpx_sad4x8x4d)
BFP(BLOCK_4X4, vpx_sad4x4, vpx_sad4x4_avg, vpx_variance4x4,
- vpx_sub_pixel_variance4x4, vpx_sub_pixel_avg_variance4x4, vpx_sad4x4x3,
- vpx_sad4x4x8, vpx_sad4x4x4d)
+ vpx_sub_pixel_variance4x4, vpx_sub_pixel_avg_variance4x4, vpx_sad4x4x4d)
#if CONFIG_VP9_HIGHBITDEPTH
highbd_set_var_fns(cpi);
ActiveMap active_map;
fractional_mv_step_fp *find_fractional_mv_step;
- vp9_full_search_fn_t full_search_sad;
vp9_diamond_search_fn_t diamond_search_sad;
vp9_variance_fn_ptr_t fn_ptr[BLOCK_SIZES];
uint64_t time_receive_data;
return bestsme;
}
-int vp9_full_search_sad_c(const MACROBLOCK *x, const MV *ref_mv,
- int sad_per_bit, int distance,
- const vp9_variance_fn_ptr_t *fn_ptr,
- const MV *center_mv, MV *best_mv) {
- int r, c;
- const MACROBLOCKD *const xd = &x->e_mbd;
- const struct buf_2d *const what = &x->plane[0].src;
- const struct buf_2d *const in_what = &xd->plane[0].pre[0];
- const int row_min = VPXMAX(ref_mv->row - distance, x->mv_limits.row_min);
- const int row_max = VPXMIN(ref_mv->row + distance, x->mv_limits.row_max);
- const int col_min = VPXMAX(ref_mv->col - distance, x->mv_limits.col_min);
- const int col_max = VPXMIN(ref_mv->col + distance, x->mv_limits.col_max);
- const MV fcenter_mv = { center_mv->row >> 3, center_mv->col >> 3 };
- int best_sad =
- fn_ptr->sdf(what->buf, what->stride, get_buf_from_mv(in_what, ref_mv),
- in_what->stride) +
- mvsad_err_cost(x, ref_mv, &fcenter_mv, sad_per_bit);
- *best_mv = *ref_mv;
-
- for (r = row_min; r < row_max; ++r) {
- for (c = col_min; c < col_max; ++c) {
- const MV mv = { r, c };
- const int sad =
- fn_ptr->sdf(what->buf, what->stride, get_buf_from_mv(in_what, &mv),
- in_what->stride) +
- mvsad_err_cost(x, &mv, &fcenter_mv, sad_per_bit);
- if (sad < best_sad) {
- best_sad = sad;
- *best_mv = mv;
- }
- }
- }
- return best_sad;
-}
-
-int vp9_full_search_sadx3(const MACROBLOCK *x, const MV *ref_mv,
- int sad_per_bit, int distance,
- const vp9_variance_fn_ptr_t *fn_ptr,
- const MV *center_mv, MV *best_mv) {
- int r;
- const MACROBLOCKD *const xd = &x->e_mbd;
- const struct buf_2d *const what = &x->plane[0].src;
- const struct buf_2d *const in_what = &xd->plane[0].pre[0];
- const int row_min = VPXMAX(ref_mv->row - distance, x->mv_limits.row_min);
- const int row_max = VPXMIN(ref_mv->row + distance, x->mv_limits.row_max);
- const int col_min = VPXMAX(ref_mv->col - distance, x->mv_limits.col_min);
- const int col_max = VPXMIN(ref_mv->col + distance, x->mv_limits.col_max);
- const MV fcenter_mv = { center_mv->row >> 3, center_mv->col >> 3 };
- unsigned int best_sad =
- fn_ptr->sdf(what->buf, what->stride, get_buf_from_mv(in_what, ref_mv),
- in_what->stride) +
- mvsad_err_cost(x, ref_mv, &fcenter_mv, sad_per_bit);
- *best_mv = *ref_mv;
-
- for (r = row_min; r < row_max; ++r) {
- int c = col_min;
- const uint8_t *check_here = &in_what->buf[r * in_what->stride + c];
-
- if (fn_ptr->sdx3f != NULL) {
- while ((c + 2) < col_max) {
- int i;
- DECLARE_ALIGNED(16, uint32_t, sads[3]);
-
- fn_ptr->sdx3f(what->buf, what->stride, check_here, in_what->stride,
- sads);
-
- for (i = 0; i < 3; ++i) {
- unsigned int sad = sads[i];
- if (sad < best_sad) {
- const MV mv = { r, c };
- sad += mvsad_err_cost(x, &mv, &fcenter_mv, sad_per_bit);
- if (sad < best_sad) {
- best_sad = sad;
- *best_mv = mv;
- }
- }
- ++check_here;
- ++c;
- }
- }
- }
-
- while (c < col_max) {
- unsigned int sad =
- fn_ptr->sdf(what->buf, what->stride, check_here, in_what->stride);
- if (sad < best_sad) {
- const MV mv = { r, c };
- sad += mvsad_err_cost(x, &mv, &fcenter_mv, sad_per_bit);
- if (sad < best_sad) {
- best_sad = sad;
- *best_mv = mv;
- }
- }
- ++check_here;
- ++c;
- }
- }
-
- return best_sad;
-}
-
-int vp9_full_search_sadx8(const MACROBLOCK *x, const MV *ref_mv,
- int sad_per_bit, int distance,
- const vp9_variance_fn_ptr_t *fn_ptr,
- const MV *center_mv, MV *best_mv) {
- int r;
- const MACROBLOCKD *const xd = &x->e_mbd;
- const struct buf_2d *const what = &x->plane[0].src;
- const struct buf_2d *const in_what = &xd->plane[0].pre[0];
- const int row_min = VPXMAX(ref_mv->row - distance, x->mv_limits.row_min);
- const int row_max = VPXMIN(ref_mv->row + distance, x->mv_limits.row_max);
- const int col_min = VPXMAX(ref_mv->col - distance, x->mv_limits.col_min);
- const int col_max = VPXMIN(ref_mv->col + distance, x->mv_limits.col_max);
- const MV fcenter_mv = { center_mv->row >> 3, center_mv->col >> 3 };
- unsigned int best_sad =
- fn_ptr->sdf(what->buf, what->stride, get_buf_from_mv(in_what, ref_mv),
- in_what->stride) +
- mvsad_err_cost(x, ref_mv, &fcenter_mv, sad_per_bit);
- *best_mv = *ref_mv;
-
- for (r = row_min; r < row_max; ++r) {
- int c = col_min;
- const uint8_t *check_here = &in_what->buf[r * in_what->stride + c];
-
- if (fn_ptr->sdx8f != NULL) {
- while ((c + 7) < col_max) {
- int i;
- DECLARE_ALIGNED(16, uint32_t, sads[8]);
-
- fn_ptr->sdx8f(what->buf, what->stride, check_here, in_what->stride,
- sads);
-
- for (i = 0; i < 8; ++i) {
- unsigned int sad = sads[i];
- if (sad < best_sad) {
- const MV mv = { r, c };
- sad += mvsad_err_cost(x, &mv, &fcenter_mv, sad_per_bit);
- if (sad < best_sad) {
- best_sad = sad;
- *best_mv = mv;
- }
- }
- ++check_here;
- ++c;
- }
- }
- }
-
- if (fn_ptr->sdx3f != NULL) {
- while ((c + 2) < col_max) {
- int i;
- DECLARE_ALIGNED(16, uint32_t, sads[3]);
-
- fn_ptr->sdx3f(what->buf, what->stride, check_here, in_what->stride,
- sads);
-
- for (i = 0; i < 3; ++i) {
- unsigned int sad = sads[i];
- if (sad < best_sad) {
- const MV mv = { r, c };
- sad += mvsad_err_cost(x, &mv, &fcenter_mv, sad_per_bit);
- if (sad < best_sad) {
- best_sad = sad;
- *best_mv = mv;
- }
- }
- ++check_here;
- ++c;
- }
- }
- }
-
- while (c < col_max) {
- unsigned int sad =
- fn_ptr->sdf(what->buf, what->stride, check_here, in_what->stride);
- if (sad < best_sad) {
- const MV mv = { r, c };
- sad += mvsad_err_cost(x, &mv, &fcenter_mv, sad_per_bit);
- if (sad < best_sad) {
- best_sad = sad;
- *best_mv = mv;
- }
- }
- ++check_here;
- ++c;
- }
- }
-
- return best_sad;
-}
-
int vp9_refining_search_sad(const MACROBLOCK *x, MV *ref_mv, int error_per_bit,
int search_range,
const vp9_variance_fn_ptr_t *fn_ptr,
else if (oxcf->mode == GOOD)
set_good_speed_feature_framesize_independent(cpi, cm, sf, oxcf->speed);
- cpi->full_search_sad = vp9_full_search_sad;
cpi->diamond_search_sad = vp9_diamond_search_sad;
// Slow quant, dct and trellis not worthwhile for first pass
sad_array[2] = HADD_UH_U32(sad2);
}
-static void sad_32width_x3_msa(const uint8_t *src, int32_t src_stride,
- const uint8_t *ref, int32_t ref_stride,
- int32_t height, uint32_t *sad_array) {
- int32_t ht_cnt;
- v16u8 src0, src1, ref0_0, ref0_1, ref0_2, ref0, ref1;
- v8u16 sad0 = { 0 };
- v8u16 sad1 = { 0 };
- v8u16 sad2 = { 0 };
-
- for (ht_cnt = height >> 1; ht_cnt--;) {
- LD_UB2(src, 16, src0, src1);
- src += src_stride;
- LD_UB3(ref, 16, ref0_0, ref0_1, ref0_2);
- ref += ref_stride;
-
- sad0 += SAD_UB2_UH(src0, src1, ref0_0, ref0_1);
-
- SLDI_B2_UB(ref0_1, ref0_2, ref0_0, ref0_1, ref0, ref1, 1);
- sad1 += SAD_UB2_UH(src0, src1, ref0, ref1);
-
- SLDI_B2_UB(ref0_1, ref0_2, ref0_0, ref0_1, ref0, ref1, 2);
- sad2 += SAD_UB2_UH(src0, src1, ref0, ref1);
-
- LD_UB2(src, 16, src0, src1);
- src += src_stride;
- LD_UB3(ref, 16, ref0_0, ref0_1, ref0_2);
- ref += ref_stride;
-
- sad0 += SAD_UB2_UH(src0, src1, ref0_0, ref0_1);
-
- SLDI_B2_UB(ref0_1, ref0_2, ref0_0, ref0_1, ref0, ref1, 1);
- sad1 += SAD_UB2_UH(src0, src1, ref0, ref1);
-
- SLDI_B2_UB(ref0_1, ref0_2, ref0_0, ref0_1, ref0, ref1, 2);
- sad2 += SAD_UB2_UH(src0, src1, ref0, ref1);
- }
-
- sad_array[0] = HADD_UH_U32(sad0);
- sad_array[1] = HADD_UH_U32(sad1);
- sad_array[2] = HADD_UH_U32(sad2);
-}
-
-static void sad_64width_x3_msa(const uint8_t *src, int32_t src_stride,
- const uint8_t *ref, int32_t ref_stride,
- int32_t height, uint32_t *sad_array) {
- int32_t ht_cnt;
- v16u8 src0, src1, src2, src3;
- v16u8 ref0_0, ref0_1, ref0_2, ref0_3, ref0_4, ref0, ref1, ref2, ref3;
- v8u16 sad0_0 = { 0 };
- v8u16 sad0_1 = { 0 };
- v8u16 sad1_0 = { 0 };
- v8u16 sad1_1 = { 0 };
- v8u16 sad2_0 = { 0 };
- v8u16 sad2_1 = { 0 };
- v4u32 sad;
-
- for (ht_cnt = height; ht_cnt--;) {
- LD_UB4(src, 16, src0, src1, src2, src3);
- src += src_stride;
- LD_UB4(ref, 16, ref0_0, ref0_1, ref0_2, ref0_3);
- ref0_4 = LD_UB(ref + 64);
- ref += ref_stride;
-
- sad0_0 += SAD_UB2_UH(src0, src1, ref0_0, ref0_1);
- sad0_1 += SAD_UB2_UH(src2, src3, ref0_2, ref0_3);
-
- SLDI_B2_UB(ref0_1, ref0_2, ref0_0, ref0_1, ref0, ref1, 1);
- SLDI_B2_UB(ref0_3, ref0_4, ref0_2, ref0_3, ref2, ref3, 1);
- sad1_0 += SAD_UB2_UH(src0, src1, ref0, ref1);
- sad1_1 += SAD_UB2_UH(src2, src3, ref2, ref3);
-
- SLDI_B2_UB(ref0_1, ref0_2, ref0_0, ref0_1, ref0, ref1, 2);
- SLDI_B2_UB(ref0_3, ref0_4, ref0_2, ref0_3, ref2, ref3, 2);
- sad2_0 += SAD_UB2_UH(src0, src1, ref0, ref1);
- sad2_1 += SAD_UB2_UH(src2, src3, ref2, ref3);
- }
-
- sad = __msa_hadd_u_w(sad0_0, sad0_0);
- sad += __msa_hadd_u_w(sad0_1, sad0_1);
- sad_array[0] = HADD_SW_S32((v4i32)sad);
-
- sad = __msa_hadd_u_w(sad1_0, sad1_0);
- sad += __msa_hadd_u_w(sad1_1, sad1_1);
- sad_array[1] = HADD_SW_S32((v4i32)sad);
-
- sad = __msa_hadd_u_w(sad2_0, sad2_0);
- sad += __msa_hadd_u_w(sad2_1, sad2_1);
- sad_array[2] = HADD_SW_S32((v4i32)sad);
-}
-
static void sad_4width_x8_msa(const uint8_t *src_ptr, int32_t src_stride,
const uint8_t *ref_ptr, int32_t ref_stride,
int32_t height, uint32_t *sad_array) {
sad_array[7] = HADD_UH_U32(sad7);
}
-static void sad_32width_x8_msa(const uint8_t *src, int32_t src_stride,
- const uint8_t *ref, int32_t ref_stride,
- int32_t height, uint32_t *sad_array) {
- int32_t ht_cnt;
- v16u8 src0, src1;
- v16u8 ref0, ref1, ref0_0, ref0_1, ref0_2;
- v8u16 sad0 = { 0 };
- v8u16 sad1 = { 0 };
- v8u16 sad2 = { 0 };
- v8u16 sad3 = { 0 };
- v8u16 sad4 = { 0 };
- v8u16 sad5 = { 0 };
- v8u16 sad6 = { 0 };
- v8u16 sad7 = { 0 };
-
- for (ht_cnt = height; ht_cnt--;) {
- LD_UB2(src, 16, src0, src1);
- src += src_stride;
- LD_UB3(ref, 16, ref0_0, ref0_1, ref0_2);
- ref += ref_stride;
-
- sad0 += SAD_UB2_UH(src0, src1, ref0_0, ref0_1);
-
- SLDI_B2_UB(ref0_1, ref0_2, ref0_0, ref0_1, ref0, ref1, 1);
- sad1 += SAD_UB2_UH(src0, src1, ref0, ref1);
-
- SLDI_B2_UB(ref0_1, ref0_2, ref0_0, ref0_1, ref0, ref1, 2);
- sad2 += SAD_UB2_UH(src0, src1, ref0, ref1);
-
- SLDI_B2_UB(ref0_1, ref0_2, ref0_0, ref0_1, ref0, ref1, 3);
- sad3 += SAD_UB2_UH(src0, src1, ref0, ref1);
-
- SLDI_B2_UB(ref0_1, ref0_2, ref0_0, ref0_1, ref0, ref1, 4);
- sad4 += SAD_UB2_UH(src0, src1, ref0, ref1);
-
- SLDI_B2_UB(ref0_1, ref0_2, ref0_0, ref0_1, ref0, ref1, 5);
- sad5 += SAD_UB2_UH(src0, src1, ref0, ref1);
-
- SLDI_B2_UB(ref0_1, ref0_2, ref0_0, ref0_1, ref0, ref1, 6);
- sad6 += SAD_UB2_UH(src0, src1, ref0, ref1);
-
- SLDI_B2_UB(ref0_1, ref0_2, ref0_0, ref0_1, ref0, ref1, 7);
- sad7 += SAD_UB2_UH(src0, src1, ref0, ref1);
- }
-
- sad_array[0] = HADD_UH_U32(sad0);
- sad_array[1] = HADD_UH_U32(sad1);
- sad_array[2] = HADD_UH_U32(sad2);
- sad_array[3] = HADD_UH_U32(sad3);
- sad_array[4] = HADD_UH_U32(sad4);
- sad_array[5] = HADD_UH_U32(sad5);
- sad_array[6] = HADD_UH_U32(sad6);
- sad_array[7] = HADD_UH_U32(sad7);
-}
-
-static void sad_64width_x8_msa(const uint8_t *src, int32_t src_stride,
- const uint8_t *ref, int32_t ref_stride,
- int32_t height, uint32_t *sad_array) {
- const uint8_t *src_dup, *ref_dup;
- int32_t ht_cnt;
- v16u8 src0, src1, src2, src3;
- v16u8 ref0_0, ref0_1, ref0_2, ref0_3, ref0_4;
- v16u8 ref0, ref1, ref2, ref3;
- v8u16 sad0_0 = { 0 };
- v8u16 sad0_1 = { 0 };
- v8u16 sad1_0 = { 0 };
- v8u16 sad1_1 = { 0 };
- v8u16 sad2_0 = { 0 };
- v8u16 sad2_1 = { 0 };
- v8u16 sad3_0 = { 0 };
- v8u16 sad3_1 = { 0 };
- v4u32 sad;
-
- src_dup = src;
- ref_dup = ref;
-
- for (ht_cnt = height; ht_cnt--;) {
- LD_UB4(src, 16, src0, src1, src2, src3);
- src += src_stride;
- LD_UB5(ref, 16, ref0_0, ref0_1, ref0_2, ref0_3, ref0_4);
- ref += ref_stride;
-
- sad0_0 += SAD_UB2_UH(src0, src1, ref0_0, ref0_1);
- sad0_1 += SAD_UB2_UH(src2, src3, ref0_2, ref0_3);
-
- SLDI_B2_UB(ref0_1, ref0_2, ref0_0, ref0_1, ref0, ref1, 1);
- SLDI_B2_UB(ref0_3, ref0_4, ref0_2, ref0_3, ref2, ref3, 1);
- sad1_0 += SAD_UB2_UH(src0, src1, ref0, ref1);
- sad1_1 += SAD_UB2_UH(src2, src3, ref2, ref3);
-
- SLDI_B2_UB(ref0_1, ref0_2, ref0_0, ref0_1, ref0, ref1, 2);
- SLDI_B2_UB(ref0_3, ref0_4, ref0_2, ref0_3, ref2, ref3, 2);
- sad2_0 += SAD_UB2_UH(src0, src1, ref0, ref1);
- sad2_1 += SAD_UB2_UH(src2, src3, ref2, ref3);
-
- SLDI_B2_UB(ref0_1, ref0_2, ref0_0, ref0_1, ref0, ref1, 3);
- SLDI_B2_UB(ref0_3, ref0_4, ref0_2, ref0_3, ref2, ref3, 3);
- sad3_0 += SAD_UB2_UH(src0, src1, ref0, ref1);
- sad3_1 += SAD_UB2_UH(src2, src3, ref2, ref3);
- }
-
- sad = __msa_hadd_u_w(sad0_0, sad0_0);
- sad += __msa_hadd_u_w(sad0_1, sad0_1);
- sad_array[0] = HADD_SW_S32(sad);
-
- sad = __msa_hadd_u_w(sad1_0, sad1_0);
- sad += __msa_hadd_u_w(sad1_1, sad1_1);
- sad_array[1] = HADD_SW_S32(sad);
-
- sad = __msa_hadd_u_w(sad2_0, sad2_0);
- sad += __msa_hadd_u_w(sad2_1, sad2_1);
- sad_array[2] = HADD_SW_S32(sad);
-
- sad = __msa_hadd_u_w(sad3_0, sad3_0);
- sad += __msa_hadd_u_w(sad3_1, sad3_1);
- sad_array[3] = HADD_SW_S32(sad);
-
- sad0_0 = (v8u16)__msa_ldi_h(0);
- sad0_1 = (v8u16)__msa_ldi_h(0);
- sad1_0 = (v8u16)__msa_ldi_h(0);
- sad1_1 = (v8u16)__msa_ldi_h(0);
- sad2_0 = (v8u16)__msa_ldi_h(0);
- sad2_1 = (v8u16)__msa_ldi_h(0);
- sad3_0 = (v8u16)__msa_ldi_h(0);
- sad3_1 = (v8u16)__msa_ldi_h(0);
-
- for (ht_cnt = 64; ht_cnt--;) {
- LD_UB4(src_dup, 16, src0, src1, src2, src3);
- src_dup += src_stride;
- LD_UB5(ref_dup, 16, ref0_0, ref0_1, ref0_2, ref0_3, ref0_4);
- ref_dup += ref_stride;
-
- SLDI_B2_UB(ref0_1, ref0_2, ref0_0, ref0_1, ref0, ref1, 4);
- SLDI_B2_UB(ref0_3, ref0_4, ref0_2, ref0_3, ref2, ref3, 4);
- sad0_0 += SAD_UB2_UH(src0, src1, ref0, ref1);
- sad0_1 += SAD_UB2_UH(src2, src3, ref2, ref3);
-
- SLDI_B2_UB(ref0_1, ref0_2, ref0_0, ref0_1, ref0, ref1, 5);
- SLDI_B2_UB(ref0_3, ref0_4, ref0_2, ref0_3, ref2, ref3, 5);
- sad1_0 += SAD_UB2_UH(src0, src1, ref0, ref1);
- sad1_1 += SAD_UB2_UH(src2, src3, ref2, ref3);
-
- SLDI_B2_UB(ref0_1, ref0_2, ref0_0, ref0_1, ref0, ref1, 6);
- SLDI_B2_UB(ref0_3, ref0_4, ref0_2, ref0_3, ref2, ref3, 6);
- sad2_0 += SAD_UB2_UH(src0, src1, ref0, ref1);
- sad2_1 += SAD_UB2_UH(src2, src3, ref2, ref3);
-
- SLDI_B2_UB(ref0_1, ref0_2, ref0_0, ref0_1, ref0, ref1, 7);
- SLDI_B2_UB(ref0_3, ref0_4, ref0_2, ref0_3, ref2, ref3, 7);
- sad3_0 += SAD_UB2_UH(src0, src1, ref0, ref1);
- sad3_1 += SAD_UB2_UH(src2, src3, ref2, ref3);
- }
-
- sad = __msa_hadd_u_w(sad0_0, sad0_0);
- sad += __msa_hadd_u_w(sad0_1, sad0_1);
- sad_array[4] = HADD_SW_S32(sad);
-
- sad = __msa_hadd_u_w(sad1_0, sad1_0);
- sad += __msa_hadd_u_w(sad1_1, sad1_1);
- sad_array[5] = HADD_SW_S32(sad);
-
- sad = __msa_hadd_u_w(sad2_0, sad2_0);
- sad += __msa_hadd_u_w(sad2_1, sad2_1);
- sad_array[6] = HADD_SW_S32(sad);
-
- sad = __msa_hadd_u_w(sad3_0, sad3_0);
- sad += __msa_hadd_u_w(sad3_1, sad3_1);
- sad_array[7] = HADD_SW_S32(sad);
-}
-
static void sad_4width_x4d_msa(const uint8_t *src_ptr, int32_t src_stride,
const uint8_t *const aref_ptr[],
int32_t ref_stride, int32_t height,
sad_16width_x3_msa(src, src_stride, ref, ref_stride, height, sads); \
}
-#define VPX_SAD_32xHEIGHTx3_MSA(height) \
- void vpx_sad32x##height##x3_msa(const uint8_t *src, int32_t src_stride, \
- const uint8_t *ref, int32_t ref_stride, \
- uint32_t *sads) { \
- sad_32width_x3_msa(src, src_stride, ref, ref_stride, height, sads); \
- }
-
-#define VPX_SAD_64xHEIGHTx3_MSA(height) \
- void vpx_sad64x##height##x3_msa(const uint8_t *src, int32_t src_stride, \
- const uint8_t *ref, int32_t ref_stride, \
- uint32_t *sads) { \
- sad_64width_x3_msa(src, src_stride, ref, ref_stride, height, sads); \
- }
-
#define VPX_SAD_4xHEIGHTx8_MSA(height) \
void vpx_sad4x##height##x8_msa(const uint8_t *src, int32_t src_stride, \
const uint8_t *ref, int32_t ref_stride, \
sad_16width_x8_msa(src, src_stride, ref, ref_stride, height, sads); \
}
-#define VPX_SAD_32xHEIGHTx8_MSA(height) \
- void vpx_sad32x##height##x8_msa(const uint8_t *src, int32_t src_stride, \
- const uint8_t *ref, int32_t ref_stride, \
- uint32_t *sads) { \
- sad_32width_x8_msa(src, src_stride, ref, ref_stride, height, sads); \
- }
-
-#define VPX_SAD_64xHEIGHTx8_MSA(height) \
- void vpx_sad64x##height##x8_msa(const uint8_t *src, int32_t src_stride, \
- const uint8_t *ref, int32_t ref_stride, \
- uint32_t *sads) { \
- sad_64width_x8_msa(src, src_stride, ref, ref_stride, height, sads); \
- }
-
#define VPX_SAD_4xHEIGHTx4D_MSA(height) \
void vpx_sad4x##height##x4d_msa(const uint8_t *src, int32_t src_stride, \
const uint8_t *const refs[], \
// 64x64
VPX_SAD_64xHEIGHT_MSA(64);
-VPX_SAD_64xHEIGHTx3_MSA(64);
-VPX_SAD_64xHEIGHTx8_MSA(64);
VPX_SAD_64xHEIGHTx4D_MSA(64);
VPX_AVGSAD_64xHEIGHT_MSA(64);
// 64x32
VPX_SAD_64xHEIGHT_MSA(32);
-VPX_SAD_64xHEIGHTx3_MSA(32);
-VPX_SAD_64xHEIGHTx8_MSA(32);
VPX_SAD_64xHEIGHTx4D_MSA(32);
VPX_AVGSAD_64xHEIGHT_MSA(32);
// 32x64
VPX_SAD_32xHEIGHT_MSA(64);
-VPX_SAD_32xHEIGHTx3_MSA(64);
-VPX_SAD_32xHEIGHTx8_MSA(64);
VPX_SAD_32xHEIGHTx4D_MSA(64);
VPX_AVGSAD_32xHEIGHT_MSA(64);
// 32x32
VPX_SAD_32xHEIGHT_MSA(32);
-VPX_SAD_32xHEIGHTx3_MSA(32);
-VPX_SAD_32xHEIGHTx8_MSA(32);
VPX_SAD_32xHEIGHTx4D_MSA(32);
VPX_AVGSAD_32xHEIGHT_MSA(32);
// 32x16
VPX_SAD_32xHEIGHT_MSA(16);
-VPX_SAD_32xHEIGHTx3_MSA(16);
-VPX_SAD_32xHEIGHTx8_MSA(16);
VPX_SAD_32xHEIGHTx4D_MSA(16);
VPX_AVGSAD_32xHEIGHT_MSA(16);
// 16x32
VPX_SAD_16xHEIGHT_MSA(32);
-VPX_SAD_16xHEIGHTx3_MSA(32);
-VPX_SAD_16xHEIGHTx8_MSA(32);
VPX_SAD_16xHEIGHTx4D_MSA(32);
VPX_AVGSAD_16xHEIGHT_MSA(32);
// 8x4
VPX_SAD_8xHEIGHT_MSA(4);
-VPX_SAD_8xHEIGHTx3_MSA(4);
-VPX_SAD_8xHEIGHTx8_MSA(4);
VPX_SAD_8xHEIGHTx4D_MSA(4);
VPX_AVGSAD_8xHEIGHT_MSA(4);
// 4x8
VPX_SAD_4xHEIGHT_MSA(8);
-VPX_SAD_4xHEIGHTx3_MSA(8);
-VPX_SAD_4xHEIGHTx8_MSA(8);
VPX_SAD_4xHEIGHTx4D_MSA(8);
VPX_AVGSAD_4xHEIGHT_MSA(8);
/* clang-format off */
// 64x64
sadMxN(64, 64)
-sadMxNxK(64, 64, 3)
-sadMxNxK(64, 64, 8)
sadMxNx4D(64, 64)
// 64x32
// 32x32
sadMxN(32, 32)
-sadMxNxK(32, 32, 3)
-sadMxNxK(32, 32, 8)
sadMxNx4D(32, 32)
// 32x16
// 8x4
sadMxN(8, 4)
-sadMxNxK(8, 4, 8)
sadMxNx4D(8, 4)
// 4x8
sadMxN(4, 8)
-sadMxNxK(4, 8, 8)
sadMxNx4D(4, 8)
// 4x4
return highbd_sadb(src, src_stride, comp_pred, m, m, n); \
}
-#define highbd_sadMxNxK(m, n, k) \
- void vpx_highbd_sad##m##x##n##x##k##_c( \
- const uint8_t *src, int src_stride, const uint8_t *ref_array, \
- int ref_stride, uint32_t *sad_array) { \
- int i; \
- for (i = 0; i < k; ++i) { \
- sad_array[i] = vpx_highbd_sad##m##x##n##_c(src, src_stride, \
- &ref_array[i], ref_stride); \
- } \
- }
-
#define highbd_sadMxNx4D(m, n) \
void vpx_highbd_sad##m##x##n##x4d_c(const uint8_t *src, int src_stride, \
const uint8_t *const ref_array[], \
/* clang-format off */
// 64x64
highbd_sadMxN(64, 64)
-highbd_sadMxNxK(64, 64, 3)
-highbd_sadMxNxK(64, 64, 8)
highbd_sadMxNx4D(64, 64)
// 64x32
// 32x32
highbd_sadMxN(32, 32)
-highbd_sadMxNxK(32, 32, 3)
-highbd_sadMxNxK(32, 32, 8)
highbd_sadMxNx4D(32, 32)
// 32x16
// 16x16
highbd_sadMxN(16, 16)
-highbd_sadMxNxK(16, 16, 3)
-highbd_sadMxNxK(16, 16, 8)
highbd_sadMxNx4D(16, 16)
// 16x8
highbd_sadMxN(16, 8)
-highbd_sadMxNxK(16, 8, 3)
-highbd_sadMxNxK(16, 8, 8)
highbd_sadMxNx4D(16, 8)
// 8x16
highbd_sadMxN(8, 16)
-highbd_sadMxNxK(8, 16, 3)
-highbd_sadMxNxK(8, 16, 8)
highbd_sadMxNx4D(8, 16)
// 8x8
highbd_sadMxN(8, 8)
-highbd_sadMxNxK(8, 8, 3)
-highbd_sadMxNxK(8, 8, 8)
highbd_sadMxNx4D(8, 8)
// 8x4
highbd_sadMxN(8, 4)
-highbd_sadMxNxK(8, 4, 8)
highbd_sadMxNx4D(8, 4)
// 4x8
highbd_sadMxN(4, 8)
-highbd_sadMxNxK(4, 8, 8)
highbd_sadMxNx4D(4, 8)
// 4x4
highbd_sadMxN(4, 4)
-highbd_sadMxNxK(4, 4, 3)
-highbd_sadMxNxK(4, 4, 8)
highbd_sadMxNx4D(4, 4)
/* clang-format on */
vpx_variance_fn_t vf;
vpx_subpixvariance_fn_t svf;
vpx_subp_avg_variance_fn_t svaf;
- vpx_sad_multi_fn_t sdx3f;
- vpx_sad_multi_fn_t sdx8f;
vpx_sad_multi_d_fn_t sdx4df;
} vp9_variance_fn_ptr_t;
#endif // CONFIG_VP9
# Multi-block SAD, comparing a reference to N blocks 1 pixel apart horizontally
#
# Blocks of 3
-add_proto qw/void vpx_sad64x64x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-specialize qw/vpx_sad64x64x3 msa/;
-
-add_proto qw/void vpx_sad32x32x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-specialize qw/vpx_sad32x32x3 msa/;
-
add_proto qw/void vpx_sad16x16x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
specialize qw/vpx_sad16x16x3 sse3 ssse3 msa/;
specialize qw/vpx_sad4x4x3 sse3 msa/;
# Blocks of 8
-add_proto qw/void vpx_sad64x64x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-specialize qw/vpx_sad64x64x8 msa/;
-
-add_proto qw/void vpx_sad32x32x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-specialize qw/vpx_sad32x32x8 msa/;
-
add_proto qw/void vpx_sad16x16x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
specialize qw/vpx_sad16x16x8 sse4_1 msa/;
add_proto qw/void vpx_sad8x8x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
specialize qw/vpx_sad8x8x8 sse4_1 msa/;
-add_proto qw/void vpx_sad8x4x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-specialize qw/vpx_sad8x4x8 msa/;
-
-add_proto qw/void vpx_sad4x8x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-specialize qw/vpx_sad4x8x8 msa/;
-
add_proto qw/void vpx_sad4x4x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
specialize qw/vpx_sad4x4x8 sse4_1 msa/;
add_proto qw/unsigned int vpx_highbd_sad4x4_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
- #
- # Multi-block SAD, comparing a reference to N blocks 1 pixel apart horizontally
- #
- # Blocks of 3
- add_proto qw/void vpx_highbd_sad64x64x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-
- add_proto qw/void vpx_highbd_sad32x32x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-
- add_proto qw/void vpx_highbd_sad16x16x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-
- add_proto qw/void vpx_highbd_sad16x8x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-
- add_proto qw/void vpx_highbd_sad8x16x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-
- add_proto qw/void vpx_highbd_sad8x8x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-
- add_proto qw/void vpx_highbd_sad4x4x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-
- # Blocks of 8
- add_proto qw/void vpx_highbd_sad64x64x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-
- add_proto qw/void vpx_highbd_sad32x32x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-
- add_proto qw/void vpx_highbd_sad16x16x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-
- add_proto qw/void vpx_highbd_sad16x8x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-
- add_proto qw/void vpx_highbd_sad8x16x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-
- add_proto qw/void vpx_highbd_sad8x8x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-
- add_proto qw/void vpx_highbd_sad8x4x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-
- add_proto qw/void vpx_highbd_sad4x8x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-
- add_proto qw/void vpx_highbd_sad4x4x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
-
#
# Multi-block SAD, comparing a reference to N independent blocks
#
projects / libvpx / commitdiff