Migrate some vp9 highbd intrapred x86 speedup to av1
Function speedup on i7-6700:
D117 sse2 ssse3
4x4 ~1.8x
8x8 ~3.4x
16x16 ~5.5x
32x32 ~2.9x
D135 sse2 ssse3
4x4 ~1.9
8x8 ~3.3x
16x16 ~5.3x
32x32 ~3.6x
D153 sse2 ssse3
4x4 ~1.9x
8x8 ~2.8x
16x16 ~5.5x
32x32 ~3.6x
Change-Id: I43ab5fa8dcbcfa51acbde554abf3e5d7d336f391
diff --git a/aom_dsp/aom_dsp.cmake b/aom_dsp/aom_dsp.cmake
index ab2e36f..dc8d24a 100644
--- a/aom_dsp/aom_dsp.cmake
+++ b/aom_dsp/aom_dsp.cmake
@@ -239,6 +239,10 @@
"${AOM_ROOT}/aom_dsp/x86/highbd_intrapred_sse2.c"
"${AOM_ROOT}/aom_dsp/x86/highbd_loopfilter_sse2.c")
+ set(AOM_DSP_COMMON_INTRIN_SSSE3
+ ${AOM_DSP_COMMON_INTRIN_SSSE3}
+ "${AOM_ROOT}/aom_dsp/x86/highbd_intrapred_ssse3.c")
+
set(AOM_DSP_COMMON_INTRIN_AVX2
${AOM_DSP_COMMON_INTRIN_AVX2}
"${AOM_ROOT}/aom_dsp/x86/highbd_convolve_avx2.c"
diff --git a/aom_dsp/aom_dsp.mk b/aom_dsp/aom_dsp.mk
index e85c9e3..7037c0e 100644
--- a/aom_dsp/aom_dsp.mk
+++ b/aom_dsp/aom_dsp.mk
@@ -84,6 +84,7 @@
DSP_SRCS-$(HAVE_SSE) += x86/highbd_intrapred_sse2.asm
DSP_SRCS-$(HAVE_SSE2) += x86/highbd_intrapred_sse2.asm
DSP_SRCS-$(HAVE_SSE2) += x86/highbd_intrapred_sse2.c
+DSP_SRCS-$(HAVE_SSSE3) += x86/highbd_intrapred_ssse3.c
endif # CONFIG_HIGHBITDEPTH
DSP_SRCS-$(HAVE_NEON_ASM) += arm/intrapred_neon_asm$(ASM)
diff --git a/aom_dsp/aom_dsp_rtcd_defs.pl b/aom_dsp/aom_dsp_rtcd_defs.pl
index 5962726..d07241a 100755
--- a/aom_dsp/aom_dsp_rtcd_defs.pl
+++ b/aom_dsp/aom_dsp_rtcd_defs.pl
@@ -242,6 +242,19 @@
specialize qw/aom_highbd_dc_left_predictor_32x32 sse2/;
specialize qw/aom_highbd_dc_top_predictor_32x32 sse2/;
specialize qw/aom_highbd_dc_128_predictor_32x32 sse2/;
+
+ specialize qw/aom_highbd_d117_predictor_4x4 sse2/;
+ specialize qw/aom_highbd_d117_predictor_8x8 ssse3/;
+ specialize qw/aom_highbd_d117_predictor_16x16 ssse3/;
+ specialize qw/aom_highbd_d117_predictor_32x32 ssse3/;
+ specialize qw/aom_highbd_d135_predictor_4x4 sse2/;
+ specialize qw/aom_highbd_d135_predictor_8x8 ssse3/;
+ specialize qw/aom_highbd_d135_predictor_16x16 ssse3/;
+ specialize qw/aom_highbd_d135_predictor_32x32 ssse3/;
+ specialize qw/aom_highbd_d153_predictor_4x4 sse2/;
+ specialize qw/aom_highbd_d153_predictor_8x8 ssse3/;
+ specialize qw/aom_highbd_d153_predictor_16x16 ssse3/;
+ specialize qw/aom_highbd_d153_predictor_32x32 ssse3/;
} # CONFIG_HIGHBITDEPTH
#
diff --git a/aom_dsp/x86/highbd_intrapred_sse2.c b/aom_dsp/x86/highbd_intrapred_sse2.c
index eb8da38..7200044 100644
--- a/aom_dsp/x86/highbd_intrapred_sse2.c
+++ b/aom_dsp/x86/highbd_intrapred_sse2.c
@@ -982,3 +982,115 @@
dst += stride;
}
}
+
+// -----------------------------------------------------------------------------
+/*
+; ------------------------------------------
+; input: x, y, z, result
+;
+; trick from pascal
+; (x+2y+z+2)>>2 can be calculated as:
+; result = avg(x,z)
+; result -= xor(x,z) & 1
+; result = avg(result,y)
+; ------------------------------------------
+*/
+static INLINE __m128i avg3_epu16(const __m128i *x, const __m128i *y,
+ const __m128i *z) {
+ const __m128i one = _mm_set1_epi16(1);
+ const __m128i a = _mm_avg_epu16(*x, *z);
+ const __m128i b =
+ _mm_subs_epu16(a, _mm_and_si128(_mm_xor_si128(*x, *z), one));
+ return _mm_avg_epu16(b, *y);
+}
+
+void aom_highbd_d117_predictor_4x4_sse2(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {
+ const int I = left[0];
+ const int J = left[1];
+ const int K = left[2];
+ const __m128i XXXXABCD = _mm_loadu_si128((const __m128i *)(above - 4));
+ const __m128i KXXXABCD = _mm_insert_epi16(XXXXABCD, K, 0);
+ const __m128i KJXXABCD = _mm_insert_epi16(KXXXABCD, J, 1);
+ const __m128i KJIXABCD = _mm_insert_epi16(KJXXABCD, I, 2);
+ const __m128i JIXABCD0 = _mm_srli_si128(KJIXABCD, 2);
+ const __m128i IXABCD00 = _mm_srli_si128(KJIXABCD, 4);
+ const __m128i avg2 = _mm_avg_epu16(KJIXABCD, JIXABCD0);
+ const __m128i avg3 = avg3_epu16(&KJIXABCD, &JIXABCD0, &IXABCD00);
+ const __m128i row0 = _mm_srli_si128(avg2, 6);
+ const __m128i row1 = _mm_srli_si128(avg3, 4);
+ const __m128i row2 = _mm_srli_si128(avg2, 4);
+ const __m128i row3 = _mm_srli_si128(avg3, 2);
+ (void)bd;
+ _mm_storel_epi64((__m128i *)dst, row0);
+ dst += stride;
+ _mm_storel_epi64((__m128i *)dst, row1);
+ dst += stride;
+ _mm_storel_epi64((__m128i *)dst, row2);
+ dst += stride;
+ _mm_storel_epi64((__m128i *)dst, row3);
+
+ dst -= stride;
+ dst[0] = _mm_extract_epi16(avg3, 1);
+ dst[stride] = _mm_extract_epi16(avg3, 0);
+}
+
+void aom_highbd_d135_predictor_4x4_sse2(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {
+ const int I = left[0];
+ const int J = left[1];
+ const int K = left[2];
+ const int L = left[3];
+ const __m128i XXXXABCD = _mm_loadu_si128((const __m128i *)(above - 4));
+ const __m128i KXXXABCD = _mm_insert_epi16(XXXXABCD, K, 0);
+ const __m128i KJXXABCD = _mm_insert_epi16(KXXXABCD, J, 1);
+ const __m128i KJIXABCD = _mm_insert_epi16(KJXXABCD, I, 2);
+ const __m128i JIXABCD0 = _mm_srli_si128(KJIXABCD, 2);
+ const __m128i LKJIXABC = _mm_insert_epi16(_mm_slli_si128(KJIXABCD, 2), L, 0);
+ const __m128i avg3 = avg3_epu16(&JIXABCD0, &KJIXABCD, &LKJIXABC);
+ const __m128i row0 = _mm_srli_si128(avg3, 6);
+ const __m128i row1 = _mm_srli_si128(avg3, 4);
+ const __m128i row2 = _mm_srli_si128(avg3, 2);
+ const __m128i row3 = avg3;
+ (void)bd;
+ _mm_storel_epi64((__m128i *)dst, row0);
+ dst += stride;
+ _mm_storel_epi64((__m128i *)dst, row1);
+ dst += stride;
+ _mm_storel_epi64((__m128i *)dst, row2);
+ dst += stride;
+ _mm_storel_epi64((__m128i *)dst, row3);
+}
+
+void aom_highbd_d153_predictor_4x4_sse2(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {
+ const int I = left[0];
+ const int J = left[1];
+ const int K = left[2];
+ const int L = left[3];
+ const __m128i XXXXXABC = _mm_loadu_si128((const __m128i *)(above - 5));
+ const __m128i LXXXXABC = _mm_insert_epi16(XXXXXABC, L, 0);
+ const __m128i LKXXXABC = _mm_insert_epi16(LXXXXABC, K, 1);
+ const __m128i LKJXXABC = _mm_insert_epi16(LKXXXABC, J, 2);
+ const __m128i LKJIXABC = _mm_insert_epi16(LKJXXABC, I, 3);
+ const __m128i KJIXABC0 = _mm_srli_si128(LKJIXABC, 2);
+ const __m128i JIXABC00 = _mm_srli_si128(LKJIXABC, 4);
+ const __m128i avg3 = avg3_epu16(&LKJIXABC, &KJIXABC0, &JIXABC00);
+ const __m128i avg2 = _mm_avg_epu16(LKJIXABC, KJIXABC0);
+ const __m128i row3 = _mm_unpacklo_epi16(avg2, avg3);
+ const __m128i row2 = _mm_srli_si128(row3, 4);
+ const __m128i row1 = _mm_srli_si128(row3, 8);
+ const __m128i row0 = _mm_srli_si128(avg3, 4);
+ (void)bd;
+ _mm_storel_epi64((__m128i *)dst, row0);
+ dst[0] = _mm_extract_epi16(avg2, 3);
+ dst += stride;
+ _mm_storel_epi64((__m128i *)dst, row1);
+ dst += stride;
+ _mm_storel_epi64((__m128i *)dst, row2);
+ dst += stride;
+ _mm_storel_epi64((__m128i *)dst, row3);
+}
diff --git a/aom_dsp/x86/highbd_intrapred_ssse3.c b/aom_dsp/x86/highbd_intrapred_ssse3.c
new file mode 100644
index 0000000..b089a3f
--- /dev/null
+++ b/aom_dsp/x86/highbd_intrapred_ssse3.c
@@ -0,0 +1,521 @@
+/*
+ * Copyright (c) 2017, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <tmmintrin.h>
+
+#include "./aom_dsp_rtcd.h"
+
+// -----------------------------------------------------------------------------
+/*
+; ------------------------------------------
+; input: x, y, z, result
+;
+; trick from pascal
+; (x+2y+z+2)>>2 can be calculated as:
+; result = avg(x,z)
+; result -= xor(x,z) & 1
+; result = avg(result,y)
+; ------------------------------------------
+*/
+static INLINE __m128i avg3_epu16(const __m128i *x, const __m128i *y,
+ const __m128i *z) {
+ const __m128i one = _mm_set1_epi16(1);
+ const __m128i a = _mm_avg_epu16(*x, *z);
+ const __m128i b =
+ _mm_subs_epu16(a, _mm_and_si128(_mm_xor_si128(*x, *z), one));
+ return _mm_avg_epu16(b, *y);
+}
+
+DECLARE_ALIGNED(16, static const uint8_t, rotate_right_epu16[16]) = {
+ 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 0, 1
+};
+
+static INLINE __m128i rotr_epu16(__m128i *a, const __m128i *rotrw) {
+ *a = _mm_shuffle_epi8(*a, *rotrw);
+ return *a;
+}
+
+void aom_highbd_d117_predictor_8x8_ssse3(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {
+ const __m128i rotrw = _mm_load_si128((const __m128i *)rotate_right_epu16);
+ const __m128i XABCDEFG = _mm_loadu_si128((const __m128i *)(above - 1));
+ const __m128i ABCDEFGH = _mm_load_si128((const __m128i *)above);
+ const __m128i IJKLMNOP = _mm_load_si128((const __m128i *)left);
+ const __m128i IXABCDEF =
+ _mm_alignr_epi8(XABCDEFG, _mm_slli_si128(IJKLMNOP, 14), 14);
+ const __m128i avg3 = avg3_epu16(&ABCDEFGH, &XABCDEFG, &IXABCDEF);
+ const __m128i avg2 = _mm_avg_epu16(ABCDEFGH, XABCDEFG);
+ const __m128i XIJKLMNO =
+ _mm_alignr_epi8(IJKLMNOP, _mm_slli_si128(XABCDEFG, 14), 14);
+ const __m128i JKLMNOP0 = _mm_srli_si128(IJKLMNOP, 2);
+ __m128i avg3_left = avg3_epu16(&XIJKLMNO, &IJKLMNOP, &JKLMNOP0);
+ __m128i rowa = avg2;
+ __m128i rowb = avg3;
+ int i;
+ (void)bd;
+ for (i = 0; i < 8; i += 2) {
+ _mm_store_si128((__m128i *)dst, rowa);
+ dst += stride;
+ _mm_store_si128((__m128i *)dst, rowb);
+ dst += stride;
+ rowa = _mm_alignr_epi8(rowa, rotr_epu16(&avg3_left, &rotrw), 14);
+ rowb = _mm_alignr_epi8(rowb, rotr_epu16(&avg3_left, &rotrw), 14);
+ }
+}
+
+void aom_highbd_d117_predictor_16x16_ssse3(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {
+ const __m128i rotrw = _mm_load_si128((const __m128i *)rotate_right_epu16);
+ const __m128i B0 = _mm_loadu_si128((const __m128i *)(above - 1));
+ const __m128i A0 = _mm_load_si128((const __m128i *)above);
+ const __m128i B1 = _mm_loadu_si128((const __m128i *)(above + 7));
+ const __m128i A1 = _mm_load_si128((const __m128i *)(above + 8));
+ const __m128i avg2_0 = _mm_avg_epu16(A0, B0);
+ const __m128i avg2_1 = _mm_avg_epu16(A1, B1);
+ const __m128i L0 = _mm_load_si128((const __m128i *)left);
+ const __m128i L1 = _mm_load_si128((const __m128i *)(left + 8));
+ const __m128i C0 = _mm_alignr_epi8(B0, _mm_slli_si128(L0, 14), 14);
+ const __m128i C1 = _mm_alignr_epi8(B1, B0, 14);
+ const __m128i avg3_0 = avg3_epu16(&A0, &B0, &C0);
+ const __m128i avg3_1 = avg3_epu16(&A1, &B1, &C1);
+ const __m128i XL0 = _mm_alignr_epi8(L0, _mm_slli_si128(B0, 14), 14);
+ const __m128i XL1 = _mm_alignr_epi8(L1, L0, 14);
+ const __m128i L0_ = _mm_alignr_epi8(L1, L0, 2);
+ const __m128i L1_ = _mm_srli_si128(L1, 2);
+ __m128i rowa_0 = avg2_0;
+ __m128i rowa_1 = avg2_1;
+ __m128i rowb_0 = avg3_0;
+ __m128i rowb_1 = avg3_1;
+ __m128i avg3_left[2];
+ int i, j;
+ (void)bd;
+ avg3_left[0] = avg3_epu16(&XL0, &L0, &L0_);
+ avg3_left[1] = avg3_epu16(&XL1, &L1, &L1_);
+ for (i = 0; i < 2; ++i) {
+ __m128i avg_left = avg3_left[i];
+ for (j = 0; j < 8; j += 2) {
+ _mm_store_si128((__m128i *)dst, rowa_0);
+ _mm_store_si128((__m128i *)(dst + 8), rowa_1);
+ dst += stride;
+ _mm_store_si128((__m128i *)dst, rowb_0);
+ _mm_store_si128((__m128i *)(dst + 8), rowb_1);
+ dst += stride;
+ rowa_1 = _mm_alignr_epi8(rowa_1, rowa_0, 14);
+ rowa_0 = _mm_alignr_epi8(rowa_0, rotr_epu16(&avg_left, &rotrw), 14);
+ rowb_1 = _mm_alignr_epi8(rowb_1, rowb_0, 14);
+ rowb_0 = _mm_alignr_epi8(rowb_0, rotr_epu16(&avg_left, &rotrw), 14);
+ }
+ }
+}
+
+void aom_highbd_d117_predictor_32x32_ssse3(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {
+ const __m128i rotrw = _mm_load_si128((const __m128i *)rotate_right_epu16);
+ const __m128i A0 = _mm_load_si128((const __m128i *)above);
+ const __m128i A1 = _mm_load_si128((const __m128i *)(above + 8));
+ const __m128i A2 = _mm_load_si128((const __m128i *)(above + 16));
+ const __m128i A3 = _mm_load_si128((const __m128i *)(above + 24));
+ const __m128i B0 = _mm_loadu_si128((const __m128i *)(above - 1));
+ const __m128i B1 = _mm_loadu_si128((const __m128i *)(above + 7));
+ const __m128i B2 = _mm_loadu_si128((const __m128i *)(above + 15));
+ const __m128i B3 = _mm_loadu_si128((const __m128i *)(above + 23));
+ const __m128i avg2_0 = _mm_avg_epu16(A0, B0);
+ const __m128i avg2_1 = _mm_avg_epu16(A1, B1);
+ const __m128i avg2_2 = _mm_avg_epu16(A2, B2);
+ const __m128i avg2_3 = _mm_avg_epu16(A3, B3);
+ const __m128i L0 = _mm_load_si128((const __m128i *)left);
+ const __m128i L1 = _mm_load_si128((const __m128i *)(left + 8));
+ const __m128i L2 = _mm_load_si128((const __m128i *)(left + 16));
+ const __m128i L3 = _mm_load_si128((const __m128i *)(left + 24));
+ const __m128i C0 = _mm_alignr_epi8(B0, _mm_slli_si128(L0, 14), 14);
+ const __m128i C1 = _mm_alignr_epi8(B1, B0, 14);
+ const __m128i C2 = _mm_alignr_epi8(B2, B1, 14);
+ const __m128i C3 = _mm_alignr_epi8(B3, B2, 14);
+ const __m128i avg3_0 = avg3_epu16(&A0, &B0, &C0);
+ const __m128i avg3_1 = avg3_epu16(&A1, &B1, &C1);
+ const __m128i avg3_2 = avg3_epu16(&A2, &B2, &C2);
+ const __m128i avg3_3 = avg3_epu16(&A3, &B3, &C3);
+ const __m128i XL0 = _mm_alignr_epi8(L0, _mm_slli_si128(B0, 14), 14);
+ const __m128i XL1 = _mm_alignr_epi8(L1, L0, 14);
+ const __m128i XL2 = _mm_alignr_epi8(L2, L1, 14);
+ const __m128i XL3 = _mm_alignr_epi8(L3, L2, 14);
+ const __m128i L0_ = _mm_alignr_epi8(L1, L0, 2);
+ const __m128i L1_ = _mm_alignr_epi8(L2, L1, 2);
+ const __m128i L2_ = _mm_alignr_epi8(L3, L2, 2);
+ const __m128i L3_ = _mm_srli_si128(L3, 2);
+ __m128i rowa_0 = avg2_0;
+ __m128i rowa_1 = avg2_1;
+ __m128i rowa_2 = avg2_2;
+ __m128i rowa_3 = avg2_3;
+ __m128i rowb_0 = avg3_0;
+ __m128i rowb_1 = avg3_1;
+ __m128i rowb_2 = avg3_2;
+ __m128i rowb_3 = avg3_3;
+ __m128i avg3_left[4];
+ int i, j;
+ (void)bd;
+ avg3_left[0] = avg3_epu16(&XL0, &L0, &L0_);
+ avg3_left[1] = avg3_epu16(&XL1, &L1, &L1_);
+ avg3_left[2] = avg3_epu16(&XL2, &L2, &L2_);
+ avg3_left[3] = avg3_epu16(&XL3, &L3, &L3_);
+ for (i = 0; i < 4; ++i) {
+ __m128i avg_left = avg3_left[i];
+ for (j = 0; j < 8; j += 2) {
+ _mm_store_si128((__m128i *)dst, rowa_0);
+ _mm_store_si128((__m128i *)(dst + 8), rowa_1);
+ _mm_store_si128((__m128i *)(dst + 16), rowa_2);
+ _mm_store_si128((__m128i *)(dst + 24), rowa_3);
+ dst += stride;
+ _mm_store_si128((__m128i *)dst, rowb_0);
+ _mm_store_si128((__m128i *)(dst + 8), rowb_1);
+ _mm_store_si128((__m128i *)(dst + 16), rowb_2);
+ _mm_store_si128((__m128i *)(dst + 24), rowb_3);
+ dst += stride;
+ rowa_3 = _mm_alignr_epi8(rowa_3, rowa_2, 14);
+ rowa_2 = _mm_alignr_epi8(rowa_2, rowa_1, 14);
+ rowa_1 = _mm_alignr_epi8(rowa_1, rowa_0, 14);
+ rowa_0 = _mm_alignr_epi8(rowa_0, rotr_epu16(&avg_left, &rotrw), 14);
+ rowb_3 = _mm_alignr_epi8(rowb_3, rowb_2, 14);
+ rowb_2 = _mm_alignr_epi8(rowb_2, rowb_1, 14);
+ rowb_1 = _mm_alignr_epi8(rowb_1, rowb_0, 14);
+ rowb_0 = _mm_alignr_epi8(rowb_0, rotr_epu16(&avg_left, &rotrw), 14);
+ }
+ }
+}
+
+void aom_highbd_d135_predictor_8x8_ssse3(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {
+ const __m128i rotrw = _mm_load_si128((const __m128i *)rotate_right_epu16);
+ const __m128i XABCDEFG = _mm_loadu_si128((const __m128i *)(above - 1));
+ const __m128i ABCDEFGH = _mm_load_si128((const __m128i *)above);
+ const __m128i BCDEFGH0 = _mm_srli_si128(ABCDEFGH, 2);
+ const __m128i IJKLMNOP = _mm_load_si128((const __m128i *)left);
+ const __m128i XIJKLMNO =
+ _mm_alignr_epi8(IJKLMNOP, _mm_slli_si128(XABCDEFG, 14), 14);
+ const __m128i AXIJKLMN =
+ _mm_alignr_epi8(XIJKLMNO, _mm_slli_si128(ABCDEFGH, 14), 14);
+ const __m128i avg3 = avg3_epu16(&XABCDEFG, &ABCDEFGH, &BCDEFGH0);
+ __m128i avg3_left = avg3_epu16(&IJKLMNOP, &XIJKLMNO, &AXIJKLMN);
+ __m128i rowa = avg3;
+ int i;
+ (void)bd;
+ for (i = 0; i < 8; ++i) {
+ rowa = _mm_alignr_epi8(rowa, rotr_epu16(&avg3_left, &rotrw), 14);
+ _mm_store_si128((__m128i *)dst, rowa);
+ dst += stride;
+ }
+}
+
+void aom_highbd_d135_predictor_16x16_ssse3(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {
+ const __m128i rotrw = _mm_load_si128((const __m128i *)rotate_right_epu16);
+ const __m128i A0 = _mm_loadu_si128((const __m128i *)(above - 1));
+ const __m128i B0 = _mm_load_si128((const __m128i *)above);
+ const __m128i A1 = _mm_loadu_si128((const __m128i *)(above + 7));
+ const __m128i B1 = _mm_load_si128((const __m128i *)(above + 8));
+ const __m128i L0 = _mm_load_si128((const __m128i *)left);
+ const __m128i L1 = _mm_load_si128((const __m128i *)(left + 8));
+ const __m128i C0 = _mm_alignr_epi8(B1, B0, 2);
+ const __m128i C1 = _mm_srli_si128(B1, 2);
+ const __m128i avg3_0 = avg3_epu16(&A0, &B0, &C0);
+ const __m128i avg3_1 = avg3_epu16(&A1, &B1, &C1);
+ const __m128i XL0 = _mm_alignr_epi8(L0, _mm_slli_si128(A0, 14), 14);
+ const __m128i XL1 = _mm_alignr_epi8(L1, L0, 14);
+ const __m128i L0_ = _mm_alignr_epi8(XL0, _mm_slli_si128(B0, 14), 14);
+ const __m128i L1_ = _mm_alignr_epi8(XL1, XL0, 14);
+ __m128i rowa_0 = avg3_0;
+ __m128i rowa_1 = avg3_1;
+ __m128i avg3_left[2];
+ int i, j;
+ (void)bd;
+ avg3_left[0] = avg3_epu16(&L0, &XL0, &L0_);
+ avg3_left[1] = avg3_epu16(&L1, &XL1, &L1_);
+ for (i = 0; i < 2; ++i) {
+ __m128i avg_left = avg3_left[i];
+ for (j = 0; j < 8; ++j) {
+ rowa_1 = _mm_alignr_epi8(rowa_1, rowa_0, 14);
+ rowa_0 = _mm_alignr_epi8(rowa_0, rotr_epu16(&avg_left, &rotrw), 14);
+ _mm_store_si128((__m128i *)dst, rowa_0);
+ _mm_store_si128((__m128i *)(dst + 8), rowa_1);
+ dst += stride;
+ }
+ }
+}
+
+void aom_highbd_d135_predictor_32x32_ssse3(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {
+ const __m128i rotrw = _mm_load_si128((const __m128i *)rotate_right_epu16);
+ const __m128i A0 = _mm_loadu_si128((const __m128i *)(above - 1));
+ const __m128i A1 = _mm_loadu_si128((const __m128i *)(above + 7));
+ const __m128i A2 = _mm_loadu_si128((const __m128i *)(above + 15));
+ const __m128i A3 = _mm_loadu_si128((const __m128i *)(above + 23));
+ const __m128i B0 = _mm_load_si128((const __m128i *)above);
+ const __m128i B1 = _mm_load_si128((const __m128i *)(above + 8));
+ const __m128i B2 = _mm_load_si128((const __m128i *)(above + 16));
+ const __m128i B3 = _mm_load_si128((const __m128i *)(above + 24));
+ const __m128i L0 = _mm_load_si128((const __m128i *)left);
+ const __m128i L1 = _mm_load_si128((const __m128i *)(left + 8));
+ const __m128i L2 = _mm_load_si128((const __m128i *)(left + 16));
+ const __m128i L3 = _mm_load_si128((const __m128i *)(left + 24));
+ const __m128i C0 = _mm_alignr_epi8(B1, B0, 2);
+ const __m128i C1 = _mm_alignr_epi8(B2, B1, 2);
+ const __m128i C2 = _mm_alignr_epi8(B3, B2, 2);
+ const __m128i C3 = _mm_srli_si128(B3, 2);
+ const __m128i avg3_0 = avg3_epu16(&A0, &B0, &C0);
+ const __m128i avg3_1 = avg3_epu16(&A1, &B1, &C1);
+ const __m128i avg3_2 = avg3_epu16(&A2, &B2, &C2);
+ const __m128i avg3_3 = avg3_epu16(&A3, &B3, &C3);
+ const __m128i XL0 = _mm_alignr_epi8(L0, _mm_slli_si128(A0, 14), 14);
+ const __m128i XL1 = _mm_alignr_epi8(L1, L0, 14);
+ const __m128i XL2 = _mm_alignr_epi8(L2, L1, 14);
+ const __m128i XL3 = _mm_alignr_epi8(L3, L2, 14);
+ const __m128i L0_ = _mm_alignr_epi8(XL0, _mm_slli_si128(B0, 14), 14);
+ const __m128i L1_ = _mm_alignr_epi8(XL1, XL0, 14);
+ const __m128i L2_ = _mm_alignr_epi8(XL2, XL1, 14);
+ const __m128i L3_ = _mm_alignr_epi8(XL3, XL2, 14);
+ __m128i rowa_0 = avg3_0;
+ __m128i rowa_1 = avg3_1;
+ __m128i rowa_2 = avg3_2;
+ __m128i rowa_3 = avg3_3;
+ __m128i avg3_left[4];
+ int i, j;
+ (void)bd;
+ avg3_left[0] = avg3_epu16(&L0, &XL0, &L0_);
+ avg3_left[1] = avg3_epu16(&L1, &XL1, &L1_);
+ avg3_left[2] = avg3_epu16(&L2, &XL2, &L2_);
+ avg3_left[3] = avg3_epu16(&L3, &XL3, &L3_);
+ for (i = 0; i < 4; ++i) {
+ __m128i avg_left = avg3_left[i];
+ for (j = 0; j < 8; ++j) {
+ rowa_3 = _mm_alignr_epi8(rowa_3, rowa_2, 14);
+ rowa_2 = _mm_alignr_epi8(rowa_2, rowa_1, 14);
+ rowa_1 = _mm_alignr_epi8(rowa_1, rowa_0, 14);
+ rowa_0 = _mm_alignr_epi8(rowa_0, rotr_epu16(&avg_left, &rotrw), 14);
+ _mm_store_si128((__m128i *)dst, rowa_0);
+ _mm_store_si128((__m128i *)(dst + 8), rowa_1);
+ _mm_store_si128((__m128i *)(dst + 16), rowa_2);
+ _mm_store_si128((__m128i *)(dst + 24), rowa_3);
+ dst += stride;
+ }
+ }
+}
+
+void aom_highbd_d153_predictor_8x8_ssse3(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {
+ const __m128i XABCDEFG = _mm_loadu_si128((const __m128i *)(above - 1));
+ const __m128i ABCDEFG0 = _mm_srli_si128(XABCDEFG, 2);
+ const __m128i BCDEFG00 = _mm_srli_si128(XABCDEFG, 4);
+ const __m128i avg3 = avg3_epu16(&BCDEFG00, &ABCDEFG0, &XABCDEFG);
+ const __m128i IJKLMNOP = _mm_load_si128((const __m128i *)left);
+ const __m128i XIJKLMNO =
+ _mm_alignr_epi8(IJKLMNOP, _mm_slli_si128(XABCDEFG, 14), 14);
+ const __m128i AXIJKLMN =
+ _mm_alignr_epi8(XIJKLMNO, _mm_slli_si128(XABCDEFG, 12), 14);
+ const __m128i avg3_left = avg3_epu16(&IJKLMNOP, &XIJKLMNO, &AXIJKLMN);
+ const __m128i avg2_left = _mm_avg_epu16(IJKLMNOP, XIJKLMNO);
+ const __m128i avg2_avg3_lo = _mm_unpacklo_epi16(avg2_left, avg3_left);
+ const __m128i avg2_avg3_hi = _mm_unpackhi_epi16(avg2_left, avg3_left);
+ const __m128i row0 =
+ _mm_alignr_epi8(avg3, _mm_slli_si128(avg2_avg3_lo, 12), 12);
+ const __m128i row1 =
+ _mm_alignr_epi8(row0, _mm_slli_si128(avg2_avg3_lo, 8), 12);
+ const __m128i row2 =
+ _mm_alignr_epi8(row1, _mm_slli_si128(avg2_avg3_lo, 4), 12);
+ const __m128i row3 = _mm_alignr_epi8(row2, avg2_avg3_lo, 12);
+ const __m128i row4 =
+ _mm_alignr_epi8(row3, _mm_slli_si128(avg2_avg3_hi, 12), 12);
+ const __m128i row5 =
+ _mm_alignr_epi8(row4, _mm_slli_si128(avg2_avg3_hi, 8), 12);
+ const __m128i row6 =
+ _mm_alignr_epi8(row5, _mm_slli_si128(avg2_avg3_hi, 4), 12);
+ const __m128i row7 = _mm_alignr_epi8(row6, avg2_avg3_hi, 12);
+ (void)bd;
+ _mm_store_si128((__m128i *)dst, row0);
+ dst += stride;
+ _mm_store_si128((__m128i *)dst, row1);
+ dst += stride;
+ _mm_store_si128((__m128i *)dst, row2);
+ dst += stride;
+ _mm_store_si128((__m128i *)dst, row3);
+ dst += stride;
+ _mm_store_si128((__m128i *)dst, row4);
+ dst += stride;
+ _mm_store_si128((__m128i *)dst, row5);
+ dst += stride;
+ _mm_store_si128((__m128i *)dst, row6);
+ dst += stride;
+ _mm_store_si128((__m128i *)dst, row7);
+}
+
+void aom_highbd_d153_predictor_16x16_ssse3(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {
+ const __m128i A0 = _mm_loadu_si128((const __m128i *)(above - 1));
+ const __m128i A1 = _mm_loadu_si128((const __m128i *)(above + 7));
+ const __m128i B0 = _mm_alignr_epi8(A1, A0, 2);
+ const __m128i B1 = _mm_srli_si128(A1, 2);
+ const __m128i C0 = _mm_alignr_epi8(A1, A0, 4);
+ const __m128i C1 = _mm_srli_si128(A1, 4);
+ const __m128i avg3_0 = avg3_epu16(&A0, &B0, &C0);
+ const __m128i avg3_1 = avg3_epu16(&A1, &B1, &C1);
+ const __m128i L0 = _mm_load_si128((const __m128i *)left);
+ const __m128i L1 = _mm_load_si128((const __m128i *)(left + 8));
+ const __m128i XL0 = _mm_alignr_epi8(L0, _mm_slli_si128(A0, 14), 14);
+ const __m128i AXL0 = _mm_alignr_epi8(XL0, _mm_slli_si128(A0, 12), 14);
+ const __m128i XL1 = _mm_alignr_epi8(L1, L0, 14);
+ const __m128i AXL1 = _mm_alignr_epi8(L1, L0, 12);
+ const __m128i avg3_left_0 = avg3_epu16(&L0, &XL0, &AXL0);
+ const __m128i avg2_left_0 = _mm_avg_epu16(L0, XL0);
+ const __m128i avg3_left_1 = avg3_epu16(&L1, &XL1, &AXL1);
+ const __m128i avg2_left_1 = _mm_avg_epu16(L1, XL1);
+ __m128i row_0 = avg3_0;
+ __m128i row_1 = avg3_1;
+ __m128i avg2_avg3_left[2][2];
+ int i, j;
+ (void)bd;
+
+ avg2_avg3_left[0][0] = _mm_unpacklo_epi16(avg2_left_0, avg3_left_0);
+ avg2_avg3_left[0][1] = _mm_unpackhi_epi16(avg2_left_0, avg3_left_0);
+ avg2_avg3_left[1][0] = _mm_unpacklo_epi16(avg2_left_1, avg3_left_1);
+ avg2_avg3_left[1][1] = _mm_unpackhi_epi16(avg2_left_1, avg3_left_1);
+
+ for (j = 0; j < 2; ++j) {
+ for (i = 0; i < 2; ++i) {
+ const __m128i avg2_avg3 = avg2_avg3_left[j][i];
+ row_1 = _mm_alignr_epi8(row_1, row_0, 12);
+ row_0 = _mm_alignr_epi8(row_0, _mm_slli_si128(avg2_avg3, 12), 12);
+ _mm_store_si128((__m128i *)dst, row_0);
+ _mm_store_si128((__m128i *)(dst + 8), row_1);
+ dst += stride;
+ row_1 = _mm_alignr_epi8(row_1, row_0, 12);
+ row_0 = _mm_alignr_epi8(row_0, _mm_slli_si128(avg2_avg3, 8), 12);
+ _mm_store_si128((__m128i *)dst, row_0);
+ _mm_store_si128((__m128i *)(dst + 8), row_1);
+ dst += stride;
+ row_1 = _mm_alignr_epi8(row_1, row_0, 12);
+ row_0 = _mm_alignr_epi8(row_0, _mm_slli_si128(avg2_avg3, 4), 12);
+ _mm_store_si128((__m128i *)dst, row_0);
+ _mm_store_si128((__m128i *)(dst + 8), row_1);
+ dst += stride;
+ row_1 = _mm_alignr_epi8(row_1, row_0, 12);
+ row_0 = _mm_alignr_epi8(row_0, avg2_avg3, 12);
+ _mm_store_si128((__m128i *)dst, row_0);
+ _mm_store_si128((__m128i *)(dst + 8), row_1);
+ dst += stride;
+ }
+ }
+}
+
+void aom_highbd_d153_predictor_32x32_ssse3(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {
+ const __m128i A0 = _mm_loadu_si128((const __m128i *)(above - 1));
+ const __m128i A1 = _mm_loadu_si128((const __m128i *)(above + 7));
+ const __m128i A2 = _mm_loadu_si128((const __m128i *)(above + 15));
+ const __m128i A3 = _mm_loadu_si128((const __m128i *)(above + 23));
+ const __m128i B0 = _mm_alignr_epi8(A1, A0, 2);
+ const __m128i B1 = _mm_alignr_epi8(A2, A1, 2);
+ const __m128i B2 = _mm_alignr_epi8(A3, A2, 2);
+ const __m128i B3 = _mm_srli_si128(A3, 2);
+ const __m128i C0 = _mm_alignr_epi8(A1, A0, 4);
+ const __m128i C1 = _mm_alignr_epi8(A2, A1, 4);
+ const __m128i C2 = _mm_alignr_epi8(A3, A2, 4);
+ const __m128i C3 = _mm_srli_si128(A3, 4);
+ const __m128i avg3_0 = avg3_epu16(&A0, &B0, &C0);
+ const __m128i avg3_1 = avg3_epu16(&A1, &B1, &C1);
+ const __m128i avg3_2 = avg3_epu16(&A2, &B2, &C2);
+ const __m128i avg3_3 = avg3_epu16(&A3, &B3, &C3);
+ const __m128i L0 = _mm_load_si128((const __m128i *)left);
+ const __m128i L1 = _mm_load_si128((const __m128i *)(left + 8));
+ const __m128i L2 = _mm_load_si128((const __m128i *)(left + 16));
+ const __m128i L3 = _mm_load_si128((const __m128i *)(left + 24));
+ const __m128i XL0 = _mm_alignr_epi8(L0, _mm_slli_si128(A0, 14), 14);
+ const __m128i XL1 = _mm_alignr_epi8(L1, L0, 14);
+ const __m128i XL2 = _mm_alignr_epi8(L2, L1, 14);
+ const __m128i XL3 = _mm_alignr_epi8(L3, L2, 14);
+ const __m128i AXL0 = _mm_alignr_epi8(XL0, _mm_slli_si128(A0, 12), 14);
+ const __m128i AXL1 = _mm_alignr_epi8(L1, L0, 12);
+ const __m128i AXL2 = _mm_alignr_epi8(L2, L1, 12);
+ const __m128i AXL3 = _mm_alignr_epi8(L3, L2, 12);
+ const __m128i avg3_left_0 = avg3_epu16(&L0, &XL0, &AXL0);
+ const __m128i avg3_left_1 = avg3_epu16(&L1, &XL1, &AXL1);
+ const __m128i avg3_left_2 = avg3_epu16(&L2, &XL2, &AXL2);
+ const __m128i avg3_left_3 = avg3_epu16(&L3, &XL3, &AXL3);
+ const __m128i avg2_left_0 = _mm_avg_epu16(L0, XL0);
+ const __m128i avg2_left_1 = _mm_avg_epu16(L1, XL1);
+ const __m128i avg2_left_2 = _mm_avg_epu16(L2, XL2);
+ const __m128i avg2_left_3 = _mm_avg_epu16(L3, XL3);
+ __m128i row_0 = avg3_0;
+ __m128i row_1 = avg3_1;
+ __m128i row_2 = avg3_2;
+ __m128i row_3 = avg3_3;
+ __m128i avg2_avg3_left[4][2];
+ int i, j;
+ (void)bd;
+
+ avg2_avg3_left[0][0] = _mm_unpacklo_epi16(avg2_left_0, avg3_left_0);
+ avg2_avg3_left[0][1] = _mm_unpackhi_epi16(avg2_left_0, avg3_left_0);
+ avg2_avg3_left[1][0] = _mm_unpacklo_epi16(avg2_left_1, avg3_left_1);
+ avg2_avg3_left[1][1] = _mm_unpackhi_epi16(avg2_left_1, avg3_left_1);
+ avg2_avg3_left[2][0] = _mm_unpacklo_epi16(avg2_left_2, avg3_left_2);
+ avg2_avg3_left[2][1] = _mm_unpackhi_epi16(avg2_left_2, avg3_left_2);
+ avg2_avg3_left[3][0] = _mm_unpacklo_epi16(avg2_left_3, avg3_left_3);
+ avg2_avg3_left[3][1] = _mm_unpackhi_epi16(avg2_left_3, avg3_left_3);
+
+ for (j = 0; j < 4; ++j) {
+ for (i = 0; i < 2; ++i) {
+ const __m128i avg2_avg3 = avg2_avg3_left[j][i];
+ row_3 = _mm_alignr_epi8(row_3, row_2, 12);
+ row_2 = _mm_alignr_epi8(row_2, row_1, 12);
+ row_1 = _mm_alignr_epi8(row_1, row_0, 12);
+ row_0 = _mm_alignr_epi8(row_0, _mm_slli_si128(avg2_avg3, 12), 12);
+ _mm_store_si128((__m128i *)dst, row_0);
+ _mm_store_si128((__m128i *)(dst + 8), row_1);
+ _mm_store_si128((__m128i *)(dst + 16), row_2);
+ _mm_store_si128((__m128i *)(dst + 24), row_3);
+ dst += stride;
+ row_3 = _mm_alignr_epi8(row_3, row_2, 12);
+ row_2 = _mm_alignr_epi8(row_2, row_1, 12);
+ row_1 = _mm_alignr_epi8(row_1, row_0, 12);
+ row_0 = _mm_alignr_epi8(row_0, _mm_slli_si128(avg2_avg3, 8), 12);
+ _mm_store_si128((__m128i *)dst, row_0);
+ _mm_store_si128((__m128i *)(dst + 8), row_1);
+ _mm_store_si128((__m128i *)(dst + 16), row_2);
+ _mm_store_si128((__m128i *)(dst + 24), row_3);
+ dst += stride;
+ row_3 = _mm_alignr_epi8(row_3, row_2, 12);
+ row_2 = _mm_alignr_epi8(row_2, row_1, 12);
+ row_1 = _mm_alignr_epi8(row_1, row_0, 12);
+ row_0 = _mm_alignr_epi8(row_0, _mm_slli_si128(avg2_avg3, 4), 12);
+ _mm_store_si128((__m128i *)dst, row_0);
+ _mm_store_si128((__m128i *)(dst + 8), row_1);
+ _mm_store_si128((__m128i *)(dst + 16), row_2);
+ _mm_store_si128((__m128i *)(dst + 24), row_3);
+ dst += stride;
+ row_3 = _mm_alignr_epi8(row_3, row_2, 12);
+ row_2 = _mm_alignr_epi8(row_2, row_1, 12);
+ row_1 = _mm_alignr_epi8(row_1, row_0, 12);
+ row_0 = _mm_alignr_epi8(row_0, avg2_avg3, 12);
+ _mm_store_si128((__m128i *)dst, row_0);
+ _mm_store_si128((__m128i *)(dst + 8), row_1);
+ _mm_store_si128((__m128i *)(dst + 16), row_2);
+ _mm_store_si128((__m128i *)(dst + 24), row_3);
+ dst += stride;
+ }
+ }
+}
diff --git a/test/intrapred_test.cc b/test/intrapred_test.cc
index 2bb446a..307b289 100644
--- a/test/intrapred_test.cc
+++ b/test/intrapred_test.cc
@@ -171,12 +171,14 @@
highbd_entry(type, 16, 32, opt, bd), \
highbd_entry(type, 32, 16, opt, bd), highbd_entry(type, 32, 32, opt, bd)
-#if HAVE_SSE2
#if CONFIG_HIGHBITDEPTH
+#if HAVE_SSE2
const IntraPredFunc<HighbdIntraPred> IntraPredTestVector8[] = {
highbd_intrapred(dc, sse2, 8), highbd_intrapred(dc_left, sse2, 8),
highbd_intrapred(dc_top, sse2, 8), highbd_intrapred(dc_128, sse2, 8),
highbd_intrapred(h, sse2, 8), highbd_intrapred(v, sse2, 8),
+ highbd_entry(d117, 4, 4, sse2, 8), highbd_entry(d135, 4, 4, sse2, 8),
+ highbd_entry(d153, 4, 4, sse2, 8),
};
INSTANTIATE_TEST_CASE_P(SSE2_TO_C_8, HighbdIntraPredTest,
@@ -186,6 +188,8 @@
highbd_intrapred(dc, sse2, 10), highbd_intrapred(dc_left, sse2, 10),
highbd_intrapred(dc_top, sse2, 10), highbd_intrapred(dc_128, sse2, 10),
highbd_intrapred(h, sse2, 10), highbd_intrapred(v, sse2, 10),
+ highbd_entry(d117, 4, 4, sse2, 10), highbd_entry(d135, 4, 4, sse2, 10),
+ highbd_entry(d153, 4, 4, sse2, 10),
};
INSTANTIATE_TEST_CASE_P(SSE2_TO_C_10, HighbdIntraPredTest,
@@ -195,14 +199,48 @@
highbd_intrapred(dc, sse2, 12), highbd_intrapred(dc_left, sse2, 12),
highbd_intrapred(dc_top, sse2, 12), highbd_intrapred(dc_128, sse2, 12),
highbd_intrapred(h, sse2, 12), highbd_intrapred(v, sse2, 12),
+ highbd_entry(d117, 4, 4, sse2, 12), highbd_entry(d135, 4, 4, sse2, 12),
+ highbd_entry(d153, 4, 4, sse2, 12),
};
INSTANTIATE_TEST_CASE_P(SSE2_TO_C_12, HighbdIntraPredTest,
::testing::ValuesIn(IntraPredTestVector12));
-#endif // CONFIG_HIGHBITDEPTH
#endif // HAVE_SSE2
+#if HAVE_SSSE3
+const IntraPredFunc<HighbdIntraPred> IntraPredTestVectorSsse3_8[] = {
+ highbd_entry(d117, 8, 8, ssse3, 8), highbd_entry(d117, 16, 16, ssse3, 8),
+ highbd_entry(d117, 32, 32, ssse3, 8), highbd_entry(d135, 8, 8, ssse3, 8),
+ highbd_entry(d135, 16, 16, ssse3, 8), highbd_entry(d135, 32, 32, ssse3, 8),
+ highbd_entry(d153, 8, 8, ssse3, 8), highbd_entry(d153, 16, 16, ssse3, 8),
+ highbd_entry(d153, 32, 32, ssse3, 8),
+};
+INSTANTIATE_TEST_CASE_P(SSSE3_TO_C_8, HighbdIntraPredTest,
+ ::testing::ValuesIn(IntraPredTestVectorSsse3_8));
+
+const IntraPredFunc<HighbdIntraPred> IntraPredTestVectorSsse3_10[] = {
+ highbd_entry(d117, 8, 8, ssse3, 10), highbd_entry(d117, 16, 16, ssse3, 10),
+ highbd_entry(d117, 32, 32, ssse3, 10), highbd_entry(d135, 8, 8, ssse3, 10),
+ highbd_entry(d135, 16, 16, ssse3, 10), highbd_entry(d135, 32, 32, ssse3, 10),
+ highbd_entry(d153, 8, 8, ssse3, 10), highbd_entry(d153, 16, 16, ssse3, 10),
+ highbd_entry(d153, 32, 32, ssse3, 10),
+};
+INSTANTIATE_TEST_CASE_P(SSSE3_TO_C_10, HighbdIntraPredTest,
+ ::testing::ValuesIn(IntraPredTestVectorSsse3_10));
+
+const IntraPredFunc<HighbdIntraPred> IntraPredTestVectorSsse3_12[] = {
+ highbd_entry(d117, 8, 8, ssse3, 12), highbd_entry(d117, 16, 16, ssse3, 12),
+ highbd_entry(d117, 32, 32, ssse3, 12), highbd_entry(d135, 8, 8, ssse3, 12),
+ highbd_entry(d135, 16, 16, ssse3, 12), highbd_entry(d135, 32, 32, ssse3, 12),
+ highbd_entry(d153, 8, 8, ssse3, 12), highbd_entry(d153, 16, 16, ssse3, 12),
+ highbd_entry(d153, 32, 32, ssse3, 12),
+};
+INSTANTIATE_TEST_CASE_P(SSSE3_TO_C_12, HighbdIntraPredTest,
+ ::testing::ValuesIn(IntraPredTestVectorSsse3_12));
+#endif // HAVE_SSSE3
+#endif // CONFIG_HIGHBITDEPTH
+
#define lowbd_entry(type, width, height, opt) \
IntraPredFunc<IntraPred>(&aom_##type##_predictor_##width##x##height##_##opt, \
&aom_##type##_predictor_##width##x##height##_c, \
diff --git a/test/test_intra_pred_speed.cc b/test/test_intra_pred_speed.cc
index 54c451e..9e9c975 100644
--- a/test/test_intra_pred_speed.cc
+++ b/test/test_intra_pred_speed.cc
@@ -1141,14 +1141,13 @@
#undef smooth_h_pred_func
#if HAVE_SSE2
-HIGHBD_INTRA_PRED_TEST(SSE2_1, TestHighbdIntraPred4, "Hbd Intra4x4",
- aom_highbd_dc_predictor_4x4_sse2,
- aom_highbd_dc_left_predictor_4x4_sse2,
- aom_highbd_dc_top_predictor_4x4_sse2,
- aom_highbd_dc_128_predictor_4x4_sse2,
- aom_highbd_v_predictor_4x4_sse2,
- aom_highbd_h_predictor_4x4_sse2, NULL, NULL, NULL, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL)
+HIGHBD_INTRA_PRED_TEST(
+ SSE2_1, TestHighbdIntraPred4, "Hbd Intra4x4",
+ aom_highbd_dc_predictor_4x4_sse2, aom_highbd_dc_left_predictor_4x4_sse2,
+ aom_highbd_dc_top_predictor_4x4_sse2, aom_highbd_dc_128_predictor_4x4_sse2,
+ aom_highbd_v_predictor_4x4_sse2, aom_highbd_h_predictor_4x4_sse2, NULL,
+ aom_highbd_d135_predictor_4x4_sse2, aom_highbd_d117_predictor_4x4_sse2,
+ aom_highbd_d153_predictor_4x4_sse2, NULL, NULL, NULL, NULL, NULL, NULL)
HIGHBD_INTRA_PRED_TEST(SSE2_2, TestHighbdIntraPred4, "Hbd Intra4x8",
aom_highbd_dc_predictor_4x8_sse2,
aom_highbd_dc_left_predictor_4x8_sse2,
@@ -1229,9 +1228,18 @@
NULL, NULL, NULL, NULL, NULL, NULL)
#endif
+#if HAVE_SSSE3
+HIGHBD_INTRA_PRED_TEST(SSSE3, TestHighbdIntraPred8, "Hbd Intra8x8", NULL, NULL,
+ NULL, NULL, NULL, NULL, NULL,
+ aom_highbd_d135_predictor_8x8_ssse3,
+ aom_highbd_d117_predictor_8x8_ssse3,
+ aom_highbd_d153_predictor_8x8_ssse3, NULL, NULL, NULL,
+ NULL, NULL, NULL)
+#endif
+
#if CONFIG_SMOOTH_HV
-#define smooth_v_pred_func aom_highbd_smooth_v_predictor_8x8_c
-#define smooth_h_pred_func aom_highbd_smooth_h_predictor_8x8_c
+#define smooth_v_pred_func aom_highbd_smooth_v_predictor_8x4_c
+#define smooth_h_pred_func aom_highbd_smooth_h_predictor_8x4_c
#else
#define smooth_v_pred_func NULL
#define smooth_h_pred_func NULL
@@ -1320,6 +1328,15 @@
NULL, NULL, NULL, NULL, NULL, NULL, NULL)
#endif
+#if HAVE_SSSE3
+HIGHBD_INTRA_PRED_TEST(SSSE3, TestHighbdIntraPred16, "Hbd Intra16x16", NULL,
+ NULL, NULL, NULL, NULL, NULL, NULL,
+ aom_highbd_d135_predictor_16x16_ssse3,
+ aom_highbd_d117_predictor_16x16_ssse3,
+ aom_highbd_d153_predictor_16x16_ssse3, NULL, NULL, NULL,
+ NULL, NULL, NULL)
+#endif
+
#if CONFIG_SMOOTH_HV
#define smooth_v_pred_func aom_highbd_smooth_v_predictor_16x8_c
#define smooth_h_pred_func aom_highbd_smooth_h_predictor_16x8_c
@@ -1404,6 +1421,15 @@
NULL, NULL, NULL, NULL, NULL, NULL, NULL)
#endif
+#if HAVE_SSSE3
+HIGHBD_INTRA_PRED_TEST(SSSE3, TestHighbdIntraPred32, "Hbd Intra32x32", NULL,
+ NULL, NULL, NULL, NULL, NULL, NULL,
+ aom_highbd_d135_predictor_32x32_ssse3,
+ aom_highbd_d117_predictor_32x32_ssse3,
+ aom_highbd_d153_predictor_32x32_ssse3, NULL, NULL, NULL,
+ NULL, NULL, NULL)
+#endif
+
#if CONFIG_SMOOTH_HV
#define smooth_v_pred_func aom_highbd_smooth_v_predictor_32x16_c
#define smooth_h_pred_func aom_highbd_smooth_h_predictor_32x16_c
