SSE optimization of motion compensation functions
SSE4_1 optimizations of av1_highbd_jnt_convolve_2d_copy_c,
av1_highbd_jnt_convolve_y_c and av1_highbd_jnt_convolve_x_c
have been added.
av1_highbd_jnt_convolve_x_sse4_1: ~8.5 times faster than its
c implementation.
av1_highbd_jnt_convolve_y_sse4_1: ~9.4 times faster than its
c implementation.
av1_highbd_jnt_convolve_2d_copy_sse4_1: ~2.6 times faster than
its c implementation.
code cleanup for highbd_convolve_rounding_sse2 and
highbd_comp_avg_sse4_1.
Change-Id: I71290d1a04e0c0654c2008689ce85a2f1700685e
diff --git a/aom_dsp/x86/convolve_sse2.h b/aom_dsp/x86/convolve_sse2.h
index a11ea4f..9790580 100644
--- a/aom_dsp/x86/convolve_sse2.h
+++ b/aom_dsp/x86/convolve_sse2.h
@@ -108,6 +108,17 @@
_mm_srai_epi16(_mm_add_epi16(res_signed, *round_const), round_shift);
return res_round;
}
+
+static INLINE __m128i highbd_convolve_rounding_sse2(
+ const __m128i *const res_unsigned, const __m128i *const offset_const,
+ const __m128i *const round_const, const int round_shift) {
+ const __m128i res_signed = _mm_sub_epi32(*res_unsigned, *offset_const);
+ const __m128i res_round =
+ _mm_srai_epi32(_mm_add_epi32(res_signed, *round_const), round_shift);
+
+ return res_round;
+}
+
#endif
#endif
diff --git a/aom_dsp/x86/convolve_sse4_1.h b/aom_dsp/x86/convolve_sse4_1.h
index 6ec9ba3..4b5a9b8 100644
--- a/aom_dsp/x86/convolve_sse4_1.h
+++ b/aom_dsp/x86/convolve_sse4_1.h
@@ -31,4 +31,25 @@
_mm_store_si128((__m128i *)dst, d);
}
+#if CONFIG_LOWPRECISION_BLEND
+static INLINE __m128i highbd_comp_avg_sse4_1(const __m128i *const data_ref_0,
+ const __m128i *const res_unsigned,
+ const __m128i *const wt0,
+ const __m128i *const wt1,
+ const int use_jnt_comp_avg) {
+ __m128i res;
+ if (use_jnt_comp_avg) {
+ const __m128i wt0_res = _mm_mullo_epi32(*data_ref_0, *wt0);
+ const __m128i wt1_res = _mm_mullo_epi32(*res_unsigned, *wt1);
+
+ const __m128i wt_res = _mm_add_epi32(wt0_res, wt1_res);
+ res = _mm_srai_epi32(wt_res, DIST_PRECISION_BITS);
+ } else {
+ const __m128i wt_res = _mm_add_epi32(*data_ref_0, *res_unsigned);
+ res = _mm_srai_epi32(wt_res, 1);
+ }
+ return res;
+}
+#endif
+
#endif // _AOM_DSP_X86_TXFM_COMMON_INTRIN_H_
diff --git a/av1/av1.cmake b/av1/av1.cmake
index 9f2191b..2c492b2 100644
--- a/av1/av1.cmake
+++ b/av1/av1.cmake
@@ -282,6 +282,10 @@
${AOM_AV1_COMMON_INTRIN_AVX2}
"${AOM_ROOT}/av1/common/x86/highbd_jnt_convolve_avx2.c")
+set(AOM_AV1_COMMON_INTRIN_SSE4_1
+ ${AOM_AV1_COMMON_INTRIN_SSE4_1}
+ "${AOM_ROOT}/av1/common/x86/highbd_jnt_convolve_sse4.c")
+
if (CONFIG_LOWPRECISION_BLEND)
set(AOM_AV1_COMMON_INTRIN_SSE2
${AOM_AV1_COMMON_INTRIN_SSE2}
diff --git a/av1/common/av1_rtcd_defs.pl b/av1/common/av1_rtcd_defs.pl
index 67a215c..1960169 100755
--- a/av1/common/av1_rtcd_defs.pl
+++ b/av1/common/av1_rtcd_defs.pl
@@ -325,9 +325,9 @@
specialize qw/av1_highbd_convolve_y_sr ssse3 avx2/;
specialize qw/av1_highbd_convolve_2d_scale sse4_1/;
specialize qw/av1_highbd_jnt_convolve_2d sse4_1 avx2/;
- specialize qw/av1_highbd_jnt_convolve_x avx2/;
- specialize qw/av1_highbd_jnt_convolve_y avx2/;
- specialize qw/av1_highbd_jnt_convolve_2d_copy avx2/;
+ specialize qw/av1_highbd_jnt_convolve_x sse4_1 avx2/;
+ specialize qw/av1_highbd_jnt_convolve_y sse4_1 avx2/;
+ specialize qw/av1_highbd_jnt_convolve_2d_copy sse4_1 avx2/;
}
else
{
diff --git a/av1/common/x86/highbd_convolve_2d_sse4.c b/av1/common/x86/highbd_convolve_2d_sse4.c
index 5214656..047f214 100644
--- a/av1/common/x86/highbd_convolve_2d_sse4.c
+++ b/av1/common/x86/highbd_convolve_2d_sse4.c
@@ -17,36 +17,154 @@
#include "aom_dsp/aom_convolve.h"
#include "aom_dsp/aom_dsp_common.h"
#include "aom_dsp/aom_filter.h"
+#include "aom_dsp/x86/convolve_sse2.h"
+#include "aom_dsp/x86/convolve_sse4_1.h"
#include "av1/common/convolve.h"
#if CONFIG_LOWPRECISION_BLEND
-static INLINE __m128i highbd_comp_avg(const __m128i *const data_ref_0,
- const __m128i *const res_unsigned,
- const __m128i *const wt0,
- const __m128i *const wt1,
- const int use_jnt_comp_avg) {
- __m128i res;
- if (use_jnt_comp_avg) {
- const __m128i wt0_res = _mm_mullo_epi32(*data_ref_0, *wt0);
- const __m128i wt1_res = _mm_mullo_epi32(*res_unsigned, *wt1);
+void av1_highbd_jnt_convolve_2d_copy_sse4_1(
+ const uint16_t *src, int src_stride, uint16_t *dst0, int dst_stride0, int w,
+ int h, InterpFilterParams *filter_params_x,
+ InterpFilterParams *filter_params_y, const int subpel_x_q4,
+ const int subpel_y_q4, ConvolveParams *conv_params, int bd) {
+ CONV_BUF_TYPE *dst = conv_params->dst;
+ int dst_stride = conv_params->dst_stride;
+ (void)filter_params_x;
+ (void)filter_params_y;
+ (void)subpel_x_q4;
+ (void)subpel_y_q4;
- const __m128i wt_res = _mm_add_epi32(wt0_res, wt1_res);
- res = _mm_srai_epi32(wt_res, DIST_PRECISION_BITS);
- } else {
- const __m128i wt_res = _mm_add_epi32(*data_ref_0, *res_unsigned);
- res = _mm_srai_epi32(wt_res, 1);
+ const int bits =
+ FILTER_BITS * 2 - conv_params->round_1 - conv_params->round_0;
+ const __m128i left_shift = _mm_cvtsi32_si128(bits);
+ const int do_average = conv_params->do_average;
+ const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+ const int w0 = conv_params->fwd_offset;
+ const int w1 = conv_params->bck_offset;
+ const __m128i wt0 = _mm_set1_epi32(w0);
+ const __m128i wt1 = _mm_set1_epi32(w1);
+ const __m128i zero = _mm_setzero_si128();
+ int i, j;
+
+ const int offset_0 =
+ bd + 2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
+ const int offset = (1 << offset_0) + (1 << (offset_0 - 1));
+ const __m128i offset_const = _mm_set1_epi32(offset);
+ const __m128i offset_const_16b = _mm_set1_epi16(offset);
+ const int rounding_shift =
+ 2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
+ const __m128i rounding_const = _mm_set1_epi32((1 << rounding_shift) >> 1);
+ const __m128i clip_pixel_to_bd =
+ _mm_set1_epi16(bd == 10 ? 1023 : (bd == 12 ? 4095 : 255));
+
+ assert(bits <= 4);
+
+ if (!(w % 8)) {
+ for (i = 0; i < h; i += 1) {
+ for (j = 0; j < w; j += 8) {
+ const __m128i src_16bit =
+ _mm_loadu_si128((__m128i *)(&src[i * src_stride + j]));
+ const __m128i res = _mm_sll_epi16(src_16bit, left_shift);
+ if (do_average) {
+ const __m128i data_0 =
+ _mm_loadu_si128((__m128i *)(&dst[i * dst_stride + j]));
+
+ const __m128i data_ref_0_lo = _mm_unpacklo_epi16(data_0, zero);
+ const __m128i data_ref_0_hi = _mm_unpackhi_epi16(data_0, zero);
+
+ const __m128i res_32b_lo = _mm_unpacklo_epi16(res, zero);
+ const __m128i res_unsigned_lo =
+ _mm_add_epi32(res_32b_lo, offset_const);
+
+ const __m128i comp_avg_res_lo = highbd_comp_avg_sse4_1(
+ &data_ref_0_lo, &res_unsigned_lo, &wt0, &wt1, use_jnt_comp_avg);
+
+ const __m128i res_32b_hi = _mm_unpackhi_epi16(res, zero);
+ const __m128i res_unsigned_hi =
+ _mm_add_epi32(res_32b_hi, offset_const);
+
+ const __m128i comp_avg_res_hi = highbd_comp_avg_sse4_1(
+ &data_ref_0_hi, &res_unsigned_hi, &wt0, &wt1, use_jnt_comp_avg);
+
+ const __m128i round_result_lo = highbd_convolve_rounding_sse2(
+ &comp_avg_res_lo, &offset_const, &rounding_const, rounding_shift);
+ const __m128i round_result_hi = highbd_convolve_rounding_sse2(
+ &comp_avg_res_hi, &offset_const, &rounding_const, rounding_shift);
+
+ const __m128i res_16b =
+ _mm_packus_epi32(round_result_lo, round_result_hi);
+ const __m128i res_clip = _mm_min_epi16(res_16b, clip_pixel_to_bd);
+
+ _mm_store_si128((__m128i *)(&dst0[i * dst_stride0 + j]), res_clip);
+ } else {
+ const __m128i res_unsigned_16b =
+ _mm_adds_epu16(res, offset_const_16b);
+
+ _mm_store_si128((__m128i *)(&dst[i * dst_stride + j]),
+ res_unsigned_16b);
+ }
+ }
+ }
+ } else if (!(w % 4)) {
+ for (i = 0; i < h; i += 2) {
+ for (j = 0; j < w; j += 4) {
+ const __m128i src_row_0 =
+ _mm_loadl_epi64((__m128i *)(&src[i * src_stride + j]));
+ const __m128i src_row_1 =
+ _mm_loadl_epi64((__m128i *)(&src[i * src_stride + j + src_stride]));
+ const __m128i src_10 = _mm_unpacklo_epi64(src_row_0, src_row_1);
+
+ const __m128i res = _mm_sll_epi16(src_10, left_shift);
+
+ if (do_average) {
+ const __m128i data_0 =
+ _mm_loadl_epi64((__m128i *)(&dst[i * dst_stride + j]));
+ const __m128i data_1 = _mm_loadl_epi64(
+ (__m128i *)(&dst[i * dst_stride + j + dst_stride]));
+
+ const __m128i data_ref_0 = _mm_unpacklo_epi16(data_0, zero);
+ const __m128i data_ref_1 = _mm_unpacklo_epi16(data_1, zero);
+
+ const __m128i res_32b = _mm_unpacklo_epi16(res, zero);
+ const __m128i res_unsigned_lo = _mm_add_epi32(res_32b, offset_const);
+
+ const __m128i res_32b_hi = _mm_unpackhi_epi16(res, zero);
+ const __m128i res_unsigned_hi =
+ _mm_add_epi32(res_32b_hi, offset_const);
+
+ const __m128i comp_avg_res_lo = highbd_comp_avg_sse4_1(
+ &data_ref_0, &res_unsigned_lo, &wt0, &wt1, use_jnt_comp_avg);
+ const __m128i comp_avg_res_hi = highbd_comp_avg_sse4_1(
+ &data_ref_1, &res_unsigned_hi, &wt0, &wt1, use_jnt_comp_avg);
+
+ const __m128i round_result_lo = highbd_convolve_rounding_sse2(
+ &comp_avg_res_lo, &offset_const, &rounding_const, rounding_shift);
+ const __m128i round_result_hi = highbd_convolve_rounding_sse2(
+ &comp_avg_res_hi, &offset_const, &rounding_const, rounding_shift);
+
+ const __m128i res_16b =
+ _mm_packus_epi32(round_result_lo, round_result_hi);
+ const __m128i res_clip = _mm_min_epi16(res_16b, clip_pixel_to_bd);
+
+ const __m128i res_1 = _mm_srli_si128(res_clip, 8);
+
+ _mm_storel_epi64((__m128i *)(&dst0[i * dst_stride0 + j]), res_clip);
+ _mm_storel_epi64(
+ (__m128i *)(&dst0[i * dst_stride0 + j + dst_stride0]), res_1);
+ } else {
+ const __m128i res_unsigned_16b =
+ _mm_adds_epu16(res, offset_const_16b);
+
+ const __m128i res_1 = _mm_srli_si128(res_unsigned_16b, 8);
+
+ _mm_storel_epi64((__m128i *)(&dst[i * dst_stride + j]),
+ res_unsigned_16b);
+ _mm_storel_epi64((__m128i *)(&dst[i * dst_stride + j + dst_stride]),
+ res_1);
+ }
+ }
+ }
}
- return res;
-}
-
-static INLINE __m128i highbd_convolve_rounding(
- const __m128i *const res_unsigned, const __m128i *const offset_const,
- const __m128i *const round_const, const int round_shift) {
- const __m128i res_signed = _mm_sub_epi32(*res_unsigned, *offset_const);
- const __m128i res_round =
- _mm_srai_epi32(_mm_add_epi32(res_signed, *round_const), round_shift);
-
- return res_round;
}
void av1_highbd_jnt_convolve_2d_sse4_1(
@@ -242,10 +360,10 @@
const __m128i data_ref_0 = _mm_cvtepu16_epi32(data_0);
- const __m128i comp_avg_res = highbd_comp_avg(
+ const __m128i comp_avg_res = highbd_comp_avg_sse4_1(
&data_ref_0, &res_unsigned_lo, &wt0, &wt1, use_jnt_comp_avg);
- const __m128i round_result = highbd_convolve_rounding(
+ const __m128i round_result = highbd_convolve_rounding_sse2(
&comp_avg_res, &offset_const, &rounding_const, rounding_shift);
const __m128i res_16b =
@@ -274,17 +392,17 @@
const __m128i data_ref_0_lo = _mm_cvtepu16_epi32(data_lo);
const __m128i data_ref_0_hi = _mm_cvtepu16_epi32(data_hi);
- const __m128i comp_avg_res_lo = highbd_comp_avg(
+ const __m128i comp_avg_res_lo = highbd_comp_avg_sse4_1(
&data_ref_0_lo, &res_unsigned_lo, &wt0, &wt1, use_jnt_comp_avg);
- const __m128i comp_avg_res_hi = highbd_comp_avg(
+ const __m128i comp_avg_res_hi = highbd_comp_avg_sse4_1(
&data_ref_0_hi, &res_unsigned_hi, &wt0, &wt1, use_jnt_comp_avg);
const __m128i round_result_lo =
- highbd_convolve_rounding(&comp_avg_res_lo, &offset_const,
- &rounding_const, rounding_shift);
+ highbd_convolve_rounding_sse2(&comp_avg_res_lo, &offset_const,
+ &rounding_const, rounding_shift);
const __m128i round_result_hi =
- highbd_convolve_rounding(&comp_avg_res_hi, &offset_const,
- &rounding_const, rounding_shift);
+ highbd_convolve_rounding_sse2(&comp_avg_res_hi, &offset_const,
+ &rounding_const, rounding_shift);
const __m128i res_16b =
_mm_packus_epi32(round_result_lo, round_result_hi);
diff --git a/av1/common/x86/highbd_jnt_convolve_sse4.c b/av1/common/x86/highbd_jnt_convolve_sse4.c
new file mode 100644
index 0000000..680bebd
--- /dev/null
+++ b/av1/common/x86/highbd_jnt_convolve_sse4.c
@@ -0,0 +1,384 @@
+/*
+ * Copyright (c) 2018, 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 <smmintrin.h>
+#include <assert.h>
+
+#include "./aom_dsp_rtcd.h"
+#include "aom_dsp/x86/convolve_sse2.h"
+#include "aom_dsp/x86/convolve_sse4_1.h"
+
+#if CONFIG_LOWPRECISION_BLEND
+void av1_highbd_jnt_convolve_y_sse4_1(
+ const uint16_t *src, int src_stride, uint16_t *dst0, int dst_stride0, int w,
+ int h, InterpFilterParams *filter_params_x,
+ InterpFilterParams *filter_params_y, const int subpel_x_q4,
+ const int subpel_y_q4, ConvolveParams *conv_params, int bd) {
+ CONV_BUF_TYPE *dst = conv_params->dst;
+ int dst_stride = conv_params->dst_stride;
+ const int fo_vert = filter_params_y->taps / 2 - 1;
+ const uint16_t *const src_ptr = src - fo_vert * src_stride;
+ const int bits = FILTER_BITS - conv_params->round_0;
+ (void)filter_params_x;
+ (void)subpel_x_q4;
+
+ assert(bits >= 0);
+ int i, j;
+ const int do_average = conv_params->do_average;
+ const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+
+ const int w0 = conv_params->fwd_offset;
+ const int w1 = conv_params->bck_offset;
+ const __m128i wt0 = _mm_set1_epi32(w0);
+ const __m128i wt1 = _mm_set1_epi32(w1);
+ const __m128i round_const_y =
+ _mm_set1_epi32(((1 << conv_params->round_1) >> 1));
+ const __m128i round_shift_y = _mm_cvtsi32_si128(conv_params->round_1);
+ const __m128i round_shift_bits = _mm_cvtsi32_si128(bits);
+
+ const int offset_0 =
+ bd + 2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
+ const int offset = (1 << offset_0) + (1 << (offset_0 - 1));
+ const __m128i offset_const = _mm_set1_epi32(offset);
+ const int rounding_shift =
+ 2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
+ const __m128i rounding_const = _mm_set1_epi32((1 << rounding_shift) >> 1);
+ const __m128i clip_pixel_to_bd =
+ _mm_set1_epi16(bd == 10 ? 1023 : (bd == 12 ? 4095 : 255));
+ const __m128i zero = _mm_setzero_si128();
+ __m128i s[16], coeffs_y[4];
+
+ prepare_coeffs(filter_params_y, subpel_y_q4, coeffs_y);
+
+ for (j = 0; j < w; j += 8) {
+ const uint16_t *data = &src_ptr[j];
+ /* Vertical filter */
+ {
+ __m128i s0 = _mm_loadu_si128((__m128i *)(data + 0 * src_stride));
+ __m128i s1 = _mm_loadu_si128((__m128i *)(data + 1 * src_stride));
+ __m128i s2 = _mm_loadu_si128((__m128i *)(data + 2 * src_stride));
+ __m128i s3 = _mm_loadu_si128((__m128i *)(data + 3 * src_stride));
+ __m128i s4 = _mm_loadu_si128((__m128i *)(data + 4 * src_stride));
+ __m128i s5 = _mm_loadu_si128((__m128i *)(data + 5 * src_stride));
+ __m128i s6 = _mm_loadu_si128((__m128i *)(data + 6 * src_stride));
+
+ s[0] = _mm_unpacklo_epi16(s0, s1);
+ s[1] = _mm_unpacklo_epi16(s2, s3);
+ s[2] = _mm_unpacklo_epi16(s4, s5);
+
+ s[4] = _mm_unpackhi_epi16(s0, s1);
+ s[5] = _mm_unpackhi_epi16(s2, s3);
+ s[6] = _mm_unpackhi_epi16(s4, s5);
+
+ s[0 + 8] = _mm_unpacklo_epi16(s1, s2);
+ s[1 + 8] = _mm_unpacklo_epi16(s3, s4);
+ s[2 + 8] = _mm_unpacklo_epi16(s5, s6);
+
+ s[4 + 8] = _mm_unpackhi_epi16(s1, s2);
+ s[5 + 8] = _mm_unpackhi_epi16(s3, s4);
+ s[6 + 8] = _mm_unpackhi_epi16(s5, s6);
+
+ for (i = 0; i < h; i += 2) {
+ data = &src_ptr[i * src_stride + j];
+
+ __m128i s7 = _mm_loadu_si128((__m128i *)(data + 7 * src_stride));
+ __m128i s8 = _mm_loadu_si128((__m128i *)(data + 8 * src_stride));
+
+ s[3] = _mm_unpacklo_epi16(s6, s7);
+ s[7] = _mm_unpackhi_epi16(s6, s7);
+
+ s[3 + 8] = _mm_unpacklo_epi16(s7, s8);
+ s[7 + 8] = _mm_unpackhi_epi16(s7, s8);
+
+ const __m128i res_a0 = convolve(s, coeffs_y);
+ __m128i res_a_round0 = _mm_sll_epi32(res_a0, round_shift_bits);
+ res_a_round0 = _mm_sra_epi32(_mm_add_epi32(res_a_round0, round_const_y),
+ round_shift_y);
+
+ const __m128i res_a1 = convolve(s + 8, coeffs_y);
+ __m128i res_a_round1 = _mm_sll_epi32(res_a1, round_shift_bits);
+ res_a_round1 = _mm_sra_epi32(_mm_add_epi32(res_a_round1, round_const_y),
+ round_shift_y);
+
+ __m128i res_unsigned_lo_0 = _mm_add_epi32(res_a_round0, offset_const);
+ __m128i res_unsigned_lo_1 = _mm_add_epi32(res_a_round1, offset_const);
+
+ if (w - j < 8) {
+ if (do_average) {
+ const __m128i data_0 =
+ _mm_loadl_epi64((__m128i *)(&dst[i * dst_stride + j]));
+ const __m128i data_1 = _mm_loadl_epi64(
+ (__m128i *)(&dst[i * dst_stride + j + dst_stride]));
+
+ const __m128i data_ref_0 = _mm_unpacklo_epi16(data_0, zero);
+ const __m128i data_ref_1 = _mm_unpacklo_epi16(data_1, zero);
+
+ const __m128i comp_avg_res_0 = highbd_comp_avg_sse4_1(
+ &data_ref_0, &res_unsigned_lo_0, &wt0, &wt1, use_jnt_comp_avg);
+ const __m128i comp_avg_res_1 = highbd_comp_avg_sse4_1(
+ &data_ref_1, &res_unsigned_lo_1, &wt0, &wt1, use_jnt_comp_avg);
+
+ const __m128i round_result_0 =
+ highbd_convolve_rounding_sse2(&comp_avg_res_0, &offset_const,
+ &rounding_const, rounding_shift);
+ const __m128i round_result_1 =
+ highbd_convolve_rounding_sse2(&comp_avg_res_1, &offset_const,
+ &rounding_const, rounding_shift);
+
+ const __m128i res_16b_0 =
+ _mm_packus_epi32(round_result_0, round_result_0);
+ const __m128i res_clip_0 =
+ _mm_min_epi16(res_16b_0, clip_pixel_to_bd);
+ const __m128i res_16b_1 =
+ _mm_packus_epi32(round_result_1, round_result_1);
+ const __m128i res_clip_1 =
+ _mm_min_epi16(res_16b_1, clip_pixel_to_bd);
+
+ _mm_storel_epi64((__m128i *)(&dst0[i * dst_stride0 + j]),
+ res_clip_0);
+ _mm_storel_epi64(
+ (__m128i *)(&dst0[i * dst_stride0 + j + dst_stride0]),
+ res_clip_1);
+
+ } else {
+ __m128i res_16b_0 =
+ _mm_packus_epi32(res_unsigned_lo_0, res_unsigned_lo_0);
+
+ __m128i res_16b_1 =
+ _mm_packus_epi32(res_unsigned_lo_1, res_unsigned_lo_1);
+
+ _mm_storel_epi64((__m128i *)&dst[i * dst_stride + j], res_16b_0);
+ _mm_storel_epi64((__m128i *)&dst[i * dst_stride + j + dst_stride],
+ res_16b_1);
+ }
+ } else {
+ const __m128i res_b0 = convolve(s + 4, coeffs_y);
+ __m128i res_b_round0 = _mm_sll_epi32(res_b0, round_shift_bits);
+ res_b_round0 = _mm_sra_epi32(
+ _mm_add_epi32(res_b_round0, round_const_y), round_shift_y);
+
+ const __m128i res_b1 = convolve(s + 4 + 8, coeffs_y);
+ __m128i res_b_round1 = _mm_sll_epi32(res_b1, round_shift_bits);
+ res_b_round1 = _mm_sra_epi32(
+ _mm_add_epi32(res_b_round1, round_const_y), round_shift_y);
+
+ __m128i res_unsigned_hi_0 = _mm_add_epi32(res_b_round0, offset_const);
+ __m128i res_unsigned_hi_1 = _mm_add_epi32(res_b_round1, offset_const);
+
+ if (do_average) {
+ const __m128i data_0 =
+ _mm_loadu_si128((__m128i *)(&dst[i * dst_stride + j]));
+ const __m128i data_1 = _mm_loadu_si128(
+ (__m128i *)(&dst[i * dst_stride + j + dst_stride]));
+ const __m128i data_ref_0_lo_0 = _mm_unpacklo_epi16(data_0, zero);
+ const __m128i data_ref_0_lo_1 = _mm_unpacklo_epi16(data_1, zero);
+
+ const __m128i data_ref_0_hi_0 = _mm_unpackhi_epi16(data_0, zero);
+ const __m128i data_ref_0_hi_1 = _mm_unpackhi_epi16(data_1, zero);
+
+ const __m128i comp_avg_res_lo_0 =
+ highbd_comp_avg_sse4_1(&data_ref_0_lo_0, &res_unsigned_lo_0,
+ &wt0, &wt1, use_jnt_comp_avg);
+ const __m128i comp_avg_res_lo_1 =
+ highbd_comp_avg_sse4_1(&data_ref_0_lo_1, &res_unsigned_lo_1,
+ &wt0, &wt1, use_jnt_comp_avg);
+ const __m128i comp_avg_res_hi_0 =
+ highbd_comp_avg_sse4_1(&data_ref_0_hi_0, &res_unsigned_hi_0,
+ &wt0, &wt1, use_jnt_comp_avg);
+ const __m128i comp_avg_res_hi_1 =
+ highbd_comp_avg_sse4_1(&data_ref_0_hi_1, &res_unsigned_hi_1,
+ &wt0, &wt1, use_jnt_comp_avg);
+
+ const __m128i round_result_lo_0 =
+ highbd_convolve_rounding_sse2(&comp_avg_res_lo_0, &offset_const,
+ &rounding_const, rounding_shift);
+ const __m128i round_result_lo_1 =
+ highbd_convolve_rounding_sse2(&comp_avg_res_lo_1, &offset_const,
+ &rounding_const, rounding_shift);
+ const __m128i round_result_hi_0 =
+ highbd_convolve_rounding_sse2(&comp_avg_res_hi_0, &offset_const,
+ &rounding_const, rounding_shift);
+ const __m128i round_result_hi_1 =
+ highbd_convolve_rounding_sse2(&comp_avg_res_hi_1, &offset_const,
+ &rounding_const, rounding_shift);
+
+ const __m128i res_16b_0 =
+ _mm_packus_epi32(round_result_lo_0, round_result_hi_0);
+ const __m128i res_clip_0 =
+ _mm_min_epi16(res_16b_0, clip_pixel_to_bd);
+
+ const __m128i res_16b_1 =
+ _mm_packus_epi32(round_result_lo_1, round_result_hi_1);
+ const __m128i res_clip_1 =
+ _mm_min_epi16(res_16b_1, clip_pixel_to_bd);
+
+ _mm_store_si128((__m128i *)(&dst0[i * dst_stride0 + j]),
+ res_clip_0);
+ _mm_store_si128(
+ (__m128i *)(&dst0[i * dst_stride0 + j + dst_stride0]),
+ res_clip_1);
+ } else {
+ __m128i res_16bit0 =
+ _mm_packus_epi32(res_unsigned_lo_0, res_unsigned_hi_0);
+ __m128i res_16bit1 =
+ _mm_packus_epi32(res_unsigned_lo_1, res_unsigned_hi_1);
+ _mm_store_si128((__m128i *)(&dst[i * dst_stride + j]), res_16bit0);
+ _mm_store_si128((__m128i *)(&dst[i * dst_stride + j + dst_stride]),
+ res_16bit1);
+ }
+ }
+ s[0] = s[1];
+ s[1] = s[2];
+ s[2] = s[3];
+
+ s[4] = s[5];
+ s[5] = s[6];
+ s[6] = s[7];
+
+ s[0 + 8] = s[1 + 8];
+ s[1 + 8] = s[2 + 8];
+ s[2 + 8] = s[3 + 8];
+
+ s[4 + 8] = s[5 + 8];
+ s[5 + 8] = s[6 + 8];
+ s[6 + 8] = s[7 + 8];
+
+ s6 = s8;
+ }
+ }
+ }
+}
+
+void av1_highbd_jnt_convolve_x_sse4_1(
+ const uint16_t *src, int src_stride, uint16_t *dst0, int dst_stride0, int w,
+ int h, InterpFilterParams *filter_params_x,
+ InterpFilterParams *filter_params_y, const int subpel_x_q4,
+ const int subpel_y_q4, ConvolveParams *conv_params, int bd) {
+ CONV_BUF_TYPE *dst = conv_params->dst;
+ int dst_stride = conv_params->dst_stride;
+ const int fo_horiz = filter_params_x->taps / 2 - 1;
+ const uint16_t *const src_ptr = src - fo_horiz;
+ const int bits = FILTER_BITS - conv_params->round_1;
+ (void)filter_params_y;
+ (void)subpel_y_q4;
+
+ int i, j;
+ __m128i s[4], coeffs_x[4];
+
+ const int do_average = conv_params->do_average;
+ const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+ const int w0 = conv_params->fwd_offset;
+ const int w1 = conv_params->bck_offset;
+ const __m128i wt0 = _mm_set1_epi32(w0);
+ const __m128i wt1 = _mm_set1_epi32(w1);
+ const __m128i zero = _mm_setzero_si128();
+
+ const __m128i round_const_x =
+ _mm_set1_epi32(((1 << conv_params->round_0) >> 1));
+ const __m128i round_shift_x = _mm_cvtsi32_si128(conv_params->round_0);
+ const __m128i round_shift_bits = _mm_cvtsi32_si128(bits);
+
+ const int offset_0 =
+ bd + 2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
+ const int offset = (1 << offset_0) + (1 << (offset_0 - 1));
+ const __m128i offset_const = _mm_set1_epi32(offset);
+ const int rounding_shift =
+ 2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
+ const __m128i rounding_const = _mm_set1_epi32((1 << rounding_shift) >> 1);
+ const __m128i clip_pixel_to_bd =
+ _mm_set1_epi16(bd == 10 ? 1023 : (bd == 12 ? 4095 : 255));
+
+ assert(bits >= 0);
+ prepare_coeffs(filter_params_x, subpel_x_q4, coeffs_x);
+
+ for (j = 0; j < w; j += 8) {
+ /* Horizontal filter */
+ for (i = 0; i < h; i += 1) {
+ const __m128i row00 =
+ _mm_loadu_si128((__m128i *)&src_ptr[i * src_stride + j]);
+ const __m128i row01 =
+ _mm_loadu_si128((__m128i *)&src_ptr[i * src_stride + (j + 8)]);
+
+ // even pixels
+ s[0] = _mm_alignr_epi8(row01, row00, 0);
+ s[1] = _mm_alignr_epi8(row01, row00, 4);
+ s[2] = _mm_alignr_epi8(row01, row00, 8);
+ s[3] = _mm_alignr_epi8(row01, row00, 12);
+
+ __m128i res_even = convolve(s, coeffs_x);
+ res_even =
+ _mm_sra_epi32(_mm_add_epi32(res_even, round_const_x), round_shift_x);
+
+ // odd pixels
+ s[0] = _mm_alignr_epi8(row01, row00, 2);
+ s[1] = _mm_alignr_epi8(row01, row00, 6);
+ s[2] = _mm_alignr_epi8(row01, row00, 10);
+ s[3] = _mm_alignr_epi8(row01, row00, 14);
+
+ __m128i res_odd = convolve(s, coeffs_x);
+ res_odd =
+ _mm_sra_epi32(_mm_add_epi32(res_odd, round_const_x), round_shift_x);
+
+ res_even = _mm_sll_epi32(res_even, round_shift_bits);
+ res_odd = _mm_sll_epi32(res_odd, round_shift_bits);
+
+ __m128i res1 = _mm_unpacklo_epi32(res_even, res_odd);
+ __m128i res_unsigned_lo = _mm_add_epi32(res1, offset_const);
+ if (w - j < 8) {
+ if (do_average) {
+ const __m128i data_0 =
+ _mm_loadl_epi64((__m128i *)(&dst[i * dst_stride + j]));
+ const __m128i data_ref_0 = _mm_unpacklo_epi16(data_0, zero);
+
+ const __m128i comp_avg_res = highbd_comp_avg_sse4_1(
+ &data_ref_0, &res_unsigned_lo, &wt0, &wt1, use_jnt_comp_avg);
+ const __m128i round_result = highbd_convolve_rounding_sse2(
+ &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
+
+ const __m128i res_16b = _mm_packus_epi32(round_result, round_result);
+ const __m128i res_clip = _mm_min_epi16(res_16b, clip_pixel_to_bd);
+ _mm_storel_epi64((__m128i *)(&dst0[i * dst_stride0 + j]), res_clip);
+ } else {
+ __m128i res_16b = _mm_packus_epi32(res_unsigned_lo, res_unsigned_lo);
+ _mm_storel_epi64((__m128i *)&dst[i * dst_stride + j], res_16b);
+ }
+ } else {
+ __m128i res2 = _mm_unpackhi_epi32(res_even, res_odd);
+ __m128i res_unsigned_hi = _mm_add_epi32(res2, offset_const);
+ if (do_average) {
+ const __m128i data_0 =
+ _mm_loadu_si128((__m128i *)(&dst[i * dst_stride + j]));
+ const __m128i data_ref_0_lo = _mm_unpacklo_epi16(data_0, zero);
+ const __m128i data_ref_0_hi = _mm_unpackhi_epi16(data_0, zero);
+
+ const __m128i comp_avg_res_lo = highbd_comp_avg_sse4_1(
+ &data_ref_0_lo, &res_unsigned_lo, &wt0, &wt1, use_jnt_comp_avg);
+ const __m128i comp_avg_res_hi = highbd_comp_avg_sse4_1(
+ &data_ref_0_hi, &res_unsigned_hi, &wt0, &wt1, use_jnt_comp_avg);
+
+ const __m128i round_result_lo = highbd_convolve_rounding_sse2(
+ &comp_avg_res_lo, &offset_const, &rounding_const, rounding_shift);
+ const __m128i round_result_hi = highbd_convolve_rounding_sse2(
+ &comp_avg_res_hi, &offset_const, &rounding_const, rounding_shift);
+
+ const __m128i res_16b =
+ _mm_packus_epi32(round_result_lo, round_result_hi);
+ const __m128i res_clip = _mm_min_epi16(res_16b, clip_pixel_to_bd);
+ _mm_store_si128((__m128i *)(&dst0[i * dst_stride0 + j]), res_clip);
+ } else {
+ __m128i res_16b = _mm_packus_epi32(res_unsigned_lo, res_unsigned_hi);
+ _mm_store_si128((__m128i *)(&dst[i * dst_stride + j]), res_16b);
+ }
+ }
+ }
+ }
+}
+#endif
diff --git a/test/av1_convolve_2d_test.cc b/test/av1_convolve_2d_test.cc
index deb05d6..2e38b07 100644
--- a/test/av1_convolve_2d_test.cc
+++ b/test/av1_convolve_2d_test.cc
@@ -226,9 +226,18 @@
av1_highbd_jnt_convolve_2d_copy_c, 0, 0));
#if CONFIG_LOWPRECISION_BLEND
#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(SSE4_1_COPY, AV1HighbdJntConvolve2DTest,
+ libaom_test::AV1HighbdConvolve2D::BuildParams(
+ av1_highbd_jnt_convolve_2d_copy_sse4_1, 0, 0));
INSTANTIATE_TEST_CASE_P(SSE4_1, AV1HighbdJntConvolve2DTest,
libaom_test::AV1HighbdConvolve2D::BuildParams(
av1_highbd_jnt_convolve_2d_sse4_1, 1, 1));
+INSTANTIATE_TEST_CASE_P(SSE4_1_X, AV1HighbdJntConvolve2DTest,
+ libaom_test::AV1HighbdConvolve2D::BuildParams(
+ av1_highbd_jnt_convolve_x_sse4_1, 1, 0));
+INSTANTIATE_TEST_CASE_P(SSE4_1_Y, AV1HighbdJntConvolve2DTest,
+ libaom_test::AV1HighbdConvolve2D::BuildParams(
+ av1_highbd_jnt_convolve_y_sse4_1, 0, 1));
#if HAVE_AVX2
INSTANTIATE_TEST_CASE_P(AVX2_COPY, AV1HighbdJntConvolve2DTest,
libaom_test::AV1HighbdConvolve2D::BuildParams(
