/src/aom/aom_dsp/x86/convolve_sse2.h

Source (jump to first uncovered line)
/*
 * 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.
 */

#ifndef AOM_AOM_DSP_X86_CONVOLVE_SSE2_H_
#define AOM_AOM_DSP_X86_CONVOLVE_SSE2_H_

#include "config/aom_scale_rtcd.h"

// Note:
//  This header file should be put below any x86 intrinsics head file
static inline void prepare_coeffs(const InterpFilterParams *const filter_params,
                                  const int subpel_q4,
                                  __m128i *const coeffs /* [4] */) {
  const int16_t *filter = av1_get_interp_filter_subpel_kernel(
      filter_params, subpel_q4 & SUBPEL_MASK);
  const __m128i coeff = _mm_loadu_si128((__m128i *)filter);

  // coeffs 0 1 0 1 0 1 0 1
  coeffs[0] = _mm_shuffle_epi32(coeff, 0x00);
  // coeffs 2 3 2 3 2 3 2 3
  coeffs[1] = _mm_shuffle_epi32(coeff, 0x55);
  // coeffs 4 5 4 5 4 5 4 5
  coeffs[2] = _mm_shuffle_epi32(coeff, 0xaa);
  // coeffs 6 7 6 7 6 7 6 7
  coeffs[3] = _mm_shuffle_epi32(coeff, 0xff);
}

static inline __m128i convolve(const __m128i *const s,
                               const __m128i *const coeffs) {
  const __m128i res_0 = _mm_madd_epi16(s[0], coeffs[0]);
  const __m128i res_1 = _mm_madd_epi16(s[1], coeffs[1]);
  const __m128i res_2 = _mm_madd_epi16(s[2], coeffs[2]);
  const __m128i res_3 = _mm_madd_epi16(s[3], coeffs[3]);

  const __m128i res =
      _mm_add_epi32(_mm_add_epi32(res_0, res_1), _mm_add_epi32(res_2, res_3));

  return res;
}

static inline __m128i convolve_lo_x(const __m128i *const s,
                                    const __m128i *const coeffs) {
  __m128i ss[4];
  ss[0] = _mm_unpacklo_epi8(s[0], _mm_setzero_si128());
  ss[1] = _mm_unpacklo_epi8(s[1], _mm_setzero_si128());
  ss[2] = _mm_unpacklo_epi8(s[2], _mm_setzero_si128());
  ss[3] = _mm_unpacklo_epi8(s[3], _mm_setzero_si128());
  return convolve(ss, coeffs);
}

static inline __m128i convolve_lo_y(const __m128i *const s,
                                    const __m128i *const coeffs) {
  __m128i ss[4];
  ss[0] = _mm_unpacklo_epi8(s[0], _mm_setzero_si128());
  ss[1] = _mm_unpacklo_epi8(s[2], _mm_setzero_si128());
  ss[2] = _mm_unpacklo_epi8(s[4], _mm_setzero_si128());
  ss[3] = _mm_unpacklo_epi8(s[6], _mm_setzero_si128());
  return convolve(ss, coeffs);
}

static inline __m128i convolve_hi_y(const __m128i *const s,
                                    const __m128i *const coeffs) {
  __m128i ss[4];
  ss[0] = _mm_unpackhi_epi8(s[0], _mm_setzero_si128());
  ss[1] = _mm_unpackhi_epi8(s[2], _mm_setzero_si128());
  ss[2] = _mm_unpackhi_epi8(s[4], _mm_setzero_si128());
  ss[3] = _mm_unpackhi_epi8(s[6], _mm_setzero_si128());
  return convolve(ss, coeffs);
}

static inline __m128i comp_avg(const __m128i *const data_ref_0,
                               const __m128i *const res_unsigned,
                               const __m128i *const wt,
                               const int use_dist_wtd_avg) {
  __m128i res;
  if (use_dist_wtd_avg) {
    const __m128i data_lo = _mm_unpacklo_epi16(*data_ref_0, *res_unsigned);
    const __m128i data_hi = _mm_unpackhi_epi16(*data_ref_0, *res_unsigned);

    const __m128i wt_res_lo = _mm_madd_epi16(data_lo, *wt);
    const __m128i wt_res_hi = _mm_madd_epi16(data_hi, *wt);

    const __m128i res_lo = _mm_srai_epi32(wt_res_lo, DIST_PRECISION_BITS);
    const __m128i res_hi = _mm_srai_epi32(wt_res_hi, DIST_PRECISION_BITS);

    res = _mm_packs_epi32(res_lo, res_hi);
  } else {
    const __m128i wt_res = _mm_add_epi16(*data_ref_0, *res_unsigned);
    res = _mm_srai_epi16(wt_res, 1);
  }
  return res;
}

static inline __m128i 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_epi16(*res_unsigned, *offset_const);
  const __m128i res_round =
      _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  // AOM_AOM_DSP_X86_CONVOLVE_SSE2_H_

Line	Count	Source (jump to first uncovered line)
1		/*
2		* Copyright (c) 2018, Alliance for Open Media. All rights reserved.
3		*
4		* This source code is subject to the terms of the BSD 2 Clause License and
5		* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6		* was not distributed with this source code in the LICENSE file, you can
7		* obtain it at www.aomedia.org/license/software. If the Alliance for Open
8		* Media Patent License 1.0 was not distributed with this source code in the
9		* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10		*/
11
12		#ifndef AOM_AOM_DSP_X86_CONVOLVE_SSE2_H_
13		#define AOM_AOM_DSP_X86_CONVOLVE_SSE2_H_
14
15		#include "config/aom_scale_rtcd.h"
16
17		// Note:
18		// This header file should be put below any x86 intrinsics head file
19		static inline void prepare_coeffs(const InterpFilterParams *const filter_params,
20		const int subpel_q4,
21	0	__m128i const coeffs / [4] */) {
22	0	const int16_t *filter = av1_get_interp_filter_subpel_kernel(
23	0	filter_params, subpel_q4 & SUBPEL_MASK);
24	0	const __m128i coeff = _mm_loadu_si128((__m128i *)filter);
25
26		// coeffs 0 1 0 1 0 1 0 1
27	0	coeffs[0] = _mm_shuffle_epi32(coeff, 0x00);
28		// coeffs 2 3 2 3 2 3 2 3
29	0	coeffs[1] = _mm_shuffle_epi32(coeff, 0x55);
30		// coeffs 4 5 4 5 4 5 4 5
31	0	coeffs[2] = _mm_shuffle_epi32(coeff, 0xaa);
32		// coeffs 6 7 6 7 6 7 6 7
33	0	coeffs[3] = _mm_shuffle_epi32(coeff, 0xff);
34	0	} Unexecuted instantiation: aom_subpixel_8t_intrin_ssse3.c:prepare_coeffs Unexecuted instantiation: highbd_convolve_ssse3.c:prepare_coeffs Unexecuted instantiation: convolve_2d_sse2.c:prepare_coeffs Unexecuted instantiation: jnt_convolve_sse2.c:prepare_coeffs Unexecuted instantiation: jnt_convolve_ssse3.c:prepare_coeffs Unexecuted instantiation: resize_ssse3.c:prepare_coeffs Unexecuted instantiation: highbd_convolve_2d_ssse3.c:prepare_coeffs Unexecuted instantiation: highbd_convolve_2d_sse4.c:prepare_coeffs Unexecuted instantiation: highbd_jnt_convolve_sse4.c:prepare_coeffs
35
36		static inline __m128i convolve(const __m128i *const s,
37	0	const __m128i *const coeffs) {
38	0	const __m128i res_0 = _mm_madd_epi16(s[0], coeffs[0]);
39	0	const __m128i res_1 = _mm_madd_epi16(s[1], coeffs[1]);
40	0	const __m128i res_2 = _mm_madd_epi16(s[2], coeffs[2]);
41	0	const __m128i res_3 = _mm_madd_epi16(s[3], coeffs[3]);
42
43	0	const __m128i res =
44	0	_mm_add_epi32(_mm_add_epi32(res_0, res_1), _mm_add_epi32(res_2, res_3));
45
46	0	return res;
47	0	} Unexecuted instantiation: aom_subpixel_8t_intrin_ssse3.c:convolve Unexecuted instantiation: highbd_convolve_ssse3.c:convolve Unexecuted instantiation: convolve_2d_sse2.c:convolve Unexecuted instantiation: jnt_convolve_sse2.c:convolve Unexecuted instantiation: jnt_convolve_ssse3.c:convolve Unexecuted instantiation: resize_ssse3.c:convolve Unexecuted instantiation: highbd_convolve_2d_ssse3.c:convolve Unexecuted instantiation: highbd_convolve_2d_sse4.c:convolve Unexecuted instantiation: highbd_jnt_convolve_sse4.c:convolve
48
49		static inline __m128i convolve_lo_x(const __m128i *const s,
50	0	const __m128i *const coeffs) {
51	0	__m128i ss[4];
52	0	ss[0] = _mm_unpacklo_epi8(s[0], _mm_setzero_si128());
53	0	ss[1] = _mm_unpacklo_epi8(s[1], _mm_setzero_si128());
54	0	ss[2] = _mm_unpacklo_epi8(s[2], _mm_setzero_si128());
55	0	ss[3] = _mm_unpacklo_epi8(s[3], _mm_setzero_si128());
56	0	return convolve(ss, coeffs);
57	0	} Unexecuted instantiation: aom_subpixel_8t_intrin_ssse3.c:convolve_lo_x Unexecuted instantiation: highbd_convolve_ssse3.c:convolve_lo_x Unexecuted instantiation: convolve_2d_sse2.c:convolve_lo_x Unexecuted instantiation: jnt_convolve_sse2.c:convolve_lo_x Unexecuted instantiation: jnt_convolve_ssse3.c:convolve_lo_x Unexecuted instantiation: resize_ssse3.c:convolve_lo_x Unexecuted instantiation: highbd_convolve_2d_ssse3.c:convolve_lo_x Unexecuted instantiation: highbd_convolve_2d_sse4.c:convolve_lo_x Unexecuted instantiation: highbd_jnt_convolve_sse4.c:convolve_lo_x
58
59		static inline __m128i convolve_lo_y(const __m128i *const s,
60	0	const __m128i *const coeffs) {
61	0	__m128i ss[4];
62	0	ss[0] = _mm_unpacklo_epi8(s[0], _mm_setzero_si128());
63	0	ss[1] = _mm_unpacklo_epi8(s[2], _mm_setzero_si128());
64	0	ss[2] = _mm_unpacklo_epi8(s[4], _mm_setzero_si128());
65	0	ss[3] = _mm_unpacklo_epi8(s[6], _mm_setzero_si128());
66	0	return convolve(ss, coeffs);
67	0	} Unexecuted instantiation: aom_subpixel_8t_intrin_ssse3.c:convolve_lo_y Unexecuted instantiation: highbd_convolve_ssse3.c:convolve_lo_y Unexecuted instantiation: convolve_2d_sse2.c:convolve_lo_y Unexecuted instantiation: jnt_convolve_sse2.c:convolve_lo_y Unexecuted instantiation: jnt_convolve_ssse3.c:convolve_lo_y Unexecuted instantiation: resize_ssse3.c:convolve_lo_y Unexecuted instantiation: highbd_convolve_2d_ssse3.c:convolve_lo_y Unexecuted instantiation: highbd_convolve_2d_sse4.c:convolve_lo_y Unexecuted instantiation: highbd_jnt_convolve_sse4.c:convolve_lo_y
68
69		static inline __m128i convolve_hi_y(const __m128i *const s,
70	0	const __m128i *const coeffs) {
71	0	__m128i ss[4];
72	0	ss[0] = _mm_unpackhi_epi8(s[0], _mm_setzero_si128());
73	0	ss[1] = _mm_unpackhi_epi8(s[2], _mm_setzero_si128());
74	0	ss[2] = _mm_unpackhi_epi8(s[4], _mm_setzero_si128());
75	0	ss[3] = _mm_unpackhi_epi8(s[6], _mm_setzero_si128());
76	0	return convolve(ss, coeffs);
77	0	} Unexecuted instantiation: aom_subpixel_8t_intrin_ssse3.c:convolve_hi_y Unexecuted instantiation: highbd_convolve_ssse3.c:convolve_hi_y Unexecuted instantiation: convolve_2d_sse2.c:convolve_hi_y Unexecuted instantiation: jnt_convolve_sse2.c:convolve_hi_y Unexecuted instantiation: jnt_convolve_ssse3.c:convolve_hi_y Unexecuted instantiation: resize_ssse3.c:convolve_hi_y Unexecuted instantiation: highbd_convolve_2d_ssse3.c:convolve_hi_y Unexecuted instantiation: highbd_convolve_2d_sse4.c:convolve_hi_y Unexecuted instantiation: highbd_jnt_convolve_sse4.c:convolve_hi_y
78
79		static inline __m128i comp_avg(const __m128i *const data_ref_0,
80		const __m128i *const res_unsigned,
81		const __m128i *const wt,
82	0	const int use_dist_wtd_avg) {
83	0	__m128i res;
84	0	if (use_dist_wtd_avg) {
85	0	const __m128i data_lo = _mm_unpacklo_epi16(data_ref_0, res_unsigned);
86	0	const __m128i data_hi = _mm_unpackhi_epi16(data_ref_0, res_unsigned);
87
88	0	const __m128i wt_res_lo = _mm_madd_epi16(data_lo, *wt);
89	0	const __m128i wt_res_hi = _mm_madd_epi16(data_hi, *wt);
90
91	0	const __m128i res_lo = _mm_srai_epi32(wt_res_lo, DIST_PRECISION_BITS);
92	0	const __m128i res_hi = _mm_srai_epi32(wt_res_hi, DIST_PRECISION_BITS);
93
94	0	res = _mm_packs_epi32(res_lo, res_hi);
95	0	} else {
96	0	const __m128i wt_res = _mm_add_epi16(data_ref_0, res_unsigned);
97	0	res = _mm_srai_epi16(wt_res, 1);
98	0	}
99	0	return res;
100	0	} Unexecuted instantiation: aom_subpixel_8t_intrin_ssse3.c:comp_avg Unexecuted instantiation: highbd_convolve_ssse3.c:comp_avg Unexecuted instantiation: convolve_2d_sse2.c:comp_avg Unexecuted instantiation: jnt_convolve_sse2.c:comp_avg Unexecuted instantiation: jnt_convolve_ssse3.c:comp_avg Unexecuted instantiation: resize_ssse3.c:comp_avg Unexecuted instantiation: highbd_convolve_2d_ssse3.c:comp_avg Unexecuted instantiation: highbd_convolve_2d_sse4.c:comp_avg Unexecuted instantiation: highbd_jnt_convolve_sse4.c:comp_avg
101
102		static inline __m128i convolve_rounding(const __m128i *const res_unsigned,
103		const __m128i *const offset_const,
104		const __m128i *const round_const,
105	0	const int round_shift) {
106	0	const __m128i res_signed = _mm_sub_epi16(res_unsigned, offset_const);
107	0	const __m128i res_round =
108	0	_mm_srai_epi16(_mm_add_epi16(res_signed, *round_const), round_shift);
109	0	return res_round;
110	0	} Unexecuted instantiation: aom_subpixel_8t_intrin_ssse3.c:convolve_rounding Unexecuted instantiation: highbd_convolve_ssse3.c:convolve_rounding Unexecuted instantiation: convolve_2d_sse2.c:convolve_rounding Unexecuted instantiation: jnt_convolve_sse2.c:convolve_rounding Unexecuted instantiation: jnt_convolve_ssse3.c:convolve_rounding Unexecuted instantiation: resize_ssse3.c:convolve_rounding Unexecuted instantiation: highbd_convolve_2d_ssse3.c:convolve_rounding Unexecuted instantiation: highbd_convolve_2d_sse4.c:convolve_rounding Unexecuted instantiation: highbd_jnt_convolve_sse4.c:convolve_rounding
111
112		static inline __m128i highbd_convolve_rounding_sse2(
113		const __m128i const res_unsigned, const __m128i const offset_const,
114	0	const __m128i *const round_const, const int round_shift) {
115	0	const __m128i res_signed = _mm_sub_epi32(res_unsigned, offset_const);
116	0	const __m128i res_round =
117	0	_mm_srai_epi32(_mm_add_epi32(res_signed, *round_const), round_shift);
118
119	0	return res_round;
120	0	} Unexecuted instantiation: aom_subpixel_8t_intrin_ssse3.c:highbd_convolve_rounding_sse2 Unexecuted instantiation: highbd_convolve_ssse3.c:highbd_convolve_rounding_sse2 Unexecuted instantiation: convolve_2d_sse2.c:highbd_convolve_rounding_sse2 Unexecuted instantiation: jnt_convolve_sse2.c:highbd_convolve_rounding_sse2 Unexecuted instantiation: jnt_convolve_ssse3.c:highbd_convolve_rounding_sse2 Unexecuted instantiation: resize_ssse3.c:highbd_convolve_rounding_sse2 Unexecuted instantiation: highbd_convolve_2d_ssse3.c:highbd_convolve_rounding_sse2 Unexecuted instantiation: highbd_convolve_2d_sse4.c:highbd_convolve_rounding_sse2 Unexecuted instantiation: highbd_jnt_convolve_sse4.c:highbd_convolve_rounding_sse2
121
122		#endif // AOM_AOM_DSP_X86_CONVOLVE_SSE2_H_

Coverage Report

Created: 2025-06-13 07:07