/src/openh264/codec/processing/src/vaacalc/vaacalculation.cpp
Line | Count | Source |
1 | | /*! |
2 | | * \copy |
3 | | * Copyright (c) 2013, Cisco Systems |
4 | | * All rights reserved. |
5 | | * |
6 | | * Redistribution and use in source and binary forms, with or without |
7 | | * modification, are permitted provided that the following conditions |
8 | | * are met: |
9 | | * |
10 | | * * Redistributions of source code must retain the above copyright |
11 | | * notice, this list of conditions and the following disclaimer. |
12 | | * |
13 | | * * Redistributions in binary form must reproduce the above copyright |
14 | | * notice, this list of conditions and the following disclaimer in |
15 | | * the documentation and/or other materials provided with the |
16 | | * distribution. |
17 | | * |
18 | | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
19 | | * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
20 | | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS |
21 | | * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE |
22 | | * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, |
23 | | * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, |
24 | | * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; |
25 | | * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER |
26 | | * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
27 | | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN |
28 | | * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE |
29 | | * POSSIBILITY OF SUCH DAMAGE. |
30 | | * |
31 | | */ |
32 | | |
33 | | #include "vaacalculation.h" |
34 | | #include "cpu.h" |
35 | | |
36 | | WELSVP_NAMESPACE_BEGIN |
37 | | |
38 | | |
39 | | /////////////////////////////////////////////////////////////////////////////////////////////////////////////// |
40 | | |
41 | 0 | CVAACalculation::CVAACalculation (int32_t iCpuFlag) { |
42 | 0 | m_iCPUFlag = iCpuFlag; |
43 | 0 | m_eMethod = METHOD_VAA_STATISTICS; |
44 | |
|
45 | 0 | WelsMemset (&m_sCalcParam, 0, sizeof (m_sCalcParam)); |
46 | 0 | WelsMemset (&m_sVaaFuncs, 0, sizeof (m_sVaaFuncs)); |
47 | 0 | InitVaaFuncs (m_sVaaFuncs, m_iCPUFlag); |
48 | 0 | } |
49 | | |
50 | 0 | CVAACalculation::~CVAACalculation() { |
51 | 0 | } |
52 | | |
53 | 0 | void CVAACalculation::InitVaaFuncs (SVaaFuncs& sVaaFuncs, int32_t iCpuFlag) { |
54 | 0 | sVaaFuncs.pfVAACalcSad = VAACalcSad_c; |
55 | 0 | sVaaFuncs.pfVAACalcSadBgd = VAACalcSadBgd_c; |
56 | 0 | sVaaFuncs.pfVAACalcSadSsd = VAACalcSadSsd_c; |
57 | 0 | sVaaFuncs.pfVAACalcSadSsdBgd = VAACalcSadSsdBgd_c; |
58 | 0 | sVaaFuncs.pfVAACalcSadVar = VAACalcSadVar_c; |
59 | 0 | #ifdef X86_ASM |
60 | 0 | if ((iCpuFlag & WELS_CPU_SSE2) == WELS_CPU_SSE2) { |
61 | 0 | sVaaFuncs.pfVAACalcSad = VAACalcSad_sse2; |
62 | 0 | sVaaFuncs.pfVAACalcSadBgd = VAACalcSadBgd_sse2; |
63 | 0 | sVaaFuncs.pfVAACalcSadSsd = VAACalcSadSsd_sse2; |
64 | 0 | sVaaFuncs.pfVAACalcSadSsdBgd = VAACalcSadSsdBgd_sse2; |
65 | 0 | sVaaFuncs.pfVAACalcSadVar = VAACalcSadVar_sse2; |
66 | 0 | } |
67 | 0 | #ifdef HAVE_AVX2 |
68 | 0 | if (iCpuFlag & WELS_CPU_AVX2) { |
69 | 0 | sVaaFuncs.pfVAACalcSad = VAACalcSad_avx2; |
70 | 0 | sVaaFuncs.pfVAACalcSadBgd = VAACalcSadBgd_avx2; |
71 | 0 | sVaaFuncs.pfVAACalcSadSsd = VAACalcSadSsd_avx2; |
72 | 0 | sVaaFuncs.pfVAACalcSadSsdBgd = VAACalcSadSsdBgd_avx2; |
73 | 0 | sVaaFuncs.pfVAACalcSadVar = VAACalcSadVar_avx2; |
74 | 0 | } |
75 | 0 | #endif |
76 | 0 | #endif//X86_ASM |
77 | | #ifdef HAVE_NEON |
78 | | if ((iCpuFlag & WELS_CPU_NEON) == WELS_CPU_NEON) { |
79 | | sVaaFuncs.pfVAACalcSad = VAACalcSad_neon; |
80 | | sVaaFuncs.pfVAACalcSadBgd = VAACalcSadBgd_neon; |
81 | | sVaaFuncs.pfVAACalcSadSsd = VAACalcSadSsd_neon; |
82 | | sVaaFuncs.pfVAACalcSadSsdBgd = VAACalcSadSsdBgd_neon; |
83 | | sVaaFuncs.pfVAACalcSadVar = VAACalcSadVar_neon; |
84 | | } |
85 | | #endif//HAVE_NEON |
86 | |
|
87 | | #if defined(HAVE_NEON_AARCH64) && defined(__aarch64__) |
88 | | if ((iCpuFlag & WELS_CPU_NEON) == WELS_CPU_NEON) { |
89 | | sVaaFuncs.pfVAACalcSad = VAACalcSad_AArch64_neon; |
90 | | sVaaFuncs.pfVAACalcSadBgd = VAACalcSadBgd_AArch64_neon; |
91 | | sVaaFuncs.pfVAACalcSadSsd = VAACalcSadSsd_AArch64_neon; |
92 | | sVaaFuncs.pfVAACalcSadSsdBgd = VAACalcSadSsdBgd_AArch64_neon; |
93 | | sVaaFuncs.pfVAACalcSadVar = VAACalcSadVar_AArch64_neon; |
94 | | } |
95 | | #endif//HAVE_NEON_AARCH64 |
96 | |
|
97 | | #ifdef HAVE_MMI |
98 | | if ((iCpuFlag & WELS_CPU_MMI) == WELS_CPU_MMI) { |
99 | | sVaaFuncs.pfVAACalcSad = VAACalcSad_mmi; |
100 | | sVaaFuncs.pfVAACalcSadBgd = VAACalcSadBgd_mmi; |
101 | | sVaaFuncs.pfVAACalcSadSsd = VAACalcSadSsd_mmi; |
102 | | sVaaFuncs.pfVAACalcSadSsdBgd = VAACalcSadSsdBgd_mmi; |
103 | | sVaaFuncs.pfVAACalcSadVar = VAACalcSadVar_mmi; |
104 | | } |
105 | | #endif//HAVE_MMI |
106 | |
|
107 | | #ifdef HAVE_LSX |
108 | | if ((iCpuFlag & WELS_CPU_LSX) == WELS_CPU_LSX) { |
109 | | sVaaFuncs.pfVAACalcSad = VAACalcSad_lsx; |
110 | | sVaaFuncs.pfVAACalcSadBgd = VAACalcSadBgd_lsx; |
111 | | } |
112 | | #endif//HAVE_LSX |
113 | | #ifdef HAVE_LASX |
114 | | if ((iCpuFlag & WELS_CPU_LASX) == WELS_CPU_LASX) { |
115 | | sVaaFuncs.pfVAACalcSadBgd = VAACalcSadBgd_lasx; |
116 | | } |
117 | | #endif//HAVE_LASX |
118 | 0 | } |
119 | | |
120 | 0 | EResult CVAACalculation::Process (int32_t iType, SPixMap* pSrcPixMap, SPixMap* pRefPixMap) { |
121 | 0 | uint8_t* pCurData = (uint8_t*)pSrcPixMap->pPixel[0]; |
122 | 0 | uint8_t* pRefData = (uint8_t*)pRefPixMap->pPixel[0]; |
123 | 0 | int32_t iPicWidth = pSrcPixMap->sRect.iRectWidth; |
124 | 0 | int32_t iPicHeight = pSrcPixMap->sRect.iRectHeight; |
125 | 0 | int32_t iPicStride = pSrcPixMap->iStride[0]; |
126 | |
|
127 | 0 | SVAACalcResult* pResult = m_sCalcParam.pCalcResult; |
128 | |
|
129 | 0 | if (pCurData == NULL || pRefData == NULL) { |
130 | 0 | return RET_INVALIDPARAM; |
131 | 0 | } |
132 | | |
133 | 0 | pResult->pCurY = pCurData; |
134 | 0 | pResult->pRefY = pRefData; |
135 | 0 | if (m_sCalcParam.iCalcBgd) { |
136 | 0 | if (m_sCalcParam.iCalcSsd) { |
137 | 0 | m_sVaaFuncs.pfVAACalcSadSsdBgd (pCurData, pRefData, iPicWidth, iPicHeight, iPicStride, &pResult->iFrameSad, |
138 | 0 | (int32_t*)pResult->pSad8x8, pResult->pSum16x16, pResult->pSumOfSquare16x16, pResult->pSsd16x16, |
139 | 0 | (int32_t*)pResult->pSumOfDiff8x8, (uint8_t*)pResult->pMad8x8); |
140 | 0 | } else { |
141 | 0 | m_sVaaFuncs.pfVAACalcSadBgd (pCurData, pRefData, iPicWidth, iPicHeight, iPicStride, &pResult->iFrameSad, |
142 | 0 | (int32_t*) (pResult->pSad8x8), (int32_t*) (pResult->pSumOfDiff8x8), (uint8_t*)pResult->pMad8x8); |
143 | 0 | } |
144 | 0 | } else { |
145 | 0 | if (m_sCalcParam.iCalcSsd) { |
146 | 0 | m_sVaaFuncs.pfVAACalcSadSsd (pCurData, pRefData, iPicWidth, iPicHeight, iPicStride, &pResult->iFrameSad, |
147 | 0 | (int32_t*)pResult->pSad8x8, pResult->pSum16x16, pResult->pSumOfSquare16x16, pResult->pSsd16x16); |
148 | 0 | } else { |
149 | 0 | if (m_sCalcParam.iCalcVar) { |
150 | 0 | m_sVaaFuncs.pfVAACalcSadVar (pCurData, pRefData, iPicWidth, iPicHeight, iPicStride, &pResult->iFrameSad, |
151 | 0 | (int32_t*)pResult->pSad8x8, pResult->pSum16x16, pResult->pSumOfSquare16x16); |
152 | 0 | } else { |
153 | 0 | m_sVaaFuncs.pfVAACalcSad (pCurData, pRefData, iPicWidth, iPicHeight, iPicStride, &pResult->iFrameSad, |
154 | 0 | (int32_t*)pResult->pSad8x8); |
155 | 0 | } |
156 | 0 | } |
157 | 0 | } |
158 | |
|
159 | 0 | return RET_SUCCESS; |
160 | 0 | } |
161 | | |
162 | 0 | EResult CVAACalculation::Set (int32_t iType, void* pParam) { |
163 | 0 | if (pParam == NULL || ((SVAACalcParam*)pParam)->pCalcResult == NULL) { |
164 | 0 | return RET_INVALIDPARAM; |
165 | 0 | } |
166 | | |
167 | 0 | m_sCalcParam = * (SVAACalcParam*)pParam; |
168 | |
|
169 | 0 | return RET_SUCCESS; |
170 | 0 | } |
171 | | |
172 | | |
173 | | WELSVP_NAMESPACE_END |