版本
menu_open

include/AK/SoundEngine/Platforms/SSE/AkSimd.h

Go to the documentation of this file.
00001 /*******************************************************************************
00002 The content of this file includes portions of the AUDIOKINETIC Wwise Technology
00003 released in source code form as part of the SDK installer package.
00004 
00005 Commercial License Usage
00006 
00007 Licensees holding valid commercial licenses to the AUDIOKINETIC Wwise Technology
00008 may use this file in accordance with the end user license agreement provided 
00009 with the software or, alternatively, in accordance with the terms contained in a
00010 written agreement between you and Audiokinetic Inc.
00011 
00012 Apache License Usage
00013 
00014 Alternatively, this file may be used under the Apache License, Version 2.0 (the 
00015 "Apache License"); you may not use this file except in compliance with the 
00016 Apache License. You may obtain a copy of the Apache License at 
00017 http://www.apache.org/licenses/LICENSE-2.0.
00018 
00019 Unless required by applicable law or agreed to in writing, software distributed
00020 under the Apache License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES
00021 OR CONDITIONS OF ANY KIND, either express or implied. See the Apache License for
00022 the specific language governing permissions and limitations under the License.
00023 
00024   Version: <VERSION>  Build: <BUILDNUMBER>
00025   Copyright (c) <COPYRIGHTYEAR> Audiokinetic Inc.
00026 *******************************************************************************/
00027 
00028 // AkSimd.h
00029 
00032 
00033 #ifndef _AK_SIMD_SSE_H_
00034 #define _AK_SIMD_SSE_H_
00035 
00036 #include <AK/SoundEngine/Common/AkTypes.h>
00037 #include <xmmintrin.h>
00038 
00041 
00042 
00043 #define AKSIMD_ARCHCACHELINESIZE    (64)                ///< Assumed cache line width for architectures on this platform
00044 #define AKSIMD_ARCHMAXPREFETCHSIZE  (512)               ///< Use this to control how much prefetching maximum is desirable (assuming 8-way cache)       
00045 
00046 #define AKSIMD_PREFETCHMEMORY( __offset__, __add__ ) _mm_prefetch(((char *)(__add__))+(__offset__), _MM_HINT_NTA ) 
00047 
00049 
00050 
00053 
00054 #define AKSIMD_ALIGNSIZE( __Size__ ) (((__Size__) + 15) & ~15)
00055 
00056 
00057 
00060 
00061 
00062 typedef float   AKSIMD_F32;     
00063 typedef __m128  AKSIMD_V4F32;   
00064 typedef AKSIMD_V4F32 AKSIMD_V4COND;  
00065 typedef AKSIMD_V4F32 AKSIMD_V4FCOND;     
00066 
00068 
00069 
00070 
00073 
00074 
00076 #define AKSIMD_LOAD_V4F32( __addr__ ) _mm_load_ps( (AkReal32*)(__addr__) )
00077 
00080 #define AKSIMD_LOADU_V4F32( __addr__ ) _mm_loadu_ps( (__addr__) )
00081 
00084 #define AKSIMD_LOAD1_V4F32( __scalar__ ) _mm_load1_ps( &(__scalar__) )
00085 
00088 #define AKSIMD_SET_V4F32( __scalar__ ) _mm_set_ps1( (__scalar__) )
00089 
00092 #define AKSIMD_SETZERO_V4F32() _mm_setzero_ps()
00093 
00097 #define AKSIMD_LOAD_SS_V4F32( __addr__ ) _mm_load_ss( (__addr__) )
00098 
00100 
00101 
00102 
00105 
00106 
00109 #define AKSIMD_STORE_V4F32( __addr__, __vec__ ) _mm_store_ps( (AkReal32*)(__addr__), (__vec__) )
00110 
00113 #define AKSIMD_STOREU_V4F32( __addr__, __vec__ ) _mm_storeu_ps( (AkReal32*)(__addr__), (__vec__) )
00114 
00117 #define AKSIMD_STORE1_V4F32( __addr__, __vec__ ) _mm_store_ss( (AkReal32*)(__addr__), (__vec__) )
00118 
00120 
00121 
00124 
00125 
00126 // Macro for shuffle parameter for AKSIMD_SHUFFLE_V4F32() (see _MM_SHUFFLE)
00127 #define AKSIMD_SHUFFLE( fp3, fp2, fp1, fp0 ) _MM_SHUFFLE( (fp3), (fp2), (fp1), (fp0) )
00128 
00131 // Usage: AKSIMD_SHUFFLE_V4F32( vec1, vec2, AKSIMD_SHUFFLE( z, y, x, w ) )
00132 #define AKSIMD_SHUFFLE_V4F32( a, b, i ) _mm_shuffle_ps( a, b, i )
00133 
00139 #define AKSIMD_MOVEHL_V4F32( a, b ) _mm_movehl_ps( a, b )
00140 
00146 #define AKSIMD_MOVELH_V4F32( a, b ) _mm_movelh_ps( a, b )
00147 
00149 #define AKSIMD_SHUFFLE_BADC( __a__ ) _mm_shuffle_ps( (__a__), (__a__), _MM_SHUFFLE(2,3,0,1))
00150 
00152 #define AKSIMD_SHUFFLE_CDAB( __a__ ) _mm_shuffle_ps( (__a__), (__a__), _MM_SHUFFLE(1,0,3,2))
00153 
00155 #define AKSIMD_SHUFFLE_BCDA( __a__ ) AKSIMD_SHUFFLE_V4F32( (__a__), (__a__), _MM_SHUFFLE(0,3,2,1))
00156 
00158 #define AKSIMD_DUP_ODD(__vv) AKSIMD_SHUFFLE_V4F32(__vv, __vv, AKSIMD_SHUFFLE(3,3,1,1))
00159 
00161 #define AKSIMD_DUP_EVEN(__vv) AKSIMD_SHUFFLE_V4F32(__vv, __vv, AKSIMD_SHUFFLE(2,2,0,0))
00162 
00164 
00165 
00166 
00169 
00170 
00173 #define AKSIMD_SUB_V4F32( a, b ) _mm_sub_ps( a, b )
00174 
00178 #define AKSIMD_SUB_SS_V4F32( a, b ) _mm_sub_ss( a, b )
00179 
00182 #define AKSIMD_ADD_V4F32( a, b ) _mm_add_ps( a, b )
00183 
00187 #define AKSIMD_ADD_SS_V4F32( a, b ) _mm_add_ss( a, b )
00188 
00191 #define AKSIMD_MUL_V4F32( a, b ) _mm_mul_ps( a, b )
00192 
00193 #define AKSIMD_DIV_V4F32( a, b ) _mm_div_ps( a, b )
00194 
00199 #define AKSIMD_MUL_SS_V4F32( a, b ) _mm_mul_ss( a, b )
00200 
00202 #define AKSIMD_MADD_V4F32( __a__, __b__, __c__ ) _mm_add_ps( _mm_mul_ps( (__a__), (__b__) ), (__c__) )
00203 #define AKSIMD_MSUB_V4F32( __a__, __b__, __c__ ) _mm_sub_ps( _mm_mul_ps( (__a__), (__b__) ), (__c__) )
00204 
00206 #define AKSIMD_MADD_SS_V4F32( __a__, __b__, __c__ ) _mm_add_ss( _mm_mul_ss( (__a__), (__b__) ), (__c__) )
00207 
00210 #define AKSIMD_MIN_V4F32( a, b ) _mm_min_ps( a, b )
00211 
00214 #define AKSIMD_MAX_V4F32( a, b ) _mm_max_ps( a, b )
00215 
00217 #define AKSIMD_ABS_V4F32( a ) _mm_andnot_ps(_mm_set1_ps(-0.f), a)
00218 
00220 #define AKSIMD_NEG_V4F32( __a__ ) _mm_xor_ps(_mm_set1_ps(-0.f), __a__)
00221 
00223 #define AKSIMD_SQRT_V4F32( __a__ ) _mm_sqrt_ps( (__a__) )
00224 
00229 static AkForceInline void AKSIMD_HORIZONTALADD(AKSIMD_V4F32 & vVec)
00230 {   
00231     __m128 vHighLow = _mm_movehl_ps(vVec, vVec);
00232     vVec = _mm_add_ps(vVec, vHighLow);
00233     vHighLow = _mm_shuffle_ps(vVec, vVec, 0x55);
00234     vVec = _mm_add_ps(vVec, vHighLow);
00235 } 
00236 
00237 static AkForceInline AKSIMD_V4F32 AKSIMD_DOTPRODUCT( AKSIMD_V4F32 & vVec, const AKSIMD_V4F32 & vfSigns )
00238 {
00239     AKSIMD_V4F32 vfDotProduct = AKSIMD_MUL_V4F32( vVec, vfSigns );
00240     AKSIMD_HORIZONTALADD( vfDotProduct );
00241     return AKSIMD_SHUFFLE_V4F32( vfDotProduct, vfDotProduct, AKSIMD_SHUFFLE(0,0,0,0) );
00242 }
00243 
00245 static AkForceInline AKSIMD_V4F32 AKSIMD_COMPLEXMUL( const AKSIMD_V4F32 vCIn1, const AKSIMD_V4F32 vCIn2 )
00246 {
00247     static const AKSIMD_V4F32 vSign = { -1.f, 1.f, -1.f, 1.f }; 
00248 
00249     AKSIMD_V4F32 vTmp1 = _mm_shuffle_ps( vCIn1, vCIn1, _MM_SHUFFLE(2,2,0,0)); 
00250     vTmp1 = AKSIMD_MUL_V4F32( vTmp1, vCIn2 );
00251     AKSIMD_V4F32 vTmp2 = _mm_shuffle_ps( vCIn1, vCIn1, _MM_SHUFFLE(3,3,1,1)); 
00252     vTmp2 = AKSIMD_MUL_V4F32( vTmp2, vSign );
00253     vTmp2 = AKSIMD_MADD_V4F32( vTmp2, AKSIMD_SHUFFLE_BADC( vCIn2 ), vTmp1 );
00254     return vTmp2;
00255 }
00256 
00257 #ifdef AK_SSE3
00258 
00259 #include <pmmintrin.h>
00260 
00262 static AKSIMD_V4F32 AKSIMD_COMPLEXMUL_SSE3( const AKSIMD_V4F32 vCIn1, const AKSIMD_V4F32 vCIn2 )
00263 {
00264     AKSIMD_V4F32 vXMM0 = _mm_moveldup_ps(vCIn1);    // multiplier real  (a1,   a1,   a0,   a0) 
00265     vXMM0 = AKSIMD_MUL_V4F32(vXMM0, vCIn2);         // temp1            (a1d1, a1c1, a0d0, a0c0) 
00266     AKSIMD_V4F32 xMM1 = _mm_shuffle_ps(vCIn2, vCIn2, 0xB1); // shuf multiplicand(c1,   d1,   c0,   d0)  
00267     AKSIMD_V4F32 xMM2 = _mm_movehdup_ps(vCIn1);     // multiplier imag  (b1,   b1,   b0,   b0) 
00268     xMM2 = AKSIMD_MUL_V4F32( xMM2, xMM1);           // temp2            (b1c1, b1d1, b0c0, b0d0) 
00269     AKSIMD_V4F32 vCOut = _mm_addsub_ps(vXMM0, xMM2);        // b1c1+a1d1, a1c1-b1d1, a0d0+b0d0, a0c0-b0c0 
00270     return vCOut;
00271 }
00272 
00273 #endif
00274 
00275 #if defined _MSC_VER && ( _MSC_VER <= 1600 )
00276     #define AKSIMD_ASSERTFLUSHZEROMODE  AKASSERT( _MM_GET_FLUSH_ZERO_MODE(dummy) == _MM_FLUSH_ZERO_ON )
00277 #else
00278     #define AKSIMD_ASSERTFLUSHZEROMODE  AKASSERT( _MM_GET_FLUSH_ZERO_MODE() == _MM_FLUSH_ZERO_ON )
00279 #endif
00280 
00282 
00283 
00284 
00287 
00288 
00290 #define AKSIMD_ADD_V4I32( a, b ) _mm_add_epi32( a, b )
00291 
00293 
00294 
00295 
00298 
00299 
00302 #define AKSIMD_UNPACKLO_V4F32( a, b ) _mm_unpacklo_ps( a, b )
00303 
00306 #define AKSIMD_UNPACKHI_V4F32( a, b ) _mm_unpackhi_ps( a, b )
00307 
00309 
00310 
00314 
00315 
00316 #define AKSIMD_CMP_CTRLMASK __m128
00317 
00319 #define AKSIMD_LTEQ_V4F32( __a__, __b__ ) _mm_cmple_ps( (__a__), (__b__) )
00320 
00322 #define AKSIMD_GTEQ_V4F32( __a__, __b__ ) _mm_cmpge_ps( (__a__), (__b__) )
00323 
00325 #define AKSIMD_EQ_V4F32( __a__, __b__ ) _mm_cmpeq_ps( (__a__), (__b__) )
00326 
00328 static AkForceInline AKSIMD_V4F32 AKSIMD_VSEL_V4F32( AKSIMD_V4F32 vA, AKSIMD_V4F32 vB, AKSIMD_V4F32 vMask )
00329 {
00330     vB = _mm_and_ps( vB, vMask );
00331     vA= _mm_andnot_ps( vMask, vA );
00332     return _mm_or_ps( vA, vB );
00333 }
00334 
00335 // (cond1 >= cond2) ? b : a.
00336 #define AKSIMD_SEL_GTEQ_V4F32( __a__, __b__, __cond1__, __cond2__ ) AKSIMD_VSEL_V4F32( __a__, __b__, AKSIMD_GTEQ_V4F32( __cond1__, __cond2__ ) )
00337 
00338 // a >= 0 ? b : c ... Written, like, you know, the normal C++ operator syntax.
00339 #define AKSIMD_SEL_GTEZ_V4F32( __a__, __b__, __c__ ) AKSIMD_VSEL_V4F32( (__c__), (__b__), AKSIMD_GTEQ_V4F32( __a__, _mm_set1_ps(0) ) )
00340 
00341 #define AKSIMD_SPLAT_V4F32(var, idx) AKSIMD_SHUFFLE_V4F32(var,var, AKSIMD_SHUFFLE(idx,idx,idx,idx))
00342 
00344 
00345 
00346 #include <emmintrin.h>
00347 
00348 typedef __m128i AKSIMD_V4I32;   
00349 
00350 typedef AKSIMD_V4I32 AKSIMD_V4ICOND;
00351 
00353 #define AKSIMD_LOADU_V4I32( __addr__ ) _mm_loadu_si128( (__addr__) )
00354 
00356 #define AKSIMD_LOAD_V4I32( __addr__ ) _mm_load_si128( (__addr__) )
00357 
00359 #define AKSIMD_SETZERO_V4I32() _mm_setzero_si128()
00360 
00361 #define AKSIMD_SET_V4I32( __scalar__ ) _mm_set1_epi32( (__scalar__) )
00362 
00363 #define AKSIMD_SETV_V4I32( _d, _c, _b, _a ) _mm_set_epi32( (_d), (_c), (_b), (_a) )
00364 
00366 #define AKSIMD_STORE_V4I32( __addr__, __vec__ ) _mm_store_si128( (__addr__), (__vec__) )
00367 
00370 #define AKSIMD_STOREU_V4I32( __addr__, __vec__ ) _mm_storeu_si128( (__addr__), (__vec__) )
00371 
00374 
00375 
00378 #define AKSIMD_CONVERT_V4I32_TO_V4F32( __vec__ ) _mm_cvtepi32_ps( (__vec__) )
00379 
00382 #define AKSIMD_CONVERT_V4F32_TO_V4I32( __vec__ ) _mm_cvtps_epi32( (__vec__) )
00383 
00386 #define AKSIMD_TRUNCATE_V4F32_TO_V4I32( __vec__ ) _mm_cvttps_epi32( (__vec__) )
00387 
00390 #define AKSIMD_AND_V4I32( __a__, __b__ ) _mm_and_si128( (__a__), (__b__) )
00391 
00394 #define AKSIMD_CMPGT_V8I16( __a__, __b__ ) _mm_cmpgt_epi16( (__a__), (__b__) )
00395 
00397 
00398 
00401 #define AKSIMD_UNPACKLO_VECTOR8I16( a, b ) _mm_unpacklo_epi16( a, b )
00402 
00405 #define AKSIMD_UNPACKHI_VECTOR8I16( a, b ) _mm_unpackhi_epi16( a, b )
00406 
00409 #define AKSIMD_PACKS_V4I32( a, b ) _mm_packs_epi32( a, b )
00410 
00413 
00414 
00417 #define AKSIMD_SHIFTLEFT_V4I32( __vec__, __shiftBy__ ) \
00418     _mm_slli_epi32( (__vec__), (__shiftBy__) )
00419 
00422 #define AKSIMD_SHIFTRIGHTARITH_V4I32( __vec__, __shiftBy__ ) \
00423     _mm_srai_epi32( (__vec__), (__shiftBy__) )
00424 
00426 
00427 
00428 #if defined( AK_CPU_X86 ) && !defined(AK_IOS) /// MMX
00429 
00430 typedef __m64   AKSIMD_V2F32;   
00431 
00432 #endif
00433 
00434 
00435 #endif //_AK_SIMD_SSE_H_

此页面对您是否有帮助？

需要技术支持？

仍有疑问？或者问题？需要更多信息？欢迎联系我们，我们可以提供帮助！

查看我们的“技术支持”页面

介绍一下自己的项目。我们会竭力为您提供帮助。

来注册自己的项目，我们帮您快速入门，不带任何附加条件！

开始 Wwise 之旅