diff options
Diffstat (limited to 'externals/grill/flext/source')
| -rw-r--r-- | externals/grill/flext/source/flext.h | 4 | ||||
| -rwxr-xr-x | externals/grill/flext/source/flsimd.cpp | 301 | ||||
| -rw-r--r-- | externals/grill/flext/source/flsupport.h | 5 |
3 files changed, 244 insertions, 66 deletions
diff --git a/externals/grill/flext/source/flext.h b/externals/grill/flext/source/flext.h index 7472e4a1..1a87e2bf 100644 --- a/externals/grill/flext/source/flext.h +++ b/externals/grill/flext/source/flext.h @@ -23,10 +23,10 @@ WARRANTIES, see the file, "license.txt," in this distribution. */ //! \brief flext version number -#define FLEXT_VERSION 404 +#define FLEXT_VERSION 500 //! \brief flext version string -#define FLEXT_VERSTR "0.4.4" +#define FLEXT_VERSTR "0.5.0pre" //! @} diff --git a/externals/grill/flext/source/flsimd.cpp b/externals/grill/flext/source/flsimd.cpp index 9786e4e7..a491bd51 100755 --- a/externals/grill/flext/source/flsimd.cpp +++ b/externals/grill/flext/source/flsimd.cpp @@ -284,15 +284,14 @@ void flext::CopySamples(t_sample *dst,const t_sample *src,int cnt) #else #ifdef FLEXT_USE_SIMD #ifdef _MSC_VER -#if 1 // t_sample is float if(GetSIMDCapabilities()&simd_sse) { // single precision int n = cnt>>4; cnt -= n<<4; - if((reinterpret_cast<unsigned long>(src)&(__alignof(t_sample)-1)) == 0 - && (reinterpret_cast<unsigned long>(dst)&(__alignof(t_sample)-1)) == 0 + if((reinterpret_cast<unsigned long>(src)&(__alignof(__m128)-1)) == 0 + && (reinterpret_cast<unsigned long>(dst)&(__alignof(__m128)-1)) == 0 ) { // aligned version while(n--) { @@ -316,41 +315,6 @@ void flext::CopySamples(t_sample *dst,const t_sample *src,int cnt) while(cnt--) *(dst++) = *(src++); } else -#elif 0 // t_sample is double - if(GetSIMDCapabilities()&simd_sse2) { - // double precision - - int n = cnt>>3; - cnt -= n<<3; - - if((reinterpret_cast<unsigned long>(src)&(__alignof(t_sample)-1)) == 0 - && (reinterpret_cast<unsigned long>(dst)&(__alignof(t_sample)-1)) == 0 - ) { - // aligned version - while(n--) { - _mm_store_pd(dst+0,_mm_load_pd(src+0)); - _mm_store_pd(dst+2,_mm_load_pd(src+2)); - _mm_store_pd(dst+4,_mm_load_pd(src+4)); - _mm_store_pd(dst+6,_mm_load_pd(src+6)); - src += 8,dst += 8; - } - } - else { - // unaligned version - while(n--) { - _mm_storeu_pd(dst+0,_mm_loadu_pd(src+0)); - _mm_storeu_pd(dst+2,_mm_loadu_pd(src+2)); - _mm_storeu_pd(dst+4,_mm_loadu_pd(src+4)); - _mm_storeu_pd(dst+6,_mm_loadu_pd(src+6)); - src += 8,dst += 8; - } - } - while(cnt--) *(dst++) = *(src++); - } - else -#else - #error t_sample data type has illegal size -#endif #elif FLEXT_OS == FLEXT_OS_MAC && defined(__VEC__) && defined(__VECTOROPS__) { int n = cnt>>2,n4 = n<<2; @@ -388,7 +352,6 @@ void flext::SetSamples(t_sample *dst,int cnt,t_sample s) #else #ifdef FLEXT_USE_SIMD #ifdef _MSC_VER -#if 1 // t_sample is float if(GetSIMDCapabilities()&simd_sse) { // single precision @@ -396,7 +359,7 @@ void flext::SetSamples(t_sample *dst,int cnt,t_sample s) int n = cnt>>4; cnt -= n<<4; - if((reinterpret_cast<unsigned long>(dst)&(__alignof(t_sample)-1)) == 0) { + if((reinterpret_cast<unsigned long>(dst)&(__alignof(__m128)-1)) == 0) { // aligned version while(n--) { _mm_store_ps(dst+0,v); @@ -419,51 +382,261 @@ void flext::SetSamples(t_sample *dst,int cnt,t_sample s) while(cnt--) *(dst++) = s; } else -#elif 0 // t_sample is double - if(GetSIMDCapabilities()&simd_sse2) { - // double precision +#endif // _MSC_VER +#endif // FLEXT_USE_SIMD + { + int n = cnt>>3; + cnt -= n<<3; + while(n--) { + dst[0] = dst[1] = dst[2] = dst[3] = dst[4] = dst[5] = dst[6] = dst[7] = s; + dst += 8; + } + + while(cnt--) *(dst++) = s; + } +#endif +} + + +void flext::MulSamples(t_sample *dst,const t_sample *src,t_sample mul,int cnt) +{ +#ifdef FLEXT_USE_IPP + if(sizeof(t_sample) == 4) { + ippsCopy_32f((const float *)src,(float *)dst,cnt); + ippsMulC_32f_I((float)mul,(float *)dst,cnt); + } + else if(sizeof(t_sample) == 8) { + ippsCopy_64f((const double *)src,(double *)dst,cnt); + ippsMulC_64f_I((double)mul,(double *)dst,cnt); + } + else + ERRINTERNAL(); +#else +#ifdef FLEXT_USE_SIMD +#ifdef _MSC_VER + if(GetSIMDCapabilities()&simd_sse) { + // single precision + __m128 a = _mm_load1_ps(&mul); - __m128 v = _mm_load1_pd(&s); - int n = cnt>>3; - cnt -= n<<3; + int n = cnt>>4; + cnt -= n<<4; - if((reinterpret_cast<unsigned long>(dst)&(__alignof(t_sample)-1)) == 0) { + if((reinterpret_cast<unsigned long>(src)&(__alignof(__m128)-1)) == 0 + && (reinterpret_cast<unsigned long>(dst)&(__alignof(__m128)-1)) == 0 + ) { // aligned version - while(n--) { - _mm_store_pd(dst+0,v); - _mm_store_pd(dst+2,v); - _mm_store_pd(dst+4,v); - _mm_store_pd(dst+8,v); - dst += 8; + while(n--) { + _mm_store_ps(dst+0,_mm_mul_ps(a,_mm_load_ps(src+0))); + _mm_store_ps(dst+4,_mm_mul_ps(a,_mm_load_ps(src+4))); + _mm_store_ps(dst+8,_mm_mul_ps(a,_mm_load_ps(src+8))); + _mm_store_ps(dst+12,_mm_mul_ps(a,_mm_load_ps(src+12))); + src += 16,dst += 16; } } else { // unaligned version while(n--) { - _mm_storeu_pd(dst+0,v); - _mm_storeu_pd(dst+2,v); - _mm_storeu_pd(dst+4,v); - _mm_storeu_pd(dst+8,v); - dst += 8; + _mm_storeu_ps(dst+0,_mm_mul_ps(a,_mm_loadu_ps(src+0))); + _mm_storeu_ps(dst+4,_mm_mul_ps(a,_mm_loadu_ps(src+4))); + _mm_storeu_ps(dst+8,_mm_mul_ps(a,_mm_loadu_ps(src+8))); + _mm_storeu_ps(dst+12,_mm_mul_ps(a,_mm_loadu_ps(src+12))); + src += 16,dst += 16; } } - while(cnt--) *(dst++) = s; + while(cnt--) *(dst++) = *(src++)*mul; } else +/* +#elif FLEXT_OS == FLEXT_OS_MAC && defined(__VEC__) && defined(__VECTOROPS__) + { + int n = cnt>>2,n4 = n<<2; + cnt -= n4; + vScopy(n4,src,dst); + src += n4,dst += n4; + while(cnt--) *(dst++) = *(src++); + } +*/ +#endif // _MSC_VER +#endif // FLEXT_USE_SIMD + { + int n = cnt>>3; + cnt -= n<<3; + while(n--) { + dst[0] = src[0]*mul; + dst[1] = src[1]*mul; + dst[2] = src[2]*mul; + dst[3] = src[3]*mul; + dst[4] = src[4]*mul; + dst[5] = src[5]*mul; + dst[6] = src[6]*mul; + dst[7] = src[7]*mul; + src += 8,dst += 8; + } + while(cnt--) *(dst++) = *(src++)*mul; + } +#endif +} + + +void flext::AddSamples(t_sample *dst,const t_sample *src,t_sample add,int cnt) +{ +#ifdef FLEXT_USE_IPP + if(sizeof(t_sample) == 4) { + ippsCopy_32f((const float *)src,(float *)dst,cnt); + ippsAddC_32f_I((float)mul,(float *)dst,cnt); + } + else if(sizeof(t_sample) == 8) { + ippsCopy_64f((const double *)src,(double *)dst,cnt); + ippsAddC_64f_I((double)mul,(double *)dst,cnt); + } + else + ERRINTERNAL(); #else - #error t_sample data type has illegal size +#ifdef FLEXT_USE_SIMD +#ifdef _MSC_VER + if(GetSIMDCapabilities()&simd_sse) { + // single precision + __m128 a = _mm_load1_ps(&add); + + int n = cnt>>4; + cnt -= n<<4; + + if((reinterpret_cast<unsigned long>(src)&(__alignof(__m128)-1)) == 0 + && (reinterpret_cast<unsigned long>(dst)&(__alignof(__m128)-1)) == 0 + ) { + // aligned version + while(n--) { + _mm_store_ps(dst+0,_mm_add_ps(a,_mm_load_ps(src+0))); + _mm_store_ps(dst+4,_mm_add_ps(a,_mm_load_ps(src+4))); + _mm_store_ps(dst+8,_mm_add_ps(a,_mm_load_ps(src+8))); + _mm_store_ps(dst+12,_mm_add_ps(a,_mm_load_ps(src+12))); + src += 16,dst += 16; + } + } + else { + // unaligned version + while(n--) { + _mm_storeu_ps(dst+0,_mm_add_ps(a,_mm_loadu_ps(src+0))); + _mm_storeu_ps(dst+4,_mm_add_ps(a,_mm_loadu_ps(src+4))); + _mm_storeu_ps(dst+8,_mm_add_ps(a,_mm_loadu_ps(src+8))); + _mm_storeu_ps(dst+12,_mm_add_ps(a,_mm_loadu_ps(src+12))); + src += 16,dst += 16; + } + } + while(cnt--) *(dst++) = *(src++)+add; + } + else +/* +#elif FLEXT_OS == FLEXT_OS_MAC && defined(__VEC__) && defined(__VECTOROPS__) + { + int n = cnt>>2,n4 = n<<2; + cnt -= n4; + vScopy(n4,src,dst); + src += n4,dst += n4; + while(cnt--) *(dst++) = *(src++); + } +*/ +#endif // _MSC_VER +#endif // FLEXT_USE_SIMD + { + int n = cnt>>3; + cnt -= n<<3; + while(n--) { + dst[0] = src[0]+add; + dst[1] = src[1]+add; + dst[2] = src[2]+add; + dst[3] = src[3]+add; + dst[4] = src[4]+add; + dst[5] = src[5]+add; + dst[6] = src[6]+add; + dst[7] = src[7]+add; + src += 8,dst += 8; + } + while(cnt--) *(dst++) = *(src++)+add; + } #endif +} + + +void flext::ScaleSamples(t_sample *dst,const t_sample *src,t_sample mul,t_sample add,int cnt) +{ +#ifdef FLEXT_USE_IPP + if(sizeof(t_sample) == 4) { + ippsCopy_32f((const float *)src,(float *)dst,cnt); + ippsMulC_32f_I((float)mul,(float *)dst,cnt); + ippsAddC_32f_I((float)add,(float *)dst,cnt); + } + else if(sizeof(t_sample) == 8) { + ippsCopy_64f((const double *)src,(double *)dst,cnt); + ippsMulC_64f_I((double)mul,(double *)dst,cnt); + ippsAddC_64f_I((double)add,(double *)dst,cnt); + } + else + ERRINTERNAL(); +#else +#ifdef FLEXT_USE_SIMD +#ifdef _MSC_VER + if(GetSIMDCapabilities()&simd_sse) { + // single precision + __m128 a = _mm_load1_ps(&add); + __m128 m = _mm_load1_ps(&mul); + + int n = cnt>>4; + cnt -= n<<4; + + if((reinterpret_cast<unsigned long>(src)&(__alignof(__m128)-1)) == 0 + && (reinterpret_cast<unsigned long>(dst)&(__alignof(__m128)-1)) == 0 + ) { + // aligned version + while(n--) { + _mm_store_ps(dst+0,_mm_add_ps(a,_mm_mul_ps(m,_mm_load_ps(src+0)))); + _mm_store_ps(dst+4,_mm_add_ps(a,_mm_mul_ps(m,_mm_load_ps(src+4)))); + _mm_store_ps(dst+8,_mm_add_ps(a,_mm_mul_ps(m,_mm_load_ps(src+8)))); + _mm_store_ps(dst+12,_mm_add_ps(a,_mm_mul_ps(m,_mm_load_ps(src+12)))); + src += 16,dst += 16; + } + } + else { + // unaligned version + while(n--) { + _mm_storeu_ps(dst+0,_mm_add_ps(a,_mm_mul_ps(m,_mm_loadu_ps(src+0)))); + _mm_storeu_ps(dst+4,_mm_add_ps(a,_mm_mul_ps(m,_mm_loadu_ps(src+4)))); + _mm_storeu_ps(dst+8,_mm_add_ps(a,_mm_mul_ps(m,_mm_loadu_ps(src+8)))); + _mm_storeu_ps(dst+12,_mm_add_ps(a,_mm_mul_ps(m,_mm_loadu_ps(src+12)))); + src += 16,dst += 16; + } + } + while(cnt--) *(dst++) = *(src++)*mul+add; + } + else +/* +#elif FLEXT_OS == FLEXT_OS_MAC && defined(__VEC__) && defined(__VECTOROPS__) + { + int n = cnt>>2,n4 = n<<2; + cnt -= n4; + vScopy(n4,src,dst); + src += n4,dst += n4; + while(cnt--) *(dst++) = *(src++); + } +*/ #endif // _MSC_VER #endif // FLEXT_USE_SIMD { int n = cnt>>3; cnt -= n<<3; while(n--) { - dst[0] = dst[1] = dst[2] = dst[3] = dst[4] = dst[5] = dst[6] = dst[7] = s; - dst += 8; + dst[0] = src[0]*mul+add; + dst[1] = src[1]*mul+add; + dst[2] = src[2]*mul+add; + dst[3] = src[3]*mul+add; + dst[4] = src[4]*mul+add; + dst[5] = src[5]*mul+add; + dst[6] = src[6]*mul+add; + dst[7] = src[7]*mul+add; + src += 8,dst += 8; } - - while(cnt--) *(dst++) = s; + while(cnt--) *(dst++) = *(src++)*mul+add; } #endif } + diff --git a/externals/grill/flext/source/flsupport.h b/externals/grill/flext/source/flsupport.h index 2235e854..fef7e16a 100644 --- a/externals/grill/flext/source/flsupport.h +++ b/externals/grill/flext/source/flsupport.h @@ -959,6 +959,11 @@ public: static unsigned long GetSIMDCapabilities(); + + static void MulSamples(t_sample *dst,const t_sample *src,t_sample mul,int cnt); + static void AddSamples(t_sample *dst,const t_sample *src,t_sample add,int cnt); + static void ScaleSamples(t_sample *dst,const t_sample *src,t_sample mul,t_sample add,int cnt); + //! @} FLEXT_S_SIMD |