diff options
Diffstat (limited to 'indra/llmath/llvector4a.cpp')
| -rwxr-xr-x[-rw-r--r--] | indra/llmath/llvector4a.cpp | 11 |
1 files changed, 10 insertions, 1 deletions
diff --git a/indra/llmath/llvector4a.cpp b/indra/llmath/llvector4a.cpp index b66b7a7076..480ccf4ed9 100644..100755 --- a/indra/llmath/llvector4a.cpp +++ b/indra/llmath/llvector4a.cpp @@ -24,6 +24,7 @@ * $/LicenseInfo$ */ +#include "llmemory.h" #include "llmath.h" #include "llquantize.h" @@ -40,11 +41,16 @@ extern const LLVector4a LL_V4A_EPSILON = reinterpret_cast<const LLVector4a&> ( F /*static */void LLVector4a::memcpyNonAliased16(F32* __restrict dst, const F32* __restrict src, size_t bytes) { +// memcpy((void*)dst,(const void*)src,bytes); +#if 1 assert(src != NULL); assert(dst != NULL); assert(bytes > 0); assert((bytes % sizeof(F32))== 0); - + ll_assert_aligned(src,16); + ll_assert_aligned(dst,16); + assert(bytes%16==0); + F32* end = dst + (bytes / sizeof(F32) ); if (bytes > 64) @@ -86,6 +92,7 @@ extern const LLVector4a LL_V4A_EPSILON = reinterpret_cast<const LLVector4a&> ( F dst += 4; src += 4; } +#endif } void LLVector4a::setRotated( const LLRotation& rot, const LLVector4a& vec ) @@ -189,6 +196,8 @@ void LLVector4a::quantize16( const LLVector4a& low, const LLVector4a& high ) LLVector4a oneOverDelta; { static LL_ALIGN_16( const F32 F_TWO_4A[4] ) = { 2.f, 2.f, 2.f, 2.f }; + ll_assert_aligned(F_TWO_4A,16); + LLVector4a two; two.load4a( F_TWO_4A ); // Here we use _mm_rcp_ps plus one round of newton-raphson |