Version History Many thanks to Uwe Rossek for designing a DLL interface that lets you access Exentia from Delphi 5! The SIMD project exports the SSE methods  as DLL entry points, then the SIMD_Import defines the necessary bindings. Click here to download it! 0.6.0 Roman "GoodOk" Gudchenko strikes again! he discovered that CISC-style coding (i.e.  the new op reg1,[mem] style instead of the old load reg2,[mem] / op reg1,reg2 style) is up to 25% faster! You can choose either coding style by defining {$DEFINE CISC_STYLE}. fixed tSSEVector.Lerp method and implemented tSSEVector and t3DNowVector.DotProduct You can disable software prefetching by defining $DEFINE NO_SOFTWARE_PREFETCHING; this is especially useful on CPUs with hardware prefetchers, such as the Pentium4. 0.5.1 Many thanks to Nelson Chu for fixing ClearArray method, and adding ClearArray to TFMatrix fixing missing initialization of vectors that are not a multiple of 4 0.5.0 Many thanks to Roman "GoodOk" Gudchenko for adding ternary operations: Add, Sub, Mul, Div, MulDiv, Square, AddSquare, Min, Max adding the following operations: Combine, Lerp, InnerSqrSum, NormL1, NormL2, MinMaxValues, IndexMinMaxValues, MinMaxAbsValues, IndexMinMaxAbsValues adding the TFMatrix and TFCube classes building a much-needed benchmarking and testing application performing a lot of testing and benchmarking Optimized prefetching instructions on P4 and Athlon. 0.4.0 Many thanks to Patrick van Laake for adding the ControlFPU unit adding the following methods to the x87 64bit vector: Deg2Rad, Rad2Deg, Sin, Cos, Tan, ArcSin, ArcCos, ArcTan Many thanks to Chris Rorden for making Exentia compilable under Kylix rewriting the SSE2 detection code Removed excessive prefetching instructions. 0.3.0 Many thanks to Robert Lee for optimizing the X87 vector code Rewritten SSE and 3DNow! detection code 0.2.0 Many thanks to Tom Womack and Chris Rorden for designing, coding and testing 64-bit x87 and SSE2 extensions for the Intel Pentium 4 processor (amazing work!!!) 0.1.5 Many thanks to Joao Paulo Schwarz Schuler for fixing the CopyArray and several X87 methods adding testing functions adding the Simple Single Vector Operations (SSVO) interface (included in UPExentia.pas) Notes Chris Rorden has reported that Pentium 4s are really slow when computing NaNs (135 times slower!): users beware! Click here to see more interesting performance data. Are you experienced in SSE and/or 3D Now! programming? Join us! Have you got a SSE, SSE2 or 3D Now! CPU and some spare time to test this library? You're welcome! Let's build together an advanced library that makes Delphi the best tool for floating-point coding!