We are striving to develop a revolutionary development tool that can really put MMX/SSE programming into mainstream. We would greatly appreciate any feedback on Quexal, as any comment, feature request and bug report would help us design a product that better fulfills your MMX/SSE programming needs. Click here to send an email! Frequently Asked Questions   Would MMX/SSE instructions accelerate my application? MMX/SSE instructions can greatly enhance the performance of the following applications: multimedia (audio/video), communications, DSP kernels, 2D and 3D graphics, image processing and speech recognition. Nowadays, all these kinds of applications need to exploit MMX/SSE instructions to offer the level of performance that users expect.  And Quexal is the only tool that can get the job done quickly and easily.   What are the differences between the Demo and the Registered versions of Quexal? The Demo version is identical to the Registered version except that:  the Save File and Save as Macro functions are disabled;  the compiler will stop after the first solution; several compiler settings are fixed; the Export to Intrinsics function is disabled; it can be used for evaluation only.  Don't overlook difference #2: the first solution is hardly optimal, and it is typical that medium-size routines enjoy tens of improvements over the first solution when compiled with the Registered version.   Why don't you add SSE floating point instructions to Quexal? The current focus is on integer SIMD programming. Extending the Quexal environment to floating-point SIMD may be a natural upgrade path, but there are several reasons that may not make it worthwhile: it is easier to write FP vector compilers than integer vector compilers: in FP code you have only one size of operands and a fixed level of parallelism (4 with SSE, 2 with 3D-Now!); another cause is that some MMX/SSE instructions do not match well with C language syntax: consider the PSADBW instruction, it replaces a whole bunch of C code, but the coding style of the C programmer may not make the match obvious. It is also quite difficult to match a C multiplication to one of MMX multiplies (high or low part?), as it requires a deep analysis of the dynamic ranges of the operands and of the following instructions; the Intel Pentium 4 processor has fast SSE but slow x87 FPU: it means that mainstream compiler vendors will have to implement FP vectorizers to fully harness P4's power. Intel offers C and Fortran SSE vectorizing compilers today, others are likely to follow.