<r>QuoteOriginally posted by: AVtif (v < 0.5f) { z = -logf(2.0f*v); return - z * CQP(z) / CQQ(z); } else { z = -logf(2.0f*(1.0f-v)); return z * CQP(z) / CQQ(z); }Each branch in this ICNDF code would compile to several dozen machine instructions, far exceeding the threshold of 7 instructions, and so ...

<t>QuoteOriginally posted by: renormI use predicate reduction with SSE too. It works only if both branches have more or less equal runtime and both are cheap at the same time.With traditional SIMD instructions (e.g. SSE), you have to manually do predicate reduction. It becomes the programmer's respo...

<t>QuoteOriginally posted by: renormI thought that OOP (a.k.a GOF design patterns) is not very relevant where CUDA is used. Does CUDA support virtual functions?Personally, I don't see much use for fancy OOP constructs in CUDA compute kernels (template programming and structs, however, are quite rele...

<t>QuoteOriginally posted by: renormSSE is easier than CUDA. CUDA vs SSE is not a clear-cut even for non-low latency algorithms. Optimization with stream processing restrictions doesn't seem to be an easy task. Learning curves are very different too. STL, Boost, templates and vanilla OOP are not ava...

<t>QuoteOriginally posted by: renormAnything which has to store and move tons of data won't speedup on GPU. Multifactor Longstaff-Schwartz is one such example. Longstaff-Schwartz method requires tons memory to store paths and significant CPU time to perform linear regression at every time step. Matr...

<r>You can think of a GPU as an ultra-wide SIMD vector processor. SIMD means `single instruction multiple data.` An Intel/AMD SSE3 core has SIMD vector width of 128 bits. This is called `short-vector` SIMD architecture. It means that each instruction in a SS3 SIMD core operates simultaneously on fou...

QuoteOriginally posted by: louist1013Numpy can be build with BLAS or LAPACK. I don't remember which I have it built with, nor have I done any speed comparisons against Matlab or straight up C/C++ stuff.You can also build NumPy with the Intel Math Kernel Library.

<t>In Monte Carlo option pricing, it is well known that the variance of finite difference Greek estimates can be reduced by sharing common random numbers between valuations. For example, if I want to estimate Delta of some Monte Carlo option pricer then I should find the value of the option at diffe...

<t>Ok I take that back. Perhaps the inner workings of the library SHOULD be written in C++, though a C interface is still useful for properly bind with other languages. My main complaint against QuantLib is not that it's written in C++, but rather that it's over-burdened with certain design patterns...

<t>Yes, QuantLib is over-engineered. An open-source finance library should be written in C, in the style of well-established numerical routines (such as those found in Numerical Recipes in C). The C implementation can then easily be augmented with bindings to higher level languages such as C++, Pyth...

<r>One option is to use HDF5 (<URL url="http://www.hdfgroup.org/">http://www.hdfgroup.org/</URL>). This is just a file format and not a database. However, it is extremely efficient for file I/O and the main bottleneck with tick data is file I/O (since hard disk access is many times slower than RAM)....

C/C++ is still the best choice of cutting-edge software and will continue to be for a very long time.

<t>Hi ClosetChartist,I thought you had some very interesting comments, and I was hoping to discuss these points a little further... QuoteOriginally posted by: ClosetChartistmarkhadley:Your general advice is reflected in several US consulting firms' annuity hedging platforms: multi-factor Hull-White,...

Hello,Does anyone have an electronic copy of the paper "The Full Monte," Boris Moro, Union Bank of Switzerland, Risk magazine 1995? I would greatly appreciate it. The paper concerns transforming Uniform(0,1) random numbers to Normal(0,1) random numbers by way of Moro's Inversion. Thanks

<t>Another related issue ... In equation (9) a formula is given for the "forward cap price." It is the difference between the cap prices computed for two consecutive maturities. The two computations of these two cap prices are based on the corresponding "flat volatility" quotes from the dealer ... B...