search

Sable / Ostrich

Benchmark suite for studying the performance of JavaScript and WebCL for numerical computation
Other
24 stars 12 forks source link

Establish structure for supporting multiple compiler projects with the suite #25

Closed elavoie closed 8 years ago

gnuvince commented 8 years ago

For MatJuice, we had a nice split of the few MATLAB benchmarks we had converted where we had the driver function in one file and the "work" function in another. That allowed me to compile the core computation without having to worry about all the extra MATLAB builtins used to read/write/gen random numbers that MATLAB used. Does it make sense to do it for the other implementations too? I know that with @taytoe we ran into a lot of weird issues compiling the C code with PNaCl and along the way we found some pretty doubtful patterns of code (e.g. a C file containing only one line of code, #define fp double)

elavoie commented 8 years ago

@gnuvince, Yes I intend to follow that structure for the different language implementations, hopefully that should simplify the PNaCl benchmark implementations.

elavoie commented 8 years ago

The lud benchmark with support for JS, C, and Matlab version, as well as the 'compiler' and 'environment' examples in the 'matlab-version' branch show how the separation was done.