noise--; do speedup comparison with lrb data |
|
More...
|
over 15 years ago
|
fix benchmark; more readable plots; speedup plot |
|
More...
|
over 15 years ago
|
add function to add points to a dotchart; make mychart3 more flexible with nicer auto-color; allow processData to skip processing of some cols; add permute function |
|
More...
|
over 15 years ago
|
try to run benchmarks with SCHED_FIFO, but with a watchdog in the backgroun |
|
More...
|
over 15 years ago
|
less mem usage |
|
More...
|
over 15 years ago
|
remove tmp file |
|
More...
|
over 15 years ago
|
add compare benchmark |
|
More...
|
over 15 years ago
|
use cpuid instruction to find the correct sizes for L2, L1 and Cacheline size |
|
More...
|
over 15 years ago
|
add a function to reduce a data set to median, mean and stddev. a sort function and another mychart function that can use those data frames and create the plots I want with very little writing work |
|
More...
|
over 15 years ago
|
enable benchmark plots generation only when compiling in Release mode |
|
More...
|
over 15 years ago
|
update plot scripts when their source files changes and get the dependecies right |
|
More...
|
over 15 years ago
|
fix using LRB data, nchar is the right function to use |
|
More...
|
over 15 years ago
|
don't fail when the LRB prototype header is not available |
|
More...
|
over 15 years ago
|
no build type needs check for -O[123] in CXX_FLAGS and set NO_OPTIMIZATION accordingly |
|
More...
|
over 15 years ago
|
allow dox creation in other directories, too; provide custom CSS |
|
More...
|
over 15 years ago
|
create dot charts from the benchmarks |
|
More...
|
over 15 years ago
|
add mask benchmark; currently benchmarks only conditional assignment |
|
More...
|
over 15 years ago
|
add support form random floats [0..1] and use +rand() to keep the compiler from optimizing too much |
|
More...
|
over 15 years ago
|
just to be sure: we don't want gcc to make statics threadsafe for us because it's not portable anyway |
|
More...
|
over 15 years ago
|
add TSC based timings |
|
More...
|
over 15 years ago
|
make the RNG useable outside of the Gather benchmark |
|
More...
|
over 15 years ago
|
also benchmark short_v and sfloat_v gathers |
|
More...
|
over 15 years ago
|
use a precomputed index/mask array instead of calculating them to get better comparability between the different implementations. TODO this kills the caches where we actually want to work on. |
|
More...
|
over 15 years ago
|
not executable |
|
More...
|
over 15 years ago
|
stupid isfinite/isnan macros; std::log(float) apparently returns double on some compilers |
|
More...
|
over 15 years ago
|
remove incorrect QueryPerformanceFrequency usage |
|
More...
|
over 15 years ago
|
add masked gathers benchmarking; use different memory for the tests to use uncached values |
|
More...
|
over 15 years ago
|
fix dependency on __m128 to be stored in memory before entering the bsf loop |
|
More...
|
over 15 years ago
|
test that masked gathers retain the old values correctly |
|
More...
|
over 15 years ago
|
add _m mask typedefs |
|
More...
|
over 15 years ago
|