Age | Commit message (Collapse) | Author |
|
Change-Id: Iffdfed00ac49a7920844fa5f9e1a3ffc6a77e641
|
|
- use %zu for size_t operands
- suppress unused argument warning messages
Change-Id: Ib39b8763414320d9e3a29408bc1b396ddb376293
|
|
- Created macros for one and two buffer benchmarks to simplify
addition of new benchmarks.
- Created cold versions of all benchmarks that work on buffer benchmarks.
This means that the benchmark is given buffers that are not in the cache.
- Verified the cold versions are working as intended by
changing it so that the same buffer is written in each loop
instead of the strided buffer. Using this method, the cold and
normal benchmark run at the same speed.
Change-Id: Ie5d1b015dd70d3935cfa2830a82955429a2f5b3c
|
|
Change-Id: I31542ecdec95c2d1d3fb9b2795581ad4d5e13c5c
|
|
List of all changes:
- Add a strcpy benchmark.
- Create a macro that is used to create all of the code for the main
loop of the benchmark.
- Fix the strcmp benchmark to use two buffers. The single buffer was
an experiment that should not have been committed. It also crashes
when a large alignment is used because it hard-coded an extra 2048
allocation regardless of the alignment chosen.
- Modify the libc benchmarks to take the function to bench as a
parameter. This will allow us to benchmark new potential libc
optimizations by adding a single line to the function table list.
Change-Id: Id7129667669318acbd001c1bfa3db82764a39029
|
|
NOTE: After this change memcpy results will be significantly lower
for larger sized copies.
Changes:
- Addition of strcmp benchmarking.
- Fix memcpy to initialize memory before copy. If this is not done, the
results of memcpy are artificially higher.
- Add a min and max when printing averages.
- Add two new options --dst_or_mask and --src_or_mask to allow more
flexibility when testing different alignments.
Change-Id: Ia2859d5569ab5ffa35e5ac0f6d5d1f248fef6533
|
|
Moving the code to cpp to access the cpuset CPU* macros (these
macros are defined in sched.h inside of __USE_GNU which is not
defined for the thumb C compiler). The C++ code is also slightly
easier to read.
Add code to set the priority of the process to the highest value.
Add code to lock the process to a single cpu.
Add the ability to compute average and standard deviation over
a number of iterations.
Change the timing code to use nanosecond resolution timing.
Add options to allow modification of the alignment of the src/dst
pointers for memcpy and the dst pointer for memset.
Add an option to change the size of the data being copied in each
iteration.
Change-Id: Ib7c50ed4463f94e638eb81690fe8fe0d0bc3ea80
|