# syntax is param_name=value1,[value2,...]
# value can be string for some parameters, or numerical for the rest
# numerical is:
# - single number: 500
# - range: 1-20 (with default linear step 1)
# - range with linear step: 0-20/4 (means 0,4,8,12,16,20)
# - range with exponential step: 1-8^2 (means 1,2,4,8)
# all numbers can have suffix 'k' or 'm'
# rib will benchmark all possible combinations of parameter values

## common parameters for all benchmarks
# warmup results to throw away before each collected sample (not used in cache_lzw)
warmup_count=10
# samples to collect for each combination of parameters
samples_count=50
# optional, list of performance events to measure
# see http://icl.cs.utk.edu/projects/papi/presets.html for supported events
# currently implemented in: cache_artifical, cache_fft
papi_events=PAPI_L1_DCM,PAPI_L1_ICM,PAPI_L2_DCM,PAPI_L2_ICM

## cache trashing parameters - common to cache_* benchmarks
# cache trashing method
# 1 - linear sweep of trashing area
# 2 - pointer-based random permutation count (trashes in random order)
#   	for cache_artifical requires cnt_method=1 || no count/trash area overlap || (cnt_step > 1 && tsh_step > 1) );
#		so the pointers for trashing and counting dont mix up	
tsh_method=1
# start of area for trashing relative to allocated buffer size (bytes)
tsh_shift=0k-512k/4k
# size of trashing area (bytes)
tsh_size=512k
# how much of memory trash (bytes), must be >=tsh_size
# for linear trash - the begining of trashing area of tsh_length is swept
# pointer-based trash - determines how many pointers to follow
tsh_length=512k
# step when advancing to next array element (bytes, aligned to sizeof(uint))
tsh_step=16
# amount of L1 instruction cache trashing, before performing general L2 trashing (bytes)
tshic_length=32k

# benchmark/thrashing procedure type
# - cache_artifical - measured workload is making sum of integers array
# - cache_lzw		- measured workload is lzw based on ncompress
# - cache_fft		- measured workload is fft
# - cache_ray		- raytracing benchmark
benchmark=cache_artifical
thrashing=cache_fft

[benchmark]
[thrashing]

## following parameters are for cache_artifical only
# 1 - old linear count of array containing random integers
# pointer-based - each integer in array is also treated as pointer to next integer
# 2 - pointer-based linear count
# 3 - pointer-based random permutation count
cnt_method=1
# start of area for counting relative to allocated buffer size (bytes)
cnt_shift=0
# size of counting area (bytes)
cnt_size=512k
# step when advancing to next array element (bytes, aligned to sizeof(uint))
cnt_step=1

## following parameters are for cache_lzw only
# name of file to compress
compress_file=something.tar
# maximal code length -> affects max dictionary size, valid values are 9 to 16
compress_maxbits=9-16
# compress reads input / writes output by 8k blocks, following parameters control after how many reads/writes trashing occurs
tsh_read_rate=4
tsh_write_rate=4

## cache_fft specific
# 1 = fft.c benchmark, 2 = fftw
fft_method=2
# fft_method = 1 => total memory allocated for fft (bytes) = 2*size of input data
# fft_method = 2 => size of input data (bytes)
fft_size=512k

## cache_ray specific
# complexity of raytraced scene
ray_levels=9
# x and y dimension of result (pixels)
ray_size=512