forked from ridiculousfish/libdivide
-
Notifications
You must be signed in to change notification settings - Fork 0
/
README.txt
42 lines (28 loc) · 3.25 KB
/
README.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
libdivide is a "library" for optimizing integer division. See http://libdivide.com for more information on libdivide.
This is summary of how to use libdivide's testing tools to develop on libdivide itself.
libdivide proper consists of a single header file, libdivide.h, which compiles as both C and C++.
libdivide has two test tools: a verification utility "tester", and a benchmarking utility "benchmark". The verification utility is used to help ensure that the division algorithm is correct, and the benchmarking utility is used to measure the speed increase. On Mac OS X, Linux, and other Unix-like systems, you can build both with the Makefile. On Windows, there is a Visual C++ 2010 project that can build both, in windows/libdivide_Win32
To build the tester via the Makefile, build one of the following targets:
debug: builds the tester without optimization
release: builds the tester with optimization
Both build an executable "tester". You can pass it one or more of the following arguments: u32, s32, u64, s64, to test the four cases (signed, unsigned, 32 bit, or 64 bit), or run it with no arguments to test all four. The tester is multithreaded so it can test multiple cases simultaneously. The tester will verify the correctness of libdivide via a set of randomly chosen denominators, by comparing the result of libdivide's division to hardware division. It may take a long time to run, but it will output as soon as it finds a discrepancy.
The benchmarking utility is built with target "benchmark." You may pass it one of the same arguments (u32, s32, u64, s64) to compare libdivide's speed against hardware division.
"benchmark" tests a simple function that inputs an array of random numerators and a single divisor, and returns the sum of their quotients. It tests this using both hardware division, and the various division approaches supported by libdivide, including vector division.
It will output data like this:
# system scalar scl_us vector vec_us gener algo
1 5.733 0.849 0.580 0.431 0.431 1.663 0
2 6.716 0.847 0.580 0.431 0.431 1.663 0
3 6.687 1.425 1.427 1.862 1.444 22.156 1
4 6.668 0.851 0.580 0.431 0.431 1.663 0
5 6.697 1.425 1.425 1.837 1.425 22.156 1
...
It will keep going as long as you let it, so it's best to stop it when you are happy with the denominators tested. These columns have the following significance. All times are in nanoseconds, and lower is better.
#: The divisor that is tested
system: Hardware divide time
scalar: libdivide time, using scalar functions
scl_us: libdivide time, using scalar unswitching functions
vector: libdivide time, using vector functions
vec_us: libdivide time, using vector unswitching
algo: The algorithm used. See libdivide_*_get_algorithm
The benchmarking utility will also verify that each function returns the same value, so "benchmark" is valuable for its verification as well.
Before sending in patches to libdivide, please run the tester to completion with all four types, and the benchmark utility for a reasonable period, to ensure that you have not introduced a regression. Happy hacking!