Most Well-rounded Hash

Qiming Hou hqm03ster@gmail.com
http://www.houqiming.net/

Judges' comments:

To use:

make

echo IOCCC | ./prog

Try:

echo "" | ./prog

Selected Judges Remarks:

A good start for understanding this entry might be https://www.ietf.org/rfc/rfc1321.txt. How does it compute this using only a single looping construct? The original reference source code for this algorithm is about 355 lines of code, yet this entry is only 199 lines with one statement per line and it isn’t using lots of macro expansion tricks like the reference code!

Some things that make you go, “huh?”

./prog < prog.c | cut -c-5
./prog < Makefile | cut -c-4

How are those values magically returned?

What happens with a large file? Something over 256K

./prog < large_file

Where does that message come from?

Author’s comments:

This looks like a normal C program.

Almost everything is a double.

There are tell-tale signs like pow(2,(?/12)) and exp(-?*?) (names redacted). You can also find a bunch of cosines and polynomial evaluations.

It looks like a ????? ???????????, right?

Run and find out.

Why obfuscated

Some poor scripting folks only have access to doubles (Jav?script, L?a, BA?IC, etc.).

As a C programmer, I feel obliged to show a gesture of sympathy, to experience their pain and provide an answer.

Except when absolutely necessary (printf, int main(), array indexing), there isn’t a single integer. Even getchar() and EOF are immediately converted.

Since bool counts as an integer type too, there is no boolean either. No comparison, no if, no ?:, no || &&. Everything is done in a single loop that terminates on printf. Straight up arithmetic.

floor and ceil are technically double-to-double functions, but they feel like cheating. Yes cos, no floor.

Limitations

Hard requirements:

The platform must implement the double type as IEEE754-compliant 64-bit floating point numbers. The 80-bit intermediate format used by x87 is considered as an violation of this. The code should print an error message on such platforms.
The program must start with the CPU / FPU in round-to-nearest mode.

Soft requirements:

The compiler must respect volatile. The code is formatted to warn about that, though.
The printed result is only correct on little-endian machines. The program takes care to warn about this issue after printing an incorrect big-endian result. Error messages become garbled, though.

Tested platforms:

Linux 3.18.7, Rasperberry Pi 2, clang 3.0-6.2, (works)
Linux 3.18.7, Rasperberry Pi 2, gcc 4.6.3, (works)
Mac OS X 10, x64, Apple LLVM version 6.1.0 (clang-602.0.53), (works)
Windows 8.1, x86, MSVC 2012 (works)
Windows 8.1, x64, MSVC 2012 (works)
Windows 8.1, x86, MinGW gcc 4.6.2 (fails, prints error message)
Windows 8.1, x86, MinGW gcc 4.6.2 with -msse2 -mfpmath=sse (works)