Add support for high-precision timers?

[juj]

I did some tests with different ways to compute the time:

• time(NULL) http://www.cplusplus.com/reference/clibrary/ctime/time/
• clock() http://www.cplusplus.com/reference/clibrary/ctime/clock/
• clock_gettime() http://www.tin.org/bin/man.cgi?section=3&topic=clock_gettime The page lists four clock types: CLOCK_REALTIME, CLOCK_MONOTONIC, CLOCK_PROCESS_CPUTIME_ID and CLOCK_THREAD_CPUTIME_ID.

I could not find CLOCK_PROCESS_CPUTIME_ID or CLOCK_THREAD_CPUTIME_ID to be defined, so they were omitted from the test. The resulting application is live at http://dl.dropbox.com/u/40949268/emcc/Geometry_time.html

The results:

• time(NULL) has precision of one second, as expected.
• clock() has precision of 1000 ticks in one second, and CLOCKS_PER_SEC has the value 1000.
• clock_gettime(CLOCK_REALTIME) and clock_gettime(CLOCK_MONOTONIC) always return 0. These are unsupported by emscripten?

The best way so far is to use clock() to get at least millisecond precision. What is the status of doing high-performance timers on web? Googling a bit:

There's at least this: http://www.w3.org/TR/hr-time/

But I suppose the reality today looks like this? http://stackoverflow.com/questions/6875625/does-javascript-provide-a-high-resolution-timer

Perhaps making two functions to enable code to adapt to different granularities: emscripten_clock_tick() returns a current wallclock time in best available timer in the browser, and emscripten_clock_ticks_per_sec() which would return the frequency that timer is running at? Alternatively (additionally?), just make use of clock_gettime(CLOCK_REALTIME) directly, instead of returning 0?

I would be happy to get access to a high-precision timer. That would allow me to bring my code performance profiler over ( http://dl.dropbox.com/u/40949268/Asteroids/Asteroids-Tegra3.png ) to measure the performance in the browser.

[kripken]

Firefox and Chrome dev versions have window.performance.now() which is high-resolution. There seems to be no libc or SDL API that is suitable to expose it, so I guess we need to provide our own.

[kripken]

Implemented in incoming as emscripten_get_now.

[juj]

Many thanks!

The commit https://github.com/kripken/emscripten/commit/354a88d763284fae92a59d311f8562542ad10a97 also cleared up for me how to go implementing my own new functions, so that I can start hacking on the API :)

[kripken]

Merged to master.

Get hacking! :)