Also, I've made a test for this function for all float values in C: https://gist.github.com/brunokim/8843039
Unfortunetely it doesn't work in my system, as it does not have other rounding modes available besides the default. If anyone suceeds in running it, please report. On Thursday, February 6, 2014 10:07:40 AM UTC-2, Bruno Kim Medeiros Cesar wrote: > > Just to add a bit to the thread: the Java compiler treats java.lang.Math > differently when more efficient alternatives are available. StrictMath is > used only as a fallback. > > From the java.lang.Math > javadoc<http://docs.oracle.com/javase/7/docs/api/java/lang/Math.html> > : > >> By default many of the Math methods simply call the equivalent method in >> StrictMath for their implementation. Code generators are encouraged to >> use platform-specific native libraries or microprocessor instructions, >> where available, to provide higher-performance implementations of Math >> methods. >> Such higher-performance implementations still must conform to the >> specification for Math. > > > Also, check this StackOverflow > question<http://stackoverflow.com/q/4232231/946814> > . > > As most probably all your versions use the same native libraries or > hardware instructions, the differences must rely either on float > configuration parameters, like rounding modes, or the order of operations. > > On Wednesday, February 5, 2014 1:58:45 PM UTC-2, Glen Fraser wrote: >> >> Thanks for the tip. After reading your comment, I looked and discovered >> the Java library called StrictMath, and tried it (replacing Math/cos and >> Math/sin by the StrictMath versions). I did indeed get different results >> than with the regular library, but unfortunately still not the same answer >> as in other languages. I guess the Java implementation(s) are indeed >> different. It's not a big deal for me, just something I found confusing, >> wondering if I'd done something wrong. >> >> Thanks, >> Glen. >> >> On Wednesday, February 5, 2014 4:06:31 PM UTC+1, Jon Harrop wrote: >>> >>> >>> >>> IIRC, Java provides unusual trigonometric functions which, I’m guessing, >>> Clojure is using. I think the Java ones are actually more accurate (and >>> slower) so you may well find the answer obtained on the JVM is more precise >>> than the others. >>> >>> >>> >>> Cheers, >>> >>> Jon. >>> >>> >>> >>> *From:* clo...@googlegroups.com [mailto:clo...@googlegroups.com] *On >>> Behalf Of *Glen Fraser >>> *Sent:* 05 February 2014 13:17 >>> *To:* clo...@googlegroups.com >>> *Subject:* Confused by Clojure floating-point differences (compared to >>> other languages) >>> >>> >>> >>> (sorry if you received an earlier mail from me that was half-formed, I >>> hit send by accident) >>> >>> >>> >>> Hi there, I'm quite new to Clojure, and was trying to do some very >>> simple benchmarking with other languages. I was surprised by the >>> floating-point results I got, which differed (for the same calculation, >>> using doubles) compared to the other languages I tried (including C++, >>> SuperCollider, Lua, Python). >>> >>> >>> >>> My benchmark iteratively runs a function 100M times: g(x) <-- sin(2.3x) >>> + cos(3.7x), starting with x of 0. >>> >>> >>> >>> In the other languages, I always got the result *0.0541718*..., but in >>> Clojure I get *0.24788989*.... I realize this is a contrived case, but >>> -- doing an identical sequence of 64-bit floating-point operations on the >>> same machine should give the same answer. Note that if you only run the >>> function for about ~110 iterations, you get the same answer in Clojure (or >>> very close), but then it diverges. >>> >>> >>> >>> I assume my confusion is due to my ignorance of Clojure and/or Java's >>> math library. I don't think I'm using 32-bit floats or the "BigDecimal" >>> type (I even explicitly converted to double, but got the same results, and >>> if I evaluate the *type* it tells me *java.lang.Double*, which seems >>> right). Maybe Clojure's answer is "better", but I do find it strange that >>> it's different. Can someone explain this to me? >>> >>> >>> >>> Here are some results: >>> >>> >>> >>> *Clojure: ~23 seconds* >>> >>> (defn g [x] (+ (Math/sin (* 2.3 x)) (Math/cos (* 3.7 x)))) >>> >>> (loop [i 100000000 x 0] (if (pos? i) (recur (dec i) (g x)) x)) >>> >>> ;; final x: *0.24788989279493556 **(???)* >>> >>> >>> >>> *C++ (g++ -O2): ~4 seconds* >>> >>> double g(double x) { >>> >>> return std::sin(2.3*x) + std::cos(3.7*x); >>> >>> } >>> >>> int main() { >>> >>> double x = 0; >>> >>> for (int i = 0; i < 100000000; ++i) { >>> >>> x = g(x); >>> >>> } >>> >>> std::cout << "final x: " << x << std::endl; >>> >>> return 0; >>> >>> } >>> >>> // final x: *0.0541718* >>> >>> >>> >>> *Lua: ~39 seconds* >>> >>> g = function(x) >>> >>> return math.sin(2.3*x) + math.cos(3.7*x) >>> >>> end >>> >>> >>> >>> x = 0; for i = 1, 100000000 do x = g(x) end >>> >>> -- Final x: *0.054171801051906* >>> >>> >>> >>> *Python: ~72 seconds* >>> >>> def g(x): >>> >>> return math.sin(2.3*x) + math.cos(3.7*x) >>> >>> >>> >>> x = 0 >>> >>> for i in xrange(100000000): >>> >>> x = g(x) >>> >>> >>> >>> # Final x: *0.05417180105190572* >>> >>> >>> >>> *SClang: ~26 seconds* >>> >>> g = { |x| sin(2.3*x) + cos(3.7*x) }; >>> >>> f = { |x| 100000000.do{ x = g.(x) }; x}; >>> >>> bench{ f.(0).postln }; >>> >>> // final x: *0.054171801051906* (same as C++, Lua, Python; different >>> from Clojure) >>> >>> >>> >>> Thanks, >>> >>> Glen. >>> >>> >>> >>> -- >>> You received this message because you are subscribed to the Google >>> Groups "Clojure" group. >>> To post to this group, send email to clo...@googlegroups.com >>> Note that posts from new members are moderated - please be patient with >>> your first post. >>> To unsubscribe from this group, send email to >>> clojure+u...@googlegroups.com >>> For more options, visit this group at >>> http://groups.google.com/group/clojure?hl=en >>> --- >>> You received this message because you are subscribed to the Google >>> Groups "Clojure" group. >>> To unsubscribe from this group and stop receiving emails from it, send >>> an email to clojure+u...@googlegroups.com. >>> For more options, visit https://groups.google.com/groups/opt_out. >>> >> -- You received this message because you are subscribed to the Google Groups "Clojure" group. To post to this group, send email to clojure@googlegroups.com Note that posts from new members are moderated - please be patient with your first post. To unsubscribe from this group, send email to clojure+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/clojure?hl=en --- You received this message because you are subscribed to the Google Groups "Clojure" group. To unsubscribe from this group and stop receiving emails from it, send an email to clojure+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/groups/opt_out.
