• d-bugmail puremagic.com (53/53) Feb 23 2018 https://issues.dlang.org/show_bug.cgi?id=18511

d-bugmail puremagic.com writes:

https://issues.dlang.org/show_bug.cgi?id=18511

          Issue ID: 18511
           Summary: Using std.range / std.algorithm templates cause big
                    slowdown in compilation time
           Product: D
           Version: D2
          Hardware: All
                OS: All
            Status: NEW
          Severity: enhancement
          Priority: P1
         Component: dmd
          Assignee: nobody puremagic.com
          Reporter: hsteoh quickfur.ath.cx

Code:
------
double dot(double[] a, double[] b) {
        version(fast) {
                assert(a.length == b.length);
                double acc = 0.0;
                foreach (i; 0 .. a.length)
                        acc += a[i] * b[i];
                return acc;
        }
        version (slow) {
                import std.range : zip;
                import std.algorithm.iteration : fold, map;

                return zip(a[], b[])
                        .map!(x => x[0]*x[1])
                        .fold!((a, b) => a + b)(0.0);
        }
}
------

Compiling with manually-written loop:
------
$ time dmd -c -version=fast test.d
real    0m0.021s
user    0m0.014s
sys     0m0.007s
------

Compiling with fancy std.algorithm / std.range templates:
------
real    0m0.499s
user    0m0.444s
sys     0m0.054s
------

Now, I understand that using fancy generic code templates requires more work on
the part of the compiler, so some degree of slowdown in compilation times is to
be expected.

However, this is an *order of magnitude* slowdown in compiling two
functionally-equivalent pieces of code, and very simple code at that.  Given
our current fast-fast-fast slogan, I think this is unacceptable.

--