Hi Damian, I think the issue here is that `anArray/aScalar` does not literally result in an array but a promoted expression as you have observed. The rationale for that is to avoid creating arrays as much as possible, and try to do that if the user explicitly asks for it (as you did in your workaround). Some more discussion about this can be found in:
https://github.com/chapel-lang/chapel/issues/14364 And to some extent some other issues that are linked from that one. Hope this helps, Engin On 7/17/20, 8:03 AM, "Damian McGuckin" <[email protected]> wrote: It would be nice to know my mistake in the below: With 1.22, the following does not compile when I pass a vector expression to the to the little routine _5tuple. // ?R is some real(?w) floating point number // returns a 5-tuple if given a 5-element 1D array inline proc _5tuple(const in t : [?tD] ?R) { return (t[0], t[1], t[2], t[3], t[4]); } // shift from bottom 2x2 minor - Golub+Reinsch but scaled as per LINPACK inline proc shift(j : int, k : int, ref e : [?D] ?R, ref q : [D] R) { const qe = [ q[j], q[k - 1], q[k], e[k - 1], e[k] ]; // const qeNormalized = qe / (max reduce abs(qe)); const (_x, _y, _z, _g, _h) = _5tuple(qe); const (x, y, z, g, h) = _5tuple(qe / (max reduce abs(qe))); const t = (half * ((y - z) * (y + z) + (g - h) * (g + h))) / (h * y); const s = sqrt(t * t + one); const f = if t < zero then t - s else t + s; : return (_x, ((_x + _z) * (x - z) + _h * ((y / f) - h)) / x); } The compiler tells me: sa.chpl:13: In function 'svd': sa.chpl:32: error: unresolved call '_5tuple(promoted expression)' sa.chpl:20: note: this candidate did not match: _5tuple(t: []) sa.chpl:32: note: because call actual argument #1 with type iterator sa.chpl:20: note: is passed to formal 'const in t: []' sa.chpl:410: Function 'svd' instantiated as: svd(a: [domain(2,int(64),false)] real(64), w: [domain(1,int(64),false)] real(64), v: [domain(2,int(64),false)] real(64)) Why does it consider that I am sending a promoted expression? An expression yes. But promoted? The expression is the normalization of a 5-element vector and yields a 5-element array. No promotion involved. If I normalize that array myself into another constant 5-slement array, it works. For example // shift from bottom 2x2 minor - Golub+Reinsch but scaled as per LINPACK inline proc shift(j : int, k : int, ref e : [?D] ?R, ref q : [D] R) { const qe = [ q[j], q[k - 1], q[k], e[k - 1], e[k] ]; const qeNormalized = qe / (max reduce abs(qe)); const (_x, _y, _z, _g, _h) = _5tuple(qe); const (x, y, z, g, h) = _5tuple(qeNormalized); const t = (half * ((y - z) * (y + z) + (g - h) * (g + h))) / (h * y); const s = sqrt(t * t + one); const f = if t < zero then t - s else t + s; return (_x, ((_x + _z) * (x - z) + _h * ((y / f) - h)) / x); } I can live with this but it seems like array algebra expressions cannot appear in parameter lists. Or is my problem that there are generic arguments? Or is my problem just me? I can send the complete code if needed. Thanks - Damian Pacific Engineering Systems International, 277-279 Broadway, Glebe NSW 2037 Ph:+61-2-8571-0847 .. Fx:+61-2-9692-9623 | unsolicited email not wanted here Views & opinions here are mine and not those of any past or present employer _______________________________________________ Chapel-developers mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/chapel-developers _______________________________________________ Chapel-developers mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/chapel-developers
