On 17 January 2012 13:01, John Gilmore <[email protected]> wrote:

> The cross-platform emphasis in PL/X discourages and was intended to
> discourage this sort of thing; but optimizing machinery, less
> important in PL/S because resort to assemb ly language was possible,
> does not appear to be much used by PL/X.   (Even such obvious things
> as moving common subexpressions out of loops and suppressing redundant
> subscript arithmetic don't seem to happen.)
>
> IBM knows and has always known how to fix this problem; what has been|is 
> lacking is the will to do it.

IBM almost a decade ago fixed the problem where it really counts - in
its millicode. That is generated by the GCC suite, with a
private-to-IBM PL8 language front end, and a published(?) middle-end
optimizer and back end code generator. One bumps into the odd
non-millicode module compiled by PL8's CMS-hosted predecessor
implementation, PL.8, but porting of the whole GCC and Linuxy
infrastructure, including ELF object format and other baggage, into
the likes of z/OS would presumably be required before the current PL/X
could be replaced and its code optimized. An outsider can only imagine
the internal geopolitical goings on with regard to all this.

Tony H.

Reply via email to