suggestion for ptxas improvement.

Not sure where to direct this kind of feedback, so I post it here.

  1. It would be nice to make ptxas reference subwords of a 64bit register with some simple syntax, for example:
    .reg.b64 %rd1;
    add.u32 %rd1.hi, %rd1.hi,%rd1.lo

  2. It would be nice to allow empty statements, i.e - a sequence of ‘;’ which is not currently allowed. This is very handy when you use preprocessor, where you can end-up with a bunch of trailing ‘;’

  3. It would be nice to support vector instructions, like add.v2.f64, to avoid repetetive typing of add.f64 for each component individually.

  1. A very cool feature would be a predicated block. Imagine you have a macro like this:
#define my_add(a,b,c) { add.f64 a.x, b.x, c.x; add.f64 a.y, b.y, c.y; }

which you can’t really use with a predicate. But if you’d have a whole block predication, which in turn just propagates the predicate into each instruction would be really handy.