In particular, the attached comment noting that JITting is the basis of our modern CPUs is poignant when we consider that it is to GPUs that we look for top performance, nowadays. While those are programmed via compilation-at-runtime, for historical reasons, there is no attempt at interpretation or runtime analysis; regular AOT code is compiled, typically at startup, and run natively. The execution model is such that tricky optimizations are not needed to get maximum performance (so compilation can be quick); instead, the key is to always have plenty of other work to do during what would otherwise be stalls, relying on what amounts to massive hyperthreading.
Massive hyperthreading is another dynamic, runtime allocation of resources, but of a wholly different order than JITting. In particular, it has no concept of data type operated on being unknown until runtime, although vtable-style late binding, dispatching according to specific, known types, works.
> [...] even when run in the most controlled of circumstances, small, deterministic, widely studied microbenchmarks often fail to reach a steady state of peak performance on a variety of common VMs.
In Davis' Matrix score there's a minor triad with its corresponding VI chord that fades in and out atop it. Plus a similar moment with a minor triad and a major triad built on its corresponding leading tone (which seems to be a staple of Marvel movie scores these days[1]).
In the Wagner link I gave, it starts with a half-diminished seventh chord oscillating between the fully-diminished neighbor that shares three common tones with it. Then you hear the trumpet play the rising Siegfried theme over that half-diminished chord. Now, Wagner could have harmonized that theme so it culminated on the same kind of VI chord that you hear in the first Matrix movie. (He probably did harmonize it that way at one of the other zillion appearances of that theme.) Instead, here Wagner uses chromatic voice leading to expand that half-diminished chord out to a minor triad. I always think of that moment as a kind of "Evil Phrygian Cadence"-- it's what you'd get if you sent a regular Phrygian cadence through that mirror Neo touches in the first Matrix movie. :)
Of course it's Wagner, so on the other side of the cadence is yet another mirror Neo touches, and everything happens again just kicked up a notch (transposed up a major third). Then three more mirrors, and finally the music slowly shifts to change character for the next scene.
Whoever filmed the Boulez version of Götterdämmerung started a bit before that excerpt with the camera zoomed in as close as possible on Hagen's face, panned out for half the ensuing instrumental section as the stage changed, then panned back in for the second half. If you can find it it's hilarious to watch how far out they have to pan to accommodate the pace of the music. I think it underscores the difference between an opera-- where the visuals serve the music[2], and film where it's the other way around.
I find it odd that I don't hear Wagner's evil Phrygian cadence used in recent film scores. Anyone have any examples I may have missed?
1: In the Marvel movies, those two chords alternate. In the Matrix, the VI chord gets stacked on top of the first chord, as if a 2nd brass section is constantly playing the 2nd chord and the mixing engineer is just fading them in and out of the mix.
2: And in the case I mentioned I wonder if it fails to work because there aren't any tried-and-true cinematic techniques to accommodate such a slow musical transition.
In particular, the attached comment noting that JITting is the basis of our modern CPUs is poignant when we consider that it is to GPUs that we look for top performance, nowadays. While those are programmed via compilation-at-runtime, for historical reasons, there is no attempt at interpretation or runtime analysis; regular AOT code is compiled, typically at startup, and run natively. The execution model is such that tricky optimizations are not needed to get maximum performance (so compilation can be quick); instead, the key is to always have plenty of other work to do during what would otherwise be stalls, relying on what amounts to massive hyperthreading.
Massive hyperthreading is another dynamic, runtime allocation of resources, but of a wholly different order than JITting. In particular, it has no concept of data type operated on being unknown until runtime, although vtable-style late binding, dispatching according to specific, known types, works.