In wasm code, we sometimes see the pattern

<some 64 bit expression>
i32.wrap/i64
i32.load

where we generate an instruction to extend the 32 bit offset into a zero
extended 64 bit value for the actual load. However, the preceeding
truncate already yields a zero extended 32 bit value, so the extra
instruction is not needed. Even more, it might get in the way of
munching more computation into the final load.

This change adds information about the zero extending behavior to
the existing optimization that avoids the zero extension.

Bug: chromium:853685
Change-Id: Iab9179379923ecb88651df6091b3d9408341cf4c
Reviewed-on: https://chromium-review.googlesource.com/c/1421839Reviewed-by: Michael Starzinger <mstarzinger@chromium.org>
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Commit-Queue: Stephan Herhut <herhut@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58998}