You may recall that moxie supports two instructions lengths: 16- and 48-bit. Today I'm introducing a few 32-bit instructions as well.
Previously, moxie's "load and store with offset" instructions were defined as...
Load offset, long. Loads into $rA the 32-bit value from memory pointed at by the address produced by adding the 32-bit value following the 16-bit opcode to $rB.
ldo.l 00001100AAAABBBB iiiiiiiiiiiiiiiiiiiiiiiiiiiiiiii
Store offset, long. Stores the 32-bit contents of $rB into memory at the address roduced by adding the 32-bit value following the 16-bit opcode to $rA.
sto.l 00001101AAAABBBB iiiiiiiiiiiiiiiiiiiiiiiiiiiiiiii
In addition to these two, there are versions to load and store 16- and 8-bit values.
In almost every ldo and sto case, however, the upper 16 bits of
the offset were either 0x0000 or 0xFFFF (for negative offsets).
You'll see a lot of these when we access local C variables accessed by
indexing off of $fp.
Shortening the offset from 32 to 16 bits has always been on my radar but I was slow to implement because it's a backwards-incompatible change, and focus has been on tool/core correctness - not optimization. But now it's time to clean things up, and I've implemented this change in all of the upstream moxie tools. I'm also testing a MoxieLite core change as well and should be pushing it to github shortly.
The new ldo and sto instructions look like this (for example)...
Load offset, long. Loads into $rA the 32-bit value from memory pointed at by the address produced by adding the 16-bit value following the 16-bit opcode to $rB.
ldo.l 00001100AAAABBBB iiiiiiiiiiiiiiii
Store offset, long. Stores the 32-bit contents of $rB into memory at the address roduced by adding the 16-bit value following the 16-bit opcode to $rA.
sto.l 00001101AAAABBBB iiiiiiiiiiiiiiii