Agner`s CPU blog

I'm really curious about how the hardware would be able to do that mapping in a very short time (ideally in a single
cycle, and in a way that the L1 cache data lines that get read don't depend on virtual address translation).

------

| Double push/pop instructions would have to be optional because I am trying to keep complexity down.

I feel that your proposal isn't that aggressive in terms of keeping complexity down (compared to a MIPS). It has many
other parts that would be more complex or would require microcode, so I'm not sure that the extra complexity in that
one place would be that salient.

| One solution is to have "save register and increment pointer" instructions in tiny form for both integer registers and full
| length vectors. Same for restore. Then you can save everything with 64 tiny instructions = 128 bytes of code.

Well, "load register and increment pointer" writes to 2 registers so it's similar in complexity to "load dual". Squeezing
a few very common types of memory loads in tiny instruction still makes sense though.

------

| I just made an experiment to see what compilers actually do with a signed 32-bit index variable in 64-bit mode (x86-64).
| Both Ms and Gnu compilers simply replaced my 32-bit integer with a 64-bit integer. I don't think it is worth the effort to
| make a special addressing mode with sign-extended 32-bit index.

I just checked msvc's x64 output, and I'm admit I'm very spooked. It does stuff like loading an index into eax, then using
rax as an array index, even though the variable is clearly signed in the code. Some code locations seem to use movsx for
sign extension, but most don't. The compiler seems to guess when the variable can never go negative, somehow. In one
place, it seems to convert from eax to rax near the beginning of a function, then use rax throughout the body of the
function. Maybe it's using the part of the C/C++ standard that says that integer overflow is "undefined".

This also affects Java (which strongly defines "int" as 32bits and signed, and all its operations as 32bit signed operations)
and there is at least one paper about figuring out how to remove as many sign extensions as possible in 64bit mode.