12

I have trouble finding out what movslq instruction does. Google isn't very helpful and there is no such instruction on this list. Here I have read that

MOVSLQ is move and sign-extend a value from a 32-bit source to a 64-bit destination.

I understand this like this. The instruction moves a value from 32-bit source to 64-bit destination but ensures that signed value of the number in the 64-bit destination is equal to the one from source, that is in case of negative number it fills first 32 bits of destination with ones, otherwise with zeros.

Is my understanding correct? If not, please explain where I am wrong.

EDIT:

This is not a duplicate of this topic: assembly cltq and movslq difference. There is no explanation for my question. And you'd have to read the whole answer for 10 minutes to convince yourself of that.

fuz
  • 88,405
  • 25
  • 200
  • 352
ibodi
  • 1,543
  • 3
  • 21
  • 40
  • 2
    It's `movsxd`: https://www.felixcloutier.com/x86/MOVSX:MOVSXD.html, and yes, it does 32->64 2's complement sign extension. You could figure this out yourself by assembling it with an AT&T assembler and disassembling with an Intel-syntax disassembler. (e.g. `objdumpd -d -Mintel foo.o`) – Peter Cordes Apr 09 '19 at 04:23
  • 1
    The information is in there, if you control-f search for `movslq` in [assembly cltq and movslq difference](//stackoverflow.com/q/37743476), the 2nd mention of it is in a sentence explaining that it's `movsx`, with a link to the Intel manual. You're right it's not a perfect duplicate, but I wanted to close it because there's a mechanical solution (assemble+disassemble) to find out Intel mnemonics for AT&T opcodes, and there's not really much else to say about it, so questions like this are somewhat low quality. – Peter Cordes Apr 09 '19 at 05:33
  • 1
    And 2's complement sign extension is something you can look up separately, or the Intel manual documents that, too. But sure, I guess that's enough steps that it's worth collecting it into an answer; you're probably not the first person to search for `movslq`. – Peter Cordes Apr 09 '19 at 05:35

1 Answers1

12

It's movsxd: https://www.felixcloutier.com/x86/MOVSX:MOVSXD.html.

You could figure this out yourself by assembling it with an AT&T assembler and disassembling with an Intel-syntax disassembler. (e.g. objdumpd -d -Mintel foo.o)

And yes, it does 32->64-bit 2's complement sign extension, extending by copying the sign-bit of the source to all the new upper bits. (i.e. dst[63:32] = src[31], and dst[31:0] = src[31:0].)

Fun fact: it's possible to use it without a REX.W prefix as just a 32-bit copy that's architecturally equivalent to mov, but there's zero advantage to ever doing that. Or even with an operand-size prefix as a 16-bit copy, but again that's totally pointless, use mov.

Related: What does cltq do in assembly? has an equivalence table for AT&T vs. Intel mnemonics for all x86 sign-extension instructions, along with description. ([ER]AX to DX:AX, within-[ER]AX, and movsx/movzx to arbitrary registers.)

assembly cltq and movslq difference has more history on how AMD64 repurposed the ARPL opcode from from 16/32-bit mode to get a 1-byte opcode for movsxd.

Peter Cordes
  • 328,167
  • 45
  • 605
  • 847