x64: convert some SSE shift instructions #10762
Conversation
| fn is_imm8_xmm(&mut self, src: &XmmMemImm) -> Option<u8> { | ||
| match src.clone().to_reg_mem_imm() { | ||
| RegMemImm::Imm { simm32 } => Some(u8::try_from(simm32).ok()?), | ||
| _ => None, | ||
| } | ||
| } |
There was a problem hiding this comment.
I don't really want to add this but, because of the type signatures of the inst.isle constructors, I have to find a way to match on a XmmMemImm instead of GprMemImm. One alternative to adding this would be to have is_imm8 examine the more general RegMemImm instead, but this would mean the constructor types would have to change. Thoughts?
There was a problem hiding this comment.
Poking around it looks like these shifting instructions are the only ones using XmmMemImm, so perhaps we could drop that type entirely? All the various helpers seem to be using mask_xmm_shift and the current lowerings could dispatch to a helper taking the RegMemImm result of mask_xmm_shift which handles the *Imm case to delegate to different constructors than when in the Reg or Mem case?
github-actions
bot
added
cranelift
Subscribe to Label Action
This issue or pull request has been labeled: "cranelift", "cranelift:area:x64", "isle"
Thus the following users have been cc'd because of the following labels:
To subscribe or unsubscribe from this label, edit the |
While reflecting on why bytecodealliance#10762 is hard, it seemed that the underlying problem to many ISLE matching issues is our inability to actually inspect the variants of the register specific versions of `RegMem` and `RegMemImm`. We had been wrapping instances of these `Reg*` types, but this change pushes the variants down to: `GprMem`, `GprMemImm`, `XmmMem`, `XmmMemAligned`, `XmmMemImm`, and `XmmMemAlignedImm`. With this, it should be possible to match directly on the variants in ISLE. This does not change any functionality.
|
that was the wrong button... |
No description provided.