bnjbvr requested cfallin for a review on PR #2258.
bnjbvr opened PR #2258 from check-sse
to main
:
This builds on top of #2255 (two commits), and adds a dynamic check that every time we emit an instruction that uses SSE features, the features were enabled by the configuration. This avoids cases where we could use e.g. SSE4.1 on a machine that doesn't have it.
The way to pass the ISA flags is a bit tricky; please see commit message for the explanation of why it was done this way. Happy to try a different way if anybody has better suggestions!
cfallin submitted PR Review.
cfallin submitted PR Review.
cfallin created PR Review Comment:
Could we rename this to
isa_requirement()
? I think "ISA" is a more standard term than "iset" as an abbreviation.
cfallin created PR Review Comment:
Are all of the below SSE2, e.g.
JmpIf
? (Perhaps they have some options or encodings that are?)(To be clear, I like this function and the explicitness of this
match
; I'm just curious about the comment is all.)
bnjbvr submitted PR Review.
bnjbvr created PR Review Comment:
Yep, I just checked the ones for which i had a few doubts:
- the conversions from int to float/doubles (and reciprocally) are SSE (for floats) and SSE2 (for doubles)
- JmpIf is just a plain old conditional jump, (e.g.
jo
), so nothing to fret about.
bnjbvr updated PR #2258 from check-sse
to main
:
This builds on top of #2255 (two commits), and adds a dynamic check that every time we emit an instruction that uses SSE features, the features were enabled by the configuration. This avoids cases where we could use e.g. SSE4.1 on a machine that doesn't have it.
The way to pass the ISA flags is a bit tricky; please see commit message for the explanation of why it was done this way. Happy to try a different way if anybody has better suggestions!
bnjbvr updated PR #2258 from check-sse
to main
:
This builds on top of #2255 (two commits), and adds a dynamic check that every time we emit an instruction that uses SSE features, the features were enabled by the configuration. This avoids cases where we could use e.g. SSE4.1 on a machine that doesn't have it.
The way to pass the ISA flags is a bit tricky; please see commit message for the explanation of why it was done this way. Happy to try a different way if anybody has better suggestions!
bnjbvr updated PR #2258 from check-sse
to main
:
This builds on top of #2255 (two commits), and adds a dynamic check that every time we emit an instruction that uses SSE features, the features were enabled by the configuration. This avoids cases where we could use e.g. SSE4.1 on a machine that doesn't have it.
The way to pass the ISA flags is a bit tricky; please see commit message for the explanation of why it was done this way. Happy to try a different way if anybody has better suggestions!
bnjbvr updated PR #2258 from check-sse
to main
:
This builds on top of #2255 (two commits), and adds a dynamic check that every time we emit an instruction that uses SSE features, the features were enabled by the configuration. This avoids cases where we could use e.g. SSE4.1 on a machine that doesn't have it.
The way to pass the ISA flags is a bit tricky; please see commit message for the explanation of why it was done this way. Happy to try a different way if anybody has better suggestions!
bnjbvr requested cfallin for a review on PR #2258.
cfallin submitted PR Review.
bnjbvr merged PR #2258.
Last updated: Nov 22 2024 at 16:03 UTC