cfallin opened Issue #2543:
The
binemit
filetest backend does not currently support new-style (MachInst) backends; it is fairly tightly coupled to the encodings scheme, because it checks the machine code emitted for each individual CLIF instruction. We should update it to at least allow bulk comparison of the entire function body, even if we don't have the fine-grained CLIF-instruction-to-emitted-bytes correspondence anymore.Noted in #2540.
cfallin labeled Issue #2543:
The
binemit
filetest backend does not currently support new-style (MachInst) backends; it is fairly tightly coupled to the encodings scheme, because it checks the machine code emitted for each individual CLIF instruction. We should update it to at least allow bulk comparison of the entire function body, even if we don't have the fine-grained CLIF-instruction-to-emitted-bytes correspondence anymore.Noted in #2540.
bjorn3 commented on Issue #2543:
I just remember that the new backends have an
#[test]
that lists all vcode instructions and tests if the generated bytes are as expected.
cfallin commented on Issue #2543:
Oh, yes, this is a better way to test the TLS emission, I agree; I'll add an emission test over in #2540. (I'm a bit slow this morning, sorry!) The proper approach to the binemit backend will be to remove it once we transition fully to new-style backends; I'll edit this issue accordingly.
Last updated: Dec 23 2024 at 12:05 UTC