bit.jinc syntax error about libjade HOT 8 CLOSED

(also the version information is not particularly helpful)

The installation instruction that you’ve followed suggested to pull the Jasmin source code from the wrong repo…

from libjade.

You need a bleeding edge version of the Jasmin compiler to compile the development version of libjade: either the head of the release-2022.09 branch (my own recommendation) or the a recent version of the main branch (what the libjade developers do).

from libjade.

I see.
I've done the instructions from Sir Cryptojedi here a couple days ago, which felt like it was pulling from main, but instead I'm getting the version @Version@.

jasminc -version
Jasmin Compiler @VERSION@

using opam install jasmin now, and it compiles with the code in the repo.
(but it wont compile with my altered code anymore, which I find surprising: It fails with

compilation error in function jade_scalarmult_curve25519_amd64_mulx_base:
asmgen: invalid rexpr for oprd (RDI >> RDX)
        invalid rexpr for word
make: *** [../../../../Makefile.common:68: scalarmult.s] Error 1

)
(also the version information is not particularly helpful)

EDIT: both of those may be better tracked in the jasmin-lang repo

from libjade.

Yes, there is no instruction on x86 that implement that kind of shift. Fortunately, there is one that does RDI >> (RDX & 63).

from libjade.

Yes, there is no instruction on x86 that implement that kind of shift. Fortunately, there is one that does RDI >> (RDX & 63).

Hm, I see shrd r/64, r/64, r32b and shr r/64, cl in the manual. Former needs BMI2.
All shifts only make sense with a one-byte COUNT.

So shouldn't jasminc be able to move p into cl and use shr rzx, cl, or use shrx (Im not sure if ISEs are enabled).
What I'm trying to understand it why does the COUNT need to be ANDed with 63, where this is clearly too big for any shifts.

The installation instruction that you’ve followed suggested to pull the Jasmin source code from the wrong repo…

Fiat, I'll let cryptojedi know, to change jasmin-lang/jasmin to jasmin-lang/jasmin-compiler
However, I still seem to face the issue that jasminc -version responds with @VERSION@. Is that because I've used opam install jasmin rather than cloning from the release-2022.09 branch?

from libjade.

why does the COUNT need to be ANDed with 63

This is what x86 processors do.

from libjade.

Yes, I'm sorry. Got confused. I my head it was AND'ing with 2^63 (which is too big). All makes sense now. Manual says COUNT AND 3FH and then successive /2s.

from libjade.

Even though this issue is closed, the following commits, which are related to this discussion, might be of interest to readers:

jasmin-lang/jasmin@678bd31

443ee3c

fecec47

from libjade.