Load/store default alignment? about spec HOT 7 CLOSED

The evaluator always ignores alignment. The intention is that None means "no hint" and therefore natural alignment. But I guess I can see confusion with None meaning "completely unaligned" (i.e., align=1), so it'd be just as well to remove the option and make align an int that parser.mly fills in when missing.

from spec.

Plus we should perhaps define this in the design no? I looked quickly and did not see much information about the lack of information on alignment and what it implied.

from spec.

If we keep alignment an option, then yes, it'd make sense for AstSemantics.md to say the hint is optional and, when not present, natural alignment should be assumed. If the alignment is changed to a non-optional int, then it's a syntactic sugar detail of the sexpr language that the align immediate is optional.

from spec.

Seems bizarre to me to say that we have to do align for a load but the immediate is optional.

(i32.load align (i32.const 0)) seems just overkill. I really think that the default should be that it is optional and if not there, it is defined in the AstSemantics.md as then meaning the load is of natural alignment.

from spec.

I think Luke was suggesting that the hint missing means that the access is
assumed to be aligned.

So:

(i32.load (i32.const 0))

and

(i32.load unaligned (i32.const 1))

?

On Tue, Nov 3, 2015 at 6:52 AM, jcbeyler [email protected] wrote:

Seems bizarre to me to say that we have to do align for a load but the
immediate is optional.

(i32.load align (i32.const 0)) seems just overkill. I really think that
the default should be that it is optional and if not there, it is defined
in the AstSemantics.md as then meaning the load is of natural alignment.

—
Reply to this email directly or view it on GitHub
#158 (comment).

from spec.

The s-expr language offers a variety of syntactic sugar wherein you can omit things and the obvious default AST node is generated. Having a missing align attribute imply natural alignment seems to be no different. Thus, (i32.load (i32.const 0)) would imply align=4. From AstSemantics.md POV (as it is currently written), there is always an alignment immediate, so no need to say what the default is. However, as I said above, we could change AstSemantics.md to say "alignment is optional, when missing natural alignment is assumed". I think it's all just a matter of subjective preference until we get to the binary encoding.

from spec.

This has been resolved.

from spec.