On Tue, Mar 7, 2017 at 9:13 AM, Marek Olšák <[email protected]> wrote: > On Tue, Mar 7, 2017 at 1:38 PM, Ilia Mirkin <[email protected]> wrote: >> Why can't the compiler perform this opt? At least we do that in nouveau... > > Our LLVM sample intrinsics don't preserve the texture target > information. It would be possible to do that with hacks, but I'd > rather avoid that, because it would need a new LLVM IR pass in > radeonsi. The information about level=0 also can't be extracted from > TGSI directly, because it can't enumerate all "defs" of a src operand.
Right, you can't "hack" this while converting the TGSI (at least not without serious analysis). I just figured the compiler could easily do it. This would also have the advantage of being able to pick up places that only become 0 after optimizations. But if not, I have no objection to these ops. (Come to think of it, due to a series of spectacular failures in codegen along with arch weirdness, this will also be nice to have on nv50 for textureLod(0).) Cheers, -ilia _______________________________________________ mesa-dev mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/mesa-dev
