For any instruction, a list of constraints can be specified, like: let Constraints = "$dst = $lhs" in ... I checked the existing backends and I didn't notice any complex constraints: basically "Constraints" keyword is used to force 1 operand register to be the same as another one. I am wondering if there is any way to express more complicated constraints like for example: - if $lhs is either r0,r1,r2,r3 then $dst should also be in the set {r0,r1,r2,r3} - if $lhs is either r4,r5,r6,r7 then $dst should also be in the set {r4, r5, r6, r7} ... Thank you ! Damien -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20110408/8e8d3ac7/attachment.html>
On Apr 8, 2011, at 10:13 AM, Damien Vincent wrote:> > For any instruction, a list of constraints can be specified, like: > let Constraints = "$dst = $lhs" in > ... > > I checked the existing backends and I didn't notice any complex constraints: basically "Constraints" keyword is used to force 1 operand register to be the same as another one. > I am wondering if there is any way to express more complicated constraints like for example: > - if $lhs is either r0,r1,r2,r3 then $dst should also be in the set {r0,r1,r2,r3} > - if $lhs is either r4,r5,r6,r7 then $dst should also be in the set {r4, r5, r6, r7}Not at the moment, no. The only constraints supported are the "tied-to" constraint you indicate above and the "@earlyclobber" constraint which indicates an output register that cannot also be allocated to one of the instruction inputs. See CodeGenInstruction.cpp in utils/TableGen for the gory details. Regards, -Jim
Seemingly Similar Threads
- {ARM} IfConversion does not detect BX instruction as a branch
- [PATCH v1 01/27] x86/crypto: Adapt assembly for PIE support
- [PATCH v1 01/27] x86/crypto: Adapt assembly for PIE support
- [LLVMdev] Add/sub with carry; widening multiply
- Loop invariant not being optimized