Hi, all On the website, it says, CCAssignToRegWithShadow <registerList, shadowList> — similar to CCAssignToReg, but with a shadow list of registers I don't know when we should use CCAssignToRegWithShadow. Because I am not a architecture expert, please bear with my naive question. Take ARMCallingConv.td as an example, CCIfType<[i32], CCIfAlign<"8", CCAssignToRegWithShadow<[R0, R2], [R0, R1]>>> If we assign a value to R0(R2), then what happen to R0(R1)? def RetCC_ARM_AAPCS_Common : CallingConv<[ CCIfType<[i32], CCAssignToReg<[R0, R1, R2, R3]>>, CCIfType<[i64], CCAssignToRegWithShadow<[R0, R2], [R1, R3]>> ]>; Does it mean the samething as above? I searched the web and found the link below, http://www.heyrick.co.uk/assembler/regs.html It says in some (ARM) processer modes, some register are shadowed. Does the word "shadow" mean the same thing? Thanks! Regards, chenwj -- Wei-Ren Chen (陳韋任) Computer Systems Lab, Institute of Information Science, Academia Sinica, Taiwan (R.O.C.) Tel:886-2-2788-3799 #1667
Hello> I don't know when we should use CCAssignToRegWithShadow. Because > I am not a architecture expert, please bear with my naive question. > Take ARMCallingConv.td as an example,ARM is not good example :) Look into win64 calling convention. Shadow means if some register is allocated for incoming argument another one cannot be used for argument allocation. On Win64 general purpose registers shadow xmm and vice versa, so, say, if you put argument into ecx, the xmm0 is marked as already occupied, if you put argument into xmm1, edx is marked as occupied, etc.> CCIfType<[i32], CCIfAlign<"8", CCAssignToRegWithShadow<[R0, R2], [R0, R1]>>> > > If we assign a value to R0(R2), then what happen to R0(R1)?It will be marked as occupied.> def RetCC_ARM_AAPCS_Common : CallingConv<[ > CCIfType<[i32], CCAssignToReg<[R0, R1, R2, R3]>>, > CCIfType<[i64], CCAssignToRegWithShadow<[R0, R2], [R1, R3]>> > ]>; > > Does it mean the samething as above?Yes. On ARM shadowing used as a trick to ensure "register alignment" for i64 arguments - they should always be allocated starting from even-numbered register and remaining (odd-numbered) one should be left unallocated. -- With best regards, Anton Korobeynikov Faculty of Mathematics and Mechanics, Saint Petersburg State University
Hi, Anton> On Win64 general purpose registers shadow xmm and vice versa, so, say, > if you put argument into ecx, the xmm0 is marked as already occupied, > if you put argument into xmm1, edx is marked as occupied, etc.I think ecx and xmm0 are physical different registers (not like ecx and cx), right? So why we cannot use xmm0 while ecx has been used? Is it just a rule or there are some considerations behind it?>> CCIfType<[i32], CCIfAlign<"8", CCAssignToRegWithShadow<[R0, R2], [R0, >> R1]>>> >> >> If we assign a value to R0(R2), then what happen to R0(R1)? >It will be marked as occupied.If we assign to R0, then R0 is marked occupied. Is it a redundant statement?> Yes. On ARM shadowing used as a trick to ensure "register alignment" > for i64 arguments - they should always be allocated starting from > even-numbered register and remaining (odd-numbered) one should be left > unallocated.If you can point me to some reference aboue "register alignment", I'll be very appreciate it. Does register alignment bring benefit? In other word, why register alignment is needed? So the word "shadow" on http://www.heyrick.co.uk/assembler/regs.html means different thing, hugh? Thanks! :-) Regards, chenwj -- Wei-Ren Chen (陳韋任) Computer Systems Lab, Institute of Information Science, Academia Sinica, Taiwan (R.O.C.) Tel:886-2-2788-3799 #1667
Seemingly Similar Threads
- [LLVMdev] What CCAssignToXXXWithShadow means?
- [LLVMdev] What CCAssignToXXXWithShadow means?
- BUG: complete misunterstanding of the MS-ABI
- [LLVMdev] llvm.x86.sse2.sqrt.pd not using sqrtpd, calling a function that modifies ECX
- [LLVMdev] RFB: Would like to flip the vector shuffle legality flag