This patch adds codegen support to generate the mrrc/mrrc2 instruction from the intrinsic.
Details
Diff Detail
Event Timeline
lib/Target/ARM/Disassembler/ARMDisassembler.cpp | ||
---|---|---|
5285 | This looks like a different change, for another patch. Also, why only validate for MRRC and MCRR? Were the checks wrong for the rest? If so, why not just have one check if MRRC || MCRR? |
lib/Target/ARM/ARMISelDAGToDAG.cpp | ||
---|---|---|
3303 | Bad name, use "isThumbV8M". |
Thanks for the review Renato. I replied back to your comments below.
lib/Target/ARM/ARMISelDAGToDAG.cpp | ||
---|---|---|
3303 | Actually I think I only need to do Subtarget->isThumb() not the rest. | |
lib/Target/ARM/Disassembler/ARMDisassembler.cpp | ||
5285 |
Decoder tests were failing because I changed the tablegen definitions for 'MCRR2' and 'MRRC2' to define the inputs and ouputs in 'lib/Target/ARM/ARMInstrInfo.td'.
This decoder method is only used by definitions instantiated from the tablegen class MovRRCopro2, there are only two definitions that use this class, they are MCRR2 and MRRC2.. |
lib/Target/ARM/Disassembler/ARMDisassembler.cpp | ||
---|---|---|
5285 | If only MCRR2 and MRRC2 use this method, there's no point for the condition, right? |
lib/Target/ARM/Disassembler/ARMDisassembler.cpp | ||
---|---|---|
5285 | The conditional checks are for controlling the order in which the operands are added to Inst. For MRRC2, the order should be [Rt, Rt2, coproc, opc1, CRm] for MCRR2 it should be [coproc, opc1, Rt, Rt2, CRm] |
lib/Target/ARM/Disassembler/ARMDisassembler.cpp | ||
---|---|---|
5285 | Just expanding on my previous answer. Previously the order in which variable Inst was constructed would be the same for both MRRC2 and MCRR2 because the tablgen description of the outputs and inputs was "...(outs), (ins p_imm:$cop, imm0_15:$opc1, GPRnopc:$Rt, GPRnopc:$Rt2, c_imm:$CRm...>". This tablegen is correct for MCRR2 but not MRRC2, MRRC2 writes to 2 registers so the tablegen description should be "...(outs GPRnopc:$Rt, GPRnopc:$Rt2), (ins p_imm:$cop, imm0_15:$opc1, c_imm:$CRm)...". This is why the operands in Inst for MRRC2 need to be [Rt, Rt2, coproc, opc1, CRm] |
Thanks for reviewing Renato, will make your suggested changes before committing. Can you also review the clang part of the patch at some point http://reviews.llvm.org/D21179 ?
Thanks,
Ranjeet
Bad name, use "isThumbV8M".