[llvm] [TableGen] Split large printInstruction() for NVPTX to avoid compiler limits (PR #218530)

[Kazu Hirata via llvm-commits]

================

----------------
kazutakahirata wrote:

Emit one `InstructionGroup` sequence per line with its starting offset comment. This shrinks `OverflowProgram[]` from ~6,450 lines to ~270 lines for NVPTX and makes sequence boundaries easy to see:
```suggestion
    << "    /* 0 */ 0,\n";
  unsigned Offset = 1;
  for (const InstructionGroup &G : Groups) {
    O << "    /* " << Offset << " */";
    for (const AsmWriterOperand &Op : G.FirstInst.Operands)
      O << " " << Commands.lookup(Op.getCode(PassSubtarget)) << ",";
    O << "\n";
    Offset += G.FirstInst.Operands.size();
  }
```


https://github.com/llvm/llvm-project/pull/218530