[LoongArch] Allow writing unaligned nop sequences

[wangleiat]

Summary

In case of unaligned nop sequences, pad to the nearest 4-byte boundary
with zeros before filling with nop instructions. This is consistent
with gas behavior, and is necessary to compile the Linux kernel with
LLVM IAS.

Replace support::endian::write with OS.write while at it. This is
simpler and correct because we only have little endian.

[xen0n]

LGTM.

I've edited the title and summary for you, mainly for correcting one semantic mistake in the summary ("switches the bytes to not be dependent on endianness, even though we have only little endian" -- you most certainly didn't mean that, we definitely want the output to be valid little-endian LoongArch machine code, and the host's endianness shouldn't matter. The semantics could be inadvertently reversed by mistranslation).

[wangleiat]

In D139285#3969668, @xen0n wrote:

LGTM.

I've edited the title and summary for you, mainly for correcting one semantic mistake in the summary ("switches the bytes to not be dependent on endianness, even though we have only little endian" -- you most certainly didn't mean that, we definitely want the output to be valid little-endian LoongArch machine code, and the host's endianness shouldn't matter. The semantics could be inadvertently reversed by mistranslation).

Thank you very much!

[SixWeining]

LGTM.

llvm/lib/Target/LoongArch/MCTargetDesc/LoongArchAsmBackend.cpp

bool LoongArchAsmBackend::shouldForceRelocation(const MCAssembler &Asm,
   case FK_Data_4:
   case FK_Data_8:
     return !Target.isAbsolute();
   }
 }
 
 bool LoongArchAsmBackend::writeNopData(raw_ostream &OS, uint64_t Count,
                                        const MCSubtargetInfo *STI) const {
-  // Check for byte count not multiple of instruction word size
-  if (Count % 4 != 0)
-    return false;
+  // We mostly follow binutils' convention here: align to 4-byte boundary with a
+  // 0-fill padding.
+  OS.write_zeros(Count % 4);
 
-  // The nop on LoongArch is andi r0, r0, 0.
+  // The remainder is now padded with 4-byte nops.
+  // nop: andi r0, r0, 0
   for (; Count >= 4; Count -= 4)
-    support::endian::write<uint32_t>(OS, 0x03400000, support::little);
+    OS.write("\0\0\x40\x03", 4);
 
   return true;
 }
 
 std::unique_ptr<MCObjectTargetWriter>
 LoongArchAsmBackend::createObjectTargetWriter() const {
   return createLoongArchELFObjectWriter(OSABI, Is64Bit);
 }

llvm/test/MC/LoongArch/Misc/unaligned-nops.s

-# RUN: not --crash llvm-mc --filetype=obj --triple=loongarch64 %s -o %t
+# RUN: llvm-mc --triple=loongarch64 --filetype=obj %s -o %t
+# RUN: llvm-objdump -d %t | FileCheck %s
+
+# CHECK:      01 00 00 00   <unknown>
+# CHECK-NEXT: 00 00 40 03   nop
 .byte 1
-# CHECK: LLVM ERROR: unable to write nop sequence of 3 bytes
-.p2align 2
+.p2align 3
 foo: