BOLT: BOLT generates invalid constant pool in .gdb-index

[itrofimow]

I think that after this commit dcfa2ab, BOLT might generate invalid constant pool table, due to CU reordering taking place here:

llvm-project/bolt/lib/Core/GDBIndex.cpp

Lines 126 to 132 in b16fe13

	using MapEntry = std::pair<uint32_t, CUInfo>;
	std::vector<MapEntry> CUVector(CUMap.begin(), CUMap.end());
	// Need to sort since we write out all of TUs in .debug_info before CUs.
	std::sort(CUVector.begin(), CUVector.end(),
	[](const MapEntry &E1, const MapEntry &E2) -> bool {
	return E1.second.Offset < E2.second.Offset;
	});

If the reordering did indeed happen, then CU-indexes in the constant pool should be updated, otherwise they point to wrong CUs, right?

cc @ayermolo

[llvmbot]

@llvm/issue-subscribers-bolt

Author: None (itrofimow)

I think that after this commit https://github.com/llvm/llvm-project/commit/dcfa2ab534c8f1b4345012a02f130a0a07060bfd, BOLT might generate invalid constant pool table, due to CU reordering taking place here: https://github.com/llvm/llvm-project/blob/b16fe132c68834a601b29c0eb53c7ec25a3cd627/bolt/lib/Core/GDBIndex.cpp#L126-L132

If the reordering did indeed happen, then CU-indexes in the constant pool should be updated, otherwise they point to wrong CUs, right?

cc @ayermolo

[itrofimow]

I just realized that address-map CU-indices should be updated as well and made a PR #151927

[itrofimow]

Resolved by #151927