[BitstreamReader] Fix 32-bit overflow

[pranavk]

This got exposed when processing large LTO-generated files leading to crashes.

[rnk]

Nice find =D

[rnk]

To me this raises a question of whether we want to support writing 4GB+ blobs in bitcode files. We could use ReadVBR64(6) above and EmitVBR64(6) on the writer side, and I think that would be backwards compatible, it just widens the maximum encodable value.

[mingmingl-llvm]

The change to {Read,Emit}VBR64(6) looks backward compatible to me as well, and https://llvm.org/docs/BitCodeFormat.html#define-abbrev-encoding needs an update around Blob (code 5): This field is emitted as a vbr6, followed by padding to a 32-bit boundary (for alignment) and an array of 8-bit objects It's not clear to me (from the doc and code comment around) how 32-bit alignment is related with value width (32 vs 64) and whether alignment needs an update for 4GB+ blobs.

If we decide not to support 4GB blob soon, I think emitting a release-build visible warning (e.g., by LLVMContext's diagnostic handler) when blob size is larger than 4GB around

llvm-project/llvm/include/llvm/Bitstream/BitstreamWriter.h

Line 550 in 76f0ff8

EmitVBR(static_cast<uint32_t>(Bytes.size()), 6);

is generally a good change.

[rnk]

I will merge this as is and file an issue covering the VBR64 change.