caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Bitcode: Collect all MDString records into a single blob

Browse Source 
Optimize output of MDStrings in bitcode.  This emits them in big blocks
(currently 1024) in a pair of records:
  - BULK_STRING_SIZES: the sizes of the strings in the block, and
  - BULK_STRING_DATA: a single blob, which is the concatenation of all
    the strings.

Inspired by Mehdi's similar patch, http://reviews.llvm.org/D18342, this
should (a) slightly reduce bitcode size, since there is less record
overhead, and (b) greatly improve reading speed, since blobs are super
cheap to deserialize.

I needed to add support for blobs to streaming input to get the test
suite passing.
  - StreamingMemoryObject::getPointer reads ahead and returns the
    address of the blob.
  - To avoid a possible reallocation of StreamingMemoryObject::Bytes,
    BitstreamCursor::readRecord needs to move the call to JumpToEnd
    forward so that getPointer is the last bitstream operation.

llvm-svn: 264409
This commit is contained in:
Duncan P. N. Exon Smith
2016-03-25 14:40:18 +00:00
parent 59bcbba6b4
commit fdbf0a5af8
11 changed files with 142 additions and 50 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
llvm/lib/Bitcode/Reader/BitcodeReader.cpp
View File

@@ -2363,7 +2363,7 @@ std::error_code BitcodeReader::parseMetadata(bool ModuleLevel) {
NextMetadataNo++);
break;
}
case bitc::METADATA_STRING: {
case bitc::METADATA_STRING_OLD: {
std::string String(Record.begin(), Record.end());
// Test for upgrading !llvm.loop.
@@ -2373,6 +2373,38 @@ std::error_code BitcodeReader::parseMetadata(bool ModuleLevel) {
MetadataList.assignValue(MD, NextMetadataNo++);
break;
}
case bitc::METADATA_BULK_STRING_SIZES: {
// This is a pair of records for an MDString block: SIZES, which is a
// list of string lengths; and DATA, which is a blob with all the strings
// concatenated together.
//
// Note: since this record type was introduced after the upgrade for
// !llvm.loop, we don't need to change HasSeenOldLoopTags.
if (Record.empty())
return error("Invalid record: missing bulk metadata string sizes");
StringRef Blob;
SmallVector<uint64_t, 1> BlobRecord;
Code = Stream.ReadCode();
unsigned BlobCode = Stream.readRecord(Code, BlobRecord, &Blob);
if (BlobCode != bitc::METADATA_BULK_STRING_DATA)
return error("Invalid record: missing bulk metadata string data");
if (!BlobRecord.empty())
return error("Invalid record: unexpected bulk metadata arguments");
for (uint64_t Size : Record) {
if (Blob.size() < Size)
return error("Invalid record: not enough bulk metadata string bytes");
// Extract the current string.
MetadataList.assignValue(MDString::get(Context, Blob.slice(0, Size)),
NextMetadataNo++);
Blob = Blob.drop_front(Size);
}
if (!Blob.empty())
return error("Invalid record: too many bulk metadata string bytes");
break;
}
case bitc::METADATA_KIND: {
// Support older bitcode files that had METADATA_KIND records in a
// block with METADATA_BLOCK_ID.