I have an LTO snapshot (for which I don't have sources) that can't
be read back by LLVM. It seems the writer emitted broken bitcode
and this assertions aims at catching such cases.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@274819 91177308-0d34-0410-b5e6-96231b3b80d8
This change is motivated by an upcoming change to the metadata representation
used for CFI. The indirect function call checker needs type information for
external function declarations in order to correctly generate jump table
entries for such declarations. We currently associate such type information
with declarations using a global metadata node, but I plan [1] to move all
such metadata to global object attachments.
In bitcode, metadata attachments for function declarations appear in the
global metadata block. This seems reasonable to me because I expect metadata
attachments on declarations to be uncommon. In the long term I'd also expect
this to be the case for CFI, because we'd want to use some specialized bitcode
format for this metadata that could be read as part of the ThinLTO thin-link
phase, which would mean that it would not appear in the global metadata block.
To solve the lazy loaded metadata issue I was seeing with D20147, I use the
same bitcode representation for metadata attachments for global variables as I
do for function declarations. Since there's a use case for metadata attachments
in the global metadata block, we might as well use that representation for
global variables as well, at least until we have a mechanism for lazy loading
global variables.
In the assembly format, the metadata attachments appear after the "declare"
keyword in order to avoid a parsing ambiguity.
[1] http://lists.llvm.org/pipermail/llvm-dev/2016-June/100462.html
Differential Revision: http://reviews.llvm.org/D21052
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@273336 91177308-0d34-0410-b5e6-96231b3b80d8
This will be necessary to allow the global merge pass to attach
multiple debug info metadata nodes to global variables once we reverse
the edge from DIGlobalVariable to GlobalVariable.
Differential Revision: http://reviews.llvm.org/D20414
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@271358 91177308-0d34-0410-b5e6-96231b3b80d8
This patch adds an IR, assembly and bitcode representation for metadata
attachments for globals. Future patches will port existing features to use
these new attachments.
Differential Revision: http://reviews.llvm.org/D20074
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@271348 91177308-0d34-0410-b5e6-96231b3b80d8
Since forward references for uniqued node operands are expensive (and
those for distinct node operands are cheap due to
DistinctMDOperandPlaceholder), minimize forward references in uniqued
node operands.
Moreover, guarantee that when a cycle is broken by a distinct node, none
of the uniqued nodes have any forward references. In
ValueEnumerator::EnumerateMetadata, enumerate uniqued node subgraphs
first, delaying distinct nodes until all uniqued nodes have been
handled. This guarantees that uniqued nodes only have forward
references when there is a uniquing cycle (since r267276 changed
ValueEnumerator::organizeMetadata to partition distinct nodes in front
of uniqued nodes as a post-pass).
Note that a single uniqued subgraph can hit multiple distinct nodes at
its leaves. Ideally these would themselves be emitted in post-order,
but this commit doesn't attempt that; I think it requires an extra pass
through the edges, which I'm not convinced is worth it (since
DistinctMDOperandPlaceholder makes forward references quite cheap
between distinct nodes).
I've added two testcases:
- test/Bitcode/mdnodes-distinct-in-post-order.ll is just like
test/Bitcode/mdnodes-in-post-order.ll, except with distinct nodes
instead of uniqued ones. This confirms that, in the absence of
uniqued nodes, distinct nodes are still emitted in post-order.
- test/Bitcode/mdnodes-distinct-nodes-break-cycles.ll is the minimal
example where a naive post-order traversal would cause one uniqued
node to forward-reference another. IOW, it's the motivating test.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@267278 91177308-0d34-0410-b5e6-96231b3b80d8
When an operand of a distinct node hasn't been read yet, the reader can
use a DistinctMDOperandPlaceholder. This is much cheaper than forward
referencing from a uniqued node. Change
ValueEnumerator::organizeMetadata to partition distinct nodes and
uniqued nodes to reduce the overhead of cycles broken by distinct nodes.
Mehdi measured this for me; this removes most of the RAUW from the
importing step of -flto=thin, even after a WIP patch that removes
string-based DITypeRefs (introducing many more cycles to the metadata
graph).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@267276 91177308-0d34-0410-b5e6-96231b3b80d8
Mehdi's pattern recognition pulled this one out. This is cleaner with
std::find_if than with the strange helper function that took an iterator
by reference and updated it.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@267271 91177308-0d34-0410-b5e6-96231b3b80d8
Re-layer the functions in the new (i.e., newly correct) post-order
traversals in ValueEnumerator (r266947) and ValueMapper (r266949).
Instead of adding a node to the worklist in a helper function and
returning a flag to say what happened, return the node itself. This
makes the code way cleaner: the worklist is local to the main function,
there is no flag for an early loop exit (since we can cleanly bury the
loop), and it's perfectly clear when pointers into the worklist might be
invalidated.
I'm fixing both algorithms in the same commit to avoid repeating the
commit message; if you take the time to understand one the other should
be easy. The diff itself isn't entirely obvious since the traversals
have some noise (i.e., things to do), but here's the high-level change:
auto helper = [&WL](T *Op) { auto helper = [](T **&I, T **E) {
=> while (I != E) {
if (shouldVisit(Op)) { T *Op = *I++;
WL.push(Op, Op->begin()); if (shouldVisit(Op)) {
return true; return Op;
} }
return false; return nullptr;
}; };
=>
WL.push(S, S->begin()); WL.push(S, S->begin());
while (!empty()) { while (!empty()) {
auto *N = WL.top().N; auto *N = WL.top().N;
auto *&I = WL.top().I; auto *&I = WL.top().I;
bool DidChange = false;
while (I != N->end())
if (helper(*I++)) { => if (T *Op = helper(I, N->end()) {
DidChange = true; WL.push(Op, Op->begin());
break; continue;
} }
if (DidChange)
continue;
POT.push(WL.pop()); => POT.push(WL.pop());
} }
Thanks to Mehdi for helping me find a better way to layer this.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@267099 91177308-0d34-0410-b5e6-96231b3b80d8
Emit metadata nodes in post-order. The iterative algorithm from r266709
failed to maintain this property. After understanding my mistake, it
wasn't too hard to write a test with llvm-bcanalyzer (and I've actually
made this change once before: see r220340).
This also reverts the "noisy" testcase change from r266709. That should
have been more of a red flag :/.
Note: The same bug crept into the ValueMapper in r265456. I'm still
working on the fix.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@266947 91177308-0d34-0410-b5e6-96231b3b80d8
Use a worklist instead of recursing through MDNode operands in
ValueEnumerator. The actual record output order has changed slightly,
but otherwise there's no functionality change.
I had to update test/Bitcode/metadata-function-blocks.ll. I renumbered
nodes so they continue to match the implicit record ids.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@266709 91177308-0d34-0410-b5e6-96231b3b80d8
I have no idea how I chose two different spellings in the space of a
couple of weeks, but now I can't remember what to use where. Choose
"Worklist".
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@266582 91177308-0d34-0410-b5e6-96231b3b80d8
Whenever metadata is only referenced by a single function, emit the
metadata just in that function block. This should improve lazy-loading
by reducing the amount of metadata in the global block.
For now, this should catch all DILocations, and anything else that
happens to be referenced only by a single function.
It's also a first step toward a couple of possible future directions
(which this commit does *not* implement):
1. Some debug info metadata is only referenced from compile units and
individual functions. If we can drop the link from the compile
unit, this optimization will get more powerful.
2. Any uniqued metadata that isn't referenced globally can in theory be
emitted in every function block that references it (trading off
bitcode size and full-parse time vs. lazy-load time).
Note: this assumes the new BitcodeReader error checking from r265223.
The metadata stored in function blocks gets purged after parsing each
function, which means unresolved forward references will get lost.
Since all the global metadata should have already been resolved by the
time we get to the function metadata blocks we just need to check for
that case. (If for some reason we need to handle bitcode that fails the
checks in r265223, the fix is to store about-to-be-dropped unresolved
nodes in MetadataList::shrinkTo until they can be handled succesfully by
a future call to MetadataList::tryToResolveCycles.)
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@265226 91177308-0d34-0410-b5e6-96231b3b80d8
Further unify the handling of function-local metadata with global
metadata, by exposing the same interface in ValueEnumerator. Both
contexts use the same accessors:
- getMDStrings(): get the strings for this block.
- getNonMDStrings(): get the non-strings for this block.
A future commit will start adding strings to the function-block.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@265224 91177308-0d34-0410-b5e6-96231b3b80d8
The caller of ValueEnumerator::EnumerateOperandType never sends in
metadata. Assert that, and remove the unnecessary logic.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@264558 91177308-0d34-0410-b5e6-96231b3b80d8
We don't really need a separate vector here; instead, point at a range
inside the main MDs array. This matches how r264551 references the
ranges of strings and non-strings.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@264552 91177308-0d34-0410-b5e6-96231b3b80d8
Spiritually reapply commit r264409 (reverted in r264410), albeit with a
bit of a redesign.
Firstly, avoid splitting the big blob into multiple chunks of strings.
r264409 imposed an arbitrary limit to avoid a massive allocation on the
shared 'Record' SmallVector. The bug with that commit only reproduced
when there were more than "chunk-size" strings. A test for this would
have been useless long-term, since we're liable to adjust the chunk-size
in the future.
Thus, eliminate the motivation for chunk-ing by storing the string sizes
in the blob. Here's the layout:
vbr6: # of strings
vbr6: offset-to-blob
blob:
[vbr6]: string lengths
[char]: concatenated strings
Secondly, make the output of llvm-bcanalyzer readable.
I noticed when debugging r264409 that llvm-bcanalyzer was outputting a
massive blob all in one line. Past a small number, the strings were
impossible to split in my head, and the lines were way too long. This
version adds support in llvm-bcanalyzer for pretty-printing.
<STRINGS abbrevid=4 op0=3 op1=9/> num-strings = 3 {
'abc'
'def'
'ghi'
}
From the original commit:
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.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@264551 91177308-0d34-0410-b5e6-96231b3b80d8
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.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@264409 91177308-0d34-0410-b5e6-96231b3b80d8
Caught by inspection while working on partitioning metadata. It's nice
to produce the same bitcode if you run the compiler twice.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@264381 91177308-0d34-0410-b5e6-96231b3b80d8
Simplify ValueEnumerator and WriteModuleMetadata by shifting the logic
for the METADATA_GENERIC_DEBUG abbreviation into WriteGenericDINode.
(This is just like r264302, but for GenericDINode.)
The only change is that the abbreviation is emitted later in the
bitcode, just before the first `GenericDINode` record. This shouldn't
be observable though.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@264303 91177308-0d34-0410-b5e6-96231b3b80d8
Simplify ValueEnumerator and WriteModuleMetadata by shifting the logic
for the METADATA_LOCATION abbreviation into WriteDILocation.
The only change is that the abbreviation is emitted later in the
bitcode, just before the first `DILocation` record. This shouldn't be
observable though.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@264302 91177308-0d34-0410-b5e6-96231b3b80d8
Malcolm Parsons
Mehdi Amini
Peter Collingbourne
Duncan P. N. Exon Smith
Dmitry Polukhin
Yaron Keren
Teresa Johnson