This models the A9 processor at the level of instruction operands, as
opposed to the itinerary, which models each operation at the level of
pipeline stages.
The two primary motivations are:
1) Allow MachineScheduler to model A9 as an out-of-order processor. It
can now distinguish between hazards that force interlocking vs.
buffered resources.
2) Reduce long-term maintenance by allowing the itinerary and target
hooks to eventually be removed. Note that almost all of the complexity
in the new model exists to model instruction variants, which the
itinerary cannot handle. Instead the scheduler previously relied on
processor-specific target hooks which are incomplete and buggy.
llvm-svn: 163921
This patch introduces a possibility for Hexagon MI scheduler
to perform some target specific post- processing on the scheduling
DAG prior to scheduling.
llvm-svn: 163903
* wrap code blocks in \code ... \endcode;
* refer to parameter names in paragraphs correctly (\arg is not what most
people want -- it starts a new paragraph);
* use \param instead of \arg to document parameters in order to be consistent
with the rest of the codebase.
llvm-svn: 163902
pointless checks in here, bad asserts, and just confusing code. I've
also added a bit more to the comment to clarify what this function is
really trying to do as it was not obvious to Duncan when studying it.
Thanks to Duncan for helping me dig through the issue.
No real functionality changed here in practical cases, and certainly no
test case. This is just cleanup spotted by inspection.
llvm-svn: 163897
inspection by Duncan during review. My suspicion is that we would still
have returned 0 anyways in this case, but doing it sooner is better.
llvm-svn: 163895
deeply suspicious and likely to go away eventually. Also fix a bogus
comment about one of the checks in the vector GEP analysis. Based on
review from Duncan.
llvm-svn: 163894
Originally I had anticipated needing to thread this through more bits of
the SROA pass itself, but that ended up not happening. In the end, this
is a much simpler way to manange the variable.
llvm-svn: 163893
This is essentially a ground up re-think of the SROA pass in LLVM. It
was initially inspired by a few problems with the existing pass:
- It is subject to the bane of my existence in optimizations: arbitrary
thresholds.
- It is overly conservative about which constructs can be split and
promoted.
- The vector value replacement aspect is separated from the splitting
logic, missing many opportunities where splitting and vector value
formation can work together.
- The splitting is entirely based around the underlying type of the
alloca, despite this type often having little to do with the reality
of how that memory is used. This is especially prevelant with unions
and base classes where we tail-pack derived members.
- When splitting fails (often due to the thresholds), the vector value
replacement (again because it is separate) can kick in for
preposterous cases where we simply should have split the value. This
results in forming i1024 and i2048 integer "bit vectors" that
tremendously slow down subsequnet IR optimizations (due to large
APInts) and impede the backend's lowering.
The new design takes an approach that fundamentally is not susceptible
to many of these problems. It is the result of a discusison between
myself and Duncan Sands over IRC about how to premptively avoid these
types of problems and how to do SROA in a more principled way. Since
then, it has evolved and grown, but this remains an important aspect: it
fixes real world problems with the SROA process today.
First, the transform of SROA actually has little to do with replacement.
It has more to do with splitting. The goal is to take an aggregate
alloca and form a composition of scalar allocas which can replace it and
will be most suitable to the eventual replacement by scalar SSA values.
The actual replacement is performed by mem2reg (and in the future
SSAUpdater).
The splitting is divided into four phases. The first phase is an
analysis of the uses of the alloca. This phase recursively walks uses,
building up a dense datastructure representing the ranges of the
alloca's memory actually used and checking for uses which inhibit any
aspects of the transform such as the escape of a pointer.
Once we have a mapping of the ranges of the alloca used by individual
operations, we compute a partitioning of the used ranges. Some uses are
inherently splittable (such as memcpy and memset), while scalar uses are
not splittable. The goal is to build a partitioning that has the minimum
number of splits while placing each unsplittable use in its own
partition. Overlapping unsplittable uses belong to the same partition.
This is the target split of the aggregate alloca, and it maximizes the
number of scalar accesses which become accesses to their own alloca and
candidates for promotion.
Third, we re-walk the uses of the alloca and assign each specific memory
access to all the partitions touched so that we have dense use-lists for
each partition.
Finally, we build a new, smaller alloca for each partition and rewrite
each use of that partition to use the new alloca. During this phase the
pass will also work very hard to transform uses of an alloca into a form
suitable for promotion, including forming vector operations, speculating
loads throguh PHI nodes and selects, etc.
After splitting is complete, each newly refined alloca that is
a candidate for promotion to a scalar SSA value is run through mem2reg.
There are lots of reasonably detailed comments in the source code about
the design and algorithms, and I'm going to be trying to improve them in
subsequent commits to ensure this is well documented, as the new pass is
in many ways more complex than the old one.
Some of this is still a WIP, but the current state is reasonbly stable.
It has passed bootstrap, the nightly test suite, and Duncan has run it
successfully through the ACATS and DragonEgg test suites. That said, it
remains behind a default-off flag until the last few pieces are in
place, and full testing can be done.
Specific areas I'm looking at next:
- Improved comments and some code cleanup from reviews.
- SSAUpdater and enabling this pass inside the CGSCC pass manager.
- Some datastructure tuning and compile-time measurements.
- More aggressive FCA splitting and vector formation.
Many thanks to Duncan Sands for the thorough final review, as well as
Benjamin Kramer for lots of review during the process of writing this
pass, and Daniel Berlin for reviewing the data structures and algorithms
and general theory of the pass. Also, several other people on IRC, over
lunch tables, etc for lots of feedback and advice.
llvm-svn: 163883
This is mostly documentation for the new machine model. It is designed
to be flexible, easy to incrementally refine for a subtarget, and
provide all the information that MachineScheduler will need.
If all goes well, I will follow up with an example of the new model in
use for ARM.
llvm-svn: 163877
- Enhance the fix to PR12312 to support wider integer, such as 256-bit
integer. If more than 1 fully evaluated vectors are found, POR them
first followed by the final PTEST.
llvm-svn: 163832
- Find a legal vector type before casting and extracting element from it.
- As the new vector type may have more than 2 elements, build the final
hi/lo pair by BFS pairing them from bottom to top.
llvm-svn: 163830
Add a PatFrag to match X86tcret using 6 fixed registers or less. This
avoids folding loads into TCRETURNmi64 using 7 or more volatile
registers.
<rdar://problem/12282281>
llvm-svn: 163819
48-bit if necessary, in order to reduce the generated code size.
We have 900 cases not covered by OpcodeInfo in ATT AsmWriter and more in Intel
AsmWriter and ARM AsmWriter.
This patch reduced the clang Release build size by 50k, running on a Mac Pro.
llvm-svn: 163814
Andrew Trick