The X86 backend only needs to know whether structure return is via an
sret pointer. This removes the categorization enumeration and
adjusts, templatizes and renames the related functions.
Differential Revision: https://reviews.llvm.org/D109966
Instead just emit a warning that analysis failed and the result will be treated conservatively.
Differential Revision: https://reviews.llvm.org/D111217
As suggested on D111024, we should treat getCmpSelInstrCost calls without a specific predicate as matching the worst case predicate cost.
These regressions will be addressed with a mixture of D111024 and fixing other specific getCmpSelInstrCost calls to have realistic predicates.
Seem to cause test failures in compiler-rt.
Revert "[SystemZ] Implement memcmp of variable length with CLC."
This reverts commit 7a4e9a0c73.
Revert "[SystemZ] Implement memcpy of variable length with MVC."
This reverts commit c6c13c58ee.
This patch allows LLDB to accept register sizes which are not aligned
to 8 bits bitsize boundary. This fixes a crash in LLDB when connecting
to OpenOCD stub. GDB xml description allows for non-aligned bit lengths
but they are rounded off to nearest byte during transfer. In case of
OpenOCD some of SOC specific system registers were less than a single
byte in length and were causing LLDB to crash.
Reviewed By: labath
Differential Revision: https://reviews.llvm.org/D111131
On Ubuntu Focal x13 is used by something in the process of calling
sched_yield. Causing the test to fail depending on when the thread
is stopped.
Adding x14 works around this and the test passes consistently.
Not switching to only x14 because that could make other platforms
fail. With both we'll always find at least one and even if both
values are present we'll only get one report.
Reviewed By: oontvoo, vitalybuka
Differential Revision: https://reviews.llvm.org/D110931
We were previously silently generating incorrect code when extracting a
fixed-width vector from a scalable vector. This is worse than crashing,
since the user will have no indication that this is currently unsupported
behaviour. I have fixed the code to only perform DAG combines when safe
to do so, i.e. the input and output vectors are both fixed-width or
both scalable.
Test added here:
CodeGen/AArch64/sve-extract-scalable-vector.ll
Differential revision: https://reviews.llvm.org/D110624
The currently implementation of funcrefs is broken since it is putting
the funcref itself on the stack before the call_indirect. Instead what
should be on the stack is the constant 0, which is the index at which
we store the funcref in __funcref_call_table.
Reviewed By: tlively
Differential Revision: https://reviews.llvm.org/D111152
This is a non-functional change to remove the duplicate
WasmAddressSpace enum and refactor reftype predicates by moving them
to the Utilities source file.
Reviewed By: tlively
Differential Revision: https://reviews.llvm.org/D111144
The builtin for vec_orc has support for the following two signatures,
but currently the compiler marks it ambiguous:
vector float vec_orc(vector float, vector float)
vector double vec_orc(vector double, vector double)
This patch implements these two builtins.
Differential revision: https://reviews.llvm.org/D110858
Script for automatic 'opt' pipeline reduction for when using the new
pass-manager (NPM). Based around the '-print-pipeline-passes' option.
The reduction algorithm consists of several phases (steps).
Step #0: Verify that input fails with the given pipeline and make note of the
error code.
Step #1: Split pipeline in two starting from front and move forward as long as
first pipeline exits normally and the second pipeline fails with the expected
error code. Move on to step #2 with the IR from the split point and the
pipeline from the second invocation.
Step #2: Remove passes from end of the pipeline as long as the pipeline fails
with the expected error code.
Step #3: Make several sweeps over the remaining pipeline trying to remove one
pass at a time. Repeat sweeps until unable to remove any more passes.
Usage example:
./utils/reduce_pipeline.py --opt-binary=./build-all-Debug/bin/opt --input=input.ll --output=output.ll --passes=PIPELINE [EXTRA-OPT-ARGS ...]
Differential Revision: https://reviews.llvm.org/D110908
Implement min and max using the newly introduced std operations instead of relying on compare and select.
Reviewed By: dcaballe
Differential Revision: https://reviews.llvm.org/D111170
These runtime callbacks are supposed to be non-instrumented,
we can't handle runtime recursion well, nor can we afford
explicit recursion checks in the hot functions (memory access,
function entry/exit).
It used to work (not crash), but it won't work with the new runtime.
Mark all runtime callbacks as non-instrumented.
Reviewed By: vitalybuka
Differential Revision: https://reviews.llvm.org/D111157
If these blocks are unreachable, then we can discard all of the instructions.
However, keep the block around because it may have an address taken or the
block may have a stale reference from a PHI somewhere. Instead of finding
those PHIs and fixing them up, just leave the block empty.
Differential Revision: https://reviews.llvm.org/D111201
PassBuilder.cpp is the slowest LLVM file to compile (if only building X86).
This makes PassBuilder.o a little faster to compile and a little smaller
as well.
These methods are not performance critical at all but are instantiated many times.
83M -> 72M instructions according to perf stat.
Reviewed By: rnk
Differential Revision: https://reviews.llvm.org/D110782
Currently we're limited to 32 bit ints in diagnostics.
With support for 4GB alignments coming soon, we need to report 4GB as the max alignment allowed.
I've tested that this does indeed properly print 2^32.
Reviewed By: rsmith
Differential Revision: https://reviews.llvm.org/D111184
This fixes round-trip / ambiguity when an operation in the standard dialect would
have the same name as an operation in the default dialect.
Differential Revision: https://reviews.llvm.org/D111204
Core changes are:
Be explicit about desired balance between missing true positives and reporting false positives.
Mention the opt-out mechanism.
Provide links to background, and give description of who to contact if needed.
Differential Revision: https://reviews.llvm.org/D110873
Without this change _term instructions can be removed during
critical edge splitting.
Reviewed By: foad
Differential Revision: https://reviews.llvm.org/D111126
In basic_string and vector, we've been encapsulating all exception
throwing code paths in helper functions of a base class, which are defined
in the compiled library. For example, __vector_base_common defines two
methods, __throw_length_error() and __throw_out_of_range(), and the class
is externally instantiated in the library. This was done a long time ago,
but after investigating, I believe the goal of the current design was to:
1. Encapsulate the code to throw an exception (which is non-trivial) in
an externally-defined function so that the important code paths that
call it (e.g. vector::at) are free from that code. Basically, the
intent is for the "hot" code path to contain a single conditional jump
(based on checking the error condition) to an externally-defined function,
which handles all the exception-throwing business.
2. Avoid defining this exception-throwing function once per instantiation
of the class template. In other words, we want a single copy of
__throw_length_error even if we have vector<int>, vector<char>, etc.
3. Encapsulate the passing of the container-specific string (i.e. "vector"
and "basic_string") to the underlying exception-throwing function
so that object files don't contain those duplicated string literals.
For example, we'd like to have a single "vector" string literal for
passing to `std::__throw_length_error` in the library, instead of
having one per translation unit.
However, the way this is achieved right now has two problems:
- Using a base class and exporting it is really weird - I've been confused
about this ever since I first saw it. It's just a really unusual way of
achieving the above goals. Also, it's made even worse by the fact that
the definitions of __throw_length_error and __throw_out_of_range appear
in the headers despite always being intended to be defined in the compiled
library (via the extern template instantiation).
- We end up exporting those functions as weak symbols, which isn't great
for load times. Instead, it would be better to export those as strong
symbols from the library.
This patch fixes those issues while retaining ABI compatibility (e.g. we
still export the exact same symbols as before). Note that we need to
keep the base classes as-is to avoid breaking the ABI of someone who
might inherit from std::basic_string or std::vector.
Differential Revision: https://reviews.llvm.org/D111173
This removes `WasmTagType`. `WasmTagType` contained an attribute and a
signature index:
```
struct WasmTagType {
uint8_t Attribute;
uint32_t SigIndex;
};
```
Currently the attribute field is not used and reserved for future use,
and always 0. And that this class contains `SigIndex` as its property is
a little weird in the place, because the tag type's signature index is
not an inherent property of a tag but rather a reference to another
section that changes after linking. This makes tag handling in the
linker also weird that tag-related methods are taking both `WasmTagType`
and `WasmSignature` even though `WasmTagType` contains a signature
index. This is because the signature index changes in linking so it
doesn't have any info at this point. This instead moves `SigIndex` to
`struct WasmTag` itself, as we did for `struct WasmFunction` in D111104.
In this CL, in lib/MC and lib/Object, this now treats tag types in the
same way as function types. Also in YAML, this removes `struct Tag`,
because now it only contains the tag index. Also tags set `SigIndex` in
`WasmImport` union, as functions do.
I think this makes things simpler and makes tag handling more in line
with function handling. These two shares similar properties in that both
of them have signatures, but they are kind of nominal so having the same
signature doesn't mean they are the same element.
Also a drive-by fix: the reserved 'attirubute' part's encoding changed
from uleb32 to uint8 a while ago. This was fixed in lib/MC and
lib/Object but not in YAML. This doesn't change object files because the
field's value is always 0 and its encoding is the same for the both
encoding.
This is effectively NFC; I didn't mark it as such just because it
changed YAML test results.
Reviewed By: sbc100, tlively
Differential Revision: https://reviews.llvm.org/D111086
This is less brittle than hand-picking the substitutions that we
pass to the test, since a config could theorically use non-base
substitutions as well (such as defining %{flags} in terms of another
substitution like %{include}).
Also, print the decoded substitutions, which makes it much easier
to debug the test when it fails.
Differential Revision: https://reviews.llvm.org/D111179
This patch extends Linalg core vectorization with support for min/max reductions
in linalg.generic ops. It enables the reduction detection for min/max combiner ops.
It also renames MIN/MAX combining kinds to MINS/MAXS to make the sign explicit for
floating point and signed integer types. MINU/MAXU should be introduce din the future
for unsigned integer types.
Reviewed By: pifon2a, ThomasRaoux
Differential Revision: https://reviews.llvm.org/D110854
These are meant to check a future patch which recurses through operands of SCEVs, but because all SCEVs are trivially bounded by function entry, we need to arrange the trivial scope not to be valid. (i.e. we specifically need a lower defining scope)
This also removes the need to disable the mandatory inlining phase in
tests.
In a departure from the previous remark, we don't output a 'cost' in
this case, because there's no such thing. We just report that inlining
happened because of the attribute.
Differential Revision: https://reviews.llvm.org/D110891
Nathan Sidwell