If the first tree element is vectorize and the second is gather, it
still might be profitable to vectorize it if the gather node contains
less scalars to vectorize than the original tree node. It might be
profitable to use shuffles.
Differential Revision: https://reviews.llvm.org/D101397
This is useful for running in batch mode.
Getting the SymbolID from via getSymbolInfo may give SymbolID
of a symbol different from that located by LocateSymbolAt (they
have different semantics of choosing the symbol.)
Differential Revision: https://reviews.llvm.org/D101388
Language options are not available when a target is being created,
thus, a new method is introduced. Also, some refactoring is done,
such as removing OpenCL feature macros setting from TargetInfo.
Reviewed By: Anastasia
Differential Revision: https://reviews.llvm.org/D101087
This was picking a concrete size for a physical register, and
enforcing exact match on the virtual register's type size. Some
targets add multiple types to a register class, and some are smaller
than the full bit width. For example x86 adds f32 to 128-bit xmm
registers, and AMDGPU adds i16/f16 to 32-bit registers.
It might be better to represent these cases as a copy of the full
register and an extraction of the subpart, but a lot of code assumes
you can directly copy. This will help fix the current usage of the DAG
calling convention infrastructure which is incompatible with how
GlobalISel is now using it.
The API is somewhat cumbersome here, but I just mirrored the existing
functions, except now with LLTs (and allow returning null on failure,
unlike the MVT version). I think the concept of selecting register
classes based on type is flawed to begin with, but I'm trying to keep
this compatible with the existing handling.
This patch simplifies the calculation of certain costs in
getInstructionCost when isScalarAfterVectorization() returns a true value.
There are a few places where we multiply a cost by a number N, i.e.
unsigned N = isScalarAfterVectorization(I, VF) ? VF.getKnownMinValue() : 1;
return N * TTI.getArithmeticInstrCost(...
After some investigation it seems that there are only these cases that occur
in practice:
1. VF is a scalar, in which case N = 1.
2. VF is a vector. We can only get here if: a) the instruction is a
GEP/bitcast/PHI with scalar uses, or b) this is an update to an induction
variable that remains scalar.
I have changed the code so that N is assumed to always be 1. For GEPs
the cost is always 0, since this is calculated later on as part of the
load/store cost. PHI nodes are costed separately and were never previously
multiplied by VF. For all other cases I have added an assert that none of
the users needs scalarising, which didn't fire in any unit tests.
Only one test required fixing and I believe the original cost for the scalar
add instruction to have been wrong, since only one copy remains after
vectorisation.
I have also added a new test for the case when a pointer PHI feeds directly
into a store that will be scalarised as we were previously never testing it.
Differential Revision: https://reviews.llvm.org/D99718
Need to respect mapping/privatization of declare target variables in the
target regions if explicitly specified by the user.
Differential Revision: https://reviews.llvm.org/D99530
Tensor inputs, if not used in the body of TiledLoopOp, can be removed.
memref::CastOp can be folded into TiledLoopOp as well.
Differential Revision: https://reviews.llvm.org/D101445
This patch also refactors the way the feasible max VF is calculated,
although this is NFC for fixed-width vectors.
After this change scalable VF hints are no longer truncated/clamped
to a shorter scalable VF, nor does it drop the 'scalable flag' from
the suggested VF to vectorize with a similar VF that is fixed.
Instead, the hint is ignored which means the vectorizer is free
to find a more suitable VF, using the CostModel to determine the
best possible VF.
Reviewed By: c-rhodes, fhahn
Differential Revision: https://reviews.llvm.org/D98509
Previously printing R_386_RELATIVE relocations would trigger
`error: can't read an entry at 0x40: it goes past the end of the section (0x40)`
I found this while writing a test case for LLD (D100490).
This also includes some minor cleanup in the elf-dynamic-relcos.test
llvm-objdump test based on the newly added test.
Reviewed By: jhenderson, MaskRay
Differential Revision: https://reviews.llvm.org/D100489
The original page no longer works, so use a web.archive.org link instead.
Reviewed By: atanasyan
Differential Revision: https://reviews.llvm.org/D100949
While implementing support for the float128 routines on x86_64, I noticed
that __builtin_isinf() was returning true for 128-bit floating point
values that are not infinite when compiling with GCC and using the
compiler-rt implementation of the soft-float comparison functions.
After stepping through the assembly, I discovered that this was caused by
GCC assuming a sign-extended 64-bit -1 result, but our implementation
returns an enum (which then has zeroes in the upper bits) and therefore
causes the comparison with -1 to fail.
Fix this by using a CMP_RESULT typedef and add a static_assert that it
matches the GCC soft-float comparison return type when compiling with GCC
(GCC has a __libgcc_cmp_return__ mode that can be used for this purpose).
Also move the 3 copies of the same code to a shared .inc file.
Reviewed By: compnerd
Differential Revision: https://reviews.llvm.org/D98205
This has been rather useful in our downstream CHERI target where we want
to run tests both with addrspace(0) and addrspace(200) pointers.
With this patch we can prefix the opt command with
`sed -e 's/addrspace(200)/addrspace(0)/g' -e 's/-A200-P200-G200//g'` to
test both cases using the same IR input.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D95137
There are post-commit notest for e4c61d5 that suggest
the test is failing on certain bots. It looks like
the code there isn't being moved, which suggests
cost-model involvement, which suggests that we need to
hardcode the target triple.
Hopefully this helps?
When using the -enable-strict-reductions flag where UF>1 we generate multiple
Phi nodes, though only one of these is used as an input to the vector.reduce.fadd
intrinsics. The unused Phi nodes are removed later by instcombine.
This patch changes widenPHIInstruction/fixReduction to only generate
one Phi, and adds an additional test for unrolling to strict-fadd.ll
Reviewed By: david-arm
Differential Revision: https://reviews.llvm.org/D100570
Checks if introspection support is available set output kind parser.
If it isn't present the auto complete will not suggest `srcloc` and an error query will be reported if a user tries to access it.
Reviewed By: steveire
Differential Revision: https://reviews.llvm.org/D101365
Prevent cases in which the start value of IV is bigger than bound for
increasing.
Prevent cases in which the start value of IV is smaller than bound for
decreasing.
Differential Revision: https://reviews.llvm.org/D101174
Both, `shape.broadcast` and `shape.cstr_broadcastable` accept dynamic and static
extent tensors. If their operands are casted, we can use the original value
instead.
Differential Revision: https://reviews.llvm.org/D101376
This patch fixes a crash in LiveDebugVariables for inputs where a
DBG_VALUE_LIST had 64 or more debug operands. This was triggering an
assert, which was added under the assumption that only bad CodeGen would
result in such a limit being hit, but relatively simple source files
that result in these incredibly long debug values have been found, so
this assert has been changed to a condition that drops the debug value
if it is not met.
Differential Revision: https://reviews.llvm.org/D101373
MAXVAL, MINVAL, MAXLOC and MINLOC are already implemented in extrema.cpp
as MaxvalCharacter, MinvalDim etc. Therefore, the interfaces in
character.h are redundant and should be removed to avoid confusion.
Differential Revision: https://reviews.llvm.org/D101354
This is a partial revert of b4537c3f51
based on the discussion in https://reviews.llvm.org/D101194. Rather
than using the getMultiarchTriple, we use the getTripleString.
Add a section attribute to LLVM_GlobalOp, during module translation attribute value is propagated to llvm
Reviewed By: sgrechanik, ftynse, mehdi_amini
Differential Revision: https://reviews.llvm.org/D100947
In terms of readability, the `enum CFIMoveType` didn't better document what it
intends to convey i.e. the type of CFI section that gets emitted.
Reviewed By: dblaikie, MaskRay
Differential Revision: https://reviews.llvm.org/D76519
Reverts parts of https://reviews.llvm.org/D17183, but keeps the
resetDataLayout() API and adds an assert that checks that datalayout string and
user label prefix are in sync.
Approach 1 in https://reviews.llvm.org/D17183#2653279
Reduces number of TUs build for 'clang-format' from 689 to 575.
I also implemented approach 2 in D100764. If someone feels motivated
to make us use DataLayout more, it's easy to revert this change here
and go with D100764 instead. I don't plan on doing more work in this
area though, so I prefer going with the smaller, more self-consistent change.
Differential Revision: https://reviews.llvm.org/D100776
This adds `mlirOperationSetOperand` to the IR C API, similar to the
function to get an operand.
In the Python API, this adds `operands[index] = value` syntax, similar
to the syntax to get an operand with `operands[index]`.
Reviewed By: mehdi_amini
Differential Revision: https://reviews.llvm.org/D101398
Add the `getCapsule()` and `createFromCapsule()` methods to the
PyValue class, as well as the necessary interoperability.
Reviewed By: stellaraccident
Differential Revision: https://reviews.llvm.org/D101090
Alexey Bataev