AMDGPU has native instructions and target intrinsics for this, but
these really should be subject to legalization and generic
optimizations. This will enable legalization of f16->f32 on targets
without f16 support.
Implement a somewhat horrible inline expansion for targets without
libcall support. This could be better if we could introduce control
flow (GlobalISel version not yet implemented). Support for strictfp
legalization is less complete but works for the simple cases.
Currently, a node and its users are added back to the worklist in reverse topological order after it is combined. This diff changes that order to be topological. This is part of a larger migration to get the DAGCombiner to process nodes in topological order.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D127115
The build failure should be fixed by de681d53. Follow-up refactor will
be done in future patches.
This reverts commit e7c5ced0b9f0551ea17e1d2b48be86f03a772c59.
This is an attempt to reland D42600 and enabling this optimisation by default.
This also resolves the issue pointed out in the context of PGO build.
Differential Revision: https://reviews.llvm.org/D42600
Commit 8064caf83fb166b709bfe0e7641c5181341cb064 added a call
to a function that performs this combine without checking whether
the target supports FPCVT. This caused asserts to trip on BE bots
as the default target does not have this feature.
There are two motivations.
`-fno-pic -fstack-protector -mstack-protector-guard=global` created
`__stack_chk_guard` is referenced directly on all ELF OSes except FreeBSD.
This patch allows referencing the symbol indirectly with
-fno-direct-access-external-data.
Some Linux kernel folks want
`-fno-pic -fstack-protector -mstack-protector-guard-reg=gs -mstack-protector-guard-symbol=__stack_chk_guard`
created `__stack_chk_guard` to be referenced directly, avoiding
R_X86_64_REX_GOTPCRELX (even if the relocation may be optimized out by the linker).
https://github.com/llvm/llvm-project/issues/60116
Why they need this isn't so clear to me.
---
Add module flag "direct-access-external-data" and set the dso_local property of
the stack protector symbol. The module flag can benefit other LLVMCodeGen
synthesized symbols that are not represented in LLVM IR.
Nowadays, with `-fno-pic` being uncommon, ideally we should set
"direct-access-external-data" when it is true. However, doing so would require
~90 clang/test tests to be updated, which are too much.
As a compromise, we set "direct-access-external-data" only when it's different
from the implied default value.
Reviewed By: nickdesaulniers
Differential Revision: https://reviews.llvm.org/D150841
On PowerPC VSX targets, fp-to-int will be transformed into xscv with
mfvsr. When the result is to be stored, mfvsr can be replaced by a
direct store.
This change simplifies the optimization by using existing fp-to-int
code, which helps CSE and handling strictfp cases.
Reviewed By: shchenz
Differential Revision: https://reviews.llvm.org/D141473
This is a follow-up to b71edfaa4ec3c998aadb35255ce2f60bba2940b0
since I forgot the lit.local.cfg files in that one.
Reformatting is done with `black`.
If you end up having problems merging this commit because you
have made changes to a python file, the best way to handle that
is to run git checkout --ours <yourfile> and then reformat it
with black.
If you run into any problems, post to discourse about it and
we will try to help.
RFC Thread below:
https://discourse.llvm.org/t/rfc-document-and-standardize-python-code-style
Reviewed By: barannikov88, kwk
Differential Revision: https://reviews.llvm.org/D150762
Summary: DWARF32 is not supported for XCOFF64 under non-integrated-as mode on AIX, because system assembler will fill the debug section lengths according to DWARF64 format. While in intergrated-as mode, XCOFF64 should be able to select the DWARF format.
Reviewed By: shchenz
Differential Revision: https://reviews.llvm.org/D150181
This commit implements the serialization and deserialization of the Machine
Function's EntryValueObjects.
Depends on D149879, D149778
Differential Revision: https://reviews.llvm.org/D149880
This reverts commit 1ddfd1c8186735c62b642df05c505dc4907ffac4.
The original commit causes a Chrome build assertion failure with
ThinLTO: https://crbug.com/1443635
This patch adds the additional step of looking through AND, OR, XOR
instructions when we check the number of leading zeros.
Reviewed By: shchenz
Differential Revision: https://reviews.llvm.org/D149223
Allow shrink-wrapping past memory accesses that only access globals or
function arguments. This patch uses getUnderlyingObject to try to
identify the accessed object by a given memory operand. If it is a
global or an argument, it does not access the stack of the current
function and should not block shrink wrapping.
Note that the caller's stack may get accessed when passing an argument
via the stack, but not the stack of the current function.
This addresses part of the TODO from D63152.
Reviewed By: thegameg
Differential Revision: https://reviews.llvm.org/D149668
D37076 makes LICM duplicate instructions into exit blocks if the
instruction is free. For GEPs, the motivation appears to be that
this allows the GEP to be folded into addressing modes, while
non-foldable users outside the loop might prevent this. TBH I don't
think LICM is the place to do this (why doesn't CGP apply this
heuristic itself?) but at least I understand the motivation.
However, the transform is also applied to all other "free"
instructions, which are just that (removed during lowering and not
"folded" in some way). For such instructions, this transform seems
somewhere between useless, counter-productive (undoing CSE/GVN) and
actively incorrect. For example, this transform can duplicate freeze
instructions, which is illegal.
This patch limits the transform to just foldable GEPs, though we
might want to drop it from LICM entirely as a followup.
This is a small compile-time improvement, because querying TTI cost
model for every single instruction is expensive.
Differential Revision: https://reviews.llvm.org/D149136
With this patch an undefined mask in a shufflevector will be printed as poison.
This change is done to support the new shufflevector semantics
for undefined mask elements.
Differential Revision: https://reviews.llvm.org/D149210
D38236 moves a redundant compare instruction from the loop body to the
preheader.
It has a bug: when `MBB1 == &MBB2`, there may be only one compare instruction in the
loop. The code will lift the compare instruction to the preheader, failing to
account for the change of the compare result in a tail call, leading to a miscompile.
Suppress the compare elimination to fix https://github.com/llvm/llvm-project/issues/62294
Reviewed By: #powerpc, nemanjai
Differential Revision: https://reviews.llvm.org/D149030
This patch adds a new test that includes a vperm instruction with xxswapd as its
vector operand on little-endian Power8. The test demonstrates the constant pool
for the mask operand, which is intended to indicate the optimization of vperm
and the modification of the constant pool in subsequent patches.
Reviewed By: amyk
Differential Revision: https://reviews.llvm.org/D148942
Added a number of functions that have a clear instruction that is not
actually required. This test is added first and then a patch will be
added later in order to remove the unnecessary instructions.
The alignment of function pointers was added to the Datalayout by
D57335 but currently is unset for the Power target. This will cause us
to compute a conservative minimum alignment of one if places like
Value::getPointerAlignment.
This patch implements the function pointer alignment in the Datalayout
for the Power backend and Power targets in clang, so we can query the
value for a particular Power target.
We come up with the correct value one of two ways:
- If the target uses function descriptor objects (i.e. ELFv1 & AIX ABIs),
then a function pointer points to the descriptor, so use the alignment
we would emit the descriptor with.
- If the target doesn't use function descriptor objects (i.e. ELFv2), a
function pointer points to the global entry point, so use the minimum
alignment for code on Power (i.e. 4-bytes).
Reviewed By: nemanjai
Differential Revision: https://reviews.llvm.org/D147016
Reassociate gep (gep ptr, idx1), idx2 to gep (gep ptr, idx2), idx1
if this would make the inner GEP loop invariant and thus hoistable.
This is intended to replace an InstCombine fold that does this (in
04f61fb73d/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp (L2006)).
The problem with the InstCombine fold is that LoopInfo is an optional
dependency, so it is not performed reliably.
Differential Revision: https://reviews.llvm.org/D146813
This patch splits a restore point to allow it to only post-dominate blocks reachable by use
or def of CSRs(Callee Saved Registers)/FI(Frame Index).
Benchmarking this on SPEC2017, this gives around 4% improvement on povray and no significant change
for others.
Co-authored-by: junbuml
Differential Revision: https://reviews.llvm.org/D42600
This patch is to update pr61315.ll what was needed as part of
D146632 and caused build failures.
Reviewed By: stefanp
Differential Revision: https://reviews.llvm.org/D147675
This patch is to fix the xxperm vector operand swap condition so that the
single-use operand is in V2 to prevent copying, it also fixes the subtarget
condition to exploit the xpperm.
Reviewed By: stefanp
Differential Revision: https://reviews.llvm.org/D146632
On Power PC some data is stored in the TOC. This pass adds statistics
to show how many entries are emitted to the TOC and what types of
entries those are.
Reviewed By: amyk
Differential Revision: https://reviews.llvm.org/D146325
Power ISA 3.0 introduced new 'test data class' instructions, which
accept flags for: NaN/Infinity/Zero/Denormal. This instruction can be
used to implement custom lowering for llvm.is.fpclass, but some extra
bits provided by the intrinsic are missing (normal and QNaN/SNaN).
For those categories not natively supported, this patch uses a two-way
or three-way combination to implement correct behavior.
Reviewed By: sepavloff, shchenz
Differential Revision: https://reviews.llvm.org/D140381
This patch renames the `mroptr` option to `mxcoff-roptr` to indicate in the option itself that it is xcoff specific.
Reviewed By: hubert.reinterpretcast
Differential Revision: https://reviews.llvm.org/D147161
This patch adds the initial support for vector functions and register banks
within GlobalISel. With this patch, we are able to support simple functions that
return vectors, and also functions that perform simple operations.
This patch also:
- Legalizes vector types for G_AND, G_OR, G_XOR, G_ADD, G_SUB, G_BITCAST, G_FADD, G_FSUB
- Introduce initial support for bitcasting (that will need to be extended upon)
- Add various different test cases to for test vector support within GlobalISel
Differential Revision: https://reviews.llvm.org/D137785
This patch partially implements the parameter passing rules outlined in the
ELFv2 ABI within TableGen. Specifically, it implements the parameter assignment
of integers, floats, and vectors within registers - where the GPR numbering will
be "skipped" depending on the ordering of floats and vectors that appear within
a parameter list.
As we begin to adopt GlobalISel to the PowerPC backend, there is a need for a
TableGen definition that encapsulates the ELFv2 parameter passing rules. Thus,
this patch also changes the default calling convention that is returned within
the ccAssignFnForCall() function used in our GlobalISel implementation, and also
adds some additional testing of the calling convention that is implemented.
Future patches that build on top of this initial TableGen definition will aim to
add more of the ABI complexities, including support for additional types and
also in-memory arguments.
Differential Revision: https://reviews.llvm.org/D137504
This DAG combine is correct on little endian targets but
is incorrect on big endian targets.
Add big endian code to correct it.
Differential revision: https://reviews.llvm.org/D146460
This patch adds an `llc` option `-mroptr` to specify storage locations for constant pointers on AIX.
When the `-mroptr` option is specified, constant pointers, virtual function tables, and virtual type tables are placed in read-only storage. Otherwise, by default, pointers, virtual function tables, and virtual type tables are placed are placed in read/write storage.
https://reviews.llvm.org/D144190 enables the `-mroptr` option for `clang`.
Reviewed By: hubert.reinterpretcast, stephenpeckham, myhsu, MaskRay, serge-sans-paille
Differential Revision: https://reviews.llvm.org/D144189
Summary: A R_REF relocation as a non-relocating reference is required to prevent garbage collection (by the binder) of the ref symbol in object generation.
Reviewed By: shchenz
Differential Revision: https://reviews.llvm.org/D144356
When doing store constant vector/scalar, some duplicated values can be reused.
Add test case and will show combiner can improve these.
Reviewed By: shchenz
Differential Revision: https://reviews.llvm.org/D146500
