This was planned since its introduction, but wasn't rolled out for a
little bit longer than intended (ahem...8 years).
All in-tree targets have now been adjusted to call
setMaxAtomicSizeInBitsSupported explicitly where required, so this
should be a no-op. The docs in docs/Atomics.rst already claimed the
default was 0, so that doesn't need updating.
Cleanup work towards removing the method Type::getPointerTo.
If a call to Type::getPointerTo is used solely to support an unneeded
pointer-cast, remove the call entirely.
This change is for AArch32 and not strictly needed, but it ensures that
we follow the model that direct accesses are only emitted for dso_local
and we do not need TargetMachine::shouldAssumeDSOLocal to force
dso_local for a dso_preemptable variable.
There is no behavior change to the arm/arm64 configurations listed in
commit 5888dee7d04748744743a35d3aef030018bdc275.
The issue #55208 noticed that std::rint is vectorized by the
SLPVectorizer, but a very similar function, std::lrint, is not.
std::lrint corresponds to ISD::LRINT in the SelectionDAG, and
std::llrint is a familiar cousin corresponding to ISD::LLRINT. Now,
neither ISD::LRINT nor ISD::LLRINT have a corresponding vector variant,
and the LangRef makes this clear in the documentation of llvm.lrint.*
and llvm.llrint.*.
This patch extends the LangRef to include vector variants of
llvm.lrint.* and llvm.llrint.*, and lays the necessary ground-work of
scalarizing it for all targets. However, this patch would be devoid of
motivation unless we show the utility of these new vector variants.
Hence, the RISCV target has been chosen to implement a custom lowering
to the vfcvt.x.f.v instruction. The patch also includes a CostModel for
RISCV, and a trivial follow-up can potentially enable the SLPVectorizer
to vectorize std::lrint and std::llrint, fixing #55208.
The patch includes tests, obviously for the RISCV target, but also for
the X86, AArch64, and PowerPC targets to justify the addition of the
vector variants to the LangRef.
We currently have log, log2, log10, exp and exp2 intrinsics. Add exp10
to fix this asymmetry. AMDGPU already has most of the code for f32
exp10 expansion implemented alongside exp, so the current
implementation is duplicating nearly identical effort between the
compiler and library which is inconvenient.
https://reviews.llvm.org/D157871
The change introduces intrinsics 'get_fpmode', 'set_fpmode' and
'reset_fpmode'. They manage all target dynamic floating-point control
modes, which include, for instance, rounding direction, precision,
treatment of denormals and so on. The intrinsics do the same
operations as the C library functions 'fegetmode' and 'fesetmode'. By
default they are lowered to calls to these functions.
Two main use cases are supported by this implementation.
1. Local modification of the control modes. In this case the code
usually has a pattern (in pseudocode):
saved_modes = get_fpmode()
set_fpmode(<new_modes>)
...
<do operations under the new modes>
...
set_fpmode(saved_modes)
In the case when it is known that the current FP environment is default,
the code may be shorter:
set_fpmode(<new_modes>)
...
<do operations under the new modes>
...
reset_fpmode()
Such patterns appear not only in user code but also in implementations
of various FP controlling pragmas. In particular, the implementation of
`#pragma STDC FENV_ROUND` requires similar code if the target does not
support static rounding mode.
2. Portable control of FP modes. Usually FP control modes are set by
writing to some control register. Different targets have different
layout of this register, the way the register is accessed also may be
different. Using set of target-specific definitions for the control
register bits together with these intrinsic functions provides enough
portable way to handle control modes across wide range of hardware.
This change defines only llvm intrinsic function, which implement the
access required for the aforementioned use cases.
Differential Revision: https://reviews.llvm.org/D82525
This adds some basic handling for bf16 constants, attempting to treat them a
lot like fp16 constants where it can. Zero immediates get lowered to FMOVH0,
others either get lowered to FMOVWHr(MOVi32imm) or use FMOVHi if they can.
Without fp16 they get expanded. This may not always be optimal, but fixes a gap
in our lowering. See llvm/test/CodeGen/AArch64/f16-imm.ll for the equivalent
fp16 test.
Differential Revision: https://reviews.llvm.org/D156649
Add an intrinsic which returns the two pieces as multiple return
values. Alternatively could introduce a pair of intrinsics to
separately return the fractional and exponent parts.
AMDGPU has native instructions to return the two halves, but could use
some generic legalization and optimization handling. For example, we
should be able to handle legalization of f16 on older targets, and for
bf16. Additionally antique targets need a hardware workaround which
would be better handled in the backend rather than in library code
where it is now.
This patch introduces the reduction intrinsic for floating point minimum
and maximum which has the same semantics (for NaN and signed zero) as
llvm.minimum and llvm.maximum.
Reviewed-By: nikic
Differential Revision: https://reviews.llvm.org/D152370
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.
The change implements intrinsics 'get_fpenv', 'set_fpenv' and 'reset_fpenv'.
They are used to read floating-point environment, set it or reset to
some default state. They do the same actions as C library functions
'fegetenv' and 'fesetenv'. By default these intrinsics are lowered to calls
to these functions.
The new intrinsics specify FP environment as a value of integer type, it
is convenient of most targets where the FP state is a content of some
register. Some targets however use long representations. On X86 the size
of FP environment is 256 bits, and even half of this size is not a legal
ibteger type. To facilitate legalization in such cases, two sets of DAG
nodes is used. Nodes GET_FPENV and SET_FPENV are used when FP
environment may be represented by a legal integer type. Nodes
GET_FPENV_MEM and SET_FPENV_MEM consider FP environment as a region in
memory, much like `fesetenv` and `fegetenv` do. They are used when
target has long representation for floationg-point state.
Differential Revision: https://reviews.llvm.org/D71742
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
This is rework of;
- rG13e77db2df94 (r328395; MVT)
Since `LowLevelType.h` has been restored to `CodeGen`, `MachinveValueType.h`
can be restored as well.
Depends on D148767
Differential Revision: https://reviews.llvm.org/D149024
Default MaxDivRemBitWidthSupported to 128, so that divisions larger
than 128 bits are always expanded, without requiring additional
configuration from the target.
Note that this may still emit calls to __udivti3 on 32-bit targets,
which likely don't have an implementation of that builtin. However,
I believe this is sufficient to fix
https://github.com/llvm/llvm-project/issues/60531, because Zig must
already be defining those builtins.
Differential Revision: https://reviews.llvm.org/D144871
This function was added for ARM targets, but aligning global/stack pointer
arguments passed to memcpy/memmove/memset can improve code size and
performance for all targets that don't have fast unaligned accesses.
This adds a generic implementation that adjusts the alignment to pointer
size if unaligned accesses are slow.
Review D134168 suggests that this significantly improves performance on
synthetic benchmarks such as Dhrystone on RV32 as it avoids memcpy() calls.
Reviewed By: efriedma
Differential Revision: https://reviews.llvm.org/D134282
This was incorrectly setting dereferenceable on unaligned
operands. getLoadMemOperandFlags does the alignment dereferenceabilty
check without alignment, and then both paths went on to check
isDereferenceableAndAlignedPointer. Make getLoadMemOperandFlags check
isDereferenceableAndAlignedPointer, and remove the second call.
Single-element vectors are legalized by splitting,
so the the memory operations would also get scalarized.
While we do have some support to reconstruct scalarized loads,
we clearly don't catch everything.
The comment for the affected AArch64 store suggests that
having two stores was the desired outcome in the first place.
This was showing as a source of *many* regressions
with more aggressive ZERO_EXTEND_VECTOR_INREG recognition.
As stated in
https://discourse.llvm.org/t/rfc-llc-add-expandlargeintfpconvert-pass-for-fp-int-conversion-of-large-bitint/65528,
this implementation is very similar to ExpandLargeDivRem, which expands
‘fptoui .. to’, ‘fptosi .. to’, ‘uitofp .. to’, ‘sitofp .. to’ instructions
with a bitwidth above a threshold into auto-generated functions. This is
useful for targets like x86_64 that cannot lower fp convertions with more
than 128 bits. The expanded nodes are referring from the IR generated by
`compiler-rt/lib/builtins/floattidf.c`, `compiler-rt/lib/builtins/fixdfti.c`,
and etc.
Corner cases:
1. For fp16: as there is no related builtins added in compliler-rt. So I
mainly utilized the fp32 <-> fp16 lib calls to implement.
2. For fp80: as this pass is soft fp emulation and no fp80 instructions can
help in this problem. I recommend users to deprecate this usage. For now, the
implementation uses fp128 as the temporary conversion type and inserts
fptrunc/ext at top/end of the function.
3. For bf16: as clang FE currently doesn't support bf16 algorithm operations
(convert to int, float, +, -, *, ...), this patch doesn't consider bf16 for
now.
4. For unsigned FPToI: since both default hardware behaviors and libgcc are
ignoring "returns 0 for negative input" spec. This pass follows this old way
to ignore unsigned FPToI. See this example:
https://gcc.godbolt.org/z/bnv3jqW1M
The end-to-end tests are uploaded at https://reviews.llvm.org/D138261
Reviewed By: LuoYuanke, mgehre-amd
Differential Revision: https://reviews.llvm.org/D137241
This patch is an alternative of D100091. It solved the problems in `f80` type lowering.
Reviewed By: LuoYuanke
Differential Revision: https://reviews.llvm.org/D137946
A target can return if a misaligned access is 'fast' as defined
by the target or not. In reality there can be different levels
of 'fast' and 'slow'. This patch changes the boolean 'Fast'
argument of the allowsMisalignedMemoryAccesses family of functions
to an unsigned representing its speed.
A target can still define it as it wants and the direct translation
of the current code uses 0 and 1 for current false and true. This
makes the change an NFC.
Subsequent patch will start using an actual value of speed in
the load/store vectorizer to compare if a vectorized access going
to be not just fast, but not slower than before.
Differential Revision: https://reviews.llvm.org/D124217
I want to default all VP operations to Expand. These 2 were blocking
because VE doesn't support them and the tests were expecting them
to fail a specific way. Using Expand caused them to fail differently.
Seemed better to emulate them using operations that are supported.
@simoll mentioned on Discord that VE has some expansion downstream. Not
sure if its done like this or in the VE target.
Reviewed By: frasercrmck, efocht
Differential Revision: https://reviews.llvm.org/D133514
Also remove new-pass-manager version of ExpandLargeDivRem because there is no way
yet to access TargetLowering in the new pass manager.
Differential Revision: https://reviews.llvm.org/D133691
LLVM contains a helpful function for getting the size of a C-style
array: `llvm::array_lengthof`. This is useful prior to C++17, but not as
helpful for C++17 or later: `std::size` already has support for C-style
arrays.
Change call sites to use `std::size` instead.
Differential Revision: https://reviews.llvm.org/D133429
The main difference is that this preserves intermediate rounding steps,
which the other route doesn't. This aligns bfloat16 more with half
floats, which use this path on most targets.
I didn't understand what the difference was between these softening
approaches when I first added bfloat lowerings, would be nice if we only
had one of them.
Based on @pengfei 's D131502
Differential Revision: https://reviews.llvm.org/D133207
TragetLowering had two last InstructionCost related `getTypeLegalizationCost()`
and `getScalingFactorCost()` members, but all other costs are processed in TTI.
E.g. it is not comfortable to use other TTI members in these two functions
overrided in a target.
Minor refactoring: `getTypeLegalizationCost()` now doesn't need DataLayout
parameter - it was always passed from TTI.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D117723
This adds similar heuristics to G_GLOBAL_VALUE, querying the cost of
materializing a specific constant in code size. Doing so prevents us from
sinking constants which require multiple instructions to generate into
use blocks.
Code size savings on CTMark -Os:
Program size.__text
before after diff
ClamAV/clamscan 381940.00 382052.00 0.0%
lencod/lencod 428408.00 428428.00 0.0%
SPASS/SPASS 411868.00 411876.00 0.0%
kimwitu++/kc 449944.00 449944.00 0.0%
Bullet/bullet 463588.00 463556.00 -0.0%
sqlite3/sqlite3 284696.00 284668.00 -0.0%
consumer-typeset/consumer-typeset 414492.00 414424.00 -0.0%
7zip/7zip-benchmark 595244.00 594972.00 -0.0%
mafft/pairlocalalign 247512.00 247368.00 -0.1%
tramp3d-v4/tramp3d-v4 372884.00 372044.00 -0.2%
Geomean difference -0.0%
Differential Revision: https://reviews.llvm.org/D130554
This is modeled after the half-precision fp support. Two new nodes are
introduced for casting from and to bf16. Since casting from bf16 is a
simple operation I opted to always directly lower it to integer
arithmetic. The other way round is more complicated if you want to
preserve IEEE semantics, so it's handled by a new __truncsfbf2
compiler-rt builtin.
This is of course very bare bones, but sufficient to get a semi-softened
fadd on x86.
Possible future improvements:
- Targets with bf16 conversion instructions can now make fp_to_bf16 legal
- The software conversion to bf16 can be replaced by a trivial
implementation under fast math.
Differential Revision: https://reviews.llvm.org/D126953
Adds MVT::v128i2, MVT::v64i4, and implied MVT::i2, MVT::i4.
Keeps MVT::i2, MVT::i4 lowering actions as expand, which should be
removed once targets set this explicitly.
Adjusts 11 lit tests to reflect slightly different behavior during
DAG combine.
Differential Revision: https://reviews.llvm.org/D125247
Adds MVT::v128i2, MVT::v64i4, and implied MVT::i2, MVT::i4.
Keeps MVT::i2, MVT::i4 lowering actions as `expand`, which should be
removed once targets set this explicitly.
Adjusts 11 lit tests to reflect slightly different behavior during
DAG combine.
Differential Revision: https://reviews.llvm.org/D125247
