Whole wave functions are functions that will run with a full EXEC mask.
They will not be invoked directly, but instead will be launched by way
of a new intrinsic, `llvm.amdgcn.call.whole.wave` (to be added in
a future patch). These functions are meant as an alternative to the
`llvm.amdgcn.init.whole.wave` or `llvm.amdgcn.strict.wwm` intrinsics.
Whole wave functions will set EXEC to -1 in the prologue and restore the
original value of EXEC in the epilogue. They must have a special first
argument, `i1 %active`, that is going to be mapped to EXEC. They may
have either the default calling convention or amdgpu_gfx. The inactive
lanes need to be preserved for all registers used, active lanes only for
the CSRs.
At the IR level, arguments to a whole wave function (other than
`%active`) contain poison in their inactive lanes. Likewise, the return
value for the inactive lanes is poison.
This patch contains the following work:
* 2 new pseudos, SI_SETUP_WHOLE_WAVE_FUNC and SI_WHOLE_WAVE_FUNC_RETURN
used for managing the EXEC mask. SI_SETUP_WHOLE_WAVE_FUNC will return
a SReg_1 representing `%active`, which needs to be passed into
SI_WHOLE_WAVE_FUNC_RETURN.
* SelectionDAG support for generating these 2 new pseudos and the
special handling of %active. Since the return may be in a different
basic block, it's difficult to add the virtual reg for %active to
SI_WHOLE_WAVE_FUNC_RETURN, so we initially generate an IMPLICIT_DEF
which is later replaced via a custom inserter.
* Expansion of the 2 pseudos during prolog/epilog insertion. PEI also
marks any used VGPRs as WWM registers, which are then spilled and
restored with the usual logic.
Future patches will include the `llvm.amdgcn.call.whole.wave` intrinsic
and a lot of optimization work (especially in order to reduce spills
around function calls).
---------
Co-authored-by: Matt Arsenault <Matthew.Arsenault@amd.com>
Co-authored-by: Shilei Tian <i@tianshilei.me>
Add `dead_on_return` attribute, which is meant to be taken advantage
by the frontend, and states that the memory pointed to by the argument
is dead upon function return. As with `byval`, it is supposed to be
used for passing aggregates by value. The difference lies in the ABI:
`byval` implies that the pointer is explicitly passed as argument to
the callee (during codegen the copy is emitted as per byval contract),
whereas a `dead_on_return`-marked argument implies that the copy
already exists in the IR, is located at a specific stack offset within
the caller, and this memory will not be read further by the caller upon
callee return – or otherwise poison, if read before being written.
RFC: https://discourse.llvm.org/t/rfc-add-dead-on-return-attribute/86871.
This flag was used to let us incrementally introduce debug records
into LLVM, however everything is now using records. It serves no
purpose now, so delete it.
For function whose vscale_range is limited to a single value we can size
scalable vectors. This aids SROA by allowing scalable vector load and
store operations to be considered for replacement whereby bitcasts
through memory can be replaced by vector insert or extract operations.
Start removing debug intrinsics support -- starting with the flag that
controls production of their replacement, debug records. This patch
removes the command-line-flag and with it the ability to switch back to
intrinsics. The module / function / block level "IsNewDbgInfoFormat"
flags get hardcoded to true, I'll to incrementally remove things that
depend on those flags.
## Purpose
This patch is one in a series of code-mods that annotate LLVM’s public
interface for export. This patch annotates the `llvm/IR`,
`llvm/IRPrinter`, and `llvm/IRReader` libraries. These annotations
currently have no meaningful impact on the LLVM build; however, they are
a prerequisite to support an LLVM Windows DLL (shared library) build.
## Background
This effort is tracked in #109483. Additional context is provided in
[this
discourse](https://discourse.llvm.org/t/psa-annotating-llvm-public-interface/85307),
and documentation for `LLVM_ABI` and related annotations is found in the
LLVM repo
[here](https://github.com/llvm/llvm-project/blob/main/llvm/docs/InterfaceExportAnnotations.rst).
The bulk of these changes were generated automatically using the
[Interface Definition Scanner (IDS)](https://github.com/compnerd/ids)
tool, followed formatting with `git clang-format`.
The following manual adjustments were also applied after running IDS on
Linux:
- Add `#include "llvm/Support/Compiler.h"` to files where it was not
auto-added by IDS due to no pre-existing block of include statements.
- Add `LLVM_ABI_FRIEND` to friend member functions declared with
`LLVM_ABI`
- Add `LLVM_TEMPLATE_ABI` and `LLVM_EXPORT_TEMPLATE` to exported
instantiated templates
- Add `LLVM_ABI` to a subset of private class methods and fields that
require export
- Add `LLVM_ABI` to a small number of symbols that require export but
are not declared in headers
- Reorder `LLVM_ABI` with `[[deprecated]]` and `[[nodiscard]]`
attributes.
## Validation
Local builds and tests to validate cross-platform compatibility. This
included llvm, clang, and lldb on the following configurations:
- Windows with MSVC
- Windows with Clang
- Linux with GCC
- Linux with Clang
- Darwin with Clang
In void functions, try to replace instruction uses
with a new non-void return. If the return type matches
the instruction, also try to directly return it.
This handles most of the cases, but doesn't try to handle
all of the weird exception related terminators.
Also doesn't try to replace argument uses, although it could. We
could also handle cases where we can insert a simple cast to an
original return value. I didn't think too hard about where to put this
in the default pass order. In many cases it obviates the need for most
of the CFG folds, but I've left it near the end initially.
I also think this is too aggressive about removing dead code, and
should leave existing dead code alone. I'm also not sure why we have
both "removeUnreachableBlocks" and EliminateUnreachableBlocks" in Utils.
Fixes#66039, fixes#107327
Currently BlockAddresses store both the Function and the BasicBlock they
reference, and the BlockAddress is part of the use list of both the
Function and BasicBlock.
This is quite awkward, because this is not really a use of the function
itself (and walks of function uses generally skip block addresses for
that reason). This also has weird implications on function RAUW (as that
will replace the function in block addresses in a way that generally
doesn't make sense), and causes other peculiar issues, like the ability
to have multiple block addresses for one block (with different
functions).
Instead, I believe it makes more sense to specify only the basic block
and let the function be implied by the BB parent. This does mean that we
may have block addresses without a function (if the BB is not inserted),
but this should only happen during IR construction.
Add support for specifying range attributes in Intrinsics.td. Use this
to specify the ucmp/scmp range [-1,2).
This case is trickier than existing intrinsic attributes, because we
need to create the attribute with the correct bitwidth. As such, the
attribute construction now needs to be aware of the function type.
We also need to be careful to no longer assign attributes on intrinsics
with invalid signatures, as we'd make invalid assumptions about the
number of arguments etc otherwise.
Fixes https://github.com/llvm/llvm-project/issues/130179.
This PR removes the old `nocapture` attribute, replacing it with the new
`captures` attribute introduced in #116990. This change is
intended to be essentially NFC, replacing existing uses of `nocapture`
with `captures(none)` without adding any new analysis capabilities.
Making use of non-`none` values is left for a followup.
Some notes:
* `nocapture` will be upgraded to `captures(none)` by the bitcode
reader.
* `nocapture` will also be upgraded by the textual IR reader. This is to
make it easier to use old IR files and somewhat reduce the test churn in
this PR.
* Helper APIs like `doesNotCapture()` will check for `captures(none)`.
* MLIR import will convert `captures(none)` into an `llvm.nocapture`
attribute. The representation in the LLVM IR dialect should be updated
separately.
In a variety of places we change the bitwidth of a parameter but don't
update the attributes.
The issue in this case is from the `range` attribute when inlining
`__memset_chk`. `optimizeMemSetChk` will replace an `i32` with an
`i8`, and if the `i32` had a `range` attr assosiated it will cause an
error.
Fixes#112633
Adds hidden kernel arguments to the function signature and marks them
inreg if they should be preloaded into user SGPRs. The normal kernarg
preloading logic then takes over with some additional checks for the
correct implicitarg_ptr alignment.
Special care is needed so that metadata for the hidden arguments is not
added twice when generating the code object.
When searching for an intrinsic name in a target specific slice of the
intrinsic name table, skip over the target prefix. For such cases,
currently the first loop iteration in `lookupLLVMIntrinsicByName` does
nothing (i.e., `Low` and `High` stay unchanged and it does not shrink
down the search window), so we can skip this useless first iteration by
skipping over the target prefix.
During inter-procedural SCCP, also infer attributes on arguments, not
just return values. This allows other non-interprocedural passes to make
use of the information later.
In `User::operator new` a single allocation is created to store the
`User` object itself, "intrusive" operands or a pointer for "hung off"
operands, and the descriptor. After allocation, details about the layout
(number of operands, how the operands are stored, if there is a
descriptor) are stored in the `User` object by settings its fields. The
`Value` and `User` constructors are then very careful not to initialize
these fields so that the values set during allocation can be
subsequently read. However, when the `User` object is returned from
`operator new` [its value is technically "indeterminate" and so reading
a field without first initializing it is undefined behavior (and will be
erroneous in
C++26)](https://en.cppreference.com/w/cpp/language/default_initialization#Indeterminate_and_erroneous_values).
We discovered this issue when trying to build LLVM using MSVC's [`/sdl`
flag](https://learn.microsoft.com/en-us/cpp/build/reference/sdl-enable-additional-security-checks?view=msvc-170)
which clears class fields after allocation (the docs say that this
feature shouldn't be turned on for custom allocators and should only
clear pointers, but that doesn't seem to match the implementation).
MSVC's behavior both with and without the `/sdl` flag is standards
conforming since a program is supposed to initialize storage before
reading from it, thus the compiler implementation changing any values
will never be observed in a well-formed program. The standard also
provides no provisions for making storage bytes not indeterminate by
setting them during allocation or `operator new`.
The fix for this is to create a set of types that encode the layout and
provide these to both `operator new` and the constructor:
* The `AllocMarker` types are used to select which `operator new` to
use.
* `AllocMarker` can then be implicitly converted to a `AllocInfo` which
tells the constructor how the type was laid out.
Change the "fixed encoding" table used for encoding intrinsic
type signature to use 16-bit encoding as opposed to 32-bit.
This results in both space and time improvements. For space,
the total static storage size (in bytes) of this info reduces by 50%:
- Current = 14193*4 (Fixed table) + 16058 + 3 (Long Table) = 72833
- New size = 14193*2 (Fixed table) + 19879 + 3 (Long Table) = 48268.
- Reduction = 50.9%
For time, with the added benchmark, we see a 7.3% speedup in
`GetIntrinsicInfoTableEntries` benchmark. Actual output of the
benchmark in included in the GitHub MR.
Every basic block that is linked into a function now has a unique
number, which can be queried using getNumber(). Numbers are densely
allocated, but not re-assigned on block removal for stability. Block
numbers are intended to be fairly stable and only be updated when
removing a several basic blocks to make sure the numbering doesn't
become too sparse.
To reduce holes in the numbering, renumberBlocks() can be called to
re-assign numbers in block order.
Additionally, getMaxBlockNumber() returns a value larger than the
largest block number, intended to pre-allocate/resize vectors.
Furthermore, this introduces the concept of a "block number epoch" --
an integer that changes after every renumbering. This is useful for
identifying use of block numbers after renumbering: on initialization,
the current epoch is stored, and on all subsequent accesses, equality
with the current epoch can be asserted.
I added a validate method to catch cases where something goes wrong,
even if I can't really imagine how invalid numbers can occur. But I
think it's better to be safe and rule out this potential source of bugs
when more things depend on the numbering.
Previous discussion in:
https://discourse.llvm.org/t/rfc-add-auxiliary-field-for-per-pass-custom-data-to-basicblock/80229
It is now translated to `<1 x i64>`, which allows the removal of a bunch
of special casing.
This _incompatibly_ changes the ABI of any LLVM IR function with
`x86_mmx` arguments or returns: instead of passing in mmx registers,
they will now be passed via integer registers. However, the real-world
incompatibility caused by this is expected to be minimal, because Clang
never uses the x86_mmx type -- it lowers `__m64` to either `<1 x i64>`
or `double`, depending on ABI.
This change does _not_ eliminate the SelectionDAG `MVT::x86mmx` type.
That type simply no longer corresponds to an IR type, and is used only
by MMX intrinsics and inline-asm operands.
Because SelectionDAGBuilder only knows how to generate the
operands/results of intrinsics based on the IR type, it thus now
generates the intrinsics with the type MVT::v1i64, instead of
MVT::x86mmx. We need to fix this before the DAG LegalizeTypes, and thus
have the X86 backend fix them up in DAGCombine. (This may be a
short-lived hack, if all the MMX intrinsics can be removed in upcoming
changes.)
Works towards issue #98272.
Functions created with createWithDefaultAttr() need to have the
correct target-{cpu,features} attributes to avoid miscompilations
such as using the wrong relocation type to access globals (missing
tagged-globals feature), clobbering registers specified via -ffixed-*
(missing reserve-* feature), and so on.
There's already a number of attributes copied from the module flags
onto functions created by createWithDefaultAttr(). I don't think
module flags are the right choice for the target attributes because
we don't need the conflict resolution logic between modules with
different target attributes, nor does it seem sensible to add it:
there's no unambiguously "correct" set of target attributes when
merging two modules with different attributes, and nor should there
be; it's perfectly valid for two modules to be compiled with different
target attributes, that's the whole reason why they are per-function.
This also implies that it's unnecessary to serialize the attributes in
bitcode, which implies that they shouldn't be stored on the module. We
can also observe that for the most part, createWithDefaultAttr()
is called from compiler passes such as sanitizers, coverage and
profiling passes that are part of the compile time pipeline, not
the LTO pipeline. This hints at a solution: we need to store the
attributes in a non-serialized location associated with the ambient
compilation context. Therefore in this patch I elected to store the
attributes on the LLVMContext.
There are calls to createWithDefaultAttr() in the NVPTX and AMDGPU
backends, and those calls would happen at LTO time. For those callers,
the bug still potentially exists and it would be necessary to refactor
them to create the functions at compile time if this issue is relevant
on those platforms.
Fixes#93633.
Reviewers: fmayer, MaskRay, eugenis
Reviewed By: MaskRay
Pull Request: https://github.com/llvm/llvm-project/pull/96721
Reapplies commit c5aeca73 (and its followup commit 21396be8), which were
reverted due to missing functionality in MLIR and Flang regarding printing
debug records. This has now been added in commit 08aa511, along with support
for printing debug records in flang.
This reverts commit 2dc2290860355dd2bac3b655eea895fe30fde257.
"[Flang] Update test to not check for tail calls on debug intrinsics" &
"Reapply#3 "[RemoveDIs] Load into new debug info format by default in LLVM (#89799)"
Recent updates to flang have added debug info generation via MLIR, a path
which currently does not support debug records. The patch that enables
debug records by default (and a small followup patch) are thus being
reverted until the MLIR path has been fixed.
This reverts commits:
21396be865b4640abf6afa0b05de6708a1a996e0
c5aeca732d1ff6769b0659efebd1cfb5f60487e4
Reapplies commit 91446e2, which was reverted due to a downstream error,
discussed on the pull request. The error could not be reproduced
upstream, and cannot be reproduced downstream as-of current main, so
until the error can be confirmed to still exist this patch should
return.
This reverts commit 23f8fac745bdde70ed4f9c585d19c4913734f1b8.
When using the -mframe-chain=aapcs or -mframe-chain=aapcs-leaf options,
we cannot use r11 as an allocatable register, even if
-fomit-frame-pointer is also used. This is so that r11 will always point
to a valid frame record, even if we don't create one in every function.
E.g. during inlining new symbol name can be duplicated and then
`ValueSymbolTable::makeUniqueName()` will add unique suffix, exceeding
the `non-global-value-max-name-size` restriction.
Also fixed `unsigned` type of the option to `int` since `ValueSymbolTable`'
constructor can use `-1` value that means unrestricted name size.
This reverts commit 91446e2aa687ec57ad88dc0df793d0c6e694a7c9 and
a unittest followup 1530f319311908b06fe935c89fca692d3e53184f (#90476).
In a stage-2 -flto=thin -gsplit-dwarf -g -fdebug-info-for-profiling
-fprofile-sample-use= build of clang, a ThinLTO backend compile has
assertion failures:
Global is external, but doesn't have external or weak linkage!
ptr @_ZN5clang12ast_matchers8internal18makeAllOfCompositeINS_8QualTypeEEENS1_15BindableMatcherIT_EEN4llvm8ArrayRefIPKNS1_7MatcherIS5_EEEE
function declaration may only have a unique !dbg attachment
ptr @_ZN5clang12ast_matchers8internal18makeAllOfCompositeINS_8QualTypeEEENS1_15BindableMatcherIT_EEN4llvm8ArrayRefIPKNS1_7MatcherIS5_EEEE
The failures somehow go away if -fprofile-sample-use= is removed.
Reapplies the original commit:
2f01fd99eb8c8ab3db9aba72c4f00e31e9e60a05
The previous application of this patch failed due to some missing
DbgVariableRecord support in clang, which has been added now by commit
8805465e.
This will probably break some downstream tools that don't already handle
debug records. If your downstream code breaks as a result of this
change, the simplest fix is to convert the module in question to the old
debug format before you process it, using
`Module::convertFromNewDbgValues()`. For more information about how to
handle debug records or about what has changed, see the migration
document:
https://llvm.org/docs/RemoveDIsDebugInfo.html
This reverts commit 4fd319ae273ed6c252f2067909c1abd9f6d97efa.
Fixes the broken tests in the original commit:
2f01fd99eb8c8ab3db9aba72c4f00e31e9e60a05
This will probably break some downstream tools that don't already handle
debug records. If your downstream code breaks as a result of this
change, the simplest fix is to convert the module in question to the old
debug format before you process it, using
`Module::convertFromNewDbgValues()`. For more information about how to
handle debug records or about what has changed, see the migration
document:
https://llvm.org/docs/RemoveDIsDebugInfo.html
This reverts commit 00821fed09969305b0003d3313c44d1e761a7131.
This patch enables parsing and creating modules directly into the new
debug info format. Prior to this patch, all modules were constructed
with the old debug info format by default, and would be converted into
the new format just before running LLVM passes. This is an important
milestone, in that this means that every tool will now be exposed to
debug records, rather than those that run LLVM passes. As far as I've
tested, all LLVM tools/projects now either handle debug records, or
convert them to the old intrinsic format.
There are a few unit tests that need updating for this patch; these are
either cases of tests that previously needed to set the debug info
format to function, or tests that depend on the old debug info format in
some way. There should be no visible change in the output of any LLVM
tool as a result of this patch, although the likelihood of this patch
breaking downstream code means an NFC tag might be a little misleading,
if not technically incorrect:
This will probably break some downstream tools that don't already handle
debug records. If your downstream code breaks as a result of this
change, the simplest fix is to convert the module in question to the old
debug format before you process it, using
`Module::convertFromNewDbgValues()`. For more information about how to
handle debug records or about what has changed, see the migration
document:
https://llvm.org/docs/RemoveDIsDebugInfo.html
I'm planning to remove StringRef::equals in favor of
StringRef::operator==.
- StringRef::operator== outnumbers StringRef::equals by a factor of 22
under llvm/ in terms of their usage.
- The elimination of StringRef::equals brings StringRef closer to
std::string_view, which has operator== but not equals.
- S == "foo" is more readable than S.equals("foo"), especially for
!Long.Expression.equals("str") vs Long.Expression != "str".
This adds proper support for calling intrinsics without mangling suffix
when parsing textual IR. This already worked (mostly by accident) when
only a single mangling suffix was in use.
This patch extends support to the case where the intrinsic is used with
multiple signatures, and as such multiple different intrinsic
declarations have to be inserted. The final IR will have intrinsics with
mangling suffix as usual.
Motivated by the discussion at:
https://discourse.llvm.org/t/recent-improvements-to-the-ir-parser/77366
Correct missing cases in a switch that result in @llvm.vp.fma.v4f32
getting lowered to a constrained fma intrinsic. Vector predicated
lowering to contrained intrinsics is not supported currently, and
there's no consensus on the path forward. We certainly shouldn't be
introducing constrained intrinsics into a function that isn't strictfp.
Problem found with D146845.
This patch adds a new flag: `--preserve-input-debuginfo-format`
This flag instructs the tool to not convert the debug info format
(intrinsics/records) of input IR, but to instead determine the format of
the input IR and overwrite the other format-determining flags so that we
process and output the file in the same format that we received it in.
This flag is turned off by llvm-link, llvm-lto, and llvm-lto2, and
should be turned off by any other tool that expects to parse multiple IR
modules and have their debug info formats match.
The motivation for this flag is to allow tools to not convert the debug
info format - verify-uselistorder and llvm-reduce, and any downstream
tools that seek to test or mutate IR as-is, without applying extraneous
modifications to the input. This is a necessary step to using debug
records by default in all (other) LLVM tools.
