We have amdgpu_gfx functions that have high register pressure. If
we do not reserve VGPR for SGPR spill, we will fall into the path
to spill the SGPR to memory, which does not only have correctness issue,
but also have really bad performance.
I don't know why there is the check for hasStackObjects(), in our case,
we don't have stack objects at the time of finalizeLowering(). So just
remove the check that we always reserve a VGPR for possible SGPR spill
in non-entry functions.
Reviewed by: arsenm
Differential Revision: https://reviews.llvm.org/D98345
I met some code generation behavior change when I tried to remove
the hasStackObject() check when reserving VGPR for SGPR spill.
For example, the function `callee_no_stack_no_fp_elim_all` in the lit
test file `callee-frame-setup.ll`.
The generated code changed from:
```
s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
s_mov_b32 s4, s33
s_mov_b32 s33, s32
s_mov_b32 s33, s4
s_setpc_b64 s[30:31]
```
into something like:
```
s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
v_writelane_b32 v63, s33, 0
s_mov_b32 s33, s32
v_readlane_b32 s33, v63, 0
s_setpc_b64 s[30:31]
```
I think we still prefer the old version where only scalar instructions are needed.
The idea here is free the reserved VGPR if no SGPR spills. So we will very likely
to use a free SGPR for fp/sp spill.
Reviewed by: arsenm
Differential Revision: https://reviews.llvm.org/D98344
Attempting to build a standalone libFuzzer in Fuchsia's default toolchain for the purpose of cross-compiling the unit tests revealed a number of not-quite-proper type conversions. Fuchsia's toolchain include `-std=c++17` and `-Werror`, among others, leading to many errors like `-Wshorten-64-to-32`, `-Wimplicit-float-conversion`, etc.
Most of these have been addressed by simply making the conversion explicit with a `static_cast`. These typically fell into one of two categories: 1) conversions between types where high precision isn't critical, e.g. the "energy" calculations for `InputInfo`, and 2) conversions where the values will never reach the bits being truncated, e.g. `DftTimeInSeconds` is not going to exceed 136 years.
The major exception to this is the number of features: there are several places that treat features as `size_t`, and others as `uint32_t`. This change makes the decision to cap the features at 32 bits. The maximum value of a feature as produced by `TracePC::CollectFeatures` is roughly:
(NumPCsInPCTables + ValueBitMap::kMapSizeInBits + ExtraCountersBegin() - ExtraCountersEnd() + log2(SIZE_MAX)) * 8
It's conceivable for extremely large targets and/or extra counters that this limit could be reached. This shouldn't break fuzzing, but it will cause certain features to collide and lower the fuzzers overall precision. To address this, this change adds a warning to TracePC::PrintModuleInfo about excessive feature size if it is detected, and recommends refactoring the fuzzer into several smaller ones.
Reviewed By: morehouse
Differential Revision: https://reviews.llvm.org/D97992
This patch adds support for vectorizing loops with 'iter_args' when those loops
are not a vector dimension. This allows vectorizing outer loops with an inner
'iter_args' loop (e.g., reductions). Vectorizing scenarios where 'iter_args'
loops are vector dimensions would require more work (e.g., analysis,
generating horizontal reduction, etc.) not included in this patch.
Reviewed By: nicolasvasilache
Differential Revision: https://reviews.llvm.org/D97892
The S_[GL]PROC32_ID symbol records are supposed to point to function ID
records. If they don't, they are corrupt. The warning message here was
very technical, but a user has encountered it in the wild. Add some more
information and some more testing.
This patch extends the matrix spec to allow matrix-by-scalar division.
Originally support for `/` was left out to avoid ambiguity for the
matrix-matrix version of `/`, which could either be elementwise or
specified as matrix multiplication M1 * (1/M2).
For the matrix-scalar version, no ambiguity exists; `*` is also
an elementwise operation in that case. Matrix-by-scalar division
is commonly supported by systems including Matlab, Mathematica
or NumPy.
Reviewed By: rjmccall
Differential Revision: https://reviews.llvm.org/D97857
This patch replaces the root-terminal vectorization approach implemented in the
Affine vectorizer with a topological order approach that vectorizes all the
operations within the target loop nest. These are the most important changes
introduced by the new algorithm:
* Removed tracking of root and terminal ops. Existing vectorization
functionality is preserved and extended so that loop nests without
root-terminal chains can be vectorized.
* Vectorizing a loop nest now only requires a single topological traversal.
* A new vector loop nest is incrementally built along the vectorization
process. The original scalar loop is kept intact. No cloning guard is needed
to recover the scalar loop if vectorization fails. This approach also
simplifies the challenging task of replacing a loop operation amid the
vectorization process without invalidating the analysis information that
depends on the original loop.
* Vectorization of specific operations has been implemented as independent,
preparing them to be moved to a potential vectorization interface.
Reviewed By: nicolasvasilache
Differential Revision: https://reviews.llvm.org/D97442
These patterns are obviously dead, they are using format
operand which is not selected and we have no corresponding
SelectMUBUF() function.
Differential Revision: https://reviews.llvm.org/D98451
Don't normalize arm architecture names; doing that loses the ability
to pick the right implementation of builtins for each architecture
variant. When building compiler-rt builtins as part of a
runtimes build, builtins for multiple armv* variants could be built
in the same directory, and with the simplified architecture name,
they'd all be built in the same directory, overlapping each other.
Splitting this out as the change is non-trivial: The way this code
handled pointer types doesn't really make sense, as GEPs can only
apply an offset to the outermost pointer, but can't drill down
into interior pointer types (which would require dereferencing
memory).
Instead give special treatment to the first (pointer) index.
I've hardcoded it to zero as that's the only way the function is
used right now, but handling non-zero indexes would be
straightforward.
The original goal here was to have an element type for CreateGEP.
When using -fprofile-list to selectively apply instrumentation only
to certain files or functions, we may end up with a binary that doesn't
have any counters in the case where no files were selected. However,
because on Linux and Fuchsia, we pass -u__llvm_profile_runtime, the
runtime would still be pulled in and incur some non-trivial overhead,
especially in the case when the continuous or runtime counter relocation
mode is being used. A better way would be to pull in the profile runtime
only when needed by declaring the __llvm_profile_runtime symbol in the
translation unit only when needed.
This approach was already used prior to 9a041a7522, but we changed it
to always generate the __llvm_profile_runtime due to a TAPI limitation.
Since TAPI is only used on Mach-O platforms, we could use the early
emission of __llvm_profile_runtime there, and on other platforms we
could change back to the earlier approach where the symbol is generated
later only when needed. We can stop passing -u__llvm_profile_runtime to
the linker on Linux and Fuchsia since the generated undefined symbol in
each translation unit that needed it serves the same purpose.
Differential Revision: https://reviews.llvm.org/D98061
Associative reduction matcher in SLP begins with select instruction but when
it reached call to llvm.umax (or alike) via def-use chain the latter also matched
as UMax kind. The routine's later code assumes matched instruction to be a select
and thus it merely died on the first encountered cast that did not fit.
Differential Revision: https://reviews.llvm.org/D98432
This allows for storage instances to store data that isn't uniqued in the context, or contain otherwise non-trivial logic, in the rare situations that they occur. Storage instances with trivial destructors will still have their destructor skipped. A consequence of this is that the storage instance definition must be visible from the place that registers the type.
Differential Revision: https://reviews.llvm.org/D98311
".llvm." suffix.
Currently IndirectCallPromotion simply strip everything after the first "."
in LTO mode, in order to match the symbol name and the name with ".llvm."
suffix in the value profile. However, if -funique-internal-linkage-names
and thinlto are both enabled, the name may have both ".__uniq." suffix and
".llvm." suffix, and the current mechanism will strip them both, which is
unexpected. The patch fixes the problem.
Differential Revision: https://reviews.llvm.org/D98389
Fix handling of skip_permission_denied on windows; after converting
the return value of GetLastError() to a standard error_code, ec.value()
is in the standard errc range, not a native windows error code. This
was missed in 156180727d.
The directory "C:\System Volume Information" does seem to exist and
have these properties on most relevant contempory setups.
Differential Revision: https://reviews.llvm.org/D98166
Use a more general strategy when splitting a vector into scalar parts (and vice-versa) to correctly handle vector types whose element size is not a power of 2 (and a multiple of 8).
Reviewed By: atanasyan
Differential Revision: https://reviews.llvm.org/D98273
The code deciding how to split the vector in register-sized integers used the integer division operator, thus rounding down the result.
Correct the computation for irregularly-sized types (non-power-of-two, non multiple of 8) by rounding the division result upwards.
Reviewed By: atanasyan
Differential Revision: https://reviews.llvm.org/D98189
In D94489 we changed the way we build the docs and now have some additional
dependencies to generate the Python API docs. As the same sphinx project is
generating the man pages for LLDB it should have in theory the same setup code
that sets up the mocked LLDB module.
However, as we don't have that setup code the man page generation just fails as
there is no mocked LLDB module and the Python API generation errors out.
The man page anyway doesn't cover the Python API so I don't think there is any
point of going through the whole process (and requiring the sphinx plugins) just
to generate the (eventually unused) Python docs.
This patch just skips the relevant Python API generation when we are building
the man page.
Reviewed By: JDevlieghere
Differential Revision: https://reviews.llvm.org/D98441
This patch fixes a heap-use-after-free introduced by the recent changes
in the vectorizer: https://reviews.llvm.org/rG95db7b4aeaad590f37720898e339a6d54313422f
The problem is due to the way candidate loops are visited. All candidate loops
are pattern-matched beforehand using the 'NestedMatch' utility. These matches may
intersect with each other so it may happen that we try to vectorize a loop that
was previously vectorized. The new vectorization algorithm replaces the original
loops that are vectorized with new loops and, therefore, any reference to the
original loops in the pre-computed matches becomes invalid.
This patch fixes the problem by classifying the candidate matches into buckets
before vectorization. Each bucket contains all the matches that intersect. The
vectorizer uses these buckets to make sure that we only vectorize *one* match from
each bucket, at most.
Differential Revision: https://reviews.llvm.org/D98382
There are 3 remaining tests that still have `REQUIRE: shell`:
* color-diagnostics.test -- seems necessary for ANSI escape sequence support
* stabs.s -- the shell part could be removed, but I don't think we can support
the test on Windows anyway due to its reliance on `touch` to set the modtime
* framework.s -- uses symlinks, I'm not sure this works on Windows
Addresses PR49512.
Reviewed By: #lld-macho, alexshap
Differential Revision: https://reviews.llvm.org/D98395
SUBTRACTOR relocations are always paired with UNSIGNED
relocations to indicate a pair of symbols whose address difference we
want. Functionally they are like a single relocation: only one pointer
gets written / relocated. Previously, we would handle these pairs by
skipping over the SUBTRACTOR relocation and writing the pointer when
handling the UNSIGNED reloc. This diff reverses things, so we write
while handling SUBTRACTORs and skip over the UNSIGNED relocs instead.
Being able to distinguish between SUBTRACTOR and UNSIGNED relocs in the
write phase (i.e. inside `relocateOne`) is useful for the upcoming range
check diff: we want to check that SUBTRACTOR relocs write signed values,
but UNSIGNED relocs (naturally) write unsigned values.
Reviewed By: #lld-macho, thakis
Differential Revision: https://reviews.llvm.org/D98386
The previous implementation miscalculated the addend, resulting
in an underflow. This meant that every SIGNED_N section relocation would
be associated with the last subsection (since the addend would now be a
huge number). We were "lucky" that this mistake was typically cancelled
out -- 64-to-32-bit-truncation meant that the final value was correct,
as long as subsections were not rearranged.
Reviewed By: #lld-macho, thakis
Differential Revision: https://reviews.llvm.org/D98385
Previously, SyntheticSections.cpp did not have a top-level `using namespace
llvm::MachO` because it caused a naming conflict: `llvm::MachO::Symbol` would
collide with `lld::macho::Symbol`.
`MachO::Symbol` represents the symbols defined in InterfaceFiles (TBDs). By
moving the inclusion of InterfaceFile.h into our .cpp files, we can avoid this
name collision in other files where we are only dealing with LLD's own symbols.
Along the way, I removed all unnecessary "MachO::" prefixes in our code.
Cons of this approach: If TextAPI/MachO/Symbol.h gets included via some other
header file in the future, we could run into this collision again.
Alternative 1: Have either TextAPI/MachO or BinaryFormat/MachO.h use a different
namespace. Most of the benefit of `using namespace llvm::MachO` comes from being
able to use things in BinaryFormat/MachO.h conveniently; if TextAPI was under a
different (and fully-qualified) namespace like `llvm::tapi` that would solve our
problems. Cons: lots of files across llvm-project will need to be updated, and
folks who own the TextAPI code need to agree to the name change.
Alternative 2: Rename our Symbol to something like `LldSymbol`. I think this is
ugly.
Personally I think alternative #1 is ideal, but I'm not sure the effort to do it is
worthwhile, this diff's halfway solution seems good enough to me. Thoughts?
Reviewed By: #lld-macho, oontvoo, MaskRay
Differential Revision: https://reviews.llvm.org/D98149
If you specify a type-bound procedure with an alternate return, there
will be no symbol associated with that dummy argument. In such cases,
the compiler's list of dummy arguments will contain a nullptr. In our
analysis of the PASS arguments of type-bound procedures, we were
assuming that all dummy arguments had non-null symbols associated with
them and were using that assumption to get the name of the dummy
argument. This caused the compiler to try to dereference a nullptr.
I fixed this by explicitly checking for a nullptr and, in such cases, emitting
an error message. I also added tests that contain type-bound procedures with
alternate returns in both legal and illegal constructs to ensure that semantic
analysis is working for them.
Differential Revision: https://reviews.llvm.org/D98430
For CGSCC inline, we need to scale down a function's branch weights and entry counts when thee it's inlined at a callsite. This is done through updateCallProfile. Additionally, we also scale the weigths for the inlined clone based on call site count in updateCallerBFI. Neither is needed for inlining during sample profile loader as it's using context profile that is separated from inlinee's own profile. This change skip the inlinee profile scaling for sample loader inlining.
Differential Revision: https://reviews.llvm.org/D98187
If --gcc-toolchain is specified, we should detect GCC installation there, and suppress other directories for detection.
Reviewed By: mgorny, manojgupta
Differential Revision: https://reviews.llvm.org/D97894
This adjusts the place that the t2DoLoopStart reg allocation hint is
inserted, adding it in the ARMTPAndVPTOptimizaionPass in a similar place
as other tail predicated loop optimizations. This removes the need for
doing so in a custom inserter, and should make the hint more accurate,
only adding it where we expect to create a DLS (not DLSTP or WLS).
Recently we improved the lowering of low overhead loops and tail
predicated loops, but concentrated first on the DLS do style loops. This
extends those improvements over to the WLS while loops, improving the
chance of lowering them successfully. To do this the lowering has to
change a little as the instructions are terminators that produce a value
- something that needs to be treated carefully.
Lowering starts at the Hardware Loop pass, inserting a new
llvm.test.start.loop.iterations that produces both an i1 to control the
loop entry and an i32 similar to the llvm.start.loop.iterations
intrinsic added for do loops. This feeds into the loop phi, properly
gluing the values together:
%wls = call { i32, i1 } @llvm.test.start.loop.iterations.i32(i32 %div)
%wls0 = extractvalue { i32, i1 } %wls, 0
%wls1 = extractvalue { i32, i1 } %wls, 1
br i1 %wls1, label %loop.ph, label %loop.exit
...
loop:
%lsr.iv = phi i32 [ %wls0, %loop.ph ], [ %iv.next, %loop ]
..
%iv.next = call i32 @llvm.loop.decrement.reg.i32(i32 %lsr.iv, i32 1)
%cmp = icmp ne i32 %iv.next, 0
br i1 %cmp, label %loop, label %loop.exit
The llvm.test.start.loop.iterations need to be lowered through ISel
lowering as a pair of WLS and WLSSETUP nodes, which each get converted
to t2WhileLoopSetup and t2WhileLoopStart Pseudos. This helps prevent
t2WhileLoopStart from being a terminator that produces a value,
something difficult to control at that stage in the pipeline. Instead
the t2WhileLoopSetup produces the value of LR (essentially acting as a
lr = subs rn, 0), t2WhileLoopStart consumes that lr value (the Bcc).
These are then converted into a single t2WhileLoopStartLR at the same
point as t2DoLoopStartTP and t2LoopEndDec. Otherwise we revert the loop
to prevent them from progressing further in the pipeline. The
t2WhileLoopStartLR is a single instruction that takes a GPR and produces
LR, similar to the WLS instruction.
%1:gprlr = t2WhileLoopStartLR %0:rgpr, %bb.3
t2B %bb.1
...
bb.2.loop:
%2:gprlr = PHI %1:gprlr, %bb.1, %3:gprlr, %bb.2
...
%3:gprlr = t2LoopEndDec %2:gprlr, %bb.2
t2B %bb.3
The t2WhileLoopStartLR can then be treated similar to the other low
overhead loop pseudos, eventually being lowered to a WLS providing the
branches are within range.
Differential Revision: https://reviews.llvm.org/D97729
1. PGOMemOPSizeOpt grabs only the first, up to five (by default) entries from
the value profile metadata and preserves the remaining entries for the fallback
memop call site. If there are more than five entries, the rest of the entries
would get dropped. This is fine for PGOMemOPSizeOpt itself as it only promotes
up to 3 (by default) values, but potentially not for other downstream passes
that may use the value profile metadata.
2. PGOMemOPSizeOpt originally assumed that only values 0 through 8 are kept
track of. When the range buckets were introduced, it was changed to skip the
range buckets, but since it does not grab all entries (only five), if some range
buckets exist in the first five entries, it could potentially cause fewer
promotion opportunities (eg. if 4 out of 5 were range buckets, it may be able to
promote up to one non-range bucket, as opposed to 3.) Also, combined with 1, it
means that wrong entries may be preserved, as it didn't correctly keep track of
which were entries were skipped.
To fix this, PGOMemOPSizeOpt now grabs all the entries (up to the maximum number
of value profile buckets), keeps track of which entries were skipped, and
preserves all the remaining entries.
Differential Revision: https://reviews.llvm.org/D97592
For the use in LLVMOps.td I used the getPointerElementType()
escape hatch, as it's not obvious to me how the load type
should be properly obtained here.
Check that appends with a path object doesn't do allocations, even
on windows.
Suggested by Marek in D98398. The patch might apply without D98398
(depending on how much of the diff context has to match), but doesn't
make much sense until after that patch has landed.
Differential Revision: https://reviews.llvm.org/D98412
This makes sure that no extra allocations happen on windows, fixing
earlier errors in the DisableAllocationGuard (in the second case that
is modified).
This is split out from D98398.
Differential Revision: https://reviews.llvm.org/D98406
Ruiling Song