Commit history in chronological order:
[BOLT] llvm-bolt-wrapper: added wrapper for bolt binary matching
Summary:
Wrapper to compare two versions of BOLT to see if they produce the same output
binary given the same input.
(cherry picked from FBD26626137)
[BOLT] llvm-bolt-wrapper: support for no-output tests and heatmap mode
Summary:
- Added an option `skip_binary_cmp` to support invocations that don't output
a binary
- Minor fixes for heatmap mode, timeout, log comparison
- Rearranged in-line config example to be copy-pasteable
(cherry picked from FBD26822016)
[BOLT] llvm-bolt-wrapper: merge stdout/stderr, search for config in script dir
(cherry picked from FBD27529335)
[BOLT] llvm-bolt-wrapper: handle /dev/null
Summary:
Fixed the wrapper to preserve `-o /dev/null` and skip binary matching for such
invocations.
(cherry picked from FBD28013747)
[BOLT] llvm-bolt-wrapper: handle cases where output binary doesn't exist
Summary:
Handle invocations where output binary is not generated (e.g. due to an expected
assertion or exit with BOLT-ERROR) and skip binary comparison in such cases.
(cherry picked from FBD28080158)
[BOLT] llvm-bolt-wrapper: handle boltdiff mode
Summary:
Handle `llvm-boltdiff` invocation similarly to `perf2bolt`
(cherry picked from FBD28080157)
[BOLT] llvm-bolt-wrapper: find section with mismatch
Summary:
For mismatching ELF files, find section with mismatch and print sections table
with highlighted mismatch section.
(cherry picked from FBD28087231)
[BOLT] llvm-bolt-wrapper: ignore-build-id in perf2bolt mode
Summary:
When perf2bolt fails to match build-id from perf output for cmp binary, we need
to use -ignore-build-id option to override the strict checking behavior.
(cherry picked from FBD28087232)
[BOLT] llvm-bolt-wrapper: suppress -bolt-info=0 in heatmap mode
Summary:
Heatmap mode is incompatible with `-bolt-info=0` used to suppress binary
differences. Remove it.
(cherry picked from FBD28087230)
[BOLT] llvm-bolt-wrapper: add config-generator mode
Summary:
llvm-bolt-wrapper config can be generated by the script itself.
It makes the workflow more reliable compared to preparing the config manually.
(cherry picked from FBD28358939)
[BOLT] llvm-bolt-wrapper: fix mismatch reporting
Summary:
1. Fixed header comparison issue where headers were skipped due to
`skip_end == 0` (`lst[:-n]` does not work if n==0).
2. Detect color support while printing mismatching section:
- use bold color if terminal supports ANSI escape codes,
- otherwise print ">" at mismatching section.
3. Remove extra 0x before mismatching offset.
(cherry picked from FBD28691979)
[BOLT] llvm-bolt-wrapper: handle perf2bolt tests with ignore-build-id
Summary:
`ignore-build-id` must be passed not more than once. Account for that.
(cherry picked from FBD29830266)
[BOLT] llvm-bolt-wrapper: fix running subprocesses in parallel
Summary:
The commands were running sequentially due to the use of blocking `communicate`
call, which is needed when stdout/stderr are directed to a pipe.
Fix this behavior by directing the output to a file.
(cherry picked from FBD29951863)
This reverts commit db49a78900. The reasoning in the patch applied to a downstream branch, and I got myself confused when trying to split apart pieces. Thankfully, the assert was simply weaker than the actual invariant currently upstream which is that ReadyInsts is not empty.
This patch adds the vector.scan op which computes the
scan for a given n-d vector. It requires specifying the operator,
the identity element and whether the scan is inclusive or
exclusive.
TEST: Added test in ops.mlir
Reviewed By: ThomasRaoux
Differential Revision: https://reviews.llvm.org/D117171
Part of the _BitInt feature in C2x
(http://www.open-std.org/jtc1/sc22/wg14/www/docs/n2763.pdf) is a new
macro in limits.h named BITINT_MAXWIDTH that can be used to determine
the maximum width of a bit-precise integer type. This macro must expand
to a value that is at least as large as ULLONG_WIDTH.
This adds an implementation-defined macro named __BITINT_MAXWIDTH__ to
specify that value, which is used by limits.h for the standard macro.
This also limits the maximum bit width to 128 bits because backends do
not currently support all mathematical operations (such as division) on
wider types yet. This maximum is expected to be increased in the future.
Functionality present in `flang/include/flang/Lower/ComplexExpr.h` are
available in `flang/include/flang/Optimizer/Builder/Complex.h`. This patch removes
the obsolete files.
Reviewed By: kiranchandramohan, schweitz
Differential Revision: https://reviews.llvm.org/D118462
Due to some complications with lifetime, and assume-like intrinsics, intrinsics were not included as outlinable instructions. This patch opens up most intrinsics, excluding lifetime and assume-like intrinsics, to be outlined. For similarity, it is required that the intrinsic IDs, and the intrinsics names match exactly, as well as the function type. This puts intrinsics in a different class than normal call instructions (https://reviews.llvm.org/D109448), where the name will no longer have to match.
This also adds an additional command line flag debug option to disable outlining intrinsics.
Recommit of: 8de76bd569
Adds extra checking of intrinsic function calls names to avoid taking the address of intrinsic calls when extracting function calls.
Reviewers: paquette, jroelofs
Differential Revision: https://reviews.llvm.org/D109450
Close#52781: for LTO, the inline asm diagnostic uses `<inline asm>` as the file
name (lib/CodeGen/AsmPrinter/AsmPrinterInlineAsm.cpp) and it is unclear which
module has the issue.
With this patch, we will see the module name (say `asm.o`) before `<inline asm>` with ThinLTO.
```
% clang -flto=thin -c asm.c && myld.lld asm.o -e f
ld.lld: error: asm.o <inline asm>:1:2: invalid instruction mnemonic 'invalid'
invalid
^~~~~~~
```
For regular LTO, unfortunately the original module name is lost and we only get
ld-temp.o.
Reviewed By: #lld-macho, ychen, Jez Ng
Differential Revision: https://reviews.llvm.org/D118434
This file was added in https://reviews.llvm.org/D74415. There was no
justification as to why it was added, and after about a year of being
in-tree, it's still unused, so this removes it.
The fact we could have a block with a valid scheduling window, but nothing to schedule was surprising to me. After digging through the code, this can only happen if we don't find anything to directly vectorize. However, the reduction handling code relies on this mode, so we can't simply consider such trees unvectorizeable. The assert conveys both that this situation can happen, but also that it can *only* happen for an immediate gather.
Context: We built the bundle before deciding that vectorization of a bundle is possible. A side effect of bundle construction is manipulating the scheduling window, so a bundle which isn't vectorizable can cause the creation or expansion of a scheduling window.
This matches GCC: https://godbolt.org/z/sM5q95PGY
I realize this is an API break for clang+clang - so I'm totally open to
discussing how we should deal with that. If Apple wants to keep the
Clang layout indefinitely, if we want to put a flag on this so non-Apple
folks can opt out of this fix/new behavior.
Differential Revision: https://reviews.llvm.org/D117616
The unit tests for PyTACO hasn't been upstreamed yet. A unit test for this
change will be added when we upstream all the unit tests for PyTACO.
Reviewed By: aartbik
Differential Revision: https://reviews.llvm.org/D118417
AIX provides additional definitions in the system libc float.h that we
would like to be available to users, so we need to include_next through,
similar to what is done on some other platforms.
We also adjust the guards for some definitions which are restricted
based on language level to also be provide with the _ALL_SOURCE feature
test macro on AIX, similar to what is done by the platform float.h
header, so we don't run into cases where we don't provide the compiler
macro but still have a different definition from the system.
Differential Revision: https://reviews.llvm.org/D117935
The existing instructions for lldb on Windows can be more explicit. This adds a few details on how to install various components and the easiest way to get to a working build.
Reviewed By: JDevlieghere
Differential Revision: https://reviews.llvm.org/D118425
Add support for Swift reflection metadata to dsymutil.
This patch adds support for copying Swift reflection metadata (__swift5_.* sections) from .o files to into the symbol-rich binary in the output .dSYM. The functionality is automatically enabled only if a .o file has reflection metadata sections and the binary doesn't. When copying dsymutil moves the section from the __TEXT segment to the __DWARF segment.
rdar://76973336
Differential Revision: https://reviews.llvm.org/D115007
mlir-cpu-runner has a dependency on ExecutionEngine which is only built for the native arch. So currently mlir-cpu-runner does not link correctly when the native arch is not targeted.
Reviewed By: mehdi_amini
Differential Revision: https://reviews.llvm.org/D118422
Same as D114481, the PCH reader looks for a `__clangast` section in the precompiled module file, which isn't present on AIX, and we don't support writing this custom section in XCOFF yet.
Reviewed By: Jake-Egan, daltenty, sfertile
Differential Revision: https://reviews.llvm.org/D118477
We were creating a truncate with the default for the type, but for
VP intrinsics we have a VL that we should use.
Reviewed By: frasercrmck
Differential Revision: https://reviews.llvm.org/D118406
`createIRLevelProfileFlagVar()` seems to be only used in
`PGOInstrumentation.cpp` so we move it to that file. Then it can also
take advantage of directly using options rather than passing them as
arguments.
Reviewed By: kyulee, phosek
Differential Revision: https://reviews.llvm.org/D118097
Fix the pseudos to have the correct size in the MCInstrDesc description.
Inspired by D118009 and D117970.
Reviewed By: asb
Differential Revision: https://reviews.llvm.org/D118175
No need to reorder the top nodes, if they are not stores or
insertelement instructions and each node should be analized only
once, when the bottom-to-top analysis is performed.
We still endup with extractelements for the top node scalars and
the final shuffle just adds an extra cost and currently
crashes the compiler for PHI nodes.
Differential Revision: https://reviews.llvm.org/D116760
In glibc before 2.33, include/sys/stat.h defines fstat/fstat64 to
`__fxstat/__fxstat64` and provides `__fxstat/__fxstat64` in libc_nonshared.a.
The symbols are glibc specific and not needed on other systems.
Reviewed By: vitalybuka, #sanitizers
Differential Revision: https://reviews.llvm.org/D118423
If we have a vector FP division with a splatted divisor, use
getVectorMinNumElements when scaling the num of uses by splat factor.
For AArch64 the combine kicks in for the <vscale x 4 x float> case since it's
above the fdiv threshold (3) when scaling num uses by splat factor, but the
codegen is worse (splat + vector fdiv + vector fmul) than the <vscale x 2 x
double> case (splat + vector fdiv).
If the combine could be converted into a scalar FP division by
scalarizeBinOpOfSplats it may be cheaper, but it looks like this is predicated
on the isExtractVecEltCheap TLI function which is implemented for x86 but not
AArch64. Perhaps for now combineRepeatedFPDivisors should only scale num uses
by splat if the division can be converted into scalar op.
Reviewed By: sdesmalen
Differential Revision: https://reviews.llvm.org/D118343
Add Thread::GetSiginfo() and SBThread::GetSiginfo() methods to retrieve
the siginfo value from server.
Differential Revision: https://reviews.llvm.org/D118055
When `-fno-integrated-as` is in effect (the default on AIX) the cc1 job produces a `.s` file instead. This patch adapts the test to accept `.s` or `.o` files.
Reviewed By: jansvoboda11
Differential Revision: https://reviews.llvm.org/D118152
This patch introduces a class LexSimplex that can currently be used to find the
lexicographically minimal rational point in an IntegerPolyhedron. This is a
series of patches leading to computing the lexicographically minimal integer
lattice point as well parametric lexicographic minimization.
Reviewed By: Groverkss
Differential Revision: https://reviews.llvm.org/D117437
We can pass this test now thanks to recent changes that allow us to use PGO without LTO, so un-XFAIL the test and update the comments.
Differential Revision: https://reviews.llvm.org/D118474
`__gnu_h2f_ieee` and `__gnu_f2h_ieee` are introduce by ARM and set that as
default name for fp16 and fp32 conversion in LLVM.
However RISC-V GCC using default naming scheme for that, which is
`__extendhfsf2` and `__truncsfhf2` for that, that cause runtime ABI
incompatible issue.
Although we didn't have formal runtime ABI spec to specify those naming
convention yet, but I think it would be great to fix the incompatible
issue first.
And I've plan to create a runtime ABI spec undere psABI spec this year.
Reviewed By: asb
Differential Revision: https://reviews.llvm.org/D118207
extract_vec_elt (load X), C --> scalar load (X+C)
As noted in the comment, DAGCombiner has this fold -- and the code in this
patch is adapted from DAGCombiner::scalarizeExtractedVectorLoad() -- but
x86 should benefit even if the loaded vector has other uses as long as we
apply some other x86-specific conditions. The motivating example from #50310
is shown in vec_int_to_fp.ll.
Fixes#50310
Differential Revision: https://reviews.llvm.org/D118376
Where the instruction mnemonic contains a dot, we name the corresponding
instruction in the .td file using a _ in the place of the dot. e.g. LR_W
rather than LRW. This commit updates RISCVInstrInfoZb.td to follow that
convention.
Amir Ayupov