This patch implements proposal https://lists.llvm.org/pipermail/llvm-dev/2020-August/144579.html
llvm-dwarfutil - is a tool that is used for processing debug info(DWARF) located in built binary files to improve debug info quality, reduce debug info size. The patch currently implements smaller set of command-line options(comparing to the proposal):
```
./llvm-dwarfutil [options] <input file> <output file>
--garbage-collection Do garbage collection for debug info(default)
-j <value> Alias for --num-threads
--no-garbage-collection Don`t do garbage collection for debug info
--no-odr-deduplication Don`t do ODR deduplication for debug types
--no-odr Alias for --no-odr-deduplication
--no-separate-debug-file
Create single output file, containing debug tables(default)
--num-threads <threads> Number of available threads for multi-threaded execution. Defaults to the number of cores on the current machine
--odr-deduplication Do ODR deduplication for debug types(default)
--odr Alias for --odr-deduplication
--separate-debug-file Create two output files: file w/o debug tables and file with debug tables
--tombstone [bfd,maxpc,exec,universal]
Tombstone value used as a marker of invalid address(default: universal)
=bfd - Zero for all addresses and [1,1] for DWARF v4 (or less) address ranges and exec
=maxpc - Minus 1 for all addresses and minus 2 for DWARF v4 (or less) address ranges
=exec - Match with address ranges of executable sections
=universal - Both: bfd and maxpc
```
Reviewed By: clayborg
Differential Revision: https://reviews.llvm.org/D86539
The request is mentioned on D129053. I feel that having this functionality is
mildly useful (not strong).
* Rename .ctors to .init_array and change sh_type to SHT_INIT_ARRAY (GNU objcopy
detects the special name but we don't).
* Craft tests for a new SHT_LLVM_* extension
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D129337
GNU objdump disassembles all unknown instructions by default. Match this user
friendly behavior with the cpu value `future`.
Differential Revision: https://reviews.llvm.org/D127824
GNU objdump disassembles all unknown instructions by default. Match this user
friendly behavior with the target feature "all" (D128029) designed for disassemblers.
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D128030
This option allows printing all sources used by an object file.
Reviewed By: dblaikie, jhenderson
Differential Revision: https://reviews.llvm.org/D87656
From binutils 2.34 onwards, ar supports --output to specify a directory
where archive members should be extracted to. Port this feature.
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D128626
This adds a --filter option to llvm-symbolizer. This takes log-bearing
symbolizer markup from stdin and writes a human-readable version to
stdout.
For now, this only implements the "symbol" markup tag; all others are
passed through unaltered. This is a proof-of-concept bit of
functionalty; implement the various tags is more-or-less just a matter
of hooking up various parts of the Symbolize library to the architecture
established here.
Reviewed By: peter.smith
Differential Revision: https://reviews.llvm.org/D126980
Automatically generate a reproducer when dsymutil crashes. We already
support generating reproducers with the --gen-reproducer flag, which
emits a reproducer on exit. This patch adds support for doing the same
on a crash and makes it the default behavior.
rdar://68357665
Differential revision: https://reviews.llvm.org/D127441
Latest GNU nm (milestone: 2.39) has added -W/--no-weak and changed -U to mean
--defined-only (instead of --unicode=). The changes match our semantics.
Close#55297
Reviewed by: jhenderson, keith
Differential Revision: https://reviews.llvm.org/D126133
These no longer exist. A few have been added since but I'm not enough of
an expert to provide a useful blurb on them outside of what you see with
`--help`.
Differential Revision: https://reviews.llvm.org/D122361
this review is extracted from D86539.
1. Rename AccelTableKind to DwarfLinkerAccelTableKind
(to differentiate from AccelTableKind from CodeGen/AsmPrinter/DwarfDebug.h)
2. Add None value to the DwarfLinkerAccelTableKind.
3. added 'None' value for 'accelerator' option of dsymutil.
Differential Revision: https://reviews.llvm.org/D125474
`--symbolize-operands` already symbolizes branch targets based on the disassembly. When the object file is created with `-fbasic-block-sections=labels` (ELF-only) it will include a SHT_LLVM_BB_ADDR_MAP section which maps basic blocks to their addresses. In such case `llvm-objdump` can annotate the disassembly based on labels inferred on this section.
In contrast to the current labels, SHT_LLVM_BB_ADDR_MAP-based labels are created for every machine basic block including empty blocks and those which are not branched into (fallthrough blocks).
The old logic is still executed even when the SHT_LLVM_BB_ADDR_MAP section is present to handle functions which have not been received an entry in this section.
Reviewed By: jhenderson, MaskRay
Differential Revision: https://reviews.llvm.org/D124560
st_size may not be of importance to the abi if you are not using
copy relocations. This is helpful when you want to check the abi
of a shared object both when instrumented and not because asan
will increase the size of objects to include the redzone.
Differential revision: https://reviews.llvm.org/D124792
st_size may not be of importance to the abi if you are not using
copy relocations. This is helpful when you want to check the abi
of a shared object both when instrumented and not because asan
will increase the size of objects to include the redzone.
Differential revision: https://reviews.llvm.org/D124792
There are many more instances of this pattern, but I chose to limit this change to .rst files (docs), anything in libcxx/include, and string literals. These have the highest chance of being seen by end users.
Reviewed By: #libc, Mordante, martong, ldionne
Differential Revision: https://reviews.llvm.org/D124708
I noticed that when --update-section was added to llvm-objcopy it was
not added to the command guide, see
25bcd94234. This change adds it to the
docs and updates the help text.
Differential Revision: https://reviews.llvm.org/D122907
https://reviews.llvm.org/D116787
This reverts commit 33b3c86afa.
New change: fixed build failures:
- in stabs-sorted:restore the the ERR-KEY statements, which were accidentally deleted during refactoring
- in ObjDumper.h/MachODumper.cpp: refactor so that current dumpers which didn't provide an impl that accept a SymCom still works
Add documentation describing how to
- Use `llvm-remark-size-diff`
- Interpret the output from the tool
Differential Revision: https://reviews.llvm.org/D122744
The functionality (and llvm-objcopy's corresponding documentation) was
added in
5ad0103d8a.
It looks like the llvm-strip docs to match were missed.
Reviewed by: gbreynoo
Differential Revision: https://reviews.llvm.org/D121902
This adds 2 new lit helpers `%{fs-src-root}` and `%{fs-sep}`, these
allow writing tests that correctly handle slashes on Windows. In the
case of tests like clang/test/CodeGen/debug-prefix-map.c, these are
unable to correctly test behavior on both platforms, unless they fork
and add OS requirements, because the relevant logic hits host specific
codepaths like checking if paths are absolute.
Differential Revision: https://reviews.llvm.org/D111457
Fails on old python (like on ubuntu bionic) otherwise with:
UnicodeDecodeError: 'ascii' codec can't decode byte 0xe2 in position 2130: ordinal not in range(128)
Add a pairs of parameters to set coverage watermark for llvm-cov, and
user can change the percentage thresholds marked with different colors
in the report.
Patch By: tanjinhua
Differential Revision: https://reviews.llvm.org/D116876
This adds a BUILD_ID prefix to the llvm-symbolizer stdin and argument
syntax. The prefix causes the given binary name to be interpreted as a
build ID instead of an object file path. The semantics are analagous to
the behavior of --obj and --build-id.
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D119901
Darwin otool implements this flag as a one-stop solution for
displaying bind and rebase info. As I am working on upstreaming
chained fixup support this command will be useful to write testcases.
Differential Revision: https://reviews.llvm.org/D113573
Summary:
the patch implement of following functionality.
1. export the symbols from archive or object files.
2. sort the export symbols. (based on same symbol name and visibility)
3. delete the duplicate export symbols (based on same symbol name and visibility)
4. print out the unique and sorted export symbols (print the symbol name and visibility).
there are two new options are add in the patch
1. --export-symbols (enable the functionality of export unique symbol)
2. --no-rsrc (exclude the symbol name begin with "__rsrc" from be exporting from xcoff object file)
Export symbol list for xcoff object file has the same functionality as
The patch has the same functionality as
https://www.ibm.com/docs/en/xl-c-aix/13.1.0?topic=library-exporting-symbols-createexportlist-utility
Reviewers: James Henderson,Fangrui Song
Differential Revision: https://reviews.llvm.org/D112735
Summary:
Added a new option "-X" to specify, which type of object file should be examine.
For example:
1. "llvm-nm -X64 archive.a" only deal with the 64bit object files in the archive.a ,ignore the all 32bit object files in the archive.a
2. "llvm-nm -X32 xcoffobj32.o xcoffobj64.o " only deal with the 32bit object file "xcoffobj32.o" , 64bit object file "xcoffobj64.o" will be ignored
Reviewers: James Henderson,Fangrui Song
Differential Revision: https://reviews.llvm.org/D118193
`opt -analyze` is legacy PM-specific. Show better ways of doing the same
thing, generally with some sort of `-passes=print<foo>`.
Reviewed By: asbirlea
Differential Revision: https://reviews.llvm.org/D119486
This change adds a pair of flags controlling whether llvm-symbolizer
attempts debuginfod lookups. Lookups are attempted if --debuginfod is
passed and disabled if --no-debuginfod is passed.
The default behavior is made more nuanced: debuginfod lookups are now
only attempted if an HTTP client is compiled in and at least one backing
debuginfod URL was configured via environment variable. Previously,
debuginfod lookups would always be attempted, even if there were no
chance that they could succeed.
Reviewed By: phosek
Differential Revision: https://reviews.llvm.org/D118665
This adds a --build-id=<hex build ID> flag to llvm-symbolizer. If --obj
is unspecified, this will attempt to look up the provided build ID using
whatever mechanisms are available to the Symbolizer (typically,
debuginfod). The semantics are then as if the found binary were given
using the --obj flag.
Reviewed By: jhenderson, phosek
Differential Revision: https://reviews.llvm.org/D118633
libtool can currently produce 2 warnings:
1. No symbols were in the object file
2. An object file with the same basename was specified multiple times
The first warning here is often harmless and may just mean you have some
translation units with no symbols for the target you're building for.
The second warning can lead to real issues like those mentioned in
https://reviews.llvm.org/D113130 where ODR violations can slip in.
This introduces a new -warnings_as_errors flag that can be used by build
systems that want to verify they never hit these warnings. For example
with bazel the libtool caller first uniques names to make sure the
duplicate base name case is not possible, but if that doesn't work as
expected, having it fail would be preferred.
It's also worth noting that llvm-libtool-darwin works around an issue
that cctools libtool experiences related to debug info and duplicate
basenames, the workaround is described here:
30baa5d2a4/llvm/lib/Object/ArchiveWriter.cpp (L424-L465)
And it avoids this bug:
f0cbbb1c37/DuplicateBasenameIssue
Differential Revision: https://reviews.llvm.org/D118931
In GNU ar (since 2008), the modifier 'T' means creating a thin archive.
In many other ar implementations (FreeBSD, macOS, elfutils, etc), -T
means "allow filename truncation of extracted files", as specified by
X/Open System Interface.
For portability, 'T' with thin archive semantics should be avoided.
See https://sourceware.org/bugzilla/show_bug.cgi?id=28759 binutils 2.38
will deprecate 'T' (without diagnostic) and add --thin.
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D116979
As suggested on the bug, to help (but not completely....) stop folded instructions crossing the inline asm barriers used for llvm-mca analysis, we should recommend tagging with memory captures/attributes.
Differential Revision: https://reviews.llvm.org/D117788
memory-barrier instructions to providing targets and developers a convenient
way to explicitly declare which instructions are memory-barriers.
Differential Revision: https://reviews.llvm.org/D116779
Although we moved to Github Issues. The bug report message refers to
Bugzilla still. This patch tries to update these URLs.
Reviewed By: MaskRay, Quuxplusone, jhenderson, libunwind, libc++
Differential Revision: https://reviews.llvm.org/D116351
To match the `--help` message and most other utilities.
While here, change `option:: -output=output` to `option:: --output=<output>` and
omit the value name for the short options (convention of other utilities).
Reviewed By: snehasish
Differential Revision: https://reviews.llvm.org/D116353
Summary: When disassembling, symbolize a branch target operand
to print a label instead of a real address.
Reviewed By: shchenz
Differential Revision: https://reviews.llvm.org/D114492
Adds JSONScopedPrinter to llvm-readelf. It includes an empty
JSONELFDumper class which will be used to override any LLVMELFDumper
methods which utilize startLine() which JSONScopedPrinter cannot
provide.
This introduces a change where calls to llvm-readelf with non-ELF object
files that specify --elf-output-style=GNU will now print file summary
information where it previously didn't.
Fixes previous Windows test failure which occured due to JSON escaping
of '\' by not relying on LIT substitution.
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D114225
Adds JSONScopedPrinter to llvm-readelf. It includes an empty
JSONELFDumper class which will be used to override any LLVMELFDumper
methods which utilize startLine() which JSONScopedPrinter cannot
provide.
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D114225
Just a simple typo fix that allows me to test landing a commit now that
I have commit access.
Reviewed By: xgupta
Differential Revision: https://reviews.llvm.org/D115414
The fields output when using --output-style=JSON has changed but the
guide wasn't updated. This change fixes up the example.
Differential Revision: https://reviews.llvm.org/D115164
Renamed the option for llvm-cov and changed variable names to use more
inclusive terms. Also changed the binary for the test.
Reviewed By: alanphipps
Differential Revision: https://reviews.llvm.org/D112816
A new tool that compares TargetLibraryInfo's opinion of the availability
of library function calls against the functions actually exported by a
specified set of libraries. Can be helpful in verifying the correctness
of TLI for a given target, and avoid mishaps such as had to be addressed
in D107509 and 94b4598d.
The tool currently supports ELF object files only, although it's unlikely
to be hard to add support for other formats.
Re-commits 62dd488 with changes to use pre-generated objects, as not all
bots have ld.lld available.
Differential Revision: https://reviews.llvm.org/D111358
A new tool that compares TargetLibraryInfo's opinion of the availability
of library function calls against the functions actually exported by a
specified set of libraries. Can be helpful in verifying the correctness
of TLI for a given target, and avoid mishaps such as had to be addressed
in D107509 and 94b4598d.
The tool currently supports ELF object files only, although it's unlikely
to be hard to add support for other formats.
Differential Revision: https://reviews.llvm.org/D111358
This change is to keep the help text and command guide of llvm-readelf
in tandem.
- In the help text mention that --section-data, --section-relocations,
--section-symbols and --stack-sizes have no effect on GNU style
output; give the accepted values for --elf-output-style and update
the description of --gnu-hash-table to use the command guide
description.
- In the command guide add the missing options -a,
--dependant-libraries,--no-demangle, --wide and -W. Also update the
description of --symbols so it matches the help text.
Differential Revision: https://reviews.llvm.org/D111240
This change is to add some missing details, clarifies some options and
brings the help text and command guide of objdump closer together.
- Added to the help that --all-headers also outputs symbols and
relocations to match the command guide.
- Added to the help that --debug-vars accepts an optional
ascii/unicode format to match the command guide.
- Changed the help descriptions for --disassemble,
--disassemble-all, --dwarf=<value>, --fault-map-section,
--line-numbers, --no-leading-addr and --source descriptions to
match the command guide.
- Added to the help that --start-address and --stop-address also
effect relocation entries and the symbol table output to match
the command guide.
- Added a note to the command guide that --unwind-info and -u
are not available for the elf format.
Differential Revision: https://reviews.llvm.org/D110633
This change is to add some missing details to the help text and command
guide:
- Added a note to the command guide that --debug-macro also dumps
.debug_macinfo.
- Added a note to the command guide that --debug-frame and --eh_frame
are aliases, and in cases where both sections are present one command
outputs both.
- Changed the wording in the help output for --ignore-case and --regex to
closer match the command guide.
This change is to keep the help text and command guide of objcopy in
tandem.
- In the help output the options --rename-section and
--set-section-flags were missing the flag exclude, which is found in
the command guide.
- In the command guide the alias -G for --keep-global-symbol was
missing, which is found in the help output.
Differential Revision: https://reviews.llvm.org/D110340
Add a new --order option to choose between available test orders:
the default "smart" order, predictable "lexical" order or "random"
order. Default to using lexical order and one job in the lit test
suite.
Differential Revision: https://reviews.llvm.org/D107695
Moved View.h and View.cpp from /tools/llvm-mca/Views/ to /lib/MCA/ and
/include/llvm/MCA/. This is so that targets can define their own Views within
the /lib/Target/ directory (so that the View can use backend functionality).
To enable these Views within mca, targets will need to add them to the vector of
Views returned by their target's CustomBehaviour::getViews() methods.
Differential Revision: https://reviews.llvm.org/D108520
For example, I need this lately in my CI config:
LIT_XFAIL_NOT='libomptarget :: nvptx64-nvidia-cuda :: unified_shared_memory/api.c'
That test specifies an XFAIL directive, but I get an XPASS result.
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D106022
Users should generally observe no difference as long as they don't use
unintended option forms. Behavior changes:
* `-t=d` is removed. Use `-t d` instead.
* `--demangle=false` and `--demangle=0` cannot be used. Omit the option or use `--no-demangle`. Other flag-style options don't have `--no-` forms.
* `--help-list` is removed. This is a `cl::` specific option.
* llvm-readobj now supports grouped short options as well.
* `--color` is removed. This is generally not useful (only apply to errors/warnings) but was inherited from Support.
Some adjustment to the canonical forms
(usually from GNU readelf; currently llvm-readobj has too many redundant aliases):
* --dyn-syms is canonical. --dyn-symbols is a hidden alias
* --file-header is canonical. --file-headers is a hidden alias
* --histogram is canonical. --elf-hash-histogram is a hidden alias
* --relocs is canonical. --relocations is a hidden alias
* --section-groups is canonical. --elf-section-groups is a hidden alias
OptTable avoids global option collision if we decide to support multiplexing for binary utilities.
* Most one-dash long options are still supported. `-dt, -sd, -st, -sr` are dropped due to their conflict with grouped short options.
* `--section-mapping=false` (D57365) is strange but is kept for now.
* Many `cl::opt` variables were unnecessarily external. I added `static` whenever appropriate.
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D105532
Part of https://lists.llvm.org/pipermail/llvm-dev/2021-July/151622.html
"Binary utilities: switch command line parsing from llvm::cl to OptTable"
* `--totals=false` and `--totals=0` cannot be used. Omit the option.
* `--help-list` is removed. This is a `cl::` specific option.
OptTable avoids global option collision if we decide to support multiplexing for binary utilities.
Note: because the tool is simple, and its long options are uncommon, I just drop
the one-dash forms except `-arch <value>` (Darwin style).
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D105598
Similar to D104889. The tool is very simple and its long options are uncommon,
so just drop the one-dash form in this patch.
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D105605
Part of https://lists.llvm.org/pipermail/llvm-dev/2021-July/151622.html
"Binary utilities: switch command line parsing from llvm::cl to OptTable"
Users should generally observe no difference as long as they only use intended
option forms. Behavior changes:
* `-t=d` is removed. Use `-t d` instead.
* `--demangle=0` cannot be used. Omit the option or use `--no-demangle` instead.
* `--help-list` is removed. This is a `cl::` specific option.
Note:
* `-t` diagnostic gets improved.
* This patch avoids cl::opt collision if we decide to support multiplexing for binary utilities
* One-dash long options are still supported.
* The `-s` collision (`-s segment section` for Mach-O) is unfortunate. `-s` means `--print-armap` in GNU nm.
* This patch removes the last `cl::multi_val` use case from the `llvm/lib/Support/CommandLine.cpp` library
`-M` (`--print-armap`), `-U` (`--defined-only`), and `-W` (`--no-weak`)
are now deprecated. They could conflict with future GNU nm options.
(--print-armap has an existing alias -s, so GNU will unlikely add a new one.
--no-weak (not in GNU nm) is rarely used anyway.)
`--just-symbol-name` is now deprecated in favor of
`--format=just-symbols` and `-j`.
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D105330
Some behavior changes:
* `-t=d` is removed. Use `-t d` instead.
* one-dash long options like `-all` are supported. Use `--all` instead.
* `--all=0` or `--all=false` cannot be used. (Note: `--all` is silently ignored anyway)
* `--help-list` is removed. This is a `cl::` specific option.
Nobody is likely leveraging any of the above.
Advantages:
* `-t` diagnostic gets improved.
* in the absence of `HideUnrelatedOptions`, `--help` will not list unrelated options if linking against libLLVM-13git.so or linker GC is not used.
* Decrease the probability of cl::opt collision if we do decide to support multiplexing
Note: because the tool is so simple, used more for forensics instead of a building
tool, and its long options are unlikely used in one-dash form, I just drop the
one-dash form in this patch.
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D104889
Summary: The patch adds the StringTable dumping to
llvm-readobj. Currently only XCOFF is supported.
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D104613
The new documentation entry gives an example use case from
libomptarget.
Reviewed By: yln, jhenderson, davezarzycki
Differential Revision: https://reviews.llvm.org/D105208
Based on the discussion in PR50922, minor changes have been done to properly
output a valid JSON. Removed "not implemented" keys.
Differential Revision: https://reviews.llvm.org/D105064
llvm-readobj is an internal testing tool for binary formats. Its output and
command line options do not need to be stable. It isn't supposed to be part of a
build process.
llvm-readelf was created as a user-facing utility and its interface intends to
be compatible with GNU readelf (unless there are good reasons not to).
The two tools have mostly compatible options. -s and -t are noticeable
exceptions due to history. I think the cost of keeping the inconsistency
overweighs the little history-compatible benefit and hinders transition from
cl::opt to OptTable, so let's change it.
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D105055
The option --no-print-imm-hex was not included in the command guide for
llvm-objdump but appears in the help text. This commit adds it to the
command guide.
Differential Revision: https://reviews.llvm.org/D104717
For now, the source variable locations are printed at about the same
space as the comments for disassembled code, which can make some ranges
for variables disappear if a line contains comments, for example:
┠─ bar = W1
0: add x0, x2, #2, lsl #12 // =8192┃
4: add z31.d, z31.d, #65280 // =0xff00
8: nop ┻
The patch shifts the report a bit to allow printing comments up to
approximately 16 characters without interferences.
Differential Revision: https://reviews.llvm.org/D104700
Fix the command line guide for -g/-s/-S.
In particular, previously it was incorrectly stating that -S is an alias for --strip-all.
Differential revision: https://reviews.llvm.org/D104888
Change --max-timeline-cycles=0 to mean no limit on the number of cycles.
Use this in AMDGPU tests to show all instructions in the timeline view
instead of having it arbitrarily truncated.
Differential Revision: https://reviews.llvm.org/D104846
The original change was pushed in main as commit f7a23ecece.
It was then reverted by commit a04f01bab2 because it caused linker failures
on buildbots that don't build the AMDGPU target.
--
Some instructions are not defined well enough within the target’s scheduling
model for llvm-mca to be able to properly simulate its behaviour. The ideal
solution to this situation is to modify the scheduling model, but that’s not
always a viable strategy. Maybe other parts of the backend depend on that
instruction being modelled the way that it is. Or maybe the instruction is quite
complex and it’s difficult to fully capture its behaviour with tablegen. The
CustomBehaviour class (which I will refer to as CB frequently) is designed to
provide intuitive scaffolding for developers to implement the correct modelling
for these instructions.
More details are available in the original commit log message (f7a23ecece).
Differential Revision: https://reviews.llvm.org/D104149
I believe that after https://reviews.llvm.org/D102355 the behaviour of --print-source-context-lines has changed.
Before: --print-source-context-lines=3 prints 4 lines.
After: --print-source-context-lines=3 prints 3 lines.
Adjust the example in the docs for this change and make the testing a little more robust.
Differential Revision: https://reviews.llvm.org/D104114
Some instructions are not defined well enough within the target’s scheduling
model for llvm-mca to be able to properly simulate its behaviour. The ideal
solution to this situation is to modify the scheduling model, but that’s not
always a viable strategy. Maybe other parts of the backend depend on that
instruction being modelled the way that it is. Or maybe the instruction is quite
complex and it’s difficult to fully capture its behaviour with tablegen. The
CustomBehaviour class (which I will refer to as CB frequently) is designed to
provide intuitive scaffolding for developers to implement the correct modelling
for these instructions.
Implementation details:
llvm-mca does its best to extract relevant register, resource, and memory
information from every MCInst when lowering them to an mca::Instruction. It then
uses this information to detect dependencies and simulate stalls within the
pipeline. For some instructions, the information that gets captured within the
mca::Instruction is not enough for mca to simulate them properly. In these
cases, there are two main possibilities:
1. The instruction has a dependency that isn’t detected by mca.
2. mca is incorrectly enforcing a dependency that shouldn’t exist.
For the rest of this discussion, I will be focusing on (1), but I have put some
thought into (2) and I may revisit it in the future.
So we have an instruction that has dependencies that aren’t picked up by mca.
The basic idea for both pipelines in mca is that when an instruction wants to be
dispatched, we first check for register hazards and then we check for resource
hazards. This is where CB is injected. If no register or resource hazards have
been detected, we make a call to CustomBehaviour::checkCustomHazard() to give
the target specific CB the chance to detect and enforce any custom dependencies.
The return value for checkCustomHazaard() is an unsigned int representing the
(minimum) number of cycles that the instruction needs to stall for. It’s fine to
underestimate this value because when StallCycles gets down to 0, we’ll end up
checking for all the hazards again before the instruction is actually
dispatched. However, it’s important not to overestimate the value and the more
accurate your estimate is, the more efficient mca’s execution can be.
In general, for checkCustomHazard() to be able to detect these custom
dependencies, it needs information about the current instruction and also all of
the instructions that are still executing within the pipeline. The mca pipeline
uses mca::Instruction rather than MCInst and the current information encoded
within each mca::Instruction isn’t sufficient for my use cases. I had to add a
few extra attributes to the mca::Instruction class and have them get set by the
MCInst during instruction building. For example, the current mca::Instruction
doesn’t know its opcode, and it also doesn’t know anything about its immediate
operands (both of which I had to add to the class).
With information about the current instruction, a list of all currently
executing instructions, and some target specific objects (MCSubtargetInfo and
MCInstrInfo which the base CB class has references to), developers should be
able to detect and enforce most custom dependencies within checkCustomHazard. If
you need more information than is present in the mca::Instruction, feel free to
add attributes to that class and have them set during the lowering sequence from
MCInst.
Fortunately, in the in-order pipeline, it’s very convenient for us to pass these
arguments to checkCustomHazard. The hazard checking is taken care of within
InOrderIssueStage::canExecute(). This function takes a const InstRef as a
parameter (representing the instruction that currently wants to be dispatched)
and the InOrderIssueStage class maintains a SmallVector<InstRef, 4> which holds
all of the currently executing instructions. For the out-of-order pipeline, it’s
a bit trickier to get the list of executing instructions and this is why I have
held off on implementing it myself. This is the main topic I will bring up when
I eventually make a post to discuss and ask for feedback.
CB is a base class where targets implement their own derived classes. If a
target specific CB does not exist (or we pass in the -disable-cb flag), the base
class is used. This base class trivially returns 0 from its checkCustomHazard()
implementation (meaning that the current instruction needs to stall for 0 cycles
aka no hazard is detected). For this reason, targets or users who choose not to
use CB shouldn’t see any negative impacts to accuracy or performance (in
comparison to pre-patch llvm-mca).
Differential Revision: https://reviews.llvm.org/D104149
Alexey Lapshin