If both flags created through BoolOption are CC1Option and the keypath has a non-default or non-implied value, the denormalizer gets called twice. If the denormalizer has the ability to generate both flags, we can end up generating the same flag twice.
Reviewed By: dexonsmith, Bigcheese
Differential Revision: https://reviews.llvm.org/D93094
We cannot be sure whether a flag is CC1Option inside the definition of `BoolOption`. Take the example below:
```
let Flags = [CC1Option] in {
defm xxx : BoolOption<...>;
}
```
where TableGen applies `Flags = [CC1Option]` to the `xxx` and `no_xxx` records **after** they have been is fully declared by `BoolOption`.
For the refactored `-f[no-]debug-pass-manager` flags (see the diff), this means `BoolOption` never adds any marshalling info, as it doesn't see either of the flags as `CC1Option`.
For that reason, we should defensively append the marshalling information to both flags inside `BoolOption`. Now the check for `CC1Option` needs to happen later, in the parsing macro, when all TableGen logic has been resolved.
However, for some flags defined through `BoolOption`, we can run into issues:
```
// Options.td
def fenable_xxx : /* ... */;
// Both flags are CC1Option, the first is implied.
defm xxx : BoolOption<"xxx,
"Opts.Xxx", DefaultsToFalse,
ChangedBy<PosFlag, [CC1Option], "", [fenable_xxx]>,
ResetBy<NegFlag, [CC1Option]>>;
```
When parsing `clang -cc1 -fenable-xxx`:
* we run parsing for `PosFlag`:
* set `Opts.Xxx` to default `false`,
* discover `PosFlag` is implied by `-fenable-xxx`: set `Opts.Xxx` to `true`,
* don't see `-fxxx` on command line: do nothing,
* we run parsing for `NegFlag`:
* set `Opts.Xxx` to default `false`,
* discover `NegFlag` cannot be implied: do nothing,
* don't see `-fno-xxx` on command line: do nothing.
Now we ended up with `Opts.Xxx` set to `false` instead of `true`. For this reason, we need to ensure to append the same `ImpliedByAnyOf` instance to both flags.
This means both parsing runs now behave identically (they set the same default value, run the same "implied by" check, and call `makeBooleanOptionNormalizer` that already has information on both flags, so it returns the same value in both calls).
The solution works well, but what might be confusing is this: you have defined a flag **A** that is not `CC1Option`, but can be implied by another flag **B** that is `CC1Option`:
* if **A** is defined manually, it will never get implied, as the code never runs
```
def no_signed_zeros : Flag<["-"], "fno-signed-zeros">, Group<f_Group>, Flags<[]>,
MarshallingInfoFlag<"LangOpts->NoSignedZero">, ImpliedByAnyOf<[menable_unsafe_fp_math]>;
```
* if **A** is defined by `BoolOption`, it could get implied, as the code is run by its `CC1Option` counterpart:
```
defm signed_zeros : BoolOption<"signed-zeros",
"LangOpts->NoSignedZero", DefaultsToFalse,
ChangedBy<NegFlag, [], "Allow optimizations that ignore the sign of floating point zeros",
[cl_no_signed_zeros, menable_unsafe_fp_math]>,
ResetBy<PosFlag, [CC1Option]>, "f">, Group<f_Group>;
```
This is a surprising inconsistency.
One solution might be to somehow propagate the final `Flags` of the implied flag in `ImpliedByAnyOf` and check whether it has `CC1Option` in the parsing macro. However, I think it doesn't make sense to spend time thinking about a corner case that won't come up in real code.
An observation: it is unfortunate that the marshalling information is a part of the flag definition. If we represented it in a separate structure, we could avoid the "double parsing" problem by having a single source of truth. This would require a lot of additional work though.
Note: the original patch missed the `CC1Option` check in the parsing macro, making my reasoning here incomplete. Moreover, it contained a change to denormalization that wasn't necessarily related to these changes, so I've extracted that to a follow-up patch: D93094.
Reviewed By: dexonsmith, Bigcheese
Differential Revision: https://reviews.llvm.org/D93008
This patch enables marshalling of the exception model options while enforcing their mutual exclusivity. The clang driver interface remains the same, this only affects the cc1 command line.
Depends on D93215.
Reviewed By: dexonsmith
Differential Revision: https://reviews.llvm.org/D93216
This squashes multiple members in LangOptions into one. This is leveraged in a follow-up patch that implements marshalling of related command-line options.
Depends on D93214.
Reviewed By: dexonsmith
Differential Revision: https://reviews.llvm.org/D93215
This introduces more flexible multiclass for declaring two flags controlling the same boolean keypath.
Compared to existing Opt{In,Out}FFlag multiclasses, the new syntax makes it easier to read option declarations and reason about the keypath.
This also makes specifying common properties of both flags possible.
I'm open to suggestions on the class names. Not 100% sure the benefits are worth the added complexity.
Depends on D92774.
Reviewed By: dexonsmith
Differential Revision: https://reviews.llvm.org/D92775
The new PM is considered stable and many downstream groups have adopted it (some
have adopted it for more than two years). Add -f[no-]legacy-pass-manager to reflect the
fact that it is no longer experimental and the legacy pass manager is something we strive to retire.
In the future, when the legacy PM eventually goes away,
-fno-experimental-new-pass-manager and -flegacy-pass-manager will be removed.
This patch also changes -f[no-]legacy-pass-manager to pass `-plugin-opt={new,legacy}-pass-manager` to the linker (supported by both ld.lld and LLVMgold.so) when -flto/-flto=thin is specified
Reviewed By: aeubanks, rsmith
Differential Revision: https://reviews.llvm.org/D92915
Currently, -ftime-report + new pass manager emits one line of report for each
pass run. This potentially causes huge output text especially with regular LTO
or large single file (Obeserved in private tests and was reported in D51276).
The behaviour of -ftime-report + legacy pass manager is
emitting one line of report for each pass object which has relatively reasonable
text output size. This patch adds a flag `-ftime-report=` to control time report
aggregation for new pass manager.
The flag is for new pass manager only. Using it with legacy pass manager gives
an error. It is a driver and cc1 flag. `per-pass` is the new default so
`-ftime-report` is aliased to `-ftime-report=per-pass`. Before this patch,
functionality-wise `-ftime-report` is aliased to `-ftime-report=per-pass-run`.
* Adds an boolean variable TimePassesHandler::PerRun to control per-pass vs per-pass-run.
* Adds a new clang CodeGen flag CodeGenOptions::TimePassesPerRun to work with the existing CodeGenOptions::TimePasses.
* Remove FrontendOptions::ShowTimers, its uses are replaced by the existing CodeGenOptions::TimePasses.
* Remove FrontendTimesIsEnabled (It was introduced in D45619 which was largely reverted.)
Differential Revision: https://reviews.llvm.org/D92436
Use lambdas with captures to replace the redundant infrastructure for marshalling of two boolean flags that control the same keypath.
Reviewed By: dexonsmith
Differential Revision: https://reviews.llvm.org/D92773
This patch implements correct hostness based overloading resolution
in isBetterOverloadCandidate.
Based on hostness, if one candidate is emittable whereas the other
candidate is not emittable, the emittable candidate is better.
If both candidates are emittable, or neither is emittable based on hostness, then
other rules should be used to determine which is better. This is because
hostness based overloading resolution is mostly for determining
viability of a function. If two functions are both viable, other factors
should take precedence in preference.
If other rules cannot determine which is better, CUDA preference will be
used again to determine which is better.
However, correct hostness based overloading resolution
requires overloading resolution diagnostics to be deferred,
which is not on by default. The rationale is that deferring
overloading resolution diagnostics may hide overloading reslolutions
issues in header files.
An option -fgpu-exclude-wrong-side-overloads is added, which is off by
default.
When -fgpu-exclude-wrong-side-overloads is off, keep the original behavior,
that is, exclude wrong side overloads only if there are same side overloads.
This may result in incorrect overloading resolution when there are no
same side candates, but is sufficient for most CUDA/HIP applications.
When -fgpu-exclude-wrong-side-overloads is on, enable deferring
overloading resolution diagnostics and enable correct hostness
based overloading resolution, i.e., always exclude wrong side overloads.
Differential Revision: https://reviews.llvm.org/D80450
This allows us to use its value everywhere, rather than just clang. Some
other places, like opt and lld, will use its value soon.
Rename it internally to LLVM_ENABLE_NEW_PASS_MANAGER.
The #define for it is now in llvm-config.h.
The initial land accidentally set the value of
LLVM_ENABLE_NEW_PASS_MANAGER to the string
ENABLE_EXPERIMENTAL_NEW_PASS_MANAGER instead of its value.
Reviewed By: rnk, hans
Differential Revision: https://reviews.llvm.org/D92072
This allows us to use its value everywhere, rather than just clang. Some
other places, like opt and lld, will use its value soon.
The #define for it is now in llvm-config.h.
Reviewed By: rnk, hans
Differential Revision: https://reviews.llvm.org/D92072
This makes the options API composable, allows boolean flags to imply non-boolean values and makes the code more logical (IMO).
Differential Revision: https://reviews.llvm.org/D91861
This is the #2 of 2 changes that make remarks hotness threshold option
available in more tools. The changes also allow the threshold to sync with
hotness threshold from profile summary with special value 'auto'.
This change expands remarks hotness threshold option
-fdiagnostics-hotness-threshold in clang and *-remarks-hotness-threshold in
other tools to utilize hotness threshold from profile summary.
Remarks hotness filtering relies on several driver options. Table below lists
how different options are correlated and affect final remarks outputs:
| profile | hotness | threshold | remarks printed |
|---------|---------|-----------|-----------------|
| No | No | No | All |
| No | No | Yes | None |
| No | Yes | No | All |
| No | Yes | Yes | None |
| Yes | No | No | All |
| Yes | No | Yes | None |
| Yes | Yes | No | All |
| Yes | Yes | Yes | >=threshold |
In the presence of profile summary, it is often more desirable to directly use
the hotness threshold from profile summary. The new argument value 'auto'
indicates threshold will be synced with hotness threshold from profile summary
during compilation. The "auto" threshold relies on the availability of profile
summary. In case of missing such information, no remarks will be generated.
Differential Revision: https://reviews.llvm.org/D85808
This change introduces a new clang switch `-fpseudo-probe-for-profiling` to enable AutoFDO with pseudo instrumentation. Please refer to https://reviews.llvm.org/D86193 for the whole story.
One implication from pseudo-probe instrumentation is that the profile is now sensitive to CFG changes. We perform the pseudo instrumentation very early in the pre-LTO pipeline, before any CFG transformation. This ensures that the CFG instrumented and annotated is stable and optimization-resilient.
The early instrumentation also allows the inliner to duplicate probes for inlined instances. When a probe along with the other instructions of a callee function are inlined into its caller function, the GUID of the callee function goes with the probe. This allows samples collected on inlined probes to be reported for the original callee function.
Reviewed By: wmi
Differential Revision: https://reviews.llvm.org/D86502
Added support for the options mabi=vec-extabi and mabi=vec-default which are analogous to qvecnvol and qnovecnvol when using XL on AIX.
The extended Altivec ABI on AIX is enabled using mabi=vec-extabi in clang and vec-extabi in llc.
Reviewed By: Xiangling_L, DiggerLin
Differential Revision: https://reviews.llvm.org/D89684
Recently HIP toolchain made a change to use clang instead of opt/llc to do compilation
(https://reviews.llvm.org/D81861). The intention is to make HIP toolchain canonical like
other toolchains.
However, this change introduced an unintentional change regarding backend fp fuse
option, which caused regressions in some HIP applications.
Basically before the change, HIP toolchain used clang to generate bitcode, then use
opt/llc to optimize bitcode and generate ISA. As such, the amdgpu backend takes
the default fp fuse mode which is 'Standard'. This mode respect contract flag of
fmul/fadd instructions and do not fuse fmul/fadd instructions without contract flag.
However, after the change, HIP toolchain now use clang to generate IR, do optimization,
and generate ISA as one process. Now amdgpu backend fp fuse option is determined
by -ffp-contract option, which is 'fast' by default. And this -ffp-contract=fast language option
is translated to 'Fast' fp fuse option in backend. Suddenly backend starts to fuse fmul/fadd
instructions without contract flag.
This causes wrong result for some device library functions, e.g. tan(-1e20), which should
return 0.8446, now returns -0.933. What is worse is that since backend with 'Fast' fp fuse
option does not respect contract flag, there is no way to use #pragma clang fp contract
directive to enforce fp contract requirements.
This patch fixes the regression by introducing a new value 'fast-honor-pragmas' for -ffp-contract
and use it for HIP by default. 'fast-honor-pragmas' is equivalent to 'fast' in frontend but
let the backend to use 'Standard' fp fuse option. 'fast-honor-pragmas' is useful since 'Fast'
fp fuse option in backend does not honor contract flag, it is of little use to HIP
applications since all code with #pragma STDC FP_CONTRACT or any IR from a
source compiled with -ffp-contract=on is broken.
Differential Revision: https://reviews.llvm.org/D90174
The dependency mechanism for C has been implemented, and we have rolled out
this to all internal users, didn't see crashy issues, we consider it is stable
enough.
Differential Revision: https://reviews.llvm.org/D89046
- The new option, -arcmt-action, is a simple enum based option.
- The driver is modified to translate the existing -ccc-acmt-* options accordingly
Depends on D83298
Reviewed By: Bigcheese
Original patch by Daniel Grumberg.
Differential Revision: https://reviews.llvm.org/D83315
As with precompiled headers, it's useful for indexers to be able to
continue through compiler errors in dependent modules.
Resolves rdar://69816264
Reviewed By: akyrtzi
Differential Revision: https://reviews.llvm.org/D91580
Add an option -munsafe-fp-atomics for AMDGPU target.
When enabled, clang adds function attribute "amdgpu-unsafe-fp-atomics"
to any functions for amdgpu target. This allows amdgpu backend to use
unsafe fp atomic instructions in these functions.
Differential Revision: https://reviews.llvm.org/D91546
This enables automatically parsing and generating CC1 arguments for options where two flags control the same field, e.g. -fexperimental-new-pass-manager and -fno-experimental new pass manager.
Reviewed By: Bigcheese, dexonsmith
Original patch by Daniel Grumberg.
Differential Revision: https://reviews.llvm.org/D83071
See discussion in https://bugs.llvm.org/show_bug.cgi?id=45073 / https://reviews.llvm.org/D66324#2334485
the implementation is known-broken for certain inputs,
the bugreport was up for a significant amount of timer,
and there has been no activity to address it.
Therefore, just completely rip out all of misexpect handling.
I suspect, fixing it requires redesigning the internals of MD_misexpect.
Should anyone commit to fixing the implementation problem,
starting from clean slate may be better anyways.
This reverts commit 7bdad08429411e7d0ecd58cd696b1efe3cff309e,
and some of it's follow-ups, that don't stand on their own.
Merge existing marhsalling info kinds and add some primitives to
express flag options that contribute to a bitfield.
Depends on D82574
Original patch by Daniel Grumberg.
Reviewed By: Bigcheese
Differential Revision: https://reviews.llvm.org/D82860
As of D80952 we are disabling strict floating point on all hosts except
those that are explicitly listed as supported. Use of strict floating point
on other hosts requires use of the -fexperimental-strict-floating-point
flag. This is to avoid bugs like "https://bugs.llvm.org/show_bug.cgi?id=45329"
(which has an incorrect link in the previous review).
In the review for D80952 I was asked to mark the -fexperimental option as
a MarshallingInfoFlag. This patch does exactly that.
Differential Revision: https://reviews.llvm.org/D88987
This reverts commit 09248a5d25bb1c9f357247fa3da8fbe4470e9c67.
Some builds are broken. I suspect a `static constexpr` in a class missing a
definition out of class (required pre-c++17).
Merge existing marhsalling info kinds and add some primitives to
express flag options that contribute to a bitfield.
Depends on D82574
Reviewed By: Bigcheese
Differential Revision: https://reviews.llvm.org/D82860
This ports a number of OpenCL and fast-math flags for floating point
over to the new marshalling infrastructure.
As part of this, `Opt{In,Out}FFlag` were enhanced to allow other flags to
imply them, via `DefaultAnyOf<>`. For example:
```
defm signed_zeros : OptOutFFlag<"signed-zeros", ...,
"LangOpts->NoSignedZero",
DefaultAnyOf<[cl_no_signed_zeros, menable_unsafe_fp_math]>>;
```
defines `-fsigned-zeros` (`false`) and `-fno-signed-zeros` (`true`)
linked to the keypath `LangOpts->NoSignedZero`, defaulting to `false`,
but set to `true` implicitly if one of `-cl-no-signed-zeros` or
`-menable-unsafe-fp-math` is on.
Note that the initial patch was written Daniel Grumberg.
Differential Revision: https://reviews.llvm.org/D82756
