In the frame index lowering we have to insert shift and add
instructions to adjust stack object access. We need to take care of the stack
object user kind and use scalar shift/add for scalar users.
Reviewed By: rampitec
Differential Revision: https://reviews.llvm.org/D121524
VPIntrinsic::getStaticVectorLength infers the operational vector length
of a VPIntrinsic instance from a type that is used with the intrinsic.
The function used the mask operand before. Yet, vp.merge|select do not
have a mask operand (in the predicating sense that the other VP
intrinsics are using them - it is a selection mask for them). Fallback
to the return type to fix this.
Reviewed By: kaz7
Differential Revision: https://reviews.llvm.org/D121913
This fixes bug <https://github.com/llvm/llvm-project/issues/54022>. For
now this means that defined functions will have two .functype directives
emitted. Given discussion in that bug has suggested interest in moving
towards using something other than .functype to mark the beginning of a
function (which would, as a side-effect, solve this issue), this patch
doesn't attempt to avoid that duplication.
Some test cases that used CHECK-LABEL: foo rather than CHECK-LABEL: foo:
are broken by this change. This patch updates those test cases to always
have a colon at the end of the CHECK-LABEL string.
Differential Revision: https://reviews.llvm.org/D122134
In MinGW mode, it's possible to build LLVM/Clang with
LLVM_LINK_LLVM_DYLIB (which implicitly enables plugins too). Other
existing ways of building plugins on Windows is to build with
LLVM_EXPORT_SYMBOLS_FOR_PLUGINS, where each executable exports its
symbols.
With LLVM_LINK_LLVM_DYLIB, we can't generally skip building plugins
even if they are set up with PLUGIN_TOOL, as some plugins (e.g.
under clang/examples) set up that way do build properly (as
they manually call clang_target_link_libraries, which links in the
libclang-cpp.dll dylib).
For CTTestTidyModule, there's no corresponding dylib that would
provide the same exports.
Differential Revision: https://reviews.llvm.org/D121687
This requires some adjustment in caller code, because there was
a confusion regarding the meaning of the PtrTy argument: This
argument is the type of the pointer being loaded, not the addresses
being loaded from.
Add the UsesMaskPolicy flag to indicate the operations result
would be effected by the mask policy. (ex. mask operations).
It means RISCVInsertVSETVLI should decide the mask policy according
by mask policy operand or passthru operand.
If UsesMaskPolicy is false (ex. unmasked, store, and reduction operations),
the mask policy could be either mask undisturbed or agnostic.
Currently, RISCVInsertVSETVLI sets UsesMaskPolicy operations default to
MA, otherwise to MU to keep the current mask policy would not be changed
for unmasked operations.
Add masked-tama, masked-tamu, masked-tuma and masked-tumu test cases.
I didn't add all operations because most of implementations are using
the same pseudo multiclass. Some tests maybe be duplicated in different
tests. (ex. masked vmacc with tumu shows in vmacc-rv32.ll and masked-tumu)
I think having different tests only for policy would make the testing
clear.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D120226
The revision introduces a affine.min and affine.max canonicalization pattern that orders the result expressions. It flattens the result expressions to arrays of dimension and symbol coefficients plus one constant coefficient and rearranges them in lexicographic order.
Without the pattern, CSE will not eliminate two affine.min / affine.max operation if the results are ordered differently. For example, the operations
```
%1 = affine.min affine_map<(d0) -> (8, -d0 + 27)>(%arg4)
%2 = affine.min affine_map<(d0) -> (-d0 + 27, 8)>(%arg4)
```
doe not CSE. After applying the pattern, the two operations are equivalent
```
%1 = affine.min affine_map<(d0) -> (8, -d0 + 27)>(%arg4)
%2 = affine.min affine_map<(d0) -> (8, -d0 + 27)>(%arg4)
```
which enables CSE.
Reviewed By: nicolasvasilache
Differential Revision: https://reviews.llvm.org/D121819
An unnecessary dep of the getenv function is removed. From the x86_64
loader, a call to __llvm_libc::memcpy is replaced with call to
__llvm_libc::inline_memcpy.
The CoroSplit pass would check the existence of coroutine intrinsic
before starting work. It is not necessary and wasteful since it would
iterate over the Module.
This patch also removes the constraint on the corresponding of the
SmallVector for the possible coroutines in the Modules. The original
value is 4. Given coroutines is used actually in practice. 4 is really
relatively a low threshold.
To perform the cost model of vector casting, the patch consider most vector
casts as their scalar form and consider those vector form of free scalr castings
as 1.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D121771
`getValueAsBitsInit` will assert when the "SoftFail" isn't presented.
But given the 'if' statement below, we should've allowed this situation.
This patch fix this.
This patch attempts to deduce when the oilist element must be printed
based on the optional arguments to it. This especially helps creating
an operation accurately because with the current implementation, the
inferred unit attributes must be manually added to print the clauses
appropriately.
Reviewed By: Mogball
Differential Revision: https://reviews.llvm.org/D121579
This error was found when analyzing MySQL with CTU enabled.
When there are space characters in the lookup name, the current
delimiter searching strategy will make the file path wrongly parsed.
And when two lookup names have the same prefix before their first space
characters, a 'multiple definitions' error will be wrongly reported.
e.g. The lookup names for the two lambda exprs in the test case are
`c:@S@G@F@G#@Sa@F@operator int (*)(char)#1` and
`c:@S@G@F@G#@Sa@F@operator bool (*)(char)#1` respectively. And their
prefixes are both `c:@S@G@F@G#@Sa@F@operator` when using the first space
character as the delimiter.
Solving the problem by adding a length for the lookup name, making the
index items in the format of `<USR-Length>:<USR File> <Path>`.
---
In the test case of this patch, we found that it will trigger a "triple
mismatch" warning when using `clang -cc1` to analyze the source file
with CTU using the on-demand-parsing strategy in Darwin systems. And
this problem is also encountered in D75665, which is the patch
introducing the on-demand parsing strategy.
We temporarily bypass this problem by using the loading-ast-file
strategy.
Refer to the [discourse topic](https://discourse.llvm.org/t/60762) for
more details.
Differential Revision: https://reviews.llvm.org/D102669
Sinking must check for interference between the block prologue
and the instruction being sunk.
Specifically check for clobbering of uses by the prologue, and
overwrites to prologue defined registers by the sunk instruction.
Reviewed By: rampitec, ruiling
Differential Revision: https://reviews.llvm.org/D121277
See https://reviews.llvm.org/D121626 for details -- this re-enables the
CTAD we removed, since it does break some stuff as well (even though it's
not nearly as bad as the removed constructors fixed by D121626).
Differential Revision: https://reviews.llvm.org/D122184
This patch uses CMAKE_OSX_SYSROOT, which should contain the SDK path
that we're building against on Apple platforms, to determine which
platform we are compiling for and set the compiler-rt suffix accordingly.
Differential Revision: https://reviews.llvm.org/D122161
This test assumes that the driver will set -object_path_lto linker
flag and it requests the platform linker with -fuse-ld=. When lld is
used as the host linker, the host linker version is unset and so
Clang won't set -object_path_lto since that flag is set conditionally
only when linker version is at least 116. We set the linker version
explicitly to make sure that -object_path_lto is set. That approach
is already used elsewhere in the test.
Differential Revision: https://reviews.llvm.org/D122110
When lld is being used as host linker, skip version detection since
lld version cannot be used interchangeably with ld64 version and lld
is already handled specially in Clang driver.
Differential Revision: https://reviews.llvm.org/D122109
The unroll pragma did not properly work as the loop bound was not known
when we optimize the runtime and we then added a "unroll disable"
metadata which prevented unrolling later when the bounds were known.
For now we manually unroll to make sure up to 16 elements are handled
nicely. This helps optimizations to look through the argument passing.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D109164
There's a bug caused when a process is relaunched: the target, which
doesn't change, keeps the Trace object from the previous process, which
is already defunct, and causes segmentation faults when it's attempted
to be used.
A fix is to clean up the Trace object when the target is disposing of
the previous process during relaunches.
A way to reproduce this:
```
lldb a.out
b main
r
process trace start
c
r
process trace start
```
Differential Revision: https://reviews.llvm.org/D122176
I incorrectly returned an ArrayRef when the underlying object didn't own
the data. Instead, returning a vector<uint8_t> is what we should do.
This fixes an issue when trying to access an intel-pt trace buffer
larger than 16 MB.
repro
```
go to a breakpoint
thread trace start -s 16777216
n
thread trace dump instructions # this doesn't fail anymore
```
Differential Revision: https://reviews.llvm.org/D122192
Fix the log and progress report message for in-memory binaries. If
there's no object file, use the name from the Module. With this patch we
correctly show the library name when attaching to a remote process
without an expanded shared cache.
Differential revision: https://reviews.llvm.org/D122177
