This implements vp_add, vp_and for the VE target by lowering them to the
VVP_* layer. We also add helper functions for VP SDNodes (isVPSDNode,
getVPMaskIdx, getVPExplicitVectorLengthIdx).
Reviewed By: kaz7
Differential Revision: https://reviews.llvm.org/D93766
As noted in PR48689, the verifier may have some kind
of exponential behavior that should be addressed
separately. For now, only run it in debug mode to
prevent problems for release+asserts.
That limit is what we had before D80401, and I'm
not sure if there was a reason to change it in that
patch.
Clean ISel patterns for LSV and LVS before upstream more hand-written
ISel patterns.
Reviewed By: simoll
Differential Revision: https://reviews.llvm.org/D94291
Fixes a crash caused by D91255, when LLVMTy is null when
calling changeExtendedVectorElementType.
Differential Revision: https://reviews.llvm.org/D94234
In the following loop:
void foo(int *a, int *b, int N) {
for (int i=0; i<N; ++i)
a[i + 4] = a[i] + b[i];
}
The loop dependence constrains the VF to a maximum of (4, fixed), which
would mean using <4 x i32> as the vector type in vectorization.
Extending this to scalable vectorization, a VF of (4, scalable) implies
a vector type of <vscale x 4 x i32>. To determine if this is legal
vscale must be taken into account. For this example, unless
max(vscale)=1, it's unsafe to vectorize.
For SVE, the number of bits in an SVE register is architecturally
defined to be a multiple of 128 bits with a maximum of 2048 bits, thus
the maximum vscale is 16. In the loop above it is therefore unfeasible
to vectorize with SVE. However, in this loop:
void foo(int *a, int *b, int N) {
#pragma clang loop vectorize_width(X, scalable)
for (int i=0; i<N; ++i)
a[i + 32] = a[i] + b[i];
}
As long as max(vscale) multiplied by the number of lanes 'X' doesn't
exceed the dependence distance, it is safe to vectorize. For SVE a VF of
(2, scalable) is within this constraint, since a vector of <16 x 2 x 32>
will have no dependencies between lanes. For any number of lanes larger
than this it would be unsafe to vectorize.
This patch extends 'computeFeasibleMaxVF' to legalize scalable VFs
specified as loop hints, implementing the following behaviour:
* If the backend does not support scalable vectors, ignore the hint.
* If scalable vectorization is unfeasible given the loop
dependence, like in the first example above for SVE, then use a
fixed VF.
* Accept scalable VFs if it's safe to do so.
* Otherwise, clamp scalable VFs that exceed the maximum safe VF.
Reviewed By: sdesmalen, fhahn, david-arm
Differential Revision: https://reviews.llvm.org/D91718
We do this mostly to be able to test the insert_vector_elt isel
patterns. As long as we don't, most single element insertions show up as
`BUILD_VECTOR` in the backend.
Reviewed By: kaz7
Differential Revision: https://reviews.llvm.org/D93759
Update CodeGen regression tests with marker at first line telling it's
auto-generated by the script, under PowerPC directory. For some reason,
these tests are generated but manually written, which makes things
unclear when someone's change affecting them.
However, some tests only show simple change after re-generated, like
extra blank lines, disappearing '.localentry', etc. Besides, some tests
are generated but added checks for debug output. This commit doesn't try
updating them.
The new test case here contains a first order recurrences and an
instruction that is replicated. The first order recurrence forces an
instruction to be sunk _into_, as opposed to after the replication
region. That causes several things to go wrong including registering
vector instructions multiple times and failing to create dominance
relations correctly.
Instead we should be sinking to after the replication region, which is
what this patch makes sure happens.
Differential Revision: https://reviews.llvm.org/D93629
This patch introduces additional infrastructure necessary to accommodate DiagnosticOptions.
DiagnosticOptions are unique in that they are parsed by the same function in cc1 AND in the Clang driver. The call to the parsing function from the driver occurs early on in the compilation process, where no proper DiagnosticEngine exists, because the diagnostic options (passed through command line) are not known yet.
To preserve the current behavior, we need to be able to selectively parse:
* all options (for -cc1),
* only diagnostic options (for driver).
This patch achieves that in the following way:
* new MacroPrefix field is added to the Option TableGen class,
* new IsDiag TableGen mixin sets MacroPrefix to "DIAG_",
* TableGen backend serializes option records into a macro with the prefix,
* CompilerInvocation parse/generate methods define the [DIAG_]OPTION_WITH_MARSHALLING macros to handle diagnostic options separately.
Depends on D93700, D93701 & D93702.
Reviewed By: dexonsmith
Differential Revision: https://reviews.llvm.org/D84673
This factors out code from MachineLICM that determines whether an instruction
is loop-invariant, which is a generally useful function. Thus this allows to
use that helper elsewhere too.
Differential Revision: https://reviews.llvm.org/D94082
As the actual windows unwinder doesn't support this case, don't
pretend that it is supported when dumping the generated unwind info
either, even if it would be possible to interpret it as something
sensible.
This should reduce the risk of us emitting such a case in code
(although it's unlikely as long as the unwind info is generated
through the SEH opcodes, as the opcodes can't describe this case).
Differential Revision: https://reviews.llvm.org/D91529
This currently blocks --print-before/after with a legacy PM pass, for
example when we use the new PM for the optimization pipeline but the
legacy PM for the codegen pipeline. Also in the future when the codegen
pipeline works with the new PM there will be multiple places to specify
passes, so even when everything is using the new PM, there will still be
multiple places that can accept different pass names.
Reviewed By: hoy, ychen
Differential Revision: https://reviews.llvm.org/D94283
When `BUILD_WITH_INSTALL_RPATH` is enabled it prevents using a custom rpath only
for the build tree as the install rpath will be used. This makes it impossible to run a
runtimes build when compiling with Clang and wanting the installed rpath to be
empty (i.e. `-DCMAKE_BUILD_RPATH="<some path>" -DCMAKE_SKIP_INSTALL_RPATH=ON`).
Disable `BUILD_WITH_INSTALL_RPATH` when `CMAKE_BUILD_RPATH` is non-empty to
allow for such build scenarios.
Reviewed By: phosek
Differential Revision: https://reviews.llvm.org/D93177
I noticed __availability was missing, so I manually diffed the
file lists and put all recently(ish) added headers:
* __availability from 2eadbc86142ba
* concepts from 601f7631827ae
* execution from 0a06eb911b830
* numbers from 4f6c4b473c4a5
Also remove libcxx_install_support_headers like the CMake build did in
6706342f48bea, and unconditionally copy
support/win32/{limits_msvc_win32.h,locale_win32.h} like the CMake
build always did as far as I can tell.
We have modules with metadata on declarations, and out-of-tree passes
use that metadata, and we need to clone those modules. We really expect
such metadata is kept during the clone operation.
Reviewed by: arsenm, aprantl
Differential Revision: https://reviews.llvm.org/D93451
Define the `vfsqrt` IR intrinsics for the respective V instructions.
Authored-by: Roger Ferrer Ibanez <rofirrim@gmail.com>
Co-Authored-by: Evandro Menezes <evandro.menezes@sifive.com>
Differential Revision: https://reviews.llvm.org/D93745
We need to handle this case before dealing with the case of constant
branch condition, because if the destinations match, latter fold
would try to remove the DomTree edge that would still be present.
This allows to make that particular DomTree update non-permissive
There are only two used in the IR optimization pipeline.
Port these and add them to the default pipeline.
Similar to https://reviews.llvm.org/D93863.
I added -mtriple to some tests since under the new PM, the passes are
only available when the TargetMachine is specified.
Reviewed By: rnk
Differential Revision: https://reviews.llvm.org/D93930
In https://reviews.llvm.org/D88138 this was incorrectly added with
registerOptimizerLastEPCallback(), when it should be
registerLoopOptimizerEndEPCallback(), matching the legacy PM's
EP_LoopOptimizerEnd.
Reviewed By: rnk
Differential Revision: https://reviews.llvm.org/D93929
