Since the divergence-driven instruction selection has been enabled for AMDGPU,
all the uniform instructions are expected to be selected to SALU form, except those not having one.
VGPR to SGPR copies appear in MIR to connect values producers and consumers. This change implements an algorithm
that evolves a reasonable tradeoff between the profit achieved from keeping the uniform instructions in SALU form
and overhead introduced by the data transfer between the VGPRs and SGPRs.
Reviewed By: rampitec
Differential Revision: https://reviews.llvm.org/D128252
Introducing the support for evaluating the constructor
of every element in an array. The idea is to record the
index of the current array member being constructed and
create a loop during the analysis. We looping over the
same CXXConstructExpr as many times as many elements
the array has.
Differential Revision: https://reviews.llvm.org/D127973
Try the shared library first, and if it doesn't exist fallback onto
the static one. When the static library is requested, skip the shared
library.
Differential Revision: https://reviews.llvm.org/D129470
On my system the date formatting is a bit different from what the test used to
support. I'm using:
Windows 11 version 21H2, build 22000.795 using the English(Canada) region.
ls from BusyBox 1.36
VS 2022 17.2.5
WinSDK 10.0.22000
After recent bufferization improvement, this test
started failing due to missed zero initialization.
Reviewed By: jpienaar
Differential Revision: https://reviews.llvm.org/D129800
Add the collective subroutine, co_sum, to the list of intrinsics.
In accordance with 16.9.50 and 16.9.137, add a check for and an
error if coindexed objects are being passed to certain arguments
in co_sum and in move_alloc. Add a semantics test to check that
this error is successfully caught in calls to move_alloc. Remove
the XFAIL directive, update the ERROR directives and add both
standard-conforming and non-standard conforming calls to the
semantics test for co_sum.
Reviewed By: jeanPerier
Differential Revision: https://reviews.llvm.org/D114134
The lds_barrier op allows workgroups to wait at a barrier for
operations to/from their local data store (LDS) to complete without
incurring the performance penalties of a full memory fence.
Reviewed By: nirvedhmeshram
Differential Revision: https://reviews.llvm.org/D129522
As we are moving towards support for multiple fragments, loops that
iterate over all basic blocks of a function, but do not depend on the
order of basic blocks in the final layout, should iterate over binary
functions directly, rather than the layout.
Eventually, all loops using the layout list should either iterate over
the function, or be aware of multiple layouts. This patch replaces
references to binary function's block layout with the binary function
itself where only little code changes are necessary.
Reviewed By: maksfb
Differential Revision: https://reviews.llvm.org/D129585
D127650 removed support for non-clang-based XL compilers, but left some
of the headers used only by this compiler and included under the
__IBMCPP__ macro. This change cleans this up by deleting these headers.
Reviewed By: hubert.reinterpretcast, fanbo-meng
Differential Revision: https://reviews.llvm.org/D129491
We already link libunwind explicitly so avoid trying to link toolchain's
default libunwind which may be missing. This matches what we already do
for libcxx.
Differential Revision: https://reviews.llvm.org/D129469
Add two options, `-fprofile-function-groups=N` and `-fprofile-selected-function-group=i` used to partition functions into `N` groups and only instrument the functions in group `i`. Similar options were added to xray in https://reviews.llvm.org/D87953 and the goal is the same; to reduce instrumented size overhead by spreading the overhead across multiple builds. Raw profiles from different groups can be added like normal using the `llvm-profdata merge` command.
Reviewed By: ianlevesque
Differential Revision: https://reviews.llvm.org/D129594
Widening a G_FCONSTANT by extending and then generating G_FPTRUNC doesn't produce
the same result all the time. Instead, we can just transform it to a G_CONSTANT
of the same bit pattern and truncate using a plain G_TRUNC instead.
Fixes https://github.com/llvm/llvm-project/issues/56454
Differential Revision: https://reviews.llvm.org/D129743
This reverts commit b7e77ff25f.
Reason: Broke sanitizer builds bots + libcxx. 'static assertion
expression is not an integral constant expression'. More details
available in the Phabricator review: https://reviews.llvm.org/D129048
Follow up to 3bc09c7da5 - remove a fixme I forgot to remove, and add test cases showing remaining work.
Note that scaled vscales show up in vectorized code from a couple of sources:
* Element types smaller than vector block size (i.e. everything under i64)
* Unrolling
* LMUL > 1
The largest scaling we can currently have is 256 (e8 in every possible vector register). More practically useful scales are in the 2-16 range.
Nobody knows if there are users of libc++ with MSVC. Let's try to find that out and encourage them to upstream their changes to make that configuration work.
Reviewed By: ldionne, #libc
Spies: libcxx-commits
Differential Revision: https://reviews.llvm.org/D129055
Remove m_inferior_prev_state that's not suitable for multiprocess
debugging and that does not seem to be really used at all.
The only use of the variable right now is to "prevent" sending the stop
reason after attach/launch. However, this code is never actually run
since none of the process plugins actually use eStateLaunching or
eStateAttaching. Through adding an assert, I've confirmed that it's
never hit in any of the LLDB tests or while attaching/launching debugged
process via lldb-server and via lldb CLI.
Differential Revision: https://reviews.llvm.org/D128878
Sponsored by: The FreeBSD Foundation
Initial optimization is to convert (i64 (zext (i32 X))) to
(i64 (sext (i32 X))) if the dominating condition for the basic block
guaranteed the sign bit of X is zero.
This frequently occurs in loop preheaders where a signed induction
variable that can never be negative has been widened. There will be
a dominating check that the 32-bit trip count isn't negative or zero.
The check here is not restricted to that specific case though.
A i32->i64 sext is cheaper than zext on RV64 without the Zba
extension. Later optimizations can often remove the sext from the
preheader basic block because the dominating block also needs a sext to
evaluate the greater than 0 check.
Reviewed By: asb
Differential Revision: https://reviews.llvm.org/D129732
clang 14 removed -gz=zlib-gnu support and ld.lld/llvm-objcopy removed zlib-gnu
support recently. Remove lldb support by migrating away from
llvm::object::Decompressor::isCompressedELFSection.
The API has another user llvm-dwp, so it is not removed in this patch.
Reviewed By: labath
Differential Revision: https://reviews.llvm.org/D129724
If an MI will not generate a target instruction, we should not compute its
latency. Then we can compute more precise instruction sequence cost, and get
better result.
Differential Revision: https://reviews.llvm.org/D129615
I happened to notice a two places where the enum was being pass
directly to the bool IsSigned argument of createExtendInst. This
was functionally ok since SignExtended in the enum has value
of 1, but the code shouldn't rely on that.
Using an enum class prevents the enum from being convertible to bool,
but does make writing the enum values more verbose. Since we now
have to write ExtendKind:: in front of them, I've shortened the
names of ZeroExtended and SignExtended.
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D129733
While testing backports of
https://reviews.llvm.org/D129572#inline-1245936
commit 2240d72f15 ("[X86] initial -mfunction-return=thunk-extern support")
I noticed that one of my unit tests mistyped a function attribute. The
unit test was intended to test fn attr merging behavior, but with the
typo it was not. Small fixup.
Reviewed By: aaron.ballman, erichkeane
Differential Revision: https://reviews.llvm.org/D129691
After 675080a453, we always create SCEVs for all operands of a
SelectInst. This can cause notable compile-time regressions compared to
the recursive algorithm, which only evaluates the operands if the select
is in a form we can create a usable expression.
This approach adds additional logic to getOperandsToCreate to only
queue operands for selects if we will later be able to construct a
usable SCEV.
Unfortunately this introduces a bit of coupling between actual SCEV
construction for selects and getOperandsToCreate, but I am not sure if
there are better alternatives to address the regression mentioned for
675080a453.
This doesn't have any notable compile-time impact on CTMark.
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D129731
We previously enabled subregister liveness by default when compiling
with RVV. This has been shown to cause miscompilations where RVV
register operand constraints are not met. A test was added for this in
D129639 which explains the issue in more detail.
Until this issue is fixed in some way, we should not be enabling
subregister liveness unless the user asks for it.
Reviewed By: craig.topper, rogfer01, kito-cheng
Differential Revision: https://reviews.llvm.org/D129646
Alexander Timofeev