This is a squash of 13 commits required in the lead up to r327898,
which fixes https://github.com/JuliaLang/julia/issues/27603. The squashed
commits are:
[SelectionDAG] Fix UpdateChains handling of TokenFactors
Summary:
In Instruction Selection UpdateChains replaces all matched Nodes'
chain references including interior token factors and deletes them.
This may allow nodes which depend on these interior nodes but are not
part of the set of matched nodes to be left with a dangling dependence.
Avoid this by doing the replacement for matched non-TokenFactor nodes.
[X86] Avoid using high register trick for test instruction
Summary:
It seems it's main effect is to create addition copies when values are inr register that do not support this trick, which increase register pressure and makes the code bigger.
[X86] Don't look for TEST instruction shrinking opportunities when the root node is a X86ISD::SUB.
I don't believe we ever create an X86ISD::SUB with a 0 constant which is what the TEST handling needs. The ternary operator at the end of this code shows up as only going one way in the llvm-cov report from the bots.
[X86] Teach LowerBUILD_VECTOR to recognize pair-wise splats of 32-bit elements and use a 64-bit broadcast
If we are splatting pairs of 32-bit elements, we can use a 64-bit broadcast to get the job done.
We could probably could probably do this with other sizes too, for example four 16-bit elements. Or we could broadcast pairs of 16-bit elements using a 32-bit element broadcast. But I've left that as a future improvement.
I've also restricted this to AVX2 only because we can only broadcast loads under AVX.
This removes tests that are failing due to codegen differences,
after the latest set of backports. Fixing thse for the backport
branch does not seem worth it.
Gbp-Pq: Topic julia
Gbp-Pq: Name llvm-rL327898.patch
[X86] Fix some isel patterns that used aligned vector load instructions with unaligned predicates.
These patterns weren't checking the alignment of the load, but were using the aligned instructions. This will cause a GP fault if the data isn't aligned.
[VNCoercion] Disallow coercion between different ni addrspaces
Summary:
I'm not sure if it would be legal by the IR reference to introduce
an addrspacecast here, since the IR reference is a bit vague on
the exact semantics, but at least for our usage of it (and I
suspect for many other's usage) it is not. For us, addrspacecasts
between non-integral address spaces carry frontend information that the
optimizer cannot deduce afterwards in a generic way (though we
have frontend specific passes in our pipline that do propagate
these). In any case, I'm sure nobody is using it this way at
the moment, since it would have introduced inttoptrs, which
are definitely illegal.
[JumpThreading] Don't restrict cast-traversal to i1
Summary:
In D17663, JumpThreading learned to look trough simple cast instructions,
but only if the source of those cast instructions was a phi/cmp i1
(in an effort to limit compile time effects). I think this condition
is too restrictive. For switches with limited value range, InstCombine
will readily introduce an extra `trunc` instruction to a smaller
integer type (e.g. from i8 to i2), leaving us in the somewhat perverse
situation that jump-threading would work before running instcombine,
but not after. Since instcombine produces this pattern, I think we
need to consider it canonical and support it in JumpThreading.
In general, for limiting recursion, I think the existing restriction
to phi and cmp nodes should be sufficient to avoid looking through
unprofitable chains of instructions.
[hurd] Make getMainExecutable get the real binary path
On GNU/Hurd, llvm-config is returning bogus value, such as:
$ llvm-config-6.0 --includedir
/usr/include
while it should be:
$ llvm-config-6.0 --includedir
/usr/lib/llvm-6.0/include
This is because getMainExecutable does not get the actual installation
path. On GNU/Hurd, /proc/self/exe is indeed a symlink to the path that
was used to start the program, and not the eventual binary file. Llvm's
getMainExecutable thus needs to run realpath over it to get the actual
place where llvm was installed (/usr/lib/llvm-6.0/bin/llvm-config), and
not /usr/bin/llvm-config-6.0. This will not change the result on Linux,
where /proc/self/exe already points to the eventual file.
Patch by Samuel Thibault!
While making changes here, I reformatted this block a bit to reduce
indentation and match 2 space indent style.
[LSR] Don't force bases of foldable formulae to the final type.
Summary:
Before emitting code for scaled registers, we prevent
SCEVExpander from hoisting any scaled addressing mode
by emitting all the bases first. However, these bases
are being forced to the final type, resulting in some
odd code.
For example, if the type of the base is an integer and
the final type is a pointer, we will emit an inttoptr
for the base, a ptrtoint for the scale, and then a
'reverse' GEP where the GEP pointer is actually the base
integer and the index is the pointer. It's more intuitive
to use the pointer as a pointer and the integer as index.
[SCEV] Don't expand Wrap predicate using inttoptr in ni addrspaces
Summary:
In non-integral address spaces, we're not allowed to introduce inttoptr/ptrtoint
intrinsics. Instead, we need to expand any pointer arithmetic as geps on the
base pointer. Luckily this is a common task for SCEV, so all we have to do here
is hook up the corresponding helper function and add test case.
James Clarke [Fri, 15 May 2020 16:25:16 +0000 (17:25 +0100)]
[Sparc] Include __tls_get_addr in symbol table for TLS calls to it
Last-Update: 2018-02-14
Global Dynamic and Local Dynamic call relocations only implicitly
reference __tls_get_addr, but it still needs to be in the symbol table
to be bound at link time otherwise it fails to link. For details, see
https://sourceware.org/bugzilla/show_bug.cgi?id=22832.
Gbp-Pq: Name sparc64-add-missing-tls-get-addr.diff
lldb-server exec users always /usr/bin/lldb-server. Server is required
for any debugging with lldb which makes it unusable unless default version
package has been installed. Small changes to code and debian/rules allows
a workaround for lldb-server start up.
To use this one needs to add cmake defination during configure. eg
-DDEBIAN_VERSION_SUFFIX=-$(LLVM_VERSION)
Better implementation would be to use /usr/share/llvm-$(VERSION)/bin but
that change seems to require a big change to the path handling code
which could then break something else.
This probably should have upstream bug but I couldn't find any existing report.
Gbp-Pq: Name lldb-addversion-suffix-to-llvm-server-exec.patch
Sylvestre [Fri, 15 May 2020 16:25:16 +0000 (17:25 +0100)]
Silent a test failing on yakkety amd64
Last-Update: 2016-08-26
/tmp/buildd/llvm-toolchain-snapshot-4.0~svn279801/test/tools/llvm-symbolizer/print_context.c:16:11: error: expected string not found in input
// CHECK: inc
^
<stdin>:1:1: note: scanning from here
_fini
^
<stdin>:1:3: note: possible intended match here
_fini
^
Peter Wu [Fri, 4 May 2018 13:55:26 +0000 (15:55 +0200)]
[PATCH] [clang] cmake: resolve symlinks in ClangConfig.cmake
Ensure that symlinks such as /usr/lib/cmake/clang-X.Y (pointing to
/usr/lib/llvm-X.Y/lib/cmake/llvm) are resolved. This ensures that
CLANG_INSTALL_PREFIX ends up to be /usr/lib/llvm-X.Y instead of /usr.
Partially addresses PR37128
Gbp-Pq: Name 0001-tools-clang-cmake-resolve-symlinks-in-ClangConfig.cmake.patch
Peter Wu [Fri, 4 May 2018 13:43:06 +0000 (15:43 +0200)]
[PATCH] [llvm] cmake: resolve symlinks in LLVMConfig.cmake
Ensure that symlinks such as /usr/lib/llvm-X.Y/cmake (pointing to
lib/cmake/llvm) are resolved. This ensures that LLVM_INSTALL_PREFIX
becomes /usr/lib/llvm-X.Y instead of /usr.
Partially addresses PR37128
Gbp-Pq: Name 0001-llvm-cmake-resolve-symlinks-in-LLVMConfig.cmake.patch
Command Output (stderr):
--
/build/llvm-toolchain-snapshot-4.0~svn279916/test/tools/gold/X86/start-lib-common.ll:22:10: error: expected string not found in input
; CHECK: @x = common global i32 0, align 8
^
<stdin>:1:1: note: scanning from here
; ModuleID = '/build/llvm-toolchain-snapshot-4.0~svn279916/build-llvm/test/tools/gold/X86/Output/start-lib-common.ll.tmp3.o'
^
<stdin>:4:1: note: possible intended match here
@x = common global i32 0, align 4
^
# Without this patch, the first local include of unwind.h might, with the
# __has_include_next, try to include the one from the system.
# It might be /usr/include/clang/3.4/include/unwind.h
# Because of the #ifndef __CLANG_UNWIND_H, it might never include any declaration
# from the system.
* Non-maintainer upload.
* Maintainer agreed NMU to drop clang-tidy-6.0 since it is not needed to
support ghc, the only remaining llvm-6.0 user, and allows progress on
python2 removal (Closes: #956903)
projects / llvm-toolchain-6.0.git / log