Transfer_ops can now work on both buffers and tensor. Right now, lowering of
the tensor case is not supported yet.
Differential Revision: https://reviews.llvm.org/D93500
For x86-64, D33100 added a diagnostic for local-exec TLS relocations referencing a preemptible symbol.
This patch generalizes it to non-preemptible symbols (see `-Bsymbolic` in `tls.s`)
on all targets.
Local-exec TLS relocations resolve to offsets relative to a fixed point within
the static TLS block, which are only meaningful for the executable.
With this change, `clang -fpic -shared -fuse-ld=bfd a.c` on the following example will be flagged for AArch64/ARM/i386/x86-64/RISC-V
```
static __attribute__((tls_model("local-exec"))) __thread long TlsVar = 42;
long bump() { return ++TlsVar; }
```
Note, in GNU ld, at least arm, riscv and x86's ports have the similar
diagnostics, but aarch64 and ppc64 do not error.
Differential Revision: https://reviews.llvm.org/D93331
Alternative to D91611.
The TLS General Dynamic/Local Dynamic code sequences need to mark
`__tls_get_addr` with R_PPC64_TLSGD or R_PPC64_TLSLD, e.g.
```
addis r3, r2, x@got@tlsgd@ha # R_PPC64_GOT_TLSGD16_HA
addi r3, r3, x@got@tlsgd@l # R_PPC64_GOT_TLSGD16_LO
bl __tls_get_addr(x@tlsgd) # R_PPC64_TLSGD followed by R_PPC64_REL24
nop
```
However, there are two deviations form the above:
1. direct call to `__tls_get_addr`. This is essential to implement ld.so in glibc/musl/FreeBSD.
```
bl __tls_get_addr
nop
```
This is only used in a -shared link, and thus not subject to the GD/LD to IE/LE
relaxation issue below.
2. Missing R_PPC64_TLSGD/R_PPC64_TLSGD for compiler generated TLS references
According to Stefan Pintille, "In the early days of the transition from the
ELFv1 ABI that is used for big endian PowerPC Linux distributions to the ELFv2
ABI that is used for little endian PowerPC Linux distributions, there was some
ambiguity in the specification of the relocations for TLS. The GNU linker has
implemented support for correct handling of calls to __tls_get_addr with a
missing relocation. Unfortunately, we didn't notice that the IBM XL compiler
did not handle TLS according to the updated ABI until we tried linking XL
compiled libraries with LLD."
In short, LLD needs to work around the old IBM XL compiler issue.
Otherwise, if the object file is linked in -no-pie or -pie mode,
the result will be incorrect because the 4 instructions are partially
rewritten (the latter 2 are not changed).
Work around the compiler bug by disable General Dynamic/Local Dynamic to
Initial Exec/Local Exec relaxation. Note, we also disable Initial Exec
to Local Exec relaxation for implementation simplicity, though technically it can be kept.
ppc64-tls-missing-gdld.s demonstrates the updated behavior.
Reviewed By: #powerpc, stefanp, grimar
Differential Revision: https://reviews.llvm.org/D92959
The fold currently only handles rotation patterns, but with the maturation of backend funnel shift handling we can now realistically handle all funnel shift patterns.
This should allow us to begin resolving PR46896 et al.
Ensure we block poison in a funnel shift value - similar to rG0fe91ad463fea9d08cbcd640a62aa9ca2d8d05e0
Reapplied with fix for PR48068 - we weren't checking that the shift values could be hoisted from their basicblocks.
Differential Revision: https://reviews.llvm.org/D90625
VE used to allocate VM1, VM2, VMP2 (VM4+VM5), and VM3. This patch
corrects to allocate VM1, VM2, VMP2 (VM4+VM5), and VM6. Also add
a regression test.
Reviewed By: simoll
Differential Revision: https://reviews.llvm.org/D93570
This patch makes VPRecipeBase a direct subclass of VPDef, moving the
SubclassID to VPDef.
Reviewed By: gilr
Differential Revision: https://reviews.llvm.org/D90564
Currently we lower invokes the same way as usual calls, e.g.:
V1 = STATEPOINT ... V (tied-def 0)
But this is incorrect is V1 is used on exceptional path.
By LLVM rules V1 neither dominates its uses in landing pad, nor
its live range is live on entry to landing pad. So compiler is
allowed to do various weird transformations like splitting live
range after statepoint and use split LR in catch block.
Until (and if) we find better solution to this problem, let's
use old lowering (spilling) for those values which are used on
exceptional path and allow VReg lowering for values used only
on normal path.
Differential Revision: https://reviews.llvm.org/D93449
Change to optimize references of elements of aggregate data. Also
add regression tests.
Reviewed By: simoll
Differential Revision: https://reviews.llvm.org/D93627
Using files with identical names leads to unexpected failures when tests
are run in parallel. This is tricky to reproduce, but has been happening
on some buildbots since merging https://reviews.llvm.org/D92854. In that
patch I added a unit test with a non-unique test file. This patch fixes
that.
This patch turns updates VPInterleaveRecipe to manage the values it defines
using VPDef. The VPValue is used during VPlan construction and
codegeneration instead of the plain IR reference where possible.
Reviewed By: gilr
Differential Revision: https://reviews.llvm.org/D90562
`LangOptions::SYCLVersion` can only have two values. This patch introduces an enum that allows us to reduce the member size from 32 bits to 1 bit.
Consequently, this also makes marshalling of this option fit into our model for enums: D84674.
Reviewed By: bader
Differential Revision: https://reviews.llvm.org/D93540
Before this patch, you needed to use `AutoNormalizeEnumJoined` whenever you wanted to **de**normalize joined enum.
Besides the naming confusion, this means the fact the option is joined is specified in two places: in the normalization multiclass and in the `Joined<["-"], ...>` multiclass.
This patch makes this work automatically, taking into account the `OptionClass` of options.
Also, the enum denormalizer now just looks up the spelling of the present enum case in a table and forwards it to the string denormalizer.
I also added more tests that exercise this.
Reviewed By: dexonsmith
Original patch by Daniel Grumberg.
Differential Revision: https://reviews.llvm.org/D84189
This replaces the existing `MarshallingInfoFlag<...>, IsNegative` with simpler `MarshallingInfoNegativeFlag`.
Reviewed By: dexonsmith
Original patch by Daniel Grumberg.
Differential Revision: https://reviews.llvm.org/D84675
An earlier patch introduced asserts that the InstructionCost is
valid because at that time the ReuseShuffleCost variable was an
unsigned. However, now that the variable is an InstructionCost
instance the asserts can be removed.
See this thread for context:
http://lists.llvm.org/pipermail/llvm-dev/2020-November/146408.html
See this patch for the introduction of the type:
https://reviews.llvm.org/D91174
This patch addresses two issues:
1. Not supported ops are updated to pick up the changes in the
SPIR-V dialect.
2. Conversion on `spv.ExecutionMode` is updated.
Reviewed By: antiagainst
Differential Revision: https://reviews.llvm.org/D91881
Errors from MCAssembler, MCObjectStreamer and *ObjectWriter typically cause a crash:
```
% cat c.c
int bar;
extern int foo __attribute__((alias("bar")));
% clang -c -fcommon c.c
fatal error: error in backend: Common symbol 'bar' cannot be used in assignment expr
PLEASE submit a bug report to ...
Stack dump:
...
```
`LLVMTargetMachine::addPassesToEmitFile` constructs `MachineModuleInfoWrapperPass`
which creates a MCContext without SourceMgr. `MCContext::reportError` calls
`report_fatal_error` which gets captured by Clang `LLVMErrorHandler` and gets translated
to the output above.
Since `MCContext::reportError` errors indicate user errors, such a crashing style error
is inappropriate. So this patch changes `report_fatal_error` to `SourceMgr().PrintMessage`.
```
% clang -c -fcommon c.c
<unknown>:0: error: Common symbol 'bar' cannot be used in assignment expr
```
Ideally we should at least recover the original filename (the line information
is generally lost). That requires general improvement to MC diagnostics,
because currently in many cases SMLoc information is lost.
Fast register allocator skips bundled MIs, as the main assignment
loop uses MachineBasicBlock::iterator (= MachineInstrBundleIterator)
This was causing SIInsertWaitcnts to crash which expects all
instructions to have registers assigned.
This patch makes sure to set everything inside bundle to the same
assignments done on BUNDLE header.
Reviewed By: qcolombet
Differential Revision: https://reviews.llvm.org/D90369
This patch base on D93366, and define vector fixed-point intrinsics.
1. vaaddu/vaadd/vasubu/vasub
2. vsmul
3. vssrl/vssra
4. vnclipu/vnclip
We work with @rogfer01 from BSC to come out this patch.
Authored-by: Roger Ferrer Ibanez <rofirrim@gmail.com>
Co-Authored-by: ShihPo Hung <shihpo.hung@sifive.com>
Differential Revision: https://reviews.llvm.org/D93508
Currently the integrated assembler only allows commas as the separator
between string arguments in .ascii. This patch adds support to using
space as separators and make IAS consistent with GNU assembler.
Link: https://github.com/ClangBuiltLinux/linux/issues/1196
Reviewed By: nickdesaulniers, jrtc27
Differential Revision: https://reviews.llvm.org/D91460
Summary: Some constants can be handled with less instructions than our current results. And it seems our original approach is not very easy to extend. Therefore this patch proposes to materialize all 64-bit constants by enumerated patterns.
I traversed almost all constants to verified the functionality of these pattens. A traversed comparison of the number of instructions used by the original method and the new method has also been completed, where no degradation was caused by this patch. This patch also passed Bootstrap test and SPEC test.
Improvements of this patch are shown in llvm/test/CodeGen/PowerPC/constants-i64.ll
Reviewed By: steven.zhang, stefanp
Differential Revision: https://reviews.llvm.org/D92089
This relands D64327 with a more specific workaround for R_386_GOTOFF
(gold<2.34 bug https://sourceware.org/bugzilla/show_bug.cgi?id=16794)
.debug_info has quite a few .debug_str relocations (R_386_32/R_ARM_ABS32).
The original workaround was too general and introduced too many .L symbols
used just as relocation targets.
From the original review:
... it reduced the size of a big ARM-32 debug image by 33%. It contained ~68M
of relocations symbols out of total ~71M symbols (96% of symbols table was
generated for relocations with symbol).
This patch does two things:
1: fix the typo that intrinsic mfvscr should be with no readmem property
2: since VSCR is not modeled yet, add has side effect for SAT bit clobber
intrinsics/instructions.
Reviewed By: steven.zhang
Differential Revision: https://reviews.llvm.org/D90807
Currently when building OpenMP along with LLVM, CMake variables for OpenMP (prefix with `LIBOMP` and `LIBOMPTARGET`) will not be passed through because by default it uses the prefix of the runtime name, aka `OPENMP` in this case. This patch fixed this issue.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D93603
This is consistent with the resolution to power-of-2 alignments.
Otherwise, emitCodeAlignment and emitValueToAlignment cannot handle alignments
larger than 2**32 and will trigger assertion failure (PR35218).
Note: GNU as as of 2.35 will use 1 for such a large byte `.align`
Currently there is an issue where the legacy pass manager uses a different OptBisect counter than the new pass manager.
This fix makes the npm OptBisectInstrumentation use the global OptBisect.
Reviewed By: aeubanks
Differential Revision: https://reviews.llvm.org/D92897
Dmitry Preobrazhensky