We don't need to populate a string table, so we don't need to pre-parse the
remarks.
Split the bitstream2yaml and yaml2bitstream cases up so that we can avoid the
extra remark traversal in the bitstream2yaml case.
This kills the TODO.
Unary shuffles such as <0,2,4,6,8,10,12,14> or <1,3,5,7,9,11,13,15>
where half the elements are returned, can be lowered using vnsrl.
SelectionDAGBuilder lowers such shuffles as a build_vector of
extract_elements since the mask has less elements than the source.
To fix this, I've enable the extractSubvectorIsCheapHook to allow
DAGCombine to rebuild the shuffle using 2 extract_subvectors preceding
the shufffle.
I've gone very conservative on extractSubvectorIsCheapHook to minimize
test impact and match what we have test coverage for. This can be
improved in the future.
Reviewed By: reames
Differential Revision: https://reviews.llvm.org/D133736
This is particularly a problem for alias construction, where you might
want to have a backtick surrounded option in the alias. Before this
patch:
command alias expression -Z \`argc\` -- argv
for instance would be rendered as:
expression -Z argc -- argv
and would fail to work.
Differential Revision: https://reviews.llvm.org/D133045
This hook is currently only used by CodeGenPrepare, which will sink *and
duplicate* an 'and' into a block that has an 'icmp 0' user of it if the
hook returns true.
This hook is less useful for RISC-V than for targets like AArch64 that
have a TBZ (test bit and branch if zero instruction), but may still be
profitable if Zbs is available and a BEXTI can be selected.
Conservatively, we return false even if Zbs is enabled for any masks
that fit in the ANDI immediate because it's possible the only use is a
branch on the result, and ANDI+BNEZ => BEXTI+BNEZ isn't a profitable
transformation.
Differential Revision: https://reviews.llvm.org/D131492
I went over the output of the following mess of a command:
(ulimit -m 2000000; ulimit -v 2000000; git ls-files -z | parallel
--xargs -0 cat | aspell list --mode=none --ignore-case | grep -E
'^[A-Za-z][a-z]*$' | sort | uniq -c | sort -n | grep -vE '.{25}' |
aspell pipe -W3 | grep : | cut -d' ' -f2 | less)
and proceeded to spend a few days looking at it to find probable typos
and fixed a few hundred of them in all of the llvm project (note, the
ones I found are not anywhere near all of them, but it seems like a
good start).
Differential revision: https://reviews.llvm.org/D131122
The following replacements have been changed to `%mlir_lib_dir`:
- `%linalg_test_lib_dir`
- `%cuda_wrapper_library_dir`
- `%spirv_wrapper_library_dir`
- `%vulkan_wrapper_library_dir`
- `%mlir_runner_utils_dir`
- `%mlir_integration_test_dir`
If there are no requests to delay submission, I plan to submit this on 9/13/2002 (Tuesday).
See also PSA: https://discourse.llvm.org/t/psa-removing-mlir-lit-test-replacements/65124
Reviewed By: herhut
Differential Revision: https://reviews.llvm.org/D133409
Previously, we would add entries to DataInCodeSection in the order they
appeared in input files. Because of this, entries would not be sorted if
sections were reordered due to e.g. `-order_file` or call graph profile
sorting. ld64 always keeps data-in-code information sorted.
This commit also fixes an incorrect assertion. The original assertion
from D103006 used to check that data-in-code entries are sorted in the
input objects -- likely because we use binary search on that data. In
D115556, the assertion was moved into `collectDataInCodeEntries`, but
the checked variable's name was not changed, so it ended up checking the
final contents of the DataInCodeSection.
We no longer crash when building LLVM with PGO using an asserts build of
LLD as the linker.
Fixes https://bugs.chromium.org/p/chromium/issues/detail?id=1265937
Numbers for linking the Chromium Framework reproducer from #48001, which
has 6829 data-in-code entries:
x before
+ after
N Min Max Median Avg Stddev
x 20 2.1076453 2.3059683 2.1132485 2.1350302 0.049905767
+ 20 2.1069031 2.3915262 2.14465 2.1728429 0.084065898
No difference proven at 95.0% confidence
Differential Revision: https://reviews.llvm.org/D133581
I encountered an issue where `p &variable` was finding an incorrect address for
32-bit PIC ELF files loaded into a running process. The problem was that the
R_386_32 ELF relocations were not being applied to the DWARF section, so all
variables in that file were reporting as being at the start of their respective
section. There is an assert that catches this on debug builds, but silently
ignores the issue on non-debug builds.
In this changeset, I added handling for the R_386_32 relocation type to
ObjectFileELF, and a supporting function to ELFRelocation to differentiate
between DT_REL & DT_RELA in ObjectFileELF::ApplyRelocations().
Demonstration of issue:
```
[dmlary@host work]$ cat rel.c
volatile char padding[32] = "make sure var isnt at .data+0";
volatile char var[] = "test";
[dmlary@host work]$ gcc -c rel.c -FPIC -fpic -g -m32
[dmlary@host work]$ lldb ./exec
(lldb) target create "./exec"
Current executable set to '/home/dmlary/src/work/exec' (i386).
(lldb) process launch --stop-at-entry
Process 21278 stopped
* thread #1, name = 'exec', stop reason = signal SIGSTOP
frame #0: 0xf7fdb150 ld-2.17.so`_start
ld-2.17.so`_start:
-> 0xf7fdb150 <+0>: movl %esp, %eax
0xf7fdb152 <+2>: calll 0xf7fdb990 ; _dl_start
ld-2.17.so`_dl_start_user:
0xf7fdb157 <+0>: movl %eax, %edi
0xf7fdb159 <+2>: calll 0xf7fdb140
Process 21278 launched: '/home/dmlary/src/work/exec' (i386)
(lldb) image add ./rel.o
(lldb) image load --file rel.o .text 0x40000000 .data 0x50000000
section '.text' loaded at 0x40000000
section '.data' loaded at 0x50000000
(lldb) image dump symtab rel.o
Symtab, file = rel.o, num_symbols = 13:
Debug symbol
|Synthetic symbol
||Externally Visible
|||
Index UserID DSX Type File Address/Value Load Address Size Flags Name
------- ------ --- --------------- ------------------ ------------------ ------------------ ---------- ----------------------------------
[ 0] 1 SourceFile 0x0000000000000000 0x0000000000000000 0x00000004 rel.c
[ 1] 2 Invalid 0x0000000000000000 0x0000000000000020 0x00000003
[ 2] 3 Invalid 0x0000000000000000 0x50000000 0x0000000000000020 0x00000003
[ 3] 4 Invalid 0x0000000000000025 0x0000000000000000 0x00000003
[ 4] 5 Invalid 0x0000000000000000 0x0000000000000020 0x00000003
[ 5] 6 Invalid 0x0000000000000000 0x0000000000000020 0x00000003
[ 6] 7 Invalid 0x0000000000000000 0x0000000000000020 0x00000003
[ 7] 8 Invalid 0x0000000000000000 0x0000000000000020 0x00000003
[ 8] 9 Invalid 0x0000000000000000 0x0000000000000020 0x00000003
[ 9] 10 Invalid 0x0000000000000000 0x0000000000000020 0x00000003
[ 10] 11 Invalid 0x0000000000000000 0x0000000000000020 0x00000003
[ 11] 12 X Data 0x0000000000000000 0x50000000 0x0000000000000020 0x00000011 padding
[ 12] 13 X Data 0x0000000000000020 0x50000020 0x0000000000000005 0x00000011 var
(lldb) p &var
(volatile char (*)[5]) $1 = 0x50000000
```
Reviewed By: labath
Differential Revision: https://reviews.llvm.org/D132954
D129370 started hoisting allocas across stacksave/stackrestore
boundaries which is wrong.
Reviewed By: chill, rnk
Differential Revision: https://reviews.llvm.org/D133730
This is the ultimate fallback code if UADDO isn't supported.
If the target uses 0/1 we used one compare, but if the target doesn't
use 0/1 we emitted two compares. Regardless of boolean constants we
should only need to check that the Result is less than one of the
original operands. So we only need one compare.
Reviewed By: spatel
Differential Revision: https://reviews.llvm.org/D133708
Introduce the additional "transform-dialect-print-top-level-after-all" debug
category that allows the user to print the paylaod IR after each transformation
performed by the transform dialect. This is useful for understanding and
debugging the effects of individual transformations in complex transformations
scripts, including in downstreams, without having to modify the transformation
script itself.
Reviewed By: mravishankar
Differential Revision: https://reviews.llvm.org/D133775
As the Zbs extension includes bext[i] for bit extract, we can
unconditionally return true from this hook. This hook causes the DAG
combiner to perform the following canonicalisation:
and (not (srl X, C)), 1 --> (and X, 1<<C) == 0
and (srl (not X), C)), 1 --> (and X, 1<<C) == 0
As simply changing the hook causes a codegen regression, this patch also
modifies a BEXTI pattern to match this canonicalised form.
As BSETINVMask is now used for BEXT as well as BSET and BINV, it has
been renamed to the more generic SingleBitSetMask.
There is one codegen change in bittest.ll for bittest_31_i64 (NOT+BEXTI
rather than NOT+SRLIW). This is neutral in terms of code quality.
Differential Revision: https://reviews.llvm.org/D131482
We use the saturating behavior of fcvt.wu.h/s/d but forgot to
take into account that fcvt.wu will sign extend the saturated
result. According to computeKnownBits a promoted FP_TO_UINT_SAT
is expected to zero extend the saturated value.
In many case the upper bits aren't be demanded so this wouldn't
be an issue. But if we computeKnownBits caused an AND to be removed
it would be a bug.
This patch inserts an AND during to zero the upper bits.
Unfortunately, this pessimizes code if we aren't able to tell if
the upper bits are demanded. To fix that we could custom type
promote the FP_TO_UINT_SAT with SEXT_INREG after it, but I'll
leave that for future work.
I haven't found a failure from this, I was revisiting the code to
add vector support and spotted it.
Differential Revision: https://reviews.llvm.org/D133746
This revision significantly improves and tests the broadcast behavior of vector.warp_execute_on_lane_0.
Previously, the implementation of the broadcast behavior of vector.warp_execute_on_lane_0
assumed that the broadcasted value was always of scalar type.
This is not necessarily the case.
Differential Revision: https://reviews.llvm.org/D133767
The original proposal was seen in Apr 2019 and we accidentally used
that date (201904L) as the feature testing value. However, WG14 N2408
was adopted at the Oct 2019 meeting and so that's the correct date for
the feature testing macro. The committee draft for C2x shows 201910L
for this value, so this changes brings us in line with the standard.
Include the transform op being applied when reporting it using an invalidated
handle. This was missing previously and made it harder for the user to
understand where the handle is being used, especially if the transform script
included some sort of iteration.
Reviewed By: guraypp
Differential Revision: https://reviews.llvm.org/D133774
If the Shuffle is a splat and the operand is a zext/sext, sinking the
operand and the s/zext can help create indexed s/umull. This is
especially useful to prevent i64 mul being scalarized.
Differential Revision: https://reviews.llvm.org/D133355
This tool is only used if `MLIR_ENABLE_SPIRV_CPU_RUNNER` was specified and should only be substituted in that case. Alternatively, we could make the substitution optional.
Reviewed By: mehdi_amini
Differential Revision: https://reviews.llvm.org/D133752
The original proposal was adopted in Apr 2019, but was subsequently
updated by WG14 N2662 in June 2021. We already supported the attribute
on a label and it behaved as expected, but we had not bumped the
feature test value.
The original proposal was adopted in Apr 2019 and so the previous value
was 201904L. However, a subsequent proposal (N2448) was adopted to add
an optional message argument to the attribute. We already support that
functionality, but had not bumped the feature test value.
The class set a SmallVector stack allocation size to 64 elements which
is uncommonly large. These structures are then used extensively and
copied often in functions which led to stack frame sizes considered
excessively large for some use cases.
Differential Revision: https://reviews.llvm.org/D133761
If one of the operands is negated in a multiplication we can optimise the operation by moving the negation to the smallest operand or to the result
Reviewed By: fhahn
Differential Revision: https://reviews.llvm.org/D133287
This reverts commit d200db3863, which causes a
clang crash. See https://reviews.llvm.org/D111283#3785755
Test case for convenience:
```
template <typename T>
using P = int T::*;
template <typename T, typename... A>
void j(P<T>, T, A...);
template <typename T>
void j(P<T>, T);
struct S {
int b;
};
void g(P<S> k, S s) { j(k, s); }
```
This extends 4658366d95 to add a note
explaining why the register is reserved.
note: x13 is clobbered by asynchronous signals when using Arm64EC.
I've added testing for w/x registers and v/q/s/d and h floating point
registers.
llvm will accept, but silently do nothing with, b registers. So they
are not tested here (clang rejects them so at least for C you're safe anyway).
Reviewed By: efriedma
Differential Revision: https://reviews.llvm.org/D133701
Jessica Paquette