platform/upstream/llvm.git
23 months ago[ELF] Simplify getRankProximity. NFC
Fangrui Song [Sat, 30 Jul 2022 23:32:42 +0000 (16:32 -0700)]
[ELF] Simplify getRankProximity. NFC

23 months ago[gn build] Port 88181375a3db more
Nico Weber [Sat, 30 Jul 2022 23:30:53 +0000 (19:30 -0400)]
[gn build] Port 88181375a3db more

23 months ago[MLIR] Add a utility to sort the operands of commutative ops
srishti-cb [Sat, 30 Jul 2022 16:53:15 +0000 (12:53 -0400)]
[MLIR] Add a utility to sort the operands of commutative ops

Added a commutativity utility pattern and a function to populate it. The pattern sorts the operands of an op in ascending order of the "key" associated with each operand iff the op is commutative. This sorting is stable.

The function is intended to be used inside passes to simplify the matching of commutative operations. After the application of the above-mentioned pattern, since the commutative operands now have a deterministic order in which they occur in an op, the matching of large DAGs becomes much simpler, i.e., requires much less number of checks to be written by a user in her/his pattern matching function.

The "key" associated with an operand is the list of the "AncestorKeys" associated with the ancestors of this operand, in a breadth-first order.

The operand of any op is produced by a set of ops and block arguments. Each of these ops and block arguments is called an "ancestor" of this operand.

Now, the "AncestorKey" associated with:
1. A block argument is `{type: BLOCK_ARGUMENT, opName: ""}`.
2. A non-constant-like op, for example, `arith.addi`, is `{type: NON_CONSTANT_OP, opName: "arith.addi"}`.
3. A constant-like op, for example, `arith.constant`, is `{type: CONSTANT_OP, opName: "arith.constant"}`.

So, if an operand, say `A`, was produced as follows:

```
`<block argument>`  `<block argument>`
             \          /
              \        /
              `arith.subi`           `arith.constant`
                         \            /
                         `arith.addi`
                                |
                           returns `A`
```

Then, the block arguments and operations present in the backward slice of `A`, in the breadth-first order are:
`arith.addi`, `arith.subi`, `arith.constant`, `<block argument>`, and `<block argument>`.

Thus, the "key" associated with operand `A` is:
```
{
 {type: NON_CONSTANT_OP, opName: "arith.addi"},
 {type: NON_CONSTANT_OP, opName: "arith.subi"},
 {type: CONSTANT_OP, opName: "arith.constant"},
 {type: BLOCK_ARGUMENT, opName: ""},
 {type: BLOCK_ARGUMENT, opName: ""}
}
```

Now, if "keyA" is the key associated with operand `A` and "keyB" is the key associated with operand `B`, then:
"keyA" < "keyB" iff:
1. In the first unequal pair of corresponding AncestorKeys, the AncestorKey in operand `A` is smaller, or,
2. Both the AncestorKeys in every pair are the same and the size of operand `A`'s "key" is smaller.

AncestorKeys of type `BLOCK_ARGUMENT` are considered the smallest, those of type `CONSTANT_OP`, the largest, and `NON_CONSTANT_OP` types come in between. Within the types `NON_CONSTANT_OP` and `CONSTANT_OP`, the smaller ones are the ones with smaller op names (lexicographically).

---

Some examples of such a sorting:

Assume that the sorting is being applied to `foo.commutative`, which is a commutative op.

Example 1:

> %1 = foo.const 0
> %2 = foo.mul <block argument>, <block argument>
> %3 = foo.commutative %1, %2

Here,
1. The key associated with %1 is:
```
    {
     {CONSTANT_OP, "foo.const"}
    }
```
2. The key associated with %2 is:
```
    {
     {NON_CONSTANT_OP, "foo.mul"},
     {BLOCK_ARGUMENT, ""},
     {BLOCK_ARGUMENT, ""}
    }
```

The key of %2 < the key of %1
Thus, the sorted `foo.commutative` is:
> %3 = foo.commutative %2, %1

Example 2:

> %1 = foo.const 0
> %2 = foo.mul <block argument>, <block argument>
> %3 = foo.mul %2, %1
> %4 = foo.add %2, %1
> %5 = foo.commutative %1, %2, %3, %4

Here,
1. The key associated with %1 is:
```
    {
     {CONSTANT_OP, "foo.const"}
    }
```
2. The key associated with %2 is:
```
    {
     {NON_CONSTANT_OP, "foo.mul"},
     {BLOCK_ARGUMENT, ""}
    }
```
3. The key associated with %3 is:
```
    {
     {NON_CONSTANT_OP, "foo.mul"},
     {NON_CONSTANT_OP, "foo.mul"},
     {CONSTANT_OP, "foo.const"},
     {BLOCK_ARGUMENT, ""},
     {BLOCK_ARGUMENT, ""}
    }
```
4. The key associated with %4 is:
```
    {
     {NON_CONSTANT_OP, "foo.add"},
     {NON_CONSTANT_OP, "foo.mul"},
     {CONSTANT_OP, "foo.const"},
     {BLOCK_ARGUMENT, ""},
     {BLOCK_ARGUMENT, ""}
    }
```

Thus, the sorted `foo.commutative` is:
> %5 = foo.commutative %4, %3, %2, %1

Signed-off-by: Srishti Srivastava <srishti.srivastava@polymagelabs.com>
Reviewed By: Mogball

Differential Revision: https://reviews.llvm.org/D124750

23 months ago[JITLink][COFF] Remove obsolete FIXMEs. (NFC)
Sunho Kim [Sat, 30 Jul 2022 23:10:14 +0000 (08:10 +0900)]
[JITLink][COFF] Remove obsolete FIXMEs. (NFC)

23 months ago[JITLInk][COFF] Remove unnecessary unique_ptr. (NFC)
Sunho Kim [Sat, 30 Jul 2022 23:08:19 +0000 (08:08 +0900)]
[JITLInk][COFF] Remove unnecessary unique_ptr. (NFC)

23 months ago[JITLink][COFF] Add explicit std::move.
Sunho Kim [Sat, 30 Jul 2022 23:00:19 +0000 (08:00 +0900)]
[JITLink][COFF] Add explicit std::move.

Since ArgList is not copyable we need to make sure it's moved explicitly.

23 months ago[gn build] Port 88181375a3db
Nico Weber [Sat, 30 Jul 2022 22:58:10 +0000 (18:58 -0400)]
[gn build] Port 88181375a3db

23 months ago[JITLink][COFF] Implement include/alternatename linker directive.
Sunho Kim [Sat, 30 Jul 2022 22:49:37 +0000 (07:49 +0900)]
[JITLink][COFF] Implement include/alternatename linker directive.

Implements include/alternatename linker directive. Alternatename is used by static msvc runtime library. Alias symbol is technically incorrect (we have to search for external definition) but we don't have a way to represent this in jitlink/orc yet, this is solved in the following up patch.

Inlcude linker directive is used in ucrt to forcelly lookup the static initializer symbols so that they will be emitted. It's implemented as extenral symbols with live flag on that cause the lookup of these symbols.

Reviewed By: lhames

Differential Revision: https://reviews.llvm.org/D130276

23 months ago[clang-repl] Disable building when LLVM_STATIC_LINK_CXX_STDLIB is ON.
Sunho Kim [Sat, 30 Jul 2022 20:42:16 +0000 (05:42 +0900)]
[clang-repl] Disable building when LLVM_STATIC_LINK_CXX_STDLIB is ON.

We have seen random symbol not found "__cxa_throw" error in fuschia build bots and out-of-tree users. The understanding have been that they are built without exception support, but it turned out that these platforms have LLVM_STATIC_LINK_CXX_STDLIB ON so that they link libstdc++ to llvm statically. The reason why this is problematic for clang-repl is that by default clang-repl tries to find symbols from symbol table of executable and dynamic libraries loaded by current process. It needs to load another libstdc++, but the platform that had LLVM_STATIC_LINK_CXX_STDLIB turned on is usally those with missing or obsolate shared libstdc++ in the first place -- trying to load it again would be destined to fail eventually with a risk to introuduce mixed libstdc++ versions.

A proper solution that doesn't take a workaround is statically link the same libstdc++ by clang-repl side, but this is not possible with old JIT linker runtimedyld. New just-in-time linker JITLink handles this relatively well, but it's not availalbe in majority of platforms. For now, this patch just disables the building of clang-repl when LLVM_STATIC_LINK_CXX_STDLIB is ON and removes the "__cxa_throw" check in exception unittest as well as reverting previous exception check flag patch.

Reviewed By: v.g.vassilev

Differential Revision: https://reviews.llvm.org/D130788

23 months ago[Libomptarget] Disable testing map_back_race.cpp
Joseph Huber [Fri, 29 Jul 2022 17:51:15 +0000 (13:51 -0400)]
[Libomptarget] Disable testing map_back_race.cpp

This test hasn't been fixed and causes spurious failures when testing.
This patch sets it as unsupported until we have a reliable fix.

Reviewed By: ronlieb

Differential Revision: https://reviews.llvm.org/D130789

23 months ago[RISCV] Remove vmerges from vector ceil, floor, trunc lowering.
Craig Topper [Sat, 30 Jul 2022 17:33:52 +0000 (10:33 -0700)]
[RISCV] Remove vmerges from vector ceil, floor, trunc lowering.

Use masked operations to suppress spurious exception bits being
set in fflags. Unfortunately, doing this adds extra copies.

23 months ago[X86] Use std::tie so we can have more meaningful variable names for demanded bits...
Simon Pilgrim [Sat, 30 Jul 2022 17:57:11 +0000 (18:57 +0100)]
[X86] Use std::tie so we can have more meaningful variable names for demanded bits/elts pairs. NFCI.

*.first + *.second were proving difficult to keep track of.

23 months ago[ELF] Support missing relocations in RISCV::getImplicitAddend()
Fangrui Song [Sat, 30 Jul 2022 17:55:59 +0000 (10:55 -0700)]
[ELF] Support missing relocations in RISCV::getImplicitAddend()

23 months ago[ELF] Support missing relocations in PPC64::getImplicitAddend()
Fangrui Song [Sat, 30 Jul 2022 17:46:58 +0000 (10:46 -0700)]
[ELF] Support missing relocations in PPC64::getImplicitAddend()

23 months agoRevert "[BOLT] Ignore functions accessing false positive jump tables"
Amir Ayupov [Sat, 30 Jul 2022 17:35:10 +0000 (10:35 -0700)]
Revert "[BOLT] Ignore functions accessing false positive jump tables"

This diff uncovers an ASAN leak in getOrCreateJumpTable:
```
Indirect leak of 264 byte(s) in 1 object(s) allocated from:
    #1 0x4f6e48c in llvm::bolt::BinaryContext::getOrCreateJumpTable ...
```
The removal of an assertion needs to be accompanied by proper deallocation of
a `JumpTable` object for which `analyzeJumpTable` was unsuccessful.

This reverts commit 52cd00cabf479aa7eb6dbb063b7ba41ea57bce9e.

23 months agoUse any_of (NFC)
Kazu Hirata [Sat, 30 Jul 2022 17:35:56 +0000 (10:35 -0700)]
Use any_of (NFC)

23 months agoUse is_contained (NFC)
Kazu Hirata [Sat, 30 Jul 2022 17:35:54 +0000 (10:35 -0700)]
Use is_contained (NFC)

23 months agoConvert for_each to range-based for loops (NFC)
Kazu Hirata [Sat, 30 Jul 2022 17:35:52 +0000 (10:35 -0700)]
Convert for_each to range-based for loops (NFC)

23 months ago[bolt] Remove redundaunt virtual specifiers (NFC)
Kazu Hirata [Sat, 30 Jul 2022 17:35:51 +0000 (10:35 -0700)]
[bolt] Remove redundaunt virtual specifiers (NFC)

Identified with modernize-use-override.

23 months ago[bolt] Remove redundaunt control-flow statements (NFC)
Kazu Hirata [Sat, 30 Jul 2022 17:35:49 +0000 (10:35 -0700)]
[bolt] Remove redundaunt control-flow statements (NFC)

Identified with readability-redundant-control-flow.

23 months agoUse nullptr instead of 0 (NFC)
Kazu Hirata [Sat, 30 Jul 2022 17:35:48 +0000 (10:35 -0700)]
Use nullptr instead of 0 (NFC)

Identified with modernize-use-nullptr.

23 months ago[lld] Remove unused forward declarations (NFC)
Kazu Hirata [Sat, 30 Jul 2022 17:35:46 +0000 (10:35 -0700)]
[lld] Remove unused forward declarations (NFC)

23 months ago[IR] Fix a header guard (NFC)
Kazu Hirata [Sat, 30 Jul 2022 17:35:45 +0000 (10:35 -0700)]
[IR] Fix a header guard (NFC)

Identified with llvm-header-guard.

23 months ago[ExecutionEngine] Ensure a newline at the end of a file (NFC)
Kazu Hirata [Sat, 30 Jul 2022 17:35:43 +0000 (10:35 -0700)]
[ExecutionEngine] Ensure a newline at the end of a file (NFC)

23 months ago[RISCV] Add merge operands to more RISCVISD::*_VL opcodes.
Craig Topper [Sat, 30 Jul 2022 16:58:31 +0000 (09:58 -0700)]
[RISCV] Add merge operands to more RISCVISD::*_VL opcodes.

This adds a merge operand to all of the binary _VL nodes. Including
integer and widening. They all share multiclasses in tablegen
so doing them all at once was easiest.

I plan to use FADD_VL in an upcoming patch. The rest are just for
consistency to keep tablegen working.

This does reduce the isel table size by about 25k so that's nice.

Reviewed By: reames

Differential Revision: https://reviews.llvm.org/D130816

23 months ago[RISCV] Swap the merge and mask operand order for VRGATHER*_VL and FCOPYSIGN_VL nodes.
Craig Topper [Sat, 30 Jul 2022 16:56:16 +0000 (09:56 -0700)]
[RISCV] Swap the merge and mask operand order for VRGATHER*_VL and FCOPYSIGN_VL nodes.

Based on review feedback from D130816.

23 months ago[X86] computeKnownBitsForTargetNode - out of range X86ISD::VSRAI doesn't fold to...
Simon Pilgrim [Sat, 30 Jul 2022 16:55:34 +0000 (17:55 +0100)]
[X86] computeKnownBitsForTargetNode - out of range X86ISD::VSRAI doesn't fold to zero

Noticed by inspection and I can't seem to make a test case, but SSE arithmetic bit shifts clamp to the max shift amount (i.e. create a sign splat) - combineVectorShiftImm already does something similar.

23 months ago[CodeGen] Fixed undeclared MISchedCutoff in case of NDEBUG and LLVM_ENABLE_ABI_BREAKI...
Dmitry Vassiliev [Sat, 30 Jul 2022 16:24:50 +0000 (18:24 +0200)]
[CodeGen] Fixed undeclared MISchedCutoff in case of NDEBUG and LLVM_ENABLE_ABI_BREAKING_CHECKS

This patch fixes the error llvm/lib/CodeGen/MachineScheduler.cpp(755): error C2065: 'MISchedCutoff': undeclared identifier in case of NDEBUG and LLVM_ENABLE_ABI_BREAKING_CHECKS.
Note MISchedCutoff is declared under #ifndef NDEBUG.

Reviewed By: RKSimon

Differential Revision: https://reviews.llvm.org/D130425

23 months ago[InstCombine] canonicalize more zext-and-of-bool compare to narrow and
Sanjay Patel [Sat, 30 Jul 2022 15:20:13 +0000 (11:20 -0400)]
[InstCombine] canonicalize more zext-and-of-bool compare to narrow and

https://alive2.llvm.org/ce/z/vBNiiM

This matches variants of patterns that were folded with:
b5a9361c90ca

23 months ago[AMDGPU] Remove unused function
Austin Kerbow [Sat, 30 Jul 2022 14:41:46 +0000 (07:41 -0700)]
[AMDGPU] Remove unused function

23 months agoOMPIRBuilder.h - fix Wdocumentation warning. NFC.
Simon Pilgrim [Sat, 30 Jul 2022 14:43:41 +0000 (15:43 +0100)]
OMPIRBuilder.h - fix Wdocumentation warning. NFC.

23 months agoFix Wdocumentation warning. NFC.
Simon Pilgrim [Sat, 30 Jul 2022 14:41:13 +0000 (15:41 +0100)]
Fix Wdocumentation warning. NFC.

warning: '\returns' command used in a comment that is attached to a function returning void

23 months ago[X86] combineSelect fold 'smin' style pattern select(pcmpgt(RHS, LHS), LHS, RHS)...
Simon Pilgrim [Sat, 30 Jul 2022 14:31:31 +0000 (15:31 +0100)]
[X86] combineSelect fold 'smin' style pattern select(pcmpgt(RHS, LHS), LHS, RHS) -> select(pcmpgt(LHS, RHS), RHS, LHS) if pcmpgt(LHS, RHS) already exists

Avoids repeated commuted comparisons when we're performing min/max and clamp patterns

23 months agoConstantFolding: fold OOB accesses to poison instead of undef
Nuno Lopes [Sat, 30 Jul 2022 14:20:32 +0000 (15:20 +0100)]
ConstantFolding: fold OOB accesses to poison instead of undef

23 months ago[InstCombine] avoid splitting a constant expression with div/rem fold
Sanjay Patel [Sat, 30 Jul 2022 13:43:09 +0000 (09:43 -0400)]
[InstCombine] avoid splitting a constant expression with div/rem fold

Follow-up to d4940c0f3d43 to further limit the transform
to avoid an unintended pattern/fold of a constant expression.

23 months ago[DAG] Pull out repeated getOperand() calls for shuffle ops. NFC.
Simon Pilgrim [Sat, 30 Jul 2022 12:59:47 +0000 (13:59 +0100)]
[DAG] Pull out repeated getOperand() calls for shuffle ops. NFC.

23 months ago[AMDGPU] Fix || vs && precedence warning. NFC.
Simon Pilgrim [Sat, 30 Jul 2022 11:19:21 +0000 (12:19 +0100)]
[AMDGPU] Fix || vs && precedence warning. NFC.

23 months ago[NFC] Switch a few uses of undef to poison as placeholders for unreachable code
Nuno Lopes [Sat, 30 Jul 2022 12:55:56 +0000 (13:55 +0100)]
[NFC] Switch a few uses of undef to poison as placeholders for unreachable code

23 months ago[X86] Add test cases for D129537
Luo, Yuanke [Sat, 30 Jul 2022 11:39:03 +0000 (19:39 +0800)]
[X86] Add test cases for D129537

23 months agoSelectionDAGNodes.h - fix Wdocumentation warnings. NFC.
Simon Pilgrim [Sat, 30 Jul 2022 10:05:33 +0000 (11:05 +0100)]
SelectionDAGNodes.h - fix Wdocumentation warnings. NFC.

23 months agoDependencyScanningTool.h - fix Wdocumentation warning. NFC.
Simon Pilgrim [Sat, 30 Jul 2022 10:00:35 +0000 (11:00 +0100)]
DependencyScanningTool.h - fix Wdocumentation warning. NFC.

23 months agoSema.h - fix Wdocumentation warnings. NFC.
Simon Pilgrim [Sat, 30 Jul 2022 10:00:16 +0000 (11:00 +0100)]
Sema.h - fix Wdocumentation warnings. NFC.

23 months ago[libc++][ranges] Implement `std::ranges::partial_sort_copy`.
Konstantin Varlamov [Sat, 30 Jul 2022 09:42:05 +0000 (02:42 -0700)]
[libc++][ranges] Implement `std::ranges::partial_sort_copy`.

Differential Revision: https://reviews.llvm.org/D130532

23 months ago[InstCombine] Add fold for redundant sign bits count comparison
Alexander Shaposhnikov [Sat, 30 Jul 2022 09:06:37 +0000 (09:06 +0000)]
[InstCombine] Add fold for redundant sign bits count comparison

For power-of-2 C:
((X s>> ShiftC) ^ X) u< C --> (X + C) u< (C << 1)
((X s>> ShiftC) ^ X) u> (C - 1) --> (X + C) u> ((C << 1) - 1)

(https://github.com/llvm/llvm-project/issues/56479)

Test plan:
0/ ninja check-llvm check-clang + bootstrap LLVM/Clang
1/ https://alive2.llvm.org/ce/z/eEUfx3

Differential revision: https://reviews.llvm.org/D130433

23 months ago[libc++] Fix reverse_iterator::iterator_concept
Nikolas Klauser [Sun, 24 Jul 2022 14:46:57 +0000 (16:46 +0200)]
[libc++] Fix reverse_iterator::iterator_concept

Fixes https://github.com/llvm/llvm-project/issues/56504

Reviewed By: ldionne, Mordante, huixie90, #libc

Spies: libcxx-commits, hewillk

Differential Revision: https://reviews.llvm.org/D129794

23 months ago[ELF] Support missing relocations in AArch64::getImplicitAddend()
Fangrui Song [Sat, 30 Jul 2022 08:24:55 +0000 (01:24 -0700)]
[ELF] Support missing relocations in AArch64::getImplicitAddend()

23 months ago[InstCombine] Add baseline tests for redundant sign bits count folds
Alexander Shaposhnikov [Sat, 30 Jul 2022 08:23:45 +0000 (08:23 +0000)]
[InstCombine] Add baseline tests for redundant sign bits count folds

Add baseline tests. NFC.
(https://github.com/llvm/llvm-project/issues/56479)

Test plan:
ninja check-all

Differential revision: https://reviews.llvm.org/D130605

23 months ago[ELF] Write R_AARCH64_IRELATIVE addends with -z rel
Fangrui Song [Sat, 30 Jul 2022 08:19:59 +0000 (01:19 -0700)]
[ELF] Write R_AARCH64_IRELATIVE addends with -z rel

23 months ago[NFC] Use more appropriate SmallVectorImpl::append call in std::initializer_list...
Dawid Jurczak [Sat, 30 Jul 2022 07:15:11 +0000 (09:15 +0200)]
[NFC] Use more appropriate SmallVectorImpl::append call in std::initializer_list SmallVector constructor

Since we are in constructor there is no need to perform redundant call to SmallVectorImpl::clear() inside assign function.
Although calling cheaper append function instead assign doesn't make any difference on optimized builds
(DSE does the job removing stores), we still save some cycles for debug binaries.

Differential Revision: https://reviews.llvm.org/D130361

23 months ago[X86][FP16] Regenerate test cases. NFC.
Phoebe Wang [Sat, 30 Jul 2022 05:13:21 +0000 (13:13 +0800)]
[X86][FP16] Regenerate test cases. NFC.

23 months ago[llvm] Use is_contained (NFC)
Kazu Hirata [Sat, 30 Jul 2022 04:18:44 +0000 (21:18 -0700)]
[llvm] Use is_contained (NFC)

23 months agoUse is_sorted (NFC)
Kazu Hirata [Sat, 30 Jul 2022 04:18:42 +0000 (21:18 -0700)]
Use is_sorted (NFC)

23 months agoUse value instead of getValue (NFC)
Kazu Hirata [Sat, 30 Jul 2022 04:18:41 +0000 (21:18 -0700)]
Use value instead of getValue (NFC)

23 months ago[clang] Use has_value instead of value (NFC)
Kazu Hirata [Sat, 30 Jul 2022 04:18:39 +0000 (21:18 -0700)]
[clang] Use has_value instead of value (NFC)

23 months ago[LLDB][RISCV] Add DWARF Registers
Emmmer [Thu, 28 Jul 2022 07:38:33 +0000 (15:38 +0800)]
[LLDB][RISCV] Add DWARF Registers

According to [RISC-V DWARF Specification](https://github.com/riscv-non-isa/riscv-elf-psabi-doc/blob/master/riscv-dwarf.adoc) add RISCV DWARF Registers.

Don't worry about the difference between riscv32 and riscv64, they just have different bytes of registers.

Reviewed By: DavidSpickett

Differential Revision: https://reviews.llvm.org/D130686

23 months ago[libc] Compile integration tests with -ffreestanding to avoid mixup with system libc.
Siva Chandra Reddy [Sat, 30 Jul 2022 01:37:55 +0000 (01:37 +0000)]
[libc] Compile integration tests with -ffreestanding to avoid mixup with system libc.

23 months agoPreserve qualifiers when getting fully qualified type
Weverything [Sat, 30 Jul 2022 02:17:20 +0000 (19:17 -0700)]
Preserve qualifiers when getting fully qualified type

15f3cd6bfc670ba6106184a903eb04be059e5977 moved the handling of UsingType
to a later point in the function getFullyQualifiedType.  This moved it
after the removal of an ElaboratedType and its qualifiers.  However,
the qualifiers were not added back, causing the fully qualified type to
have a qualifier mismatch with the original type.  Make sure the
qualifers are added before continuing to fully qualify the type.

23 months ago[AMDGPU] Extend SILoadStoreOptimizer to s_load instructions
Carl Ritson [Sat, 30 Jul 2022 02:13:20 +0000 (11:13 +0900)]
[AMDGPU] Extend SILoadStoreOptimizer to s_load instructions

Apply merging to s_load as is done for s_buffer_load.

Reviewed By: foad

Differential Revision: https://reviews.llvm.org/D130742

23 months ago[InstCombine] Refactor foldICmpMulConstant
Alexander Shaposhnikov [Sat, 30 Jul 2022 01:43:25 +0000 (01:43 +0000)]
[InstCombine] Refactor foldICmpMulConstant

This is a follow-up to 2ebfda2417
(replace "if" with "else if" since the cases nuw/nsw
were meant to be handled separately).

Test plan:
1/ ninja check-llvm check-clang check-lld
2/ Bootstrapped LLVM/Clang pass tests

23 months ago[Clang] Fix handling of Max from getValueRange(...) in IntExprEvaluator::VisitCastExp...
Shafik Yaghmour [Sat, 30 Jul 2022 02:17:42 +0000 (19:17 -0700)]
[Clang] Fix handling of Max from getValueRange(...) in IntExprEvaluator::VisitCastExpr(...)

This is a follow-up to D130058 to fix how we handle the Max value we obtain from
getValueRange(...) in IntExprEvaluator::VisitCastExpr(...) which in the case of
an enum that contains an enumerator with the max integer value will overflow by
one.

The fix is to decrement the value of Max and use slt and ult for comparison Vs
sle and ule.`

Differential Revision: https://reviews.llvm.org/D130811

23 months agoFix lack of cc1 flag in llvmcmd sections when assertions are enabled
Aiden Grossman [Sat, 30 Jul 2022 01:38:34 +0000 (18:38 -0700)]
Fix lack of cc1 flag in llvmcmd sections when assertions are enabled

Currently when assertions are enabled, the cc1 flag is not
inserted into the llvmcmd section of object files with embedded
bitcode. This deviates from the normal behavior where this is
the first flag that is inserted. This error stems from incorrect
use of the function generateCC1CommandLine() which requires
manually adding in the -cc1 flag which is currently not done.

Reviewed By: jansvoboda11

Differential Revision: https://reviews.llvm.org/D130620

23 months ago[MLIR] Fix getCommonBlock utility in affine analysis
Uday Bondhugula [Sat, 30 Jul 2022 01:39:02 +0000 (07:09 +0530)]
[MLIR] Fix getCommonBlock utility in affine analysis

Fix the hardcoded check for `FuncOp` in `getCommonBlock` utility: the
check should have been for an op that starts an affine scope. The
incorrect block returned in turn causes dependence analysis to function
incorrectly.

This change allows affine store-load forwarding to work correctly inside
any ops that start an affine scope.

Reviewed By: ftynse, dcaballe

Differential Revision: https://reviews.llvm.org/D130749

23 months agoRevert "[compiler-rt] [builtins] Detect _Float16 support at compile time"
Phoebe Wang [Sat, 30 Jul 2022 01:41:33 +0000 (09:41 +0800)]
Revert "[compiler-rt] [builtins] Detect _Float16 support at compile time"

This reverts commit ce6d40f5c23923a807388c58b82b4c343eced0ce.

Revert due to it broke the sanitizer-windows buildbot: https://lab.llvm.org/buildbot/#/builders/127/builds/33583/steps/4/logs/stdio

23 months ago[libc] Implement cosf function that is correctly rounded to all rounding modes.
Tue Ly [Wed, 27 Jul 2022 16:22:27 +0000 (12:22 -0400)]
[libc] Implement cosf function that is correctly rounded to all rounding modes.

Implement cosf function that is correctly rounded to all rounding
modes.

Performance benchmark using perf tool from CORE-MATH project

(https://gitlab.inria.fr/core-math/core-math/-/tree/master) on Ryzen 1700:
Before this patch (not correctly rounded):
```
$ CORE_MATH_PERF_MODE="rdtsc" ./perf.sh cosf
CORE-MATH reciprocal throughput   : 19.043
System LIBC reciprocal throughput : 26.328
LIBC reciprocal throughput        : 30.955

$ CORE_MATH_PERF_MODE="rdtsc" ./perf.sh cosf --latency
GNU libc version: 2.31
GNU libc release: stable
CORE-MATH latency   : 49.995
System LIBC latency : 59.286
LIBC latency        : 60.174

```
After this patch (correctly rounded):
```
$ CORE_MATH_PERF_MODE="rdtsc" ./perf.sh cosf
GNU libc version: 2.31
GNU libc release: stable
CORE-MATH reciprocal throughput   : 19.072
System LIBC reciprocal throughput : 26.286
LIBC reciprocal throughput        : 13.631

$ CORE_MATH_PERF_MODE="rdtsc" ./perf.sh cosf --latency
GNU libc version: 2.31
GNU libc release: stable
CORE-MATH latency   : 49.872
System LIBC latency : 59.468
LIBC latency        : 56.119
```

Reviewed By: orex, zimmermann6

Differential Revision: https://reviews.llvm.org/D130644

23 months ago[flang] Add semantics test for EVENT WAIT statement
Naje George [Mon, 25 Jul 2022 23:07:07 +0000 (16:07 -0700)]
[flang] Add semantics test for EVENT WAIT statement

Reviewed By: ktras

Differential Revision: https://reviews.llvm.org/D130608

23 months ago[ELF] Move addDependentLibrary/handleSectionGroup. NFC
Fangrui Song [Sat, 30 Jul 2022 00:07:09 +0000 (17:07 -0700)]
[ELF] Move addDependentLibrary/handleSectionGroup. NFC

To reduce diff for my upcoming parallel input section initialization patch.

23 months ago[mlir][sparse]Add more test cases for sparse reshape
Anlun Xu [Fri, 29 Jul 2022 23:40:47 +0000 (16:40 -0700)]
[mlir][sparse]Add more test cases for sparse reshape

Add more test cases for sparse reshape

Reviewed By: Peiming

Differential Revision: https://reviews.llvm.org/D130603

23 months ago[clang-repl] Add missing link component.
Sunho Kim [Fri, 29 Jul 2022 22:59:43 +0000 (07:59 +0900)]
[clang-repl] Add missing link component.

OrcJIT was missing in LLVM_LINK_COMPONENTS.

23 months agoUpdate references to mailing lists that have moved to Discourse.
tlattner [Fri, 29 Jul 2022 22:55:14 +0000 (15:55 -0700)]
Update references to mailing lists that have moved to Discourse.

23 months agoRevert D130458 "[llvm-objcopy] Support --{,de}compress-debug-sections for zstd"
Fangrui Song [Fri, 29 Jul 2022 22:46:51 +0000 (15:46 -0700)]
Revert D130458 "[llvm-objcopy] Support --{,de}compress-debug-sections for zstd"

This reverts commit c26dc2904b95b3685d883e760e84046ea6c33d7f.

The new Zstd dispatch has an ongoing design discussion related to https://reviews.llvm.org/D130516#3688123 .
Revert for now before it is resolved.

23 months ago[ASTWriter] Provide capability to output a PCM/PCH file that does not write out infor...
Argyrios Kyrtzidis [Thu, 28 Jul 2022 07:48:56 +0000 (00:48 -0700)]
[ASTWriter] Provide capability to output a PCM/PCH file that does not write out information about its output path

This is useful to enable sharing of the same PCH file even when it's intended for a different output path.

The only information this option disables writing is for `ORIGINAL_PCH_DIR` record which is treated as optional and (when present) used as fallback for resolving input file paths relative to it.

Differential Revision: https://reviews.llvm.org/D130710

23 months ago[clang-repl] Disable exectuion unitests on unsupported platform by lljit instance...
Sunho Kim [Fri, 29 Jul 2022 22:17:05 +0000 (07:17 +0900)]
[clang-repl] Disable exectuion unitests on unsupported platform by lljit instance test.

The method used in 4191d661c74622c6fa72c1643e4567f45e6c9e1b was fragile because it didn't consider cross-platform builds and rely on enlisting unsupported targets. Uses the host-supports-jit mechanism to make an escape path. This should fix buildbot failures happening in upstream as well as out-of-tree.

23 months ago[RISCV] Precommit test for D123265
Luís Marques [Fri, 29 Jul 2022 21:56:26 +0000 (23:56 +0200)]
[RISCV] Precommit test for D123265

Differential Revision: https://reviews.llvm.org/D128562

23 months agoAdd BAT testing framework
Rafael Auler [Wed, 6 Jul 2022 22:53:27 +0000 (15:53 -0700)]
Add BAT testing framework

This patch refactors BAT to be testable as a library, so we
can have open-source tests on it. This further fixes an issue with
basic blocks that lack a valid input offset, making BAT omit those
when writing translation tables.

Test Plan: new testcases added, new testing tool added (llvm-bat-dump)

Differential Revision: https://reviews.llvm.org/D129382

23 months ago[InstCombine] fix miscompile from urem/udiv transform with constant expression
Sanjay Patel [Fri, 29 Jul 2022 21:02:47 +0000 (17:02 -0400)]
[InstCombine] fix miscompile from urem/udiv transform with constant expression

The isa<Constant> check could misfire on an instruction with 2 constant
operands. This bug was introduced with bb789381fc11cce (D36988).

See issue #56810 for a C source example that exposed the bug.

23 months ago[RISCV] Add isel pattern for (setne/eq GPR, -2048)
Craig Topper [Fri, 29 Jul 2022 21:01:14 +0000 (14:01 -0700)]
[RISCV] Add isel pattern for (setne/eq GPR, -2048)

For constants in the range [-2047, 2048] we use addi. If the constant
is -2048 we can use xori. If we don't match this explicitly, we'll
emit an LI for the -2048 followed by an XOR.

23 months ago[lld][WebAssemby] Demote LazySymbols back to undefined symbols if they are not loaded
Sam Clegg [Thu, 28 Jul 2022 22:45:02 +0000 (15:45 -0700)]
[lld][WebAssemby] Demote LazySymbols back to undefined symbols if they are not loaded

A LazySymbol is one that lives in `.a` archive and gets pulled in by a
strong reference.  However, weak references to such symbols do not
result in them be loaded from the archive.  In this case we want to
treat such symbols at undefined rather then lazy, once symbols
resolution is complete.

This fixes a crash bug in the linker when weakly referenced symbol that
lives in an archive file is live at the end of the link.  In the case of
dynamic linking this is expected to turn into an import with (in the
case of a function symbol) a function index.

Differential Revision: https://reviews.llvm.org/D130736

23 months ago[lld][WebAsssembly] Convert more tests from IR to assembly. NFC
Sam Clegg [Thu, 28 Jul 2022 22:45:02 +0000 (15:45 -0700)]
[lld][WebAsssembly] Convert more tests from IR to assembly. NFC

Differential Revision: https://reviews.llvm.org/D130738

23 months ago[InstCombine] add tests for udiv/urem miscompile; NFC
Sanjay Patel [Fri, 29 Jul 2022 20:24:52 +0000 (16:24 -0400)]
[InstCombine] add tests for udiv/urem miscompile; NFC

See issue #56810

23 months ago[InstCombine] add test for icmp with zext-bool logic; NFC
Sanjay Patel [Fri, 29 Jul 2022 17:34:21 +0000 (13:34 -0400)]
[InstCombine] add test for icmp with zext-bool logic; NFC

23 months ago[LLDB][NFC][Reliability] Fixes for int overflow and uninitialized state
Slava Gurevich [Fri, 29 Jul 2022 19:23:03 +0000 (12:23 -0700)]
[LLDB][NFC][Reliability] Fixes for int overflow and uninitialized state

Fixing potential int overflow and uninitialized variables.
These were found by Coverity static code inspection.

Differential Revision: https://reviews.llvm.org/D130795

23 months ago[IRBuilder] Make createCallHelper a member function. NFC.
Jay Foad [Fri, 29 Jul 2022 08:59:16 +0000 (09:59 +0100)]
[IRBuilder] Make createCallHelper a member function. NFC.

This just avoids explicitly passing in 'this' as an argument in a bunch
of places.

Differential Revision: https://reviews.llvm.org/D130752

23 months ago[LangRef] Reduce depth of table of contents
Jay Foad [Wed, 27 Jul 2022 14:08:00 +0000 (15:08 +0100)]
[LangRef] Reduce depth of table of contents

The table of contents in the HTML version of this doc takes up 25 pages
(in my browser, on my 4K monitor) and is too long for me to navigate
comfortably. And most of it is irrelevant detail like this:

- Bitwise Binary Operations
  - 'shl' Instruction
    - Syntax:
    - Overview:
    - Arguments:
    - Semantics:
    - Example:
  - 'lshr' Instruction
    - Syntax:
    - Overview:
    - Arguments:
    - Semantics:
    - Example:

Reducing the contents depth from 4 to 3 removes most of this detail,
leaving just a list of instructions, which only takes up 7 pages and I
find it much easier to navigate.

Incidentally the depth was set to 3 when this document was first
converted to reST and was only increased to 4 in what looks like an
accidental change: 080133453bae62a43466e881ffc03fb7cffd7288

Differential Revision: https://reviews.llvm.org/D130635

23 months ago[clang][dataflow] Handle multiple context-sensitive calls to the same function
Sam Estep [Fri, 29 Jul 2022 19:39:52 +0000 (19:39 +0000)]
[clang][dataflow] Handle multiple context-sensitive calls to the same function

This patch enables context-sensitive analysis of multiple different calls to the same function (see the `ContextSensitiveSetBothTrueAndFalse` example in the `TransferTest` suite) by replacing the `Environment` copy-assignment with a call to the new `popCall` method, which  `std::move`s some fields but specifically does not move `DeclToLoc` and `ExprToLoc` from the callee back to the caller.

To enable this, the `StorageLocation` for a given parameter needs to be stable across different calls to the same function, so this patch also improves the modeling of parameter initialization, using `ReferenceValue` when necessary (for arguments passed by reference).

This approach explicitly does not work for recursive calls, because we currently only plan to use this context-sensitive machinery to support specialized analysis models we write, not analysis of arbitrary callees.

Reviewed By: ymandel, xazax.hun

Differential Revision: https://reviews.llvm.org/D130726

23 months agoRevert "Revert "[Clang][Attribute] Introduce maybe_undef attribute for function argum...
skc7 [Tue, 19 Jul 2022 09:53:31 +0000 (09:53 +0000)]
Revert "Revert "[Clang][Attribute] Introduce maybe_undef attribute for function arguments which accepts undef values""

This reverts commit 4e1fe96.

Reverting this commit and fix the tests that caused failures due to
a35c64c.

23 months ago[compiler-rt] [builtins] Detect _Float16 support at compile time
Dimitry Andric [Thu, 28 Jul 2022 17:52:16 +0000 (19:52 +0200)]
[compiler-rt] [builtins] Detect _Float16 support at compile time

Instead of detecting `_Float16` support at CMake configuration time,
detect it at compile time by checking for the predefined (by the
compiler) macro `__FLT16_MAX__` instead.

This solves the issue where compiler-rt is built simultaneously for both
x86_64 and i386 targets, and the CMake configuration uses x86_64
compilation to detect `_Float16` support, while it may not be supported
by the i386 target (if it does not have SSE2).

While here, rename `COMPILERT_RT_HAS_FLOAT16` to `CRT_HAS_FLOAT16`, to
conform more to the naming style used in `int_lib.h` and `int_types.h`.

Reviewed By: MaskRay

Differential Revision: https://reviews.llvm.org/D130718

23 months agoRevert "[Clang][Attribute] Introduce maybe_undef attribute for function arguments...
Amy Kwan [Fri, 29 Jul 2022 18:16:37 +0000 (13:16 -0500)]
Revert "[Clang][Attribute] Introduce maybe_undef attribute for function arguments which accepts undef values"

This reverts commit a35c64ce23b7c7e4972c89b224b9363639dddea2.

Reverting this commit as it causes various failures on LE and BE PPC bots.

23 months ago[TSan][Darwin] Additional TSAN test requiring weak symbol for dyld64
Blue Gaston [Thu, 28 Jul 2022 21:41:32 +0000 (14:41 -0700)]
[TSan][Darwin] Additional TSAN test requiring weak symbol for dyld64

Tests require a weak symbol for dyld weak-def coalescing.

Differential Revision: https://reviews.llvm.org/D130732

Adding additional weak attribute

23 months ago[AMDGPU] Omit unnecessary waitcnt before barriers
Austin Kerbow [Fri, 25 Feb 2022 07:26:51 +0000 (23:26 -0800)]
[AMDGPU] Omit unnecessary waitcnt before barriers

It is not necessary to wait for all outstanding memory operations before
barriers on hardware that can back off of the barrier in the event of an
exception when traps are enabled. Add a new subtarget feature which
tracks which HW has this ability.

Reviewed By: #amdgpu, rampitec

Differential Revision: https://reviews.llvm.org/D130722

23 months ago[gn build] (manually) port 507125af3d0b more
Nico Weber [Fri, 29 Jul 2022 18:06:10 +0000 (14:06 -0400)]
[gn build] (manually) port 507125af3d0b more

23 months ago[gn build] (manually) port 507125af3d0b
Nico Weber [Fri, 29 Jul 2022 18:03:32 +0000 (14:03 -0400)]
[gn build] (manually) port 507125af3d0b

23 months agoReplace Optional::hasValue with has_value or operator bool. NFC
Fangrui Song [Fri, 29 Jul 2022 17:57:25 +0000 (10:57 -0700)]
Replace Optional::hasValue with has_value or operator bool. NFC

23 months ago[libc++] Rename __libcpp_assertion_handler to __libcpp_verbose_abort
Louis Dionne [Mon, 25 Jul 2022 17:43:47 +0000 (13:43 -0400)]
[libc++] Rename __libcpp_assertion_handler to __libcpp_verbose_abort

With the goal of reusing that handler to do other things besides
handling assertions (such as terminating when an exception is thrown
under -fno-exceptions), the name `__libcpp_assertion_handler` doesn't
really make sense anymore.

Furthermore, I didn't want to use the name `__libcpp_abort_handler`,
since that would give the impression that the handler is called
whenever `std::abort()` is called, which is not the case at all.

Differential Revision: https://reviews.llvm.org/D130562

23 months ago[clang-repl] Disable execution unittests on unsupported platforms.
Sunho Kim [Fri, 29 Jul 2022 17:27:45 +0000 (02:27 +0900)]
[clang-repl] Disable execution unittests on unsupported platforms.

After the intoduction of global destructor support, there is a possiblity to run invalid instructions in the destructor of Interpreter class. Completely disable tests in platforms with failing test cases.

Differential Revision: https://reviews.llvm.org/D130786

23 months agoDiagnose use of _Noreturn on a struct/union field
Aaron Ballman [Fri, 29 Jul 2022 17:16:10 +0000 (13:16 -0400)]
Diagnose use of _Noreturn on a struct/union field

C99 6.7.4p2 clarifies that a function specifier can only be used in the
declaration of a function. _Noreturn is a function specifier, so it is
a constraint violation to write it on a structure or union field, but
we missed that case.

Fixes #56800

23 months agoworkflows: Pass phab token to github-automation.py when creating a pull request
Tom Stellard [Fri, 29 Jul 2022 16:59:55 +0000 (09:59 -0700)]
workflows: Pass phab token to github-automation.py when creating a pull request

The script needs this in order to automatically assign a reviewer.

23 months ago[lld/mac] Comment changes requested on https://reviews.llvm.org/D130725
Nico Weber [Fri, 29 Jul 2022 16:55:04 +0000 (12:55 -0400)]
[lld/mac] Comment changes requested on https://reviews.llvm.org/D130725

No behavior change.

23 months ago[mlir][LLVM] Rework the API of GEPOp
Markus Böck [Thu, 28 Jul 2022 20:42:37 +0000 (22:42 +0200)]
[mlir][LLVM] Rework the API of GEPOp

The implementation and API of GEP Op has gotten a bit convoluted over the time. Issues with it are:
* Misleading naming: `indices` actually only contains the dynamic indices, not all of them. To get the amount of indices you need to query the size of `structIndices`
* Very difficult to iterate over all indices properly: One had to iterate over `structIndices`, check whether it contains the magic constant `kDynamicIndex`, if it does, access the next value in `index` etc.
* Inconvenient to build: One either has create lots of constant ops for every index or have an odd split of passing both a `ValueRange` as well as a `ArrayRef<int32_t>` filled with `kDynamicIndex` at the correct places.
* Implementation doing verification in the build method
and more.

This patch attempts to address all these issues via convenience classes and reworking the way GEP Op works:
* Adds `GEPArg` class which is a sum type of a `int32_t` and `Value` and is used to have a single convenient easy to use `ArrayRef<GEPArg>` in the builders instead of the previous `ValueRange` + `ArrayRef<int32_t>` builders.
* Adds `GEPIndicesAdapter` which is a class used for easy random access and iteration over the indices of a GEP. It is generic and flexible enough to also instead return eg. a corresponding `Attribute` for an operand inside of `fold`.
*  Rename `structIndices` to `rawConstantIndices` and `indices` to `dynamicIndices`: `rawConstantIndices` signifies one shouldn't access it directly as it is encoded, and `dynamicIndices` is more accurate and also frees up the `indices` name.
* Add `getIndices` returning a `GEPIndicesAdapter` to easily iterate over the GEP Ops indices.
* Move the verification/asserts out of the build method and into the `verify` method emitting op error messages.
* Add convenient builder methods making use of `GEPArg`.
* Add canonicalizer turning dynamic indices with constant values into constant indices to have a canonical representation.

The only breaking change is for any users building GEPOps that have so far used the old `ValueRange` + `ArrayRef<int32_t>` builder as well as those using the generic syntax.

Another follow up patch then goes through upstream and makes use of the new `ArrayRef<GEPArg>` builder to remove a lot of code building constants for GEP indices.

Differential Revision: https://reviews.llvm.org/D130730

23 months ago[ELF] Strip directories for -Map when emitting reproducer rsp
Alex Brachet [Fri, 29 Jul 2022 16:17:33 +0000 (16:17 +0000)]
[ELF] Strip directories for -Map when emitting reproducer rsp

Similarly to -o output directories will not be created so -Map being
copied verbatim will likely cause ld.lld @response.txt to fail.

Differential Revision: https://reviews.llvm.org/D130681

23 months ago[mlir][complex] Canonicalize complex.add zero
lewuathe [Fri, 29 Jul 2022 12:31:55 +0000 (14:31 +0200)]
[mlir][complex] Canonicalize complex.add zero

Adding complex value with 0 for real and imaginary part can be ignored.

NOTE: This type of canonicalization can be written in an easy and tidy format using `complex.number` after constant op supports custom attribute.

Differential Revision: https://reviews.llvm.org/D130748