Commit Graph

357071 Commits

Author SHA1 Message Date
Xing GUO 613c4a87ba [ObjectYAML][DWARF] Add one helper function `writeInitialLength()`. NFC. 2020-06-12 21:10:58 +08:00
Kirstóf Umann 33fb9cbe21 [analyzer][NFC] Don't allow dependency checkers to emit diagnostics
The thrilling conclusion to the barrage of patches I uploaded lately! This is a
big milestone towards the goal set out in http://lists.llvm.org/pipermail/cfe-dev/2019-August/063070.html.
I hope to accompany this with a patch where the a coreModeling package is added,
from which package diagnostics aren't allowed either, is an implicit dependency
of all checkers, and the core package for the first time can be safely disabled.

Differential Revision: https://reviews.llvm.org/D78126
2020-06-12 14:59:48 +02:00
Erich Keane 82a21229da (PR46111) Properly handle elaborated types in an implicit deduction guide
As reported in PR46111, implicit instantiation of a deduction guide
causes us to have an elaborated type as the parameter, rather than the
dependent type.

After review and feedback from @rsmith, this patch solves this problem
by wrapping the value in an uninstantiated typedef/type-alias that is
instantiated when required later.

Differential Revision: https://reviews.llvm.org/D80743
2020-06-12 05:32:13 -07:00
Kirstóf Umann e22f1c02a2 [analyzer] Introduce weak dependencies to express *preferred* checker callback evaluation order
Checker dependencies were added D54438 to solve a bug where the checker names
were incorrectly registered, for example, InnerPointerChecker would incorrectly
emit diagnostics under the name MallocChecker, or vice versa [1]. Since the
system over the course of about a year matured, our expectations of what a role
of a dependency and a dependent checker should be crystallized a bit more --
D77474 and its summary, as well as a variety of patches in the stack
demonstrates how we try to keep dependencies to play a purely modeling role. In
fact, D78126 outright forbids diagnostics under a dependency checkers name.

These dependencies ensured the registration order and enabling only when all
dependencies are satisfied. This was a very "strong" contract however, that
doesn't fit the dependency added in D79420. As its summary suggests, this
relation is directly in between diagnostics, not modeling -- we'd prefer a more
specific warning over a general one.

To support this, I added a new dependency kind, weak dependencies. These are not
as strict of a contract, they only express a preference in registration order.
If a weak dependency isn't satisfied, the checker may still be enabled, but if
it is, checker registration, and transitively, checker callback evaluation order
is ensured.

If you are not familiar with the TableGen changes, a rather short description
can be found in the summary of D75360. A lengthier one is in D58065.

[1] https://www.youtube.com/watch?v=eqKeqHRAhQM

Differential Revision: https://reviews.llvm.org/D80905
2020-06-12 14:08:38 +02:00
Simon Pilgrim 8d30945ab9 [X86][SSE] combineX86ShuffleChain - combine INSERT_VECTOR_ELT patterns to INSERTPS
Noticed while trying to cleanup D66004 - if a shuffle operand came from a scalar, we're better off using INSERTPS vs UNPCKLPS as this is more likely to load fold later on. It also matches our existing BUILD_VECTOR lowering.

We can extend this to other PINSRB/D/Q/W cases in the future as the need arises.
2020-06-12 11:59:01 +01:00
Frederik Gossen 6196c37969 [MLIR] Add missing traits and assembly format to `shape.from/to_extent_tensor`
Add `NoSideEffect` trait to `shape.to_extent_tensor` and
`shape.from_extent_tensor` and defined custom assembly format for the
operations.

Differential Revision: https://reviews.llvm.org/D81158
2020-06-12 10:54:57 +00:00
Florian Hahn 4495a6b141 [BreakCritEdges] Add option to opt-out of perserving loop-simplify.
This patch adds a new option to CriticalEdgeSplittingOptions to control
whether loop-simplify form must be preserved. It is them used by GVN to
indicate that loop-simplify form does not have to be preserved.

This fixes a crash exposed by 189efe295b.

If the critical edge we are splitting goes from a block inside a loop to
a block outside the loop, splitting the edge will create a new exit
block. As a result, the new block will branch to the original exit
block, which will add a non-loop predecessor, breaking loop-simplify
form. To preserve loop-simplify form, the predecessor blocks of the
original exit are split, but that does not work for blocks with
indirectbr terminators. If preserving loop-simplify form is requested,
bail out , before making any changes.

Reviewers: reames, hfinkel, davide, efriedma

Reviewed By: efriedma

Differential Revision: https://reviews.llvm.org/D81582
2020-06-12 11:47:13 +01:00
Kirstóf Umann d61b1f8534 [analyzer][NFC] Change checker dependency unit tests to check for the registration order
Exactly what it says on the tin! "Strong" dependencies are mentioned in contrast
to a new kind of dependency introduced in a followup patch.

Differential Revision: https://reviews.llvm.org/D80901
2020-06-12 12:43:56 +02:00
Xing GUO 7e0827e86f [ObjectYAML][test] Use a single test file to test the empty 'DWARF' entry.
This patch addresses comments in [D81450](https://reviews.llvm.org/D81450#inline-748745)

Reviewed By: jhenderson

Differential Revision: https://reviews.llvm.org/D81529
2020-06-12 17:56:22 +08:00
Alex Bradbury 3dcfd482cb [CodeGen] Increase applicability of ffine-grained-bitfield-accesses for targets with limited native integer widths
As pointed out in PR45708, -ffine-grained-bitfield-accesses doesn't
trigger in all cases you think it might for RISC-V. The logic in
CGRecordLowering::accumulateBitFields checks OffsetInRecord is a legal
integer according to the datalayout. RISC targets will typically only
have the native width as a legal integer type so this check will fail
for OffsetInRecord of 8 or 16 when you would expect the transformation
is still worthwhile.

This patch changes the logic to check for an OffsetInRecord of a at
least 1 byte, that fits in a legal integer, and is a power of 2. We
would prefer to query whether native load/store operations are
available, but I don't believe that is possible.

Differential Revision: https://reviews.llvm.org/D79155
2020-06-12 10:33:47 +01:00
Raphael Isemann f257c2138f Add missing IRBuilder include to IndirectThunks.h
This file didn't build with enabled modules.
2020-06-12 11:31:58 +02:00
Florian Hahn 3a846d4d92 [VPlan] Reject loops without computable backedge taken counts
getOrCreateTripCount is used to generate code for the outer loop, but it
requires a computable backedge taken counts. Check that in the VPlan
native path.

Reviewers: Ayal, gilr, rengolin, sguggill

Reviewed By: sguggill

Differential Revision: https://reviews.llvm.org/D81088
2020-06-12 10:31:18 +01:00
Sebastian Neubauer 29a6ad94fd [AMDGPU] Add G16 support to image instructions
Add G16 feature for GFX10 and support A16 and G16 in GlobalISel.

Differential Revision: https://reviews.llvm.org/D76836
2020-06-12 11:26:31 +02:00
Georgii Rymar d95f8e7aef [yaml2obj][MachO] - Fix PubName/PubType handling.
`PubName` and `PubType` are optional fields since D80722.

They are defined as:
  Optional<PubSection> PubNames;
  Optional<PubSection> PubTypes;

And initialized in the following way:
  IO.mapOptional("debug_pubnames", DWARF.PubNames);
  IO.mapOptional("debug_pubtypes", DWARF.PubTypes);

But problem is that because of the issue in `YAMLTraits.cpp`,
when there are no `debug_pubnames`/`debug_pubtypes` keys in a YAML description,
they are not initialized to `Optional::None` as the code expects, but they
are initialized to default `PubSection()` instances.

Because of this, the `if` condition in the following code is always true:

if (Obj.DWARF.PubNames)
  Err = DWARFYAML::emitPubSection(OS, *Obj.DWARF.PubNames,
                                  Obj.IsLittleEndian);

What means `emitPubSection` is always called and it writes few values.

This patch fixes the issue. I've reduced `sizeofcmds` by size of data
previously written because of this bug.

Differential revision: https://reviews.llvm.org/D81686
2020-06-12 12:03:51 +03:00
Raphael Isemann 91728b9172 [lldb] Don't print IRForTarget errors directly to the console
Summary:

When we get an error back from IRForTarget we directly print that error to the
debugger output stream instead of putting it in the result object. The result
object only gets a vague "The expression could not be prepared to run in the
target" error message that doesn't actually tell the user what went wrong.

This patch just puts the IRForTarget errors into the status object that is
returned to the caller instead of directly printing it to the debugger. Also
updates one test that now can actually check for the error message it is
supposed to check for (instead of the default error which is all we had before).

Reviewers: JDevlieghere

Reviewed By: JDevlieghere

Differential Revision: https://reviews.llvm.org/D81654
2020-06-12 10:27:25 +02:00
Chen Zheng 9b6e86a1a5 [PowerPC] refactor convertToImmediateForm - NFC
This is a NFC patch to make convertToImmediateForm a light wrapper
for converting xform and imm form instructions on PowerPC.

Reviewed By: Steven.zhang

Differential Revision: https://reviews.llvm.org/D80907
2020-06-12 03:57:54 -04:00
Rainer Orth 2feb72bcd5 [cmake] Don't pass -z discard-unused to Illumos ld
I'm currently working to port `libc++` to Solaris.  There exists a slightly
bitrotten port already, which was done on Illumos, an OpenSolaris
derivative.  In order not to break that port with my work, I need to test
the result on both Solaris and Illumos.  While doing so, it turned out that
Illumos `ld` doesn't support the `-z discard-sections=unused` option
currently used on SunOS unconditionally.

While there exists a patch
<https://github.com/OpenIndiana/oi-userland/blob/oi/hipster/components/developer/clang-90/patches/02-cmake_modules_AddLLVM.cmake.patch>
for LLVM 9.0 in the OpenIndiana repository, it apparently hasn't been
submitted upstream and is completely wrong: it replaces
`-z discard-sections=unused` with `-z ignore`.  In terms of the equivalent
`gld` options, this means replacing `--gc-sections` with `--as-needed`.

This patch instead tests if the linker actually supports the option before
using it.

Tested on `amd64-pc-solaris2.11` (all of Solaris 11.4, 11.3 and OpenIndiana
2020.04).

Differential Revision: https://reviews.llvm.org/D81545
2020-06-12 09:56:42 +02:00
Martin Boehme 2e92b397ae [clang] Rename Decl::isHidden() to isUnconditionallyVisible().
Also invert the sense of the return value.

As pointed out by the FIXME that this change resolves, isHidden() wasn't
a very accurate name for this function.

I haven't yet changed any of the strings that are output in
ASTDumper.cpp / JSONNodeDumper.cpp / TextNodeDumper.cpp in response to
whether isHidden() is set because

a) I'm not sure whether it's actually desired to change these strings
   (would appreciate feedback on this), and

b) In any case, I'd like to get this pure rename out of the way first,
   without any changes to tests. Changing the strings that are output in
   the various ...Dumper.cpp files will require changes to quite a few
   tests, and I'd like to make those in a separate change.

Differential Revision: https://reviews.llvm.org/D81392

Reviewed By: rsmith
2020-06-12 09:33:42 +02:00
EgorBo 012909dcaf
[InstCombine] "X - (X / C) * C == 0" to "X & C-1 == 0"
Summary:
"X % C == 0" is optimized to "X & C-1 == 0" (where C is a power-of-two)
However, "X % Y" can also be represented as "X - (X / Y) * Y" so if I rewrite the initial expression:
"X - (X / C) * C == 0" it's not currently optimized to "X & C-1 == 0", see godbolt: https://godbolt.org/z/KzuXUj

This is my first contribution to LLVM so I hope I didn't mess things up

Reviewers: lebedev.ri, spatel

Reviewed By: lebedev.ri

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D79369
2020-06-12 10:20:06 +03:00
EgorBo 6538b3adbe
[NFC][InstCombine] Tests for "X - (X / C) * C == 0" pattern
See https://reviews.llvm.org/D79369
2020-06-12 10:20:06 +03:00
Jonas Devlieghere 425c6f079b [llvm/Object] Reimplment basic_symbol_iterator in TapiFile
Use indices into the Symbols vector instead of casting the objects in
the vector and dereferencing std::vector::end().

This change is NFC modulo the Windows failure reported by
llvm-clang-x86_64-expensive-checks-win.

Differential revision: https://reviews.llvm.org/D81717
2020-06-12 00:03:32 -07:00
Kristof Beyls c35ed40f4f [AArch64] Extend AArch64SLSHardeningPass to harden BLR instructions.
To make sure that no barrier gets placed on the architectural execution
path, each
  BLR x<N>
instruction gets transformed to a
  BL __llvm_slsblr_thunk_x<N>
instruction, with __llvm_slsblr_thunk_x<N> a thunk that contains
__llvm_slsblr_thunk_x<N>:
  BR x<N>
  <speculation barrier>

Therefore, the BLR instruction gets split into 2; one BL and one BR.
This transformation results in not inserting a speculation barrier on
the architectural execution path.

The mitigation is off by default and can be enabled by the
harden-sls-blr subtarget feature.

As a linker is allowed to clobber X16 and X17 on function calls, the
above code transformation would not be correct in case a linker does so
when N=16 or N=17. Therefore, when the mitigation is enabled, generation
of BLR x16 or BLR x17 is avoided.

As BLRA* indirect calls are not produced by LLVM currently, this does
not aim to implement support for those.

Differential Revision:  https://reviews.llvm.org/D81402
2020-06-12 07:34:33 +01:00
Mehdi Amini 6f0ce46873 Revert "[mlir][spirv] Enhance structure type member decoration handling"
This reverts commit 5d74df5b03.

This broke the MSVC build:  <bits/stdint-uintn.h> isn't available on Windows
2020-06-12 05:01:24 +00:00
Yevgeny Rouban 707836ed4e [JumpThreading] Handle zero !prof branch_weights
Avoid division by zero in updatePredecessorProfileMetadata().

Reviewers: yamauchi
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D81499
2020-06-12 11:55:15 +07:00
Craig Topper 0ce9bf6eed [X86] Add a helper lambda to getIntelProcessorTypeAndSubtype to select feature bits from the correct 32-bit feature variable.
We have three 32 bit variables containing feature bits. But our
enum is a flat 96 bit space. So we need to pick which of the
variables to use based on the bit value. We used to do this
manually by mentioning the correct variable and subtracting an
offset from the enum. But this is error prone.
2020-06-11 21:14:46 -07:00
Vitaly Buka 999307323a [StackSafety] Fix byval handling
We don't need process paramenters which marked as
byval as we are not going to pass interested allocas
without copying.

If we pass value into byval argument, we just handle that
as Load of corresponding type and stop that branch of analysis.
2020-06-11 20:58:36 -07:00
Alexander Shaposhnikov c966ed8dc7 [llvm-objcopy][MachO] Fix cmdsize of LC_RPATH
Fix the calculation of the field cmdsize (in the function buildRPathLoadCommand)
to account for the null byte terminator.

Patch by Sameer Arora!

Test plan: make check-all

Differential revision: https://reviews.llvm.org/D81575
2020-06-11 19:55:04 -07:00
Jonas Devlieghere 526e0c8d15 [lldb/Test] Fix ASan/TSan workaround for Xcode Python 3
The Python 3 interpreter in Xcode has a relative RPATH and dyld fails to
load it when we copy it into the build directory.

This patch adds an additional check that the copied binary can be
executed. If it doesn't, we assume we're dealing with the Xcode python
interpreter and return the path to the real executable. That is
sufficient for the sanitizers because only system binaries need to be
copied to work around SIP.

This patch also moves all that logic out of LLDBTest and into the lit
configuration so that it's executed only once per test run, instead of
once for every test. Although I didn't benchmark the difference this
should result in a mild speedup.

Differential revision: https://reviews.llvm.org/D81696
2020-06-11 19:36:42 -07:00
Yonghong Song 4db1878158 [BPF] fix incorrect type in BPFISelDAGToDAG readonly load optimization
In BPF Instruction Selection DAGToDAG transformation phase,
BPF backend had an optimization to turn load from readonly data
section to direct load of the values. This phase is implemented
before libbpf has readonly section support and before alu32
is supported.

This phase however may generate incorrect type when alu32 is
enabled. The following is an example,
  -bash-4.4$ cat ~/tmp2/t.c
  struct t {
    unsigned char a;
    unsigned char b;
    unsigned char c;
  };
  extern void foo(void *);
  int test() {
    struct t v = {
      .b = 2,
    };
    foo(&v);
    return 0;
  }

The compiler will turn local variable "v" into a readonly section.
During instruction selection phase, the compiler generates two
loads from readonly section, one 2 byte load or 1 byte load, e.g., for 2 loads,
  t8: i32,ch = load<(dereferenceable load 2 from `i8* getelementptr inbounds
       (%struct.t, %struct.t* @__const.test.v, i64 0, i32 0)`, align 1),
       anyext from i16> t3, GlobalAddress:i64<%struct.t* @__const.test.v> 0, undef:i64
  t9: ch = store<(store 2 into %ir.v1.sub1), trunc to i16> t3, t8,
    FrameIndex:i64<0>, undef:i64

BPF backend changed t8 to i64 = Constant<2> and eventually the generated machine IR:
  t10: i64 = MOV_ri TargetConstant:i64<2>
  t40: i32 = SLL_ri_32 t10, TargetConstant:i32<8>
  t41: i32 = OR_ri_32 t40, TargetConstant:i64<0>
  t9: ch = STH32<Mem:(store 2 into %ir.v1.sub1)> t41, TargetFrameIndex:i64<0>,
      TargetConstant:i64<0>, t3

Note that t10 in the above is not correct. The type should be i32 and instruction
should be MOV_ri_32. The reason for incorrect insn selection is BPF insn selection
generated an i64 constant instead of an i32 constant as specified in the original
load instruction. Such incorrect insn sequence eventually caused the following
fatal error when a COPY insn tries to copy a 64bit register to a 32bit subregister.
  Impossible reg-to-reg copy
  UNREACHABLE executed at ../lib/Target/BPF/BPFInstrInfo.cpp:42!

This patch fixed the issue by using the load result type instead of always i64
when doing readonly load optimization.

Differential Revision: https://reviews.llvm.org/D81630
2020-06-11 19:31:06 -07:00
Esme-Yi af9f8c24a0 Revert "[PowerPC][NFC] Testing ROTL of v1i128."
This reverts commit 174192af01.
2020-06-12 02:23:52 +00:00
Cyndy Ishida 28fefcc83c [llvm][llvm-nm] add TextAPI/MachO support
Summary:
This completes the needed glueing to support reading tbd files from nm.
This includes specifying which slice filtering with `--arch` and a new
option specifically for tbd files `--add-inlinedinfo` which will show
the reexported libraries that are appended in the tbd file.

Reviewers: ributzka, steven_wu, JDevlieghere, jhenderson

Reviewed By: JDevlieghere

Subscribers: hiraditya, MaskRay, dexonsmith, rupprecht, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D81614
2020-06-11 18:54:16 -07:00
Alina Sbirlea 519b019a0a Verify MemorySSA after all updates.
Verify after completing all updates.
Resolves PR46275.
2020-06-11 18:48:41 -07:00
Stella Laurenzo 10bd67c1bd Install the MLIRTableGen static library.
Summary: * This library is special because of its dependencies so seems to have been inadvertently left out of installs.

Reviewers: antiagainst

Subscribers: mgorny, mehdi_amini, rriddle, jpienaar, shauheen, antiagainst, nicolasvasilache, arpith-jacob, mgester, lucyrfox, aartbik, liufengdb, stephenneuendorffer, Joonsoo, grosul1, frgossen, Kayjukh, jurahul, msifontes

Tags: #mlir

Differential Revision: https://reviews.llvm.org/D81693
2020-06-11 18:23:24 -07:00
Mehdi Amini 95371ce9c2 Enable FileCheck -enable-var-scope by default in MLIR test
This option avoids to accidentally reuse variable across -LABEL match,
it can be explicitly opted-in by prefixing the variable name with $

Differential Revision: https://reviews.llvm.org/D81531
2020-06-12 00:43:09 +00:00
Wen-Heng (Jack) Chung 7bfcb9120f Use python3 in generate-test-checks.py.
Summary: Upgrade to python3 in generate-test-checks.py .

Subscribers: mehdi_amini, rriddle, jpienaar, shauheen, antiagainst, nicolasvasilache, arpith-jacob, mgester, lucyrfox, liufengdb, stephenneuendorffer, Joonsoo, grosul1, frgossen, Kayjukh, jurahul, msifontes

Tags: #mlir

Differential Revision: https://reviews.llvm.org/D81701
2020-06-11 19:19:22 -05:00
HazemAbdelhafez 5d74df5b03 [mlir][spirv] Enhance structure type member decoration handling
Modify structure type in SPIR-V dialect to support:
1) Multiple decorations per structure member
2) Key-value based decorations (e.g., MatrixStride)

This commit kept the Offset decoration separate from members'
decorations container for easier implementation and logical clarity.
As such, all references to Structure layoutinfo are now offsetinfo,
and any member layout defining decoration (e.g., RowMajor for Matrix)
will be add to the members' decorations container along with its
value if any.

Differential Revision: https://reviews.llvm.org/D81426
2020-06-11 19:52:13 -04:00
Eric Christopher 3ff8f61930 Tidy up unsigned -> Register fixups. 2020-06-11 16:50:58 -07:00
Jacques Pienaar 8472ae1773 [mlir] Add additional header
Fix gcc10 build and closes: https://bugs.llvm.org/show_bug.cgi?id=45680
2020-06-11 16:46:49 -07:00
Eric Christopher cb21b16822 Add a diagnostic string to an assert. 2020-06-11 16:34:55 -07:00
Matt Arsenault 7d913becfc AMDGPU/GlobalISel: Fix select of private <2 x s16> load 2020-06-11 19:25:25 -04:00
Sanjay Patel 039ff29ef6 [VectorCombine] remove unused parameters; NFC 2020-06-11 19:15:03 -04:00
Vitaly Buka a10fc165f5 [StackSafety,NFC] Fix use of CallBase API
Code does not need iterate arguments and can get ArgNo from
CallBase::getArgOperandNo.
2020-06-11 16:11:30 -07:00
Matt Arsenault 27f8bd94cb AMDGPU/GlobalISel: Fix select of <8 x s64> scalar load 2020-06-11 19:09:43 -04:00
Akira Hatanaka c9a52de002 [CodeGen] Simplify the way lifetime of block captures is extended
Rather than pushing inactive cleanups for the block captures at the
entry of a full expression and activating them during the creation of
the block literal, just call pushLifetimeExtendedDestroy to ensure the
cleanups are popped at the end of the scope enclosing the block
expression.

rdar://problem/63996471

Differential Revision: https://reviews.llvm.org/D81624
2020-06-11 16:06:22 -07:00
Jonas Devlieghere 8d8ec55035 [lldb/Test] Unify DYLD_INSERT_LIBRARIES solution for ASan and TSan
Add the same fix for loading the sanitizer runtime for TSan as we
currently have for ASan and unify the code with a helper function.
2020-06-11 16:04:46 -07:00
Matt Arsenault 2247072b65 AMDGPU/GlobalISel: Set insert point when emitting control flow pseudos
This was implicitly assuming the branch instruction was the next after
the pseudo. It's possible for another non-terminator instruction to be
inserted between the intrinsic and the branch, so adjust the insertion
point. Fixes a non-terminator after terminator verifier error (which
without the verifier, manifested itself as an infinite loop in
analyzeBranch much later on).
2020-06-11 18:53:26 -04:00
George Mitenkov fc148a4c88 [MLIR][SPIRVToLLVM] Added conversion for SPIR-V comparison ops
Implemented `FComparePattern` and `IComparePattern` classes
that provide conversion of SPIR-V comparison ops (such as
`spv.FOrdGreaterThanEqual` and others) to LLVM dialect.
Also added tests in `comparison-ops-to-llvm.mlir`.

Differential Revision: https://reviews.llvm.org/D81487
2020-06-11 18:46:17 -04:00
Kirill Naumov 1022b5eb5b [InlineCost] Preparational patch for creation of Printer pass.
- Renaming the printer class, flag
- Refactoring
- Changing some tests

This patch is a preparational stage for introducing a new printing pass and new
functionality to the existing Annotation Writer. I plan to extend
this functionality for this tool to be more useful when looking at the inline
process.
2020-06-11 22:29:03 +00:00
Fangrui Song 030897523d [Support] Don't tie errs() to outs() by default
This reverts part of D81156.

Accessing errs() concurrently was safe before and racy after D81156.
(`errs() << 'a'` is always racy)

Accessing outs() and errs() concurrently was safe before and racy after D81156.

Don't tie errs() to outs() by default to fix the fallout.
llvm-dwarfdump is single-threaded and opting in the tie behavior is safe.
2020-06-11 15:19:56 -07:00
John McCall 7fac1acc61 Set the LLVM FP optimization flags conservatively.
Functions can have local pragmas that override the global settings.
We set the flags eagerly based on global settings, but if we emit
an expression under the influence of a pragma, we clear the
appropriate flags from the function.

In order to avoid doing a ton of redundant work whenever we emit
an FP expression, configure the IRBuilder to default to global
settings, and only reconfigure it when we see an FP expression
that's not using the global settings.

Patch by Michele Scandale!

https://reviews.llvm.org/D80462
2020-06-11 18:16:41 -04:00