New pass manager for LICM.
Summary: Port LICM to the new pass manager.
Reviewers: davidxl, silvas
Subscribers: krasin, vitalybuka, silvas, davide, sanjoy, llvm-commits, mehdi_amini
Differential Revision: http://reviews.llvm.org/D21772
llvm-svn: 275224
The lookup tables can get out of date during the lifetime of the object so we
need to preserve LLDB's ability to answer questions about TagDecls' contents.
<rdar://problem/20751935>
llvm-svn: 275223
We can freeze the registers after the MachineFrameInfo has been configured (by
telling it about calls, inline asm, ...). This doesn't happen at all yet, but
will be part of IR translation.
Fixes -verify-machineinstrs assertion.
llvm-svn: 275221
The LCSSA pass itself will not generate several redundant PHI nodes in a single
exit block. However, such redundant PHI nodes don't violate LCSSA form, and may
be introduced by passes that preserve LCSSA, and/or preserved by the LCSSA pass
itself. So, assuming a single PHI node per exit block is not safe.
llvm-svn: 275217
Summary:
Refactored the profitability analysis out of the IC promotion pass and
into lib/Analysis so that it can be accessed by the summary index
builder in a follow-on patch to enable IC promotion in ThinLTO (D21932).
Reviewers: davidxl, xur
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D22182
llvm-svn: 275216
This patch corresponds to review:
http://reviews.llvm.org/D20239
It adds exploitation of XXINSERTW and XXEXTRACTUW instructions that
are useful in some cases for inserting and extracting vector elements of
v4[if]32 vectors.
llvm-svn: 275215
The thread registry test was failing to allocate 25 threads with stack
size 134MB, which is pretty reasonable.
Also print the error code in our pthread wrappers in case this happens
again.
llvm-svn: 275209
mov edi,edi is _not_ NOP in 64-bit, use 66,90h instead.
This bug was causing interception unittest to crash on
Windows64 (windows 8 and windows 10).
Credits to etienneb for finding the root cause.
Patch by: Wei Wang
Differential Revision: http://reviews.llvm.org/D22274
llvm-svn: 275207
With r274952 and r275201 in place there are no cases left where a
forward liveness analysis yields different results than a backward one.
So we can remove the forward stepping logic.
Differential Revision: http://reviews.llvm.org/D22083
llvm-svn: 275204
Use LivePhysRegs with a backwards walking algorithm to update live in
lists, this way the results do not depend on the presence of kill flags
anymore.
This patch also reduces the number of registers added as live-in.
Previously all pristine registers as well as all sub registers of a
super register were added resulting in unnecessarily large live in
lists. This fixed https://llvm.org/PR25263.
Differential Revision: http://reviews.llvm.org/D22027
llvm-svn: 275201
A BuiltinTemplateDecl has no underlying templated decl and as such they
cannot be relied upon for mangling. The ItaniumMangler had some bugs
here which lead to crashes.
This fixes PR28519.
llvm-svn: 275190
Check not only that the compiler is not crashing, but also whether the
probablematic part (The sequence of instructions simplified to '4') is reflected
in the output.
Thanks to Tobias for the hint.
llvm-svn: 275189
Loading a dynamic library can take quite a long time, since it triggers a number of
shared-library-event stops for dependent libraries. This is especially true for remote targets
due to communication latency. Increase the default 500ms timeout to account for that.
Committing as obvious.
llvm-svn: 275185
I am not sure exactly which test breakage Martin was trying to fix in
r273694. For now, fix the behavior for top-level conditionals, which
(surprisingly) are actually used somewhat commonly.
llvm-svn: 275183
Summary:
The unittest is not working on windows. The "regexp.h" is not available and
gtest-port won't support regexp with operator '|'.
The test wasn't failing on 32-bits because of this check:
```
if (SANITIZER_WORDSIZE != 64 || ASAN_AVOID_EXPENSIVE_TESTS) return;
```
```
[ RUN ] AddressSanitizer.HugeMallocTest
C:/src/llvm/llvm/utils/unittest/googletest\src/gtest-port.cc(272): error: Failed
Syntax error at index 30 in simple regular expression "is located 1 bytes to the left|AddressSanitizer failed to allocate": '|' is unsupported.
C:/src/llvm/llvm/projects/compiler-rt/lib/asan/tests/asan_test.cc(308): error: Death test: Ident((char*)malloc(n_megs << 20))[-1] = 0
Result: died but not with expected error.
Expected: is located 1 bytes to the left|AddressSanitizer failed to allocate
```
See gtest-port.h:
```
#elif GTEST_OS_WINDOWS
// <regex.h> is not available on Windows. Use our own simple regex
// implementation instead.
# define GTEST_USES_SIMPLE_RE 1
```
Reviewers: rnk
Subscribers: llvm-commits, wang0109, kubabrecka, chrisha
Differential Revision: http://reviews.llvm.org/D22257
llvm-svn: 275181
Summary:
Many CRT (64-bits) functions contains a "hint-nop". The current padding
detection is not able to recognize the 10-bytes padding and the HotPatch
hooking technique cannot be used.
Other patterns may be discover and may be added later.
Reviewers: rnk
Subscribers: llvm-commits, wang0109, chrisha
Differential Revision: http://reviews.llvm.org/D22258
llvm-svn: 275180
If a subtarget has both ZCZeroing and CustomCheapAsMoveHandling features (now
only Kryo has both), set FMOVS0 and FMOVD0 isAsCheapAsAMove.
Differential Revision: http://reviews.llvm.org/D22256
llvm-svn: 275178
Greg Clayton