Before, loop unrolling was only enabled for loops with a single
block. This restriction has been removed and replaced by:
- allow a maximum of two exiting blocks,
- a four basic block limit for cores with a branch predictor.
Differential Revision: https://reviews.llvm.org/D38952
llvm-svn: 316313
r315292 introduced a change that's supposed to consistently ignore
"dead" output sections when placing orphans. Unfortunately, that
change doesn't handle the special case when the orphan section is
second to last section and the last section is dead (e.g. because
it's being discarded) introducing a regression in some cases.
This change handles this case by using the same predicate when
checking the last section.
Differential Revision: https://reviews.llvm.org/D39172
llvm-svn: 316307
The support of R_PPC_ADDR16_HI improves ld compatibility and makes
things on par with RuntimeDyldELF that already implements this
relocation.
Patch by vit9696.
llvm-svn: 316306
We used to have a map from section piece offsets to section pieces
as a cache for binary search. But I found that the map took quite a
large amount of memory and didn't make linking faster. So, in this
patch, I removed the map.
This patch saves 566 MiB of RAM (2.019 GiB -> 1.453 GiB) when linking
clang with debug info, and the link time is 4% faster in that test case.
Thanks for Sean Silva for pointing this out.
llvm-svn: 316305
The overflow detection assertions were tautological due to truncation.
Adjust them to no longer be tautological.
Patch by Alex Langford!
llvm-svn: 316303
Summary:
It's unclear if this is the only thing we can do but at least this is consistent with the check
of address space agreement in `isBitCastable`.
The code is used at least in both instcombine and jumpthreading though
I could only find a way to trigger the invalid cast in instcombine.
Reviewers: loladiro, sanjoy, majnemer
Reviewed By: sanjoy
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D34335
llvm-svn: 316302
This avoids a hack for making it a no-op for windows.
Also explicitly check for _WIN32 instead of assuming it.
Differential Revision: https://reviews.llvm.org/D39156
llvm-svn: 316300
As discussed in D39011:
https://reviews.llvm.org/D39011
...replacing constants with a variable is inverting the transform done
by other IR passes, so we definitely don't want to do this early.
In fact, it's questionable whether this transform belongs in SimplifyCFG
at all. I'll look at moving this to codegen as a follow-up step.
llvm-svn: 316298
Summary: test/CodeGen/PowerPC/pr33093.ll uses both powerpc64 (big-endian) and powerpc64le while the former was unsupported.
Subscribers: nemanjai
Differential Revision: https://reviews.llvm.org/D39164
llvm-svn: 316297
This fixes bugzilla 26810
https://bugs.llvm.org/show_bug.cgi?id=26810
This is intended to prevent sequences like:
movl %ebp, 8(%esp) # 4-byte Spill
movl %ecx, %ebp
movl %ebx, %ecx
movl %edi, %ebx
movl %edx, %edi
cltd
idivl %esi
movl %edi, %edx
movl %ebx, %edi
movl %ecx, %ebx
movl %ebp, %ecx
movl 16(%esp), %ebp # 4 - byte Reload
Such sequences are created in 2 scenarios:
Scenario #1:
vreg0 is evicted from physreg0 by vreg1
Evictee vreg0 is intended for region splitting with split candidate physreg0 (the reg vreg0 was evicted from)
Region splitting creates a local interval because of interference with the evictor vreg1 (normally region spliiting creates 2 interval, the "by reg" and "by stack" intervals. Local interval created when interference occurs.)
one of the split intervals ends up evicting vreg2 from physreg1
Evictee vreg2 is intended for region splitting with split candidate physreg1
one of the split intervals ends up evicting vreg3 from physreg2 etc.. until someone spills
Scenario #2
vreg0 is evicted from physreg0 by vreg1
vreg2 is evicted from physreg2 by vreg3 etc
Evictee vreg0 is intended for region splitting with split candidate physreg1
Region splitting creates a local interval because of interference with the evictor vreg1
one of the split intervals ends up evicting back original evictor vreg1 from physreg0 (the reg vreg0 was evicted from)
Another evictee vreg2 is intended for region splitting with split candidate physreg1
one of the split intervals ends up evicting vreg3 from physreg2 etc.. until someone spills
As compile time was a concern, I've added a flag to control weather we do cost calculations for local intervals we expect to be created (it's on by default for X86 target, off for the rest).
Differential Revision: https://reviews.llvm.org/D35816
Change-Id: Id9411ff7bbb845463d289ba2ae97737a1ee7cc39
llvm-svn: 316295
This is similar to how we generate the VEX tables.
More fixes are still needed for the instructions that use EVEX.b (broadcast and embedded rounding).
llvm-svn: 316294
The missed canonicalization/optimization in the motivating test from PR34471 leads to very different codegen:
int switcher(int x) {
switch(x) {
case 17: return 17;
case 19: return 19;
case 42: return 42;
default: break;
}
return 0;
}
int comparator(int x) {
if (x == 17) return 17;
if (x == 19) return 19;
if (x == 42) return 42;
return 0;
}
For the first example, we use a bit-test optimization to avoid a series of compare-and-branch:
https://godbolt.org/g/BivDsw
Differential Revision: https://reviews.llvm.org/D39011
llvm-svn: 316293
In order to identify the copy deduction candidate, I considered two approaches:
- attempt to determine whether an implicit guide is a copy deduction candidate by checking certain properties of its subsituted parameter during overload-resolution.
- using one of the many bits (WillHaveBody) from FunctionDecl (that CXXDeductionGuideDecl inherits from) that are otherwise irrelevant for deduction guides
After some brittle gymnastics w the first strategy, I settled on the second, although to avoid confusion and to give that bit a better name, i turned it into a member of an anonymous union.
Given this identification 'bit', the tweak to overload resolution was a simple reordering of the deduction guide checks (in SemaOverload.cpp::isBetterOverloadCandidate), in-line with Jason Merrill's p0620r0 drafting which made it into the working paper. Concordant with that, I made sure the copy deduction candidate is always added.
References:
See https://bugs.llvm.org/show_bug.cgi?id=34970
See http://wg21.link/p0620r0
llvm-svn: 316292
This patch implements dynamic stack (re-)alignment for 16-bit Thumb. When
targeting processors, which support only the 16-bit Thumb instruction set
the compiler ignores the alignment attributes of automatic variables and may
silently generate incorrect code.
Differential revision: https://reviews.llvm.org/D38143
llvm-svn: 316289
The pass scans the function to find instruction chains that define
registers in the same domain (closures).
It then calculates the cost of converting the closure to another domain.
If found profitable, the instructions are converted to instructions in
the other domain and the register classes are changed accordingly.
This commit adds the pass infrastructure and a simple conversion from
the GPR domain to the Mask domain.
Differential Revision:
https://reviews.llvm.org/D37251
Change-Id: Ic2cf1d76598110401168326d411128ae2580a604
llvm-svn: 316288
By assuming that mergeable input sections are smaller than 4 GiB,
lld's memory usage when linking clang with debug info drops from
2.788 GiB to 2.019 GiB (measured by valgrind, and that does not include
memory space for mmap'ed files). I think that's a reasonable assumption
given such a large RAM savings, so this patch.
According to valgrind, gold needs 3.54 GiB of RAM to do the same thing.
NB: This patch does not introduce a limitation on the size of
output sections. You can still create sections larger than 4 GiB.
llvm-svn: 316280
o) Add a 'Location' class that represents the four properties of a
physical location
o) Enhance 'SourceLocation' to provide 'expansion' and 'spelling'
locations, maintaining backwards compatibility with existing code by
forwarding the four properties to 'expansion'.
o) Update the implementation to use 'clang_getExpansionLocation'
instead of the deprecated 'clang_getInstantiationLocation', which
has been present since 2011.
o) Update the implementation of 'clang_getSpellingLocation' to actually
obtain spelling location instead of file location.
llvm-svn: 316278
This introduces a new operand type to encode the whether the index register should be XMM/YMM/ZMM. And new code to fixup the results created by readSIB.
This has the nice effect of removing a bunch of code that hard coded the name of every GATHER and SCATTER instruction to map the index type.
This fixes PR32807.
llvm-svn: 316273
Summary:
As Mattias Eriksson has reported in PR35009, in C, for enums, the underlying type should
be used when checking for the tautological comparison, unlike C++, where the enumerator
values define the value range. So if not in CPlusPlus mode, use the enum underlying type.
Also, i have discovered a problem (a crash) when evaluating tautological-ness of the following comparison:
```
enum A { A_a = 0 };
if (a < 0) // expected-warning {{comparison of unsigned enum expression < 0 is always false}}
return 0;
```
This affects both the C and C++, but after the first fix, only C++ code was affected.
That was also fixed, while preserving (i think?) the proper diagnostic output.
And while there, attempt to enhance the test coverage.
Yes, some tests got moved around, sorry about that :)
Fixes PR35009
Reviewers: aaron.ballman, rsmith, rjmccall
Reviewed By: aaron.ballman
Subscribers: Rakete1111, efriedma, materi, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D39122
llvm-svn: 316268
Neither of these cases really require a temporary APInt outside the loop. For the ConstantDataSequential case the APInt will never be larger than 64-bits so its fine to just call getElementAsAPInt. For ConstantVector we can get the APInt by reference and only make a copy where the inversion is needed.
llvm-svn: 316265
Some API calls accept 'NULL' instead of a char array (e.g. the second
argument to 'clang_ParseTranslationUnit'). For Python 3 compatibility,
all strings are passed through 'c_interop_string' which expects to
receive only 'bytes' or 'str' objects. This change extends this
behavior to additionally allow 'None' to be supplied.
A test case was added which breaks in Python 3, and is fixed by this
change. All the test cases pass in both, Python 2 and Python 3.
llvm-svn: 316264
Sam Parker