Implements the various TTI functions to enable constant hoisting on PPC. The
only significant test-suite change is this:
MultiSource/Benchmarks/VersaBench/bmm/bmm - 20% speedup
(which essentially reverses the slowdown from r206120).
llvm-svn: 206141
If multiplication involves zero-extended arguments and the result is
compared as in the patterns:
%mul32 = trunc i64 %mul64 to i32
%zext = zext i32 %mul32 to i64
%overflow = icmp ne i64 %mul64, %zext
or
%overflow = icmp ugt i64 %mul64 , 0xffffffff
then the multiplication may be replaced by call to umul.with.overflow.
This change fixes PR4917 and PR4918.
Differential Revision: http://llvm-reviews.chandlerc.com/D2814
llvm-svn: 206137
Originally the cost model would give up for large constants and just return the
maximum cost. This is not what we want for constant hoisting, because some of
these constants are large in bitwidth, but are still cheap to materialize.
This commit fixes the cost model to either return TCC_Free if the cost cannot be
determined, or accurately calculate the cost even for large constants
(bitwidth > 128).
This fixes <rdar://problem/16591573>.
llvm-svn: 206100
The current memory-instruction optimization logic in CGP, which sinks parts of
the address computation that can be adsorbed by the addressing mode, does this
by explicitly converting the relevant part of the address computation into
IR-level integer operations (making use of ptrtoint and inttoptr). For most
targets this is currently not a problem, but for targets wishing to make use of
IR-level aliasing analysis during CodeGen, the use of ptrtoint/inttoptr is a
problem for two reasons:
1. BasicAA becomes less powerful in the face of the ptrtoint/inttoptr
2. In cases where type-punning was used, and BasicAA was used
to override TBAA, BasicAA may no longer do so. (this had forced us to disable
all use of TBAA in CodeGen; something which we can now enable again)
This (use of GEPs instead of ptrtoint/inttoptr) is not currently enabled by
default (except for those targets that use AA during CodeGen), and so aside
from some PowerPC subtargets and SystemZ, there should be no change in
behavior. We may be able to switch completely away from the ptrtoint/inttoptr
sinking on all targets, but further testing is required.
I've doubled-up on a number of existing tests that are sensitive to the
address sinking behavior (including some store-merging tests that are
sensitive to the order of the resulting ADD operations at the SDAG level).
llvm-svn: 206092
The immediate cost calculation code was hitting an assertion in the included
test case, because APInt was still internally 128-bits. Truncating it to 64-bits
fixed the issue.
Fixes <rdar://problem/16572521>.
llvm-svn: 205947
This implements the target-hooks for ARM64 to enable constant hoisting.
This fixes <rdar://problem/14774662> and <rdar://problem/16381500>.
llvm-svn: 205791
into a constant size alloca by inlining.
Ran a run over the testsuite, no results out of the noise, fixes
the testcase in the PR.
PR19115.
llvm-svn: 205710
Some Intrinsics are overloaded to the extent that return type equality (all
that's been checked up to now) does not guarantee that the arguments are the
same. In these cases SLP vectorizer should not recurse into the operands, which
can be achieved by comparing them as "Function *" rather than simply the ID.
llvm-svn: 205424
For the purpose of calculating the cost of the loop at various vectorization
factors, we need to count dependencies of consecutive pointers as uniforms
(which means that the VF = 1 cost is used for all overall VF values).
For example, the TSVC benchmark function s173 has:
...
%3 = add nsw i64 %indvars.iv, 16000
%arrayidx8 = getelementptr inbounds %struct.GlobalData* @global_data, i64 0, i32 0, i64 %3
...
and we must realize that the add will be a scalar in order to correctly deduce
it to be profitable to vectorize this on PowerPC with VSX enabled. In fact, all
dependencies of a consecutive pointer must be a scalar (uniform), and so we
simply need to add all consecutive pointers to the worklist that currently
detects collects uniforms.
Fixes PR19296.
llvm-svn: 205387
This provides an initial implementation of getUnrollingPreferences for x86.
getUnrollingPreferences is used by the generic (concatenation) unroller, which
is distinct from the unrolling done by the loop vectorizer. Many modern x86
cores have some kind of uop cache and loop-stream detector (LSD) used to
efficiently dispatch small loops, and taking full advantage of this requires
unrolling small loops (small here means 10s of uops).
These caches also have limits on the number of taken branches in the loop, and
so we also cap the loop unrolling factor based on the maximum "depth" of the
loop. This is currently calculated with a partial DFS traversal (partial
because it will stop early if the path length grows too much). This is still an
approximation, and one that is both conservative (because it does not account
for branches eliminated via block placement) and optimistic (because it is only
recording the maximum depth over minimum paths). Nevertheless, because the
loops that fit in these uop caches are so small, it is not clear how much the
details matter.
The original set of patches posted for review produced the following test-suite
performance results (from the TSVC benchmark) at that time:
ControlLoops-dbl - 13% speedup
ControlLoops-flt - 15% speedup
Reductions-dbl - 7.5% speedup
llvm-svn: 205348
The generic (concatenation) loop unroller is currently placed early in the
standard optimization pipeline. This is a good place to perform full unrolling,
but not the right place to perform partial/runtime unrolling. However, most
targets don't enable partial/runtime unrolling, so this never mattered.
However, even some x86 cores benefit from partial/runtime unrolling of very
small loops, and follow-up commits will enable this. First, we need to move
partial/runtime unrolling late in the optimization pipeline (importantly, this
is after SLP and loop vectorization, as vectorization can drastically change
the size of a loop), while keeping the full unrolling where it is now. This
change does just that.
llvm-svn: 205264
This reverts commit r205018.
Conflicts:
lib/Transforms/Vectorize/SLPVectorizer.cpp
test/Transforms/SLPVectorizer/X86/insert-element-build-vector.ll
This is breaking libclc build.
llvm-svn: 205260
There is no direct AVX instruction to convert to unsigned. I have some ideas
how we may be able to do this with three vector instructions but the current
backend just bails on this to get it scalarized.
See the comment why we need to adjust the cost returned by BasicTTI.
The test is a bit roundabout (and checks assembly rather than bit code) because
I'd like it to work even if at some point we could vectorize this conversion.
Fixes <rdar://problem/16371920>
llvm-svn: 205159
This adds a second implementation of the AArch64 architecture to LLVM,
accessible in parallel via the "arm64" triple. The plan over the
coming weeks & months is to merge the two into a single backend,
during which time thorough code review should naturally occur.
Everything will be easier with the target in-tree though, hence this
commit.
llvm-svn: 205090
This reverts commit r204912, and follow-up commit r204948.
This introduced a performance regression, and the fix is not completely
clear yet.
llvm-svn: 205010
This reverts commit r203553, and follow-up commits r203558 and r203574.
I will follow this up on the mailinglist to do it in a way that won't
cause subtle PRE bugs.
llvm-svn: 205009
Fixes a miscompile introduced in r204912. It would miscompile code like
(unsigned)(a + -49) <= 5U. The transform would turn this into
(unsigned)a < 55U, which would return true for values in [0, 49], when
it should not.
llvm-svn: 204948
This adds back r204781.
Original message:
Aliases are just another name for a position in a file. As such, the
regular symbol resolutions are not applied. For example, given
define void @my_func() {
ret void
}
@my_alias = alias weak void ()* @my_func
@my_alias2 = alias void ()* @my_alias
We produce without this patch:
.weak my_alias
my_alias = my_func
.globl my_alias2
my_alias2 = my_alias
That is, in the resulting ELF file my_alias, my_func and my_alias are
just 3 names pointing to offset 0 of .text. That is *not* the
semantics of IR linking. For example, linking in a
@my_alias = alias void ()* @other_func
would require the strong my_alias to override the weak one and
my_alias2 would end up pointing to other_func.
There is no way to represent that with aliases being just another
name, so the best solution seems to be to just disallow it, converting
a miscompile into an error.
llvm-svn: 204934
This reverts commit r204781.
I will follow up to with msan folks to see what is what they
were trying to do with aliases to weak aliases.
llvm-svn: 204784
Aliases are just another name for a position in a file. As such, the
regular symbol resolutions are not applied. For example, given
define void @my_func() {
ret void
}
@my_alias = alias weak void ()* @my_func
@my_alias2 = alias void ()* @my_alias
We produce without this patch:
.weak my_alias
my_alias = my_func
.globl my_alias2
my_alias2 = my_alias
That is, in the resulting ELF file my_alias, my_func and my_alias are
just 3 names pointing to offset 0 of .text. That is *not* the
semantics of IR linking. For example, linking in a
@my_alias = alias void ()* @other_func
would require the strong my_alias to override the weak one and
my_alias2 would end up pointing to other_func.
There is no way to represent that with aliases being just another
name, so the best solution seems to be to just disallow it, converting
a miscompile into an error.
llvm-svn: 204781
Summary:
Previously the code didn't check if the before and after types for the
store were pointers to different address spaces. This resulted in
instcombine using a bitcast to convert between pointers to different
address spaces, causing an assertion due to the invalid cast.
It is not be appropriate to use addrspacecast this case because it is
not guaranteed to be a no-op cast. Instead bail out and do not do the
transformation.
CC: llvm-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D3117
llvm-svn: 204733
A PHI node usually has only one value/basic block pair per incoming basic block.
In the case of a switch statement it is possible that a following PHI node may
have more than one such pair per incoming basic block. E.g.:
%0 = phi i64 [ 123456, %case2 ], [ 654321, %Entry ], [ 654321, %Entry ]
This is valid and the verfier doesn't complain, because both values are the
same.
Constant hoisting materializes the constant for each operand separately and the
value is still the same, but the variable names have changed. As a result the
verfier can't recognize anymore that they are the same value and complains.
This fix adds special update code for PHI node in constant hoisting to prevent
this corner case.
This fixes <rdar://problem/16394449>
llvm-svn: 204537
Extend the target hook to take also the operand index into account when
calculating the cost of the constant materialization.
Related to <rdar://problem/16381500>
llvm-svn: 204435
Originally the algorithm would search for expensive constants and track their
users, which could be instructions and constant expressions. This change only
tracks the constants for instructions, but constant expressions are indirectly
covered too. If an operand is an constant expression, then we look through the
expression to find anny expensive constants.
The algorithm keep now track of the instruction and the operand index where the
constant is used. This allows more precise hoisting of constant materialization
code for PHI instructions, because we only hoist to the basic block of the
incoming operand. Before we had to find the idom of all PHI operands and hoist
the materialization code there.
This also makes updating of instructions easier. Before we had to keep track of
the original constant, find it in the instructions, and then replace it. Now we
can just simply update the operand.
Related to <rdar://problem/16381500>
llvm-svn: 204433
Matt Arsenault