The code assumed that we always had a preheader without making the pass
dependent on LoopSimplify.
Thanks to Mattias Eriksson V for reporting this.
llvm-svn: 263173
Looking at the IR definition of a masked load made me realize
there was no reason to use a shuffle here, so we don't need
to convert the format of the mask at all.
llvm-svn: 263167
Generalise the existing SIGN_EXTEND to SIGN_EXTEND_VECTOR_INREG combine to support zero extension as well and get rid of a lot of unnecessary ANY_EXTEND + mask patterns.
Reapplied with a fix for PR26870 (avoid premature use of TargetConstant in ZERO_EXTEND_VECTOR_INREG expansion).
Differential Revision: http://reviews.llvm.org/D17691
llvm-svn: 263159
This patch fixes the problem which occurs when loop-vectorize tries to use @llvm.masked.load/store intrinsic for a non-default addrspace pointer. It fails with "Calling a function with a bad signature!" assertion in CallInst constructor because it tries to pass a non-default addrspace pointer to the pointer argument which has default addrspace.
The fix is to add pointer type as another overloaded type to @llvm.masked.load/store intrinsics.
Reviewed By: reames
Differential Revision: http://reviews.llvm.org/D17270
llvm-svn: 263158
Summary:
Unless we plan to do later postpass metadata linking (ThinLTO special mode),
always invoke metadata materialization at the start of IRLinker::run().
This avoids the need for clients who use lazy metadata loading to
explicitly invoke materializeMetadata before the IRMover, which in
turn invokes IRLinker::run and needs materialized metadata for mapping.
Came up in the context of an LLD issue (D17982).
Reviewers: rafael
Subscribers: silvas, llvm-commits
Differential Revision: http://reviews.llvm.org/D17992
llvm-svn: 263143
Summary:
They correspond to BUFFER_LOAD/STORE_FORMAT_XYZW and will be used by Mesa
to implement the GL_ARB_shader_image_load_store extension.
The intention is that for llvm.amdgcn.buffer.load.format, LLVM will decide
whether one of the _X/_XY/_XYZ opcodes can be used (similar to image sampling
and loads). However, this is not currently implemented.
For llvm.amdgcn.buffer.store, LLVM cannot decide to use one of the "smaller"
opcodes and therefore the intrinsic is overloaded. Currently, only the v4f32
is actually implemented since GLSL also only has a vec4 variant of the store
instructions, although it's conceivable that Mesa will want to be smarter
about this in the future.
BUFFER_LOAD_FORMAT_XYZW is already exposed via llvm.SI.vs.load.input, which
has a legacy name, pretends not to access memory, and does not capture the
full flexibility of the instruction.
Reviewers: arsenm, tstellarAMD, mareko
Subscribers: arsenm, llvm-commits
Differential Revision: http://reviews.llvm.org/D17277
llvm-svn: 263140
When trying to replace an add to esp with pops, we need to choose dead
registers to pop into. Registers clobbered by the call and not imp-def'd
by it should be safe. Except that it's not enough to check the register
itself isn't defined, we also need to make sure no overlapping registers
are defined either.
This fixes PR26711.
Differential Revision: http://reviews.llvm.org/D18029
llvm-svn: 263139
Summary:
Peephole optimization that generates a single TBZ/TBNZ instruction
for test and branch sequences like in the example below. This handles
the cases that miss folding of AND into TBZ/TBNZ during ISelLowering of BR_CC
Examples:
and w8, w8, #0x400
cbnz w8, L1
to
tbnz w8, #10, L1
Reviewers: MatzeB, jmolloy, mcrosier, t.p.northover
Subscribers: aemerson, rengolin, mcrosier, llvm-commits
Differential Revision: http://reviews.llvm.org/D17942
llvm-svn: 263136
This patch adds Cortex-R8 to Target Parser and TableGen.
It also adds CodeGen tests for the build attributes.
Patch by Pablo Barrio.
Differential Revision: http://reviews.llvm.org/D17925
llvm-svn: 263132
This is avoiding a naming conflict with opt and llc.
While opt and llc don't link to LTO usually, users that are building a
monolithic libLLVM.dylib and linking the tools to it would have a
runtime error because of the duplicate cl::opt registration.
From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 263127
The initial change was insufficiently complete for always getting the semantics
of __builtin_longjmp correct. The builtin is translated into a
`tInt_eh_sjlj_longjmp` DAG node. This node set R7 as clobbered. However, the
code would then follow up with a clobber of R11. I had failed to notice the
imp-def,kill on R7 in the isel. Unfortunately, it seems that it is not possible
to conditionalise the Defs list via an !if. Instead, construct a new parallel
WIN node and prefer that when targeting windows. This ensures that we now both
correctly model the __builtin_longjmp as well as construct the frame in a more
ABI conformant manner.
llvm-svn: 263123
of, and I misdiagnosed for months and months.
Andrea has had a patch for this forever, but I just couldn't see how
it was fixing the root cause of the problem. It didn't make sense to me,
even though the patch was perfectly good and the analysis of the actual
failure event was *fantastic*.
Well, I came back to it today because the patch has sat for *far* too
long and needs attention and decided I wouldn't let it go until I really
understood what was going on. After quite some time in the debugger,
I finally realized that in fact I had just missed an important case with
my previous attempt to fix PR22093 in r225149. Not only do we need to
handle loads that won't be split, but stores-of-loads that we won't
split. We *do* actually have enough logic in the presplitting to form
new slices for split stores.... *unless* we decided not to split them!
I'm so sorry that it took me this long to come to the realization that
this is the issue. It seems so obvious in hind sight (of course).
Anyways, the fix becomes *much* smaller and more focused. The fact that
we're left doing integer smashing is related to the FIXME in my original
commit: fundamentally, we're not aggressive about pre-splitting for
loads and stores to the same alloca. If we want to get aggressive about
this, it'll need both what Andrea had put into the proposed fix, but
also a *lot* more logic to essentially iteratively pre-split the alloca
until we can't do any more. As I said in that commit log, its really
unclear that this is the right call. Instead, the integer blending and
letting targets lower this to narrower stores seems slightly better. But
we definitely shouldn't really go down that path just to fix this bug.
Again, tons of thanks are owed to Andrea and others at Sony for working
on this bug. I really should have seen what was going on here and
re-directed them sooner. =////
llvm-svn: 263121
WoA uses r11 as the FP even though it is a pure thumb-2 environment in contrast
to AAPCS which states r7. This adjusts __builtin_longjmp to not clobber r7 and
to properly restore the frame pointer on execution.
llvm-svn: 263118
We already have the instruction extracted into 'I', just cast that to
a store the way we do for loads. Also, we don't enter the if unless SI
is non-null, so don't test it again for null.
I'm pretty sure the entire test there can be nuked, but this is just the
trivial cleanup.
llvm-svn: 263112
actually finish wiring up the old call graph.
There were bugs in the old call graph that hadn't been caught because it
wasn't being tested. It wasn't being tested because it wasn't in the
pipeline system and we didn't have a printing pass to run in tests. This
fixes all of that.
As for why I'm still keeping the old call graph alive its so that I can
port GlobalsAA to the new pass manager with out forking it to work with
the lazy call graph. That's clearly the right eventual design, but it
seems pragmatic to defer that until its necessary. The old call graph
works just fine for GlobalsAA.
llvm-svn: 263104
This patch reorders the combining of target shuffle masks so that when a unary shuffle takes a binary shuffle as its input but only references one of its inputs it can correctly combine into a unary shuffle mask.
This is starting to encroach on the purpose of resolveTargetShuffleInputs, but I don't want to remove it until we definitely know we won't need it for full binary shuffle combining.
There is a lot more work before we can properly support binary target shuffle masks but this was an easy case to add support for.
Differential Revision: http://reviews.llvm.org/D17858
llvm-svn: 263102
location in the opt tool to live along side the analysis in LLVM's
libraries.
No functionality changed here, but this will allow me to port the
printer to the new pass manager as well.
llvm-svn: 263101
There is another pass by the generic name 'CallGraphPrinter' which is
actually just a call graph printer tucked away inside the opt tool. I'd
like to bring it out and make it follow the same patterns as the rest of
the CallGraph code, but doing so would end up conflicting with the name
of the DOT printing pass. So this makes the DOT printing pass name be
more precise.
No functionality changed here.
llvm-svn: 263100
Operation SCALAR_TO_VECTOR for v64i8 and v32i16 should be lowered if BW feature is "on".
Differential Revision: http://reviews.llvm.org/D17994
llvm-svn: 263097
This change adds a support for a preserve_most calling convention to the AArch64 backend, similar to how it was done for X86-64.
There is also a subsequent patch on top of this one to add a tail-calls support for this calling convention.
Differential Revision: http://reviews.llvm.org/D18016
llvm-svn: 263092
MinVecRegSize is currently hardcoded to 128; this patch adds a cl::opt
to allow changing it. I tried not to change any existing behavior for the default
case.
Differential revision: http://reviews.llvm.org/D13278
llvm-svn: 263089
Summary:
This is intended to be a performance flag, on the same level as clang
cc1 option "--disable-free". LLVM will never initialize it by default,
it will be up to the client creating the LLVMContext to request this
behavior. Clang will do it by default in Release build (just like
--disable-free).
"opt" and "llc" can opt-in using -disable-named-value command line
option.
When performing LTO on llvm-tblgen, the initial merging of IR peaks
at 92MB without this patch, and 86MB after this patch,setNameImpl()
drops from 6.5MB to 0.5MB.
The total link time goes from ~29.5s to ~27.8s.
Compared to a compile-time flag (like the IRBuilder one), it performs
very close. I profiled on SROA and obtain these results:
420ms with IRBuilder that preserve name
372ms with IRBuilder that strip name
375ms with IRBuilder that preserve name, and a runtime flag to strip
Reviewers: chandlerc, dexonsmith, bogner
Subscribers: joker.eph, llvm-commits
Differential Revision: http://reviews.llvm.org/D17946
From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 263086
need to be changed for porting to the new pass manager.
Also sink the comment on the ValueTable class back to that class instead
of it dangling on an anonymous namespace.
No functionality changed.
llvm-svn: 263084
This is a fairly straightforward port to the new pass manager with one
exception. It removes a very questionable use of releaseMemory() in
the old pass to invalidate its caches between runs on a function.
I don't think this is really guaranteed to be safe. I've just used the
more direct port to the new PM to address this by nuking the results
object each time the pass runs. While this could cause some minor malloc
traffic increase, I don't expect the compile time performance hit to be
noticable, and it makes the correctness and other aspects of the pass
much easier to reason about. In some cases, it may make things faster by
making the sets and maps smaller with better locality. Indeed, the
measurements collected by Bruno (thanks!!!) show mostly compile time
improvements.
There is sadly very limited testing at this point as there are only two
tests of memdep, and both rely on GVN. I'll be porting GVN next and that
will exercise this heavily though.
Differential Revision: http://reviews.llvm.org/D17962
llvm-svn: 263082
MemoryDependenceAnalysis had a hard-coded exception to the general aliasing rules for malloc and calloc. The reasoning that applied there is equally valid in BasicAA and clarifies the remaining logic in MDA.
In principal, this can expose slightly more optimization opportunities, but since essentially all of our aliasing aware memory optimization passes go through MDA, this will likely be NFC in practice.
Differential Revision: http://reviews.llvm.org/D15912
llvm-svn: 263075
This patch teaches CGP to duplicate addressing mode computations into cold paths (detected via explicit cold attribute on calls) if required to let addressing mode be safely sunk into the basic block containing each load and store.
In general, duplicating code into cold blocks may result in code growth, but should not effect performance. In this case, it's better to duplicate some code than to put extra pressure on the register allocator by making it keep the address through the entirely of the fast path.
This patch only handles addressing computations, but in principal, we could implement a more general cold cold scheduling heuristic which tries to reduce register pressure in the fast path by duplicating code into the cold path. Getting the profitability of the general case right seemed likely to be challenging, so I stuck to the existing case (addressing computation) we already had.
Differential Revision: http://reviews.llvm.org/D17652
llvm-svn: 263074
This patch teaches LICM's implementation of store promotion to exploit the fact that the memory location being accessed might be provable thread local. The fact it's thread local weakens the requirements for where we can insert stores since no other thread can observe the write. This allows us perform store promotion even in cases where the store is not guaranteed to execute in the loop.
Two key assumption worth drawing out is that this assumes a) no-capture is strong enough to imply no-escape, and b) standard allocation functions like malloc, calloc, and operator new return values which can be assumed not to have previously escaped.
In future work, it would be nice to generalize this so that it works without directly seeing the allocation site. I believe that the nocapture return attribute should be suitable for this purpose, but haven't investigated carefully. It's also likely that we could support unescaped allocas with similar reasoning, but since SROA and Mem2Reg should destroy those, they're less interesting than they first might seem.
Differential Revision: http://reviews.llvm.org/D16783
llvm-svn: 263072
The irony of this patch is that one CPU that is affected is AMD Jaguar, and Jaguar
has a completely double-pumped AVX implementation. But getting the cost model to
reflect that is a much bigger problem. The small goal here is simply to improve on
the lie that !AVX2 == SandyBridge.
Differential Revision: http://reviews.llvm.org/D18000
llvm-svn: 263069
Instead of a variable-blend instruction, form a blend with immediate because those are always cheaper.
Differential Revision: http://reviews.llvm.org/D17899
llvm-svn: 263067
When checking whether an smin is positive, we can move the comparison to one of the inputs if the other is known positive. If the known positive one is the min, then the other can't be negative. If the other is the min, then we compute the min.
Differential Revision: http://reviews.llvm.org/D17873
llvm-svn: 263059
I somehow missed this. The case in GCC (global_alloc) was similar to
the new testcase except it had an array of structs rather than a two
dimensional array.
Fixes RP26885.
llvm-svn: 263058
As part of r251146 InstCombine was extended to call computeKnownBits on
every value in the function to determine whether it happens to be
constant. This increases typical compiletime by 1-3% (5% in irgen+opt
time) in my measurements. On the other hand this case did not trigger
once in the whole llvm-testsuite.
This patch introduces the notion of ExpensiveCombines which are only
enabled for OptLevel > 2. I removed the check in InstructionSimplify as
that is called from various places where the OptLevel is not known but
given the rarity of the situation I think a check in InstCombine is
enough.
Differential Revision: http://reviews.llvm.org/D16835
llvm-svn: 263047
This will allow inline assembler code to utilize these features, but no automatic lowering is provided, except for the previously provided @llvm.trap, which lowers to "ta 5".
The change also separates out the different assembly language syntaxes for V8 and V9 Sparc. Previously, only V9 Sparc assembly syntax was provided.
The change also corrects the selection order of trap disassembly, allowing, e.g. "ta %g0 + 15" to be rendered, more readably, as "ta 15", ignoring the %g0 register. This is per the sparc v8 and v9 manuals.
Check-in includes many extra unit tests to check this works correctly on both V8 and V9 Sparc processors.
Code Reviewed at http://reviews.llvm.org/D17960.
llvm-svn: 263044
Adam Nemet