Summary:
The target frame lowering's concrete type is always known in RegisterInfo, yet it's only sometimes devirtualized through a static_cast. This change adds an auto-generated static function <Target>GenRegisterInfo::getFrameLowering(const MachineFunction &MF) which does this devirtualization, and uses this function in all targets which can.
This change was suggested by sunfish in D11070 for WebAssembly, I figure that I may as well improve the other targets while I'm here.
This improves the logic in several ways and is a preparation for
followup patches:
- First perform an analysis and create a list of merge candidates, then
transform. This simplifies the code in that you have don't have to
care to much anymore that you may be holding iterators to
MachineInstrs that get removed.
- Analyze/Transform basic blocks in reverse order. This allows to use
LivePhysRegs to find free registers instead of the RegisterScavenger.
The RegisterScavenger will become less precise in the future as it
relies on the deprecated kill-flags.
- Return the newly created node in MergeOps so there's no need to look
around in the schedule to find it.
- Rename some MBBI iterators to InsertBefore to make their role clear.
- General code cleanup.
James Molloy [Fri, 10 Jul 2015 12:52:00 +0000 (12:52 +0000)]
Add support for fast-math flags to the FCmp instruction.
FCmp behaves a lot like a floating-point binary operator in many ways,
and can benefit from fast-math information. Flags such as nsz and nnan
can affect if this fcmp (in combination with a select) can be treated
as a fminnum/fmaxnum operation.
This adds backwards-compatible bitcode support, IR parsing and writing,
LangRef changes and IRBuilder changes. I'll need to audit InstSimplify
and InstCombine in a followup to find places where flags should be
copied.
Disable loop re-rotation for -Oz (patch by Andrey Turetsky)
After changes in rL231820 loop re-rotation is performed even in -Oz mode. Since loop rotation is disabled for -Oz, it seems loop re-rotation should be disabled too.
Differential Revision: http://reviews.llvm.org/D10961
[llvm-stress] Enhance scalar type selection from command line.
llvm-stress command line options like -generate-x86-fp80 has been replaced with one list-like option -types. E.g. llvm-stress -types=x86_fp80,i100,i256,half. Default types (i1, i8, i16, i32, i64, float, double) are always added at the beginning of that list.
Daniel Jasper [Fri, 10 Jul 2015 07:09:20 +0000 (07:09 +0000)]
Add missing 'const'. I don't think this is strictly required, but some
compiler configuration is giving me an error and it seems to be
recommended anyway.
David Majnemer [Fri, 10 Jul 2015 07:00:44 +0000 (07:00 +0000)]
New EH representation for MSVC compatibility
Summary:
This introduces new instructions neccessary to implement MSVC-compatible
exception handling support. Most of the middle-end and none of the
back-end haven't been audited or updated to take them into account.
[InstCombine] Properly combine metadata when replacing a load with another
Not doing this can lead to misoptimizations down the line, e.g. because
of range metadata on the replacing load excluding values that are valid
for the load that is being replaced.
Summary:
Without this patch, LoopVectorizer in certain cases (see loop-vectorize.ll)
produces code with complex control flow which hurts later optimizations. Since
NVPTX doesn't have vector registers in LLVM's sense
(NVPTXTTI::getRegisterBitWidth(true) == 32), we for now declare no vector
registers to effectively disable loop vectorization.
Alex Lorenz [Thu, 9 Jul 2015 22:23:13 +0000 (22:23 +0000)]
MIR Serialization: Serialize the virtual register definitions.
The virtual registers are serialized using a YAML sequence of YAML inline
mappings. Each mapping has the id of the virtual register and the register
class.
Currently canCheckPtrAtRT returns two flags NeedRTCheck and CanDoRT.
NeedRTCheck says whether we need checks and CanDoRT whether we can
generate the checks. The idea is to encode three states with these:
Need/Can:
(1) false/dont-care: no checks are needed
(2) true/false: we need checks but can't generate them
(3) true/true: we need checks and we can generate them
This is pretty unnecessary since the caller (analyzeLoop) is only
interested in whether we can generate the checks if we actually need
them (i.e. 1 or 3).
So this change cleans up to return just that (CanDoRTIfNeeded) and pulls
all the underlying logic into canCheckPtrAtRT.
By doing all this, we simplify analyzeLoop which is the complex function
in LAA.
There is further room for improvement here by using RtCheck.Need
directly rather than a new local variable NeedRTCheck but that's for a
later patch.
[WinEH] Give up on using CSRs across 32-bit invokes for now
The runtime does not restore CSRs when transferring control back to the
function handling the exception. According to the experts on IRC, LLVM's
register allocator has no way to model register clobbers that only
happen on one edge of the CFG. For now, don't worry about trying to use
the meager three CSRs available on 32-bit X86 and just say that such
invokes preserve nothing.
Alex Lorenz [Thu, 9 Jul 2015 21:21:33 +0000 (21:21 +0000)]
MIR Parser: Report an error when parsing machine function with an empty body.
This commit adds a new error which is reported when the MIR Parser encounters
a machine function without any machine basic blocks. The machine verifier
expects that the machine functions have at least one MBB, and this error will
prevent machine functions without MBBs from reaching the machine verifier and
crashing with an assertion.
This change teaches ImplicitNullChecks to look through "unrelated"
instructions like `inc Reg3` when searching for a load instruction
to convert to a trapping load.
Alex Lorenz [Thu, 9 Jul 2015 19:55:27 +0000 (19:55 +0000)]
MIR Serialization: Serialize the simple MachineFrameInfo attributes.
This commit serializes the 13 scalar boolean and integer attributes from the
MachineFrameInfo class: IsFrameAddressTaken, IsReturnAddressTaken, HasStackMap,
HasPatchPoint, StackSize, OffsetAdjustment, MaxAlignment, AdjustsStack,
HasCalls, MaxCallFrameSize, HasOpaqueSPAdjustment, HasVAStart, and
HasMustTailInVarArgFunc. These attributes are serialized as part
of the frameInfo YAML mapping, which itself is a part of the machine function's
YAML mapping.
Sanjoy Das [Thu, 9 Jul 2015 18:46:12 +0000 (18:46 +0000)]
[IndVars] Try to use existing values in RewriteLoopExitValues.
Summary:
In RewriteLoopExitValues, before expanding out an SCEV expression using
SCEVExpander, try to see if an existing LLVM IR expression already
computes the value we're interested in. If so use that existing
expression.
Apart from reducing IndVars' reliance on the rest of the compilation
pipeline, this also prevents IndVars from concluding some expressions as
"high cost" when they're not. For instance,
`InductiveRangeCheckElimination` often emits code of the following form:
```
len = umin(len_A, len_B)
loop:
...
if (i++ < len)
goto loop
outside_loop:
use(i)
```
`SCEVExpander` refuses to rewrite the use of `i` in `outside_loop`,
since it thinks the value of `i` on loop exit, `len`, is a high cost
expansion since it contains an `umax` in it. With this change,
`IndVars` can see that it can re-use `len` instead of creating a new
expression to compute `umin(len_A, len_B)`.
I considered putting this cleverness in `SCEVExpander`, but I was
worried that it may then have a deterimental effect on other passes
that use it. So I decided it was better to just do this in the one
place where it seems like an obviously good idea, with the intent of
generalizing later if needed.
Pat Gavlin [Thu, 9 Jul 2015 17:40:29 +0000 (17:40 +0000)]
Allow {e,r}bp as the target of {read,write}_register.
This patch allows the read_register and write_register intrinsics to
read/write the RBP/EBP registers on X86 iff the targeted register is
the frame pointer for the containing function.
fix an invisible bug when combining repeated FP divisors
This patch fixes bugs that were exposed by the addition of fast-math-flags in the DAG:
r237046 ( http://reviews.llvm.org/rL237046 ):
1. When replacing a division node, it's not enough to RAUW.
We should call CombineTo() to delete dead nodes and combine again.
2. Because we are changing the DAG, we can't return an empty SDValue
after the transform. As the code comments say:
Visitation implementation - Implement dag node combining for different node types.
The semantics are as follows: Return Value:
SDValue.getNode() == 0 - No change was made
SDValue.getNode() == N - N was replaced, is dead and has been handled.
otherwise - N should be replaced by the returned Operand.
The new test case shows no difference with or without this patch, but it will crash if
we re-apply r237046 or enable FMF via the current -enable-fmf-dag cl::opt.
Don't rely on the DepCands iteration order when constructing checking pointer groups
Summary:
The checking pointer group construction algorithm relied on the iteration on DepCands.
We would need the same leaders across runs and the same iteration order over the underlying std::set for determinism.
This changes the algorithm to process the pointers in the order in which they were added to the runtime check, which is deterministic.
We need to update the tests, since the order in which pointers appear has changed.
No new tests were added, since it is impossible to test for non-determinism.
Add a helper to printing BE of LE depending on the format.
The gnu ar format uses BE numbers. The BSD one uses LE. Add a helper for one or the
other. NFC for now, just removes some noise from the following patch.
Reapply fixed r241790: Fix shift legalization and lowering for big constants.
Summary: If shift amount is a constant value > 64 bit it is handled incorrectly during type legalization and X86 lowering. This patch the type of shift amount argument in function DAGTypeLegalizer::ExpandShiftByConstant from unsigned to APInt.
The nest attribute is currently supported on the x86 (32-bit) and x86-64
backends, but not on ARM (32-bit) or AArch64. This patch adds support for
nest to the AArch64 backend.
Register x18 is used by GCC for this purpose and hence is used here.
As discussed on the GCC mailing list the register choice is an ABI issue
and so choosing the same register as GCC means __builtin_call_with_static_chain
is compatible.
Fix shift legalization and lowering for big constants.
Summary: If shift amount is a constant value > 64 bit it is handled incorrectly during type legalization and X86 lowering. This patch the type of shift amount argument in function DAGTypeLegalizer::ExpandShiftByConstant from unsigned to APInt.
Extended syntax of vector version of getelementptr instruction.
The justification of this change is here: http://lists.cs.uiuc.edu/pipermail/llvmdev/2015-March/082989.html
According to the current GEP syntax, vector GEP requires that each index must be a vector with the same number of elements.
%A = getelementptr i8, <4 x i8*> %ptrs, <4 x i64> %offsets
In this implementation I let each index be or vector or scalar. All vector indices must have the same number of elements. The scalar value will mean the splat vector value.
(1) %A = getelementptr i8, i8* %ptr, <4 x i64> %offsets
or
(2) %A = getelementptr i8, <4 x i8*> %ptrs, i64 %offset
In all cases the %A type is <4 x i8*>
In the case (2) we add the same offset to all pointers.
The case (1) covers C[B[i]] case, when we have the same base C and different offsets B[i].
Adam Nemet [Thu, 9 Jul 2015 06:47:18 +0000 (06:47 +0000)]
[LAA] Rename IsRTNeeded to IsRTCheckAnalysisNeeded
The original name was too close to NeedRTCheck which is what the actual
memcheck analysis returns. This flag, as the new name suggests, is only
used to whether to initiate that analysis.
Also a comment is added to answer one question I had about this code for
a long time. Namely, how does this flag differ from
isDependencyCheckNeeded since they are seemingly set at the same time.
Mehdi Amini [Thu, 9 Jul 2015 02:09:52 +0000 (02:09 +0000)]
Remove getDataLayout() from TargetLowering
Summary:
This change is part of a series of commits dedicated to have a single
DataLayout during compilation by using always the one owned by the
module.
Mehdi Amini [Thu, 9 Jul 2015 02:09:40 +0000 (02:09 +0000)]
Make isLegalAddressingMode() taking DataLayout as an argument
Summary:
This change is part of a series of commits dedicated to have a single
DataLayout during compilation by using always the one owned by the
module.
Mehdi Amini [Thu, 9 Jul 2015 02:09:28 +0000 (02:09 +0000)]
Make getByValTypeAlignment() taking DataLayout as an argument
Summary:
This change is part of a series of commits dedicated to have a single
DataLayout during compilation by using always the one owned by the
module.
Mehdi Amini [Thu, 9 Jul 2015 02:09:20 +0000 (02:09 +0000)]
Make TargetLowering::getShiftAmountTy() taking DataLayout as an argument
Summary:
This change is part of a series of commits dedicated to have a single
DataLayout during compilation by using always the one owned by the
module.
Mehdi Amini [Thu, 9 Jul 2015 02:09:04 +0000 (02:09 +0000)]
Make TargetLowering::getPointerTy() taking DataLayout as an argument
Summary:
This change is part of a series of commits dedicated to have a single
DataLayout during compilation by using always the one owned by the
module.
Mehdi Amini [Thu, 9 Jul 2015 02:08:42 +0000 (02:08 +0000)]
Make TargetTransformInfo keeping a reference to the Module DataLayout
DataLayout is no longer optional. It was initialized with or without
a DataLayout, and the DataLayout when supplied could have been the
one from the TargetMachine.
Summary:
This change is part of a series of commits dedicated to have a single
DataLayout during compilation by using always the one owned by the
module.
Mehdi Amini [Thu, 9 Jul 2015 01:57:34 +0000 (01:57 +0000)]
Redirect DataLayout from TargetMachine to Module in ComputeValueVTs()
Summary:
Avoid using the TargetMachine owned DataLayout and use the Module owned
one instead. This requires passing the DataLayout up the stack to
ComputeValueVTs().
This change is part of a series of commits dedicated to have a single
DataLayout during compilation by using always the one owned by the
module.
David Majnemer [Thu, 9 Jul 2015 00:19:51 +0000 (00:19 +0000)]
[CodeView] Add support for emitting column information
Column information is present in CodeView when the line table subsection
has bit 0 set to 1 in it's flags field. The column information is
represented as a pair of 16-bit quantities: a starting and ending
column. This information is present at the end of the chunk, after all
the line-PC pairs.
Adam Nemet [Wed, 8 Jul 2015 22:58:48 +0000 (22:58 +0000)]
[LAA] Revert a small part of r239295
This commit ([LAA] Fix estimation of number of memchecks) regressed the
logic a bit. We shouldn't quit the analysis if we encounter a pointer
without known bounds *unless* we actually need to emit a memcheck for
it.
The original code was using NumComparisons which is now computed
differently. Instead I compute NeedRTCheck from NumReadPtrChecks and
NumWritePtrChecks.
As side note, I find the separation of NeedRTCheck and CanDoRT
confusing, so I will try to merge them in a follow-up patch.
projects / llvm / log