Reed Kotler [Sat, 12 Oct 2013 02:19:08 +0000 (02:19 +0000)]
For Mips16, start to consolidate all forms of 32 bit literal loading so that
they can be better handled and optimized in the Mips16 constant island code.
Andrew Kaylor [Fri, 11 Oct 2013 22:47:10 +0000 (22:47 +0000)]
Fixing problems in lli's RemoteMemoryManager.
This fixes a problem from a previous check-in where a return value was omitted.
Previously the remote/stubs-remote.ll and remote/stubs-sm-pic.ll tests were reporting passes, but they should have been failing. Those tests attempt to link against an external symbol and remote symbol resolution is not supported. The old RemoteMemoryManager implementation resulted in local symbols being used for resolution and the child process crashed but the test didn't notice. With this check-in remote symbol resolution fails, and so the test (correctly) fails.
Matthias Braun [Fri, 11 Oct 2013 19:04:37 +0000 (19:04 +0000)]
Remove kill flags after if conversion if necessary
When if converting something like:
true:
... = R0<kill>
false:
... = R0<kill>
then the instructions of the true block must not have a <kill> flag
anymore, as the instruction of the false block follow and do still read
the R0 value.
Specifically this patch determines the set of register live-in in the
false block (possibly after simulating the liveness changes of the
duplicated instructions). Each of these live-in registers mustn't be
killed.
Stephen Lin [Fri, 11 Oct 2013 18:38:36 +0000 (18:38 +0000)]
Really fix CHECK-LABEL and CHECK-DAG interaction. This actually just restores the initial implementation that was in r186162 but got lost in some subsequent refactoring. More explicit variable names and comments are present now to hopefully prevent repeat regression, as well as another test.
Quentin Colombet [Fri, 11 Oct 2013 18:29:42 +0000 (18:29 +0000)]
[DAGCombiner] Reapply load slicing (192471) with a test that explicitly set sse4.2 support.
This should fix the buildbots.
Original commit message:
[DAGCombiner] Slice a big load in two loads when the element are next to each
other in memory and the target has paired load and performs post-isel loads
combining.
E.g., this optimization will transform something like this:
a = load i64* addr
b = trunc i64 a to i32
c = lshr i64 a, 32
d = trunc i64 c to i32
into:
b = load i32* addr1
d = load i32* addr2
Where addr1 = addr2 +/- sizeof(i32), if the target supports paired load and
performs post-isel loads combining.
One should overload TargetLowering::hasPairedLoad to provide this information.
The default is false.
Quentin Colombet [Fri, 11 Oct 2013 18:01:14 +0000 (18:01 +0000)]
[DAGCombiner] Slice a big load in two loads when the element are next to each
other in memory and the target has paired load and performs post-isel loads
combining.
E.g., this optimization will transform something like this:
a = load i64* addr
b = trunc i64 a to i32
c = lshr i64 a, 32
d = trunc i64 c to i32
into:
b = load i32* addr1
d = load i32* addr2
Where addr1 = addr2 +/- sizeof(i32), if the target supports paired load and
performs post-isel loads combining.
One should overload TargetLowering::hasPairedLoad to provide this information.
The default is false.
Matheus Almeida [Fri, 11 Oct 2013 13:29:36 +0000 (13:29 +0000)]
[mips][msa] Direct Object Emission of INSERT.{B,H,W} instruction.
INSERT is the first type of MSA instruction that requires a change to the way MSA registers are parsed.
This happens because MSA registers may be suffixed by an index in the form of an immediate or a
general purpose register. The changes to parseMSARegs reflect that requirement.
Make AsmPrinter::emitImplicitDef a virtual method so targets can emit custom comments for implicit defs
For NVPTX, this fixes a crash where the emitImplicitDef implementation was expecting physical registers,
while NVPTX uses virtual registers (with a couple of exceptions). Now, the implicit def comment will be
emitted as a true PTX register name. Other targets can use this to customize the output of implicit def
comments.
Robert Lytton [Fri, 11 Oct 2013 10:26:48 +0000 (10:26 +0000)]
XCore target: fix bug in XCoreLowerThreadLocal.cpp
When a ConstantExpr which uses a thread local is part of a PHI node
instruction, the insruction that replaces the ConstantExpr must
be inserted in the predecessor block, in front of the terminator instruction.
If the predecessor block has multiple successors, the edge is first split.
Matthias Braun [Thu, 10 Oct 2013 21:28:47 +0000 (21:28 +0000)]
Refactor LiveInterval: introduce new LiveRange class
LiveRange just manages a list of segments and a list of value numbers
now as LiveInterval did previously, but without having details like spill
weight or a fixed register number.
LiveInterval is now a subclass of LiveRange and simply adds the spill weight
and the register number.
Matthias Braun [Thu, 10 Oct 2013 21:28:43 +0000 (21:28 +0000)]
Rename LiveRange to LiveInterval::Segment
The Segment struct contains a single interval; multiple instances of this struct
are used to construct a live range, but the struct is not a live range by
itself.
Tom Stellard [Thu, 10 Oct 2013 17:11:46 +0000 (17:11 +0000)]
R600/SI: Use -verify-machineinstrs for most tests
We can't enable the verifier for tests with SI_IF and SI_ELSE, because
these instructions are always followed by a COPY which copies their
result to the next basic block. This violates the machine verifier's
rule that non-terminators can not folow terminators.
Reviewed-by: Vincent Lejeune<vljn at ovi.com>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192366 91177308-0d34-0410-b5e6-96231b3b80d8
Tom Stellard [Thu, 10 Oct 2013 17:11:24 +0000 (17:11 +0000)]
R600/SI: Define a separate MIMG instruction for each possible output value type
During instruction selection, we rewrite the destination register
class for MIMG instructions based on their writemasks. This creates
machine verifier errors since the new register class does not match
the register class in the MIMG instruction definition.
We can avoid this by defining different MIMG instructions for each
possible destination type and then switching to the correct instruction
when we change the register class.
Tom Stellard [Thu, 10 Oct 2013 17:11:12 +0000 (17:11 +0000)]
R600: Use StructurizeCFGPass for non SI targets
StructurizeCFG pass allows to make complex cfg reducible ; it allows a lot of
shader from shadertoy (which exhibits complex control flow constructs) to works
correctly with respect to CFG handling (and allow us to detect potential bug in
other part of the backend).
We provide a cmd line argument to disable the pass for debug purpose.
Patch by: Vincent Lejeune
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192363 91177308-0d34-0410-b5e6-96231b3b80d8
we were gathering the {s0, s1} loads below the s3 load. This is fine,
but confused the verifier since now the s3 load had Q0<imp-use> with
no definition above it.
This should mark such uses <undef> as well. The liveness structure at
the beginning and end of the block is unaffected, and the true sN
definitions should prevent any dodgy reorderings being introduced
elsewhere.
Rui Ueyama [Wed, 9 Oct 2013 23:15:49 +0000 (23:15 +0000)]
Revert "llvm-c: Make target initializer functions external functions in lib."
This reverts commit r192316. The original change introduced circular
dependencies between libTarget and backends. That would broke a build unless
link everything into one big binary.
llvm-c: Make target initializer functions external functions in lib.
Making them proper functions defined in the (shared)lib instead of
static inlines defined in the header files makes it possible to
actually distribute a binary compiled against the shared library
without having to worry about getting undefined symbol errors when
calling e.g LLVMInitializeAllTargetInfos because the shared library on
the other system was compiled with different targets.
Shuxin Yang [Wed, 9 Oct 2013 17:21:44 +0000 (17:21 +0000)]
Fix a bug in Dead Argument Elimination.
If a function seen at compile time is not necessarily the one linked to
the binary being built, it is illegal to change the actual arguments
passing to it.
e.g.
--------------------------
void foo(int lol) {
// foo() has linkage satisifying isWeakForLinker()
// "lol" is not used at all.
}
void bar(int lo2) {
// xform to foo(undef) is illegal, as compiler dose not know which
// instance of foo() will be linked to the the binary being built.
foo(lol2);
}
-----------------------------
Such functions can be captured by isWeakForLinker(). NOTE that
mayBeOverridden() is insufficient for this purpose as it dosen't include
linkage types like AvailableExternallyLinkage and LinkOnceODRLinkage.
Take link_odr* as an example, it indicates a set of *EQUIVALENT* globals
that can be merged at link-time. However, the semantic of
*EQUIVALENT*-functions includes parameters. Changing parameters breaks
the assumption.
Thank John McCall for help, especially for the explanation of subtle
difference between linkage types.
Tim Northover [Wed, 9 Oct 2013 07:53:57 +0000 (07:53 +0000)]
AArch64: enable MISched by default.
Substantial SelectionDAG scheduling is going away soon, and is
interfering with Hao's attempts to implement LDn/STn instructions, so
I say we make the leap first.
There were a few reorderings (inevitably) which broke some tests. I
tried to replace them with CHECK-DAG variants mostly, but some too
complex for that to be useful and I just reordered them.
Craig Topper [Wed, 9 Oct 2013 04:54:21 +0000 (04:54 +0000)]
Replace a couple instructions with patterns referring to other instructions with same encoding and operands. Mark a couple other instructions as CodeGenOnly since we have FR and VR instructions and only one of them is needed by the assembler/disassembler.
Craig Topper [Wed, 9 Oct 2013 03:56:16 +0000 (03:56 +0000)]
Mark some instructions as CodeGenOnly since they aren't needed by the assembler or disassembler. Disassembler already filtered them, but asm parser still had them in its tables.
Craig Topper [Wed, 9 Oct 2013 02:18:34 +0000 (02:18 +0000)]
Add in64BitMode/in32BitMode to the MMX/SSE2/AVX maskmovq/dq instructions. This way the asm parser will pick the right one based on the mode. Instruction selection already did the right thing based on the pointer size.
projects / llvm / log