Simon Pilgrim [Wed, 23 Dec 2015 13:10:07 +0000 (13:10 +0000)]
[X86][AVX] Only shuffle the lower half of vectors if the upper half is undefined
First step towards making better use of AVX's implicit zeroing of the upper half of a 256-bit vector by instructions that only act on the lower 128-bit vector - discussed on D14151.
As well as the fact that 128-bit shuffle instructions are generally more capable, this can be performant for older CPUs with 128-bit ALUs (e.g. Jaguar, Sandy Bridge) that must treat 256-bit vectors as multiple micro-ops.
Moved the similar subvector extraction shuffle combines from PerformShuffleCombine256 to lowerVectorShuffle as well.
Note: I've avoided combining shuffles that reference elements from the upper halves of the input vectors - this may be reviewed in future work as well (AVX1 would probably always gain, but AVX2 does have some cross-lane shuffle instructions).
David Majnemer [Wed, 23 Dec 2015 03:59:04 +0000 (03:59 +0000)]
[WinEH] Don't visit the same catchswitch twice
We visited the same catchswitch twice because it was both the child of
another funclet and the predecessor of a cleanuppad.
Instead, change the numbering algorithm to only recurse if the unwind
destination of the inner funclet agrees with the unwind destination of
the catchswitch.
Nico Weber [Wed, 23 Dec 2015 02:38:31 +0000 (02:38 +0000)]
win: Pass /W4 in front of all the -wd flags.
This should fix many many -Wunused-parameter warnings in self-host builds on
Windows after r255382. cl.exe doesn't care about the order of /W4 and
/wd flags, but clang-cl currently does (just like -Wno-foo -Wall order
matters for clang). We might want to change how clang-cl behaves in
the future, but until then this change makes self-host builds much more
silent.
Philip Reames [Wed, 23 Dec 2015 01:42:15 +0000 (01:42 +0000)]
[GC] Make GCStrategy::isGCManagedPointer a type predicate not a value predicate [NFC]
Reasons:
1) The existing form was a form of false generality. None of the implemented GCStrategies use anything other than a type. Its becoming more and more clear we're going to need some type of strong GC pointer in the type system and we shouldn't pretend otherwise at this point.
2) The API was awkward when applied to vectors-of-pointers. The old one could have been made to work, but calling isGCManagedPointer(Ty->getScalarType()) is much cleaner than the Value alternatives.
3) The rewriting implementation effectively assumes the type based predicate as well. We should be consistent.
Akira Hatanaka [Tue, 22 Dec 2015 23:57:37 +0000 (23:57 +0000)]
Provide a way to specify inliner's attribute compatibility and merging.
This reapplies r256277 with two changes:
- In emitFnAttrCompatCheck, change FuncName's type to std::string to fix
a use-after-free bug.
- Remove an unnecessary install-local target in lib/IR/Makefile.
Original commit message for r252949:
Provide a way to specify inliner's attribute compatibility and merging
rules using table-gen. NFC.
This commit adds new classes CompatRule and MergeRule to Attributes.td,
which are used to generate code to check attribute compatibility and
merge attributes of the caller and callee.
Dan Gohman [Tue, 22 Dec 2015 23:37:37 +0000 (23:37 +0000)]
Add an OperandNamespace field to Target.td's Operand.
For targets to add their own operand types as needed, as advertised in
Operand's comment, they need to be able to specify an alternate namespace
for OperandType names too. This matches the RegisterOperand class.
Changpeng Fang [Tue, 22 Dec 2015 20:55:23 +0000 (20:55 +0000)]
AMDGPU/SI: Use flat for global load/store when targeting HSA
Summary:
For some reason doing executing an MUBUF instruction with the addr64
bit set and a zero base pointer in the resource descriptor causes
the memory operation to be dropped when the shader is executed using
the HSA runtime.
This kind of MUBUF instruction is commonly used when the pointer is
stored in VGPRs. The base pointer field in the resource descriptor
is set to zero and and the pointer is stored in the vaddr field.
This patch resolves the issue by only using flat instructions for
global memory operations when targeting HSA. This is an overly
conservative fix as all other configurations of MUBUF instructions
appear to work.
NOTE: re-commit by fixing a failure in Codegen/AMDGPU/llvm.dbg.value.ll
Akira Hatanaka [Tue, 22 Dec 2015 20:00:05 +0000 (20:00 +0000)]
Provide a way to specify inliner's attribute compatibility and merging.
This reapplies r252990 and r252949. I've added member function getKind
to the Attr classes which returns the enum or string of the attribute.
Original commit message for r252949:
Provide a way to specify inliner's attribute compatibility and merging
rules using table-gen. NFC.
This commit adds new classes CompatRule and MergeRule to Attributes.td,
which are used to generate code to check attribute compatibility and
merge attributes of the caller and callee.
Changpeng Fang [Tue, 22 Dec 2015 19:32:28 +0000 (19:32 +0000)]
AMDGPU/SI: Use flat for global load/store when targeting HSA
Summary:
For some reason doing executing an MUBUF instruction with the addr64
bit set and a zero base pointer in the resource descriptor causes
the memory operation to be dropped when the shader is executed using
the HSA runtime.
This kind of MUBUF instruction is commonly used when the pointer is
stored in VGPRs. The base pointer field in the resource descriptor
is set to zero and and the pointer is stored in the vaddr field.
This patch resolves the issue by only using flat instructions for
global memory operations when targeting HSA. This is an overly
conservative fix as all other configurations of MUBUF instructions
appear to work.
Cong Hou [Tue, 22 Dec 2015 18:56:14 +0000 (18:56 +0000)]
[BPI] Replace weights by probabilities in BPI.
This patch removes all weight-related interfaces from BPI and replace
them by probability versions. With this patch, we won't use edge weight
anymore in either IR or MC passes. Edge probabilitiy is a better
representation in terms of CFG update and validation.
Manuel Jacob [Tue, 22 Dec 2015 16:50:44 +0000 (16:50 +0000)]
[RS4GC] Fix crash in the case that a live variable has a constant base.
Summary:
Previously, RS4GC crashed in CreateGCRelocates() because it assumed
that every base is also in the array of live variables, which isn't true if a
live variable has a constant base.
This change fixes the crash by making sure CreateGCRelocates() won't try to
relocate a live variable with a constant base. This would be unnecessary
anyway because anything with a constant base won't move.
Jun Bum Lim [Tue, 22 Dec 2015 16:36:16 +0000 (16:36 +0000)]
[AArch64] Promote loads from stored
This is a recommit of r256004 which was reverted in r256160. The issue was the
incorrect promotion for half and byte loads transformed into mov instructions.
This fix will replace half and byte type loads only with bit field extracts.
Original commit message:
This change promotes load instructions which directly read from stored by
replacing them with mov instructions. If the store is wider than the load,
the load will be replaced with a bitfield extract.
For example :
STRWui %W1, %X0, 1
%W0 = LDRHHui %X0, 3
becomes
STRWui %W1, %X0, 1
%W0 = UBFMWri %W1, 16, 31
Keno Fischer [Tue, 22 Dec 2015 07:14:50 +0000 (07:14 +0000)]
[ASMPrinter] Fix missing handling of DW_OP_bit_piece
In r256077, I added printing for DIExpressions in DEBUG_VALUE comments,
but neglected to handle DW_OP_bit_piece operands. Thanks to
Mikael Holmen and Joerg Sonnenberger for spotting this.
David Majnemer [Tue, 22 Dec 2015 01:39:04 +0000 (01:39 +0000)]
[MC] Don't use the architecture to govern which object file format to use
InitMCObjectFileInfo was trying to override the triple in awkward ways.
For example, a triple specifying COFF but not Windows was forced as ELF.
This makes it easy for internal invariants to get violated, such as
those which triggered PR25912.
Easwaran Raman [Tue, 22 Dec 2015 00:32:35 +0000 (00:32 +0000)]
Determine callee's hotness and adjust threshold based on that. NFC.
This uses the same criteria used in CFE's CodeGenPGO to identify hot and cold
callees and uses values of inlinehint-threshold and inlinecold-threshold
respectively as the thresholds for such callees.
Evgeniy Stepanov [Tue, 22 Dec 2015 00:13:11 +0000 (00:13 +0000)]
[safestack] Add option for non-TLS unsafe stack pointer.
This patch adds an option, -safe-stack-no-tls, for using normal
storage instead of thread-local storage for the unsafe stack pointer.
This can be useful when SafeStack is applied to an operating system
kernel.
The linker requires that a comdat section must be associated
with a another comdat section that precedes it. This
means the comdat section's name needs to use the profile name
var's name.
David Majnemer [Mon, 21 Dec 2015 22:09:27 +0000 (22:09 +0000)]
[MC, COFF] Support link /incremental conditionally
Today, we always take into account the possibility that object files
produced by MC may be consumed by an incremental linker. This results
in us initialing fields which vary with time (TimeDateStamp) which harms
hermetic builds (e.g. verifying a self-host went well) and produces
sub-optimal code because we cannot assume anything about the relative
position of functions within a section (call sites can get redirected
through incremental linker thunks).
Let's provide an MCTargetOption which controls this behavior so that we
can disable this functionality if we know a-priori that the build will
not rely on /incremental.
Jun Bum Lim [Mon, 21 Dec 2015 22:00:51 +0000 (22:00 +0000)]
Enhance BranchProbabilityInfo::calcUnreachableHeuristics for InvokeInst
This is recommit of r256028 with minor fixes in unittests:
CodeGen/Mips/eh.ll
CodeGen/Mips/insn-zero-size-bb.ll
Original commit message:
When identifying blocks post-dominated by an unreachable-terminated block
in BranchProbabilityInfo, consider only the edge to the normal destination
block if the terminator is InvokeInst and let calcInvokeHeuristics() decide
edge weights for the InvokeInst.
Cong Hou [Mon, 21 Dec 2015 20:42:43 +0000 (20:42 +0000)]
[X86][SSE] Transform truncations between vectors of integers into X86ISD::PACKUS/PACKSS operations during DAG combine.
This patch transforms truncation between vectors of integers into
X86ISD::PACKUS/PACKSS operations during DAG combine. We don't do it in
lowering phase because after type legalization, the original truncation
will be turned into a BUILD_VECTOR with each element that is extracted
from a vector and then truncated, and from them it is difficult to do
this optimization. This greatly improves the performance of truncations
on some specific types.
[PGO] Fix profile var comdat generation problem with COFF
When targeting COFF, it is required that a comdat section to
have a global obj with the same name as the comdat (except for
comdats with select kind to be associative). This fix makes
sure that the comdat is keyed on the data variable for COFF.
Adrian Prantl [Mon, 21 Dec 2015 20:03:00 +0000 (20:03 +0000)]
Fix PR24563 (LiveDebugVariables unconditionally propagates all DBG_VALUEs)
LiveDebugVariables unconditionally propagates all DBG_VALUE down the
dominator tree, which happens to work fine if there already is another
DBG_VALUE or the DBG_VALUE happends to describe a single-assignment vreg
but is otherwise wrong if the DBG_VALUE is coming from only one of the
predecessors.
In r255759 we introduced a proper data flow analysis scheduled after
LiveDebugVariables that correctly propagates DBG_VALUEs across basic block
boundaries. With the new pass in place, the incorrect propagation in
LiveDebugVariables can be retired witout loosing any of the benefits
where LiveDebugVariables happened to do the right thing.
Tom Stellard [Mon, 21 Dec 2015 18:44:27 +0000 (18:44 +0000)]
AMDGPU/SI: Fix encoding for FLAT_SCRATCH registers on VI
Summary:
These register has different encodings on CI and VI, so we add pseudo
FLAT_SCRACTH registers to be used before MC, and subtarget specific
registers to be used by the MC layer.
Matthew Simpson [Mon, 21 Dec 2015 18:31:25 +0000 (18:31 +0000)]
[AArch64] Add additional extract-extend patterns for smov
This patch adds to the target description two additional patterns for matching
extract-extend operations to SMOV. The patterns catch the v16i8-to-i64 and
v8i16-to-i64 cases. The existing patterns miss these cases because the
extracted elements must first be legalized to i32, resulting in any_extend
nodes.
This was originally implemented as a DAG combine (r255895), but was reverted
due to failing out-of-tree tests.
Teresa Johnson [Mon, 21 Dec 2015 15:38:13 +0000 (15:38 +0000)]
Remove overly strict new assert in BitcodeReader.
This fixes a bug introduced by the ThinLTO metadata linking patch
r255909. The assert is overly-strict and while useful in development of
the patch, doesn't seem interesting to keep.
Zlatko Buljan [Mon, 21 Dec 2015 13:08:58 +0000 (13:08 +0000)]
[mips][microMIPS] Implement DERET and DI instructions and check size operand for EXT and DEXT* instructions
Differential Revision: http://reviews.llvm.org/D15570
David Majnemer [Mon, 21 Dec 2015 08:03:07 +0000 (08:03 +0000)]
[MC, COFF] Unbreak support for COFF timestamps
Support for COFF timestamps was unintentionally broken in r246905 when
it was conditionally available depending on whether or not LLVM was
configured with LLVM_ENABLE_TIMESTAMPS. However, Config/config.h was
never included which essentially broke the feature. Due to lax testing,
the breakage was never identified until we observed strange failures
during incremental links of Chromium.
This issue is resolved by simply including Config/config.h in
WinCOFFObjectWriter and teaching lit that the MC/COFF/timestamp.s test
is conditionally supported depending on LLVM_ENABLE_TIMESTAMPS. With
this in place, we can strengthen the test to ensure that it will not
accidentally get broken in the future.
Craig Topper [Sun, 20 Dec 2015 18:41:54 +0000 (18:41 +0000)]
[X86] Prevent constant hoisting for a couple compare immediates that the selection DAG knows how to optimize into a shift.
This allows "icmp ugt %a, 4294967295" and "icmp uge %a, 4294967296" to be optimized into right shifts by 32 which can fold the immediate into the shift instruction. These patterns show up with some regularity in real code.
Unfortunately, since getImmCost can't see the icmp predicate we can't be tell if we're only catching these specific cases.
projects / llvm / log