Sam Kolton [Mon, 17 Jul 2017 14:23:38 +0000 (14:23 +0000)]
[AMDGPU] CodeGen: check dst operand type to determine if omod is supported for VOP3 instructions
Summary:
Previously, CodeGen checked first src operand type to determine if omod is supported by instruction. This isn't correct for some instructions: e.g. V_CMP_EQ_F32 has floating-point src operands but desn't support omod.
Changed .td files to check if dst operand instead of src operand.
Alex Bradbury [Mon, 17 Jul 2017 11:41:30 +0000 (11:41 +0000)]
[YAMLTraits] Add filename support to yaml::Input
Summary:
The current yaml::Input constructor takes a StringRef of data as its
first parameter, discarding any filename information that may have been
present when a YAML file was opened. Add an alterate yaml::Input
constructor that takes a MemoryBufferRef, which can have a filename
associated with it. This leads to clearer diagnostic messages.
[X86] Use MSVC's __cpuidex intrinsic instead of inline assembly in getHostCPUName/getHostCPUFeatures for 32-bit builds too.
We're already using it in 64-bit builds because 64-bit MSVC doesn't support inline assembly.
As far as I know we were using inline assembly because at the time the code was added we had to support MSVC 2008 pre-SP1 while the intrinsic was added to MSVC in SP1. Now that we don't have to support that we should be able to just use the intrinsic.
[X86] X86::CMOV to Branch heuristic based optimization.
LLVM compiler recognizes opportunities to transform a branch into IR select instruction(s) - later it will be lowered into X86::CMOV instruction, assuming no other optimization eliminated the SelectInst.
However, it is not always profitable to emit X86::CMOV instruction. For example, branch is preferable over an X86::CMOV instruction when:
1. Branch is well predicted
2. Condition operand is expensive, compared to True-value and the False-value operands
In CodeGenPrepare pass there is a shallow optimization that tries to convert SelectInst into branch, but it is not enough.
This commit, implements machine optimization pass that converts X86::CMOV instruction(s) into branch, based on a conservative heuristic.
Some platforms have problems with emmiting constructors when class
templates get explicitly instantiated.
This patch fixes the bug reported in D35315 by replacing `= default`
with an empty constructor body.
Teresa Johnson [Sun, 16 Jul 2017 00:28:22 +0000 (00:28 +0000)]
Fix bot failures from r308114
Finally figured out that some bots were failing from r308114
with the message:
llvm-lto2: LTO::run failed: No available targets are compatible with this triple.
after adding in some other checking that finally caused this to show up
in the FileCheck output.
Added "REQUIRES: x86-registered-target" which should fix it.
Teresa Johnson [Sat, 15 Jul 2017 22:58:06 +0000 (22:58 +0000)]
Restore with fix "[ThinLTO] Ensure we always select the same function copy to import"
This restores r308078/r308079 with a fix for bot non-determinisim (make
sure we run llvm-lto in single threaded mode so the debug output doesn't get
interleaved).
[IR] Implement Constant::isNegativeZeroValue/isZeroValue/isAllOnesValue/isOneValue/isMinSignedValue for ConstantDataVector without going through getElementAsConstant
Summary:
Currently these methods call ConstantDataVector::getSplatValue which uses getElementsAsConstant to create a Constant object representing the element value. This method incurs a map lookup to see if we already have created such a Constant before and if not allocates a new Constant object.
This patch changes these methods to use getElementAsAPFloat and getElementAsInteger so we can just examine the data values directly.
[InstCombine] Improve the expansion in SimplifyUsingDistributiveLaws to handle cases where one side doesn't simplify, but the other side resolves to an identity value
Summary:
If one side simplifies to the identity value for inner opcode, we can replace the value with just the operation that can't be simplified.
I've removed a couple now unneeded special cases in visitAnd and visitOr. There are probably other cases I missed.
This fold hit the trifecta:
1. It was untested.
2. It oversteps (multiuse is not checked, so increases instruction count).
3. It is incomplete (doesn't work for vectors).
Revert r308078 (and subsequent tweak in r308079) which introduces a test
that appears to exhibit non-determinism and is flaking on the bots
pretty consistently.
r308078: [ThinLTO] Ensure we always select the same function copy to import
r308079: Require asserts in new test that uses debug flag
[PM/LCG] Teach the LazyCallGraph to maintain reference edges from every
function to every defined function known to LLVM as a library function.
LLVM can introduce calls to these functions either by replacing other
library calls or by recognizing patterns (such as memset_pattern or
vector math patterns) and replacing those with calls. When these library
functions are actually defined in the module, we need to have reference
edges to them initially so that we visit them during the CGSCC walk in
the right order and can effectively rebuild the call graph afterward.
This was discovered when building code with Fortify enabled as that is
a common case of both inline definitions of library calls and
simplifications of code into calling them.
This can in extreme cases of LTO-ing with libc introduce *many* more
reference edges. I discussed a bunch of different options with folks but
all of them are unsatisfying. They either make the graph operations
substantially more complex even when there are *no* defined libfuncs, or
they introduce some other complexity into the callgraph. So this patch
goes with the simplest possible solution of actual synthetic reference
edges. If this proves to be a memory problem, I'm happy to implement one
of the clever techniques to save memory here.
Simon Atanasyan [Sat, 15 Jul 2017 07:14:25 +0000 (07:14 +0000)]
[mips] Handle the `long-calls` feature flags in the MIPS backend
If the `long-calls` feature flags is enabled, disable use of the `jal`
instruction. Instead of that call a function by by first loading its
address into a register, and then using the contents of that register.
Matt Arsenault [Sat, 15 Jul 2017 05:52:59 +0000 (05:52 +0000)]
AMDGPU: Return correct type during argument lowering
The type needs to be casted back to the original argument type.
Fixes an assert that for some reason is only run when
using -debug.
Includes an additional combine to avoid test regressions
from having conversions mixed with multiple Assert[SZ]ext
nodes. On subtargets where i16 is legal, this was producing an i32
register with an i16 AssertZExt, truncated to i16 with another i8
AssertZExt.
Yonghong Song [Sat, 15 Jul 2017 05:41:42 +0000 (05:41 +0000)]
bpf: generate better lowering code for certain select/setcc instructions
Currently, for code like below,
===
inner_map = bpf_map_lookup_elem(outer_map, &port_key);
if (!inner_map) {
inner_map = &fallback_map;
}
===
the compiler generates (pseudo) code like the below:
===
I1: r1 = bpf_map_lookup_elem(outer_map, &port_key);
I2: r2 = 0
I3: if (r1 == r2)
I4: r6 = &fallback_map
I5: ...
===
During kernel verification process, After I1, r1 holds a state
map_ptr_or_null. If I3 condition is not taken
(path [I1, I2, I3, I5]), supposedly r1 should become map_ptr.
Unfortunately, kernel does not recognize this pattern
and r1 remains map_ptr_or_null at insn I5. This will cause
verificaiton failure later on.
Kernel, however, is able to recognize pattern "if (r1 == 0)"
properly and give a map_ptr state to r1 in the above case.
LLVM here generates suboptimal code which causes kernel verification
failure. This patch fixes the issue by changing BPF insn pattern
matching and lowering to generate proper codes if the righthand
parameter of the above condition is a constant. A test case
is also added.
Signed-off-by: Yonghong Song <yhs@fb.com>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@308080 91177308-0d34-0410-b5e6-96231b3b80d8
Teresa Johnson [Sat, 15 Jul 2017 04:53:05 +0000 (04:53 +0000)]
[ThinLTO] Ensure we always select the same function copy to import
Summary:
Check if the first eligible callee is under the instruction threshold.
Checking this on the first eligible callee ensures that we don't end
up selecting different callees to import when we invoke this routine
with different thresholds due to reaching the callee via paths that
are shallower or hotter (when there are multiple copies, i.e. with
weak or linkonce linkage). We don't want to leave the decision of which
copy to import up to the backend.
Now, getUserCost() only checks the src and dst types of EXT to decide it is free
or not. This change first checks the types, then calls isExtFreeImpl(), and
check if EXT can form ExtLoad at last. Currently, only AArch64 has customized
implementation of isExtFreeImpl() to check if EXT can be folded into its use.
Jakub Kuderski [Sat, 15 Jul 2017 01:27:16 +0000 (01:27 +0000)]
[Dominators] Fix reachable visitation and reenable a unit test
This fixes a minor bug in insertion to a reachable node that caused
DominatorTree.InsertDeleteExhaustive flakiness. The patch also adds
a new testcase for this exact failure.
Jakub Kuderski [Fri, 14 Jul 2017 23:49:12 +0000 (23:49 +0000)]
[Dominators] Temporarily disable a flaky unit test
The DominatorTree.InsertDeleteExhaustive uses a RNG with a
constant seed to generate different sequences of updates. The test
fails on some buildbots and this patch disables it for now.
[libFuzzer] Allow non-fuzzer args after -ignore_remaining_args=1
With this change, libFuzzer will ignore any arguments after a sigil
argument, but it will preserve these arguments at the end of the
command line when launching subprocesses. Using this, its possible to
handle positional and single-dash arguments to the program under test
by discarding everything up to -ignore_remaining_args=1 in
LLVMFuzzerInitialize.
Jakub Kuderski [Fri, 14 Jul 2017 21:58:53 +0000 (21:58 +0000)]
[Dominators] Implement incremental deletions
Summary:
This patch implements incremental edge deletions.
It also makes DominatorTreeBase store a pointer to the parent function. The parent function is needed to perform full rebuilts during some deletions, but it is also used to verify that inserted and deleted edges come from the same function.
[AArch64][Falkor] Avoid HW prefetcher tag collisions (step 1)
Summary:
This patch is the first step in reducing HW prefetcher instruction tag
collisions in inner loops for Falkor. It adds a pass that annotates IR
loads with metadata to indicate that they are known to be strided loads,
and adds a target lowering hook that translates this metadata to a
target-specific MachineMemOperand flag.
A follow on change will use this MachineMemOperand flag to re-write
instructions to reduce tag collisions.
Summary:
When checking for memory dependencies between calls using MemorySSA,
handle cases where the calls have no MemoryAccess associated with them
because the AA analysis being used has determined that the call does not
read/write memory.
The Select in the above pattern will be unfolded and then jump-threaded. The
current implementation does not allow CMP in the middle of PHI and Select.
[TableGen][MC] Fix a few places where we didn't hide the underlying type of LaneBitmask very well.
One place compared with 32, which I've replaced with LaneBitmask::BitWidth.
The other places are shifts of a constant 1 by a lane number. But if LaneBitmask were to be a larger type than 32-bits like 64-bits, the 1 would need to be 1ULL to do a 64-bit shift. To hide this I've added a LanebitMask::getLane that hides the shift and make sures the 1 is casted to correct type first.
Jakub Kuderski [Fri, 14 Jul 2017 18:26:09 +0000 (18:26 +0000)]
[Dominators] Make IsPostDominator a template parameter
Summary:
DominatorTreeBase used to have IsPostDominators (bool) member to indicate if the tree is a dominator or a postdominator tree. This made it possible to switch between the two 'modes' at runtime, but it isn't used in practice anywhere.
This patch makes IsPostDominator a template argument. This way, it is easier to switch between different algorithms at compile-time based on this argument and design external utilities around it. It also makes it impossible to incidentally assign a postdominator tree to a dominator tree (and vice versa), and to further simplify template code in GenericDominatorTreeConstruction.
Alfred Huang [Fri, 14 Jul 2017 17:56:55 +0000 (17:56 +0000)]
[AMDGPU] Do not insert an instruction into worklist twice in movetovalu
In moveToVALU(), move to vector ALU is performed, all instrs in
the use chain will be visited. We do not want the same node to be
pushed to the visit worklist more than once.
Jakub Kuderski [Fri, 14 Jul 2017 16:56:35 +0000 (16:56 +0000)]
[Dominators] Simplify block and node printing
Summary:
This patch adds `BlockPrinter`-- a small wrapper for printing CFG nodes and DomTree nodes to `raw_ostream`. It is meant to be only used internally, for debugging and printing errors.
[Hexagon] Add intrinsics for data cache operations
This is the LLVM part, adding definitions for
void @llvm.hexagon.Y2.dccleana(i8*)
void @llvm.hexagon.Y2.dccleaninva(i8*)
void @llvm.hexagon.Y2.dcinva(i8*)
void @llvm.hexagon.Y2.dczeroa(i8*)
void @llvm.hexagon.Y4.l2fetch(i8*, i32)
void @llvm.hexagon.Y5.l2fetch(i8*, i64)
The clang part will follow.
Nirav Dave [Fri, 14 Jul 2017 13:56:21 +0000 (13:56 +0000)]
Improve Aliasing of operations to static alloca
Recommiting after adding check to avoid miscomputing alias information
on addresses of the same base but different subindices.
Memory accesses offset from frame indices may alias, e.g., we
may merge write from function arguments passed on the stack when they
are contiguous. As a result, when checking aliasing, we consider the
underlying frame index's offset from the stack pointer.
Static allocs are realized as stack objects in SelectionDAG, but its
offset is not set until post-DAG causing DAGCombiner's alias check to
consider access to static allocas to frequently alias. Modify isAlias
to consider access between static allocas and access from other frame
objects to be considered aliasing.
Many test changes are included here. Most are fixes for tests which
indirectly relied on our aliasing ability and needed to be modified to
preserve their original intent.
The remaining tests have minor improvements due to relaxed
ordering. The exception is CodeGen/X86/2011-10-19-widen_vselect.ll
which has a minor degradation dispite though the pre-legalized DAG is
improved.
projects / llvm / log