Current TargetTransformInfo can support throughput cost model and code size model, but sometimes we also need instruction latency cost model in different optimizations. Hal suggested we need a single public interface to query the different cost of an instruction. So I proposed following interface:
enum TargetCostKind {
TCK_RecipThroughput, ///< Reciprocal throughput.
TCK_Latency, ///< The latency of instruction.
TCK_CodeSize ///< Instruction code size.
};
int getInstructionCost(const Instruction *I, enum TargetCostKind kind) const;
All clients should mainly use this function to query the cost of an instruction, parameter <kind> specifies the desired cost model.
This patch also provides a simple default implementation of getInstructionLatency.
The default getInstructionLatency provides latency numbers for only small number of instruction classes, those latency numbers are only reasonable for modern OOO processors. It can be extended in following ways:
Add more detail into this function.
Add getXXXLatency function and call it from here.
Implement target specific getInstructionLatency function.
Differential Revision: https://reviews.llvm.org/D37170
llvm-svn: 312832
On a Windows bot, I see a FileCheck error where the source being matched
over no longer exists, i.e it seems like it's FileCheck'ing some stale
output:
http://lab.llvm.org:8011/builders/llvm-clang-x86_64-expensive-checks-win/builds/4747
You can see "// CHECK: [[@LINE]]|{{ +}Marker at 19:3 = 1" in the
FileCheck stderr, but that CHECK line doesn't exist.
Remove the input file to FileCheck before running the test, to try and
appease the bot.
llvm-svn: 312825
The various scalar bit operations set SCC,
so one is erased or moved it needs to be recomputed.
Not sure why the existing tests don't fail on this.
llvm-svn: 312819
A coverage segment contains a starting line and column, an execution
count, and some other metadata. Clients of the coverage library use
segments to prepare line-oriented reports.
Users of the coverage library depend on segments being unique and sorted
in source order. Currently this is not guaranteed (this is why the clang
change which introduced deferred regions was reverted).
This commit documents the "unique and sorted" condition and asserts that
it holds. It also fixes the SegmentBuilder so that it produces correct
output in some edge cases.
Testing: I've added unit tests for some edge cases. I've also checked
that the new SegmentBuilder implementation is fully covered. Apart from
running check-profile and the llvm-cov tests, I've successfully used a
stage1 llvm-cov to prepare a coverage report for an instrumented clang
binary.
Differential Revision: https://reviews.llvm.org/D36813
llvm-svn: 312817
Each source region has a start and end location. Report an error when
the end location does not precede the begin location.
The old lineExecutionCounts.covmapping test actually had a buggy source
region in it. This commit introduces a regenerated copy of the coverage
and moves the old copy to malformedRegions.covmapping, for a test.
Differential Revision: https://reviews.llvm.org/D37387
llvm-svn: 312814
Make sure that the text and html emitters always emit the same set of
region markers, and avoid emitting redundant markers for line segments
which don't end on the line they start on.
This is related to D35925, and depends on D36014
Differential Revision: https://reviews.llvm.org/D36020
llvm-svn: 312813
rL312641 Allowed llvm.memcpy/memset/memmove to be tail calls when parent
function return the intrinsics's first argument. However on arm-none-eabi
platform, llvm.memcpy will be expanded to __aeabi_memcpy which doesn't
have return value. The fix is to check the libcall name after expansion
to match "memcpy/memset/memmove" before allowing those intrinsic to be
tail calls.
llvm-svn: 312799
SLP vectorizer supports horizontal reductions for Add/FAdd binary
operations. Patch adds support for horizontal min/max reductions.
Function getReductionCost() is split to getArithmeticReductionCost() for
binary operation reductions and getMinMaxReductionCost() for min/max
reductions.
Patch fixes PR26956.
Differential revision: https://reviews.llvm.org/D27846
llvm-svn: 312791
Summary:
This fixes code-gen for XRay in PPC. The regression wasn't caught by
codegen tests which we add in this change.
What happened was the following:
- For tail exits, we used to unconditionally prepend the returns/exits
with a pseudo-instruction that gets lowered to the instrumentation
sled (and leave the actual return/exit instruction as-is).
- Changes to the XRay instrumentation pass caused the tail exits to
suddenly also emit the tail exit pseudo-instruction, since the check
for whether a return instruction was also a call instruction meant it
was a tail exit instruction.
- None of the tests caught the regression either due to non-existent
tests, or the tests being disabled/removed for continuous breakage.
This change re-introduces some of the basic tests and verifies that
we're back to a state that allows the back-end to generate appropriate
XRay instrumented binaries for PPC in the presence of tail exits.
Reviewers: echristo, timshen
Subscribers: nemanjai, kbarton, llvm-commits
Differential Revision: https://reviews.llvm.org/D37570
llvm-svn: 312772
cover the bitwise operators.
Nothing really exciting here, this just stamps out the rest of the core
operations that can RMW memory and set flags.
Still not implemented here: ADC, SBB. Those will require more
interesting logic to channel the flags *in*, and I'm not currently
planning to try to tackle that. It might be interesting for someone who
wants to improve our code generation for bignum implementations.
Differential Revision: https://reviews.llvm.org/D37141
llvm-svn: 312768
This is required when targeting COFF, as the comdat name must match
one of the names of the symbols in the comdat.
Differential Revision: https://reviews.llvm.org/D37550
llvm-svn: 312767
operands and used flags to support matching immediate operands.
This is a bit trickier than register operands, and we still want to fall
back on a register operands even for things that appear to be
"immediates" when they won't actually select into the operation's
immediate operand. This also requires us to handle things like selecting
`sub` vs. `add` to minimize the number of bits needed to represent the
immediate, and picking the shortest immediate encoding. In order to
that, we in turn need to scan to make sure that CF isn't used as it will
get inverted.
The end result seems very nice though, and we're now generating
optimal instruction sequences for these patterns IMO.
A follow-up patch will further expand this to other operations with RMW
memory operands. But handing `add` and `sub` are useful starting points
to flesh out the machinery and make sure interesting and complex cases
can be handled.
Thanks to Craig Topper who provided a few fixes and improvements to this
patch in addition to the review!
Differential Revision: https://reviews.llvm.org/D37139
llvm-svn: 312764
Most callers were not expecting the exit(0) and trying to exit with a
different value.
This also adds back the call to cl::PrintHelpMessage in llvm-ar.
llvm-svn: 312761
r312318 - Debug info for variables whose type is shrinked to bool
r312325, r312424, r312489 - Test case for r312318
Revision 312318 introduced a null dereference bug.
Details in https://bugs.llvm.org/show_bug.cgi?id=34490
llvm-svn: 312758
As is indexes above SHN_LORESERVE will not be handled correctly because
they'll be treated as indexes of sections rather than special values
that should just be copied. This change adds support to copy them
though.
Patch by Jake Ehrlich
Differential Revision: https://reviews.llvm.org/D37393
llvm-svn: 312756
We already uses pipefail to detect failure of a redirected command, so
the "|| echo failure" construct was unnecessary.
These tests run and pass on Windows now.
llvm-svn: 312747
Right now Symbols must be either undefined or defined in a specific
section. Some symbols have section indexes like SHN_ABS however. This
change adds support for outputting symbols that have such section
indexes.
Patch by Jake Ehrlich
Differential Revision: https://reviews.llvm.org/D37391
llvm-svn: 312745
The tests are filechecking against stderr and use some magic to make stdout go
away and pipe stderr to FileCheck. This broke bots on windows.
llvm-svn: 312739
For now CUDA-9 is not included in the list of CUDA versions clang
searches for, so the path to CUDA-9 must be explicitly passed
via --cuda-path=.
On LLVM side NVPTX added sm_70 GPU type which bumps required
PTX version to 6.0, but otherwise is equivalent to sm_62 at the moment.
Differential Revision: https://reviews.llvm.org/D37576
llvm-svn: 312734
Second try after fixing a code san problem with iterator reference types.
This change introduces a subcommand to the llvm-xray tool called
"stacks" which allows for analysing XRay traces provided as inputs and
accounting time to stacks instead of just individual functions. This
gives us a more precise view of where in a program the latency is
actually attributed.
The tool uses a trie data structure to keep track of the caller-callee
relationships as we process the XRay traces. In particular, we keep
track of the function call stack as we enter functions. While we're
doing this we're adding nodes in a trie and indicating a "calls"
relatinship between the caller (current top of the stack) and the callee
(the new top of the stack). When we push function ids onto the stack, we
keep track of the timestamp (TSC) for the enter event.
When exiting functions, we are able to account the duration by getting
the difference between the timestamp of the exit event and the
corresponding entry event in the stack. This works even if we somehow
miss the exit events for intermediary functions (i.e. if the exit event
is not cleanly associated with the enter event at the top of the stack).
The output of the tool currently provides just the top N leaf functions
that contribute the most latency, and the top N stacks that have the
most frequency. In the future we can provide more sophisticated query
mechanisms and potentially an export to database feature to make offline
analysis of the stack traces possible with existing tools.
Differential revision: D34863
llvm-svn: 312733
These don't add any value as they're just compositions of existing
patterns. However, they can confuse the cost logic in ISel, leading to
duplicated vcvt instructions like in PR33199.
llvm-svn: 312724
This patch expands the support of lowerInterleavedload to {8|16|32}x8i stride 3.
LLVM creates suboptimal shuffle code-gen for AVX2. In overall, this patch is a specific fix for the pattern (Strid=3 VF={8|16|32}) and we plan to include the store (deinterleved side).
The patch goal is to optimize the following sequence:
a0 b0 c0 a1 b1 c1 a2 b2
c2 a3 b3 c3 a4 b4 c4 a5
b5 c5 a6 b6 c6 a7 b7 c7
into
a0 a1 a2 a3 a4 a5 a6 a7
b0 b1 b2 b3 b4 b5 b6 b7
c0 c1 c2 c3 c4 c5 c6 c7
Reviewers
1. zvi
2. igor
3. guyblank
4. dorit
5. Ayal
llvm-svn: 312722
Summary:
For large basic blocks with lots of combinable instructions, the
MachineTraceMetrics computations in MachineCombiner can dominate the compile
time, as computing the trace information is quadratic in the number of
instructions in a BB and it's relevant successors/predecessors.
In most cases, knowing the instruction depth should be enough to make
combination decisions. As we already iterate over all instructions in a basic
block, the instruction depth can be computed incrementally. This reduces the
cost of machine-combine drastically in cases where lots of instructions
are combined. The major drawback is that AFAIK, computing the critical path
length cannot be done incrementally. Therefore we only compute
instruction depths incrementally, for basic blocks with more
instructions than inc_threshold. The -machine-combiner-inc-threshold
option can be used to set the threshold and allows for easier
experimenting and checking if using incremental updates for all basic
blocks has any impact on the performance.
Reviewers: sanjoy, Gerolf, MatzeB, efriedma, fhahn
Reviewed By: fhahn
Subscribers: kiranchandramohan, javed.absar, efriedma, llvm-commits
Differential Revision: https://reviews.llvm.org/D36619
llvm-svn: 312719
Summary:
Add patterns for
fptoui <16 x float> to <16 x i8>
fptoui <16 x float> to <16 x i16>
Reviewers: igorb, delena, craig.topper
Reviewed By: craig.topper
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D37505
llvm-svn: 312704
The current code that handles personality functions when creating a
module summary does not correctly handle the case where a function's
personality function operand refers to the function indirectly
(e.g. via a bitcast). This patch handles such cases by treating
personality function references like any other reference, i.e. by
adding them to the function's reference list. This has the minor side
benefit of allowing personality functions to participate in early
dead stripping.
We do this by calling findRefEdges on the function itself. This way
we also end up handling other function operands (specifically prefix
data and prologue data) for free.
Differential Revision: https://reviews.llvm.org/D37553
llvm-svn: 312698
Globals that are promoted to an ARM constant pool may alias with another
existing constant pool entry. We need to keep a reference to all globals
that were promoted to each constant pool value so that we can emit a
distinct label for each promoted global. These labels are necessary so
that debug info can refer to the promoted global without an undefined
reference during linking.
Patch by Stephen Crane!
llvm-svn: 312692
This change adds support for SHT_REL and SHT_RELA sections in
llvm-objcopy.
Patch by Jake Ehrlich
Differential Revision: https://reviews.llvm.org/D36554
llvm-svn: 312680
This change only treats imported and exports functions and globals
as symbol table entries the object has a "linking" section (i.e. it is
relocatable object file).
In this case all globals must be of type I32 and initialized with
i32.const. This was previously being assumed but not checked for and
was causing a failure on big endian machines due to using the wrong
value of then union.
See: https://bugs.llvm.org/show_bug.cgi?id=34487
Differential Revision: https://reviews.llvm.org/D37497
llvm-svn: 312674
Tail merging can convert an undef use into a normal one when creating a
common tail. Doing so can make the register live out from a block which
previously contained the undef use. To keep the liveness up-to-date,
insert IMPLICIT_DEFs in such blocks when necessary.
To enable this patch the computeLiveIns() function which used to
compute live-ins for a block and set them immediately is split into new
functions:
- computeLiveIns() just computes the live-ins in a LivePhysRegs set.
- addLiveIns() applies the live-ins to a block live-in list.
- computeAndAddLiveIns() is a convenience function combining the other
two functions and behaving like computeLiveIns() before this patch.
Based on a patch by Krzysztof Parzyszek <kparzysz@codeaurora.org>
Differential Revision: https://reviews.llvm.org/D37034
llvm-svn: 312668
Guozhi Wei