I really needed this, like, factually, yesterday,
when verifying dependency breaking idioms for AMD Zen 3 scheduler model.
Consider the following example:
```
$ ./bin/llvm-exegesis --mode=inverse_throughput --snippets-file=/tmp/snippet.s --num-repetitions=1000000 --repetition-mode=duplicate
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-4a7e50.o
---
mode: inverse_throughput
key:
instructions:
- 'VPXORYrr YMM0 YMM0 YMM0'
config: ''
register_initial_values: []
cpu_name: znver3
llvm_triple: x86_64-unknown-linux-gnu
num_repetitions: 1000000
measurements:
- { key: inverse_throughput, value: 0.31025, per_snippet_value: 0.31025 }
error: ''
info: ''
assembled_snippet: C5FDEFC0C5FDEFC0C5FDEFC0C5FDEFC0C5FDEFC0C5FDEFC0C5FDEFC0C5FDEFC0C5FDEFC0C5FDEFC0C5FDEFC0C5FDEFC0C5FDEFC0C5FDEFC0C5FDEFC0C5FDEFC0C3
...
```
What does it tell us?
So wait, it can only execute ~3 x86 AVX YMM PXOR zero-idioms per cycle?
That doesn't seem right. That's even less than there are pipes supporting this type of op.
Now, second example:
```
$ ./bin/llvm-exegesis --mode=inverse_throughput --snippets-file=/tmp/snippet.s --num-repetitions=1000000 --repetition-mode=loop
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-2418b5.o
---
mode: inverse_throughput
key:
instructions:
- 'VPXORYrr YMM0 YMM0 YMM0'
config: ''
register_initial_values: []
cpu_name: znver3
llvm_triple: x86_64-unknown-linux-gnu
num_repetitions: 1000000
measurements:
- { key: inverse_throughput, value: 1.00011, per_snippet_value: 1.00011 }
error: ''
info: ''
assembled_snippet: 49B80800000000000000C5FDEFC0C5FDEFC04983C0FF75F2C3
...
```
Now that's just worse. Due to the looping, the throughput completely plummeted,
and now we can only do a single instruction/cycle!?
That's not great.
And final example:
```
$ ./bin/llvm-exegesis --mode=inverse_throughput --snippets-file=/tmp/snippet.s --num-repetitions=1000000 --repetition-mode=loop --loop-body-size=1000
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-c402e2.o
---
mode: inverse_throughput
key:
instructions:
- 'VPXORYrr YMM0 YMM0 YMM0'
config: ''
register_initial_values: []
cpu_name: znver3
llvm_triple: x86_64-unknown-linux-gnu
num_repetitions: 1000000
measurements:
- { key: inverse_throughput, value: 0.167087, per_snippet_value: 0.167087 }
error: ''
info: ''
assembled_snippet: 49B80800000000000000C5FDEFC0C5FDEFC04983C0FF75F2C3
...
```
So if we merge the previous two approaches, do duplicate this single-instruction snippet 1000x
(loop-body-size/instruction count in snippet), and run a loop with 1000 iterations
over that duplicated/unrolled snippet, the measured throughput goes through the roof,
up to 5.9 instructions/cycle, which finally tells us that this idiom is zero-cycle!
Reviewed By: courbet
Differential Revision: https://reviews.llvm.org/D102522
CommandLine.h is indirectly included in ~50% of TUs when building
clang, and VirtualFileSystem.h is large.
(Already remarked by jhenderson on D70769.)
No behavior change.
Differential Revision: https://reviews.llvm.org/D100957
This attempts to move all tools over to using `add_llvm_library` for
better consistency. After doing this, I noticed it ended up as nearly a
reimplementation of https://reviews.llvm.org/rL342148, which later got
reverted in r342336 (b09a8c9bd9b819741b38071a7ccd95042ef2643a).
With ccache and ninja on a large core machine (40), I haven't run into
build errors, so I'm hopeful it's better now, though it doesn't seem to
be any different / new.
Reviewed By: stephenneuendorffer
Differential Revision: https://reviews.llvm.org/D90970
As mentioned in TODO comment, casting double to float causes NaNs to change bits.
To avoid the change, this patch adds support for single-floating-point immediate value on MachineCode.
Patch by Yuta Saito.
Differential Revision: https://reviews.llvm.org/D77384
This change compresses the context string by removing cycles due to recursive function for CS profile generation. Removing recursion cycles is a way to normalize the calling context which will be better for the sample aggregation and also make the context promoting deterministic.
Specifically for implementation, we recognize adjacent repeated frames as cycles and deduplicated them through multiple round of iteration.
For example:
Considering a input context string stack:
[“a”, “a”, “b”, “c”, “a”, “b”, “c”, “b”, “c”, “d”]
For first iteration,, it removed all adjacent repeated frames of size 1:
[“a”, “b”, “c”, “a”, “b”, “c”, “b”, “c”, “d”]
For second iteration, it removed all adjacent repeated frames of size 2:
[“a”, “b”, “c”, “a”, “b”, “c”, “d”]
So in the end, we get compressed output:
[“a”, “b”, “c”, “d”]
Compression will be called in two place: one for sample's context key right after unwinding, one is for the eventual context string id in the ProfileGenerator.
Added a switch `compress-recursion` to control the size of duplicated frames, default -1 means no size limit.
Added unit tests and regression test for this.
Differential Revision: https://reviews.llvm.org/D93556
This change compresses the context string by removing cycles due to recursive function for CS profile generation. Removing recursion cycles is a way to normalize the calling context which will be better for the sample aggregation and also make the context promoting deterministic.
Specifically for implementation, we recognize adjacent repeated frames as cycles and deduplicated them through multiple round of iteration.
For example:
Considering a input context string stack:
[“a”, “a”, “b”, “c”, “a”, “b”, “c”, “b”, “c”, “d”]
For first iteration,, it removed all adjacent repeated frames of size 1:
[“a”, “b”, “c”, “a”, “b”, “c”, “b”, “c”, “d”]
For second iteration, it removed all adjacent repeated frames of size 2:
[“a”, “b”, “c”, “a”, “b”, “c”, “d”]
So in the end, we get compressed output:
[“a”, “b”, “c”, “d”]
Compression will be called in two place: one for sample's context key right after unwinding, one is for the eventual context string id in the ProfileGenerator.
Added a switch `compress-recursion` to control the size of duplicated frames, default -1 means no size limit.
Added unit tests and regression test for this.
Differential Revision: https://reviews.llvm.org/D93556
This PR adds more register class support in PowerPC,
mark OperandType for imm and memory operands.
Also added more unit tests for SnippetGenerator.
Reviewed By: #powerpc, steven.zhang
Differential Revision: https://reviews.llvm.org/D88044
Some LLVM unit tests forget to clean up temporary files and
directories. Introduce RAII classes for cleaning them up.
Refactor the tests to use those classes.
Differential Revision: https://reviews.llvm.org/D83228
A perf helper is always only ever cretaed to be checked for validity
then passed as Counter ctor argument, never to be touched again.
Its lifetime should outlive that of the counter, and there is never any
reason to have two different counters of top of the perf helper.
Make sure these assumptions always hold by making the Counter consume the
PerfHelper.
Summary: Commit 63bb9fee525f8f29fd9c2174fa7f15573c3d1fd7 was reverted in
7603bfb4b0a6a90137d47f0182a490fe54bf7ca3 because it broke builds that treat
warnings as errors.
This commit updates the calls to `assembleToStream()` in tests to check that
the return value is valid.
Original commit message:
Followup to D74084.
Replace the use of `report_fatal_error()` with returning the error to
`llvm-exegesis.cpp` and handling it there.
Differential Revision: https://reviews.llvm.org/D74325
function_ref is non-owning, so if we get it as a parameter in constructor,
our reference goes out-of-scope as soon as constructor returns.
Instead, let's just take it as a parameter to the actual `generate()` call
Summary:
Currently, we only have nice exploration for LEA instruction,
while for the rest, we rely on `randomizeUnsetVariables()`
to sometimes generate something interesting.
While that works, it isn't very reliable in coverage :)
Here, i'm making an assumption that while we may want to explore
multi-instruction configs, we are most interested in the
characteristics of the main instruction we were asked about.
Which we can do, by taking the existing `randomizeMCOperand()`,
and turning it on it's head - instead of relying on it to randomly fill
one of the interesting values, let's pregenerate all the possible interesting
values for the variable, and then generate as much `InstructionTemplate`
combinations of these possible values for variables as needed/possible.
Of course, that requires invasive changes to no longer pass just the
naked `Instruction`, but sometimes partially filled `InstructionTemplate`.
As it can be seen from the test, this allows us to explore
`X86::OperandType::OPERAND_COND_CODE` for instructions
that take such an operand.
I'm hoping this will greatly simplify exploration.
Reviewers: courbet, gchatelet
Reviewed By: gchatelet
Subscribers: orodley, mgorny, sdardis, tschuett, jrtc27, atanasyan, mstojanovic, andreadb, RKSimon, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D74156
This is how it should've been and brings it more in line with
std::string_view. There should be no functional change here.
This is mostly mechanical from a custom clang-tidy check, with a lot of
manual fixups. It uncovers a lot of minor inefficiencies.
This doesn't actually modify StringRef yet, I'll do that in a follow-up.
Summary:
... instead of crashing.
On typical exmaple is when there are no available registers.
Reviewers: gchatelet
Subscribers: tschuett, mstojanovic, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D73196
Summary:
Right now when picking a back-to-back instruction at random, we might select
instructions that we do not know how to handle.
Add a ExegesisTarget hook to possibly filter instructions.
Reviewers: gchatelet
Subscribers: tschuett, mstojanovic, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D73161
The addition of `inverse_throughput` mode highlighted the disjointedness
of snippet generators and benchmark runners because it used the
`UopsSnippetGenerator` with the `LatencyBenchmarkRunner`.
To keep the code consistent tie the snippet generators to
parallelization/serialization rather than their benchmark runners.
Renaming `LatencySnippetGenerator` -> `SerialSnippetGenerator`.
Renaming `UopsSnippetGenerator` -> `ParallelSnippetGenerator`.
Differential Revision: https://reviews.llvm.org/D72928
It's been an empty target since r360498 and friends
(`git log --grep='Move InstPrinter files to MCTargetDesc.' llvm/lib/Target`),
but due to hwo the way these targets are structured it was silently
an empty target without anyone noticing.
No behavior change.
Summary:
For builds with LLVM_BUILD_LLVM_DYLIB=ON and BUILD_SHARED_LIBS=OFF
this change makes all symbols in the target specific libraries hidden
by default.
A new macro called LLVM_EXTERNAL_VISIBILITY has been added to mark symbols in these
libraries public, which is mainly needed for the definitions of the
LLVMInitialize* functions.
This patch reduces the number of public symbols in libLLVM.so by about
25%. This should improve load times for the dynamic library and also
make abi checker tools, like abidiff require less memory when analyzing
libLLVM.so
One side-effect of this change is that for builds with
LLVM_BUILD_LLVM_DYLIB=ON and LLVM_LINK_LLVM_DYLIB=ON some unittests that
access symbols that are no longer public will need to be statically linked.
Before and after public symbol counts (using gcc 8.2.1, ld.bfd 2.31.1):
nm before/libLLVM-9svn.so | grep ' [A-Zuvw] ' | wc -l
36221
nm after/libLLVM-9svn.so | grep ' [A-Zuvw] ' | wc -l
26278
Reviewers: chandlerc, beanz, mgorny, rnk, hans
Reviewed By: rnk, hans
Subscribers: merge_guards_bot, luismarques, smeenai, ldionne, lenary, s.egerton, pzheng, sameer.abuasal, MaskRay, wuzish, echristo, Jim, hiraditya, michaelplatings, chapuni, jholewinski, arsenm, dschuff, jyknight, dylanmckay, sdardis, nemanjai, jvesely, javed.absar, sbc100, jgravelle-google, aheejin, kbarton, fedor.sergeev, asb, rbar, johnrusso, simoncook, apazos, sabuasal, niosHD, jrtc27, zzheng, edward-jones, mgrang, atanasyan, rogfer01, MartinMosbeck, brucehoult, the_o, PkmX, jocewei, kristina, jsji, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D54439
Summary: This is a follow up on D71137 properly setting up the AsmTargetStreamer prior to AsmParser::Run call.
Reviewers: courbet, mstojanovic
Subscribers: tschuett, mikhail.ramalho, llvm-commits, petarj, atanasyan
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D71468
Test writing and reading benchmark instructions to and from disc, and
check calculations of min, max and avg values from a list of benchmark
measures.
Differential Revision: https://reviews.llvm.org/D71265
The target does just enough to be able to run llvm-exegesis in latency
mode for at least some opcodes.
Patch by Miloš Stojanović.
Differential Revision: https://reviews.llvm.org/D68649
llvm-svn: 374590
Summary:
This adds a `-max-configs-per-opcode` option to limit the number of
configs per opcode.
Reviewers: gchatelet
Subscribers: tschuett, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68642
llvm-svn: 374054
Summary:
Right now there are no snippet generators that emit the `Config` Field,
but I plan to add it to investigate LEA operands for PR32326.
What was broken was:
- `Config` Was not propagated up until the BenchmarkResult::Key.
- Clustering should really consider different configs as measuring
different things, so we should stabilize on (Opcode, Config) instead of
just Opcode.
Reviewers: gchatelet
Subscribers: tschuett, llvm-commits, lebedev.ri
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68629
llvm-svn: 374031
Summary:
Before this change the Executable function was made by duplicating the
snippet. This change adds a --repetion-mode={loop|duplicate} flag that
allows choosing between this behaviour and wrapping the snippet instructions
in a loop.
The new mode can help measurements when the snippet fits in the DSB by
short-cirtcuiting decoding. The loop adds a dec + jmp to the measurements, but
since these are not part of the critical path, they execute in parallel
with the measured code and do not impact measurements in practice.
Overview of the change:
- New SnippetRepetitor abstraction that handles repeating the snippet.
The assembler delegates repeating the instructions to this class.
- ExegesisTarget learns how to decrement loop counter and jump.
- Some refactoring of the assembler into FunctionFiller/BasicBlockFiller.
Reviewers: gchatelet
Subscribers: mgorny, tschuett, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68125
llvm-svn: 373083
Summary:
Right now latency generation can incorrectly select the scratch register
as a dependency-carrying register.
- Move the logic for preventing register selection from Uops
implementation to common SnippetGenerator class.
- Aliasing detection now takes a set of forbidden registers just like
random register assignment does.
Reviewers: gchatelet
Subscribers: tschuett, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68084
llvm-svn: 373048
This reverts r362990 (git commit 374571301dc8e9bc9fdd1d70f86015de198673bd)
This was causing linker warnings on Darwin:
ld: warning: direct access in function 'llvm::initializeEvexToVexInstPassPass(llvm::PassRegistry&)'
from file '../../lib/libLLVMX86CodeGen.a(X86EvexToVex.cpp.o)' to global weak symbol
'void std::__1::__call_once_proxy<std::__1::tuple<void* (&)(llvm::PassRegistry&),
std::__1::reference_wrapper<llvm::PassRegistry>&&> >(void*)' from file '../../lib/libLLVMCore.a(Verifier.cpp.o)'
means the weak symbol cannot be overridden at runtime. This was likely caused by different translation
units being compiled with different visibility settings.
llvm-svn: 363028
Summary:
For builds with LLVM_BUILD_LLVM_DYLIB=ON and BUILD_SHARED_LIBS=OFF
this change makes all symbols in the target specific libraries hidden
by default.
A new macro called LLVM_EXTERNAL_VISIBILITY has been added to mark symbols in these
libraries public, which is mainly needed for the definitions of the
LLVMInitialize* functions.
This patch reduces the number of public symbols in libLLVM.so by about
25%. This should improve load times for the dynamic library and also
make abi checker tools, like abidiff require less memory when analyzing
libLLVM.so
One side-effect of this change is that for builds with
LLVM_BUILD_LLVM_DYLIB=ON and LLVM_LINK_LLVM_DYLIB=ON some unittests that
access symbols that are no longer public will need to be statically linked.
Before and after public symbol counts (using gcc 8.2.1, ld.bfd 2.31.1):
nm before/libLLVM-9svn.so | grep ' [A-Zuvw] ' | wc -l
36221
nm after/libLLVM-9svn.so | grep ' [A-Zuvw] ' | wc -l
26278
Reviewers: chandlerc, beanz, mgorny, rnk, hans
Reviewed By: rnk, hans
Subscribers: Jim, hiraditya, michaelplatings, chapuni, jholewinski, arsenm, dschuff, jyknight, dylanmckay, sdardis, nemanjai, jvesely, nhaehnle, javed.absar, sbc100, jgravelle-google, aheejin, kbarton, fedor.sergeev, asb, rbar, johnrusso, simoncook, apazos, sabuasal, niosHD, jrtc27, zzheng, edward-jones, mgrang, atanasyan, rogfer01, MartinMosbeck, brucehoult, the_o, PkmX, jocewei, kristina, jsji, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D54439
llvm-svn: 362990
Summary:
Reorder the condition code enum to match their encodings. Move it to MC layer so it can be used by the scheduler models.
This avoids needing an isel pattern for each condition code. And it removes
translation switches for converting between CMOV instructions and condition
codes.
Now the printer, encoder and disassembler take care of converting the immediate.
We use InstAliases to handle the assembly matching. But we print using the
asm string in the instruction definition. The instruction itself is marked
IsCodeGenOnly=1 to hide it from the assembly parser.
This does complicate the scheduler models a little since we can't assign the
A and BE instructions to a separate class now.
I plan to make similar changes for SETcc and Jcc.
Reviewers: RKSimon, spatel, lebedev.ri, andreadb, courbet
Reviewed By: RKSimon
Subscribers: gchatelet, hiraditya, kristina, lebedev.ri, jdoerfert, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D60041
llvm-svn: 357800
Summary:
`ResolvedSchedClass` will need to be used outside of `Analysis`
(before `InstructionBenchmarkClustering` even), therefore promote
it into a non-private top-level class, and while there also
move all of the functions that are only called by `ResolvedSchedClass`
into that same new file.
Reviewers: courbet, gchatelet
Reviewed By: courbet
Subscribers: mgorny, tschuett, mgrang, jdoerfert, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D59993
llvm-svn: 357259
Summary:
This is an alternative to D59539.
Let's suppose we have measured 4 different opcodes, and got: `0.5`, `1.0`, `1.5`, `2.0`.
Let's suppose we are using `-analysis-clustering-epsilon=0.5`.
By default now we will start processing the `0.5` point, find that `1.0` is it's neighbor, add them to a new cluster.
Then we will notice that `1.5` is a neighbor of `1.0` and add it to that same cluster.
Then we will notice that `2.0` is a neighbor of `1.5` and add it to that same cluster.
So all these points ended up in the same cluster.
This may or may not be a correct implementation of dbscan clustering algorithm.
But this is rather horribly broken for the reasons of comparing the clusters with the LLVM sched data.
Let's suppose all those opcodes are currently in the same sched cluster.
If i specify `-analysis-inconsistency-epsilon=0.5`, then no matter
the LLVM values this cluster will **never** match the LLVM values,
and thus this cluster will **always** be displayed as inconsistent.
The solution is obviously to split off some of these opcodes into different sched cluster.
But how do i do that? Out of 4 opcodes displayed in the inconsistency report,
which ones are the "bad ones"? Which ones are the most different from the checked-in data?
I'd need to go in to the `.yaml` and look it up manually.
The trivial solution is to, when creating clusters, don't use the full dbscan algorithm,
but instead "pick some unclustered point, pick all unclustered points that are it's neighbor,
put them all into a new cluster, repeat". And just so as it happens, we can arrive
at that algorithm by not performing the "add neighbors of a neighbor to the cluster" step.
But that won't work well once we teach analyze mode to operate in on-1D mode
(i.e. on more than a single measurement type at a time), because the clustering would
depend on the order of the measurements.
Instead, let's just create a single cluster per opcode, and put all the points of that opcode into said cluster.
And simultaneously check that every point in that cluster is a neighbor of every other point in the cluster,
and if they are not, the cluster (==opcode) is unstable.
This is //yet another// step to bring me closer to being able to continue cleanup of bdver2 sched model..
Fixes [[ https://bugs.llvm.org/show_bug.cgi?id=40880 | PR40880 ]].
Reviewers: courbet, gchatelet
Reviewed By: courbet
Subscribers: tschuett, jdoerfert, RKSimon, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D59820
llvm-svn: 357152
Summary:
This prevents "Cannot encode high byte register in REX-prefixed instruction"
from happening on instructions that require REX encoding when AH & co
get selected.
On the down side, these 4 registers can no longer be selected
automatically, but this avoids having to expose all the X86 encoding
complexity.
Reviewers: gchatelet
Subscribers: tschuett, jdoerfert, llvm-commits, bdb
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D59821
llvm-svn: 357003
Summary: To show that dbscan is insensitive to the order of the points.
Subscribers: tschuett, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D59693
llvm-svn: 356747
That patch is the fix for https://bugs.llvm.org/show_bug.cgi?id=40703
"wrong line number info for obj file compiled with -ffunction-sections"
bug. The problem happened with only .o files. If object file contains
several .text sections then line number information showed incorrectly.
The reason for this is that DwarfLineTable could not detect section which
corresponds to specified address(because address is the local to the
section). And as the result it could not select proper sequence in the
line table. The fix is to pass SectionIndex with the address. So that it
would be possible to differentiate addresses from various sections. With
this fix llvm-objdump shows correct line numbers for disassembled code.
Differential review: https://reviews.llvm.org/D58194
llvm-svn: 354972
