2018-04-04 19:37:06 +08:00
|
|
|
//===-- llvm-exegesis.cpp ---------------------------------------*- C++ -*-===//
|
|
|
|
//
|
2019-01-19 16:50:56 +08:00
|
|
|
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
|
|
|
|
// See https://llvm.org/LICENSE.txt for license information.
|
|
|
|
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
|
2018-04-04 19:37:06 +08:00
|
|
|
//
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
///
|
|
|
|
/// \file
|
|
|
|
/// Measures execution properties (latencies/uops) of an instruction.
|
|
|
|
///
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
2018-05-15 20:08:00 +08:00
|
|
|
#include "lib/Analysis.h"
|
2018-04-04 19:37:06 +08:00
|
|
|
#include "lib/BenchmarkResult.h"
|
|
|
|
#include "lib/BenchmarkRunner.h"
|
2018-05-15 20:08:00 +08:00
|
|
|
#include "lib/Clustering.h"
|
2019-09-30 21:53:50 +08:00
|
|
|
#include "lib/Error.h"
|
2018-04-04 19:37:06 +08:00
|
|
|
#include "lib/LlvmState.h"
|
|
|
|
#include "lib/PerfHelper.h"
|
2019-09-30 20:50:25 +08:00
|
|
|
#include "lib/SnippetFile.h"
|
2019-09-27 20:56:24 +08:00
|
|
|
#include "lib/SnippetRepetitor.h"
|
2018-06-26 16:49:30 +08:00
|
|
|
#include "lib/Target.h"
|
2019-05-22 21:50:16 +08:00
|
|
|
#include "lib/TargetSelect.h"
|
2018-04-04 19:37:06 +08:00
|
|
|
#include "llvm/ADT/StringExtras.h"
|
|
|
|
#include "llvm/ADT/Twine.h"
|
|
|
|
#include "llvm/MC/MCInstBuilder.h"
|
2018-09-25 15:31:44 +08:00
|
|
|
#include "llvm/MC/MCObjectFileInfo.h"
|
|
|
|
#include "llvm/MC/MCParser/MCAsmParser.h"
|
|
|
|
#include "llvm/MC/MCParser/MCTargetAsmParser.h"
|
2018-04-04 19:37:06 +08:00
|
|
|
#include "llvm/MC/MCRegisterInfo.h"
|
2018-05-15 20:08:00 +08:00
|
|
|
#include "llvm/MC/MCSubtargetInfo.h"
|
2018-09-25 15:31:44 +08:00
|
|
|
#include "llvm/Object/ObjectFile.h"
|
2018-04-04 19:37:06 +08:00
|
|
|
#include "llvm/Support/CommandLine.h"
|
2018-05-15 20:08:00 +08:00
|
|
|
#include "llvm/Support/Format.h"
|
2018-04-04 19:37:06 +08:00
|
|
|
#include "llvm/Support/Path.h"
|
2018-09-25 15:31:44 +08:00
|
|
|
#include "llvm/Support/SourceMgr.h"
|
2018-05-15 20:08:00 +08:00
|
|
|
#include "llvm/Support/TargetRegistry.h"
|
2018-04-04 19:37:06 +08:00
|
|
|
#include "llvm/Support/TargetSelect.h"
|
|
|
|
#include <algorithm>
|
|
|
|
#include <string>
|
|
|
|
|
2018-10-23 01:10:47 +08:00
|
|
|
namespace llvm {
|
|
|
|
namespace exegesis {
|
|
|
|
|
2019-03-18 19:32:37 +08:00
|
|
|
static cl::OptionCategory Options("llvm-exegesis options");
|
|
|
|
static cl::OptionCategory BenchmarkOptions("llvm-exegesis benchmark options");
|
|
|
|
static cl::OptionCategory AnalysisOptions("llvm-exegesis analysis options");
|
|
|
|
|
2020-02-13 17:45:15 +08:00
|
|
|
static cl::opt<int> OpcodeIndex(
|
|
|
|
"opcode-index",
|
|
|
|
cl::desc("opcode to measure, by index, or -1 to measure all opcodes"),
|
|
|
|
cl::cat(BenchmarkOptions), cl::init(0));
|
2018-10-23 01:10:47 +08:00
|
|
|
|
|
|
|
static cl::opt<std::string>
|
|
|
|
OpcodeNames("opcode-name",
|
|
|
|
cl::desc("comma-separated list of opcodes to measure, by name"),
|
2019-03-18 19:32:37 +08:00
|
|
|
cl::cat(BenchmarkOptions), cl::init(""));
|
2018-10-23 01:10:47 +08:00
|
|
|
|
|
|
|
static cl::opt<std::string> SnippetsFile("snippets-file",
|
|
|
|
cl::desc("code snippets to measure"),
|
2019-03-18 19:32:37 +08:00
|
|
|
cl::cat(BenchmarkOptions),
|
2018-10-23 01:10:47 +08:00
|
|
|
cl::init(""));
|
|
|
|
|
2019-03-18 19:32:37 +08:00
|
|
|
static cl::opt<std::string>
|
|
|
|
BenchmarkFile("benchmarks-file",
|
|
|
|
cl::desc("File to read (analysis mode) or write "
|
|
|
|
"(latency/uops/inverse_throughput modes) benchmark "
|
|
|
|
"results. “-” uses stdin/stdout."),
|
|
|
|
cl::cat(Options), cl::init(""));
|
2018-10-23 01:10:47 +08:00
|
|
|
|
2019-01-31 00:02:20 +08:00
|
|
|
static cl::opt<exegesis::InstructionBenchmark::ModeE> BenchmarkMode(
|
[llvm-exegesis] Introduce a 'naive' clustering algorithm (PR40880)
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
2019-03-28 16:55:01 +08:00
|
|
|
"mode", cl::desc("the mode to run"), cl::cat(Options),
|
2019-01-31 00:02:20 +08:00
|
|
|
cl::values(clEnumValN(exegesis::InstructionBenchmark::Latency, "latency",
|
|
|
|
"Instruction Latency"),
|
|
|
|
clEnumValN(exegesis::InstructionBenchmark::InverseThroughput,
|
|
|
|
"inverse_throughput",
|
|
|
|
"Instruction Inverse Throughput"),
|
|
|
|
clEnumValN(exegesis::InstructionBenchmark::Uops, "uops",
|
|
|
|
"Uop Decomposition"),
|
|
|
|
// When not asking for a specific benchmark mode,
|
|
|
|
// we'll analyse the results.
|
|
|
|
clEnumValN(exegesis::InstructionBenchmark::Unknown, "analysis",
|
|
|
|
"Analysis")));
|
2018-10-23 01:10:47 +08:00
|
|
|
|
2019-09-27 20:56:24 +08:00
|
|
|
static cl::opt<exegesis::InstructionBenchmark::RepetitionModeE> RepetitionMode(
|
|
|
|
"repetition-mode", cl::desc("how to repeat the instruction snippet"),
|
|
|
|
cl::cat(BenchmarkOptions),
|
[llvm-exegesis] 'Min' repetition mode
Summary:
As noted in documentation, different repetition modes have different trade-offs:
> .. option:: -repetition-mode=[duplicate|loop]
>
> Specify the repetition mode. `duplicate` will create a large, straight line
> basic block with `num-repetitions` copies of the snippet. `loop` will wrap
> the snippet in a loop which will be run `num-repetitions` times. The `loop`
> mode tends to better hide the effects of the CPU frontend on architectures
> that cache decoded instructions, but consumes a register for counting
> iterations.
Indeed. Example:
>>! In D74156#1873657, @lebedev.ri wrote:
> At least for `CMOV`, i'm seeing wildly different results
> | | Latency | RThroughput |
> | duplicate | 1 | 0.8 |
> | loop | 2 | 0.6 |
> where latency=1 seems correct, and i'd expect the througput to be close to 1/2 (since there are two execution units).
This isn't great for analysis, at least for schedule model development.
As discussed in excruciating detail in
>>! In D74156#1924514, @gchatelet wrote:
>>>! In D74156#1920632, @lebedev.ri wrote:
>> ... did that explanation of the question i'm having made any sense?
>
> Thx for digging in the conversation !
> Ok it makes more sense now.
>
> I discussed it a bit with @courbet:
> - We want the analysis tool to stay simple so we'd rather not make it knowledgeable of the repetition mode.
> - We'd like to still be able to select either repetition mode to dig into special cases
>
> So we could add a third `min` repetition mode that would run both and take the minimum. It could be the default option.
> Would you have some time to look what it would take to add this third mode?
there appears to be an agreement that it is indeed sub-par,
and that we should provide an optional, measurement (not analysis!) -time
way to rectify the situation.
However, the solutions isn't entirely straight-forward.
We can just add an actual 'multiplexer' `MinSnippetRepetitor`, because
if we just concatenate snippets produced by `DuplicateSnippetRepetitor`
and `LoopSnippetRepetitor` and run+measure that, the measurement will
naturally be different from what we'd get by running+measuring
them separately and taking the min.
([[ https://www.wolframalpha.com/input/?i=%28x%2By%29%2F2+%21%3D+min%28x%2C+y%29 | `time(D+L)/2 != min(time(D), time(L))` ]])
Also, it seems best to me to have a single snippet instead of generating
a snippet per repetition mode, since the only difference here is that the
loop repetition mode reserves one register for loop counter.
As far as i can tell, we can either teach `BenchmarkRunner::runConfiguration()`
to produce a single report given multiple repetitors (as in the patch),
or do that one layer higher - don't modify `BenchmarkRunner::runConfiguration()`,
produce multiple reports, don't actually print each one, but aggregate them somehow
and only print the final one.
Initially i've gone ahead with the latter approach, but it didn't look like a natural fit;
the former (as in the diff) does seem like a better fit to me.
There's also a question of the test coverage. It sure currently does work here:
```
$ ./bin/llvm-exegesis --opcode-name=CMOV64rr --mode=inverse_throughput --repetition-mode=duplicate
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-8fb949.o
---
mode: inverse_throughput
key:
instructions:
- 'CMOV64rr RAX RAX R11 i_0x0'
- 'CMOV64rr RBP RBP R15 i_0x0'
- 'CMOV64rr RBX RBX RBX i_0x0'
- 'CMOV64rr RCX RCX RBX i_0x0'
- 'CMOV64rr RDI RDI R10 i_0x0'
- 'CMOV64rr RDX RDX RAX i_0x0'
- 'CMOV64rr RSI RSI RAX i_0x0'
- 'CMOV64rr R8 R8 R8 i_0x0'
- 'CMOV64rr R9 R9 RDX i_0x0'
- 'CMOV64rr R10 R10 RBX i_0x0'
- 'CMOV64rr R11 R11 R14 i_0x0'
- 'CMOV64rr R12 R12 R9 i_0x0'
- 'CMOV64rr R13 R13 R12 i_0x0'
- 'CMOV64rr R14 R14 R15 i_0x0'
- 'CMOV64rr R15 R15 R13 i_0x0'
config: ''
register_initial_values:
- 'RAX=0x0'
- 'R11=0x0'
- 'EFLAGS=0x0'
- 'RBP=0x0'
- 'R15=0x0'
- 'RBX=0x0'
- 'RCX=0x0'
- 'RDI=0x0'
- 'R10=0x0'
- 'RDX=0x0'
- 'RSI=0x0'
- 'R8=0x0'
- 'R9=0x0'
- 'R14=0x0'
- 'R12=0x0'
- 'R13=0x0'
cpu_name: bdver2
llvm_triple: x86_64-unknown-linux-gnu
num_repetitions: 10000
measurements:
- { key: inverse_throughput, value: 0.819, per_snippet_value: 12.285 }
error: ''
info: instruction has tied variables, using static renaming.
assembled_snippet: 5541574156415541545348B8000000000000000049BB00000000000000004883EC08C7042400000000C7442404000000009D48BD000000000000000049BF000000000000000048BB000000000000000048B9000000000000000048BF000000000000000049BA000000000000000048BA000000000000000048BE000000000000000049B8000000000000000049B9000000000000000049BE000000000000000049BC000000000000000049BD0000000000000000490F40C3490F40EF480F40DB480F40CB490F40FA480F40D0480F40F04D0F40C04C0F40CA4C0F40D34D0F40DE4D0F40E14D0F40EC4D0F40F74D0F40FD490F40C35B415C415D415E415F5DC3
...
$ ./bin/llvm-exegesis --opcode-name=CMOV64rr --mode=inverse_throughput --repetition-mode=loop
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-051eb3.o
---
mode: inverse_throughput
key:
instructions:
- 'CMOV64rr RAX RAX R11 i_0x0'
- 'CMOV64rr RBP RBP RSI i_0x0'
- 'CMOV64rr RBX RBX R9 i_0x0'
- 'CMOV64rr RCX RCX RSI i_0x0'
- 'CMOV64rr RDI RDI RBP i_0x0'
- 'CMOV64rr RDX RDX R9 i_0x0'
- 'CMOV64rr RSI RSI RDI i_0x0'
- 'CMOV64rr R9 R9 R12 i_0x0'
- 'CMOV64rr R10 R10 R11 i_0x0'
- 'CMOV64rr R11 R11 R9 i_0x0'
- 'CMOV64rr R12 R12 RBP i_0x0'
- 'CMOV64rr R13 R13 RSI i_0x0'
- 'CMOV64rr R14 R14 R14 i_0x0'
- 'CMOV64rr R15 R15 R10 i_0x0'
config: ''
register_initial_values:
- 'RAX=0x0'
- 'R11=0x0'
- 'EFLAGS=0x0'
- 'RBP=0x0'
- 'RSI=0x0'
- 'RBX=0x0'
- 'R9=0x0'
- 'RCX=0x0'
- 'RDI=0x0'
- 'RDX=0x0'
- 'R12=0x0'
- 'R10=0x0'
- 'R13=0x0'
- 'R14=0x0'
- 'R15=0x0'
cpu_name: bdver2
llvm_triple: x86_64-unknown-linux-gnu
num_repetitions: 10000
measurements:
- { key: inverse_throughput, value: 0.6083, per_snippet_value: 8.5162 }
error: ''
info: instruction has tied variables, using static renaming.
assembled_snippet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
...
$ ./bin/llvm-exegesis --opcode-name=CMOV64rr --mode=inverse_throughput --repetition-mode=min
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-c7a47d.o
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-2581f1.o
---
mode: inverse_throughput
key:
instructions:
- 'CMOV64rr RAX RAX R11 i_0x0'
- 'CMOV64rr RBP RBP R10 i_0x0'
- 'CMOV64rr RBX RBX R10 i_0x0'
- 'CMOV64rr RCX RCX RDX i_0x0'
- 'CMOV64rr RDI RDI RAX i_0x0'
- 'CMOV64rr RDX RDX R9 i_0x0'
- 'CMOV64rr RSI RSI RAX i_0x0'
- 'CMOV64rr R9 R9 RBX i_0x0'
- 'CMOV64rr R10 R10 R12 i_0x0'
- 'CMOV64rr R11 R11 RDI i_0x0'
- 'CMOV64rr R12 R12 RDI i_0x0'
- 'CMOV64rr R13 R13 RDI i_0x0'
- 'CMOV64rr R14 R14 R9 i_0x0'
- 'CMOV64rr R15 R15 RBP i_0x0'
config: ''
register_initial_values:
- 'RAX=0x0'
- 'R11=0x0'
- 'EFLAGS=0x0'
- 'RBP=0x0'
- 'R10=0x0'
- 'RBX=0x0'
- 'RCX=0x0'
- 'RDX=0x0'
- 'RDI=0x0'
- 'R9=0x0'
- 'RSI=0x0'
- 'R12=0x0'
- 'R13=0x0'
- 'R14=0x0'
- 'R15=0x0'
cpu_name: bdver2
llvm_triple: x86_64-unknown-linux-gnu
num_repetitions: 10000
measurements:
- { key: inverse_throughput, value: 0.6073, per_snippet_value: 8.5022 }
error: ''
info: instruction has tied variables, using static renaming.
assembled_snippet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
...
```
but i open to suggestions as to how test that.
I also have gone with the suggestion to default to this new mode.
This was irking me for some time, so i'm happy to finally see progress here.
Looking forward to feedback.
Reviewers: courbet, gchatelet
Reviewed By: courbet, gchatelet
Subscribers: mstojanovic, RKSimon, llvm-commits, courbet, gchatelet
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D76921
2020-04-02 14:28:35 +08:00
|
|
|
cl::values(
|
|
|
|
clEnumValN(exegesis::InstructionBenchmark::Duplicate, "duplicate",
|
|
|
|
"Duplicate the snippet"),
|
|
|
|
clEnumValN(exegesis::InstructionBenchmark::Loop, "loop",
|
|
|
|
"Loop over the snippet"),
|
|
|
|
clEnumValN(exegesis::InstructionBenchmark::AggregateMin, "min",
|
|
|
|
"All of the above and take the minimum of measurements")),
|
|
|
|
cl::init(exegesis::InstructionBenchmark::Duplicate));
|
2019-09-27 20:56:24 +08:00
|
|
|
|
2018-10-23 01:10:47 +08:00
|
|
|
static cl::opt<unsigned>
|
2018-04-04 19:37:06 +08:00
|
|
|
NumRepetitions("num-repetitions",
|
2018-10-23 01:10:47 +08:00
|
|
|
cl::desc("number of time to repeat the asm snippet"),
|
2019-03-18 19:32:37 +08:00
|
|
|
cl::cat(BenchmarkOptions), cl::init(10000));
|
2018-04-04 19:37:06 +08:00
|
|
|
|
2019-10-08 22:30:24 +08:00
|
|
|
static cl::opt<unsigned> MaxConfigsPerOpcode(
|
|
|
|
"max-configs-per-opcode",
|
|
|
|
cl::desc(
|
|
|
|
"allow to snippet generator to generate at most that many configs"),
|
|
|
|
cl::cat(BenchmarkOptions), cl::init(1));
|
|
|
|
|
2018-10-23 01:10:47 +08:00
|
|
|
static cl::opt<bool> IgnoreInvalidSchedClass(
|
2018-06-18 19:27:47 +08:00
|
|
|
"ignore-invalid-sched-class",
|
2018-10-23 01:10:47 +08:00
|
|
|
cl::desc("ignore instructions that do not define a sched class"),
|
2019-03-18 19:32:37 +08:00
|
|
|
cl::cat(BenchmarkOptions), cl::init(false));
|
2018-06-18 19:27:47 +08:00
|
|
|
|
[llvm-exegesis] Introduce a 'naive' clustering algorithm (PR40880)
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
2019-03-28 16:55:01 +08:00
|
|
|
static cl::opt<exegesis::InstructionBenchmarkClustering::ModeE>
|
|
|
|
AnalysisClusteringAlgorithm(
|
|
|
|
"analysis-clustering", cl::desc("the clustering algorithm to use"),
|
|
|
|
cl::cat(AnalysisOptions),
|
|
|
|
cl::values(clEnumValN(exegesis::InstructionBenchmarkClustering::Dbscan,
|
|
|
|
"dbscan", "use DBSCAN/OPTICS algorithm"),
|
|
|
|
clEnumValN(exegesis::InstructionBenchmarkClustering::Naive,
|
|
|
|
"naive", "one cluster per opcode")),
|
|
|
|
cl::init(exegesis::InstructionBenchmarkClustering::Dbscan));
|
|
|
|
|
|
|
|
static cl::opt<unsigned> AnalysisDbscanNumPoints(
|
2018-05-15 20:08:00 +08:00
|
|
|
"analysis-numpoints",
|
[llvm-exegesis] Introduce a 'naive' clustering algorithm (PR40880)
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
2019-03-28 16:55:01 +08:00
|
|
|
cl::desc("minimum number of points in an analysis cluster (dbscan only)"),
|
2019-03-18 19:32:37 +08:00
|
|
|
cl::cat(AnalysisOptions), cl::init(3));
|
2018-05-15 20:08:00 +08:00
|
|
|
|
[llvm-exegesis] Split Epsilon param into two (PR40787)
Summary:
This eps param is used for two distinct things:
* initial point clusterization
* checking clusters against the llvm values
What if one wants to only look at highly different clusters, without changing
the clustering itself? In particular, this helps to weed out noisy measurements
(since the clusterization epsilon is still small, so there is a better chance
that noisy measurements from the same opcode will go into different clusters)
By splitting it into two params it is now possible.
This is nearly-free performance-wise:
Old:
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency-1.yaml -analysis-inconsistencies-output-file=/tmp/clusters-old.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 10099 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-old.html'
...
Performance counter stats for './bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency-1.yaml -analysis-inconsistencies-output-file=/tmp/clusters-old.html' (25 runs):
390.01 msec task-clock # 0.998 CPUs utilized ( +- 0.25% )
12 context-switches # 31.735 M/sec ( +- 27.38% )
0 cpu-migrations # 0.000 K/sec
4745 page-faults # 12183.732 M/sec ( +- 0.54% )
1562711900 cycles # 4012303.327 GHz ( +- 0.24% ) (82.90%)
185567822 stalled-cycles-frontend # 11.87% frontend cycles idle ( +- 0.52% ) (83.30%)
392106234 stalled-cycles-backend # 25.09% backend cycles idle ( +- 1.31% ) (33.79%)
1839236666 instructions # 1.18 insn per cycle
# 0.21 stalled cycles per insn ( +- 0.15% ) (50.37%)
407035764 branches # 1045074878.710 M/sec ( +- 0.12% ) (66.80%)
10896459 branch-misses # 2.68% of all branches ( +- 0.17% ) (83.20%)
0.390629 +- 0.000972 seconds time elapsed ( +- 0.25% )
```
```
$ perf stat -r 9 ./bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency.yml -analysis-inconsistencies-output-file=/tmp/clusters-old.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 50572 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-old.html'
...
Performance counter stats for './bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency.yml -analysis-inconsistencies-output-file=/tmp/clusters-old.html' (9 runs):
6803.36 msec task-clock # 0.999 CPUs utilized ( +- 0.96% )
262 context-switches # 38.546 M/sec ( +- 23.06% )
0 cpu-migrations # 0.065 M/sec ( +- 76.03% )
13287 page-faults # 1953.206 M/sec ( +- 0.32% )
27252537904 cycles # 4006024.257 GHz ( +- 0.95% ) (83.31%)
1496314935 stalled-cycles-frontend # 5.49% frontend cycles idle ( +- 0.97% ) (83.32%)
16128404524 stalled-cycles-backend # 59.18% backend cycles idle ( +- 0.30% ) (33.37%)
17611143370 instructions # 0.65 insn per cycle
# 0.92 stalled cycles per insn ( +- 0.05% ) (50.04%)
3894906599 branches # 572537147.437 M/sec ( +- 0.03% ) (66.69%)
116314514 branch-misses # 2.99% of all branches ( +- 0.20% ) (83.35%)
6.8118 +- 0.0689 seconds time elapsed ( +- 1.01%)
```
New:
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency-1.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 10099 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new.html'
...
Performance counter stats for './bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency-1.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new.html' (25 runs):
400.14 msec task-clock # 0.998 CPUs utilized ( +- 0.66% )
12 context-switches # 29.429 M/sec ( +- 25.95% )
0 cpu-migrations # 0.100 M/sec ( +-100.00% )
4714 page-faults # 11796.496 M/sec ( +- 0.55% )
1603131306 cycles # 4011840.105 GHz ( +- 0.66% ) (82.85%)
199538509 stalled-cycles-frontend # 12.45% frontend cycles idle ( +- 2.40% ) (83.10%)
402249109 stalled-cycles-backend # 25.09% backend cycles idle ( +- 1.19% ) (34.05%)
1847783963 instructions # 1.15 insn per cycle
# 0.22 stalled cycles per insn ( +- 0.18% ) (50.64%)
407162722 branches # 1018925730.631 M/sec ( +- 0.12% ) (67.02%)
10932779 branch-misses # 2.69% of all branches ( +- 0.51% ) (83.28%)
0.40077 +- 0.00267 seconds time elapsed ( +- 0.67% )
lebedevri@pini-pini:/build/llvm-build-Clang-release$ perf stat -r 9 ./bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency.yml -analysis-inconsistencies-output-file=/tmp/clusters-new.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 50572 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new.html'
...
Performance counter stats for './bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency.yml -analysis-inconsistencies-output-file=/tmp/clusters-new.html' (9 runs):
6947.79 msec task-clock # 1.000 CPUs utilized ( +- 0.90% )
217 context-switches # 31.236 M/sec ( +- 36.16% )
1 cpu-migrations # 0.096 M/sec ( +- 50.00% )
13258 page-faults # 1908.389 M/sec ( +- 0.34% )
27830796523 cycles # 4006032.286 GHz ( +- 0.89% ) (83.30%)
1504554006 stalled-cycles-frontend # 5.41% frontend cycles idle ( +- 2.10% ) (83.32%)
16716574843 stalled-cycles-backend # 60.07% backend cycles idle ( +- 0.65% ) (33.38%)
17755545931 instructions # 0.64 insn per cycle
# 0.94 stalled cycles per insn ( +- 0.09% ) (50.04%)
3897255686 branches # 560980426.597 M/sec ( +- 0.06% ) (66.70%)
117045395 branch-misses # 3.00% of all branches ( +- 0.47% ) (83.34%)
6.9507 +- 0.0627 seconds time elapsed ( +- 0.90% )
```
I.e. it's +2.6% slowdown for one whole sweep, or +2% for 5 whole sweeps.
Within noise i'd say.
Should help with [[ https://bugs.llvm.org/show_bug.cgi?id=40787 | PR40787 ]].
Reviewers: courbet, gchatelet
Reviewed By: courbet
Subscribers: tschuett, RKSimon, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D58476
llvm-svn: 354767
2019-02-25 17:36:12 +08:00
|
|
|
static cl::opt<float> AnalysisClusteringEpsilon(
|
|
|
|
"analysis-clustering-epsilon",
|
[llvm-exegesis] Introduce a 'naive' clustering algorithm (PR40880)
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
2019-03-28 16:55:01 +08:00
|
|
|
cl::desc("epsilon for benchmark point clustering"),
|
2019-03-18 19:32:37 +08:00
|
|
|
cl::cat(AnalysisOptions), cl::init(0.1));
|
[llvm-exegesis] Split Epsilon param into two (PR40787)
Summary:
This eps param is used for two distinct things:
* initial point clusterization
* checking clusters against the llvm values
What if one wants to only look at highly different clusters, without changing
the clustering itself? In particular, this helps to weed out noisy measurements
(since the clusterization epsilon is still small, so there is a better chance
that noisy measurements from the same opcode will go into different clusters)
By splitting it into two params it is now possible.
This is nearly-free performance-wise:
Old:
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency-1.yaml -analysis-inconsistencies-output-file=/tmp/clusters-old.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 10099 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-old.html'
...
Performance counter stats for './bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency-1.yaml -analysis-inconsistencies-output-file=/tmp/clusters-old.html' (25 runs):
390.01 msec task-clock # 0.998 CPUs utilized ( +- 0.25% )
12 context-switches # 31.735 M/sec ( +- 27.38% )
0 cpu-migrations # 0.000 K/sec
4745 page-faults # 12183.732 M/sec ( +- 0.54% )
1562711900 cycles # 4012303.327 GHz ( +- 0.24% ) (82.90%)
185567822 stalled-cycles-frontend # 11.87% frontend cycles idle ( +- 0.52% ) (83.30%)
392106234 stalled-cycles-backend # 25.09% backend cycles idle ( +- 1.31% ) (33.79%)
1839236666 instructions # 1.18 insn per cycle
# 0.21 stalled cycles per insn ( +- 0.15% ) (50.37%)
407035764 branches # 1045074878.710 M/sec ( +- 0.12% ) (66.80%)
10896459 branch-misses # 2.68% of all branches ( +- 0.17% ) (83.20%)
0.390629 +- 0.000972 seconds time elapsed ( +- 0.25% )
```
```
$ perf stat -r 9 ./bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency.yml -analysis-inconsistencies-output-file=/tmp/clusters-old.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 50572 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-old.html'
...
Performance counter stats for './bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency.yml -analysis-inconsistencies-output-file=/tmp/clusters-old.html' (9 runs):
6803.36 msec task-clock # 0.999 CPUs utilized ( +- 0.96% )
262 context-switches # 38.546 M/sec ( +- 23.06% )
0 cpu-migrations # 0.065 M/sec ( +- 76.03% )
13287 page-faults # 1953.206 M/sec ( +- 0.32% )
27252537904 cycles # 4006024.257 GHz ( +- 0.95% ) (83.31%)
1496314935 stalled-cycles-frontend # 5.49% frontend cycles idle ( +- 0.97% ) (83.32%)
16128404524 stalled-cycles-backend # 59.18% backend cycles idle ( +- 0.30% ) (33.37%)
17611143370 instructions # 0.65 insn per cycle
# 0.92 stalled cycles per insn ( +- 0.05% ) (50.04%)
3894906599 branches # 572537147.437 M/sec ( +- 0.03% ) (66.69%)
116314514 branch-misses # 2.99% of all branches ( +- 0.20% ) (83.35%)
6.8118 +- 0.0689 seconds time elapsed ( +- 1.01%)
```
New:
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency-1.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 10099 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new.html'
...
Performance counter stats for './bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency-1.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new.html' (25 runs):
400.14 msec task-clock # 0.998 CPUs utilized ( +- 0.66% )
12 context-switches # 29.429 M/sec ( +- 25.95% )
0 cpu-migrations # 0.100 M/sec ( +-100.00% )
4714 page-faults # 11796.496 M/sec ( +- 0.55% )
1603131306 cycles # 4011840.105 GHz ( +- 0.66% ) (82.85%)
199538509 stalled-cycles-frontend # 12.45% frontend cycles idle ( +- 2.40% ) (83.10%)
402249109 stalled-cycles-backend # 25.09% backend cycles idle ( +- 1.19% ) (34.05%)
1847783963 instructions # 1.15 insn per cycle
# 0.22 stalled cycles per insn ( +- 0.18% ) (50.64%)
407162722 branches # 1018925730.631 M/sec ( +- 0.12% ) (67.02%)
10932779 branch-misses # 2.69% of all branches ( +- 0.51% ) (83.28%)
0.40077 +- 0.00267 seconds time elapsed ( +- 0.67% )
lebedevri@pini-pini:/build/llvm-build-Clang-release$ perf stat -r 9 ./bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency.yml -analysis-inconsistencies-output-file=/tmp/clusters-new.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 50572 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new.html'
...
Performance counter stats for './bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency.yml -analysis-inconsistencies-output-file=/tmp/clusters-new.html' (9 runs):
6947.79 msec task-clock # 1.000 CPUs utilized ( +- 0.90% )
217 context-switches # 31.236 M/sec ( +- 36.16% )
1 cpu-migrations # 0.096 M/sec ( +- 50.00% )
13258 page-faults # 1908.389 M/sec ( +- 0.34% )
27830796523 cycles # 4006032.286 GHz ( +- 0.89% ) (83.30%)
1504554006 stalled-cycles-frontend # 5.41% frontend cycles idle ( +- 2.10% ) (83.32%)
16716574843 stalled-cycles-backend # 60.07% backend cycles idle ( +- 0.65% ) (33.38%)
17755545931 instructions # 0.64 insn per cycle
# 0.94 stalled cycles per insn ( +- 0.09% ) (50.04%)
3897255686 branches # 560980426.597 M/sec ( +- 0.06% ) (66.70%)
117045395 branch-misses # 3.00% of all branches ( +- 0.47% ) (83.34%)
6.9507 +- 0.0627 seconds time elapsed ( +- 0.90% )
```
I.e. it's +2.6% slowdown for one whole sweep, or +2% for 5 whole sweeps.
Within noise i'd say.
Should help with [[ https://bugs.llvm.org/show_bug.cgi?id=40787 | PR40787 ]].
Reviewers: courbet, gchatelet
Reviewed By: courbet
Subscribers: tschuett, RKSimon, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D58476
llvm-svn: 354767
2019-02-25 17:36:12 +08:00
|
|
|
|
|
|
|
static cl::opt<float> AnalysisInconsistencyEpsilon(
|
|
|
|
"analysis-inconsistency-epsilon",
|
|
|
|
cl::desc("epsilon for detection of when the cluster is different from the "
|
|
|
|
"LLVM schedule profile values"),
|
2019-03-18 19:32:37 +08:00
|
|
|
cl::cat(AnalysisOptions), cl::init(0.1));
|
2018-05-15 20:08:00 +08:00
|
|
|
|
2018-10-23 01:10:47 +08:00
|
|
|
static cl::opt<std::string>
|
|
|
|
AnalysisClustersOutputFile("analysis-clusters-output-file", cl::desc(""),
|
2019-03-18 19:32:37 +08:00
|
|
|
cl::cat(AnalysisOptions), cl::init(""));
|
2018-10-23 01:10:47 +08:00
|
|
|
static cl::opt<std::string>
|
2018-05-17 21:41:28 +08:00
|
|
|
AnalysisInconsistenciesOutputFile("analysis-inconsistencies-output-file",
|
2019-03-18 19:32:37 +08:00
|
|
|
cl::desc(""), cl::cat(AnalysisOptions),
|
|
|
|
cl::init(""));
|
2018-04-04 19:37:06 +08:00
|
|
|
|
[llvm-exegesis] Opcode stabilization / reclusterization (PR40715)
Summary:
Given an instruction `Opcode`, we can make benchmarks (measurements) of the
instruction characteristics/performance. Then, to facilitate further analysis
we group the benchmarks with *similar* characteristics into clusters.
Now, this is all not entirely deterministic. Some instructions have variable
characteristics, depending on their arguments. And thus, if we do several
benchmarks of the same instruction `Opcode`, we may end up with *different*
performance characteristics measurements. And when we then do clustering,
these several benchmarks of the same instruction `Opcode` may end up being
clustered into *different* clusters. This is not great for further analysis.
We shall find every `Opcode` with benchmarks not in just one cluster, and move
*all* the benchmarks of said `Opcode` into one new unstable cluster per `Opcode`.
I have solved this by making `ClusterId` a bit field, adding a `IsUnstable` bit,
and introducing `-analysis-display-unstable-clusters` switch to toggle between
displaying stable-only clusters and unstable-only clusters.
The reclusterization is deterministically stable, produces identical reports
between runs. (Or at least that is what i'm seeing, maybe it isn't)
Timings/comparisons:
old (current trunk/head) {F8303582}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-old.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-old.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-old.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-old.html' (25 runs):
6624.73 msec task-clock # 0.999 CPUs utilized ( +- 0.53% )
172 context-switches # 25.965 M/sec ( +- 29.89% )
0 cpu-migrations # 0.042 M/sec ( +- 56.54% )
31073 page-faults # 4690.754 M/sec ( +- 0.08% )
26538711696 cycles # 4006230.292 GHz ( +- 0.53% ) (83.31%)
2017496807 stalled-cycles-frontend # 7.60% frontend cycles idle ( +- 0.93% ) (83.32%)
13403650062 stalled-cycles-backend # 50.51% backend cycles idle ( +- 0.33% ) (33.37%)
19770706799 instructions # 0.74 insn per cycle
# 0.68 stalled cycles per insn ( +- 0.04% ) (50.04%)
4419821812 branches # 667207369.714 M/sec ( +- 0.03% ) (66.69%)
121741669 branch-misses # 2.75% of all branches ( +- 0.28% ) (83.34%)
6.6283 +- 0.0358 seconds time elapsed ( +- 0.54% )
```
patch, with reclustering but without filtering (i.e. outputting all the stable *and* unstable clusters) {F8303586}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-all.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-all.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-all.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-all.html' (25 runs):
6475.29 msec task-clock # 0.999 CPUs utilized ( +- 0.31% )
213 context-switches # 32.952 M/sec ( +- 23.81% )
1 cpu-migrations # 0.130 M/sec ( +- 43.84% )
31287 page-faults # 4832.057 M/sec ( +- 0.08% )
25939086577 cycles # 4006160.279 GHz ( +- 0.31% ) (83.31%)
1958812858 stalled-cycles-frontend # 7.55% frontend cycles idle ( +- 0.68% ) (83.32%)
13218961512 stalled-cycles-backend # 50.96% backend cycles idle ( +- 0.29% ) (33.37%)
19752995402 instructions # 0.76 insn per cycle
# 0.67 stalled cycles per insn ( +- 0.04% ) (50.04%)
4417079244 branches # 682195472.305 M/sec ( +- 0.03% ) (66.70%)
121510065 branch-misses # 2.75% of all branches ( +- 0.19% ) (83.34%)
6.4832 +- 0.0229 seconds time elapsed ( +- 0.35% )
```
Funnily, *this* measurement shows that said reclustering actually improved performance.
patch, with reclustering, only the stable clusters {F8303594}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-stable.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-stable.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-stable.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-stable.html' (25 runs):
6387.71 msec task-clock # 0.999 CPUs utilized ( +- 0.13% )
133 context-switches # 20.792 M/sec ( +- 23.39% )
0 cpu-migrations # 0.063 M/sec ( +- 61.24% )
31318 page-faults # 4903.256 M/sec ( +- 0.08% )
25591984967 cycles # 4006786.266 GHz ( +- 0.13% ) (83.31%)
1881234904 stalled-cycles-frontend # 7.35% frontend cycles idle ( +- 0.25% ) (83.33%)
13209749965 stalled-cycles-backend # 51.62% backend cycles idle ( +- 0.16% ) (33.36%)
19767554347 instructions # 0.77 insn per cycle
# 0.67 stalled cycles per insn ( +- 0.04% ) (50.03%)
4417480305 branches # 691618858.046 M/sec ( +- 0.03% ) (66.68%)
118676358 branch-misses # 2.69% of all branches ( +- 0.07% ) (83.33%)
6.3954 +- 0.0118 seconds time elapsed ( +- 0.18% )
```
Performance improved even further?! Makes sense i guess, less clusters to print.
patch, with reclustering, only the unstable clusters {F8303601}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-unstable.html -analysis-display-unstable-clusters
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-unstable.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-unstable.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-unstable.html -analysis-display-unstable-clusters' (25 runs):
6124.96 msec task-clock # 1.000 CPUs utilized ( +- 0.20% )
194 context-switches # 31.709 M/sec ( +- 20.46% )
0 cpu-migrations # 0.039 M/sec ( +- 49.77% )
31413 page-faults # 5129.261 M/sec ( +- 0.06% )
24536794267 cycles # 4006425.858 GHz ( +- 0.19% ) (83.31%)
1676085087 stalled-cycles-frontend # 6.83% frontend cycles idle ( +- 0.46% ) (83.32%)
13035595603 stalled-cycles-backend # 53.13% backend cycles idle ( +- 0.16% ) (33.36%)
18260877653 instructions # 0.74 insn per cycle
# 0.71 stalled cycles per insn ( +- 0.05% ) (50.03%)
4112411983 branches # 671484364.603 M/sec ( +- 0.03% ) (66.68%)
114066929 branch-misses # 2.77% of all branches ( +- 0.11% ) (83.32%)
6.1278 +- 0.0121 seconds time elapsed ( +- 0.20% )
```
This tells us that the actual `-analysis-inconsistencies-output-file=` outputting only takes ~0.4 sec for 43970 benchmark points (3 whole sweeps)
(Also, wow this is fast, it used to take several minutes originally)
Fixes [[ https://bugs.llvm.org/show_bug.cgi?id=40715 | PR40715 ]].
Reviewers: courbet, gchatelet
Reviewed By: courbet
Subscribers: tschuett, jdoerfert, llvm-commits, RKSimon
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D58355
llvm-svn: 354441
2019-02-20 17:14:04 +08:00
|
|
|
static cl::opt<bool> AnalysisDisplayUnstableOpcodes(
|
|
|
|
"analysis-display-unstable-clusters",
|
|
|
|
cl::desc("if there is more than one benchmark for an opcode, said "
|
|
|
|
"benchmarks may end up not being clustered into the same cluster "
|
|
|
|
"if the measured performance characteristics are different. by "
|
|
|
|
"default all such opcodes are filtered out. this flag will "
|
|
|
|
"instead show only such unstable opcodes"),
|
2019-03-18 19:32:37 +08:00
|
|
|
cl::cat(AnalysisOptions), cl::init(false));
|
[llvm-exegesis] Opcode stabilization / reclusterization (PR40715)
Summary:
Given an instruction `Opcode`, we can make benchmarks (measurements) of the
instruction characteristics/performance. Then, to facilitate further analysis
we group the benchmarks with *similar* characteristics into clusters.
Now, this is all not entirely deterministic. Some instructions have variable
characteristics, depending on their arguments. And thus, if we do several
benchmarks of the same instruction `Opcode`, we may end up with *different*
performance characteristics measurements. And when we then do clustering,
these several benchmarks of the same instruction `Opcode` may end up being
clustered into *different* clusters. This is not great for further analysis.
We shall find every `Opcode` with benchmarks not in just one cluster, and move
*all* the benchmarks of said `Opcode` into one new unstable cluster per `Opcode`.
I have solved this by making `ClusterId` a bit field, adding a `IsUnstable` bit,
and introducing `-analysis-display-unstable-clusters` switch to toggle between
displaying stable-only clusters and unstable-only clusters.
The reclusterization is deterministically stable, produces identical reports
between runs. (Or at least that is what i'm seeing, maybe it isn't)
Timings/comparisons:
old (current trunk/head) {F8303582}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-old.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-old.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-old.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-old.html' (25 runs):
6624.73 msec task-clock # 0.999 CPUs utilized ( +- 0.53% )
172 context-switches # 25.965 M/sec ( +- 29.89% )
0 cpu-migrations # 0.042 M/sec ( +- 56.54% )
31073 page-faults # 4690.754 M/sec ( +- 0.08% )
26538711696 cycles # 4006230.292 GHz ( +- 0.53% ) (83.31%)
2017496807 stalled-cycles-frontend # 7.60% frontend cycles idle ( +- 0.93% ) (83.32%)
13403650062 stalled-cycles-backend # 50.51% backend cycles idle ( +- 0.33% ) (33.37%)
19770706799 instructions # 0.74 insn per cycle
# 0.68 stalled cycles per insn ( +- 0.04% ) (50.04%)
4419821812 branches # 667207369.714 M/sec ( +- 0.03% ) (66.69%)
121741669 branch-misses # 2.75% of all branches ( +- 0.28% ) (83.34%)
6.6283 +- 0.0358 seconds time elapsed ( +- 0.54% )
```
patch, with reclustering but without filtering (i.e. outputting all the stable *and* unstable clusters) {F8303586}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-all.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-all.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-all.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-all.html' (25 runs):
6475.29 msec task-clock # 0.999 CPUs utilized ( +- 0.31% )
213 context-switches # 32.952 M/sec ( +- 23.81% )
1 cpu-migrations # 0.130 M/sec ( +- 43.84% )
31287 page-faults # 4832.057 M/sec ( +- 0.08% )
25939086577 cycles # 4006160.279 GHz ( +- 0.31% ) (83.31%)
1958812858 stalled-cycles-frontend # 7.55% frontend cycles idle ( +- 0.68% ) (83.32%)
13218961512 stalled-cycles-backend # 50.96% backend cycles idle ( +- 0.29% ) (33.37%)
19752995402 instructions # 0.76 insn per cycle
# 0.67 stalled cycles per insn ( +- 0.04% ) (50.04%)
4417079244 branches # 682195472.305 M/sec ( +- 0.03% ) (66.70%)
121510065 branch-misses # 2.75% of all branches ( +- 0.19% ) (83.34%)
6.4832 +- 0.0229 seconds time elapsed ( +- 0.35% )
```
Funnily, *this* measurement shows that said reclustering actually improved performance.
patch, with reclustering, only the stable clusters {F8303594}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-stable.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-stable.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-stable.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-stable.html' (25 runs):
6387.71 msec task-clock # 0.999 CPUs utilized ( +- 0.13% )
133 context-switches # 20.792 M/sec ( +- 23.39% )
0 cpu-migrations # 0.063 M/sec ( +- 61.24% )
31318 page-faults # 4903.256 M/sec ( +- 0.08% )
25591984967 cycles # 4006786.266 GHz ( +- 0.13% ) (83.31%)
1881234904 stalled-cycles-frontend # 7.35% frontend cycles idle ( +- 0.25% ) (83.33%)
13209749965 stalled-cycles-backend # 51.62% backend cycles idle ( +- 0.16% ) (33.36%)
19767554347 instructions # 0.77 insn per cycle
# 0.67 stalled cycles per insn ( +- 0.04% ) (50.03%)
4417480305 branches # 691618858.046 M/sec ( +- 0.03% ) (66.68%)
118676358 branch-misses # 2.69% of all branches ( +- 0.07% ) (83.33%)
6.3954 +- 0.0118 seconds time elapsed ( +- 0.18% )
```
Performance improved even further?! Makes sense i guess, less clusters to print.
patch, with reclustering, only the unstable clusters {F8303601}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-unstable.html -analysis-display-unstable-clusters
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-unstable.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-unstable.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-unstable.html -analysis-display-unstable-clusters' (25 runs):
6124.96 msec task-clock # 1.000 CPUs utilized ( +- 0.20% )
194 context-switches # 31.709 M/sec ( +- 20.46% )
0 cpu-migrations # 0.039 M/sec ( +- 49.77% )
31413 page-faults # 5129.261 M/sec ( +- 0.06% )
24536794267 cycles # 4006425.858 GHz ( +- 0.19% ) (83.31%)
1676085087 stalled-cycles-frontend # 6.83% frontend cycles idle ( +- 0.46% ) (83.32%)
13035595603 stalled-cycles-backend # 53.13% backend cycles idle ( +- 0.16% ) (33.36%)
18260877653 instructions # 0.74 insn per cycle
# 0.71 stalled cycles per insn ( +- 0.05% ) (50.03%)
4112411983 branches # 671484364.603 M/sec ( +- 0.03% ) (66.68%)
114066929 branch-misses # 2.77% of all branches ( +- 0.11% ) (83.32%)
6.1278 +- 0.0121 seconds time elapsed ( +- 0.20% )
```
This tells us that the actual `-analysis-inconsistencies-output-file=` outputting only takes ~0.4 sec for 43970 benchmark points (3 whole sweeps)
(Also, wow this is fast, it used to take several minutes originally)
Fixes [[ https://bugs.llvm.org/show_bug.cgi?id=40715 | PR40715 ]].
Reviewers: courbet, gchatelet
Reviewed By: courbet
Subscribers: tschuett, jdoerfert, llvm-commits, RKSimon
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D58355
llvm-svn: 354441
2019-02-20 17:14:04 +08:00
|
|
|
|
2019-03-18 19:32:37 +08:00
|
|
|
static cl::opt<std::string> CpuName(
|
|
|
|
"mcpu",
|
|
|
|
cl::desc("cpu name to use for pfm counters, leave empty to autodetect"),
|
|
|
|
cl::cat(Options), cl::init(""));
|
2018-10-25 15:44:01 +08:00
|
|
|
|
2019-04-05 23:18:59 +08:00
|
|
|
static cl::opt<bool>
|
|
|
|
DumpObjectToDisk("dump-object-to-disk",
|
|
|
|
cl::desc("dumps the generated benchmark object to disk "
|
|
|
|
"and prints a message to access it"),
|
|
|
|
cl::cat(BenchmarkOptions), cl::init(true));
|
|
|
|
|
2020-02-07 00:18:42 +08:00
|
|
|
static ExitOnError ExitOnErr("llvm-exegesis error: ");
|
|
|
|
|
|
|
|
// Helper function that logs the error(s) and exits.
|
|
|
|
template <typename... ArgTs> static void ExitWithError(ArgTs &&... Args) {
|
|
|
|
ExitOnErr(make_error<Failure>(std::forward<ArgTs>(Args)...));
|
|
|
|
}
|
|
|
|
|
|
|
|
// Check Err. If it's in a failure state log the file error(s) and exit.
|
|
|
|
static void ExitOnFileError(const Twine &FileName, Error Err) {
|
|
|
|
if (Err) {
|
|
|
|
ExitOnErr(createFileError(FileName, std::move(Err)));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Check E. If it's in a success state then return the contained value.
|
|
|
|
// If it's in a failure state log the file error(s) and exit.
|
|
|
|
template <typename T>
|
|
|
|
T ExitOnFileError(const Twine &FileName, Expected<T> &&E) {
|
|
|
|
ExitOnFileError(FileName, E.takeError());
|
|
|
|
return std::move(*E);
|
|
|
|
}
|
2018-06-07 15:51:16 +08:00
|
|
|
|
2018-10-17 23:04:15 +08:00
|
|
|
// Checks that only one of OpcodeNames, OpcodeIndex or SnippetsFile is provided,
|
|
|
|
// and returns the opcode indices or {} if snippets should be read from
|
2018-09-25 15:31:44 +08:00
|
|
|
// `SnippetsFile`.
|
2019-10-09 19:58:42 +08:00
|
|
|
static std::vector<unsigned> getOpcodesOrDie(const MCInstrInfo &MCInstrInfo) {
|
2018-10-17 23:04:15 +08:00
|
|
|
const size_t NumSetFlags = (OpcodeNames.empty() ? 0 : 1) +
|
2018-09-25 15:31:44 +08:00
|
|
|
(OpcodeIndex == 0 ? 0 : 1) +
|
|
|
|
(SnippetsFile.empty() ? 0 : 1);
|
2020-02-07 00:18:42 +08:00
|
|
|
if (NumSetFlags != 1) {
|
|
|
|
ExitOnErr.setBanner("llvm-exegesis: ");
|
|
|
|
ExitWithError("please provide one and only one of 'opcode-index', "
|
|
|
|
"'opcode-name' or 'snippets-file'");
|
|
|
|
}
|
2018-09-25 15:31:44 +08:00
|
|
|
if (!SnippetsFile.empty())
|
2018-10-17 23:04:15 +08:00
|
|
|
return {};
|
2018-05-17 18:52:18 +08:00
|
|
|
if (OpcodeIndex > 0)
|
2018-10-17 23:04:15 +08:00
|
|
|
return {static_cast<unsigned>(OpcodeIndex)};
|
|
|
|
if (OpcodeIndex < 0) {
|
|
|
|
std::vector<unsigned> Result;
|
2018-10-18 16:20:50 +08:00
|
|
|
for (unsigned I = 1, E = MCInstrInfo.getNumOpcodes(); I < E; ++I)
|
2018-10-17 23:04:15 +08:00
|
|
|
Result.push_back(I);
|
|
|
|
return Result;
|
|
|
|
}
|
2018-05-17 18:52:18 +08:00
|
|
|
// Resolve opcode name -> opcode.
|
2019-10-09 19:58:42 +08:00
|
|
|
const auto ResolveName = [&MCInstrInfo](StringRef OpcodeName) -> unsigned {
|
2018-10-17 23:04:15 +08:00
|
|
|
for (unsigned I = 1, E = MCInstrInfo.getNumOpcodes(); I < E; ++I)
|
|
|
|
if (MCInstrInfo.getName(I) == OpcodeName)
|
|
|
|
return I;
|
|
|
|
return 0u;
|
|
|
|
};
|
2019-10-09 19:58:42 +08:00
|
|
|
SmallVector<StringRef, 2> Pieces;
|
|
|
|
StringRef(OpcodeNames.getValue())
|
2018-10-17 23:04:15 +08:00
|
|
|
.split(Pieces, ",", /* MaxSplit */ -1, /* KeepEmpty */ false);
|
|
|
|
std::vector<unsigned> Result;
|
2019-12-23 02:11:17 +08:00
|
|
|
for (const StringRef &OpcodeName : Pieces) {
|
2018-10-17 23:04:15 +08:00
|
|
|
if (unsigned Opcode = ResolveName(OpcodeName))
|
|
|
|
Result.push_back(Opcode);
|
|
|
|
else
|
2020-02-07 00:18:42 +08:00
|
|
|
ExitWithError(Twine("unknown opcode ").concat(OpcodeName));
|
2018-10-17 23:04:15 +08:00
|
|
|
}
|
|
|
|
return Result;
|
2018-05-17 18:52:18 +08:00
|
|
|
}
|
|
|
|
|
2018-09-13 15:40:53 +08:00
|
|
|
// Generates code snippets for opcode `Opcode`.
|
2019-10-09 19:58:42 +08:00
|
|
|
static Expected<std::vector<BenchmarkCode>>
|
2019-09-27 20:56:24 +08:00
|
|
|
generateSnippets(const LLVMState &State, unsigned Opcode,
|
2019-10-09 19:58:42 +08:00
|
|
|
const BitVector &ForbiddenRegs) {
|
2018-10-24 19:55:06 +08:00
|
|
|
const Instruction &Instr = State.getIC().getInstr(Opcode);
|
2019-12-18 19:08:38 +08:00
|
|
|
const MCInstrDesc &InstrDesc = Instr.Description;
|
2018-09-13 15:40:53 +08:00
|
|
|
// Ignore instructions that we cannot run.
|
|
|
|
if (InstrDesc.isPseudo())
|
2019-09-30 21:53:50 +08:00
|
|
|
return make_error<Failure>("Unsupported opcode: isPseudo");
|
2018-09-13 15:40:53 +08:00
|
|
|
if (InstrDesc.isBranch() || InstrDesc.isIndirectBranch())
|
2019-09-30 21:53:50 +08:00
|
|
|
return make_error<Failure>("Unsupported opcode: isBranch/isIndirectBranch");
|
2018-09-13 15:40:53 +08:00
|
|
|
if (InstrDesc.isCall() || InstrDesc.isReturn())
|
2019-09-30 21:53:50 +08:00
|
|
|
return make_error<Failure>("Unsupported opcode: isCall/isReturn");
|
2018-09-13 15:40:53 +08:00
|
|
|
|
[llvm-exegesis] Exploring X86::OperandType::OPERAND_COND_CODE
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
2020-02-13 01:54:39 +08:00
|
|
|
const std::vector<InstructionTemplate> InstructionVariants =
|
|
|
|
State.getExegesisTarget().generateInstructionVariants(
|
|
|
|
Instr, MaxConfigsPerOpcode);
|
|
|
|
|
2019-11-09 21:43:09 +08:00
|
|
|
SnippetGenerator::Options SnippetOptions;
|
|
|
|
SnippetOptions.MaxConfigsPerOpcode = MaxConfigsPerOpcode;
|
2018-10-10 22:57:32 +08:00
|
|
|
const std::unique_ptr<SnippetGenerator> Generator =
|
2019-10-08 22:30:24 +08:00
|
|
|
State.getExegesisTarget().createSnippetGenerator(BenchmarkMode, State,
|
2019-11-09 21:43:09 +08:00
|
|
|
SnippetOptions);
|
2018-10-10 22:57:32 +08:00
|
|
|
if (!Generator)
|
2020-02-07 00:18:42 +08:00
|
|
|
ExitWithError("cannot create snippet generator");
|
[llvm-exegesis] Exploring X86::OperandType::OPERAND_COND_CODE
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
2020-02-13 01:54:39 +08:00
|
|
|
|
|
|
|
std::vector<BenchmarkCode> Benchmarks;
|
|
|
|
for (const InstructionTemplate &Variant : InstructionVariants) {
|
|
|
|
if (Benchmarks.size() >= MaxConfigsPerOpcode)
|
|
|
|
break;
|
|
|
|
if (auto Err = Generator->generateConfigurations(Variant, Benchmarks,
|
|
|
|
ForbiddenRegs))
|
|
|
|
return std::move(Err);
|
|
|
|
}
|
|
|
|
return Benchmarks;
|
2018-09-13 15:40:53 +08:00
|
|
|
}
|
|
|
|
|
2018-05-15 20:08:00 +08:00
|
|
|
void benchmarkMain() {
|
2019-04-08 18:50:31 +08:00
|
|
|
#ifndef HAVE_LIBPFM
|
2020-02-07 00:18:42 +08:00
|
|
|
ExitWithError("benchmarking unavailable, LLVM was built without libpfm.");
|
2019-04-08 18:50:31 +08:00
|
|
|
#endif
|
|
|
|
|
2018-05-15 20:08:00 +08:00
|
|
|
if (exegesis::pfm::pfmInitialize())
|
2020-02-07 00:18:42 +08:00
|
|
|
ExitWithError("cannot initialize libpfm");
|
2018-05-15 20:08:00 +08:00
|
|
|
|
2019-10-09 19:58:42 +08:00
|
|
|
InitializeNativeTarget();
|
|
|
|
InitializeNativeTargetAsmPrinter();
|
|
|
|
InitializeNativeTargetAsmParser();
|
2019-05-22 21:50:16 +08:00
|
|
|
InitializeNativeExegesisTarget();
|
2018-04-04 19:37:06 +08:00
|
|
|
|
2018-10-25 15:44:01 +08:00
|
|
|
const LLVMState State(CpuName);
|
2019-12-31 21:14:41 +08:00
|
|
|
|
2020-02-07 01:21:01 +08:00
|
|
|
const std::unique_ptr<BenchmarkRunner> Runner = ExitOnErr(
|
|
|
|
State.getExegesisTarget().createBenchmarkRunner(BenchmarkMode, State));
|
2019-12-31 21:14:41 +08:00
|
|
|
if (!Runner) {
|
2020-02-07 00:18:42 +08:00
|
|
|
ExitWithError("cannot create benchmark runner");
|
2019-12-31 21:14:41 +08:00
|
|
|
}
|
|
|
|
|
2018-10-17 23:04:15 +08:00
|
|
|
const auto Opcodes = getOpcodesOrDie(State.getInstrInfo());
|
2018-04-04 19:37:06 +08:00
|
|
|
|
[llvm-exegesis] 'Min' repetition mode
Summary:
As noted in documentation, different repetition modes have different trade-offs:
> .. option:: -repetition-mode=[duplicate|loop]
>
> Specify the repetition mode. `duplicate` will create a large, straight line
> basic block with `num-repetitions` copies of the snippet. `loop` will wrap
> the snippet in a loop which will be run `num-repetitions` times. The `loop`
> mode tends to better hide the effects of the CPU frontend on architectures
> that cache decoded instructions, but consumes a register for counting
> iterations.
Indeed. Example:
>>! In D74156#1873657, @lebedev.ri wrote:
> At least for `CMOV`, i'm seeing wildly different results
> | | Latency | RThroughput |
> | duplicate | 1 | 0.8 |
> | loop | 2 | 0.6 |
> where latency=1 seems correct, and i'd expect the througput to be close to 1/2 (since there are two execution units).
This isn't great for analysis, at least for schedule model development.
As discussed in excruciating detail in
>>! In D74156#1924514, @gchatelet wrote:
>>>! In D74156#1920632, @lebedev.ri wrote:
>> ... did that explanation of the question i'm having made any sense?
>
> Thx for digging in the conversation !
> Ok it makes more sense now.
>
> I discussed it a bit with @courbet:
> - We want the analysis tool to stay simple so we'd rather not make it knowledgeable of the repetition mode.
> - We'd like to still be able to select either repetition mode to dig into special cases
>
> So we could add a third `min` repetition mode that would run both and take the minimum. It could be the default option.
> Would you have some time to look what it would take to add this third mode?
there appears to be an agreement that it is indeed sub-par,
and that we should provide an optional, measurement (not analysis!) -time
way to rectify the situation.
However, the solutions isn't entirely straight-forward.
We can just add an actual 'multiplexer' `MinSnippetRepetitor`, because
if we just concatenate snippets produced by `DuplicateSnippetRepetitor`
and `LoopSnippetRepetitor` and run+measure that, the measurement will
naturally be different from what we'd get by running+measuring
them separately and taking the min.
([[ https://www.wolframalpha.com/input/?i=%28x%2By%29%2F2+%21%3D+min%28x%2C+y%29 | `time(D+L)/2 != min(time(D), time(L))` ]])
Also, it seems best to me to have a single snippet instead of generating
a snippet per repetition mode, since the only difference here is that the
loop repetition mode reserves one register for loop counter.
As far as i can tell, we can either teach `BenchmarkRunner::runConfiguration()`
to produce a single report given multiple repetitors (as in the patch),
or do that one layer higher - don't modify `BenchmarkRunner::runConfiguration()`,
produce multiple reports, don't actually print each one, but aggregate them somehow
and only print the final one.
Initially i've gone ahead with the latter approach, but it didn't look like a natural fit;
the former (as in the diff) does seem like a better fit to me.
There's also a question of the test coverage. It sure currently does work here:
```
$ ./bin/llvm-exegesis --opcode-name=CMOV64rr --mode=inverse_throughput --repetition-mode=duplicate
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-8fb949.o
---
mode: inverse_throughput
key:
instructions:
- 'CMOV64rr RAX RAX R11 i_0x0'
- 'CMOV64rr RBP RBP R15 i_0x0'
- 'CMOV64rr RBX RBX RBX i_0x0'
- 'CMOV64rr RCX RCX RBX i_0x0'
- 'CMOV64rr RDI RDI R10 i_0x0'
- 'CMOV64rr RDX RDX RAX i_0x0'
- 'CMOV64rr RSI RSI RAX i_0x0'
- 'CMOV64rr R8 R8 R8 i_0x0'
- 'CMOV64rr R9 R9 RDX i_0x0'
- 'CMOV64rr R10 R10 RBX i_0x0'
- 'CMOV64rr R11 R11 R14 i_0x0'
- 'CMOV64rr R12 R12 R9 i_0x0'
- 'CMOV64rr R13 R13 R12 i_0x0'
- 'CMOV64rr R14 R14 R15 i_0x0'
- 'CMOV64rr R15 R15 R13 i_0x0'
config: ''
register_initial_values:
- 'RAX=0x0'
- 'R11=0x0'
- 'EFLAGS=0x0'
- 'RBP=0x0'
- 'R15=0x0'
- 'RBX=0x0'
- 'RCX=0x0'
- 'RDI=0x0'
- 'R10=0x0'
- 'RDX=0x0'
- 'RSI=0x0'
- 'R8=0x0'
- 'R9=0x0'
- 'R14=0x0'
- 'R12=0x0'
- 'R13=0x0'
cpu_name: bdver2
llvm_triple: x86_64-unknown-linux-gnu
num_repetitions: 10000
measurements:
- { key: inverse_throughput, value: 0.819, per_snippet_value: 12.285 }
error: ''
info: instruction has tied variables, using static renaming.
assembled_snippet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
...
$ ./bin/llvm-exegesis --opcode-name=CMOV64rr --mode=inverse_throughput --repetition-mode=loop
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-051eb3.o
---
mode: inverse_throughput
key:
instructions:
- 'CMOV64rr RAX RAX R11 i_0x0'
- 'CMOV64rr RBP RBP RSI i_0x0'
- 'CMOV64rr RBX RBX R9 i_0x0'
- 'CMOV64rr RCX RCX RSI i_0x0'
- 'CMOV64rr RDI RDI RBP i_0x0'
- 'CMOV64rr RDX RDX R9 i_0x0'
- 'CMOV64rr RSI RSI RDI i_0x0'
- 'CMOV64rr R9 R9 R12 i_0x0'
- 'CMOV64rr R10 R10 R11 i_0x0'
- 'CMOV64rr R11 R11 R9 i_0x0'
- 'CMOV64rr R12 R12 RBP i_0x0'
- 'CMOV64rr R13 R13 RSI i_0x0'
- 'CMOV64rr R14 R14 R14 i_0x0'
- 'CMOV64rr R15 R15 R10 i_0x0'
config: ''
register_initial_values:
- 'RAX=0x0'
- 'R11=0x0'
- 'EFLAGS=0x0'
- 'RBP=0x0'
- 'RSI=0x0'
- 'RBX=0x0'
- 'R9=0x0'
- 'RCX=0x0'
- 'RDI=0x0'
- 'RDX=0x0'
- 'R12=0x0'
- 'R10=0x0'
- 'R13=0x0'
- 'R14=0x0'
- 'R15=0x0'
cpu_name: bdver2
llvm_triple: x86_64-unknown-linux-gnu
num_repetitions: 10000
measurements:
- { key: inverse_throughput, value: 0.6083, per_snippet_value: 8.5162 }
error: ''
info: instruction has tied variables, using static renaming.
assembled_snippet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
...
$ ./bin/llvm-exegesis --opcode-name=CMOV64rr --mode=inverse_throughput --repetition-mode=min
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-c7a47d.o
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-2581f1.o
---
mode: inverse_throughput
key:
instructions:
- 'CMOV64rr RAX RAX R11 i_0x0'
- 'CMOV64rr RBP RBP R10 i_0x0'
- 'CMOV64rr RBX RBX R10 i_0x0'
- 'CMOV64rr RCX RCX RDX i_0x0'
- 'CMOV64rr RDI RDI RAX i_0x0'
- 'CMOV64rr RDX RDX R9 i_0x0'
- 'CMOV64rr RSI RSI RAX i_0x0'
- 'CMOV64rr R9 R9 RBX i_0x0'
- 'CMOV64rr R10 R10 R12 i_0x0'
- 'CMOV64rr R11 R11 RDI i_0x0'
- 'CMOV64rr R12 R12 RDI i_0x0'
- 'CMOV64rr R13 R13 RDI i_0x0'
- 'CMOV64rr R14 R14 R9 i_0x0'
- 'CMOV64rr R15 R15 RBP i_0x0'
config: ''
register_initial_values:
- 'RAX=0x0'
- 'R11=0x0'
- 'EFLAGS=0x0'
- 'RBP=0x0'
- 'R10=0x0'
- 'RBX=0x0'
- 'RCX=0x0'
- 'RDX=0x0'
- 'RDI=0x0'
- 'R9=0x0'
- 'RSI=0x0'
- 'R12=0x0'
- 'R13=0x0'
- 'R14=0x0'
- 'R15=0x0'
cpu_name: bdver2
llvm_triple: x86_64-unknown-linux-gnu
num_repetitions: 10000
measurements:
- { key: inverse_throughput, value: 0.6073, per_snippet_value: 8.5022 }
error: ''
info: instruction has tied variables, using static renaming.
assembled_snippet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
...
```
but i open to suggestions as to how test that.
I also have gone with the suggestion to default to this new mode.
This was irking me for some time, so i'm happy to finally see progress here.
Looking forward to feedback.
Reviewers: courbet, gchatelet
Reviewed By: courbet, gchatelet
Subscribers: mstojanovic, RKSimon, llvm-commits, courbet, gchatelet
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D76921
2020-04-02 14:28:35 +08:00
|
|
|
SmallVector<std::unique_ptr<const SnippetRepetitor>, 2> Repetitors;
|
|
|
|
if (RepetitionMode != InstructionBenchmark::RepetitionModeE::AggregateMin)
|
|
|
|
Repetitors.emplace_back(SnippetRepetitor::Create(RepetitionMode, State));
|
|
|
|
else {
|
|
|
|
for (InstructionBenchmark::RepetitionModeE RepMode :
|
|
|
|
{InstructionBenchmark::RepetitionModeE::Duplicate,
|
|
|
|
InstructionBenchmark::RepetitionModeE::Loop})
|
|
|
|
Repetitors.emplace_back(SnippetRepetitor::Create(RepMode, State));
|
|
|
|
}
|
|
|
|
|
|
|
|
BitVector AllReservedRegs;
|
|
|
|
llvm::for_each(Repetitors,
|
|
|
|
[&AllReservedRegs](
|
|
|
|
const std::unique_ptr<const SnippetRepetitor> &Repetitor) {
|
|
|
|
AllReservedRegs |= Repetitor->getReservedRegs();
|
|
|
|
});
|
2019-09-27 20:56:24 +08:00
|
|
|
|
2018-09-25 15:31:44 +08:00
|
|
|
std::vector<BenchmarkCode> Configurations;
|
2018-10-17 23:04:15 +08:00
|
|
|
if (!Opcodes.empty()) {
|
|
|
|
for (const unsigned Opcode : Opcodes) {
|
|
|
|
// Ignore instructions without a sched class if
|
|
|
|
// -ignore-invalid-sched-class is passed.
|
|
|
|
if (IgnoreInvalidSchedClass &&
|
|
|
|
State.getInstrInfo().get(Opcode).getSchedClass() == 0) {
|
2019-10-09 19:58:42 +08:00
|
|
|
errs() << State.getInstrInfo().getName(Opcode)
|
|
|
|
<< ": ignoring instruction without sched class\n";
|
2018-10-17 23:04:15 +08:00
|
|
|
continue;
|
|
|
|
}
|
[llvm-exegesis] 'Min' repetition mode
Summary:
As noted in documentation, different repetition modes have different trade-offs:
> .. option:: -repetition-mode=[duplicate|loop]
>
> Specify the repetition mode. `duplicate` will create a large, straight line
> basic block with `num-repetitions` copies of the snippet. `loop` will wrap
> the snippet in a loop which will be run `num-repetitions` times. The `loop`
> mode tends to better hide the effects of the CPU frontend on architectures
> that cache decoded instructions, but consumes a register for counting
> iterations.
Indeed. Example:
>>! In D74156#1873657, @lebedev.ri wrote:
> At least for `CMOV`, i'm seeing wildly different results
> | | Latency | RThroughput |
> | duplicate | 1 | 0.8 |
> | loop | 2 | 0.6 |
> where latency=1 seems correct, and i'd expect the througput to be close to 1/2 (since there are two execution units).
This isn't great for analysis, at least for schedule model development.
As discussed in excruciating detail in
>>! In D74156#1924514, @gchatelet wrote:
>>>! In D74156#1920632, @lebedev.ri wrote:
>> ... did that explanation of the question i'm having made any sense?
>
> Thx for digging in the conversation !
> Ok it makes more sense now.
>
> I discussed it a bit with @courbet:
> - We want the analysis tool to stay simple so we'd rather not make it knowledgeable of the repetition mode.
> - We'd like to still be able to select either repetition mode to dig into special cases
>
> So we could add a third `min` repetition mode that would run both and take the minimum. It could be the default option.
> Would you have some time to look what it would take to add this third mode?
there appears to be an agreement that it is indeed sub-par,
and that we should provide an optional, measurement (not analysis!) -time
way to rectify the situation.
However, the solutions isn't entirely straight-forward.
We can just add an actual 'multiplexer' `MinSnippetRepetitor`, because
if we just concatenate snippets produced by `DuplicateSnippetRepetitor`
and `LoopSnippetRepetitor` and run+measure that, the measurement will
naturally be different from what we'd get by running+measuring
them separately and taking the min.
([[ https://www.wolframalpha.com/input/?i=%28x%2By%29%2F2+%21%3D+min%28x%2C+y%29 | `time(D+L)/2 != min(time(D), time(L))` ]])
Also, it seems best to me to have a single snippet instead of generating
a snippet per repetition mode, since the only difference here is that the
loop repetition mode reserves one register for loop counter.
As far as i can tell, we can either teach `BenchmarkRunner::runConfiguration()`
to produce a single report given multiple repetitors (as in the patch),
or do that one layer higher - don't modify `BenchmarkRunner::runConfiguration()`,
produce multiple reports, don't actually print each one, but aggregate them somehow
and only print the final one.
Initially i've gone ahead with the latter approach, but it didn't look like a natural fit;
the former (as in the diff) does seem like a better fit to me.
There's also a question of the test coverage. It sure currently does work here:
```
$ ./bin/llvm-exegesis --opcode-name=CMOV64rr --mode=inverse_throughput --repetition-mode=duplicate
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-8fb949.o
---
mode: inverse_throughput
key:
instructions:
- 'CMOV64rr RAX RAX R11 i_0x0'
- 'CMOV64rr RBP RBP R15 i_0x0'
- 'CMOV64rr RBX RBX RBX i_0x0'
- 'CMOV64rr RCX RCX RBX i_0x0'
- 'CMOV64rr RDI RDI R10 i_0x0'
- 'CMOV64rr RDX RDX RAX i_0x0'
- 'CMOV64rr RSI RSI RAX i_0x0'
- 'CMOV64rr R8 R8 R8 i_0x0'
- 'CMOV64rr R9 R9 RDX i_0x0'
- 'CMOV64rr R10 R10 RBX i_0x0'
- 'CMOV64rr R11 R11 R14 i_0x0'
- 'CMOV64rr R12 R12 R9 i_0x0'
- 'CMOV64rr R13 R13 R12 i_0x0'
- 'CMOV64rr R14 R14 R15 i_0x0'
- 'CMOV64rr R15 R15 R13 i_0x0'
config: ''
register_initial_values:
- 'RAX=0x0'
- 'R11=0x0'
- 'EFLAGS=0x0'
- 'RBP=0x0'
- 'R15=0x0'
- 'RBX=0x0'
- 'RCX=0x0'
- 'RDI=0x0'
- 'R10=0x0'
- 'RDX=0x0'
- 'RSI=0x0'
- 'R8=0x0'
- 'R9=0x0'
- 'R14=0x0'
- 'R12=0x0'
- 'R13=0x0'
cpu_name: bdver2
llvm_triple: x86_64-unknown-linux-gnu
num_repetitions: 10000
measurements:
- { key: inverse_throughput, value: 0.819, per_snippet_value: 12.285 }
error: ''
info: instruction has tied variables, using static renaming.
assembled_snippet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
...
$ ./bin/llvm-exegesis --opcode-name=CMOV64rr --mode=inverse_throughput --repetition-mode=loop
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-051eb3.o
---
mode: inverse_throughput
key:
instructions:
- 'CMOV64rr RAX RAX R11 i_0x0'
- 'CMOV64rr RBP RBP RSI i_0x0'
- 'CMOV64rr RBX RBX R9 i_0x0'
- 'CMOV64rr RCX RCX RSI i_0x0'
- 'CMOV64rr RDI RDI RBP i_0x0'
- 'CMOV64rr RDX RDX R9 i_0x0'
- 'CMOV64rr RSI RSI RDI i_0x0'
- 'CMOV64rr R9 R9 R12 i_0x0'
- 'CMOV64rr R10 R10 R11 i_0x0'
- 'CMOV64rr R11 R11 R9 i_0x0'
- 'CMOV64rr R12 R12 RBP i_0x0'
- 'CMOV64rr R13 R13 RSI i_0x0'
- 'CMOV64rr R14 R14 R14 i_0x0'
- 'CMOV64rr R15 R15 R10 i_0x0'
config: ''
register_initial_values:
- 'RAX=0x0'
- 'R11=0x0'
- 'EFLAGS=0x0'
- 'RBP=0x0'
- 'RSI=0x0'
- 'RBX=0x0'
- 'R9=0x0'
- 'RCX=0x0'
- 'RDI=0x0'
- 'RDX=0x0'
- 'R12=0x0'
- 'R10=0x0'
- 'R13=0x0'
- 'R14=0x0'
- 'R15=0x0'
cpu_name: bdver2
llvm_triple: x86_64-unknown-linux-gnu
num_repetitions: 10000
measurements:
- { key: inverse_throughput, value: 0.6083, per_snippet_value: 8.5162 }
error: ''
info: instruction has tied variables, using static renaming.
assembled_snippet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
...
$ ./bin/llvm-exegesis --opcode-name=CMOV64rr --mode=inverse_throughput --repetition-mode=min
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-c7a47d.o
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-2581f1.o
---
mode: inverse_throughput
key:
instructions:
- 'CMOV64rr RAX RAX R11 i_0x0'
- 'CMOV64rr RBP RBP R10 i_0x0'
- 'CMOV64rr RBX RBX R10 i_0x0'
- 'CMOV64rr RCX RCX RDX i_0x0'
- 'CMOV64rr RDI RDI RAX i_0x0'
- 'CMOV64rr RDX RDX R9 i_0x0'
- 'CMOV64rr RSI RSI RAX i_0x0'
- 'CMOV64rr R9 R9 RBX i_0x0'
- 'CMOV64rr R10 R10 R12 i_0x0'
- 'CMOV64rr R11 R11 RDI i_0x0'
- 'CMOV64rr R12 R12 RDI i_0x0'
- 'CMOV64rr R13 R13 RDI i_0x0'
- 'CMOV64rr R14 R14 R9 i_0x0'
- 'CMOV64rr R15 R15 RBP i_0x0'
config: ''
register_initial_values:
- 'RAX=0x0'
- 'R11=0x0'
- 'EFLAGS=0x0'
- 'RBP=0x0'
- 'R10=0x0'
- 'RBX=0x0'
- 'RCX=0x0'
- 'RDX=0x0'
- 'RDI=0x0'
- 'R9=0x0'
- 'RSI=0x0'
- 'R12=0x0'
- 'R13=0x0'
- 'R14=0x0'
- 'R15=0x0'
cpu_name: bdver2
llvm_triple: x86_64-unknown-linux-gnu
num_repetitions: 10000
measurements:
- { key: inverse_throughput, value: 0.6073, per_snippet_value: 8.5022 }
error: ''
info: instruction has tied variables, using static renaming.
assembled_snippet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
...
```
but i open to suggestions as to how test that.
I also have gone with the suggestion to default to this new mode.
This was irking me for some time, so i'm happy to finally see progress here.
Looking forward to feedback.
Reviewers: courbet, gchatelet
Reviewed By: courbet, gchatelet
Subscribers: mstojanovic, RKSimon, llvm-commits, courbet, gchatelet
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D76921
2020-04-02 14:28:35 +08:00
|
|
|
|
|
|
|
auto ConfigsForInstr = generateSnippets(State, Opcode, AllReservedRegs);
|
2018-10-17 23:04:15 +08:00
|
|
|
if (!ConfigsForInstr) {
|
2019-10-09 19:58:42 +08:00
|
|
|
logAllUnhandledErrors(
|
|
|
|
ConfigsForInstr.takeError(), errs(),
|
|
|
|
Twine(State.getInstrInfo().getName(Opcode)).concat(": "));
|
2018-10-17 23:04:15 +08:00
|
|
|
continue;
|
|
|
|
}
|
|
|
|
std::move(ConfigsForInstr->begin(), ConfigsForInstr->end(),
|
|
|
|
std::back_inserter(Configurations));
|
2018-09-25 15:31:44 +08:00
|
|
|
}
|
|
|
|
} else {
|
|
|
|
Configurations = ExitOnErr(readSnippets(State, SnippetsFile));
|
|
|
|
}
|
2018-09-13 15:40:53 +08:00
|
|
|
|
2020-02-07 00:18:42 +08:00
|
|
|
if (NumRepetitions == 0) {
|
|
|
|
ExitOnErr.setBanner("llvm-exegesis: ");
|
|
|
|
ExitWithError("--num-repetitions must be greater than zero");
|
|
|
|
}
|
2018-05-17 18:52:18 +08:00
|
|
|
|
2018-06-07 15:51:16 +08:00
|
|
|
// Write to standard output if file is not set.
|
|
|
|
if (BenchmarkFile.empty())
|
|
|
|
BenchmarkFile = "-";
|
|
|
|
|
2018-09-13 15:40:53 +08:00
|
|
|
for (const BenchmarkCode &Conf : Configurations) {
|
2020-02-07 20:45:10 +08:00
|
|
|
InstructionBenchmark Result = ExitOnErr(Runner->runConfiguration(
|
[llvm-exegesis] 'Min' repetition mode
Summary:
As noted in documentation, different repetition modes have different trade-offs:
> .. option:: -repetition-mode=[duplicate|loop]
>
> Specify the repetition mode. `duplicate` will create a large, straight line
> basic block with `num-repetitions` copies of the snippet. `loop` will wrap
> the snippet in a loop which will be run `num-repetitions` times. The `loop`
> mode tends to better hide the effects of the CPU frontend on architectures
> that cache decoded instructions, but consumes a register for counting
> iterations.
Indeed. Example:
>>! In D74156#1873657, @lebedev.ri wrote:
> At least for `CMOV`, i'm seeing wildly different results
> | | Latency | RThroughput |
> | duplicate | 1 | 0.8 |
> | loop | 2 | 0.6 |
> where latency=1 seems correct, and i'd expect the througput to be close to 1/2 (since there are two execution units).
This isn't great for analysis, at least for schedule model development.
As discussed in excruciating detail in
>>! In D74156#1924514, @gchatelet wrote:
>>>! In D74156#1920632, @lebedev.ri wrote:
>> ... did that explanation of the question i'm having made any sense?
>
> Thx for digging in the conversation !
> Ok it makes more sense now.
>
> I discussed it a bit with @courbet:
> - We want the analysis tool to stay simple so we'd rather not make it knowledgeable of the repetition mode.
> - We'd like to still be able to select either repetition mode to dig into special cases
>
> So we could add a third `min` repetition mode that would run both and take the minimum. It could be the default option.
> Would you have some time to look what it would take to add this third mode?
there appears to be an agreement that it is indeed sub-par,
and that we should provide an optional, measurement (not analysis!) -time
way to rectify the situation.
However, the solutions isn't entirely straight-forward.
We can just add an actual 'multiplexer' `MinSnippetRepetitor`, because
if we just concatenate snippets produced by `DuplicateSnippetRepetitor`
and `LoopSnippetRepetitor` and run+measure that, the measurement will
naturally be different from what we'd get by running+measuring
them separately and taking the min.
([[ https://www.wolframalpha.com/input/?i=%28x%2By%29%2F2+%21%3D+min%28x%2C+y%29 | `time(D+L)/2 != min(time(D), time(L))` ]])
Also, it seems best to me to have a single snippet instead of generating
a snippet per repetition mode, since the only difference here is that the
loop repetition mode reserves one register for loop counter.
As far as i can tell, we can either teach `BenchmarkRunner::runConfiguration()`
to produce a single report given multiple repetitors (as in the patch),
or do that one layer higher - don't modify `BenchmarkRunner::runConfiguration()`,
produce multiple reports, don't actually print each one, but aggregate them somehow
and only print the final one.
Initially i've gone ahead with the latter approach, but it didn't look like a natural fit;
the former (as in the diff) does seem like a better fit to me.
There's also a question of the test coverage. It sure currently does work here:
```
$ ./bin/llvm-exegesis --opcode-name=CMOV64rr --mode=inverse_throughput --repetition-mode=duplicate
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-8fb949.o
---
mode: inverse_throughput
key:
instructions:
- 'CMOV64rr RAX RAX R11 i_0x0'
- 'CMOV64rr RBP RBP R15 i_0x0'
- 'CMOV64rr RBX RBX RBX i_0x0'
- 'CMOV64rr RCX RCX RBX i_0x0'
- 'CMOV64rr RDI RDI R10 i_0x0'
- 'CMOV64rr RDX RDX RAX i_0x0'
- 'CMOV64rr RSI RSI RAX i_0x0'
- 'CMOV64rr R8 R8 R8 i_0x0'
- 'CMOV64rr R9 R9 RDX i_0x0'
- 'CMOV64rr R10 R10 RBX i_0x0'
- 'CMOV64rr R11 R11 R14 i_0x0'
- 'CMOV64rr R12 R12 R9 i_0x0'
- 'CMOV64rr R13 R13 R12 i_0x0'
- 'CMOV64rr R14 R14 R15 i_0x0'
- 'CMOV64rr R15 R15 R13 i_0x0'
config: ''
register_initial_values:
- 'RAX=0x0'
- 'R11=0x0'
- 'EFLAGS=0x0'
- 'RBP=0x0'
- 'R15=0x0'
- 'RBX=0x0'
- 'RCX=0x0'
- 'RDI=0x0'
- 'R10=0x0'
- 'RDX=0x0'
- 'RSI=0x0'
- 'R8=0x0'
- 'R9=0x0'
- 'R14=0x0'
- 'R12=0x0'
- 'R13=0x0'
cpu_name: bdver2
llvm_triple: x86_64-unknown-linux-gnu
num_repetitions: 10000
measurements:
- { key: inverse_throughput, value: 0.819, per_snippet_value: 12.285 }
error: ''
info: instruction has tied variables, using static renaming.
assembled_snippet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
...
$ ./bin/llvm-exegesis --opcode-name=CMOV64rr --mode=inverse_throughput --repetition-mode=loop
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-051eb3.o
---
mode: inverse_throughput
key:
instructions:
- 'CMOV64rr RAX RAX R11 i_0x0'
- 'CMOV64rr RBP RBP RSI i_0x0'
- 'CMOV64rr RBX RBX R9 i_0x0'
- 'CMOV64rr RCX RCX RSI i_0x0'
- 'CMOV64rr RDI RDI RBP i_0x0'
- 'CMOV64rr RDX RDX R9 i_0x0'
- 'CMOV64rr RSI RSI RDI i_0x0'
- 'CMOV64rr R9 R9 R12 i_0x0'
- 'CMOV64rr R10 R10 R11 i_0x0'
- 'CMOV64rr R11 R11 R9 i_0x0'
- 'CMOV64rr R12 R12 RBP i_0x0'
- 'CMOV64rr R13 R13 RSI i_0x0'
- 'CMOV64rr R14 R14 R14 i_0x0'
- 'CMOV64rr R15 R15 R10 i_0x0'
config: ''
register_initial_values:
- 'RAX=0x0'
- 'R11=0x0'
- 'EFLAGS=0x0'
- 'RBP=0x0'
- 'RSI=0x0'
- 'RBX=0x0'
- 'R9=0x0'
- 'RCX=0x0'
- 'RDI=0x0'
- 'RDX=0x0'
- 'R12=0x0'
- 'R10=0x0'
- 'R13=0x0'
- 'R14=0x0'
- 'R15=0x0'
cpu_name: bdver2
llvm_triple: x86_64-unknown-linux-gnu
num_repetitions: 10000
measurements:
- { key: inverse_throughput, value: 0.6083, per_snippet_value: 8.5162 }
error: ''
info: instruction has tied variables, using static renaming.
assembled_snippet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
...
$ ./bin/llvm-exegesis --opcode-name=CMOV64rr --mode=inverse_throughput --repetition-mode=min
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-c7a47d.o
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-2581f1.o
---
mode: inverse_throughput
key:
instructions:
- 'CMOV64rr RAX RAX R11 i_0x0'
- 'CMOV64rr RBP RBP R10 i_0x0'
- 'CMOV64rr RBX RBX R10 i_0x0'
- 'CMOV64rr RCX RCX RDX i_0x0'
- 'CMOV64rr RDI RDI RAX i_0x0'
- 'CMOV64rr RDX RDX R9 i_0x0'
- 'CMOV64rr RSI RSI RAX i_0x0'
- 'CMOV64rr R9 R9 RBX i_0x0'
- 'CMOV64rr R10 R10 R12 i_0x0'
- 'CMOV64rr R11 R11 RDI i_0x0'
- 'CMOV64rr R12 R12 RDI i_0x0'
- 'CMOV64rr R13 R13 RDI i_0x0'
- 'CMOV64rr R14 R14 R9 i_0x0'
- 'CMOV64rr R15 R15 RBP i_0x0'
config: ''
register_initial_values:
- 'RAX=0x0'
- 'R11=0x0'
- 'EFLAGS=0x0'
- 'RBP=0x0'
- 'R10=0x0'
- 'RBX=0x0'
- 'RCX=0x0'
- 'RDX=0x0'
- 'RDI=0x0'
- 'R9=0x0'
- 'RSI=0x0'
- 'R12=0x0'
- 'R13=0x0'
- 'R14=0x0'
- 'R15=0x0'
cpu_name: bdver2
llvm_triple: x86_64-unknown-linux-gnu
num_repetitions: 10000
measurements:
- { key: inverse_throughput, value: 0.6073, per_snippet_value: 8.5022 }
error: ''
info: instruction has tied variables, using static renaming.
assembled_snippet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
...
```
but i open to suggestions as to how test that.
I also have gone with the suggestion to default to this new mode.
This was irking me for some time, so i'm happy to finally see progress here.
Looking forward to feedback.
Reviewers: courbet, gchatelet
Reviewed By: courbet, gchatelet
Subscribers: mstojanovic, RKSimon, llvm-commits, courbet, gchatelet
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D76921
2020-04-02 14:28:35 +08:00
|
|
|
Conf, NumRepetitions, Repetitors, DumpObjectToDisk));
|
2020-02-07 00:18:42 +08:00
|
|
|
ExitOnFileError(BenchmarkFile, Result.writeYaml(State, BenchmarkFile));
|
2018-09-13 15:40:53 +08:00
|
|
|
}
|
2018-05-15 20:08:00 +08:00
|
|
|
exegesis::pfm::pfmTerminate();
|
|
|
|
}
|
|
|
|
|
2018-05-17 21:41:28 +08:00
|
|
|
// Prints the results of running analysis pass `Pass` to file `OutputFilename`
|
|
|
|
// if OutputFilename is non-empty.
|
|
|
|
template <typename Pass>
|
|
|
|
static void maybeRunAnalysis(const Analysis &Analyzer, const std::string &Name,
|
2018-06-05 18:56:19 +08:00
|
|
|
const std::string &OutputFilename) {
|
2018-05-17 21:41:28 +08:00
|
|
|
if (OutputFilename.empty())
|
|
|
|
return;
|
|
|
|
if (OutputFilename != "-") {
|
2019-10-09 19:58:42 +08:00
|
|
|
errs() << "Printing " << Name << " results to file '" << OutputFilename
|
|
|
|
<< "'\n";
|
2018-05-17 21:41:28 +08:00
|
|
|
}
|
|
|
|
std::error_code ErrorCode;
|
2019-10-09 19:58:42 +08:00
|
|
|
raw_fd_ostream ClustersOS(OutputFilename, ErrorCode,
|
|
|
|
sys::fs::FA_Read | sys::fs::FA_Write);
|
2018-06-08 03:58:58 +08:00
|
|
|
if (ErrorCode)
|
2020-02-07 00:18:42 +08:00
|
|
|
ExitOnFileError(OutputFilename, errorCodeToError(ErrorCode));
|
2018-06-08 03:58:58 +08:00
|
|
|
if (auto Err = Analyzer.run<Pass>(ClustersOS))
|
2020-02-07 00:18:42 +08:00
|
|
|
ExitOnFileError(OutputFilename, std::move(Err));
|
2018-05-17 21:41:28 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static void analysisMain() {
|
2020-02-07 00:18:42 +08:00
|
|
|
ExitOnErr.setBanner("llvm-exegesis: ");
|
2018-06-07 15:51:16 +08:00
|
|
|
if (BenchmarkFile.empty())
|
2020-02-07 00:18:42 +08:00
|
|
|
ExitWithError("--benchmarks-file must be set");
|
2018-06-07 15:51:16 +08:00
|
|
|
|
2019-02-04 17:12:08 +08:00
|
|
|
if (AnalysisClustersOutputFile.empty() &&
|
|
|
|
AnalysisInconsistenciesOutputFile.empty()) {
|
2020-02-07 00:18:42 +08:00
|
|
|
ExitWithError(
|
|
|
|
"for --mode=analysis: At least one of --analysis-clusters-output-file"
|
|
|
|
"and --analysis-inconsistencies-output-file must be specified");
|
2019-02-04 17:12:08 +08:00
|
|
|
}
|
|
|
|
|
2019-10-09 19:58:42 +08:00
|
|
|
InitializeNativeTarget();
|
|
|
|
InitializeNativeTargetAsmPrinter();
|
|
|
|
InitializeNativeTargetDisassembler();
|
2020-02-07 00:18:42 +08:00
|
|
|
|
2018-05-15 20:08:00 +08:00
|
|
|
// Read benchmarks.
|
2018-10-25 15:44:01 +08:00
|
|
|
const LLVMState State("");
|
2020-02-07 00:18:42 +08:00
|
|
|
const std::vector<InstructionBenchmark> Points = ExitOnFileError(
|
|
|
|
BenchmarkFile, InstructionBenchmark::readYamls(State, BenchmarkFile));
|
|
|
|
|
2019-10-09 19:58:42 +08:00
|
|
|
outs() << "Parsed " << Points.size() << " benchmark points\n";
|
2018-05-15 20:08:00 +08:00
|
|
|
if (Points.empty()) {
|
2019-10-09 19:58:42 +08:00
|
|
|
errs() << "no benchmarks to analyze\n";
|
2018-05-15 20:08:00 +08:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
// FIXME: Check that all points have the same triple/cpu.
|
|
|
|
// FIXME: Merge points from several runs (latency and uops).
|
|
|
|
|
|
|
|
std::string Error;
|
|
|
|
const auto *TheTarget =
|
2019-10-09 19:58:42 +08:00
|
|
|
TargetRegistry::lookupTarget(Points[0].LLVMTriple, Error);
|
2018-05-15 20:08:00 +08:00
|
|
|
if (!TheTarget) {
|
2019-10-09 19:58:42 +08:00
|
|
|
errs() << "unknown target '" << Points[0].LLVMTriple << "'\n";
|
2018-05-15 20:08:00 +08:00
|
|
|
return;
|
|
|
|
}
|
[llvm-exegesis] Opcode stabilization / reclusterization (PR40715)
Summary:
Given an instruction `Opcode`, we can make benchmarks (measurements) of the
instruction characteristics/performance. Then, to facilitate further analysis
we group the benchmarks with *similar* characteristics into clusters.
Now, this is all not entirely deterministic. Some instructions have variable
characteristics, depending on their arguments. And thus, if we do several
benchmarks of the same instruction `Opcode`, we may end up with *different*
performance characteristics measurements. And when we then do clustering,
these several benchmarks of the same instruction `Opcode` may end up being
clustered into *different* clusters. This is not great for further analysis.
We shall find every `Opcode` with benchmarks not in just one cluster, and move
*all* the benchmarks of said `Opcode` into one new unstable cluster per `Opcode`.
I have solved this by making `ClusterId` a bit field, adding a `IsUnstable` bit,
and introducing `-analysis-display-unstable-clusters` switch to toggle between
displaying stable-only clusters and unstable-only clusters.
The reclusterization is deterministically stable, produces identical reports
between runs. (Or at least that is what i'm seeing, maybe it isn't)
Timings/comparisons:
old (current trunk/head) {F8303582}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-old.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-old.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-old.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-old.html' (25 runs):
6624.73 msec task-clock # 0.999 CPUs utilized ( +- 0.53% )
172 context-switches # 25.965 M/sec ( +- 29.89% )
0 cpu-migrations # 0.042 M/sec ( +- 56.54% )
31073 page-faults # 4690.754 M/sec ( +- 0.08% )
26538711696 cycles # 4006230.292 GHz ( +- 0.53% ) (83.31%)
2017496807 stalled-cycles-frontend # 7.60% frontend cycles idle ( +- 0.93% ) (83.32%)
13403650062 stalled-cycles-backend # 50.51% backend cycles idle ( +- 0.33% ) (33.37%)
19770706799 instructions # 0.74 insn per cycle
# 0.68 stalled cycles per insn ( +- 0.04% ) (50.04%)
4419821812 branches # 667207369.714 M/sec ( +- 0.03% ) (66.69%)
121741669 branch-misses # 2.75% of all branches ( +- 0.28% ) (83.34%)
6.6283 +- 0.0358 seconds time elapsed ( +- 0.54% )
```
patch, with reclustering but without filtering (i.e. outputting all the stable *and* unstable clusters) {F8303586}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-all.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-all.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-all.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-all.html' (25 runs):
6475.29 msec task-clock # 0.999 CPUs utilized ( +- 0.31% )
213 context-switches # 32.952 M/sec ( +- 23.81% )
1 cpu-migrations # 0.130 M/sec ( +- 43.84% )
31287 page-faults # 4832.057 M/sec ( +- 0.08% )
25939086577 cycles # 4006160.279 GHz ( +- 0.31% ) (83.31%)
1958812858 stalled-cycles-frontend # 7.55% frontend cycles idle ( +- 0.68% ) (83.32%)
13218961512 stalled-cycles-backend # 50.96% backend cycles idle ( +- 0.29% ) (33.37%)
19752995402 instructions # 0.76 insn per cycle
# 0.67 stalled cycles per insn ( +- 0.04% ) (50.04%)
4417079244 branches # 682195472.305 M/sec ( +- 0.03% ) (66.70%)
121510065 branch-misses # 2.75% of all branches ( +- 0.19% ) (83.34%)
6.4832 +- 0.0229 seconds time elapsed ( +- 0.35% )
```
Funnily, *this* measurement shows that said reclustering actually improved performance.
patch, with reclustering, only the stable clusters {F8303594}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-stable.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-stable.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-stable.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-stable.html' (25 runs):
6387.71 msec task-clock # 0.999 CPUs utilized ( +- 0.13% )
133 context-switches # 20.792 M/sec ( +- 23.39% )
0 cpu-migrations # 0.063 M/sec ( +- 61.24% )
31318 page-faults # 4903.256 M/sec ( +- 0.08% )
25591984967 cycles # 4006786.266 GHz ( +- 0.13% ) (83.31%)
1881234904 stalled-cycles-frontend # 7.35% frontend cycles idle ( +- 0.25% ) (83.33%)
13209749965 stalled-cycles-backend # 51.62% backend cycles idle ( +- 0.16% ) (33.36%)
19767554347 instructions # 0.77 insn per cycle
# 0.67 stalled cycles per insn ( +- 0.04% ) (50.03%)
4417480305 branches # 691618858.046 M/sec ( +- 0.03% ) (66.68%)
118676358 branch-misses # 2.69% of all branches ( +- 0.07% ) (83.33%)
6.3954 +- 0.0118 seconds time elapsed ( +- 0.18% )
```
Performance improved even further?! Makes sense i guess, less clusters to print.
patch, with reclustering, only the unstable clusters {F8303601}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-unstable.html -analysis-display-unstable-clusters
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-unstable.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-unstable.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-unstable.html -analysis-display-unstable-clusters' (25 runs):
6124.96 msec task-clock # 1.000 CPUs utilized ( +- 0.20% )
194 context-switches # 31.709 M/sec ( +- 20.46% )
0 cpu-migrations # 0.039 M/sec ( +- 49.77% )
31413 page-faults # 5129.261 M/sec ( +- 0.06% )
24536794267 cycles # 4006425.858 GHz ( +- 0.19% ) (83.31%)
1676085087 stalled-cycles-frontend # 6.83% frontend cycles idle ( +- 0.46% ) (83.32%)
13035595603 stalled-cycles-backend # 53.13% backend cycles idle ( +- 0.16% ) (33.36%)
18260877653 instructions # 0.74 insn per cycle
# 0.71 stalled cycles per insn ( +- 0.05% ) (50.03%)
4112411983 branches # 671484364.603 M/sec ( +- 0.03% ) (66.68%)
114066929 branch-misses # 2.77% of all branches ( +- 0.11% ) (83.32%)
6.1278 +- 0.0121 seconds time elapsed ( +- 0.20% )
```
This tells us that the actual `-analysis-inconsistencies-output-file=` outputting only takes ~0.4 sec for 43970 benchmark points (3 whole sweeps)
(Also, wow this is fast, it used to take several minutes originally)
Fixes [[ https://bugs.llvm.org/show_bug.cgi?id=40715 | PR40715 ]].
Reviewers: courbet, gchatelet
Reviewed By: courbet
Subscribers: tschuett, jdoerfert, llvm-commits, RKSimon
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D58355
llvm-svn: 354441
2019-02-20 17:14:04 +08:00
|
|
|
|
2019-10-09 19:58:42 +08:00
|
|
|
std::unique_ptr<MCInstrInfo> InstrInfo(TheTarget->createMCInstrInfo());
|
[llvm-exegesis] Opcode stabilization / reclusterization (PR40715)
Summary:
Given an instruction `Opcode`, we can make benchmarks (measurements) of the
instruction characteristics/performance. Then, to facilitate further analysis
we group the benchmarks with *similar* characteristics into clusters.
Now, this is all not entirely deterministic. Some instructions have variable
characteristics, depending on their arguments. And thus, if we do several
benchmarks of the same instruction `Opcode`, we may end up with *different*
performance characteristics measurements. And when we then do clustering,
these several benchmarks of the same instruction `Opcode` may end up being
clustered into *different* clusters. This is not great for further analysis.
We shall find every `Opcode` with benchmarks not in just one cluster, and move
*all* the benchmarks of said `Opcode` into one new unstable cluster per `Opcode`.
I have solved this by making `ClusterId` a bit field, adding a `IsUnstable` bit,
and introducing `-analysis-display-unstable-clusters` switch to toggle between
displaying stable-only clusters and unstable-only clusters.
The reclusterization is deterministically stable, produces identical reports
between runs. (Or at least that is what i'm seeing, maybe it isn't)
Timings/comparisons:
old (current trunk/head) {F8303582}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-old.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-old.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-old.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-old.html' (25 runs):
6624.73 msec task-clock # 0.999 CPUs utilized ( +- 0.53% )
172 context-switches # 25.965 M/sec ( +- 29.89% )
0 cpu-migrations # 0.042 M/sec ( +- 56.54% )
31073 page-faults # 4690.754 M/sec ( +- 0.08% )
26538711696 cycles # 4006230.292 GHz ( +- 0.53% ) (83.31%)
2017496807 stalled-cycles-frontend # 7.60% frontend cycles idle ( +- 0.93% ) (83.32%)
13403650062 stalled-cycles-backend # 50.51% backend cycles idle ( +- 0.33% ) (33.37%)
19770706799 instructions # 0.74 insn per cycle
# 0.68 stalled cycles per insn ( +- 0.04% ) (50.04%)
4419821812 branches # 667207369.714 M/sec ( +- 0.03% ) (66.69%)
121741669 branch-misses # 2.75% of all branches ( +- 0.28% ) (83.34%)
6.6283 +- 0.0358 seconds time elapsed ( +- 0.54% )
```
patch, with reclustering but without filtering (i.e. outputting all the stable *and* unstable clusters) {F8303586}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-all.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-all.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-all.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-all.html' (25 runs):
6475.29 msec task-clock # 0.999 CPUs utilized ( +- 0.31% )
213 context-switches # 32.952 M/sec ( +- 23.81% )
1 cpu-migrations # 0.130 M/sec ( +- 43.84% )
31287 page-faults # 4832.057 M/sec ( +- 0.08% )
25939086577 cycles # 4006160.279 GHz ( +- 0.31% ) (83.31%)
1958812858 stalled-cycles-frontend # 7.55% frontend cycles idle ( +- 0.68% ) (83.32%)
13218961512 stalled-cycles-backend # 50.96% backend cycles idle ( +- 0.29% ) (33.37%)
19752995402 instructions # 0.76 insn per cycle
# 0.67 stalled cycles per insn ( +- 0.04% ) (50.04%)
4417079244 branches # 682195472.305 M/sec ( +- 0.03% ) (66.70%)
121510065 branch-misses # 2.75% of all branches ( +- 0.19% ) (83.34%)
6.4832 +- 0.0229 seconds time elapsed ( +- 0.35% )
```
Funnily, *this* measurement shows that said reclustering actually improved performance.
patch, with reclustering, only the stable clusters {F8303594}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-stable.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-stable.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-stable.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-stable.html' (25 runs):
6387.71 msec task-clock # 0.999 CPUs utilized ( +- 0.13% )
133 context-switches # 20.792 M/sec ( +- 23.39% )
0 cpu-migrations # 0.063 M/sec ( +- 61.24% )
31318 page-faults # 4903.256 M/sec ( +- 0.08% )
25591984967 cycles # 4006786.266 GHz ( +- 0.13% ) (83.31%)
1881234904 stalled-cycles-frontend # 7.35% frontend cycles idle ( +- 0.25% ) (83.33%)
13209749965 stalled-cycles-backend # 51.62% backend cycles idle ( +- 0.16% ) (33.36%)
19767554347 instructions # 0.77 insn per cycle
# 0.67 stalled cycles per insn ( +- 0.04% ) (50.03%)
4417480305 branches # 691618858.046 M/sec ( +- 0.03% ) (66.68%)
118676358 branch-misses # 2.69% of all branches ( +- 0.07% ) (83.33%)
6.3954 +- 0.0118 seconds time elapsed ( +- 0.18% )
```
Performance improved even further?! Makes sense i guess, less clusters to print.
patch, with reclustering, only the unstable clusters {F8303601}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-unstable.html -analysis-display-unstable-clusters
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-unstable.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-unstable.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-unstable.html -analysis-display-unstable-clusters' (25 runs):
6124.96 msec task-clock # 1.000 CPUs utilized ( +- 0.20% )
194 context-switches # 31.709 M/sec ( +- 20.46% )
0 cpu-migrations # 0.039 M/sec ( +- 49.77% )
31413 page-faults # 5129.261 M/sec ( +- 0.06% )
24536794267 cycles # 4006425.858 GHz ( +- 0.19% ) (83.31%)
1676085087 stalled-cycles-frontend # 6.83% frontend cycles idle ( +- 0.46% ) (83.32%)
13035595603 stalled-cycles-backend # 53.13% backend cycles idle ( +- 0.16% ) (33.36%)
18260877653 instructions # 0.74 insn per cycle
# 0.71 stalled cycles per insn ( +- 0.05% ) (50.03%)
4112411983 branches # 671484364.603 M/sec ( +- 0.03% ) (66.68%)
114066929 branch-misses # 2.77% of all branches ( +- 0.11% ) (83.32%)
6.1278 +- 0.0121 seconds time elapsed ( +- 0.20% )
```
This tells us that the actual `-analysis-inconsistencies-output-file=` outputting only takes ~0.4 sec for 43970 benchmark points (3 whole sweeps)
(Also, wow this is fast, it used to take several minutes originally)
Fixes [[ https://bugs.llvm.org/show_bug.cgi?id=40715 | PR40715 ]].
Reviewers: courbet, gchatelet
Reviewed By: courbet
Subscribers: tschuett, jdoerfert, llvm-commits, RKSimon
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D58355
llvm-svn: 354441
2019-02-20 17:14:04 +08:00
|
|
|
|
2018-06-11 17:18:01 +08:00
|
|
|
const auto Clustering = ExitOnErr(InstructionBenchmarkClustering::create(
|
[llvm-exegesis] Introduce a 'naive' clustering algorithm (PR40880)
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
2019-03-28 16:55:01 +08:00
|
|
|
Points, AnalysisClusteringAlgorithm, AnalysisDbscanNumPoints,
|
|
|
|
AnalysisClusteringEpsilon, InstrInfo->getNumOpcodes()));
|
2018-05-16 16:47:21 +08:00
|
|
|
|
[llvm-exegesis] Opcode stabilization / reclusterization (PR40715)
Summary:
Given an instruction `Opcode`, we can make benchmarks (measurements) of the
instruction characteristics/performance. Then, to facilitate further analysis
we group the benchmarks with *similar* characteristics into clusters.
Now, this is all not entirely deterministic. Some instructions have variable
characteristics, depending on their arguments. And thus, if we do several
benchmarks of the same instruction `Opcode`, we may end up with *different*
performance characteristics measurements. And when we then do clustering,
these several benchmarks of the same instruction `Opcode` may end up being
clustered into *different* clusters. This is not great for further analysis.
We shall find every `Opcode` with benchmarks not in just one cluster, and move
*all* the benchmarks of said `Opcode` into one new unstable cluster per `Opcode`.
I have solved this by making `ClusterId` a bit field, adding a `IsUnstable` bit,
and introducing `-analysis-display-unstable-clusters` switch to toggle between
displaying stable-only clusters and unstable-only clusters.
The reclusterization is deterministically stable, produces identical reports
between runs. (Or at least that is what i'm seeing, maybe it isn't)
Timings/comparisons:
old (current trunk/head) {F8303582}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-old.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-old.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-old.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-old.html' (25 runs):
6624.73 msec task-clock # 0.999 CPUs utilized ( +- 0.53% )
172 context-switches # 25.965 M/sec ( +- 29.89% )
0 cpu-migrations # 0.042 M/sec ( +- 56.54% )
31073 page-faults # 4690.754 M/sec ( +- 0.08% )
26538711696 cycles # 4006230.292 GHz ( +- 0.53% ) (83.31%)
2017496807 stalled-cycles-frontend # 7.60% frontend cycles idle ( +- 0.93% ) (83.32%)
13403650062 stalled-cycles-backend # 50.51% backend cycles idle ( +- 0.33% ) (33.37%)
19770706799 instructions # 0.74 insn per cycle
# 0.68 stalled cycles per insn ( +- 0.04% ) (50.04%)
4419821812 branches # 667207369.714 M/sec ( +- 0.03% ) (66.69%)
121741669 branch-misses # 2.75% of all branches ( +- 0.28% ) (83.34%)
6.6283 +- 0.0358 seconds time elapsed ( +- 0.54% )
```
patch, with reclustering but without filtering (i.e. outputting all the stable *and* unstable clusters) {F8303586}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-all.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-all.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-all.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-all.html' (25 runs):
6475.29 msec task-clock # 0.999 CPUs utilized ( +- 0.31% )
213 context-switches # 32.952 M/sec ( +- 23.81% )
1 cpu-migrations # 0.130 M/sec ( +- 43.84% )
31287 page-faults # 4832.057 M/sec ( +- 0.08% )
25939086577 cycles # 4006160.279 GHz ( +- 0.31% ) (83.31%)
1958812858 stalled-cycles-frontend # 7.55% frontend cycles idle ( +- 0.68% ) (83.32%)
13218961512 stalled-cycles-backend # 50.96% backend cycles idle ( +- 0.29% ) (33.37%)
19752995402 instructions # 0.76 insn per cycle
# 0.67 stalled cycles per insn ( +- 0.04% ) (50.04%)
4417079244 branches # 682195472.305 M/sec ( +- 0.03% ) (66.70%)
121510065 branch-misses # 2.75% of all branches ( +- 0.19% ) (83.34%)
6.4832 +- 0.0229 seconds time elapsed ( +- 0.35% )
```
Funnily, *this* measurement shows that said reclustering actually improved performance.
patch, with reclustering, only the stable clusters {F8303594}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-stable.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-stable.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-stable.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-stable.html' (25 runs):
6387.71 msec task-clock # 0.999 CPUs utilized ( +- 0.13% )
133 context-switches # 20.792 M/sec ( +- 23.39% )
0 cpu-migrations # 0.063 M/sec ( +- 61.24% )
31318 page-faults # 4903.256 M/sec ( +- 0.08% )
25591984967 cycles # 4006786.266 GHz ( +- 0.13% ) (83.31%)
1881234904 stalled-cycles-frontend # 7.35% frontend cycles idle ( +- 0.25% ) (83.33%)
13209749965 stalled-cycles-backend # 51.62% backend cycles idle ( +- 0.16% ) (33.36%)
19767554347 instructions # 0.77 insn per cycle
# 0.67 stalled cycles per insn ( +- 0.04% ) (50.03%)
4417480305 branches # 691618858.046 M/sec ( +- 0.03% ) (66.68%)
118676358 branch-misses # 2.69% of all branches ( +- 0.07% ) (83.33%)
6.3954 +- 0.0118 seconds time elapsed ( +- 0.18% )
```
Performance improved even further?! Makes sense i guess, less clusters to print.
patch, with reclustering, only the unstable clusters {F8303601}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-unstable.html -analysis-display-unstable-clusters
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-unstable.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-unstable.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-unstable.html -analysis-display-unstable-clusters' (25 runs):
6124.96 msec task-clock # 1.000 CPUs utilized ( +- 0.20% )
194 context-switches # 31.709 M/sec ( +- 20.46% )
0 cpu-migrations # 0.039 M/sec ( +- 49.77% )
31413 page-faults # 5129.261 M/sec ( +- 0.06% )
24536794267 cycles # 4006425.858 GHz ( +- 0.19% ) (83.31%)
1676085087 stalled-cycles-frontend # 6.83% frontend cycles idle ( +- 0.46% ) (83.32%)
13035595603 stalled-cycles-backend # 53.13% backend cycles idle ( +- 0.16% ) (33.36%)
18260877653 instructions # 0.74 insn per cycle
# 0.71 stalled cycles per insn ( +- 0.05% ) (50.03%)
4112411983 branches # 671484364.603 M/sec ( +- 0.03% ) (66.68%)
114066929 branch-misses # 2.77% of all branches ( +- 0.11% ) (83.32%)
6.1278 +- 0.0121 seconds time elapsed ( +- 0.20% )
```
This tells us that the actual `-analysis-inconsistencies-output-file=` outputting only takes ~0.4 sec for 43970 benchmark points (3 whole sweeps)
(Also, wow this is fast, it used to take several minutes originally)
Fixes [[ https://bugs.llvm.org/show_bug.cgi?id=40715 | PR40715 ]].
Reviewers: courbet, gchatelet
Reviewed By: courbet
Subscribers: tschuett, jdoerfert, llvm-commits, RKSimon
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D58355
llvm-svn: 354441
2019-02-20 17:14:04 +08:00
|
|
|
const Analysis Analyzer(*TheTarget, std::move(InstrInfo), Clustering,
|
[llvm-exegesis] Split Epsilon param into two (PR40787)
Summary:
This eps param is used for two distinct things:
* initial point clusterization
* checking clusters against the llvm values
What if one wants to only look at highly different clusters, without changing
the clustering itself? In particular, this helps to weed out noisy measurements
(since the clusterization epsilon is still small, so there is a better chance
that noisy measurements from the same opcode will go into different clusters)
By splitting it into two params it is now possible.
This is nearly-free performance-wise:
Old:
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency-1.yaml -analysis-inconsistencies-output-file=/tmp/clusters-old.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 10099 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-old.html'
...
Performance counter stats for './bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency-1.yaml -analysis-inconsistencies-output-file=/tmp/clusters-old.html' (25 runs):
390.01 msec task-clock # 0.998 CPUs utilized ( +- 0.25% )
12 context-switches # 31.735 M/sec ( +- 27.38% )
0 cpu-migrations # 0.000 K/sec
4745 page-faults # 12183.732 M/sec ( +- 0.54% )
1562711900 cycles # 4012303.327 GHz ( +- 0.24% ) (82.90%)
185567822 stalled-cycles-frontend # 11.87% frontend cycles idle ( +- 0.52% ) (83.30%)
392106234 stalled-cycles-backend # 25.09% backend cycles idle ( +- 1.31% ) (33.79%)
1839236666 instructions # 1.18 insn per cycle
# 0.21 stalled cycles per insn ( +- 0.15% ) (50.37%)
407035764 branches # 1045074878.710 M/sec ( +- 0.12% ) (66.80%)
10896459 branch-misses # 2.68% of all branches ( +- 0.17% ) (83.20%)
0.390629 +- 0.000972 seconds time elapsed ( +- 0.25% )
```
```
$ perf stat -r 9 ./bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency.yml -analysis-inconsistencies-output-file=/tmp/clusters-old.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 50572 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-old.html'
...
Performance counter stats for './bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency.yml -analysis-inconsistencies-output-file=/tmp/clusters-old.html' (9 runs):
6803.36 msec task-clock # 0.999 CPUs utilized ( +- 0.96% )
262 context-switches # 38.546 M/sec ( +- 23.06% )
0 cpu-migrations # 0.065 M/sec ( +- 76.03% )
13287 page-faults # 1953.206 M/sec ( +- 0.32% )
27252537904 cycles # 4006024.257 GHz ( +- 0.95% ) (83.31%)
1496314935 stalled-cycles-frontend # 5.49% frontend cycles idle ( +- 0.97% ) (83.32%)
16128404524 stalled-cycles-backend # 59.18% backend cycles idle ( +- 0.30% ) (33.37%)
17611143370 instructions # 0.65 insn per cycle
# 0.92 stalled cycles per insn ( +- 0.05% ) (50.04%)
3894906599 branches # 572537147.437 M/sec ( +- 0.03% ) (66.69%)
116314514 branch-misses # 2.99% of all branches ( +- 0.20% ) (83.35%)
6.8118 +- 0.0689 seconds time elapsed ( +- 1.01%)
```
New:
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency-1.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 10099 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new.html'
...
Performance counter stats for './bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency-1.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new.html' (25 runs):
400.14 msec task-clock # 0.998 CPUs utilized ( +- 0.66% )
12 context-switches # 29.429 M/sec ( +- 25.95% )
0 cpu-migrations # 0.100 M/sec ( +-100.00% )
4714 page-faults # 11796.496 M/sec ( +- 0.55% )
1603131306 cycles # 4011840.105 GHz ( +- 0.66% ) (82.85%)
199538509 stalled-cycles-frontend # 12.45% frontend cycles idle ( +- 2.40% ) (83.10%)
402249109 stalled-cycles-backend # 25.09% backend cycles idle ( +- 1.19% ) (34.05%)
1847783963 instructions # 1.15 insn per cycle
# 0.22 stalled cycles per insn ( +- 0.18% ) (50.64%)
407162722 branches # 1018925730.631 M/sec ( +- 0.12% ) (67.02%)
10932779 branch-misses # 2.69% of all branches ( +- 0.51% ) (83.28%)
0.40077 +- 0.00267 seconds time elapsed ( +- 0.67% )
lebedevri@pini-pini:/build/llvm-build-Clang-release$ perf stat -r 9 ./bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency.yml -analysis-inconsistencies-output-file=/tmp/clusters-new.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 50572 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new.html'
...
Performance counter stats for './bin/llvm-exegesis -mode=analysis -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-latency.yml -analysis-inconsistencies-output-file=/tmp/clusters-new.html' (9 runs):
6947.79 msec task-clock # 1.000 CPUs utilized ( +- 0.90% )
217 context-switches # 31.236 M/sec ( +- 36.16% )
1 cpu-migrations # 0.096 M/sec ( +- 50.00% )
13258 page-faults # 1908.389 M/sec ( +- 0.34% )
27830796523 cycles # 4006032.286 GHz ( +- 0.89% ) (83.30%)
1504554006 stalled-cycles-frontend # 5.41% frontend cycles idle ( +- 2.10% ) (83.32%)
16716574843 stalled-cycles-backend # 60.07% backend cycles idle ( +- 0.65% ) (33.38%)
17755545931 instructions # 0.64 insn per cycle
# 0.94 stalled cycles per insn ( +- 0.09% ) (50.04%)
3897255686 branches # 560980426.597 M/sec ( +- 0.06% ) (66.70%)
117045395 branch-misses # 3.00% of all branches ( +- 0.47% ) (83.34%)
6.9507 +- 0.0627 seconds time elapsed ( +- 0.90% )
```
I.e. it's +2.6% slowdown for one whole sweep, or +2% for 5 whole sweeps.
Within noise i'd say.
Should help with [[ https://bugs.llvm.org/show_bug.cgi?id=40787 | PR40787 ]].
Reviewers: courbet, gchatelet
Reviewed By: courbet
Subscribers: tschuett, RKSimon, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D58476
llvm-svn: 354767
2019-02-25 17:36:12 +08:00
|
|
|
AnalysisInconsistencyEpsilon,
|
[llvm-exegesis] Opcode stabilization / reclusterization (PR40715)
Summary:
Given an instruction `Opcode`, we can make benchmarks (measurements) of the
instruction characteristics/performance. Then, to facilitate further analysis
we group the benchmarks with *similar* characteristics into clusters.
Now, this is all not entirely deterministic. Some instructions have variable
characteristics, depending on their arguments. And thus, if we do several
benchmarks of the same instruction `Opcode`, we may end up with *different*
performance characteristics measurements. And when we then do clustering,
these several benchmarks of the same instruction `Opcode` may end up being
clustered into *different* clusters. This is not great for further analysis.
We shall find every `Opcode` with benchmarks not in just one cluster, and move
*all* the benchmarks of said `Opcode` into one new unstable cluster per `Opcode`.
I have solved this by making `ClusterId` a bit field, adding a `IsUnstable` bit,
and introducing `-analysis-display-unstable-clusters` switch to toggle between
displaying stable-only clusters and unstable-only clusters.
The reclusterization is deterministically stable, produces identical reports
between runs. (Or at least that is what i'm seeing, maybe it isn't)
Timings/comparisons:
old (current trunk/head) {F8303582}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-old.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-old.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-old.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-old.html' (25 runs):
6624.73 msec task-clock # 0.999 CPUs utilized ( +- 0.53% )
172 context-switches # 25.965 M/sec ( +- 29.89% )
0 cpu-migrations # 0.042 M/sec ( +- 56.54% )
31073 page-faults # 4690.754 M/sec ( +- 0.08% )
26538711696 cycles # 4006230.292 GHz ( +- 0.53% ) (83.31%)
2017496807 stalled-cycles-frontend # 7.60% frontend cycles idle ( +- 0.93% ) (83.32%)
13403650062 stalled-cycles-backend # 50.51% backend cycles idle ( +- 0.33% ) (33.37%)
19770706799 instructions # 0.74 insn per cycle
# 0.68 stalled cycles per insn ( +- 0.04% ) (50.04%)
4419821812 branches # 667207369.714 M/sec ( +- 0.03% ) (66.69%)
121741669 branch-misses # 2.75% of all branches ( +- 0.28% ) (83.34%)
6.6283 +- 0.0358 seconds time elapsed ( +- 0.54% )
```
patch, with reclustering but without filtering (i.e. outputting all the stable *and* unstable clusters) {F8303586}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-all.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-all.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-all.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-all.html' (25 runs):
6475.29 msec task-clock # 0.999 CPUs utilized ( +- 0.31% )
213 context-switches # 32.952 M/sec ( +- 23.81% )
1 cpu-migrations # 0.130 M/sec ( +- 43.84% )
31287 page-faults # 4832.057 M/sec ( +- 0.08% )
25939086577 cycles # 4006160.279 GHz ( +- 0.31% ) (83.31%)
1958812858 stalled-cycles-frontend # 7.55% frontend cycles idle ( +- 0.68% ) (83.32%)
13218961512 stalled-cycles-backend # 50.96% backend cycles idle ( +- 0.29% ) (33.37%)
19752995402 instructions # 0.76 insn per cycle
# 0.67 stalled cycles per insn ( +- 0.04% ) (50.04%)
4417079244 branches # 682195472.305 M/sec ( +- 0.03% ) (66.70%)
121510065 branch-misses # 2.75% of all branches ( +- 0.19% ) (83.34%)
6.4832 +- 0.0229 seconds time elapsed ( +- 0.35% )
```
Funnily, *this* measurement shows that said reclustering actually improved performance.
patch, with reclustering, only the stable clusters {F8303594}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-stable.html
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-stable.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-stable.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-stable.html' (25 runs):
6387.71 msec task-clock # 0.999 CPUs utilized ( +- 0.13% )
133 context-switches # 20.792 M/sec ( +- 23.39% )
0 cpu-migrations # 0.063 M/sec ( +- 61.24% )
31318 page-faults # 4903.256 M/sec ( +- 0.08% )
25591984967 cycles # 4006786.266 GHz ( +- 0.13% ) (83.31%)
1881234904 stalled-cycles-frontend # 7.35% frontend cycles idle ( +- 0.25% ) (83.33%)
13209749965 stalled-cycles-backend # 51.62% backend cycles idle ( +- 0.16% ) (33.36%)
19767554347 instructions # 0.77 insn per cycle
# 0.67 stalled cycles per insn ( +- 0.04% ) (50.03%)
4417480305 branches # 691618858.046 M/sec ( +- 0.03% ) (66.68%)
118676358 branch-misses # 2.69% of all branches ( +- 0.07% ) (83.33%)
6.3954 +- 0.0118 seconds time elapsed ( +- 0.18% )
```
Performance improved even further?! Makes sense i guess, less clusters to print.
patch, with reclustering, only the unstable clusters {F8303601}
```
$ perf stat -r 25 ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-unstable.html -analysis-display-unstable-clusters
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-unstable.html'
...
no exegesis target for x86_64-unknown-linux-gnu, using default
Parsed 43970 benchmark points
Printing sched class consistency analysis results to file '/tmp/clusters-new-unstable.html'
Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=0.5 -benchmarks-file=/home/lebedevri/PileDriver-Sched/benchmarks-inverse_throughput.yaml -analysis-inconsistencies-output-file=/tmp/clusters-new-unstable.html -analysis-display-unstable-clusters' (25 runs):
6124.96 msec task-clock # 1.000 CPUs utilized ( +- 0.20% )
194 context-switches # 31.709 M/sec ( +- 20.46% )
0 cpu-migrations # 0.039 M/sec ( +- 49.77% )
31413 page-faults # 5129.261 M/sec ( +- 0.06% )
24536794267 cycles # 4006425.858 GHz ( +- 0.19% ) (83.31%)
1676085087 stalled-cycles-frontend # 6.83% frontend cycles idle ( +- 0.46% ) (83.32%)
13035595603 stalled-cycles-backend # 53.13% backend cycles idle ( +- 0.16% ) (33.36%)
18260877653 instructions # 0.74 insn per cycle
# 0.71 stalled cycles per insn ( +- 0.05% ) (50.03%)
4112411983 branches # 671484364.603 M/sec ( +- 0.03% ) (66.68%)
114066929 branch-misses # 2.77% of all branches ( +- 0.11% ) (83.32%)
6.1278 +- 0.0121 seconds time elapsed ( +- 0.20% )
```
This tells us that the actual `-analysis-inconsistencies-output-file=` outputting only takes ~0.4 sec for 43970 benchmark points (3 whole sweeps)
(Also, wow this is fast, it used to take several minutes originally)
Fixes [[ https://bugs.llvm.org/show_bug.cgi?id=40715 | PR40715 ]].
Reviewers: courbet, gchatelet
Reviewed By: courbet
Subscribers: tschuett, jdoerfert, llvm-commits, RKSimon
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D58355
llvm-svn: 354441
2019-02-20 17:14:04 +08:00
|
|
|
AnalysisDisplayUnstableOpcodes);
|
2018-05-16 16:47:21 +08:00
|
|
|
|
2018-05-17 21:41:28 +08:00
|
|
|
maybeRunAnalysis<Analysis::PrintClusters>(Analyzer, "analysis clusters",
|
|
|
|
AnalysisClustersOutputFile);
|
|
|
|
maybeRunAnalysis<Analysis::PrintSchedClassInconsistencies>(
|
|
|
|
Analyzer, "sched class consistency analysis",
|
|
|
|
AnalysisInconsistenciesOutputFile);
|
2018-04-04 19:37:06 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
} // namespace exegesis
|
2018-10-23 01:10:47 +08:00
|
|
|
} // namespace llvm
|
2018-04-04 19:37:06 +08:00
|
|
|
|
|
|
|
int main(int Argc, char **Argv) {
|
2018-10-23 01:10:47 +08:00
|
|
|
using namespace llvm;
|
|
|
|
cl::ParseCommandLineOptions(Argc, Argv, "");
|
2018-04-04 19:37:06 +08:00
|
|
|
|
2019-10-09 19:58:42 +08:00
|
|
|
exegesis::ExitOnErr.setExitCodeMapper([](const Error &Err) {
|
2020-02-07 00:08:05 +08:00
|
|
|
if (Err.isA<exegesis::ClusteringError>())
|
2018-06-11 17:18:01 +08:00
|
|
|
return EXIT_SUCCESS;
|
|
|
|
return EXIT_FAILURE;
|
|
|
|
});
|
|
|
|
|
2018-10-23 01:10:47 +08:00
|
|
|
if (exegesis::BenchmarkMode == exegesis::InstructionBenchmark::Unknown) {
|
2018-05-15 20:08:00 +08:00
|
|
|
exegesis::analysisMain();
|
|
|
|
} else {
|
|
|
|
exegesis::benchmarkMain();
|
2018-04-04 19:37:06 +08:00
|
|
|
}
|
|
|
|
return EXIT_SUCCESS;
|
|
|
|
}
|