[ctx_prof] CtxProfAnalysis #102084
Merged
[ctx_prof] CtxProfAnalysis #102084
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This is an immutable analysis that loads and makes the contextual profile available to other passes. This patch just introduces the analysis and an analysis printer pass. Subsequent patches will introduce the APIs that IPO passes will call to modify the profile as result of their changes.
|
@llvm/pr-subscribers-llvm-analysis Author: Mircea Trofin (mtrofin) ChangesThis is an immutable analysis that loads and makes the contextual profile available to other passes. This patch just introduces the analysis and an analysis printer pass. Subsequent patches will introduce the APIs that IPO passes will call to modify the profile as result of their changes. Full diff: https://github.com/llvm/llvm-project/pull/102084.diff 7 Files Affected:
diff --git a/llvm/include/llvm/Analysis/CtxProfAnalysis.h b/llvm/include/llvm/Analysis/CtxProfAnalysis.h
new file mode 100644
index 0000000000000..dbdbc3a64a0ac
--- /dev/null
+++ b/llvm/include/llvm/Analysis/CtxProfAnalysis.h
@@ -0,0 +1,53 @@
+//===- CtxProfAnalysis.h - maintain contextual profile info -*- C++ ---*-===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+//
+#ifndef LLVM_ANALYSIS_CTXPROFANALYSIS_H
+#define LLVM_ANALYSIS_CTXPROFANALYSIS_H
+
+#include "llvm/IR/GlobalValue.h"
+#include "llvm/IR/PassManager.h"
+#include "llvm/ProfileData/PGOCtxProfReader.h"
+#include <map>
+
+namespace llvm {
+class CtxProfAnalysis : public AnalysisInfoMixin<CtxProfAnalysis> {
+ StringRef Profile;
+public:
+ static AnalysisKey Key;
+ explicit CtxProfAnalysis(StringRef Profile) : Profile(Profile) {};
+
+ class Result {
+ std::optional<PGOContextualProfile::CallTargetMapTy> Profiles;
+ public:
+ explicit Result(PGOContextualProfile::CallTargetMapTy &&Profiles)
+ : Profiles(std::move(Profiles)) {}
+ Result() = default;
+ Result(const Result&) = delete;
+ Result(Result &&) = default;
+
+ operator bool() const { return !!Profiles; }
+ const PGOContextualProfile::CallTargetMapTy &profiles() const {
+ return *Profiles;
+ }
+ };
+
+ Result run(Module &M, ModuleAnalysisManager &MAM);
+};
+
+class CtxProfAnalysisPrinterPass
+ : public PassInfoMixin<CtxProfAnalysisPrinterPass> {
+ raw_ostream &OS;
+
+public:
+ explicit CtxProfAnalysisPrinterPass(raw_ostream &OS) : OS(OS) {}
+
+ PreservedAnalyses run(Module &M, ModuleAnalysisManager &MAM);
+ static bool isRequired() { return true; }
+};
+} // namespace llvm
+#endif // LLVM_ANALYSIS_CTXPROFANALYSIS_H
diff --git a/llvm/lib/Analysis/CMakeLists.txt b/llvm/lib/Analysis/CMakeLists.txt
index 997bb7a0bb178..2cb3547ec4047 100644
--- a/llvm/lib/Analysis/CMakeLists.txt
+++ b/llvm/lib/Analysis/CMakeLists.txt
@@ -46,6 +46,7 @@ add_llvm_component_library(LLVMAnalysis
CostModel.cpp
CodeMetrics.cpp
ConstantFolding.cpp
+ CtxProfAnalysis.cpp
CycleAnalysis.cpp
DDG.cpp
DDGPrinter.cpp
diff --git a/llvm/lib/Analysis/CtxProfAnalysis.cpp b/llvm/lib/Analysis/CtxProfAnalysis.cpp
new file mode 100644
index 0000000000000..0ac67ac863ee2
--- /dev/null
+++ b/llvm/lib/Analysis/CtxProfAnalysis.cpp
@@ -0,0 +1,91 @@
+//===- CtxProfAnalysis.cpp - contextual profile analysis ------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+//
+// Implementation of the contextual profile analysis, which maintains contextual
+// profiling info through IPO passes.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/CtxProfAnalysis.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/IR/Analysis.h"
+#include "llvm/IR/Module.h"
+#include "llvm/IR/PassManager.h"
+#include "llvm/ProfileData/PGOCtxProfReader.h"
+#include "llvm/Support/MemoryBuffer.h"
+#include "llvm/Support/JSON.h"
+
+namespace llvm {
+namespace json {
+Value toJSON(const PGOContextualProfile &P) {
+ Object Ret;
+ Ret["Guid"] = P.guid();
+ Ret["Counters"] = Array(P.counters());
+ auto AllCS =
+ ::llvm::map_range(P.callsites(), [](const auto &P) { return P.first; });
+ auto MaxIt = ::llvm::max_element(AllCS);
+ if (MaxIt != AllCS.end()) {
+ Array CSites;
+ // Iterate to, and including, the maximum index.
+ for (auto I = 0U; I <= *MaxIt; ++I) {
+ CSites.push_back(Array());
+ Array &Targets = *CSites.back().getAsArray();
+ if (P.hasCallsite(I))
+ for (const auto &[_, Ctx] : P.callsite(I))
+ Targets.push_back(toJSON(Ctx));
+ }
+ Ret["Callsites"] = std::move(CSites);
+ }
+ return Ret;
+}
+
+Value toJSON(const PGOContextualProfile::CallTargetMapTy &P) {
+ Array Ret;
+ for (const auto &[_, Ctx] : P)
+ Ret.push_back(toJSON(Ctx));
+ return Ret;
+}
+} // namespace json
+} // namespace llvm
+
+using namespace llvm;
+#define DEBUG_TYPE "ctx_prof"
+
+AnalysisKey CtxProfAnalysis::Key;
+
+CtxProfAnalysis::Result CtxProfAnalysis::run(Module &M,
+ ModuleAnalysisManager &MAM) {
+ ErrorOr<std::unique_ptr<MemoryBuffer>> MB = MemoryBuffer::getFile(Profile);
+ if (auto EC = MB.getError()) {
+ M.getContext().emitError("could not open contextual profile file: " +
+ EC.message());
+ return {};
+ }
+ PGOCtxProfileReader Reader(MB.get()->getBuffer());
+ auto MaybeCtx = Reader.loadContexts();
+ if (!MaybeCtx) {
+ M.getContext().emitError("contextual profile file is invalid: " +
+ toString(MaybeCtx.takeError()));
+ return {};
+ }
+ return Result(std::move(*MaybeCtx));
+}
+
+PreservedAnalyses CtxProfAnalysisPrinterPass::run(Module &M,
+ ModuleAnalysisManager &MAM) {
+ CtxProfAnalysis::Result &C = MAM.getResult<CtxProfAnalysis>(M);
+ if (!C) {
+ M.getContext().emitError("Invalid CtxProfAnalysis");
+ return PreservedAnalyses::all();
+ }
+ const auto JSONed = ::llvm::json::toJSON(C.profiles());
+
+ OS << formatv("{0:2}", JSONed);
+ OS << "\n";
+ return PreservedAnalyses::all();
+}
diff --git a/llvm/lib/Passes/PassBuilder.cpp b/llvm/lib/Passes/PassBuilder.cpp
index 5dbb1e2f49871..bcc69d5ac3db6 100644
--- a/llvm/lib/Passes/PassBuilder.cpp
+++ b/llvm/lib/Passes/PassBuilder.cpp
@@ -28,6 +28,7 @@
#include "llvm/Analysis/CallGraph.h"
#include "llvm/Analysis/CallPrinter.h"
#include "llvm/Analysis/CostModel.h"
+#include "llvm/Analysis/CtxProfAnalysis.h"
#include "llvm/Analysis/CycleAnalysis.h"
#include "llvm/Analysis/DDG.h"
#include "llvm/Analysis/DDGPrinter.h"
@@ -330,6 +331,8 @@ cl::opt<bool> PrintPipelinePasses(
"(best-effort only)."));
} // namespace llvm
+extern cl::opt<std::string> UseCtxProfile;
+
AnalysisKey NoOpModuleAnalysis::Key;
AnalysisKey NoOpCGSCCAnalysis::Key;
AnalysisKey NoOpFunctionAnalysis::Key;
diff --git a/llvm/lib/Passes/PassBuilderPipelines.cpp b/llvm/lib/Passes/PassBuilderPipelines.cpp
index adebbb5eeba32..c175ee8980984 100644
--- a/llvm/lib/Passes/PassBuilderPipelines.cpp
+++ b/llvm/lib/Passes/PassBuilderPipelines.cpp
@@ -304,7 +304,7 @@ static cl::opt<bool> UseLoopVersioningLICM(
"enable-loop-versioning-licm", cl::init(false), cl::Hidden,
cl::desc("Enable the experimental Loop Versioning LICM pass"));
-static cl::opt<std::string>
+cl::opt<std::string>
UseCtxProfile("use-ctx-profile", cl::init(""), cl::Hidden,
cl::desc("Use the specified contextual profile file"));
diff --git a/llvm/lib/Passes/PassRegistry.def b/llvm/lib/Passes/PassRegistry.def
index 3b92823cd283b..2365ca4d3d88a 100644
--- a/llvm/lib/Passes/PassRegistry.def
+++ b/llvm/lib/Passes/PassRegistry.def
@@ -20,6 +20,7 @@
#endif
MODULE_ANALYSIS("callgraph", CallGraphAnalysis())
MODULE_ANALYSIS("collector-metadata", CollectorMetadataAnalysis())
+MODULE_ANALYSIS("ctx-prof-analysis", CtxProfAnalysis(UseCtxProfile))
MODULE_ANALYSIS("inline-advisor", InlineAdvisorAnalysis())
MODULE_ANALYSIS("ir-similarity", IRSimilarityAnalysis())
MODULE_ANALYSIS("lcg", LazyCallGraphAnalysis())
@@ -79,6 +80,7 @@ MODULE_PASS("insert-gcov-profiling", GCOVProfilerPass())
MODULE_PASS("instrorderfile", InstrOrderFilePass())
MODULE_PASS("instrprof", InstrProfilingLoweringPass())
MODULE_PASS("ctx-instr-lower", PGOCtxProfLoweringPass())
+MODULE_PASS("print<ctx-prof-analysis>", CtxProfAnalysisPrinterPass(dbgs()))
MODULE_PASS("invalidate<all>", InvalidateAllAnalysesPass())
MODULE_PASS("iroutliner", IROutlinerPass())
MODULE_PASS("jmc-instrumenter", JMCInstrumenterPass())
diff --git a/llvm/test/Analysis/CtxProfAnalysis/load.ll b/llvm/test/Analysis/CtxProfAnalysis/load.ll
new file mode 100644
index 0000000000000..fdf40d1b7f136
--- /dev/null
+++ b/llvm/test/Analysis/CtxProfAnalysis/load.ll
@@ -0,0 +1,56 @@
+; RUN: split-file %s %t
+; RUN: llvm-ctxprof-util fromJSON --input=%t/profile.json --output=%t/profile.ctxprofdata
+; RUN: not opt -passes='require<ctx-prof-analysis>,print<ctx-prof-analysis>' \
+; RUN: %t/empty.ll -S 2>&1 | FileCheck %s --check-prefix=NO-FILE
+
+; RUN: not opt -passes='require<ctx-prof-analysis>,print<ctx-prof-analysis>' \
+; RUN: -use-ctx-profile=does_not_exist.ctxprofdata %t/empty.ll -S 2>&1 | FileCheck %s --check-prefix=NO-FILE
+
+; RUN: opt -passes='require<ctx-prof-analysis>,print<ctx-prof-analysis>' \
+; RUN: -use-ctx-profile=%t/profile.ctxprofdata %t/empty.ll -S 2> %t/output.json
+; RUN: diff %t/profile.json %t/output.json
+
+; NO-FILE: error: could not open contextual profile file
+;
+; This is the reference profile, laid out in the format the json formatter will
+; output it from opt.
+;--- profile.json
+[
+ {
+ "Callsites": [
+ [],
+ [
+ {
+ "Counters": [
+ 4,
+ 5
+ ],
+ "Guid": 2000
+ },
+ {
+ "Counters": [
+ 6,
+ 7,
+ 8
+ ],
+ "Guid": 18446744073709551613
+ }
+ ]
+ ],
+ "Counters": [
+ 1,
+ 2,
+ 3
+ ],
+ "Guid": 1000
+ },
+ {
+ "Counters": [
+ 5,
+ 9,
+ 10
+ ],
+ "Guid": 18446744073709551612
+ }
+]
+;--- empty.ll
|
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
snehasish
reviewed
Aug 7, 2024
to be a standalone class with a nice name.
snehasish
approved these changes
Aug 7, 2024
Was confused what you meant, until I realized I forgot to push my replies :) |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/59/builds/2798 Here is the relevant piece of the build log for the reference: |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/46/builds/2859 Here is the relevant piece of the build log for the reference: |
mtrofin
added a commit
that referenced
this pull request
Aug 7, 2024
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/155/builds/1269 Here is the relevant piece of the build log for the reference: |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/54/builds/1184 Here is the relevant piece of the build log for the reference: |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/35/builds/1686 Here is the relevant piece of the build log for the reference: |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/193/builds/1665 Here is the relevant piece of the build log for the reference: |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/161/builds/1132 Here is the relevant piece of the build log for the reference: |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/81/builds/1302 Here is the relevant piece of the build log for the reference: |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/141/builds/1375 Here is the relevant piece of the build log for the reference: |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/63/builds/928 Here is the relevant piece of the build log for the reference: |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/38/builds/21 Here is the relevant piece of the build log for the reference: |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/14/builds/596 Here is the relevant piece of the build log for the reference: |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/146/builds/380 Here is the relevant piece of the build log for the reference: |
mtrofin
added a commit
to mtrofin/llvm-project
that referenced
this pull request
Aug 12, 2024
Continuing from llvm#102084, which introduced the analysis, we now populate it with info about functions contained in the module. When we will update the profile due to e.g. inlined callsites, we'll ingest the callee's counters and callsites to the caller. We'll move those to the caller's respective index space (counter and callers), so we need to know and maintain where those currently end. We also don't need to keep profiles not pertinent to this module. This patch also introduces an arguably much simpler way to track the GUID of a function from the frontend compilation, through ThinLTO, and into the post-thinlink compilation step, which doesn't rely on keeping names around. A separate RFC and patches will discuss extending this to the current PGO (instrumented and sampled) and other consumers as an infrastructural component.
mtrofin
added a commit
to mtrofin/llvm-project
that referenced
this pull request
Aug 12, 2024
Continuing from llvm#102084, which introduced the analysis, we now populate it with info about functions contained in the module. When we will update the profile due to e.g. inlined callsites, we'll ingest the callee's counters and callsites to the caller. We'll move those to the caller's respective index space (counter and callers), so we need to know and maintain where those currently end. We also don't need to keep profiles not pertinent to this module. This patch also introduces an arguably much simpler way to track the GUID of a function from the frontend compilation, through ThinLTO, and into the post-thinlink compilation step, which doesn't rely on keeping names around. A separate RFC and patches will discuss extending this to the current PGO (instrumented and sampled) and other consumers as an infrastructural component.
mtrofin
added a commit
that referenced
this pull request
Aug 15, 2024
Continuing from #102084, which introduced the analysis, we now populate it with info about functions contained in the module. When we will update the profile due to e.g. inlined callsites, we'll ingest the callee's counters and callsites to the caller. We'll move those to the caller's respective index space (counter and callers), so we need to know and maintain where those currently end. We also don't need to keep profiles not pertinent to this module. This patch also introduces an arguably much simpler way to track the GUID of a function from the frontend compilation, through ThinLTO, and into the post-thinlink compilation step, which doesn't rely on keeping names around. A separate RFC and patches will discuss extending this to the current PGO (instrumented and sampled) and other consumers as an infrastructural component.
bwendling
pushed a commit
to bwendling/llvm-project
that referenced
this pull request
Aug 15, 2024
Continuing from llvm#102084, which introduced the analysis, we now populate it with info about functions contained in the module. When we will update the profile due to e.g. inlined callsites, we'll ingest the callee's counters and callsites to the caller. We'll move those to the caller's respective index space (counter and callers), so we need to know and maintain where those currently end. We also don't need to keep profiles not pertinent to this module. This patch also introduces an arguably much simpler way to track the GUID of a function from the frontend compilation, through ThinLTO, and into the post-thinlink compilation step, which doesn't rely on keeping names around. A separate RFC and patches will discuss extending this to the current PGO (instrumented and sampled) and other consumers as an infrastructural component.
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/124/builds/125 Here is the relevant piece of the build log for the reference: |
kstoimenov
pushed a commit
to kstoimenov/llvm-project
that referenced
this pull request
Aug 15, 2024
This is an immutable analysis that loads and makes the contextual profile available to other passes. This patch introduces the analysis and an analysis printer pass. Subsequent patches will introduce the APIs that IPO passes will call to modify the profile as result of their changes.
kstoimenov
pushed a commit
to kstoimenov/llvm-project
that referenced
this pull request
Aug 15, 2024
searlmc1
pushed a commit
to ROCm/llvm-project
that referenced
this pull request
Aug 22, 2024
Revert "[ctx_prof] CtxProfAnalysis (llvm#102084)" Change-Id: Ibd6d6f1811e713df4b70d9f68eb027498568ea21
searlmc1
pushed a commit
to ROCm/llvm-project
that referenced
this pull request
Sep 6, 2024
This is an immutable analysis that loads and makes the contextual profile available to other passes. This patch introduces the analysis and an analysis printer pass. Subsequent patches will introduce the APIs that IPO passes will call to modify the profile as result of their changes. Change-Id: I1a1047a73074f143a2df79afdbc06b075473feb8
searlmc1
pushed a commit
to ROCm/llvm-project
that referenced
this pull request
Sep 6, 2024
Continuing from llvm#102084, which introduced the analysis, we now populate it with info about functions contained in the module. When we will update the profile due to e.g. inlined callsites, we'll ingest the callee's counters and callsites to the caller. We'll move those to the caller's respective index space (counter and callers), so we need to know and maintain where those currently end. We also don't need to keep profiles not pertinent to this module. This patch also introduces an arguably much simpler way to track the GUID of a function from the frontend compilation, through ThinLTO, and into the post-thinlink compilation step, which doesn't rely on keeping names around. A separate RFC and patches will discuss extending this to the current PGO (instrumented and sampled) and other consumers as an infrastructural component. Change-Id: I587eb6837788741abfee47f9d8370addfda46265
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This is an immutable analysis that loads and makes the contextual profile available to other passes. This patch just introduces the analysis and an analysis printer pass. Subsequent patches will introduce the APIs that IPO passes will call to modify the profile as result of their changes.
Issue #89287