[PGO][SampledInstr] Correct off by 1s and allow 100% sampling #113350
Conversation
This file contains hidden or 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
|
Thank you for submitting a Pull Request (PR) to the LLVM Project! This PR will be automatically labeled and the relevant teams will be notified. If you wish to, you can add reviewers by using the "Reviewers" section on this page. If this is not working for you, it is probably because you do not have write permissions for the repository. In which case you can instead tag reviewers by name in a comment by using If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers. If you have further questions, they may be answered by the LLVM GitHub User Guide. You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums. |
|
@llvm/pr-subscribers-pgo @llvm/pr-subscribers-llvm-transforms Author: Michael O'Farrell (mofarrell) ChangesThis corrects a couple off by ones related to the sampling of instrumented counters, and enables setting 100% rates for burst sampling (burst duration = period). Off by ones: 100% sampling: Full diff: https://github.com/llvm/llvm-project/pull/113350.diff 7 Files Affected:
diff --git a/llvm/lib/Transforms/Instrumentation/InstrProfiling.cpp b/llvm/lib/Transforms/Instrumentation/InstrProfiling.cpp
index d7d809dfdd5f65..08e6543208b68a 100644
--- a/llvm/lib/Transforms/Instrumentation/InstrProfiling.cpp
+++ b/llvm/lib/Transforms/Instrumentation/InstrProfiling.cpp
@@ -184,21 +184,48 @@ static cl::opt<unsigned> SampledInstrPeriod(
cl::desc("Set the profile instrumentation sample period. For each sample "
"period, a fixed number of consecutive samples will be recorded. "
"The number is controlled by 'sampled-instr-burst-duration' flag. "
- "The default sample period of 65535 is optimized for generating "
+ "The default sample period of 65536 is optimized for generating "
"efficient code that leverages unsigned integer wrapping in "
"overflow."),
- cl::init(65535));
+ cl::init(USHRT_MAX + 1));
static cl::opt<unsigned> SampledInstrBurstDuration(
"sampled-instr-burst-duration",
cl::desc("Set the profile instrumentation burst duration, which can range "
- "from 0 to one less than the value of 'sampled-instr-period'. "
- "This number of samples will be recorded for each "
- "'sampled-instr-period' count update. Setting to 1 enables "
- "simple sampling, in which case it is recommended to set "
- "'sampled-instr-period' to a prime number."),
+ "from 1 to the value of 'sampled-instr-period'. This number of "
+ "samples will be recorded for each 'sampled-instr-period' count "
+ "update. Setting to 1 enables simple sampling, in which case it "
+ "is recommended to set 'sampled-instr-period' to a prime "
+ "number."),
cl::init(200));
+struct SampledInstrumentationConfig {
+ unsigned BurstDuration;
+ unsigned Period;
+ bool UseShort;
+ bool IsSimpleSampling;
+ bool IsFastSampling;
+};
+
+static SampledInstrumentationConfig getSampledInstrumentationConfig() {
+ SampledInstrumentationConfig config;
+ config.BurstDuration = SampledInstrBurstDuration.getValue();
+ config.Period = SampledInstrPeriod.getValue();
+ if (config.BurstDuration > config.Period)
+ report_fatal_error(
+ "SampledBurstDuration must be less than or equal to SampledPeriod");
+ if (config.Period == 0 || config.BurstDuration == 0)
+ report_fatal_error(
+ "SampledPeriod and SampledBurstDuration must be greater than 0");
+ config.IsSimpleSampling = (config.BurstDuration == 1);
+ // If (BurstDuration == 1 && Period == 65536), generate the simple sampling
+ // style code.
+ config.IsFastSampling =
+ (!config.IsSimpleSampling && config.Period == USHRT_MAX + 1);
+ config.UseShort = (config.Period <= USHRT_MAX) || config.IsFastSampling;
+ return config;
+}
+
using LoadStorePair = std::pair<Instruction *, Instruction *>;
static uint64_t getIntModuleFlagOrZero(const Module &M, StringRef Flag) {
@@ -665,7 +692,7 @@ PreservedAnalyses InstrProfilingLoweringPass::run(Module &M,
// (1) Full burst sampling: We transform:
// Increment_Instruction;
// to:
-// if (__llvm_profile_sampling__ < SampledInstrBurstDuration) {
+// if (__llvm_profile_sampling__ <= SampledInstrBurstDuration - 1) {
// Increment_Instruction;
// }
// __llvm_profile_sampling__ += 1;
@@ -680,14 +707,14 @@ PreservedAnalyses InstrProfilingLoweringPass::run(Module &M,
// "__llvm_profile_sampling__" variable is an unsigned type, meaning it will
// wrap around to zero when overflows. In this case, the second check is
// unnecessary, so we won't generate check2 when the SampledInstrPeriod is
-// set to 65535 (64K - 1). The code after:
-// if (__llvm_profile_sampling__ < SampledInstrBurstDuration) {
+// set to 65536 (64K). The code after:
+// if (__llvm_profile_sampling__ <= SampledInstrBurstDuration - 1) {
// Increment_Instruction;
// }
// __llvm_profile_sampling__ += 1;
//
// (3) Simple sampling:
-// When SampledInstrBurstDuration sets to 1, we do a simple sampling:
+// When SampledInstrBurstDuration is set to 1, we do a simple sampling:
// __llvm_profile_sampling__ += 1;
// if (__llvm_profile_sampling__ >= SampledInstrPeriod) {
// __llvm_profile_sampling__ = 0;
@@ -706,27 +733,16 @@ void InstrLowerer::doSampling(Instruction *I) {
if (!isSamplingEnabled())
return;
- unsigned SampledBurstDuration = SampledInstrBurstDuration.getValue();
- unsigned SampledPeriod = SampledInstrPeriod.getValue();
- if (SampledBurstDuration >= SampledPeriod) {
- report_fatal_error(
- "SampledPeriod needs to be greater than SampledBurstDuration");
- }
- bool UseShort = (SampledPeriod <= USHRT_MAX);
- bool IsSimpleSampling = (SampledBurstDuration == 1);
- // If (SampledBurstDuration == 1 && SampledPeriod == 65535), generate
- // the simple sampling style code.
- bool IsFastSampling = (!IsSimpleSampling && SampledPeriod == 65535);
-
- auto GetConstant = [UseShort](IRBuilder<> &Builder, uint32_t C) {
- if (UseShort)
+ SampledInstrumentationConfig config = getSampledInstrumentationConfig();
+ auto GetConstant = [&config](IRBuilder<> &Builder, uint32_t C) {
+ if (config.UseShort)
return Builder.getInt16(C);
else
return Builder.getInt32(C);
};
IntegerType *SamplingVarTy;
- if (UseShort)
+ if (config.UseShort)
SamplingVarTy = Type::getInt16Ty(M.getContext());
else
SamplingVarTy = Type::getInt32Ty(M.getContext());
@@ -741,18 +757,18 @@ void InstrLowerer::doSampling(Instruction *I) {
MDNode *BranchWeight;
IRBuilder<> CondBuilder(I);
auto *LoadSamplingVar = CondBuilder.CreateLoad(SamplingVarTy, SamplingVar);
- if (IsSimpleSampling) {
+ if (config.IsSimpleSampling) {
// For the simple sampling, just create the load and increments.
IRBuilder<> IncBuilder(I);
NewSamplingVarVal =
IncBuilder.CreateAdd(LoadSamplingVar, GetConstant(IncBuilder, 1));
SamplingVarIncr = IncBuilder.CreateStore(NewSamplingVarVal, SamplingVar);
} else {
- // For the bust-sampling, create the conditonal update.
+ // For the burst-sampling, create the conditional update.
auto *DurationCond = CondBuilder.CreateICmpULE(
- LoadSamplingVar, GetConstant(CondBuilder, SampledBurstDuration));
+ LoadSamplingVar, GetConstant(CondBuilder, config.BurstDuration - 1));
BranchWeight = MDB.createBranchWeights(
- SampledBurstDuration, SampledPeriod + 1 - SampledBurstDuration);
+ config.BurstDuration, config.Period - config.BurstDuration);
Instruction *ThenTerm = SplitBlockAndInsertIfThen(
DurationCond, I, /* Unreachable */ false, BranchWeight);
IRBuilder<> IncBuilder(I);
@@ -762,20 +778,20 @@ void InstrLowerer::doSampling(Instruction *I) {
I->moveBefore(ThenTerm);
}
- if (IsFastSampling)
+ if (config.IsFastSampling)
return;
- // Create the condtion for checking the period.
+ // Create the condition for checking the period.
Instruction *ThenTerm, *ElseTerm;
IRBuilder<> PeriodCondBuilder(SamplingVarIncr);
auto *PeriodCond = PeriodCondBuilder.CreateICmpUGE(
- NewSamplingVarVal, GetConstant(PeriodCondBuilder, SampledPeriod));
- BranchWeight = MDB.createBranchWeights(1, SampledPeriod);
+ NewSamplingVarVal, GetConstant(PeriodCondBuilder, config.Period));
+ BranchWeight = MDB.createBranchWeights(1, config.Period - 1);
SplitBlockAndInsertIfThenElse(PeriodCond, SamplingVarIncr, &ThenTerm,
&ElseTerm, BranchWeight);
// For the simple sampling, the counter update happens in sampling var reset.
- if (IsSimpleSampling)
+ if (config.IsSimpleSampling)
I->moveBefore(ThenTerm);
IRBuilder<> ResetBuilder(ThenTerm);
@@ -2138,7 +2154,7 @@ void createProfileSamplingVar(Module &M) {
const StringRef VarName(INSTR_PROF_QUOTE(INSTR_PROF_PROFILE_SAMPLING_VAR));
IntegerType *SamplingVarTy;
Constant *ValueZero;
- if (SampledInstrPeriod.getValue() <= USHRT_MAX) {
+ if (getSampledInstrumentationConfig().UseShort) {
SamplingVarTy = Type::getInt16Ty(M.getContext());
ValueZero = Constant::getIntegerValue(SamplingVarTy, APInt(16, 0));
} else {
diff --git a/llvm/test/Transforms/PGOProfile/counter_promo_sampling.ll b/llvm/test/Transforms/PGOProfile/counter_promo_sampling.ll
index 9d083fe04015e6..43377f695be675 100644
--- a/llvm/test/Transforms/PGOProfile/counter_promo_sampling.ll
+++ b/llvm/test/Transforms/PGOProfile/counter_promo_sampling.ll
@@ -6,7 +6,7 @@
define void @foo(i32 %n, i32 %N) {
; SAMPLING-LABEL: @foo
; SAMPLING: %[[VV0:[0-9]+]] = load i16, ptr @__llvm_profile_sampling, align 2
-; SAMPLING: %[[VV1:[0-9]+]] = icmp ule i16 %[[VV0]], 200
+; SAMPLING: %[[VV1:[0-9]+]] = icmp ule i16 %[[VV0]], 199
; SAMPLING: br i1 %[[VV1]], label {{.*}}, label {{.*}}, !prof !0
; SAMPLING: {{.*}} = load {{.*}} @__profc_foo{{.*}} 3)
; SAMPLING-NEXT: add
diff --git a/llvm/test/Transforms/PGOProfile/cspgo_sample.ll b/llvm/test/Transforms/PGOProfile/cspgo_sample.ll
index 97ad4d00c9d9c0..07f1e2d8a09ee6 100644
--- a/llvm/test/Transforms/PGOProfile/cspgo_sample.ll
+++ b/llvm/test/Transforms/PGOProfile/cspgo_sample.ll
@@ -53,7 +53,7 @@ for.end:
; CSGEN-LABEL: @foo
; CSGEN: [[TMP0:%.*]] = load i16, ptr @__llvm_profile_sampling, align 2
-; CSGEN-NEXT: [[TMP1:%.*]] = icmp ult i16 [[TMP0]], 201
+; CSGEN-NEXT: [[TMP1:%.*]] = icmp ult i16 [[TMP0]], 200
; CSGEN-NEXT: br i1 [[TMP1]], label %{{.*}}, label %{{.*}}, !prof [[PROF:![0-9]+]]
; CSGEN: [[TMP2:%.*]] = add i16 {{.*}}, 1
; CSGEN-NEXT: store i16 [[TMP2]], ptr @__llvm_profile_sampling, align 2
@@ -67,7 +67,7 @@ entry:
}
; CSGEN-LABEL: @main
; CSGEN: [[TMP0:%.*]] = load i16, ptr @__llvm_profile_sampling, align 2
-; CSGEN-NEXT: [[TMP1:%.*]] = icmp ult i16 [[TMP0]], 201
+; CSGEN-NEXT: [[TMP1:%.*]] = icmp ult i16 [[TMP0]], 200
; CSGEN-NEXT: br i1 [[TMP1]], label %{{.*}}, label %{{.*}}, !prof [[PROF:![0-9]+]]
; CSGEN: [[TMP2:%.*]] = add i16 {{.*}}, 1
; CSGEN-NEXT: store i16 [[TMP2]], ptr @__llvm_profile_sampling, align 2
diff --git a/llvm/test/Transforms/PGOProfile/instrprof_burst_sampling_fast.ll b/llvm/test/Transforms/PGOProfile/instrprof_burst_sampling_fast.ll
index dcc1e805ba6f64..56d8364d8f5431 100644
--- a/llvm/test/Transforms/PGOProfile/instrprof_burst_sampling_fast.ll
+++ b/llvm/test/Transforms/PGOProfile/instrprof_burst_sampling_fast.ll
@@ -1,5 +1,6 @@
; RUN: opt < %s --passes=instrprof --sampled-instrumentation -S | FileCheck %s --check-prefixes=SAMPLE-VAR,SAMPLE-CODE,SAMPLE-DURATION,SAMPLE-WEIGHT
; RUN: opt < %s --passes=instrprof --sampled-instrumentation --sampled-instr-burst-duration=100 -S | FileCheck %s --check-prefixes=SAMPLE-VAR,SAMPLE-CODE,SAMPLE-DURATION100,SAMPLE-WEIGHT100
+; RUN: opt < %s --passes=instrprof --sampled-instrumentation --sampled-instr-burst-duration=65536 -S | FileCheck %s --check-prefixes=SAMPLE-VAR,SAMPLE-CODE,UNSAMPLED-DURATION,UNSAMPLED-WEIGHT
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"
@@ -23,8 +24,9 @@ define void @f() {
; SAMPLE-CODE-LABEL: @f(
; SAMPLE-CODE: entry:
; SAMPLE-CODE-NEXT: [[TMP0:%.*]] = load i16, ptr @__llvm_profile_sampling, align 2
-; SAMPLE-DURATION: [[TMP1:%.*]] = icmp ule i16 [[TMP0]], 200
-; SAMPLE-DURATION100: [[TMP1:%.*]] = icmp ule i16 [[TMP0]], 100
+; SAMPLE-DURATION: [[TMP1:%.*]] = icmp ule i16 [[TMP0]], 199
+; SAMPLE-DURATION100: [[TMP1:%.*]] = icmp ule i16 [[TMP0]], 99
+; UNSAMPLED-DURATION: [[TMP1:%.*]] = icmp ule i16 [[TMP0]], -1
; SAMPLE-CODE: br i1 [[TMP1]], label %[[TMP2:.*]], label %[[TMP4:.*]], !prof !0
; SAMPLE-CODE: [[TMP2]]:
; SAMPLE-CODE-NEXT: [[PGOCOUNT:%.*]] = load i64, ptr @__profc_f
@@ -43,5 +45,6 @@ entry:
; SAMPLE-WEIGHT: !0 = !{!"branch_weights", i32 200, i32 65336}
; SAMPLE-WEIGHT100: !0 = !{!"branch_weights", i32 100, i32 65436}
+; UNSAMPLED-WEIGHT: !0 = !{!"branch_weights", i32 65536, i32 0}
declare void @llvm.instrprof.increment(i8*, i64, i32, i32)
diff --git a/llvm/test/Transforms/PGOProfile/instrprof_burst_sampling_full.ll b/llvm/test/Transforms/PGOProfile/instrprof_burst_sampling_full.ll
index 57d1a0cd33fbe8..726df2886ca840 100644
--- a/llvm/test/Transforms/PGOProfile/instrprof_burst_sampling_full.ll
+++ b/llvm/test/Transforms/PGOProfile/instrprof_burst_sampling_full.ll
@@ -1,5 +1,6 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
-; RUN: opt < %s --passes=instrprof -sampled-instrumentation --sampled-instr-period=1009 --sampled-instr-burst-duration=32 -S | FileCheck %s
+; RUN: opt < %s --passes=instrprof -sampled-instrumentation --sampled-instr-period=1009 --sampled-instr-burst-duration=32 -S | FileCheck %s --check-prefixes=CHECK,CHECK-32
+; RUN: opt < %s --passes=instrprof -sampled-instrumentation --sampled-instr-period=1009 --sampled-instr-burst-duration=1009 -S | FileCheck %s --check-prefixes=CHECK,CHECK-UNSAMPLED
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"
@@ -13,7 +14,8 @@ define void @f() {
; CHECK-LABEL: define void @f() {
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[TMP0:%.*]] = load i16, ptr @__llvm_profile_sampling, align 2
-; CHECK-NEXT: [[TMP1:%.*]] = icmp ule i16 [[TMP0]], 32
+; CHECK-32-NEXT: [[TMP1:%.*]] = icmp ule i16 [[TMP0]], 31
+; CHECK-UNSAMPLED-NEXT: [[TMP1:%.*]] = icmp ule i16 [[TMP0]], 1008
; CHECK-NEXT: br i1 [[TMP1]], label %[[BB2:.*]], label %[[BB4:.*]], !prof [[PROF0:![0-9]+]]
; CHECK: [[BB2]]:
; CHECK-NEXT: [[PGOCOUNT:%.*]] = load i64, ptr @__profc_f, align 8
@@ -40,6 +42,8 @@ entry:
declare void @llvm.instrprof.increment(i8*, i64, i32, i32)
;.
-; CHECK: [[PROF0]] = !{!"branch_weights", i32 32, i32 978}
-; CHECK: [[PROF1]] = !{!"branch_weights", i32 1, i32 1009}
+; CHECK-32: [[PROF0]] = !{!"branch_weights", i32 32, i32 977}
+; CHECK-32: [[PROF1]] = !{!"branch_weights", i32 1, i32 1008}
+; CHECK-UNSAMPLED: [[PROF0]] = !{!"branch_weights", i32 1009, i32 0}
+; CHECK-UNSAMPLED: [[PROF1]] = !{!"branch_weights", i32 1, i32 1008}
;.
diff --git a/llvm/test/Transforms/PGOProfile/instrprof_burst_sampling_full_intsize.ll b/llvm/test/Transforms/PGOProfile/instrprof_burst_sampling_full_intsize.ll
index 1ad889524bc6a8..2d6323c1034715 100644
--- a/llvm/test/Transforms/PGOProfile/instrprof_burst_sampling_full_intsize.ll
+++ b/llvm/test/Transforms/PGOProfile/instrprof_burst_sampling_full_intsize.ll
@@ -13,7 +13,7 @@ define void @f() {
; CHECK-LABEL: define void @f() {
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[TMP0:%.*]] = load i32, ptr @__llvm_profile_sampling, align 4
-; CHECK-NEXT: [[TMP1:%.*]] = icmp ule i32 [[TMP0]], 3000
+; CHECK-NEXT: [[TMP1:%.*]] = icmp ule i32 [[TMP0]], 2999
; CHECK-NEXT: br i1 [[TMP1]], label %[[BB2:.*]], label %[[BB4:.*]], !prof [[PROF0:![0-9]+]]
; CHECK: [[BB2]]:
; CHECK-NEXT: [[PGOCOUNT:%.*]] = load i64, ptr @__profc_f, align 8
@@ -40,6 +40,6 @@ entry:
declare void @llvm.instrprof.increment(i8*, i64, i32, i32)
;.
-; CHECK: [[PROF0]] = !{!"branch_weights", i32 3000, i32 997020}
-; CHECK: [[PROF1]] = !{!"branch_weights", i32 1, i32 1000019}
+; CHECK: [[PROF0]] = !{!"branch_weights", i32 3000, i32 997019}
+; CHECK: [[PROF1]] = !{!"branch_weights", i32 1, i32 1000018}
;.
diff --git a/llvm/test/Transforms/PGOProfile/instrprof_simple_sampling.ll b/llvm/test/Transforms/PGOProfile/instrprof_simple_sampling.ll
index 8e846bbf1d9821..5ef93af881c4b3 100644
--- a/llvm/test/Transforms/PGOProfile/instrprof_simple_sampling.ll
+++ b/llvm/test/Transforms/PGOProfile/instrprof_simple_sampling.ll
@@ -31,18 +31,18 @@ define void @f() {
;
; DEFAULTPERIOD-LABEL: define void @f() {
; DEFAULTPERIOD-NEXT: [[ENTRY:.*:]]
-; DEFAULTPERIOD-NEXT: [[TMP0:%.*]] = load i16, ptr @__llvm_profile_sampling, align 2
-; DEFAULTPERIOD-NEXT: [[TMP1:%.*]] = add i16 [[TMP0]], 1
-; DEFAULTPERIOD-NEXT: [[TMP2:%.*]] = icmp uge i16 [[TMP1]], -1
+; DEFAULTPERIOD-NEXT: [[TMP0:%.*]] = load i32, ptr @__llvm_profile_sampling, align 4
+; DEFAULTPERIOD-NEXT: [[TMP1:%.*]] = add i32 [[TMP0]], 1
+; DEFAULTPERIOD-NEXT: [[TMP2:%.*]] = icmp uge i32 [[TMP1]], 65536
; DEFAULTPERIOD-NEXT: br i1 [[TMP2]], label %[[BB3:.*]], label %[[BB5:.*]], !prof [[PROF0:![0-9]+]]
; DEFAULTPERIOD: [[BB3]]:
; DEFAULTPERIOD-NEXT: [[PGOCOUNT:%.*]] = load i64, ptr @__profc_f, align 8
; DEFAULTPERIOD-NEXT: [[TMP4:%.*]] = add i64 [[PGOCOUNT]], 1
; DEFAULTPERIOD-NEXT: store i64 [[TMP4]], ptr @__profc_f, align 8
-; DEFAULTPERIOD-NEXT: store i16 0, ptr @__llvm_profile_sampling, align 2
+; DEFAULTPERIOD-NEXT: store i32 0, ptr @__llvm_profile_sampling, align 4
; DEFAULTPERIOD-NEXT: br label %[[BB6:.*]]
; DEFAULTPERIOD: [[BB5]]:
-; DEFAULTPERIOD-NEXT: store i16 [[TMP1]], ptr @__llvm_profile_sampling, align 2
+; DEFAULTPERIOD-NEXT: store i32 [[TMP1]], ptr @__llvm_profile_sampling, align 4
; DEFAULTPERIOD-NEXT: br label %[[BB6]]
; DEFAULTPERIOD: [[BB6]]:
; DEFAULTPERIOD-NEXT: ret void
@@ -54,7 +54,7 @@ entry:
declare void @llvm.instrprof.increment(i8*, i64, i32, i32)
;.
-; PERIOD1009: [[PROF0]] = !{!"branch_weights", i32 1, i32 1009}
+; PERIOD1009: [[PROF0]] = !{!"branch_weights", i32 1, i32 1008}
;.
; DEFAULTPERIOD: [[PROF0]] = !{!"branch_weights", i32 1, i32 65535}
;.
|
xur-llvm
approved these changes
Oct 22, 2024
This corrects a couple off by ones related to the sampling of instrumented counters, and enables setting 100% rates for burst sampling (burst duration = period). Off by ones: Prior to this change it was impossible to set a period of 65535 because this was converted to fast sampling which rollsover at USHRT_MAX + 1 (65536). Similarly the burst durations would collect burst duration + 1 counts as they used an ULE comparison. 100% sampling: Although this is not useful for a productionized use case, it does allow for more deterministic testing with the sampling checks in place. After all the off by ones are fixed, allowing for 100% sampling is a matter of letting burst duration = period.
mofarrell
force-pushed
the
fix-off-by-one-sampling
branch
from
dc3434b to
b3cfb13
Compare
|
@mofarrell Congratulations on having your first Pull Request (PR) merged into the LLVM Project! Your changes will be combined with recent changes from other authors, then tested by our build bots. If there is a problem with a build, you may receive a report in an email or a comment on this PR. Please check whether problems have been caused by your change specifically, as the builds can include changes from many authors. It is not uncommon for your change to be included in a build that fails due to someone else's changes, or infrastructure issues. How to do this, and the rest of the post-merge process, is covered in detail here. If your change does cause a problem, it may be reverted, or you can revert it yourself. This is a normal part of LLVM development. You can fix your changes and open a new PR to merge them again. If you don't get any reports, no action is required from you. Your changes are working as expected, well done! |
Some correction: I meant to say "simple" sampling. |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/55/builds/3002 Here is the relevant piece of the build log for the reference |
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 corrects a couple off by ones related to the sampling of instrumented counters, and enables setting 100% rates for burst sampling (burst duration = period).
Off by ones:
Prior to this change it was impossible to set a period of 65535 because this was converted to fast sampling which rollsover at USHRT_MAX + 1 (65536). Similarly the burst durations would collect burst duration + 1 counts as they used an ULE comparison.
100% sampling:
Although this is not useful for a productionized use case, it does allow for more deterministic testing with the sampling checks in place. After all the off by ones are fixed, allowing for 100% sampling is a matter of letting burst duration = period.