[llvm] r302844 - Restrict call metadata based hotness detection to Sample PGO mode
Teresa Johnson via llvm-commits
llvm-commits at lists.llvm.org
Thu May 11 16:18:06 PDT 2017
Author: tejohnson
Date: Thu May 11 18:18:05 2017
New Revision: 302844
URL: http://llvm.org/viewvc/llvm-project?rev=302844&view=rev
Log:
Restrict call metadata based hotness detection to Sample PGO mode
Summary:
Don't use the metadata on call instructions for determining hotness
unless we are in sample PGO mode, where it is needed because profile
counts are not accurate. In instrumentation mode this is not necessary
and does more harm than good when calls have VP metadata that hasn't
been properly scaled after transformations or dropped after constant
prop based devirtualization (both should be fixed, but we don't need
to do this in the first place for instrumentation PGO).
This required adjusting a number of tests to distinguish between sample
and instrumentation PGO handling, and to add in profile summary metadata
so that getProfileCount can get the summary.
Reviewers: davidxl, danielcdh
Subscribers: aemerson, rengolin, mehdi_amini, Prazek, llvm-commits
Differential Revision: https://reviews.llvm.org/D32877
Added:
llvm/trunk/test/Bitcode/thinlto-function-summary-callgraph-sample-profile-summary.ll
- copied, changed from r302820, llvm/trunk/test/Bitcode/thinlto-function-summary-callgraph-profile-summary.ll
llvm/trunk/test/Transforms/CodeGenPrepare/section-samplepgo.ll
- copied, changed from r302820, llvm/trunk/test/Transforms/CodeGenPrepare/section.ll
Modified:
llvm/trunk/lib/Analysis/ProfileSummaryInfo.cpp
llvm/trunk/test/Bitcode/thinlto-function-summary-callgraph-profile-summary.ll
llvm/trunk/test/Transforms/CodeGenPrepare/section.ll
llvm/trunk/test/Transforms/Inline/prof-update.ll
llvm/trunk/unittests/Analysis/ProfileSummaryInfoTest.cpp
Modified: llvm/trunk/lib/Analysis/ProfileSummaryInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/ProfileSummaryInfo.cpp?rev=302844&r1=302843&r2=302844&view=diff
==============================================================================
--- llvm/trunk/lib/Analysis/ProfileSummaryInfo.cpp (original)
+++ llvm/trunk/lib/Analysis/ProfileSummaryInfo.cpp Thu May 11 18:18:05 2017
@@ -75,11 +75,14 @@ ProfileSummaryInfo::getProfileCount(cons
return None;
assert((isa<CallInst>(Inst) || isa<InvokeInst>(Inst)) &&
"We can only get profile count for call/invoke instruction.");
- // Check if there is a profile metadata on the instruction. If it is present,
- // determine hotness solely based on that.
- uint64_t TotalCount;
- if (Inst->extractProfTotalWeight(TotalCount))
- return TotalCount;
+ if (computeSummary() && Summary->getKind() == ProfileSummary::PSK_Sample) {
+ // In sample PGO mode, check if there is a profile metadata on the
+ // instruction. If it is present, determine hotness solely based on that,
+ // since the sampled entry count may not be accurate.
+ uint64_t TotalCount;
+ if (Inst->extractProfTotalWeight(TotalCount))
+ return TotalCount;
+ }
if (BFI)
return BFI->getBlockProfileCount(Inst->getParent());
return None;
Modified: llvm/trunk/test/Bitcode/thinlto-function-summary-callgraph-profile-summary.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Bitcode/thinlto-function-summary-callgraph-profile-summary.ll?rev=302844&r1=302843&r2=302844&view=diff
==============================================================================
--- llvm/trunk/test/Bitcode/thinlto-function-summary-callgraph-profile-summary.ll (original)
+++ llvm/trunk/test/Bitcode/thinlto-function-summary-callgraph-profile-summary.ll Thu May 11 18:18:05 2017
@@ -29,7 +29,7 @@
; CHECK-NEXT: <VERSION
; CHECK-NEXT: <VALUE_GUID op0=25 op1=123/>
; op4=hot1 op6=cold op8=hot2 op10=hot4 op12=none1 op14=hot3 op16=none2 op18=none3 op20=123
-; CHECK-NEXT: <PERMODULE_PROFILE {{.*}} op4=1 op5=3 op6=5 op7=1 op8=2 op9=3 op10=4 op11=3 op12=6 op13=2 op14=3 op15=3 op16=7 op17=2 op18=8 op19=2 op20=25 op21=3/>
+; CHECK-NEXT: <PERMODULE_PROFILE {{.*}} op4=1 op5=3 op6=5 op7=1 op8=2 op9=3 op10=4 op11=1 op12=6 op13=2 op14=3 op15=3 op16=7 op17=2 op18=8 op19=2 op20=25 op21=3/>
; CHECK-NEXT: </GLOBALVAL_SUMMARY_BLOCK>
; CHECK: <STRTAB_BLOCK
Copied: llvm/trunk/test/Bitcode/thinlto-function-summary-callgraph-sample-profile-summary.ll (from r302820, llvm/trunk/test/Bitcode/thinlto-function-summary-callgraph-profile-summary.ll)
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Bitcode/thinlto-function-summary-callgraph-sample-profile-summary.ll?p2=llvm/trunk/test/Bitcode/thinlto-function-summary-callgraph-sample-profile-summary.ll&p1=llvm/trunk/test/Bitcode/thinlto-function-summary-callgraph-profile-summary.ll&r1=302820&r2=302844&rev=302844&view=diff
==============================================================================
--- llvm/trunk/test/Bitcode/thinlto-function-summary-callgraph-profile-summary.ll (original)
+++ llvm/trunk/test/Bitcode/thinlto-function-summary-callgraph-sample-profile-summary.ll Thu May 11 18:18:05 2017
@@ -106,7 +106,7 @@ declare void @none3() #1
!1 = !{i32 1, !"ProfileSummary", !2}
!2 = !{!3, !4, !5, !6, !7, !8, !9, !10}
-!3 = !{!"ProfileFormat", !"InstrProf"}
+!3 = !{!"ProfileFormat", !"SampleProfile"}
!4 = !{!"TotalCount", i64 10000}
!5 = !{!"MaxCount", i64 10}
!6 = !{!"MaxInternalCount", i64 1}
Copied: llvm/trunk/test/Transforms/CodeGenPrepare/section-samplepgo.ll (from r302820, llvm/trunk/test/Transforms/CodeGenPrepare/section.ll)
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/CodeGenPrepare/section-samplepgo.ll?p2=llvm/trunk/test/Transforms/CodeGenPrepare/section-samplepgo.ll&p1=llvm/trunk/test/Transforms/CodeGenPrepare/section.ll&r1=302820&r2=302844&rev=302844&view=diff
==============================================================================
--- llvm/trunk/test/Transforms/CodeGenPrepare/section.ll (original)
+++ llvm/trunk/test/Transforms/CodeGenPrepare/section-samplepgo.ll Thu May 11 18:18:05 2017
@@ -39,7 +39,7 @@ define void @cold_func() !prof !16 {
!llvm.module.flags = !{!1}
!1 = !{i32 1, !"ProfileSummary", !2}
!2 = !{!3, !4, !5, !6, !7, !8, !9, !10}
-!3 = !{!"ProfileFormat", !"InstrProf"}
+!3 = !{!"ProfileFormat", !"SampleProfile"}
!4 = !{!"TotalCount", i64 10000}
!5 = !{!"MaxCount", i64 1000}
!6 = !{!"MaxInternalCount", i64 1}
Modified: llvm/trunk/test/Transforms/CodeGenPrepare/section.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/CodeGenPrepare/section.ll?rev=302844&r1=302843&r2=302844&view=diff
==============================================================================
--- llvm/trunk/test/Transforms/CodeGenPrepare/section.ll (original)
+++ llvm/trunk/test/Transforms/CodeGenPrepare/section.ll Thu May 11 18:18:05 2017
@@ -10,26 +10,26 @@ define void @hot_func() !prof !15 {
ret void
}
-; CHECK: hot_call_func{{.*}}!section_prefix ![[HOT_ID]]
-; The sum of 2 callsites are hot
-define void @hot_call_func() !prof !16 {
+; For instrumentation based PGO, we should only look at entry counts,
+; not call site VP metadata (which can exist on value profiled memcpy,
+; or possibly left behind after static analysis based devirtualization).
+; CHECK: cold_func1{{.*}}!section_prefix ![[COLD_ID:[0-9]+]]
+define void @cold_func1() !prof !16 {
call void @hot_func(), !prof !17
call void @hot_func(), !prof !17
ret void
}
-; CHECK-NOT: normal_func{{.*}}!section_prefix
-; The sum of all callsites are neither hot or cold
-define void @normal_func() !prof !16 {
+; CHECK: cold_func2{{.*}}!section_prefix
+define void @cold_func2() !prof !16 {
call void @hot_func(), !prof !17
call void @hot_func(), !prof !18
call void @hot_func(), !prof !18
ret void
}
-; CHECK: cold_func{{.*}}!section_prefix ![[COLD_ID:[0-9]+]]
-; The entry and the callsite are both cold
-define void @cold_func() !prof !16 {
+; CHECK: cold_func3{{.*}}!section_prefix ![[COLD_ID]]
+define void @cold_func3() !prof !16 {
call void @hot_func(), !prof !18
ret void
}
Modified: llvm/trunk/test/Transforms/Inline/prof-update.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/Inline/prof-update.ll?rev=302844&r1=302843&r2=302844&view=diff
==============================================================================
--- llvm/trunk/test/Transforms/Inline/prof-update.ll (original)
+++ llvm/trunk/test/Transforms/Inline/prof-update.ll Thu May 11 18:18:05 2017
@@ -6,21 +6,21 @@ declare void @ext1();
@func = global void ()* null
; CHECK: define void @callee(i32 %n) !prof ![[ENTRY_COUNT:[0-9]*]]
-define void @callee(i32 %n) !prof !1 {
+define void @callee(i32 %n) !prof !15 {
%cond = icmp sle i32 %n, 10
br i1 %cond, label %cond_true, label %cond_false
cond_true:
; ext1 is optimized away, thus not updated.
; CHECK: call void @ext1(), !prof ![[COUNT_CALLEE1:[0-9]*]]
- call void @ext1(), !prof !2
+ call void @ext1(), !prof !16
ret void
cond_false:
; ext is cloned and updated.
; CHECK: call void @ext(), !prof ![[COUNT_CALLEE:[0-9]*]]
- call void @ext(), !prof !2
+ call void @ext(), !prof !16
%f = load void ()*, void ()** @func
; CHECK: call void %f(), !prof ![[COUNT_IND_CALLEE:[0-9]*]]
- call void %f(), !prof !4
+ call void %f(), !prof !18
ret void
}
@@ -28,16 +28,29 @@ cond_false:
define void @caller() {
; CHECK: call void @ext(), !prof ![[COUNT_CALLER:[0-9]*]]
; CHECK: call void %f.i(), !prof ![[COUNT_IND_CALLER:[0-9]*]]
- call void @callee(i32 15), !prof !3
+ call void @callee(i32 15), !prof !17
ret void
}
-!llvm.module.flags = !{!0}
-!0 = !{i32 1, !"MaxFunctionCount", i32 2000}
-!1 = !{!"function_entry_count", i64 1000}
-!2 = !{!"branch_weights", i64 2000}
-!3 = !{!"branch_weights", i64 400}
-!4 = !{!"VP", i32 0, i64 140, i64 111, i64 80, i64 222, i64 40, i64 333, i64 20}
+!llvm.module.flags = !{!1}
+!1 = !{i32 1, !"ProfileSummary", !2}
+!2 = !{!3, !4, !5, !6, !7, !8, !9, !10}
+!3 = !{!"ProfileFormat", !"SampleProfile"}
+!4 = !{!"TotalCount", i64 10000}
+!5 = !{!"MaxCount", i64 10}
+!6 = !{!"MaxInternalCount", i64 1}
+!7 = !{!"MaxFunctionCount", i64 2000}
+!8 = !{!"NumCounts", i64 2}
+!9 = !{!"NumFunctions", i64 2}
+!10 = !{!"DetailedSummary", !11}
+!11 = !{!12, !13, !14}
+!12 = !{i32 10000, i64 100, i32 1}
+!13 = !{i32 999000, i64 100, i32 1}
+!14 = !{i32 999999, i64 1, i32 2}
+!15 = !{!"function_entry_count", i64 1000}
+!16 = !{!"branch_weights", i64 2000}
+!17 = !{!"branch_weights", i64 400}
+!18 = !{!"VP", i32 0, i64 140, i64 111, i64 80, i64 222, i64 40, i64 333, i64 20}
attributes #0 = { alwaysinline }
; CHECK: ![[ENTRY_COUNT]] = !{!"function_entry_count", i64 600}
; CHECK: ![[COUNT_CALLEE1]] = !{!"branch_weights", i64 2000}
Modified: llvm/trunk/unittests/Analysis/ProfileSummaryInfoTest.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/unittests/Analysis/ProfileSummaryInfoTest.cpp?rev=302844&r1=302843&r2=302844&view=diff
==============================================================================
--- llvm/trunk/unittests/Analysis/ProfileSummaryInfoTest.cpp (original)
+++ llvm/trunk/unittests/Analysis/ProfileSummaryInfoTest.cpp Thu May 11 18:18:05 2017
@@ -162,6 +162,12 @@ TEST_F(ProfileSummaryInfoTest, InstrProf
EXPECT_TRUE(PSI.isHotCallSite(CS1, &BFI));
EXPECT_FALSE(PSI.isHotCallSite(CS2, &BFI));
+
+ // Test that adding an MD_prof metadata with a hot count on CS2 does not
+ // change its hotness as it has no effect in instrumented profiling.
+ MDBuilder MDB(M->getContext());
+ CI2->setMetadata(llvm::LLVMContext::MD_prof, MDB.createBranchWeights({400}));
+ EXPECT_FALSE(PSI.isHotCallSite(CS2, &BFI));
}
TEST_F(ProfileSummaryInfoTest, SampleProf) {
More information about the llvm-commits
mailing list