[llvm] [TypeProf][ICP]Allow vtable-comparison as long as vtable count is comparable with function count for each candidate (PR #106260)

[Mingming Liu via llvm-commits]

https://github.com/minglotus-6 created https://github.com/llvm/llvm-project/pull/106260

The current cost-benefit analysis between vtable comparison and function comparison require the indirect fallback branch to be cold. This is too conservative. 

This change allows vtable-comparison as long as vtable count is comparable with function count for each function candidate and removes the cold indirect fallback requirement.

Testing this on benchmarks uplifts the measurable performance wins. Counting the (non-duplicated) remarks show the number of vtable remarks increases from ~30k-ish to 50k-ish.

>From d074d8ad3ddeff35e1fc9d4123d2d3fe2fe9c128 Mon Sep 17 00:00:00 2001
From: mingmingl <mingmingl at google.com>
Date: Tue, 27 Aug 2024 10:13:02 -0700
Subject: [PATCH] [TypeProf][ICP]Allow vtable-comparison when the indirect
 fallback is not cold, as long as vtable count is comparable with function
 count for each candidate

---
 .../Instrumentation/IndirectCallPromotion.cpp  | 10 +---------
 .../Transforms/PGOProfile/icp_vtable_cmp.ll    | 18 ++++++++++++++++++
 2 files changed, 19 insertions(+), 9 deletions(-)

diff --git a/llvm/lib/Transforms/Instrumentation/IndirectCallPromotion.cpp b/llvm/lib/Transforms/Instrumentation/IndirectCallPromotion.cpp
index 0d1f506986379d..c9007498ee2331 100644
--- a/llvm/lib/Transforms/Instrumentation/IndirectCallPromotion.cpp
+++ b/llvm/lib/Transforms/Instrumentation/IndirectCallPromotion.cpp
@@ -117,7 +117,7 @@ static cl::opt<bool>
 // Indirect call promotion pass will fall back to function-based comparison if
 // vtable-count / function-count is smaller than this threshold.
 static cl::opt<float> ICPVTablePercentageThreshold(
-    "icp-vtable-percentage-threshold", cl::init(0.99), cl::Hidden,
+    "icp-vtable-percentage-threshold", cl::init(0.995), cl::Hidden,
     cl::desc("The percentage threshold of vtable-count / function-count for "
              "cost-benefit analysis."));
 
@@ -888,14 +888,6 @@ bool IndirectCallPromoter::isProfitableToCompareVTables(
     }
   }
 
-  // If the indirect fallback is not cold, don't compare vtables.
-  if (PSI && PSI->hasProfileSummary() &&
-      !PSI->isColdCount(RemainingVTableCount)) {
-    LLVM_DEBUG(dbgs() << "    Indirect fallback basic block is not cold. Bail "
-                         "out for vtable comparison.\n");
-    return false;
-  }
-
   return true;
 }
 
diff --git a/llvm/test/Transforms/PGOProfile/icp_vtable_cmp.ll b/llvm/test/Transforms/PGOProfile/icp_vtable_cmp.ll
index c77be3b1ed2444..b6afce3d7c6d5d 100644
--- a/llvm/test/Transforms/PGOProfile/icp_vtable_cmp.ll
+++ b/llvm/test/Transforms/PGOProfile/icp_vtable_cmp.ll
@@ -120,6 +120,7 @@ declare i32 @Base2_foo(ptr)
 declare i32 @Base1_foo(ptr)
 declare void @Base3_foo(ptr)
 
+!llvm.module.flags = !{!11}
 !0 = !{i64 16, !"Base1"}
 !1 = !{i64 40, !"Base1"}
 !2 = !{i64 16, !"Base2"}
@@ -131,6 +132,23 @@ declare void @Base3_foo(ptr)
 !8 = !{i64 16, !"Derived3"}
 !9 = !{!"VP", i32 2, i64 1600, i64 -4123858694673519054, i64 600, i64 -7211198353767973908, i64 500, i64 -3574436251470806727, i64 200, i64 6288809125658696740, i64 200, i64 12345678, i64 100}
 !10 = !{!"VP", i32 0, i64 1600, i64 3827408714133779784, i64 600, i64 5837445539218476403, i64 500, i64 -9064955852395570538, i64 400,  i64 56781234, i64 100}
+
+!11 = !{i32 1, !"ProfileSummary", !12}
+!12 = !{!13, !14, !15, !16, !17, !18, !19, !20}
+!13 = !{!"ProfileFormat", !"InstrProf"}
+!14 = !{!"TotalCount", i64 10000}
+!15 = !{!"MaxCount", i64 10}
+!16 = !{!"MaxInternalCount", i64 1}
+!17 = !{!"MaxFunctionCount", i64 1000}
+!18 = !{!"NumCounts", i64 3}
+!19 = !{!"NumFunctions", i64 3}
+!20 = !{!"DetailedSummary", !21}
+!21 = !{!22, !23, !24}
+!22 = !{i32 10000, i64 101, i32 1}
+!23 = !{i32 990000, i64 101, i32 1}
+!24 = !{i32 999999, i64 1, i32 2}
+
+
 ;.
 ; VTABLE-COMMON: [[PROF9]] = !{!"VP", i32 2, i64 100, i64 12345678, i64 100}
 ; VTABLE-COMMON: [[PROF10]] = !{!"branch_weights", i32 600, i32 1000}