[llvm] r329459 - Runtime flag to control branch funnel threshold
Vitaly Buka via llvm-commits
llvm-commits at lists.llvm.org
Fri Apr 6 14:32:36 PDT 2018
Author: vitalybuka
Date: Fri Apr 6 14:32:36 2018
New Revision: 329459
URL: http://llvm.org/viewvc/llvm-project?rev=329459&view=rev
Log:
Runtime flag to control branch funnel threshold
Reviewers: pcc
Subscribers: hiraditya, llvm-commits
Differential Revision: https://reviews.llvm.org/D45193
Added:
llvm/trunk/test/Transforms/WholeProgramDevirt/branch-funnel-threshold.ll
Modified:
llvm/trunk/lib/Transforms/IPO/WholeProgramDevirt.cpp
Modified: llvm/trunk/lib/Transforms/IPO/WholeProgramDevirt.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/WholeProgramDevirt.cpp?rev=329459&r1=329458&r2=329459&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/IPO/WholeProgramDevirt.cpp (original)
+++ llvm/trunk/lib/Transforms/IPO/WholeProgramDevirt.cpp Fri Apr 6 14:32:36 2018
@@ -111,6 +111,11 @@ static cl::opt<std::string> ClWriteSumma
cl::desc("Write summary to given YAML file after running pass"),
cl::Hidden);
+static cl::opt<int> ClThreshold("wholeprogramdevirt-branch-funnel-threshold",
+ cl::Hidden, cl::init(10), cl::ZeroOrMore,
+ cl::desc("Maximum number of call targets per "
+ "call site to enable branch funnels"));
+
// Find the minimum offset that we may store a value of size Size bits at. If
// IsAfter is set, look for an offset before the object, otherwise look for an
// offset after the object.
@@ -820,8 +825,7 @@ void DevirtModule::tryICallBranchFunnel(
if (T.getArch() != Triple::x86_64)
return;
- const unsigned kBranchFunnelThreshold = 10;
- if (TargetsForSlot.size() > kBranchFunnelThreshold)
+ if (TargetsForSlot.size() > ClThreshold)
return;
bool HasNonDevirt = !SlotInfo.CSInfo.AllCallSitesDevirted;
Added: llvm/trunk/test/Transforms/WholeProgramDevirt/branch-funnel-threshold.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/WholeProgramDevirt/branch-funnel-threshold.ll?rev=329459&view=auto
==============================================================================
--- llvm/trunk/test/Transforms/WholeProgramDevirt/branch-funnel-threshold.ll (added)
+++ llvm/trunk/test/Transforms/WholeProgramDevirt/branch-funnel-threshold.ll Fri Apr 6 14:32:36 2018
@@ -0,0 +1,100 @@
+; RUN: opt -wholeprogramdevirt -wholeprogramdevirt-summary-action=export -wholeprogramdevirt-read-summary=%S/Inputs/export.yaml -wholeprogramdevirt-write-summary=%t -wholeprogramdevirt-branch-funnel-threshold=1 -S -o - %s | not grep @llvm.icall.branch.funnel | count 0
+
+; RUN: opt -wholeprogramdevirt -wholeprogramdevirt-summary-action=export -wholeprogramdevirt-read-summary=%S/Inputs/export.yaml -wholeprogramdevirt-write-summary=%t -wholeprogramdevirt-branch-funnel-threshold=10 -S -o - %s | grep @llvm.icall.branch.funnel | count 4
+
+; RUN: opt -wholeprogramdevirt -wholeprogramdevirt-summary-action=export -wholeprogramdevirt-read-summary=%S/Inputs/export.yaml -wholeprogramdevirt-write-summary=%t -wholeprogramdevirt-branch-funnel-threshold=100 -S -o - %s | grep @llvm.icall.branch.funnel | count 5
+
+target datalayout = "e-p:64:64"
+target triple = "x86_64-unknown-linux-gnu"
+
+ at vt1_1 = constant [1 x i8*] [i8* bitcast (i32 (i8*, i32)* @vf1_1 to i8*)], !type !0
+ at vt1_2 = constant [1 x i8*] [i8* bitcast (i32 (i8*, i32)* @vf1_2 to i8*)], !type !0
+
+declare i32 @vf1_1(i8* %this, i32 %arg)
+declare i32 @vf1_2(i8* %this, i32 %arg)
+
+ at vt2_1 = constant [1 x i8*] [i8* bitcast (i32 (i8*, i32)* @vf2_1 to i8*)], !type !1
+ at vt2_2 = constant [1 x i8*] [i8* bitcast (i32 (i8*, i32)* @vf2_2 to i8*)], !type !1
+ at vt2_3 = constant [1 x i8*] [i8* bitcast (i32 (i8*, i32)* @vf2_3 to i8*)], !type !1
+ at vt2_4 = constant [1 x i8*] [i8* bitcast (i32 (i8*, i32)* @vf2_4 to i8*)], !type !1
+ at vt2_5 = constant [1 x i8*] [i8* bitcast (i32 (i8*, i32)* @vf2_5 to i8*)], !type !1
+ at vt2_6 = constant [1 x i8*] [i8* bitcast (i32 (i8*, i32)* @vf2_6 to i8*)], !type !1
+ at vt2_7 = constant [1 x i8*] [i8* bitcast (i32 (i8*, i32)* @vf2_7 to i8*)], !type !1
+ at vt2_8 = constant [1 x i8*] [i8* bitcast (i32 (i8*, i32)* @vf2_8 to i8*)], !type !1
+ at vt2_9 = constant [1 x i8*] [i8* bitcast (i32 (i8*, i32)* @vf2_9 to i8*)], !type !1
+ at vt2_10 = constant [1 x i8*] [i8* bitcast (i32 (i8*, i32)* @vf2_10 to i8*)], !type !1
+ at vt2_11 = constant [1 x i8*] [i8* bitcast (i32 (i8*, i32)* @vf2_11 to i8*)], !type !1
+
+declare i32 @vf2_1(i8* %this, i32 %arg)
+declare i32 @vf2_2(i8* %this, i32 %arg)
+declare i32 @vf2_3(i8* %this, i32 %arg)
+declare i32 @vf2_4(i8* %this, i32 %arg)
+declare i32 @vf2_5(i8* %this, i32 %arg)
+declare i32 @vf2_6(i8* %this, i32 %arg)
+declare i32 @vf2_7(i8* %this, i32 %arg)
+declare i32 @vf2_8(i8* %this, i32 %arg)
+declare i32 @vf2_9(i8* %this, i32 %arg)
+declare i32 @vf2_10(i8* %this, i32 %arg)
+declare i32 @vf2_11(i8* %this, i32 %arg)
+
+ at vt3_1 = constant [1 x i8*] [i8* bitcast (i32 (i8*, i32)* @vf3_1 to i8*)], !type !2
+ at vt3_2 = constant [1 x i8*] [i8* bitcast (i32 (i8*, i32)* @vf3_2 to i8*)], !type !2
+
+declare i32 @vf3_1(i8* %this, i32 %arg)
+declare i32 @vf3_2(i8* %this, i32 %arg)
+
+ at vt4_1 = constant [1 x i8*] [i8* bitcast (i32 (i8*, i32)* @vf4_1 to i8*)], !type !3
+ at vt4_2 = constant [1 x i8*] [i8* bitcast (i32 (i8*, i32)* @vf4_2 to i8*)], !type !3
+
+declare i32 @vf4_1(i8* %this, i32 %arg)
+declare i32 @vf4_2(i8* %this, i32 %arg)
+
+define i32 @fn1(i8* %obj) #0 {
+ %vtableptr = bitcast i8* %obj to [1 x i8*]**
+ %vtable = load [1 x i8*]*, [1 x i8*]** %vtableptr
+ %vtablei8 = bitcast [1 x i8*]* %vtable to i8*
+ %p = call i1 @llvm.type.test(i8* %vtablei8, metadata !"typeid1")
+ call void @llvm.assume(i1 %p)
+ %fptrptr = getelementptr [1 x i8*], [1 x i8*]* %vtable, i32 0, i32 0
+ %fptr = load i8*, i8** %fptrptr
+ %fptr_casted = bitcast i8* %fptr to i32 (i8*, i32)*
+ %result = call i32 %fptr_casted(i8* %obj, i32 1)
+ ret i32 %result
+}
+
+define i32 @fn2(i8* %obj) #0 {
+ %vtableptr = bitcast i8* %obj to [1 x i8*]**
+ %vtable = load [1 x i8*]*, [1 x i8*]** %vtableptr
+ %vtablei8 = bitcast [1 x i8*]* %vtable to i8*
+ %p = call i1 @llvm.type.test(i8* %vtablei8, metadata !"typeid2")
+ call void @llvm.assume(i1 %p)
+ %fptrptr = getelementptr [1 x i8*], [1 x i8*]* %vtable, i32 0, i32 0
+ %fptr = load i8*, i8** %fptrptr
+ %fptr_casted = bitcast i8* %fptr to i32 (i8*, i32)*
+ %result = call i32 %fptr_casted(i8* %obj, i32 1)
+ ret i32 %result
+}
+
+define i32 @fn3(i8* %obj) #0 {
+ %vtableptr = bitcast i8* %obj to [1 x i8*]**
+ %vtable = load [1 x i8*]*, [1 x i8*]** %vtableptr
+ %vtablei8 = bitcast [1 x i8*]* %vtable to i8*
+ %p = call i1 @llvm.type.test(i8* %vtablei8, metadata !4)
+ call void @llvm.assume(i1 %p)
+ %fptrptr = getelementptr [1 x i8*], [1 x i8*]* %vtable, i32 0, i32 0
+ %fptr = load i8*, i8** %fptrptr
+ %fptr_casted = bitcast i8* %fptr to i32 (i8*, i32)*
+ %result = call i32 %fptr_casted(i8* %obj, i32 1)
+ ret i32 %result
+}
+
+declare i1 @llvm.type.test(i8*, metadata)
+declare void @llvm.assume(i1)
+
+!0 = !{i32 0, !"typeid1"}
+!1 = !{i32 0, !"typeid2"}
+!2 = !{i32 0, !"typeid3"}
+!3 = !{i32 0, !4}
+!4 = distinct !{}
+
+attributes #0 = { "target-features"="+retpoline" }
More information about the llvm-commits
mailing list