[PATCH] D12781: PGO IR-level instrumentation infrastructure
Justin Bogner via llvm-commits
llvm-commits at lists.llvm.org
Tue Dec 1 14:38:10 PST 2015
Rong Xu <xur at google.com> writes:
> xur updated this revision to Diff 41428.
> xur marked 2 inline comments as done.
> xur added a comment.
>
> Integrated Sean's most recently review comments.
This is looking pretty reasonable. Some style nitpicks and minor
comments below.
>
> http://reviews.llvm.org/D12781
>
> Files:
> include/llvm/IR/DiagnosticInfo.h
> include/llvm/InitializePasses.h
> include/llvm/LinkAllPasses.h
> include/llvm/Transforms/Instrumentation.h
> lib/IR/DiagnosticInfo.cpp
> lib/Transforms/Instrumentation/CFGMST.h
> lib/Transforms/Instrumentation/CMakeLists.txt
> lib/Transforms/Instrumentation/Instrumentation.cpp
> lib/Transforms/Instrumentation/LLVMBuild.txt
> lib/Transforms/Instrumentation/PGOInstrumentation.cpp
> test/Transforms/PGOProfile/Inputs/branch1.proftext
> test/Transforms/PGOProfile/Inputs/branch2.proftext
> test/Transforms/PGOProfile/Inputs/criticaledge.proftext
> test/Transforms/PGOProfile/Inputs/landingpad.proftext
> test/Transforms/PGOProfile/Inputs/loop1.proftext
> test/Transforms/PGOProfile/Inputs/loop2.proftext
> test/Transforms/PGOProfile/Inputs/single_bb.proftext
> test/Transforms/PGOProfile/Inputs/switch.proftext
> test/Transforms/PGOProfile/branch1.ll
> test/Transforms/PGOProfile/branch2.ll
> test/Transforms/PGOProfile/checksum_mismatch.ll
> test/Transforms/PGOProfile/criticaledge.ll
> test/Transforms/PGOProfile/landingpad.ll
> test/Transforms/PGOProfile/loop1.ll
> test/Transforms/PGOProfile/loop2.ll
> test/Transforms/PGOProfile/noprofile_use.ll
> test/Transforms/PGOProfile/single_bb.ll
> test/Transforms/PGOProfile/switch.ll
>
> Index: test/Transforms/PGOProfile/switch.ll
> ===================================================================
> --- /dev/null
> +++ test/Transforms/PGOProfile/switch.ll
> @@ -0,0 +1,48 @@
> +; RUN: opt < %s -pgo-instr-gen -S | FileCheck %s --check-prefix=GEN
> +; RUN: llvm-profdata merge %S/Inputs/switch.proftext -o %T/switch.profdata
> +; RUN: opt < %s -pgo-instr-use -pgo-test-profile-file=%T/switch.profdata -S | FileCheck %s --check-prefix=USE
> +target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
> +target triple = "x86_64-unknown-linux-gnu"
> +
> +; GEN: @__llvm_profile_name_test_switch = private constant [11 x i8] c"test_switch"
> +
> +define void @test_switch(i32 %i) {
> +entry:
> +; GEN: entry:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + switch i32 %i, label %sw.default [
> + i32 1, label %sw.bb
> + i32 2, label %sw.bb1
> + i32 3, label %sw.bb2
> + ]
> +; USE: ]
> +; USE-SAME: !prof ![[BW_SWITCH:[0-9]+]]
> +
> +sw.bb:
> +; GEN: sw.bb:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([11 x i8], [11 x i8]* @__llvm_profile_name_test_switch, i32 0, i32 0), i64 46200943743, i32 4, i32 2)
> + br label %sw.epilog
> +
> +sw.bb1:
> +; GEN: sw.bb1:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([11 x i8], [11 x i8]* @__llvm_profile_name_test_switch, i32 0, i32 0), i64 46200943743, i32 4, i32 0)
> + br label %sw.epilog
> +
> +sw.bb2:
> +; GEN: sw.bb2:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([11 x i8], [11 x i8]* @__llvm_profile_name_test_switch, i32 0, i32 0), i64 46200943743, i32 4, i32 1)
> + br label %sw.epilog
> +
> +sw.default:
> +; GEN: sw.default:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([11 x i8], [11 x i8]* @__llvm_profile_name_test_switch, i32 0, i32 0), i64 46200943743, i32 4, i32 3)
> + br label %sw.epilog
> +
> +sw.epilog:
> +; GEN: sw.epilog:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + ret void
> +; GEN: ret void
> +}
> +
> +; USE: ![[BW_SWITCH]] = !{!"branch_weights", i32 3, i32 2, i32 0, i32 5}
> Index: test/Transforms/PGOProfile/single_bb.ll
> ===================================================================
> --- /dev/null
> +++ test/Transforms/PGOProfile/single_bb.ll
> @@ -0,0 +1,12 @@
> +; RUN: opt < %s -pgo-instr-gen -S | FileCheck %s --check-prefix=GEN
> +target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
> +target triple = "x86_64-unknown-linux-gnu"
> +
> +; GEN: @__llvm_profile_name_single_bb = private constant [9 x i8] c"single_bb"
> +
> +define i32 @single_bb() {
> +entry:
> +; GEN: entry:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([9 x i8], [9 x i8]* @__llvm_profile_name_single_bb, i32 0, i32 0), i64 12884901887, i32 1, i32 0)
> + ret i32 0
> +}
> Index: test/Transforms/PGOProfile/noprofile_use.ll
> ===================================================================
> --- /dev/null
> +++ test/Transforms/PGOProfile/noprofile_use.ll
> @@ -0,0 +1,9 @@
> +; RUN: not opt < %s -pgo-instr-use -pgo-test-profile-file=%T/notexisting.profdata -S 2>&1
Could you FileCheck for the error message here please? It makes the
purpose of the test a bit clearer.
> +
> +target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
> +target triple = "x86_64-unknown-linux-gnu"
> +
> +define i32 @foo() {
> +entry:
> + ret i32 0
> +}
> Index: test/Transforms/PGOProfile/loop2.ll
> ===================================================================
> --- /dev/null
> +++ test/Transforms/PGOProfile/loop2.ll
> @@ -0,0 +1,97 @@
> +; RUN: opt < %s -pgo-instr-gen -S | FileCheck %s --check-prefix=GEN
> +; RUN: llvm-profdata merge %S/Inputs/loop2.proftext -o %T/loop2.profdata
> +; RUN: opt < %s -pgo-instr-use -pgo-test-profile-file=%T/loop2.profdata -S | FileCheck %s --check-prefix=USE
> +target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
> +target triple = "x86_64-unknown-linux-gnu"
> +
> +; GEN: @__llvm_profile_name_test_nested_for = private constant [15 x i8] c"test_nested_for"
> +
> +define i32 @test_nested_for(i32 %r, i32 %s, i32 %t) {
> +entry:
> +; GEN: entry:
> + br label %for.cond
> +
> +for.cond:
> +; GEN: for.cond:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + %i.0 = phi i32 [ 0, %entry ], [ %inc12, %for.inc11 ]
> + %nested_for_sum.0 = phi i32 [ 1, %entry ], [ %nested_for_sum.1, %for.inc11 ]
> + %cmp = icmp slt i32 %i.0, %r
> + br i1 %cmp, label %for.body, label %for.end13
> +; USE: br i1 %cmp, label %for.body, label %for.end13
> +; USE-SAME: !prof ![[BW_FOR_COND:[0-9]+]]
> +
> +for.body:
> +; GEN: for.body:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + br label %for.cond1
> +
> +for.cond1:
> +; GEN: for.cond1:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + %j.0 = phi i32 [ 0, %for.body ], [ %inc9, %for.inc8 ]
> + %nested_for_sum.1 = phi i32 [ %nested_for_sum.0, %for.body ], [ %nested_for_sum.2, %for.inc8 ]
> + %cmp2 = icmp slt i32 %j.0, %s
> + br i1 %cmp2, label %for.body3, label %for.end10
> +; USE: br i1 %cmp2, label %for.body3, label %for.end10
> +; USE-SAME: !prof ![[BW_FOR_COND1:[0-9]+]]
> +
> +for.body3:
> +; GEN: for.body3:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + br label %for.cond4
> +
> +for.cond4:
> +; GEN: for.cond4:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + %k.0 = phi i32 [ 0, %for.body3 ], [ %inc7, %for.inc ]
> + %nested_for_sum.2 = phi i32 [ %nested_for_sum.1, %for.body3 ], [ %inc, %for.inc ]
> + %cmp5 = icmp slt i32 %k.0, %t
> + br i1 %cmp5, label %for.body6, label %for.end
> +; USE: br i1 %cmp5, label %for.body6, label %for.end
> +; USE-SAME: !prof ![[BW_FOR_COND4:[0-9]+]]
> +
> +for.body6:
> +; GEN: for.body6:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + %inc = add nsw i32 %nested_for_sum.2, 1
> + br label %for.inc
> +
> +for.inc:
> +; GEN: for.inc:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([15 x i8], [15 x i8]* @__llvm_profile_name_test_nested_for, i32 0, i32 0), i64 74483645111, i32 4, i32 0)
> + %inc7 = add nsw i32 %k.0, 1
> + br label %for.cond4
> +
> +for.end:
> +; GEN: for.end:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + br label %for.inc8
> +
> +for.inc8:
> +; GEN: for.inc8:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([15 x i8], [15 x i8]* @__llvm_profile_name_test_nested_for, i32 0, i32 0), i64 74483645111, i32 4, i32 1)
> + %inc9 = add nsw i32 %j.0, 1
> + br label %for.cond1
> +
> +for.end10:
> +; GEN: for.end10:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + br label %for.inc11
> +
> +for.inc11:
> +; GEN: for.inc11:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([15 x i8], [15 x i8]* @__llvm_profile_name_test_nested_for, i32 0, i32 0), i64 74483645111, i32 4, i32 2)
> + %inc12 = add nsw i32 %i.0, 1
> + br label %for.cond
> +
> +for.end13:
> +; GEN: for.end13:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([15 x i8], [15 x i8]* @__llvm_profile_name_test_nested_for, i32 0, i32 0), i64 74483645111, i32 4, i32 3)
> + ret i32 %nested_for_sum.0
> +}
> +
> +; USE: ![[BW_FOR_COND]] = !{!"branch_weights", i32 10, i32 6}
> +; USE: ![[BW_FOR_COND1]] = !{!"branch_weights", i32 33, i32 10}
> +; USE: ![[BW_FOR_COND4]] = !{!"branch_weights", i32 186, i32 33}
> +
> Index: test/Transforms/PGOProfile/loop1.ll
> ===================================================================
> --- /dev/null
> +++ test/Transforms/PGOProfile/loop1.ll
> @@ -0,0 +1,43 @@
> +; RUN: opt < %s -pgo-instr-gen -S | FileCheck %s --check-prefix=GEN
> +; RUN: llvm-profdata merge %S/Inputs/loop1.proftext -o %T/loop1.profdata
> +; RUN: opt < %s -pgo-instr-use -pgo-test-profile-file=%T/loop1.profdata -S | FileCheck %s --check-prefix=USE
> +target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
> +target triple = "x86_64-unknown-linux-gnu"
> +
> +; GEN: @__llvm_profile_name_test_simple_for = private constant [15 x i8] c"test_simple_for"
> +
> +define i32 @test_simple_for(i32 %n) {
> +entry:
> +; GEN: entry:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + br label %for.cond
> +
> +for.cond:
> +; GEN: for.cond:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + %i = phi i32 [ 0, %entry ], [ %inc1, %for.inc ]
> + %sum = phi i32 [ 1, %entry ], [ %inc, %for.inc ]
> + %cmp = icmp slt i32 %i, %n
> + br i1 %cmp, label %for.body, label %for.end
> +; USE: br i1 %cmp, label %for.body, label %for.end
> +; USE-SAME: !prof ![[BW_FOR_COND:[0-9]+]]
> +
> +for.body:
> +; GEN: for.body:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + %inc = add nsw i32 %sum, 1
> + br label %for.inc
> +
> +for.inc:
> +; GEN: for.inc:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([15 x i8], [15 x i8]* @__llvm_profile_name_test_simple_for, i32 0, i32 0), i64 34137660316, i32 2, i32 0)
> + %inc1 = add nsw i32 %i, 1
> + br label %for.cond
> +
> +for.end:
> +; GEN: for.end:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([15 x i8], [15 x i8]* @__llvm_profile_name_test_simple_for, i32 0, i32 0), i64 34137660316, i32 2, i32 1)
> + ret i32 %sum
> +}
> +
> +; USE: ![[BW_FOR_COND]] = !{!"branch_weights", i32 96, i32 4}
> Index: test/Transforms/PGOProfile/landingpad.ll
> ===================================================================
> --- /dev/null
> +++ test/Transforms/PGOProfile/landingpad.ll
> @@ -0,0 +1,124 @@
> +; RUN: opt < %s -pgo-instr-gen -S | FileCheck %s --check-prefix=GEN
> +; RUN: llvm-profdata merge %S/Inputs/landingpad.proftext -o %T/landingpad.profdata
> +; RUN: opt < %s -pgo-instr-use -pgo-test-profile-file=%T/landingpad.profdata -S | FileCheck %s --check-prefix=USE
> +target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
> +target triple = "x86_64-unknown-linux-gnu"
> +
> + at val = global i32 0, align 4
> + at _ZTIi = external constant i8*
> +; GEN: @__llvm_profile_name_bar = private constant [3 x i8] c"bar"
> +; GEN: @__llvm_profile_name_foo = private constant [3 x i8] c"foo"
> +
> +define i32 @bar(i32 %i) {
> +entry:
> +; GEN: entry:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + %rem = srem i32 %i, 3
> + %tobool = icmp ne i32 %rem, 0
> + br i1 %tobool, label %if.then, label %if.end
> +; USE: br i1 %tobool, label %if.then, label %if.end
> +; USE-SAME: !prof ![[BW_BAR_ENTRY:[0-9]+]]
> +
> +if.then:
> +; GEN: if.then:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([3 x i8], [3 x i8]* @__llvm_profile_name_bar, i32 0, i32 0), i64 24868915205, i32 2, i32 1)
> + %exception = call i8* @__cxa_allocate_exception(i64 4)
> + %tmp = bitcast i8* %exception to i32*
> + store i32 %i, i32* %tmp, align 16
> + call void @__cxa_throw(i8* %exception, i8* bitcast (i8** @_ZTIi to i8*), i8* null)
> + unreachable
> +
> +if.end:
> +; GEN: if.end:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([3 x i8], [3 x i8]* @__llvm_profile_name_bar, i32 0, i32 0), i64 24868915205, i32 2, i32 0)
> + ret i32 0
> +}
> +
> +declare i8* @__cxa_allocate_exception(i64)
> +
> +declare void @__cxa_throw(i8*, i8*, i8*)
> +
> +define i32 @foo(i32 %i) personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
> +entry:
> +; GEN: entry:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + %rem = srem i32 %i, 2
> + %tobool = icmp ne i32 %rem, 0
> + br i1 %tobool, label %if.then, label %if.end
> +; USE: br i1 %tobool, label %if.then, label %if.end
> +; USE-SAME: !prof ![[BW_FOO_ENTRY:[0-9]+]]
> +
> +if.then:
> +; GEN: if.then:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + %mul = mul nsw i32 %i, 7
> + %call = invoke i32 @bar(i32 %mul)
> + to label %invoke.cont unwind label %lpad
> +
> +invoke.cont:
> +; GEN: invoke.cont:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([3 x i8], [3 x i8]* @__llvm_profile_name_foo, i32 0, i32 0), i64 59130013419, i32 4, i32 1)
> + br label %if.end
> +
> +lpad:
> +; GEN: lpad:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + %tmp = landingpad { i8*, i32 }
> + catch i8* bitcast (i8** @_ZTIi to i8*)
> + %tmp1 = extractvalue { i8*, i32 } %tmp, 0
> + %tmp2 = extractvalue { i8*, i32 } %tmp, 1
> + br label %catch.dispatch
> +
> +catch.dispatch:
> +; GEN: catch.dispatch:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + %tmp3 = call i32 @llvm.eh.typeid.for(i8* bitcast (i8** @_ZTIi to i8*))
> + %matches = icmp eq i32 %tmp2, %tmp3
> + br i1 %matches, label %catch, label %eh.resume
> +; USE: br i1 %matches, label %catch, label %eh.resume
> +; USE-SAME: !prof ![[BW_CATCH_DISPATCH:[0-9]+]]
> +
> +catch:
> +; GEN: catch:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([3 x i8], [3 x i8]* @__llvm_profile_name_foo, i32 0, i32 0), i64 59130013419, i32 4, i32 2)
> + %tmp4 = call i8* @__cxa_begin_catch(i8* %tmp1)
> + %tmp5 = bitcast i8* %tmp4 to i32*
> + %tmp6 = load i32, i32* %tmp5, align 4
> + %tmp7 = load i32, i32* @val, align 4
> + %sub = sub nsw i32 %tmp7, %tmp6
> + store i32 %sub, i32* @val, align 4
> + call void @__cxa_end_catch()
> + br label %try.cont
> +
> +try.cont:
> +; GEN: try.cont:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + ret i32 -1
> +
> +if.end:
> +; GEN: if.end:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([3 x i8], [3 x i8]* @__llvm_profile_name_foo, i32 0, i32 0), i64 59130013419, i32 4, i32 0)
> + %tmp8 = load i32, i32* @val, align 4
> + %add = add nsw i32 %tmp8, %i
> + store i32 %add, i32* @val, align 4
> + br label %try.cont
> +
> +eh.resume:
> +; GEN: eh.resume:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([3 x i8], [3 x i8]* @__llvm_profile_name_foo, i32 0, i32 0), i64 59130013419, i32 4, i32 3)
> + %lpad.val = insertvalue { i8*, i32 } undef, i8* %tmp1, 0
> + %lpad.val3 = insertvalue { i8*, i32 } %lpad.val, i32 %tmp2, 1
> + resume { i8*, i32 } %lpad.val3
> +}
> +
> +declare i32 @__gxx_personality_v0(...)
> +
> +declare i32 @llvm.eh.typeid.for(i8*)
> +
> +declare i8* @__cxa_begin_catch(i8*)
> +
> +declare void @__cxa_end_catch()
> +
> +; USE: ![[BW_BAR_ENTRY]] = !{!"branch_weights", i32 2, i32 1}
> +; USE: ![[BW_FOO_ENTRY]] = !{!"branch_weights", i32 3, i32 2}
> +; USE: ![[BW_CATCH_DISPATCH]] = !{!"branch_weights", i32 2, i32 0}
> Index: test/Transforms/PGOProfile/criticaledge.ll
> ===================================================================
> --- /dev/null
> +++ test/Transforms/PGOProfile/criticaledge.ll
> @@ -0,0 +1,109 @@
> +; RUN: opt < %s -pgo-instr-gen -S | FileCheck %s --check-prefix=GEN
> +; RUN: llvm-profdata merge %S/Inputs/criticaledge.proftext -o %T/criticaledge.profdata
> +; RUN: opt < %s -pgo-instr-use -pgo-test-profile-file=%T/criticaledge.profdata -S | FileCheck %s --check-prefix=USE
> +target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
> +target triple = "x86_64-unknown-linux-gnu"
> +
> +; GEN: @__llvm_profile_name_test_criticalEdge = private constant [17 x i8] c"test_criticalEdge"
> +; GEN: @"__llvm_profile_name_<stdin>:bar" = private constant [11 x i8] c"<stdin>:bar"
> +
> +define i32 @test_criticalEdge(i32 %i, i32 %j) {
> +entry:
> +; CHECK: entry:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + switch i32 %i, label %sw.default [
> + i32 1, label %sw.bb
> + i32 2, label %sw.bb1
> + i32 3, label %sw.bb2
> + i32 4, label %sw.bb2
> +; CHECK: i32 3, label %entry.sw.bb2_crit_edge
> +; CHECK: i32 4, label %entry.sw.bb2_crit_edge1
> + i32 5, label %sw.bb2
> + ]
> +; USE: ]
> +; USE-SAME: !prof ![[BW_SWITCH:[0-9]+]]
> +
> +; CHECK: entry.sw.bb2_crit_edge1:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([17 x i8], [17 x i8]* @__llvm_profile_name_test_criticalEdge, i32 0, i32 0), i64 82323253069, i32 8, i32 1)
> +; CHECK: br label %sw.bb2
> +
> +; CHECK: entry.sw.bb2_crit_edge:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([17 x i8], [17 x i8]* @__llvm_profile_name_test_criticalEdge, i32 0, i32 0), i64 82323253069, i32 8, i32 0)
> +; CHECK: br label %sw.bb2
> +
> +sw.bb:
> +; GEN: sw.bb:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([17 x i8], [17 x i8]* @__llvm_profile_name_test_criticalEdge, i32 0, i32 0), i64 82323253069, i32 8, i32 5)
> + %call = call i32 @bar(i32 2)
> + br label %sw.epilog
> +
> +sw.bb1:
> +; GEN: sw.bb1:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([17 x i8], [17 x i8]* @__llvm_profile_name_test_criticalEdge, i32 0, i32 0), i64 82323253069, i32 8, i32 4)
> + %call2 = call i32 @bar(i32 1024)
> + br label %sw.epilog
> +
> +sw.bb2:
> +; GEN: sw.bb2:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + %cmp = icmp eq i32 %j, 2
> + br i1 %cmp, label %if.then, label %if.end
> +; USE: br i1 %cmp, label %if.then, label %if.end
> +; USE-SAME: !prof ![[BW_SW_BB2:[0-9]+]]
> +
> +if.then:
> +; GEN: if.then:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([17 x i8], [17 x i8]* @__llvm_profile_name_test_criticalEdge, i32 0, i32 0), i64 82323253069, i32 8, i32 2)
> + %call4 = call i32 @bar(i32 4)
> + br label %return
> +
> +if.end:
> +; GEN: if.end:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([17 x i8], [17 x i8]* @__llvm_profile_name_test_criticalEdge, i32 0, i32 0), i64 82323253069, i32 8, i32 3)
> + %call5 = call i32 @bar(i32 8)
> + br label %sw.epilog
> +
> +sw.default:
> +; GEN: sw.default:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + %call6 = call i32 @bar(i32 32)
> + %cmp7 = icmp sgt i32 %j, 10
> + br i1 %cmp7, label %if.then8, label %if.end9
> +; USE: br i1 %cmp7, label %if.then8, label %if.end9
> +; USE-SAME: !prof ![[BW_SW_DEFAULT:[0-9]+]]
> +
> +if.then8:
> +; GEN: if.then8:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([17 x i8], [17 x i8]* @__llvm_profile_name_test_criticalEdge, i32 0, i32 0), i64 82323253069, i32 8, i32 7)
> + %add = add nsw i32 %call6, 10
> + br label %if.end9
> +
> +if.end9:
> +; GEN: if.end9:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([17 x i8], [17 x i8]* @__llvm_profile_name_test_criticalEdge, i32 0, i32 0), i64 82323253069, i32 8, i32 6)
> + %res.0 = phi i32 [ %add, %if.then8 ], [ %call6, %sw.default ]
> + br label %sw.epilog
> +
> +sw.epilog:
> +; GEN: sw.epilog:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + %res.1 = phi i32 [ %res.0, %if.end9 ], [ %call5, %if.end ], [ %call2, %sw.bb1 ], [ %call, %sw.bb ]
> + br label %return
> +
> +return:
> +; GEN: return:
> +; GEN-NOT: call void @llvm.instrprof.increment
> + %retval = phi i32 [ %res.1, %sw.epilog ], [ %call4, %if.then ]
> + ret i32 %retval
> +}
> +
> +define internal i32 @bar(i32 %i) {
> +entry:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([11 x i8], [11 x i8]* @"__llvm_profile_name_<stdin>:bar", i32 0, i32 0), i64 12884901887, i32 1, i32 0)
> + ret i32 %i
> +}
> +
> +; USE: ![[BW_SWITCH]] = !{!"branch_weights", i32 2, i32 1, i32 0, i32 2, i32 1, i32 1}
> +; USE: ![[BW_SW_BB2]] = !{!"branch_weights", i32 2, i32 2}
> +; USE: ![[BW_SW_DEFAULT]] = !{!"branch_weights", i32 1, i32 1}
> +
> Index: test/Transforms/PGOProfile/checksum_mismatch.ll
> ===================================================================
> --- /dev/null
> +++ test/Transforms/PGOProfile/checksum_mismatch.ll
> @@ -0,0 +1,18 @@
> +; RUN: llvm-profdata merge %S/Inputs/single_bb.proftext -o %T/single_bb.profdata
> +; RUN: opt < %s -pgo-instr-use -pgo-test-profile-file=%T/single_bb.profdata -S 2>&1 | FileCheck %s
> +
> +; CHECK: Function control flow change detected (hash mismatch) single_bb
> +; CHECK: No profile data available for function uncalled
> +
> +target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
> +target triple = "x86_64-unknown-linux-gnu"
> +
> +define i32 @single_bb() {
> +entry:
> + ret i32 0
> +}
> +
> +define i32 @uncalled(i32 %i, i32 %j) {
> + %mul = mul nsw i32 %i, %j
> + ret i32 %mul
> +}
> Index: test/Transforms/PGOProfile/branch2.ll
> ===================================================================
> --- /dev/null
> +++ test/Transforms/PGOProfile/branch2.ll
> @@ -0,0 +1,38 @@
> +; RUN: opt < %s -pgo-instr-gen -S | FileCheck %s --check-prefix=GEN
> +; RUN: llvm-profdata merge %S/Inputs/branch2.proftext -o %T/branch2.profdata
> +; RUN: opt < %s -pgo-instr-use -pgo-test-profile-file=%T/branch2.profdata -S | FileCheck %s --check-prefix=USE
> +target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
> +target triple = "x86_64-unknown-linux-gnu"
> +
> +; GEN: @__llvm_profile_name_test_br_2 = private constant [9 x i8] c"test_br_2"
> +
> +define i32 @test_br_2(i32 %i) {
> +entry:
> +; GEN: entry:
> +; GEN-NOT: llvm.instrprof.increment
> + %cmp = icmp sgt i32 %i, 0
> + br i1 %cmp, label %if.then, label %if.else
> +; USE: br i1 %cmp, label %if.then, label %if.else
> +; USE-SAME: !prof ![[BW_ENTRY:[0-9]+]]
> +
> +if.then:
> +; GEN: if.then:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([9 x i8], [9 x i8]* @__llvm_profile_name_test_br_2, i32 0, i32 0), i64 29667547796, i32 2, i32 0)
> + %add = add nsw i32 %i, 2
> + br label %if.end
> +
> +if.else:
> +; GEN: if.else:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([9 x i8], [9 x i8]* @__llvm_profile_name_test_br_2, i32 0, i32 0), i64 29667547796, i32 2, i32 1)
> + %sub = sub nsw i32 %i, 2
> + br label %if.end
> +
> +if.end:
> +; GEN: if.end:
> +; GEN-NOT: llvm.instrprof.increment
> + %retv = phi i32 [ %add, %if.then ], [ %sub, %if.else ]
> + ret i32 %retv
> +; GEN: ret
> +}
> +
> +; USE: ![[BW_ENTRY]] = !{!"branch_weights", i32 1, i32 1}
> Index: test/Transforms/PGOProfile/branch1.ll
> ===================================================================
> --- /dev/null
> +++ test/Transforms/PGOProfile/branch1.ll
> @@ -0,0 +1,31 @@
> +; RUN: opt < %s -pgo-instr-gen -S | FileCheck %s --check-prefix=GEN
> +; RUN: llvm-profdata merge %S/Inputs/branch1.proftext -o %T/branch1.profdata
> +; RUN: opt < %s -pgo-instr-use -pgo-test-profile-file=%T/branch1.profdata -S | FileCheck %s --check-prefix=USE
> +target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
> +target triple = "x86_64-unknown-linux-gnu"
> +
> +; GEN: @__llvm_profile_name_test_br_1 = private constant [9 x i8] c"test_br_1"
> +
> +define i32 @test_br_1(i32 %i) {
> +entry:
> +; GEN: entry:
> +; GEN-NOT: llvm.instrprof.increment
> + %cmp = icmp sgt i32 %i, 0
> + br i1 %cmp, label %if.then, label %if.end
> +; USE: br i1 %cmp, label %if.then, label %if.end
> +; USE-SAME: !prof ![[BW_ENTRY:[0-9]+]]
> +
> +if.then:
> +; GEN: if.then:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([9 x i8], [9 x i8]* @__llvm_profile_name_test_br_1, i32 0, i32 0), i64 25571299074, i32 2, i32 1)
> + %add = add nsw i32 %i, 2
> + br label %if.end
> +
> +if.end:
> +; GEN: if.end:
> +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds ([9 x i8], [9 x i8]* @__llvm_profile_name_test_br_1, i32 0, i32 0), i64 25571299074, i32 2, i32 0)
> + %retv = phi i32 [ %add, %if.then ], [ %i, %entry ]
> + ret i32 %retv
> +}
> +
> +; USE: ![[BW_ENTRY]] = !{!"branch_weights", i32 2, i32 1}
> Index: test/Transforms/PGOProfile/Inputs/switch.proftext
> ===================================================================
> --- /dev/null
> +++ test/Transforms/PGOProfile/Inputs/switch.proftext
> @@ -0,0 +1,8 @@
> +test_switch
> +46200943743
> +4
> +0
> +5
> +2
> +3
> +
> Index: test/Transforms/PGOProfile/Inputs/single_bb.proftext
> ===================================================================
> --- /dev/null
> +++ test/Transforms/PGOProfile/Inputs/single_bb.proftext
> @@ -0,0 +1,5 @@
> +single_bb
> +12884999999
> +1
> +1
> +
> Index: test/Transforms/PGOProfile/Inputs/loop2.proftext
> ===================================================================
> --- /dev/null
> +++ test/Transforms/PGOProfile/Inputs/loop2.proftext
> @@ -0,0 +1,8 @@
> +test_nested_for
> +74483645111
> +4
> +186
> +33
> +10
> +6
> +
> Index: test/Transforms/PGOProfile/Inputs/loop1.proftext
> ===================================================================
> --- /dev/null
> +++ test/Transforms/PGOProfile/Inputs/loop1.proftext
> @@ -0,0 +1,6 @@
> +test_simple_for
> +34137660316
> +2
> +96
> +4
> +
> Index: test/Transforms/PGOProfile/Inputs/landingpad.proftext
> ===================================================================
> --- /dev/null
> +++ test/Transforms/PGOProfile/Inputs/landingpad.proftext
> @@ -0,0 +1,14 @@
> +foo
> +59130013419
> +4
> +3
> +1
> +2
> +0
> +
> +bar
> +24868915205
> +2
> +1
> +2
> +
> Index: test/Transforms/PGOProfile/Inputs/criticaledge.proftext
> ===================================================================
> --- /dev/null
> +++ test/Transforms/PGOProfile/Inputs/criticaledge.proftext
> @@ -0,0 +1,17 @@
> +test_criticalEdge
> +82323253069
> +8
> +2
> +1
> +2
> +2
> +0
> +1
> +2
> +1
> +
> +<stdin>:bar
> +12884901887
> +1
> +7
> +
> Index: test/Transforms/PGOProfile/Inputs/branch2.proftext
> ===================================================================
> --- /dev/null
> +++ test/Transforms/PGOProfile/Inputs/branch2.proftext
> @@ -0,0 +1,6 @@
> +test_br_2
> +29667547796
> +2
> +1
> +1
> +
> Index: test/Transforms/PGOProfile/Inputs/branch1.proftext
> ===================================================================
> --- /dev/null
> +++ test/Transforms/PGOProfile/Inputs/branch1.proftext
> @@ -0,0 +1,6 @@
> +test_br_1
> +25571299074
> +2
> +3
> +2
> +
> Index: lib/Transforms/Instrumentation/PGOInstrumentation.cpp
> ===================================================================
> --- /dev/null
> +++ lib/Transforms/Instrumentation/PGOInstrumentation.cpp
> @@ -0,0 +1,721 @@
> +//===-- PGOInstru.cpp - PGO Instrumentation -------------------------------===//
This should match the filename (PGOInstrumentation.cpp). I might even go
so far as to say "MST-based PGO instrumentation" or something like that
in the top comment, just to differentiate it from the frontend
instrumentation.
> +//
> +// The LLVM Compiler Infrastructure
> +//
> +// This file is distributed under the University of Illinois Open Source
> +// License. See LICENSE.TXT for details.
> +//
> +//===----------------------------------------------------------------------===//
> +//
> +// This file implements PGO instrumentation using a minimum spanning tree based
> +// on the following paper:
> +// [1] Donald E. Knuth, Francis R. Stevenson. Optimal measurement of points
> +// for program frequency counts. BIT Numerical Mathematics 1973, Volume 13,
> +// Issue 3, pp 313-322
> +// The idea of the algorithm based on the fact that for each node (except for
> +// the entry and exit), the sum of incoming edge counts equals the sum of
> +// outgoing edge counts. The count of edge on spanning tree can be derived from
> +// those edges not on the spanning tree. Knuth proves this method instruments
> +// the minimum number of edges.
> +//
> +// The minimal spanning tree here is actually a maximum weight tree -- on-tree
> +// edges have higher frequencies (more likely to execute). The idea is to
> +// instrument those less frequently executed edges to reduce the runtime
> +// overhead of instrumented binaries.
> +//
> +// This file contains two passes:
> +// (1) Pass PGOInstrumentationGen which instruments the IR to generate edge
> +// count profile, and
> +// (2) Pass PGOInstrumentationUse which reads the edge count profile and
> +// annotates the branch weights.
> +// To get the precise counter information, These two passes need to invoke at
> +// the same compilation point (so they see the same IR). For pass
> +// PGOInstrumentationGen, the real work is done in instrumentOneFunc(). For
> +// pass PGOInstrumentationUse, the real work in done in class PGOUseFunc and
> +// the profile is opened in module level and passed to each PGOUseFunc instance.
> +// The shared code for PGOInstrumentationGen and PGOInstrumentationUse is put
> +// in class FuncPGOInstrumentation.
> +//
> +// Class PGOEdge represents a CFG edge and some auxiliary information. Class
> +// BBInfo contains auxiliary information for each BB. These two classes are used
> +// in pass PGOInstrumentationGen. Class PGOUseEdge and UseBBInfo are the derived
> +// class of PGOEdge and BBInfo, respectively. They contains extra data structure
> +// used in populating profile counters.
> +// The MST implementation is in Class CFGMST (CFGMST.h).
> +//
> +//===----------------------------------------------------------------------===//
> +
> +#include "llvm/Transforms/Instrumentation.h"
> +#include "CFGMST.h"
> +#include "llvm/ADT/DenseMap.h"
> +#include "llvm/ADT/STLExtras.h"
> +#include "llvm/ADT/Statistic.h"
> +#include "llvm/Analysis/BlockFrequencyInfo.h"
> +#include "llvm/Analysis/BranchProbabilityInfo.h"
> +#include "llvm/Analysis/CFG.h"
> +#include "llvm/IR/DiagnosticInfo.h"
> +#include "llvm/IR/IRBuilder.h"
> +#include "llvm/IR/InstIterator.h"
> +#include "llvm/IR/Instructions.h"
> +#include "llvm/IR/IntrinsicInst.h"
> +#include "llvm/IR/MDBuilder.h"
> +#include "llvm/IR/Module.h"
> +#include "llvm/Pass.h"
> +#include "llvm/ProfileData/InstrProfReader.h"
> +#include "llvm/Support/BranchProbability.h"
> +#include "llvm/Support/Debug.h"
> +#include "llvm/Support/JamCRC.h"
> +#include "llvm/Transforms/Utils/BasicBlockUtils.h"
> +#include <string>
> +#include <utility>
> +#include <vector>
> +
> +using namespace llvm;
> +
> +#define DEBUG_TYPE "pgo-instr"
It's probably more discoverable to use "pgo-instrumentation" to match
the filename.
> +
> +STATISTIC(NumOfPGOInstrument, "Number of edges instrumented.");
> +STATISTIC(NumOfPGOEdge, "Number of edges.");
> +STATISTIC(NumOfPGOBB, "Number of basic-blocks.");
> +STATISTIC(NumOfPGOSplit, "Number of critical edge splits.");
> +STATISTIC(NumOfPGOFunc, "Number of functions having valid profile counts.");
> +STATISTIC(NumOfPGOMismatch, "Number of functions having mismatch profile.");
> +STATISTIC(NumOfPGOMissing, "Number of functions without profile.");
> +
> +// Command line option to specify the file to read profile from. This is
> +// mainly used for testing.
> +static cl::opt<std::string>
> + PGOProfileFile("pgo-test-profile-file", cl::init(""), cl::Hidden,
> + cl::value_desc("filename"),
> + cl::desc("Specify the path of profile data file. This is"
> + "mainly for test purpose."));
> +
> +namespace {
> +class PGOInstrumentationGen : public ModulePass {
> +public:
> + static char ID;
> +
> + PGOInstrumentationGen() : ModulePass(ID) {
> + initializePGOInstrumentationGenPass(*PassRegistry::getPassRegistry());
> + }
> +
> + const char *getPassName() const override {
> + return "PGOInstrumentationGenPass";
> + }
> +
> +private:
> + bool runOnModule(Module &M) override;
> +
> + void getAnalysisUsage(AnalysisUsage &AU) const override {
> + AU.addRequired<BlockFrequencyInfoWrapperPass>();
> + AU.addRequired<BranchProbabilityInfoWrapperPass>();
> + }
> +};
> +
> +class PGOInstrumentationUse : public ModulePass {
> +public:
> + static char ID;
> +
> + // Provide the profile filename as the parameter.
> + PGOInstrumentationUse(StringRef Filename = StringRef(""))
StringRef seems like a poor argument type here, since it isn't clear
that the string needs to outlive the class.
> + : ModulePass(ID), ProfileFileName(Filename) {
> + if (!PGOProfileFile.empty())
> + ProfileFileName = StringRef(PGOProfileFile);
> + initializePGOInstrumentationUsePass(*PassRegistry::getPassRegistry());
> + }
> +
> + const char *getPassName() const override {
> + return "PGOInstrumentationUsePass";
> + }
> +
> +private:
> + StringRef ProfileFileName;
> + std::unique_ptr<IndexedInstrProfReader> PGOReader;
> + bool runOnModule(Module &M) override;
> +
> + void getAnalysisUsage(AnalysisUsage &AU) const override {
> + AU.addRequired<BlockFrequencyInfoWrapperPass>();
> + AU.addRequired<BranchProbabilityInfoWrapperPass>();
> + }
> +};
> +} // end anonymous namespace
> +
> +char PGOInstrumentationGen::ID = 0;
> +INITIALIZE_PASS_BEGIN(PGOInstrumentationGen, "pgo-instr-gen",
> + "PGO instrumentation.", false, false)
> +INITIALIZE_PASS_DEPENDENCY(BlockFrequencyInfoWrapperPass)
> +INITIALIZE_PASS_DEPENDENCY(BranchProbabilityInfoWrapperPass)
> +INITIALIZE_PASS_END(PGOInstrumentationGen, "pgo-instr-gen",
> + "PGO instrumentation.", false, false)
> +
> +ModulePass *llvm::createPGOInstrumentationGenPass() {
> + return new PGOInstrumentationGen();
> +}
> +
> +char PGOInstrumentationUse::ID = 0;
> +INITIALIZE_PASS_BEGIN(PGOInstrumentationUse, "pgo-instr-use",
> + "Read PGO instrumentation profile.", false, false)
> +INITIALIZE_PASS_DEPENDENCY(BlockFrequencyInfoWrapperPass)
> +INITIALIZE_PASS_DEPENDENCY(BranchProbabilityInfoWrapperPass)
> +INITIALIZE_PASS_END(PGOInstrumentationUse, "pgo-instr-use",
> + "Read PGO instrumentation profile.", false, false)
> +
> +ModulePass *llvm::createPGOInstrumentationUsePass(StringRef Filename) {
> + return new PGOInstrumentationUse(Filename);
> +}
> +
> +namespace {
> +/// \brief An MST based instrumentation for PGO
> +///
> +/// Implements a Minimum Spanning Tree (MST) based instrumentation for PGO
> +/// in the function level.
> +//
> +// This class implements the CFG edges. Note the CFG can be a multi-graph.
This seems to switch from a doc-comment to a regular comment halfway
through.
> +struct PGOEdge {
> + const BasicBlock *SrcBB;
> + const BasicBlock *DestBB;
> + uint64_t Weight;
> + bool InMST;
> + bool Removed;
> + bool IsCritical;
> + PGOEdge(const BasicBlock *Src, const BasicBlock *Dest, unsigned W = 1)
> + : SrcBB(Src), DestBB(Dest), Weight(W), InMST(false), Removed(false),
> + IsCritical(false) {}
> + // Return the information string of an edge.
> + const std::string infoString() const {
> + std::string Str = (Removed ? "-" : " ");
> + Str += (InMST ? " " : "*");
> + Str += (IsCritical ? "c" : " ");
> + Str += " W=" + std::to_string(Weight);
ISTR that std::to_string isn't available in all of the standard
libraries we support, so this might fail on some of the bots as is.
> + return Str;
> + }
> +};
> +
> +// This class stores the auxiliary information for each BB.
> +struct BBInfo {
> + BBInfo *Group;
> + uint32_t Index;
> + uint32_t Rank;
> +
> + BBInfo(unsigned IX) : Group(this), Index(IX), Rank(0) {}
> +
> + // Return the information string of this object.
> + const std::string infoString() const {
> + return "Index=" + std::to_string(Index);
> + }
> +};
> +
> +// This class implements the CFG edges. Note the CFG can be a multi-graph.
> +template <class Edge, class BBInfo> class FuncPGOInstrumentation {
> +private:
> + Function &F;
> + void computeCFGHash();
> +
> +public:
> + std::string FuncName;
> + GlobalVariable *FuncNameVar;
> + // CFG hash value for this function.
> + uint64_t FunctionHash;
> +
> + // The Minimum Spanning Tree of function CFG.
> + CFGMST<Edge, BBInfo> MST;
> +
> + // Give an edge, find the BB that will be instrumented.
> + // Return nullptr if there is no BB to be instrumented.
> + BasicBlock *getInstrBB(Edge *E);
> +
> + // Return the auxiliary BB information.
> + BBInfo &getBBInfo(const BasicBlock *BB) const { return MST.getBBInfo(BB); }
> +
> + // Dump edges and BB information.
> + void dumpInfo(std::string Str = "") const {
> + std::string Message = "Dump Function " + FuncName + " Hash: " +
> + std::to_string(FunctionHash) + "\t" + Str;
> + MST.dumpEdges(dbgs(), Message);
Would it make more sense for dumpEdges to take a Twine?
> + }
> +
> + FuncPGOInstrumentation(Function &Func, bool CreateGlobalVar = false,
> + BranchProbabilityInfo *BPI_ = nullptr,
> + BlockFrequencyInfo *BFI_ = nullptr)
> + : F(Func), FunctionHash(0), MST(F, BPI_, BFI_) {
No need for underscores in BPI_ and BFI_ - the shadowing is unambiguous.
> + FuncName = getPGOFuncName(F);
> + computeCFGHash();
> + DEBUG(dumpInfo("after CFGMST"));
> +
> + NumOfPGOBB += MST.BBInfos.size();
> + for (auto &Ei : MST.AllEdges) {
Why "Ei"? "E" for edge would be more typical.
> + if (Ei->Removed)
> + continue;
> + NumOfPGOEdge++;
> + if (!Ei->InMST)
> + NumOfPGOInstrument++;
> + }
> +
> + if (CreateGlobalVar)
> + FuncNameVar = createPGOFuncNameVar(F, FuncName);
> + };
> +};
> +
> +// Compute Hash value for the CFG: the lower 32 bits are CRC32 of the index
> +// value of each BB in the CFG. The higher 32 bits record the number of edges.
> +template <class Edge, class BBInfo>
> +void FuncPGOInstrumentation<Edge, BBInfo>::computeCFGHash() {
> + std::vector<char> Indexes;
> + JamCRC JC;
> + for (auto &BB : F) {
> + const TerminatorInst *TI = BB.getTerminator();
> + for (unsigned s = 0, e = TI->getNumSuccessors(); s != e; ++s) {
> + BasicBlock *Succ = TI->getSuccessor(s);
> + uint32_t Index = getBBInfo(Succ).Index;
> + for (int i = 0; i < 4; i++)
> + Indexes.push_back((char)(Index >> (i * 8)));
> + }
> + }
> + JC.update(Indexes);
> + FunctionHash = (uint64_t)MST.AllEdges.size() << 32 | JC.getCRC();
> +}
> +
> +template <class Edge, class BBInfo>
> +BasicBlock *FuncPGOInstrumentation<Edge, BBInfo>::getInstrBB(Edge *E) {
> + if (E->InMST || E->Removed)
> + return nullptr;
> +
> + BasicBlock *SrcBB = const_cast<BasicBlock *>(E->SrcBB);
> + BasicBlock *DestBB = const_cast<BasicBlock *>(E->DestBB);
> + // For a fake edge, instrument the real BB.
> + if (SrcBB == nullptr)
> + return DestBB;
> + if (DestBB == nullptr)
> + return SrcBB;
> +
> + // Instrument the SrcBB if it has a single successor,
> + // otherwise, the DestBB if this is not a critical edge.
> + TerminatorInst *TI = SrcBB->getTerminator();
> + if (TI->getNumSuccessors() <= 1)
> + return SrcBB;
> + if (!E->IsCritical)
> + return DestBB;
> +
> + // For a critical edge, we have to split. Instrument the newly
> + // created BB.
> + NumOfPGOSplit++;
> + DEBUG(dbgs() << "Split critical edge: " << getBBInfo(SrcBB).Index << " --> "
> + << getBBInfo(DestBB).Index << "\n");
> + unsigned SuccNum = GetSuccessorNumber(SrcBB, DestBB);
> + BasicBlock *InstrBB = SplitCriticalEdge(TI, SuccNum);
> + assert(InstrBB && "Critical edge is not split");
> +
> + E->Removed = true;
> + return InstrBB;
> +}
> +
> +// Visit all edge and instrument the edges not in MST.
> +// Critical edges will be split.
> +static void instrumentOneFunc(Function &F, Module *M,
> + BranchProbabilityInfo *BPI,
> + BlockFrequencyInfo *BFI) {
> + unsigned NumCounters = 0;
> + FuncPGOInstrumentation<PGOEdge, BBInfo> FuncInfo(F, true, BPI, BFI);
> + for (auto &Ei : FuncInfo.MST.AllEdges) {
> + if (!Ei->InMST && !Ei->Removed)
> + NumCounters++;
> + }
> +
> + uint32_t j = 0;
The style says this should be a capital J (also, why not I?).
> + for (auto &Ei : FuncInfo.MST.AllEdges) {
> + BasicBlock *InstrBB = FuncInfo.getInstrBB(Ei.get());
> + if (!InstrBB)
> + continue;
> +
> + IRBuilder<> Builder(InstrBB, InstrBB->getFirstInsertionPt());
> + assert(Builder.GetInsertPoint() != InstrBB->end() &&
> + "Cannot get the Instrumentation point");
> + auto *I8PtrTy = Type::getInt8PtrTy(M->getContext());
Not worth using auto here, `Type` is the same number of characters.
> + Builder.CreateCall(
> + Intrinsic::getDeclaration(M, Intrinsic::instrprof_increment),
> + {llvm::ConstantExpr::getBitCast(FuncInfo.FuncNameVar, I8PtrTy),
> + Builder.getInt64(FuncInfo.FunctionHash), Builder.getInt32(NumCounters),
> + Builder.getInt32(j++)});
> + }
> +}
> +
> +struct PGOUseEdge : public PGOEdge {
> + bool CountValid;
> + uint64_t CountValue;
> + PGOUseEdge(const BasicBlock *Src, const BasicBlock *Dest, unsigned W = 1)
> + : PGOEdge(Src, Dest, W), CountValid(false), CountValue(0) {}
> +
> + // Set edge count value
> + void setEdgeCount(uint64_t Value) {
> + CountValue = Value;
> + CountValid = true;
> + }
> +
> + // Return the information string for this object.
> + const std::string infoString() const {
> + if (!CountValid)
> + return PGOEdge::infoString();
> + return PGOEdge::infoString() + " Count=" + std::to_string(CountValue);
> + }
> +};
> +
> +typedef SmallVector<PGOUseEdge *, 2> DirectEdges;
> +
> +// This class stores the auxiliary information for each BB.
> +struct UseBBInfo : public BBInfo {
> + uint64_t CountValue;
> + bool CountValid;
> + int32_t UnknownCountInEdge;
> + int32_t UnknownCountOutEdge;
> + DirectEdges InEdges;
> + DirectEdges OutEdges;
> + UseBBInfo(unsigned IX)
> + : BBInfo(IX), CountValue(0), CountValid(false), UnknownCountInEdge(0),
> + UnknownCountOutEdge(0) {}
> + UseBBInfo(unsigned IX, uint64_t C)
> + : BBInfo(IX), CountValue(C), CountValid(true), UnknownCountInEdge(0),
> + UnknownCountOutEdge(0) {}
> +
> + // Set the profile count value for this BB.
> + void setBBInfoCount(uint64_t Value) {
> + CountValue = Value;
> + CountValid = true;
> + }
> +
> + // Return the information string of this object.
> + const std::string infoString() const {
> + if (!CountValid)
> + return BBInfo::infoString();
> + return BBInfo::infoString() + " Count=" + std::to_string(CountValue);
> + }
> +};
> +
> +// Sum up the count values for all the edges.
> +static uint64_t sumEdgeCount(const ArrayRef<PGOUseEdge *> Edges) {
> + uint64_t Total = 0;
> + for (auto &Ei : Edges) {
> + if (Ei->Removed)
> + continue;
> + Total += Ei->CountValue;
> + }
> + return Total;
> +}
> +
> +class PGOUseFunc {
> +private:
> + Function &F;
> + Module *M;
> + // This member stores the shared information with class PGOGenFunc.
> + FuncPGOInstrumentation<PGOUseEdge, UseBBInfo> FuncInfo;
> +
> + // Return the auxiliary BB information.
> + UseBBInfo &getBBInfo(const BasicBlock *BB) const {
> + return FuncInfo.getBBInfo(BB);
> + }
> +
> + // The maximum count value in the profile. This is only used in PGO use
> + // compilation.
> + uint64_t ProgramMaxCount;
> +
> + // Find the Instrumented BB and set the value.
> + void setInstrumentedCounts(const std::vector<uint64_t> &CountFromProfile);
> +
> + // Set the edge counter value for the unknown edge -- there should be only
> + // one unknown edge.
> + void setEdgeCount(DirectEdges &Edges, uint64_t Value);
> +
> + // Return FuncName string;
> + const std::string getFuncName() const { return FuncInfo.FuncName; }
> +
> + // Set the hot/cold inline hints based on the count values.
> + void applyFunctionAttributes(uint64_t EntryCount, uint64_t MaxCount) {
> + if (ProgramMaxCount == 0)
> + return;
> + // Threshold of the hot functions.
> + const BranchProbability HotFunctionThreshold(1, 100);
> + // Threshold of the cold functions.
> + const BranchProbability ColdFunctionThreshold(2, 10000);
> + if (EntryCount >= HotFunctionThreshold.scale(ProgramMaxCount))
> + F.addFnAttr(llvm::Attribute::InlineHint);
> + else if (MaxCount <= ColdFunctionThreshold.scale(ProgramMaxCount))
> + F.addFnAttr(llvm::Attribute::Cold);
This is more or less a workaround for limitations of the inliner and
should eventually be removed in favour of making the inliner smarter,
correct?
> + }
> +
> +public:
> + PGOUseFunc(Function &Func, Module *Modu,
> + BranchProbabilityInfo *BPI_ = nullptr,
> + BlockFrequencyInfo *BFI_ = nullptr)
> + : F(Func), M(Modu), FuncInfo(Func, false, BPI_, BFI_) {}
> +
> + // Read counts for the instrumented BB from profile.
> + bool readCounters(IndexedInstrProfReader *PGOReader);
> +
> + // Populate the counts for all BBs.
> + void populateCounters();
> +
> + // Set the branch weights based on the count values.
> + void setBranchWeights();
> +};
> +
> +// Visit all the edges and assign the count value for the instrumented
> +// edges and the BB.
> +void PGOUseFunc::setInstrumentedCounts(
> + const std::vector<uint64_t> &CountFromProfile) {
> +
> + // Use a worklist as we will update the vector during the iteration.
> + std::vector<PGOUseEdge *> WorkList;
> + for (auto &Ei : FuncInfo.MST.AllEdges)
> + WorkList.push_back(Ei.get());
> +
> + uint32_t j = 0;
> + for (auto &Ei : WorkList) {
Same as before: j->J, Ei->E.
> + BasicBlock *InstrBB = FuncInfo.getInstrBB(Ei);
> + if (!InstrBB)
> + continue;
> + uint64_t CountValue = CountFromProfile[j++];
> + if (!Ei->Removed) {
> + getBBInfo(InstrBB).setBBInfoCount(CountValue);
> + Ei->setEdgeCount(CountValue);
> + continue;
> + }
> +
> + // Need to add two new edges.
> + BasicBlock *SrcBB = const_cast<BasicBlock *>(Ei->SrcBB);
> + BasicBlock *DestBB = const_cast<BasicBlock *>(Ei->DestBB);
> + // Add new edge of SrcBB->InstrBB.
> + PGOUseEdge &NewEdge = FuncInfo.MST.addEdge(SrcBB, InstrBB, 0);
> + NewEdge.setEdgeCount(CountValue);
> + // Add new edge of InstrBB->DestBB.
> + PGOUseEdge &NewEdge1 = FuncInfo.MST.addEdge(InstrBB, DestBB, 0);
Why not just reuse the NewEdge variable here?
> + NewEdge1.setEdgeCount(CountValue);
> + NewEdge1.InMST = true;
> + getBBInfo(InstrBB).setBBInfoCount(CountValue);
> + }
> +}
> +
> +// Set the count value for the unknown edge. There should be one and only one
> +// unknown edge in Edges vector.
> +void PGOUseFunc::setEdgeCount(DirectEdges &Edges, uint64_t Value) {
> + for (auto &Ei : Edges) {
> + if (Ei->CountValid)
> + continue;
> + Ei->setEdgeCount(Value);
> +
> + getBBInfo(Ei->SrcBB).UnknownCountOutEdge--;
> + getBBInfo(Ei->DestBB).UnknownCountInEdge--;
> + return;
> + }
> + llvm_unreachable("Cannot find the unknown count edge");
> +}
> +
> +// Read the profile from ProfileFileName and assign the value to the
> +// instrumented BB and the edges. This function also updates ProgramMaxCount.
> +// Return true if the profile are successfully read, and false on errors.
> +bool PGOUseFunc::readCounters(IndexedInstrProfReader *PGOReader) {
> + auto &Ctx = M->getContext();
> + ErrorOr<InstrProfRecord> Result =
> + PGOReader->getInstrProfRecord(FuncInfo.FuncName, FuncInfo.FunctionHash);
> + if (std::error_code EC = Result.getError()) {
> + if (EC == instrprof_error::unknown_function)
> + NumOfPGOMissing++;
> + else if (EC == instrprof_error::hash_mismatch ||
> + EC == llvm::instrprof_error::malformed)
> + NumOfPGOMismatch++;
> +
> + std::string Msg = EC.message() + std::string(" ") + F.getName().str();
> + Ctx.diagnose(
> + DiagnosticInfoPGOProfile(M->getName().data(), Msg, DS_Warning));
> + return false;
> + }
> + std::vector<uint64_t> &CountFromProfile = Result.get().Counts;
> +
> + NumOfPGOFunc++;
> + DEBUG(dbgs() << CountFromProfile.size() << " counts\n");
> + uint64_t ValueSum = 0;
> + for (unsigned i = 0, e = CountFromProfile.size(); i < e; i++) {
> + DEBUG(dbgs() << " " << i << ": " << CountFromProfile[i] << "\n");
> + ValueSum += CountFromProfile[i];
> + }
> +
> + DEBUG(dbgs() << "SUM = " << ValueSum << "\n");
> +
> + getBBInfo(nullptr).UnknownCountOutEdge = 2;
> + getBBInfo(nullptr).UnknownCountInEdge = 2;
> +
> + setInstrumentedCounts(CountFromProfile);
> + ProgramMaxCount = PGOReader->getMaximumFunctionCount();
> + return true;
> +}
> +
> +// Populate the counters from instrumented BBs to all BBs.
> +// In the end of this operation, all BBs should have a valid count value.
> +void PGOUseFunc::populateCounters() {
> + // First set up Count variable for all BBs.
> + for (auto &Ei : FuncInfo.MST.AllEdges) {
> + if (Ei->Removed)
> + continue;
> +
> + const BasicBlock *SrcBB = Ei->SrcBB;
> + const BasicBlock *DestBB = Ei->DestBB;
> + UseBBInfo &SrcInfo = getBBInfo(SrcBB);
> + UseBBInfo &DestInfo = getBBInfo(DestBB);
> + SrcInfo.OutEdges.push_back(Ei.get());
> + DestInfo.InEdges.push_back(Ei.get());
> + SrcInfo.UnknownCountOutEdge++;
> + DestInfo.UnknownCountInEdge++;
> +
> + if (!Ei->CountValid)
> + continue;
> + DestInfo.UnknownCountInEdge--;
> + SrcInfo.UnknownCountOutEdge--;
> + }
> +
> + bool Changes = true;
> + unsigned NumPasses = 0;
> + while (Changes) {
> + NumPasses++;
> + Changes = false;
> +
> + // For efficient traversal, it's better to start from the end as most
> + // of the instrumented edges are at the end.
> + for (auto &BB : reverse(F)) {
> + UseBBInfo &Count = getBBInfo(&BB);
> + if (!Count.CountValid) {
> + if (Count.UnknownCountOutEdge == 0) {
> + Count.CountValue = sumEdgeCount(Count.OutEdges);
> + Count.CountValid = true;
> + Changes = true;
> + } else if (Count.UnknownCountInEdge == 0) {
> + Count.CountValue = sumEdgeCount(Count.InEdges);
> + Count.CountValid = true;
> + Changes = true;
> + }
> + }
> + if (Count.CountValid) {
> + if (Count.UnknownCountOutEdge == 1) {
> + uint64_t Total = Count.CountValue - sumEdgeCount(Count.OutEdges);
> + setEdgeCount(Count.OutEdges, Total);
> + Changes = true;
> + }
> + if (Count.UnknownCountInEdge == 1) {
> + uint64_t Total = Count.CountValue - sumEdgeCount(Count.InEdges);
> + setEdgeCount(Count.InEdges, Total);
> + Changes = true;
> + }
> + }
> + }
> + }
> +
> + DEBUG(dbgs() << "Populate counts in " << NumPasses << " passes.\n");
> + // Assert every BB has a valid counter.
> + uint64_t FuncEntryCount = getBBInfo(&*F.begin()).CountValue;
> + uint64_t FuncMaxCount = FuncEntryCount;
> + for (auto &BB : F) {
> + assert(getBBInfo(&BB).CountValid && "BB count is not valid");
> + uint64_t Count = getBBInfo(&BB).CountValue;
> + if (Count > FuncMaxCount)
> + FuncMaxCount = Count;
> + }
> + applyFunctionAttributes(FuncEntryCount, FuncMaxCount);
> +
> + DEBUG(FuncInfo.dumpInfo("after reading profile."));
> +}
> +
> +// Assign the scaled count values to the BB with multiple out edges.
> +void PGOUseFunc::setBranchWeights() {
> + // Generate MD_prof metadata for every branch instruction.
> + DEBUG(dbgs() << "\nSetting branch weights.\n");
> + MDBuilder MDB(M->getContext());
> + for (auto &BB : F) {
> + TerminatorInst *TI = BB.getTerminator();
> + if (TI->getNumSuccessors() < 2)
> + continue;
> + if (!isa<BranchInst>(TI) && !isa<SwitchInst>(TI))
> + continue;
> + if (getBBInfo(&BB).CountValue == 0)
> + continue;
> +
> + // We have a non-zero Branch BB.
> + const UseBBInfo &BBCountInfo = getBBInfo(&BB);
> + unsigned Size = BBCountInfo.OutEdges.size();
> + SmallVector<unsigned, 2> EdgeCounts(Size, 0);
> + uint64_t MaxCount = 0;
> + for (unsigned s = 0; s < Size; s++) {
> + const PGOUseEdge *E = BBCountInfo.OutEdges[s];
> + const BasicBlock *SrcBB = E->SrcBB;
> + const BasicBlock *DestBB = E->DestBB;
> + if (DestBB == 0)
> + continue;
> + unsigned SuccNum = GetSuccessorNumber(SrcBB, DestBB);
> + uint64_t EdgeCount = E->CountValue;
> + if (EdgeCount > MaxCount)
> + MaxCount = EdgeCount;
> + EdgeCounts[SuccNum] = EdgeCount;
> + }
> + assert(MaxCount > 0 && "Bad max count");
> + uint64_t Scale = calculateCountScale(MaxCount);
> + SmallVector<unsigned, 4> Weights;
> + for (const auto &ECI : EdgeCounts)
> + Weights.push_back(scaleBranchCount(ECI, Scale));
> +
> + TI->setMetadata(llvm::LLVMContext::MD_prof,
> + MDB.createBranchWeights(Weights));
> + DEBUG(dbgs() << "Weight is: "; for (const auto &W
> + : Weights) dbgs()
> + << W << " ";
> + dbgs() << "\n";);
This is formatted really strangely. It's possible that clang-format
doesn't know how to deal with this code-in-DEBUG() stuff and you just
need to format this manually (or file a bug on clang-format ;)
> + }
> +}
> +} // end anonymous namespace
> +
> +bool PGOInstrumentationGen::runOnModule(Module &M) {
> + for (auto &F : M) {
> + if (F.isDeclaration())
> + continue;
> + BranchProbabilityInfo *BPI =
> + &(getAnalysis<BranchProbabilityInfoWrapperPass>(F).getBPI());
> + BlockFrequencyInfo *BFI =
> + &(getAnalysis<BlockFrequencyInfoWrapperPass>(F).getBFI());
> + instrumentOneFunc(F, &M, BPI, BFI);
> + }
> + return true;
> +}
> +
> +static void setPGOCountOnFunc(PGOUseFunc &Func,
> + IndexedInstrProfReader *PGOReader) {
> + if (Func.readCounters(PGOReader)) {
> + Func.populateCounters();
> + Func.setBranchWeights();
> + }
> +}
> +
> +bool PGOInstrumentationUse::runOnModule(Module &M) {
> + DEBUG(dbgs() << "Read in profile counters: ");
> + auto &Ctx = M.getContext();
> + // Read the counter array from file.
> + auto ReaderOrErr = IndexedInstrProfReader::create(ProfileFileName);
> + if (std::error_code EC = ReaderOrErr.getError()) {
> + Ctx.diagnose(
> + DiagnosticInfoPGOProfile(ProfileFileName.data(), EC.message()));
> + return false;
> + }
> +
> + PGOReader = std::move(ReaderOrErr.get());
> + if (!PGOReader) {
> + Ctx.diagnose(DiagnosticInfoPGOProfile(ProfileFileName.data(),
> + "Cannot get PGOReader"));
> + return false;
> + }
> +
> + for (auto &F : M) {
> + if (F.isDeclaration())
> + continue;
> + BranchProbabilityInfo *BPI =
> + &(getAnalysis<BranchProbabilityInfoWrapperPass>(F).getBPI());
> + BlockFrequencyInfo *BFI =
> + &(getAnalysis<BlockFrequencyInfoWrapperPass>(F).getBFI());
> + PGOUseFunc Func(F, &M, BPI, BFI);
> + setPGOCountOnFunc(Func, PGOReader.get());
> + }
> + return true;
> +}
> Index: lib/Transforms/Instrumentation/LLVMBuild.txt
> ===================================================================
> --- lib/Transforms/Instrumentation/LLVMBuild.txt
> +++ lib/Transforms/Instrumentation/LLVMBuild.txt
> @@ -19,4 +19,4 @@
> type = Library
> name = Instrumentation
> parent = Transforms
> -required_libraries = Analysis Core MC Support TransformUtils
> +required_libraries = Analysis Core MC Support TransformUtils ProfileData
> Index: lib/Transforms/Instrumentation/Instrumentation.cpp
> ===================================================================
> --- lib/Transforms/Instrumentation/Instrumentation.cpp
> +++ lib/Transforms/Instrumentation/Instrumentation.cpp
> @@ -60,6 +60,8 @@
> initializeAddressSanitizerModulePass(Registry);
> initializeBoundsCheckingPass(Registry);
> initializeGCOVProfilerPass(Registry);
> + initializePGOInstrumentationGenPass(Registry);
> + initializePGOInstrumentationUsePass(Registry);
> initializeInstrProfilingPass(Registry);
> initializeMemorySanitizerPass(Registry);
> initializeThreadSanitizerPass(Registry);
> Index: lib/Transforms/Instrumentation/CMakeLists.txt
> ===================================================================
> --- lib/Transforms/Instrumentation/CMakeLists.txt
> +++ lib/Transforms/Instrumentation/CMakeLists.txt
> @@ -6,6 +6,7 @@
> MemorySanitizer.cpp
> Instrumentation.cpp
> InstrProfiling.cpp
> + PGOInstrumentation.cpp
> SafeStack.cpp
> SanitizerCoverage.cpp
> ThreadSanitizer.cpp
> Index: lib/Transforms/Instrumentation/CFGMST.h
> ===================================================================
> --- /dev/null
> +++ lib/Transforms/Instrumentation/CFGMST.h
> @@ -0,0 +1,211 @@
> +//===-- CFGMST.h - Minimum Spanning Tree for CFG -------*- C++ -*-===//
> +//
> +// The LLVM Compiler Infrastructure
> +//
> +// This file is distributed under the University of Illinois Open Source
> +// License. See LICENSE.TXT for details.
> +//
> +//===----------------------------------------------------------------------===//
> +//
> +// This file implements Union-find algorithm to compute Minimum Spanning Tree
> +// for a given CFG.
> +//
> +//===----------------------------------------------------------------------===//
> +
> +#include "llvm/ADT/DenseMap.h"
> +#include "llvm/ADT/STLExtras.h"
> +#include "llvm/Analysis/BlockFrequencyInfo.h"
> +#include "llvm/Analysis/BranchProbabilityInfo.h"
> +#include "llvm/Analysis/CFG.h"
> +#include "llvm/Support/BranchProbability.h"
> +#include "llvm/Support/Debug.h"
> +#include "llvm/Support/raw_ostream.h"
> +#include "llvm/Transforms/Utils/BasicBlockUtils.h"
> +#include <string>
> +#include <utility>
> +#include <vector>
> +
> +namespace llvm {
> +
> +#define DEBUG_TYPE "cfgmst"
> +
> +template <class Edge, class BBInfo> class CFGMST {
> +public:
> + Function &F;
> +
> + // Store all the edges in CFG. It may contain some stale edges
> + // when Removed is set.
Might as well use doc comments here.
> + std::vector<std::unique_ptr<Edge>> AllEdges;
> +
> + // This map records the auxiliary information for each BB.
> + DenseMap<const BasicBlock *, std::unique_ptr<BBInfo>> BBInfos;
> +
> + // Find the root group of the G and compress the path from G to the root.
> + BBInfo *findAndCompressGroup(BBInfo *G) {
> + if (G->Group != G)
> + G->Group = findAndCompressGroup(static_cast<BBInfo *>(G->Group));
> + return static_cast<BBInfo *>(G->Group);
> + }
> +
> + // Union BB1 and BB2 into the same group and return true.
> + // Returns false if BB1 and BB2 are already in the same group.
> + bool unionGroups(const BasicBlock *BB1, const BasicBlock *BB2) {
> + BBInfo *BB1G = findAndCompressGroup(&getBBInfo(BB1));
> + BBInfo *BB2G = findAndCompressGroup(&getBBInfo(BB2));
> +
> + if (BB1G == BB2G)
> + return false;
> +
> + // Make the smaller rank tree a direct child or the root of high rank tree.
> + if (BB1G->Rank < BB2G->Rank)
> + BB1G->Group = BB2G;
> + else {
> + BB2G->Group = BB1G;
> + // If the ranks are the same, increment root of one tree by one.
> + if (BB1G->Rank == BB2G->Rank)
> + BB1G->Rank++;
> + }
> + return true;
> + }
> +
> + // Give BB, return the auxiliary information.
> + BBInfo &getBBInfo(const BasicBlock *BB) const {
> + auto It = BBInfos.find(BB);
> + assert(It->second.get() != nullptr);
> + return *It->second.get();
> + }
> +
> + // Traverse the CFG using a stack. Find all the edges and assign the weight.
> + // Edges with large weight will be put into MST first so they are less likely
> + // to be instrumented.
> + void buildEdges() {
> + DEBUG(dbgs() << "Build Edge on " << F.getName() << "\n");
> +
> + const BasicBlock *BB = &(F.getEntryBlock());
> + // Add a fake edge to the entry.
> + addEdge(nullptr, BB, BFI->getEntryFreq());
> +
> + // Special handling for single BB functions.
> + if (succ_empty(BB)) {
> + addEdge(BB, nullptr, BFI->getEntryFreq());
> + return;
> + }
> +
> + static const uint32_t CriticalEdgeMultiplier = 1000;
> +
> + for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
> + TerminatorInst *TI = BB->getTerminator();
> + uint64_t BBWeight = BFI->getBlockFreq(&*BB).getFrequency();
> + uint64_t Weight;
> + if (int successors = TI->getNumSuccessors()) {
> + for (int i = 0; i != successors; ++i) {
> + BasicBlock *TargetBB = TI->getSuccessor(i);
> + bool Critical = isCriticalEdge(TI, i);
> + uint64_t scaleFactor = BBWeight;
> + if (Critical) {
> + if (scaleFactor < UINT64_MAX / CriticalEdgeMultiplier)
> + scaleFactor *= CriticalEdgeMultiplier;
> + else
> + scaleFactor = UINT64_MAX;
> + }
> + Weight = BPI->getEdgeProbability(&*BB, TargetBB).scale(scaleFactor);
> + addEdge(&*BB, TargetBB, Weight).IsCritical = Critical;
> + DEBUG(dbgs() << " Edge: from " << BB->getName() << " to "
> + << TargetBB->getName() << " w=" << Weight << "\n");
> + }
> + } else {
> + addEdge(&*BB, nullptr, BBWeight);
> + DEBUG(dbgs() << " Edge: from " << BB->getName() << " to exit"
> + << " w = " << BBWeight << "\n");
> + }
> + }
> + }
> +
> + // Sort CFG edges based on its weight.
> + void sortEdgesByWeight() {
> + std::stable_sort(
> + AllEdges.begin(), AllEdges.end(),
> + [](const std::unique_ptr<Edge> &Edge1, const std::unique_ptr<Edge> &Edge2) {
> + return Edge1->Weight > Edge2->Weight;
> + });
> + }
> +
> + // Traverse all the edges and compute the Minimum Weight Spanning Tree
> + // using union-find algorithm.
> + void computeMinimumSpanningTree() {
> + // First, put all the critical edge with landing-pad as the Dest to MST.
> + // This works around the insufficient support of critical edges split
> + // when destination BB is a landing pad.
> + for (auto &Ei : AllEdges) {
> + if (Ei->Removed)
> + continue;
> + if (Ei->IsCritical) {
> + if (Ei->DestBB && Ei->DestBB->isLandingPad()) {
> + if (unionGroups(Ei->SrcBB, Ei->DestBB))
> + Ei->InMST = true;
> + }
> + }
> + }
> +
> + for (auto &Ei : AllEdges) {
> + if (Ei->Removed)
> + continue;
> + if (unionGroups(Ei->SrcBB, Ei->DestBB))
> + Ei->InMST = true;
> + }
> + }
> +
> + // Dump the Debug information about the instrumentation.
> + void dumpEdges(raw_ostream &OS, const StringRef Message = StringRef()) const {
> + if (!Message.empty())
> + OS << Message << "\n";
> + OS << " Number of Basic Blocks: " << BBInfos.size() << "\n";
> + for (auto &BI : BBInfos) {
> + const BasicBlock *BB = BI.first;
> + OS << " BB: " << (BB == nullptr ? "FakeNode" : BB->getName()) << " "
> + << BI.second->infoString() << "\n";
> + }
> +
> + OS << " Number of Edges: " << AllEdges.size()
> + << " (*: Instrument, C: CriticalEdge, -: Removed)\n";
> + uint32_t Count = 0;
> + for (auto &EI : AllEdges) {
> + OS << " Edge " << Count++ << ": " << getBBInfo(EI->SrcBB).Index << "-->"
> + << getBBInfo(EI->DestBB).Index << EI->infoString() << "\n";
> + };
> + }
> +
> + // Add an edge to AllEdges with weight W.
> + Edge &addEdge(const BasicBlock *Src, const BasicBlock *Dest, uint64_t W) {
> + uint32_t Index = BBInfos.size();
> + auto Iter = BBInfos.end();
> + bool Inserted;
> + std::tie(Iter, Inserted) = BBInfos.insert(std::make_pair(Src, nullptr));
> + if (Inserted) {
> + // Newly inserted, update the real info.
> + Iter->second = std::move(llvm::make_unique<BBInfo>(Index));
> + Index++;
> + }
> + std::tie(Iter, Inserted) = BBInfos.insert(std::make_pair(Dest, nullptr));
> + if (Inserted)
> + // Newly inserted, update the real info.
> + Iter->second = std::move(llvm::make_unique<BBInfo>(Index));
> + AllEdges.emplace_back(new Edge(Src, Dest, W));
> + return *AllEdges.back();
> + }
> +
> + BranchProbabilityInfo *BPI;
> + BlockFrequencyInfo *BFI;
> +
> +public:
> + CFGMST(Function &Func, BranchProbabilityInfo *BPI_ = nullptr,
> + BlockFrequencyInfo *BFI_ = nullptr)
> + : F(Func), BPI(BPI_), BFI(BFI_) {
> + buildEdges();
> + sortEdgesByWeight();
> + computeMinimumSpanningTree();
> + }
> +};
> +
> +#undef DEBUG_TYPE // "cfgmst"
> +} // end namespace llvm
> Index: lib/IR/DiagnosticInfo.cpp
> ===================================================================
> --- lib/IR/DiagnosticInfo.cpp
> +++ lib/IR/DiagnosticInfo.cpp
> @@ -132,6 +132,12 @@
> DP << getMsg();
> }
>
> +void DiagnosticInfoPGOProfile::print(DiagnosticPrinter &DP) const {
> + if (getFileName())
> + DP << getFileName() << ": ";
> + DP << getMsg();
> +}
> +
> bool DiagnosticInfoOptimizationBase::isLocationAvailable() const {
> return getDebugLoc();
> }
> Index: include/llvm/Transforms/Instrumentation.h
> ===================================================================
> --- include/llvm/Transforms/Instrumentation.h
> +++ include/llvm/Transforms/Instrumentation.h
> @@ -79,6 +79,11 @@
> ModulePass *createGCOVProfilerPass(const GCOVOptions &Options =
> GCOVOptions::getDefault());
>
> +// PGO Instrumention
> +ModulePass *createPGOInstrumentationGenPass();
> +ModulePass *
> +createPGOInstrumentationUsePass(StringRef Filename = StringRef(""));
> +
> /// Options for the frontend instrumentation based profiling pass.
> struct InstrProfOptions {
> InstrProfOptions() : NoRedZone(false) {}
> @@ -149,6 +154,24 @@
> /// protect against stack-based overflow vulnerabilities.
> FunctionPass *createSafeStackPass(const TargetMachine *TM = nullptr);
>
> +/// \brief Calculate what to divide by to scale counts.
> +///
> +/// Given the maximum count, calculate a divisor that will scale all the
> +/// weights to strictly less than UINT32_MAX.
> +static inline uint64_t calculateCountScale(uint64_t MaxCount) {
> + return MaxCount < UINT32_MAX ? 1 : MaxCount / UINT32_MAX + 1;
> +}
> +
> +/// \brief Scale an individual branch count.
> +///
> +/// Scale a 64-bit weight down to 32-bits using \c Scale.
> +///
> +static inline uint32_t scaleBranchCount(uint64_t Count, uint64_t Scale) {
> + uint64_t Scaled = Count / Scale;
> + assert(Scaled <= UINT32_MAX && "overflow 32-bits");
> + return Scaled;
> +}
> +
> } // End llvm namespace
>
> #endif
> Index: include/llvm/LinkAllPasses.h
> ===================================================================
> --- include/llvm/LinkAllPasses.h
> +++ include/llvm/LinkAllPasses.h
> @@ -85,6 +85,8 @@
> (void) llvm::createDomOnlyViewerPass();
> (void) llvm::createDomViewerPass();
> (void) llvm::createGCOVProfilerPass();
> + (void) llvm::createPGOInstrumentationGenPass();
> + (void) llvm::createPGOInstrumentationUsePass();
> (void) llvm::createInstrProfilingPass();
> (void) llvm::createFunctionInliningPass();
> (void) llvm::createAlwaysInlinerPass();
> Index: include/llvm/InitializePasses.h
> ===================================================================
> --- include/llvm/InitializePasses.h
> +++ include/llvm/InitializePasses.h
> @@ -117,6 +117,8 @@
> void initializeExpandPostRAPass(PassRegistry&);
> void initializeAAResultsWrapperPassPass(PassRegistry &);
> void initializeGCOVProfilerPass(PassRegistry&);
> +void initializePGOInstrumentationGenPass(PassRegistry&);
> +void initializePGOInstrumentationUsePass(PassRegistry&);
> void initializeInstrProfilingPass(PassRegistry&);
> void initializeAddressSanitizerPass(PassRegistry&);
> void initializeAddressSanitizerModulePass(PassRegistry&);
> Index: include/llvm/IR/DiagnosticInfo.h
> ===================================================================
> --- include/llvm/IR/DiagnosticInfo.h
> +++ include/llvm/IR/DiagnosticInfo.h
> @@ -60,6 +60,7 @@
> DK_OptimizationRemarkAnalysisAliasing,
> DK_OptimizationFailure,
> DK_MIRParser,
> + DK_PGOProfile,
> DK_FirstPluginKind
> };
>
> @@ -250,6 +251,31 @@
> const Twine &Msg;
> };
>
> +/// Diagnostic information for the PGO profiler.
> +class DiagnosticInfoPGOProfile : public DiagnosticInfo {
> +public:
> + DiagnosticInfoPGOProfile(const char *FileName, const Twine &Msg,
> + DiagnosticSeverity Severity = DS_Error)
> + : DiagnosticInfo(DK_PGOProfile, Severity), FileName(FileName), Msg(Msg) {}
> +
> + /// \see DiagnosticInfo::print.
> + void print(DiagnosticPrinter &DP) const override;
> +
> + static bool classof(const DiagnosticInfo *DI) {
> + return DI->getKind() == DK_PGOProfile;
> + }
> +
> + const char *getFileName() const { return FileName; }
> + const Twine &getMsg() const { return Msg; }
> +
> +private:
> + /// Name of the input file associated with this diagnostic.
> + const char *FileName;
> +
> + /// Message to report.
> + const Twine &Msg;
> +};
> +
> /// Common features for diagnostics dealing with optimization remarks.
> class DiagnosticInfoOptimizationBase : public DiagnosticInfo {
> public:
More information about the llvm-commits
mailing list