LLVM Optimization pass breakdown example

a.cc

inline int square(int num) {
    return num*num*num*num*num*num*num*num;
}

int foo(int n, int num) {
    int res = 0;
    for(int i = 0; i < n; i++)
        res += square(num);
    return res;
}

b0.ll

; ModuleID = 'a.ll'
source_filename = "a.cc"
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

$_Z6squarei = comdat any

; Function Attrs: uwtable
define i32 @_Z3fooii(i32, i32) #0 {
  %3 = alloca i32, align 4
  %4 = alloca i32, align 4
  %5 = alloca i32, align 4
  %6 = alloca i32, align 4
  store i32 %0, i32* %3, align 4, !tbaa !2
  store i32 %1, i32* %4, align 4, !tbaa !2
  %7 = bitcast i32* %5 to i8*
  call void @llvm.lifetime.start.p0i8(i64 4, i8* %7) #3
  store i32 0, i32* %5, align 4, !tbaa !2
  %8 = bitcast i32* %6 to i8*
  call void @llvm.lifetime.start.p0i8(i64 4, i8* %8) #3
  store i32 0, i32* %6, align 4, !tbaa !2
  br label %9

; <label>:9:                                      ; preds = %20, %2
  %10 = load i32, i32* %6, align 4, !tbaa !2
  %11 = load i32, i32* %3, align 4, !tbaa !2
  %12 = icmp slt i32 %10, %11
  br i1 %12, label %15, label %13

; <label>:13:                                     ; preds = %9
  %14 = bitcast i32* %6 to i8*
  call void @llvm.lifetime.end.p0i8(i64 4, i8* %14) #3
  br label %23

; <label>:15:                                     ; preds = %9
  %16 = load i32, i32* %4, align 4, !tbaa !2
  %17 = call i32 @_Z6squarei(i32 %16)
  %18 = load i32, i32* %5, align 4, !tbaa !2
  %19 = add nsw i32 %18, %17
  store i32 %19, i32* %5, align 4, !tbaa !2
  br label %20

; <label>:20:                                     ; preds = %15
  %21 = load i32, i32* %6, align 4, !tbaa !2
  %22 = add nsw i32 %21, 1
  store i32 %22, i32* %6, align 4, !tbaa !2
  br label %9

; <label>:23:                                     ; preds = %13
  %24 = load i32, i32* %5, align 4, !tbaa !2
  %25 = bitcast i32* %5 to i8*
  call void @llvm.lifetime.end.p0i8(i64 4, i8* %25) #3
  ret i32 %24
}

; Function Attrs: argmemonly nounwind
declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture) #1

; Function Attrs: inlinehint nounwind uwtable
define linkonce_odr i32 @_Z6squarei(i32) #2 comdat {
  %2 = alloca i32, align 4
  store i32 %0, i32* %2, align 4, !tbaa !2
  %3 = load i32, i32* %2, align 4, !tbaa !2
  %4 = load i32, i32* %2, align 4, !tbaa !2
  %5 = mul nsw i32 %3, %4
  %6 = load i32, i32* %2, align 4, !tbaa !2
  %7 = mul nsw i32 %5, %6
  %8 = load i32, i32* %2, align 4, !tbaa !2
  %9 = mul nsw i32 %7, %8
  %10 = load i32, i32* %2, align 4, !tbaa !2
  %11 = mul nsw i32 %9, %10
  %12 = load i32, i32* %2, align 4, !tbaa !2
  %13 = mul nsw i32 %11, %12
  %14 = load i32, i32* %2, align 4, !tbaa !2
  %15 = mul nsw i32 %13, %14
  %16 = load i32, i32* %2, align 4, !tbaa !2
  %17 = mul nsw i32 %15, %16
  ret i32 %17
}

; Function Attrs: argmemonly nounwind
declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture) #1

attributes #0 = { uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
attributes #1 = { argmemonly nounwind }
attributes #2 = { inlinehint nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
attributes #3 = { nounwind }

!llvm.module.flags = !{!0}
!llvm.ident = !{!1}

!0 = !{i32 1, !"wchar_size", i32 4}
!1 = !{!"clang version 5.0.2 (tags/RELEASE_502/final)"}
!2 = !{!3, !3, i64 0}
!3 = !{!"int", !4, i64 0}
!4 = !{!"omnipotent char", !5, i64 0}
!5 = !{!"Simple C++ TBAA"}

b1.ll

; ModuleID = 'a.ll'
source_filename = "a.cc"
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

$_Z6squarei = comdat any

; Function Attrs: uwtable
define i32 @_Z3fooii(i32, i32) #0 {
  br label %3

; <label>:3:                                      ; preds = %9, %2
  %.01 = phi i32 [ 0, %2 ], [ %8, %9 ]
  %.0 = phi i32 [ 0, %2 ], [ %10, %9 ]
  %4 = icmp slt i32 %.0, %0
  br i1 %4, label %6, label %5

; <label>:5:                                      ; preds = %3
  br label %11

; <label>:6:                                      ; preds = %3
  %7 = call i32 @_Z6squarei(i32 %1)
  %8 = add nsw i32 %.01, %7
  br label %9

; <label>:9:                                      ; preds = %6
  %10 = add nsw i32 %.0, 1
  br label %3

; <label>:11:                                     ; preds = %5
  ret i32 %.01
}

; Function Attrs: argmemonly nounwind
declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture) #1

; Function Attrs: inlinehint nounwind uwtable
define linkonce_odr i32 @_Z6squarei(i32) #2 comdat {
  %2 = mul nsw i32 %0, %0
  %3 = mul nsw i32 %2, %0
  %4 = mul nsw i32 %3, %0
  %5 = mul nsw i32 %4, %0
  %6 = mul nsw i32 %5, %0
  %7 = mul nsw i32 %6, %0
  %8 = mul nsw i32 %7, %0
  ret i32 %8
}

; Function Attrs: argmemonly nounwind
declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture) #1

attributes #0 = { uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
attributes #1 = { argmemonly nounwind }
attributes #2 = { inlinehint nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }

!llvm.module.flags = !{!0}
!llvm.ident = !{!1}

!0 = !{i32 1, !"wchar_size", i32 4}
!1 = !{!"clang version 5.0.2 (tags/RELEASE_502/final)"}

b2.ll

; ModuleID = 'a.ll'
source_filename = "a.cc"
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

$_Z6squarei = comdat any

; Function Attrs: uwtable
define i32 @_Z3fooii(i32, i32) #0 {
  br label %3

; <label>:3:                                      ; preds = %9, %2
  %.01 = phi i32 [ 0, %2 ], [ %8, %9 ]
  %.0 = phi i32 [ 0, %2 ], [ %10, %9 ]
  %4 = icmp slt i32 %.0, %0
  br i1 %4, label %6, label %5

; <label>:5:                                      ; preds = %3
  br label %11

; <label>:6:                                      ; preds = %3
  %7 = call i32 @_Z6squarei(i32 %1)
  %8 = add nsw i32 %7, %.01
  br label %9

; <label>:9:                                      ; preds = %6
  %10 = add nsw i32 %.0, 1
  br label %3

; <label>:11:                                     ; preds = %5
  ret i32 %.01
}

; Function Attrs: argmemonly nounwind
declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture) #1

; Function Attrs: inlinehint nounwind uwtable
define linkonce_odr i32 @_Z6squarei(i32) #2 comdat {
  %2 = mul i32 %0, %0
  %3 = mul i32 %2, %2
  %4 = mul i32 %3, %3
  ret i32 %4
}

; Function Attrs: argmemonly nounwind
declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture) #1

attributes #0 = { uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
attributes #1 = { argmemonly nounwind }
attributes #2 = { inlinehint nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }

!llvm.module.flags = !{!0}
!llvm.ident = !{!1}

!0 = !{i32 1, !"wchar_size", i32 4}
!1 = !{!"clang version 5.0.2 (tags/RELEASE_502/final)"}

b3.ll

; ModuleID = 'a.ll'
source_filename = "a.cc"
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

; Function Attrs: uwtable
define i32 @_Z3fooii(i32, i32) #0 {
  br label %3

; <label>:3:                                      ; preds = %11, %2
  %.01 = phi i32 [ 0, %2 ], [ %10, %11 ]
  %.0 = phi i32 [ 0, %2 ], [ %12, %11 ]
  %4 = icmp slt i32 %.0, %0
  br i1 %4, label %6, label %5

; <label>:5:                                      ; preds = %3
  br label %13

; <label>:6:                                      ; preds = %3
  %7 = mul i32 %1, %1
  %8 = mul i32 %7, %7
  %9 = mul i32 %8, %8
  %10 = add nsw i32 %9, %.01
  br label %11

; <label>:11:                                     ; preds = %6
  %12 = add nsw i32 %.0, 1
  br label %3

; <label>:13:                                     ; preds = %5
  ret i32 %.01
}

; Function Attrs: argmemonly nounwind
declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture) #1

; Function Attrs: argmemonly nounwind
declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture) #1

attributes #0 = { uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
attributes #1 = { argmemonly nounwind }

!llvm.module.flags = !{!0}
!llvm.ident = !{!1}

!0 = !{i32 1, !"wchar_size", i32 4}
!1 = !{!"clang version 5.0.2 (tags/RELEASE_502/final)"}

b4.ll

; ModuleID = 'a.ll'
source_filename = "a.cc"
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

; Function Attrs: uwtable
define i32 @_Z3fooii(i32, i32) #0 {
  %3 = mul i32 %1, %1
  %4 = mul i32 %3, %3
  %5 = mul i32 %4, %4
  br label %6

; <label>:6:                                      ; preds = %11, %2
  %.01 = phi i32 [ 0, %2 ], [ %10, %11 ]
  %.0 = phi i32 [ 0, %2 ], [ %12, %11 ]
  %7 = icmp slt i32 %.0, %0
  br i1 %7, label %9, label %8

; <label>:8:                                      ; preds = %6
  %.01.lcssa = phi i32 [ %.01, %6 ]
  br label %13

; <label>:9:                                      ; preds = %6
  %10 = add nsw i32 %5, %.01
  br label %11

; <label>:11:                                     ; preds = %9
  %12 = add nsw i32 %.0, 1
  br label %6

; <label>:13:                                     ; preds = %8
  ret i32 %.01.lcssa
}

; Function Attrs: argmemonly nounwind
declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture) #1

; Function Attrs: argmemonly nounwind
declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture) #1

attributes #0 = { uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
attributes #1 = { argmemonly nounwind }

!llvm.module.flags = !{!0}
!llvm.ident = !{!1}

!0 = !{i32 1, !"wchar_size", i32 4}
!1 = !{!"clang version 5.0.2 (tags/RELEASE_502/final)"}

b5.ll

; ModuleID = 'a.ll'
source_filename = "a.cc"
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

; Function Attrs: uwtable
define i32 @_Z3fooii(i32, i32) #0 {
  %3 = mul i32 %1, %1
  %4 = mul i32 %3, %3
  %5 = mul i32 %4, %4
  %6 = icmp sgt i32 %0, 0
  %smax = select i1 %6, i32 %0, i32 0
  br label %7

; <label>:7:                                      ; preds = %2
  %8 = mul i32 %5, %smax
  br label %9

; <label>:9:                                      ; preds = %7
  ret i32 %8
}

; Function Attrs: argmemonly nounwind
declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture) #1

; Function Attrs: argmemonly nounwind
declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture) #1

attributes #0 = { uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
attributes #1 = { argmemonly nounwind }

!llvm.module.flags = !{!0}
!llvm.ident = !{!1}

!0 = !{i32 1, !"wchar_size", i32 4}
!1 = !{!"clang version 5.0.2 (tags/RELEASE_502/final)"}

b6.ll

; ModuleID = 'a.ll'
source_filename = "a.cc"
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

; Function Attrs: uwtable
define i32 @_Z3fooii(i32, i32) #0 {
  %3 = mul i32 %1, %1
  %4 = mul i32 %3, %3
  %5 = icmp sgt i32 %0, 0
  %smax = select i1 %5, i32 %0, i32 0
  %6 = mul i32 %4, %smax
  %7 = mul i32 %6, %4
  ret i32 %7
}

; Function Attrs: argmemonly nounwind
declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture) #1

; Function Attrs: argmemonly nounwind
declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture) #1

attributes #0 = { uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
attributes #1 = { argmemonly nounwind }

!llvm.module.flags = !{!0}
!llvm.ident = !{!1}

!0 = !{i32 1, !"wchar_size", i32 4}
!1 = !{!"clang version 5.0.2 (tags/RELEASE_502/final)"}

doit.sh

#!/usr/bin/env bash

# Adding "-O1 -Xclang -disable-llvm-passes" is disables optimziations like "-O0",
# but doing "-O0" also adds a optnone attribute to each function.
# This attribute would tell opt not to optimize each function, which would break this example.
clang a.cc -O1 -Xclang -disable-llvm-passes -emit-llvm -S -o a.ll

opt -S a.ll -o b0.ll
opt -mem2reg -S a.ll -o b1.ll
opt -mem2reg -reassociate -S a.ll -o b2.ll
opt -mem2reg -reassociate -inline -S a.ll -o b3.ll
opt -mem2reg -reassociate -inline -licm -S a.ll -o b4.ll
opt -mem2reg -reassociate -inline -licm -indvars -loop-deletion -S a.ll -o b5.ll
opt -mem2reg -reassociate -inline -licm -indvars -loop-deletion -simplifycfg -S a.ll -o b6.ll