Profile: set branch weight metadata with data generated from profiling.

This patch implements ProfileDataLoader which loads profile data generated by
-insert-edge-profiling and updates branch weight metadata accordingly.

Patch by Alastair Murray.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@162799 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 466 insertions, 0 deletions

diff --git a/test/Analysis/Profiling/load-branch-weights-ifs.ll b/test/Analysis/Profiling/load-branch-weights-ifs.ll
new file mode 100644
index 0000000000..b5ee2f3466
--- /dev/null
+++ b/test/Analysis/Profiling/load-branch-weights-ifs.ll
@@ -0,0 +1,119 @@
+; RUN: opt -insert-edge-profiling -o %t1 < %s
+; RUN: rm -f %t1.prof_data
+; RUN: lli -load %llvmshlibdir/libprofile_rt%shlibext %t1 \
+; RUN:     -llvmprof-output %t1.prof_data
+; RUN: opt -profile-file %t1.prof_data -profile-metadata-loader -S -o - < %s \
+; RUN:     | FileCheck %s
+; RUN: rm -f %t1.prof_data
+
+;; func_mod - Branch taken 6 times in 7.
+define i32 @func_mod(i32 %N) nounwind uwtable {
+entry:
+  %retval = alloca i32, align 4
+  %N.addr = alloca i32, align 4
+  store i32 %N, i32* %N.addr, align 4
+  %0 = load i32* %N.addr, align 4
+  %rem = srem i32 %0, 7
+  %tobool = icmp ne i32 %rem, 0
+  br i1 %tobool, label %if.then, label %if.else
+; CHECK: br i1 %tobool, label %if.then, label %if.else, !prof !0
+
+if.then:
+  store i32 1, i32* %retval
+  br label %return
+
+if.else:
+  store i32 0, i32* %retval
+  br label %return
+
+return:
+  %1 = load i32* %retval
+  ret i32 %1
+}
+
+;; func_const_true - conditional branch which 100% taken probability.
+define i32 @func_const_true(i32 %N) nounwind uwtable {
+entry:
+  %retval = alloca i32, align 4
+  %N.addr = alloca i32, align 4
+  store i32 %N, i32* %N.addr, align 4
+  %0 = load i32* %N.addr, align 4
+  %cmp = icmp eq i32 %0, 1
+  br i1 %cmp, label %if.then, label %if.end
+; CHECK: br i1 %cmp, label %if.then, label %if.end, !prof !1
+
+if.then:
+  store i32 1, i32* %retval
+  br label %return
+
+if.end:
+  store i32 0, i32* %retval
+  br label %return
+
+return:
+  %1 = load i32* %retval
+  ret i32 %1
+}
+
+;; func_const_true - conditional branch which 100% not-taken probability.
+define i32 @func_const_false(i32 %N) nounwind uwtable {
+entry:
+  %retval = alloca i32, align 4
+  %N.addr = alloca i32, align 4
+  store i32 %N, i32* %N.addr, align 4
+  %0 = load i32* %N.addr, align 4
+  %cmp = icmp eq i32 %0, 1
+  br i1 %cmp, label %if.then, label %if.end
+; CHECK: br i1 %cmp, label %if.then, label %if.end, !prof !2
+
+if.then:
+  store i32 1, i32* %retval
+  br label %return
+
+if.end:
+  store i32 0, i32* %retval
+  br label %return
+
+return:
+  %1 = load i32* %retval
+  ret i32 %1
+}
+
+define i32 @main(i32 %argc, i8** %argv) nounwind uwtable {
+entry:
+  %retval = alloca i32, align 4
+  %argc.addr = alloca i32, align 4
+  %argv.addr = alloca i8**, align 8
+  %loop = alloca i32, align 4
+  store i32 0, i32* %retval
+  store i32 0, i32* %loop, align 4
+  br label %for.cond
+
+for.cond:
+  %0 = load i32* %loop, align 4
+  %cmp = icmp slt i32 %0, 7000
+  br i1 %cmp, label %for.body, label %for.end
+; CHECK: br i1 %cmp, label %for.body, label %for.end, !prof !3
+
+for.body:
+  %1 = load i32* %loop, align 4
+  %call = call i32 @func_mod(i32 %1)
+  br label %for.inc
+
+for.inc:
+  %2 = load i32* %loop, align 4
+  %inc = add nsw i32 %2, 1
+  store i32 %inc, i32* %loop, align 4
+  br label %for.cond
+
+for.end:
+  %call1 = call i32 @func_const_true(i32 1)
+  %call2 = call i32 @func_const_false(i32 0)
+  ret i32 0
+}
+
+; CHECK: !0 = metadata !{metadata !"branch_weights", i32 6000, i32 1000}
+; CHECK: !1 = metadata !{metadata !"branch_weights", i32 1, i32 0}
+; CHECK: !2 = metadata !{metadata !"branch_weights", i32 0, i32 1}
+; CHECK: !3 = metadata !{metadata !"branch_weights", i32 7000, i32 1}
+; CHECK-NOT: !4
diff --git a/test/Analysis/Profiling/load-branch-weights-loops.ll b/test/Analysis/Profiling/load-branch-weights-loops.ll
new file mode 100644
index 0000000000..26c89b7c9b
--- /dev/null
+++ b/test/Analysis/Profiling/load-branch-weights-loops.ll
@@ -0,0 +1,185 @@
+; RUN: opt -insert-edge-profiling -o %t1 < %s
+; RUN: rm -f %t1.prof_data
+; RUN: lli -load %llvmshlibdir/libprofile_rt%shlibext %t1 \
+; RUN:     -llvmprof-output %t1.prof_data
+; RUN: opt -profile-file %t1.prof_data -profile-metadata-loader -S -o - < %s \
+; RUN:     | FileCheck %s
+; RUN: rm -f %t1.prof_data
+
+;; func_for - Test branch probabilities for a vanilla for loop.
+define i32 @func_for(i32 %N) nounwind uwtable {
+entry:
+  %N.addr = alloca i32, align 4
+  %ret = alloca i32, align 4
+  %loop = alloca i32, align 4
+  store i32 %N, i32* %N.addr, align 4
+  store i32 0, i32* %ret, align 4
+  store i32 0, i32* %loop, align 4
+  br label %for.cond
+
+for.cond:
+  %0 = load i32* %loop, align 4
+  %1 = load i32* %N.addr, align 4
+  %cmp = icmp slt i32 %0, %1
+  br i1 %cmp, label %for.body, label %for.end
+; CHECK: br i1 %cmp, label %for.body, label %for.end, !prof !0
+
+for.body:
+  %2 = load i32* %N.addr, align 4
+  %3 = load i32* %ret, align 4
+  %add = add nsw i32 %3, %2
+  store i32 %add, i32* %ret, align 4
+  br label %for.inc
+
+for.inc:
+  %4 = load i32* %loop, align 4
+  %inc = add nsw i32 %4, 1
+  store i32 %inc, i32* %loop, align 4
+  br label %for.cond
+
+for.end:
+  %5 = load i32* %ret, align 4
+  ret i32 %5
+}
+
+;; func_for_odd - Test branch probabilities for a for loop with a continue and
+;; a break.
+define i32 @func_for_odd(i32 %N) nounwind uwtable {
+entry:
+  %N.addr = alloca i32, align 4
+  %ret = alloca i32, align 4
+  %loop = alloca i32, align 4
+  store i32 %N, i32* %N.addr, align 4
+  store i32 0, i32* %ret, align 4
+  store i32 0, i32* %loop, align 4
+  br label %for.cond
+
+for.cond:
+  %0 = load i32* %loop, align 4
+  %1 = load i32* %N.addr, align 4
+  %cmp = icmp slt i32 %0, %1
+  br i1 %cmp, label %for.body, label %for.end
+; CHECK: br i1 %cmp, label %for.body, label %for.end, !prof !1
+
+for.body:
+  %2 = load i32* %loop, align 4
+  %rem = srem i32 %2, 10
+  %tobool = icmp ne i32 %rem, 0
+  br i1 %tobool, label %if.then, label %if.end
+; CHECK: br i1 %tobool, label %if.then, label %if.end, !prof !2
+
+if.then:
+  br label %for.inc
+
+if.end:
+  %3 = load i32* %loop, align 4
+  %cmp1 = icmp eq i32 %3, 500
+  br i1 %cmp1, label %if.then2, label %if.end3
+; CHECK: br i1 %cmp1, label %if.then2, label %if.end3, !prof !3
+
+if.then2:
+  br label %for.end
+
+if.end3:
+  %4 = load i32* %N.addr, align 4
+  %5 = load i32* %ret, align 4
+  %add = add nsw i32 %5, %4
+  store i32 %add, i32* %ret, align 4
+  br label %for.inc
+
+for.inc:
+  %6 = load i32* %loop, align 4
+  %inc = add nsw i32 %6, 1
+  store i32 %inc, i32* %loop, align 4
+  br label %for.cond
+
+for.end:
+  %7 = load i32* %ret, align 4
+  ret i32 %7
+}
+
+;; func_while - Test branch probability in a vanilla while loop.
+define i32 @func_while(i32 %N) nounwind uwtable {
+entry:
+  %N.addr = alloca i32, align 4
+  %ret = alloca i32, align 4
+  %loop = alloca i32, align 4
+  store i32 %N, i32* %N.addr, align 4
+  store i32 0, i32* %ret, align 4
+  store i32 0, i32* %loop, align 4
+  br label %while.cond
+
+while.cond:
+  %0 = load i32* %loop, align 4
+  %1 = load i32* %N.addr, align 4
+  %cmp = icmp slt i32 %0, %1
+  br i1 %cmp, label %while.body, label %while.end
+; CHECK: br i1 %cmp, label %while.body, label %while.end, !prof !0
+
+while.body:
+  %2 = load i32* %N.addr, align 4
+  %3 = load i32* %ret, align 4
+  %add = add nsw i32 %3, %2
+  store i32 %add, i32* %ret, align 4
+  %4 = load i32* %loop, align 4
+  %inc = add nsw i32 %4, 1
+  store i32 %inc, i32* %loop, align 4
+  br label %while.cond
+
+while.end:
+  %5 = load i32* %ret, align 4
+  ret i32 %5
+}
+
+;; func_while - Test branch probability in a vanilla do-while loop.
+define i32 @func_do_while(i32 %N) nounwind uwtable {
+entry:
+  %N.addr = alloca i32, align 4
+  %ret = alloca i32, align 4
+  %loop = alloca i32, align 4
+  store i32 %N, i32* %N.addr, align 4
+  store i32 0, i32* %ret, align 4
+  store i32 0, i32* %loop, align 4
+  br label %do.body
+
+do.body:
+  %0 = load i32* %N.addr, align 4
+  %1 = load i32* %ret, align 4
+  %add = add nsw i32 %1, %0
+  store i32 %add, i32* %ret, align 4
+  %2 = load i32* %loop, align 4
+  %inc = add nsw i32 %2, 1
+  store i32 %inc, i32* %loop, align 4
+  br label %do.cond
+
+do.cond:
+  %3 = load i32* %loop, align 4
+  %4 = load i32* %N.addr, align 4
+  %cmp = icmp slt i32 %3, %4
+  br i1 %cmp, label %do.body, label %do.end
+; CHECK: br i1 %cmp, label %do.body, label %do.end, !prof !4
+
+do.end:
+  %5 = load i32* %ret, align 4
+  ret i32 %5
+}
+
+define i32 @main(i32 %argc, i8** %argv) nounwind uwtable {
+entry:
+  %retval = alloca i32, align 4
+  %argc.addr = alloca i32, align 4
+  %argv.addr = alloca i8**, align 8
+  store i32 0, i32* %retval
+  %call = call i32 @func_for(i32 1000)
+  %call1 = call i32 @func_for_odd(i32 1000)
+  %call2 = call i32 @func_while(i32 1000)
+  %call3 = call i32 @func_do_while(i32 1000)
+  ret i32 0
+}
+
+!0 = metadata !{metadata !"branch_weights", i32 1000, i32 1}
+!1 = metadata !{metadata !"branch_weights", i32 501, i32 0}
+!2 = metadata !{metadata !"branch_weights", i32 450, i32 51}
+!3 = metadata !{metadata !"branch_weights", i32 1, i32 50}
+!4 = metadata !{metadata !"branch_weights", i32 999, i32 1}
+; CHECK-NOT: !5
diff --git a/test/Analysis/Profiling/load-branch-weights-switches.ll b/test/Analysis/Profiling/load-branch-weights-switches.ll
new file mode 100644
index 0000000000..9efe6d0fe4
--- /dev/null
+++ b/test/Analysis/Profiling/load-branch-weights-switches.ll
@@ -0,0 +1,162 @@
+; RUN: opt -insert-edge-profiling -o %t1 < %s
+; RUN: rm -f %t1.prof_data
+; RUN: lli -load %llvmshlibdir/libprofile_rt%shlibext %t1 \
+; RUN:     -llvmprof-output %t1.prof_data
+; RUN: opt -profile-file %t1.prof_data -profile-metadata-loader -S -o - < %s \
+; RUN:     | FileCheck %s
+; RUN: rm -f %t1.prof_data
+
+;; func_switch - Test branch probabilities for a switch instruction with an
+;; even chance of taking each case (or no case).
+define i32 @func_switch(i32 %N) nounwind uwtable {
+entry:
+  %retval = alloca i32, align 4
+  %N.addr = alloca i32, align 4
+  store i32 %N, i32* %N.addr, align 4
+  %0 = load i32* %N.addr, align 4
+  %rem = srem i32 %0, 4
+  switch i32 %rem, label %sw.epilog [
+    i32 0, label %sw.bb
+    i32 1, label %sw.bb1
+    i32 2, label %sw.bb2
+  ]
+; CHECK: ], !prof !0
+
+sw.bb:
+  store i32 5, i32* %retval
+  br label %return
+
+sw.bb1:
+  store i32 6, i32* %retval
+  br label %return
+
+sw.bb2:
+  store i32 7, i32* %retval
+  br label %return
+
+sw.epilog:
+  store i32 8, i32* %retval
+  br label %return
+
+return:
+  %1 = load i32* %retval
+  ret i32 %1
+}
+
+;; func_switch_switch - Test branch probabilities in a switch-instruction that
+;; leads to further switch instructions.  The first-tier switch occludes some
+;; possibilities in the second-tier switches, leading to some branches having a
+;; 0 probability.
+define i32 @func_switch_switch(i32 %N) nounwind uwtable {
+entry:
+  %retval = alloca i32, align 4
+  %N.addr = alloca i32, align 4
+  store i32 %N, i32* %N.addr, align 4
+  %0 = load i32* %N.addr, align 4
+  %rem = srem i32 %0, 2
+  switch i32 %rem, label %sw.default11 [
+    i32 0, label %sw.bb
+    i32 1, label %sw.bb5
+  ]
+; CHECK: ], !prof !1
+
+sw.bb:
+  %1 = load i32* %N.addr, align 4
+  %rem1 = srem i32 %1, 4
+  switch i32 %rem1, label %sw.default [
+    i32 0, label %sw.bb2
+    i32 1, label %sw.bb3
+    i32 2, label %sw.bb4
+  ]
+; CHECK: ], !prof !2
+
+sw.bb2:
+  store i32 5, i32* %retval
+  br label %return
+
+sw.bb3:
+  store i32 6, i32* %retval
+  br label %return
+
+sw.bb4:
+  store i32 7, i32* %retval
+  br label %return
+
+sw.default:
+  store i32 8, i32* %retval
+  br label %return
+
+sw.bb5:
+  %2 = load i32* %N.addr, align 4
+  %rem6 = srem i32 %2, 4
+  switch i32 %rem6, label %sw.default10 [
+    i32 0, label %sw.bb7
+    i32 1, label %sw.bb8
+    i32 2, label %sw.bb9
+  ]
+; CHECK: ], !prof !3
+
+sw.bb7:
+  store i32 9, i32* %retval
+  br label %return
+
+sw.bb8:
+  store i32 10, i32* %retval
+  br label %return
+
+sw.bb9:
+  store i32 11, i32* %retval
+  br label %return
+
+sw.default10:
+  store i32 12, i32* %retval
+  br label %return
+
+sw.default11:
+  store i32 13, i32* %retval
+  br label %return
+
+return:
+  %3 = load i32* %retval
+  ret i32 %3
+}
+
+define i32 @main(i32 %argc, i8** %argv) nounwind uwtable {
+entry:
+  %retval = alloca i32, align 4
+  %argc.addr = alloca i32, align 4
+  %argv.addr = alloca i8**, align 8
+  %loop = alloca i32, align 4
+  store i32 0, i32* %retval
+  store i32 0, i32* %loop, align 4
+  br label %for.cond
+
+for.cond:
+  %0 = load i32* %loop, align 4
+  %cmp = icmp slt i32 %0, 4000
+  br i1 %cmp, label %for.body, label %for.end
+; CHECK: br i1 %cmp, label %for.body, label %for.end, !prof !4
+
+for.body:
+  %1 = load i32* %loop, align 4
+  %call = call i32 @func_switch(i32 %1)
+  %2 = load i32* %loop, align 4
+  %call1 = call i32 @func_switch_switch(i32 %2)
+  br label %for.inc
+
+for.inc:
+  %3 = load i32* %loop, align 4
+  %inc = add nsw i32 %3, 1
+  store i32 %inc, i32* %loop, align 4
+  br label %for.cond
+
+for.end:
+  ret i32 0
+}
+
+; CHECK: !0 = metadata !{metadata !"branch_weights", i32 1000, i32 1000, i32 1000, i32 1000}
+; CHECK: !1 = metadata !{metadata !"branch_weights", i32 0, i32 2000, i32 2000}
+; CHECK: !2 = metadata !{metadata !"branch_weights", i32 0, i32 1000, i32 0, i32 1000}
+; CHECK: !3 = metadata !{metadata !"branch_weights", i32 1000, i32 0, i32 1000, i32 0}
+; CHECK: !4 = metadata !{metadata !"branch_weights", i32 4000, i32 1}
+; CHECK-NOT: !5