[SLU][profcheck] create likely branch weights for guard->branch (llvm#164271)

The `llvm.experimental.guard` intrinsic is a `call`, so its metadata - if present - would be one value (as per `Verifier::visitProfMetadata`). That wouldn't be a correct `branch_weights` metadata. Likely, `GI->getMetadata(LLVMContext::MD_prof)` was always `nullptr`.

We can bias away from deopt instead.

Issue llvm#147390

Author: mtrofin

llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp

@@ -40,6 +40,7 @@
 #include "llvm/IR/Instruction.h"
 #include "llvm/IR/Instructions.h"
 #include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/MDBuilder.h"
 #include "llvm/IR/Module.h"
 #include "llvm/IR/PatternMatch.h"
 #include "llvm/IR/ProfDataUtils.h"
@@ -2831,9 +2832,14 @@ static BranchInst *turnGuardIntoBranch(IntrinsicInst *GI, Loop &L,
      MSSAU->getMemorySSA()->verifyMemorySSA();
 
   DomTreeUpdater DTU(DT, DomTreeUpdater::UpdateStrategy::Eager);
-  Instruction *DeoptBlockTerm =
-      SplitBlockAndInsertIfThen(GI->getArgOperand(0), GI, true,
-                                GI->getMetadata(LLVMContext::MD_prof), &DTU, &LI);
+  // llvm.experimental.guard doesn't have branch weights. We can assume,
+  // however, that the deopt path is unlikely.
+  Instruction *DeoptBlockTerm = SplitBlockAndInsertIfThen(
+      GI->getArgOperand(0), GI, true,
+      !ProfcheckDisableMetadataFixes && EstimateProfile
+          ? MDBuilder(GI->getContext()).createUnlikelyBranchWeights()
+          : nullptr,
+      &DTU, &LI);
   BranchInst *CheckBI = cast<BranchInst>(CheckBB->getTerminator());
   // SplitBlockAndInsertIfThen inserts control flow that branches to
   // DeoptBlockTerm if the condition is true.  We want the opposite.

llvm/test/Transforms/SimpleLoopUnswitch/guards.ll

@@ -1,15 +1,15 @@
-; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals all --version 5
 ; RUN: opt -passes='loop(simple-loop-unswitch<nontrivial>),verify<loops>' -simple-loop-unswitch-guards -S < %s | FileCheck %s
 ; RUN: opt -passes='simple-loop-unswitch<nontrivial>' -simple-loop-unswitch-guards -S < %s | FileCheck %s
 ; RUN: opt -passes='loop-mssa(simple-loop-unswitch<nontrivial>),verify<loops>' -simple-loop-unswitch-guards  -verify-memoryssa -verify-loop-info -S < %s | FileCheck %s
 
 declare void @llvm.experimental.guard(i1, ...)
 
-define void @test_simple_case(i1 %cond, i32 %N) {
+define void @test_simple_case(i1 %cond, i32 %N) !prof !0 {
 ; CHECK-LABEL: define void @test_simple_case(
-; CHECK-SAME: i1 [[COND:%.*]], i32 [[N:%.*]]) {
+; CHECK-SAME: i1 [[COND:%.*]], i32 [[N:%.*]]) !prof [[PROF0:![0-9]+]] {
 ; CHECK-NEXT:  [[ENTRY:.*:]]
-; CHECK-NEXT:    br i1 [[COND]], label %[[ENTRY_SPLIT_US:.*]], label %[[ENTRY_SPLIT:.*]]
+; CHECK-NEXT:    br i1 [[COND]], label %[[ENTRY_SPLIT_US:.*]], label %[[ENTRY_SPLIT:.*]], !prof [[PROF1:![0-9]+]]
 ; CHECK:       [[ENTRY_SPLIT_US]]:
 ; CHECK-NEXT:    br label %[[LOOP_US:.*]]
 ; CHECK:       [[LOOP_US]]:
@@ -50,9 +50,9 @@ define void @test_two_guards(i1 %cond1, i1 %cond2, i32 %N) {
 ; CHECK-LABEL: define void @test_two_guards(
 ; CHECK-SAME: i1 [[COND1:%.*]], i1 [[COND2:%.*]], i32 [[N:%.*]]) {
 ; CHECK-NEXT:  [[ENTRY:.*:]]
-; CHECK-NEXT:    br i1 [[COND1]], label %[[ENTRY_SPLIT_US:.*]], label %[[ENTRY_SPLIT:.*]]
+; CHECK-NEXT:    br i1 [[COND1]], label %[[ENTRY_SPLIT_US:.*]], label %[[ENTRY_SPLIT:.*]], !prof [[PROF1]]
 ; CHECK:       [[ENTRY_SPLIT_US]]:
-; CHECK-NEXT:    br i1 [[COND2]], label %[[ENTRY_SPLIT_US_SPLIT_US:.*]], label %[[ENTRY_SPLIT_US_SPLIT:.*]]
+; CHECK-NEXT:    br i1 [[COND2]], label %[[ENTRY_SPLIT_US_SPLIT_US:.*]], label %[[ENTRY_SPLIT_US_SPLIT:.*]], !prof [[PROF1]]
 ; CHECK:       [[ENTRY_SPLIT_US_SPLIT_US]]:
 ; CHECK-NEXT:    br label %[[LOOP_US_US:.*]]
 ; CHECK:       [[LOOP_US_US]]:
@@ -108,7 +108,7 @@ define void @test_conditional_guards(i1 %cond, i32 %N) {
 ; CHECK-SAME: i1 [[COND:%.*]], i32 [[N:%.*]]) {
 ; CHECK-NEXT:  [[ENTRY:.*:]]
 ; CHECK-NEXT:    [[COND_FR:%.*]] = freeze i1 [[COND]]
-; CHECK-NEXT:    br i1 [[COND_FR]], label %[[ENTRY_SPLIT_US:.*]], label %[[ENTRY_SPLIT:.*]]
+; CHECK-NEXT:    br i1 [[COND_FR]], label %[[ENTRY_SPLIT_US:.*]], label %[[ENTRY_SPLIT:.*]], !prof [[PROF1]]
 ; CHECK:       [[ENTRY_SPLIT_US]]:
 ; CHECK-NEXT:    br label %[[LOOP_US:.*]]
 ; CHECK:       [[LOOP_US]]:
@@ -171,7 +171,7 @@ define void @test_nested_loop(i1 %cond, i32 %N, i1 %arg) {
 ; CHECK-LABEL: define void @test_nested_loop(
 ; CHECK-SAME: i1 [[COND:%.*]], i32 [[N:%.*]], i1 [[ARG:%.*]]) {
 ; CHECK-NEXT:  [[ENTRY:.*:]]
-; CHECK-NEXT:    br i1 [[COND]], label %[[ENTRY_SPLIT:.*]], label %[[OUTER_LOOP_SPLIT:.*]]
+; CHECK-NEXT:    br i1 [[COND]], label %[[ENTRY_SPLIT:.*]], label %[[OUTER_LOOP_SPLIT:.*]], !prof [[PROF1]]
 ; CHECK:       [[ENTRY_SPLIT]]:
 ; CHECK-NEXT:    br i1 [[ARG]], label %[[ENTRY_SPLIT_SPLIT_US:.*]], label %[[ENTRY_SPLIT_SPLIT:.*]]
 ; CHECK:       [[ENTRY_SPLIT_SPLIT_US]]:
@@ -243,7 +243,7 @@ define void @test_sibling_loops(i1 %cond1, i1 %cond2, i32 %N) {
 ; CHECK-LABEL: define void @test_sibling_loops(
 ; CHECK-SAME: i1 [[COND1:%.*]], i1 [[COND2:%.*]], i32 [[N:%.*]]) {
 ; CHECK-NEXT:  [[ENTRY:.*:]]
-; CHECK-NEXT:    br i1 [[COND1]], label %[[ENTRY_SPLIT_US:.*]], label %[[ENTRY_SPLIT:.*]]
+; CHECK-NEXT:    br i1 [[COND1]], label %[[ENTRY_SPLIT_US:.*]], label %[[ENTRY_SPLIT:.*]], !prof [[PROF1]]
 ; CHECK:       [[ENTRY_SPLIT_US]]:
 ; CHECK-NEXT:    br label %[[LOOP1_US:.*]]
 ; CHECK:       [[LOOP1_US]]:
@@ -263,7 +263,7 @@ define void @test_sibling_loops(i1 %cond1, i1 %cond2, i32 %N) {
 ; CHECK-NEXT:    call void (i1, ...) @llvm.experimental.guard(i1 false) [ "deopt"() ]
 ; CHECK-NEXT:    unreachable
 ; CHECK:       [[BETWEEN]]:
-; CHECK-NEXT:    br i1 [[COND2]], label %[[BETWEEN_SPLIT_US2:.*]], label %[[BETWEEN_SPLIT:.*]]
+; CHECK-NEXT:    br i1 [[COND2]], label %[[BETWEEN_SPLIT_US2:.*]], label %[[BETWEEN_SPLIT:.*]], !prof [[PROF1]]
 ; CHECK:       [[BETWEEN_SPLIT_US2]]:
 ; CHECK-NEXT:    br label %[[LOOP2_US:.*]]
 ; CHECK:       [[LOOP2_US]]:
@@ -343,3 +343,9 @@ exit:
 
 declare void @may_throw(i32 %i)
 declare i32 @__CxxFrameHandler3(...)
+
+!0 = !{!"function_entry_count", i32 10}
+;.
+; CHECK: [[PROF0]] = !{!"function_entry_count", i32 10}
+; CHECK: [[PROF1]] = !{!"branch_weights", i32 1048575, i32 1}
+;.