[SLP] Recalculate dependencies for potential control dependencies if cleared

If the control dependecies are cleared after calcellation of the
copyables, need to reclculate them unconditionally.

Fixes #153754 #153676

Author: alexey-bataev

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

@@ -20975,11 +20975,11 @@ BoUpSLP::BlockScheduling::tryScheduleBundle(ArrayRef<Value *> VL, BoUpSLP *SLP,
         continue;
       }
       ScheduledBundles.find(I)->getSecond().pop_back();
-      if (!ControlDependentMembers.empty()) {
-        ScheduleBundle Invalid = ScheduleBundle::invalid();
-        calculateDependencies(Invalid, /*InsertInReadyList=*/false, SLP,
-                              ControlDependentMembers);
-      }
+    }
+    if (!ControlDependentMembers.empty()) {
+      ScheduleBundle Invalid = ScheduleBundle::invalid();
+      calculateDependencies(Invalid, /*InsertInReadyList=*/false, SLP,
+                            ControlDependentMembers);
     }
     return std::nullopt;
   }

llvm/test/Transforms/SLPVectorizer/X86/schedule-control-deps-after-copyable.ll

@@ -0,0 +1,44 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S --passes=slp-vectorizer -mtriple=x86_64-unknown-linux-gnu < %s | FileCheck %s
+
+define void @test(i32 %e, i32 %0) {
+; CHECK-LABEL: define void @test(
+; CHECK-SAME: i32 [[E:%.*]], i32 [[TMP0:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[H:%.*]] = alloca [4 x i32], align 16
+; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP0]], [[E]]
+; CHECK-NEXT:    [[ADD2:%.*]] = add i32 [[ADD]], 1
+; CHECK-NEXT:    [[ADD4:%.*]] = add i32 [[ADD]], [[ADD2]]
+; CHECK-NEXT:    store i32 [[ADD4]], ptr [[H]], align 16
+; CHECK-NEXT:    [[ARRAYINIT_ELEMENT:%.*]] = getelementptr i8, ptr [[H]], i64 4
+; CHECK-NEXT:    [[ADD5:%.*]] = add i32 [[E]], 1
+; CHECK-NEXT:    [[ADD6:%.*]] = add i32 [[ADD5]], [[ADD2]]
+; CHECK-NEXT:    store i32 [[ADD6]], ptr [[ARRAYINIT_ELEMENT]], align 4
+; CHECK-NEXT:    [[ARRAYINIT_ELEMENT7:%.*]] = getelementptr i8, ptr [[H]], i64 8
+; CHECK-NEXT:    store i32 [[ADD6]], ptr [[ARRAYINIT_ELEMENT7]], align 8
+; CHECK-NEXT:    [[ARRAYINIT_ELEMENT10:%.*]] = getelementptr i8, ptr [[H]], i64 12
+; CHECK-NEXT:    [[CALL12:%.*]] = tail call i32 @fn1(i32 [[ADD]])
+; CHECK-NEXT:    store i32 [[CALL12]], ptr [[ARRAYINIT_ELEMENT10]], align 4
+; CHECK-NEXT:    store volatile i32 0, ptr [[H]], align 4
+; CHECK-NEXT:    ret void
+;
+entry:
+  %h = alloca [4 x i32], align 16
+  %add = add i32 %0, %e
+  %add2 = add i32 %add, 1
+  %add4 = add i32 %add, %add2
+  store i32 %add4, ptr %h, align 16
+  %arrayinit.element = getelementptr i8, ptr %h, i64 4
+  %add5 = add i32 %e, 1
+  %add6 = add i32 %add5, %add2
+  store i32 %add6, ptr %arrayinit.element, align 4
+  %arrayinit.element7 = getelementptr i8, ptr %h, i64 8
+  store i32 %add6, ptr %arrayinit.element7, align 8
+  %arrayinit.element10 = getelementptr i8, ptr %h, i64 12
+  %call12 = tail call i32 @fn1(i32 %add)
+  store i32 %call12, ptr %arrayinit.element10, align 4
+  store volatile i32 0, ptr %h, align 4
+  ret void
+}
+
+declare i32 @fn1(i32)