[OpenMP][LLVM] Update alloca IP after PrivCB in OMPIRBUIlder (#93920)

Fixes a crash uncovered by
[pr77666.f90](https://github.com/llvm/llvm-test-suite/blob/main/Fortran/gfortran/regression/gomp/pr77666.f90)
in the test suite when delayed privatization is enabled by default.

In particular, whenever `PrivCB` (the callback responsible for
generating privatizaiton logic for an OMP variable) generates a
multi-block privatization region, the insertion point diverges: the BB
component of the IP can become a different BB from the parent block of
the instruction iterator component of the IP. This PR updates the IP to
make sure that the BB is the parent block of the instruction iterator.

Author: ergawy

clang/test/OpenMP/parallel_codegen.cpp

@@ -822,8 +822,8 @@ int main (int argc, char **argv) {
 // CHECK3-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TID_ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[TMP1]], ptr [[TID_ADDR_LOCAL]], align 4
 // CHECK3-NEXT:    [[TID:%.*]] = load i32, ptr [[TID_ADDR_LOCAL]], align 4
-// CHECK3-NEXT:    [[TMP2:%.*]] = load i64, ptr [[LOADGEP__RELOADED]], align 8
 // CHECK3-NEXT:    [[VAR:%.*]] = alloca ptr, align 8
+// CHECK3-NEXT:    [[TMP2:%.*]] = load i64, ptr [[LOADGEP__RELOADED]], align 8
 // CHECK3-NEXT:    br label [[OMP_PAR_REGION:%.*]]
 // CHECK3:       omp.par.region:
 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[LOADGEP_ARGC_ADDR]], align 8
@@ -966,8 +966,8 @@ int main (int argc, char **argv) {
 // CHECK4-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TID_ADDR]], align 4
 // CHECK4-NEXT:    store i32 [[TMP1]], ptr [[TID_ADDR_LOCAL]], align 4
 // CHECK4-NEXT:    [[TID:%.*]] = load i32, ptr [[TID_ADDR_LOCAL]], align 4
-// CHECK4-NEXT:    [[TMP2:%.*]] = load i64, ptr [[LOADGEP__RELOADED]], align 8
 // CHECK4-NEXT:    [[VAR:%.*]] = alloca ptr, align 8
+// CHECK4-NEXT:    [[TMP2:%.*]] = load i64, ptr [[LOADGEP__RELOADED]], align 8
 // CHECK4-NEXT:    br label [[OMP_PAR_REGION:%.*]]
 // CHECK4:       omp.par.region:
 // CHECK4-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[LOADGEP_ARGC_ADDR]], align 8, !dbg [[DBG58:![0-9]+]]

flang/test/Lower/OpenMP/delayed-privatization-lower-allocatable-to-llvm.f90

@@ -0,0 +1,23 @@
+! Tests the OMPIRBuilder can handle multiple privatization regions that contain
+! multiple BBs (for example, for allocatables).
+
+! RUN: %flang -S -emit-llvm -fopenmp -mmlir --openmp-enable-delayed-privatization \
+! RUN:   -o - %s 2>&1 | FileCheck %s
+
+subroutine foo(x)
+  integer, allocatable :: x, y
+!$omp parallel private(x, y)
+  x = y
+!$omp end parallel
+end
+
+! CHECK-LABEL: define void @foo_
+! CHECK:         ret void
+! CHECK-NEXT:  }
+
+! CHECK-LABEL: define internal void @foo_..omp_par
+! CHECK-DAG:     call ptr @malloc
+! CHECK-DAG:     call ptr @malloc
+! CHECK-DAG:     call void @free
+! CHECK-DAG:     call void @free
+! CHECK:       }

llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp

@@ -1583,6 +1583,10 @@ IRBuilder<>::InsertPoint OpenMPIRBuilder::createParallel(
     } else {
       Builder.restoreIP(
           PrivCB(InnerAllocaIP, Builder.saveIP(), V, *Inner, ReplacementValue));
+      InnerAllocaIP = {
+          InnerAllocaIP.getBlock(),
+          InnerAllocaIP.getBlock()->getTerminator()->getIterator()};
+
       assert(ReplacementValue &&
              "Expected copy/create callback to set replacement value!");
       if (ReplacementValue == &V)