[OpenMP] Naturally align internal global variables in the OpenMPIRBuilder

[jhuber6]

Summary

We use this helper to make several internal global variables during
codegen. currently we do not specify any alignment which allows the
alignment to be set incorrectly after some changes in how alignment was
handled. This patch explicitly aligns these variables to the natural
alignment as specified by the data layout

Fixes https://github.com/llvm/llvm-project/issues/62668

[tianshilei1992]

LG

[gchatelet]

You beat me to it :)
Thx for the fix.

[gchatelet]

quick question, did you try to build the openmp runtime as well to check if the tests still pass?

[jhuber6]

In D150461#4338498, @gchatelet wrote:

quick question, did you try to build the openmp runtime as well to check if the tests still pass?

I'm an OpenMP developer so I always have it built : ). Yes they still pass.

[gchatelet]

In D150461#4338499, @jhuber6 wrote:

In D150461#4338498, @gchatelet wrote:

quick question, did you try to build the openmp runtime as well to check if the tests still pass?

I'm an OpenMP developer so I always have it built : ). Yes they still pass.

👍

clang/test/OpenMP/distribute_parallel_for_reduction_task_codegen.cpp

 // CHECK1-LABEL: define {{[^@]+}}@.red_init..1
 // CHECK1-SAME: (ptr noalias noundef [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
-// CHECK1-NEXT:    [[TMP3:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP3:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i64, ptr [[TMP3]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr i8, ptr [[TMP2]], i64 [[TMP4]]
 // CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq ptr [[TMP2]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYINIT_ISEMPTY]], label [[OMP_ARRAYINIT_DONE:%.*]], label [[OMP_ARRAYINIT_BODY:%.*]]
 // CHECK1:       omp.arrayinit.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[TMP2]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
 // CHECK1-NEXT:    store i8 0, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr i8, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYINIT_DONE]], label [[OMP_ARRAYINIT_BODY]]
 // CHECK1:       omp.arrayinit.done:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_comb..2
 // CHECK1-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP2:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr i8, ptr [[TMP4]], i64 [[TMP3]]
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq ptr [[TMP4]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 // CHECK1:       omp.arraycpy.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi ptr [ [[TMP5]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]

clang/test/OpenMP/for_reduction_task_codegen.cpp

 #pragma omp parallel
 #pragma omp for reduction(task, +: argc, argv[0:10][0:argc])
   for (long long i = 0; i < 10; ++i) {
 #pragma omp task in_reduction(+: argc, argv[0:10][0:argc])
     ;
   }
 }
 
-
-
-// Init firstprivate copy of argc
-
-// Init firstprivate copy of argv[0:10][0:argc]
-
-// Register task reduction.
-
-
-
-
-
-
-
-
-
 #endif
 // CHECK1-LABEL: define {{[^@]+}}@main
 // CHECK1-SAME: (i32 noundef [[ARGC:%.*]], ptr noundef [[ARGV:%.*]]) #[[ATTR0:[0-9]+]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[ARGC_ADDR:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[ARGV_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store i32 [[ARGC]], ptr [[ARGC_ADDR]], align 4
 // CHECK1-NEXT:    store ptr [[ARGV]], ptr [[ARGV_ADDR]], align 8

 // CHECK1-LABEL: define {{[^@]+}}@.red_init..1
 // CHECK1-SAME: (ptr noalias noundef [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
-// CHECK1-NEXT:    [[TMP3:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP3:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i64, ptr [[TMP3]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr i8, ptr [[TMP2]], i64 [[TMP4]]
 // CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq ptr [[TMP2]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYINIT_ISEMPTY]], label [[OMP_ARRAYINIT_DONE:%.*]], label [[OMP_ARRAYINIT_BODY:%.*]]
 // CHECK1:       omp.arrayinit.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[TMP2]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
 // CHECK1-NEXT:    store i8 0, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr i8, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYINIT_DONE]], label [[OMP_ARRAYINIT_BODY]]
 // CHECK1:       omp.arrayinit.done:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_comb..2
 // CHECK1-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP2:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr i8, ptr [[TMP4]], i64 [[TMP3]]
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq ptr [[TMP4]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 // CHECK1:       omp.arraycpy.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi ptr [ [[TMP5]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]

clang/test/OpenMP/parallel_for_reduction_task_codegen.cpp

 // CHECK1-LABEL: define {{[^@]+}}@.red_init..1
 // CHECK1-SAME: (ptr noalias noundef [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
-// CHECK1-NEXT:    [[TMP3:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP3:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i64, ptr [[TMP3]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr i8, ptr [[TMP2]], i64 [[TMP4]]
 // CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq ptr [[TMP2]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYINIT_ISEMPTY]], label [[OMP_ARRAYINIT_DONE:%.*]], label [[OMP_ARRAYINIT_BODY:%.*]]
 // CHECK1:       omp.arrayinit.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[TMP2]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
 // CHECK1-NEXT:    store i8 0, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr i8, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYINIT_DONE]], label [[OMP_ARRAYINIT_BODY]]
 // CHECK1:       omp.arrayinit.done:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_comb..2
 // CHECK1-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP2:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr i8, ptr [[TMP4]], i64 [[TMP3]]
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq ptr [[TMP4]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 // CHECK1:       omp.arraycpy.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi ptr [ [[TMP5]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]

clang/test/OpenMP/parallel_master_reduction_task_codegen.cpp

 // CHECK1-LABEL: define {{[^@]+}}@.red_init..1
 // CHECK1-SAME: (ptr noalias noundef [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
-// CHECK1-NEXT:    [[TMP3:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP3:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i64, ptr [[TMP3]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr i8, ptr [[TMP2]], i64 [[TMP4]]
 // CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq ptr [[TMP2]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYINIT_ISEMPTY]], label [[OMP_ARRAYINIT_DONE:%.*]], label [[OMP_ARRAYINIT_BODY:%.*]]
 // CHECK1:       omp.arrayinit.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[TMP2]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
 // CHECK1-NEXT:    store i8 0, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr i8, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYINIT_DONE]], label [[OMP_ARRAYINIT_BODY]]
 // CHECK1:       omp.arrayinit.done:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_comb..2
 // CHECK1-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP2:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr i8, ptr [[TMP4]], i64 [[TMP3]]
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq ptr [[TMP4]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 // CHECK1:       omp.arraycpy.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi ptr [ [[TMP5]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]

clang/test/OpenMP/parallel_reduction_task_codegen.cpp

 // CHECK1-LABEL: define {{[^@]+}}@.red_init..1
 // CHECK1-SAME: (ptr noalias noundef [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
-// CHECK1-NEXT:    [[TMP3:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP3:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i64, ptr [[TMP3]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr i8, ptr [[TMP2]], i64 [[TMP4]]
 // CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq ptr [[TMP2]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYINIT_ISEMPTY]], label [[OMP_ARRAYINIT_DONE:%.*]], label [[OMP_ARRAYINIT_BODY:%.*]]
 // CHECK1:       omp.arrayinit.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[TMP2]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
 // CHECK1-NEXT:    store i8 0, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr i8, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYINIT_DONE]], label [[OMP_ARRAYINIT_BODY]]
 // CHECK1:       omp.arrayinit.done:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_comb..2
 // CHECK1-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP2:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr i8, ptr [[TMP4]], i64 [[TMP3]]
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq ptr [[TMP4]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 // CHECK1:       omp.arraycpy.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi ptr [ [[TMP5]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]

clang/test/OpenMP/parallel_sections_reduction_task_codegen.cpp

 // CHECK1-LABEL: define {{[^@]+}}@.red_init..1
 // CHECK1-SAME: (ptr noalias noundef [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
-// CHECK1-NEXT:    [[TMP3:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP3:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i64, ptr [[TMP3]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr i8, ptr [[TMP2]], i64 [[TMP4]]
 // CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq ptr [[TMP2]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYINIT_ISEMPTY]], label [[OMP_ARRAYINIT_DONE:%.*]], label [[OMP_ARRAYINIT_BODY:%.*]]
 // CHECK1:       omp.arrayinit.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[TMP2]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
 // CHECK1-NEXT:    store i8 0, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr i8, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYINIT_DONE]], label [[OMP_ARRAYINIT_BODY]]
 // CHECK1:       omp.arrayinit.done:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_comb..2
 // CHECK1-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP2:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr i8, ptr [[TMP4]], i64 [[TMP3]]
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq ptr [[TMP4]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 // CHECK1:       omp.arraycpy.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi ptr [ [[TMP5]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]

clang/test/OpenMP/reduction_implicit_map.cpp

 // CHECK1-LABEL: define {{[^@]+}}@.red_init.
 // CHECK1-SAME: (ptr noalias noundef [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR4]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
-// CHECK1-NEXT:    [[TMP3:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP3:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i64, ptr [[TMP3]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr double, ptr [[TMP2]], i64 [[TMP4]]
 // CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq ptr [[TMP2]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYINIT_ISEMPTY]], label [[OMP_ARRAYINIT_DONE:%.*]], label [[OMP_ARRAYINIT_BODY:%.*]]
 // CHECK1:       omp.arrayinit.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[TMP2]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
 // CHECK1-NEXT:    store double 0.000000e+00, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 8
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr double, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYINIT_DONE]], label [[OMP_ARRAYINIT_BODY]]
 // CHECK1:       omp.arrayinit.done:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_comb.
 // CHECK1-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR4]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP2:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr double, ptr [[TMP4]], i64 [[TMP3]]
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq ptr [[TMP4]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE2:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 // CHECK1:       omp.arraycpy.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi ptr [ [[TMP5]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]

clang/test/OpenMP/sections_reduction_task_codegen.cpp

 // CHECK1-LABEL: define {{[^@]+}}@.red_init..1
 // CHECK1-SAME: (ptr noalias noundef [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
-// CHECK1-NEXT:    [[TMP3:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP3:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i64, ptr [[TMP3]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr i8, ptr [[TMP2]], i64 [[TMP4]]
 // CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq ptr [[TMP2]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYINIT_ISEMPTY]], label [[OMP_ARRAYINIT_DONE:%.*]], label [[OMP_ARRAYINIT_BODY:%.*]]
 // CHECK1:       omp.arrayinit.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[TMP2]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
 // CHECK1-NEXT:    store i8 0, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr i8, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYINIT_DONE]], label [[OMP_ARRAYINIT_BODY]]
 // CHECK1:       omp.arrayinit.done:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_comb..2
 // CHECK1-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP2:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr i8, ptr [[TMP4]], i64 [[TMP3]]
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq ptr [[TMP4]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 // CHECK1:       omp.arraycpy.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi ptr [ [[TMP5]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]

clang/test/OpenMP/target_parallel_for_reduction_task_codegen.cpp

 // CHECK1-LABEL: define {{[^@]+}}@.red_init..1
 // CHECK1-SAME: (ptr noalias noundef [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
-// CHECK1-NEXT:    [[TMP3:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP3:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i64, ptr [[TMP3]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr i8, ptr [[TMP2]], i64 [[TMP4]]
 // CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq ptr [[TMP2]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYINIT_ISEMPTY]], label [[OMP_ARRAYINIT_DONE:%.*]], label [[OMP_ARRAYINIT_BODY:%.*]]
 // CHECK1:       omp.arrayinit.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[TMP2]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
 // CHECK1-NEXT:    store i8 0, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr i8, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYINIT_DONE]], label [[OMP_ARRAYINIT_BODY]]
 // CHECK1:       omp.arrayinit.done:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_comb..2
 // CHECK1-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP2:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr i8, ptr [[TMP4]], i64 [[TMP3]]
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq ptr [[TMP4]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 // CHECK1:       omp.arraycpy.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi ptr [ [[TMP5]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]

clang/test/OpenMP/target_parallel_reduction_task_codegen.cpp

 // CHECK1-LABEL: define {{[^@]+}}@.red_init..1
 // CHECK1-SAME: (ptr noalias noundef [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
-// CHECK1-NEXT:    [[TMP3:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP3:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i64, ptr [[TMP3]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr i8, ptr [[TMP2]], i64 [[TMP4]]
 // CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq ptr [[TMP2]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYINIT_ISEMPTY]], label [[OMP_ARRAYINIT_DONE:%.*]], label [[OMP_ARRAYINIT_BODY:%.*]]
 // CHECK1:       omp.arrayinit.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[TMP2]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
 // CHECK1-NEXT:    store i8 0, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr i8, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYINIT_DONE]], label [[OMP_ARRAYINIT_BODY]]
 // CHECK1:       omp.arrayinit.done:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_comb..2
 // CHECK1-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP2:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr i8, ptr [[TMP4]], i64 [[TMP3]]
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq ptr [[TMP4]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 // CHECK1:       omp.arraycpy.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi ptr [ [[TMP5]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]

clang/test/OpenMP/target_teams_distribute_parallel_for_reduction_task_codegen.cpp

 // CHECK1-LABEL: define {{[^@]+}}@.red_init..1
 // CHECK1-SAME: (ptr noalias noundef [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
-// CHECK1-NEXT:    [[TMP3:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP3:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i64, ptr [[TMP3]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr i8, ptr [[TMP2]], i64 [[TMP4]]
 // CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq ptr [[TMP2]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYINIT_ISEMPTY]], label [[OMP_ARRAYINIT_DONE:%.*]], label [[OMP_ARRAYINIT_BODY:%.*]]
 // CHECK1:       omp.arrayinit.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[TMP2]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
 // CHECK1-NEXT:    store i8 0, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr i8, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYINIT_DONE]], label [[OMP_ARRAYINIT_BODY]]
 // CHECK1:       omp.arrayinit.done:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_comb..2
 // CHECK1-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP2:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr i8, ptr [[TMP4]], i64 [[TMP3]]
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq ptr [[TMP4]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 // CHECK1:       omp.arraycpy.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi ptr [ [[TMP5]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]

 // CHECK1-LABEL: define {{[^@]+}}@.red_init..5
 // CHECK1-SAME: (ptr noalias noundef [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
-// CHECK1-NEXT:    [[TMP3:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP3:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i64, ptr [[TMP3]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr i8, ptr [[TMP2]], i64 [[TMP4]]
 // CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq ptr [[TMP2]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYINIT_ISEMPTY]], label [[OMP_ARRAYINIT_DONE:%.*]], label [[OMP_ARRAYINIT_BODY:%.*]]
 // CHECK1:       omp.arrayinit.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[TMP2]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
 // CHECK1-NEXT:    store i8 0, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr i8, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYINIT_DONE]], label [[OMP_ARRAYINIT_BODY]]
 // CHECK1:       omp.arrayinit.done:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_comb..6
 // CHECK1-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP2:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr i8, ptr [[TMP4]], i64 [[TMP3]]
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq ptr [[TMP4]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 // CHECK1:       omp.arraycpy.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi ptr [ [[TMP5]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]

clang/test/OpenMP/taskloop_reduction_codegen.cpp

 
 // CHECK: define internal void [[OMP_COMB1:@.+]](ptr noalias noundef %0, ptr noalias noundef %1)
 // CHECK: fadd float %
 
 // CHECK: define internal void [[OMP_INIT1]](ptr noalias noundef %0, ptr noalias noundef %1)
 // CHECK: call void @llvm.memcpy.p0.p0.i64(
 
 // CHECK: define internal void @[[RED_FINI2]](ptr noundef %0)
-// CHECK: [[RED_SIZE1_ADDR:%.+]] = call ptr @llvm.threadlocal.address.p0(ptr [[RED_SIZE1]]
+// CHECK: [[RED_SIZE1_ADDR:%.+]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 [[RED_SIZE1]]
 // CHECK: load i64, ptr [[RED_SIZE1_ADDR]]
 // CHECK: call void @
 // CHECK: ret void
 
 // CHECK: define internal void @[[RED_COMB2]](ptr noundef %0, ptr noundef %1)
-// CHECK: [[RED_SIZE1_ADDR2:%.+]] = call ptr @llvm.threadlocal.address.p0(ptr [[RED_SIZE1]]
+// CHECK: [[RED_SIZE1_ADDR2:%.+]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 [[RED_SIZE1]]
 // CHECK: load i64, ptr [[RED_SIZE1_ADDR2]]
 // CHECK: call void [[OMP_COMB1]](
 // CHECK: ret void
 
 // CHECK: define internal void @[[RED_INIT3]](ptr noalias noundef %{{.+}}, ptr noalias noundef %{{.+}})
 // CHECK: store float 0.000000e+00, ptr %
 // CHECK: ret void
 
 // CHECK: define internal void @[[RED_COMB3]](ptr noundef %0, ptr noundef %1)
 // CHECK: fadd float %
 // CHECK: store float %{{.+}}, ptr %
 // CHECK: ret void
 
 // CHECK: define internal void @[[RED_INIT4]](ptr noalias noundef %{{.+}}, ptr noalias noundef %{{.+}})
-// CHECK: [[RED_SIZE2_ADDR:%.+]] = call ptr @llvm.threadlocal.address.p0(ptr [[RED_SIZE2]]
+// CHECK: [[RED_SIZE2_ADDR:%.+]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 [[RED_SIZE2]]
 // CHECK: load i64, ptr [[RED_SIZE2_ADDR]]
 // CHECK: store float 0.000000e+00, ptr %
 // CHECK: ret void
 
 // CHECK: define internal void @[[RED_COMB4]](ptr noundef %0, ptr noundef %1)
-// CHECK: [[RED_SIZE2_ADDR2:%.+]] = call ptr @llvm.threadlocal.address.p0(ptr [[RED_SIZE2]]
+// CHECK: [[RED_SIZE2_ADDR2:%.+]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 [[RED_SIZE2]]
 // CHECK: load i64, ptr [[RED_SIZE2_ADDR2]]
 // CHECK: fadd float %
 // CHECK: store float %{{.+}}, ptr %
 // CHECK: ret void
 
 // CHECK: call ptr @__kmpc_task_reduction_get_th_data(
 // CHECK: call ptr @__kmpc_task_reduction_get_th_data(
 // CHECK: call ptr @__kmpc_task_reduction_get_th_data(

clang/test/OpenMP/teams_distribute_parallel_for_reduction_task_codegen.cpp

 // CHECK1-LABEL: define {{[^@]+}}@.red_init..1
 // CHECK1-SAME: (ptr noalias noundef [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
-// CHECK1-NEXT:    [[TMP3:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP3:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i64, ptr [[TMP3]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr i8, ptr [[TMP2]], i64 [[TMP4]]
 // CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq ptr [[TMP2]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYINIT_ISEMPTY]], label [[OMP_ARRAYINIT_DONE:%.*]], label [[OMP_ARRAYINIT_BODY:%.*]]
 // CHECK1:       omp.arrayinit.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[TMP2]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
 // CHECK1-NEXT:    store i8 0, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr i8, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYINIT_DONE]], label [[OMP_ARRAYINIT_BODY]]
 // CHECK1:       omp.arrayinit.done:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_comb..2
 // CHECK1-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP2:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr i8, ptr [[TMP4]], i64 [[TMP3]]
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq ptr [[TMP4]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 // CHECK1:       omp.arraycpy.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi ptr [ [[TMP5]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]

 // CHECK1-LABEL: define {{[^@]+}}@.red_init..5
 // CHECK1-SAME: (ptr noalias noundef [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
-// CHECK1-NEXT:    [[TMP3:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP3:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i64, ptr [[TMP3]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr i8, ptr [[TMP2]], i64 [[TMP4]]
 // CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq ptr [[TMP2]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYINIT_ISEMPTY]], label [[OMP_ARRAYINIT_DONE:%.*]], label [[OMP_ARRAYINIT_BODY:%.*]]
 // CHECK1:       omp.arrayinit.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[TMP2]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
 // CHECK1-NEXT:    store i8 0, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr i8, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP5]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYINIT_DONE]], label [[OMP_ARRAYINIT_BODY]]
 // CHECK1:       omp.arrayinit.done:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_comb..6
 // CHECK1-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR3]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = call ptr @llvm.threadlocal.address.p0(ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP2:%.*]] = call align 8 ptr @llvm.threadlocal.address.p0(ptr align 8 @{{reduction_size[.].+[.]}})
 // CHECK1-NEXT:    [[TMP3:%.*]] = load i64, ptr [[TMP2]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr i8, ptr [[TMP4]], i64 [[TMP3]]
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq ptr [[TMP4]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 // CHECK1:       omp.arraycpy.body:
 // CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi ptr [ [[TMP5]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]

llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp

if (Elem.second) {
     assert(cast<PointerType>(Elem.second->getType())
                ->isOpaqueOrPointeeTypeMatches(Ty) &&
            "OMP internal variable has different type than requested");
   } else {
     // TODO: investigate the appropriate linkage type used for the global
     // variable for possibly changing that to internal or private, or maybe
     // create different versions of the function for different OMP internal
     // variables.
-    Elem.second = new GlobalVariable(
+    auto *GV = new GlobalVariable(
         M, Ty, /*IsConstant=*/false, GlobalValue::CommonLinkage,
         Constant::getNullValue(Ty), Elem.first(),
         /*InsertBefore=*/nullptr, GlobalValue::NotThreadLocal, AddressSpace);
+    GV->setAlignment(M.getDataLayout().getABITypeAlign(Ty));
+    Elem.second = GV;
   }
 
   return cast<GlobalVariable>(&*Elem.second);
 }
 
 Value *OpenMPIRBuilder::getOMPCriticalRegionLock(StringRef CriticalName) {
   std::string Prefix = Twine("gomp_critical_user_", CriticalName).str();
   std::string Name = getNameWithSeparators({Prefix, "var"}, ".", ".");