[OpenMP] Restore backwards compatibility for libomptarget

[jhuber6]

Summary

The changes introduced in D87946 changed the API for libomptarget
functions. __kmpc_push_target_tripcount was a function in Clang 11.x
but was not given a backward-compatible interface accidentally. This
will make it backwards compatible, but will require Clang 12 to recompile.

[jdoerfert]

LG. We should have a test that references the entire interface so we realize if something breaks.

[cchen]

I think some existing Clang OpenMP codegen tests will break if they are testing against __kmpc_push_target_tripcount.

[jhuber6]

In D98358#2617152, @cchen wrote:

I think some existing Clang OpenMP codegen tests will break if they are testing against __kmpc_push_target_tripcount.

I'm working on fixing the tests, I also forgot to include it in the exports

[jhuber6]

Updating tests and adding the new function to the exports.

[cchen]

LGTM.

clang/lib/CodeGen/CGOpenMPRuntime.cpp

       llvm::Value *Args[] = {RTLoc, DeviceID, NumIterations};
       CGF.EmitRuntimeCall(
           OMPBuilder.getOrCreateRuntimeFunction(
-              CGM.getModule(), OMPRTL___kmpc_push_target_tripcount),
+              CGM.getModule(), OMPRTL___kmpc_push_target_tripcount_mapper),
           Args);
     }
   };

clang/test/OpenMP/target_teams_distribute_parallel_for_codegen.cpp

 
 #ifdef CK1
 
-// HCK_NO_TGT-NOT: @__kmpc_push_target_tripcount
+// HCK_NO_TGT-NOT: @__kmpc_push_target_tripcount_mapper
 
 // HCK1: define{{.*}} i32 @{{.+}}target_teams_fun{{.*}}(
 int target_teams_fun(int *g){

   // HCK1: [[N_PAR:%.+]] = load{{.+}}, {{.+}} [[N_CAST]],
   // HCK1: [[TE_PAR:%.+]] = load{{.+}}, {{.+}} [[TE_CAST]],
   // HCK1: [[TH_PAR:%.+]] = load{{.+}}, {{.+}} [[TH_CAST]],
-  // HCK1: call void @__kmpc_push_target_tripcount(%struct.ident_t* @{{.+}}, i64 -1, i64 %{{.+}})
+  // HCK1: call void @__kmpc_push_target_tripcount_mapper(%struct.ident_t* @{{.+}}, i64 -1, i64 %{{.+}})
   // HCK1: call i32 @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}}, i64 -1, i8* @{{[^,]+}}, i32 4, i8** %{{[^,]+}}, i8** %{{[^,]+}},
 
   // HCK1: call void @[[OFFL1:.+]](i{{32|64}} [[N_PAR]], {{.+}}, i{{32|64}} [[TE_PAR]], i{{32|64}} [[TH_PAR]])

clang/test/OpenMP/target_teams_distribute_parallel_for_if_codegen.cpp

 
 // CHECK-LABEL: define {{.*}}void @{{.+}}gtid_test
 void gtid_test() {
-// CHECK: call void @__kmpc_push_target_tripcount(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
+// CHECK: call void @__kmpc_push_target_tripcount_mapper(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
 // CHECK: call i{{[0-9]+}} @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}},
 // CHECK: call void [[OFFLOADING_FUN_0:@.+]](
-// CHECK: call void @__kmpc_push_target_tripcount(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
+// CHECK: call void @__kmpc_push_target_tripcount_mapper(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
 // CHECK: call i{{[0-9]+}} @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}},
 // CHECK: call void [[OFFLOADING_FUN_1:@.+]](
 #pragma omp target teams distribute parallel for

 
 // CHECK-LABEL: define {{.*}}i{{[0-9]+}} @main()
 int main() {
-// CHECK: call void @__kmpc_push_target_tripcount(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
+// CHECK: call void @__kmpc_push_target_tripcount_mapper(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
 // CHECK: call i{{[0-9]+}} @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}},
 // CHECK: call void [[OFFLOADING_FUN_0:@.+]](
-// CHECK-NOT: call void @__kmpc_push_target_tripcount(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
+// CHECK-NOT: call void @__kmpc_push_target_tripcount_mapper(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
 // CHECK: call void [[OFFLOADING_FUN_1:@.+]](
-// CHECK: call void @__kmpc_push_target_tripcount(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
+// CHECK: call void @__kmpc_push_target_tripcount_mapper(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
 // CHECK: call i{{[0-9]+}} @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}},
 // CHECK: call void [[OFFLOADING_FUN_2:@.+]](
 // CHECK: = call {{.*}}i{{.+}} @{{.+}}tmain

clang/test/OpenMP/target_teams_distribute_parallel_for_order_codegen.cpp

 
 // CHECK-LABEL: define {{.*}}void @{{.+}}gtid_test
 void gtid_test() {
-// CHECK: call void @__kmpc_push_target_tripcount(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
+// CHECK: call void @__kmpc_push_target_tripcount_mapper(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
 // CHECK: %0 = call i32 @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}}, i64 -1, i8* @{{.+}}, i32 0, i8** null, i8** null, i64* null, i64* null, i8** null, i8** null, i32 0, i32 0)
 // CHECK: call void [[TARGET_OUTLINE:@.+]]()
 // CHECK: ret void

clang/test/OpenMP/target_teams_distribute_parallel_for_simd_codegen.cpp

 // HCK1: [[N_PAR:%.+]] = load{{.+}}, {{.+}} [[N_CAST]],
 // HCK1: [[TE_PAR:%.+]] = load{{.+}}, {{.+}} [[TE_CAST]],
 // HCK1: [[TH_PAR:%.+]] = load{{.+}}, {{.+}} [[TH_CAST]],
-// HCK1: call void @__kmpc_push_target_tripcount(%struct.ident_t* @{{.+}}, i64 -1, i64 %{{.+}})
+// HCK1: call void @__kmpc_push_target_tripcount_mapper(%struct.ident_t* @{{.+}}, i64 -1, i64 %{{.+}})
 // HCK1: call i32 @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}}, i64 -1, i8* @{{[^,]+}}, i32 5, i8** %{{[^,]+}}, i8** %{{[^,]+}},
 
 // HCK1: call void @[[OFFL1:.+]](i{{32|64}} [[I_PAR]], i{{32|64}} [[N_PAR]], {{.+}}, i{{32|64}} [[TE_PAR]], i{{32|64}} [[TH_PAR]])

clang/test/OpenMP/target_teams_distribute_parallel_for_simd_if_codegen.cpp

 
 // CHECK-LABEL: define {{.*}}void @{{.+}}gtid_test
 void gtid_test() {
-// CHECK: call void @__kmpc_push_target_tripcount(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
+// CHECK: call void @__kmpc_push_target_tripcount_mapper(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
 // CHECK: call i{{[0-9]+}} @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}},
 // CHECK: call void [[OFFLOADING_FUN_0:@.+]](
-// CHECK: call void @__kmpc_push_target_tripcount(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
+// CHECK: call void @__kmpc_push_target_tripcount_mapper(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
 // CHECK: call i{{[0-9]+}} @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}},
 // CHECK: call void [[OFFLOADING_FUN_1:@.+]](
 #ifdef OMP5

 
 // CHECK-LABEL: define {{.*}}i{{[0-9]+}} @main()
 int main() {
-// CHECK: call void @__kmpc_push_target_tripcount(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
+// CHECK: call void @__kmpc_push_target_tripcount_mapper(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
 // CHECK: call i{{[0-9]+}} @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}},
 // CHECK: call void [[OFFLOADING_FUN_0:@.+]](
-// CHECK-NOT: call void @__kmpc_push_target_tripcount(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
+// CHECK-NOT: call void @__kmpc_push_target_tripcount_mapper(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
 // CHECK: call void [[OFFLOADING_FUN_1:@.+]](
-// CHECK: call void @__kmpc_push_target_tripcount(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
+// CHECK: call void @__kmpc_push_target_tripcount_mapper(%struct.ident_t* @{{.+}}, i64 -1, i64 100)
 // CHECK: call i{{[0-9]+}} @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}},
 // CHECK: call void [[OFFLOADING_FUN_2:@.+]](
 // CHECK: = call {{.*}}i{{.+}} @{{.+}}tmain

clang/test/OpenMP/teams_distribute_codegen.cpp

   // CK1: [[TE_PAR:%.+]] = load{{.+}}, {{.+}} [[TE_CAST]],
   // CK1: [[TH_PAR:%.+]] = load{{.+}}, {{.+}} [[TH_CAST]],
 
-  // CK1: call void @__kmpc_push_target_tripcount(%struct.ident_t* @{{.+}}, i64 -1, i64 %{{.+}})
+  // CK1: call void @__kmpc_push_target_tripcount_mapper(%struct.ident_t* @{{.+}}, i64 -1, i64 %{{.+}})
   // CK1: call i32 @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}}, i64 -1, i8* @{{[^,]+}}, i32 4, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i8** null, i8** null, i32 {{.+}}, i32 {{.+}})
 
   // CK1: call void @[[OFFL1:.+]](i{{32|64}} [[TE_PAR]], i{{32|64}} [[TH_PAR]],

clang/test/OpenMP/teams_distribute_parallel_for_codegen.cpp

   // CK1: [[TH_CAST:%.+]] = alloca i{{32|64}},
   // CK1: [[TE_PAR:%.+]] = load{{.+}}, {{.+}} [[TE_CAST]],
   // CK1: [[TH_PAR:%.+]] = load{{.+}}, {{.+}} [[TH_CAST]],
-  // CK1: call void @__kmpc_push_target_tripcount(%struct.ident_t* @{{.+}}, i64 -1, i64 %{{.+}})
+  // CK1: call void @__kmpc_push_target_tripcount_mapper(%struct.ident_t* @{{.+}}, i64 -1, i64 %{{.+}})
   // CK1: call i32 @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}}, i64 -1, i8* @{{[^,]+}}, i32 4, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i8** null, i8** null, i32 {{.+}}, i32 {{.+}})
 
   // CK1: call void @[[OFFL1:.+]](i{{32|64}} [[TE_PAR]], i{{32|64}} [[TH_PAR]],

clang/test/OpenMP/teams_distribute_parallel_for_simd_codegen.cpp

   // CK1: [[TE_PAR:%.+]] = load{{.+}}, {{.+}} [[TE_CAST]],
   // CK1: [[TH_PAR:%.+]] = load{{.+}}, {{.+}} [[TH_CAST]],
 
-  // CK1: call void @__kmpc_push_target_tripcount(%struct.ident_t* @{{.+}}, i64 -1, i64 %{{.+}})
+  // CK1: call void @__kmpc_push_target_tripcount_mapper(%struct.ident_t* @{{.+}}, i64 -1, i64 %{{.+}})
   // CK1: call i32 @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}}, i64 -1, i8* @{{[^,]+}}, i32 4, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i8** null
 
   // CK1: call void @[[OFFL1:.+]](i{{32|64}} [[TE_PAR]], i{{32|64}} [[TH_PAR]],

clang/test/OpenMP/teams_distribute_simd_codegen.cpp

   // CK1: [[TE_PAR:%.+]] = load{{.+}}, {{.+}} [[TE_CAST]],
   // CK1: [[TH_PAR:%.+]] = load{{.+}}, {{.+}} [[TH_CAST]],
 
-  // CK1: call void @__kmpc_push_target_tripcount(%struct.ident_t* @{{.+}}, i64 -1, i64 %{{.+}})
+  // CK1: call void @__kmpc_push_target_tripcount_mapper(%struct.ident_t* @{{.+}}, i64 -1, i64 %{{.+}})
   // CK1: call i32 @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}}, i64 -1, i8* @{{[^,]+}}, i32 5, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i8** null, i8** null, i32 {{.+}}, i32 1)
 
   // CK1: call void @[[OFFL1:.+]](i{{32|64}} [[TE_PAR]], i{{32|64}} [[TH_PAR]],

llvm/include/llvm/Frontend/OpenMP/OMPKinds.def

 __OMP_RTL(__kmpc_destroy_allocator, false, Void, /* Int */ Int32,
           /* omp_allocator_handle_t */ VoidPtr)
 
-__OMP_RTL(__kmpc_push_target_tripcount, false, Void, IdentPtr, Int64, Int64)
+__OMP_RTL(__kmpc_push_target_tripcount_mapper, false, Void, IdentPtr, Int64, Int64)
 __OMP_RTL(__tgt_target_mapper, false, Int32, IdentPtr, Int64, VoidPtr, Int32, VoidPtrPtr,
           VoidPtrPtr, Int64Ptr, Int64Ptr, VoidPtrPtr, VoidPtrPtr)
 __OMP_RTL(__tgt_target_nowait_mapper, false, Int32, IdentPtr, Int64, VoidPtr, Int32,

 __OMP_RTL_ATTRS(__kmpc_init_allocator, DefaultAttrs, ReturnPtrAttrs, {})
 __OMP_RTL_ATTRS(__kmpc_destroy_allocator, AllocAttrs, AttributeSet(), {})
 
-__OMP_RTL_ATTRS(__kmpc_push_target_tripcount, SetterAttrs, AttributeSet(), {})
+__OMP_RTL_ATTRS(__kmpc_push_target_tripcount_mapper, SetterAttrs, AttributeSet(), {})
 __OMP_RTL_ATTRS(__tgt_target_mapper, ForkAttrs, AttributeSet(), {})
 __OMP_RTL_ATTRS(__tgt_target_nowait_mapper, ForkAttrs, AttributeSet(), {})
 __OMP_RTL_ATTRS(__tgt_target_teams_mapper, ForkAttrs, AttributeSet(), {})

llvm/test/Transforms/OpenMP/add_attributes.ll

 
 declare void @__kmpc_destroy_allocator(i32, i8*)
 
-declare void @__kmpc_push_target_tripcount(%struct.ident_t*, i64, i64)
+declare void @__kmpc_push_target_tripcount_mapper(%struct.ident_t*, i64, i64)
 
 declare i32 @__kmpc_warp_active_thread_mask()
 

 ; CHECK-NEXT: declare void @__kmpc_destroy_allocator(i32, i8*)
 
 ; CHECK: ; Function Attrs: nounwind
-; CHECK-NEXT: declare void @__kmpc_push_target_tripcount(%struct.ident_t*, i64, i64)
+; CHECK-NEXT: declare void @__kmpc_push_target_tripcount_mapper(%struct.ident_t*, i64, i64)
 
 ; CHECK: ; Function Attrs: convergent nounwind
 ; CHECK-NEXT: declare i32 @__kmpc_warp_active_thread_mask()

 ; OPTIMISTIC-NEXT: declare void @__kmpc_destroy_allocator(i32, i8*)
 
 ; OPTIMISTIC: ; Function Attrs: inaccessiblememonly nofree nosync nounwind willreturn writeonly
-; OPTIMISTIC-NEXT: declare void @__kmpc_push_target_tripcount(%struct.ident_t*, i64, i64)
+; OPTIMISTIC-NEXT: declare void @__kmpc_push_target_tripcount_mapper(%struct.ident_t*, i64, i64)
 
 ; OPTIMISTIC: ; Function Attrs: convergent nounwind
 ; OPTIMISTIC-NEXT: declare i32 @__kmpc_warp_active_thread_mask()

openmp/libomptarget/include/omptarget.h

     int32_t thread_limit, int32_t depNum, void *depList, int32_t noAliasDepNum,
     void *noAliasDepList);
 
-void __kmpc_push_target_tripcount(ident_t *loc, int64_t device_id,
-                                  uint64_t loop_tripcount);
+void __kmpc_push_target_tripcount(int64_t device_id, uint64_t loop_tripcount);
+
+void __kmpc_push_target_tripcount_mapper(ident_t *loc, int64_t device_id,
+                                         uint64_t loop_tripcount);
 
 #ifdef __cplusplus
 }

openmp/libomptarget/src/exports

     __tgt_target_teams_nowait_mapper;
     __tgt_mapper_num_components;
     __tgt_push_mapper_component;
+    __kmpc_push_target_tripcount;
+    __kmpc_push_target_tripcount_mapper;
     omp_get_num_devices;
     omp_get_initial_device;
     omp_target_alloc;

     omp_target_memcpy_rect;
     omp_target_associate_ptr;
     omp_target_disassociate_ptr;
-    __kmpc_push_target_tripcount;
   local:
     *;
 };

openmp/libomptarget/src/interface.cpp

       MapComponentInfoTy(base, begin, size, type, name));
 }
 
-EXTERN void __kmpc_push_target_tripcount(ident_t *loc, int64_t device_id,
+EXTERN void __kmpc_push_target_tripcount(int64_t device_id,
                                          uint64_t loop_tripcount) {
+  __kmpc_push_target_tripcount_mapper(nullptr, device_id, loop_tripcount);
+}
+
+EXTERN void __kmpc_push_target_tripcount_mapper(ident_t *loc, int64_t device_id,
+                                                uint64_t loop_tripcount) {
   TIMESCOPE_WITH_IDENT(loc);
   if (checkDeviceAndCtors(device_id, loc) != OFFLOAD_SUCCESS) {
     DP("Not offloading to device %" PRId64 "\n", device_id);

openmp/libomptarget/src/omptarget.cpp

 
 /// Get loop trip count
 /// FIXME: This function will not work right if calling
-/// __kmpc_push_target_tripcount in one thread but doing offloading in another
-/// thread, which might occur when we call task yield.
+/// __kmpc_push_target_tripcount_mapper in one thread but doing offloading in
+/// another thread, which might occur when we call task yield.
 uint64_t getLoopTripCount(int64_t DeviceId) {
   DeviceTy &Device = PM->Devices[DeviceId];
   uint64_t LoopTripCount = 0;

openmp/libomptarget/test/offloading/host_as_target.c

   printf("omp_is_initial_device() = %d\n", omp_is_initial_device());
   CHECK_DATA();
 
-  // Check that __kmpc_push_target_tripcount doesn't fail.  I'm not sure how to
-  // check that it actually pushes to the initial device.
+  // Check that __kmpc_push_target_tripcount_mapper doesn't fail. I'm not sure
+  // how to check that it actually pushes to the initial device.
   #pragma omp target teams device(DevInit) num_teams(1)
   #pragma omp distribute
   for (int i = 0; i < 2; ++i)

   printf("omp_is_initial_device() = %d\n", omp_is_initial_device());
   CHECK_DATA();
 
-  // Check that __kmpc_push_target_tripcount doesn't fail.  I'm not sure how to
-  // check that it actually pushes to the initial device.
+  // Check that __kmpc_push_target_tripcount_mapper doesn't fail. I'm not sure
+  // how to check that it actually pushes to the initial device.
   #pragma omp target teams num_teams(1)
   #pragma omp distribute
   for (int i = 0; i < 2; ++i)