Revert "[OpenMP] Remove noinline attributes in the device runtime"

The behaviour of this patch is not great, but it has some side-effects
that are required for OpenMPOpt to work. The problem is that when we use
`-mlink-builtin-bitcode` we only import used symbols from the runtime.
Then OpenMPOpt will insert calls to symbols that were not previously
included. This patch removed this implicit behaviour as these functions
were kept alive by the `noinline` simply because it kept calls to them
in the module. This caused regression in some tests that relied on some
OpenMPOpt passes without using LTO. Reverting for the LLVM15 release but
will try to fix it more correctly on main.

This reverts commit d61d72d.

Fixes #56752

Author: jhuber6

llvm/lib/Transforms/IPO/OpenMPOpt.cpp

@@ -499,6 +499,18 @@ struct OMPInformationCache : public InformationCache {
   }
 #include "llvm/Frontend/OpenMP/OMPKinds.def"
 
+    // Remove the `noinline` attribute from `__kmpc`, `_OMP::` and `omp_`
+    // functions, except if `optnone` is present.
+    if (isOpenMPDevice(M)) {
+      for (Function &F : M) {
+        for (StringRef Prefix : {"__kmpc", "_ZN4_OMP", "omp_"})
+          if (F.hasFnAttribute(Attribute::NoInline) &&
+              F.getName().startswith(Prefix) &&
+              !F.hasFnAttribute(Attribute::OptimizeNone))
+            F.removeFnAttr(Attribute::NoInline);
+      }
+    }
+
     // TODO: We should attach the attributes defined in OMPKinds.def.
   }
 

llvm/test/Transforms/OpenMP/remove_noinline_attributes.ll

@@ -0,0 +1,99 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-attributes
+; RUN: opt < %s -S -openmp-opt-cgscc        | FileCheck %s
+; RUN: opt < %s -S -passes=openmp-opt-cgscc | FileCheck %s
+
+declare void @unknown()
+
+; __kmpc functions
+define void @__kmpc_noinline() noinline nounwind {
+; CHECK: Function Attrs: nounwind
+; CHECK-LABEL: @__kmpc_noinline(
+; CHECK-NEXT:    call void @unknown()
+; CHECK-NEXT:    ret void
+;
+  call void @unknown()
+  ret void
+}
+; omp_X functions
+define void @omp_noinline() noinline nounwind {
+; CHECK: Function Attrs: nounwind
+; CHECK-LABEL: @omp_noinline(
+; CHECK-NEXT:    call void @unknown()
+; CHECK-NEXT:    ret void
+;
+  call void @unknown()
+  ret void
+}
+; _OMP namespace
+define void @_ZN4_OMP_noinline() noinline nounwind {
+; CHECK: Function Attrs: nounwind
+; CHECK-LABEL: @_ZN4_OMP_noinline(
+; CHECK-NEXT:    call void @unknown()
+; CHECK-NEXT:    ret void
+;
+  call void @unknown()
+  ret void
+}
+
+; Negative tests:
+
+define void @__kmpc_noinline_optnone() noinline optnone nounwind {
+; CHECK: Function Attrs: noinline nounwind optnone
+; CHECK-LABEL: @__kmpc_noinline_optnone(
+; CHECK-NEXT:    call void @unknown()
+; CHECK-NEXT:    ret void
+;
+  call void @unknown()
+  ret void
+}
+define void @omp_noinline_optnone() noinline optnone nounwind {
+; CHECK: Function Attrs: noinline nounwind optnone
+; CHECK-LABEL: @omp_noinline_optnone(
+; CHECK-NEXT:    call void @unknown()
+; CHECK-NEXT:    ret void
+;
+  call void @unknown()
+  ret void
+}
+; _OMP namespace
+define void @_ZN4_OMP_noinline_optnone() noinline optnone nounwind {
+; CHECK: Function Attrs: noinline nounwind optnone
+; CHECK-LABEL: @_ZN4_OMP_noinline_optnone(
+; CHECK-NEXT:    call void @unknown()
+; CHECK-NEXT:    ret void
+;
+  call void @unknown()
+  ret void
+}
+define void @a___kmpc_noinline() noinline nounwind {
+; CHECK: Function Attrs: noinline nounwind
+; CHECK-LABEL: @a___kmpc_noinline(
+; CHECK-NEXT:    call void @unknown()
+; CHECK-NEXT:    ret void
+;
+  call void @unknown()
+  ret void
+}
+define void @a_omp_noinline() noinline nounwind {
+; CHECK: Function Attrs: noinline nounwind
+; CHECK-LABEL: @a_omp_noinline(
+; CHECK-NEXT:    call void @unknown()
+; CHECK-NEXT:    ret void
+;
+  call void @unknown()
+  ret void
+}
+define void @a__ZN4_OMP_noinline() noinline nounwind {
+; CHECK: Function Attrs: noinline nounwind
+; CHECK-LABEL: @a__ZN4_OMP_noinline(
+; CHECK-NEXT:    call void @unknown()
+; CHECK-NEXT:    ret void
+;
+  call void @unknown()
+  ret void
+}
+
+!llvm.module.flags = !{!0, !1}
+
+!0 = !{i32 7, !"openmp", i32 50}
+!1 = !{i32 7, !"openmp-device", i32 50}

openmp/libomptarget/DeviceRTL/include/Synchronization.h

@@ -29,13 +29,15 @@ void threads();
 
 /// Synchronizing threads is allowed even if they all hit different instances of
 /// `synchronize::threads()`. However, `synchronize::threadsAligned()` is more
-/// restrictive in that it requires all threads to hit the same instance.
+/// restrictive in that it requires all threads to hit the same instance. The
+/// noinline is removed by the openmp-opt pass and helps to preserve the
+/// information till then.
 ///{
 #pragma omp begin assumes ext_aligned_barrier
 
 /// Synchronize all threads in a block, they are are reaching the same
 /// instruction (hence all threads in the block are "aligned").
-void threadsAligned();
+__attribute__((noinline)) void threadsAligned();
 
 #pragma omp end assumes
 ///}

openmp/libomptarget/DeviceRTL/src/Mapping.cpp

@@ -289,17 +289,17 @@ bool mapping::isGenericMode() { return !isSPMDMode(); }
 ///}
 
 extern "C" {
-uint32_t __kmpc_get_hardware_thread_id_in_block() {
+__attribute__((noinline)) uint32_t __kmpc_get_hardware_thread_id_in_block() {
   FunctionTracingRAII();
   return mapping::getThreadIdInBlock();
 }
 
-uint32_t __kmpc_get_hardware_num_threads_in_block() {
+__attribute__((noinline)) uint32_t __kmpc_get_hardware_num_threads_in_block() {
   FunctionTracingRAII();
   return impl::getNumHardwareThreadsInBlock();
 }
 
-uint32_t __kmpc_get_warp_size() {
+__attribute__((noinline)) uint32_t __kmpc_get_warp_size() {
   FunctionTracingRAII();
   return impl::getWarpSize();
 }

openmp/libomptarget/DeviceRTL/src/Parallelism.cpp

@@ -243,7 +243,8 @@ void __kmpc_parallel_51(IdentTy *ident, int32_t, int32_t if_expr,
     __kmpc_end_sharing_variables();
 }
 
-bool __kmpc_kernel_parallel(ParallelRegionFnTy *WorkFn) {
+__attribute__((noinline)) bool
+__kmpc_kernel_parallel(ParallelRegionFnTy *WorkFn) {
   FunctionTracingRAII();
   // Work function and arguments for L1 parallel region.
   *WorkFn = state::ParallelRegionFn;
@@ -258,7 +259,7 @@ bool __kmpc_kernel_parallel(ParallelRegionFnTy *WorkFn) {
   return ThreadIsActive;
 }
 
-void __kmpc_kernel_end_parallel() {
+__attribute__((noinline)) void __kmpc_kernel_end_parallel() {
   FunctionTracingRAII();
   // In case we have modified an ICV for this thread before a ThreadState was
   // created. We drop it now to not contaminate the next parallel region.

openmp/libomptarget/DeviceRTL/src/State.cpp

@@ -393,12 +393,12 @@ int omp_get_initial_device(void) { return -1; }
 }
 
 extern "C" {
-void *__kmpc_alloc_shared(uint64_t Bytes) {
+__attribute__((noinline)) void *__kmpc_alloc_shared(uint64_t Bytes) {
   FunctionTracingRAII();
   return memory::allocShared(Bytes, "Frontend alloc shared");
 }
 
-void __kmpc_free_shared(void *Ptr, uint64_t Bytes) {
+__attribute__((noinline)) void __kmpc_free_shared(void *Ptr, uint64_t Bytes) {
   FunctionTracingRAII();
   memory::freeShared(Ptr, Bytes, "Frontend free shared");
 }

openmp/libomptarget/DeviceRTL/src/Synchronization.cpp

@@ -358,12 +358,14 @@ void __kmpc_barrier(IdentTy *Loc, int32_t TId) {
   impl::namedBarrier();
 }
 
-void __kmpc_barrier_simple_spmd(IdentTy *Loc, int32_t TId) {
+__attribute__((noinline)) void __kmpc_barrier_simple_spmd(IdentTy *Loc,
+                                                          int32_t TId) {
   FunctionTracingRAII();
   synchronize::threadsAligned();
 }
 
-void __kmpc_barrier_simple_generic(IdentTy *Loc, int32_t TId) {
+__attribute__((noinline)) void __kmpc_barrier_simple_generic(IdentTy *Loc,
+                                                             int32_t TId) {
   FunctionTracingRAII();
   synchronize::threads();
 }