[OpenMP][FIX] Disabled optimizations have to be made known

To avoid simplification with wrong constants we need to make sure we
know that we won't perform specific optimizations based on the users
request. The non-SPMDzation and non-CustomStateMachine flags did only
prevent the final transformation but allowed to value simplification
to go ahead.

Differential Revision: https://reviews.llvm.org/D107862

Author: jdoerfert

llvm/lib/Transforms/IPO/OpenMPOpt.cpp

@@ -2834,6 +2834,10 @@ struct AAKernelInfoFunction : AAKernelInfo {
     // to avoid other parts using the current constant value for simpliication.
     auto &OMPInfoCache = static_cast<OMPInformationCache &>(A.getInfoCache());
 
+    // If we have disabled SPMD-ization, stop
+    if (DisableOpenMPOptSPMDization)
+      SPMDCompatibilityTracker.indicatePessimisticFixpoint();
+
     Function *Fn = getAnchorScope();
     if (!OMPInfoCache.Kernels.count(Fn))
       return;
@@ -2891,6 +2895,9 @@ struct AAKernelInfoFunction : AAKernelInfo {
       // in an invalid state, we won't change the value that is in the IR.
       if (!isValidState())
         return nullptr;
+      // If we have disabled state machine rewrites, don't make a custom one.
+      if (DisableOpenMPOptStateMachineRewrite)
+        return nullptr;
       if (AA)
         A.recordDependence(*this, *AA, DepClassTy::OPTIONAL);
       UsedAssumedInformation = !isAtFixpoint();
@@ -2992,10 +2999,6 @@ struct AAKernelInfoFunction : AAKernelInfo {
   }
 
   bool changeToSPMDMode(Attributor &A) {
-    // If we have disabled SPMD-ization, stop
-    if (DisableOpenMPOptSPMDization)
-      indicatePessimisticFixpoint();
-
     auto &OMPInfoCache = static_cast<OMPInformationCache &>(A.getInfoCache());
 
     if (!SPMDCompatibilityTracker.isAssumed()) {

llvm/test/Transforms/OpenMP/spmdization.ll

@@ -138,6 +138,7 @@ define weak void @__omp_offloading_14_a34ca11_sequential_loop_l5() #0 {
 ; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
 ; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 ; CHECK-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 ; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
 ; CHECK-DISABLED-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
 ; CHECK-DISABLED-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
@@ -394,6 +395,7 @@ define weak void @__omp_offloading_14_a34ca11_sequential_loop_to_stack_var_l20()
 ; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
 ; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 ; CHECK-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 ; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
 ; CHECK-DISABLED-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
 ; CHECK-DISABLED-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
@@ -652,6 +654,7 @@ define weak void @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_l35(
 ; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
 ; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 ; CHECK-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 ; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
 ; CHECK-DISABLED-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
 ; CHECK-DISABLED-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
@@ -932,6 +935,7 @@ define weak void @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_guar
 ; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
 ; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 ; CHECK-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 ; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
 ; CHECK-DISABLED-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
 ; CHECK-DISABLED-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
@@ -1253,6 +1257,7 @@ define weak void @__omp_offloading_14_a34ca11_do_not_spmdize_target_l65() #0 {
 ; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
 ; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 ; CHECK-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 ; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
 ; CHECK-DISABLED-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
 ; CHECK-DISABLED-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]