[OpenMP] Guarding restrictions are required only for guarding

If we do not guard code during SPMDzation, we do not need to check
conditions for successfull guarding. That is, even if some code is
executed in different modes, it does not prevent SPMDzation if there is
no guarded code in there.

diff --git a/llvm/lib/Transforms/IPO/OpenMPOpt.cpp b/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
index 8a25148..67014bd 100644 (file)
--- a/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
+++ b/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
@@ -4149,28 +4149,30 @@ struct AAKernelInfoFunction : AAKernelInfo {
       updateReachingKernelEntries(A, AllReachingKernelsKnown);
       UsedAssumedInformationFromReachingKernels = !AllReachingKernelsKnown;
 
-      if (!ParallelLevels.isValidState())
-        SPMDCompatibilityTracker.indicatePessimisticFixpoint();
-      else if (!ReachingKernelEntries.isValidState())
-        SPMDCompatibilityTracker.indicatePessimisticFixpoint();
-      else if (!SPMDCompatibilityTracker.empty()) {
-        // Check if all reaching kernels agree on the mode as we can otherwise
-        // not guard instructions. We might not be sure about the mode so we
-        // we cannot fix the internal spmd-zation state either.
-        int SPMD = 0, Generic = 0;
-        for (auto *Kernel : ReachingKernelEntries) {
-          auto &CBAA = A.getAAFor<AAKernelInfo>(
-              *this, IRPosition::function(*Kernel), DepClassTy::OPTIONAL);
-          if (CBAA.SPMDCompatibilityTracker.isValidState() &&
-              CBAA.SPMDCompatibilityTracker.isAssumed())
-            ++SPMD;
-          else
-            ++Generic;
-          if (!CBAA.SPMDCompatibilityTracker.isAtFixpoint())
-            UsedAssumedInformationFromReachingKernels = true;
-        }
-        if (SPMD != 0 && Generic != 0)
+      if (!SPMDCompatibilityTracker.empty()) {
+        if (!ParallelLevels.isValidState())
           SPMDCompatibilityTracker.indicatePessimisticFixpoint();
+        else if (!ReachingKernelEntries.isValidState())
+          SPMDCompatibilityTracker.indicatePessimisticFixpoint();
+        else {
+          // Check if all reaching kernels agree on the mode as we can otherwise
+          // not guard instructions. We might not be sure about the mode so we
+          // we cannot fix the internal spmd-zation state either.
+          int SPMD = 0, Generic = 0;
+          for (auto *Kernel : ReachingKernelEntries) {
+            auto &CBAA = A.getAAFor<AAKernelInfo>(
+                *this, IRPosition::function(*Kernel), DepClassTy::OPTIONAL);
+            if (CBAA.SPMDCompatibilityTracker.isValidState() &&
+                CBAA.SPMDCompatibilityTracker.isAssumed())
+              ++SPMD;
+            else
+              ++Generic;
+            if (!CBAA.SPMDCompatibilityTracker.isAtFixpoint())
+              UsedAssumedInformationFromReachingKernels = true;
+          }
+          if (SPMD != 0 && Generic != 0)
+            SPMDCompatibilityTracker.indicatePessimisticFixpoint();
+        }
       }
     }
 

diff --git a/llvm/test/Transforms/OpenMP/spmdization_no_guarding_two_reaching_kernels.ll b/llvm/test/Transforms/OpenMP/spmdization_no_guarding_two_reaching_kernels.ll
index e1971dd..686d487 100644 (file)
--- a/llvm/test/Transforms/OpenMP/spmdization_no_guarding_two_reaching_kernels.ll
+++ b/llvm/test/Transforms/OpenMP/spmdization_no_guarding_two_reaching_kernels.ll
@@ -43,14 +43,13 @@ target triple = "nvptx64"
 ;.
 ; CHECK: @[[GLOB0:[0-9]+]] = private unnamed_addr constant [23 x i8] c"
 ; CHECK: @[[GLOB1:[0-9]+]] = private unnamed_addr constant [[STRUCT_IDENT_T:%.*]] { i32 0, i32 2, i32 0, i32 0, ptr @[[GLOB0]] }, align 8
-; CHECK: @[[__OMP_OFFLOADING_2B_10393B5_SPMD_L12_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 1
+; CHECK: @[[__OMP_OFFLOADING_2B_10393B5_SPMD_L12_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 3
 ; CHECK: @[[__OMP_OFFLOADING_2B_10393B5_GENERIC_L20_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 1
 ; CHECK: @[[GLOB2:[0-9]+]] = private unnamed_addr constant [[STRUCT_IDENT_T:%.*]] { i32 0, i32 2, i32 2, i32 0, ptr @[[GLOB0]] }, align 8
 ; CHECK: @[[G:[a-zA-Z0-9_$"\\.-]+]] = external addrspace(5) global i32, align 4
 ; CHECK: @[[LLVM_COMPILER_USED:[a-zA-Z0-9_$"\\.-]+]] = appending global [2 x ptr] [ptr @__omp_offloading_2b_10393b5_spmd_l12_exec_mode, ptr @__omp_offloading_2b_10393b5_generic_l20_exec_mode], section "llvm.metadata"
 ; CHECK: @[[__OMP_OFFLOADING_2B_10393B5_SPMD_L12_NESTED_PARALLELISM:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 0
 ; CHECK: @[[__OMP_OFFLOADING_2B_10393B5_GENERIC_L20_NESTED_PARALLELISM:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 0
-; CHECK: @[[__OMP_OUTLINED___WRAPPER_ID:[a-zA-Z0-9_$"\\.-]+]] = private constant i8 undef
 ;.
 ; CHECK-DISABLE-SPMDIZATION: @[[GLOB0:[0-9]+]] = private unnamed_addr constant [23 x i8] c"
 ; CHECK-DISABLE-SPMDIZATION: @[[GLOB1:[0-9]+]] = private unnamed_addr constant [[STRUCT_IDENT_T:%.*]] { i32 0, i32 2, i32 0, i32 0, ptr @[[GLOB0]] }, align 8
@@ -67,46 +66,12 @@ define weak void @__omp_offloading_2b_10393b5_spmd_l12() #0 {
 ; CHECK-LABEL: define {{[^@]+}}@__omp_offloading_2b_10393b5_spmd_l12
 ; CHECK-SAME: () #[[ATTR0:[0-9]+]] {
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca ptr, align 8
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 false)
-; CHECK-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
-; CHECK-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[IS_WORKER_CHECK:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
-; CHECK:       is_worker_check:
-; CHECK-NEXT:    [[BLOCK_HW_SIZE:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
-; CHECK-NEXT:    [[WARP_SIZE:%.*]] = call i32 @__kmpc_get_warp_size()
-; CHECK-NEXT:    [[BLOCK_SIZE:%.*]] = sub i32 [[BLOCK_HW_SIZE]], [[WARP_SIZE]]
-; CHECK-NEXT:    [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[BLOCK_SIZE]]
-; CHECK-NEXT:    br i1 [[THREAD_IS_MAIN_OR_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]]
-; CHECK:       worker_state_machine.begin:
-; CHECK-NEXT:    call void @__kmpc_barrier_simple_generic(ptr @[[GLOB1]], i32 [[TMP0]])
-; CHECK-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(ptr [[WORKER_WORK_FN_ADDR]])
-; CHECK-NEXT:    [[WORKER_WORK_FN:%.*]] = load ptr, ptr [[WORKER_WORK_FN_ADDR]], align 8
-; CHECK-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast ptr [[WORKER_WORK_FN]] to ptr
-; CHECK-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq ptr [[WORKER_WORK_FN]], null
-; CHECK-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
-; CHECK:       worker_state_machine.finished:
-; CHECK-NEXT:    ret void
-; CHECK:       worker_state_machine.is_active.check:
-; CHECK-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
-; CHECK:       worker_state_machine.parallel_region.check:
-; CHECK-NEXT:    br i1 true, label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK1:%.*]]
-; CHECK:       worker_state_machine.parallel_region.execute:
-; CHECK-NEXT:    call void @__omp_outlined___wrapper(i16 0, i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
-; CHECK:       worker_state_machine.parallel_region.check1:
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
-; CHECK:       worker_state_machine.parallel_region.end:
-; CHECK-NEXT:    call void @__kmpc_kernel_end_parallel()
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
-; CHECK:       worker_state_machine.done.barrier:
-; CHECK-NEXT:    call void @__kmpc_barrier_simple_generic(ptr @[[GLOB1]], i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
-; CHECK:       thread.user_code.check:
+; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
 ; CHECK-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 ; CHECK-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 ; CHECK:       user_code.entry:
 ; CHECK-NEXT:    call void @spmd_helper() #[[ATTR6:[0-9]+]]
-; CHECK-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+; CHECK-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
 ; CHECK-NEXT:    ret void
 ; CHECK:       worker.exit:
 ; CHECK-NEXT:    ret void
@@ -297,7 +262,7 @@ define internal void @spmd_helper() #1 {
 ; CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x ptr], align 8
 ; CHECK-NEXT:    call void @leaf() #[[ATTR3:[0-9]+]]
 ; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB2]]) #[[ATTR3]]
-; CHECK-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB2]], i32 [[TMP0]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr @__omp_outlined___wrapper.ID, ptr [[CAPTURED_VARS_ADDRS]], i64 0)
+; CHECK-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB2]], i32 [[TMP0]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr @__omp_outlined___wrapper, ptr [[CAPTURED_VARS_ADDRS]], i64 0)
 ; CHECK-NEXT:    ret void
 ;
 ; CHECK-DISABLE-SPMDIZATION-LABEL: define {{[^@]+}}@spmd_helper