[llvm] 7bc88cb - [OpenMP] Simplify `llvm.assume` operands in device code

[Johannes Doerfert via llvm-commits]

Author: Johannes Doerfert
Date: 2023-01-22T01:27:41-08:00
New Revision: 7bc88cbe5c0365bba0873ed896a435e57402a5e0

URL: https://github.com/llvm/llvm-project/commit/7bc88cbe5c0365bba0873ed896a435e57402a5e0
DIFF: https://github.com/llvm/llvm-project/commit/7bc88cbe5c0365bba0873ed896a435e57402a5e0.diff

LOG: [OpenMP] Simplify `llvm.assume` operands in device code

Added: 
    

Modified: 
    llvm/lib/Transforms/IPO/OpenMPOpt.cpp
    llvm/test/Transforms/OpenMP/value-simplify-openmp-opt.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/IPO/OpenMPOpt.cpp b/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
index 89723a3d6c0df..12202dd5335ce 100644
--- a/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
+++ b/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
@@ -4899,8 +4899,18 @@ void OpenMPOpt::registerAAsForFunction(Attributor &A, const Function &F) {
       bool UsedAssumedInformation = false;
       A.getAssumedSimplified(IRPosition::value(*LI), /* AA */ nullptr,
                              UsedAssumedInformation, AA::Interprocedural);
-    } else if (auto *SI = dyn_cast<StoreInst>(&I)) {
+      continue;
+    }
+    if (auto *SI = dyn_cast<StoreInst>(&I)) {
       A.getOrCreateAAFor<AAIsDead>(IRPosition::value(*SI));
+      continue;
+    }
+    if (auto *II = dyn_cast<IntrinsicInst>(&I)) {
+      if (II->getIntrinsicID() == Intrinsic::assume) {
+        A.getOrCreateAAFor<AAPotentialValues>(
+            IRPosition::value(*II->getArgOperand(0)));
+        continue;
+      }
     }
   }
 }

diff  --git a/llvm/test/Transforms/OpenMP/value-simplify-openmp-opt.ll b/llvm/test/Transforms/OpenMP/value-simplify-openmp-opt.ll
index 7a45eb5e8be01..5bc6704ffc67e 100644
--- a/llvm/test/Transforms/OpenMP/value-simplify-openmp-opt.ll
+++ b/llvm/test/Transforms/OpenMP/value-simplify-openmp-opt.ll
@@ -7,10 +7,16 @@ target triple = "amdgcn-amd-amdhsa"
 %struct.ident_t = type { i32, i32, i32, i32, ptr }
 
 @G = internal addrspace(3) global i32 undef, align 4
+ at str = private unnamed_addr addrspace(4) constant [1 x i8] c"\00", align 1
 
 ; Make sure we do not delete the stores to @G without also replacing the load with `1`.
 ;.
-; CHECK: @[[G:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global i32 undef, align 4
+; TUNIT: @[[G:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global i32 undef, align 4
+; TUNIT: @[[STR:[a-zA-Z0-9_$"\\.-]+]] = private unnamed_addr addrspace(4) constant [1 x i8] zeroinitializer, align 1
+; TUNIT: @[[KERNEL_NESTED_PARALLELISM:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 0
+;.
+; CGSCC: @[[G:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global i32 undef, align 4
+; CGSCC: @[[STR:[a-zA-Z0-9_$"\\.-]+]] = private unnamed_addr addrspace(4) constant [1 x i8] zeroinitializer, align 1
 ;.
 define void @kernel() "kernel" {
 ; TUNIT: Function Attrs: norecurse
@@ -23,15 +29,15 @@ define void @kernel() "kernel" {
 ; TUNIT-NEXT:    store i32 1, ptr addrspace(3) @G, align 4
 ; TUNIT-NEXT:    br label [[IF_MERGE:%.*]]
 ; TUNIT:       if.else:
-; TUNIT-NEXT:    call void @barrier() #[[ATTR4:[0-9]+]]
+; TUNIT-NEXT:    call void @barrier() #[[ATTR5:[0-9]+]]
 ; TUNIT-NEXT:    [[L:%.*]] = load i32, ptr addrspace(3) @G, align 4
-; TUNIT-NEXT:    call void @use1(i32 [[L]]) #[[ATTR4]]
+; TUNIT-NEXT:    call void @use1(i32 [[L]]) #[[ATTR5]]
 ; TUNIT-NEXT:    br label [[IF_MERGE]]
 ; TUNIT:       if.merge:
 ; TUNIT-NEXT:    br i1 [[CMP]], label [[IF_THEN2:%.*]], label [[IF_END:%.*]]
 ; TUNIT:       if.then2:
 ; TUNIT-NEXT:    store i32 2, ptr addrspace(3) @G, align 4
-; TUNIT-NEXT:    call void @barrier() #[[ATTR4]]
+; TUNIT-NEXT:    call void @barrier() #[[ATTR5]]
 ; TUNIT-NEXT:    br label [[IF_END]]
 ; TUNIT:       if.end:
 ; TUNIT-NEXT:    call void @__kmpc_target_deinit(ptr undef, i8 1)
@@ -83,10 +89,20 @@ if.end:
   ret void
 }
 
+define void @test_assume() {
+; CHECK-LABEL: define {{[^@]+}}@test_assume() {
+; CHECK-NEXT:    call void @llvm.assume(i1 true)
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.assume(i1 icmp ne (ptr addrspacecast (ptr addrspace(4) @str to ptr), ptr null))
+  ret void
+}
+
 declare void @barrier() norecurse nounwind nocallback
 declare void @use1(i32) nosync norecurse nounwind nocallback
 declare i32 @__kmpc_target_init(ptr, i8, i1) nocallback
 declare void @__kmpc_target_deinit(ptr, i8) nocallback
+declare void @llvm.assume(i1)
 
 !llvm.module.flags = !{!0, !1}
 !nvvm.annotations = !{!2}
@@ -100,16 +116,16 @@ declare void @__kmpc_target_deinit(ptr, i8) nocallback
 ; TUNIT: attributes #[[ATTR1:[0-9]+]] = { nocallback norecurse nounwind }
 ; TUNIT: attributes #[[ATTR2:[0-9]+]] = { nocallback norecurse nosync nounwind }
 ; TUNIT: attributes #[[ATTR3:[0-9]+]] = { nocallback }
-; TUNIT: attributes #[[ATTR4]] = { nounwind }
+; TUNIT: attributes #[[ATTR4:[0-9]+]] = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: readwrite) }
+; TUNIT: attributes #[[ATTR5]] = { nounwind }
 ;.
 ; CGSCC: attributes #[[ATTR0]] = { norecurse "kernel" }
 ; CGSCC: attributes #[[ATTR1:[0-9]+]] = { nocallback norecurse nounwind }
 ; CGSCC: attributes #[[ATTR2:[0-9]+]] = { nocallback norecurse nosync nounwind }
 ; CGSCC: attributes #[[ATTR3:[0-9]+]] = { nocallback }
+; CGSCC: attributes #[[ATTR4:[0-9]+]] = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: readwrite) }
 ;.
 ; CHECK: [[META0:![0-9]+]] = !{i32 7, !"openmp", i32 50}
 ; CHECK: [[META1:![0-9]+]] = !{i32 7, !"openmp-device", i32 50}
 ; CHECK: [[META2:![0-9]+]] = !{ptr @kernel, !"kernel", i32 1}
 ;.
-;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
-; CHECK: {{.*}}