[CilkABI] Fix declaration of __cilkrts_get_nworkers() to use MemoryEf…

…fects.
OpenCilk · Nov 1, 2024 · d7f0f82 · d7f0f82
1 parent f78505e
commit d7f0f82
Show file tree

Hide file tree

Showing 2 changed files with 109 additions and 1 deletion.
diff --git a/llvm/lib/Transforms/Tapir/CilkABI.cpp b/llvm/lib/Transforms/Tapir/CilkABI.cpp
@@ -163,10 +163,13 @@ FunctionCallee CilkABI::Get__cilkrts_get_nworkers() {
 
   LLVMContext &C = M.getContext();
   AttributeList AL;
-  AL = AL.addFnAttribute(C, Attribute::ReadNone);
   AL = AL.addFnAttribute(C, Attribute::NoUnwind);
   FunctionType *FTy = FunctionType::get(Type::getInt32Ty(C), {}, false);
   CilkRTSGetNworkers = M.getOrInsertFunction("__cilkrts_get_nworkers", FTy, AL);
+  if (Function *Fn = dyn_cast<Function>(CilkRTSGetNworkers.getCallee())) {
+      MemoryEffects ME = MemoryEffects::inaccessibleMemOnly(ModRefInfo::Ref);
+      Fn->setMemoryEffects(ME);
+  }
   return CilkRTSGetNworkers;
 }
 

diff --git a/llvm/test/Transforms/Tapir/cilkplus-get-nworkers.ll b/llvm/test/Transforms/Tapir/cilkplus-get-nworkers.ll
@@ -0,0 +1,105 @@
+; Check that the Intel Cilk Plus function __cilkrts_get_nworkers()
+; is properly declared in the IR.
+;
+; RUN: opt < %s -passes="tapir2target" -tapir-target=cilk -S | FileCheck %s
+target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"
+target triple = "arm64-apple-macosx15.0.0"
+
+; Function Attrs: nounwind ssp uwtable(sync)
+define void @normalize(ptr noalias nocapture noundef writeonly %out, ptr noalias noundef %in, i32 noundef %n) local_unnamed_addr #0 {
+entry:
+  %cmp = icmp sgt i32 %n, 0
+  br i1 %cmp, label %pfor.cond.preheader, label %cleanup
+
+pfor.cond.preheader:                              ; preds = %entry
+  %wide.trip.count = zext nneg i32 %n to i64
+  %0 = call i64 @llvm.tapir.loop.grainsize.i64(i64 %wide.trip.count)
+  call fastcc void @normalize.outline_pfor.cond.ls1(i64 0, i64 %wide.trip.count, i64 %0, ptr %in, ptr %out, i32 %n) #4
+  br label %cleanup
+
+cleanup:                                          ; preds = %pfor.cond.preheader, %entry
+  ret void
+}
+
+; CHECK: define {{.*}}void @normalize(
+; CHECK: call i32 @__cilkrts_get_nworkers()
+; CHECK: call {{.*}}void @normalize.outline_pfor.cond.ls1(
+
+; CHECK: declare i32 @__cilkrts_get_nworkers()
+
+; Function Attrs: nounwind willreturn memory(argmem: readwrite)
+declare token @llvm.syncregion.start() #1
+
+declare double @norm(ptr noundef, i32 noundef) local_unnamed_addr #2
+
+; Function Attrs: nounwind speculatable willreturn memory(none)
+declare i64 @llvm.tapir.loop.grainsize.i64(i64) #3
+
+; Function Attrs: nounwind ssp uwtable(sync)
+define internal fastcc void @normalize.outline_pfor.cond.ls1(i64 %indvars.iv.start.ls1, i64 %end.ls1, i64 %grainsize.ls1, ptr noalias noundef align 1 %in.ls1, ptr noalias nocapture noundef writeonly align 1 %out.ls1, i32 noundef %n.ls1) unnamed_addr #0 {
+pfor.cond.preheader.ls1:
+  %0 = tail call token @llvm.syncregion.start()
+  %itercount1 = sub i64 %end.ls1, %indvars.iv.start.ls1
+  %1 = icmp ugt i64 %itercount1, %grainsize.ls1
+  br i1 %1, label %.lr.ph, label %pfor.cond.ls1.preheader
+
+pfor.cond.ls1.preheader:                          ; preds = %pfor.cond.preheader.ls1.dac.cont, %pfor.cond.preheader.ls1
+  %indvars.iv.ls1.dac.lcssa = phi i64 [ %indvars.iv.start.ls1, %pfor.cond.preheader.ls1 ], [ %miditer, %pfor.cond.preheader.ls1.dac.cont ]
+  br label %pfor.cond.ls1
+
+.lr.ph:                                           ; preds = %pfor.cond.preheader.ls1, %pfor.cond.preheader.ls1.dac.cont
+  %itercount3 = phi i64 [ %itercount, %pfor.cond.preheader.ls1.dac.cont ], [ %itercount1, %pfor.cond.preheader.ls1 ]
+  %indvars.iv.ls1.dac2 = phi i64 [ %miditer, %pfor.cond.preheader.ls1.dac.cont ], [ %indvars.iv.start.ls1, %pfor.cond.preheader.ls1 ]
+  %halfcount = lshr i64 %itercount3, 1
+  %miditer = add nuw nsw i64 %halfcount, %indvars.iv.ls1.dac2
+  detach within %0, label %pfor.cond.preheader.ls1.dac.detach, label %pfor.cond.preheader.ls1.dac.cont
+
+pfor.cond.preheader.ls1.dac.detach:               ; preds = %.lr.ph
+  call fastcc void @normalize.outline_pfor.cond.ls1(i64 %indvars.iv.ls1.dac2, i64 %miditer, i64 %grainsize.ls1, ptr %in.ls1, ptr %out.ls1, i32 %n.ls1) #4
+  reattach within %0, label %pfor.cond.preheader.ls1.dac.cont
+
+pfor.cond.preheader.ls1.dac.cont:                 ; preds = %pfor.cond.preheader.ls1.dac.detach, %.lr.ph
+  %itercount = sub i64 %end.ls1, %miditer
+  %2 = icmp ugt i64 %itercount, %grainsize.ls1
+  br i1 %2, label %.lr.ph, label %pfor.cond.ls1.preheader
+
+pfor.cond.ls1:                                    ; preds = %pfor.cond.ls1.preheader, %pfor.cond.ls1
+  %indvars.iv.ls1 = phi i64 [ %indvars.iv.next.ls1, %pfor.cond.ls1 ], [ %indvars.iv.ls1.dac.lcssa, %pfor.cond.ls1.preheader ]
+  %indvars.iv.next.ls1 = add nuw nsw i64 %indvars.iv.ls1, 1
+  %arrayidx.ls1 = getelementptr inbounds double, ptr %in.ls1, i64 %indvars.iv.ls1
+  %3 = load double, ptr %arrayidx.ls1, align 8, !tbaa !6
+  %call.ls1 = tail call double @norm(ptr noundef %in.ls1, i32 noundef %n.ls1) #4
+  %div.ls1 = fdiv double %3, %call.ls1
+  %arrayidx2.ls1 = getelementptr inbounds double, ptr %out.ls1, i64 %indvars.iv.ls1
+  store double %div.ls1, ptr %arrayidx2.ls1, align 8, !tbaa !6
+  %exitcond.not.ls1 = icmp eq i64 %indvars.iv.next.ls1, %end.ls1
+  br i1 %exitcond.not.ls1, label %pfor.cond.cleanup.ls1, label %pfor.cond.ls1, !llvm.loop !10
+
+pfor.cond.cleanup.ls1:                            ; preds = %pfor.cond.ls1
+  sync within %0, label %pfor.cond.cleanup.ls1.synced
+
+pfor.cond.cleanup.ls1.synced:                     ; preds = %pfor.cond.cleanup.ls1
+  ret void
+}
+
+attributes #0 = { nounwind ssp uwtable(sync) "frame-pointer"="non-leaf" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="apple-m1" "target-features"="+aes,+complxnum,+crc,+dotprod,+fp-armv8,+fp16fml,+fullfp16,+jsconv,+lse,+neon,+pauth,+ras,+rcpc,+rdm,+sha2,+sha3,+v8.1a,+v8.2a,+v8.3a,+v8.4a,+v8.5a,+v8a,+zcm,+zcz" }
+attributes #1 = { nounwind willreturn memory(argmem: readwrite) }
+attributes #2 = { "frame-pointer"="non-leaf" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="apple-m1" "target-features"="+aes,+complxnum,+crc,+dotprod,+fp-armv8,+fp16fml,+fullfp16,+jsconv,+lse,+neon,+pauth,+ras,+rcpc,+rdm,+sha2,+sha3,+v8.1a,+v8.2a,+v8.3a,+v8.4a,+v8.5a,+v8a,+zcm,+zcz" }
+attributes #3 = { nounwind speculatable willreturn memory(none) }
+attributes #4 = { nounwind }
+
+!llvm.module.flags = !{!0, !1, !2, !3, !4}
+!llvm.ident = !{!5}
+
+!0 = !{i32 2, !"SDK Version", [2 x i32] [i32 15, i32 0]}
+!1 = !{i32 1, !"wchar_size", i32 4}
+!2 = !{i32 8, !"PIC Level", i32 2}
+!3 = !{i32 7, !"uwtable", i32 1}
+!4 = !{i32 7, !"frame-pointer", i32 1}
+!5 = !{!"clang version 18.1.8 ([email protected]:OpenCilk/opencilk-project.git f78505e37b91651d57f439f34cbcb441c98b59a1)"}
+!6 = !{!7, !7, i64 0}
+!7 = !{!"double", !8, i64 0}
+!8 = !{!"omnipotent char", !9, i64 0}
+!9 = !{!"Simple C/C++ TBAA"}
+!10 = distinct !{!10, !11}
+!11 = !{!"tapir.loop.spawn.strategy", i32 0}