PR21403: CGSCC should not treat intrinsics as function calls

[spatel]

Summary

This patch tries to make the handling of calls to intrinsics in CGSCC consistent: they are not treated like regular function calls (because they are never lowered to function calls?).

Without this patch, we can get dangling pointer asserts from the subsequent loop that processes callsites because it already ignores intrinsics.

Please see http://llvm.org/bugs/show_bug.cgi?id=21403 for more details / discussion.

[spatel]

Ping.

Note: http://llvm.org/bugs/show_bug.cgi?id=21527 has been marked as a dup of PR21403.

[hfinkel]

Based on an IRC conversation with Chandler a few minutes ago, I think this is okay (because CGSCC should not actually visit the intrinsic call because there isn't really a called function). LGTM.

[spatel]

Closed by commit rL221802 (authored by @spatel).

[spatel]

Thanks, Hal and Chandler. Committed at r221802.

llvm/trunk/lib/Analysis/IPA/CallGraphSCCPass.cpp

for (CallGraphNode::iterator I = CGN->begin(), E = CGN->end(); I != E; ) {
         if (WasLast)
           break;
         E = CGN->end();
         continue;
       }
       
       assert(!CallSites.count(I->first) &&
              "Call site occurs in node multiple times");
+      
+      CallSite CS(I->first);
+      if (CS) {
+        Function *Callee = CS.getCalledFunction();
+        // Ignore intrinsics because they're not really function calls.
+        if (!Callee || !(Callee->isIntrinsic()))
           CallSites.insert(std::make_pair(I->first, I->second));
+      }
       ++I;
     }
     
     // Loop over all of the instructions in the function, getting the callsites.
     // Keep track of the number of direct/indirect calls added.
     unsigned NumDirectAdded = 0, NumIndirectAdded = 0;
     
     for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)

llvm/trunk/test/Transforms/InstCombine/no_cgscc_assert.ll

+; RUN: opt < %s -inline -instcombine -S | FileCheck %s
+
+; PR21403: http://llvm.org/bugs/show_bug.cgi?id=21403
+; When the call to sqrtf is replaced by an intrinsic call to fabs,
+; it should not cause a problem in CGSCC. 
+
+define float @bar(float %f) #0 {
+  %mul = fmul fast float %f, %f
+  %call1 = call float @sqrtf(float %mul) #0
+  ret float %call1
+
+; CHECK-LABEL: @bar(
+; CHECK-NEXT: call float @llvm.fabs.f32
+; CHECK-NEXT: ret float
+}
+
+declare float @sqrtf(float) #0
+
+attributes #0 = { readnone "unsafe-fp-math"="true" }