[llvm] r240039 - [CallGraph] Teach the CallGraph about non-leaf intrinsics.
Sanjoy Das
sanjoy at playingwithpointers.com
Fri Jun 19 14:18:17 PDT 2015
On Fri, Jun 19, 2015 at 1:28 PM, Sanjoy Das
<sanjoy at playingwithpointers.com> wrote:
> I think -print-callgraph (all print functions, actually) should be
> deterministic within reason. I'll put up a review fixing this
> shortly.
PTAL at http://reviews.llvm.org/D10575
>
>
> On Fri, Jun 19, 2015 at 1:12 PM, Justin Bogner <mail at justinbogner.com> wrote:
>> Sanjoy Das <sanjoy at playingwithpointers.com> writes:
>>> Author: sanjoy
>>> Date: Thu Jun 18 14:28:26 2015
>>> New Revision: 240039
>>>
>>> URL: http://llvm.org/viewvc/llvm-project?rev=240039&view=rev
>>> Log:
>>> [CallGraph] Teach the CallGraph about non-leaf intrinsics.
>>>
>>> Summary:
>>> Currently intrinsics don't affect the creation of the call graph.
>>> This is not accurate with respect to statepoint and patchpoint
>>> intrinsics -- these do call (or invoke) LLVM level functions.
>>>
>>> This change fixes this inconsistency by adding a call to the external
>>> node for call sites that call these non-leaf intrinsics. This coupled
>>> with the fact that these intrinsics also escape the function pointer
>>> they call gives us a conservatively correct call graph.
>>>
>>> Reviewers: reames, chandlerc, atrick, pgavlin
>>>
>>> Subscribers: llvm-commits
>>>
>>> Differential Revision: http://reviews.llvm.org/D10526
>>>
>>> Added:
>>> llvm/trunk/test/Analysis/CallGraph/non-leaf-intrinsics.ll
>>> Modified:
>>> llvm/trunk/include/llvm/Analysis/CallGraph.h
>>> llvm/trunk/include/llvm/IR/Intrinsics.h
>>> llvm/trunk/lib/Analysis/IPA/CallGraph.cpp
>>> llvm/trunk/lib/Analysis/IPA/CallGraphSCCPass.cpp
>>> llvm/trunk/lib/IR/Function.cpp
>>>
>>> Modified: llvm/trunk/include/llvm/Analysis/CallGraph.h
>>> URL:
>>> http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Analysis/CallGraph.h?rev=240039&r1=240038&r2=240039&view=diff
>>> ==============================================================================
>>> --- llvm/trunk/include/llvm/Analysis/CallGraph.h (original)
>>> +++ llvm/trunk/include/llvm/Analysis/CallGraph.h Thu Jun 18 14:28:26 2015
>>> @@ -56,6 +56,7 @@
>>> #include "llvm/ADT/STLExtras.h"
>>> #include "llvm/IR/CallSite.h"
>>> #include "llvm/IR/Function.h"
>>> +#include "llvm/IR/Intrinsics.h"
>>> #include "llvm/IR/ValueHandle.h"
>>> #include "llvm/Pass.h"
>>> #include <map>
>>> @@ -229,7 +230,8 @@ public:
>>> /// \brief Adds a function to the list of functions called by this one.
>>> void addCalledFunction(CallSite CS, CallGraphNode *M) {
>>> assert(!CS.getInstruction() || !CS.getCalledFunction() ||
>>> - !CS.getCalledFunction()->isIntrinsic());
>>> + !CS.getCalledFunction()->isIntrinsic() ||
>>> + !Intrinsic::isLeaf(CS.getCalledFunction()->getIntrinsicID()));
>>> CalledFunctions.emplace_back(CS.getInstruction(), M);
>>> M->AddRef();
>>> }
>>>
>>> Modified: llvm/trunk/include/llvm/IR/Intrinsics.h
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/IR/Intrinsics.h?rev=240039&r1=240038&r2=240039&view=diff
>>> ==============================================================================
>>> --- llvm/trunk/include/llvm/IR/Intrinsics.h (original)
>>> +++ llvm/trunk/include/llvm/IR/Intrinsics.h Thu Jun 18 14:28:26 2015
>>> @@ -52,6 +52,11 @@ namespace Intrinsic {
>>> /// Returns true if the intrinsic can be overloaded.
>>> bool isOverloaded(ID id);
>>>
>>> + /// Returns true if the intrinsic is a leaf, i.e. it does not make any calls
>>> + /// itself. Most intrinsics are leafs, the exceptions being the patchpoint
>>> + /// and statepoint intrinsics. These call (or invoke) their "target" argument.
>>> + bool isLeaf(ID id);
>>> +
>>> /// Return the attributes for an intrinsic.
>>> AttributeSet getAttributes(LLVMContext &C, ID id);
>>>
>>>
>>> Modified: llvm/trunk/lib/Analysis/IPA/CallGraph.cpp
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/IPA/CallGraph.cpp?rev=240039&r1=240038&r2=240039&view=diff
>>> ==============================================================================
>>> --- llvm/trunk/lib/Analysis/IPA/CallGraph.cpp (original)
>>> +++ llvm/trunk/lib/Analysis/IPA/CallGraph.cpp Thu Jun 18 14:28:26 2015
>>> @@ -79,8 +79,10 @@ void CallGraph::addToCallGraph(Function
>>> CallSite CS(cast<Value>(II));
>>> if (CS) {
>>> const Function *Callee = CS.getCalledFunction();
>>> - if (!Callee)
>>> + if (!Callee || !Intrinsic::isLeaf(Callee->getIntrinsicID()))
>>> // Indirect calls of intrinsics are not allowed so no need to check.
>>> + // We can be more precise here by using TargetArg returned by
>>> + // Intrinsic::isLeaf.
>>> Node->addCalledFunction(CS, CallsExternalNode);
>>> else if (!Callee->isIntrinsic())
>>> Node->addCalledFunction(CS, getOrInsertFunction(Callee));
>>>
>>> Modified: llvm/trunk/lib/Analysis/IPA/CallGraphSCCPass.cpp
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/IPA/CallGraphSCCPass.cpp?rev=240039&r1=240038&r2=240039&view=diff
>>> ==============================================================================
>>> --- llvm/trunk/lib/Analysis/IPA/CallGraphSCCPass.cpp (original)
>>> +++ llvm/trunk/lib/Analysis/IPA/CallGraphSCCPass.cpp Thu Jun 18 14:28:26 2015
>>> @@ -217,8 +217,10 @@ bool CGPassManager::RefreshCallGraph(Cal
>>> // another value. This can happen when constant folding happens
>>> // of well known functions etc.
>>> !CallSite(I->first) ||
>>> - (CallSite(I->first).getCalledFunction() &&
>>> - CallSite(I->first).getCalledFunction()->isIntrinsic())) {
>>> + (CallSite(I->first).getCalledFunction() &&
>>> + CallSite(I->first).getCalledFunction()->isIntrinsic() &&
>>> + Intrinsic::isLeaf(
>>> + CallSite(I->first).getCalledFunction()->getIntrinsicID()))) {
>>> assert(!CheckingMode &&
>>> "CallGraphSCCPass did not update the CallGraph correctly!");
>>>
>>>
>>> Modified: llvm/trunk/lib/IR/Function.cpp
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/IR/Function.cpp?rev=240039&r1=240038&r2=240039&view=diff
>>> ==============================================================================
>>> --- llvm/trunk/lib/IR/Function.cpp (original)
>>> +++ llvm/trunk/lib/IR/Function.cpp Thu Jun 18 14:28:26 2015
>>> @@ -846,6 +846,18 @@ bool Intrinsic::isOverloaded(ID id) {
>>> #undef GET_INTRINSIC_OVERLOAD_TABLE
>>> }
>>>
>>> +bool Intrinsic::isLeaf(ID id) {
>>> + switch (id) {
>>> + default:
>>> + return true;
>>> +
>>> + case Intrinsic::experimental_gc_statepoint:
>>> + case Intrinsic::experimental_patchpoint_void:
>>> + case Intrinsic::experimental_patchpoint_i64:
>>> + return false;
>>> + }
>>> +}
>>> +
>>> /// This defines the "Intrinsic::getAttributes(ID id)" method.
>>> #define GET_INTRINSIC_ATTRIBUTES
>>> #include "llvm/IR/Intrinsics.gen"
>>>
>>> Added: llvm/trunk/test/Analysis/CallGraph/non-leaf-intrinsics.ll
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Analysis/CallGraph/non-leaf-intrinsics.ll?rev=240039&view=auto
>>> ==============================================================================
>>> --- llvm/trunk/test/Analysis/CallGraph/non-leaf-intrinsics.ll (added)
>>> +++ llvm/trunk/test/Analysis/CallGraph/non-leaf-intrinsics.ll Thu Jun 18 14:28:26 2015
>>> @@ -0,0 +1,32 @@
>>> +; RUN: opt -S -print-callgraph -disable-output < %s 2>&1 | FileCheck %s
>>> +
>>> +declare void @llvm.experimental.patchpoint.void(i64, i32, i8*, i32, ...)
>>> +declare i32 @llvm.experimental.gc.statepoint.p0f_isVoidf(i64, i32, void ()*, i32, i32, ...)
>>> +
>>> +define private void @f() {
>>> + ret void
>>> +}
>>> +
>>> +define void @calls_statepoint(i8 addrspace(1)* %arg) gc "statepoint-example" {
>>> +entry:
>>> + %cast = bitcast i8 addrspace(1)* %arg to i64 addrspace(1)*
>>> + %safepoint_token = call i32 (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 0, i32 0, void ()* @f, i32 0, i32 0, i32 0, i32 5, i32 0, i32 0, i32 0, i32 10, i32 0, i8 addrspace(1)* %arg, i64 addrspace(1)* %cast, i8 addrspace(1)* %arg, i8 addrspace(1)* %arg)
>>> + ret void
>>> +}
>>> +
>>> +define void @calls_patchpoint() {
>>> +entry:
>>> + %c = bitcast void()* @f to i8*
>>> + tail call void (i64, i32, i8*, i32, ...) @llvm.experimental.patchpoint.void(i64 1, i32 15, i8* %c, i32 0, i16 65535, i16 -1, i32 65536, i32 2000000000, i32 2147483647, i32 -1, i32 4294967295, i32 4294967296, i64 2147483648, i64 4294967295, i64 4294967296, i64 -1)
>>> + ret void
>>> +}
>>> +
>>> +
>>> +; CHECK: Call graph node <<null function>>
>>> +; CHECK: CS<0x0> calls function 'f'
>>> +
>>> +; CHECK: Call graph node for function: 'calls_statepoint'
>>> +; CHECK-NEXT: CS<[[addr_0:[^>]+]]> calls external node
>>> +
>>> +; CHECK: Call graph node for function: 'calls_patchpoint'
>>> +; CHECK-NEXT: CS<[[addr_1:[^>]+]]> calls external node
>>
>> The order that these call graph nodes are emitted seems to change when I
>> run the tests under asan+ubsan. That is, the test outputs:
>>
>> Call graph node <<null function>><<0x6040000041d0>> #uses=0
>> CS<0x0> calls function 'llvm.experimental.patchpoint.void'
>> CS<0x0> calls function 'llvm.experimental.gc.statepoint.p0f_isVoidf'
>> CS<0x0> calls function 'f'
>> CS<0x0> calls function 'calls_statepoint'
>> CS<0x0> calls function 'calls_patchpoint'
>>
>> Call graph node for function: 'calls_patchpoint'<<0x604000003f10>> #uses=1
>> CS<0x61500000f218> calls external node
>>
>> Call graph node for function: 'calls_statepoint'<<0x604000003f90>> #uses=1
>> CS<0x61500000f498> calls external node
>>
>> Does it make sense loosen this test to accept the output in either
>> order, or does this look like an issue with -print-callgraph?
>>
>>>
>>> _______________________________________________
>>> llvm-commits mailing list
>>> llvm-commits at cs.uiuc.edu
>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
More information about the llvm-commits
mailing list