Enable recursion and convert all recursive calls to stack calls

Change-Id: Idd91b2406c5c3691280a3d4daf3d992cb9e59cff

Author: dlei6g

IGC/AdaptorCommon/AddImplicitArgs.cpp

@@ -517,59 +517,16 @@ BuiltinCallGraphAnalysis::BuiltinCallGraphAnalysis() : ModulePass(ID)
     initializeBuiltinCallGraphAnalysisPass(*PassRegistry::getPassRegistry());
 }
 
-/// Check whether there are recursions.
-static bool hasRecursion(CallGraph &CG)
-{
-    // Use Tarjan's algorithm to detect recursions.
-    for (auto I = scc_begin(&CG), E = scc_end(&CG); I != E; ++I)
-    {
-        const std::vector<CallGraphNode *> &SCCNodes = *I;
-        if (SCCNodes.size() >= 2)
-        {
-            return true;
-        }
-
-        // Check self-recursion.
-        auto Node = SCCNodes.back();
-        for (auto Callee : *Node)
-        {
-            if (Callee.second == Node)
-            {
-                return true;
-            }
-        }
-    }
-
-    // No recursion.
-    return false;
-}
-
 bool BuiltinCallGraphAnalysis::runOnModule(Module &M)
 {
     if (IGC_GET_FLAG_VALUE(FunctionControl) == FLAG_FCALL_FORCE_INLINE)
     {
         return false;
     }
 
-    CodeGenContext* ctx = getAnalysis<CodeGenContextWrapper>().getCodeGenContext();
     m_pMdUtils = getAnalysis<MetaDataUtilsWrapper>().getMetaDataUtils();
     CallGraph &CG = getAnalysis<CallGraphWrapperPass>().getCallGraph();
 
-    if (IGC_IS_FLAG_DISABLED(EnableRecursionOpenCL) &&
-        !ctx->m_DriverInfo.AllowRecursion() &&
-        hasRecursion(CG))
-    {
-        IGC_ASSERT_MESSAGE(0, "Recursion detected!");
-        ctx->EmitError(" undefined reference to `jmp()' ");
-        return false;
-    }
-
-    //Return if any error
-    if (!(ctx->oclErrorMessage.empty()))
-    {
-        return false;
-    }
-
     for (auto I = scc_begin(&CG), IE = scc_end(&CG); I != IE; ++I)
     {
         const std::vector<CallGraphNode *> &SCCNodes = *I;
@@ -614,7 +571,7 @@ void BuiltinCallGraphAnalysis::traveseCallGraphSCC(const std::vector<CallGraphNo
             if (argMapIter != argMap.end())
             {
                 IGC_ASSERT(nullptr != argMapIter->second);
-                combineTwoArgDetail(*argData, *(argMapIter->second), 
+                combineTwoArgDetail(*argData, *(argMapIter->second),
 #if LLVM_VERSION_MAJOR <= 10
                     N.first
 #else

IGC/AdaptorCommon/ProcessFuncAttributes.cpp

@@ -45,6 +45,8 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include <llvm/IR/Value.h>
 #include <llvm/IR/Attributes.h>
 #include <llvm/Support/raw_ostream.h>
+#include <llvm/Analysis/CallGraph.h>
+#include "llvm/ADT/SCCIterator.h"
 #include "common/LLVMWarningsPop.hpp"
 #include "common/igc_regkeys.hpp"
 #include <string>
@@ -66,6 +68,7 @@ class ProcessFuncAttributes : public ModulePass
         AU.addRequired<MetaDataUtilsWrapper>();
         AU.addRequired<CodeGenContextWrapper>();
         AU.addRequired<EstimateFunctionSize>();
+        AU.addRequired<llvm::CallGraphWrapperPass>();
     }
 
     ProcessFuncAttributes();
@@ -96,6 +99,7 @@ IGC_INITIALIZE_PASS_BEGIN(ProcessFuncAttributes, PASS_FLAG, PASS_DESCRIPTION, PA
 IGC_INITIALIZE_PASS_DEPENDENCY(MetaDataUtilsWrapper)
 IGC_INITIALIZE_PASS_DEPENDENCY(CodeGenContextWrapper)
 IGC_INITIALIZE_PASS_DEPENDENCY(EstimateFunctionSize)
+IGC_INITIALIZE_PASS_DEPENDENCY(CallGraphWrapperPass)
 IGC_INITIALIZE_PASS_END(ProcessFuncAttributes, PASS_FLAG, PASS_DESCRIPTION, PASS_CFG_ONLY, PASS_ANALYSIS)
 
 char ProcessFuncAttributes::ID = 0;
@@ -166,6 +170,37 @@ static bool containsOpaque(llvm::Type *T)
     return false;
 }
 
+// Convert functions with recursion to stackcall
+static bool convertRecursionToStackCall(CallGraph& CG, CodeGenContext* pCtx, IGCMD::MetaDataUtils* pM)
+{
+    bool hasRecursion = false;
+    // Use Tarjan's algorithm to detect recursions.
+    for (auto I = scc_begin(&CG), E = scc_end(&CG); I != E; ++I)
+    {
+        const std::vector<CallGraphNode*>& SCCNodes = *I;
+        if (SCCNodes.size() >= 2)
+        {
+            hasRecursion = true;
+            // Convert all functions in the recursion call graph to stackcall
+            for (auto Node : SCCNodes)
+            {
+                Node->getFunction()->addFnAttr("visaStackCall");
+            }
+        }
+        // Check self-recursion.
+        auto Node = SCCNodes.back();
+        for (auto Callee : *Node)
+        {
+            if (Callee.second == Node)
+            {
+                hasRecursion = true;
+                Node->getFunction()->addFnAttr("visaStackCall");
+            }
+        }
+    }
+    return hasRecursion;
+}
+
 // __builtin_spirv related OpGroup call implementations contain both
 // workgroup and subgroup code in them that is switched on based on the
 // 'Execution' and 'Operation' parameters and these will almost always
@@ -517,6 +552,18 @@ bool ProcessFuncAttributes::runOnModule(Module& M)
             }
         }
     }
+
+    // Detect recursive calls, and convert them to stack calls, since subroutines does not support recursion
+    CallGraph& CG = getAnalysis<CallGraphWrapperPass>().getCallGraph();
+    if (convertRecursionToStackCall(CG, pCtx, pMdUtils))
+    {
+        if (IGC_IS_FLAG_DISABLED(EnableRecursionOpenCL) && !pCtx->m_DriverInfo.AllowRecursion())
+        {
+            IGC_ASSERT_MESSAGE(0, "Recursion detected!");
+        }
+        pCtx->m_enableStackCall = true;
+    }
+
     return Changed;
 }
 

IGC/Compiler/CISACodeGen/GenCodeGenModule.cpp

@@ -194,20 +194,6 @@ inline Function* getCallerFunc(Value* user)
 
 void GenXCodeGenModule::processFunction(Function& F)
 {
-    // force stack-call for self-recursion
-    for (auto U : F.users())
-    {
-        if (CallInst * CI = dyn_cast<CallInst>(U))
-        {
-            Function* Caller = CI->getParent()->getParent();
-            if (Caller == &F)
-            {
-                F.addFnAttr("visaStackCall");
-                break;
-            }
-        }
-    }
-
     // See what FunctionGroups this Function is called from.
     SetVector<std::pair<FunctionGroup*, Function*>> CallerFGs;
     for (auto U : F.users())

IGC/common/igc_flags.def

@@ -300,7 +300,7 @@ DECLARE_IGC_REGKEY(bool, Enable64BitEmulation,          false, "Enable 64-bit em
 DECLARE_IGC_REGKEY(bool, Enable64BitEmulationOnSelectedPlatform, true, "Enable 64-bit emulation on selected platforms", false)
 DECLARE_IGC_REGKEY(DWORD, EnableConstIntDivReduction,   0x1,   "Enables strength reduction on integer division/remainder with constant divisors/moduli", true)
 DECLARE_IGC_REGKEY(DWORD, EnableIntDivRemCombine,       0x0,   "Given div/rem pairs with same operands merged; replace rem with mul+sub on quotient; 0x3 (set bit[1]) forces this on constant power of two divisors as well", true)
-DECLARE_IGC_REGKEY(bool, EnableRecursionOpenCL,         false, "Enable recursion with OpenCL user functions", false)
+DECLARE_IGC_REGKEY(bool, EnableRecursionOpenCL,         true,  "Enable recursion with OpenCL user functions", false)
 DECLARE_IGC_REGKEY(bool, ForceDPEmulation,              false, "Force double emulation for testing purpose", false)
 DECLARE_IGC_REGKEY(bool, EnableDPEmulation,             false, "Enforce double precision floating point operations emulation on platforms that do not support it natively", true)
 DECLARE_IGC_REGKEY(bool, DPEmuNeedI64Emu,               true,  "Double Emulation needs I64 emulation. Unsetting it to disable I64 Emulation for testing.", false)