[AMDGPU] Tune inlining parameters for AMDGPU target

Summary:
Since the target has no significant advantage of vectorization,
vector instructions bous threshold bonus should be optional.

amdgpu-inline-arg-alloca-cost parameter default value and the target
InliningThresholdMultiplier value tuned then respectively.

Reviewers: arsenm, rampitec

Subscribers: kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye, eraman, hiraditya, haicheng, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D64642

llvm-svn: 366348

include/llvm/Analysis/TargetTransformInfo.h

@@ -263,6 +263,18 @@ public:
   /// individual classes of instructions would be better.
   unsigned getInliningThresholdMultiplier() const;
 
+  /// \returns Vector bonus in percent.
+  ///
+  /// Vector bonuses: We want to more aggressively inline vector-dense kernels
+  /// and apply this bonus based on the percentage of vector instructions. A
+  /// bonus is applied if the vector instructions exceed 50% and half that amount
+  /// is applied if it exceeds 10%. Note that these bonuses are some what
+  /// arbitrary and evolved over time by accident as much as because they are
+  /// principled bonuses.
+  /// FIXME: It would be nice to base the bonus values on something more
+  /// scientific. A target may has no bonus on vector instructions.
+  int getInlinerVectorBonusPercent() const;
+
   /// Estimate the cost of an intrinsic when lowered.
   ///
   /// Mirrors the \c getCallCost method but uses an intrinsic identifier.
@@ -1128,6 +1140,7 @@ public:
   virtual int getCallCost(const Function *F,
                           ArrayRef<const Value *> Arguments, const User *U) = 0;
   virtual unsigned getInliningThresholdMultiplier() = 0;
+  virtual int getInlinerVectorBonusPercent() = 0;
   virtual int getIntrinsicCost(Intrinsic::ID IID, Type *RetTy,
                                ArrayRef<Type *> ParamTys, const User *U) = 0;
   virtual int getIntrinsicCost(Intrinsic::ID IID, Type *RetTy,
@@ -1351,6 +1364,9 @@ public:
   unsigned getInliningThresholdMultiplier() override {
     return Impl.getInliningThresholdMultiplier();
   }
+  int getInlinerVectorBonusPercent() override {
+    return Impl.getInlinerVectorBonusPercent();
+  }
   int getIntrinsicCost(Intrinsic::ID IID, Type *RetTy,
                        ArrayRef<Type *> ParamTys, const User *U = nullptr) override {
     return Impl.getIntrinsicCost(IID, RetTy, ParamTys, U);

include/llvm/Analysis/TargetTransformInfoImpl.h

@@ -140,6 +140,8 @@ public:
 
   unsigned getInliningThresholdMultiplier() { return 1; }
 
+  int getInlinerVectorBonusPercent() { return 150; }
+
   unsigned getMemcpyCost(const Instruction *I) {
     return TTI::TCC_Expensive;
   }

include/llvm/CodeGen/BasicTTIImpl.h

@@ -427,6 +427,8 @@ public:
 
   unsigned getInliningThresholdMultiplier() { return 1; }
 
+  int getInlinerVectorBonusPercent() { return 150; }
+
   void getUnrollingPreferences(Loop *L, ScalarEvolution &SE,
                                TTI::UnrollingPreferences &UP) {
     // This unrolling functionality is target independent, but to provide some

lib/Analysis/InlineCost.cpp

@@ -880,15 +880,6 @@ void CallAnalyzer::updateThreshold(CallBase &Call, Function &Callee) {
   // basic block at the given callsite context. This is speculatively applied
   // and withdrawn if more than one basic block is seen.
   //
-  // Vector bonuses: We want to more aggressively inline vector-dense kernels
-  // and apply this bonus based on the percentage of vector instructions. A
-  // bonus is applied if the vector instructions exceed 50% and half that amount
-  // is applied if it exceeds 10%. Note that these bonuses are some what
-  // arbitrary and evolved over time by accident as much as because they are
-  // principled bonuses.
-  // FIXME: It would be nice to base the bonus values on something more
-  // scientific.
-  //
   // LstCallToStaticBonus: This large bonus is applied to ensure the inlining
   // of the last call to a static function as inlining such functions is
   // guaranteed to reduce code size.
@@ -896,7 +887,7 @@ void CallAnalyzer::updateThreshold(CallBase &Call, Function &Callee) {
   // These bonus percentages may be set to 0 based on properties of the caller
   // and the callsite.
   int SingleBBBonusPercent = 50;
-  int VectorBonusPercent = 150;
+  int VectorBonusPercent = TTI.getInlinerVectorBonusPercent();
   int LastCallToStaticBonus = InlineConstants::LastCallToStaticBonus;
 
   // Lambda to set all the above bonus and bonus percentages to 0.

lib/Analysis/TargetTransformInfo.cpp

@@ -176,6 +176,10 @@ unsigned TargetTransformInfo::getInliningThresholdMultiplier() const {
   return TTIImpl->getInliningThresholdMultiplier();
 }
 
+int TargetTransformInfo::getInlinerVectorBonusPercent() const {
+  return TTIImpl->getInlinerVectorBonusPercent();
+}
+
 int TargetTransformInfo::getGEPCost(Type *PointeeType, const Value *Ptr,
                                     ArrayRef<const Value *> Operands) const {
   return TTIImpl->getGEPCost(PointeeType, Ptr, Operands);

lib/Target/AMDGPU/AMDGPUInline.cpp

@@ -39,7 +39,7 @@ using namespace llvm;
 #define DEBUG_TYPE "inline"
 
 static cl::opt<int>
-ArgAllocaCost("amdgpu-inline-arg-alloca-cost", cl::Hidden, cl::init(2200),
+ArgAllocaCost("amdgpu-inline-arg-alloca-cost", cl::Hidden, cl::init(1500),
               cl::desc("Cost of alloca argument"));
 
 // If the amount of scratch memory to eliminate exceeds our ability to allocate

lib/Target/AMDGPU/AMDGPUTargetTransformInfo.h

@@ -191,7 +191,9 @@ public:
   bool areInlineCompatible(const Function *Caller,
                            const Function *Callee) const;
 
-  unsigned getInliningThresholdMultiplier() { return 9; }
+  unsigned getInliningThresholdMultiplier() { return 7; }
+
+  int getInlinerVectorBonusPercent() { return 0; }
 
   int getArithmeticReductionCost(unsigned Opcode,
                                  Type *Ty,

test/CodeGen/AMDGPU/amdgpu-inline.ll

@@ -28,15 +28,8 @@ if.end:                                           ; preds = %if.then, %entry
 define coldcc void @foo_private_ptr2(float addrspace(5)* nocapture %p1, float addrspace(5)* nocapture %p2) {
 entry:
   %tmp1 = load float, float addrspace(5)* %p1, align 4
-  %cmp = fcmp ogt float %tmp1, 1.000000e+00
-  br i1 %cmp, label %if.then, label %if.end
-
-if.then:                                          ; preds = %entry
   %div = fdiv float 2.000000e+00, %tmp1
   store float %div, float addrspace(5)* %p2, align 4
-  br label %if.end
-
-if.end:                                           ; preds = %if.then, %entry
   ret void
 }
 

test/Transforms/Inline/AMDGPU/inline-amdgpu-vecbonus.ll

@@ -0,0 +1,31 @@
+; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -amdgpu-inline --inline-threshold=1 < %s | FileCheck %s
+
+define hidden <16 x i32> @div_vecbonus(<16 x i32> %x, <16 x i32> %y) {
+entry:
+  %div.1 = udiv <16 x i32> %x, %y
+  %div.2 = udiv <16 x i32> %div.1, %y
+  %div.3 = udiv <16 x i32> %div.2, %y
+  %div.4 = udiv <16 x i32> %div.3, %y
+  %div.5 = udiv <16 x i32> %div.4, %y
+  %div.6 = udiv <16 x i32> %div.5, %y
+  %div.7 = udiv <16 x i32> %div.6, %y
+  %div.8 = udiv <16 x i32> %div.7, %y
+  %div.9 = udiv <16 x i32> %div.8, %y
+  %div.10 = udiv <16 x i32> %div.9, %y
+  %div.11 = udiv <16 x i32> %div.10, %y
+  %div.12 = udiv <16 x i32> %div.11, %y
+  ret <16 x i32> %div.12
+}
+
+; CHECK-LABEL: define amdgpu_kernel void @caller_vecbonus
+; CHECK-NOT: udiv
+; CHECK: tail call <16 x i32> @div_vecbonus
+; CHECK: ret void
+define amdgpu_kernel void @caller_vecbonus(<16 x i32> addrspace(1)* nocapture %x, <16 x i32> addrspace(1)* nocapture readonly %y) {
+entry:
+  %tmp = load <16 x i32>, <16 x i32> addrspace(1)* %x
+  %tmp1 = load <16 x i32>, <16 x i32> addrspace(1)* %y
+  %div.i = tail call <16 x i32> @div_vecbonus(<16 x i32> %tmp, <16 x i32> %tmp1)
+  store <16 x i32> %div.i, <16 x i32> addrspace(1)* %x
+  ret void
+}