[PowerPC] Increase the user cost of vector instructions by their legalization cost

[bnemanich]

Summary

The cost of splitting a large vector instruction is not being taken into account by the getUserCost function. This was leading to some loops being over unrolled. The cost of a vector instruction is now being multiplied by the cost of the type legalization. This will return a more accurate cost.

[echristo]

lib/Target/PowerPC/PPCTargetTransformInfo.cpp

194–197 ↗

(On Diff #119171)

I have concerns about this being Power only. And to what costs we're applying this to, etc.

Thoughts?

[bnemanich]

lib/Target/PowerPC/PPCTargetTransformInfo.cpp

194–197 ↗

(On Diff #119171)

It could be done for all platforms, but I was worried that it would hurt some tuning that I wasn't able to test. This should only affect cases where there are vector instructions in the IR that are larger than the machine can handle. In those cases, getUserCost will not give a more accurate description of the cost. For example, if there is a 16 wide add of i32s, getUserCost will now return a value of 4 instead of 1, since it will eventually be split into 4 instructions on a machine that only has 128bit wide vector instructions.

[echristo]

I guess number of instructions is rather hard coded in. I'd definitely like to see this across other targets, but it works for now.

Thanks!

-eric

llvm/trunk/lib/Target/PowerPC/PPCTargetTransformInfo.h

public:
 
   using BaseT::getIntImmCost;
   int getIntImmCost(const APInt &Imm, Type *Ty);
 
   int getIntImmCost(unsigned Opcode, unsigned Idx, const APInt &Imm, Type *Ty);
   int getIntImmCost(Intrinsic::ID IID, unsigned Idx, const APInt &Imm,
                     Type *Ty);
 
+  unsigned getUserCost(const User *U, ArrayRef<const Value *> Operands);
+
   TTI::PopcntSupportKind getPopcntSupport(unsigned TyWidth);
   void getUnrollingPreferences(Loop *L, ScalarEvolution &SE,
                                TTI::UnrollingPreferences &UP);
 
   /// @}
 
   /// \name Vector TTI Implementations
   /// @{

llvm/trunk/lib/Target/PowerPC/PPCTargetTransformInfo.cpp

if (Idx == ImmIdx && Imm.getBitWidth() <= 64) {
 
     if (ShiftedFree && (Imm.getZExtValue() & 0xFFFF) == 0)
       return TTI::TCC_Free;
   }
 
   return PPCTTIImpl::getIntImmCost(Imm, Ty);
 }
 
+unsigned PPCTTIImpl::getUserCost(const User *U,
+                                 ArrayRef<const Value *> Operands) {
+  if (U->getType()->isVectorTy()) {
+    // Instructions that need to be split should cost more.
+    std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, U->getType());
+    return LT.first * BaseT::getUserCost(U, Operands);
+  }
+  
+  return BaseT::getUserCost(U, Operands);
+}
+
 void PPCTTIImpl::getUnrollingPreferences(Loop *L, ScalarEvolution &SE,
                                          TTI::UnrollingPreferences &UP) {
   if (ST->getDarwinDirective() == PPC::DIR_A2) {
     // The A2 is in-order with a deep pipeline, and concatenation unrolling
     // helps expose latency-hiding opportunities to the instruction scheduler.
     UP.Partial = UP.Runtime = true;
 
     // We unroll a lot on the A2 (hundreds of instructions), and the benefits

llvm/trunk/test/Transforms/LoopUnroll/PowerPC/p8-unrolling-legalize-vectors.ll

+; RUN: opt < %s -S -mtriple=powerpc64le-unknown-linux-gnu -mcpu=pwr8 -loop-unroll | FileCheck %s
+; RUN: opt < %s -S -mtriple=powerpc64le-unknown-linux-gnu -mcpu=pwr9 -loop-unroll | FileCheck %s
+
+target datalayout = "e-m:e-i64:64-n32:64"
+target triple = "powerpc64le-unknown-linux-gnu"
+
+; Function Attrs: norecurse nounwind
+define i8* @f(i8* returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_addr #0 {
+entry:
+  %cmp10 = icmp sgt i32 %k, 0
+  br i1 %cmp10, label %for.body.lr.ph, label %for.end
+
+for.body.lr.ph:                                   ; preds = %entry
+  %wide.trip.count = zext i32 %k to i64
+  %min.iters.check = icmp ult i32 %k, 16
+  br i1 %min.iters.check, label %for.body.preheader, label %vector.ph
+
+vector.ph:                                        ; preds = %for.body.lr.ph
+  %n.vec = and i64 %wide.trip.count, 4294967280
+  %broadcast.splatinsert = insertelement <16 x i32> undef, i32 %x, i32 0
+  %broadcast.splat = shufflevector <16 x i32> %broadcast.splatinsert, <16 x i32> undef, <16 x i32> zeroinitializer
+  br label %vector.body
+
+vector.body:                                      ; preds = %vector.body, %vector.ph
+  %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
+  %vec.ind12 = phi <16 x i32> [ <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>, %vector.ph ], [ %vec.ind.next13, %vector.body ]
+  %0 = shl <16 x i32> <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>, %vec.ind12
+  %1 = and <16 x i32> %0, %broadcast.splat
+  %2 = icmp eq <16 x i32> %1, zeroinitializer
+  %3 = select <16 x i1> %2, <16 x i8> <i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48>, <16 x i8> <i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49>
+  %4 = getelementptr inbounds i8, i8* %s, i64 %index
+  %5 = bitcast i8* %4 to <16 x i8>*
+  store <16 x i8> %3, <16 x i8>* %5, align 1
+  %index.next = add i64 %index, 16
+  %vec.ind.next13 = add <16 x i32> %vec.ind12, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+  %6 = icmp eq i64 %index.next, %n.vec
+  br i1 %6, label %middle.block, label %vector.body
+
+middle.block:                                     ; preds = %vector.body
+  %cmp.n = icmp eq i64 %n.vec, %wide.trip.count
+  br i1 %cmp.n, label %for.end, label %for.body.preheader
+
+for.body.preheader:                               ; preds = %middle.block, %for.body.lr.ph
+  %indvars.iv.ph = phi i64 [ 0, %for.body.lr.ph ], [ %n.vec, %middle.block ]
+  br label %for.body
+
+for.body:                                         ; preds = %for.body.preheader, %for.body
+  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ %indvars.iv.ph, %for.body.preheader ]
+  %7 = trunc i64 %indvars.iv to i32
+  %shl = shl i32 1, %7
+  %and = and i32 %shl, %x
+  %tobool = icmp eq i32 %and, 0
+  %conv = select i1 %tobool, i8 48, i8 49
+  %arrayidx = getelementptr inbounds i8, i8* %s, i64 %indvars.iv
+  store i8 %conv, i8* %arrayidx, align 1
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond = icmp eq i64 %indvars.iv.next, %wide.trip.count
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body, %middle.block, %entry
+  %idxprom1 = sext i32 %k to i64
+  %arrayidx2 = getelementptr inbounds i8, i8* %s, i64 %idxprom1
+  store i8 0, i8* %arrayidx2, align 1
+  ret i8* %s
+}
+
+
+; CHECK-LABEL: vector.body
+; CHECK:      shl
+; CHECK-NEXT: and
+; CHECK: shl
+; CHECK-NEXT: and
+; CHECK: label %vector.body
+