[llvm] c963bde - [CostModel] remove cost-kind predicate for scatter/gather cost

[Sanjay Patel via llvm-commits]

Author: Sanjay Patel
Date: 2020-10-21T14:26:05-04:00
New Revision: c963bde0152a741d58b9e1e9ac485d8f2e6c6d58

URL: https://github.com/llvm/llvm-project/commit/c963bde0152a741d58b9e1e9ac485d8f2e6c6d58
DIFF: https://github.com/llvm/llvm-project/commit/c963bde0152a741d58b9e1e9ac485d8f2e6c6d58.diff

LOG: [CostModel] remove cost-kind predicate for scatter/gather cost

This is similar in spirit to 01ea93d85d6e (memcpy) except that
here the underlying caller assumptions were created for vectorizer
use (throughput) rather than other passes.

That meant ARM could have an enormous throughput cost with no
corresponding size, latency, or blended cost increase. X86 has
the same throughput restriction as the basic implementation, so
it is still unchanged.

Paraphrasing from the previous commit:
This may not make sense for some callers, but at least now the
costs will be consistently wrong instead of mysteriously wrong.

Targets should provide better overrides if the current modeling
is not accurate.

Added: 
    

Modified: 
    llvm/include/llvm/CodeGen/BasicTTIImpl.h
    llvm/test/Analysis/CostModel/ARM/intrinsic-cost-kinds.ll

Removed: 
    


################################################################################
diff  --git a/llvm/include/llvm/CodeGen/BasicTTIImpl.h b/llvm/include/llvm/CodeGen/BasicTTIImpl.h
index 2eec38bbcc5d..817cabd6344e 100644
--- a/llvm/include/llvm/CodeGen/BasicTTIImpl.h
+++ b/llvm/include/llvm/CodeGen/BasicTTIImpl.h
@@ -1169,9 +1169,6 @@ class BasicTTIImplBase : public TargetTransformInfoImplCRTPBase<T> {
       return thisT()->getMemcpyCost(ICA.getInst());
 
     case Intrinsic::masked_scatter: {
-      // FIXME: all cost kinds should default to the same thing?
-      if (CostKind != TTI::TCK_RecipThroughput)
-        return BaseT::getIntrinsicInstrCost(ICA, CostKind);
       assert(VF == 1 && "Can't vectorize types here.");
       const Value *Mask = Args[3];
       bool VarMask = !isa<Constant>(Mask);
@@ -1181,9 +1178,6 @@ class BasicTTIImplBase : public TargetTransformInfoImplCRTPBase<T> {
                                              VarMask, Alignment, CostKind, I);
     }
     case Intrinsic::masked_gather: {
-      // FIXME: all cost kinds should default to the same thing?
-      if (CostKind != TTI::TCK_RecipThroughput)
-        return BaseT::getIntrinsicInstrCost(ICA, CostKind);
       assert(VF == 1 && "Can't vectorize types here.");
       const Value *Mask = Args[2];
       bool VarMask = !isa<Constant>(Mask);

diff  --git a/llvm/test/Analysis/CostModel/ARM/intrinsic-cost-kinds.ll b/llvm/test/Analysis/CostModel/ARM/intrinsic-cost-kinds.ll
index 9e02d6f149ca..bffaa98c82aa 100644
--- a/llvm/test/Analysis/CostModel/ARM/intrinsic-cost-kinds.ll
+++ b/llvm/test/Analysis/CostModel/ARM/intrinsic-cost-kinds.ll
@@ -171,11 +171,11 @@ define void @maskedgather(<16 x float*> %va, <16 x i1> %vb, <16 x float> %vc) {
 ; LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
 ; SIZE-LABEL: 'maskedgather'
-; SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> %va, i32 1, <16 x i1> %vb, <16 x float> %vc)
+; SIZE-NEXT:  Cost Model: Found an estimated cost of 576 for instruction: %v = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> %va, i32 1, <16 x i1> %vb, <16 x float> %vc)
 ; SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
 ; SIZE_LATE-LABEL: 'maskedgather'
-; SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> %va, i32 1, <16 x i1> %vb, <16 x float> %vc)
+; SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 576 for instruction: %v = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> %va, i32 1, <16 x i1> %vb, <16 x float> %vc)
 ; SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   %v = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> %va, i32 1, <16 x i1> %vb, <16 x float> %vc)
@@ -192,11 +192,11 @@ define void @maskedscatter(<16 x float> %va, <16 x float*> %vb, <16 x i1> %vc) {
 ; LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
 ; SIZE-LABEL: 'maskedscatter'
-; SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> %va, <16 x float*> %vb, i32 1, <16 x i1> %vc)
+; SIZE-NEXT:  Cost Model: Found an estimated cost of 576 for instruction: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> %va, <16 x float*> %vb, i32 1, <16 x i1> %vc)
 ; SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
 ; SIZE_LATE-LABEL: 'maskedscatter'
-; SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> %va, <16 x float*> %vb, i32 1, <16 x i1> %vc)
+; SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 576 for instruction: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> %va, <16 x float*> %vb, i32 1, <16 x i1> %vc)
 ; SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> %va, <16 x float*> %vb, i32 1, <16 x i1> %vc)