[llvm] ea7cc12 - [ARM] Add fallback fptoi_sat costs.

[David Green via llvm-commits]

Author: David Green
Date: 2024-07-28T23:38:21+01:00
New Revision: ea7cc12f612ef5c5f70082ebfdee378d80ec44bd

URL: https://github.com/llvm/llvm-project/commit/ea7cc12f612ef5c5f70082ebfdee378d80ec44bd
DIFF: https://github.com/llvm/llvm-project/commit/ea7cc12f612ef5c5f70082ebfdee378d80ec44bd.diff

LOG: [ARM] Add fallback fptoi_sat costs.

This makes sure that the custom operations get a fallback cost, even if they
are not perfect.

Added: 
    

Modified: 
    llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp b/llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp
index b5ca045058cbc..e940dce7faa14 100644
--- a/llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp
+++ b/llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp
@@ -1963,7 +1963,7 @@ ARMTTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
         LT.second.getScalarSizeInBits() == MTy.getScalarSizeInBits())
       return LT.first * ST->getMVEVectorCostFactor(CostKind);
 
-    // Otherwise we use a legal convert followed by a min+max
+    // If we can we use a legal convert followed by a min+max
     if (((ST->hasVFP2Base() && LT.second == MVT::f32) ||
          (ST->hasFP64() && LT.second == MVT::f64) ||
          (ST->hasFullFP16() && LT.second == MVT::f16) ||
@@ -1984,7 +1984,25 @@ ARMTTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
       Cost += getIntrinsicInstrCost(Attrs2, CostKind);
       return LT.first * Cost;
     }
-    break;
+    // Otherwise we need to follow the default expansion that clamps the value
+    // using a float min/max with a fcmp+sel for nan handling when signed.
+    Type *FPTy = ICA.getArgTypes()[0];
+    Type *RetTy = ICA.getReturnType();
+    IntrinsicCostAttributes Attrs1(Intrinsic::minnum, FPTy, {FPTy, FPTy});
+    InstructionCost Cost = getIntrinsicInstrCost(Attrs1, CostKind);
+    IntrinsicCostAttributes Attrs2(Intrinsic::maxnum, FPTy, {FPTy, FPTy});
+    Cost += getIntrinsicInstrCost(Attrs2, CostKind);
+    Cost +=
+        getCastInstrCost(IsSigned ? Instruction::FPToSI : Instruction::FPToUI,
+                         RetTy, FPTy, TTI::CastContextHint::None, CostKind);
+    if (IsSigned) {
+      Type *CondTy = RetTy->getWithNewBitWidth(1);
+      Cost += getCmpSelInstrCost(BinaryOperator::FCmp, FPTy, CondTy,
+                                 CmpInst::FCMP_UNO, CostKind);
+      Cost += getCmpSelInstrCost(BinaryOperator::Select, RetTy, CondTy,
+                                 CmpInst::FCMP_UNO, CostKind);
+    }
+    return Cost;
   }
   }