[llvm] [RISCV][TTI] Scale the cost of FP-Int conversion with LMUL (PR #87506)

Shih-Po Hung via llvm-commits llvm-commits at lists.llvm.org
Wed Jul 24 01:07:41 PDT 2024


================
@@ -988,31 +988,97 @@ InstructionCost RISCVTTIImpl::getCastInstrCost(unsigned Opcode, Type *Dst,
     return Cost;
   }
   case ISD::FP_TO_SINT:
-  case ISD::FP_TO_UINT:
+  case ISD::FP_TO_UINT: {
+    unsigned IsSigned = ISD == ISD::FP_TO_SINT;
+    unsigned FCVT = IsSigned ? RISCV::VFCVT_RTZ_X_F_V : RISCV::VFCVT_RTZ_XU_F_V;
+    unsigned FWCVT =
+        IsSigned ? RISCV::VFWCVT_RTZ_X_F_V : RISCV::VFWCVT_RTZ_XU_F_V;
+    unsigned FNCVT =
+        IsSigned ? RISCV::VFNCVT_RTZ_X_F_W : RISCV::VFNCVT_RTZ_XU_F_W;
+    unsigned SrcEltSize = Src->getScalarSizeInBits();
+    unsigned DstEltSize = Dst->getScalarSizeInBits();
+    if ((SrcEltSize == 16) &&
+        (!ST->hasVInstructionsF16() || ((DstEltSize >> 1) > SrcEltSize))) {
+      // pre-widening to f32 and then convert f32 to integer
+      VectorType *VecF32Ty =
+          VectorType::get(Type::getFloatTy(Dst->getContext()),
+                          cast<VectorType>(Dst)->getElementCount());
+      std::pair<InstructionCost, MVT> VecF32LT =
+          getTypeLegalizationCost(VecF32Ty);
+      InstructionCost WidenCost = getRISCVInstructionCost(
+          RISCV::VFWCVT_F_F_V, VecF32LT.second, CostKind);
+      InstructionCost ConvCost =
+          getCastInstrCost(Opcode, Dst, VecF32Ty, CCH, CostKind, I);
+      return VecF32LT.first * WidenCost + ConvCost;
+    }
+    if (DstEltSize == SrcEltSize)
+      return getRISCVInstructionCost(FCVT, DstLT.second, CostKind);
+    if ((DstEltSize >> 1) == SrcEltSize)
+      return getRISCVInstructionCost(FWCVT, DstLT.second, CostKind);
+    InstructionCost TruncCost = 0;
+    if ((SrcEltSize >> 1) > DstEltSize) {
+      VectorType *VecTy =
+          VectorType::get(IntegerType::get(Dst->getContext(), SrcEltSize >> 1),
+                          cast<VectorType>(Dst)->getElementCount());
+      TruncCost =
+          getCastInstrCost(Instruction::Trunc, Dst, VecTy, CCH, CostKind, I);
+    }
+    if (SrcEltSize > DstEltSize) {
+      // First do a narrowing conversion to an integer half the size, then
+      // truncate if needed.
+      MVT ElementVT = MVT::getIntegerVT(SrcEltSize >> 1);
+      MVT VecVT = DstLT.second.changeVectorElementType(ElementVT);
+      InstructionCost ConvCost =
+          getRISCVInstructionCost(FNCVT, VecVT, CostKind);
+      return ConvCost + TruncCost;
+    }
+
+    return BaseT::getCastInstrCost(Opcode, Dst, Src, CCH, CostKind, I);
+  }
   case ISD::SINT_TO_FP:
-  case ISD::UINT_TO_FP:
-    if (Src->getScalarSizeInBits() == 1 || Dst->getScalarSizeInBits() == 1) {
-      // The cost of convert from or to mask vector is different from other
-      // cases. We could not use PowDiff to calculate it.
-      // For mask vector to fp, we should use the following instructions:
-      // vmv.v.i v8, 0
-      // vmerge.vim v8, v8, -1, v0
-      // vfcvt.f.x.v v8, v8
+  case ISD::UINT_TO_FP: {
+    unsigned IsSigned = ISD == ISD::SINT_TO_FP;
+    unsigned FCVT = IsSigned ? RISCV::VFCVT_F_X_V : RISCV::VFCVT_F_XU_V;
+    unsigned FWCVT = IsSigned ? RISCV::VFWCVT_F_X_V : RISCV::VFWCVT_F_XU_V;
+    unsigned FNCVT = IsSigned ? RISCV::VFNCVT_F_X_W : RISCV::VFNCVT_F_XU_W;
+    unsigned SrcEltSize = Src->getScalarSizeInBits();
+    unsigned DstEltSize = Dst->getScalarSizeInBits();
 
-      // And for fp vector to mask, we use:
-      // vfncvt.rtz.x.f.w v9, v8
-      // vand.vi v8, v9, 1
-      // vmsne.vi v0, v8, 0
-      return 3;
----------------
arcbbb wrote:

Conversions to mask vectors  (vfncvt.rtz.x.f + vand.vi +vmsne.vi ) are handled by a call to `getCastInstrCost(Instruction::Trunc,..)` which generates the cost of ` vand.vi +vmsne.vi` and a series of `getRISCVInstructionCost(FNCVT, ..)` in case ISD::FP_TO_*INT.
and conversions from mask vectors (vmv.v.i+vmerge.vim + vfcvt.f.x.v) are handled by `PreWidenCost  + getRISCVInstructionCost`. Prewiden cost `vmv.v.i+vmerge.vim` is handled by getCastInstrCost(SEXT|ZEXT, ..) in case ISD::*INT_TO_FP.

https://github.com/llvm/llvm-project/pull/87506


More information about the llvm-commits mailing list