Promote the bf16/f16 to fp32 to calculate the cost

Author: LiqinWeng

llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp

@@ -1057,18 +1057,22 @@ RISCVTTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
   }
   case Intrinsic::sqrt: {
     auto LT = getTypeLegalizationCost(RetTy);
+    auto NVT = LT.second;
     if (ST->hasVInstructions() && LT.second.isVector()) {
       SmallVector<unsigned, 3> Opcodes;
       // f16 with zvfhmin and bf16 with zvfbfmin
-      if (LT.second.getVectorElementType() == MVT::bf16)
+      if (LT.second.getVectorElementType() == MVT::bf16) {
         Opcodes = {RISCV::VFWCVTBF16_F_F_V, RISCV::VFSQRT_V,
                    RISCV::VFNCVTBF16_F_F_W};
-      else if (LT.second.getVectorElementType() == MVT::f16 &&
-               !ST->hasVInstructionsF16())
+        NVT = TLI->getTypeToPromoteTo(ISD::FSQRT, NVT);
+      } else if (LT.second.getVectorElementType() == MVT::f16 &&
+                 !ST->hasVInstructionsF16()) {
         Opcodes = {RISCV::VFWCVT_F_F_V, RISCV::VFSQRT_V, RISCV::VFNCVT_F_F_W};
-      else
+        NVT = TLI->getTypeToPromoteTo(ISD::FSQRT, NVT);
+      } else {
         Opcodes = {RISCV::VFSQRT_V};
-      return LT.first * getRISCVInstructionCost(Opcodes, LT.second, CostKind);
+      }
+      return LT.first * getRISCVInstructionCost(Opcodes, NVT, CostKind);
     }
     break;
   }

llvm/test/Analysis/CostModel/RISCV/fp-sqrt-pow.ll

@@ -7,12 +7,12 @@ define void @sqrt() {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %1 = call bfloat @llvm.sqrt.bf16(bfloat undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %2 = call <2 x bfloat> @llvm.sqrt.v2bf16(<2 x bfloat> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %3 = call <4 x bfloat> @llvm.sqrt.v4bf16(<4 x bfloat> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %4 = call <8 x bfloat> @llvm.sqrt.v8bf16(<8 x bfloat> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %5 = call <16 x bfloat> @llvm.sqrt.v16bf16(<16 x bfloat> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %4 = call <8 x bfloat> @llvm.sqrt.v8bf16(<8 x bfloat> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %5 = call <16 x bfloat> @llvm.sqrt.v16bf16(<16 x bfloat> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %6 = call <vscale x 2 x bfloat> @llvm.sqrt.nxv2bf16(<vscale x 2 x bfloat> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %7 = call <vscale x 4 x bfloat> @llvm.sqrt.nxv4bf16(<vscale x 4 x bfloat> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %8 = call <vscale x 8 x bfloat> @llvm.sqrt.nxv8bf16(<vscale x 8 x bfloat> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %9 = call <vscale x 16 x bfloat> @llvm.sqrt.nxv16bf16(<vscale x 16 x bfloat> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %7 = call <vscale x 4 x bfloat> @llvm.sqrt.nxv4bf16(<vscale x 4 x bfloat> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %8 = call <vscale x 8 x bfloat> @llvm.sqrt.nxv8bf16(<vscale x 8 x bfloat> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %9 = call <vscale x 16 x bfloat> @llvm.sqrt.nxv16bf16(<vscale x 16 x bfloat> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %10 = call float @llvm.sqrt.f32(float undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %11 = call <2 x float> @llvm.sqrt.v2f32(<2 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %12 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> undef)
@@ -82,12 +82,12 @@ define void @sqrt_f16() {
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %1 = call half @llvm.sqrt.f16(half undef)
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %2 = call <2 x half> @llvm.sqrt.v2f16(<2 x half> undef)
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %3 = call <4 x half> @llvm.sqrt.v4f16(<4 x half> undef)
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %4 = call <8 x half> @llvm.sqrt.v8f16(<8 x half> undef)
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %5 = call <16 x half> @llvm.sqrt.v16f16(<16 x half> undef)
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %4 = call <8 x half> @llvm.sqrt.v8f16(<8 x half> undef)
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %5 = call <16 x half> @llvm.sqrt.v16f16(<16 x half> undef)
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %6 = call <vscale x 2 x half> @llvm.sqrt.nxv2f16(<vscale x 2 x half> undef)
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %7 = call <vscale x 4 x half> @llvm.sqrt.nxv4f16(<vscale x 4 x half> undef)
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %8 = call <vscale x 8 x half> @llvm.sqrt.nxv8f16(<vscale x 8 x half> undef)
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %9 = call <vscale x 16 x half> @llvm.sqrt.nxv16f16(<vscale x 16 x half> undef)
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %7 = call <vscale x 4 x half> @llvm.sqrt.nxv4f16(<vscale x 4 x half> undef)
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %8 = call <vscale x 8 x half> @llvm.sqrt.nxv8f16(<vscale x 8 x half> undef)
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %9 = call <vscale x 16 x half> @llvm.sqrt.nxv16f16(<vscale x 16 x half> undef)
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   call half @llvm.sqrt.f16(half undef)