[SVE] Fix LoopVectorizer test scalalable-call.ll

[sdesmalen]

Summary

This marks FSIN and other operations to EXPAND for scalable
vectors, so that they are not assumed to be legal by the cost-model.

Depends on D97470

[dmgreen]

Are codegen test for these impossible? If so can costmodel tests be added, showing invalid costs?

[sdesmalen]

Added tests showing unknown costs.

The only two ISD nodes not covered with cost tests are:

• SINCOS as this can't be expressed in IR directly.
• FREM as this requires getArithmeticInstrCost to be migrated to InstructionCost.

[dmgreen]

Thanks for the fix. LGTM.

(We might want to consider change it from "Unknown" to "Invalid", if the meaning is now that the cost is essentially infinite.)

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

for (auto VT :
       setCondCodeAction(ISD::SETOLE, VT, Expand);
       setCondCodeAction(ISD::SETLE, VT, Expand);
       setCondCodeAction(ISD::SETULT, VT, Expand);
       setCondCodeAction(ISD::SETULE, VT, Expand);
       setCondCodeAction(ISD::SETUGE, VT, Expand);
       setCondCodeAction(ISD::SETUGT, VT, Expand);
       setCondCodeAction(ISD::SETUEQ, VT, Expand);
       setCondCodeAction(ISD::SETUNE, VT, Expand);
+
+      setOperationAction(ISD::FREM, VT, Expand);
+      setOperationAction(ISD::FPOW, VT, Expand);
+      setOperationAction(ISD::FPOWI, VT, Expand);
+      setOperationAction(ISD::FCOS, VT, Expand);
+      setOperationAction(ISD::FSIN, VT, Expand);
+      setOperationAction(ISD::FSINCOS, VT, Expand);
+      setOperationAction(ISD::FEXP, VT, Expand);
+      setOperationAction(ISD::FEXP2, VT, Expand);
+      setOperationAction(ISD::FLOG, VT, Expand);
+      setOperationAction(ISD::FLOG2, VT, Expand);
+      setOperationAction(ISD::FLOG10, VT, Expand);
     }
   }
 
   // Compute derived properties from the register classes
   computeRegisterProperties(Subtarget->getRegisterInfo());
 
   // Provide all sorts of operation actions
   setOperationAction(ISD::GlobalAddress, MVT::i64, Custom);

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

void InnerLoopVectorizer::widenCallInstruction(CallInst &I, VPValue *Def,
   // version of the instruction.
   // Is it beneficial to perform intrinsic call compared to lib call?
   bool NeedToScalarize = false;
   InstructionCost CallCost = Cost->getVectorCallCost(CI, VF, NeedToScalarize);
   InstructionCost IntrinsicCost = ID ? Cost->getVectorIntrinsicCost(CI, VF) : 0;
   bool UseVectorIntrinsic = ID && IntrinsicCost <= CallCost;
   assert((UseVectorIntrinsic || !NeedToScalarize) &&
          "Instruction should be scalarized elsewhere.");
-  assert(IntrinsicCost.isValid() && CallCost.isValid() &&
-         "Cannot have invalid costs while widening");
+  assert((IntrinsicCost.isValid() || CallCost.isValid()) &&
+         "Either the intrinsic cost or vector call cost must be valid");
 
   for (unsigned Part = 0; Part < UF; ++Part) {
     SmallVector<Value *, 4> Args;
     for (auto &I : enumerate(ArgOperands.operands())) {
       // Some intrinsics have a scalar argument - don't replace it with a
       // vector.
       Value *Arg;
       if (!UseVectorIntrinsic || !hasVectorInstrinsicScalarOpd(ID, I.index()))
auto willWiden = [&](ElementCount VF) -> bool {
     // The following case may be scalarized depending on the VF.
     // The flag shows whether we use Intrinsic or a usual Call for vectorized
     // version of the instruction.
     // Is it beneficial to perform intrinsic call compared to lib call?
     bool NeedToScalarize = false;
     InstructionCost CallCost = CM.getVectorCallCost(CI, VF, NeedToScalarize);
     InstructionCost IntrinsicCost = ID ? CM.getVectorIntrinsicCost(CI, VF) : 0;
     bool UseVectorIntrinsic = ID && IntrinsicCost <= CallCost;
-    assert(IntrinsicCost.isValid() && CallCost.isValid() &&
-           "Cannot have invalid costs while widening");
+    assert((IntrinsicCost.isValid() || CallCost.isValid()) &&
+           "Either the intrinsic cost or vector call cost must be valid");
     return UseVectorIntrinsic || !NeedToScalarize;
   };
 
   if (!LoopVectorizationPlanner::getDecisionAndClampRange(willWiden, Range))
     return nullptr;
 
   return new VPWidenCallRecipe(*CI, Plan.mapToVPValues(CI->arg_operands()));
 }

llvm/test/Analysis/CostModel/AArch64/sve-intrinsics.ll

 declare <vscale x 16 x half> @llvm.experimental.vector.reverse.nxv16f16(<vscale x 16 x half>)
 declare <vscale x 4 x float> @llvm.experimental.vector.reverse.nxv4f32(<vscale x 4 x float>)
 declare <vscale x 8 x float> @llvm.experimental.vector.reverse.nxv8f32(<vscale x 8 x float>)
 declare <vscale x 2 x double> @llvm.experimental.vector.reverse.nxv2f64(<vscale x 2 x double>)
 declare <vscale x 4 x double> @llvm.experimental.vector.reverse.nxv4f64(<vscale x 4 x double>)
 declare <vscale x 8 x bfloat> @llvm.experimental.vector.reverse.nxv8bf16(<vscale x 8 x bfloat>)
 declare <vscale x 16 x bfloat> @llvm.experimental.vector.reverse.nxv16bf16(<vscale x 16 x bfloat>)
 
+define void @unsupported_fp_ops(<vscale x 4 x float> %vec) {
+; CHECK-LABEL: 'unsupported_fp_ops'
+; CHECK-NEXT: Cost Model: Unknown cost for instruction: %sin = call <vscale x 4 x float> @llvm.sin.nxv4f32(<vscale x 4 x float> %vec)
+; CHECK-NEXT: Cost Model: Unknown cost for instruction: %cos = call <vscale x 4 x float> @llvm.cos.nxv4f32(<vscale x 4 x float> %vec)
+; CHECK-NEXT: Cost Model: Unknown cost for instruction: %pow = call <vscale x 4 x float> @llvm.pow.nxv4f32(<vscale x 4 x float> %vec, <vscale x 4 x float> %vec)
+; CHECK-NEXT: Cost Model: Unknown cost for instruction: %powi = call <vscale x 4 x float> @llvm.powi.nxv4f32(<vscale x 4 x float> %vec, i32 42)
+; CHECK-NEXT: Cost Model: Unknown cost for instruction: %exp = call <vscale x 4 x float> @llvm.exp.nxv4f32(<vscale x 4 x float> %vec)
+; CHECK-NEXT: Cost Model: Unknown cost for instruction: %exp2 = call <vscale x 4 x float> @llvm.exp2.nxv4f32(<vscale x 4 x float> %vec)
+; CHECK-NEXT: Cost Model: Unknown cost for instruction: %log = call <vscale x 4 x float> @llvm.log.nxv4f32(<vscale x 4 x float> %vec)
+; CHECK-NEXT: Cost Model: Unknown cost for instruction: %log2 = call <vscale x 4 x float> @llvm.log2.nxv4f32(<vscale x 4 x float> %vec)
+; CHECK-NEXT: Cost Model: Unknown cost for instruction: %log10 = call <vscale x 4 x float> @llvm.log10.nxv4f32(<vscale x 4 x float> %vec)
+
+  %sin = call <vscale x 4 x float> @llvm.sin.nxv4f32(<vscale x 4 x float> %vec)
+  %cos = call <vscale x 4 x float> @llvm.cos.nxv4f32(<vscale x 4 x float> %vec)
+  %pow = call <vscale x 4 x float> @llvm.pow.nxv4f32(<vscale x 4 x float> %vec, <vscale x 4 x float> %vec)
+  %powi = call <vscale x 4 x float> @llvm.powi.nxv4f32(<vscale x 4 x float> %vec, i32 42)
+  %exp = call <vscale x 4 x float> @llvm.exp.nxv4f32(<vscale x 4 x float> %vec)
+  %exp2 = call <vscale x 4 x float> @llvm.exp2.nxv4f32(<vscale x 4 x float> %vec)
+  %log = call <vscale x 4 x float> @llvm.log.nxv4f32(<vscale x 4 x float> %vec)
+  %log2 = call <vscale x 4 x float> @llvm.log2.nxv4f32(<vscale x 4 x float> %vec)
+  %log10 = call <vscale x 4 x float> @llvm.log10.nxv4f32(<vscale x 4 x float> %vec)
+  ret void
+}
+
+declare <vscale x 4 x float> @llvm.sin.nxv4f32(<vscale x 4 x float>)
+declare <vscale x 4 x float> @llvm.cos.nxv4f32(<vscale x 4 x float>)
+declare <vscale x 4 x float> @llvm.pow.nxv4f32(<vscale x 4 x float>, <vscale x 4 x float>)
+declare <vscale x 4 x float> @llvm.powi.nxv4f32(<vscale x 4 x float>, i32)
+declare <vscale x 4 x float> @llvm.exp.nxv4f32(<vscale x 4 x float>)
+declare <vscale x 4 x float> @llvm.exp2.nxv4f32(<vscale x 4 x float>)
+declare <vscale x 4 x float> @llvm.log.nxv4f32(<vscale x 4 x float>)
+declare <vscale x 4 x float> @llvm.log2.nxv4f32(<vscale x 4 x float>)
+declare <vscale x 4 x float> @llvm.log10.nxv4f32(<vscale x 4 x float>)
+
 attributes #0 = { "target-features"="+sve,+bf16" }

llvm/test/Transforms/LoopVectorize/AArch64/scalable-call.ll

for.body:
   %exitcond = icmp eq i64 %iv.next, 1024
   br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !1
 
 for.end:
   ret void
 }
 
 define void @vec_intrinsic(i64 %N, double* nocapture readonly %a) {
-;; FIXME: Should be calling sin_vec, once the cost of scalarizing is handled.
 ; CHECK-LABEL: @vec_intrinsic
 ; CHECK: vector.body:
 ; CHECK: %[[LOAD:.*]] = load <vscale x 2 x double>, <vscale x 2 x double>*
-; CHECK: call fast <vscale x 2 x double> @llvm.sin.nxv2f64(<vscale x 2 x double> %[[LOAD]])
+; CHECK: call fast <vscale x 2 x double> @sin_vec(<vscale x 2 x double> %[[LOAD]])
 entry:
   %cmp7 = icmp sgt i64 %N, 0
   br i1 %cmp7, label %for.body, label %for.end
 
 for.body:
   %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
   %arrayidx = getelementptr inbounds double, double* %a, i64 %iv
   %0 = load double, double* %arrayidx, align 8