PreISelIntrinsicLowering: Lower llvm.log to a loop if scalable vec arg (#129744)

Similar to ab976a1, but for llvm.log.

Author: steplong

llvm/lib/CodeGen/PreISelIntrinsicLowering.cpp

@@ -587,12 +587,14 @@ bool PreISelIntrinsicLowering::lowerIntrinsics(Module &M) const {
       break;
     case Intrinsic::exp:
     case Intrinsic::exp2:
+    case Intrinsic::log:
       Changed |= forEachCall(F, [&](CallInst *CI) {
         Type *Ty = CI->getArgOperand(0)->getType();
         if (!isa<ScalableVectorType>(Ty))
           return false;
         const TargetLowering *TL = TM->getSubtargetImpl(F)->getTargetLowering();
         unsigned Op = TL->IntrinsicIDToISD(F.getIntrinsicID());
+        assert(Op != ISD::DELETED_NODE && "unsupported intrinsic");
         if (!TL->isOperationExpand(Op, EVT::getEVT(Ty)))
           return false;
         return lowerUnaryVectorIntrinsicAsLoop(M, CI);

llvm/lib/CodeGen/TargetLoweringBase.cpp

@@ -1924,6 +1924,8 @@ int TargetLoweringBase::IntrinsicIDToISD(Intrinsic::ID ID) const {
     return ISD::FEXP;
   case Intrinsic::exp2:
     return ISD::FEXP2;
+  case Intrinsic::log:
+    return ISD::FLOG;
   default:
     return ISD::DELETED_NODE;
   }

llvm/test/Transforms/PreISelIntrinsicLowering/AArch64/expand-log.ll

@@ -0,0 +1,40 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -passes=pre-isel-intrinsic-lowering -S < %s | FileCheck %s
+target triple = "aarch64"
+
+define <vscale x 4 x float> @scalable_vec_log(<vscale x 4 x float> %input) {
+; CHECK-LABEL: define <vscale x 4 x float> @scalable_vec_log(
+; CHECK-SAME: <vscale x 4 x float> [[INPUT:%.*]]) {
+; CHECK-NEXT:    [[TMP1:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT:    [[TMP2:%.*]] = mul nuw i64 [[TMP1]], 4
+; CHECK-NEXT:    br label %[[BB3:.*]]
+; CHECK:       [[BB3]]:
+; CHECK-NEXT:    [[TMP4:%.*]] = phi i64 [ 0, [[TMP0:%.*]] ], [ [[TMP9:%.*]], %[[BB3]] ]
+; CHECK-NEXT:    [[TMP5:%.*]] = phi <vscale x 4 x float> [ [[INPUT]], [[TMP0]] ], [ [[TMP8:%.*]], %[[BB3]] ]
+; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <vscale x 4 x float> [[TMP5]], i64 [[TMP4]]
+; CHECK-NEXT:    [[TMP7:%.*]] = call float @llvm.log.f32(float [[TMP6]])
+; CHECK-NEXT:    [[TMP8]] = insertelement <vscale x 4 x float> [[TMP5]], float [[TMP7]], i64 [[TMP4]]
+; CHECK-NEXT:    [[TMP9]] = add i64 [[TMP4]], 1
+; CHECK-NEXT:    [[TMP10:%.*]] = icmp eq i64 [[TMP9]], [[TMP2]]
+; CHECK-NEXT:    br i1 [[TMP10]], label %[[BB11:.*]], label %[[BB3]]
+; CHECK:       [[BB11]]:
+; CHECK-NEXT:    ret <vscale x 4 x float> [[TMP8]]
+;
+  %output = call <vscale x 4 x float> @llvm.log.nxv4f32(<vscale x 4 x float> %input)
+  ret <vscale x 4 x float> %output
+}
+
+define <4 x float> @fixed_vec_log(<4 x float> %input) {
+; CHECK-LABEL: define <4 x float> @fixed_vec_log(
+; CHECK-SAME: <4 x float> [[INPUT:%.*]]) {
+; CHECK-NEXT:    [[OUTPUT:%.*]] = call <4 x float> @llvm.log.v4f32(<4 x float> [[INPUT]])
+; CHECK-NEXT:    ret <4 x float> [[OUTPUT]]
+;
+  %output = call <4 x float> @llvm.log.v4f32(<4 x float> %input)
+  ret <4 x float> %output
+}
+
+declare <4 x float> @llvm.log.v4f32(<4 x float>) #0
+declare <vscale x 4 x float> @llvm.log.nxv4f32(<vscale x 4 x float>) #0
+
+attributes #0 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }