[RISCV] Scalarize binop followed by extractelement.

[jacquesguan]

Summary

This patch adds shouldScalarizeBinop to RISCV target in order to convert an extract element of a vector binary operation into an extract element followed by a scalar binary operation.
For now, the DAGCombiner only supports to combine the fixed vectors. I will change it in the next revision.

[reames]

LGTM.

This is a straight forward port of the X86 logic, and is a reasonable stepping stone. However, I think framing this in terms of legality leaves some potential cases on the floor. Specifically, we can have cases where the scalar op isn't legal, but is still cheaper than the legal vector op.

An example might be a 32 bit udiv. Masking off the high bits requires two extra instructions, but that's probably still profitable over doing the vector op on a long vector.

Long term, I'd propose we frame this in terms of the expected cost of the two forms, not the legality of either side directly.

[jacquesguan]

ping.

[reames]

In D129545#3670364, @jacquesguan wrote:

ping.

This has already been approved. Do you have commit access or do you need someone to land on your behalf?

[craig.topper]

In D129545#3670596, @reames wrote:

In D129545#3670364, @jacquesguan wrote:

ping.

This has already been approved. Do you have commit access or do you need someone to land on your behalf?

Phabricator does not say it’s approved. Guess you missed changing the status?

[reames]

In D129545#3670606, @craig.topper wrote:

In D129545#3670596, @reames wrote:

In D129545#3670364, @jacquesguan wrote:

ping.

This has already been approved. Do you have commit access or do you need someone to land on your behalf?

Phabricator does not say it’s approved. Guess you missed changing the status?

Per our dev policy, review status is not authoritative. I'd previous given an "LGTM". That means "approved". This is long standing policy.

[craig.topper]

LGTM

llvm/lib/Target/RISCV/RISCVISelLowering.h

public:
   bool hasAndNotCompare(SDValue Y) const override;
   bool hasBitTest(SDValue X, SDValue Y) const override;
   bool shouldProduceAndByConstByHoistingConstFromShiftsLHSOfAnd(
       SDValue X, ConstantSDNode *XC, ConstantSDNode *CC, SDValue Y,
       unsigned OldShiftOpcode, unsigned NewShiftOpcode,
       SelectionDAG &DAG) const override;
   bool shouldSinkOperands(Instruction *I,
                           SmallVectorImpl<Use *> &Ops) const override;
+  bool shouldScalarizeBinop(SDValue VecOp) const override;
   bool isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const override;
   bool isFPImmLegal(const APFloat &Imm, EVT VT,
                     bool ForCodeSize) const override;
 
   bool softPromoteHalfType() const override { return true; }
 
   /// Return the register type for a given MVT, ensuring vectors are treated
   /// as a series of gpr sized integers.

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

for (auto OpIdx : enumerate(I->operands())) {
     }
 
     Ops.push_back(&Op->getOperandUse(0));
     Ops.push_back(&OpIdx.value());
   }
   return true;
 }
 
+bool RISCVTargetLowering::shouldScalarizeBinop(SDValue VecOp) const {
+  unsigned Opc = VecOp.getOpcode();
+
+  // Assume target opcodes can't be scalarized.
+  // TODO - do we have any exceptions?
+  if (Opc >= ISD::BUILTIN_OP_END)
+    return false;
+
+  // If the vector op is not supported, try to convert to scalar.
+  EVT VecVT = VecOp.getValueType();
+  if (!isOperationLegalOrCustomOrPromote(Opc, VecVT))
+    return true;
+
+  // If the vector op is supported, but the scalar op is not, the transform may
+  // not be worthwhile.
+  EVT ScalarVT = VecVT.getScalarType();
+  return isOperationLegalOrCustomOrPromote(Opc, ScalarVT);
+}
+
 bool RISCVTargetLowering::isOffsetFoldingLegal(
     const GlobalAddressSDNode *GA) const {
   // In order to maximise the opportunity for common subexpression elimination,
   // keep a separate ADD node for the global address offset instead of folding
   // it in the global address node. Later peephole optimisations may choose to
   // fold it back in when profitable.
   return false;
 }

llvm/test/CodeGen/RISCV/rvv/fixed-vectors-extract.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=riscv32 -target-abi=ilp32d -mattr=+v,+zfh,+experimental-zvfh,+f,+d -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV32
+; RUN: llc -mtriple=riscv32 -target-abi=ilp32d -mattr=+v,+zfh,+experimental-zvfh,+f,+d -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV32,RV32NOM
+; RUN: llc -mtriple=riscv32 -target-abi=ilp32d -mattr=+v,+zfh,+experimental-zvfh,+f,+d,+m -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV32,RV32M
 ; RUN: llc -mtriple=riscv64 -target-abi=lp64d -mattr=+v,+zfh,+experimental-zvfh,+f,+d -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV64
+; RUN: llc -mtriple=riscv64 -target-abi=lp64d -mattr=+v,+zfh,+experimental-zvfh,+f,+d,+m -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV64
 
 define i8 @extractelt_v16i8(<16 x i8>* %x) nounwind {
 ; CHECK-LABEL: extractelt_v16i8:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vsetivli zero, 16, e8, m1, ta, mu
 ; CHECK-NEXT:    vle8.v v8, (a0)
 ; CHECK-NEXT:    vsetivli zero, 1, e8, m1, ta, mu
 ; CHECK-NEXT:    vslidedown.vi v8, v8, 7

 ; CHECK-NEXT:    ret
   %a = load <2 x double>, <2 x double>* %x
   %b = extractelement <2 x double> %a, i64 1
   store double %b, double* %p
   ret void
 }
 
 define i32 @extractelt_add_v4i32(<4 x i32> %x) {
-; CHECK-LABEL: extractelt_add_v4i32:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
-; CHECK-NEXT:    vadd.vi v8, v8, 13
-; CHECK-NEXT:    vsetivli zero, 1, e32, m1, ta, mu
-; CHECK-NEXT:    vslidedown.vi v8, v8, 2
-; CHECK-NEXT:    vmv.x.s a0, v8
-; CHECK-NEXT:    ret
+; RV32-LABEL: extractelt_add_v4i32:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli zero, 1, e32, m1, ta, mu
+; RV32-NEXT:    vslidedown.vi v8, v8, 2
+; RV32-NEXT:    vmv.x.s a0, v8
+; RV32-NEXT:    addi a0, a0, 13
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: extractelt_add_v4i32:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
+; RV64-NEXT:    vadd.vi v8, v8, 13
+; RV64-NEXT:    vsetivli zero, 1, e32, m1, ta, mu
+; RV64-NEXT:    vslidedown.vi v8, v8, 2
+; RV64-NEXT:    vmv.x.s a0, v8
+; RV64-NEXT:    ret
   %bo = add <4 x i32> %x, <i32 11, i32 12, i32 13, i32 14>
   %ext = extractelement <4 x i32> %bo, i32 2
   ret i32 %ext
 }
 
 define i32 @extractelt_sub_v4i32(<4 x i32> %x) {
-; CHECK-LABEL: extractelt_sub_v4i32:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
-; CHECK-NEXT:    vrsub.vi v8, v8, 13
-; CHECK-NEXT:    vsetivli zero, 1, e32, m1, ta, mu
-; CHECK-NEXT:    vslidedown.vi v8, v8, 2
-; CHECK-NEXT:    vmv.x.s a0, v8
-; CHECK-NEXT:    ret
+; RV32-LABEL: extractelt_sub_v4i32:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli zero, 1, e32, m1, ta, mu
+; RV32-NEXT:    vslidedown.vi v8, v8, 2
+; RV32-NEXT:    vmv.x.s a0, v8
+; RV32-NEXT:    li a1, 13
+; RV32-NEXT:    sub a0, a1, a0
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: extractelt_sub_v4i32:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
+; RV64-NEXT:    vrsub.vi v8, v8, 13
+; RV64-NEXT:    vsetivli zero, 1, e32, m1, ta, mu
+; RV64-NEXT:    vslidedown.vi v8, v8, 2
+; RV64-NEXT:    vmv.x.s a0, v8
+; RV64-NEXT:    ret
   %bo = sub <4 x i32> <i32 11, i32 12, i32 13, i32 14>, %x
   %ext = extractelement <4 x i32> %bo, i32 2
   ret i32 %ext
 }
 
 define i32 @extractelt_mul_v4i32(<4 x i32> %x) {
-; CHECK-LABEL: extractelt_mul_v4i32:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    li a0, 13
-; CHECK-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
-; CHECK-NEXT:    vmul.vx v8, v8, a0
-; CHECK-NEXT:    vsetivli zero, 1, e32, m1, ta, mu
-; CHECK-NEXT:    vslidedown.vi v8, v8, 2
-; CHECK-NEXT:    vmv.x.s a0, v8
-; CHECK-NEXT:    ret
+; RV32NOM-LABEL: extractelt_mul_v4i32:
+; RV32NOM:       # %bb.0:
+; RV32NOM-NEXT:    li a0, 13
+; RV32NOM-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
+; RV32NOM-NEXT:    vmul.vx v8, v8, a0
+; RV32NOM-NEXT:    vsetivli zero, 1, e32, m1, ta, mu
+; RV32NOM-NEXT:    vslidedown.vi v8, v8, 2
+; RV32NOM-NEXT:    vmv.x.s a0, v8
+; RV32NOM-NEXT:    ret
+;
+; RV32M-LABEL: extractelt_mul_v4i32:
+; RV32M:       # %bb.0:
+; RV32M-NEXT:    vsetivli zero, 1, e32, m1, ta, mu
+; RV32M-NEXT:    vslidedown.vi v8, v8, 2
+; RV32M-NEXT:    vmv.x.s a0, v8
+; RV32M-NEXT:    li a1, 13
+; RV32M-NEXT:    mul a0, a0, a1
+; RV32M-NEXT:    ret
+;
+; RV64-LABEL: extractelt_mul_v4i32:
+; RV64:       # %bb.0:
+; RV64-NEXT:    li a0, 13
+; RV64-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
+; RV64-NEXT:    vmul.vx v8, v8, a0
+; RV64-NEXT:    vsetivli zero, 1, e32, m1, ta, mu
+; RV64-NEXT:    vslidedown.vi v8, v8, 2
+; RV64-NEXT:    vmv.x.s a0, v8
+; RV64-NEXT:    ret
   %bo = mul <4 x i32> %x, <i32 11, i32 12, i32 13, i32 14>
   %ext = extractelement <4 x i32> %bo, i32 2
   ret i32 %ext
 }
 
 define i32 @extractelt_sdiv_v4i32(<4 x i32> %x) {
-; RV32-LABEL: extractelt_sdiv_v4i32:
-; RV32:       # %bb.0:
-; RV32-NEXT:    li a0, -1
-; RV32-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
-; RV32-NEXT:    vmv.s.x v9, a0
-; RV32-NEXT:    vmv.v.i v10, 0
-; RV32-NEXT:    vsetvli zero, zero, e32, m1, tu, mu
-; RV32-NEXT:    vslideup.vi v10, v9, 3
-; RV32-NEXT:    vsetvli zero, zero, e32, m1, ta, mu
-; RV32-NEXT:    lui a0, %hi(.LCPI38_0)
-; RV32-NEXT:    addi a0, a0, %lo(.LCPI38_0)
-; RV32-NEXT:    vle32.v v9, (a0)
-; RV32-NEXT:    lui a0, %hi(.LCPI38_1)
-; RV32-NEXT:    addi a0, a0, %lo(.LCPI38_1)
-; RV32-NEXT:    vle32.v v11, (a0)
-; RV32-NEXT:    vand.vv v10, v8, v10
-; RV32-NEXT:    vmulh.vv v8, v8, v9
-; RV32-NEXT:    vadd.vv v8, v8, v10
-; RV32-NEXT:    vsra.vv v9, v8, v11
-; RV32-NEXT:    vsrl.vi v8, v8, 31
-; RV32-NEXT:    vadd.vv v8, v9, v8
-; RV32-NEXT:    vsetivli zero, 1, e32, m1, ta, mu
-; RV32-NEXT:    vslidedown.vi v8, v8, 2
-; RV32-NEXT:    vmv.x.s a0, v8
-; RV32-NEXT:    ret
+; RV32NOM-LABEL: extractelt_sdiv_v4i32:
+; RV32NOM:       # %bb.0:
+; RV32NOM-NEXT:    li a0, -1
+; RV32NOM-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
+; RV32NOM-NEXT:    vmv.s.x v9, a0
+; RV32NOM-NEXT:    vmv.v.i v10, 0
+; RV32NOM-NEXT:    vsetvli zero, zero, e32, m1, tu, mu
+; RV32NOM-NEXT:    vslideup.vi v10, v9, 3
+; RV32NOM-NEXT:    vsetvli zero, zero, e32, m1, ta, mu
+; RV32NOM-NEXT:    lui a0, %hi(.LCPI38_0)
+; RV32NOM-NEXT:    addi a0, a0, %lo(.LCPI38_0)
+; RV32NOM-NEXT:    vle32.v v9, (a0)
+; RV32NOM-NEXT:    lui a0, %hi(.LCPI38_1)
+; RV32NOM-NEXT:    addi a0, a0, %lo(.LCPI38_1)
+; RV32NOM-NEXT:    vle32.v v11, (a0)
+; RV32NOM-NEXT:    vand.vv v10, v8, v10
+; RV32NOM-NEXT:    vmulh.vv v8, v8, v9
+; RV32NOM-NEXT:    vadd.vv v8, v8, v10
+; RV32NOM-NEXT:    vsra.vv v9, v8, v11
+; RV32NOM-NEXT:    vsrl.vi v8, v8, 31
+; RV32NOM-NEXT:    vadd.vv v8, v9, v8
+; RV32NOM-NEXT:    vsetivli zero, 1, e32, m1, ta, mu
+; RV32NOM-NEXT:    vslidedown.vi v8, v8, 2
+; RV32NOM-NEXT:    vmv.x.s a0, v8
+; RV32NOM-NEXT:    ret
+;
+; RV32M-LABEL: extractelt_sdiv_v4i32:
+; RV32M:       # %bb.0:
+; RV32M-NEXT:    vsetivli zero, 1, e32, m1, ta, mu
+; RV32M-NEXT:    vslidedown.vi v8, v8, 2
+; RV32M-NEXT:    vmv.x.s a0, v8
+; RV32M-NEXT:    lui a1, 322639
+; RV32M-NEXT:    addi a1, a1, -945
+; RV32M-NEXT:    mulh a0, a0, a1
+; RV32M-NEXT:    srli a1, a0, 31
+; RV32M-NEXT:    srai a0, a0, 2
+; RV32M-NEXT:    add a0, a0, a1
+; RV32M-NEXT:    ret
 ;
 ; RV64-LABEL: extractelt_sdiv_v4i32:
 ; RV64:       # %bb.0:
 ; RV64-NEXT:    li a0, -1
 ; RV64-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
 ; RV64-NEXT:    vmv.s.x v9, a0
 ; RV64-NEXT:    vmv.v.i v10, 0
 ; RV64-NEXT:    vsetvli zero, zero, e32, m1, tu, mu

 ; RV64-NEXT:    vmv.x.s a0, v8
 ; RV64-NEXT:    ret
   %bo = sdiv <4 x i32> %x, <i32 11, i32 12, i32 13, i32 14>
   %ext = extractelement <4 x i32> %bo, i32 2
   ret i32 %ext
 }
 
 define i32 @extractelt_udiv_v4i32(<4 x i32> %x) {
-; CHECK-LABEL: extractelt_udiv_v4i32:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    li a0, 1
-; CHECK-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
-; CHECK-NEXT:    vmv.s.x v9, a0
-; CHECK-NEXT:    vmv.v.i v10, 0
-; CHECK-NEXT:    vsetvli zero, zero, e32, m1, tu, mu
-; CHECK-NEXT:    vslideup.vi v10, v9, 3
-; CHECK-NEXT:    vsetvli zero, zero, e32, m1, ta, mu
-; CHECK-NEXT:    lui a0, %hi(.LCPI39_0)
-; CHECK-NEXT:    addi a0, a0, %lo(.LCPI39_0)
-; CHECK-NEXT:    vle32.v v9, (a0)
-; CHECK-NEXT:    vsrl.vv v8, v8, v10
-; CHECK-NEXT:    vmulhu.vv v8, v8, v9
-; CHECK-NEXT:    vsrl.vi v8, v8, 2
-; CHECK-NEXT:    vsetivli zero, 1, e32, m1, ta, mu
-; CHECK-NEXT:    vslidedown.vi v8, v8, 2
-; CHECK-NEXT:    vmv.x.s a0, v8
-; CHECK-NEXT:    ret
+; RV32NOM-LABEL: extractelt_udiv_v4i32:
+; RV32NOM:       # %bb.0:
+; RV32NOM-NEXT:    li a0, 1
+; RV32NOM-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
+; RV32NOM-NEXT:    vmv.s.x v9, a0
+; RV32NOM-NEXT:    vmv.v.i v10, 0
+; RV32NOM-NEXT:    vsetvli zero, zero, e32, m1, tu, mu
+; RV32NOM-NEXT:    vslideup.vi v10, v9, 3
+; RV32NOM-NEXT:    vsetvli zero, zero, e32, m1, ta, mu
+; RV32NOM-NEXT:    lui a0, %hi(.LCPI39_0)
+; RV32NOM-NEXT:    addi a0, a0, %lo(.LCPI39_0)
+; RV32NOM-NEXT:    vle32.v v9, (a0)
+; RV32NOM-NEXT:    vsrl.vv v8, v8, v10
+; RV32NOM-NEXT:    vmulhu.vv v8, v8, v9
+; RV32NOM-NEXT:    vsetivli zero, 1, e32, m1, ta, mu
+; RV32NOM-NEXT:    vslidedown.vi v8, v8, 2
+; RV32NOM-NEXT:    vmv.x.s a0, v8
+; RV32NOM-NEXT:    srli a0, a0, 2
+; RV32NOM-NEXT:    ret
+;
+; RV32M-LABEL: extractelt_udiv_v4i32:
+; RV32M:       # %bb.0:
+; RV32M-NEXT:    vsetivli zero, 1, e32, m1, ta, mu
+; RV32M-NEXT:    vslidedown.vi v8, v8, 2
+; RV32M-NEXT:    vmv.x.s a0, v8
+; RV32M-NEXT:    lui a1, 322639
+; RV32M-NEXT:    addi a1, a1, -945
+; RV32M-NEXT:    mulhu a0, a0, a1
+; RV32M-NEXT:    srli a0, a0, 2
+; RV32M-NEXT:    ret
+;
+; RV64-LABEL: extractelt_udiv_v4i32:
+; RV64:       # %bb.0:
+; RV64-NEXT:    li a0, 1
+; RV64-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
+; RV64-NEXT:    vmv.s.x v9, a0
+; RV64-NEXT:    vmv.v.i v10, 0
+; RV64-NEXT:    vsetvli zero, zero, e32, m1, tu, mu
+; RV64-NEXT:    vslideup.vi v10, v9, 3
+; RV64-NEXT:    vsetvli zero, zero, e32, m1, ta, mu
+; RV64-NEXT:    lui a0, %hi(.LCPI39_0)
+; RV64-NEXT:    addi a0, a0, %lo(.LCPI39_0)
+; RV64-NEXT:    vle32.v v9, (a0)
+; RV64-NEXT:    vsrl.vv v8, v8, v10
+; RV64-NEXT:    vmulhu.vv v8, v8, v9
+; RV64-NEXT:    vsrl.vi v8, v8, 2
+; RV64-NEXT:    vsetivli zero, 1, e32, m1, ta, mu
+; RV64-NEXT:    vslidedown.vi v8, v8, 2
+; RV64-NEXT:    vmv.x.s a0, v8
+; RV64-NEXT:    ret
   %bo = udiv <4 x i32> %x, <i32 11, i32 12, i32 13, i32 14>
   %ext = extractelement <4 x i32> %bo, i32 2
   ret i32 %ext
 }
 
 define float @extractelt_fadd_v4f32(<4 x float> %x) {
 ; CHECK-LABEL: extractelt_fadd_v4f32:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    lui a0, %hi(.LCPI40_0)
 ; CHECK-NEXT:    flw ft0, %lo(.LCPI40_0)(a0)
-; CHECK-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
-; CHECK-NEXT:    vfadd.vf v8, v8, ft0
 ; CHECK-NEXT:    vsetivli zero, 1, e32, m1, ta, mu
 ; CHECK-NEXT:    vslidedown.vi v8, v8, 2
-; CHECK-NEXT:    vfmv.f.s fa0, v8
+; CHECK-NEXT:    vfmv.f.s ft1, v8
+; CHECK-NEXT:    fadd.s fa0, ft1, ft0
 ; CHECK-NEXT:    ret
   %bo = fadd <4 x float> %x, <float 11.0, float 12.0, float 13.0, float 14.0>
   %ext = extractelement <4 x float> %bo, i32 2
   ret float %ext
 }
 
 define float @extractelt_fsub_v4f32(<4 x float> %x) {
 ; CHECK-LABEL: extractelt_fsub_v4f32:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    lui a0, %hi(.LCPI41_0)
 ; CHECK-NEXT:    flw ft0, %lo(.LCPI41_0)(a0)
-; CHECK-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
-; CHECK-NEXT:    vfrsub.vf v8, v8, ft0
 ; CHECK-NEXT:    vsetivli zero, 1, e32, m1, ta, mu
 ; CHECK-NEXT:    vslidedown.vi v8, v8, 2
-; CHECK-NEXT:    vfmv.f.s fa0, v8
+; CHECK-NEXT:    vfmv.f.s ft1, v8
+; CHECK-NEXT:    fsub.s fa0, ft0, ft1
 ; CHECK-NEXT:    ret
   %bo = fsub <4 x float> <float 11.0, float 12.0, float 13.0, float 14.0>, %x
   %ext = extractelement <4 x float> %bo, i32 2
   ret float %ext
 }
 
 define float @extractelt_fmul_v4f32(<4 x float> %x) {
 ; CHECK-LABEL: extractelt_fmul_v4f32:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    lui a0, %hi(.LCPI42_0)
 ; CHECK-NEXT:    flw ft0, %lo(.LCPI42_0)(a0)
-; CHECK-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
-; CHECK-NEXT:    vfmul.vf v8, v8, ft0
 ; CHECK-NEXT:    vsetivli zero, 1, e32, m1, ta, mu
 ; CHECK-NEXT:    vslidedown.vi v8, v8, 2
-; CHECK-NEXT:    vfmv.f.s fa0, v8
+; CHECK-NEXT:    vfmv.f.s ft1, v8
+; CHECK-NEXT:    fmul.s fa0, ft1, ft0
 ; CHECK-NEXT:    ret
   %bo = fmul <4 x float> %x, <float 11.0, float 12.0, float 13.0, float 14.0>
   %ext = extractelement <4 x float> %bo, i32 2
   ret float %ext
 }
 
 define float @extractelt_fdiv_v4f32(<4 x float> %x) {
 ; CHECK-LABEL: extractelt_fdiv_v4f32:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    lui a0, %hi(.LCPI43_0)
 ; CHECK-NEXT:    flw ft0, %lo(.LCPI43_0)(a0)
-; CHECK-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
-; CHECK-NEXT:    vfdiv.vf v8, v8, ft0
 ; CHECK-NEXT:    vsetivli zero, 1, e32, m1, ta, mu
 ; CHECK-NEXT:    vslidedown.vi v8, v8, 2
-; CHECK-NEXT:    vfmv.f.s fa0, v8
+; CHECK-NEXT:    vfmv.f.s ft1, v8
+; CHECK-NEXT:    fdiv.s fa0, ft1, ft0
 ; CHECK-NEXT:    ret
   %bo = fdiv <4 x float> %x, <float 11.0, float 12.0, float 13.0, float 14.0>
   %ext = extractelement <4 x float> %bo, i32 2
   ret float %ext
 }