[SVE] Fix incorrect predicate for fixed length int/fp conversion.

[paulwalker-arm]

Summary

When performing shrinking int/fp conversions the predicate should
be created to match the original fixed length vector type so the
unused lanes don't trigger side effects.

This patch also includes related refactoring to better detect such
issues and streamline the code a little.

[paulwalker-arm]

@hassnaa-arm: Just a heads up that this patch will affect the output for some of the tests you're adding as part of D136858.

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

23037

NOTE: Although using bitsGE maintains the original behaviour, I figure for the EQ case the else block is simpler and so might save a few cycles of compile time whilst having no affect on the generated code.

[david-arm]

LGTM! A work of art Paul. :)

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

unsigned Opcode = IsSigned ? AArch64ISD::SINT_TO_FP_MERGE_PASSTHRU
                              : AArch64ISD::UINT_TO_FP_MERGE_PASSTHRU;
 
   SDLoc DL(Op);
   SDValue Val = Op.getOperand(0);
   EVT SrcVT = Val.getValueType();
   EVT ContainerDstVT = getContainerForFixedLengthVector(DAG, VT);
   EVT ContainerSrcVT = getContainerForFixedLengthVector(DAG, SrcVT);
 
-  if (ContainerSrcVT.getVectorElementType().getSizeInBits() <=
-      ContainerDstVT.getVectorElementType().getSizeInBits()) {
-    SDValue Pg = getPredicateForVector(DAG, DL, VT);
+  if (VT.bitsGE(SrcVT)) {
+    SDValue Pg = getPredicateForFixedLengthVector(DAG, DL, VT);
 
     Val = DAG.getNode(IsSigned ? ISD::SIGN_EXTEND : ISD::ZERO_EXTEND, DL,
                       VT.changeTypeToInteger(), Val);
 
     Val = convertToScalableVector(DAG, ContainerSrcVT, Val);
     Val = getSVESafeBitCast(ContainerDstVT.changeTypeToInteger(), Val, DAG);
     // Safe to use a larger than specified operand since we just unpacked the
     // data, hence the upper bits are zero.
     Val = DAG.getNode(Opcode, DL, ContainerDstVT, Pg, Val,
                       DAG.getUNDEF(ContainerDstVT));
     return convertFromScalableVector(DAG, VT, Val);
   } else {
     EVT CvtVT = ContainerSrcVT.changeVectorElementType(
         ContainerDstVT.getVectorElementType());
-    SDValue Pg = getPredicateForVector(DAG, DL, CvtVT);
+    SDValue Pg = getPredicateForFixedLengthVector(DAG, DL, SrcVT);
 
     Val = convertToScalableVector(DAG, ContainerSrcVT, Val);
     Val = DAG.getNode(Opcode, DL, CvtVT, Pg, Val, DAG.getUNDEF(CvtVT));
     Val = getSVESafeBitCast(ContainerSrcVT, Val, DAG);
     Val = convertFromScalableVector(DAG, SrcVT, Val);
 
     Val = DAG.getNode(ISD::TRUNCATE, DL, VT.changeTypeToInteger(), Val);
     return DAG.getNode(ISD::BITCAST, DL, VT, Val);
unsigned Opcode = IsSigned ? AArch64ISD::FCVTZS_MERGE_PASSTHRU
                              : AArch64ISD::FCVTZU_MERGE_PASSTHRU;
 
   SDLoc DL(Op);
   SDValue Val = Op.getOperand(0);
   EVT SrcVT = Val.getValueType();
   EVT ContainerDstVT = getContainerForFixedLengthVector(DAG, VT);
   EVT ContainerSrcVT = getContainerForFixedLengthVector(DAG, SrcVT);
 
-  if (ContainerSrcVT.getVectorElementType().getSizeInBits() <=
-      ContainerDstVT.getVectorElementType().getSizeInBits()) {
+  if (VT.bitsGT(SrcVT)) {
     EVT CvtVT = ContainerDstVT.changeVectorElementType(
       ContainerSrcVT.getVectorElementType());
-    SDValue Pg = getPredicateForVector(DAG, DL, VT);
+    SDValue Pg = getPredicateForFixedLengthVector(DAG, DL, VT);
 
     Val = DAG.getNode(ISD::BITCAST, DL, SrcVT.changeTypeToInteger(), Val);
     Val = DAG.getNode(ISD::ANY_EXTEND, DL, VT, Val);
 
     Val = convertToScalableVector(DAG, ContainerSrcVT, Val);
     Val = getSVESafeBitCast(CvtVT, Val, DAG);
     Val = DAG.getNode(Opcode, DL, ContainerDstVT, Pg, Val,
                       DAG.getUNDEF(ContainerDstVT));
     return convertFromScalableVector(DAG, VT, Val);
   } else {
     EVT CvtVT = ContainerSrcVT.changeTypeToInteger();
-    SDValue Pg = getPredicateForVector(DAG, DL, CvtVT);
+    SDValue Pg = getPredicateForFixedLengthVector(DAG, DL, SrcVT);
 
     // Safe to use a larger than specified result since an fp_to_int where the
     // result doesn't fit into the destination is undefined.
     Val = convertToScalableVector(DAG, ContainerSrcVT, Val);
     Val = DAG.getNode(Opcode, DL, CvtVT, Pg, Val, DAG.getUNDEF(CvtVT));
     Val = convertFromScalableVector(DAG, SrcVT.changeTypeToInteger(), Val);
 
     return DAG.getNode(ISD::TRUNCATE, DL, VT, Val);

llvm/test/CodeGen/AArch64/sve-fixed-length-fp-to-int.ll

; CHECK-NEXT:    ret
   ret <4 x i16> %res
 }
 
 define <8 x i16> @fcvtzu_v8f32_v8i16(ptr %a) vscale_range(2,0) #0 {
 ; CHECK-LABEL: fcvtzu_v8f32_v8i16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.s, vl8
 ; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.s
 ; CHECK-NEXT:    fcvtzu z0.s, p0/m, z0.s
 ; CHECK-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; CHECK-NEXT:    // kill: def $q0 killed $q0 killed $z0
 ; CHECK-NEXT:    ret
   %op1 = load <8 x float>, ptr %a
   %res = fptoui <8 x float> %op1 to <8 x i16>
   ret <8 x i16> %res
 }
 
 define void @fcvtzu_v16f32_v16i16(ptr %a, ptr %b) #0 {
 ; VBITS_GE_256-LABEL: fcvtzu_v16f32_v16i16:
 ; VBITS_GE_256:       // %bb.0:
 ; VBITS_GE_256-NEXT:    mov x8, #8
 ; VBITS_GE_256-NEXT:    ptrue p0.s, vl8
 ; VBITS_GE_256-NEXT:    ld1w { z0.s }, p0/z, [x0, x8, lsl #2]
 ; VBITS_GE_256-NEXT:    ld1w { z1.s }, p0/z, [x0]
-; VBITS_GE_256-NEXT:    ptrue p0.s
 ; VBITS_GE_256-NEXT:    fcvtzu z0.s, p0/m, z0.s
 ; VBITS_GE_256-NEXT:    fcvtzu z1.s, p0/m, z1.s
 ; VBITS_GE_256-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; VBITS_GE_256-NEXT:    uzp1 z1.h, z1.h, z1.h
 ; VBITS_GE_256-NEXT:    ptrue p0.h, vl8
 ; VBITS_GE_256-NEXT:    splice z1.h, p0, z1.h, z0.h
 ; VBITS_GE_256-NEXT:    ptrue p0.h, vl16
 ; VBITS_GE_256-NEXT:    st1h { z1.h }, p0, [x1]
 ; VBITS_GE_256-NEXT:    ret
 ;
 ; VBITS_GE_512-LABEL: fcvtzu_v16f32_v16i16:
 ; VBITS_GE_512:       // %bb.0:
 ; VBITS_GE_512-NEXT:    ptrue p0.s, vl16
-; VBITS_GE_512-NEXT:    ptrue p1.s
 ; VBITS_GE_512-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; VBITS_GE_512-NEXT:    fcvtzu z0.s, p1/m, z0.s
+; VBITS_GE_512-NEXT:    fcvtzu z0.s, p0/m, z0.s
 ; VBITS_GE_512-NEXT:    st1h { z0.s }, p0, [x1]
 ; VBITS_GE_512-NEXT:    ret
   %op1 = load <16 x float>, ptr %a
   %res = fptoui <16 x float> %op1 to <16 x i16>
   store <16 x i16> %res, ptr %b
   ret void
 }
 
 define void @fcvtzu_v32f32_v32i16(ptr %a, ptr %b) vscale_range(8,0) #0 {
 ; CHECK-LABEL: fcvtzu_v32f32_v32i16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.s, vl32
-; CHECK-NEXT:    ptrue p1.s
 ; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    fcvtzu z0.s, p1/m, z0.s
+; CHECK-NEXT:    fcvtzu z0.s, p0/m, z0.s
 ; CHECK-NEXT:    st1h { z0.s }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <32 x float>, ptr %a
   %res = fptoui <32 x float> %op1 to <32 x i16>
   store <32 x i16> %res, ptr %b
   ret void
 }
 
 define void @fcvtzu_v64f32_v64i16(ptr %a, ptr %b) vscale_range(16,0) #0 {
 ; CHECK-LABEL: fcvtzu_v64f32_v64i16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.s, vl64
-; CHECK-NEXT:    ptrue p1.s
 ; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    fcvtzu z0.s, p1/m, z0.s
+; CHECK-NEXT:    fcvtzu z0.s, p0/m, z0.s
 ; CHECK-NEXT:    st1h { z0.s }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <64 x float>, ptr %a
   %res = fptoui <64 x float> %op1 to <64 x i16>
   store <64 x i16> %res, ptr %b
   ret void
 }
 

 ; FCVTZU D -> H
 ;
 
 ; v1f64 is perfered to be widened to v4f64, so use SVE
 define <1 x i16> @fcvtzu_v1f64_v1i16(<1 x double> %op1) vscale_range(2,0) #0 {
 ; CHECK-LABEL: fcvtzu_v1f64_v1i16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    // kill: def $d0 killed $d0 def $z0
-; CHECK-NEXT:    ptrue p0.d
+; CHECK-NEXT:    ptrue p0.d, vl4
 ; CHECK-NEXT:    fcvtzu z0.d, p0/m, z0.d
 ; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; CHECK-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $z0
 ; CHECK-NEXT:    ret
   %res = fptoui <1 x double> %op1 to <1 x i16>
   ret <1 x i16> %res
 }
; CHECK-NEXT:    ret
   ret <2 x i16> %res
 }
 
 define <4 x i16> @fcvtzu_v4f64_v4i16(ptr %a) vscale_range(2,0) #0 {
 ; CHECK-LABEL: fcvtzu_v4f64_v4i16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl4
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.d
 ; CHECK-NEXT:    fcvtzu z0.d, p0/m, z0.d
 ; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; CHECK-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $z0
 ; CHECK-NEXT:    ret
   %op1 = load <4 x double>, ptr %a
   %res = fptoui <4 x double> %op1 to <4 x i16>
   ret <4 x i16> %res
 }
 
 define <8 x i16> @fcvtzu_v8f64_v8i16(ptr %a) #0 {
 ; VBITS_GE_256-LABEL: fcvtzu_v8f64_v8i16:
 ; VBITS_GE_256:       // %bb.0:
 ; VBITS_GE_256-NEXT:    mov x8, #4
 ; VBITS_GE_256-NEXT:    ptrue p0.d, vl4
 ; VBITS_GE_256-NEXT:    ld1d { z0.d }, p0/z, [x0, x8, lsl #3]
 ; VBITS_GE_256-NEXT:    ld1d { z1.d }, p0/z, [x0]
-; VBITS_GE_256-NEXT:    ptrue p0.d
 ; VBITS_GE_256-NEXT:    fcvtzu z0.d, p0/m, z0.d
 ; VBITS_GE_256-NEXT:    fcvtzu z1.d, p0/m, z1.d
 ; VBITS_GE_256-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; VBITS_GE_256-NEXT:    uzp1 z1.s, z1.s, z1.s
 ; VBITS_GE_256-NEXT:    uzp1 z2.h, z0.h, z0.h
 ; VBITS_GE_256-NEXT:    uzp1 z0.h, z1.h, z1.h
 ; VBITS_GE_256-NEXT:    mov v0.d[1], v2.d[0]
 ; VBITS_GE_256-NEXT:    // kill: def $q0 killed $q0 killed $z0
 ; VBITS_GE_256-NEXT:    ret
 ;
 ; VBITS_GE_512-LABEL: fcvtzu_v8f64_v8i16:
 ; VBITS_GE_512:       // %bb.0:
 ; VBITS_GE_512-NEXT:    ptrue p0.d, vl8
 ; VBITS_GE_512-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; VBITS_GE_512-NEXT:    ptrue p0.d
 ; VBITS_GE_512-NEXT:    fcvtzu z0.d, p0/m, z0.d
 ; VBITS_GE_512-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; VBITS_GE_512-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; VBITS_GE_512-NEXT:    // kill: def $q0 killed $q0 killed $z0
 ; VBITS_GE_512-NEXT:    ret
   %op1 = load <8 x double>, ptr %a
   %res = fptoui <8 x double> %op1 to <8 x i16>
   ret <8 x i16> %res
 }
 
 define void @fcvtzu_v16f64_v16i16(ptr %a, ptr %b) vscale_range(8,0) #0 {
 ; CHECK-LABEL: fcvtzu_v16f64_v16i16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl16
-; CHECK-NEXT:    ptrue p1.d
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    fcvtzu z0.d, p1/m, z0.d
+; CHECK-NEXT:    fcvtzu z0.d, p0/m, z0.d
 ; CHECK-NEXT:    st1h { z0.d }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <16 x double>, ptr %a
   %res = fptoui <16 x double> %op1 to <16 x i16>
   store <16 x i16> %res, ptr %b
   ret void
 }
 
 define void @fcvtzu_v32f64_v32i16(ptr %a, ptr %b) vscale_range(16,0) #0 {
 ; CHECK-LABEL: fcvtzu_v32f64_v32i16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl32
-; CHECK-NEXT:    ptrue p1.d
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    fcvtzu z0.d, p1/m, z0.d
+; CHECK-NEXT:    fcvtzu z0.d, p0/m, z0.d
 ; CHECK-NEXT:    st1h { z0.d }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <32 x double>, ptr %a
   %res = fptoui <32 x double> %op1 to <32 x i16>
   store <32 x i16> %res, ptr %b
   ret void
 }
 
; CHECK-NEXT:    ret
   ret <2 x i32> %res
 }
 
 define <4 x i32> @fcvtzu_v4f64_v4i32(ptr %a) vscale_range(2,0) #0 {
 ; CHECK-LABEL: fcvtzu_v4f64_v4i32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl4
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.d
 ; CHECK-NEXT:    fcvtzu z0.d, p0/m, z0.d
 ; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; CHECK-NEXT:    // kill: def $q0 killed $q0 killed $z0
 ; CHECK-NEXT:    ret
   %op1 = load <4 x double>, ptr %a
   %res = fptoui <4 x double> %op1 to <4 x i32>
   ret <4 x i32> %res
 }
 
 define void @fcvtzu_v8f64_v8i32(ptr %a, ptr %b) #0 {
 ; VBITS_GE_256-LABEL: fcvtzu_v8f64_v8i32:
 ; VBITS_GE_256:       // %bb.0:
 ; VBITS_GE_256-NEXT:    mov x8, #4
 ; VBITS_GE_256-NEXT:    ptrue p0.d, vl4
 ; VBITS_GE_256-NEXT:    ld1d { z0.d }, p0/z, [x0, x8, lsl #3]
 ; VBITS_GE_256-NEXT:    ld1d { z1.d }, p0/z, [x0]
-; VBITS_GE_256-NEXT:    ptrue p0.d
 ; VBITS_GE_256-NEXT:    fcvtzu z0.d, p0/m, z0.d
 ; VBITS_GE_256-NEXT:    fcvtzu z1.d, p0/m, z1.d
 ; VBITS_GE_256-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; VBITS_GE_256-NEXT:    uzp1 z1.s, z1.s, z1.s
 ; VBITS_GE_256-NEXT:    ptrue p0.s, vl4
 ; VBITS_GE_256-NEXT:    splice z1.s, p0, z1.s, z0.s
 ; VBITS_GE_256-NEXT:    ptrue p0.s, vl8
 ; VBITS_GE_256-NEXT:    st1w { z1.s }, p0, [x1]
 ; VBITS_GE_256-NEXT:    ret
 ;
 ; VBITS_GE_512-LABEL: fcvtzu_v8f64_v8i32:
 ; VBITS_GE_512:       // %bb.0:
 ; VBITS_GE_512-NEXT:    ptrue p0.d, vl8
-; VBITS_GE_512-NEXT:    ptrue p1.d
 ; VBITS_GE_512-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; VBITS_GE_512-NEXT:    fcvtzu z0.d, p1/m, z0.d
+; VBITS_GE_512-NEXT:    fcvtzu z0.d, p0/m, z0.d
 ; VBITS_GE_512-NEXT:    st1w { z0.d }, p0, [x1]
 ; VBITS_GE_512-NEXT:    ret
   %op1 = load <8 x double>, ptr %a
   %res = fptoui <8 x double> %op1 to <8 x i32>
   store <8 x i32> %res, ptr %b
   ret void
 }
 
 define void @fcvtzu_v16f64_v16i32(ptr %a, ptr %b) vscale_range(8,0) #0 {
 ; CHECK-LABEL: fcvtzu_v16f64_v16i32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl16
-; CHECK-NEXT:    ptrue p1.d
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    fcvtzu z0.d, p1/m, z0.d
+; CHECK-NEXT:    fcvtzu z0.d, p0/m, z0.d
 ; CHECK-NEXT:    st1w { z0.d }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <16 x double>, ptr %a
   %res = fptoui <16 x double> %op1 to <16 x i32>
   store <16 x i32> %res, ptr %b
   ret void
 }
 
 define void @fcvtzu_v32f64_v32i32(ptr %a, ptr %b) vscale_range(16,0) #0 {
 ; CHECK-LABEL: fcvtzu_v32f64_v32i32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl32
-; CHECK-NEXT:    ptrue p1.d
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    fcvtzu z0.d, p1/m, z0.d
+; CHECK-NEXT:    fcvtzu z0.d, p0/m, z0.d
 ; CHECK-NEXT:    st1w { z0.d }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <32 x double>, ptr %a
   %res = fptoui <32 x double> %op1 to <32 x i32>
   store <32 x i32> %res, ptr %b
   ret void
 }
 
; CHECK-NEXT:    ret
   ret <4 x i16> %res
 }
 
 define <8 x i16> @fcvtzs_v8f32_v8i16(ptr %a) vscale_range(2,0) #0 {
 ; CHECK-LABEL: fcvtzs_v8f32_v8i16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.s, vl8
 ; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.s
 ; CHECK-NEXT:    fcvtzs z0.s, p0/m, z0.s
 ; CHECK-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; CHECK-NEXT:    // kill: def $q0 killed $q0 killed $z0
 ; CHECK-NEXT:    ret
   %op1 = load <8 x float>, ptr %a
   %res = fptosi <8 x float> %op1 to <8 x i16>
   ret <8 x i16> %res
 }
 
 define void @fcvtzs_v16f32_v16i16(ptr %a, ptr %b) #0 {
 ; VBITS_GE_256-LABEL: fcvtzs_v16f32_v16i16:
 ; VBITS_GE_256:       // %bb.0:
 ; VBITS_GE_256-NEXT:    mov x8, #8
 ; VBITS_GE_256-NEXT:    ptrue p0.s, vl8
 ; VBITS_GE_256-NEXT:    ld1w { z0.s }, p0/z, [x0, x8, lsl #2]
 ; VBITS_GE_256-NEXT:    ld1w { z1.s }, p0/z, [x0]
-; VBITS_GE_256-NEXT:    ptrue p0.s
 ; VBITS_GE_256-NEXT:    fcvtzs z0.s, p0/m, z0.s
 ; VBITS_GE_256-NEXT:    fcvtzs z1.s, p0/m, z1.s
 ; VBITS_GE_256-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; VBITS_GE_256-NEXT:    uzp1 z1.h, z1.h, z1.h
 ; VBITS_GE_256-NEXT:    ptrue p0.h, vl8
 ; VBITS_GE_256-NEXT:    splice z1.h, p0, z1.h, z0.h
 ; VBITS_GE_256-NEXT:    ptrue p0.h, vl16
 ; VBITS_GE_256-NEXT:    st1h { z1.h }, p0, [x1]
 ; VBITS_GE_256-NEXT:    ret
 ;
 ; VBITS_GE_512-LABEL: fcvtzs_v16f32_v16i16:
 ; VBITS_GE_512:       // %bb.0:
 ; VBITS_GE_512-NEXT:    ptrue p0.s, vl16
-; VBITS_GE_512-NEXT:    ptrue p1.s
 ; VBITS_GE_512-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; VBITS_GE_512-NEXT:    fcvtzs z0.s, p1/m, z0.s
+; VBITS_GE_512-NEXT:    fcvtzs z0.s, p0/m, z0.s
 ; VBITS_GE_512-NEXT:    st1h { z0.s }, p0, [x1]
 ; VBITS_GE_512-NEXT:    ret
   %op1 = load <16 x float>, ptr %a
   %res = fptosi <16 x float> %op1 to <16 x i16>
   store <16 x i16> %res, ptr %b
   ret void
 }
 
 define void @fcvtzs_v32f32_v32i16(ptr %a, ptr %b) vscale_range(8,0) #0 {
 ; CHECK-LABEL: fcvtzs_v32f32_v32i16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.s, vl32
-; CHECK-NEXT:    ptrue p1.s
 ; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    fcvtzs z0.s, p1/m, z0.s
+; CHECK-NEXT:    fcvtzs z0.s, p0/m, z0.s
 ; CHECK-NEXT:    st1h { z0.s }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <32 x float>, ptr %a
   %res = fptosi <32 x float> %op1 to <32 x i16>
   store <32 x i16> %res, ptr %b
   ret void
 }
 
 define void @fcvtzs_v64f32_v64i16(ptr %a, ptr %b) vscale_range(16,0) #0 {
 ; CHECK-LABEL: fcvtzs_v64f32_v64i16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.s, vl64
-; CHECK-NEXT:    ptrue p1.s
 ; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    fcvtzs z0.s, p1/m, z0.s
+; CHECK-NEXT:    fcvtzs z0.s, p0/m, z0.s
 ; CHECK-NEXT:    st1h { z0.s }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <64 x float>, ptr %a
   %res = fptosi <64 x float> %op1 to <64 x i16>
   store <64 x i16> %res, ptr %b
   ret void
 }
 

 ; FCVTZS D -> H
 ;
 
 ; v1f64 is perfered to be widened to v4f64, so use SVE
 define <1 x i16> @fcvtzs_v1f64_v1i16(<1 x double> %op1) vscale_range(2,0) #0 {
 ; CHECK-LABEL: fcvtzs_v1f64_v1i16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    // kill: def $d0 killed $d0 def $z0
-; CHECK-NEXT:    ptrue p0.d
+; CHECK-NEXT:    ptrue p0.d, vl4
 ; CHECK-NEXT:    fcvtzs z0.d, p0/m, z0.d
 ; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; CHECK-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $z0
 ; CHECK-NEXT:    ret
   %res = fptosi <1 x double> %op1 to <1 x i16>
   ret <1 x i16> %res
 }
; CHECK-NEXT:    ret
   ret <2 x i16> %res
 }
 
 define <4 x i16> @fcvtzs_v4f64_v4i16(ptr %a) vscale_range(2,0) #0 {
 ; CHECK-LABEL: fcvtzs_v4f64_v4i16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl4
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.d
 ; CHECK-NEXT:    fcvtzs z0.d, p0/m, z0.d
 ; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; CHECK-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $z0
 ; CHECK-NEXT:    ret
   %op1 = load <4 x double>, ptr %a
   %res = fptosi <4 x double> %op1 to <4 x i16>
   ret <4 x i16> %res
 }
 
 define <8 x i16> @fcvtzs_v8f64_v8i16(ptr %a) #0 {
 ; VBITS_GE_256-LABEL: fcvtzs_v8f64_v8i16:
 ; VBITS_GE_256:       // %bb.0:
 ; VBITS_GE_256-NEXT:    mov x8, #4
 ; VBITS_GE_256-NEXT:    ptrue p0.d, vl4
 ; VBITS_GE_256-NEXT:    ld1d { z0.d }, p0/z, [x0, x8, lsl #3]
 ; VBITS_GE_256-NEXT:    ld1d { z1.d }, p0/z, [x0]
-; VBITS_GE_256-NEXT:    ptrue p0.d
 ; VBITS_GE_256-NEXT:    fcvtzs z0.d, p0/m, z0.d
 ; VBITS_GE_256-NEXT:    fcvtzs z1.d, p0/m, z1.d
 ; VBITS_GE_256-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; VBITS_GE_256-NEXT:    uzp1 z1.s, z1.s, z1.s
 ; VBITS_GE_256-NEXT:    uzp1 z2.h, z0.h, z0.h
 ; VBITS_GE_256-NEXT:    uzp1 z0.h, z1.h, z1.h
 ; VBITS_GE_256-NEXT:    mov v0.d[1], v2.d[0]
 ; VBITS_GE_256-NEXT:    // kill: def $q0 killed $q0 killed $z0
 ; VBITS_GE_256-NEXT:    ret
 ;
 ; VBITS_GE_512-LABEL: fcvtzs_v8f64_v8i16:
 ; VBITS_GE_512:       // %bb.0:
 ; VBITS_GE_512-NEXT:    ptrue p0.d, vl8
 ; VBITS_GE_512-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; VBITS_GE_512-NEXT:    ptrue p0.d
 ; VBITS_GE_512-NEXT:    fcvtzs z0.d, p0/m, z0.d
 ; VBITS_GE_512-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; VBITS_GE_512-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; VBITS_GE_512-NEXT:    // kill: def $q0 killed $q0 killed $z0
 ; VBITS_GE_512-NEXT:    ret
   %op1 = load <8 x double>, ptr %a
   %res = fptosi <8 x double> %op1 to <8 x i16>
   ret <8 x i16> %res
 }
 
 define void @fcvtzs_v16f64_v16i16(ptr %a, ptr %b) vscale_range(8,0) #0 {
 ; CHECK-LABEL: fcvtzs_v16f64_v16i16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl16
-; CHECK-NEXT:    ptrue p1.d
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    fcvtzs z0.d, p1/m, z0.d
+; CHECK-NEXT:    fcvtzs z0.d, p0/m, z0.d
 ; CHECK-NEXT:    st1h { z0.d }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <16 x double>, ptr %a
   %res = fptosi <16 x double> %op1 to <16 x i16>
   store <16 x i16> %res, ptr %b
   ret void
 }
 
 define void @fcvtzs_v32f64_v32i16(ptr %a, ptr %b) vscale_range(16,0) #0 {
 ; CHECK-LABEL: fcvtzs_v32f64_v32i16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl32
-; CHECK-NEXT:    ptrue p1.d
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    fcvtzs z0.d, p1/m, z0.d
+; CHECK-NEXT:    fcvtzs z0.d, p0/m, z0.d
 ; CHECK-NEXT:    st1h { z0.d }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <32 x double>, ptr %a
   %res = fptosi <32 x double> %op1 to <32 x i16>
   store <32 x i16> %res, ptr %b
   ret void
 }
 
; CHECK-NEXT:    ret
   ret <2 x i32> %res
 }
 
 define <4 x i32> @fcvtzs_v4f64_v4i32(ptr %a) vscale_range(2,0) #0 {
 ; CHECK-LABEL: fcvtzs_v4f64_v4i32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl4
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.d
 ; CHECK-NEXT:    fcvtzs z0.d, p0/m, z0.d
 ; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; CHECK-NEXT:    // kill: def $q0 killed $q0 killed $z0
 ; CHECK-NEXT:    ret
   %op1 = load <4 x double>, ptr %a
   %res = fptosi <4 x double> %op1 to <4 x i32>
   ret <4 x i32> %res
 }
 
 define void @fcvtzs_v8f64_v8i32(ptr %a, ptr %b) #0 {
 ; VBITS_GE_256-LABEL: fcvtzs_v8f64_v8i32:
 ; VBITS_GE_256:       // %bb.0:
 ; VBITS_GE_256-NEXT:    mov x8, #4
 ; VBITS_GE_256-NEXT:    ptrue p0.d, vl4
 ; VBITS_GE_256-NEXT:    ld1d { z0.d }, p0/z, [x0, x8, lsl #3]
 ; VBITS_GE_256-NEXT:    ld1d { z1.d }, p0/z, [x0]
-; VBITS_GE_256-NEXT:    ptrue p0.d
 ; VBITS_GE_256-NEXT:    fcvtzs z0.d, p0/m, z0.d
 ; VBITS_GE_256-NEXT:    fcvtzs z1.d, p0/m, z1.d
 ; VBITS_GE_256-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; VBITS_GE_256-NEXT:    uzp1 z1.s, z1.s, z1.s
 ; VBITS_GE_256-NEXT:    ptrue p0.s, vl4
 ; VBITS_GE_256-NEXT:    splice z1.s, p0, z1.s, z0.s
 ; VBITS_GE_256-NEXT:    ptrue p0.s, vl8
 ; VBITS_GE_256-NEXT:    st1w { z1.s }, p0, [x1]
 ; VBITS_GE_256-NEXT:    ret
 ;
 ; VBITS_GE_512-LABEL: fcvtzs_v8f64_v8i32:
 ; VBITS_GE_512:       // %bb.0:
 ; VBITS_GE_512-NEXT:    ptrue p0.d, vl8
-; VBITS_GE_512-NEXT:    ptrue p1.d
 ; VBITS_GE_512-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; VBITS_GE_512-NEXT:    fcvtzs z0.d, p1/m, z0.d
+; VBITS_GE_512-NEXT:    fcvtzs z0.d, p0/m, z0.d
 ; VBITS_GE_512-NEXT:    st1w { z0.d }, p0, [x1]
 ; VBITS_GE_512-NEXT:    ret
   %op1 = load <8 x double>, ptr %a
   %res = fptosi <8 x double> %op1 to <8 x i32>
   store <8 x i32> %res, ptr %b
   ret void
 }
 
 define void @fcvtzs_v16f64_v16i32(ptr %a, ptr %b) vscale_range(8,0) #0 {
 ; CHECK-LABEL: fcvtzs_v16f64_v16i32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl16
-; CHECK-NEXT:    ptrue p1.d
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    fcvtzs z0.d, p1/m, z0.d
+; CHECK-NEXT:    fcvtzs z0.d, p0/m, z0.d
 ; CHECK-NEXT:    st1w { z0.d }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <16 x double>, ptr %a
   %res = fptosi <16 x double> %op1 to <16 x i32>
   store <16 x i32> %res, ptr %b
   ret void
 }
 
 define void @fcvtzs_v32f64_v32i32(ptr %a, ptr %b) vscale_range(16,0) #0 {
 ; CHECK-LABEL: fcvtzs_v32f64_v32i32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl32
-; CHECK-NEXT:    ptrue p1.d
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    fcvtzs z0.d, p1/m, z0.d
+; CHECK-NEXT:    fcvtzs z0.d, p0/m, z0.d
 ; CHECK-NEXT:    st1w { z0.d }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <32 x double>, ptr %a
   %res = fptosi <32 x double> %op1 to <32 x i32>
   store <32 x i32> %res, ptr %b
   ret void
 }
 

llvm/test/CodeGen/AArch64/sve-fixed-length-int-to-fp.ll

; CHECK-NEXT:    ret
   ret <4 x half> %res
 }
 
 define <8 x half> @ucvtf_v8i32_v8f16(ptr %a) vscale_range(2,0) #0 {
 ; CHECK-LABEL: ucvtf_v8i32_v8f16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.s, vl8
 ; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.s
 ; CHECK-NEXT:    ucvtf z0.h, p0/m, z0.s
 ; CHECK-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; CHECK-NEXT:    // kill: def $q0 killed $q0 killed $z0
 ; CHECK-NEXT:    ret
   %op1 = load <8 x i32>, ptr %a
   %res = uitofp <8 x i32> %op1 to <8 x half>
   ret <8 x half> %res
 }
 
 define void @ucvtf_v16i32_v16f16(ptr %a, ptr %b) #0 {
 ; VBITS_GE_256-LABEL: ucvtf_v16i32_v16f16:
 ; VBITS_GE_256:       // %bb.0:
 ; VBITS_GE_256-NEXT:    mov x8, #8
 ; VBITS_GE_256-NEXT:    ptrue p0.s, vl8
 ; VBITS_GE_256-NEXT:    ld1w { z0.s }, p0/z, [x0, x8, lsl #2]
 ; VBITS_GE_256-NEXT:    ld1w { z1.s }, p0/z, [x0]
-; VBITS_GE_256-NEXT:    ptrue p0.s
 ; VBITS_GE_256-NEXT:    ucvtf z0.h, p0/m, z0.s
 ; VBITS_GE_256-NEXT:    ucvtf z1.h, p0/m, z1.s
 ; VBITS_GE_256-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; VBITS_GE_256-NEXT:    uzp1 z1.h, z1.h, z1.h
 ; VBITS_GE_256-NEXT:    ptrue p0.h, vl8
 ; VBITS_GE_256-NEXT:    splice z1.h, p0, z1.h, z0.h
 ; VBITS_GE_256-NEXT:    ptrue p0.h, vl16
 ; VBITS_GE_256-NEXT:    st1h { z1.h }, p0, [x1]
 ; VBITS_GE_256-NEXT:    ret
 ;
 ; VBITS_GE_512-LABEL: ucvtf_v16i32_v16f16:
 ; VBITS_GE_512:       // %bb.0:
 ; VBITS_GE_512-NEXT:    ptrue p0.s, vl16
 ; VBITS_GE_512-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; VBITS_GE_512-NEXT:    ptrue p0.s
 ; VBITS_GE_512-NEXT:    ucvtf z0.h, p0/m, z0.s
 ; VBITS_GE_512-NEXT:    ptrue p0.h, vl16
 ; VBITS_GE_512-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; VBITS_GE_512-NEXT:    st1h { z0.h }, p0, [x1]
 ; VBITS_GE_512-NEXT:    ret
   %op1 = load <16 x i32>, ptr %a
   %res = uitofp <16 x i32> %op1 to <16 x half>
   store <16 x half> %res, ptr %b
   ret void
 }
 
 define void @ucvtf_v32i32_v32f16(ptr %a, ptr %b) vscale_range(8,0) #0 {
 ; CHECK-LABEL: ucvtf_v32i32_v32f16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.s, vl32
 ; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.s
 ; CHECK-NEXT:    ucvtf z0.h, p0/m, z0.s
 ; CHECK-NEXT:    ptrue p0.h, vl32
 ; CHECK-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; CHECK-NEXT:    st1h { z0.h }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <32 x i32>, ptr %a
   %res = uitofp <32 x i32> %op1 to <32 x half>
   store <32 x half> %res, ptr %b
   ret void
 }
 
 define void @ucvtf_v64i32_v64f16(ptr %a, ptr %b) vscale_range(16,0) #0 {
 ; CHECK-LABEL: ucvtf_v64i32_v64f16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.s, vl64
 ; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.s
 ; CHECK-NEXT:    ucvtf z0.h, p0/m, z0.s
 ; CHECK-NEXT:    ptrue p0.h, vl64
 ; CHECK-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; CHECK-NEXT:    st1h { z0.h }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <64 x i32>, ptr %a
   %res = uitofp <64 x i32> %op1 to <64 x half>
   store <64 x half> %res, ptr %b
; CHECK-NEXT:    ret
   ret <1 x half> %res
 }
 
 ; v2f16 is not legal for NEON, so use SVE
 define <2 x half> @ucvtf_v2i64_v2f16(<2 x i64> %op1) vscale_range(2,0) #0 {
 ; CHECK-LABEL: ucvtf_v2i64_v2f16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    // kill: def $q0 killed $q0 def $z0
-; CHECK-NEXT:    ptrue p0.d
+; CHECK-NEXT:    ptrue p0.d, vl4
 ; CHECK-NEXT:    ucvtf z0.h, p0/m, z0.d
 ; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; CHECK-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $z0
 ; CHECK-NEXT:    ret
   %res = uitofp <2 x i64> %op1 to <2 x half>
   ret <2 x half> %res
 }
 
 define <4 x half> @ucvtf_v4i64_v4f16(ptr %a) vscale_range(2,0) #0 {
 ; CHECK-LABEL: ucvtf_v4i64_v4f16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl4
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.d
 ; CHECK-NEXT:    ucvtf z0.h, p0/m, z0.d
 ; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; CHECK-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $z0
 ; CHECK-NEXT:    ret
   %op1 = load <4 x i64>, ptr %a
   %res = uitofp <4 x i64> %op1 to <4 x half>
   ret <4 x half> %res
 }
 
 define <8 x half> @ucvtf_v8i64_v8f16(ptr %a) #0 {
 ; VBITS_GE_256-LABEL: ucvtf_v8i64_v8f16:
 ; VBITS_GE_256:       // %bb.0:
 ; VBITS_GE_256-NEXT:    mov x8, #4
 ; VBITS_GE_256-NEXT:    ptrue p0.d, vl4
 ; VBITS_GE_256-NEXT:    ld1d { z0.d }, p0/z, [x0, x8, lsl #3]
 ; VBITS_GE_256-NEXT:    ld1d { z1.d }, p0/z, [x0]
-; VBITS_GE_256-NEXT:    ptrue p0.d
 ; VBITS_GE_256-NEXT:    ucvtf z0.h, p0/m, z0.d
 ; VBITS_GE_256-NEXT:    ucvtf z1.h, p0/m, z1.d
 ; VBITS_GE_256-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; VBITS_GE_256-NEXT:    uzp1 z1.s, z1.s, z1.s
 ; VBITS_GE_256-NEXT:    uzp1 z2.h, z0.h, z0.h
 ; VBITS_GE_256-NEXT:    uzp1 z0.h, z1.h, z1.h
 ; VBITS_GE_256-NEXT:    mov v0.d[1], v2.d[0]
 ; VBITS_GE_256-NEXT:    // kill: def $q0 killed $q0 killed $z0
 ; VBITS_GE_256-NEXT:    ret
 ;
 ; VBITS_GE_512-LABEL: ucvtf_v8i64_v8f16:
 ; VBITS_GE_512:       // %bb.0:
 ; VBITS_GE_512-NEXT:    ptrue p0.d, vl8
 ; VBITS_GE_512-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; VBITS_GE_512-NEXT:    ptrue p0.d
 ; VBITS_GE_512-NEXT:    ucvtf z0.h, p0/m, z0.d
 ; VBITS_GE_512-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; VBITS_GE_512-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; VBITS_GE_512-NEXT:    // kill: def $q0 killed $q0 killed $z0
 ; VBITS_GE_512-NEXT:    ret
   %op1 = load <8 x i64>, ptr %a
   %res = uitofp <8 x i64> %op1 to <8 x half>
   ret <8 x half> %res
 }
 
 define void @ucvtf_v16i64_v16f16(ptr %a, ptr %b) vscale_range(8,0) #0 {
 ; CHECK-LABEL: ucvtf_v16i64_v16f16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl16
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.d
 ; CHECK-NEXT:    ucvtf z0.h, p0/m, z0.d
 ; CHECK-NEXT:    ptrue p0.s, vl16
 ; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; CHECK-NEXT:    st1h { z0.s }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <16 x i64>, ptr %a
   %res = uitofp <16 x i64> %op1 to <16 x half>
   store <16 x half> %res, ptr %b
   ret void
 }
 
 define void @ucvtf_v32i64_v32f16(ptr %a, ptr %b) vscale_range(16,0) #0 {
 ; CHECK-LABEL: ucvtf_v32i64_v32f16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl32
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.d
 ; CHECK-NEXT:    ucvtf z0.h, p0/m, z0.d
 ; CHECK-NEXT:    ptrue p0.s, vl32
 ; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; CHECK-NEXT:    st1h { z0.s }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <32 x i64>, ptr %a
   %res = uitofp <32 x i64> %op1 to <32 x half>
   store <32 x half> %res, ptr %b
; CHECK-NEXT:    ret
   ret <2 x float> %res
 }
 
 define <4 x float> @ucvtf_v4i64_v4f32(ptr %a) vscale_range(2,0) #0 {
 ; CHECK-LABEL: ucvtf_v4i64_v4f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl4
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.d
 ; CHECK-NEXT:    ucvtf z0.s, p0/m, z0.d
 ; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; CHECK-NEXT:    // kill: def $q0 killed $q0 killed $z0
 ; CHECK-NEXT:    ret
   %op1 = load <4 x i64>, ptr %a
   %res = uitofp <4 x i64> %op1 to <4 x float>
   ret <4 x float> %res
 }
 
 define void @ucvtf_v8i64_v8f32(ptr %a, ptr %b) #0 {
 ; VBITS_GE_256-LABEL: ucvtf_v8i64_v8f32:
 ; VBITS_GE_256:       // %bb.0:
 ; VBITS_GE_256-NEXT:    mov x8, #4
 ; VBITS_GE_256-NEXT:    ptrue p0.d, vl4
 ; VBITS_GE_256-NEXT:    ld1d { z0.d }, p0/z, [x0, x8, lsl #3]
 ; VBITS_GE_256-NEXT:    ld1d { z1.d }, p0/z, [x0]
-; VBITS_GE_256-NEXT:    ptrue p0.d
 ; VBITS_GE_256-NEXT:    ucvtf z0.s, p0/m, z0.d
 ; VBITS_GE_256-NEXT:    ucvtf z1.s, p0/m, z1.d
 ; VBITS_GE_256-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; VBITS_GE_256-NEXT:    uzp1 z1.s, z1.s, z1.s
 ; VBITS_GE_256-NEXT:    ptrue p0.s, vl4
 ; VBITS_GE_256-NEXT:    splice z1.s, p0, z1.s, z0.s
 ; VBITS_GE_256-NEXT:    ptrue p0.s, vl8
 ; VBITS_GE_256-NEXT:    st1w { z1.s }, p0, [x1]
 ; VBITS_GE_256-NEXT:    ret
 ;
 ; VBITS_GE_512-LABEL: ucvtf_v8i64_v8f32:
 ; VBITS_GE_512:       // %bb.0:
 ; VBITS_GE_512-NEXT:    ptrue p0.d, vl8
 ; VBITS_GE_512-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; VBITS_GE_512-NEXT:    ptrue p0.d
 ; VBITS_GE_512-NEXT:    ucvtf z0.s, p0/m, z0.d
 ; VBITS_GE_512-NEXT:    ptrue p0.s, vl8
 ; VBITS_GE_512-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; VBITS_GE_512-NEXT:    st1w { z0.s }, p0, [x1]
 ; VBITS_GE_512-NEXT:    ret
   %op1 = load <8 x i64>, ptr %a
   %res = uitofp <8 x i64> %op1 to <8 x float>
   store <8 x float> %res, ptr %b
   ret void
 }
 
 define void @ucvtf_v16i64_v16f32(ptr %a, ptr %b) vscale_range(8,0) #0 {
 ; CHECK-LABEL: ucvtf_v16i64_v16f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl16
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.d
 ; CHECK-NEXT:    ucvtf z0.s, p0/m, z0.d
 ; CHECK-NEXT:    ptrue p0.s, vl16
 ; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; CHECK-NEXT:    st1w { z0.s }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <16 x i64>, ptr %a
   %res = uitofp <16 x i64> %op1 to <16 x float>
   store <16 x float> %res, ptr %b
   ret void
 }
 
 define void @ucvtf_v32i64_v32f32(ptr %a, ptr %b) vscale_range(16,0) #0 {
 ; CHECK-LABEL: ucvtf_v32i64_v32f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl32
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.d
 ; CHECK-NEXT:    ucvtf z0.s, p0/m, z0.d
 ; CHECK-NEXT:    ptrue p0.s, vl32
 ; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; CHECK-NEXT:    st1w { z0.s }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <32 x i64>, ptr %a
   %res = uitofp <32 x i64> %op1 to <32 x float>
   store <32 x float> %res, ptr %b
; CHECK-NEXT:    ret
   ret <4 x half> %res
 }
 
 define <8 x half> @scvtf_v8i32_v8f16(ptr %a) vscale_range(2,0) #0 {
 ; CHECK-LABEL: scvtf_v8i32_v8f16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.s, vl8
 ; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.s
 ; CHECK-NEXT:    scvtf z0.h, p0/m, z0.s
 ; CHECK-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; CHECK-NEXT:    // kill: def $q0 killed $q0 killed $z0
 ; CHECK-NEXT:    ret
   %op1 = load <8 x i32>, ptr %a
   %res = sitofp <8 x i32> %op1 to <8 x half>
   ret <8 x half> %res
 }
 
 define void @scvtf_v16i32_v16f16(ptr %a, ptr %b) #0 {
 ; VBITS_GE_256-LABEL: scvtf_v16i32_v16f16:
 ; VBITS_GE_256:       // %bb.0:
 ; VBITS_GE_256-NEXT:    mov x8, #8
 ; VBITS_GE_256-NEXT:    ptrue p0.s, vl8
 ; VBITS_GE_256-NEXT:    ld1w { z0.s }, p0/z, [x0, x8, lsl #2]
 ; VBITS_GE_256-NEXT:    ld1w { z1.s }, p0/z, [x0]
-; VBITS_GE_256-NEXT:    ptrue p0.s
 ; VBITS_GE_256-NEXT:    scvtf z0.h, p0/m, z0.s
 ; VBITS_GE_256-NEXT:    scvtf z1.h, p0/m, z1.s
 ; VBITS_GE_256-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; VBITS_GE_256-NEXT:    uzp1 z1.h, z1.h, z1.h
 ; VBITS_GE_256-NEXT:    ptrue p0.h, vl8
 ; VBITS_GE_256-NEXT:    splice z1.h, p0, z1.h, z0.h
 ; VBITS_GE_256-NEXT:    ptrue p0.h, vl16
 ; VBITS_GE_256-NEXT:    st1h { z1.h }, p0, [x1]
 ; VBITS_GE_256-NEXT:    ret
 ;
 ; VBITS_GE_512-LABEL: scvtf_v16i32_v16f16:
 ; VBITS_GE_512:       // %bb.0:
 ; VBITS_GE_512-NEXT:    ptrue p0.s, vl16
 ; VBITS_GE_512-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; VBITS_GE_512-NEXT:    ptrue p0.s
 ; VBITS_GE_512-NEXT:    scvtf z0.h, p0/m, z0.s
 ; VBITS_GE_512-NEXT:    ptrue p0.h, vl16
 ; VBITS_GE_512-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; VBITS_GE_512-NEXT:    st1h { z0.h }, p0, [x1]
 ; VBITS_GE_512-NEXT:    ret
   %op1 = load <16 x i32>, ptr %a
   %res = sitofp <16 x i32> %op1 to <16 x half>
   store <16 x half> %res, ptr %b
   ret void
 }
 
 define void @scvtf_v32i32_v32f16(ptr %a, ptr %b) vscale_range(8,0) #0 {
 ; CHECK-LABEL: scvtf_v32i32_v32f16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.s, vl32
 ; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.s
 ; CHECK-NEXT:    scvtf z0.h, p0/m, z0.s
 ; CHECK-NEXT:    ptrue p0.h, vl32
 ; CHECK-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; CHECK-NEXT:    st1h { z0.h }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <32 x i32>, ptr %a
   %res = sitofp <32 x i32> %op1 to <32 x half>
   store <32 x half> %res, ptr %b
   ret void
 }
 
 define void @scvtf_v64i32_v64f16(ptr %a, ptr %b) vscale_range(16,0) #0 {
 ; CHECK-LABEL: scvtf_v64i32_v64f16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.s, vl64
 ; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.s
 ; CHECK-NEXT:    scvtf z0.h, p0/m, z0.s
 ; CHECK-NEXT:    ptrue p0.h, vl64
 ; CHECK-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; CHECK-NEXT:    st1h { z0.h }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <64 x i32>, ptr %a
   %res = sitofp <64 x i32> %op1 to <64 x half>
   store <64 x half> %res, ptr %b
; CHECK-NEXT:    ret
   ret <1 x half> %res
 }
 
 ; v2f16 is not legal for NEON, so use SVE
 define <2 x half> @scvtf_v2i64_v2f16(<2 x i64> %op1) vscale_range(2,0) #0 {
 ; CHECK-LABEL: scvtf_v2i64_v2f16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    // kill: def $q0 killed $q0 def $z0
-; CHECK-NEXT:    ptrue p0.d
+; CHECK-NEXT:    ptrue p0.d, vl4
 ; CHECK-NEXT:    scvtf z0.h, p0/m, z0.d
 ; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; CHECK-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $z0
 ; CHECK-NEXT:    ret
   %res = sitofp <2 x i64> %op1 to <2 x half>
   ret <2 x half> %res
 }
 
 define <4 x half> @scvtf_v4i64_v4f16(ptr %a) vscale_range(2,0) #0 {
 ; CHECK-LABEL: scvtf_v4i64_v4f16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl4
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.d
 ; CHECK-NEXT:    scvtf z0.h, p0/m, z0.d
 ; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; CHECK-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $z0
 ; CHECK-NEXT:    ret
   %op1 = load <4 x i64>, ptr %a
   %res = sitofp <4 x i64> %op1 to <4 x half>
   ret <4 x half> %res
 }
 
 define <8 x half> @scvtf_v8i64_v8f16(ptr %a) #0 {
 ; VBITS_GE_256-LABEL: scvtf_v8i64_v8f16:
 ; VBITS_GE_256:       // %bb.0:
 ; VBITS_GE_256-NEXT:    mov x8, #4
 ; VBITS_GE_256-NEXT:    ptrue p0.d, vl4
 ; VBITS_GE_256-NEXT:    ld1d { z0.d }, p0/z, [x0, x8, lsl #3]
 ; VBITS_GE_256-NEXT:    ld1d { z1.d }, p0/z, [x0]
-; VBITS_GE_256-NEXT:    ptrue p0.d
 ; VBITS_GE_256-NEXT:    scvtf z0.h, p0/m, z0.d
 ; VBITS_GE_256-NEXT:    scvtf z1.h, p0/m, z1.d
 ; VBITS_GE_256-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; VBITS_GE_256-NEXT:    uzp1 z1.s, z1.s, z1.s
 ; VBITS_GE_256-NEXT:    uzp1 z2.h, z0.h, z0.h
 ; VBITS_GE_256-NEXT:    uzp1 z0.h, z1.h, z1.h
 ; VBITS_GE_256-NEXT:    mov v0.d[1], v2.d[0]
 ; VBITS_GE_256-NEXT:    // kill: def $q0 killed $q0 killed $z0
 ; VBITS_GE_256-NEXT:    ret
 ;
 ; VBITS_GE_512-LABEL: scvtf_v8i64_v8f16:
 ; VBITS_GE_512:       // %bb.0:
 ; VBITS_GE_512-NEXT:    ptrue p0.d, vl8
 ; VBITS_GE_512-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; VBITS_GE_512-NEXT:    ptrue p0.d
 ; VBITS_GE_512-NEXT:    scvtf z0.h, p0/m, z0.d
 ; VBITS_GE_512-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; VBITS_GE_512-NEXT:    uzp1 z0.h, z0.h, z0.h
 ; VBITS_GE_512-NEXT:    // kill: def $q0 killed $q0 killed $z0
 ; VBITS_GE_512-NEXT:    ret
   %op1 = load <8 x i64>, ptr %a
   %res = sitofp <8 x i64> %op1 to <8 x half>
   ret <8 x half> %res
 }
 
 define void @scvtf_v16i64_v16f16(ptr %a, ptr %b) vscale_range(8,0) #0 {
 ; CHECK-LABEL: scvtf_v16i64_v16f16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl16
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.d
 ; CHECK-NEXT:    scvtf z0.h, p0/m, z0.d
 ; CHECK-NEXT:    ptrue p0.s, vl16
 ; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; CHECK-NEXT:    st1h { z0.s }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <16 x i64>, ptr %a
   %res = sitofp <16 x i64> %op1 to <16 x half>
   store <16 x half> %res, ptr %b
   ret void
 }
 
 define void @scvtf_v32i64_v32f16(ptr %a, ptr %b) vscale_range(16,0) #0 {
 ; CHECK-LABEL: scvtf_v32i64_v32f16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl32
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.d
 ; CHECK-NEXT:    scvtf z0.h, p0/m, z0.d
 ; CHECK-NEXT:    ptrue p0.s, vl32
 ; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; CHECK-NEXT:    st1h { z0.s }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <32 x i64>, ptr %a
   %res = sitofp <32 x i64> %op1 to <32 x half>
   store <32 x half> %res, ptr %b
; CHECK-NEXT:    ret
   ret <2 x float> %res
 }
 
 define <4 x float> @scvtf_v4i64_v4f32(ptr %a) vscale_range(2,0) #0 {
 ; CHECK-LABEL: scvtf_v4i64_v4f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl4
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.d
 ; CHECK-NEXT:    scvtf z0.s, p0/m, z0.d
 ; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; CHECK-NEXT:    // kill: def $q0 killed $q0 killed $z0
 ; CHECK-NEXT:    ret
   %op1 = load <4 x i64>, ptr %a
   %res = sitofp <4 x i64> %op1 to <4 x float>
   ret <4 x float> %res
 }
 
 define void @scvtf_v8i64_v8f32(ptr %a, ptr %b) #0 {
 ; VBITS_GE_256-LABEL: scvtf_v8i64_v8f32:
 ; VBITS_GE_256:       // %bb.0:
 ; VBITS_GE_256-NEXT:    mov x8, #4
 ; VBITS_GE_256-NEXT:    ptrue p0.d, vl4
 ; VBITS_GE_256-NEXT:    ld1d { z0.d }, p0/z, [x0, x8, lsl #3]
 ; VBITS_GE_256-NEXT:    ld1d { z1.d }, p0/z, [x0]
-; VBITS_GE_256-NEXT:    ptrue p0.d
 ; VBITS_GE_256-NEXT:    scvtf z0.s, p0/m, z0.d
 ; VBITS_GE_256-NEXT:    scvtf z1.s, p0/m, z1.d
 ; VBITS_GE_256-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; VBITS_GE_256-NEXT:    uzp1 z1.s, z1.s, z1.s
 ; VBITS_GE_256-NEXT:    ptrue p0.s, vl4
 ; VBITS_GE_256-NEXT:    splice z1.s, p0, z1.s, z0.s
 ; VBITS_GE_256-NEXT:    ptrue p0.s, vl8
 ; VBITS_GE_256-NEXT:    st1w { z1.s }, p0, [x1]
 ; VBITS_GE_256-NEXT:    ret
 ;
 ; VBITS_GE_512-LABEL: scvtf_v8i64_v8f32:
 ; VBITS_GE_512:       // %bb.0:
 ; VBITS_GE_512-NEXT:    ptrue p0.d, vl8
 ; VBITS_GE_512-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; VBITS_GE_512-NEXT:    ptrue p0.d
 ; VBITS_GE_512-NEXT:    scvtf z0.s, p0/m, z0.d
 ; VBITS_GE_512-NEXT:    ptrue p0.s, vl8
 ; VBITS_GE_512-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; VBITS_GE_512-NEXT:    st1w { z0.s }, p0, [x1]
 ; VBITS_GE_512-NEXT:    ret
   %op1 = load <8 x i64>, ptr %a
   %res = sitofp <8 x i64> %op1 to <8 x float>
   store <8 x float> %res, ptr %b
   ret void
 }
 
 define void @scvtf_v16i64_v16f32(ptr %a, ptr %b) vscale_range(8,0) #0 {
 ; CHECK-LABEL: scvtf_v16i64_v16f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl16
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.d
 ; CHECK-NEXT:    scvtf z0.s, p0/m, z0.d
 ; CHECK-NEXT:    ptrue p0.s, vl16
 ; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; CHECK-NEXT:    st1w { z0.s }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <16 x i64>, ptr %a
   %res = sitofp <16 x i64> %op1 to <16 x float>
   store <16 x float> %res, ptr %b
   ret void
 }
 
 define void @scvtf_v32i64_v32f32(ptr %a, ptr %b) vscale_range(16,0) #0 {
 ; CHECK-LABEL: scvtf_v32i64_v32f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d, vl32
 ; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ptrue p0.d
 ; CHECK-NEXT:    scvtf z0.s, p0/m, z0.d
 ; CHECK-NEXT:    ptrue p0.s, vl32
 ; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; CHECK-NEXT:    st1w { z0.s }, p0, [x1]
 ; CHECK-NEXT:    ret
   %op1 = load <32 x i64>, ptr %a
   %res = sitofp <32 x i64> %op1 to <32 x float>
   store <32 x float> %res, ptr %b