[SVE][CodeGen] Legalisation of vsetcc with scalable types

[kmclaughlin]

Summary

Changes SplitVecOp_VSETCC to use getVectorElementCount()

[kmclaughlin]

• Rebased patch & updated checks in the tests with update_llc_test_checks.py

[efriedma]

Is it possible to write tests for this that don't result in a "max" or "min" operation? Or does that fail for some other reason?

Otherwise LGTM.

[kmclaughlin]

Added tests to llvm-ir-to-intrinsic.ll which check the results of compare instructions

[kmclaughlin]

In D79167#2098774, @efriedma wrote:

Is it possible to write tests for this that don't result in a "max" or "min" operation? Or does that fail for some other reason?

Otherwise LGTM.

Thanks for reviewing this, @efriedma. It was possible to add tests which don't result in a min/max by returning the result of the compare.

[efriedma]

LGTM

llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp

SDValue DAGTypeLegalizer::SplitVecOp_VSETCC(SDNode *N) {
   assert(N->getValueType(0).isVector() &&
          N->getOperand(0).getValueType().isVector() &&
          "Operand types must be vectors");
   // The result has a legal vector type, but the input needs splitting.
   SDValue Lo0, Hi0, Lo1, Hi1, LoRes, HiRes;
   SDLoc DL(N);
   GetSplitVector(N->getOperand(0), Lo0, Hi0);
   GetSplitVector(N->getOperand(1), Lo1, Hi1);
-  unsigned PartElements = Lo0.getValueType().getVectorNumElements();
-  EVT PartResVT = EVT::getVectorVT(*DAG.getContext(), MVT::i1, PartElements);
-  EVT WideResVT = EVT::getVectorVT(*DAG.getContext(), MVT::i1, 2*PartElements);
+  auto PartEltCnt = Lo0.getValueType().getVectorElementCount();
+
+  LLVMContext &Context = *DAG.getContext();
+  EVT PartResVT = EVT::getVectorVT(Context, MVT::i1, PartEltCnt);
+  EVT WideResVT = EVT::getVectorVT(Context, MVT::i1, PartEltCnt*2);
 
   LoRes = DAG.getNode(ISD::SETCC, DL, PartResVT, Lo0, Lo1, N->getOperand(2));
   HiRes = DAG.getNode(ISD::SETCC, DL, PartResVT, Hi0, Hi1, N->getOperand(2));
   SDValue Con = DAG.getNode(ISD::CONCAT_VECTORS, DL, WideResVT, LoRes, HiRes);
 
   EVT OpVT = N->getOperand(0).getValueType();
   ISD::NodeType ExtendCode =
       TargetLowering::getExtendForContent(TLI.getBooleanContents(OpVT));

llvm/test/CodeGen/AArch64/llvm-ir-to-intrinsic.ll

 ; CHECK-NEXT:    ptrue p0.d
 ; CHECK-NEXT:    smin z0.d, p0/m, z0.d, z1.d
 ; CHECK-NEXT:    ret
   %cmp = icmp slt <vscale x 2 x i64> %a, %b
   %min = select <vscale x 2 x i1> %cmp, <vscale x 2 x i64> %a, <vscale x 2 x i64> %b
   ret <vscale x 2 x i64> %min
 }
 
+define <vscale x 32 x i8> @smin_split_i8(<vscale x 32 x i8> %a, <vscale x 32 x i8> %b, <vscale x 32 x i8> %c) {
+; CHECK-LABEL: smin_split_i8:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p0.b
+; CHECK-NEXT:    smin z0.b, p0/m, z0.b, z2.b
+; CHECK-NEXT:    smin z1.b, p0/m, z1.b, z3.b
+; CHECK-NEXT:    ret
+  %cmp = icmp slt <vscale x 32 x i8> %a, %b
+  %min = select <vscale x 32 x i1> %cmp, <vscale x 32 x i8> %a, <vscale x 32 x i8> %b
+  ret <vscale x 32 x i8> %min
+}
+
+define <vscale x 32 x i16> @smin_split_i16(<vscale x 32 x i16> %a, <vscale x 32 x i16> %b, <vscale x 32 x i16> %c) {
+; CHECK-LABEL: smin_split_i16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p0.h
+; CHECK-NEXT:    smin z0.h, p0/m, z0.h, z4.h
+; CHECK-NEXT:    smin z1.h, p0/m, z1.h, z5.h
+; CHECK-NEXT:    smin z2.h, p0/m, z2.h, z6.h
+; CHECK-NEXT:    smin z3.h, p0/m, z3.h, z7.h
+; CHECK-NEXT:    ret
+  %cmp = icmp slt <vscale x 32 x i16> %a, %b
+  %min = select <vscale x 32 x i1> %cmp, <vscale x 32 x i16> %a, <vscale x 32 x i16> %b
+  ret <vscale x 32 x i16> %min
+}
+
+define <vscale x 8 x i32> @smin_split_i32(<vscale x 8 x i32> %a, <vscale x 8 x i32> %b, <vscale x 8 x i32> %c) {
+; CHECK-LABEL: smin_split_i32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p0.s
+; CHECK-NEXT:    smin z0.s, p0/m, z0.s, z2.s
+; CHECK-NEXT:    smin z1.s, p0/m, z1.s, z3.s
+; CHECK-NEXT:    ret
+  %cmp = icmp slt <vscale x 8 x i32> %a, %b
+  %min = select <vscale x 8 x i1> %cmp, <vscale x 8 x i32> %a, <vscale x 8 x i32> %b
+  ret <vscale x 8 x i32> %min
+}
+
+define <vscale x 4 x i64> @smin_split_i64(<vscale x 4 x i64> %a, <vscale x 4 x i64> %b, <vscale x 4 x i64> %c) {
+; CHECK-LABEL: smin_split_i64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p0.d
+; CHECK-NEXT:    smin z0.d, p0/m, z0.d, z2.d
+; CHECK-NEXT:    smin z1.d, p0/m, z1.d, z3.d
+; CHECK-NEXT:    ret
+  %cmp = icmp slt <vscale x 4 x i64> %a, %b
+  %min = select <vscale x 4 x i1> %cmp, <vscale x 4 x i64> %a, <vscale x 4 x i64> %b
+  ret <vscale x 4 x i64> %min
+}
+
+define <vscale x 8 x i8> @smin_promote_i8(<vscale x 8 x i8> %a, <vscale x 8 x i8> %b, <vscale x 8 x i8> %c) {
+; CHECK-LABEL: smin_promote_i8:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p0.h
+; CHECK-NEXT:    sxtb z1.h, p0/m, z1.h
+; CHECK-NEXT:    sxtb z0.h, p0/m, z0.h
+; CHECK-NEXT:    smin z0.h, p0/m, z0.h, z1.h
+; CHECK-NEXT:    ret
+  %cmp = icmp slt <vscale x 8 x i8> %a, %b
+  %min = select <vscale x 8 x i1> %cmp, <vscale x 8 x i8> %a, <vscale x 8 x i8> %b
+  ret <vscale x 8 x i8> %min
+}
+
+define <vscale x 4 x i16> @smin_promote_i16(<vscale x 4 x i16> %a, <vscale x 4 x i16> %b, <vscale x 4 x i16> %c) {
+; CHECK-LABEL: smin_promote_i16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p0.s
+; CHECK-NEXT:    sxth z1.s, p0/m, z1.s
+; CHECK-NEXT:    sxth z0.s, p0/m, z0.s
+; CHECK-NEXT:    smin z0.s, p0/m, z0.s, z1.s
+; CHECK-NEXT:    ret
+  %cmp = icmp slt <vscale x 4 x i16> %a, %b
+  %min = select <vscale x 4 x i1> %cmp, <vscale x 4 x i16> %a, <vscale x 4 x i16> %b
+  ret <vscale x 4 x i16> %min
+}
+
+define <vscale x 2 x i32> @smin_promote_i32(<vscale x 2 x i32> %a, <vscale x 2 x i32> %b, <vscale x 2 x i32> %c) {
+; CHECK-LABEL: smin_promote_i32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p0.d
+; CHECK-NEXT:    sxtw z1.d, p0/m, z1.d
+; CHECK-NEXT:    sxtw z0.d, p0/m, z0.d
+; CHECK-NEXT:    smin z0.d, p0/m, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %cmp = icmp slt <vscale x 2 x i32> %a, %b
+  %min = select <vscale x 2 x i1> %cmp, <vscale x 2 x i32> %a, <vscale x 2 x i32> %b
+  ret <vscale x 2 x i32> %min
+}
+
 ;
 ; UMIN
 ;
 
 define <vscale x 16 x i8> @umin_i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b, <vscale x 16 x i8> %c) {
 ; CHECK-LABEL: umin_i8:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.b

 ; CHECK-NEXT:    ptrue p0.d
 ; CHECK-NEXT:    umin z0.d, p0/m, z0.d, z1.d
 ; CHECK-NEXT:    ret
   %cmp = icmp ult <vscale x 2 x i64> %a, %b
   %min = select <vscale x 2 x i1> %cmp, <vscale x 2 x i64> %a, <vscale x 2 x i64> %b
   ret <vscale x 2 x i64> %min
 }
 
+define <vscale x 4 x i64> @umin_split_i64(<vscale x 4 x i64> %a, <vscale x 4 x i64> %b, <vscale x 4 x i64> %c) {
+; CHECK-LABEL: umin_split_i64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p0.d
+; CHECK-NEXT:    umin z0.d, p0/m, z0.d, z2.d
+; CHECK-NEXT:    umin z1.d, p0/m, z1.d, z3.d
+; CHECK-NEXT:    ret
+  %cmp = icmp ult <vscale x 4 x i64> %a, %b
+  %min = select <vscale x 4 x i1> %cmp, <vscale x 4 x i64> %a, <vscale x 4 x i64> %b
+  ret <vscale x 4 x i64> %min
+}
+
+define <vscale x 8 x i8> @umin_promote_i8(<vscale x 8 x i8> %a, <vscale x 8 x i8> %b, <vscale x 8 x i8> %c) {
+; CHECK-LABEL: umin_promote_i8:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p0.h
+; CHECK-NEXT:    and z1.h, z1.h, #0xff
+; CHECK-NEXT:    and z0.h, z0.h, #0xff
+; CHECK-NEXT:    umin z0.h, p0/m, z0.h, z1.h
+; CHECK-NEXT:    ret
+  %cmp = icmp ult <vscale x 8 x i8> %a, %b
+  %min = select <vscale x 8 x i1> %cmp, <vscale x 8 x i8> %a, <vscale x 8 x i8> %b
+  ret <vscale x 8 x i8> %min
+}
+
 ;
 ; SMAX
 ;
 
 define <vscale x 16 x i8> @smax_i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b, <vscale x 16 x i8> %c) {
 ; CHECK-LABEL: smax_i8:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.b
 ; CHECK-NEXT:    smax z0.b, p0/m, z0.b, z1.b
 ; CHECK-NEXT:    ret
   %cmp = icmp sgt <vscale x 16 x i8> %a, %b
-  %min = select <vscale x 16 x i1> %cmp, <vscale x 16 x i8> %a, <vscale x 16 x i8> %b
-  ret <vscale x 16 x i8> %min
+  %max = select <vscale x 16 x i1> %cmp, <vscale x 16 x i8> %a, <vscale x 16 x i8> %b
+  ret <vscale x 16 x i8> %max
 }
 
 define <vscale x 8 x i16> @smax_i16(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b, <vscale x 8 x i16> %c) {
 ; CHECK-LABEL: smax_i16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.h
 ; CHECK-NEXT:    smax z0.h, p0/m, z0.h, z1.h
 ; CHECK-NEXT:    ret
   %cmp = icmp sgt <vscale x 8 x i16> %a, %b
-  %min = select <vscale x 8 x i1> %cmp, <vscale x 8 x i16> %a, <vscale x 8 x i16> %b
-  ret <vscale x 8 x i16> %min
+  %max = select <vscale x 8 x i1> %cmp, <vscale x 8 x i16> %a, <vscale x 8 x i16> %b
+  ret <vscale x 8 x i16> %max
 }
 
 define <vscale x 4 x i32> @smax_i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b, <vscale x 4 x i32> %c) {
 ; CHECK-LABEL: smax_i32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.s
 ; CHECK-NEXT:    smax z0.s, p0/m, z0.s, z1.s
 ; CHECK-NEXT:    ret
   %cmp = icmp sgt <vscale x 4 x i32> %a, %b
-  %min = select <vscale x 4 x i1> %cmp, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b
-  ret <vscale x 4 x i32> %min
+  %max = select <vscale x 4 x i1> %cmp, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b
+  ret <vscale x 4 x i32> %max
 }
 
 define <vscale x 2 x i64> @smax_i64(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b, <vscale x 2 x i64> %c) {
 ; CHECK-LABEL: smax_i64:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d
 ; CHECK-NEXT:    smax z0.d, p0/m, z0.d, z1.d
 ; CHECK-NEXT:    ret
   %cmp = icmp sgt <vscale x 2 x i64> %a, %b
-  %min = select <vscale x 2 x i1> %cmp, <vscale x 2 x i64> %a, <vscale x 2 x i64> %b
-  ret <vscale x 2 x i64> %min
+  %max = select <vscale x 2 x i1> %cmp, <vscale x 2 x i64> %a, <vscale x 2 x i64> %b
+  ret <vscale x 2 x i64> %max
+}
+
+define <vscale x 8 x i32> @smax_split_i32(<vscale x 8 x i32> %a, <vscale x 8 x i32> %b, <vscale x 8 x i32> %c) {
+; CHECK-LABEL: smax_split_i32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p0.s
+; CHECK-NEXT:    smax z0.s, p0/m, z0.s, z2.s
+; CHECK-NEXT:    smax z1.s, p0/m, z1.s, z3.s
+; CHECK-NEXT:    ret
+  %cmp = icmp sgt <vscale x 8 x i32> %a, %b
+  %max = select <vscale x 8 x i1> %cmp, <vscale x 8 x i32> %a, <vscale x 8 x i32> %b
+  ret <vscale x 8 x i32> %max
+}
+
+define <vscale x 4 x i16> @smax_promote_i16(<vscale x 4 x i16> %a, <vscale x 4 x i16> %b, <vscale x 4 x i16> %c) {
+; CHECK-LABEL: smax_promote_i16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p0.s
+; CHECK-NEXT:    sxth z1.s, p0/m, z1.s
+; CHECK-NEXT:    sxth z0.s, p0/m, z0.s
+; CHECK-NEXT:    smax z0.s, p0/m, z0.s, z1.s
+; CHECK-NEXT:    ret
+  %cmp = icmp sgt <vscale x 4 x i16> %a, %b
+  %max = select <vscale x 4 x i1> %cmp, <vscale x 4 x i16> %a, <vscale x 4 x i16> %b
+  ret <vscale x 4 x i16> %max
 }
 
 ;
 ; UMAX
 ;
 
 define <vscale x 16 x i8> @umax_i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b, <vscale x 16 x i8> %c) {
 ; CHECK-LABEL: umax_i8:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.b
 ; CHECK-NEXT:    umax z0.b, p0/m, z0.b, z1.b
 ; CHECK-NEXT:    ret
   %cmp = icmp ugt <vscale x 16 x i8> %a, %b
-  %min = select <vscale x 16 x i1> %cmp, <vscale x 16 x i8> %a, <vscale x 16 x i8> %b
-  ret <vscale x 16 x i8> %min
+  %max = select <vscale x 16 x i1> %cmp, <vscale x 16 x i8> %a, <vscale x 16 x i8> %b
+  ret <vscale x 16 x i8> %max
 }
 
 define <vscale x 8 x i16> @umax_i16(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b, <vscale x 8 x i16> %c) {
 ; CHECK-LABEL: umax_i16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.h
 ; CHECK-NEXT:    umax z0.h, p0/m, z0.h, z1.h
 ; CHECK-NEXT:    ret
   %cmp = icmp ugt <vscale x 8 x i16> %a, %b
-  %min = select <vscale x 8 x i1> %cmp, <vscale x 8 x i16> %a, <vscale x 8 x i16> %b
-  ret <vscale x 8 x i16> %min
+  %max = select <vscale x 8 x i1> %cmp, <vscale x 8 x i16> %a, <vscale x 8 x i16> %b
+  ret <vscale x 8 x i16> %max
 }
 
 define <vscale x 4 x i32> @umax_i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b, <vscale x 4 x i32> %c) {
 ; CHECK-LABEL: umax_i32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.s
 ; CHECK-NEXT:    umax z0.s, p0/m, z0.s, z1.s
 ; CHECK-NEXT:    ret
   %cmp = icmp ugt <vscale x 4 x i32> %a, %b
-  %min = select <vscale x 4 x i1> %cmp, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b
-  ret <vscale x 4 x i32> %min
+  %max = select <vscale x 4 x i1> %cmp, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b
+  ret <vscale x 4 x i32> %max
 }
 
 define <vscale x 2 x i64> @umax_i64(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b, <vscale x 2 x i64> %c) {
 ; CHECK-LABEL: umax_i64:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.d
 ; CHECK-NEXT:    umax z0.d, p0/m, z0.d, z1.d
 ; CHECK-NEXT:    ret
   %cmp = icmp ugt <vscale x 2 x i64> %a, %b
-  %min = select <vscale x 2 x i1> %cmp, <vscale x 2 x i64> %a, <vscale x 2 x i64> %b
-  ret <vscale x 2 x i64> %min
+  %max = select <vscale x 2 x i1> %cmp, <vscale x 2 x i64> %a, <vscale x 2 x i64> %b
+  ret <vscale x 2 x i64> %max
+}
+
+define <vscale x 16 x i16> @umax_split_i16(<vscale x 16 x i16> %a, <vscale x 16 x i16> %b, <vscale x 16 x i16> %c) {
+; CHECK-LABEL: umax_split_i16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p0.h
+; CHECK-NEXT:    umax z0.h, p0/m, z0.h, z2.h
+; CHECK-NEXT:    umax z1.h, p0/m, z1.h, z3.h
+; CHECK-NEXT:    ret
+  %cmp = icmp ugt <vscale x 16 x i16> %a, %b
+  %max = select <vscale x 16 x i1> %cmp, <vscale x 16 x i16> %a, <vscale x 16 x i16> %b
+  ret <vscale x 16 x i16> %max
+}
+
+define <vscale x 2 x i32> @umax_promote_i32(<vscale x 2 x i32> %a, <vscale x 2 x i32> %b, <vscale x 2 x i32> %c) {
+; CHECK-LABEL: umax_promote_i32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p0.d
+; CHECK-NEXT:    and z1.d, z1.d, #0xffffffff
+; CHECK-NEXT:    and z0.d, z0.d, #0xffffffff
+; CHECK-NEXT:    umax z0.d, p0/m, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %cmp = icmp ugt <vscale x 2 x i32> %a, %b
+  %max = select <vscale x 2 x i1> %cmp, <vscale x 2 x i32> %a, <vscale x 2 x i32> %b
+  ret <vscale x 2 x i32> %max
 }
 
 ;
 ; ASR
 ;
 
 define <vscale x 16 x i8> @asr_i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b){
 ; CHECK-LABEL: asr_i8:

 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ptrue p0.s
 ; CHECK-NEXT:    lsr z0.s, p0/m, z0.s, z2.s
 ; CHECK-NEXT:    lsr z1.s, p0/m, z1.s, z3.s
 ; CHECK-NEXT:    ret
   %shr = lshr <vscale x 8 x i32> %a, %b
   ret <vscale x 8 x i32> %shr
 }
+
+;
+; CMP
+;
+
+define <vscale x 32 x i1> @cmp_split_32(<vscale x 32 x i8> %a, <vscale x 32 x i8> %b) {
+; CHECK-LABEL: cmp_split_32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p1.b
+; CHECK-NEXT:    cmpgt p0.b, p1/z, z2.b, z0.b
+; CHECK-NEXT:    cmpgt p1.b, p1/z, z3.b, z1.b
+; CHECK-NEXT:    ret
+  %cmp = icmp slt <vscale x 32 x i8> %a, %b
+  ret <vscale x 32 x i1> %cmp
+}
+
+define <vscale x 64 x i1> @cmp_split_64(<vscale x 64 x i8> %a, <vscale x 64 x i8> %b) {
+; CHECK-LABEL: cmp_split_64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p3.b
+; CHECK-NEXT:    cmpgt p0.b, p3/z, z0.b, z4.b
+; CHECK-NEXT:    cmpgt p1.b, p3/z, z1.b, z5.b
+; CHECK-NEXT:    cmpgt p2.b, p3/z, z2.b, z6.b
+; CHECK-NEXT:    cmpgt p3.b, p3/z, z3.b, z7.b
+; CHECK-NEXT:    ret
+  %cmp = icmp sgt <vscale x 64 x i8> %a, %b
+  ret <vscale x 64 x i1> %cmp
+}