[AArch64] Generate zeroing forms of certain SVE2.2 instructions (11/11) (#116837)

SVE2.2 introduces instructions with predicated forms with zeroing of
the inactive lanes. This allows in some cases to save a `movprfx` or
a `mov` instruction when emitting code for `_x` or `_z` variants of
intrinsics.

This patch adds support for emitting the zeroing forms of certain
`SXTB`, `UXTB`, `SXTH`, `UXTH`, `SXTW`, and `UXTW` instructions.

Author: momchil-velikov

llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td

@@ -4325,14 +4325,14 @@ let Predicates = [HasSVE2p2_or_SME2p2] in {
   defm FNEG_ZPzZ : sve_int_un_pred_arit_bitwise_fp_z<0b101, "fneg", AArch64fneg_mt>;
 
   // SVE2p2 integer unary arithmetic, zeroing predicate
-  defm SXTB_ZPzZ  : sve_int_un_pred_arit_h_z<0b000, "sxtb">;
-  defm UXTB_ZPzZ  : sve_int_un_pred_arit_h_z<0b001, "uxtb">;
-  defm SXTH_ZPzZ  : sve_int_un_pred_arit_w_z<0b010, "sxth">;
-  defm UXTH_ZPzZ  : sve_int_un_pred_arit_w_z<0b011, "uxth">;
+  defm SXTB_ZPzZ  : sve_int_un_pred_arit_h_z<0b000, "sxtb", AArch64sxt_mt>;
+  defm UXTB_ZPzZ  : sve_int_un_pred_arit_h_z<0b001, "uxtb", AArch64uxt_mt>;
+  defm SXTH_ZPzZ  : sve_int_un_pred_arit_w_z<0b010, "sxth", AArch64sxt_mt>;
+  defm UXTH_ZPzZ  : sve_int_un_pred_arit_w_z<0b011, "uxth", AArch64uxt_mt>;
   defm ABS_ZPzZ   : sve_int_un_pred_arit_z<  0b110, "abs", AArch64abs_mt>;
   defm NEG_ZPzZ   : sve_int_un_pred_arit_z<  0b111, "neg", AArch64neg_mt>;
-  def SXTW_ZPzZ_D : sve_int_un_pred_arit_z<0b11, 0b1000, "sxtw", ZPR64>;
-  def UXTW_ZPzZ_D : sve_int_un_pred_arit_z<0b11, 0b1010, "uxtw", ZPR64>;
+  defm SXTW_ZPzZ  : sve_int_un_pred_arit_d_z<0b100, "sxtw", AArch64sxt_mt>;
+  defm UXTW_ZPzZ  : sve_int_un_pred_arit_d_z<0b101, "uxtw", AArch64uxt_mt>;
 
   // SVE predicate count
   defm FIRSTP_XPP : sve_int_pcount_pred_tmp<0b001, "firstp">;

llvm/lib/Target/AArch64/SVEInstrFormats.td

@@ -662,6 +662,17 @@ multiclass SVE_InReg_Extend_PassthruUndef<ValueType vt, SDPatternOperator op, Va
             (inst $PassThru, $Pg, $Src)>;
 }
 
+multiclass SVE_InReg_Extend_PassthruUndefZero<ValueType vt, SDPatternOperator op, ValueType pt,
+                                         ValueType inreg_vt, Instruction inst> {
+  let AddedComplexity = 1 in {
+    def : Pat<(vt (op pt:$Pg, vt:$Src, inreg_vt, (vt (SVEDup0Undef)))),
+               (inst $Pg, $Src)>;
+
+    def : Pat<(vt (op (pt (SVEAllActive:$Pg)), vt:$Src, inreg_vt, (vt (SVEAny)))),
+              (inst $Pg, $Src)>;
+  }
+}
+
 class SVE_Shift_DupImm_Pred_Pat<ValueType vt, SDPatternOperator op,
                                 ValueType pt, ValueType it,
                                 ComplexPattern cast, Instruction inst>
@@ -4930,10 +4941,14 @@ multiclass sve_int_un_pred_arit_h<bits<3> opc, string asm,
   defm : SVE_InReg_Extend_PassthruUndef<nxv2i64, op, nxv2i1, nxv2i8, !cast<Pseudo>(NAME # _D_UNDEF)>;
 }
 
-multiclass sve_int_un_pred_arit_h_z<bits<3> opc, string asm> {
+multiclass sve_int_un_pred_arit_h_z<bits<3> opc, string asm, SDPatternOperator op> {
   def _H : sve_int_un_pred_arit_z<0b01, { opc, 0b0 }, asm, ZPR16>;
   def _S : sve_int_un_pred_arit_z<0b10, { opc, 0b0 }, asm, ZPR32>;
   def _D : sve_int_un_pred_arit_z<0b11, { opc, 0b0 }, asm, ZPR64>;
+
+  defm : SVE_InReg_Extend_PassthruUndefZero<nxv8i16, op, nxv8i1, nxv8i8, !cast<Instruction>(NAME # _H)>;
+  defm : SVE_InReg_Extend_PassthruUndefZero<nxv4i32, op, nxv4i1, nxv4i8, !cast<Instruction>(NAME # _S)>;
+  defm : SVE_InReg_Extend_PassthruUndefZero<nxv2i64, op, nxv2i1, nxv2i8, !cast<Instruction>(NAME # _D)>;
 }
 
 multiclass sve_int_un_pred_arit_w<bits<3> opc, string asm,
@@ -4953,9 +4968,12 @@ multiclass sve_int_un_pred_arit_w<bits<3> opc, string asm,
   defm : SVE_InReg_Extend_PassthruUndef<nxv2i64, op, nxv2i1, nxv2i16, !cast<Pseudo>(NAME # _D_UNDEF)>;
 }
 
-multiclass sve_int_un_pred_arit_w_z<bits<3> opc, string asm> {
+multiclass sve_int_un_pred_arit_w_z<bits<3> opc, string asm, SDPatternOperator op> {
   def _S : sve_int_un_pred_arit_z<0b10, { opc, 0b0 }, asm, ZPR32>;
   def _D : sve_int_un_pred_arit_z<0b11, { opc, 0b0 }, asm, ZPR64>;
+
+  defm : SVE_InReg_Extend_PassthruUndefZero<nxv4i32, op, nxv4i1, nxv4i16, !cast<Instruction>(NAME # _S)>;
+  defm : SVE_InReg_Extend_PassthruUndefZero<nxv2i64, op, nxv2i1, nxv2i16, !cast<Instruction>(NAME # _D)>;
 }
 
 multiclass sve_int_un_pred_arit_d<bits<3> opc, string asm,
@@ -4970,6 +4988,12 @@ multiclass sve_int_un_pred_arit_d<bits<3> opc, string asm,
   defm : SVE_InReg_Extend_PassthruUndef<nxv2i64, op, nxv2i1, nxv2i32, !cast<Pseudo>(NAME # _D_UNDEF)>;
 }
 
+multiclass sve_int_un_pred_arit_d_z<bits<3> opc, string asm, SDPatternOperator op> {
+  def _D : sve_int_un_pred_arit_z<0b11, {opc, 0b0}, asm, ZPR64>;
+
+  defm : SVE_InReg_Extend_PassthruUndefZero<nxv2i64, op, nxv2i1, nxv2i32, !cast<Instruction>(NAME # _D)>;
+}
+
 multiclass sve_int_un_pred_arit_bitwise<bits<3> opc, string asm,
                                         SDPatternOperator op> {
   def _B : sve_int_un_pred_arit<0b00, { opc, 0b1 }, asm, ZPR8>,