[AArch64][SVE2] Load/store instruction fixes

[c-rhodes]

Summary

• Loads and stores in SVE2 are gather/scatter not contiguous, fixed by renaming multiclasses to reflect this and also updated comments.
• Remove aliases from load/store multiclasses that reflect the behaviour of the original form.
• Fix bug in scatter store implementation, vector list should be used as input, not output.

[sdesmalen]

LGTM, these seem like straightforward fixes.
I doubt there is any way to test the scatter-fix with only the assembler at the moment.

llvm/trunk/lib/Target/AArch64/AArch64SVEInstrInfo.td

let Predicates = [HasSVE2] in {
   def NBSL_ZZZZ_D   : sve2_int_bitwise_ternary_op_d<0b111, "nbsl">;
 
   // SVE2 bitwise xor and rotate right by immediate
   defm XAR_ZZZI : sve2_int_rotate_right_imm<"xar">;
 
   // SVE2 extract vector (immediate offset, constructive)
   def EXT_ZZI_B : sve2_int_perm_extract_i_cons<"ext">;
 
-
-  // Non-temporal contiguous loads (vector + register)
-  defm LDNT1SB_ZZR_S : sve2_mem_cldnt_vs<0b00000, "ldnt1sb", Z_s, ZPR32>;
-  defm LDNT1B_ZZR_S  : sve2_mem_cldnt_vs<0b00001, "ldnt1b",  Z_s, ZPR32>;
-  defm LDNT1SH_ZZR_S : sve2_mem_cldnt_vs<0b00100, "ldnt1sh", Z_s, ZPR32>;
-  defm LDNT1H_ZZR_S  : sve2_mem_cldnt_vs<0b00101, "ldnt1h",  Z_s, ZPR32>;
-  defm LDNT1W_ZZR_S  : sve2_mem_cldnt_vs<0b01001, "ldnt1w",  Z_s, ZPR32>;
-
-  defm LDNT1SB_ZZR_D : sve2_mem_cldnt_vs<0b10000, "ldnt1sb", Z_d, ZPR64>;
-  defm LDNT1B_ZZR_D  : sve2_mem_cldnt_vs<0b10010, "ldnt1b",  Z_d, ZPR64>;
-  defm LDNT1SH_ZZR_D : sve2_mem_cldnt_vs<0b10100, "ldnt1sh", Z_d, ZPR64>;
-  defm LDNT1H_ZZR_D  : sve2_mem_cldnt_vs<0b10110, "ldnt1h",  Z_d, ZPR64>;
-  defm LDNT1SW_ZZR_D : sve2_mem_cldnt_vs<0b11000, "ldnt1sw", Z_d, ZPR64>;
-  defm LDNT1W_ZZR_D  : sve2_mem_cldnt_vs<0b11010, "ldnt1w",  Z_d, ZPR64>;
-  defm LDNT1D_ZZR_D  : sve2_mem_cldnt_vs<0b11110, "ldnt1d",  Z_d, ZPR64>;
+  // SVE2 non-temporal gather loads
+  defm LDNT1SB_ZZR_S : sve2_mem_gldnt_vs<0b00000, "ldnt1sb", Z_s, ZPR32>;
+  defm LDNT1B_ZZR_S  : sve2_mem_gldnt_vs<0b00001, "ldnt1b",  Z_s, ZPR32>;
+  defm LDNT1SH_ZZR_S : sve2_mem_gldnt_vs<0b00100, "ldnt1sh", Z_s, ZPR32>;
+  defm LDNT1H_ZZR_S  : sve2_mem_gldnt_vs<0b00101, "ldnt1h",  Z_s, ZPR32>;
+  defm LDNT1W_ZZR_S  : sve2_mem_gldnt_vs<0b01001, "ldnt1w",  Z_s, ZPR32>;
+
+  defm LDNT1SB_ZZR_D : sve2_mem_gldnt_vs<0b10000, "ldnt1sb", Z_d, ZPR64>;
+  defm LDNT1B_ZZR_D  : sve2_mem_gldnt_vs<0b10010, "ldnt1b",  Z_d, ZPR64>;
+  defm LDNT1SH_ZZR_D : sve2_mem_gldnt_vs<0b10100, "ldnt1sh", Z_d, ZPR64>;
+  defm LDNT1H_ZZR_D  : sve2_mem_gldnt_vs<0b10110, "ldnt1h",  Z_d, ZPR64>;
+  defm LDNT1SW_ZZR_D : sve2_mem_gldnt_vs<0b11000, "ldnt1sw", Z_d, ZPR64>;
+  defm LDNT1W_ZZR_D  : sve2_mem_gldnt_vs<0b11010, "ldnt1w",  Z_d, ZPR64>;
+  defm LDNT1D_ZZR_D  : sve2_mem_gldnt_vs<0b11110, "ldnt1d",  Z_d, ZPR64>;
 
   // SVE2 vector splice (constructive)
   defm SPLICE_ZPZZ : sve2_int_perm_splice_cons<"splice">;
 
-
-  // Non-temporal contiguous stores (vector + register)
-  defm STNT1B_ZZR_S : sve2_mem_cstnt_vs<0b001, "stnt1b", Z_s, ZPR32>;
-  defm STNT1H_ZZR_S : sve2_mem_cstnt_vs<0b011, "stnt1h", Z_s, ZPR32>;
-  defm STNT1W_ZZR_S : sve2_mem_cstnt_vs<0b101, "stnt1w", Z_s, ZPR32>;
-
-  defm STNT1B_ZZR_D : sve2_mem_cstnt_vs<0b000, "stnt1b", Z_d, ZPR64>;
-  defm STNT1H_ZZR_D : sve2_mem_cstnt_vs<0b010, "stnt1h", Z_d, ZPR64>;
-  defm STNT1W_ZZR_D : sve2_mem_cstnt_vs<0b100, "stnt1w", Z_d, ZPR64>;
-  defm STNT1D_ZZR_D : sve2_mem_cstnt_vs<0b110, "stnt1d", Z_d, ZPR64>;
+  // SVE2 non-temporal scatter stores
+  defm STNT1B_ZZR_S : sve2_mem_sstnt_vs<0b001, "stnt1b", Z_s, ZPR32>;
+  defm STNT1H_ZZR_S : sve2_mem_sstnt_vs<0b011, "stnt1h", Z_s, ZPR32>;
+  defm STNT1W_ZZR_S : sve2_mem_sstnt_vs<0b101, "stnt1w", Z_s, ZPR32>;
+
+  defm STNT1B_ZZR_D : sve2_mem_sstnt_vs<0b000, "stnt1b", Z_d, ZPR64>;
+  defm STNT1H_ZZR_D : sve2_mem_sstnt_vs<0b010, "stnt1h", Z_d, ZPR64>;
+  defm STNT1W_ZZR_D : sve2_mem_sstnt_vs<0b100, "stnt1w", Z_d, ZPR64>;
+  defm STNT1D_ZZR_D : sve2_mem_sstnt_vs<0b110, "stnt1d", Z_d, ZPR64>;
 
   // SVE2 table lookup (three sources)
   defm TBL_ZZZZ : sve2_int_perm_tbl<"tbl">;
   defm TBX_ZZZ  : sve2_int_perm_tbx<"tbx">;
 
   // SVE2 integer compare scalar count and limit
   defm WHILEGE_PWW : sve_int_while4_rr<0b000, "whilege">;
   defm WHILEGT_PWW : sve_int_while4_rr<0b001, "whilegt">;

llvm/trunk/lib/Target/AArch64/SVEInstrFormats.td

 multiclass sve_mem_cstnt_ss<bits<2> msz, string asm, RegisterOperand listty,
                             ZPRRegOp zprty, RegisterOperand gprty> {
   def NAME : sve_mem_cstnt_ss_base<msz, asm, listty, gprty>;
 
   def : InstAlias<asm # "\t$Zt, $Pg, [$Rn, $Rm]",
                  (!cast<Instruction>(NAME) zprty:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, gprty:$Rm), 0>;
 }
 
-class sve2_mem_cstnt_vs_base<bits<3> opc, dag iops, string asm,
-                             RegisterOperand VecList>
-: I<(outs VecList:$Zt), iops,
+class sve2_mem_sstnt_vs_base<bits<3> opc, string asm,
+                             RegisterOperand listty, ZPRRegOp zprty>
+: I<(outs), (ins listty:$Zt, PPR3bAny:$Pg, zprty:$Zn, GPR64:$Rm),
   asm, "\t$Zt, $Pg, [$Zn, $Rm]",
   "",
   []>, Sched<[]> {
   bits<3> Pg;
   bits<5> Rm;
   bits<5> Zn;
   bits<5> Zt;
   let Inst{31-25} = 0b1110010;
   let Inst{24-22} = opc;
   let Inst{21}    = 0b0;
   let Inst{20-16} = Rm;
   let Inst{15-13} = 0b001;
   let Inst{12-10} = Pg;
   let Inst{9-5}   = Zn;
   let Inst{4-0}   = Zt;
 
   let mayStore = 1;
 }
 
-multiclass sve2_mem_cstnt_vs<bits<3> opc, string asm,
+multiclass sve2_mem_sstnt_vs<bits<3> opc, string asm,
                              RegisterOperand listty, ZPRRegOp zprty> {
-  def _REAL : sve2_mem_cstnt_vs_base<opc, (ins PPR3bAny:$Pg, zprty:$Zn, GPR64:$Rm),
-                                     asm, listty>;
+  def _REAL : sve2_mem_sstnt_vs_base<opc, asm, listty, zprty>;
 
   def : InstAlias<asm # "\t$Zt, $Pg, [$Zn, $Rm]",
                  (!cast<Instruction>(NAME # _REAL) zprty:$Zt, PPR3bAny:$Pg, zprty:$Zn, GPR64:$Rm), 0>;
   def : InstAlias<asm # "\t$Zt, $Pg, [$Zn]",
                  (!cast<Instruction>(NAME # _REAL) zprty:$Zt, PPR3bAny:$Pg, zprty:$Zn, XZR), 0>;
-  def : InstAlias<asm # "\t$Zt, $Pg, [$Zn, $Rm]",
-                 (!cast<Instruction>(NAME # _REAL) listty:$Zt, PPR3bAny:$Pg, zprty:$Zn, GPR64:$Rm), 0>;
   def : InstAlias<asm # "\t$Zt, $Pg, [$Zn]",
                  (!cast<Instruction>(NAME # _REAL) listty:$Zt, PPR3bAny:$Pg, zprty:$Zn, XZR), 1>;
 }
 
 class sve_mem_sst_sv<bits<3> opc, bit xs, bit scaled, string asm,
                      RegisterOperand VecList, RegisterOperand zprext>
 : I<(outs), (ins VecList:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, zprext:$Zm),
   asm, "\t$Zt, $Pg, [$Rn, $Zm]",

 
 multiclass sve_mem_p_fill<string asm> {
   def NAME : sve_mem_p_fill<asm>;
 
   def : InstAlias<asm # "\t$Pt, [$Rn]",
                   (!cast<Instruction>(NAME) PPRAny:$Pt, GPR64sp:$Rn, 0), 1>;
 }
 
-class sve2_mem_cldnt_vs_base<bits<5> opc, dag iops, string asm,
+class sve2_mem_gldnt_vs_base<bits<5> opc, dag iops, string asm,
                              RegisterOperand VecList>
 : I<(outs VecList:$Zt), iops,
   asm, "\t$Zt, $Pg/z, [$Zn, $Rm]",
   "",
   []>, Sched<[]> {
   bits<3> Pg;
   bits<5> Rm;
   bits<5> Zn;
   bits<5> Zt;
   let Inst{31}    = 0b1;
   let Inst{30}    = opc{4};
   let Inst{29-25} = 0b00010;
   let Inst{24-23} = opc{3-2};
   let Inst{22-21} = 0b00;
   let Inst{20-16} = Rm;
   let Inst{15}    = 0b1;
   let Inst{14-13} = opc{1-0};
   let Inst{12-10} = Pg;
   let Inst{9-5}   = Zn;
   let Inst{4-0}   = Zt;
 
   let mayLoad = 1;
 }
 
-multiclass sve2_mem_cldnt_vs<bits<5> opc, string asm,
+multiclass sve2_mem_gldnt_vs<bits<5> opc, string asm,
                              RegisterOperand listty, ZPRRegOp zprty> {
-  def _REAL : sve2_mem_cldnt_vs_base<opc, (ins PPR3bAny:$Pg, zprty:$Zn, GPR64:$Rm),
+  def _REAL : sve2_mem_gldnt_vs_base<opc, (ins PPR3bAny:$Pg, zprty:$Zn, GPR64:$Rm),
                                      asm, listty>;
 
   def : InstAlias<asm # "\t$Zt, $Pg/z, [$Zn, $Rm]",
                  (!cast<Instruction>(NAME # _REAL) zprty:$Zt, PPR3bAny:$Pg, zprty:$Zn, GPR64:$Rm), 0>;
   def : InstAlias<asm # "\t$Zt, $Pg/z, [$Zn]",
                  (!cast<Instruction>(NAME # _REAL) zprty:$Zt, PPR3bAny:$Pg, zprty:$Zn, XZR), 0>;
-  def : InstAlias<asm # "\t$Zt, $Pg/z, [$Zn, $Rm]",
-                 (!cast<Instruction>(NAME # _REAL) listty:$Zt, PPR3bAny:$Pg, zprty:$Zn, GPR64:$Rm), 0>;
   def : InstAlias<asm # "\t$Zt, $Pg/z, [$Zn]",
                  (!cast<Instruction>(NAME # _REAL) listty:$Zt, PPR3bAny:$Pg, zprty:$Zn, XZR), 1>;
 }
 
 //===----------------------------------------------------------------------===//
 // SVE Memory - 64-bit Gather Group
 //===----------------------------------------------------------------------===//