[LLVM][CodeGen][AArch64] Add isel for i8/i16 sve.compact intrinsics. (#143139)

The i8/i16 instruction variants are available for:
  * normal functions using sve2p2
  * streaming functions using sme2p2

Author: paulwalker-arm

llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td

@@ -4499,7 +4499,7 @@ let Predicates = [HasNonStreamingSVE2p2_or_SME2p2] in {
   // SVE2 EXPAND
   defm EXPAND_ZPZ : sve2_int_perm_expand<"expand">;
   // SVE COMPACT - byte and halfword
-  defm COMPACT_ZPZ : sve_int_perm_compact_bh<"compact">;
+  defm COMPACT_ZPZ : sve_int_perm_compact_bh<"compact", int_aarch64_sve_compact>;
 }
 
 //===----------------------------------------------------------------------===//

llvm/lib/Target/AArch64/SVEInstrFormats.td

@@ -7884,9 +7884,14 @@ multiclass sve_int_perm_compact_sd<string asm, SDPatternOperator op> {
   def : SVE_2_Op_Pat<nxv2f64, op, nxv2i1, nxv2f64, !cast<Instruction>(NAME # _D)>;
 }
 
-multiclass sve_int_perm_compact_bh<string asm> {
+multiclass sve_int_perm_compact_bh<string asm, SDPatternOperator op> {
   def _B : sve_int_perm_compact<0b00, asm, ZPR8>;
   def _H : sve_int_perm_compact<0b01, asm, ZPR16>;
+
+  def : SVE_2_Op_Pat<nxv16i8,  op, nxv16i1, nxv16i8,  !cast<Instruction>(NAME # _B)>;
+  def : SVE_2_Op_Pat<nxv8i16,  op, nxv8i1,  nxv8i16,  !cast<Instruction>(NAME # _H)>;
+  def : SVE_2_Op_Pat<nxv8f16,  op, nxv8i1,  nxv8f16,  !cast<Instruction>(NAME # _H)>;
+  def : SVE_2_Op_Pat<nxv8bf16, op, nxv8i1,  nxv8bf16, !cast<Instruction>(NAME # _H)>;
 }
 
 //===----------------------------------------------------------------------===//

llvm/test/CodeGen/AArch64/sve2p2-intrinsics.ll

@@ -0,0 +1,52 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve2p2 < %s | FileCheck %s --check-prefixes=CHECK
+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sme2p2 -force-streaming < %s | FileCheck %s --check-prefixes=CHECK
+
+;
+; COMPACT
+;
+
+define <vscale x 16 x i8> @compact_i8(<vscale x 16 x i1> %pg, <vscale x 16 x i8> %a) {
+; CHECK-LABEL: compact_i8:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    compact z0.b, p0, z0.b
+; CHECK-NEXT:    ret
+  %out = call <vscale x 16 x i8> @llvm.aarch64.sve.compact.nxv16i8(<vscale x 16 x i1> %pg,
+                                                                   <vscale x 16 x i8> %a)
+  ret <vscale x 16 x i8> %out
+}
+
+define <vscale x 8 x i16> @compact_i16(<vscale x 8 x i1> %pg, <vscale x 8 x i16> %a) {
+; CHECK-LABEL: compact_i16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    compact z0.h, p0, z0.h
+; CHECK-NEXT:    ret
+  %out = call <vscale x 8 x i16> @llvm.aarch64.sve.compact.nxv8i16(<vscale x 8 x i1> %pg,
+                                                                   <vscale x 8 x i16> %a)
+  ret <vscale x 8 x i16> %out
+}
+
+define <vscale x 8 x half> @compact_f16(<vscale x 8 x i1> %pg, <vscale x 8 x half> %a) {
+; CHECK-LABEL: compact_f16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    compact z0.h, p0, z0.h
+; CHECK-NEXT:    ret
+  %out = call <vscale x 8 x half> @llvm.aarch64.sve.compact.nxv8f16(<vscale x 8 x i1> %pg,
+                                                                    <vscale x 8 x half> %a)
+  ret <vscale x 8 x half> %out
+}
+
+define <vscale x 8 x bfloat> @compact_bf16(<vscale x 8 x i1> %pg, <vscale x 8 x bfloat> %a) {
+; CHECK-LABEL: compact_bf16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    compact z0.h, p0, z0.h
+; CHECK-NEXT:    ret
+  %out = call <vscale x 8 x bfloat> @llvm.aarch64.sve.compact.nxv8bf16(<vscale x 8 x i1> %pg,
+                                                                       <vscale x 8 x bfloat> %a)
+  ret <vscale x 8 x bfloat> %out
+}
+
+declare <vscale x 16 x i8> @llvm.aarch64.sve.compact.nxv16i8(<vscale x 16 x i1>, <vscale x 16 x i8>)
+declare <vscale x 8 x i16> @llvm.aarch64.sve.compact.nxv8i16(<vscale x 8 x i1>, <vscale x 8 x i16>)
+declare <vscale x 8 x half> @llvm.aarch64.sve.compact.nxv8f16(<vscale x 8 x i1>, <vscale x 8 x half>)
+declare <vscale x 8 x bfloat> @llvm.aarch64.sve.compact.nxv8bf16(<vscale x 8 x i1>, <vscale x 8 x bfloat>)