[Clang][SVE2.1] Add builtins for Multi-vector load and store

 Patch by : David Sherwood <[email protected]>

As described in: ARM-software/acle#257

Reviewed By: kmclaughlin

Differential Revision: https://reviews.llvm.org/D151433

Author: CarolineConcatto

clang/include/clang/Basic/arm_sve.td

@@ -1880,6 +1880,78 @@ def SVWHILELO_COUNT  : SInst<"svwhilelo_{d}",  "}nni", "QcQsQiQl", MergeNone, "a
 def SVWHILELS_COUNT  : SInst<"svwhilels_{d}",  "}nni", "QcQsQiQl", MergeNone, "aarch64_sve_whilels_{d}", [IsOverloadNone], [ImmCheck<2, ImmCheck2_4_Mul2>]>;
 def SVWHILEHI_COUNT  : SInst<"svwhilehi_{d}",  "}nni", "QcQsQiQl", MergeNone, "aarch64_sve_whilehi_{d}", [IsOverloadNone], [ImmCheck<2, ImmCheck2_4_Mul2>]>;
 def SVWHILEHS_COUNT  : SInst<"svwhilehs_{d}",  "}nni", "QcQsQiQl", MergeNone, "aarch64_sve_whilehs_{d}", [IsOverloadNone], [ImmCheck<2, ImmCheck2_4_Mul2>]>;
+
+def SVLD1B_X2 : MInst<"svld1[_{2}]_x2", "2}c", "cUc", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ld1_pn_x2">;
+def SVLD1H_X2 : MInst<"svld1[_{2}]_x2", "2}c", "sUshb", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ld1_pn_x2">;
+def SVLD1W_X2 : MInst<"svld1[_{2}]_x2", "2}c", "iUif", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ld1_pn_x2">;
+def SVLD1D_X2 : MInst<"svld1[_{2}]_x2", "2}c", "lUld", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ld1_pn_x2">;
+def SVLD1B_X4 : MInst<"svld1[_{2}]_x4", "4}c", "cUc", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ld1_pn_x4">;
+def SVLD1H_X4 : MInst<"svld1[_{2}]_x4", "4}c", "sUshb", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ld1_pn_x4">;
+def SVLD1W_X4 : MInst<"svld1[_{2}]_x4", "4}c", "iUif", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ld1_pn_x4">;
+def SVLD1D_X4 : MInst<"svld1[_{2}]_x4", "4}c", "lUld", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ld1_pn_x4">;
+
+def SVLDNT1B_X2 : MInst<"svldnt1[_{2}]_x2", "2}c", "cUc", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ldnt1_pn_x2">;
+def SVLDNT1H_X2 : MInst<"svldnt1[_{2}]_x2", "2}c", "sUshb", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ldnt1_pn_x2">;
+def SVLDNT1W_X2 : MInst<"svldnt1[_{2}]_x2", "2}c", "iUif", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ldnt1_pn_x2">;
+def SVLDNT1D_X2 : MInst<"svldnt1[_{2}]_x2", "2}c", "lUld", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ldnt1_pn_x2">;
+def SVLDNT1B_X4 : MInst<"svldnt1[_{2}]_x4", "4}c", "cUc", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ldnt1_pn_x4">;
+def SVLDNT1H_X4 : MInst<"svldnt1[_{2}]_x4", "4}c", "sUshb", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ldnt1_pn_x4">;
+def SVLDNT1W_X4 : MInst<"svldnt1[_{2}]_x4", "4}c", "iUif", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ldnt1_pn_x4">;
+def SVLDNT1D_X4 : MInst<"svldnt1[_{2}]_x4", "4}c", "lUld", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ldnt1_pn_x4">;
+
+def SVLD1B_VNUM_X2 : MInst<"svld1_vnum[_{2}]_x2", "2}cl", "cUc", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ld1_pn_x2">;
+def SVLD1H_VNUM_X2 : MInst<"svld1_vnum[_{2}]_x2", "2}cl", "sUshb", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ld1_pn_x2">;
+def SVLD1W_VNUM_X2 : MInst<"svld1_vnum[_{2}]_x2", "2}cl", "iUif", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ld1_pn_x2">;
+def SVLD1D_VNUM_X2 : MInst<"svld1_vnum[_{2}]_x2", "2}cl", "lUld", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ld1_pn_x2">;
+def SVLD1B_VNUM_X4 : MInst<"svld1_vnum[_{2}]_x4", "4}cl", "cUc", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ld1_pn_x4">;
+def SVLD1H_VNUM_X4 : MInst<"svld1_vnum[_{2}]_x4", "4}cl", "sUshb", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ld1_pn_x4">;
+def SVLD1W_VNUM_X4 : MInst<"svld1_vnum[_{2}]_x4", "4}cl", "iUif", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ld1_pn_x4">;
+def SVLD1D_VNUM_X4 : MInst<"svld1_vnum[_{2}]_x4", "4}cl", "lUld", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ld1_pn_x4">;
+
+def SVLDNT1B_VNUM_X2 : MInst<"svldnt1_vnum[_{2}]_x2", "2}cl", "cUc", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ldnt1_pn_x2">;
+def SVLDNT1H_VNUM_X2 : MInst<"svldnt1_vnum[_{2}]_x2", "2}cl", "sUshb", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ldnt1_pn_x2">;
+def SVLDNT1W_VNUM_X2 : MInst<"svldnt1_vnum[_{2}]_x2", "2}cl", "iUif", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ldnt1_pn_x2">;
+def SVLDNT1D_VNUM_X2 : MInst<"svldnt1_vnum[_{2}]_x2", "2}cl", "lUld", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ldnt1_pn_x2">;
+def SVLDNT1B_VNUM_X4 : MInst<"svldnt1_vnum[_{2}]_x4", "4}cl", "cUc", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ldnt1_pn_x4">;
+def SVLDNT1H_VNUM_X4 : MInst<"svldnt1_vnum[_{2}]_x4", "4}cl", "sUshb", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ldnt1_pn_x4">;
+def SVLDNT1W_VNUM_X4 : MInst<"svldnt1_vnum[_{2}]_x4", "4}cl", "iUif", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ldnt1_pn_x4">;
+def SVLDNT1D_VNUM_X4 : MInst<"svldnt1_vnum[_{2}]_x4", "4}cl", "lUld", [IsStructLoad], MemEltTyDefault, "aarch64_sve_ldnt1_pn_x4">;
+
+def SVST1B_X2 : MInst<"svst1[_{2}_x2]", "v}p2", "cUc", [IsStructStore], MemEltTyDefault, "aarch64_sve_st1_pn_x2">;
+def SVST1H_X2 : MInst<"svst1[_{2}_x2]", "v}p2", "sUshb", [IsStructStore], MemEltTyDefault, "aarch64_sve_st1_pn_x2">;
+def SVST1W_X2 : MInst<"svst1[_{2}_x2]", "v}p2", "iUif", [IsStructStore], MemEltTyDefault, "aarch64_sve_st1_pn_x2">;
+def SVST1D_X2 : MInst<"svst1[_{2}_x2]", "v}p2", "lUld", [IsStructStore], MemEltTyDefault, "aarch64_sve_st1_pn_x2">;
+def SVST1B_X4 : MInst<"svst1[_{2}_x4]", "v}p4", "cUc", [IsStructStore], MemEltTyDefault, "aarch64_sve_st1_pn_x4">;
+def SVST1H_X4 : MInst<"svst1[_{2}_x4]", "v}p4", "sUshb", [IsStructStore], MemEltTyDefault, "aarch64_sve_st1_pn_x4">;
+def SVST1W_X4 : MInst<"svst1[_{2}_x4]", "v}p4", "iUif", [IsStructStore], MemEltTyDefault, "aarch64_sve_st1_pn_x4">;
+def SVST1D_X4 : MInst<"svst1[_{2}_x4]", "v}p4", "lUld", [IsStructStore], MemEltTyDefault, "aarch64_sve_st1_pn_x4">;
+
+def SVST1B_VNUM_X2 : MInst<"svst1_vnum[_{2}_x2]", "v}pl2", "cUc", [IsStructStore], MemEltTyDefault, "aarch64_sve_st1_pn_x2">;
+def SVST1H_VNUM_X2 : MInst<"svst1_vnum[_{2}_x2]", "v}pl2", "sUshb", [IsStructStore], MemEltTyDefault, "aarch64_sve_st1_pn_x2">;
+def SVST1W_VNUM_X2 : MInst<"svst1_vnum[_{2}_x2]", "v}pl2", "iUif", [IsStructStore], MemEltTyDefault, "aarch64_sve_st1_pn_x2">;
+def SVST1D_VNUM_X2 : MInst<"svst1_vnum[_{2}_x2]", "v}pl2", "lUld", [IsStructStore], MemEltTyDefault, "aarch64_sve_st1_pn_x2">;
+def SVST1B_VNUM_X4 : MInst<"svst1_vnum[_{2}_x4]", "v}pl4", "cUc", [IsStructStore], MemEltTyDefault, "aarch64_sve_st1_pn_x4">;
+def SVST1H_VNUM_X4 : MInst<"svst1_vnum[_{2}_x4]", "v}pl4", "sUshb", [IsStructStore], MemEltTyDefault, "aarch64_sve_st1_pn_x4">;
+def SVST1W_VNUM_X4 : MInst<"svst1_vnum[_{2}_x4]", "v}pl4", "iUif", [IsStructStore], MemEltTyDefault, "aarch64_sve_st1_pn_x4">;
+def SVST1D_VNUM_X4 : MInst<"svst1_vnum[_{2}_x4]", "v}pl4", "lUld", [IsStructStore], MemEltTyDefault, "aarch64_sve_st1_pn_x4">;
+
+def SVSTNT1B_X2 : MInst<"svstnt1[_{2}_x2]", "v}p2", "cUc", [IsStructStore], MemEltTyDefault, "aarch64_sve_stnt1_pn_x2">;
+def SVSTNT1H_X2 : MInst<"svstnt1[_{2}_x2]", "v}p2", "sUshb", [IsStructStore], MemEltTyDefault, "aarch64_sve_stnt1_pn_x2">;
+def SVSTNT1W_X2 : MInst<"svstnt1[_{2}_x2]", "v}p2", "iUif", [IsStructStore], MemEltTyDefault, "aarch64_sve_stnt1_pn_x2">;
+def SVSTNT1D_X2 : MInst<"svstnt1[_{2}_x2]", "v}p2", "lUld", [IsStructStore], MemEltTyDefault, "aarch64_sve_stnt1_pn_x2">;
+def SVSTNT1B_X4 : MInst<"svstnt1[_{2}_x4]", "v}p4", "cUc", [IsStructStore], MemEltTyDefault, "aarch64_sve_stnt1_pn_x4">;
+def SVSTNT1H_X4 : MInst<"svstnt1[_{2}_x4]", "v}p4", "sUshb", [IsStructStore], MemEltTyDefault, "aarch64_sve_stnt1_pn_x4">;
+def SVSTNT1W_X4 : MInst<"svstnt1[_{2}_x4]", "v}p4", "iUif", [IsStructStore], MemEltTyDefault, "aarch64_sve_stnt1_pn_x4">;
+def SVSTNT1D_X4 : MInst<"svstnt1[_{2}_x4]", "v}p4", "lUld", [IsStructStore], MemEltTyDefault, "aarch64_sve_stnt1_pn_x4">;
+
+def SVSTNT1B_VNUM_X2 : MInst<"svstnt1_vnum[_{2}_x2]", "v}pl2", "cUc", [IsStructStore], MemEltTyDefault, "aarch64_sve_stnt1_pn_x2">;
+def SVSTNT1H_VNUM_X2 : MInst<"svstnt1_vnum[_{2}_x2]", "v}pl2", "sUshb", [IsStructStore], MemEltTyDefault, "aarch64_sve_stnt1_pn_x2">;
+def SVSTNT1W_VNUM_X2 : MInst<"svstnt1_vnum[_{2}_x2]", "v}pl2", "iUif", [IsStructStore], MemEltTyDefault, "aarch64_sve_stnt1_pn_x2">;
+def SVSTNT1D_VNUM_X2 : MInst<"svstnt1_vnum[_{2}_x2]", "v}pl2", "lUld", [IsStructStore], MemEltTyDefault, "aarch64_sve_stnt1_pn_x2">;
+def SVSTNT1B_VNUM_X4 : MInst<"svstnt1_vnum[_{2}_x4]", "v}pl4", "cUc", [IsStructStore], MemEltTyDefault, "aarch64_sve_stnt1_pn_x4">;
+def SVSTNT1H_VNUM_X4 : MInst<"svstnt1_vnum[_{2}_x4]", "v}pl4", "sUshb", [IsStructStore], MemEltTyDefault, "aarch64_sve_stnt1_pn_x4">;
+def SVSTNT1W_VNUM_X4 : MInst<"svstnt1_vnum[_{2}_x4]", "v}pl4", "iUif", [IsStructStore], MemEltTyDefault, "aarch64_sve_stnt1_pn_x4">;
+def SVSTNT1D_VNUM_X4 : MInst<"svstnt1_vnum[_{2}_x4]", "v}pl4", "lUld", [IsStructStore], MemEltTyDefault, "aarch64_sve_stnt1_pn_x4">;
 }
 
 let TargetGuard = "sve2p1" in {

clang/lib/CodeGen/CGBuiltin.cpp

@@ -9340,6 +9340,11 @@ static llvm::ScalableVectorType *getSVEVectorForElementType(llvm::Type *EltTy) {
 // the elements of the specified datatype.
 Value *CodeGenFunction::EmitSVEPredicateCast(Value *Pred,
                                              llvm::ScalableVectorType *VTy) {
+
+  if (isa<TargetExtType>(Pred->getType()) &&
+      cast<TargetExtType>(Pred->getType())->getName() == "aarch64.svcount")
+    return Pred;
+
   auto *RTy = llvm::VectorType::get(IntegerType::get(getLLVMContext(), 1), VTy);
   if (Pred->getType() == RTy)
     return Pred;
@@ -9514,12 +9519,16 @@ Value *CodeGenFunction::EmitSVEStructLoad(const SVETypeFlags &TypeFlags,
   unsigned N;
   switch (IntID) {
   case Intrinsic::aarch64_sve_ld2_sret:
+  case Intrinsic::aarch64_sve_ld1_pn_x2:
+  case Intrinsic::aarch64_sve_ldnt1_pn_x2:
     N = 2;
     break;
   case Intrinsic::aarch64_sve_ld3_sret:
     N = 3;
     break;
   case Intrinsic::aarch64_sve_ld4_sret:
+  case Intrinsic::aarch64_sve_ld1_pn_x4:
+  case Intrinsic::aarch64_sve_ldnt1_pn_x4:
     N = 4;
     break;
   default:
@@ -9555,12 +9564,16 @@ Value *CodeGenFunction::EmitSVEStructStore(const SVETypeFlags &TypeFlags,
   unsigned N;
   switch (IntID) {
   case Intrinsic::aarch64_sve_st2:
+  case Intrinsic::aarch64_sve_st1_pn_x2:
+  case Intrinsic::aarch64_sve_stnt1_pn_x2:
     N = 2;
     break;
   case Intrinsic::aarch64_sve_st3:
     N = 3;
     break;
   case Intrinsic::aarch64_sve_st4:
+  case Intrinsic::aarch64_sve_st1_pn_x4:
+  case Intrinsic::aarch64_sve_stnt1_pn_x4:
     N = 4;
     break;
   default: