[Hexagon] Make conversions to vector predicate types explicit for builtins

HVX does not have load/store instructions for vector predicates (i.e. bool
vectors). Because of that, vector predicates need to be converted to another
type before being stored, and the most convenient representation is an HVX
vector.
As a consequence, in C/C++, source-level builtins that either take or
produce vector predicates take or return regular vectors instead. On the
other hand, the corresponding LLVM intrinsics do have boolean types that,
and so a conversion of the operand or the return value was necessary.
This conversion would happen inside clang's codegen, but was somewhat
fragile.

This patch changes the strategy: a builtin that takes a vector predicate
now really expects a vector predicate. Since such a predicate cannot be
provided via a variable, this builtin must be composed with other builtins
that either convert vector to a predicate (V6_vandvrt) or predicate to a
vector (V6_vandqrt).

For users using builtins defined in hvx_hexagon_protos.h there is no impact:
the conversions were added to that file. Other users will need to insert
- __builtin_HEXAGON_V6_vandvrt[_128B](V, -1) to convert vector V to a
  vector predicate, or
- __builtin_HEXAGON_V6_vandqrt[_128B](Q, -1) to convert vector predicate Q
  to a vector.

Builtins __builtin_HEXAGON_V6_vmaskedstore.* are a temporary exception to
that, but they are deprecated and should not be used anyway. In the future
they will either follow the same rule, or be removed.

Author: Krzysztof Parzyszek

clang/include/clang/Basic/BuiltinsHexagonMapCustomDep.def

@@ -8,199 +8,7 @@
 // Automatically generated file, do not edit!
 //===----------------------------------------------------------------------===//
 
-CUSTOM_BUILTIN_MAPPING(A2_add, 0)
-CUSTOM_BUILTIN_MAPPING(A2_addi, 0)
-CUSTOM_BUILTIN_MAPPING(A2_addp, 0)
-CUSTOM_BUILTIN_MAPPING(A2_and, 0)
-CUSTOM_BUILTIN_MAPPING(A2_andir, 0)
-CUSTOM_BUILTIN_MAPPING(A2_neg, 0)
-CUSTOM_BUILTIN_MAPPING(A2_not, 0)
-CUSTOM_BUILTIN_MAPPING(A2_or, 0)
-CUSTOM_BUILTIN_MAPPING(A2_orir, 0)
-CUSTOM_BUILTIN_MAPPING(A2_sub, 0)
-CUSTOM_BUILTIN_MAPPING(A2_subp, 0)
-CUSTOM_BUILTIN_MAPPING(A2_subri, 0)
-CUSTOM_BUILTIN_MAPPING(A2_sxtb, 0)
-CUSTOM_BUILTIN_MAPPING(A2_sxth, 0)
-CUSTOM_BUILTIN_MAPPING(A2_xor, 0)
-CUSTOM_BUILTIN_MAPPING(A2_zxtb, 0)
-CUSTOM_BUILTIN_MAPPING(A2_zxth, 0)
-CUSTOM_BUILTIN_MAPPING(M2_dpmpyss_s0, 0)
-CUSTOM_BUILTIN_MAPPING(M2_dpmpyuu_s0, 0)
-CUSTOM_BUILTIN_MAPPING(M2_mpyi, 0)
-CUSTOM_BUILTIN_MAPPING(M2_mpysmi, 0)
-CUSTOM_BUILTIN_MAPPING(M2_mpyui, 0)
-CUSTOM_BUILTIN_MAPPING(S2_asl_i_p, 0)
-CUSTOM_BUILTIN_MAPPING(S2_asl_i_r, 0)
-CUSTOM_BUILTIN_MAPPING(S2_asr_i_p, 0)
-CUSTOM_BUILTIN_MAPPING(S2_asr_i_r, 0)
-CUSTOM_BUILTIN_MAPPING(S2_lsr_i_p, 0)
-CUSTOM_BUILTIN_MAPPING(S2_lsr_i_r, 0)
-CUSTOM_BUILTIN_MAPPING(V6_pred_and, 64)
-CUSTOM_BUILTIN_MAPPING(V6_pred_and_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_pred_and_n, 64)
-CUSTOM_BUILTIN_MAPPING(V6_pred_and_n_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_pred_not, 64)
-CUSTOM_BUILTIN_MAPPING(V6_pred_not_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_pred_or, 64)
-CUSTOM_BUILTIN_MAPPING(V6_pred_or_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_pred_or_n, 64)
-CUSTOM_BUILTIN_MAPPING(V6_pred_or_n_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_pred_scalar2, 64)
-CUSTOM_BUILTIN_MAPPING(V6_pred_scalar2_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_pred_xor, 64)
-CUSTOM_BUILTIN_MAPPING(V6_pred_xor_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vS32b_nqpred_ai, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vS32b_nqpred_ai_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vS32b_nt_nqpred_ai, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vS32b_nt_nqpred_ai_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vS32b_nt_qpred_ai, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vS32b_nt_qpred_ai_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vS32b_qpred_ai, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vS32b_qpred_ai_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vaddbnq, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vaddbnq_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vaddbq, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vaddbq_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vaddhnq, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vaddhnq_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vaddhq, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vaddhq_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vaddwnq, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vaddwnq_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vaddwq, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vaddwq_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vandqrt, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vandqrt_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vandqrt_acc, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vandqrt_acc_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vandvrt, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vandvrt_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vandvrt_acc, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vandvrt_acc_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_veqb, 64)
-CUSTOM_BUILTIN_MAPPING(V6_veqb_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_veqb_and, 64)
-CUSTOM_BUILTIN_MAPPING(V6_veqb_and_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_veqb_or, 64)
-CUSTOM_BUILTIN_MAPPING(V6_veqb_or_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_veqb_xor, 64)
-CUSTOM_BUILTIN_MAPPING(V6_veqb_xor_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_veqh, 64)
-CUSTOM_BUILTIN_MAPPING(V6_veqh_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_veqh_and, 64)
-CUSTOM_BUILTIN_MAPPING(V6_veqh_and_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_veqh_or, 64)
-CUSTOM_BUILTIN_MAPPING(V6_veqh_or_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_veqh_xor, 64)
-CUSTOM_BUILTIN_MAPPING(V6_veqh_xor_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_veqw, 64)
-CUSTOM_BUILTIN_MAPPING(V6_veqw_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_veqw_and, 64)
-CUSTOM_BUILTIN_MAPPING(V6_veqw_and_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_veqw_or, 64)
-CUSTOM_BUILTIN_MAPPING(V6_veqw_or_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_veqw_xor, 64)
-CUSTOM_BUILTIN_MAPPING(V6_veqw_xor_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgtb, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgtb_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgtb_and, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgtb_and_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgtb_or, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgtb_or_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgtb_xor, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgtb_xor_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgth, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgth_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgth_and, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgth_and_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgth_or, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgth_or_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgth_xor, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgth_xor_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgtub, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgtub_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgtub_and, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgtub_and_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgtub_or, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgtub_or_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgtub_xor, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgtub_xor_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgtuh, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgtuh_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgtuh_and, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgtuh_and_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgtuh_or, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgtuh_or_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgtuh_xor, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgtuh_xor_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgtuw, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgtuw_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgtuw_and, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgtuw_and_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgtuw_or, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgtuw_or_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgtuw_xor, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgtuw_xor_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgtw, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgtw_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgtw_and, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgtw_and_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgtw_or, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgtw_or_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgtw_xor, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgtw_xor_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vmux, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vmux_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vsubbnq, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vsubbnq_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vsubbq, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vsubbq_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vsubhnq, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vsubhnq_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vsubhq, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vsubhq_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vsubwnq, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vsubwnq_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vsubwq, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vsubwq_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vswap, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vswap_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_pred_scalar2v2, 64)
-CUSTOM_BUILTIN_MAPPING(V6_pred_scalar2v2_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_shuffeqh, 64)
-CUSTOM_BUILTIN_MAPPING(V6_shuffeqh_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_shuffeqw, 64)
-CUSTOM_BUILTIN_MAPPING(V6_shuffeqw_128B, 128)
 CUSTOM_BUILTIN_MAPPING(V6_vaddcarry, 64)
 CUSTOM_BUILTIN_MAPPING(V6_vaddcarry_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vandnqrt, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vandnqrt_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vandnqrt_acc, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vandnqrt_acc_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vandvnqv, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vandvnqv_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vandvqv, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vandvqv_128B, 128)
 CUSTOM_BUILTIN_MAPPING(V6_vsubcarry, 64)
 CUSTOM_BUILTIN_MAPPING(V6_vsubcarry_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgathermhq, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgathermhq_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgathermhwq, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgathermhwq_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vgathermwq, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vgathermwq_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vprefixqb, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vprefixqb_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vprefixqh, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vprefixqh_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vprefixqw, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vprefixqw_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vscattermhq, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vscattermhq_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vscattermhwq, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vscattermhwq_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vscattermwq, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vscattermwq_128B, 128)
-CUSTOM_BUILTIN_MAPPING(V6_vaddcarrysat, 64)
-CUSTOM_BUILTIN_MAPPING(V6_vaddcarrysat_128B, 128)

clang/lib/CodeGen/CGBuiltin.cpp

@@ -18596,6 +18596,7 @@ getIntrinsicForHexagonNonGCCBuiltin(unsigned BuiltinID) {
     CUSTOM_BUILTIN_MAPPING(S2_storerf_pcr, 0)
     CUSTOM_BUILTIN_MAPPING(S2_storeri_pcr, 0)
     CUSTOM_BUILTIN_MAPPING(S2_storerd_pcr, 0)
+    // Legacy builtins that take a vector in place of a vector predicate.
     CUSTOM_BUILTIN_MAPPING(V6_vmaskedstoreq, 64)
     CUSTOM_BUILTIN_MAPPING(V6_vmaskedstorenq, 64)
     CUSTOM_BUILTIN_MAPPING(V6_vmaskedstorentq, 64)
@@ -18733,6 +18734,27 @@ Value *CodeGenFunction::EmitHexagonBuiltinExpr(unsigned BuiltinID,
     return Builder.CreateExtractValue(Result, 0);
   }
 
+  case Hexagon::BI__builtin_HEXAGON_V6_vmaskedstoreq:
+  case Hexagon::BI__builtin_HEXAGON_V6_vmaskedstorenq:
+  case Hexagon::BI__builtin_HEXAGON_V6_vmaskedstorentq:
+  case Hexagon::BI__builtin_HEXAGON_V6_vmaskedstorentnq:
+  case Hexagon::BI__builtin_HEXAGON_V6_vmaskedstoreq_128B:
+  case Hexagon::BI__builtin_HEXAGON_V6_vmaskedstorenq_128B:
+  case Hexagon::BI__builtin_HEXAGON_V6_vmaskedstorentq_128B:
+  case Hexagon::BI__builtin_HEXAGON_V6_vmaskedstorentnq_128B: {
+    SmallVector<llvm::Value*,4> Ops;
+    const Expr *PredOp = E->getArg(0);
+    // There will be an implicit cast to a boolean vector. Strip it.
+    if (auto *Cast = dyn_cast<ImplicitCastExpr>(PredOp)) {
+      if (Cast->getCastKind() == CK_BitCast)
+        PredOp = Cast->getSubExpr();
+      Ops.push_back(V2Q(EmitScalarExpr(PredOp)));
+    }
+    for (int i = 1, e = E->getNumArgs(); i != e; ++i)
+      Ops.push_back(EmitScalarExpr(E->getArg(i)));
+    return Builder.CreateCall(CGM.getIntrinsic(ID), Ops);
+  }
+
   case Hexagon::BI__builtin_HEXAGON_L2_loadrub_pci:
   case Hexagon::BI__builtin_HEXAGON_L2_loadrb_pci:
   case Hexagon::BI__builtin_HEXAGON_L2_loadruh_pci:
@@ -18769,40 +18791,6 @@ Value *CodeGenFunction::EmitHexagonBuiltinExpr(unsigned BuiltinID,
     return MakeBrevLd(Intrinsic::hexagon_L2_loadri_pbr, Int32Ty);
   case Hexagon::BI__builtin_brev_ldd:
     return MakeBrevLd(Intrinsic::hexagon_L2_loadrd_pbr, Int64Ty);
-
-  default: {
-    if (ID == Intrinsic::not_intrinsic)
-      return nullptr;
-
-    auto IsVectorPredTy = [](llvm::Type *T) {
-      return T->isVectorTy() &&
-             cast<llvm::VectorType>(T)->getElementType()->isIntegerTy(1);
-    };
-
-    llvm::Function *IntrFn = CGM.getIntrinsic(ID);
-    llvm::FunctionType *IntrTy = IntrFn->getFunctionType();
-    SmallVector<llvm::Value*,4> Ops;
-    for (unsigned i = 0, e = IntrTy->getNumParams(); i != e; ++i) {
-      llvm::Type *T = IntrTy->getParamType(i);
-      const Expr *A = E->getArg(i);
-      if (IsVectorPredTy(T)) {
-        // There will be an implicit cast to a boolean vector. Strip it.
-        if (auto *Cast = dyn_cast<ImplicitCastExpr>(A)) {
-          if (Cast->getCastKind() == CK_BitCast)
-            A = Cast->getSubExpr();
-        }
-        Ops.push_back(V2Q(EmitScalarExpr(A)));
-      } else {
-        Ops.push_back(EmitScalarExpr(A));
-      }
-    }
-
-    llvm::Value *Call = Builder.CreateCall(IntrFn, Ops);
-    if (IsVectorPredTy(IntrTy->getReturnType()))
-      Call = Q2V(Call);
-
-    return Call;
-  } // default
   } // switch
 
   return nullptr;