[PowerPC] Fix __builtin_ppc_load2r to return short instead of int.

[stefanp]

Summary

This patch fixes the return value of the builtin __builtin_ppc_load2r to
correctly return short instead of int.

[nemanjai]

LGTM as long as the sign-extending test is added.

llvm/test/CodeGen/PowerPC/builtins-ppc-xlcompat-load-store-reversed.ll

52

Please add an equivalent test with signext to ensure there is a extsh prior to the return.

[stefanp]

Rebased and added signext test case.

clang/include/clang/Basic/BuiltinsPPC.def

 BUILTIN(__builtin_ppc_maddhd, "LLiLLiLLiLLi", "")
 BUILTIN(__builtin_ppc_maddhdu, "ULLiULLiULLiULLi", "")
 BUILTIN(__builtin_ppc_maddld, "LLiLLiLLiLLi", "")
 // Rotate
 BUILTIN(__builtin_ppc_rlwnm, "UiUiIUiIUi", "")
 BUILTIN(__builtin_ppc_rlwimi, "UiUiUiIUiIUi", "")
 BUILTIN(__builtin_ppc_rldimi, "ULLiULLiULLiIUiIULLi", "")
 // load
-BUILTIN(__builtin_ppc_load2r, "UiUs*", "")
+BUILTIN(__builtin_ppc_load2r, "UsUs*", "")
 BUILTIN(__builtin_ppc_load4r, "UiUi*", "")
 BUILTIN(__builtin_ppc_load8r, "ULLiULLi*", "")
 // store
 BUILTIN(__builtin_ppc_store2r, "vUiUs*", "")
 BUILTIN(__builtin_ppc_store4r, "vUiUi*", "")
 BUILTIN(__builtin_ppc_store8r, "vULLiULLi*", "")
 BUILTIN(__builtin_ppc_extract_exp, "Uid", "")
 BUILTIN(__builtin_ppc_extract_sig, "ULLid", "")

clang/lib/CodeGen/CGBuiltin.cpp

Value *CodeGenFunction::EmitPPCBuiltinExpr(unsigned BuiltinID,
   }
   case PPC::BI__builtin_ppc_rdlam: {
     llvm::Type *Ty = Ops[0]->getType();
     Value *ShiftAmt = Builder.CreateIntCast(Ops[1], Ty, false);
     Function *F = CGM.getIntrinsic(Intrinsic::fshl, Ty);
     Value *Rotate = Builder.CreateCall(F, {Ops[0], Ops[0], ShiftAmt});
     return Builder.CreateAnd(Rotate, Ops[2]);
   }
+  case PPC::BI__builtin_ppc_load2r: {
+    Function *F = CGM.getIntrinsic(Intrinsic::ppc_load2r);
+    Ops[0] = Builder.CreateBitCast(Ops[0], Int8PtrTy);
+    Value *LoadIntrinsic = Builder.CreateCall(F, Ops);
+    return Builder.CreateTrunc(LoadIntrinsic, Int16Ty);
+  }
   // FMA variations
   case PPC::BI__builtin_vsx_xvmaddadp:
   case PPC::BI__builtin_vsx_xvmaddasp:
   case PPC::BI__builtin_vsx_xvnmaddadp:
   case PPC::BI__builtin_vsx_xvnmaddasp:
   case PPC::BI__builtin_vsx_xvmsubadp:
   case PPC::BI__builtin_vsx_xvmsubasp:
   case PPC::BI__builtin_vsx_xvnmsubadp:

clang/test/CodeGen/builtins-ppc-xlcompat-load-store-reversed.c

-// REQUIRES: powerpc-registered-target.
+// REQUIRES: powerpc-registered-target
 // RUN: %clang_cc1 -triple powerpc64-unknown-linux-gnu \
 // RUN:   -emit-llvm %s -o - -target-cpu pwr7 | FileCheck %s
 // RUN: %clang_cc1 -triple powerpc64le-unknown-linux-gnu \
 // RUN:   -emit-llvm %s -o - -target-cpu pwr8 | FileCheck %s
 // RUN: %clang_cc1 -triple powerpc-unknown-aix \
 // RUN:   -emit-llvm %s -o - -target-cpu pwr7 | FileCheck %s --check-prefixes=CHECK-32B
 // RUN: %clang_cc1 -triple powerpc64-unknown-aix \
 // RUN:   -emit-llvm %s -o - -target-cpu pwr7 | FileCheck %s

llvm/include/llvm/IR/IntrinsicsPowerPC.td

let TargetPrefix = "ppc" in {
   def int_ppc_maddhdu
       : GCCBuiltin<"__builtin_ppc_maddhdu">,
         Intrinsic<[llvm_i64_ty], [llvm_i64_ty, llvm_i64_ty, llvm_i64_ty], [IntrNoMem]>;
   def int_ppc_maddld
       : GCCBuiltin<"__builtin_ppc_maddld">,
         Intrinsic<[llvm_i64_ty], [llvm_i64_ty, llvm_i64_ty, llvm_i64_ty], [IntrNoMem]>;
   // load
   def int_ppc_load2r
-      : GCCBuiltin<"__builtin_ppc_load2r">,
-        Intrinsic<[llvm_i32_ty], [llvm_ptr_ty], [IntrReadMem, IntrArgMemOnly]>;
+      : Intrinsic<[llvm_i32_ty], [llvm_ptr_ty], [IntrReadMem, IntrArgMemOnly]>;
   def int_ppc_load4r
       : GCCBuiltin<"__builtin_ppc_load4r">,
         Intrinsic<[llvm_i32_ty], [llvm_ptr_ty], [IntrReadMem, IntrArgMemOnly]>;
   def int_ppc_load8r
       : GCCBuiltin<"__builtin_ppc_load8r">,
         Intrinsic<[llvm_i64_ty], [llvm_ptr_ty], [IntrReadMem, IntrArgMemOnly]>;
   // store
   def int_ppc_store2r

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

case ISD::INTRINSIC_WO_CHAIN: {
     case Intrinsic::ppc_altivec_vcmpgtub_p:
     case Intrinsic::ppc_altivec_vcmpgtuh_p:
     case Intrinsic::ppc_altivec_vcmpgtuw_p:
     case Intrinsic::ppc_altivec_vcmpgtud_p:
     case Intrinsic::ppc_altivec_vcmpgtuq_p:
       Known.Zero = ~1U;  // All bits but the low one are known to be zero.
       break;
     }
+    break;
+  }
+  case ISD::INTRINSIC_W_CHAIN: {
+    switch (cast<ConstantSDNode>(Op.getOperand(1))->getZExtValue()) {
+    default:
+      break;
+    case Intrinsic::ppc_load2r:
+      // Top bits are cleared for load2r (which is the same as lhbrx).
+      Known.Zero = 0xFFFF0000;
+      break;
+    }
+    break;
   }
   }
 }
 
 Align PPCTargetLowering::getPrefLoopAlignment(MachineLoop *ML) const {
   switch (Subtarget.getCPUDirective()) {
   default: break;
   case PPC::DIR_970:

llvm/test/CodeGen/PowerPC/builtins-ppc-xlcompat-load-store-reversed.ll

entry:
   %1 = load i32*, i32** @ui_addr, align 8
   %2 = bitcast i32* %1 to i8*
   call void @llvm.ppc.store4r(i32 %0, i8* %2)
   ret void
 }
 
 declare void @llvm.ppc.store4r(i32, i8*)
 
 define dso_local zeroext i16 @test_builtin_ppc_load2r() {
 ; CHECK-64B-LABEL: test_builtin_ppc_load2r:
 ; CHECK-64B:         lhbrx 3, 0, 3
-; CHECK-64B-NEXT:    clrldi 3, 3, 48
 ; CHECK-64B-NEXT:    blr
 
 ; CHECK-32B-LABEL: test_builtin_ppc_load2r:
 ; CHECK-32B:         lhbrx 3, 0, 3
-; CHECK-32B-NEXT:    clrlwi 3, 3, 16
+; CHECK-32B-NEXT:    blr
+entry:
+  %0 = load i16*, i16** @us_addr, align 8
+  %1 = bitcast i16* %0 to i8*
+  %2 = call i32 @llvm.ppc.load2r(i8* %1)
+  %conv = trunc i32 %2 to i16
+  ret i16 %conv
+}
+
+define dso_local signext i16 @test_builtin_ppc_load2r_signext() {
+; CHECK-64B-LABEL: test_builtin_ppc_load2r_signext:
+; CHECK-64B:         lhbrx 3, 0, 3
+; CHECK-64B-NEXT:    extsh 3, 3
+; CHECK-64B-NEXT:    blr
+
+; CHECK-32B-LABEL: test_builtin_ppc_load2r_signext:
+; CHECK-32B:         lhbrx 3, 0, 3
+; CHECK-32B-NEXT:    extsh 3, 3
 ; CHECK-32B-NEXT:    blr
 entry:
   %0 = load i16*, i16** @us_addr, align 8
   %1 = bitcast i16* %0 to i8*
   %2 = call i32 @llvm.ppc.load2r(i8* %1)
   %conv = trunc i32 %2 to i16
   ret i16 %conv
 }