[PATCH] D13127: [ARM] Upgrade codegen for vld[234] and vst[234] to to communicate a 0 address space

[Jeroen Ketema via cfe-commits]

jketema created this revision.
jketema added a reviewer: sbaranga.
jketema added a subscriber: cfe-commits.
Herald added subscribers: rengolin, aemerson.

This if the clang companion patch for http://reviews.llvm.org/D12985, which upgrades the vld[234] and vst[234] to take pointers annotated with an address space.

http://reviews.llvm.org/D13127

Files:
  lib/CodeGen/CGBuiltin.cpp
  test/CodeGen/vld_dup.c

Index: test/CodeGen/vld_dup.c
===================================================================
--- test/CodeGen/vld_dup.c
+++ test/CodeGen/vld_dup.c
@@ -41,10 +41,10 @@
 // CHECK-NEXT: [[T190:%.*]] = insertvalue { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } [[T187]], <2 x i32> [[T189]], 3
 
     v4 = vld3_dup_s64(v6);
-// CHECK: {{%.*}} = call { <1 x i64>, <1 x i64>, <1 x i64> } @llvm.arm.neon.vld3.v1i64(i8* {{.*}}, i32 {{[0-9]+}})
+// CHECK: {{%.*}} = call { <1 x i64>, <1 x i64>, <1 x i64> } @llvm.arm.neon.vld3.v1i64.p0i8(i8* {{.*}}, i32 {{[0-9]+}})
 
     v5 = vld4_dup_s64(v7);
-// CHECK: {{%.*}} = call { <1 x i64>, <1 x i64>, <1 x i64>, <1 x i64> } @llvm.arm.neon.vld4.v1i64(i8* {{.*}}, i32 {{[0-9]+}})
+// CHECK: {{%.*}} = call { <1 x i64>, <1 x i64>, <1 x i64>, <1 x i64> } @llvm.arm.neon.vld4.v1i64.p0i8(i8* {{.*}}, i32 {{[0-9]+}})
 
     return 0;
 }
Index: lib/CodeGen/CGBuiltin.cpp
===================================================================
--- lib/CodeGen/CGBuiltin.cpp
+++ lib/CodeGen/CGBuiltin.cpp
@@ -2890,7 +2890,8 @@
   case NEON::BI__builtin_neon_vld3q_v:
   case NEON::BI__builtin_neon_vld4_v:
   case NEON::BI__builtin_neon_vld4q_v: {
-    Function *F = CGM.getIntrinsic(LLVMIntrinsic, Ty);
+    llvm::Type *Tys[] = {Ty, Int8PtrTy};
+    Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys);
     Value *Align = getAlignmentValue32(PtrOp1);
     Ops[1] = Builder.CreateCall(F, {Ops[1], Align}, NameHint);
     Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
@@ -3019,14 +3020,18 @@
   case NEON::BI__builtin_neon_vshr_n_v:
   case NEON::BI__builtin_neon_vshrq_n_v:
     return EmitNeonRShiftImm(Ops[0], Ops[1], Ty, Usgn, "vshr_n");
-  case NEON::BI__builtin_neon_vst1_v:
-  case NEON::BI__builtin_neon_vst1q_v:
   case NEON::BI__builtin_neon_vst2_v:
   case NEON::BI__builtin_neon_vst2q_v:
   case NEON::BI__builtin_neon_vst3_v:
   case NEON::BI__builtin_neon_vst3q_v:
   case NEON::BI__builtin_neon_vst4_v:
-  case NEON::BI__builtin_neon_vst4q_v:
+  case NEON::BI__builtin_neon_vst4q_v: {
+    llvm::Type *Tys[] = {Int8PtrTy, Ty};
+    Ops.push_back(getAlignmentValue32(PtrOp0));
+    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "");
+  }
+  case NEON::BI__builtin_neon_vst1_v:
+  case NEON::BI__builtin_neon_vst1q_v:
   case NEON::BI__builtin_neon_vst2_lane_v:
   case NEON::BI__builtin_neon_vst2q_lane_v:
   case NEON::BI__builtin_neon_vst3_lane_v:
@@ -3794,7 +3799,8 @@
         break;
       default: llvm_unreachable("unknown vld_dup intrinsic?");
       }
-      Function *F = CGM.getIntrinsic(Int, Ty);
+      llvm::Type *Tys[] = {Ty, Int8PtrTy};
+      Function *F = CGM.getIntrinsic(Int, Tys);
       llvm::Value *Align = getAlignmentValue32(PtrOp1);
       Ops[1] = Builder.CreateCall(F, {Ops[1], Align}, "vld_dup");
       Ty = llvm::PointerType::getUnqual(Ops[1]->getType());


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D13127.35600.patch
Type: text/x-patch
Size: 2841 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20150924/accd2758/attachment.bin>