[Polly] Do not introduce address space cast

[etherzhhb]

Summary

Do not introduce address space cast in IslNodeBuilder::preloadUnconditionally.

[etherzhhb]

I want to also include a test case, if you can point me to the test case related to IslNodeBuilder::preloadUnconditionally, I can add one.
Let me search polly test dir at the same time

[grosser]

Hi ether,

thank you for the patch. Related test cases are:

Polly :: Isl/CodeGen/inv-load-lnt-crash-wrong-order-3.ll
Polly :: Isl/CodeGen/inv-load-lnt-crash-wrong-order.ll
Polly :: Isl/CodeGen/invariant_cannot_handle_void.ll
Polly :: Isl/CodeGen/invariant_load_complex_condition.ll
Polly :: Isl/CodeGen/invariant_load_different_sized_types.ll
Polly :: Isl/CodeGen/invariant_loads_from_struct_with_different_types_1.ll
Polly :: Isl/CodeGen/invariant_loads_from_struct_with_different_types_2.ll
Polly :: Isl/CodeGen/multiple-types-invariant-load.ll
Polly :: ScopInfo/invariant-loads-leave-read-only-statements.ll
Polly :: ScopInfo/invariant_load_access_classes_different_base_type.ll
Polly :: ScopInfo/invariant_load_access_classes_different_base_type_escaping.ll
Polly :: ScopInfo/invariant_load_access_classes_different_base_type_same_pointer.ll
Polly :: ScopInfo/invariant_load_access_classes_different_base_type_same_pointer_escaping.ll

[etherzhhb]

Add testcase

[grosser]

Also, please add llvm-commits and pollydev in CC (see other polly reviews) and also mark this as #Polly project.

[grosser]

Hi ether,

I just run your patch, but the test case you added breaks:

opt  -polly-process-unprofitable  -polly-remarks-minimal  -polly-codegen -polly-invariant-load-hoisting=true -S < /home/grosser/Projects/polly/git/tools/polly/test/Isl/CodeGen/invariant_load_address_space.ll | FileCheck /home/grosser/Projects/polly/git/tools/polly/test/Isl/CodeGen/invariant_load_address_space.ll
--
Exit Code: 1

Command Output (stderr):
--
/home/grosser/Projects/polly/git/tools/polly/test/Isl/CodeGen/invariant_load_address_space.ll:4:15: error: expected string not found in input
; CHECK-NEXT: %polly.access.B = getelementptr i32, i32 addrspace(5)* %B, i64 0
              ^
<stdin>:13:22: note: scanning from here
polly.preload.begin: ; preds = %polly.split_new_and_old
                     ^
<stdin>:14:2: note: possible intended match here
 %polly.access.B = getelementptr i32, i32 addrspace(1)* %B, i64 0

The input code does not contain any element from address space 5? Is this intended?

[etherzhhb]

Fix the test case.

[grosser]

LGTM. Thank you!

polly/trunk/lib/CodeGen/IslNodeBuilder.cpp

Value *IslNodeBuilder::preloadUnconditionally(isl_set *AccessRange,
   Value *PreloadVal;
 
   // Correct the type as the SAI might have a different type than the user
   // expects, especially if the base pointer is a struct.
   Type *Ty = AccInst->getType();
 
   auto *Ptr = AddressValue;
   auto Name = Ptr->getName();
-  Ptr = Builder.CreatePointerCast(Ptr, Ty->getPointerTo(), Name + ".cast");
+  auto AS = Ptr->getType()->getPointerAddressSpace();
+  Ptr = Builder.CreatePointerCast(Ptr, Ty->getPointerTo(AS), Name + ".cast");
   PreloadVal = Builder.CreateLoad(Ptr, Name + ".load");
   if (LoadInst *PreloadInst = dyn_cast<LoadInst>(PreloadVal))
     PreloadInst->setAlignment(dyn_cast<LoadInst>(AccInst)->getAlignment());
 
   // TODO: This is only a hot fix for SCoP sequences that use the same load
   //       instruction contained and hoisted by one of the SCoPs.
   if (SE.isSCEVable(Ty))
     SE.forgetValue(AccInst);

polly/trunk/test/Isl/CodeGen/invariant_load_address_space.ll

+; RUN: opt %loadPolly -polly-codegen -polly-invariant-load-hoisting=true -S < %s | FileCheck %s
+;
+; CHECK-LABEL: polly.preload.begin:
+; CHECK-NEXT:    %polly.access.B = getelementptr i32, i32 addrspace(1)* %B, i64 0
+; CHECK-NOT:     addrspacecast 
+; CHECK-NEXT:    %polly.access.B.load = load i32, i32 addrspace(1)* %polly.access.B
+;
+; CHECK-LABEL: polly.stmt.bb2:
+; CHECK-NEXT:    %scevgep = getelementptr i32, i32* %A, i64 %polly.indvar
+; CHECK-NEXT:    store i32 %polly.access.B.load, i32* %scevgep, align 4
+;
+;    void f(int *restrict A, int *restrict B) {
+;      for (int i = 0; i < 1024; i++)
+;        A[i] = *B;
+;    }
+;
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+
+define void @f(i32* noalias %A, i32 addrspace(1)* noalias %B) {
+bb:
+  br label %bb1
+
+bb1:                                              ; preds = %bb4, %bb
+  %indvars.iv = phi i64 [ %indvars.iv.next, %bb4 ], [ 0, %bb ]
+  %exitcond = icmp ne i64 %indvars.iv, 1024
+  br i1 %exitcond, label %bb2, label %bb5
+
+bb2:                                              ; preds = %bb1
+  %tmp = load i32, i32 addrspace(1)* %B, align 4
+  %tmp3 = getelementptr inbounds i32, i32* %A, i64 %indvars.iv
+  store i32 %tmp, i32* %tmp3, align 4
+  br label %bb4
+
+bb4:                                              ; preds = %bb2
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  br label %bb1
+
+bb5:                                              ; preds = %bb1
+  ret void
+}