[llvm] r293859 - InferAddressSpaces: Handle more cases with constant select operands
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Wed Feb 1 19:37:23 PST 2017
Author: arsenm
Date: Wed Feb 1 21:37:22 2017
New Revision: 293859
URL: http://llvm.org/viewvc/llvm-project?rev=293859&view=rev
Log:
InferAddressSpaces: Handle more cases with constant select operands
Modified:
llvm/trunk/lib/Transforms/Scalar/InferAddressSpaces.cpp
llvm/trunk/test/Transforms/InferAddressSpaces/AMDGPU/select.ll
Modified: llvm/trunk/lib/Transforms/Scalar/InferAddressSpaces.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/InferAddressSpaces.cpp?rev=293859&r1=293858&r2=293859&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/InferAddressSpaces.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/InferAddressSpaces.cpp Wed Feb 1 21:37:22 2017
@@ -347,12 +347,18 @@ static Value *operandWithNewAddressSpace
const ValueToValueMapTy &ValueWithNewAddrSpace,
SmallVectorImpl<const Use *> *UndefUsesToFix) {
Value *Operand = OperandUse.get();
+
+ Type *NewPtrTy =
+ Operand->getType()->getPointerElementType()->getPointerTo(NewAddrSpace);
+
+ if (Constant *C = dyn_cast<Constant>(Operand))
+ return ConstantExpr::getAddrSpaceCast(C, NewPtrTy);
+
if (Value *NewOperand = ValueWithNewAddrSpace.lookup(Operand))
return NewOperand;
UndefUsesToFix->push_back(&OperandUse);
- return UndefValue::get(
- Operand->getType()->getPointerElementType()->getPointerTo(NewAddrSpace));
+ return UndefValue::get(NewPtrTy);
}
// Returns a clone of `I` with its operands converted to those specified in
@@ -442,6 +448,18 @@ static Value *cloneConstantExprWithNewAd
return ConstantExpr::getBitCast(CE->getOperand(0), TargetType);
}
+ if (CE->getOpcode() == Instruction::Select) {
+ Constant *Src0 = CE->getOperand(1);
+ Constant *Src1 = CE->getOperand(2);
+ if (Src0->getType()->getPointerAddressSpace() ==
+ Src1->getType()->getPointerAddressSpace()) {
+
+ return ConstantExpr::getSelect(
+ CE->getOperand(0), ConstantExpr::getAddrSpaceCast(Src0, TargetType),
+ ConstantExpr::getAddrSpaceCast(Src1, TargetType));
+ }
+ }
+
// Computes the operands of the new constant expression.
SmallVector<Constant *, 4> NewOperands;
for (unsigned Index = 0; Index < CE->getNumOperands(); ++Index) {
@@ -589,15 +607,47 @@ Optional<unsigned> InferAddressSpaces::u
// The new inferred address space equals the join of the address spaces
// of all its pointer operands.
unsigned NewAS = UninitializedAddressSpace;
- for (Value *PtrOperand : getPointerOperands(V)) {
- auto I = InferredAddrSpace.find(PtrOperand);
- unsigned OperandAS = I != InferredAddrSpace.end() ?
- I->second : PtrOperand->getType()->getPointerAddressSpace();
-
- // join(flat, *) = flat. So we can break if NewAS is already flat.
- NewAS = joinAddressSpaces(NewAS, OperandAS);
- if (NewAS == FlatAddrSpace)
- break;
+
+ const Operator &Op = cast<Operator>(V);
+ if (Op.getOpcode() == Instruction::Select) {
+ Value *Src0 = Op.getOperand(1);
+ Value *Src1 = Op.getOperand(2);
+
+ auto I = InferredAddrSpace.find(Src0);
+ unsigned Src0AS = (I != InferredAddrSpace.end()) ?
+ I->second : Src0->getType()->getPointerAddressSpace();
+
+ auto J = InferredAddrSpace.find(Src1);
+ unsigned Src1AS = (J != InferredAddrSpace.end()) ?
+ J->second : Src1->getType()->getPointerAddressSpace();
+
+ auto *C0 = dyn_cast<Constant>(Src0);
+ auto *C1 = dyn_cast<Constant>(Src1);
+
+ // If one of the inputs is a constant, we may be able to do a constant
+ // addrspacecast of it. Defer inferring the address space until the input
+ // address space is known.
+ if ((C1 && Src0AS == UninitializedAddressSpace) ||
+ (C0 && Src1AS == UninitializedAddressSpace))
+ return None;
+
+ if (C0 && isSafeToCastConstAddrSpace(C0, Src1AS))
+ NewAS = Src1AS;
+ else if (C1 && isSafeToCastConstAddrSpace(C1, Src0AS))
+ NewAS = Src0AS;
+ else
+ NewAS = joinAddressSpaces(Src0AS, Src1AS);
+ } else {
+ for (Value *PtrOperand : getPointerOperands(V)) {
+ auto I = InferredAddrSpace.find(PtrOperand);
+ unsigned OperandAS = I != InferredAddrSpace.end() ?
+ I->second : PtrOperand->getType()->getPointerAddressSpace();
+
+ // join(flat, *) = flat. So we can break if NewAS is already flat.
+ NewAS = joinAddressSpaces(NewAS, OperandAS);
+ if (NewAS == FlatAddrSpace)
+ break;
+ }
}
unsigned OldAS = InferredAddrSpace.lookup(&V);
@@ -680,6 +730,8 @@ static bool handleMemIntrinsicPtrUse(Mem
// \p returns true if it is OK to change the address space of constant \p C with
// a ConstantExpr addrspacecast.
bool InferAddressSpaces::isSafeToCastConstAddrSpace(Constant *C, unsigned NewAS) const {
+ assert(NewAS != UninitializedAddressSpace);
+
unsigned SrcAS = C->getType()->getPointerAddressSpace();
if (SrcAS == NewAS || isa<UndefValue>(C))
return true;
Modified: llvm/trunk/test/Transforms/InferAddressSpaces/AMDGPU/select.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InferAddressSpaces/AMDGPU/select.ll?rev=293859&r1=293858&r2=293859&view=diff
==============================================================================
--- llvm/trunk/test/Transforms/InferAddressSpaces/AMDGPU/select.ll (original)
+++ llvm/trunk/test/Transforms/InferAddressSpaces/AMDGPU/select.ll Wed Feb 1 21:37:22 2017
@@ -55,18 +55,24 @@ define void @store_select_mismatch_group
@lds1 = internal addrspace(3) global i32 456, align 4
; CHECK-LABEL: @constexpr_select_group_flat(
-; CHCK: %tmp = load i32, i32 addrspace(3)* select (i1 icmp eq (i32 ptrtoint (i32 addrspace(3)* @lds1 to i32), i32 4), i32 addrspace(3)* @lds0, i32 addrspace(3)* @lds1)
+; CHECK: %tmp = load i32, i32 addrspace(3)* select (i1 icmp eq (i32 ptrtoint (i32 addrspace(3)* @lds1 to i32), i32 4), i32 addrspace(3)* @lds0, i32 addrspace(3)* @lds1)
define i32 @constexpr_select_group_flat() #0 {
bb:
%tmp = load i32, i32 addrspace(4)* select (i1 icmp eq (i32 ptrtoint (i32 addrspace(3)* @lds1 to i32), i32 4), i32 addrspace(4)* addrspacecast (i32 addrspace(3)* @lds0 to i32 addrspace(4)*), i32 addrspace(4)* addrspacecast (i32 addrspace(3)* @lds1 to i32 addrspace(4)*))
ret i32 %tmp
}
-; FIXME: Should be able to cast the constants
+; CHECK-LABEL: @constexpr_select_group_global_flat_mismatch(
+; CHECK: %tmp = load i32, i32 addrspace(4)* select (i1 icmp eq (i32 ptrtoint (i32 addrspace(3)* @lds1 to i32), i32 4), i32 addrspace(4)* addrspacecast (i32 addrspace(3)* @lds0 to i32 addrspace(4)*), i32 addrspace(4)* addrspacecast (i32 addrspace(1)* @global0 to i32 addrspace(4)*))
+define i32 @constexpr_select_group_global_flat_mismatch() #0 {
+bb:
+ %tmp = load i32, i32 addrspace(4)* select (i1 icmp eq (i32 ptrtoint (i32 addrspace(3)* @lds1 to i32), i32 4), i32 addrspace(4)* addrspacecast (i32 addrspace(3)* @lds0 to i32 addrspace(4)*), i32 addrspace(4)* addrspacecast (i32 addrspace(1)* @global0 to i32 addrspace(4)*))
+ ret i32 %tmp
+}
+
; CHECK-LABEL: @store_select_group_flat_null(
-; CHECK: %1 = addrspacecast i32 addrspace(3)* %group.ptr.0 to i32 addrspace(4)*
-; CHECK: %select = select i1 %c, i32 addrspace(4)* %1, i32 addrspace(4)* null
-; CHECK: store i32 -1, i32 addrspace(4)* %select
+; CHECK: %select = select i1 %c, i32 addrspace(3)* %group.ptr.0, i32 addrspace(3)* addrspacecast (i32 addrspace(4)* null to i32 addrspace(3)*)
+; CHECK: store i32 -1, i32 addrspace(3)* %select
define void @store_select_group_flat_null(i1 %c, i32 addrspace(3)* %group.ptr.0) #0 {
%cast0 = addrspacecast i32 addrspace(3)* %group.ptr.0 to i32 addrspace(4)*
%select = select i1 %c, i32 addrspace(4)* %cast0, i32 addrspace(4)* null
@@ -75,9 +81,8 @@ define void @store_select_group_flat_nul
}
; CHECK-LABEL: @store_select_group_flat_null_swap(
-; CHECK: %1 = addrspacecast i32 addrspace(3)* %group.ptr.0 to i32 addrspace(4)*
-; CHECK: %select = select i1 %c, i32 addrspace(4)* null, i32 addrspace(4)* %1
-; CHECK: store i32 -1, i32 addrspace(4)* %select
+; CHECK: %select = select i1 %c, i32 addrspace(3)* addrspacecast (i32 addrspace(4)* null to i32 addrspace(3)*), i32 addrspace(3)* %group.ptr.0
+; CHECK: store i32 -1, i32 addrspace(3)* %select
define void @store_select_group_flat_null_swap(i1 %c, i32 addrspace(3)* %group.ptr.0) #0 {
%cast0 = addrspacecast i32 addrspace(3)* %group.ptr.0 to i32 addrspace(4)*
%select = select i1 %c, i32 addrspace(4)* null, i32 addrspace(4)* %cast0
@@ -85,11 +90,9 @@ define void @store_select_group_flat_nul
ret void
}
-
; CHECK-LABEL: @store_select_group_flat_undef(
-; CHECK: %1 = addrspacecast i32 addrspace(3)* %group.ptr.0 to i32 addrspace(4)*
-; CHECK: %select = select i1 %c, i32 addrspace(4)* %1, i32 addrspace(4)* undef
-; CHECK: store i32 -1, i32 addrspace(4)* %select
+; CHECK: %select = select i1 %c, i32 addrspace(3)* %group.ptr.0, i32 addrspace(3)* undef
+; CHECK: store i32 -1, i32 addrspace(3)* %select
define void @store_select_group_flat_undef(i1 %c, i32 addrspace(3)* %group.ptr.0) #0 {
%cast0 = addrspacecast i32 addrspace(3)* %group.ptr.0 to i32 addrspace(4)*
%select = select i1 %c, i32 addrspace(4)* %cast0, i32 addrspace(4)* undef
@@ -98,9 +101,8 @@ define void @store_select_group_flat_und
}
; CHECK-LABEL: @store_select_group_flat_undef_swap(
-; CHECK: %1 = addrspacecast i32 addrspace(3)* %group.ptr.0 to i32 addrspace(4)*
-; CHECK: %select = select i1 %c, i32 addrspace(4)* undef, i32 addrspace(4)* %1
-; CHECK: store i32 -1, i32 addrspace(4)* %select
+; CHECK: %select = select i1 %c, i32 addrspace(3)* undef, i32 addrspace(3)* %group.ptr.0
+; CHECK: store i32 -1, i32 addrspace(3)* %select
define void @store_select_group_flat_undef_swap(i1 %c, i32 addrspace(3)* %group.ptr.0) #0 {
%cast0 = addrspacecast i32 addrspace(3)* %group.ptr.0 to i32 addrspace(4)*
%select = select i1 %c, i32 addrspace(4)* undef, i32 addrspace(4)* %cast0
@@ -108,6 +110,18 @@ define void @store_select_group_flat_und
ret void
}
+; CHECK-LABEL: @store_select_gep_group_flat_null(
+; CHECK: %select = select i1 %c, i32 addrspace(3)* %group.ptr.0, i32 addrspace(3)* addrspacecast (i32 addrspace(4)* null to i32 addrspace(3)*)
+; CHECK: %gep = getelementptr i32, i32 addrspace(3)* %select, i64 16
+; CHECK: store i32 -1, i32 addrspace(3)* %gep
+define void @store_select_gep_group_flat_null(i1 %c, i32 addrspace(3)* %group.ptr.0) #0 {
+ %cast0 = addrspacecast i32 addrspace(3)* %group.ptr.0 to i32 addrspace(4)*
+ %select = select i1 %c, i32 addrspace(4)* %cast0, i32 addrspace(4)* null
+ %gep = getelementptr i32, i32 addrspace(4)* %select, i64 16
+ store i32 -1, i32 addrspace(4)* %gep
+ ret void
+}
+
@global0 = internal addrspace(1) global i32 123, align 4
; CHECK-LABEL: @store_select_group_flat_constexpr(
@@ -120,6 +134,26 @@ define void @store_select_group_flat_con
ret void
}
+; CHECK-LABEL: @store_select_group_flat_inttoptr_flat(
+; CHECK: %select = select i1 %c, i32 addrspace(3)* %group.ptr.0, i32 addrspace(3)* addrspacecast (i32 addrspace(4)* inttoptr (i64 12345 to i32 addrspace(4)*) to i32 addrspace(3)*)
+; CHECK: store i32 7, i32 addrspace(3)* %select
+define void @store_select_group_flat_inttoptr_flat(i1 %c, i32 addrspace(3)* %group.ptr.0) #0 {
+ %cast0 = addrspacecast i32 addrspace(3)* %group.ptr.0 to i32 addrspace(4)*
+ %select = select i1 %c, i32 addrspace(4)* %cast0, i32 addrspace(4)* inttoptr (i64 12345 to i32 addrspace(4)*)
+ store i32 7, i32 addrspace(4)* %select
+ ret void
+}
+
+; CHECK-LABEL: @store_select_group_flat_inttoptr_group(
+; CHECK: %select = select i1 %c, i32 addrspace(3)* %group.ptr.0, i32 addrspace(3)* inttoptr (i32 400 to i32 addrspace(3)*)
+; CHECK-NEXT: store i32 7, i32 addrspace(3)* %select
+define void @store_select_group_flat_inttoptr_group(i1 %c, i32 addrspace(3)* %group.ptr.0) #0 {
+ %cast0 = addrspacecast i32 addrspace(3)* %group.ptr.0 to i32 addrspace(4)*
+ %select = select i1 %c, i32 addrspace(4)* %cast0, i32 addrspace(4)* addrspacecast (i32 addrspace(3)* inttoptr (i32 400 to i32 addrspace(3)*) to i32 addrspace(4)*)
+ store i32 7, i32 addrspace(4)* %select
+ ret void
+}
+
; CHECK-LABEL: @store_select_group_global_mismatch_flat_constexpr(
; CHECK: %1 = addrspacecast i32 addrspace(3)* %group.ptr.0 to i32 addrspace(4)*
; CHECK: %select = select i1 %c, i32 addrspace(4)* %1, i32 addrspace(4)* addrspacecast (i32 addrspace(1)* @global0 to i32 addrspace(4)*)
@@ -142,6 +176,70 @@ define void @store_select_group_global_m
ret void
}
+; CHECK-LABEL: @store_select_group_global_mismatch_null_null(
+; CHECK: %select = select i1 %c, i32 addrspace(4)* addrspacecast (i32 addrspace(3)* null to i32 addrspace(4)*), i32 addrspace(4)* addrspacecast (i32 addrspace(1)* null to i32 addrspace(4)*)
+; CHECK: store i32 7, i32 addrspace(4)* %select
+define void @store_select_group_global_mismatch_null_null(i1 %c) #0 {
+ %select = select i1 %c, i32 addrspace(4)* addrspacecast (i32 addrspace(3)* null to i32 addrspace(4)*), i32 addrspace(4)* addrspacecast (i32 addrspace(1)* null to i32 addrspace(4)*)
+ store i32 7, i32 addrspace(4)* %select
+ ret void
+}
+
+; CHECK-LABEL: @store_select_group_global_mismatch_null_null_constexpr(
+; CHECK: store i32 7, i32 addrspace(4)* select (i1 icmp eq (i32 ptrtoint (i32 addrspace(3)* @lds1 to i32), i32 4), i32 addrspace(4)* addrspacecast (i32 addrspace(3)* null to i32 addrspace(4)*), i32 addrspace(4)* addrspacecast (i32 addrspace(1)* null to i32 addrspace(4)*)), align 4
+define void @store_select_group_global_mismatch_null_null_constexpr() #0 {
+ store i32 7, i32 addrspace(4)* select (i1 icmp eq (i32 ptrtoint (i32 addrspace(3)* @lds1 to i32), i32 4), i32 addrspace(4)* addrspacecast (i32 addrspace(3)* null to i32 addrspace(4)*), i32 addrspace(4)* addrspacecast (i32 addrspace(1)* null to i32 addrspace(4)*)), align 4
+ ret void
+}
+
+; CHECK-LABEL: @store_select_group_global_mismatch_gv_null_constexpr(
+; CHECK: store i32 7, i32 addrspace(4)* select (i1 icmp eq (i32 ptrtoint (i32 addrspace(3)* @lds1 to i32), i32 4), i32 addrspace(4)* addrspacecast (i32 addrspace(3)* @lds0 to i32 addrspace(4)*), i32 addrspace(4)* addrspacecast (i32 addrspace(1)* null to i32 addrspace(4)*)), align 4
+define void @store_select_group_global_mismatch_gv_null_constexpr() #0 {
+ store i32 7, i32 addrspace(4)* select (i1 icmp eq (i32 ptrtoint (i32 addrspace(3)* @lds1 to i32), i32 4), i32 addrspace(4)* addrspacecast (i32 addrspace(3)* @lds0 to i32 addrspace(4)*), i32 addrspace(4)* addrspacecast (i32 addrspace(1)* null to i32 addrspace(4)*)), align 4
+ ret void
+}
+
+; CHECK-LABEL: @store_select_group_global_mismatch_null_gv_constexpr(
+; CHECK: store i32 7, i32 addrspace(4)* select (i1 icmp eq (i32 ptrtoint (i32 addrspace(3)* @lds1 to i32), i32 4), i32 addrspace(4)* addrspacecast (i32 addrspace(3)* null to i32 addrspace(4)*), i32 addrspace(4)* addrspacecast (i32 addrspace(1)* @global0 to i32 addrspace(4)*)), align 4
+define void @store_select_group_global_mismatch_null_gv_constexpr() #0 {
+ store i32 7, i32 addrspace(4)* select (i1 icmp eq (i32 ptrtoint (i32 addrspace(3)* @lds1 to i32), i32 4), i32 addrspace(4)* addrspacecast (i32 addrspace(3)* null to i32 addrspace(4)*), i32 addrspace(4)* addrspacecast (i32 addrspace(1)* @global0 to i32 addrspace(4)*)), align 4
+ ret void
+}
+
+; CHECK-LABEL: @store_select_group_global_mismatch_inttoptr_null_constexpr(
+; CHECK: store i32 7, i32 addrspace(4)* select (i1 icmp eq (i32 ptrtoint (i32 addrspace(3)* @lds1 to i32), i32 4), i32 addrspace(4)* addrspacecast (i32 addrspace(3)* inttoptr (i64 123 to i32 addrspace(3)*) to i32 addrspace(4)*), i32 addrspace(4)* addrspacecast (i32 addrspace(1)* null to i32 addrspace(4)*)), align 4
+define void @store_select_group_global_mismatch_inttoptr_null_constexpr() #0 {
+ store i32 7, i32 addrspace(4)* select (i1 icmp eq (i32 ptrtoint (i32 addrspace(3)* @lds1 to i32), i32 4), i32 addrspace(4)* addrspacecast (i32 addrspace(3)* inttoptr (i64 123 to i32 addrspace(3)*) to i32 addrspace(4)*), i32 addrspace(4)* addrspacecast (i32 addrspace(1)* null to i32 addrspace(4)*)), align 4
+ ret void
+}
+
+; CHECK-LABEL: @store_select_group_global_mismatch_inttoptr_flat_null_constexpr(
+; CHECK: store i32 7, i32 addrspace(1)* select (i1 icmp eq (i32 ptrtoint (i32 addrspace(3)* @lds1 to i32), i32 4), i32 addrspace(1)* addrspacecast (i32 addrspace(4)* inttoptr (i64 123 to i32 addrspace(4)*) to i32 addrspace(1)*), i32 addrspace(1)* null), align 4
+define void @store_select_group_global_mismatch_inttoptr_flat_null_constexpr() #0 {
+ store i32 7, i32 addrspace(4)* select (i1 icmp eq (i32 ptrtoint (i32 addrspace(3)* @lds1 to i32), i32 4), i32 addrspace(4)* inttoptr (i64 123 to i32 addrspace(4)*), i32 addrspace(4)* addrspacecast (i32 addrspace(1)* null to i32 addrspace(4)*)), align 4
+ ret void
+}
+
+; CHECK-LABEL: @store_select_group_global_mismatch_undef_undef_constexpr(
+; CHECK: store i32 7, i32 addrspace(3)* null
+define void @store_select_group_global_mismatch_undef_undef_constexpr() #0 {
+ store i32 7, i32 addrspace(4)* select (i1 icmp eq (i32 ptrtoint (i32 addrspace(3)* @lds1 to i32), i32 4), i32 addrspace(4)* addrspacecast (i32 addrspace(3)* null to i32 addrspace(4)*), i32 addrspace(4)* addrspacecast (i32 addrspace(1)* undef to i32 addrspace(4)*)), align 4
+ ret void
+}
+
+ at lds2 = external addrspace(3) global [1024 x i32], align 4
+
+; CHECK-LABEL: @store_select_group_constexpr_ptrtoint(
+; CHECK: %1 = addrspacecast i32 addrspace(3)* %group.ptr.0 to i32 addrspace(4)*
+; CHECK: %select = select i1 %c, i32 addrspace(4)* %1, i32 addrspace(4)* addrspacecast (i32 addrspace(1)* inttoptr (i32 add (i32 ptrtoint ([1024 x i32] addrspace(3)* @lds2 to i32), i32 124) to i32 addrspace(1)*) to i32 addrspace(4)*)
+; CHECK: store i32 7, i32 addrspace(4)* %select
+define void @store_select_group_constexpr_ptrtoint(i1 %c, i32 addrspace(3)* %group.ptr.0) #0 {
+ %cast0 = addrspacecast i32 addrspace(3)* %group.ptr.0 to i32 addrspace(4)*
+ %select = select i1 %c, i32 addrspace(4)* %cast0, i32 addrspace(4)* addrspacecast (i32 addrspace(1)* inttoptr (i32 add (i32 ptrtoint ([1024 x i32] addrspace(3)* @lds2 to i32), i32 124) to i32 addrspace(1)*) to i32 addrspace(4)*)
+ store i32 7, i32 addrspace(4)* %select
+ ret void
+}
+
; CHECK-LABEL: @store_select_group_flat_vector(
; CHECK: %cast0 = addrspacecast <2 x i32 addrspace(3)*> %group.ptr.0 to <2 x i32 addrspace(4)*>
; CHECK: %cast1 = addrspacecast <2 x i32 addrspace(3)*> %group.ptr.1 to <2 x i32 addrspace(4)*>
@@ -160,6 +258,7 @@ define void @store_select_group_flat_vec
store i32 -2, i32 addrspace(4)* %extract1
ret void
}
+
attributes #0 = { nounwind }
!0 = !{!"branch_weights", i32 2, i32 10}
More information about the llvm-commits
mailing list