RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-25 12:12:47 +01:00
Code Issues Projects Releases Wiki Activity

[GVN] teach ConstantFolding correct handling of non-integral addrspace casts

Browse Source 
Here we teach the ConstantFolding analysis pass that it is not legal to
replace a load of a bitcast constant (having a non-integral addrspace)
with a bitcast of the value of that constant (with a different
non-integral addrspace).

But also teach it that certain bit patterns are always known and
convertable (a fact it already uses elsewhere). This required us to also
fix a globalopt test, since, after this change, LLVM is able to realize
that the test actually is a valid transform (NULL is always a known
bit-pattern) and so it doesn't need to emit the failure remarks for it.

Also simplify some of the negative tests for transforms by avoiding a
type change in their bitcast, and add positive versions of the same
tests, to show that they otherwise should work.

Differential Revision: https://reviews.llvm.org/D59730
This commit is contained in:
Jameson Nash 2019-11-06 12:01:57 -05:00
parent 3f3cb11075
commit 202eae7f23
3 changed files with 44 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
lib/Analysis/ConstantFolding.cpp
View File

@ -338,10 +338,20 @@ Constant *llvm::ConstantFoldLoadThroughBitcast(Constant *C, Type *DestTy,
if (SrcSize < DestSize)
return nullptr;
// Catch the obvious splat cases (since all-zeros can coerce non-integral
// pointers legally).
if (C->isNullValue() && !DestTy->isX86_MMXTy())
return Constant::getNullValue(DestTy);
if (C->isAllOnesValue() && !DestTy->isX86_MMXTy() &&
!DestTy->isPtrOrPtrVectorTy()) // Don't get ones for ptr types!
return Constant::getAllOnesValue(DestTy);
// If the type sizes are the same and a cast is legal, just directly
// cast the constant.
// But be careful not to coerce non-integral pointers illegally.
if (SrcSize == DestSize) {
if (SrcSize == DestSize &&
DL.isNonIntegralPointerType(SrcTy->getScalarType()) ==
DL.isNonIntegralPointerType(DestTy->getScalarType())) {
Instruction::CastOps Cast = Instruction::BitCast;
// If we are going from a pointer to int or vice versa, we spell the cast
// differently.

6
lib/Transforms/Utils/VNCoercion.cpp
View File

@ -403,12 +403,6 @@ int analyzeLoadFromClobberingMemInst(Type *LoadTy, Value *LoadPtr,
if (Offset == -1)
return Offset;
// Don't coerce non-integral pointers to integers or vice versa, and the
// memtransfer is implicitly a raw byte code
if (DL.isNonIntegralPointerType(LoadTy->getScalarType()))
// TODO: Can allow nullptrs from constant zeros
return -1;
unsigned AS = Src->getType()->getPointerAddressSpace();
// Otherwise, see if we can constant fold a load from the constant with the
// offset applied as appropriate.

36
test/Transforms/GVN/non-integral-pointers.ll
View File

@ -230,6 +230,21 @@ entry:
ret i8 addrspace(4)* %ref
}
define i64 addrspace(4)* @forward_memcopy2(i64 addrspace(4)* addrspace(4)* %loc) {
; CHECK-LABEL: @forward_memcopy2(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[LOC_BC:%.*]] = bitcast i64 addrspace(4)* addrspace(4)* [[LOC:%.*]] to i8 addrspace(4)*
; CHECK-NEXT: call void @llvm.memcpy.p4i8.p0i8.i64(i8 addrspace(4)* align 4 [[LOC_BC]], i8* bitcast (<4 x i64 addrspace(4)*>* @NonZeroConstant2 to i8*), i64 8, i1 false)
; CHECK-NEXT: ret i64 addrspace(4)* getelementptr (i64, i64 addrspace(4)* null, i32 3)
;
entry:
%loc.bc = bitcast i64 addrspace(4)* addrspace(4)* %loc to i8 addrspace(4)*
%src.bc = bitcast <4 x i64 addrspace(4)*>* @NonZeroConstant2 to i8*
call void @llvm.memcpy.p4i8.p0i8.i64(i8 addrspace(4)* align 4 %loc.bc, i8* %src.bc, i64 8, i1 false)
%ref = load i64 addrspace(4)*, i64 addrspace(4)* addrspace(4)* %loc
ret i64 addrspace(4)* %ref
}
define <1 x i8 addrspace(4)*> @neg_forward_memcpy_vload(<1 x i8 addrspace(4)*> addrspace(4)* %loc) {
; CHECK-LABEL: @neg_forward_memcpy_vload(
; CHECK-NEXT: entry:
@ -278,15 +293,30 @@ entry:
ret <4 x i64> %ref
}
define <1 x i64 addrspace(4)*> @forward_memcpy_vload3(<4 x i64 addrspace(4)*> addrspace(4)* %loc) {
; CHECK-LABEL: @forward_memcpy_vload3(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[LOC_BC:%.*]] = bitcast <4 x i64 addrspace(4)*> addrspace(4)* [[LOC:%.*]] to i8 addrspace(4)*
; CHECK-NEXT: call void @llvm.memcpy.p4i8.p0i8.i64(i8 addrspace(4)* align 4 [[LOC_BC]], i8* bitcast (<4 x i64 addrspace(4)*>* @NonZeroConstant2 to i8*), i64 32, i1 false)
; CHECK-NEXT: ret <1 x i64 addrspace(4)*> <i64 addrspace(4)* getelementptr (i64, i64 addrspace(4)* null, i32 3)>
;
entry:
%loc.bc = bitcast <4 x i64 addrspace(4)*> addrspace(4)* %loc to i8 addrspace(4)*
%src.bc = bitcast <4 x i64 addrspace(4)*>* @NonZeroConstant2 to i8*
call void @llvm.memcpy.p4i8.p0i8.i64(i8 addrspace(4)* align 4 %loc.bc, i8* %src.bc, i64 32, i1 false)
%ref = load <4 x i64 addrspace(4)*>, <4 x i64 addrspace(4)*> addrspace(4)* %loc
%val = extractelement <4 x i64 addrspace(4)*> %ref, i32 0
%ret = insertelement <1 x i64 addrspace(4)*> undef, i64 addrspace(4)* %val, i32 0
ret <1 x i64 addrspace(4)*> %ret
}
; Can forward since we can do so w/o breaking types
; TODO: missed optimization
define i8 addrspace(4)* @forward_memcpy_zero(i8 addrspace(4)* addrspace(4)* %loc) {
; CHECK-LABEL: @forward_memcpy_zero(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[LOC_BC:%.*]] = bitcast i8 addrspace(4)* addrspace(4)* [[LOC:%.*]] to i8 addrspace(4)*
; CHECK-NEXT: call void @llvm.memcpy.p4i8.p0i8.i64(i8 addrspace(4)* align 4 [[LOC_BC]], i8* bitcast (<4 x i64>* @ZeroConstant to i8*), i64 8, i1 false)
; CHECK-NEXT: [[REF:%.*]] = load i8 addrspace(4)*, i8 addrspace(4)* addrspace(4)* [[LOC]], align 8
; CHECK-NEXT: ret i8 addrspace(4)* [[REF]]
; CHECK-NEXT: ret i8 addrspace(4)* null
;
entry:
%loc.bc = bitcast i8 addrspace(4)* addrspace(4)* %loc to i8 addrspace(4)*