[llvm] f29645e - [gvn] Handle a corner case w/vectors of non-integral pointers
Philip Reames via llvm-commits
llvm-commits at lists.llvm.org
Thu Oct 1 19:20:28 PDT 2020
Author: Philip Reames
Date: 2020-10-01T19:20:21-07:00
New Revision: f29645e7afdbb8d1fc2dd603c0b128bac055625c
URL: https://github.com/llvm/llvm-project/commit/f29645e7afdbb8d1fc2dd603c0b128bac055625c
DIFF: https://github.com/llvm/llvm-project/commit/f29645e7afdbb8d1fc2dd603c0b128bac055625c.diff
LOG: [gvn] Handle a corner case w/vectors of non-integral pointers
If we try to coerce a vector of non-integral pointers to a narrower type (either narrower vector or single pointer), we use inttoptr and violate the semantics of non-integral pointers. In theory, we can handle many of these cases, we just need to use a different code idiom to convert without going through inttoptr and back.
This shows up as wrong code bugs, and in some cases, crashes due to failed asserts. Modeled after a change which has lived downstream for a couple years, though completely rewritten to be more idiomatic.
Added:
Modified:
llvm/lib/Transforms/Utils/VNCoercion.cpp
llvm/test/Transforms/GVN/non-integral-pointers.ll
Removed:
################################################################################
diff --git a/llvm/lib/Transforms/Utils/VNCoercion.cpp b/llvm/lib/Transforms/Utils/VNCoercion.cpp
index 1939c0e3b504..11b42eca4960 100644
--- a/llvm/lib/Transforms/Utils/VNCoercion.cpp
+++ b/llvm/lib/Transforms/Utils/VNCoercion.cpp
@@ -17,6 +17,7 @@ static bool isFirstClassAggregateOrScalableType(Type *Ty) {
bool canCoerceMustAliasedValueToLoad(Value *StoredVal, Type *LoadTy,
const DataLayout &DL) {
Type *StoredTy = StoredVal->getType();
+
if (StoredTy == LoadTy)
return true;
@@ -46,6 +47,14 @@ bool canCoerceMustAliasedValueToLoad(Value *StoredVal, Type *LoadTy,
return CI->isNullValue();
return false;
}
+
+
+ // The implementation below uses inttoptr for vectors of unequal size; we
+ // can't allow this for non integral pointers. Wecould teach it to extract
+ // exact subvectors if desired.
+ if (DL.isNonIntegralPointerType(StoredTy->getScalarType()) &&
+ StoreSize != DL.getTypeSizeInBits(LoadTy).getFixedSize())
+ return false;
return true;
}
@@ -223,14 +232,8 @@ int analyzeLoadFromClobberingStore(Type *LoadTy, Value *LoadPtr,
if (isFirstClassAggregateOrScalableType(StoredVal->getType()))
return -1;
- // Don't coerce non-integral pointers to integers or vice versa.
- if (DL.isNonIntegralPointerType(StoredVal->getType()->getScalarType()) !=
- DL.isNonIntegralPointerType(LoadTy->getScalarType())) {
- // Allow casts of zero values to null as a special case
- auto *CI = dyn_cast<Constant>(StoredVal);
- if (!CI || !CI->isNullValue())
- return -1;
- }
+ if (!canCoerceMustAliasedValueToLoad(StoredVal, LoadTy, DL))
+ return -1;
Value *StorePtr = DepSI->getPointerOperand();
uint64_t StoreSize =
@@ -333,9 +336,7 @@ int analyzeLoadFromClobberingLoad(Type *LoadTy, Value *LoadPtr, LoadInst *DepLI,
if (DepLI->getType()->isStructTy() || DepLI->getType()->isArrayTy())
return -1;
- // Don't coerce non-integral pointers to integers or vice versa.
- if (DL.isNonIntegralPointerType(DepLI->getType()->getScalarType()) !=
- DL.isNonIntegralPointerType(LoadTy->getScalarType()))
+ if (!canCoerceMustAliasedValueToLoad(DepLI, LoadTy, DL))
return -1;
Value *DepPtr = DepLI->getPointerOperand();
diff --git a/llvm/test/Transforms/GVN/non-integral-pointers.ll b/llvm/test/Transforms/GVN/non-integral-pointers.ll
index a017dda926e3..872b6648084e 100644
--- a/llvm/test/Transforms/GVN/non-integral-pointers.ll
+++ b/llvm/test/Transforms/GVN/non-integral-pointers.ll
@@ -202,7 +202,7 @@ define i64 addrspace(4)* @neg_forward_memcopy2(i64 addrspace(4)* addrspace(4)* %
; 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>* @NonZeroConstant to i8*), i64 8, i1 false)
-; CHECK-NEXT: [[REF:%.*]] = load i64 addrspace(4)*, i64 addrspace(4)* addrspace(4)* [[LOC]]
+; CHECK-NEXT: [[REF:%.*]] = load i64 addrspace(4)*, i64 addrspace(4)* addrspace(4)* [[LOC]], align 8
; CHECK-NEXT: ret i64 addrspace(4)* [[REF]]
;
entry:
@@ -219,7 +219,7 @@ define i8 addrspace(4)* @forward_memcopy(i8 addrspace(4)* addrspace(4)* %loc) {
; 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 addrspace(4)*>* @NonZeroConstant2 to i8*), i64 8, i1 false)
-; CHECK-NEXT: [[REF:%.*]] = load i8 addrspace(4)*, i8 addrspace(4)* addrspace(4)* %loc
+; CHECK-NEXT: [[REF:%.*]] = load i8 addrspace(4)*, i8 addrspace(4)* addrspace(4)* [[LOC]], align 8
; CHECK-NEXT: ret i8 addrspace(4)* [[REF]]
;
entry:
@@ -266,7 +266,7 @@ define <4 x i64 addrspace(4)*> @neg_forward_memcpy_vload2(<4 x i64 addrspace(4)*
; 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>* @NonZeroConstant to i8*), i64 32, i1 false)
-; CHECK-NEXT: [[REF:%.*]] = load <4 x i64 addrspace(4)*>, <4 x i64 addrspace(4)*> addrspace(4)* [[LOC]]
+; CHECK-NEXT: [[REF:%.*]] = load <4 x i64 addrspace(4)*>, <4 x i64 addrspace(4)*> addrspace(4)* [[LOC]], align 32
; CHECK-NEXT: ret <4 x i64 addrspace(4)*> [[REF]]
;
entry:
@@ -282,7 +282,7 @@ define <4 x i64> @neg_forward_memcpy_vload3(<4 x i64> addrspace(4)* %loc) {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[LOC_BC:%.*]] = bitcast <4 x i64> 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: [[REF:%.*]] = load <4 x i64>, <4 x i64> addrspace(4)* [[LOC]]
+; CHECK-NEXT: [[REF:%.*]] = load <4 x i64>, <4 x i64> addrspace(4)* [[LOC]], align 32
; CHECK-NEXT: ret <4 x i64> [[REF]]
;
entry:
@@ -386,3 +386,47 @@ entry:
%ref = load i8 addrspace(4)*, i8 addrspace(4)* addrspace(4)* %loc.off
ret i8 addrspace(4)* %ref
}
+
+
+define void @smaller_vector(i8* %p) {
+; CHECK-LABEL: @smaller_vector(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[A:%.*]] = bitcast i8* [[P:%.*]] to <4 x i64 addrspace(4)*>*
+; CHECK-NEXT: [[B:%.*]] = bitcast i8* [[P]] to <2 x i64 addrspace(4)*>*
+; CHECK-NEXT: [[V4:%.*]] = load <4 x i64 addrspace(4)*>, <4 x i64 addrspace(4)*>* [[A]], align 32
+; CHECK-NEXT: [[V2:%.*]] = load <2 x i64 addrspace(4)*>, <2 x i64 addrspace(4)*>* [[B]], align 32
+; CHECK-NEXT: call void @use.v2(<2 x i64 addrspace(4)*> [[V2]])
+; CHECK-NEXT: call void @use.v4(<4 x i64 addrspace(4)*> [[V4]])
+; CHECK-NEXT: ret void
+;
+entry:
+ %a = bitcast i8* %p to <4 x i64 addrspace(4)*>*
+ %b = bitcast i8* %p to <2 x i64 addrspace(4)*>*
+ %v4 = load <4 x i64 addrspace(4)*>, <4 x i64 addrspace(4)*>* %a, align 32
+ %v2 = load <2 x i64 addrspace(4)*>, <2 x i64 addrspace(4)*>* %b, align 32
+ call void @use.v2(<2 x i64 addrspace(4)*> %v2)
+ call void @use.v4(<4 x i64 addrspace(4)*> %v4)
+ ret void
+}
+
+define i64 addrspace(4)* @vector_extract(i8* %p) {
+; CHECK-LABEL: @vector_extract(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[A:%.*]] = bitcast i8* [[P:%.*]] to <4 x i64 addrspace(4)*>*
+; CHECK-NEXT: [[B:%.*]] = bitcast i8* [[P]] to i64 addrspace(4)**
+; CHECK-NEXT: [[V4:%.*]] = load <4 x i64 addrspace(4)*>, <4 x i64 addrspace(4)*>* [[A]], align 32
+; CHECK-NEXT: [[RES:%.*]] = load i64 addrspace(4)*, i64 addrspace(4)** [[B]], align 32
+; CHECK-NEXT: call void @use.v4(<4 x i64 addrspace(4)*> [[V4]])
+; CHECK-NEXT: ret i64 addrspace(4)* [[RES]]
+;
+entry:
+ %a = bitcast i8* %p to <4 x i64 addrspace(4)*>*
+ %b = bitcast i8* %p to i64 addrspace(4)**
+ %v4 = load <4 x i64 addrspace(4)*>, <4 x i64 addrspace(4)*>* %a, align 32
+ %res = load i64 addrspace(4)*, i64 addrspace(4)** %b, align 32
+ call void @use.v4(<4 x i64 addrspace(4)*> %v4)
+ ret i64 addrspace(4)* %res
+}
+
+declare void @use.v2(<2 x i64 addrspace(4)*>)
+declare void @use.v4(<4 x i64 addrspace(4)*>)
More information about the llvm-commits
mailing list