[compiler-rt] [llvm] [nsan] Emit calls to optimized functions (PR #98900)
via llvm-commits
llvm-commits at lists.llvm.org
Mon Jul 15 06:39:40 PDT 2024
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-compiler-rt-sanitizer
Author: Dmitriy Chestnykh (chestnykh)
<details>
<summary>Changes</summary>
As previously noted in nsan.cpp we can implement
optimized variants of `__nsan_copy_values` and
`__nsan_set_value_unknown` if a memory operation
size is known.
Now the instrumentation creates calls to optimized functions if there is 4, 8 or 16-byte memory operation like
`memset(X, value, 4/8/16)` or `memcpy(dst, src, 4/8/16)`
nsan.cpp provides definition of the optimized functions.
---
Full diff: https://github.com/llvm/llvm-project/pull/98900.diff
3 Files Affected:
- (modified) compiler-rt/lib/nsan/nsan.cpp (+22-4)
- (modified) llvm/lib/Transforms/Instrumentation/NumericalStabilitySanitizer.cpp (+71-16)
- (modified) llvm/test/Instrumentation/NumericalStabilitySanitizer/memory.ll (+86-3)
``````````diff
diff --git a/compiler-rt/lib/nsan/nsan.cpp b/compiler-rt/lib/nsan/nsan.cpp
index e5a9b7a036e0e..e60dc4f5c3a70 100644
--- a/compiler-rt/lib/nsan/nsan.cpp
+++ b/compiler-rt/lib/nsan/nsan.cpp
@@ -54,8 +54,6 @@ using namespace __nsan;
constexpr int kMaxVectorWidth = 8;
// When copying application memory, we also copy its shadow and shadow type.
-// FIXME: We could provide fixed-size versions that would nicely
-// vectorize for known sizes.
extern "C" SANITIZER_INTERFACE_ATTRIBUTE void
__nsan_copy_values(const u8 *daddr, const u8 *saddr, uptr size) {
internal_memmove((void *)GetShadowTypeAddrFor(daddr),
@@ -64,13 +62,33 @@ __nsan_copy_values(const u8 *daddr, const u8 *saddr, uptr size) {
size * kShadowScale);
}
-// FIXME: We could provide fixed-size versions that would nicely
-// vectorize for known sizes.
+#define NSAN_COPY_VALUES_N(N) \
+ extern "C" SANITIZER_INTERFACE_ATTRIBUTE void __nsan_copy_##N( \
+ const u8 *daddr, const u8 *saddr) { \
+ __builtin_memmove((void *)GetShadowTypeAddrFor(daddr), \
+ GetShadowTypeAddrFor(saddr), N); \
+ __builtin_memmove((void *)GetShadowAddrFor(daddr), \
+ GetShadowAddrFor(saddr), N *kShadowScale); \
+ }
+
+NSAN_COPY_VALUES_N(4);
+NSAN_COPY_VALUES_N(8);
+NSAN_COPY_VALUES_N(16);
+
extern "C" SANITIZER_INTERFACE_ATTRIBUTE void
__nsan_set_value_unknown(const u8 *addr, uptr size) {
internal_memset((void *)GetShadowTypeAddrFor(addr), 0, size);
}
+#define NSAN_SET_VALUE_UNKNOWN_N(N) \
+ extern "C" SANITIZER_INTERFACE_ATTRIBUTE void __nsan_set_value_unknown_##N( \
+ const u8 *daddr) { \
+ __builtin_memset((void *)GetShadowTypeAddrFor(daddr), 0, N); \
+ }
+
+NSAN_SET_VALUE_UNKNOWN_N(4);
+NSAN_SET_VALUE_UNKNOWN_N(8);
+NSAN_SET_VALUE_UNKNOWN_N(16);
const char *FTInfo<float>::kCppTypeName = "float";
const char *FTInfo<double>::kCppTypeName = "double";
diff --git a/llvm/lib/Transforms/Instrumentation/NumericalStabilitySanitizer.cpp b/llvm/lib/Transforms/Instrumentation/NumericalStabilitySanitizer.cpp
index 99b1c779f3167..607b4384b5136 100644
--- a/llvm/lib/Transforms/Instrumentation/NumericalStabilitySanitizer.cpp
+++ b/llvm/lib/Transforms/Instrumentation/NumericalStabilitySanitizer.cpp
@@ -550,12 +550,15 @@ class NumericalStabilitySanitizer {
LLVMContext &Context;
MappingConfig Config;
IntegerType *IntptrTy = nullptr;
+
+ // TODO: Use std::array instead?
FunctionCallee NsanGetShadowPtrForStore[FTValueType::kNumValueTypes] = {};
FunctionCallee NsanGetShadowPtrForLoad[FTValueType::kNumValueTypes] = {};
FunctionCallee NsanCheckValue[FTValueType::kNumValueTypes] = {};
FunctionCallee NsanFCmpFail[FTValueType::kNumValueTypes] = {};
- FunctionCallee NsanCopyValues;
- FunctionCallee NsanSetValueUnknown;
+
+ std::array<FunctionCallee, 4> NsanCopyFunction;
+ std::array<FunctionCallee, 4> NsanSetValueUnknownFunction;
FunctionCallee NsanGetRawShadowTypePtr;
FunctionCallee NsanGetRawShadowPtr;
GlobalValue *NsanShadowRetTag = nullptr;
@@ -634,10 +637,27 @@ NumericalStabilitySanitizer::NumericalStabilitySanitizer(Module &M)
Attr, VoidTy, VTTy, VTTy, ShadowTy, ShadowTy, Int32Ty, Int1Ty, Int1Ty);
}
- NsanCopyValues = M.getOrInsertFunction("__nsan_copy_values", Attr, VoidTy,
- PtrTy, PtrTy, IntptrTy);
- NsanSetValueUnknown = M.getOrInsertFunction("__nsan_set_value_unknown", Attr,
- VoidTy, PtrTy, IntptrTy);
+ NsanCopyFunction[0] =
+ M.getOrInsertFunction("__nsan_copy_4", Attr, VoidTy, PtrTy, PtrTy);
+ NsanCopyFunction[1] =
+ M.getOrInsertFunction("__nsan_copy_8", Attr, VoidTy, PtrTy, PtrTy);
+ NsanCopyFunction[2] =
+ M.getOrInsertFunction("__nsan_copy_16", Attr, VoidTy, PtrTy, PtrTy);
+
+ NsanCopyFunction[3] = M.getOrInsertFunction("__nsan_copy_values", Attr,
+ VoidTy, PtrTy, PtrTy, IntptrTy);
+
+ NsanSetValueUnknownFunction[0] =
+ M.getOrInsertFunction("__nsan_set_value_unknown_4", Attr, VoidTy, PtrTy);
+
+ NsanSetValueUnknownFunction[1] =
+ M.getOrInsertFunction("__nsan_set_value_unknown_8", Attr, VoidTy, PtrTy);
+
+ NsanSetValueUnknownFunction[2] =
+ M.getOrInsertFunction("__nsan_set_value_unknown_16", Attr, VoidTy, PtrTy);
+
+ NsanSetValueUnknownFunction[3] = M.getOrInsertFunction(
+ "__nsan_set_value_unknown", Attr, VoidTy, PtrTy, IntptrTy);
// TODO: Add attributes nofree, nosync, readnone, readonly,
NsanGetRawShadowTypePtr = M.getOrInsertFunction(
@@ -1880,7 +1900,7 @@ void NumericalStabilitySanitizer::propagateNonFTStore(
}
}
// All other stores just reset the shadow value to unknown.
- Builder.CreateCall(NsanSetValueUnknown, {Dst, ValueSize});
+ Builder.CreateCall(NsanSetValueUnknownFunction.back(), {Dst, ValueSize});
}
void NumericalStabilitySanitizer::propagateShadowValues(
@@ -2123,21 +2143,56 @@ bool NumericalStabilitySanitizer::sanitizeFunction(
return !ValueToShadow.empty();
}
+static size_t GetInstrumentationCalleeIdxForMemOp(Value *V,
+ std::size_t MaxIdx) {
+ uint64_t OpSize = 0;
+ if (Constant *C = dyn_cast<Constant>(V)) {
+ auto *CInt = dyn_cast<ConstantInt>(C);
+ if (CInt && CInt->getValue().getBitWidth() <= 64)
+ OpSize = CInt->getValue().getZExtValue();
+ }
+
+ size_t CandidateIdx =
+ OpSize == 4 ? 0 : (OpSize == 8 ? 1 : (OpSize == 16 ? 2 : MaxIdx));
+
+ return CandidateIdx <= MaxIdx ? CandidateIdx : MaxIdx;
+}
+
// Instrument the memory intrinsics so that they properly modify the shadow
// memory.
bool NumericalStabilitySanitizer::instrumentMemIntrinsic(MemIntrinsic *MI) {
IRBuilder<> Builder(MI);
if (auto *M = dyn_cast<MemSetInst>(MI)) {
- Builder.CreateCall(
- NsanSetValueUnknown,
- {/*Address=*/M->getArgOperand(0),
- /*Size=*/Builder.CreateIntCast(M->getArgOperand(2), IntptrTy, false)});
+ Value *OpSizeValue = M->getArgOperand(2);
+ size_t NrSetValueUnknownFns = NsanSetValueUnknownFunction.size();
+ size_t NsanSetValueUnknownFunctionIdx = GetInstrumentationCalleeIdxForMemOp(
+ OpSizeValue, NrSetValueUnknownFns - 1);
+ if (NsanSetValueUnknownFunctionIdx != NrSetValueUnknownFns - 1)
+ Builder.CreateCall(
+ NsanSetValueUnknownFunction[NsanSetValueUnknownFunctionIdx],
+ {/*Address=*/M->getArgOperand(0)});
+ else
+ Builder.CreateCall(NsanSetValueUnknownFunction.back(),
+ {/*Address=*/M->getArgOperand(0),
+ /*Size=*/Builder.CreateIntCast(M->getArgOperand(2),
+ IntptrTy, false)});
+
} else if (auto *M = dyn_cast<MemTransferInst>(MI)) {
- Builder.CreateCall(
- NsanCopyValues,
- {/*Destination=*/M->getArgOperand(0),
- /*Source=*/M->getArgOperand(1),
- /*Size=*/Builder.CreateIntCast(M->getArgOperand(2), IntptrTy, false)});
+ Value *OpSizeValue = M->getArgOperand(2);
+ size_t NrCopyFns = NsanCopyFunction.size();
+ size_t NsanCopyFunctionIdx =
+ GetInstrumentationCalleeIdxForMemOp(OpSizeValue, NrCopyFns - 1);
+
+ if (NsanCopyFunctionIdx != NrCopyFns - 1)
+ Builder.CreateCall(NsanCopyFunction[NsanCopyFunctionIdx],
+ {/*Destination=*/M->getArgOperand(0),
+ /*Source=*/M->getArgOperand(1)});
+ else
+ Builder.CreateCall(
+ NsanCopyFunction.back(),
+ {/*Destination=*/M->getArgOperand(0),
+ /*Source=*/M->getArgOperand(1),
+ /*Size=*/Builder.CreateIntCast(OpSizeValue, IntptrTy, false)});
}
return false;
}
diff --git a/llvm/test/Instrumentation/NumericalStabilitySanitizer/memory.ll b/llvm/test/Instrumentation/NumericalStabilitySanitizer/memory.ll
index 3fe78d8b19b0a..1f238b88f49a4 100644
--- a/llvm/test/Instrumentation/NumericalStabilitySanitizer/memory.ll
+++ b/llvm/test/Instrumentation/NumericalStabilitySanitizer/memory.ll
@@ -7,10 +7,10 @@ target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f3
declare void @llvm.memcpy.p0i8.p0i8.i64(i8*, i8*, i64, i1)
-define void @call_memcpy_intrinsic(i8* nonnull align 8 dereferenceable(16) %a, i8* nonnull align 8 dereferenceable(16) %b) sanitize_numerical_stability {
-; CHECK-LABEL: @call_memcpy_intrinsic(
+define void @call_memcpy_intrinsic_16bytes(i8* nonnull align 8 dereferenceable(16) %a, i8* nonnull align 8 dereferenceable(16) %b) sanitize_numerical_stability {
+; CHECK-LABEL: @call_memcpy_intrinsic_16bytes(
; CHECK-NEXT: entry:
-; CHECK-NEXT: call void @__nsan_copy_values(ptr [[A:%.*]], ptr [[B:%.*]], i64 16)
+; CHECK-NEXT: call void @__nsan_copy_16(ptr [[A:%.*]], ptr [[B:%.*]])
; CHECK-NEXT: tail call void @llvm.memcpy.p0.p0.i64(ptr nonnull align 8 dereferenceable(16) [[A]], ptr nonnull align 8 dereferenceable(16) [[B]], i64 16, i1 false)
; CHECK-NEXT: ret void
;
@@ -19,6 +19,42 @@ entry:
ret void
}
+define void @call_memcpy_intrinsic_8bytes(i8* nonnull align 8 dereferenceable(16) %a, i8* nonnull align 8 dereferenceable(16) %b) sanitize_numerical_stability {
+; CHECK-LABEL: @call_memcpy_intrinsic_8bytes(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: call void @__nsan_copy_8(ptr [[A:%.*]], ptr [[B:%.*]])
+; CHECK-NEXT: tail call void @llvm.memcpy.p0.p0.i64(ptr nonnull align 8 dereferenceable(16) [[A]], ptr nonnull align 8 dereferenceable(16) [[B]], i64 8, i1 false)
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @llvm.memcpy.p0i8.p0i8.i64(ptr nonnull align 8 dereferenceable(16) %a, ptr nonnull align 8 dereferenceable(16) %b, i64 8, i1 false)
+ ret void
+}
+
+define void @call_memcpy_intrinsic_4bytes(i8* nonnull align 8 dereferenceable(16) %a, i8* nonnull align 8 dereferenceable(16) %b) sanitize_numerical_stability {
+; CHECK-LABEL: @call_memcpy_intrinsic_4bytes(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: call void @__nsan_copy_4(ptr [[A:%.*]], ptr [[B:%.*]])
+; CHECK-NEXT: tail call void @llvm.memcpy.p0.p0.i64(ptr nonnull align 8 dereferenceable(16) [[A]], ptr nonnull align 8 dereferenceable(16) [[B]], i64 4, i1 false)
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @llvm.memcpy.p0i8.p0i8.i64(ptr nonnull align 8 dereferenceable(16) %a, ptr nonnull align 8 dereferenceable(16) %b, i64 4, i1 false)
+ ret void
+}
+
+define void @call_memcpy_intrinsic(i8* nonnull align 8 dereferenceable(16) %a, i8* nonnull align 8 dereferenceable(16) %b) sanitize_numerical_stability {
+; CHECK-LABEL: @call_memcpy_intrinsic(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: call void @__nsan_copy_values(ptr [[A:%.*]], ptr [[B:%.*]], i64 15)
+; CHECK-NEXT: tail call void @llvm.memcpy.p0.p0.i64(ptr nonnull align 8 dereferenceable(16) [[A]], ptr nonnull align 8 dereferenceable(16) [[B]], i64 15, i1 false)
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @llvm.memcpy.p0i8.p0i8.i64(ptr nonnull align 8 dereferenceable(16) %a, ptr nonnull align 8 dereferenceable(16) %b, i64 15, i1 false)
+ ret void
+}
+
declare dso_local i8* @memcpy(i8*, i8*, i64) local_unnamed_addr
define void @call_memcpy(i8* nonnull align 8 dereferenceable(16) %a, i8* nonnull align 8 dereferenceable(16) %b) sanitize_numerical_stability {
@@ -32,6 +68,53 @@ entry:
ret void
}
+define void @call_memset_intrinsic_16bytes(i8* nonnull align 8 dereferenceable(16) %a) sanitize_numerical_stability {
+; CHECK-LABEL: @call_memset_intrinsic_16bytes(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: call void @__nsan_set_value_unknown_16(ptr [[A:%.*]])
+; CHECK-NEXT: tail call void @llvm.memset.p0.i64(ptr nonnull align 8 dereferenceable(16) [[A]], i8 0, i64 16, i1 false)
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @llvm.memset.p0.i64(ptr nonnull align 8 dereferenceable(16) %a, i8 0, i64 16, i1 false)
+ ret void
+}
+
+define void @call_memset_intrinsic_8bytes(i8* nonnull align 8 dereferenceable(16) %a) sanitize_numerical_stability {
+; CHECK-LABEL: @call_memset_intrinsic_8bytes(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: call void @__nsan_set_value_unknown_8(ptr [[A:%.*]])
+; CHECK-NEXT: tail call void @llvm.memset.p0.i64(ptr nonnull align 8 dereferenceable(16) [[A]], i8 0, i64 8, i1 false)
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @llvm.memset.p0.i64(ptr nonnull align 8 dereferenceable(16) %a, i8 0, i64 8, i1 false)
+ ret void
+}
+
+define void @call_memset_intrinsic_4bytes(i8* nonnull align 8 dereferenceable(16) %a) sanitize_numerical_stability {
+; CHECK-LABEL: @call_memset_intrinsic_4bytes(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: call void @__nsan_set_value_unknown_4(ptr [[A:%.*]])
+; CHECK-NEXT: tail call void @llvm.memset.p0.i64(ptr nonnull align 8 dereferenceable(16) [[A]], i8 0, i64 4, i1 false)
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @llvm.memset.p0.i64(ptr nonnull align 8 dereferenceable(16) %a, i8 0, i64 4, i1 false)
+ ret void
+}
+
+define void @call_memset_intrinsic(i8* nonnull align 8 dereferenceable(16) %a) sanitize_numerical_stability {
+; CHECK-LABEL: @call_memset_intrinsic(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: call void @__nsan_set_value_unknown(ptr [[A:%.*]], i64 15)
+; CHECK-NEXT: tail call void @llvm.memset.p0.i64(ptr nonnull align 8 dereferenceable(16) [[A]], i8 0, i64 15, i1 false)
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @llvm.memset.p0.i64(ptr nonnull align 8 dereferenceable(16) %a, i8 0, i64 15, i1 false)
+ ret void
+}
define void @transfer_float(float* %dst, float* %src) sanitize_numerical_stability {
; CHECK-LABEL: @transfer_float(
``````````
</details>
https://github.com/llvm/llvm-project/pull/98900
More information about the llvm-commits
mailing list