diff options
23 files changed, 272 insertions, 47 deletions
diff --git a/clang/lib/Sema/SemaTemplate.cpp b/clang/lib/Sema/SemaTemplate.cpp index dcf2876..419f3e1 100644 --- a/clang/lib/Sema/SemaTemplate.cpp +++ b/clang/lib/Sema/SemaTemplate.cpp @@ -3822,14 +3822,19 @@ QualType Sema::CheckTemplateIdType(ElaboratedTypeKeyword Keyword, AliasTemplate->getTemplateParameters()->getDepth()); LocalInstantiationScope Scope(*this); - InstantiatingTemplate Inst( - *this, /*PointOfInstantiation=*/TemplateLoc, - /*Entity=*/AliasTemplate, - /*TemplateArgs=*/TemplateArgLists.getInnermost()); // Diagnose uses of this alias. (void)DiagnoseUseOfDecl(AliasTemplate, TemplateLoc); + // FIXME: The TemplateArgs passed here are not used for the context note, + // nor they should, because this note will be pointing to the specialization + // anyway. These arguments are needed for a hack for instantiating lambdas + // in the pattern of the alias. In getTemplateInstantiationArgs, these + // arguments will be used for collating the template arguments needed to + // instantiate the lambda. + InstantiatingTemplate Inst(*this, /*PointOfInstantiation=*/TemplateLoc, + /*Entity=*/AliasTemplate, + /*TemplateArgs=*/CTAI.SugaredConverted); if (Inst.isInvalid()) return QualType(); diff --git a/clang/lib/Sema/SemaTemplateInstantiate.cpp b/clang/lib/Sema/SemaTemplateInstantiate.cpp index 1f762ca..7b05e4c 100644 --- a/clang/lib/Sema/SemaTemplateInstantiate.cpp +++ b/clang/lib/Sema/SemaTemplateInstantiate.cpp @@ -1271,6 +1271,12 @@ void Sema::PrintInstantiationStack(InstantiationContextDiagFuncRef DiagFunc) { PDiag(diag::note_building_deduction_guide_here)); break; case CodeSynthesisContext::TypeAliasTemplateInstantiation: + // Workaround for a workaround: don't produce a note if we are merely + // instantiating some other template which contains this alias template. + // This would be redundant either with the error itself, or some other + // context note attached to it. + if (Active->NumTemplateArgs == 0) + break; DiagFunc(Active->PointOfInstantiation, PDiag(diag::note_template_type_alias_instantiation_here) << cast<TypeAliasTemplateDecl>(Active->Entity) diff --git a/clang/lib/Sema/SemaTemplateInstantiateDecl.cpp b/clang/lib/Sema/SemaTemplateInstantiateDecl.cpp index e2dc703..3819f77 100644 --- a/clang/lib/Sema/SemaTemplateInstantiateDecl.cpp +++ b/clang/lib/Sema/SemaTemplateInstantiateDecl.cpp @@ -1580,17 +1580,19 @@ Decl *TemplateDeclInstantiator::InstantiateTypeAliasTemplateDecl( if (!InstParams) return nullptr; - TypeAliasDecl *Pattern = D->getTemplatedDecl(); - Sema::InstantiatingTemplate InstTemplate( - SemaRef, D->getBeginLoc(), D, - D->getTemplateDepth() >= TemplateArgs.getNumLevels() - ? ArrayRef<TemplateArgument>() - : (TemplateArgs.begin() + TemplateArgs.getNumLevels() - 1 - - D->getTemplateDepth()) - ->Args); + // FIXME: This is a hack for instantiating lambdas in the pattern of the + // alias. We are not really instantiating the alias at its template level, + // that only happens in CheckTemplateId, this is only for outer templates + // which contain it. In getTemplateInstantiationArgs, the template arguments + // used here would be used for collating the template arguments needed to + // instantiate the lambda. Pass an empty argument list, so this workaround + // doesn't get confused if there is an outer alias being instantiated. + Sema::InstantiatingTemplate InstTemplate(SemaRef, D->getBeginLoc(), D, + ArrayRef<TemplateArgument>()); if (InstTemplate.isInvalid()) return nullptr; + TypeAliasDecl *Pattern = D->getTemplatedDecl(); TypeAliasTemplateDecl *PrevAliasTemplate = nullptr; if (getPreviousDeclForInstantiation<TypedefNameDecl>(Pattern)) { DeclContext::lookup_result Found = Owner->lookup(Pattern->getDeclName()); diff --git a/clang/test/SemaTemplate/alias-template-deprecated.cpp b/clang/test/SemaTemplate/alias-template-deprecated.cpp index 7418e222..6dffd37 100644 --- a/clang/test/SemaTemplate/alias-template-deprecated.cpp +++ b/clang/test/SemaTemplate/alias-template-deprecated.cpp @@ -46,23 +46,19 @@ using UsingInstWithCPPAttr [[deprecated("Do not use this")]] = NoAttr<int>; void bar() { NoAttr<int> obj; // Okay - // expected-warning@+2 {{'UsingWithAttr' is deprecated}} - // expected-note@+1 {{in instantiation of template type alias 'UsingWithAttr' requested here}} + // expected-warning@+1 {{'UsingWithAttr' is deprecated}} UsingWithAttr<int> objUsingWA; - // expected-warning@+2 {{'UsingWithAttr' is deprecated}} - // expected-note@+1 {{in instantiation of template type alias 'UsingWithAttr' requested here}} + // expected-warning@+1 {{'UsingWithAttr' is deprecated}} NoAttr<UsingWithAttr<int>> s; // expected-note@+1 {{'DepInt' has been explicitly marked deprecated here}} using DepInt [[deprecated]] = int; - // expected-warning@+3 {{'UsingWithAttr' is deprecated}} - // expected-warning@+2 {{'DepInt' is deprecated}} - // expected-note@+1 {{in instantiation of template type alias 'UsingWithAttr' requested here}} + // expected-warning@+2 {{'UsingWithAttr' is deprecated}} + // expected-warning@+1 {{'DepInt' is deprecated}} using X = UsingWithAttr<DepInt>; - // expected-warning@+2 {{'UsingWithAttr' is deprecated}} - // expected-note@+1 {{in instantiation of template type alias 'UsingWithAttr' requested here}} + // expected-warning@+1 {{'UsingWithAttr' is deprecated}} UsingWithAttr<int>().foo(); // expected-warning@+1 {{'UsingInstWithAttr' is deprecated}} @@ -74,8 +70,7 @@ void bar() { // expected-warning@+1 {{'UsingTDWithAttr' is deprecated}} UsingTDWithAttr objUTDWA; - // expected-warning@+2 {{'UsingWithCPPAttr' is deprecated}} - // expected-note@+1 {{in instantiation of template type alias 'UsingWithCPPAttr' requested here}} + // expected-warning@+1 {{'UsingWithCPPAttr' is deprecated}} UsingWithCPPAttr<int> objUsingWCPPA; // expected-warning@+1 {{'UsingInstWithCPPAttr' is deprecated: Do not use this}} diff --git a/clang/test/SemaTemplate/alias-templates.cpp b/clang/test/SemaTemplate/alias-templates.cpp index ab5cad7..09fe72e 100644 --- a/clang/test/SemaTemplate/alias-templates.cpp +++ b/clang/test/SemaTemplate/alias-templates.cpp @@ -312,3 +312,11 @@ namespace resolved_nttp { using TC2 = decltype(C<bool, 2, 3>::p); // expected-note {{instantiation of}} } + +namespace OuterSubstFailure { + template <class T> struct A { + template <class> using B = T&; + // expected-error@-1 {{cannot form a reference to 'void'}} + }; + template struct A<void>; // expected-note {{requested here}} +} // namespace OuterSubstFailure diff --git a/compiler-rt/lib/asan/asan_interceptors.cpp b/compiler-rt/lib/asan/asan_interceptors.cpp index 7c9a08b..0f613f0 100644 --- a/compiler-rt/lib/asan/asan_interceptors.cpp +++ b/compiler-rt/lib/asan/asan_interceptors.cpp @@ -58,13 +58,20 @@ namespace __asan { static inline uptr MaybeRealStrnlen(const char *s, uptr maxlen) { #if SANITIZER_INTERCEPT_STRNLEN - if (REAL(strnlen)) { + if (REAL(strnlen)) return REAL(strnlen)(s, maxlen); - } -#endif +# endif return internal_strnlen(s, maxlen); } +static inline uptr MaybeRealWcsnlen(const wchar_t* s, uptr maxlen) { +# if SANITIZER_INTERCEPT_WCSNLEN + if (REAL(wcsnlen)) + return REAL(wcsnlen)(s, maxlen); +# endif + return internal_wcsnlen(s, maxlen); +} + void SetThreadName(const char *name) { AsanThread *t = GetCurrentThread(); if (t) @@ -570,6 +577,20 @@ INTERCEPTOR(char *, strcpy, char *to, const char *from) { return REAL(strcpy)(to, from); } +INTERCEPTOR(wchar_t*, wcscpy, wchar_t* to, const wchar_t* from) { + void* ctx; + ASAN_INTERCEPTOR_ENTER(ctx, wcscpy); + if (!TryAsanInitFromRtl()) + return REAL(wcscpy)(to, from); + if (flags()->replace_str) { + uptr size = (internal_wcslen(from) + 1) * sizeof(wchar_t); + CHECK_RANGES_OVERLAP("wcscpy", to, size, from, size); + ASAN_READ_RANGE(ctx, from, size); + ASAN_WRITE_RANGE(ctx, to, size); + } + return REAL(wcscpy)(to, from); +} + // Windows doesn't always define the strdup identifier, // and when it does it's a macro defined to either _strdup // or _strdup_dbg, _strdup_dbg ends up calling _strdup, so @@ -633,6 +654,20 @@ INTERCEPTOR(char*, strncpy, char *to, const char *from, usize size) { return REAL(strncpy)(to, from, size); } +INTERCEPTOR(wchar_t*, wcsncpy, wchar_t* to, const wchar_t* from, uptr size) { + void* ctx; + ASAN_INTERCEPTOR_ENTER(ctx, wcsncpy); + AsanInitFromRtl(); + if (flags()->replace_str) { + uptr from_size = + Min(size, MaybeRealWcsnlen(from, size) + 1) * sizeof(wchar_t); + CHECK_RANGES_OVERLAP("wcsncpy", to, from_size, from, from_size); + ASAN_READ_RANGE(ctx, from, from_size); + ASAN_WRITE_RANGE(ctx, to, size * sizeof(wchar_t)); + } + return REAL(wcsncpy)(to, from, size); +} + template <typename Fn> static ALWAYS_INLINE auto StrtolImpl(void *ctx, Fn real, const char *nptr, char **endptr, int base) @@ -809,6 +844,11 @@ void InitializeAsanInterceptors() { ASAN_INTERCEPT_FUNC(strncat); ASAN_INTERCEPT_FUNC(strncpy); ASAN_INTERCEPT_FUNC(strdup); + + // Intercept wcs* functions. + ASAN_INTERCEPT_FUNC(wcscpy); + ASAN_INTERCEPT_FUNC(wcsncpy); + # if ASAN_INTERCEPT___STRDUP ASAN_INTERCEPT_FUNC(__strdup); #endif diff --git a/compiler-rt/lib/asan/asan_interceptors.h b/compiler-rt/lib/asan/asan_interceptors.h index 3e2386e..2d551cf 100644 --- a/compiler-rt/lib/asan/asan_interceptors.h +++ b/compiler-rt/lib/asan/asan_interceptors.h @@ -129,6 +129,7 @@ DECLARE_REAL(char*, strchr, const char *str, int c) DECLARE_REAL(SIZE_T, strlen, const char *s) DECLARE_REAL(char*, strncpy, char *to, const char *from, SIZE_T size) DECLARE_REAL(SIZE_T, strnlen, const char *s, SIZE_T maxlen) +DECLARE_REAL(SIZE_T, wcsnlen, const wchar_t* s, SIZE_T maxlen) DECLARE_REAL(char*, strstr, const char *s1, const char *s2) # if !SANITIZER_APPLE diff --git a/compiler-rt/lib/asan/asan_win_static_runtime_thunk.cpp b/compiler-rt/lib/asan/asan_win_static_runtime_thunk.cpp index 4a69b66..46e0e90 100644 --- a/compiler-rt/lib/asan/asan_win_static_runtime_thunk.cpp +++ b/compiler-rt/lib/asan/asan_win_static_runtime_thunk.cpp @@ -63,6 +63,10 @@ INTERCEPT_LIBRARY_FUNCTION_ASAN(strpbrk); INTERCEPT_LIBRARY_FUNCTION_ASAN(strspn); INTERCEPT_LIBRARY_FUNCTION_ASAN(strstr); INTERCEPT_LIBRARY_FUNCTION_ASAN(strtok); +INTERCEPT_LIBRARY_FUNCTION_ASAN(wcscat); +INTERCEPT_LIBRARY_FUNCTION_ASAN(wcscpy); +INTERCEPT_LIBRARY_FUNCTION_ASAN(wcsncat); +INTERCEPT_LIBRARY_FUNCTION_ASAN(wcsncpy); INTERCEPT_LIBRARY_FUNCTION_ASAN(wcslen); INTERCEPT_LIBRARY_FUNCTION_ASAN(wcsnlen); diff --git a/compiler-rt/lib/sanitizer_common/sanitizer_platform_interceptors.h b/compiler-rt/lib/sanitizer_common/sanitizer_platform_interceptors.h index 29987de..88ecd7e 100644 --- a/compiler-rt/lib/sanitizer_common/sanitizer_platform_interceptors.h +++ b/compiler-rt/lib/sanitizer_common/sanitizer_platform_interceptors.h @@ -551,7 +551,7 @@ SANITIZER_WEAK_IMPORT void *aligned_alloc(__sanitizer::usize __alignment, #define SANITIZER_INTERCEPT_MALLOC_USABLE_SIZE (!SI_MAC && !SI_NETBSD) #define SANITIZER_INTERCEPT_MCHECK_MPROBE SI_LINUX_NOT_ANDROID #define SANITIZER_INTERCEPT_WCSLEN 1 -#define SANITIZER_INTERCEPT_WCSCAT SI_POSIX +#define SANITIZER_INTERCEPT_WCSCAT (SI_POSIX || SI_WINDOWS) #define SANITIZER_INTERCEPT_WCSDUP SI_POSIX #define SANITIZER_INTERCEPT_SIGNAL_AND_SIGACTION (!SI_WINDOWS && SI_NOT_FUCHSIA) #define SANITIZER_INTERCEPT_BSD_SIGNAL SI_ANDROID diff --git a/compiler-rt/test/asan/TestCases/wcscat.cpp b/compiler-rt/test/asan/TestCases/wcscat.cpp new file mode 100644 index 0000000..fd0b5a4 --- /dev/null +++ b/compiler-rt/test/asan/TestCases/wcscat.cpp @@ -0,0 +1,26 @@ +// RUN: %clangxx_asan -O0 %s -o %t && not %env_asan_opts=log_to_stderr=1 %run %t 2>&1 | FileCheck %s --check-prefix=CHECK +// RUN: %clangxx_asan -O1 %s -o %t && not %env_asan_opts=log_to_stderr=1 %run %t 2>&1 | FileCheck %s --check-prefix=CHECK +// RUN: %clangxx_asan -O2 %s -o %t && not %env_asan_opts=log_to_stderr=1 %run %t 2>&1 | FileCheck %s --check-prefix=CHECK +// RUN: %clangxx_asan -O3 %s -o %t && not %env_asan_opts=log_to_stderr=1 %run %t 2>&1 | FileCheck %s --check-prefix=CHECK + +#include <stdio.h> +#include <wchar.h> + +int main() { + const wchar_t *start = L"X means "; + const wchar_t *append = L"dog"; + wchar_t goodDst[12]; + wcscpy(goodDst, start); + wcscat(goodDst, append); + + wchar_t badDst[9]; + wcscpy(badDst, start); + fprintf(stderr, "Good so far.\n"); + // CHECK-DAG: Good so far. + fflush(stderr); + wcscat(badDst, append); // Boom! + // CHECK-DAG: ERROR: AddressSanitizer: stack-buffer-overflow on address [[ADDR:0x[0-9a-f]+]] at pc {{0x[0-9a-f]+}} bp {{0x[0-9a-f]+}} sp {{0x[0-9a-f]+}} + // CHECK-DAG: WRITE of size {{[0-9]+}} at [[ADDR]] thread T0 + // CHECK-DAG: #0 {{0x[0-9a-f]+}} in wcscat + printf("Should have failed with ASAN error.\n"); +} diff --git a/compiler-rt/test/asan/TestCases/wcscpy.cpp b/compiler-rt/test/asan/TestCases/wcscpy.cpp new file mode 100644 index 0000000..8133a58 --- /dev/null +++ b/compiler-rt/test/asan/TestCases/wcscpy.cpp @@ -0,0 +1,23 @@ +// RUN: %clangxx_asan -O0 %s -o %t && not %env_asan_opts=log_to_stderr=1 %run %t 2>&1 | FileCheck %s --check-prefix=CHECK +// RUN: %clangxx_asan -O1 %s -o %t && not %env_asan_opts=log_to_stderr=1 %run %t 2>&1 | FileCheck %s --check-prefix=CHECK +// RUN: %clangxx_asan -O2 %s -o %t && not %env_asan_opts=log_to_stderr=1 %run %t 2>&1 | FileCheck %s --check-prefix=CHECK +// RUN: %clangxx_asan -O3 %s -o %t && not %env_asan_opts=log_to_stderr=1 %run %t 2>&1 | FileCheck %s --check-prefix=CHECK + +#include <stdio.h> +#include <wchar.h> + +int main() { + const wchar_t *src = L"X means dog"; + wchar_t goodDst[12]; + wcscpy(goodDst, src); + + wchar_t badDst[7]; + fprintf(stderr, "Good so far.\n"); + // CHECK-DAG: Good so far. + fflush(stderr); + wcscpy(badDst, src); // Boom! + // CHECK-DAG: ERROR: AddressSanitizer: stack-buffer-overflow on address [[ADDR:0x[0-9a-f]+]] at pc {{0x[0-9a-f]+}} bp {{0x[0-9a-f]+}} sp {{0x[0-9a-f]+}} + // CHECK-DAG: WRITE of size {{[0-9]+}} at [[ADDR]] thread T0 + // CHECK-DAG: #0 {{0x[0-9a-f]+}} in wcscpy + printf("Should have failed with ASAN error.\n"); +} diff --git a/compiler-rt/test/asan/TestCases/wcsncat.cpp b/compiler-rt/test/asan/TestCases/wcsncat.cpp new file mode 100644 index 0000000..365e732 --- /dev/null +++ b/compiler-rt/test/asan/TestCases/wcsncat.cpp @@ -0,0 +1,27 @@ +// RUN: %clangxx_asan -O0 %s -o %t && not %env_asan_opts=log_to_stderr=1 %run %t 2>&1 | FileCheck %s --check-prefix=CHECK +// RUN: %clangxx_asan -O1 %s -o %t && not %env_asan_opts=log_to_stderr=1 %run %t 2>&1 | FileCheck %s --check-prefix=CHECK +// RUN: %clangxx_asan -O2 %s -o %t && not %env_asan_opts=log_to_stderr=1 %run %t 2>&1 | FileCheck %s --check-prefix=CHECK +// RUN: %clangxx_asan -O3 %s -o %t && not %env_asan_opts=log_to_stderr=1 %run %t 2>&1 | FileCheck %s --check-prefix=CHECK + +#include <stdio.h> +#include <wchar.h> + +int main() { + const wchar_t *start = L"X means "; + const wchar_t *append = L"dog"; + wchar_t goodDst[15]; + wcscpy(goodDst, start); + wcsncat(goodDst, append, 5); + + wchar_t badDst[11]; + wcscpy(badDst, start); + wcsncat(badDst, append, 1); + fprintf(stderr, "Good so far.\n"); + // CHECK-DAG: Good so far. + fflush(stderr); + wcsncat(badDst, append, 3); // Boom! + // CHECK-DAG: ERROR: AddressSanitizer: stack-buffer-overflow on address [[ADDR:0x[0-9a-f]+]] at pc {{0x[0-9a-f]+}} bp {{0x[0-9a-f]+}} sp {{0x[0-9a-f]+}} + // CHECK-DAG: WRITE of size {{[0-9]+}} at [[ADDR]] thread T0 + // CHECK-DAG: #0 {{0x[0-9a-f]+}} in wcsncat + printf("Should have failed with ASAN error.\n"); +} diff --git a/compiler-rt/test/asan/TestCases/wcsncpy.cpp b/compiler-rt/test/asan/TestCases/wcsncpy.cpp new file mode 100644 index 0000000..485ddc4 --- /dev/null +++ b/compiler-rt/test/asan/TestCases/wcsncpy.cpp @@ -0,0 +1,24 @@ +// RUN: %clangxx_asan -O0 %s -o %t && not %env_asan_opts=log_to_stderr=1 %run %t 2>&1 | FileCheck %s --check-prefix=CHECK +// RUN: %clangxx_asan -O1 %s -o %t && not %env_asan_opts=log_to_stderr=1 %run %t 2>&1 | FileCheck %s --check-prefix=CHECK +// RUN: %clangxx_asan -O2 %s -o %t && not %env_asan_opts=log_to_stderr=1 %run %t 2>&1 | FileCheck %s --check-prefix=CHECK +// RUN: %clangxx_asan -O3 %s -o %t && not %env_asan_opts=log_to_stderr=1 %run %t 2>&1 | FileCheck %s --check-prefix=CHECK + +#include <stdio.h> +#include <wchar.h> + +int main() { + const wchar_t *src = L"X means dog"; + wchar_t goodDst[12]; + wcsncpy(goodDst, src, 12); + + wchar_t badDst[7]; + wcsncpy(badDst, src, 7); // This should still work. + fprintf(stderr, "Good so far.\n"); + // CHECK-DAG: Good so far. + fflush(stderr); + wcsncpy(badDst, src, 15); // Boom! + // CHECK-DAG: ERROR: AddressSanitizer: stack-buffer-overflow on address [[ADDR:0x[0-9a-f]+]] at pc {{0x[0-9a-f]+}} bp {{0x[0-9a-f]+}} sp {{0x[0-9a-f]+}} + // CHECK-DAG: WRITE of size {{[0-9]+}} at [[ADDR]] thread T0 + // CHECK-DAG: #0 {{0x[0-9a-f]+}} in wcsncpy + printf("Should have failed with ASAN error.\n"); +} diff --git a/flang/include/flang/Optimizer/Builder/FIRBuilder.h b/flang/include/flang/Optimizer/Builder/FIRBuilder.h index d3af3ba..2ce0d86 100644 --- a/flang/include/flang/Optimizer/Builder/FIRBuilder.h +++ b/flang/include/flang/Optimizer/Builder/FIRBuilder.h @@ -372,6 +372,8 @@ public: return createCommonLinkage(getContext()); } + mlir::StringAttr createExternalLinkage() { return getStringAttr("external"); } + mlir::StringAttr createInternalLinkage() { return getStringAttr("internal"); } mlir::StringAttr createLinkOnceLinkage() { return getStringAttr("linkonce"); } diff --git a/flang/lib/Optimizer/Transforms/CUFComputeSharedMemoryOffsetsAndSize.cpp b/flang/lib/Optimizer/Transforms/CUFComputeSharedMemoryOffsetsAndSize.cpp index 6e04c71..09126e0 100644 --- a/flang/lib/Optimizer/Transforms/CUFComputeSharedMemoryOffsetsAndSize.cpp +++ b/flang/lib/Optimizer/Transforms/CUFComputeSharedMemoryOffsetsAndSize.cpp @@ -143,7 +143,11 @@ struct CUFComputeSharedMemoryOffsetsAndSize auto sharedMemType = fir::SequenceType::get(sharedMemSize, i8Ty); std::string sharedMemGlobalName = (funcOp.getName() + llvm::Twine(cudaSharedMemSuffix)).str(); - mlir::StringAttr linkage = builder.createInternalLinkage(); + // Dynamic shared memory needs an external linkage while static shared + // memory needs an internal linkage. + mlir::StringAttr linkage = nbDynamicSharedVariables > 0 + ? builder.createExternalLinkage() + : builder.createInternalLinkage(); builder.setInsertionPointToEnd(gpuMod.getBody()); llvm::SmallVector<mlir::NamedAttribute> attrs; auto globalOpName = mlir::OperationName(fir::GlobalOp::getOperationName(), diff --git a/flang/test/Fir/CUDA/cuda-shared-offset.mlir b/flang/test/Fir/CUDA/cuda-shared-offset.mlir index 29316c9..9c057d0 100644 --- a/flang/test/Fir/CUDA/cuda-shared-offset.mlir +++ b/flang/test/Fir/CUDA/cuda-shared-offset.mlir @@ -17,7 +17,7 @@ module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<!llvm.ptr, dense< // CHECK: %{{.*}} = cuf.shared_memory[%c0{{.*}} : i32] !fir.array<?xf32>, %c-1 : index {bindc_name = "r", uniq_name = "_QFdynsharedEr"} -> !fir.ref<!fir.array<?xf32>> // CHECK: gpu.return // CHECK: } -// CHECK: fir.global internal @_QPdynshared__shared_mem {alignment = 4 : i64, data_attr = #cuf.cuda<shared>} : !fir.array<0xi8> +// CHECK: fir.global external @_QPdynshared__shared_mem {alignment = 4 : i64, data_attr = #cuf.cuda<shared>} : !fir.array<0xi8> // ----- @@ -158,3 +158,5 @@ module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<!llvm.ptr, dense< // CHECK-LABEL: gpu.func @_QMmtestsPtestany // CHECK: %{{.*}} = cuf.shared_memory[%c0{{.*}} : i32] !fir.array<?xf64>, %c-1{{.*}} : index {bindc_name = "dmasks", uniq_name = "_QMmtestsFtestanyEdmasks"} -> !fir.ref<!fir.array<?xf64>> // CHECK: %{{.*}} = cuf.shared_memory[%c0{{.*}} : i32] !fir.array<?xf32>, %c-1{{.*}} : index {bindc_name = "smasks", uniq_name = "_QMmtestsFtestanyEsmasks"} -> !fir.ref<!fir.array<?xf32>> + +// CHECK: fir.global external @_QMmtestsPtestany__shared_mem {alignment = 8 : i64, data_attr = #cuf.cuda<shared>} : !fir.array<0xi8> diff --git a/libclc/CMakeLists.txt b/libclc/CMakeLists.txt index 7960f34..ba0fc4b 100644 --- a/libclc/CMakeLists.txt +++ b/libclc/CMakeLists.txt @@ -225,7 +225,7 @@ set( tahiti_aliases pitcairn verde oland hainan bonaire kabini kaveri hawaii gfx1030 gfx1031 gfx1032 gfx1033 gfx1034 gfx1035 gfx1036 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1153 - gfx1200 gfx1201 + gfx1200 gfx1201 gfx1250 gfx1251 ) # pkg-config file diff --git a/llvm/include/llvm/Analysis/MemoryProfileInfo.h b/llvm/include/llvm/Analysis/MemoryProfileInfo.h index 571caf9..be690a4 100644 --- a/llvm/include/llvm/Analysis/MemoryProfileInfo.h +++ b/llvm/include/llvm/Analysis/MemoryProfileInfo.h @@ -59,6 +59,14 @@ LLVM_ABI std::string getAllocTypeAttributeString(AllocationType Type); /// True if the AllocTypes bitmask contains just a single type. LLVM_ABI bool hasSingleAllocType(uint8_t AllocTypes); +/// Removes any existing "ambiguous" memprof attribute. Called before we apply a +/// specific allocation type such as "cold", "notcold", or "hot". +LLVM_ABI void removeAnyExistingAmbiguousAttribute(CallBase *CB); + +/// Adds an "ambiguous" memprof attribute to call with a matched allocation +/// profile but that we haven't yet been able to disambiguate. +LLVM_ABI void addAmbiguousAttribute(CallBase *CB); + /// Class to build a trie of call stack contexts for a particular profiled /// allocation call, along with their associated allocation types. /// The allocation will be at the root of the trie, which is then used to diff --git a/llvm/lib/Analysis/MemoryProfileInfo.cpp b/llvm/lib/Analysis/MemoryProfileInfo.cpp index 0c1f8db..92a5b6f 100644 --- a/llvm/lib/Analysis/MemoryProfileInfo.cpp +++ b/llvm/lib/Analysis/MemoryProfileInfo.cpp @@ -54,6 +54,10 @@ cl::opt<unsigned> MinPercentMaxColdSize( "memprof-min-percent-max-cold-size", cl::init(100), cl::Hidden, cl::desc("Min percent of max cold bytes for critical cold context")); +LLVM_ABI cl::opt<bool> MemProfUseAmbiguousAttributes( + "memprof-ambiguous-attributes", cl::init(true), cl::Hidden, + cl::desc("Apply ambiguous memprof attribute to ambiguous allocations")); + } // end namespace llvm bool llvm::memprof::metadataIncludesAllContextSizeInfo() { @@ -125,6 +129,26 @@ bool llvm::memprof::hasSingleAllocType(uint8_t AllocTypes) { return NumAllocTypes == 1; } +void llvm::memprof::removeAnyExistingAmbiguousAttribute(CallBase *CB) { + if (!CB->hasFnAttr("memprof")) + return; + assert(CB->getFnAttr("memprof").getValueAsString() == "ambiguous"); + CB->removeFnAttr("memprof"); +} + +void llvm::memprof::addAmbiguousAttribute(CallBase *CB) { + if (!MemProfUseAmbiguousAttributes) + return; + // We may have an existing ambiguous attribute if we are reanalyzing + // after inlining. + if (CB->hasFnAttr("memprof")) { + assert(CB->getFnAttr("memprof").getValueAsString() == "ambiguous"); + } else { + auto A = llvm::Attribute::get(CB->getContext(), "memprof", "ambiguous"); + CB->addFnAttr(A); + } +} + void CallStackTrie::addCallStack( AllocationType AllocType, ArrayRef<uint64_t> StackIds, std::vector<ContextTotalSize> ContextSizeInfo) { @@ -470,6 +494,9 @@ void CallStackTrie::addSingleAllocTypeAttribute(CallBase *CI, AllocationType AT, StringRef Descriptor) { auto AllocTypeString = getAllocTypeAttributeString(AT); auto A = llvm::Attribute::get(CI->getContext(), "memprof", AllocTypeString); + // After inlining we may be able to convert an existing ambiguous allocation + // to an unambiguous one. + removeAnyExistingAmbiguousAttribute(CI); CI->addFnAttr(A); if (MemProfReportHintedSizes) { std::vector<ContextTotalSize> ContextSizeInfo; @@ -529,6 +556,7 @@ bool CallStackTrie::buildAndAttachMIBMetadata(CallBase *CI) { assert(MIBCallStack.size() == 1 && "Should only be left with Alloc's location in stack"); CI->setMetadata(LLVMContext::MD_memprof, MDNode::get(Ctx, MIBNodes)); + addAmbiguousAttribute(CI); return true; } // If there exists corner case that CallStackTrie has one chain to leaf diff --git a/llvm/lib/Transforms/IPO/MemProfContextDisambiguation.cpp b/llvm/lib/Transforms/IPO/MemProfContextDisambiguation.cpp index faeab95..cfdfd94 100644 --- a/llvm/lib/Transforms/IPO/MemProfContextDisambiguation.cpp +++ b/llvm/lib/Transforms/IPO/MemProfContextDisambiguation.cpp @@ -3986,6 +3986,7 @@ void CallsiteContextGraph<DerivedCCG, FuncTy, CallTy>::identifyClones( void ModuleCallsiteContextGraph::updateAllocationCall( CallInfo &Call, AllocationType AllocType) { std::string AllocTypeString = getAllocTypeAttributeString(AllocType); + removeAnyExistingAmbiguousAttribute(cast<CallBase>(Call.call())); auto A = llvm::Attribute::get(Call.call()->getFunction()->getContext(), "memprof", AllocTypeString); cast<CallBase>(Call.call())->addFnAttr(A); @@ -5661,9 +5662,10 @@ bool MemProfContextDisambiguation::applyImport(Module &M) { auto *MemProfMD = I.getMetadata(LLVMContext::MD_memprof); // Include allocs that were already assigned a memprof function - // attribute in the statistics. - if (CB->getAttributes().hasFnAttr("memprof")) { - assert(!MemProfMD); + // attribute in the statistics. Only do this for those that do not have + // memprof metadata, since we add an "ambiguous" memprof attribute by + // default. + if (CB->getAttributes().hasFnAttr("memprof") && !MemProfMD) { CB->getAttributes().getFnAttr("memprof").getValueAsString() == "cold" ? AllocTypeColdThinBackend++ : AllocTypeNotColdThinBackend++; @@ -5740,6 +5742,7 @@ bool MemProfContextDisambiguation::applyImport(Module &M) { // clone J-1 (J==0 is the original clone and does not have a VMaps // entry). CBClone = cast<CallBase>((*VMaps[J - 1])[CB]); + removeAnyExistingAmbiguousAttribute(CBClone); CBClone->addFnAttr(A); ORE.emit(OptimizationRemark(DEBUG_TYPE, "MemprofAttribute", CBClone) << ore::NV("AllocationCall", CBClone) << " in clone " diff --git a/llvm/test/ThinLTO/X86/memprof-basic.ll b/llvm/test/ThinLTO/X86/memprof-basic.ll index 0ff0ce0..537e1b8 100644 --- a/llvm/test/ThinLTO/X86/memprof-basic.ll +++ b/llvm/test/ThinLTO/X86/memprof-basic.ll @@ -103,7 +103,9 @@ declare i32 @sleep() define internal ptr @_Z3barv() #0 !dbg !15 { entry: - %call = call ptr @_Znam(i64 0), !memprof !2, !callsite !7 + ;; Use an ambiguous attribute for this allocation, which is now added to such + ;; allocations during matching. It should not affect cloning. + %call = call ptr @_Znam(i64 0) #1, !memprof !2, !callsite !7 ret ptr null } @@ -125,6 +127,7 @@ entry: uselistorder ptr @_Z3foov, { 1, 0 } attributes #0 = { noinline optnone } +attributes #1 = { "memprof"="ambiguous" } !llvm.dbg.cu = !{!13} !llvm.module.flags = !{!20, !21} diff --git a/llvm/test/Transforms/PGOProfile/memprof.ll b/llvm/test/Transforms/PGOProfile/memprof.ll index c69d031..f6a89a8 100644 --- a/llvm/test/Transforms/PGOProfile/memprof.ll +++ b/llvm/test/Transforms/PGOProfile/memprof.ll @@ -38,7 +38,7 @@ ; ALL-NOT: no profile data available for function ;; Using a memprof-only profile for memprof-use should only give memprof metadata -; RUN: opt < %s -passes='memprof-use<profile-filename=%t.memprofdata>' -pgo-warn-missing-function -S -memprof-print-match-info -stats 2>&1 | FileCheck %s --check-prefixes=MEMPROF,ALL,MEMPROFONLY,MEMPROFMATCHINFO,MEMPROFSTATS +; RUN: opt < %s -passes='memprof-use<profile-filename=%t.memprofdata>' -pgo-warn-missing-function -S -memprof-print-match-info -stats 2>&1 | FileCheck %s --check-prefixes=MEMPROF,ALL,MEMPROFONLY,MEMPROFMATCHINFO,MEMPROFSTATS,AMBIG ; There should not be any PGO metadata ; MEMPROFONLY-NOT: !prof @@ -51,10 +51,10 @@ ;; Test the same thing but by passing the memory profile through to a default ;; pipeline via -memory-profile-file=, which should cause the necessary field ;; of the PGOOptions structure to be populated with the profile filename. -; RUN: opt < %s -passes='default<O2>' -memory-profile-file=%t.memprofdata -pgo-warn-missing-function -S 2>&1 | FileCheck %s --check-prefixes=MEMPROF,ALL,MEMPROFONLY +; RUN: opt < %s -passes='default<O2>' -memory-profile-file=%t.memprofdata -pgo-warn-missing-function -S 2>&1 | FileCheck %s --check-prefixes=MEMPROF,ALL,MEMPROFONLY,AMBIG ;; Using a pgo+memprof profile for memprof-use should only give memprof metadata -; RUN: opt < %s -passes='memprof-use<profile-filename=%t.pgomemprofdata>' -pgo-warn-missing-function -S 2>&1 | FileCheck %s --check-prefixes=MEMPROF,ALL,MEMPROFONLY +; RUN: opt < %s -passes='memprof-use<profile-filename=%t.pgomemprofdata>' -pgo-warn-missing-function -S 2>&1 | FileCheck %s --check-prefixes=MEMPROF,ALL,MEMPROFONLY,AMBIG ;; Using a pgo-only profile for memprof-use should give an error ; RUN: not opt < %s -passes='memprof-use<profile-filename=%t.pgoprofdata>' -S 2>&1 | FileCheck %s --check-prefixes=MEMPROFWITHPGOONLY @@ -72,7 +72,7 @@ ;; Using a pgo+memprof profile for both memprof-use and pgo-instr-use should ;; give both memprof and pgo metadata. -; RUN: opt < %s -passes='pgo-instr-use,memprof-use<profile-filename=%t.pgomemprofdata>' -pgo-test-profile-file=%t.pgomemprofdata -pgo-warn-missing-function -S 2>&1 | FileCheck %s --check-prefixes=MEMPROF,ALL,PGO +; RUN: opt < %s -passes='pgo-instr-use,memprof-use<profile-filename=%t.pgomemprofdata>' -pgo-test-profile-file=%t.pgomemprofdata -pgo-warn-missing-function -S 2>&1 | FileCheck %s --check-prefixes=MEMPROF,ALL,PGO,AMBIG ;; Check that the total sizes are reported if requested. A message should be ;; emitted for the pruned context. Also check that remarks are emitted for the @@ -108,7 +108,11 @@ ;; However, with the same threshold, but hot hints not enabled, it should be ;; notcold again. -; RUN: opt < %s -passes='memprof-use<profile-filename=%t.memprofdata>' -pgo-warn-missing-function -S -memprof-min-ave-lifetime-access-density-hot-threshold=0 2>&1 | FileCheck %s --check-prefixes=MEMPROF,ALL +; RUN: opt < %s -passes='memprof-use<profile-filename=%t.memprofdata>' -pgo-warn-missing-function -S -memprof-min-ave-lifetime-access-density-hot-threshold=0 2>&1 | FileCheck %s --check-prefixes=MEMPROF,ALL,AMBIG + +;; Test that we don't get an ambiguous memprof attribute when +;; -memprof-ambiguous-attributes is disabled. +; RUN: opt < %s -passes='memprof-use<profile-filename=%t.memprofdata>' -pgo-warn-missing-function -S -memprof-ambiguous-attributes=false 2>&1 | FileCheck %s --check-prefixes=MEMPROF,ALL,NOAMBIG ; MEMPROFMATCHINFO: MemProf notcold context with id 1093248920606587996 has total profiled size 10 is matched with 1 frames ; MEMPROFMATCHINFO: MemProf notcold context with id 5725971306423925017 has total profiled size 10 is matched with 1 frames @@ -140,7 +144,7 @@ target triple = "x86_64-unknown-linux-gnu" ; PGO: !prof define dso_local noundef ptr @_Z3foov() #0 !dbg !10 { entry: - ; MEMPROF: call {{.*}} @_Znam{{.*}} !memprof ![[M1:[0-9]+]], !callsite ![[C1:[0-9]+]] + ; MEMPROF: call {{.*}} @_Znam{{.*}} #[[A0:[0-9]+]]{{.*}} !memprof ![[M1:[0-9]+]], !callsite ![[C1:[0-9]+]] ; MEMPROFNOCOLINFO: call {{.*}} @_Znam{{.*}} !memprof ![[M1:[0-9]+]], !callsite ![[C1:[0-9]+]] %call = call noalias noundef nonnull ptr @_Znam(i64 noundef 10) #6, !dbg !13 ret ptr %call, !dbg !14 @@ -364,6 +368,9 @@ for.end: ; preds = %for.cond ret i32 0, !dbg !103 } +;; We optionally apply an ambiguous memprof attribute to ambiguous allocations +; AMBIG: #[[A0]] = { builtin allocsize(0) "memprof"="ambiguous" } +; NOAMBIG: #[[A0]] = { builtin allocsize(0) } ; MEMPROF: #[[A1]] = { builtin allocsize(0) "memprof"="notcold" } ; MEMPROF: #[[A2]] = { builtin allocsize(0) "memprof"="cold" } ; MEMPROF: ![[M1]] = !{![[MIB1:[0-9]+]], ![[MIB2:[0-9]+]], ![[MIB3:[0-9]+]], ![[MIB4:[0-9]+]]} diff --git a/llvm/unittests/Analysis/MemoryProfileInfoTest.cpp b/llvm/unittests/Analysis/MemoryProfileInfoTest.cpp index d8457a3..d1c0f64 100644 --- a/llvm/unittests/Analysis/MemoryProfileInfoTest.cpp +++ b/llvm/unittests/Analysis/MemoryProfileInfoTest.cpp @@ -230,7 +230,8 @@ declare dso_local noalias noundef i8* @malloc(i64 noundef) CallBase *Call = findCall(*Func, "call"); Trie.buildAndAttachMIBMetadata(Call); - EXPECT_FALSE(Call->hasFnAttr("memprof")); + EXPECT_TRUE(Call->hasFnAttr("memprof")); + EXPECT_EQ(Call->getFnAttr("memprof").getValueAsString(), "ambiguous"); EXPECT_TRUE(Call->hasMetadata(LLVMContext::MD_memprof)); MDNode *MemProfMD = Call->getMetadata(LLVMContext::MD_memprof); ASSERT_EQ(MemProfMD->getNumOperands(), 2u); @@ -279,7 +280,8 @@ declare dso_local noalias noundef i8* @malloc(i64 noundef) CallBase *Call = findCall(*Func, "call"); Trie.buildAndAttachMIBMetadata(Call); - EXPECT_FALSE(Call->hasFnAttr("memprof")); + EXPECT_TRUE(Call->hasFnAttr("memprof")); + EXPECT_EQ(Call->getFnAttr("memprof").getValueAsString(), "ambiguous"); EXPECT_TRUE(Call->hasMetadata(LLVMContext::MD_memprof)); MDNode *MemProfMD = Call->getMetadata(LLVMContext::MD_memprof); ASSERT_EQ(MemProfMD->getNumOperands(), 2u); @@ -333,7 +335,8 @@ declare dso_local noalias noundef i8* @malloc(i64 noundef) CallBase *Call = findCall(*Func, "call"); Trie.buildAndAttachMIBMetadata(Call); - EXPECT_FALSE(Call->hasFnAttr("memprof")); + EXPECT_TRUE(Call->hasFnAttr("memprof")); + EXPECT_EQ(Call->getFnAttr("memprof").getValueAsString(), "ambiguous"); EXPECT_TRUE(Call->hasMetadata(LLVMContext::MD_memprof)); MDNode *MemProfMD = Call->getMetadata(LLVMContext::MD_memprof); ASSERT_EQ(MemProfMD->getNumOperands(), 2u); @@ -392,7 +395,8 @@ declare dso_local noalias noundef i8* @malloc(i64 noundef) CallBase *Call = findCall(*Func, "call"); Trie.buildAndAttachMIBMetadata(Call); - EXPECT_FALSE(Call->hasFnAttr("memprof")); + EXPECT_TRUE(Call->hasFnAttr("memprof")); + EXPECT_EQ(Call->getFnAttr("memprof").getValueAsString(), "ambiguous"); EXPECT_TRUE(Call->hasMetadata(LLVMContext::MD_memprof)); MDNode *MemProfMD = Call->getMetadata(LLVMContext::MD_memprof); ASSERT_EQ(MemProfMD->getNumOperands(), 2u); @@ -463,7 +467,8 @@ declare dso_local noalias noundef i8* @malloc(i64 noundef) ASSERT_NE(Call, nullptr); Trie.buildAndAttachMIBMetadata(Call); - EXPECT_FALSE(Call->hasFnAttr("memprof")); + EXPECT_TRUE(Call->hasFnAttr("memprof")); + EXPECT_EQ(Call->getFnAttr("memprof").getValueAsString(), "ambiguous"); EXPECT_TRUE(Call->hasMetadata(LLVMContext::MD_memprof)); MDNode *MemProfMD = Call->getMetadata(LLVMContext::MD_memprof); EXPECT_THAT(MemProfMD, MemprofMetadataEquals(ExpectedVals)); @@ -536,7 +541,8 @@ declare dso_local noalias noundef i8* @malloc(i64 noundef) // Restore original option value. MemProfKeepAllNotColdContexts = OrigMemProfKeepAllNotColdContexts; - EXPECT_FALSE(Call->hasFnAttr("memprof")); + EXPECT_TRUE(Call->hasFnAttr("memprof")); + EXPECT_EQ(Call->getFnAttr("memprof").getValueAsString(), "ambiguous"); EXPECT_TRUE(Call->hasMetadata(LLVMContext::MD_memprof)); MDNode *MemProfMD = Call->getMetadata(LLVMContext::MD_memprof); EXPECT_THAT(MemProfMD, MemprofMetadataEquals(ExpectedVals)); @@ -664,7 +670,8 @@ declare dso_local noalias noundef i8* @malloc(i64 noundef) // The hot allocations will be converted to NotCold and pruned as they // are unnecessary to determine how to clone the cold allocation. - EXPECT_FALSE(Call->hasFnAttr("memprof")); + EXPECT_TRUE(Call->hasFnAttr("memprof")); + EXPECT_EQ(Call->getFnAttr("memprof").getValueAsString(), "ambiguous"); EXPECT_TRUE(Call->hasMetadata(LLVMContext::MD_memprof)); MemProfMD = Call->getMetadata(LLVMContext::MD_memprof); ASSERT_EQ(MemProfMD->getNumOperands(), 2u); |