[Clang] Enable -fpointer-tbaa by default.

Support for more precise TBAA metadata has been added a while ago (behind the -fpointer-tbaa flag). The more precise TBAA metadata allows treating accesses of different pointer types as no-alias. This helps to remove more redundant loads and stores in a number of workloads. Some highlights on the impact across llvm-test-suite's MultiSource, SPEC2006 & SPEC2017 include: * +2% more NoAlias results for memory access * +4% more loops vectorized * +3% more stores removed by DSE. This closes a relatively big gap to GCC, which has been supporting disambiguating based on pointer types for a long time. (https://clang.godbolt.org/z/K7Wbhrz4q) Pointer-TBAA support for pointers to builtin types has been added in llvm#76612. Support for user-defined types has been added in llvm#110569. There are 2 pending PRs with bug fixes for special cases uncovered during some of my testing: * llvm#116991 * llvm#116596
fhahn · Dec 3, 2024 · 1815f81 · 1815f81
1 parent 56ab56c
commit 1815f81
Show file tree

Hide file tree

Showing 11 changed files with 459 additions and 439 deletions.
diff --git a/clang/docs/ReleaseNotes.rst b/clang/docs/ReleaseNotes.rst
@@ -240,6 +240,10 @@ C++ Language Changes
 - The builtin type alias ``__builtin_common_type`` has been added to improve the
   performance of ``std::common_type``.
 
+- Clang now emits distinct type-based alias analysis tags for incompatible
+  pointers by default, enabling more powerful alias analysis when accessing
+  pointer types. The new default behavior can be disabled using ``-fno-pointer-tbaa``.
+
 C++2c Feature Support
 ^^^^^^^^^^^^^^^^^^^^^
 

diff --git a/clang/include/clang/Basic/CodeGenOptions.def b/clang/include/clang/Basic/CodeGenOptions.def
@@ -242,7 +242,7 @@ ENUM_CODEGENOPT(StructReturnConvention, StructReturnConventionKind, 2, SRCK_Defa
 
 CODEGENOPT(RelaxAll          , 1, 0) ///< Relax all machine code instructions.
 CODEGENOPT(RelaxedAliasing   , 1, 0) ///< Set when -fno-strict-aliasing is enabled.
-CODEGENOPT(PointerTBAA, 1, 0)        ///< Whether or not to use distinct TBAA tags for pointers.
+CODEGENOPT(PointerTBAA       , 1, 1) ///< Whether or not to use distinct TBAA tags for pointers.
 CODEGENOPT(StructPathTBAA    , 1, 0) ///< Whether or not to use struct-path TBAA.
 CODEGENOPT(NewStructPathTBAA , 1, 0) ///< Whether or not to use enhanced struct-path TBAA.
 CODEGENOPT(SaveTempLabels    , 1, 0) ///< Save temporary labels.

diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
@@ -7387,9 +7387,12 @@ def fuse_register_sized_bitfield_access: Flag<["-"], "fuse-register-sized-bitfie
 def relaxed_aliasing : Flag<["-"], "relaxed-aliasing">,
   HelpText<"Turn off Type Based Alias Analysis">,
   MarshallingInfoFlag<CodeGenOpts<"RelaxedAliasing">>;
-def pointer_tbaa: Flag<["-"], "pointer-tbaa">,
-  HelpText<"Turn on Type Based Alias Analysis for pointer accesses">,
-  MarshallingInfoFlag<CodeGenOpts<"PointerTBAA">>;
+defm pointer_tbaa: BoolOption<"", "pointer-tbaa", CodeGenOpts<"PointerTBAA">, 
+DefaultTrue,
+  PosFlag<SetTrue, [], [ClangOption], "Enable">,
+  NegFlag<SetFalse, [], [ClangOption], "Disable">,
+  BothFlags<[], [ClangOption], " that single precision floating-point divide and sqrt used in ">>
+  ;
 def no_struct_path_tbaa : Flag<["-"], "no-struct-path-tbaa">,
   HelpText<"Turn off struct-path aware Type Based Alias Analysis">,
   MarshallingInfoNegativeFlag<CodeGenOpts<"StructPathTBAA">>;

diff --git a/clang/lib/Driver/ToolChains/Clang.cpp b/clang/lib/Driver/ToolChains/Clang.cpp
@@ -5937,9 +5937,9 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   if (!Args.hasFlag(options::OPT_fstrict_aliasing, StrictAliasingAliasOption,
                     options::OPT_fno_strict_aliasing, !IsWindowsMSVC))
     CmdArgs.push_back("-relaxed-aliasing");
-  if (Args.hasFlag(options::OPT_fpointer_tbaa, options::OPT_fno_pointer_tbaa,
+  if (Args.hasFlag(options::OPT_fno_pointer_tbaa, options::OPT_fpointer_tbaa,
                    false))
-    CmdArgs.push_back("-pointer-tbaa");
+    CmdArgs.push_back("-no-pointer-tbaa");
   if (!Args.hasFlag(options::OPT_fstruct_path_tbaa,
                     options::OPT_fno_struct_path_tbaa, true))
     CmdArgs.push_back("-no-struct-path-tbaa");

diff --git a/clang/test/CodeGen/attr-counted-by.c b/clang/test/CodeGen/attr-counted-by.c
diff --git a/clang/test/CodeGen/tbaa-pointers.c b/clang/test/CodeGen/tbaa-pointers.c
diff --git a/clang/test/CodeGen/tbaa-reference.cpp b/clang/test/CodeGen/tbaa-reference.cpp
@@ -1,7 +1,7 @@
-// RUN: %clang_cc1 -triple x86_64-linux -O1 -disable-llvm-passes %s -emit-llvm -o - | FileCheck %s  -check-prefixes=CHECK,OLD-PATH
-// RUN: %clang_cc1 -triple x86_64-linux -O1 -disable-llvm-passes -pointer-tbaa %s -emit-llvm -o - | FileCheck %s  -check-prefixes=CHECK,OLD-PATH-POINTER
-// RUN: %clang_cc1 -triple x86_64-linux -O1 -disable-llvm-passes %s -emit-llvm -new-struct-path-tbaa -o - | FileCheck %s -check-prefixes=CHECK,NEW-PATH
-// RUN: %clang_cc1 -triple x86_64-linux -O1 -disable-llvm-passes %s -pointer-tbaa -emit-llvm -new-struct-path-tbaa -o - | FileCheck %s -check-prefixes=CHECK,NEW-PATH-POINTER
+// RUN: %clang_cc1 -triple x86_64-linux -O1 -disable-llvm-passes -no-pointer-tbaa %s -emit-llvm -o - | FileCheck %s  -check-prefixes=CHECK,OLD-PATH
+// RUN: %clang_cc1 -triple x86_64-linux -O1 -disable-llvm-passes %s -emit-llvm -o - | FileCheck %s  -check-prefixes=CHECK,OLD-PATH-POINTER
+// RUN: %clang_cc1 -triple x86_64-linux -O1 -disable-llvm-passes %s -no-pointer-tbaa -emit-llvm -new-struct-path-tbaa -o - | FileCheck %s -check-prefixes=CHECK,NEW-PATH
+// RUN: %clang_cc1 -triple x86_64-linux -O1 -disable-llvm-passes %s -emit-llvm -new-struct-path-tbaa -o - | FileCheck %s -check-prefixes=CHECK,NEW-PATH-POINTER
 //
 // Check that we generate correct TBAA information for reference accesses.
 

diff --git a/clang/test/CodeGenCXX/template-instantiation.cpp b/clang/test/CodeGenCXX/template-instantiation.cpp
@@ -17,6 +17,7 @@
 
 // CHECK2-NOT: _ZTVN5test31SIiEE
 // CHECK2-NOT: _ZTSN5test31SIiEE
+// CHECK2:  !{!"p1 _ZTSN5test31SIiEE",
 
 // CHECK-LABEL: define linkonce_odr void @_ZN5test21CIiEC1Ev(ptr {{[^,]*}} %this) unnamed_addr
 // CHECK-LABEL: define linkonce_odr void @_ZN5test21CIiE6foobarIdEEvT_(

diff --git a/clang/test/CodeGenOpenCL/amdgpu-enqueue-kernel.cl b/clang/test/CodeGenOpenCL/amdgpu-enqueue-kernel.cl
diff --git a/clang/test/OpenMP/nvptx_target_parallel_reduction_codegen_tbaa_PR46146.cpp b/clang/test/OpenMP/nvptx_target_parallel_reduction_codegen_tbaa_PR46146.cpp
diff --git a/clang/unittests/CodeGen/TBAAMetadataTest.cpp b/clang/unittests/CodeGen/TBAAMetadataTest.cpp
@@ -39,6 +39,11 @@ auto OmnipotentCharC = MMTuple(
   MConstInt(0, 64)
 );
 
+auto AnyPtr = MMTuple(
+  MMString("any pointer"),
+  OmnipotentCharC,
+  MConstInt(0, 64)
+);
 
 auto OmnipotentCharCXX = MMTuple(
   MMString("omnipotent char"),
@@ -116,8 +121,8 @@ TEST(TBAAMetadataTest, BasicTypes) {
         MValType(PointerType::getUnqual(Compiler.Context)),
         MMTuple(
           MMTuple(
-            MMString("any pointer"),
-            OmnipotentCharC,
+            MMString("p1 void"),
+            AnyPtr,
             MConstInt(0)),
           MSameAs(0),
           MConstInt(0))));
@@ -128,8 +133,8 @@ TEST(TBAAMetadataTest, BasicTypes) {
         MValType(PointerType::getUnqual(Compiler.Context)),
         MMTuple(
           MMTuple(
-            MMString("any pointer"),
-            OmnipotentCharC,
+            MMString("p1 int"),
+            AnyPtr,
             MConstInt(0)),
           MSameAs(0),
           MConstInt(0))));