[Driver] Accept the -fno-shrink-wrap option for GCC compatibility

[sdardis]

Summary

As reported in GCC bug #86069, LLVM currently provokes a bug in GCC where
objects compiled for MIPS with PIC and optimizations where shrink wrapped
functions can attempt to access the global pointer's spill slot before the global
pointer is spilled. This bug is present in GCC since at least 4.9.2, and affects the
AArch64 backend when LLVM is built for a MIPS host.

However, setting CMAKE_C_FLAGS and CMAKE_CXX_FLAGS affects the compiler-rt
tests which rely on the just built clang rather than the host or cross GCC.
Since Clang doesn't support this flag, so the tests spuriously fail.

Unfortunately, not all targets support shrink wrapping in LLVM at this time,
so providing -fshrink-wrap would expose users to any number of wrong-codegen
issues.

For targets that do support shrink wrapping, this option performs as expected.

[efriedma]

Is this something which is actually useful to control? From your description, you want to add the flag to clang not because you actually want to use it, but just because you can't figure out how to pass the right flags to your clang build.

If it is useful, it should be implemented as a function attribute, not a global flag.

[sdardis]

Modify implementation to produce a function attribute.

[sdardis]

In D47829#1124040, @efriedma wrote:

Is this something which is actually useful to control? From your description, you want to add the flag to clang not because you actually want to use it, but just because you can't figure out how to pass the right flags to your clang build.

If it is useful, it should be implemented as a function attribute, not a global flag.

This is useful to control in the sense of being able to opt-in or opt-out of the shrink wrapping pass for clang, especially in the cases of where we might discover a triggerable bug in a released version of clang.

Currently, the compiler-rt sanitizer tests make use of the CMAKE_C_FLAGS and CMAKE_CXX_FLAGS when generating the test objects. Working around this would require stripping the incompatible arguments from those variables when building the test objects. This seems to me to worse choice, as I'm also posting a patch to automatically add -fno-shrink-wrap when building LLVM with GCC for MIPS (D48069).

include/clang/Driver/Options.td

def fno_rtti : Flag<["-"], "fno-rtti">, Group<f_Group>, Flags<[CC1Option]>,
   HelpText<"Disable generation of rtti information">;
 def fno_rtti_data : Flag<["-"], "fno-rtti-data">, Group<f_Group>, Flags<[CC1Option]>,
   HelpText<"Control emission of RTTI data">;
 def fno_short_enums : Flag<["-"], "fno-short-enums">, Group<f_Group>;
 def fno_show_column : Flag<["-"], "fno-show-column">, Group<f_Group>, Flags<[CC1Option]>,
   HelpText<"Do not include column number on diagnostics">;
 def fno_show_source_location : Flag<["-"], "fno-show-source-location">, Group<f_Group>,
   Flags<[CC1Option]>, HelpText<"Do not include source location information with diagnostics">;
+def fno_shrink_wrap : Flag<["-"], "fno-shrink-wrap">, Group<f_Group>,
+  Flags<[CC1Option]>, HelpText<"Disable shrink wrapping">;
 def fdiagnostics_absolute_paths : Flag<["-"], "fdiagnostics-absolute-paths">, Group<f_Group>,
   Flags<[CC1Option, CoreOption]>, HelpText<"Print absolute paths in diagnostics">;
 def fno_spell_checking : Flag<["-"], "fno-spell-checking">, Group<f_Group>,
   Flags<[CC1Option]>, HelpText<"Disable spell-checking">;
 def fno_stack_protector : Flag<["-"], "fno-stack-protector">, Group<f_Group>,
   HelpText<"Disable the use of stack protectors">;
 def fno_strict_aliasing : Flag<["-"], "fno-strict-aliasing">, Group<f_Group>,
   Flags<[DriverOption, CoreOption]>;

include/clang/Frontend/CodeGenOptions.def

 CODEGENOPT(SimplifyLibCalls  , 1, 1) ///< Set when -fbuiltin is enabled.
 CODEGENOPT(SoftFloat         , 1, 0) ///< -soft-float.
 CODEGENOPT(FineGrainedBitfieldAccesses, 1, 0) ///< Enable fine-grained bitfield accesses.
 CODEGENOPT(StrictEnums       , 1, 0) ///< Optimize based on strict enum definition.
 CODEGENOPT(StrictVTablePointers, 1, 0) ///< Optimize based on the strict vtable pointers
 CODEGENOPT(TimePasses        , 1, 0) ///< Set when -ftime-report is enabled.
 CODEGENOPT(UnrollLoops       , 1, 0) ///< Control whether loops are unrolled.
 CODEGENOPT(RerollLoops       , 1, 0) ///< Control whether loops are rerolled.
+CODEGENOPT(NoShrinkWrap      , 1, 0) ///< Set when -fno-shrink-wrap is enabled.
 CODEGENOPT(NoUseJumpTables   , 1, 0) ///< Set when -fno-jump-tables is enabled.
 CODEGENOPT(UnsafeFPMath      , 1, 0) ///< Allow unsafe floating point optzns.
 CODEGENOPT(UnwindTables      , 1, 0) ///< Emit unwind tables.
 CODEGENOPT(VectorizeLoop     , 1, 0) ///< Run loop vectorizer.
 CODEGENOPT(VectorizeSLP      , 1, 0) ///< Run SLP vectorizer.
 CODEGENOPT(ProfileSampleAccurate, 1, 0) ///< Sample profile is accurate.
 
   /// Attempt to use register sized accesses to bit-fields in structures, when

lib/CodeGen/CodeGenFunction.cpp

if (D && InstrumentXray) {
     } else {
       if (!CGM.imbueXRayAttrs(Fn, Loc))
         Fn->addFnAttr(
             "xray-instruction-threshold",
             llvm::itostr(CGM.getCodeGenOpts().XRayInstructionThreshold));
     }
   }
 
+  // Add no-shrink-wrap value.
+  if (CGM.getCodeGenOpts().NoShrinkWrap)
+    Fn->addFnAttr("no-shrink-wrap",
+                  llvm::toStringRef(CGM.getCodeGenOpts().NoShrinkWrap));
+
   // Add no-jump-tables value.
   Fn->addFnAttr("no-jump-tables",
                 llvm::toStringRef(CGM.getCodeGenOpts().NoUseJumpTables));
 
   // Add profile-sample-accurate value.
   if (CGM.getCodeGenOpts().ProfileSampleAccurate)
     Fn->addFnAttr("profile-sample-accurate");
 

lib/Driver/ToolChains/Clang.cpp

#endif
 
   if (Arg *A = Args.getLastArg(options::OPT_ftrapv_handler_EQ)) {
     CmdArgs.push_back("-ftrapv-handler");
     CmdArgs.push_back(A->getValue());
   }
 
   Args.AddLastArg(CmdArgs, options::OPT_ftrap_function_EQ);
 
+  if (Args.hasArg(options::OPT_fno_shrink_wrap))
+    CmdArgs.push_back("-fno-shrink-wrap");
+
   // -fno-strict-overflow implies -fwrapv if it isn't disabled, but
   // -fstrict-overflow won't turn off an explicitly enabled -fwrapv.
   if (Arg *A = Args.getLastArg(options::OPT_fwrapv, options::OPT_fno_wrapv)) {
     if (A->getOption().matches(options::OPT_fwrapv))
       CmdArgs.push_back("-fwrapv");
   } else if (Arg *A = Args.getLastArg(options::OPT_fstrict_overflow,
                                       options::OPT_fno_strict_overflow)) {
     if (A->getOption().matches(options::OPT_fno_strict_overflow))

lib/Frontend/CompilerInvocation.cpp

Opts.DataSections = Args.hasFlag(OPT_fdata_sections,
                                    OPT_fno_data_sections, false);
   Opts.StackSizeSection =
       Args.hasFlag(OPT_fstack_size_section, OPT_fno_stack_size_section, false);
   Opts.UniqueSectionNames = Args.hasFlag(OPT_funique_section_names,
                                          OPT_fno_unique_section_names, true);
 
   Opts.MergeFunctions = Args.hasArg(OPT_fmerge_functions);
 
+  Opts.NoShrinkWrap = Args.hasArg(OPT_fno_shrink_wrap);
+
   Opts.NoUseJumpTables = Args.hasArg(OPT_fno_jump_tables);
 
   Opts.ProfileSampleAccurate = Args.hasArg(OPT_fprofile_sample_accurate);
 
   Opts.PrepareForLTO = Args.hasArg(OPT_flto, OPT_flto_EQ);
   Opts.EmitSummaryIndex = false;
   if (Arg *A = Args.getLastArg(OPT_flto_EQ)) {
     StringRef S = A->getValue();

test/CodeGen/noshrinkwrapping.c

+// RUN: %clang_cc1 -S -fno-shrink-wrapping %s -emit-llvm -o - | FileCheck %s
+
+// CHECK-LABEL: main
+// CHECK: attributes #0 = {{.*}}"no-shrink-wrapping"="true"{{.*}}
+
+int main() {
+  return 0;
+}