[PowerPC] Support -mabi=ieeelongdouble and -mabi=ibmlongdouble

[MaskRay]

Summary

gcc PowerPC supports 3 representations of long double:

• -mlong-double-64
  
  long double has the same representation of double but is mangled as e. In clang, this is the default on AIX, FreeBSD and Linux musl.

• -mlong-double-128
  
  long double has 2 possible 128-bit floating point representations:
  
  + -mabi=ibmlongdouble IBM extended double format. Mangled as g In clang, this is the default on Linux glibc. + -mabi=ieeelongdouble IEEE 754 quadruple-precision format. Mangled as u9__ieee128 (U10__float128 before gcc 8.2) This is currently unavailable.

This patch adds -mabi=ibmlongdouble and -mabi=ieeelongdouble, and thus
makes the IEEE 754 quadruple-precision long double available for
languages supported by clang.

[MaskRay]

Update description.

Fix the mangled code: U10float128 -> u9ieee128

[hfinkel]

Ah, fun with overloaded, legacy command-line options...

lib/Driver/ToolChains/Clang.cpp

1836

This seems like an unintentional behavior change on top of the existing behavior (which may have also been not quite right). As best I can tell, we're trying to set ABIName equal to the last ABI type specified, unless that type is "altivec", in which case the ABI name should be its default value (either a nullptr or something like "elfv2"). With this change, we'll now take the first ABI name seen, not the last (as we'll get SeenOther to true after the first name).

Maybe we should loop over the list multiple times, once to get the long-double format, once to get the basic ABI name (where we save the default, and then if the last name found is "altivec", then reset to the default)?

[MaskRay]

lib/Driver/ToolChains/Clang.cpp

1836

My understanding (could be wrong) is that "ibmlongdouble", "ieeelongdouble", and "altivec" cannot be ABIName, so I write it this way (filtered_reverse + SeenLongDouble + SeenOther).

e.g. with -mabi=elfv2 -mabi=ieeelongdouble, ABIName should be "elfv2". With -mabi=elfv1 -mabi=ieeelongdouble, ABIName should be "elfv1" (no local entry). This matches the behavior I observed from powerpc64le-linux-gnu-gcc.

Before the change, ABIName is "ieeelongdouble". I think that is not desired.

[MaskRay]

Add another test: -mabi=elfv2 -mabi=elfv1 -mabi=ibmlongdouble -mabi=ieeelongdouble

Try fixing the markdown in the description

[MaskRay]

Make -mabi=ieeelongdouble a pure CC1 option so that we can simplify the -mabi= loop

[MaskRay]

Woohoo, this is the last :)

[hfinkel]

LGTM

include/clang/Basic/LangOptions.def

 COMPATIBLE_LANGOPT(OptimizeSize      , 1, 0, "__OPTIMIZE_SIZE__ predefined macro")
 COMPATIBLE_LANGOPT(Static            , 1, 0, "__STATIC__ predefined macro (as opposed to __DYNAMIC__)")
 VALUE_LANGOPT(PackStruct  , 32, 0,
               "default struct packing maximum alignment")
 VALUE_LANGOPT(MaxTypeAlign  , 32, 0,
               "default maximum alignment for types")
 VALUE_LANGOPT(AlignDouble            , 1, 0, "Controls if doubles should be aligned to 8 bytes (x86 only)")
 VALUE_LANGOPT(LongDoubleSize        , 32, 0, "width of long double")
+LANGOPT(PPCIEEELongDouble            , 1, 0, "use IEEE 754 quadruple-precision for long double")
 COMPATIBLE_VALUE_LANGOPT(PICLevel    , 2, 0, "__PIC__ level")
 COMPATIBLE_VALUE_LANGOPT(PIE         , 1, 0, "is pie")
 LANGOPT(ROPI                         , 1, 0, "Read-only position independence")
 LANGOPT(RWPI                         , 1, 0, "Read-write position independence")
 COMPATIBLE_LANGOPT(GNUInline         , 1, 0, "GNU inline semantics")
 COMPATIBLE_LANGOPT(NoInlineDefine    , 1, 0, "__NO_INLINE__ predefined macro")
 COMPATIBLE_LANGOPT(Deprecated        , 1, 0, "__DEPRECATED predefined macro")
 COMPATIBLE_LANGOPT(FastMath          , 1, 0, "fast FP math optimizations, and __FAST_MATH__ predefined macro")

include/clang/Driver/CC1Options.td

def menable_no_infinities : Flag<["-"], "menable-no-infs">,
   HelpText<"Allow optimization to assume there are no infinities.">;
 def menable_no_nans : Flag<["-"], "menable-no-nans">,
   HelpText<"Allow optimization to assume there are no NaNs.">;
 def menable_unsafe_fp_math : Flag<["-"], "menable-unsafe-fp-math">,
   HelpText<"Allow unsafe floating-point math optimizations which may decrease "
            "precision">;
 def mreassociate : Flag<["-"], "mreassociate">,
   HelpText<"Allow reassociation transformations for floating-point instructions">;
+def mabi_EQ_ieeelongdouble : Flag<["-"], "mabi=ieeelongdouble">,
+  HelpText<"Use IEEE 754 quadruple-precision for long double">;
 def mfloat_abi : Separate<["-"], "mfloat-abi">,
   HelpText<"The float ABI to use">;
 def mtp : Separate<["-"], "mtp">,
   HelpText<"Mode for reading thread pointer">;
 def mlimit_float_precision : Separate<["-"], "mlimit-float-precision">,
   HelpText<"Limit float precision to the given value">;
 def split_stacks : Flag<["-"], "split-stacks">,
   HelpText<"Try to use a split stack if possible.">;

lib/Basic/Targets/PPC.cpp

void PPCTargetInfo::fillValidCPUList(SmallVectorImpl<StringRef> &Values) const {
   Values.append(std::begin(ValidCPUNames), std::end(ValidCPUNames));
 }
 
 void PPCTargetInfo::adjust(LangOptions &Opts) {
   if (HasAltivec)
     Opts.AltiVec = 1;
   TargetInfo::adjust(Opts);
   if (LongDoubleFormat != &llvm::APFloat::IEEEdouble())
-    LongDoubleFormat = &llvm::APFloat::PPCDoubleDouble();
+    LongDoubleFormat = Opts.PPCIEEELongDouble
+                           ? &llvm::APFloat::IEEEquad()
+                           : &llvm::APFloat::PPCDoubleDouble();
 }
 
 ArrayRef<Builtin::Info> PPCTargetInfo::getTargetBuiltins() const {
   return llvm::makeArrayRef(BuiltinInfo, clang::PPC::LastTSBuiltin -
                                              Builtin::FirstTSBuiltin);
 }

lib/Driver/ToolChains/Clang.cpp

if (getToolChain().getTriple().isOSLinux())
     }
     case llvm::Triple::ppc64le:
       ABIName = "elfv2";
       break;
     default:
       break;
     }
 
-  if (Arg *A = Args.getLastArg(options::OPT_mabi_EQ))
+  bool IEEELongDouble = false;
+  for (const Arg *A : Args.filtered(options::OPT_mabi_EQ)) {
+    StringRef V = A->getValue();
+    if (V == "ieeelongdouble")
+      IEEELongDouble = true;
+    else if (V == "ibmlongdouble")
+      IEEELongDouble = false;
+    else if (V != "altivec")
       // The ppc64 linux abis are all "altivec" abis by default. Accept and ignore
       // the option if given as we don't have backend support for any targets
       // that don't use the altivec abi.
-    if (StringRef(A->getValue()) != "altivec")
       ABIName = A->getValue();
+  }
+  if (IEEELongDouble)
+    CmdArgs.push_back("-mabi=ieeelongdouble");
 
   ppc::FloatABI FloatABI =
       ppc::getPPCFloatABI(getToolChain().getDriver(), Args);
 
   if (FloatABI == ppc::FloatABI::Soft) {
     // Floating point operations and argument passing are soft.
     CmdArgs.push_back("-msoft-float");
     CmdArgs.push_back("-mfloat-abi");
     CmdArgs.push_back("soft");
   } else {
     // Floating point operations and argument passing are hard.
     assert(FloatABI == ppc::FloatABI::Hard && "Invalid float abi!");
     CmdArgs.push_back("-mfloat-abi");
     CmdArgs.push_back("hard");
   }
 
   if (ABIName) {
     CmdArgs.push_back("-target-abi");
     CmdArgs.push_back(ABIName);

lib/Frontend/CompilerInvocation.cpp

Opts.EncodeExtendedBlockSig =
     Args.hasArg(OPT_fencode_extended_block_signature);
   Opts.EmitAllDecls = Args.hasArg(OPT_femit_all_decls);
   Opts.PackStruct = getLastArgIntValue(Args, OPT_fpack_struct_EQ, 0, Diags);
   Opts.MaxTypeAlign = getLastArgIntValue(Args, OPT_fmax_type_align_EQ, 0, Diags);
   Opts.AlignDouble = Args.hasArg(OPT_malign_double);
   Opts.LongDoubleSize = Args.hasArg(OPT_mlong_double_128)
                             ? 128
                             : Args.hasArg(OPT_mlong_double_64) ? 64 : 0;
+  Opts.PPCIEEELongDouble = Args.hasArg(OPT_mabi_EQ_ieeelongdouble);
   Opts.PICLevel = getLastArgIntValue(Args, OPT_pic_level, 0, Diags);
   Opts.ROPI = Args.hasArg(OPT_fropi);
   Opts.RWPI = Args.hasArg(OPT_frwpi);
   Opts.PIE = Args.hasArg(OPT_pic_is_pie);
   Opts.Static = Args.hasArg(OPT_static_define);
   Opts.DumpRecordLayoutsSimple = Args.hasArg(OPT_fdump_record_layouts_simple);
   Opts.DumpRecordLayouts = Opts.DumpRecordLayoutsSimple
                         || Args.hasArg(OPT_fdump_record_layouts);

test/CodeGen/ppc64-long-double.cpp

 // RUN: %clang_cc1 -triple powerpc64-linux-musl -emit-llvm -o - %s | \
 // RUN:   FileCheck --check-prefix=FP64 %s
 // RUN: %clang_cc1 -triple powerpc64-linux-gnu -emit-llvm -o - %s -mlong-double-64 | \
 // RUN:   FileCheck --check-prefix=FP64 %s
 
+// musl defaults to -mlong-double-64, so -mlong-double-128 is needed to make
+// -mabi=ieeelongdouble effective.
+// RUN: %clang_cc1 -triple powerpc64-linux-musl -emit-llvm -o - %s -mlong-double-128 \
+// RUN:   -mabi=ieeelongdouble | FileCheck --check-prefix=FP128 %s
+// RUN: %clang_cc1 -triple powerpc64-linux-gnu -emit-llvm -o - %s \
+// RUN:   -mabi=ieeelongdouble | FileCheck --check-prefix=FP128 %s
+
+// IBM extended double is the default.
 // RUN: %clang_cc1 -triple powerpc64-linux-gnu -emit-llvm -o - %s | \
 // RUN:   FileCheck --check-prefix=IBM128 %s
 // RUN: %clang_cc1 -triple powerpc64-linux-musl -emit-llvm -o - -mlong-double-128 %s | \
 // RUN:   FileCheck --check-prefix=IBM128 %s
 
 long double x = 0;
 int size = sizeof(x);
 
 // FP64: @x = global double {{.*}}, align 8
 // FP64: @size = global i32 8
+// FP128: @x = global fp128 {{.*}}, align 16
+// FP128: @size = global i32 16
 // IBM128: @x = global ppc_fp128 {{.*}}, align 16
 // IBM128: @size = global i32 16
 
 long double foo(long double d) { return d; }
 
 // FP64: double @_Z3fooe(double %d)
+// FP128: fp128 @_Z3foou9__ieee128(fp128 %d)
 // IBM128: ppc_fp128 @_Z3foog(ppc_fp128 %d)

test/Driver/ppc-abi.c

 
 // CHECK-ELFv1-PIC: "-mrelocation-model" "pic" "-pic-level" "2"
 // CHECK-ELFv1-PIC: "-target-abi" "elfv1"
 // CHECK-ELFv1-QPX-PIC: "-mrelocation-model" "pic" "-pic-level" "2"
 // CHECK-ELFv1-QPX-PIC: "-target-abi" "elfv1-qpx"
 // CHECK-ELFv2-PIC: "-mrelocation-model" "pic" "-pic-level" "2"
 // CHECK-ELFv2-PIC: "-target-abi" "elfv2"
 
+// Check -mabi=ieeelongdouble is passed through but it does not change -target-abi.
+// RUN: %clang -target powerpc64le-linux-gnu %s -mabi=ieeelongdouble -mabi=elfv1 -### 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHECK-ELFv1-IEEE %s
+// RUN: %clang -target powerpc64le-linux-gnu %s -mabi=elfv1 -mabi=ieeelongdouble -### 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHECK-ELFv1-IEEE %s
+// RUN: %clang -target powerpc64le-linux-gnu %s -mabi=elfv2 -mabi=elfv1 -mabi=ibmlongdouble -mabi=ieeelongdouble -### 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHECK-ELFv1-IEEE %s
+
+// CHECK-ELFv1-IEEE: "-mabi=ieeelongdouble"
+// CHECK-ELFv1-IEEE: "-target-abi" "elfv1"
+
+// Check -mabi=ibmlongdouble is the default.
+// RUN: %clang -target powerpc64le-linux-gnu %s -### 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHECK-ELFv2-IBM128 %s
+// RUN: %clang -target powerpc64le-linux-gnu %s -mabi=ibmlongdouble -### 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHECK-ELFv2-IBM128 %s
 
+// CHECK-ELFv2-IBM128-NOT: "-mabi=ieeelongdouble"
+// CHECK-ELFv2-IBM128: "-target-abi" "elfv2"