[OpenCL][AMDGPU] Add support for -cl-denorms-are-zero

[yaxunl]

Summary

Adjust target features for amdgcn target when -cl-denorms-are-zero is set.

Denormal support is controlled by feature strings fp32-denormals fp64-denormals in amdgcn target. If -cl-denorms-are-zero is not set and the command line does not set fp32/64-denormals feature string, +fp32-denormals +fp64-denormals will be on for GPU's supporting them.

A new virtual function virtual void TargetInfo::adjustTargetOptions(const CodeGenOptions &CGOpts, TargetOptions &TargetOpts) const is introduced to allow adjusting target option by codegen option.

[Anastasia]

LGTM!

cfe/trunk/include/clang/Basic/TargetInfo.h

 
 namespace llvm {
 struct fltSemantics;
 }
 
 namespace clang {
 class DiagnosticsEngine;
 class LangOptions;
+class CodeGenOptions;
 class MacroBuilder;
 class SourceLocation;
 class SourceManager;
 
 namespace Builtin { struct Info; }
 
 /// \brief Exposes information about the current target.
 ///
public:
   }
 
   /// \brief Set forced language options.
   ///
   /// Apply changes to the target information with respect to certain
   /// language options which change the target configuration.
   virtual void adjust(const LangOptions &Opts);
 
+  /// \brief Adjust target options based on codegen options.
+  virtual void adjustTargetOptions(const CodeGenOptions &CGOpts,
+                                   TargetOptions &TargetOpts) const {}
+
   /// \brief Initialize the map with the default set of target features for the
   /// CPU this should include all legal feature strings on the target.
   ///
   /// \return False on error (invalid features).
   virtual bool initFeatureMap(llvm::StringMap<bool> &Features,
                               DiagnosticsEngine &Diags, StringRef CPU,
                               const std::vector<std::string> &FeatureVec) const;
 

cfe/trunk/include/clang/Frontend/CodeGenOptions.def

 CODEGENOPT(EnableSegmentedStacks , 1, 0) ///< Set when -fsplit-stack is enabled.
 CODEGENOPT(NoImplicitFloat   , 1, 0) ///< Set when -mno-implicit-float is enabled.
 CODEGENOPT(NoInfsFPMath      , 1, 0) ///< Assume FP arguments, results not +-Inf.
 CODEGENOPT(NoSignedZeros     , 1, 0) ///< Allow ignoring the signedness of FP zero
 CODEGENOPT(ReciprocalMath    , 1, 0) ///< Allow FP divisions to be reassociated.
 CODEGENOPT(NoInline          , 1, 0) ///< Set when -fno-inline is enabled.
                                      ///< Disables use of the inline keyword.
 CODEGENOPT(NoNaNsFPMath      , 1, 0) ///< Assume FP arguments, results not NaN.
+CODEGENOPT(FlushDenorm       , 1, 0) ///< Allow FP denorm numbers to be flushed to zero
 CODEGENOPT(NoZeroInitializedInBSS , 1, 0) ///< -fno-zero-initialized-in-bss.
 /// \brief Method of Objective-C dispatch to use.
 ENUM_CODEGENOPT(ObjCDispatchMethod, ObjCDispatchMethodKind, 2, Legacy)
 CODEGENOPT(OmitLeafFramePointer , 1, 0) ///< Set when -momit-leaf-frame-pointer is
                                         ///< enabled.
 VALUE_CODEGENOPT(OptimizationLevel, 2, 0) ///< The -O[0-3] option specified.
 VALUE_CODEGENOPT(OptimizeSize, 2, 0) ///< If -Os (==1) or -Oz (==2) is specified.
 

cfe/trunk/lib/Basic/Targets.cpp

 #include "clang/Basic/Cuda.h"
 #include "clang/Basic/Diagnostic.h"
 #include "clang/Basic/LangOptions.h"
 #include "clang/Basic/MacroBuilder.h"
 #include "clang/Basic/TargetBuiltins.h"
 #include "clang/Basic/TargetInfo.h"
 #include "clang/Basic/TargetOptions.h"
 #include "clang/Basic/Version.h"
+#include "clang/Frontend/CodeGenOptions.h"
 #include "llvm/ADT/APFloat.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/ADT/StringRef.h"
 #include "llvm/ADT/StringSwitch.h"
 #include "llvm/ADT/Triple.h"
 #include "llvm/MC/MCSectionMachO.h"
 #include "llvm/Support/ErrorHandling.h"
enum GPUKind {
     GK_SOUTHERN_ISLANDS,
     GK_SEA_ISLANDS,
     GK_VOLCANIC_ISLANDS
   } GPU;
 
   bool hasFP64:1;
   bool hasFMAF:1;
   bool hasLDEXPF:1;
+  bool hasDenormSupport:1;
 
   static bool isAMDGCN(const llvm::Triple &TT) {
     return TT.getArch() == llvm::Triple::amdgcn;
   }
 
 public:
-  AMDGPUTargetInfo(const llvm::Triple &Triple, const TargetOptions &)
+  AMDGPUTargetInfo(const llvm::Triple &Triple, const TargetOptions &Opts)
     : TargetInfo(Triple) ,
       GPU(isAMDGCN(Triple) ? GK_SOUTHERN_ISLANDS : GK_R600),
       hasFP64(false),
       hasFMAF(false),
-      hasLDEXPF(false) {
+      hasLDEXPF(false),
+      hasDenormSupport(false){
     if (getTriple().getArch() == llvm::Triple::amdgcn) {
       hasFP64 = true;
       hasFMAF = true;
       hasLDEXPF = true;
     }
+    if (Opts.CPU == "fiji")
+      hasDenormSupport = true;
 
     resetDataLayout(getTriple().getArch() == llvm::Triple::amdgcn ?
                     DataLayoutStringSI : DataLayoutStringR600);
 
     AddrSpaceMap = &AMDGPUAddrSpaceMap;
     UseAddrSpaceMapMangling = true;
   }
 
bool validateAsmConstraint(const char *&Name,
     }
     return false;
   }
 
   bool initFeatureMap(llvm::StringMap<bool> &Features,
                       DiagnosticsEngine &Diags, StringRef CPU,
                       const std::vector<std::string> &FeatureVec) const override;
 
+  void adjustTargetOptions(const CodeGenOptions &CGOpts,
+                           TargetOptions &TargetOpts) const override {
+    if (!hasDenormSupport)
+      return;
+    bool hasFP32Denormals = false;
+    bool hasFP64Denormals = false;
+    for (auto &I : TargetOpts.FeaturesAsWritten) {
+      if (I == "+fp32-denormals" || I == "-fp32-denormals")
+        hasFP32Denormals = true;
+      if (I == "+fp64-denormals" || I == "-fp64-denormals")
+        hasFP64Denormals = true;
+    }
+    if (!hasFP32Denormals)
+      TargetOpts.Features.push_back((Twine(CGOpts.FlushDenorm ? '-' : '+') +
+                                     Twine("fp32-denormals")).str());
+    if (!hasFP64Denormals && hasFP64)
+      TargetOpts.Features.push_back((Twine(CGOpts.FlushDenorm ? '-' : '+') +
+                                     Twine("fp64-denormals")).str());
+  }
+
   ArrayRef<Builtin::Info> getTargetBuiltins() const override {
     return llvm::makeArrayRef(BuiltinInfo,
                         clang::AMDGPU::LastTSBuiltin - Builtin::FirstTSBuiltin);
   }
 
   void getTargetDefines(const LangOptions &Opts,
                         MacroBuilder &Builder) const override {
     if (getTriple().getArch() == llvm::Triple::amdgcn)

cfe/trunk/lib/Frontend/CompilerInstance.cpp

bool CompilerInstance::ExecuteAction(FrontendAction &Act) {
   }
 
   // Inform the target of the language options.
   //
   // FIXME: We shouldn't need to do this, the target should be immutable once
   // created. This complexity should be lifted elsewhere.
   getTarget().adjust(getLangOpts());
 
+  // Adjust target options based on codegen options.
+  getTarget().adjustTargetOptions(getCodeGenOpts(), getTargetOpts());
+
   // rewriter project will change target built-in bool type from its default. 
   if (getFrontendOpts().ProgramAction == frontend::RewriteObjC)
     getTarget().noSignedCharForObjCBool();
 
   // Validate/process some options.
   if (getHeaderSearchOpts().Verbose)
     OS << "clang -cc1 version " CLANG_VERSION_STRING
        << " based upon " << BACKEND_PACKAGE_STRING

cfe/trunk/lib/Frontend/CompilerInvocation.cpp

Opts.NoInfsFPMath = (Args.hasArg(OPT_menable_no_infinities) ||
                        Args.hasArg(OPT_cl_finite_math_only) ||
                        Args.hasArg(OPT_cl_fast_relaxed_math));
   Opts.NoNaNsFPMath = (Args.hasArg(OPT_menable_no_nans) ||
                        Args.hasArg(OPT_cl_unsafe_math_optimizations) ||
                        Args.hasArg(OPT_cl_finite_math_only) ||
                        Args.hasArg(OPT_cl_fast_relaxed_math));
   Opts.NoSignedZeros = (Args.hasArg(OPT_fno_signed_zeros) ||
                         Args.hasArg(OPT_cl_no_signed_zeros));
+  Opts.FlushDenorm = Args.hasArg(OPT_cl_denorms_are_zero);
   Opts.ReciprocalMath = Args.hasArg(OPT_freciprocal_math);
   Opts.NoZeroInitializedInBSS = Args.hasArg(OPT_mno_zero_initialized_in_bss);
   Opts.BackendOptions = Args.getAllArgValues(OPT_backend_option);
   Opts.NumRegisterParameters = getLastArgIntValue(Args, OPT_mregparm, 0, Diags);
   Opts.NoExecStack = Args.hasArg(OPT_mno_exec_stack);
   Opts.FatalWarnings = Args.hasArg(OPT_massembler_fatal_warnings);
   Opts.EnableSegmentedStacks = Args.hasArg(OPT_split_stacks);
   Opts.RelaxAll = Args.hasArg(OPT_mrelax_all);

cfe/trunk/test/CodeGenOpenCL/denorms-are-zero.cl

 // RUN: %clang_cc1 -S -cl-denorms-are-zero -o - %s 2>&1
+// RUN: %clang_cc1 -emit-llvm -cl-denorms-are-zero -o - -triple amdgcn--amdhsa -target-cpu fiji %s | FileCheck %s
+// RUN: %clang_cc1 -emit-llvm -o - -triple amdgcn--amdhsa -target-cpu fiji %s | FileCheck %s --check-prefix=CHECK-DENORM
 
-// This test just checks that the -cl-denorms-are-zero argument is accepted
+// For non-amdgcn targets, this test just checks that the -cl-denorms-are-zero argument is accepted
 // by clang.  This option is currently a no-op, which is allowed by the
 // OpenCL specification.
+
+// CHECK-DENORM-LABEL: define void @f()
+// CHECK-DENORM: attributes #{{[0-9]*}} = {{{[^}]*}} "target-features"="{{[^"]*}}+fp32-denormals,+fp64-denormals{{[^"]*}}"
+// CHECK-LABEL: define void @f()
+// CHECK-NOT: attributes #{{[0-9]*}} = {{{[^}]*}} "target-features"="{{[^"]*}}+fp32-denormals,+fp64-denormals{{[^"]*}}"
+void f() {}