Add FMA4 feature flag. Intrinsics coming soon. Also make sse4a feature flag imply...

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@147370 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/clang/Driver/Options.td b/include/clang/Driver/Options.td
index 4dfda54c8e36bfe5ece78a41a7c0548e94b58636..892a972a620d811a4608bab4d41fe29603549494 100644 (file)
--- a/include/clang/Driver/Options.td
+++ b/include/clang/Driver/Options.td
@@ -616,6 +616,7 @@ def mno_lzcnt : Flag<"-mno-lzcnt">, Group<m_x86_Features_Group>;
 def mno_bmi : Flag<"-mno-bmi">, Group<m_x86_Features_Group>;
 def mno_bmi2 : Flag<"-mno-bmi2">, Group<m_x86_Features_Group>;
 def mno_popcnt : Flag<"-mno-popcnt">, Group<m_x86_Features_Group>;
+def mno_fma4 : Flag<"-mno-fma4">, Group<m_x86_Features_Group>;
 
 def mno_thumb : Flag<"-mno-thumb">, Group<m_Group>;
 def marm : Flag<"-marm">, Alias<mno_thumb>;
@@ -645,6 +646,7 @@ def mlzcnt : Flag<"-mlzcnt">, Group<m_x86_Features_Group>;
 def mbmi : Flag<"-mbmi">, Group<m_x86_Features_Group>;
 def mbmi2 : Flag<"-mbmi2">, Group<m_x86_Features_Group>;
 def mpopcnt : Flag<"-mpopcnt">, Group<m_x86_Features_Group>;
+def mfma4 : Flag<"-mfma4">, Group<m_x86_Features_Group>;
 def mthumb : Flag<"-mthumb">, Group<m_Group>;
 def mtune_EQ : Joined<"-mtune=">, Group<m_Group>;
 def multi__module : Flag<"-multi_module">;

diff --git a/lib/Basic/Targets.cpp b/lib/Basic/Targets.cpp
index a768226b40c43a6db66033035e86b2f7a2d438fc..aebb28a68e56b1fd884b816acab963fb86cea66b 100644 (file)
--- a/lib/Basic/Targets.cpp
+++ b/lib/Basic/Targets.cpp
@@ -1199,6 +1199,7 @@ class X86TargetInfo : public TargetInfo {
   bool HasBMI;
   bool HasBMI2;
   bool HasPOPCNT;
+  bool HasFMA4;
 
   /// \brief Enumeration of all of the X86 CPUs supported by Clang.
   ///
@@ -1336,7 +1337,8 @@ public:
   X86TargetInfo(const std::string& triple)
     : TargetInfo(triple), SSELevel(NoSSE), MMX3DNowLevel(NoMMX3DNow),
       HasAES(false), HasAVX(false), HasAVX2(false), HasLZCNT(false),
-      HasBMI(false), HasBMI2(false), HasPOPCNT(false), CPU(CK_Generic) {
+      HasBMI(false), HasBMI2(false), HasPOPCNT(false), HasFMA4(false),
+      CPU(CK_Generic) {
     BigEndian = false;
     LongDoubleFormat = &llvm::APFloat::x87DoubleExtended;
   }
@@ -1521,6 +1523,7 @@ void X86TargetInfo::getDefaultFeatures(llvm::StringMap<bool> &Features) const {
   Features["bmi"] = false;
   Features["bmi2"] = false;
   Features["popcnt"] = false;
+  Features["fma4"] = false;
 
   // FIXME: This *really* should not be here.
 
@@ -1690,8 +1693,13 @@ bool X86TargetInfo::setFeatureEnabled(llvm::StringMap<bool> &Features,
       Features["mmx"] = Features["sse"] = Features["sse2"] = Features["sse3"] =
         Features["ssse3"] = Features["sse41"] = Features["sse42"] =
         Features["popcnt"] = Features["avx"] = Features["avx2"] = true;
+    else if (Name == "fma4")
+        Features["mmx"] = Features["sse"] = Features["sse2"] = Features["sse3"] =
+        Features["ssse3"] = Features["sse41"] = Features["sse42"] =
+        Features["popcnt"] = Features["avx"] = Features["fma4"] = true;
     else if (Name == "sse4a")
-      Features["mmx"] = Features["sse4a"] = true;
+      Features["mmx"] = Features["sse"] = Features["sse2"] = Features["sse3"] =
+        Features["sse4a"] = true;
     else if (Name == "lzcnt")
       Features["lzcnt"] = true;
     else if (Name == "bmi")
@@ -1705,13 +1713,14 @@ bool X86TargetInfo::setFeatureEnabled(llvm::StringMap<bool> &Features,
       Features["mmx"] = Features["3dnow"] = Features["3dnowa"] = false;
     else if (Name == "sse")
       Features["sse"] = Features["sse2"] = Features["sse3"] =
-        Features["ssse3"] = Features["sse41"] = Features["sse42"] = false;
+        Features["ssse3"] = Features["sse41"] = Features["sse42"] =
+        Features["sse4a"] = false;
     else if (Name == "sse2")
       Features["sse2"] = Features["sse3"] = Features["ssse3"] =
-        Features["sse41"] = Features["sse42"] = false;
+        Features["sse41"] = Features["sse42"] = Features["sse4a"] = false;
     else if (Name == "sse3")
       Features["sse3"] = Features["ssse3"] = Features["sse41"] =
-        Features["sse42"] = false;
+        Features["sse42"] = Features["sse4a"] = false;
     else if (Name == "ssse3")
       Features["ssse3"] = Features["sse41"] = Features["sse42"] = false;
     else if (Name == "sse4" || Name == "sse4.1")
@@ -1725,7 +1734,7 @@ bool X86TargetInfo::setFeatureEnabled(llvm::StringMap<bool> &Features,
     else if (Name == "aes")
       Features["aes"] = false;
     else if (Name == "avx")
-      Features["avx"] = Features["avx2"] = false;
+      Features["avx"] = Features["avx2"] = Features["fma4"] = false;
     else if (Name == "avx2")
       Features["avx2"] = false;
     else if (Name == "sse4a")
@@ -1738,6 +1747,8 @@ bool X86TargetInfo::setFeatureEnabled(llvm::StringMap<bool> &Features,
       Features["bmi2"] = false;
     else if (Name == "popcnt")
       Features["popcnt"] = false;
+    else if (Name == "fma4")
+      Features["fma4"] = false;
   }
 
   return true;
@@ -1777,10 +1788,14 @@ void X86TargetInfo::HandleTargetFeatures(std::vector<std::string> &Features) {
       continue;
     }
 
+    if (Features[i].substr(1) == "fma4") {
+      HasFMA4 = true;
+      continue;
+    }
+
     // FIXME: Not sure yet how to treat AVX in regard to SSE levels.
     // For now let it be enabled together with other SSE levels.
     if (Features[i].substr(1) == "avx2") {
-      HasAVX = true;
       HasAVX2 = true;
       continue;
     }
@@ -2011,6 +2026,9 @@ void X86TargetInfo::getTargetDefines(const LangOptions &Opts,
   if (HasPOPCNT)
     Builder.defineMacro("__POPCNT__");
 
+  if (HasFMA4)
+    Builder.defineMacro("__FMA4__");
+
   // Each case falls through to the previous one here.
   switch (SSELevel) {
   case SSE42:

diff --git a/lib/Headers/x86intrin.h b/lib/Headers/x86intrin.h
index 181330d0bb636c1d0ef0578cbe74697ca7cfdbb5..5f9bea7107a25f28968ba4597906d4ba7d05a79b 100644 (file)
--- a/lib/Headers/x86intrin.h
+++ b/lib/Headers/x86intrin.h
@@ -42,6 +42,10 @@
 #include <popcntintrin.h>
 #endif
 
-// FIXME: SSE4A, 3dNOW, FMA4, XOP, LWP, ABM
+#ifdef __FMA4__
+#include <fma4intrin.h>
+#endif
+
+// FIXME: SSE4A, 3dNOW, XOP, LWP, ABM
 
 #endif /* __X86INTRIN_H */