Add support for emitting ARM file attributes.

[oota-llvm.git] / lib / Target / ARM / ARMSubtarget.cpp

diff --git a/lib/Target/ARM/ARMSubtarget.cpp b/lib/Target/ARM/ARMSubtarget.cpp
index d6b072b6c277df6d71f21d9dd132312efaf89210..787bc305dfbcb5b49eec3890e2c8ead5a7d8e648 100644 (file)
--- a/lib/Target/ARM/ARMSubtarget.cpp
+++ b/lib/Target/ARM/ARMSubtarget.cpp
@@ -22,30 +22,42 @@ using namespace llvm;
 static cl::opt<bool>
 ReserveR9("arm-reserve-r9", cl::Hidden,
           cl::desc("Reserve R9, making it unavailable as GPR"));
-static cl::opt<bool>
-UseNEONFP("arm-use-neon-fp",
-          cl::desc("Use NEON for single-precision FP"),
-          cl::init(false), cl::Hidden);
 
 static cl::opt<bool>
 UseMOVT("arm-use-movt",
         cl::init(true), cl::Hidden);
 
+static cl::opt<bool>
+StrictAlign("arm-strict-align", cl::Hidden,
+            cl::desc("Disallow all unaligned memory accesses"));
+
 ARMSubtarget::ARMSubtarget(const std::string &TT, const std::string &FS,
                            bool isT)
-  : ARMArchVersion(V4T)
+  : ARMArchVersion(V4)
+  , ARMProcFamily(Others)
   , ARMFPUType(None)
-  , UseNEONForSinglePrecisionFP(UseNEONFP)
+  , UseNEONForSinglePrecisionFP(false)
+  , SlowVMLx(false)
+  , SlowFPBrcc(false)
   , IsThumb(isT)
   , ThumbMode(Thumb1)
+  , NoARM(false)
   , PostRAScheduler(false)
   , IsR9Reserved(ReserveR9)
   , UseMovt(UseMOVT)
+  , HasFP16(false)
+  , HasD16(false)
+  , HasHardwareDivide(false)
+  , HasT2ExtractPack(false)
+  , HasDataBarrier(false)
+  , Pref32BitThumb(false)
+  , FPOnlySP(false)
+  , AllowsUnalignedMem(false)
   , stackAlignment(4)
   , CPUString("generic")
   , TargetType(isELF) // Default to ELF unless otherwise specified.
   , TargetABI(ARM_ABI_APCS) {
-  // default to soft float ABI
+  // Default to soft float ABI
   if (FloatABIType == FloatABI::Default)
     FloatABIType = FloatABI::Soft;
 
@@ -54,6 +66,11 @@ ARMSubtarget::ARMSubtarget(const std::string &TT, const std::string &FS,
   // Parse features string.
   CPUString = ParseSubtargetFeatures(FS, CPUString);
 
+  // When no arch is specified either by CPU or by attributes, make the default
+  // ARMv4T.
+  if (CPUString == "generic" && (FS.empty() || FS == "generic"))
+    ARMArchVersion = V4T;
+
   // Set the boolean corresponding to the current target triple, or the default
   // if one cannot be determined, to true.
   unsigned Len = TT.length();
@@ -68,25 +85,30 @@ ARMSubtarget::ARMSubtarget(const std::string &TT, const std::string &FS,
   }
   if (Idx) {
     unsigned SubVer = TT[Idx];
-    if (SubVer > '4' && SubVer <= '9') {
-      if (SubVer >= '7') {
-        ARMArchVersion = V7A;
-      } else if (SubVer == '6') {
-        ARMArchVersion = V6;
-        if (Len >= Idx+3 && TT[Idx+1] == 't' && TT[Idx+2] == '2')
-          ARMArchVersion = V6T2;
-      } else if (SubVer == '5') {
-        ARMArchVersion = V5T;
-        if (Len >= Idx+3 && TT[Idx+1] == 't' && TT[Idx+2] == 'e')
-          ARMArchVersion = V5TE;
-      }
-      if (ARMArchVersion >= V6T2)
-        ThumbMode = Thumb2;
+    if (SubVer >= '7' && SubVer <= '9') {
+      ARMArchVersion = V7A;
+      if (Len >= Idx+2 && TT[Idx+1] == 'm')
+        ARMArchVersion = V7M;
+    } else if (SubVer == '6') {
+      ARMArchVersion = V6;
+      if (Len >= Idx+3 && TT[Idx+1] == 't' && TT[Idx+2] == '2')
+        ARMArchVersion = V6T2;
+    } else if (SubVer == '5') {
+      ARMArchVersion = V5T;
+      if (Len >= Idx+3 && TT[Idx+1] == 't' && TT[Idx+2] == 'e')
+        ARMArchVersion = V5TE;
+    } else if (SubVer == '4') {
+      if (Len >= Idx+2 && TT[Idx+1] == 't')
+        ARMArchVersion = V4T;
+      else
+        ARMArchVersion = V4;
     }
   }
 
   // Thumb2 implies at least V6T2.
-  if (ARMArchVersion < V6T2 && ThumbMode >= Thumb2)
+  if (ARMArchVersion >= V6T2)
+    ThumbMode = Thumb2;
+  else if (ThumbMode >= Thumb2)
     ARMArchVersion = V6T2;
 
   if (Len >= 10) {
@@ -107,26 +129,22 @@ ARMSubtarget::ARMSubtarget(const std::string &TT, const std::string &FS,
   if (!isThumb() || hasThumb2())
     PostRAScheduler = true;
 
-  // Set CPU specific features.
-  if (CPUString == "cortex-a8") {
-    // On Cortex-a8, it's faster to perform some single-precision FP
-    // operations with NEON instructions.
-    if (UseNEONFP.getPosition() == 0)
-      UseNEONForSinglePrecisionFP = true;
-  }
-  HasBranchTargetBuffer = (CPUString == "cortex-a8" ||
-                           CPUString == "cortex-a9");
+  // v6+ may or may not support unaligned mem access depending on the system
+  // configuration.
+  if (!StrictAlign && hasV6Ops() && isTargetDarwin())
+    AllowsUnalignedMem = true;
 }
 
 /// GVIsIndirectSymbol - true if the GV will be accessed via an indirect symbol.
 bool
-ARMSubtarget::GVIsIndirectSymbol(GlobalValue *GV, Reloc::Model RelocM) const {
+ARMSubtarget::GVIsIndirectSymbol(const GlobalValue *GV,
+                                 Reloc::Model RelocM) const {
   if (RelocM == Reloc::Static)
     return false;
 
-  // GV with ghost linkage (in JIT lazy compilation mode) do not require an
-  // extra load from stub.
-  bool isDecl = GV->isDeclaration() && !GV->hasNotBeenReadFromBitcode();
+  // Materializable GVs (in JIT lazy compilation mode) do not require an extra
+  // load from stub.
+  bool isDecl = GV->isDeclaration() && !GV->isMaterializable();
 
   if (!isTargetDarwin()) {
     // Extra load is needed for all externally visible.
@@ -168,6 +186,18 @@ ARMSubtarget::GVIsIndirectSymbol(GlobalValue *GV, Reloc::Model RelocM) const {
   return false;
 }
 
+unsigned ARMSubtarget::getMispredictionPenalty() const {
+  // If we have a reasonable estimate of the pipeline depth, then we can
+  // estimate the penalty of a misprediction based on that.
+  if (isCortexA8())
+    return 13;
+  else if (isCortexA9())
+    return 8;
+  
+  // Otherwise, just return a sensible default.
+  return 10;
+}
+
 bool ARMSubtarget::enablePostRAScheduler(
            CodeGenOpt::Level OptLevel,
            TargetSubtarget::AntiDepBreakMode& Mode,