Fix a x86-64 codegen deficiency. Allow gv + offset when using rip addressing mode.

Before:
_main:
        subq    $8, %rsp
        leaq    _X(%rip), %rax
        movsd   8(%rax), %xmm1
        movss   _X(%rip), %xmm0
        call    _t
        xorl    %ecx, %ecx
        movl    %ecx, %eax
        addq    $8, %rsp
        ret
Now:
_main:
        subq    $8, %rsp
        movsd   _X+8(%rip), %xmm1
        movss   _X(%rip), %xmm0
        call    _t
        xorl    %ecx, %ecx
        movl    %ecx, %eax
        addq    $8, %rsp
        ret

Notice there is another idiotic codegen issue that needs to be fixed asap:
xorl    %ecx, %ecx
movl    %ecx, %eax


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46850 91177308-0d34-0410-b5e6-96231b3b80d8

7 files changed, 76 insertions, 82 deletions

diff --git a/lib/Target/X86/X86ISelDAGToDAG.cpp b/lib/Target/X86/X86ISelDAGToDAG.cpp
index c51cd80cb9..0703563f01 100644
--- a/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -63,7 +63,7 @@ namespace {
       int FrameIndex;
     } Base;
 
-    bool isRIPRel;     // RIP relative?
+    bool isRIPRel;     // RIP as base?
     unsigned Scale;
     SDOperand IndexReg; 
     unsigned Disp;
@@ -664,7 +664,9 @@ bool X86DAGToDAGISel::MatchAddress(SDOperand N, X86ISelAddressMode &AM,
   case X86ISD::Wrapper: {
     bool is64Bit = Subtarget->is64Bit();
     // Under X86-64 non-small code model, GV (and friends) are 64-bits.
-    if (is64Bit && TM.getCodeModel() != CodeModel::Small)
+    // Also, base and index reg must be 0 in order to use rip as base.
+    if (is64Bit && (TM.getCodeModel() != CodeModel::Small ||
+                    AM.Base.Reg.Val || AM.IndexReg.Val))
       break;
     if (AM.GV != 0 || AM.CP != 0 || AM.ES != 0 || AM.JT != -1)
       break;
@@ -672,39 +674,27 @@ bool X86DAGToDAGISel::MatchAddress(SDOperand N, X86ISelAddressMode &AM,
     // been picked, we can't fit the result available in the register in the
     // addressing mode. Duplicate GlobalAddress or ConstantPool as displacement.
     if (!AlreadySelected || (AM.Base.Reg.Val && AM.IndexReg.Val)) {
-      bool isStatic = TM.getRelocationModel() == Reloc::Static;
       SDOperand N0 = N.getOperand(0);
-      // Mac OS X X86-64 lower 4G address is not available.
-      bool isAbs32 = !is64Bit ||
-        (isStatic && Subtarget->hasLow4GUserSpaceAddress());
       if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(N0)) {
         GlobalValue *GV = G->getGlobal();
-        if (isAbs32 || isRoot) {
-          AM.GV = GV;
-          AM.Disp += G->getOffset();
-          AM.isRIPRel = !isAbs32;
-          return false;
-        }
+        AM.GV = GV;
+        AM.Disp += G->getOffset();
+        AM.isRIPRel = is64Bit;
+        return false;
       } else if (ConstantPoolSDNode *CP = dyn_cast<ConstantPoolSDNode>(N0)) {
-        if (isAbs32 || isRoot) {
-          AM.CP = CP->getConstVal();
-          AM.Align = CP->getAlignment();
-          AM.Disp += CP->getOffset();
-          AM.isRIPRel = !isAbs32;
-          return false;
-        }
+        AM.CP = CP->getConstVal();
+        AM.Align = CP->getAlignment();
+        AM.Disp += CP->getOffset();
+        AM.isRIPRel = is64Bit;
+        return false;
       } else if (ExternalSymbolSDNode *S =dyn_cast<ExternalSymbolSDNode>(N0)) {
-        if (isAbs32 || isRoot) {
-          AM.ES = S->getSymbol();
-          AM.isRIPRel = !isAbs32;
-          return false;
-        }
+        AM.ES = S->getSymbol();
+        AM.isRIPRel = is64Bit;
+        return false;
       } else if (JumpTableSDNode *J = dyn_cast<JumpTableSDNode>(N0)) {
-        if (isAbs32 || isRoot) {
-          AM.JT = J->getIndex();
-          AM.isRIPRel = !isAbs32;
-          return false;
-        }
+        AM.JT = J->getIndex();
+        AM.isRIPRel = is64Bit;
+        return false;
       }
     }
     break;
@@ -719,7 +709,7 @@ bool X86DAGToDAGISel::MatchAddress(SDOperand N, X86ISelAddressMode &AM,
     break;
 
   case ISD::SHL:
-    if (AlreadySelected || AM.IndexReg.Val != 0 || AM.Scale != 1)
+    if (AlreadySelected || AM.IndexReg.Val != 0 || AM.Scale != 1 || AM.isRIPRel)
       break;
       
     if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(N.Val->getOperand(1))) {
@@ -759,7 +749,8 @@ bool X86DAGToDAGISel::MatchAddress(SDOperand N, X86ISelAddressMode &AM,
     if (!AlreadySelected &&
         AM.BaseType == X86ISelAddressMode::RegBase &&
         AM.Base.Reg.Val == 0 &&
-        AM.IndexReg.Val == 0) {
+        AM.IndexReg.Val == 0 &&
+        !AM.isRIPRel) {
       if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(N.Val->getOperand(1)))
         if (CN->getValue() == 3 || CN->getValue() == 5 || CN->getValue() == 9) {
           AM.Scale = unsigned(CN->getValue())-1;
@@ -834,6 +825,9 @@ bool X86DAGToDAGISel::MatchAddress(SDOperand N, X86ISelAddressMode &AM,
     
     // Scale must not be used already.
     if (AM.IndexReg.Val != 0 || AM.Scale != 1) break;
+
+    // Not when RIP is used as the base.
+    if (AM.isRIPRel) break;
       
     ConstantSDNode *C2 = dyn_cast<ConstantSDNode>(N.getOperand(1));
     ConstantSDNode *C1 = dyn_cast<ConstantSDNode>(Shift.getOperand(1));
@@ -874,7 +868,7 @@ bool X86DAGToDAGISel::MatchAddressBase(SDOperand N, X86ISelAddressMode &AM,
   // Is the base register already occupied?
   if (AM.BaseType != X86ISelAddressMode::RegBase || AM.Base.Reg.Val) {
     // If so, check to see if the scale index register is set.
-    if (AM.IndexReg.Val == 0) {
+    if (AM.IndexReg.Val == 0 && !AM.isRIPRel) {
       AM.IndexReg = N;
       AM.Scale = 1;
       return false;
diff --git a/lib/Target/X86/X86Instr64bit.td b/lib/Target/X86/X86Instr64bit.td
index 9528dbd2c9..9895853b28 100644
--- a/lib/Target/X86/X86Instr64bit.td
+++ b/lib/Target/X86/X86Instr64bit.td
@@ -1125,16 +1125,16 @@ def : Pat<(i64 (X86Wrapper texternalsym:$dst)),
 
 def : Pat<(store (i64 (X86Wrapper tconstpool:$src)), addr:$dst),
           (MOV64mi32 addr:$dst, tconstpool:$src)>,
-          Requires<[SmallCode, HasLow4G, IsStatic]>;
+          Requires<[SmallCode, IsStatic]>;
 def : Pat<(store (i64 (X86Wrapper tjumptable:$src)), addr:$dst),
           (MOV64mi32 addr:$dst, tjumptable:$src)>,
-          Requires<[SmallCode, HasLow4G, IsStatic]>;
+          Requires<[SmallCode, IsStatic]>;
 def : Pat<(store (i64 (X86Wrapper tglobaladdr:$src)), addr:$dst),
           (MOV64mi32 addr:$dst, tglobaladdr:$src)>,
-          Requires<[SmallCode, HasLow4G, IsStatic]>;
+          Requires<[SmallCode, IsStatic]>;
 def : Pat<(store (i64 (X86Wrapper texternalsym:$src)), addr:$dst),
           (MOV64mi32 addr:$dst, texternalsym:$src)>,
-          Requires<[SmallCode, HasLow4G, IsStatic]>;
+          Requires<[SmallCode, IsStatic]>;
 
 // Calls
 // Direct PC relative function call for small code model. 32-bit displacement
diff --git a/lib/Target/X86/X86InstrInfo.td b/lib/Target/X86/X86InstrInfo.td
index a79947b5c0..f40847f1e4 100644
--- a/lib/Target/X86/X86InstrInfo.td
+++ b/lib/Target/X86/X86InstrInfo.td
@@ -172,7 +172,6 @@ def FPStackf32   : Predicate<"!Subtarget->hasSSE1()">;
 def FPStackf64   : Predicate<"!Subtarget->hasSSE2()">;
 def In32BitMode  : Predicate<"!Subtarget->is64Bit()">;
 def In64BitMode  : Predicate<"Subtarget->is64Bit()">;
-def HasLow4G     : Predicate<"Subtarget->hasLow4GUserSpaceAddress()">;
 def SmallCode    : Predicate<"TM.getCodeModel() == CodeModel::Small">;
 def NotSmallCode : Predicate<"TM.getCodeModel() != CodeModel::Small">;
 def IsStatic     : Predicate<"TM.getRelocationModel() == Reloc::Static">;
diff --git a/lib/Target/X86/X86Subtarget.cpp b/lib/Target/X86/X86Subtarget.cpp
index 35a83e4ef7..abd756c45c 100644
--- a/lib/Target/X86/X86Subtarget.cpp
+++ b/lib/Target/X86/X86Subtarget.cpp
@@ -228,7 +228,6 @@ X86Subtarget::X86Subtarget(const Module &M, const std::string &FS, bool is64Bit)
   // FIXME: this is a known good value for Yonah. How about others?
   , MaxInlineSizeThreshold(128)
   , Is64Bit(is64Bit)
-  , HasLow4GUserAddress(true)
   , TargetType(isELF) { // Default to ELF unless otherwise specified.
 
   // Determine default and user specified characteristics
@@ -300,9 +299,6 @@ X86Subtarget::X86Subtarget(const Module &M, const std::string &FS, bool is64Bit)
       ? X86Subtarget::Intel : X86Subtarget::ATT;
   }
 
-  if (TargetType == isDarwin && Is64Bit)
-    HasLow4GUserAddress = false;
-
   if (TargetType == isDarwin ||
       TargetType == isCygwin ||
       TargetType == isMingw  ||
diff --git a/lib/Target/X86/X86Subtarget.h b/lib/Target/X86/X86Subtarget.h
index c2687265af..789ea3eccf 100644
--- a/lib/Target/X86/X86Subtarget.h
+++ b/lib/Target/X86/X86Subtarget.h
@@ -82,10 +82,6 @@ private:
   /// pointer size is 64 bit.
   bool Is64Bit;
 
-  /// HasLow4GUserAddress - True if the low 4G user-space address is available.
-  ///
-  bool HasLow4GUserAddress;
-
 public:
   enum {
     isELF, isCygwin, isDarwin, isWindows, isMingw
@@ -115,10 +111,6 @@ public:
 
   bool is64Bit() const { return Is64Bit; }
 
-  /// hasLow4GUserSpaceAddress - True if lower 4G user-space address is
-  /// available.
-  bool hasLow4GUserSpaceAddress() const { return HasLow4GUserAddress; }
-
   PICStyle::Style getPICStyle() const { return PICStyle; }
   void setPICStyle(PICStyle::Style Style)  { PICStyle = Style; }
 
diff --git a/test/CodeGen/X86/pic_jumptable.ll b/test/CodeGen/X86/pic_jumptable.ll
index d074514476..553021dc90 100644
--- a/test/CodeGen/X86/pic_jumptable.ll
+++ b/test/CodeGen/X86/pic_jumptable.ll
@@ -1,36 +1,35 @@
-; RUN: llvm-upgrade < %s | llvm-as | llc -relocation-model=pic -mtriple=i386-linux-gnu | not grep -F .text
-; RUN: llvm-upgrade < %s | llvm-as | llc -relocation-model=pic -mtriple=i686-apple-darwin | not grep lea
+; RUN: llvm-as < %s | llc -relocation-model=pic -mtriple=i386-linux-gnu | not grep -F .text
+; RUN: llvm-as < %s | llc -relocation-model=pic -mtriple=i686-apple-darwin | not grep lea
+; RUN: llvm-as < %s | llc -relocation-model=pic -mtriple=i686-apple-darwin | grep add | count 2
 
-implementation   ; Functions:
+declare void @_Z3bari(i32)
 
-declare void %_Z3bari( int  )
-
-linkonce void %_Z3fooILi1EEvi(int %Y) {
+define linkonce void @_Z3fooILi1EEvi(i32 %Y) {
 entry:
-	%Y_addr = alloca int		; <int*> [#uses=2]
-	"alloca point" = cast int 0 to int		; <int> [#uses=0]
-	store int %Y, int* %Y_addr
-	%tmp = load int* %Y_addr		; <int> [#uses=1]
-	switch int %tmp, label %bb10 [
-		 int 0, label %bb3
-		 int 1, label %bb
-		 int 2, label %bb
-		 int 3, label %bb
-		 int 4, label %bb
-		 int 5, label %bb
-		 int 6, label %bb
-		 int 7, label %bb
-		 int 8, label %bb
-		 int 9, label %bb
-		 int 10, label %bb
-		 int 12, label %bb1
-		 int 13, label %bb5
-		 int 14, label %bb6
-		 int 16, label %bb2
-		 int 17, label %bb4
-		 int 23, label %bb8
-		 int 27, label %bb7
-		 int 34, label %bb9
+	%Y_addr = alloca i32		; <i32*> [#uses=2]
+	%"alloca point" = bitcast i32 0 to i32		; <i32> [#uses=0]
+	store i32 %Y, i32* %Y_addr
+	%tmp = load i32* %Y_addr		; <i32> [#uses=1]
+	switch i32 %tmp, label %bb10 [
+		 i32 0, label %bb3
+		 i32 1, label %bb
+		 i32 2, label %bb
+		 i32 3, label %bb
+		 i32 4, label %bb
+		 i32 5, label %bb
+		 i32 6, label %bb
+		 i32 7, label %bb
+		 i32 8, label %bb
+		 i32 9, label %bb
+		 i32 10, label %bb
+		 i32 12, label %bb1
+		 i32 13, label %bb5
+		 i32 14, label %bb6
+		 i32 16, label %bb2
+		 i32 17, label %bb4
+		 i32 23, label %bb8
+		 i32 27, label %bb7
+		 i32 34, label %bb9
 	]
 
 bb:		; preds = %entry, %entry, %entry, %entry, %entry, %entry, %entry, %entry, %entry, %entry
@@ -40,7 +39,7 @@ bb1:		; preds = %bb, %entry
 	br label %bb2
 
 bb2:		; preds = %bb1, %entry
-	call void %_Z3bari( int 1 )
+	call void @_Z3bari( i32 1 )
 	br label %bb11
 
 bb3:		; preds = %entry
@@ -53,7 +52,7 @@ bb5:		; preds = %bb4, %entry
 	br label %bb6
 
 bb6:		; preds = %bb5, %entry
-	call void %_Z3bari( int 2 )
+	call void @_Z3bari( i32 2 )
 	br label %bb11
 
 bb7:		; preds = %entry
@@ -63,7 +62,7 @@ bb8:		; preds = %bb7, %entry
 	br label %bb9
 
 bb9:		; preds = %bb8, %entry
-	call void %_Z3bari( int 3 )
+	call void @_Z3bari( i32 3 )
 	br label %bb11
 
 bb10:		; preds = %entry
diff --git a/test/CodeGen/X86/x86-64-gv-offset.ll b/test/CodeGen/X86/x86-64-gv-offset.ll
new file mode 100644
index 0000000000..b89e1b9536
--- /dev/null
+++ b/test/CodeGen/X86/x86-64-gv-offset.ll
@@ -0,0 +1,14 @@
+; RUN: llvm-as < %s | llc -mtriple=x86_64-apple-darwin | not grep lea
+
+	%struct.x = type { float, double }
+@X = global %struct.x { float 1.000000e+00, double 2.000000e+00 }, align 16		; <%struct.x*> [#uses=2]
+
+define i32 @main() nounwind  {
+entry:
+	%tmp2 = load float* getelementptr (%struct.x* @X, i32 0, i32 0), align 16		; <float> [#uses=1]
+	%tmp4 = load double* getelementptr (%struct.x* @X, i32 0, i32 1), align 8		; <double> [#uses=1]
+	tail call void @t( float %tmp2, double %tmp4 ) nounwind 
+	ret i32 0
+}
+
+declare void @t(float, double)