Reapply 132424 with fixes. This fixes PR10068.

[oota-llvm.git] / lib / Target / X86 / X86CallingConv.td

diff --git a/lib/Target/X86/X86CallingConv.td b/lib/Target/X86/X86CallingConv.td
index 56863d9492e2fcc3e844d8e33433231cc900c0f0..56351756e8ddfce3170e0e6708877e5de5cad516 100644 (file)
--- a/lib/Target/X86/X86CallingConv.td
+++ b/lib/Target/X86/X86CallingConv.td
@@ -61,7 +61,7 @@ def RetCC_X86_32_C : CallingConv<[
   // weirdly; this is really the sse-regparm calling convention) in which
   // case they use XMM0, otherwise it is the same as the common X86 calling
   // conv.
-  CCIfInReg<CCIfSubtarget<"hasSSE2()",
+  CCIfInReg<CCIfSubtarget<"hasXMMInt()",
     CCIfType<[f32, f64], CCAssignToReg<[XMM0,XMM1,XMM2]>>>>,
   CCIfType<[f32,f64], CCAssignToReg<[ST0, ST1]>>,
   CCDelegateTo<RetCC_X86Common>
@@ -73,8 +73,8 @@ def RetCC_X86_32_Fast : CallingConv<[
   // SSE2.
   // This can happen when a float, 2 x float, or 3 x float vector is split by
   // target lowering, and is returned in 1-3 sse regs.
-  CCIfType<[f32], CCIfSubtarget<"hasSSE2()", CCAssignToReg<[XMM0,XMM1,XMM2]>>>,
-  CCIfType<[f64], CCIfSubtarget<"hasSSE2()", CCAssignToReg<[XMM0,XMM1,XMM2]>>>,
+  CCIfType<[f32], CCIfSubtarget<"hasXMMInt()", CCAssignToReg<[XMM0,XMM1,XMM2]>>>,
+  CCIfType<[f64], CCIfSubtarget<"hasXMMInt()", CCAssignToReg<[XMM0,XMM1,XMM2]>>>,
 
   // For integers, ECX can be used as an extra return register
   CCIfType<[i8],  CCAssignToReg<[AL, DL, CL]>>,
@@ -163,12 +163,12 @@ def CC_X86_64_C : CallingConv<[
   // registers on Darwin.
   CCIfType<[x86mmx],
             CCIfSubtarget<"isTargetDarwin()",
-            CCIfSubtarget<"hasSSE2()",
+            CCIfSubtarget<"hasXMMInt()",
             CCPromoteToType<v2i64>>>>,
 
   // The first 8 FP/Vector arguments are passed in XMM registers.
   CCIfType<[f32, f64, v16i8, v8i16, v4i32, v2i64, v4f32, v2f64],
-            CCIfSubtarget<"hasSSE1()",
+            CCIfSubtarget<"hasXMM()",
             CCAssignToReg<[XMM0, XMM1, XMM2, XMM3, XMM4, XMM5, XMM6, XMM7]>>>,
 
   // The first 8 256-bit vector arguments are passed in YMM registers.
@@ -215,6 +215,13 @@ def CC_X86_Win64_C : CallingConv<[
   // The first 4 integer arguments are passed in integer registers.
   CCIfType<[i32], CCAssignToRegWithShadow<[ECX , EDX , R8D , R9D ],
                                           [XMM0, XMM1, XMM2, XMM3]>>,
+  
+  // Do not pass the sret argument in RCX, the Win64 thiscall calling
+  // convention requires "this" to be passed in RCX.                                        
+  CCIfCC<"CallingConv::X86_ThisCall", 
+    CCIfSRet<CCIfType<[i64], CCAssignToRegWithShadow<[RDX , R8  , R9  ],
+                                                     [XMM1, XMM2, XMM3]>>>>,
+
   CCIfType<[i64], CCAssignToRegWithShadow<[RCX , RDX , R8  , R9  ],
                                           [XMM0, XMM1, XMM2, XMM3]>>,
 
@@ -245,7 +252,7 @@ def CC_X86_64_GHC : CallingConv<[
 
   // Pass in STG registers: F1, F2, F3, F4, D1, D2
   CCIfType<[f32, f64, v16i8, v8i16, v4i32, v2i64, v4f32, v2f64],
-            CCIfSubtarget<"hasSSE1()",
+            CCIfSubtarget<"hasXMM()",
             CCAssignToReg<[XMM1, XMM2, XMM3, XMM4, XMM5, XMM6]>>>
 ]>;
 
@@ -263,7 +270,7 @@ def CC_X86_32_Common : CallingConv<[
   // The first 3 float or double arguments, if marked 'inreg' and if the call
   // is not a vararg call and if SSE2 is available, are passed in SSE registers.
   CCIfNotVarArg<CCIfInReg<CCIfType<[f32,f64],
-                CCIfSubtarget<"hasSSE2()",
+                CCIfSubtarget<"hasXMMInt()",
                 CCAssignToReg<[XMM0,XMM1,XMM2]>>>>>,
 
   // The first 3 __m64 (except for v1i64) vector arguments are passed in mmx
@@ -362,7 +369,7 @@ def CC_X86_32_FastCC : CallingConv<[
   // The first 3 float or double arguments, if the call is not a vararg
   // call and if SSE2 is available, are passed in SSE registers.
   CCIfNotVarArg<CCIfType<[f32,f64],
-                CCIfSubtarget<"hasSSE2()",
+                CCIfSubtarget<"hasXMMInt()",
                 CCAssignToReg<[XMM0,XMM1,XMM2]>>>>,
 
   // Doubles get 8-byte slots that are 8-byte aligned.