- added from gcc/

[packages/gcc4.git] / gcc4-pr24879.patch

--- gcc/config/i386/i386.c.sse3	2005-11-15 17:33:02.000000000 -0800
+++ gcc/config/i386/i386.c	2005-11-17 09:44:47.000000000 -0800
@@ -14347,10 +14347,20 @@ ix86_init_mmx_sse_builtins (void)
   tree void_ftype_unsigned_unsigned
     = build_function_type_list (void_type_node, unsigned_type_node,
 				unsigned_type_node, NULL_TREE);
+  tree void_ftype_di_di
+    = build_function_type_list (void_type_node,
+				long_long_unsigned_type_node,
+				long_long_unsigned_type_node,
+				NULL_TREE);
   tree void_ftype_pcvoid_unsigned_unsigned
     = build_function_type_list (void_type_node, const_ptr_type_node,
 				unsigned_type_node, unsigned_type_node,
 				NULL_TREE);
+  tree void_ftype_pcvoid_di_di
+    = build_function_type_list (void_type_node, const_ptr_type_node,
+				long_long_unsigned_type_node,
+				long_long_unsigned_type_node,
+				NULL_TREE);
   tree unsigned_ftype_void
     = build_function_type (unsigned_type_node, void_list_node);
   tree v2si_ftype_v4sf
@@ -14811,12 +14821,24 @@ ix86_init_mmx_sse_builtins (void)
   def_builtin (MASK_SSE2, "__builtin_ia32_pmaddwd128", v4si_ftype_v8hi_v8hi, IX86_BUILTIN_PMADDWD128);
 
   /* Prescott New Instructions.  */
-  def_builtin (MASK_SSE3, "__builtin_ia32_monitor",
-	       void_ftype_pcvoid_unsigned_unsigned,
-	       IX86_BUILTIN_MONITOR);
-  def_builtin (MASK_SSE3, "__builtin_ia32_mwait",
-	       void_ftype_unsigned_unsigned,
-	       IX86_BUILTIN_MWAIT);
+  if (TARGET_64BIT)
+    {
+      def_builtin (MASK_SSE3, "__builtin_ia32_monitor",
+		   void_ftype_pcvoid_di_di,
+		   IX86_BUILTIN_MONITOR);
+      def_builtin (MASK_SSE3, "__builtin_ia32_mwait",
+		   void_ftype_di_di,
+		   IX86_BUILTIN_MWAIT);
+    }
+  else
+    {
+      def_builtin (MASK_SSE3, "__builtin_ia32_monitor",
+		   void_ftype_pcvoid_unsigned_unsigned,
+		   IX86_BUILTIN_MONITOR);
+      def_builtin (MASK_SSE3, "__builtin_ia32_mwait",
+		   void_ftype_unsigned_unsigned,
+		   IX86_BUILTIN_MWAIT);
+    }
   def_builtin (MASK_SSE3, "__builtin_ia32_movshdup",
 	       v4sf_ftype_v4sf,
 	       IX86_BUILTIN_MOVSHDUP);
@@ -15660,13 +15682,17 @@ ix86_expand_builtin (tree exp, rtx targe
       op0 = expand_expr (arg0, NULL_RTX, VOIDmode, 0);
       op1 = expand_expr (arg1, NULL_RTX, VOIDmode, 0);
       op2 = expand_expr (arg2, NULL_RTX, VOIDmode, 0);
+      mode = TARGET_64BIT ? DImode : SImode;
       if (!REG_P (op0))
-	op0 = copy_to_mode_reg (SImode, op0);
+	op0 = copy_to_mode_reg (mode, op0);
       if (!REG_P (op1))
-	op1 = copy_to_mode_reg (SImode, op1);
+	op1 = copy_to_mode_reg (mode, op1);
       if (!REG_P (op2))
-	op2 = copy_to_mode_reg (SImode, op2);
-      emit_insn (gen_sse3_monitor (op0, op1, op2));
+	op2 = copy_to_mode_reg (mode, op2);
+      if (TARGET_64BIT)
+	emit_insn (gen_sse3_monitor64 (op0, op1, op2));
+      else
+	emit_insn (gen_sse3_monitor (op0, op1, op2));
       return 0;
 
     case IX86_BUILTIN_MWAIT:
@@ -15674,11 +15700,15 @@ ix86_expand_builtin (tree exp, rtx targe
       arg1 = TREE_VALUE (TREE_CHAIN (arglist));
       op0 = expand_expr (arg0, NULL_RTX, VOIDmode, 0);
       op1 = expand_expr (arg1, NULL_RTX, VOIDmode, 0);
+      mode = TARGET_64BIT ? DImode : SImode;
       if (!REG_P (op0))
-	op0 = copy_to_mode_reg (SImode, op0);
+	op0 = copy_to_mode_reg (mode, op0);
       if (!REG_P (op1))
-	op1 = copy_to_mode_reg (SImode, op1);
-      emit_insn (gen_sse3_mwait (op0, op1));
+	op1 = copy_to_mode_reg (mode, op1);
+      if (TARGET_64BIT)
+	emit_insn (gen_sse3_mwait64 (op0, op1));
+      else
+	emit_insn (gen_sse3_mwait (op0, op1));
       return 0;
 
     case IX86_BUILTIN_LDDQU:
--- gcc/config/i386/pmmintrin.h.sse3	2005-11-04 14:13:48.000000000 -0800
+++ gcc/config/i386/pmmintrin.h	2005-11-15 22:44:02.000000000 -0800
@@ -110,7 +110,6 @@ _mm_lddqu_si128 (__m128i const *__P)
   return (__m128i) __builtin_ia32_lddqu ((char const *)__P);
 }
 
-#if 0
 static __inline void __attribute__((__always_inline__))
 _mm_monitor (void const * __P, unsigned int __E, unsigned int __H)
 {
@@ -122,10 +121,6 @@ _mm_mwait (unsigned int __E, unsigned in
 {
   __builtin_ia32_mwait (__E, __H);
 }
-#else
-#define _mm_monitor(P, E, H)	__builtin_ia32_monitor ((P), (E), (H))
-#define _mm_mwait(E, H)		__builtin_ia32_mwait ((E), (H))
-#endif
 
 #endif /* __SSE3__ */
 
--- gcc/config/i386/sse.md.sse3	2005-11-04 14:13:48.000000000 -0800
+++ gcc/config/i386/sse.md	2005-11-17 09:25:31.000000000 -0800
@@ -3890,15 +3890,36 @@
   [(unspec_volatile [(match_operand:SI 0 "register_operand" "a")
 		     (match_operand:SI 1 "register_operand" "c")]
 		    UNSPECV_MWAIT)]
-  "TARGET_SSE3"
+  "TARGET_SSE3 && !TARGET_64BIT"
   "mwait\t%0, %1"
   [(set_attr "length" "3")])
 
+(define_insn "sse3_mwait64"
+  [(unspec_volatile [(match_operand:DI 0 "register_operand" "a")
+		     (match_operand:DI 1 "register_operand" "c")]
+		    UNSPECV_MWAIT)]
+  "TARGET_SSE3 && TARGET_64BIT"
+;; Older assembler doesn't support "mwait %rax,%rcx".
+;;  "mwait\t%0, %1"
+  "mwait"
+  [(set_attr "length" "3")])
+
 (define_insn "sse3_monitor"
   [(unspec_volatile [(match_operand:SI 0 "register_operand" "a")
 		     (match_operand:SI 1 "register_operand" "c")
 		     (match_operand:SI 2 "register_operand" "d")]
 		    UNSPECV_MONITOR)]
-  "TARGET_SSE3"
+  "TARGET_SSE3 && !TARGET_64BIT"
   "monitor\t%0, %1, %2"
   [(set_attr "length" "3")])
+
+(define_insn "sse3_monitor64"
+  [(unspec_volatile [(match_operand:DI 0 "register_operand" "a")
+		     (match_operand:DI 1 "register_operand" "c")
+		     (match_operand:DI 2 "register_operand" "d")]
+		    UNSPECV_MONITOR)]
+  "TARGET_SSE3 && TARGET_64BIT"
+;; Older assembler doesn't support "monitor %rax,%rcx,%rdx".
+;;  "monitor\t%0, %1, %2"
+  "monitor"
+  [(set_attr "length" "3")])
--- gcc/testsuite/gcc.target/i386/monitor.c.sse3	2005-11-15 22:59:54.000000000 -0800
+++ gcc/testsuite/gcc.target/i386/monitor.c	2005-11-15 23:25:23.000000000 -0800
@@ -0,0 +1,27 @@
+/* { dg-do compile { target i?86-*-* x86_64-*-* } } */
+/* { dg-options "-O2 -msse3" } */
+
+/* Verify that they work in both 32bit and 64bit.  */
+
+#include <pmmintrin.h>
+
+void
+foo (char *p, int x, int y, int z)
+{
+   _mm_monitor (p, y, x);
+   _mm_mwait (z, y);
+}
+
+void
+bar (char *p, long x, long y, long z)
+{
+   _mm_monitor (p, y, x);
+   _mm_mwait (z, y);
+}
+
+void
+foo1 (char *p)
+{
+   _mm_monitor (p, 0, 0);
+   _mm_mwait (0, 0);
+}