re PR target/24879 ([4.1]: SSE3 monitor intrinsic doesn't work in 64bit)

gcc/

2006-05-07  H.J. Lu  <hongjiu.lu@intel.com>

	PR target/24879
	* config/i386/pmmintrin.h (_mm_monitor): Remove macro. Use
	inline function.
	(_mm_mwait): Likewise.

	* config/i386/sse.md (sse3_mwait): Replace "mwait\t%0, %1" with
	"mwait".
	(sse3_monitor): Make it 32bit only.
	(sse3_monitor64): New. 64bit monitor.

gcc/testsuite/

2006-05-07  H.J. Lu  <hongjiu.lu@intel.com>

	PR target/24879
	* gcc.target/i386/monitor.c: New file.

From-SVN: r113617

5 files changed, 63 insertions, 9 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 5f35ff3..4882a20 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,15 @@
+2006-05-07  H.J. Lu  <hongjiu.lu@intel.com>
+
+	PR target/24879
+	* config/i386/pmmintrin.h (_mm_monitor): Remove macro. Use
+	inline function.
+	(_mm_mwait): Likewise.
+
+	* config/i386/sse.md (sse3_mwait): Replace "mwait\t%0, %1" with
+	"mwait".
+	(sse3_monitor): Make it 32bit only.
+	(sse3_monitor64): New. 64bit monitor.
+
 2006-05-07  Volker Reichelt  <reichelt@igpm.rwth-aachen.de>
 
 	PR target/27421
diff --git a/gcc/config/i386/pmmintrin.h b/gcc/config/i386/pmmintrin.h
index ca29559..7dbf030 100644
--- a/gcc/config/i386/pmmintrin.h
+++ b/gcc/config/i386/pmmintrin.h
@@ -1,4 +1,4 @@
-/* Copyright (C) 2003, 2004, 2005 Free Software Foundation, Inc.
+/* Copyright (C) 2003, 2004, 2005, 2006 Free Software Foundation, Inc.
 
    This file is part of GCC.
 
@@ -25,7 +25,7 @@
    Public License.  */
 
 /* Implemented from the specification included in the Intel C++ Compiler
-   User Guide and Reference, version 8.0.  */
+   User Guide and Reference, version 9.0.  */
 
 #ifndef _PMMINTRIN_H_INCLUDED
 #define _PMMINTRIN_H_INCLUDED
@@ -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 int __H)
 {
   __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__ */
 
diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
index 50eced2..ed36276 100644
--- a/gcc/config/i386/sse.md
+++ b/gcc/config/i386/sse.md
@@ -3972,7 +3972,10 @@
 		     (match_operand:SI 1 "register_operand" "c")]
 		    UNSPECV_MWAIT)]
   "TARGET_SSE3"
-  "mwait\t%0, %1"
+;; 64bit version is "mwait %rax,%rcx". But only lower 32bits are used.
+;; Since 32bit register operands are implicitly zero extended to 64bit,
+;; we only need to set up 32bit registers.
+  "mwait"
   [(set_attr "length" "3")])
 
 (define_insn "sse3_monitor"
@@ -3980,6 +3983,18 @@
 		     (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:SI 1 "register_operand" "c")
+		     (match_operand:SI 2 "register_operand" "d")]
+		    UNSPECV_MONITOR)]
+  "TARGET_SSE3 && TARGET_64BIT"
+;; 64bit version is "monitor %rax,%rcx,%rdx". But only lower 32bits in
+;; RCX and RDX are used.  Since 32bit register operands are implicitly
+;; zero extended to 64bit, we only need to set up 32bit registers.
+  "monitor"
+  [(set_attr "length" "3")])
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index 5e608d9..5da14b5 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,8 @@
+2006-05-07  H.J. Lu  <hongjiu.lu@intel.com>
+
+	PR target/24879
+	* gcc.target/i386/monitor.c: New file.
+
 2006-05-08  Alan Modra  <amodra@bigpond.net.au>
 
 	* gcc.dg/pr27095.c: xfail *-*-darwin*.
diff --git a/gcc/testsuite/gcc.target/i386/monitor.c b/gcc/testsuite/gcc.target/i386/monitor.c
new file mode 100644
index 0000000..d98a2fd
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/monitor.c
@@ -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);
+}