Sync SH port with FSF; enable regmove for SH.

From-SVN: r16371

gcc/ChangeLog

+Sat Nov  8 18:20:21 1997  J"orn Rennecke <amylaar@cygnus.co.uk>
+
+	* sh.h (ENABLE_REGMOVE_PASS): Define.
+
+	Bring over from FSF:
+
+	Thu Oct 30 12:21:06 1997  J"orn Rennecke <amylaar@cygnus.co.uk>
+
+	* va-sh.h (__va_arg_sh1): Define.
+	(va_arg): Use it.
+	SH3E doesn't use any integer registers for subsequent arguments
+	once a non-float value was passed in the stack.
+	* sh.c (machine_dependent_reorg): If optimizing, put explicit
+	alignment in front label for ADDR_DIFF_VEC.
+	* sh.h (PASS_IN_REG_P): Fix SH3E case.
+	(ADJUST_INSN_LENGTH): If not optimizing, add two extra bytes length.
+
+	Tue Oct 28 15:06:44 1997  J"orn Rennecke <amylaar@cygnus.co.uk>
+
+	* sh/elf.h (PREFERRED_DEBUGGING_TYPE): Undefine before including
+	svr4.h.
+
+	Mon Oct 27 16:11:52 1997  J"orn Rennecke <amylaar@cygnus.co.uk>
+
+	* sh.c (machine_dependent_reorg): When -flag_delayed_branches,
+	put an use_sfunc_addr before each sfunc.
+	* sh.md (use_sfunc_addr, dummy_jump): New insns.
+	(casesi): For TARGET_SH2, emit a dummy_jump after LAB.
+
+	Tue Oct 21 07:12:28 1997  J"orn Rennecke <amylaar@cygnus.co.uk>
+
+	* sh/elf.h (PREFERRED_DEBUGGING_TYPE): Don't redefine.
+
 Fri Nov  7 10:22:24 1997  Jason Merrill  <jason@yorick.cygnus.com>
 
 	* frame.c (add_fdes, count_fdes): Go back to checking pc_begin for

gcc/config/sh/elf.h

@@ -25,10 +25,6 @@ Boston, MA 02111-1307, USA.  */
 /* No SDB debugging info.  */
 #undef SDB_DEBUGGING_INFO
 
-/* Prefer stabs.  */
-#undef PREFERRED_DEBUGGING_TYPE
-#define PREFERRED_DEBUGGING_TYPE DBX_DEBUG
-
 /* Undefine some macros defined in both sh.h and svr4.h.  */
 #undef IDENT_ASM_OP
 #undef ASM_FILE_END
@@ -40,10 +36,30 @@ Boston, MA 02111-1307, USA.  */
 #undef ASM_OUTPUT_CONSTRUCTOR
 #undef ASM_OUTPUT_DESTRUCTOR
 #undef ASM_DECLARE_FUNCTION_NAME
+#undef PREFERRED_DEBUGGING_TYPE
 
 /* Be ELF-like.  */
 #include "svr4.h"
 
+/* The prefix to add to user-visible assembler symbols.
+   Note that svr4.h redefined it from the original value (that we want)
+   in sh.h */
+
+#undef USER_LABEL_PREFIX
+#define USER_LABEL_PREFIX "_"
+
+#undef LOCAL_LABEL_PREFIX
+#define LOCAL_LABEL_PREFIX "."
+
+#undef ASM_FILE_START
+#define ASM_FILE_START(FILE) do {				\
+  output_file_directive ((FILE), main_input_filename);		\
+  if (TARGET_LITTLE_ENDIAN)					\
+    fprintf ((FILE), "\t.little\n");				\
+} while (0)
+
+
+
 /* Let code know that this is ELF.  */
 #define CPP_PREDEFINES "-D__sh__ -D__ELF__ -Acpu(sh) -Amachine(sh)"
 
@@ -52,7 +68,7 @@ Boston, MA 02111-1307, USA.  */
 #define ASM_SPEC  "%{ml:-little} %{mrelax:-relax}"
 
 #undef LINK_SPEC
-#define LINK_SPEC "%{ml:-m shl} %{mrelax:-relax}"
+#define LINK_SPEC "%{ml:-m shlelf} %{mrelax:-relax}"
 
 /* svr4.h undefined DBX_REGISTER_NUMBER, so we need to define it
    again.  */
@@ -63,29 +79,26 @@ Boston, MA 02111-1307, USA.  */
    symbol names.  */
 #undef ASM_OUTPUT_LABELREF
 #define ASM_OUTPUT_LABELREF(STREAM,NAME) \
-  fprintf (STREAM, "_%s", NAME)
-
-/* Because SH ELF uses underscores, we don't put a '.' before local
-   labels, for easy compatibility with the COFF implementation.  */
+  asm_fprintf (STREAM, "%U%s", NAME)
 
 #undef ASM_GENERATE_INTERNAL_LABEL
 #define ASM_GENERATE_INTERNAL_LABEL(STRING, PREFIX, NUM) \
-  sprintf (STRING, "*%s%d", PREFIX, NUM)
+  sprintf ((STRING), "*%s%s%d", LOCAL_LABEL_PREFIX, (PREFIX), (NUM))
 
 #undef ASM_OUTPUT_INTERNAL_LABEL
 #define ASM_OUTPUT_INTERNAL_LABEL(FILE,PREFIX,NUM) \
-  fprintf (FILE, "%s%d:\n", PREFIX, NUM)
+  asm_fprintf ((FILE), "%L%s%d:\n", (PREFIX), (NUM))
 
 #undef  ASM_OUTPUT_SOURCE_LINE
 #define ASM_OUTPUT_SOURCE_LINE(file, line)				\
 do									\
   {									\
     static int sym_lineno = 1;						\
-    fprintf (file, ".stabn 68,0,%d,LM%d-",				\
-	     line, sym_lineno);						\
-    assemble_name (file,						\
+    asm_fprintf ((file), ".stabn 68,0,%d,%LLM%d-",			\
+	     (line), sym_lineno);					\
+    assemble_name ((file),						\
 		   XSTR (XEXP (DECL_RTL (current_function_decl), 0), 0));\
-    fprintf (file, "\nLM%d:\n", sym_lineno);				\
+    asm_fprintf ((file), "\n%LLM%d:\n", sym_lineno);			\
     sym_lineno += 1;							\
   }									\
 while (0)
@@ -94,7 +107,7 @@ while (0)
 #define DBX_OUTPUT_MAIN_SOURCE_FILE_END(FILE, FILENAME)			\
 do {									\
   text_section ();							\
-  fprintf (FILE, "\t.stabs \"\",%d,0,0,Letext\nLetext:\n", N_SO);	\
+  fprintf ((FILE), "\t.stabs \"\",%d,0,0,Letext\nLetext:\n", N_SO);	\
 } while (0)
 
 /* Arrange to call __main, rather than using crtbegin.o and crtend.o
@@ -103,3 +116,7 @@ do {									\
 #undef FINI_SECTION_ASM_OP
 #undef STARTFILE_SPEC
 #undef ENDFILE_SPEC
+
+/* HANDLE_SYSV_PRAGMA (defined by svr4.h) takes precedence over HANDLE_PRAGMA.
+   We want to use the HANDLE_PRAGMA from sh.h.  */
+#undef HANDLE_SYSV_PRAGMA

gcc/config/sh/sh.h

@@ -804,9 +804,16 @@ struct sh_args {
    This macro is only used in this file. */
 
 #define PASS_IN_REG_P(CUM, MODE, TYPE) \
-  (ROUND_REG ((CUM), (MODE)) < NPARM_REGS (MODE)		\
-   && ((TYPE) == 0 || ! TREE_ADDRESSABLE ((tree)(TYPE)))	\
-   && (! TARGET_SH3E || (ROUND_REG((CUM), (MODE)) + (GET_MODE_SIZE(MODE)/4) <= NPARM_REGS (MODE))))
+  (((TYPE) == 0 || ! TREE_ADDRESSABLE ((tree)(TYPE))) \
+   && (TARGET_SH3E \
+       ? ((MODE) == BLKmode \
+	  ? (((CUM).arg_count[(int) SH_ARG_INT] * UNITS_PER_WORD \
+	      + int_size_in_bytes (TYPE)) \
+	     <= NPARM_REGS (SImode) * UNITS_PER_WORD) \
+	  : ((ROUND_REG((CUM), (MODE)) \
+	      + HARD_REGNO_NREGS (BASE_ARG_REG (MODE), (MODE))) \
+	     <= NPARM_REGS (MODE))) \
+       : ROUND_REG ((CUM), (MODE)) < NPARM_REGS (MODE)))
 
 /* Define where to put the arguments to a function.
    Value is zero to push the argument on the stack,
@@ -1755,9 +1762,14 @@ sh_valid_machine_decl_attribute (DECL, ATTRIBUTES, IDENTIFIER, ARGS)
     (LENGTH) = align_length (X);				\
   if (GET_CODE (X) == JUMP_INSN					\
       && GET_CODE (PATTERN (X)) == ADDR_DIFF_VEC)		\
-    /* The code before an ADDR_DIFF_VEC is even aligned, thus	\
-       any odd estimate is wrong.  */				\
-    (LENGTH) &= ~1;
+    {								\
+      /* The code before an ADDR_DIFF_VEC is even aligned,	\
+	 thus any odd estimate is wrong.  */			\
+      (LENGTH) &= ~1;						\
+      /* If not optimizing, the alignment is implicit.  */	\
+      if (! optimize)						\
+	(LENGTH) += 2;						\
+    }
 
 /* Enable a bug fix for the shorten_branches pass.  */
 #define SHORTEN_WITH_ADJUST_INSN_LENGTH
@@ -1859,4 +1871,7 @@ do {									\
 /* For the sake of libgcc2.c, indicate target supports atexit.  */
 #define HAVE_ATEXIT
 
+/* Enable the register move pass to improve code.  */
+#define ENABLE_REGMOVE_PASS
+
 #define SH_DYNAMIC_SHIFT_COST (TARGET_SH3 ? (TARGET_SMALLCODE ? 1 : 2) : 20)

gcc/ginclude/va-sh.h

@@ -23,9 +23,6 @@ typedef struct {
 
 typedef void *__gnuc_va_list;
 
-#define __va_rounded_size(TYPE)  \
-  (((sizeof (TYPE) + sizeof (int) - 1) / sizeof (int)) * sizeof (int))
-
 #endif /* ! SH3E */
 
 #endif /* __GNUC_VA_LIST */
@@ -116,104 +113,83 @@ enum __va_type_classes {
 #endif
 #define va_end(pvar)	((void)0)
 
+#ifdef __LITTLE_ENDIAN__
+#define __LITTLE_ENDIAN_P 1
+#else
+#define __LITTLE_ENDIAN_P 0
+#endif
+
+#define __SCALAR_TYPE(TYPE)					\
+  ((TYPE) == __integer_type_class				\
+   || (TYPE) == __char_type_class				\
+   || (TYPE) == __enumeral_type_class)
+
 /* RECORD_TYPE args passed using the C calling convention are
    passed by invisible reference.  ??? RECORD_TYPE args passed
    in the stack are made to be word-aligned; for an aggregate that is
    not word-aligned, we advance the pointer to the first non-reg slot.  */
 
+  /* When this is a smaller-than-int integer, using
+     auto-increment in the promoted (SImode) is fastest;
+     however, there is no way to express that is C.  Therefore,
+     we use an asm.
+     We want the MEM_IN_STRUCT_P bit set in the emitted RTL, therefore we
+     use unions even when it would otherwise be unnecessary.  */
+
+#define __va_arg_sh1(AP, TYPE) __extension__ 				\
+__extension__								\
+({(sizeof (TYPE) == 1							\
+   ? ({union {TYPE t; char c;} __t;					\
+       asm(""								\
+	   : "=r" (__t.c)						\
+	   : "0" ((((union { int i, j; } *) (AP))++)->i));		\
+       __t.t;})								\
+   : sizeof (TYPE) == 2							\
+   ? ({union {TYPE t; short s;} __t;					\
+       asm(""								\
+	   : "=r" (__t.s)						\
+	   : "0" ((((union { int i, j; } *) (AP))++)->i));		\
+       __t.t;})								\
+   : sizeof (TYPE) >= 4 || __LITTLE_ENDIAN_P				\
+   ? (((union { TYPE t; int i;} *) (AP))++)->t				\
+   : ((union {TYPE t;TYPE u;}*) ((char *)++(int *)(AP) - sizeof (TYPE)))->t);})
+
 #ifdef __SH3E__
 
-#ifdef __LITTLE_ENDIAN__
+#define __PASS_AS_FLOAT(TYPE_CLASS,SIZE) \
+  (TYPE_CLASS == __real_type_class && SIZE == 4)
 
 #define va_arg(pvar,TYPE)					\
 __extension__							\
-(*({int __type = __builtin_classify_type (* (TYPE *) 0);	\
-  void * __result;						\
-  if (__type == __real_type_class && sizeof(TYPE) == 4)		\
-						/* float? */	\
+({int __type = __builtin_classify_type (* (TYPE *) 0);		\
+  void * __result_p;						\
+  if (__PASS_AS_FLOAT (__type, sizeof(TYPE)))			\
     {								\
-      __va_freg *__r;						\
       if (pvar.__va_next_fp < pvar.__va_next_fp_limit)		\
-	__r = (__va_freg *) pvar.__va_next_fp++;		\
-      else							\
-	__r = (__va_freg *) pvar.__va_next_stack++;		\
-      __result = (char *) __r;					\
-    }								\
-  else								\
-    {								\
-      __va_greg *_r;						\
-      if (pvar.__va_next_o + ((sizeof (TYPE) + 3) / 4)		\
-	  <= pvar.__va_next_o_limit) 				\
 	{							\
-	  _r = pvar.__va_next_o;				\
-	  pvar.__va_next_o += (sizeof (TYPE) + 3) / 4;		\
+	  __result_p = &pvar.__va_next_fp;			\
 	}							\
       else							\
-	{							\
-	  _r = pvar.__va_next_stack;				\
-	  pvar.__va_next_stack += (sizeof (TYPE) + 3) / 4;	\
-	}							\
-      __result = (char *) _r;					\
-    } 								\
-  (TYPE *) __result;}))
-
-#else /* ! __LITTLE_ENDIAN__ */
-
-#define va_arg(pvar,TYPE)					\
-__extension__							\
-(*({int __type = __builtin_classify_type (* (TYPE *) 0);	\
-  void * __result;						\
-  if (__type == __real_type_class && sizeof(TYPE) == 4)		\
-						/* float? */	\
-    {								\
-      __va_freg *__r;						\
-      if (pvar.__va_next_fp < pvar.__va_next_fp_limit)		\
-	__r = (__va_freg *) pvar.__va_next_fp++;		\
-      else							\
-	__r = (__va_freg *) pvar.__va_next_stack++;		\
-      __result = (char *) __r;					\
+	__result_p = &pvar.__va_next_stack;			\
     }								\
   else								\
     {								\
-      __va_greg *_r;						\
       if (pvar.__va_next_o + ((sizeof (TYPE) + 3) / 4)		\
 	  <= pvar.__va_next_o_limit) 				\
-	{							\
-	  pvar.__va_next_o += (sizeof (TYPE) + 3) / 4;		\
-	  _r = pvar.__va_next_o;				\
-	}							\
+	__result_p = &pvar.__va_next_o;				\
       else							\
 	{							\
-	  pvar.__va_next_stack += (sizeof (TYPE) + 3) / 4;	\
-	  _r = pvar.__va_next_stack;				\
+	  if (sizeof (TYPE) > 4)				\
+	    pvar.__va_next_o = pvar.__va_next_o_limit;		\
+								\
+	  __result_p = &pvar.__va_next_stack;			\
 	}							\
-      __result = ((char *) _r					\
-		  - (sizeof (TYPE) < 4 ? sizeof (TYPE)		\
-		     : ((sizeof (TYPE) + 3) / 4) * 4));		\
     } 								\
-  (TYPE *) __result;}))
-
-#endif /* __LITTLE_ENDIAN__ */
+  __va_arg_sh1(*(void **)__result_p, TYPE);})
 
 #else /* ! SH3E */
 
-#ifdef __LITTLE_ENDIAN__
-
-/* This is for little-endian machines; small args are padded upward.  */
-#define va_arg(AP, TYPE)						\
- (AP = (__gnuc_va_list) ((char *) (AP) + __va_rounded_size (TYPE)),	\
-  *((TYPE *) (void *) ((char *) (AP) - __va_rounded_size (TYPE))))
-
-#else /* ! __LITTLE_ENDIAN__ */
-
-/* This is for big-endian machines; small args are padded downward.  */
-#define va_arg(AP, TYPE)						\
- (AP = (__gnuc_va_list) ((char *) (AP) + __va_rounded_size (TYPE)),	\
-  *((TYPE *) (void *) ((char *) (AP)					\
-		       - ((sizeof (TYPE) < __va_rounded_size (char)	\
-			   ? sizeof (TYPE) : __va_rounded_size (TYPE))))))
-
-#endif /* __LITTLE_ENDIAN__ */
+#define va_arg(AP, TYPE) __va_arg_sh1((AP), TYPE)
 
 #endif /* SH3E */