Add inline assembler hacks for restoring the $gp register on the alpha,

Estimated hours taken: 24

Add inline assembler hacks for restoring the $gp register on the alpha,
to make the `asm_fast.gc' grade work on the alpha.

runtime/goto.h:
	For the alpha, change the GOTO macro so that it sets up
	register $27 (the "procedure value" register) to contain the
	address to which you are going to jump, and change the
	Define_{entry,local,static} macros so that they have a `ldgp'
	instruction after the label which restores the gp register from
	$27.

runtime/call.mod:
	For the alpha, change the call macro so that it uses a `ldgp'
	instruction to restore the gp register on return from a call.
	Add new call_localret macro, for the case when the continuation
	label is a local label (this is like the inverse of the localcall
	macro), because a call_localret is two instructions cheaper
	than the equivalent call would be.

runtime/engine.mod:
	In call_engine(), we need to use `Define_label(engine_done)'
	rather than just `engine_done:', so that it works on the alpha.

These changes improve the speed of the compiler by about 21%,
and cut executable size by about 24%.

There's still a fair bit of room for improving efficiency by
avoiding moves to $27 and `ldgp' instructions in situations
where they aren't needed.

runtime/engine.mod

@@ -80,7 +80,7 @@ static void init_registers(void)
 /* The gcc-specific version */
 
 void call_engine(Code *entry_point)
-{
+{{
 	/*
 	** Allocate some space for local variables in other
 	** procedures. This used to be done by just calling
@@ -134,7 +134,7 @@ void call_engine(Code *entry_point)
 
 	noprof_call(entry_point, LABEL(engine_done));
 
-engine_done:
+Define_label(engine_done);
 	/*
 	** Save any registers which will get clobbered by the normal
 	** C function call / return mechanism
@@ -180,7 +180,7 @@ engine_done:
 	}
 #endif
 
-}
+}}
 
 /* with nonlocal gotos, we don't save the previous locations */
 void dump_prev_locations(void) {}

runtime/goto.h

@@ -35,6 +35,7 @@
 #define paste(a,b) a##b
 #define stringify(string) #string
 #define entry(label) paste(entry_,label)
+#define skip(label) paste(skip_,label)
 
 #ifdef SPLIT_C_FILES
 #define MODULE_STATIC_OR_EXTERN extern
@@ -42,6 +43,84 @@
 #define MODULE_STATIC_OR_EXTERN static
 #endif
 
+/*---------------------------------------------------------------------------*/
+
+/* MACHINE SPECIFIC STUFF REQUIRED FOR NON-LOCAL GOTOS */
+
+#if defined(__alpha__)
+
+  /* when doing a jump, we need to set $27, the "procedure value" register,
+     to the address we are jumping to, so that we can use an `ldgp'
+     instruction in ASM_ENTRY to set up the right gp value.
+  */
+  #define ASM_JUMP(address)				\
+	__asm__("bis %0, %0, $27\n\t" 			\
+		: : "r"(address) : "$27");		\
+	goto *(address)
+	/* Explanation:
+		Move `address' to register $27
+		Jump to `address'
+	*/
+
+  /* on entry to a procedure, we need to load the $gp register
+     with the correct value relative to the current address in $27 */
+  #define ASM_ENTRY(label) \
+	/* on fall-thru, we need to skip the ldgp instruction */ \
+  	goto skip(label);				\
+	entry(label):					\
+	__asm__ __volatile__ (				\
+		".globl entry_" stringify(label) "\n"	\
+		"entry_" stringify(label) ":\n\t" 	\
+		"ldgp $gp, 0($27)"			\
+		: : : "memory"				\
+	);						\
+	skip(label):
+  /* even static entry points must fix up the $gp register, since
+     although there won't be any direct calls to them from another
+     C file, their address may be taken and so there may be indirect
+     calls */
+  #define ASM_STATIC_ENTRY(label) \
+	/* on fall-thru, we need to skip the ldgp instruction */ \
+  	goto skip(label);				\
+	entry(label):					\
+	__asm__ __volatile__ (				\
+		"entry_" stringify(label) ":\n\t" 	\
+		"ldgp $gp, 0($27)"			\
+		: : : "memory"				\
+	);						\
+	skip(label):
+  /* even local entry points must fix up the $gp register, since
+     although there won't be any direct calls to them from another
+     C file, their address may be taken and so there may be indirect
+     calls */
+  #define ASM_LOCAL_ENTRY(label) \
+	/* on fall-thru, we need to skip the ldgp instruction */ \
+  	goto skip(label);				\
+	entry(label):					\
+	__asm__ __volatile__ (				\
+		"ldgp $gp, 0($27)"			\
+		: : : "memory"				\
+	);						\
+	skip(label):
+#else
+
+  #define ASM_JUMP(label)	goto *(label)
+  #define ASM_ENTRY(label) 				\
+  	entry(label):					\
+	__asm__(".globl entry_" stringify(label) "\n\t"	\
+		"entry_" stringify(label) ":" 		\
+		);
+  #define ASM_STATIC_ENTRY(label) 			\
+  	entry(label):					\
+	__asm__ (					\
+		"entry_" stringify(label) ":" 		\
+		);
+  #define ASM_LOCAL_ENTRY(label) 			\
+  	entry(label): ;
+#endif
+
+/*---------------------------------------------------------------------------*/
+
 #if defined(USE_GCC_NONLOCAL_GOTOS)
 
   #ifndef __GNUC__
@@ -97,22 +176,24 @@
   /* body of module goes here */
   #define END_MODULE } }
 
+
   #if defined(USE_ASM_LABELS)
     #define Declare_entry(label)	\
 	extern void label(void) __asm__("entry_" stringify(label))
     #define Declare_static(label)	\
 	static void label(void) __asm__("entry_" stringify(label))
     #define Define_extern_entry(label)	Declare_entry(label)
-    #define Define_entry(label)	\
-	}	\
-	label:	\
-		__asm__(".globl entry_" stringify(label) "\n"	\
-			"entry_" stringify(label) ":");	\
+    #define Define_entry(label)		\
+		ASM_ENTRY(label)	\
+	}				\
+	label:				\
+	PRETEND_ADDRESS_IS_USED(&&entry(label));	\
 	{
     #define Define_static(label)	\
-	}	\
-	label:	\
-		__asm__("entry_" stringify(label) ":");	\
+		ASM_STATIC_ENTRY(label) \
+	}				\
+	label:				\
+	PRETEND_ADDRESS_IS_USED(&&entry(label));	\
 	{
     /*
        The PRETEND_ADDRESS_IS_USED macro is necessary to 
@@ -125,6 +206,8 @@
 
     #define ENTRY(label) 	(&label)
 
+    #define JUMP(label)		ASM_JUMP(label)
+
   #else
     /* !defined(USE_ASM_LABELS) */
 
@@ -144,24 +227,24 @@
 	entry(label) = &&label
     #define ENTRY(label) 	(entry(label))
 
+    #define JUMP(label)		goto (*label)
+
   #endif
 
   #define Declare_local(label)	/* no declaration required */
   #define Define_local(label)	\
+  		ASM_LOCAL_ENTRY(label)	\
 	}	\
 	label:	\
 	{
   #define init_local(label)	make_local(stringify(label), &&label)
   #define Declare_label(label)	/* no declaration required */
-  #define Define_label(label)	\
-	}	\
-	label:	\
-	{
+  #define Define_label(label)	Define_local(label)
   #define init_label(label)	make_label(stringify(label), &&label)
 
-  #define LOCAL(label)		(&&label)
-  #define LABEL(label)		(&&label)
-  #define GOTO(label)		do { debuggoto(label); goto *(label); } while(0)
+  #define LOCAL(label)		(&&entry(label))
+  #define LABEL(label)		(&&entry(label))
+  #define GOTO(label)		do { debuggoto(label); JUMP(label); } while(0)
   #define GOTO_ENTRY(label) 	GOTO(ENTRY(label))
   #define GOTO_LOCAL(label) 	GOTO_LABEL(label)
   #define GOTO_LABEL(label) 	do { debuggoto(&&label); goto label; } while(0)

runtime/imp.h

@@ -66,6 +66,37 @@ typedef void (*Cont) (void);
 			GOTO_LABEL(label);			\
 		} while (0)
 
+#if defined(__alpha__) && defined(USE_ASM_LABELS)
+#define	noprof_call(proc, succ_cont)				\
+		({						\
+			__label__ fixup_gp;			\
+			debugcall((proc), (succ_cont));		\
+			succip = (&&fixup_gp);			\
+			set_prof_current_proc(proc);		\
+			GOTO(proc);				\
+		fixup_gp:					\
+			__asm__ __volatile__ (			\
+				"ldgp $gp, 0($27)"		\
+				: : : "memory"			\
+			);					\
+			GOTO(succ_cont); 			\
+		})
+	/* same as above, but with GOTO_LABEL rather than GOTO */
+#define	noprof_call_localret(proc, succ_cont)			\
+		({						\
+			__label__ fixup_gp;			\
+			debugcall((proc), (succ_cont));		\
+			succip = (&&fixup_gp);			\
+			set_prof_current_proc(proc);		\
+			GOTO(proc);				\
+		fixup_gp:					\
+			__asm__ __volatile__ (			\
+				"ldgp $gp, 0($27)"		\
+				: : : "memory"			\
+			);					\
+			GOTO_LABEL(succ_cont); 			\
+		})
+#else
 #define	noprof_call(proc, succ_cont)				\
 		do {						\
 			debugcall((proc), (succ_cont));		\
@@ -73,6 +104,9 @@ typedef void (*Cont) (void);
 			set_prof_current_proc(proc);		\
 			GOTO(proc);				\
 		} while (0)
+#define noprof_call_localret(proc, succ_cont) 			\
+		noprof_call((proc), LABEL(succ_cont))
+#endif
 
 #define	localcall(label, succ_cont, current_label)		\
 		do {						\
@@ -85,11 +119,14 @@ typedef void (*Cont) (void);
 
 #define	call(proc, succ_cont, current_label)			\
 		do {						\
-			debugcall((proc), (succ_cont));		\
-			succip = (succ_cont);			\
 			PROFILE((proc), (current_label));	\
-			set_prof_current_proc(proc);		\
-			GOTO(proc);				\
+			noprof_call((proc), (succ_cont));	\
+		} while (0)
+
+#define	call_localret(proc, succ_cont, current_label)		\
+		do {						\
+			PROFILE((proc), (current_label));	\
+			noprof_call_localret(proc, succ_cont); \
 		} while (0)
 
 #define	call_det_closure(succ_cont, current_label)		\