Build and pass all test cases on a multiprocessor HP-UX.

* configure.ac, check/Makefile.am, doc/Makefile.am: Do not
	explicitly link to -ldl, but instead autodetect the library
	with dlopen, dlsym, etc.

	* check/lightning.c: Add workaround to apparently buggy
	getopt in HP-UX that sets optind to the wrong index, and
	use RTLD_NEXT on HP-UX instead of RTLD_DEFAULT to dlsym
	global symbols.

	* include/lightning.h: Rework definitions of wordsize and
	byte order to detect proper values on HP-UX.

	* lib/lightning.c: Minor correction to use MAP_ANONYMOUS
	instead of MAP_ANON on HP-UX.

	* lib/jit_hppa.c: Float arguments must be passed on integer
	registers on HP-UX, not only for varargs functions.
	  Add code to properly clear instruction cache. This was
	not required on Debian hppa port, but may have been working
	by accident.

	* lib/jit_hppa-cpu.c: Follow pattern of HP-UX binaries and
	use bve,n instead of bv,n to return from functions.

	* lib/jit_hppa-fpu.c: For some reason "fst? frX,rX,(rY)" did
	not work on the tested computer	(HP-UX B.11.23 U 9000/785 HP-UX)
	so the code was changed, at first for __hpux only to add the
	base and offset register and use the instruction with an
	immediate (zero) offset.

ChangeLog

@@ -1,4 +1,36 @@
-2013-07-01 Paulo Andrade <pcpa@gnu.org>
+2013-06-09 Paulo Andrade <pcpa@gnu.org>
+
+	* configure.ac, check/Makefile.am, doc/Makefile.am: Do not
+	explicitly link to -ldl, but instead autodetect the library
+	with dlopen, dlsym, etc.
+
+	* check/lightning.c: Add workaround to apparently buggy
+	getopt in HP-UX that sets optind to the wrong index, and
+	use RTLD_NEXT on HP-UX instead of RTLD_DEFAULT to dlsym
+	global symbols.
+
+	* include/lightning.h: Rework definitions of wordsize and
+	byte order to detect proper values on HP-UX.
+
+	* lib/lightning.c: Minor correction to use MAP_ANONYMOUS
+	instead of MAP_ANON on HP-UX.
+
+	* lib/jit_hppa.c: Float arguments must be passed on integer
+	registers on HP-UX, not only for varargs functions.
+	  Add code to properly clear instruction cache. This was
+	not required on Debian hppa port, but may have been working
+	by accident.
+
+	* lib/jit_hppa-cpu.c: Follow pattern of HP-UX binaries and
+	use bve,n instead of bv,n to return from functions.
+
+	* lib/jit_hppa-fpu.c: For some reason "fst? frX,rX,(rY)" did
+	not work on the tested computer	(HP-UX B.11.23 U 9000/785 HP-UX)
+	so the code was changed, at first for __hpux only to add the
+	base and offset register and use the instruction with an
+	immediate (zero) offset.
+
+2013-06-07 Paulo Andrade <pcpa@gnu.org>
 
 	* check/lightning.c, lib/jit_disasm.c, lib/jit_ppc-cpu.c,
 	lib/jit_ppc-fpu.c, lib/jit_ppc.c, include/lightning.h,

check/Makefile.am

@@ -16,10 +16,10 @@ AM_CFLAGS = -I$(top_srcdir)/include -D_GNU_SOURCE
 
 check_PROGRAMS = lightning ccall
 
-lightning_LDADD = $(top_builddir)/lib/liblightning.la -lm -ldl
+lightning_LDADD = $(top_builddir)/lib/liblightning.la -lm $(SHLIB)
 lightning_SOURCES = lightning.c
 
-ccall_LDADD = $(top_builddir)/lib/liblightning.la -lm -ldl
+ccall_LDADD = $(top_builddir)/lib/liblightning.la -lm $(SHLIB)
 ccall_SOURCES = ccall.c
 
 $(top_builddir)/lib/liblightning.la:

check/lightning.c

@@ -38,6 +38,12 @@
 #  define label_t		l_label_t
 #endif
 
+#if defined(__hpux)
+#  define DL_HANDLE		RTLD_NEXT
+#else
+#  define DL_HANDLE		RTLD_DEFAULT
+#endif
+
 #if defined(__GNUC__)
 #  define noreturn		__attribute__ ((noreturn))
 #  define printf_format(f, v)	__attribute__ ((format (printf, f, v)))
@@ -2295,7 +2301,7 @@ dynamic(void)
     char	*string;
     (void)identifier('@');
     if ((label = get_label_by_name(parser.string)) == NULL) {
-	value = dlsym(RTLD_DEFAULT, parser.string + 1);
+	value = dlsym(DL_HANDLE, parser.string + 1);
 	if ((string = dlerror()))
 	    error("%s", string);
 	label = new_label(label_kind_dynamic, parser.string, value);
@@ -3939,6 +3945,11 @@ main(int argc, char *argv[])
 #endif
 
     opt_index = optind;
+#if defined(__hpux)
+    /* Workaround */
+    if (opt_index < argc && argv[opt_index][0] == '-')
+	++opt_index;
+#endif
     if (opt_index < 0 || opt_index >= argc)
 	usage();
     if (strcmp(argv[opt_index], "-") == 0)

configure.ac

@@ -68,6 +68,12 @@ else
     DEBUG=no
 fi
 
+case "$host_os" in
+    *hpux*)		SHLIB="-ldld"	;;
+    *)			SHLIB="-ldl"	;;
+esac
+AC_SUBST(SHLIB)
+
 cpu=
 case "$target_cpu" in
     i?86|x86_64)	cpu=x86		;;

doc/Makefile.am

@@ -24,17 +24,17 @@ noinst_PROGRAMS = incr printf rpn rfib ifib
 $(top_builddir)/lib/liblightning.la:
 	cd $(top_builddir)/lib; $(MAKE) $(AM_MAKEFLAGS) liblightning.la
 
-incr_LDADD = $(top_builddir)/lib/liblightning.la -lm -ldl
+incr_LDADD = $(top_builddir)/lib/liblightning.la -lm $(SHLIB)
 incr_SOURCES = incr.c
 
-printf_LDADD = $(top_builddir)/lib/liblightning.la -lm -ldl
+printf_LDADD = $(top_builddir)/lib/liblightning.la -lm $(SHLIB)
 printf_SOURCES = printf.c
 
-rpn_LDADD = $(top_builddir)/lib/liblightning.la -lm -ldl
+rpn_LDADD = $(top_builddir)/lib/liblightning.la -lm $(SHLIB)
 rpn_SOURCES = rpn.c
 
-rfib_LDADD = $(top_builddir)/lib/liblightning.la -lm -ldl
+rfib_LDADD = $(top_builddir)/lib/liblightning.la -lm $(SHLIB)
 rfib_SOURCES = rfib.c
 
-ifib_LDADD = $(top_builddir)/lib/liblightning.la -lm -ldl
+ifib_LDADD = $(top_builddir)/lib/liblightning.la -lm $(SHLIB)
 ifib_SOURCES = ifib.c

include/lightning.h

@@ -27,32 +27,48 @@
 #include <stdint.h>
 #include <string.h>
 
-#ifndef __WORDSIZE
+#ifdef __hpux
-#  if defined(_AIX)
+#  include <machine/param.h>
-#    define __WORDSIZE		(__SIZEOF_POINTER__ << 3)
-#  else
-#    define __WORDSIZE		WORDSIZE
-#  endif
 #endif
-#ifndef __BYTE_ORDER
+
-#  if defined(_AIX)
+#ifndef __WORDSIZE
-#    define __BYTE_ORDER	__BYTE_ORDER__
+#  if defined(WORDSIZE)
+#    define __WORDSIZE		WORDSIZE
+#  elif defined(__SIZEOF_POINTER__)
+#    define __WORDSIZE		(__SIZEOF_POINTER__ << 3)
+#  elif defined(_ILP32)
+#    define __WORDSIZE		32
 #  else
-#    define __BYTE_ORDER	BYTE_ORDER
+#    error cannot figure __WORDSIZE
 #  endif
 #endif
 #ifndef __LITTLE_ENDIAN
-#  if defined(_AIX)
+#  if defined(LITTLE_ENDIAN)
+#    define __LITTLE_ENDIAN	LITTLE_ENDIAN
+#  elif defined(__ORDER_LITTLE_ENDIAN__)
 #    define __LITTLE_ENDIAN	__ORDER_LITTLE_ENDIAN__
 #  else
-#    define __LITTLE_ENDIAN	LITTLE_ENDIAN
+#    define __LITTLE_ENDIAN	1234
 #  endif
 #endif
 #ifndef __BIG_ENDIAN
-#  if defined(_AIX)
+#  if defined(BIG_ENDIAN)
+#    define __BIG_ENDIAN	BIG_ENDIAN
+#  elif defined(__ORDER_BIG_ENDIAN__)
 #    define __BIG_ENDIAN	__ORDER_BIG_ENDIAN__
 #  else
-#    define __BIG_ENDIAN	BIG_ENDIAN
+#    define __BIG_ENDIAN	4321
+#  endif
+#endif
+#ifndef __BYTE_ORDER
+#  if defined(BYTE_ORDER)
+#    define __BYTE_ORDER	BYTE_ORDER
+#  elif defined(__BYTE_ORDER__)
+#    define __BYTE_ORDER	__BYTE_ORDER__
+#  elif defined(_BIG_ENDIAN)
+#    define __BYTE_ORDER	__BIG_ENDIAN
+#  else
+#    error cannot figure __BYTE_ORDER
 #  endif
 #endif
 

lib/jit_hppa-cpu.c

@@ -322,6 +322,7 @@ static void _f38(jit_state_t*,jit_int32_t,
 #define BV(x,b)			f21(0x3a,b,x,6,0)
 #define BV_N(x,b)		f21(0x3a,b,x,6,1)
 #define BVE(b)			f22(0x3a,b,6,0,0,0)
+#define BVE_N(b)		f22(0x3a,b,6,0,1,0)
 #define BVE_L(b)		f22(0x3a,b,7,0,0,0)
 #define II_C_NONE		0
 #define II_C_M			(1<<5)
@@ -2652,7 +2653,11 @@ _epilog(jit_state_t *_jit, jit_node_t *node)
     LDWL(-20, _FP_REGNO, _RP_REGNO);
     LDO(_jitc->function->stack, _FP_REGNO, _SP_REGNO);
     LDWL_MB(-_jitc->function->stack, _SP_REGNO, _FP_REGNO);
+#if defined(__hpux)
+    BVE_N(_RP_REGNO);
+#else
     BV_N(_R0_REGNO, _RP_REGNO);
+#endif
 }
 
 static void

lib/jit_hppa-fpu.c

@@ -16,6 +16,15 @@
  */
 
 #if PROTO
+
+/* FIXME should actually be hw model/version/etc or other constraint
+ * that causes a SIGSEGV if using these instructions */
+#if defined(__hpux)
+#  define FSTXR				0
+#else
+#  define FSTXR				1
+#endif
+
 #define f39(o,b,x,t)			_f39(_jit,o,b,x,t)
 static void _f39(jit_state_t*,jit_int32_t,jit_int32_t,jit_int32_t,jit_int32_t);
 #define f40(o,b,x,r)			_f40(_jit,o,b,x,r)
@@ -385,13 +394,20 @@ static void _ldxi_d(jit_state_t*,jit_int32_t,jit_int32_t,jit_word_t);
 #define str_f(r0,r1)			FSTWI(r1,0,r0)
 #define sti_f(i0,r0)			_sti_f(_jit,i0,r0)
 static void _sti_f(jit_state_t*,jit_word_t,jit_int32_t);
-#define stxr_f(r0,r1,r2)		FSTW(r2,r1,r0)
+#if FSTXR
+#  define stxr_f(r0,r1,r2)		FSTW(r2,r1,r0)
+#  define stxr_d(r0,r1,r2)		FSTD(r2,r1,r0)
+#else
+#  define stxr_f(r0,r1,r2)		_stxr_f(_jit,r0,r1,r2)
+static void _stxr_f(jit_state_t*,jit_int32_t,jit_int32_t,jit_int32_t);
+#  define stxr_d(r0,r1,r2)		_stxr_d(_jit,r0,r1,r2)
+static void _stxr_d(jit_state_t*,jit_int32_t,jit_int32_t,jit_int32_t);
+#endif
 #define stxi_f(i0,r0,r1)		_stxi_f(_jit,i0,r0,r1)
 static void _stxi_f(jit_state_t*,jit_word_t,jit_int32_t,jit_int32_t);
 #define str_d(r0,r1)			FSTDI(r1,0,r0)
 #define sti_d(i0,r0)			_sti_d(_jit,i0,r0)
 static void _sti_d(jit_state_t*,jit_word_t,jit_int32_t);
-#define stxr_d(r0,r1,r2)		FSTD(r2,r1,r0)
 #define stxi_d(i0,r0,r1)		_stxi_d(_jit,i0,r0,r1)
 static void _stxi_d(jit_state_t*,jit_word_t,jit_int32_t,jit_int32_t);
 #define bcmpr_f(c,i0,r0,r1)		_bcmpr_f(_jit,c,i0,r0,r1)
@@ -813,6 +829,28 @@ _sti_f(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0)
     }
 }
 
+#if !FSTXR
+static void
+_stxr_f(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_int32_t r2)
+{
+    jit_int32_t		reg;
+    reg = jit_get_reg(jit_class_gpr);
+    addr(rn(reg), r0, r1);
+    str_f(rn(reg), r2);
+    jit_unget_reg(reg);
+}
+
+static void
+_stxr_d(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_int32_t r2)
+{
+    jit_int32_t		reg;
+    reg = jit_get_reg(jit_class_gpr);
+    addr(rn(reg), r0, r1);
+    str_d(rn(reg), r2);
+    jit_unget_reg(reg);
+}
+#endif
+
 static void
 _stxi_f(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, jit_int32_t r1)
 {
@@ -824,8 +862,13 @@ _stxi_f(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, jit_int32_t r1)
 	FSTWL(r1, i0, r0);
     else {
 	reg = jit_get_reg(jit_class_gpr);
+#if FSTXR
+	movi(rn(reg), i0);
+	stxr_f(rn(reg), r0, r1);
+#else
 	addi(rn(reg), r0, i0);
 	str_f(rn(reg), r1);
+#endif
 	jit_unget_reg(reg);
     }
 }
@@ -856,8 +899,13 @@ _stxi_d(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, jit_int32_t r1)
 	FSTDL(r1, i0, r0);
     else {
 	reg = jit_get_reg(jit_class_gpr);
+#if FSTXR
 	movi(rn(reg), i0);
 	stxr_d(rn(reg), r0, r1);
+#else
+	addi(rn(reg), r0, i0);
+	str_d(rn(reg), r1);
+#endif
 	jit_unget_reg(reg);
     }
 }

lib/jit_hppa.c

@@ -32,6 +32,9 @@
 #define patch(instr, node)		_patch(_jit, instr, node)
 static void _patch(jit_state_t*,jit_word_t,jit_node_t*);
 
+/* libgcc */
+extern void __clear_cache(void *, void *);
+
 /*
  * Initialization
  */
@@ -149,7 +152,7 @@ _jit_prolog(jit_state_t *_jit)
     _jitc->function->self.size = params_offset;
     _jitc->function->self.argi = _jitc->function->self.alen = 0;
     /* float conversion */
-    _jitc->function->self.aoff = alloca_offset - 8;
+    _jitc->function->self.aoff = alloca_offset;
     _jitc->function->self.call = jit_call_default;
     jit_alloc((jit_pointer_t *)&_jitc->function->regoff,
 	      _jitc->reglen * sizeof(jit_int32_t));
@@ -409,7 +412,10 @@ _jit_pushargr_f(jit_state_t *_jit, jit_int32_t u)
     _jitc->function->call.size -= sizeof(jit_word_t);
     if (_jitc->function->call.argi < 4) {
 	jit_movr_f(_F4 - _jitc->function->call.argi, u);
-	if (_jitc->function->call.call & jit_call_varargs) {
+#if !defined(__hpux)
+	if (_jitc->function->call.call & jit_call_varargs)
+#endif
+	{
 	    jit_stxi_f(alloca_offset - 8, JIT_FP, u);
 	    jit_ldxi(_R26 - _jitc->function->call.argi, JIT_FP,
 		     alloca_offset - 8);
@@ -428,7 +434,10 @@ _jit_pushargi_f(jit_state_t *_jit, jit_float32_t u)
     _jitc->function->call.size -= sizeof(jit_word_t);
     if (_jitc->function->call.argi < 4) {
 	jit_movi_f(_F4 - _jitc->function->call.argi, u);
-	if (_jitc->function->call.call & jit_call_varargs) {
+#if !defined(__hpux)
+	if (_jitc->function->call.call & jit_call_varargs)
+#endif
+	{
 	    jit_stxi_f(alloca_offset - 8, JIT_FP,
 		       _F4 - _jitc->function->call.argi);
 	    jit_ldxi(_R26 - _jitc->function->call.argi,
@@ -455,7 +464,10 @@ _jit_pushargr_d(jit_state_t *_jit, jit_int32_t u)
     }
     if (_jitc->function->call.argi < 4) {
 	jit_movr_d(_F4 - (_jitc->function->call.argi + 1), u);
-	if (_jitc->function->call.call & jit_call_varargs) {
+#if !defined(__hpux)
+	if (_jitc->function->call.call & jit_call_varargs)
+#endif
+	{
 	    jit_stxi_d(alloca_offset - 8, JIT_FP, u);
 	    jit_ldxi(_R26 - _jitc->function->call.argi,
 		     JIT_FP, alloca_offset - 4);
@@ -484,7 +496,10 @@ _jit_pushargi_d(jit_state_t *_jit, jit_float64_t u)
     }
     if (_jitc->function->call.argi < 4) {
 	jit_movi_d(_F4 - (_jitc->function->call.argi + 1), u);
-	if (_jitc->function->call.call & jit_call_varargs) {
+#if !defined(__hpux)
+	if (_jitc->function->call.call & jit_call_varargs)
+#endif
+	{
 	    jit_stxi_d(alloca_offset - 8, JIT_FP,
 		       _F4 - (_jitc->function->call.argi + 1));
 	    jit_ldxi(_R26 - _jitc->function->call.argi,
@@ -1131,6 +1146,83 @@ _emit_code(jit_state_t *_jit)
 	patch_at(_jitc->patches.ptr[offset].inst, word);
     }
 
+#if defined(__hppa)
+/* --- parisc2.0.pdf ---
+		Programming Note
+
+The minimum spacing that is guaranteed to work for "self-modifying code" is
+shown in the code segment below. Since instruction prefetching is permitted,
+any data cache flushes must be separated from any instruction cache flushes
+by a SYNC. This will ensure that the "new" instruction will be written to
+memory prior to any attempts at prefetching it as an instruction.
+
+	LDIL	l%newinstr,rnew
+	LDW	r%newinstr(0,rnew),temp
+	LDIL	l%instr,rinstr
+	STW	temp,r%instr(0,rinstr)
+	FDC	r%instr(0,rinstr)
+	SYNC	
+	FIC	r%instr(rinstr)
+	SYNC
+	instr	...
+	(at least seven instructions)
+
+This sequence assumes a uniprocessor system. In a multiprocessor system,
+software must ensure no processor is executing code which is in the process
+of being modified.
+*/
+    
+/*
+  Adapted from ffcall/trampoline/cache-hppa.c:__TR_clear_cache to
+loop over addresses as it is unlikely from and to addresses would fit in
+at most two cachelines.
+ */
+    /*
+     * Copyright 1995-1997 Bruno Haible, <bruno@clisp.org>
+     *
+     * This is free software distributed under the GNU General Public Licence
+     * described in the file COPYING. Contact the author if you don't have this
+     * or can't live with it. There is ABSOLUTELY NO WARRANTY, explicit or implied,
+     * on this software.
+     */
+    {
+	/* FIXME this may be required on Linux or other OSes with
+	 * multiprocessor support (was not required for the hppa
+	 * port done on Debian hppa...) */
+	jit_word_t	f = (jit_word_t)_jit->code.ptr;
+	jit_word_t	t = f + _jit->code.length;
+	register int	u, v;
+	for (; f <= t; f += 32) {
+	    asm volatile ("fdc 0(0,%0)"
+			  "\n\t" "fdc 0(0,%1)"
+			  "\n\t" "sync"
+			  :
+			  : "r" (f), "r" (t)
+			  );
+	    asm volatile ("mfsp %%sr0,%1"
+			  "\n\t" "ldsid (0,%4),%0"
+			  "\n\t" "mtsp %0,%%sr0"
+			  "\n\t" "fic 0(%%sr0,%2)"
+			  "\n\t" "fic 0(%%sr0,%3)"
+			  "\n\t" "sync"
+			  "\n\t" "mtsp %1,%%sr0"
+			  "\n\t" "nop"
+			  "\n\t" "nop"
+			  "\n\t" "nop"
+			  "\n\t" "nop"
+			  "\n\t" "nop"
+			  "\n\t" "nop"
+			  : "=r" (u), "=r" (v)
+			  : "r" (f), "r" (t), "r" (f)
+			  );
+	}
+    }
+#else
+    /* This is supposed to work but appears to fail on multiprocessor systems */
+    word = sysconf(_SC_PAGE_SIZE);
+    __clear_cache(_jit->code.ptr, (void *)((_jit->pc.w + word) & -word));
+#endif
+
     return (_jit->code.ptr);
 }
 

lib/lightning.c

@@ -19,6 +19,10 @@
 #include <lightning/jit_private.h>
 #include <sys/mman.h>
 
+#ifndef MAP_ANON
+#  define MAP_ANON			MAP_ANONYMOUS
+#endif
+
 #define jit_regload_reload		0	/* convert to reload */
 #define jit_regload_delete		1	/* just remove node */
 #define jit_regload_isdead		2	/* delete and unset live bit */