Add support to test different/alternate code generation setups.

* check/lightning.c: Remove the ".cpu name value" syntax,
	as it was not able to do proper changes before the jit
	internal data structure was initialized. Now it supports
	several getopt options to force using different jit
	generation options, effectively replacing the previous
	syntax.

	* check/run-test: Add simple extra logic to handle differently
	named test scripts, used to test things like x87 coprocessor
	in ix86, and arm instruction set or software float in armv7l.

	* configure.ac: Add some AC_RUN_IFELSE calls to figure at
	compile time if can test different code generation options,
	and update Makefile generation accordingly.

	* check/Makefile.am, lib/jit_arm.c, lib/jit_x86.c: Update to
	properly work with the test tool updating the jit_cpu global
	information.

	* check/check.arm.sh, check/check.swf.sh, check/check.x87.sh:
	New wrapper files passing -mthumb=0, mvfp=0 and -mx87=1 to
	the test tool, if applicable, so that it can validate alternate
	code generation options on test hosts that support them.

ChangeLog

@@ -1,3 +1,29 @@
+2012-12-14 Paulo Andrade <pcpa@gnu.org>
+
+	* check/lightning.c: Remove the ".cpu name value" syntax,
+	as it was not able to do proper changes before the jit
+	internal data structure was initialized. Now it supports
+	several getopt options to force using different jit
+	generation options, effectively replacing the previous
+	syntax.
+
+	* check/run-test: Add simple extra logic to handle differently
+	named test scripts, used to test things like x87 coprocessor
+	in ix86, and arm instruction set or software float in armv7l.
+
+	* configure.ac: Add some AC_RUN_IFELSE calls to figure at
+	compile time if can test different code generation options,
+	and update Makefile generation accordingly.
+
+	* check/Makefile.am, lib/jit_arm.c, lib/jit_x86.c: Update to
+	properly work with the test tool updating the jit_cpu global
+	information.
+
+	* check/check.arm.sh, check/check.swf.sh, check/check.x87.sh:
+	New wrapper files passing -mthumb=0, mvfp=0 and -mx87=1 to
+	the test tool, if applicable, so that it can validate alternate
+	code generation options on test hosts that support them.
+
 2012-12-14 Paulo Andrade <pcpa@gnu.org>
 
 	* lib/jit_x86-x87.c, lib/jit_x86.c: Correct test cases in ix86

check/Makefile.am

@@ -58,10 +58,13 @@ EXTRA_DIST =				\
 	alu_com.tst	alu_com.ok	\
 	alu_neg.tst	alu_neg.ok	\
 	varargs.tst	varargs.ok	\
-	check.sh	run-test	\
-	all.tst
+	check.sh			\
+	check.x87.sh			\
+	check.arm.sh	check.swf.sh	\
+	run-test	all.tst
 
-TESTS = 3to2 add allocai		\
+base_TESTS =				\
+	3to2 add allocai		\
 	bp divi fib rpn			\
 	ldstr ldsti			\
 	ldstxr ldstxi			\
@@ -75,12 +78,79 @@ TESTS = 3to2 add allocai		\
 	alu_com alu_neg			\
 	varargs
 
-CLEANFILES = $(TESTS)
-
-# Not so pretty but good for a prototype
-$(TESTS):	check.sh
+$(base_TESTS):	check.sh
 	$(LN_S) $(srcdir)/check.sh $@
 
+TESTS = $(base_TESTS)
+
+if test_x86_x87
+x87_TESTS =					\
+	3to2.x87 add.x87 allocai.x87		\
+	bp.x87 divi.x87 fib.x87 rpn.x87		\
+	ldstr.x87 ldsti.x87			\
+	ldstxr.x87 ldstxi.x87			\
+	ldstr-c.x87 ldstxr-c.x87 ldstxi-c.x87	\
+	cvt.x87 branch.x87			\
+	alu_add.x87 alux_add.x87		\
+	alu_sub.x87 alux_sub.x87		\
+	alu_mul.x87 alu_div.x87 alu_rem.x87	\
+	alu_and.x87 alu_or.x87 alu_xor.x87	\
+	alu_lsh.x87 alu_rsh.x87			\
+	alu_com.x87 alu_neg.x87			\
+	varargs.x87
+
+$(x87_TESTS):	check.x87.sh
+	$(LN_S) $(srcdir)/check.x87.sh $@
+
+TESTS += $(x87_TESTS)
+endif
+
+if test_arm_arm
+arm_TESTS =					\
+	3to2.arm add.arm allocai.arm		\
+	bp.arm divi.arm fib.arm rpn.arm		\
+	ldstr.arm ldsti.arm			\
+	ldstxr.arm ldstxi.arm			\
+	ldstr-c.arm ldstxr-c.arm ldstxi-c.arm	\
+	cvt.arm branch.arm			\
+	alu_add.arm alux_add.arm		\
+	alu_sub.arm alux_sub.arm		\
+	alu_mul.arm alu_div.arm alu_rem.arm	\
+	alu_and.arm alu_or.arm alu_xor.arm	\
+	alu_lsh.arm alu_rsh.arm			\
+	alu_com.arm alu_neg.arm			\
+	varargs.arm
+
+$(arm_TESTS):	check.arm.sh
+	$(LN_S) $(srcdir)/check.arm.sh $@
+
+TESTS += $(arm_TESTS)
+endif
+
+if test_arm_swf
+swf_TESTS =					\
+	3to2.swf add.swf allocai.swf		\
+	bp.swf divi.swf fib.swf rpn.swf		\
+	ldstr.swf ldsti.swf			\
+	ldstxr.swf ldstxi.swf			\
+	ldstr-c.swf ldstxr-c.swf ldstxi-c.swf	\
+	cvt.swf branch.swf			\
+	alu_add.swf alux_add.swf		\
+	alu_sub.swf alux_sub.swf		\
+	alu_mul.swf alu_div.swf alu_rem.swf	\
+	alu_and.swf alu_or.swf alu_xor.swf	\
+	alu_lsh.swf alu_rsh.swf			\
+	alu_com.swf alu_neg.swf			\
+	varargs.swf
+
+$(swf_TESTS):	check.swf.sh
+	$(LN_S) $(srcdir)/check.swf.sh $@
+
+TESTS += $(swf_TESTS)
+endif
+
+CLEANFILES = $(TESTS)
+
 TESTS_ENVIRONMENT=$(srcdir)/run-test
 
 debug:		$(check_PROGRAMS)

check/check.arm.sh

@@ -0,0 +1,2 @@
+#!/bin/sh
+./lightning -mthumb=0 $srcdir/`basename $0 | sed -e 's|\.arm$||'`.tst

check/check.swf.sh

@@ -0,0 +1,2 @@
+#!/bin/sh
+./lightning -mvfp=0 $srcdir/`basename $0 | sed -e 's|\.swf$||'`.tst

check/check.x87.sh

@@ -0,0 +1,2 @@
+#!/bin/sh
+./lightning -mx87=1 $srcdir/`basename $0 | sed -e 's|\.x87$||'`.tst

check/lightning.c

@@ -2061,46 +2061,6 @@ dot(void)
 	check_data(length);
 	data_offset += length;
     }
-    else if (strcmp(parser.string, "cpu") == 0) {
-	if (primary(skip_ws) != tok_symbol)
-	    error("expecting cpu flag");
-	ch = get_int(skip_ws);
-	if (strcmp(parser.string, "sse2") == 0)
-#if defined(__i386__)
-	    /* only meaningful for i386 as there is no x87 path for x86_64
-	     * and should only use just after jit_prolog and not mix with
-	     * code that uses xmm registers */
-	    jit_cpu.sse2 = !!ch
-#endif
-	;
-	else if (strcmp(parser.string, "sse4_1") == 0)
-#if defined(__i386__) || defined(__x86_64__)
-	    jit_cpu.sse4_1 = !!ch
-#endif
-	;
-	else if (strcmp(parser.string, "version") == 0)
-#if defined(__arm__)
-	    jit_cpu.version = ch
-#endif
-	;
-	else if (strcmp(parser.string, "thumb") == 0)
-#if defined(__arm__)
-	    jit_cpu.thumb = ch
-#endif
-	;
-	else if (strcmp(parser.string, "vfp") == 0)
-#if defined(__arm__)
-	    jit_cpu.vfp = ch
-#endif
-	;
-	else if (strcmp(parser.string, "neon") == 0)
-#if defined(__arm__)
-	    jit_cpu.neon = !!ch
-#endif
-	;
-	else
-	    warn("ignoring \".cpu %s %d\"", parser.string, ch);
-    }
     else if (strcmp(parser.string, "disasm") == 0)
 	flag_disasm = 1;
     else
@@ -3776,21 +3736,46 @@ static void
 usage(void)
 {
     fprintf(stderr, "\
-Usage: %s [jit-assembler-options] file [jit-program-options]\n\
-Options:\n\
+Usage: %s [jit assembler options] file [jit program options]\n\
+Jit assembler options:\n\
   -help                    Display this information\n\
   -v[0-3]                  Verbose output level\n\
   -D<macro>[=<val>]        Preprocessor options\n"
+#if defined(__i386__) && __WORDSIZE == 32
+"  -mx87=1                  Force using x87 when sse2 available\n"
+#endif
+#if defined(__i386__) || defined(__x86_64__)
+"  -msse4_1=0               Do not use sse4_1 instructions when available\n"
+#endif
+#if defined(__arm__)
+"  -mcpu=<val>              Force cpu version (4, 5, 6 or 7)\n\
+  -mthumb[=0|1]            Enable or disable thumb\n\
+  -mvfp=<val>              Set vpf version (0 to disable)\n\
+  -mneon[=0|1]             Enable or disable neon\n"
+#endif
 	    , progname);
+    finish_jit();
     exit(1);
 }
 
 int
 main(int argc, char *argv[])
 {
-    static const char	*short_options = "v::";
+    static const char	*short_options = "v:";
     static struct option long_options[] = {
 	{ "help",		0, 0, 'h' },
+#if defined(__i386__) && __WORDSIZE == 32
+	{ "mx87",		1, 0, '7' },
+#endif
+#if defined(__i386__) || defined(__x86_64__)
+	{ "msse4_1",		1, 0, '4' },
+#endif
+#if defined(__arm__)
+	{ "mcpu",		1, 0, 'c' },
+	{ "mthumb",		1, 0, 't' },
+	{ "mvfp",		1, 0, 'f' },
+	{ "mneon",		1, 0, 'n' },
+#endif
 	{ 0,			0, 0, 0   }
     };
     int			 offset;
@@ -3799,6 +3784,8 @@ main(int argc, char *argv[])
     int			 opt_short;
     char		 cmdline[8192];
 
+    init_jit();
+
     progname = argv[0];
     for (;;) {
 	if ((opt_short = getopt_long_only(argc, argv, short_options,
@@ -3818,6 +3805,67 @@ main(int argc, char *argv[])
 		else
 		    flag_verbose = 1;
 		break;
+#if defined(__i386__) && __WORDSIZE == 32
+	    case '7':
+		if (optarg) {
+		    if (strcmp(optarg, "") == 0 || strcmp(optarg, "1") == 0)
+			jit_cpu.sse2 = 0;
+		    else if (strcmp(optarg, "0"))
+			usage();
+		}
+		else
+		    jit_cpu.sse2 = 0;
+		break;
+#endif
+#if defined(__i386__) || defined(__x86_64__)
+	    case '4':
+		if (optarg) {
+		    if (strcmp(optarg, "0") == 0)
+			jit_cpu.sse4_2 = 0;
+		    else if (strcmp(optarg, "1"))
+			usage();
+		}
+		break;
+#endif
+#if defined(__arm__)
+	    case 'c':
+		if (optarg) {
+		    offset = strtol(optarg, &endptr, 10);
+		    if (*endptr || offset < 0)
+			usage();
+		    if (offset < jit_cpu.version)
+			jit_cpu.version = offset;
+		}
+		break;
+	    case 't':
+		if (optarg) {
+		    if (strcmp(optarg, "0") == 0)
+			jit_cpu.thumb = 0;
+		    else if (strcmp(optarg, "1") && strcmp(optarg, "2"))
+			usage();
+		}
+		break;
+	    case 'f':
+#  if !defined(__ARM_PCS_VFP)
+		/* Do not allow overrinding hard float abi */
+		if (optarg) {
+		    offset = strtol(optarg, &endptr, 10);
+		    if (*endptr || offset < 0)
+			usage();
+		    if (offset < jit_cpu.vfp)
+			jit_cpu.vfp = offset;
+		}
+#  endif
+		break;
+	    case 'n':
+		if (optarg) {
+		    if (strcmp(optarg, "0") == 0)
+			jit_cpu.neon = 0;
+		    else if (strcmp(optarg, "1"))
+			usage();
+		}
+		break;
+#endif
 	}
     }
 
@@ -3899,7 +3947,6 @@ main(int argc, char *argv[])
     }
 #endif
 
-    init_jit();
     _jit = jit_new_state();
 
     instrs = new_hash();

check/run-test

@@ -1,9 +1,10 @@
 #! /bin/sh
 
+ok=`echo $1 | sed -e 's@\.\(x87\|arm\|swf\)@@'`
 $1 | tr -d \\r > $1.log
-if cmp -s $srcdir/$1.ok $1.log; then
+if cmp -s $srcdir/$ok.ok $1.log; then
   rm $1.log
 else
-  diff $srcdir/$1.ok $1.log
+  diff $srcdir/$ok.ok $1.log
   exit 1
 fi

configure.ac

@@ -67,16 +67,104 @@ case "$target_cpu" in
     *powerpc*)		cpu=ppc		;;
     *)					;;
 esac
-if test x$cpu = x; then
-    AC_MSG_ERROR([cpu $target_cpu not supported])
-fi
 AM_CONDITIONAL(cpu_arm,  [test cpu-$cpu = cpu-arm])
 AM_CONDITIONAL(cpu_mips, [test cpu-$cpu = cpu-mips])
 AM_CONDITIONAL(cpu_ppc,  [test cpu-$cpu = cpu-ppc])
 AM_CONDITIONAL(cpu_x86,  [test cpu-$cpu = cpu-x86])
-if test $cpu = arm; then
-    AC_CHECK_LIB(m, sqrtf, ,
-	[AC_MSG_ERROR([sqrtf required but not available])])
+
+# Test x87 if both, x87 and sse2 available
+ac_cv_test_x86_x87=
+# Test arm instruction set if thumb instruction set available
+ac_cv_test_arm_arm=
+# Test sofware float if vfp available and not using hard float abi
+ac_cv_test_arm_swf=
+
+save_CFLAGS=$CFLAGS
+CFLAGS="-I$PWD/include -D_GNU_SOURCE"
+if test x$cpu = x; then
+    AC_MSG_ERROR([cpu $target_cpu not supported])
+elif test $cpu = x86; then
+    AC_RUN_IFELSE([AC_LANG_SOURCE([[
+    #include <lightning.h>
+    int main(void) {
+	int		    ac, flags;
+	unsigned int	    eax, ebx, ecx, edx;
+	if (__WORDSIZE == 64)
+	    return 1;
+	__asm__ volatile ("pushfl;\n\t"
+			  "popl %0;\n\t"
+			  "movl \$0x240000, %1;\n\t"
+			  "xorl %0, %1;\n\t"
+			  "pushl %1;\n\t"
+			  "popfl;\n\t"
+			  "pushfl;\n\t"
+			  "popl %1;\n\t"
+			  "xorl %0, %1;\n\t"
+			  "pushl %0;\n\t"
+			  "popfl"
+			  : "=r" (flags), "=r" (ac));
+	if ((ac & (1 << 21)) == 0)
+	    return 1;
+	__asm__ volatile ("xchgl %%ebx, %1; cpuid; xchgl %%ebx, %1"
+			  : "=a" (eax), "=r" (ebx),
+			  "=c" (ecx), "=d" (edx)
+			  : "0" (1));
+	return (edx & 1 << 26) ? 0 : 1;
+    }
+    ]])],[ac_cv_test_x86_x87=yes],[][])
+elif test $cpu = arm; then
+    AC_RUN_IFELSE([AC_LANG_SOURCE([[
+    #include <stdio.h>
+    int main(void) {
+    #if defined(__linux__)
+	FILE	*fp;
+	char	 buf[128];
+	if ((fp = fopen("/proc/cpuinfo", "r")) == NULL)
+	    return 1;
+	while (fgets(buf, sizeof(buf), fp)) {
+	    if (strncmp(buf, "Features\t:", 10) == 0 &&
+		strstr(buf + 10, "thumb")) {
+		fclose(fp);
+		return 0;
+	    }
+	}
+	fclose(fp);
+    #endif
+	return 1;
+    }
+    ]])],[ac_cv_test_arm_arm=yes],[][])
+    AC_RUN_IFELSE([AC_LANG_SOURCE([[
+    #include <stdio.h>
+    int main(void) {
+    #if defined(__linux__)
+	FILE	*fp;
+	char	 buf[128];
+    #  if !defined(__ARM_PCS_VFP)
+	if ((fp = fopen("/proc/cpuinfo", "r")) == NULL)
+	    return 1;
+	while (fgets(buf, sizeof(buf), fp)) {
+	    if (strncmp(buf, "Features\t:", 10) == 0 &&
+		strstr(buf + 10, "vfp")) {
+		fclose(fp);
+		return 0;
+	    }
+	}
+	fclose(fp);
+    #  endif
+    #endif
+	return 1;
+    }
+    ]])],[ac_cv_test_arm_swf=yes],[][])
+fi
+CFLAGS=$save_CFLAGS
+
+AM_CONDITIONAL(test_x86_x87, [test x$ac_cv_test_x86_x87 = xyes])
+AM_CONDITIONAL(test_arm_arm, [test x$ac_cv_test_arm_arm = xyes])
+AM_CONDITIONAL(test_arm_swf, [test x$ac_cv_test_arm_swf = xyes])
+
+if test $cpu=arm; then
+     AC_CHECK_LIB(m, sqrtf, ,
+        [AC_MSG_ERROR([sqrtf required but not available])])
 fi
 
 AC_SUBST([LIGHTNING_CFLAGS])

lib/jit_arm.c

@@ -185,17 +185,22 @@ void
 _jit_init(jit_state_t *_jit)
 {
     jit_int32_t		regno;
+    static jit_bool_t	first = 1;
+
     _jit->reglen = jit_size(_rvs) - 1;
-    /* jit_get_cpu() should have been already called, and only once */
-    if (!jit_cpu.vfp) {
-	/* cause register to never be allocated, because simple
-	 * software float only allocates stack space for 8 slots  */
-	for (regno = _D8; regno < _D7; regno++)
-	    _rvs[regno].spec = 0;
-    }
-    if (!jit_cpu.abi) {
-	for (regno = _S15; regno <= _D0; regno++)
-	    _rvs[regno].spec &= ~rc(arg);
+    if (first) {
+	/* jit_get_cpu() should have been already called, and only once */
+	if (!jit_cpu.vfp) {
+	    /* cause register to never be allocated, because simple
+	     * software float only allocates stack space for 8 slots  */
+	    for (regno = _D8; regno < _D7; regno++)
+		_rvs[regno].spec = 0;
+	}
+	if (!jit_cpu.abi) {
+	    for (regno = _S15; regno <= _D0; regno++)
+		_rvs[regno].spec &= ~rc(arg);
+	}
+	first = 0;
     }
 }
 

lib/jit_x86.c

@@ -268,15 +268,19 @@ jit_get_cpu(void)
 void
 _jit_init(jit_state_t *_jit)
 {
+    jit_int32_t		regno;
+    static jit_bool_t	first = 1;
+
     _jit->reglen = jit_size(_rvs) - 1;
 #if __WORDSIZE == 32
-    if (!jit_cpu.sse2) {
-	jit_int32_t	regno;
-
-	for (regno = _jit->reglen; regno >= 0; regno--) {
-	    if (_rvs[regno].spec & jit_class_xpr)
-		_rvs[regno].spec = 0;
+    if (first) {
+	if (!jit_cpu.sse2) {
+	    for (regno = _jit->reglen; regno >= 0; regno--) {
+		if (_rvs[regno].spec & jit_class_xpr)
+		    _rvs[regno].spec = 0;
+	    }
 	}
+	first = 0;
     }
 #endif
 }