add JIT_NEED_PUSH_POP and merge correct implementation of push/pop for SPARC

2006-11-20  Paolo Bonzini  <bonzini@gnu.org>
	    Ludovic Courtes  <ludo@chbouib.org>

	* lightning/i386/core-32.h: Disable jit_push and jit_pop if stack not needed.
	* lightning/i386/core-64.h: Disable jit_push and jit_pop if stack not needed.
	* lightning/sparc/core.h: Merge final implementation of jit_pushr and jit_popr.
	* lightning/ppc/core.h: Fix implementation of jit_pushr and jit_popr to
	work (more or less) across function calls.

	* tests/push-pop.c, tests/push-pop.ok: New test.
	* tests/Makefile.am: Run it.

Patches applied:

 * lcourtes@laas.fr--2005-libre/lightning--stable--1.2--patch-32
   Added missing SPARC changes from lightning@sv.gnu.org/lightning--stable--1.2--patch-10.

 * lcourtes@laas.fr--2005-libre/lightning--stable--1.2--patch-33
   SPARC: Fixed pushr/popr using %l7 to hold the pushed register offset.

 * lcourtes@laas.fr--2005-libre/lightning--stable--1.2--patch-34
   tests/push-pop.c: define `JIT_NEED_PUSH_POP'.

git-archimport-id: bonzini@gnu.org--2004b/lightning--stable--1.2--patch-40

ChangeLog

@@ -1,3 +1,15 @@
+2006-11-20  Paolo Bonzini  <bonzini@gnu.org>
+	    Ludovic Courtes  <ludo@chbouib.org>
+
+	* lightning/i386/core-32.h: Disable jit_push and jit_pop if stack not needed.
+	* lightning/i386/core-64.h: Disable jit_push and jit_pop if stack not needed.
+	* lightning/sparc/core.h: Merge final implementation of jit_pushr and jit_popr.
+	* lightning/ppc/core.h: Fix implementation of jit_pushr and jit_popr to
+	work (more or less) across function calls.
+
+	* tests/push-pop.c, tests/push-pop.ok: New test.
+	* tests/Makefile.am: Run it.
+
 2006-11-20  Paolo Bonzini  <bonzini@gnu.org>
 
 	* lightning/asm-common.h: Make 64-bit safe.

doc/using.texi

@@ -768,16 +768,16 @@ variables, and pass the address around.
 typedef int (*pifi)(int);       @rem{/* Pointer to Int Function of Int */}
 
 void stack_push(int reg, int *sp)
-{
+@{
   jit_stxi_i (*sp, JIT_FP, reg);
   *sp += sizeof (int);
-}
+@}
 
 void stack_pop(int reg, int *sp)
-{
+@{
   *sp -= sizeof (int);
   jit_ldxi_i (reg, JIT_FP, *sp);
-}
+@}
 
 pifi compile_rpn(char *expr)
 @{

lightning/i386/core-32.h

@@ -63,8 +63,10 @@ struct jit_local_state {
    _jitl.alloca_offset -= (amount))
    
 /* Stack */
+#ifdef JIT_NEED_PUSH_POP
 #define jit_pushr_i(rs)		PUSHLr(rs)
 #define jit_popr_i(rs)		POPLr(rs)
+#endif
 
 /* The += in argssize allows for stack pollution */
 

lightning/i386/core-64.h

@@ -106,8 +106,10 @@ struct jit_local_state {
 #define jit_rshr_ul(d, r1, r2)	jit_replace((r1), (r2), _ECX, 				jit_qop_ ((d), (r1), SHRQrr(_CL,  (d)) ))
 
 /* Stack */
+#ifdef JIT_NEED_PUSH_POP
 #define jit_pushr_i(rs)		PUSHQr(rs)
 #define jit_popr_i(rs)		POPQr(rs)
+#endif
 
 #define jit_base_prolog() (PUSHQr(_EBP), MOVQrr(_ESP, _EBP), PUSHQr(_EBX), PUSHQr(_R12), PUSHQr(_R13))
 #define jit_prolog(n) (_jitl.nextarg_geti = 0, _jitl.alloca_offset = -24, jit_base_prolog())

lightning/ppc/core.h

@@ -257,12 +257,16 @@ struct jit_local_state {
 #define jit_nop()			NOP()
 #define jit_ori_i(d, rs, is)		jit_chk_imu((is), ORIrri((d), (rs), (is)), ORrrr((d), (rs), JIT_AUX))
 #define jit_orr_i(d, s1, s2)				  ORrrr((d), (s1), (s2))
-#define jit_popr_i(rs)			(LWZrm((rs), 0, 1), ADDIrri(1, 1, 4))
+
+#ifdef JIT_NEED_PUSH_POP
+#define jit_popr_i(rs)			(LWZrm((rs), 8, 1), ADDIrri(1, 1, 4))
+#define jit_pushr_i(rs)			(STWrm((rs), 4, 1), ADDIrri (1, 1, -4))
+#endif
+
 #define jit_prepare_i(numi)		(_jitl.nextarg_puti = numi)
 #define jit_prepare_f(numf)		(_jitl.nextarg_putf = numf)
 #define jit_prepare_d(numd)		(_jitl.nextarg_putd = numd)
 #define jit_prolog(n)			_jit_prolog(&_jit, (n))
-#define jit_pushr_i(rs)			STWUrm((rs), -4, 1)
 #define jit_pusharg_i(rs)		(--_jitl.nextarg_puti, MRrr((3 + _jitl.nextarg_putd * 2 + _jitl.nextarg_putf + _jitl.nextarg_puti), (rs)))
 #define jit_ret()			_jit_epilog(&_jit)
 #define jit_retval_i(rd)		MRrr((rd), 3)

lightning/sparc/core.h

@@ -34,9 +34,14 @@
 #define __lightning_core_h
 
 #define JIT_R_NUM		3
-#define JIT_V_NUM		6
+#ifdef JIT_NEED_PUSH_POP
-#define JIT_R(i)		((i) ? _Rl((i) - 1) : _Rg(2))
+# define JIT_V_NUM		7
-#define JIT_V(i)		_Rl((i)+2)
+#else
+# define JIT_V_NUM		8
+#endif
+
+#define JIT_R(i)		_Rg(2 + (i))
+#define JIT_V(i)		_Rl(i)
 
 #define JIT_BIG			_Rg(1)	/* %g1 used to make 32-bit operands */
 #define JIT_BIG2		_Ro(7)	/* %o7 used to make 32-bit compare operands */
@@ -62,7 +67,7 @@
  */
 
 
-/* Implementation of `allocai'.
+/* Implementation of `allocai', `pushr' and `popr'.
  *
  * The SysV ABI for SPARC is specified in "System V Application Binary
  * Interface, SPARC Processor Supplement, Third Edition", available from
@@ -77,14 +82,20 @@
  *
  * Stack space may be allocated dynamically as decribed in Section
  * "Allocating Stack Space Dynamically", p. 3-36, and shown in Figure 3-49.
- * `allocai' is implementing by patching a function prolog's `save'
+ * `allocai' is implemented by patching a function prolog's `save'
- * instruction in order to increase the initial frame size.  Thus,
+ * instruction in order to increase the initial frame size.  `pushr' is
- * %fp and below is used for the memory allocated via `allocai'.  */
+ * implemented by storing register into a fixed-size region right below %fp
+ * (thereby limiting the number of registers that may be pushed).
+ *
+ * Thus, %fp to %fp - JIT_SPARC_MAX_STACK_REGISTER_AREA is reserved for
+ * pushed registers, and %fp - JIT_SPARC_MAX_STACK_REGISTER_AREA and below is
+ * used for the memory allocated via `allocai'.  */
 
 
 struct jit_local_state {
   int	nextarg_put;	/* Next %o reg. to be written */
   int	nextarg_get;	/* Next %i reg. to be read */
+
   jit_insn *save;	/* Pointer to the `save' instruction */
   unsigned  frame_size;	/* Current frame size as allocated by `save' */
   int       alloca_offset; /* Current offset to the alloca'd memory (negative
@@ -92,8 +103,24 @@ struct jit_local_state {
   jit_insn delay;
 };
 
+#ifdef JIT_NEED_PUSH_POP
+/* Maximum size for the "automatic variables" area on the stack (the area
+   that starts from %fp-1 and ends at %sp+104, see the ABI doc).  This must
+   be a multiple of 8 so that %sp remains double-word aligned as required by
+   SysV ABI (see page 3-10).  In lightning, this effectively limits the
+   number of registers that can be pushed with `pushr'.
+
+   Initialize %l7 to contain the (negative) offset relative to %fp of the last
+   register pushed.  */
+# define JIT_SPARC_MAX_STACK_REGISTER_AREA  (8 * sizeof (void *))
+# define JIT_SPARC_INIT_PUSH_POP()          (MOVrr (_Rg(0), _Rl(7)))
+#else
+# define JIT_SPARC_MAX_STACK_REGISTER_AREA  0
+# define JIT_SPARC_INIT_PUSH_POP()          ((void)0)
+#endif
+
 /* Minimum size of a stack frame.  */
-#define JIT_SPARC_MIN_FRAME_SIZE  104
+#define JIT_SPARC_MIN_FRAME_SIZE  (104 + JIT_SPARC_MAX_STACK_REGISTER_AREA)
 
 
 /* Round AMOUNT to the closest higher multiple of 2^ALIGNMENT.  */
@@ -288,10 +315,9 @@ struct jit_local_state {
 #define jit_ori_i(d, rs, is)		jit_chk_imm((is), ORrir((rs), (is), (d)), ORrrr((rs), JIT_BIG, (d)))
 #define jit_orr_i(d, s1, s2)				  ORrrr((s1), (s2), (d))
 #define jit_patch_at(delay_pc, pv)	jit_patch_ (((delay_pc) - 1) , (pv))
-#define jit_popr_i(rs)			(LDmr(JIT_SP, 0, (rs)), ADDrir(JIT_SP, 8, JIT_SP))
 #define jit_prepare_i(num)		(_jitl.nextarg_put += (num))
-#define jit_prolog(numargs)		(_jitl.save = (jit_insn *) _jit.x.pc, SAVErir (JIT_SP, -JIT_SPARC_MIN_FRAME_SIZE, JIT_SP), _jitl.frame_size = JIT_SPARC_MIN_FRAME_SIZE, _jitl.alloca_offset = 0, _jitl.nextarg_get = _Ri(0), _jitl.next_push = 0)
+#define jit_prolog(numargs)		(_jitl.save = (jit_insn *) _jit.x.pc, SAVErir (JIT_SP, -JIT_SPARC_MIN_FRAME_SIZE, JIT_SP), _jitl.frame_size = JIT_SPARC_MIN_FRAME_SIZE, _jitl.alloca_offset = -JIT_SPARC_MAX_STACK_REGISTER_AREA, _jitl.nextarg_get = _Ri(0), JIT_SPARC_INIT_PUSH_POP ())
-#define jit_pushr_i(rs)			(STrm((rs), JIT_SP, -8), SUBrir(JIT_SP, 8, JIT_SP))
+
 #define jit_pusharg_i(rs)		(--_jitl.nextarg_put, MOVrr((rs), _Ro(_jitl.nextarg_put)))
 #define jit_ret()			(RET(), RESTORE())
 #define jit_retval_i(rd)		MOVrr(_Ro(0), (rd))
@@ -312,4 +338,9 @@ struct jit_local_state {
 #define jit_xori_i(d, rs, is)		jit_chk_imm((is), XORrir((rs), (is), (d)), XORrrr((rs), JIT_BIG, (d)))
 #define jit_xorr_i(d, s1, s2)				  XORrrr((s1), (s2), (d))
 
+#ifdef JIT_NEED_PUSH_POP
+# define jit_pushr_i(rs)		(SUBrir(_Rl(7), sizeof (void *), _Rl(7)), STWrx((rs), JIT_FP, _Rl(7)))
+# define jit_popr_i(rd)			(LDSWxr(_Rl(7), JIT_FP, (rd)), ADDrir(_Rl(7), sizeof (void *), _Rl(7)))
+#endif
+
 #endif /* __lightning_core_h */

tests/Makefile.am

@@ -2,12 +2,12 @@ AM_CPPFLAGS = -I$(top_builddir) -I$(top_srcdir) -I$(top_srcdir)/lightning/$(cpu)
 
 check_PROGRAMS = fibit incr printf printf2 rpn fib fibdelay	\
 	add bp testfp funcfp rpnfp modi ldxi divi movi ret	\
-	allocai
+	allocai push-pop
 
 noinst_DATA = fibit.ok incr.ok printf.ok printf2.ok rpn.ok	\
 	fib.ok fibdelay.ok testfp.ok funcfp.ok rpnfp.ok add.ok	\
 	bp.ok modi.ok ldxi.ok divi.ok movi.ok ret.ok		\
-	allocai.ok
+	allocai.ok push-pop.ok
 
 EXTRA_DIST = $(noinst_DATA) run-test
 
@@ -17,7 +17,8 @@ endif
 
 if REGRESSION_TESTING
 TESTS = fib fibit fibdelay incr printf printf2 rpn add bp	\
-	testfp funcfp rpnfp modi ldxi divi movi ret allocai
+	testfp funcfp rpnfp modi ldxi divi movi ret allocai	\
+	push-pop
 
 TESTS_ENVIRONMENT=$(srcdir)/run-test
 endif

tests/Makefile.in

@@ -41,7 +41,7 @@ check_PROGRAMS = fibit$(EXEEXT) incr$(EXEEXT) printf$(EXEEXT) \
 	printf2$(EXEEXT) rpn$(EXEEXT) fib$(EXEEXT) fibdelay$(EXEEXT) \
 	add$(EXEEXT) bp$(EXEEXT) testfp$(EXEEXT) funcfp$(EXEEXT) \
 	rpnfp$(EXEEXT) modi$(EXEEXT) ldxi$(EXEEXT) divi$(EXEEXT) \
-	movi$(EXEEXT) ret$(EXEEXT) allocai$(EXEEXT)
+	movi$(EXEEXT) ret$(EXEEXT) allocai$(EXEEXT) push-pop$(EXEEXT)
 subdir = tests
 DIST_COMMON = $(srcdir)/Makefile.am $(srcdir)/Makefile.in
 ACLOCAL_M4 = $(top_srcdir)/aclocal.m4
@@ -111,6 +111,11 @@ printf2_OBJECTS = printf2.$(OBJEXT)
 printf2_LDADD = $(LDADD)
 @DISASS_TRUE@printf2_DEPENDENCIES =  \
 @DISASS_TRUE@	$(top_builddir)/opcode/libdisass.a
+push_pop_SOURCES = push-pop.c
+push_pop_OBJECTS = push-pop.$(OBJEXT)
+push_pop_LDADD = $(LDADD)
+@DISASS_TRUE@push_pop_DEPENDENCIES =  \
+@DISASS_TRUE@	$(top_builddir)/opcode/libdisass.a
 ret_SOURCES = ret.c
 ret_OBJECTS = ret.$(OBJEXT)
 ret_LDADD = $(LDADD)
@@ -135,11 +140,11 @@ COMPILE = $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) \
 CCLD = $(CC)
 LINK = $(CCLD) $(AM_CFLAGS) $(CFLAGS) $(AM_LDFLAGS) $(LDFLAGS) -o $@
 SOURCES = add.c allocai.c bp.c divi.c fib.c fibdelay.c fibit.c \
-	funcfp.c incr.c ldxi.c modi.c movi.c printf.c printf2.c ret.c \
+	funcfp.c incr.c ldxi.c modi.c movi.c printf.c printf2.c \
-	rpn.c rpnfp.c testfp.c
+	push-pop.c ret.c rpn.c rpnfp.c testfp.c
 DIST_SOURCES = add.c allocai.c bp.c divi.c fib.c fibdelay.c fibit.c \
-	funcfp.c incr.c ldxi.c modi.c movi.c printf.c printf2.c ret.c \
+	funcfp.c incr.c ldxi.c modi.c movi.c printf.c printf2.c \
-	rpn.c rpnfp.c testfp.c
+	push-pop.c ret.c rpn.c rpnfp.c testfp.c
 DATA = $(noinst_DATA)
 ETAGS = etags
 CTAGS = ctags
@@ -248,12 +253,13 @@ AM_CPPFLAGS = -I$(top_builddir) -I$(top_srcdir) -I$(top_srcdir)/lightning/$(cpu)
 noinst_DATA = fibit.ok incr.ok printf.ok printf2.ok rpn.ok	\
 	fib.ok fibdelay.ok testfp.ok funcfp.ok rpnfp.ok add.ok	\
 	bp.ok modi.ok ldxi.ok divi.ok movi.ok ret.ok		\
-	allocai.ok
+	allocai.ok push-pop.ok
 
 EXTRA_DIST = $(noinst_DATA) run-test
 @DISASS_TRUE@LDADD = $(top_builddir)/opcode/libdisass.a
 @REGRESSION_TESTING_TRUE@TESTS = fib fibit fibdelay incr printf printf2 rpn add bp	\
-@REGRESSION_TESTING_TRUE@	testfp funcfp rpnfp modi ldxi divi movi ret allocai
+@REGRESSION_TESTING_TRUE@	testfp funcfp rpnfp modi ldxi divi movi ret allocai	\
+@REGRESSION_TESTING_TRUE@	push-pop
 
 @REGRESSION_TESTING_TRUE@TESTS_ENVIRONMENT = $(srcdir)/run-test
 all: all-am
@@ -334,6 +340,9 @@ printf$(EXEEXT): $(printf_OBJECTS) $(printf_DEPENDENCIES)
 printf2$(EXEEXT): $(printf2_OBJECTS) $(printf2_DEPENDENCIES) 
 	@rm -f printf2$(EXEEXT)
 	$(LINK) $(printf2_LDFLAGS) $(printf2_OBJECTS) $(printf2_LDADD) $(LIBS)
+push-pop$(EXEEXT): $(push_pop_OBJECTS) $(push_pop_DEPENDENCIES) 
+	@rm -f push-pop$(EXEEXT)
+	$(LINK) $(push_pop_LDFLAGS) $(push_pop_OBJECTS) $(push_pop_LDADD) $(LIBS)
 ret$(EXEEXT): $(ret_OBJECTS) $(ret_DEPENDENCIES) 
 	@rm -f ret$(EXEEXT)
 	$(LINK) $(ret_LDFLAGS) $(ret_OBJECTS) $(ret_LDADD) $(LIBS)
@@ -367,6 +376,7 @@ distclean-compile:
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/movi.Po@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/printf.Po@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/printf2.Po@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/push-pop.Po@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/ret.Po@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/rpn.Po@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/rpnfp.Po@am__quote@

tests/push-pop.c

@@ -0,0 +1,91 @@
+/******************************** -*- C -*- ****************************
+ *
+ *	Test `jit_pushr_i' and `jit_popr_i'
+ *
+ ***********************************************************************/
+
+
+/* Contributed by Ludovic Courtès.  */
+
+#ifdef HAVE_CONFIG_H
+# include "config.h"
+#endif
+
+#define JIT_NEED_PUSH_POP
+
+#include <stdio.h>
+#include <string.h>
+#include "lightning.h"
+
+typedef int (* stakumilo_t) (int);
+
+static void
+display_message (const char *fmt, int i)
+{
+  printf (fmt, i);
+}
+
+static stakumilo_t
+generate_push_pop (void)
+{
+  static const char msg[] = "we got %i\n";
+  static char buffer[1024];
+  stakumilo_t result;
+  int arg;
+
+  result = (stakumilo_t)(jit_set_ip (buffer).ptr);
+  jit_prolog (1);
+  arg = jit_arg_i ();
+  jit_getarg_i (JIT_R1, arg);
+
+  /* Save R1 on the stack.  */
+  jit_pushr_i (JIT_R1);
+
+  /* Save two other registers just for the sake of using the stack.  */
+  jit_movi_i (JIT_R0, -1);
+  jit_movi_i (JIT_R2, -1);
+  jit_pushr_i (JIT_R0);
+  jit_pushr_i (JIT_R2);
+
+  jit_movr_i (JIT_R0, JIT_R1);
+  jit_movi_p (JIT_R1, msg);
+
+  /* Invoke a function that may modify R1.  */
+  jit_prepare (2);
+  jit_pusharg_i (JIT_R0);
+  jit_pusharg_p (JIT_R1);
+  (void)jit_finish (display_message);
+
+  /* Restore the dummy registers.  */
+  jit_popr_i (JIT_R2);
+  jit_popr_i (JIT_R0);
+
+  /* Restore R1.  */
+  jit_popr_i (JIT_R1);
+
+  jit_movr_i (JIT_RET, JIT_R1);
+
+  jit_ret ();
+  jit_flush_code (buffer, jit_get_ip ().ptr);
+
+  return result;
+}
+
+
+int
+main (int argc, char *argv[])
+{
+  stakumilo_t stakumilo;
+
+  stakumilo = generate_push_pop ();
+  if (stakumilo (7777) != 7777)
+    {
+      printf ("failed: got %i instead of %i\n",
+	      stakumilo (7777), 7777);
+      return 1;
+    }
+  else
+    printf ("succeeded\n");
+
+  return 0;
+}

tests/push-pop.ok

@@ -0,0 +1,2 @@
+we got 7777
+succeeded