diff --git a/ChangeLog b/ChangeLog
index f6fb8c479..c81dd8802 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,30 @@
+2013-10-07 Paulo Andrade <pcpa@gnu.org>
+
+	* check/self.c: Extend tests to validate jit_callee_save_p
+	does not cause an assertion on valid arguments, and test
+	extra registers defined on some backends.
+
+	* configure.ac: Do not ignore environment CFLAGS when
+	checking if need to test runtime configurable options,
+	like use x87 when sse2 is available, arm instruction set
+	instead of thumb, etc.
+
+	* include/lightning/jit_arm.h: Correct wrong jit_f macro
+	definition.
+
+	* include/lightning/jit_ia64.h, include/lightning/jit_ppc.h: 
+	Correct wrong jit_r macro definition.
+
+	* lib/jit_x86-x87.c, lib/jit_x86.c: Actually use the
+	reserved stack space for integer to/from float conversion.
+	The stack space was also changed to ensure it is 8 bytes
+	aligned. Also, for Solaris x86 in 32 bit mode, an alternate
+	truncr_d was implemented because for some reason it is
+	failing with SIGILL if using the "fisttpl" instructions,
+	that must be available on p6 or newer, but for the sake of
+	making all tests pass, implement a 486 or newer sequence
+	if "sun" is defined.
+
 2013-10-03 Paulo Andrade <pcpa@gnu.org>
 
 	* include/lightning/jit_mips.h, lib/jit_mips-cpu.c,
diff --git a/check/self.c b/check/self.c
index 42758251a..7cfbb94ce 100644
--- a/check/self.c
+++ b/check/self.c
@@ -5,44 +5,93 @@
 int
 main(int argc, char *argv[])
 {
+    jit_state_t		*_jit;
+
     /* Same JIT_XY are not constants */
     init_jit(argv[0]);
 
+    _jit = jit_new_state();
     assert(JIT_R0 == jit_r(0));
+    (void)jit_callee_save_p(JIT_R0);
     assert(JIT_R1 == jit_r(1));
+    (void)jit_callee_save_p(JIT_R1);
     assert(JIT_R2 == jit_r(2));
+    (void)jit_callee_save_p(JIT_R2);
 #if defined(JIT_R3)
     assert(JIT_R3 == jit_r(3));
+    (void)jit_callee_save_p(JIT_R3);
 #  if defined(JIT_R4)
     assert(JIT_R4 == jit_r(4));
+    (void)jit_callee_save_p(JIT_R4);
 #    if defined(JIT_R5)
     assert(JIT_R5 == jit_r(5));
+    (void)jit_callee_save_p(JIT_R5);
 #      if defined(JIT_R6)
     assert(JIT_R6 == jit_r(6));
+    (void)jit_callee_save_p(JIT_R6);
 #        if defined(JIT_R7)
     assert(JIT_R7 == jit_r(7));
+    (void)jit_callee_save_p(JIT_R7);
 #        endif
 #      endif
 #    endif
 #  endif
 #endif
     assert(JIT_V0 == jit_v(0));
+    assert(jit_callee_save_p(JIT_V0));
     assert(JIT_V1 == jit_v(1));
+    assert(jit_callee_save_p(JIT_V1));
     assert(JIT_V2 == jit_v(2));
+    assert(jit_callee_save_p(JIT_V2));
 #if defined(JIT_V3)
     assert(JIT_V3 == jit_v(3));
+    assert(jit_callee_save_p(JIT_V3));
 #  if defined(JIT_V4)
     assert(JIT_V4 == jit_v(4));
+    assert(jit_callee_save_p(JIT_V4));
 #    if defined(JIT_V5)
     assert(JIT_V5 == jit_v(5));
+    assert(jit_callee_save_p(JIT_V5));
 #      if defined(JIT_V6)
     assert(JIT_V6 == jit_v(6));
+    assert(jit_callee_save_p(JIT_V6));
 #        if defined(JIT_V7)
     assert(JIT_V7 == jit_v(7));
+    assert(jit_callee_save_p(JIT_V7));
 #          if defined(JIT_V8)
     assert(JIT_V8 == jit_v(8));
+    assert(jit_callee_save_p(JIT_V8));
 #            if defined(JIT_V9)
     assert(JIT_V9 == jit_v(9));
+    assert(jit_callee_save_p(JIT_V9));
+#              if defined(JIT_V10)
+    assert(JIT_V10 == jit_v(10));
+    assert(jit_callee_save_p(JIT_V10));
+#                if defined(JIT_V11)
+    assert(JIT_V11 == jit_v(11));
+    assert(jit_callee_save_p(JIT_V11));
+#                  if defined(JIT_V12)
+    assert(JIT_V12 == jit_v(12));
+    assert(jit_callee_save_p(JIT_V12));
+#                    if defined(JIT_V13)
+    assert(JIT_V13 == jit_v(13));
+    assert(jit_callee_save_p(JIT_V13));
+#                      if defined(JIT_V14)
+    assert(JIT_V14 == jit_v(14));
+    assert(jit_callee_save_p(JIT_V14));
+#                        if defined(JIT_V15)
+    assert(JIT_V15 == jit_v(15));
+    assert(jit_callee_save_p(JIT_V15));
+#                          if defined(JIT_V16)
+    assert(JIT_V16 == jit_v(16));
+    assert(jit_callee_save_p(JIT_V16));
+#                          endif
+#                        endif
+#                      endif
+#                    endif
+#                  endif
+#                endif
+#              endif
 #            endif
 #          endif
 #        endif
@@ -51,18 +100,36 @@ main(int argc, char *argv[])
 #  endif
 #endif
     assert(JIT_F0 == jit_f(0));
+    (void)jit_callee_save_p(JIT_F0);
     assert(JIT_F1 == jit_f(1));
+    (void)jit_callee_save_p(JIT_F1);
     assert(JIT_F2 == jit_f(2));
+    (void)jit_callee_save_p(JIT_F2);
     assert(JIT_F3 == jit_f(3));
+    (void)jit_callee_save_p(JIT_F3);
     assert(JIT_F4 == jit_f(4));
+    (void)jit_callee_save_p(JIT_F4);
     assert(JIT_F5 == jit_f(5));
+    (void)jit_callee_save_p(JIT_F5);
 #if defined(JIT_F6)
     assert(JIT_F6 == jit_f(6));
+    (void)jit_callee_save_p(JIT_F6);
 #  if defined(JIT_F7)
     assert(JIT_F7 == jit_f(7));
+    (void)jit_callee_save_p(JIT_F7);
+#    if defined(JIT_F8)
+    assert(JIT_F8 == jit_f(8));
+    (void)jit_callee_save_p(JIT_F8);
+#      if defined(JIT_F9)
+    assert(JIT_F9 == jit_f(9));
+    (void)jit_callee_save_p(JIT_F9);
+#      endif
+#    endif
 #  endif
 #endif
 
+    jit_clear_state();
+    jit_destroy_state();
     finish_jit();
 
     return (0);
diff --git a/configure.ac b/configure.ac
index 11b0063e2..6b9e7a703 100644
--- a/configure.ac
+++ b/configure.ac
@@ -137,7 +137,7 @@ ac_cv_test_arm_arm=
 ac_cv_test_arm_swf=
 
 save_CFLAGS=$CFLAGS
-CFLAGS="-I$PWD/include -D_GNU_SOURCE"
+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
diff --git a/include/lightning.h b/include/lightning.h
index 561c559bb..9fab1386f 100644
--- a/include/lightning.h
+++ b/include/lightning.h
@@ -84,7 +84,9 @@
 #    define __BYTE_ORDER	__BIG_ENDIAN
 #  elif defined(__BIG_ENDIAN__)			/* ia64 hp-ux */
 #    define __BYTE_ORDER	__BIG_ENDIAN
-#  elif defined(__i386__)			/* x86 solaris */
+#  elif defined(__i386__)			/* 32 bit x86 solaris */
+#    define __BYTE_ORDER	__LITTLE_ENDIAN
+#  elif defined(__x86_64__)			/* 64 bit x86 solaris */
 #    define __BYTE_ORDER	__LITTLE_ENDIAN
 #  elif defined(__MIPSEB)			/* mips irix */
 #    define __BYTE_ORDER	__BIG_ENDIAN
diff --git a/include/lightning/jit_arm.h b/include/lightning/jit_arm.h
index d217a4bc6..6c8ede848 100644
--- a/include/lightning/jit_arm.h
+++ b/include/lightning/jit_arm.h
@@ -37,7 +37,7 @@ typedef enum {
 #define jit_v(i)		(_R7 + (i))
 #define jit_v_num()		3
 #define jit_arg_f_reg_p(i)	((i) >= 0 && (i) < 4)
-#define jit_f(i)		(jit_cpu.abi ? _D8 + (i) : _D7 + (i))
+#define jit_f(i)		(jit_cpu.abi ? _D8 + ((i)<<1) : _D0 - ((i)<<1))
 #define jit_f_num()		8
     _R12,			/* ip - temporary */
 #define JIT_R0			_R4
diff --git a/include/lightning/jit_ia64.h b/include/lightning/jit_ia64.h
index 05397496a..29f4189c0 100644
--- a/include/lightning/jit_ia64.h
+++ b/include/lightning/jit_ia64.h
@@ -37,7 +37,7 @@ typedef enum {
 #define JIT_V2		_R46
 #define JIT_V3		_R47
 #define jit_r_num()	4
-#define jit_r(n)	(_R40 - (n))
+#define jit_r(n)	(_R40 + (n))
 #define jit_v_num()	4
 #define jit_v(n)	(_R44 + (n))
     _R0,	/* constant - Always 0 */
diff --git a/include/lightning/jit_ppc.h b/include/lightning/jit_ppc.h
index 9976b095e..4daf9de76 100644
--- a/include/lightning/jit_ppc.h
+++ b/include/lightning/jit_ppc.h
@@ -28,7 +28,11 @@
  */
 typedef enum {
 #define jit_arg_reg_p(i)	((i) >= 0 && (i) < 8)
-#define jit_r(i)		(_R11 + (i))
+#if __ppc__
+#  define jit_r(i)		(_R11 + (i))
+#else
+#  define jit_r(i)		(_R28 + (i))
+#endif
 #define jit_r_num()		3
 #if __ppc__
 #  define jit_v(i)		(_R30 - (i))
diff --git a/lib/jit_x86-x87.c b/lib/jit_x86-x87.c
index 69355934c..63043c818 100644
--- a/lib/jit_x86-x87.c
+++ b/lib/jit_x86-x87.c
@@ -18,6 +18,11 @@
  */
 
 #if PROTO
+#  if __WORDSIZE == 32
+#    define CVT_OFFSET			-12
+#  else
+#    define CVT_OFFSET			-8
+#  endif
 #  define _ST0_REGNO			0
 #  define _ST1_REGNO			1
 #  define _ST2_REGNO			2
@@ -27,11 +32,16 @@
 #  define _ST6_REGNO			6
 #  define _ST7_REGNO			7
 #  define x87rx(code, md, rb, ri, ms)	_x87rx(_jit, code, md, rb, ri, ms)
+#  define fldcwm(md, rb, ri, ms)	x87rx(015, md, rb, ri, ms)
+#  define fstcwm(md, rb, ri, ms)	_fstcwm(_jit, md, rb, ri, ms)
+static void
+_fstcwm(jit_state_t*, jit_int32_t, jit_int32_t, jit_int32_t, jit_int32_t);
 #  define fldsm(md, rb, ri, ms)		x87rx(010, md, rb, ri, ms)
 #  define fstsm(md, rb, ri, ms)		x87rx(012, md, rb, ri, ms)
 #  define fldlm(md, rb, ri, ms)		x87rx(050, md, rb, ri, ms)
 #  define fstlm(md, rb, ri, ms)		x87rx(052, md, rb, ri, ms)
 #  define fisttplm(md, rb, ri, ms)	x87rx(031, md, rb, ri, ms)
+#  define fistlm(md, rb, ri, ms)	x87rx(032, md, rb, ri, ms)
 #  define fisttpqm(md, rb, ri, ms)	x87rx(071, md, rb, ri, ms)
 #  define fildlm(md, rb, ri, ms)	x87rx(030, md, rb,ri, ms)
 #  define fildqm(md, rb, ri, ms)	x87rx(075, md, rb,ri, ms)
@@ -402,6 +412,15 @@ _x87_b##name##i_##type(jit_state_t *_jit,				\
 #  define dopi(name)			fpr_opi(name, d, 64)
 #  define dbopi(name)			fpr_bopi(name, d, 64)
 
+static void
+_fstcwm(jit_state_t *_jit, jit_int32_t md,
+	jit_int32_t rb,	jit_int32_t ri, jit_int32_t ms)
+{
+    ic(0x9b);
+    rex(0, 1, rb, ri, _NOREG);
+    x87rx(017, md, rb, ri, ms);
+}
+
 static void
 _x87rx(jit_state_t *_jit, jit_int32_t code, jit_int32_t md,
        jit_int32_t rb, jit_int32_t ri, jit_int32_t ms)
@@ -630,34 +649,52 @@ _x87_sqrtr_d(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
 static void
 _x87_truncr_d_i(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
 {
-    pushr(_RAX_REGNO);
+#if defined(sun)
+    /* for the sake of passing test cases in x87 mode, otherwise only sse
+     * is supported */
+    fstcwm(-4, _RBP_REGNO, _NOREG, _SCL1);
+    ldxi_s(r0, _RBP_REGNO, -4);
+    extr_uc(r0, r0);
+#  define FPCW_CHOP	0xc00
+    ori(r0, r0, FPCW_CHOP);
+    stxi_s(-8, _RBP_REGNO, r0);
+    fldcwm(-8, _RBP_REGNO, _NOREG, _SCL1);
+    if (r1 == _ST0_REGNO)
+	fistlm(CVT_OFFSET, _RBP_REGNO, _NOREG, _SCL1);
+    else {
+	fxchr(r1);
+	fistlm(CVT_OFFSET, _RBP_REGNO, _NOREG, _SCL1);
+	fxchr(r1);
+    }
+    fldcwm(-4, _RBP_REGNO, _NOREG, _SCL1);
+    ldxi(r0, _RBP_REGNO, CVT_OFFSET);
+#else
     fldr(r1);
-    fisttplm(0, _RSP_REGNO, _NOREG, _SCL1);
-    popr(r0);
+    fisttplm(CVT_OFFSET, _RBP_REGNO, _NOREG, _SCL1);
+    ldxi_i(r0, _RBP_REGNO, CVT_OFFSET);
+#endif
 }
 
 #  if __WORDSIZE == 64
 static void
 _x87_truncr_d_l(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
 {
-    pushr(_RAX_REGNO);
     fldr(r1);
-    fisttpqm(0, _RSP_REGNO, _NOREG, _SCL1);
-    popr(r0);
+    fisttpqm(CVT_OFFSET, _RBP_REGNO, _NOREG, _SCL1);
+    ldxi(r0, _RBP_REGNO, CVT_OFFSET);
 }
 #  endif
 
 static void
 _x87_extr_d(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
 {
-    pushr(r1);
+    stxi(CVT_OFFSET, _RBP_REGNO, r1);
 #  if __WORDSIZE == 32
-    fildlm(0, _RSP_REGNO, _NOREG, _SCL1);
+    fildlm(CVT_OFFSET, _RBP_REGNO, _NOREG, _SCL1);
 #  else
-    fildqm(0, _RSP_REGNO, _NOREG, _SCL1);
+    fildqm(CVT_OFFSET, _RBP_REGNO, _NOREG, _SCL1);
 #  endif
     fstpr(r0 + 1);
-    popr(r1);
 }
 
 static void
diff --git a/lib/jit_x86.c b/lib/jit_x86.c
index b2c1b4006..28b61665d 100644
--- a/lib/jit_x86.c
+++ b/lib/jit_x86.c
@@ -310,7 +310,11 @@ _jit_prolog(jit_state_t *_jit)
     _jitc->function->self.argi = _jitc->function->self.argf =
 	_jitc->function->self.aoff = _jitc->function->self.alen = 0;
     /* sse/x87 conversion */
+#if __WORDSIZE == 32
+    _jitc->function->self.aoff = -12;
+#else
     _jitc->function->self.aoff = -8;
+#endif
     _jitc->function->self.call = jit_call_default;
     jit_alloc((jit_pointer_t *)&_jitc->function->regoff,
 	      _jitc->reglen * sizeof(jit_int32_t));