diff --git a/ChangeLog b/ChangeLog
index 746863192..0f51dea97 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,27 @@
+2013-04-27 Paulo Andrade <pcpa@gnu.org>
+
+	* include/lightning.h: Add new backend specific movr_w_d,
+	movr_d_w and movi_d_w codes as helpers to ia64 varargs
+	functions arguments.
+
+	* lib/jit_ia64-cpu.c:
+	  Correct wrong encoding of A5 small integers.
+	  Correct define of "mux" instruction modifiers.
+	  Correct ordering of arguments and predicates of cmp_xy
+	implementation with immediate arguments; like most other
+	codes with an immediate, the immediate is the second, not
+	the third argument.
+
+	* lib/jit_ia64-fpu.c: Actual implementation of the code
+	to move to/from gpr to/from fpr, to implement varargs abi.
+
+	* lib/jit_ia64.c: Make fpr argument registers not allocatable
+	as temporaries, no need for the extra checks when there are
+	plenty registers.
+
+	* lib/jit_print.c, lib/lightning.c: Minor updates for the
+	new movr_w_d, movr_d_w and movi_d_w codes.
+
 2013-04-26 Paulo Andrade <pcpa@gnu.org>
 
 	* lib/jit_ia64-cpu.c, lib/jit_ia64-fpu.c: Correct code to
diff --git a/include/lightning.h b/include/lightning.h
index 88029b861..7cbaf3011 100644
--- a/include/lightning.h
+++ b/include/lightning.h
@@ -739,6 +739,8 @@ typedef enum {
     jit_code_movr_w_f,		jit_code_movr_ww_d,	/* w* -> f|d */
 #define jit_movr_w_f(u, v)	jit_new_node_ww(jit_code_movr_w_f, u, v)
 #define jit_movr_ww_d(u, v, w)	jit_new_node_www(jit_code_movr_ww_d, u, v, w)
+    jit_code_movr_w_d,					/* w -> d */
+#define jit_movr_w_d(u, v)	jit_new_node_ww(jit_code_movr_w_d, u, v)
 
     jit_code_movr_f_w,		jit_code_movi_f_w,	/* f|d -> w* */
 #define jit_movr_f_w(u, v)	jit_new_node_ww(jit_code_movr_f_w, u, v)
@@ -746,6 +748,11 @@ typedef enum {
     jit_code_movr_d_ww,		jit_code_movi_d_ww,
 #define jit_movr_d_ww(u, v, w)	jit_new_node_www(jit_code_movr_d_ww, u, v, w)
 #define jit_movi_d_ww(u, v, w)	jit_new_node_wwd(jit_code_movi_d_ww, u, v, w)
+
+    jit_code_movr_d_w,		jit_code_movi_d_w,	/* d -> w */
+#define jit_movr_d_w(u, v)	jit_new_node_ww(jit_code_movr_d_w, u, v)
+#define jit_movi_d_w(u, v)	jit_new_node_wd(jit_code_movi_d_w, u, v)
+
     jit_code_x86_retval_f,	jit_code_x86_retval_d,
 } jit_code_t;
 
diff --git a/lib/jit_ia64-cpu.c b/lib/jit_ia64-cpu.c
index dc4f4f1d7..0182d95d7 100644
--- a/lib/jit_ia64-cpu.c
+++ b/lib/jit_ia64-cpu.c
@@ -202,11 +202,10 @@ typedef enum {
 #define BR_INDWH_DPTK			2
 
 #define MUX_BRCST			0
-/* FIXME only @name values in manual and only @brcst disassembled by binutils */
-#define MUX_REV				1	/* not disassembled as @rev */
-#define MUX_MIX				2	/* not disassembled as @mix */
-#define MUX_SHUF			3	/* not disassembled as @shuf */
-#define MUX_ALT				4	/* not disassembled as @alt */
+#define MUX_REV				11
+#define MUX_MIX				8
+#define MUX_SHUF			9
+#define MUX_ALT				10
 
 #define ldr(r0,r1)			ldr_l(r0,r1)
 #define ldi(r0,i0)			ldi_l(r0,i0)
@@ -855,8 +854,8 @@ static void _X5(jit_state_t*,jit_word_t,
 #define MOV(r0,r1)			ADDS(r0,0,r1)
 #define MOV_p(r0,r1,_p)			ADDS_p(r0,0,r1,_p)
 /* mov - Move Immediate */
-#define MOVI(r1,im)			ADDL(r1,im,0)
-#define MOVI_p(r1,im,_p)		ADDL_p(r1,im,0,_p)
+#define MOVI(r1,im)			ADDL(r1,im,GR_0)
+#define MOVI_p(r1,im,_p)		ADDL_p(r1,im,GR_0,_p)
 /* mov - Move Indirect Register */
 #define MOV_rn_RR(r1,r3)		M43(0x10,r3,r1)
 #define MOV_rn_DBR(r1,r3)		M43(0x11,r3,r1)
@@ -2075,15 +2074,21 @@ static void
 _A5(jit_state_t *_jit, jit_word_t _p,
     jit_word_t r3, jit_word_t im, jit_word_t r1)
 {
+    jit_word_t		s, i5, i9, i7;
     assert(!(_p & ~0x3fL));
     assert(!(r3  & ~0x3L));
     assert(im >= -2097152 && im < 2097151);
     assert(!(r1  & ~0x7fL));
+    /* imm22 = sign_ext(s << 21 | imm5c << 16 | imm9d << 7 | imm7b, 22) */
+    s  = (im & 0x200000) >> 21;
+    i5 = (im & 0x1f0000) >> 16;
+    i9 = (im &   0xff80) >>  7;
+    i7 =  im &     0x7f;
     TSTREG1(r3);
     TSTPRED(_p);
     TSTREG1(r1);
-    inst((9L<<37)|(((im>>7)&0x7fffL)<<22)|(r3<<20)|
-	 ((im&0x7fL)<<13)|(r1<<6)|_p, INST_A);
+    inst((9L<<37)|(s<<36)|(i9<<27)|(i5<<22)|
+	 (r3<<20)|(i7<<13)|(r1<<6)|_p, INST_A);
     SETREG(r1);
 }
 
@@ -3685,7 +3690,7 @@ _iqmulr(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1,
     }
     else
 	mulr(r0, r2, r3);
-    mulh(sign, r1, r2, r3);
+    mulh(r1, r2, r3, sign);
     if (r0 == r2 || r0 == r3) {
 	movr(r0, rn(reg));
 	jit_unget_reg(reg);
@@ -3837,7 +3842,7 @@ _lti(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_word_t i0)
 {
     jit_int32_t		reg;
     if (i0 >= -128 && i0 <= 127)
-	CMPI_LT(PR_6, PR_7, i0, r1);
+	CMPI_LT(PR_7, PR_6, i0 - 1, r1);
     else {
 	reg = jit_get_reg(jit_class_gpr);
 	movi(rn(reg), i0);
@@ -3861,7 +3866,7 @@ _lti_u(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_word_t i0)
 {
     jit_int32_t		reg;
     if (i0 >= -128 && i0 <= 127)
-	CMPI_LTU(PR_6, PR_7, i0, r1);
+	CMPI_LTU(PR_7, PR_6, i0 - 1, r1);
     else {
 	reg = jit_get_reg(jit_class_gpr);
 	movi(rn(reg), i0);
@@ -3945,7 +3950,7 @@ _gei(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_word_t i0)
 {
     jit_int32_t		reg;
     if (i0 >= -128 && i0 <= 127)
-	CMPI_LT(PR_6, PR_7, i0, r1);
+	CMPI_LT(PR_7, PR_6, i0 - 1, r1);
     else {
 	reg = jit_get_reg(jit_class_gpr);
 	movi(rn(reg), i0);
@@ -3969,7 +3974,7 @@ _gei_u(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_word_t i0)
 {
     jit_int32_t		reg;
     if (i0 >= -128 && i0 <= 127)
-	CMPI_LTU(PR_6, PR_7, i0, r1);
+	CMPI_LTU(PR_7, PR_6, i0, r1);
     else {
 	reg = jit_get_reg(jit_class_gpr);
 	movi(rn(reg), i0);
@@ -4387,7 +4392,7 @@ _blti(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, jit_word_t i1)
     jit_word_t		w;
     jit_int32_t		reg;
     if (i1 >= -128 && i1 <= 127)
-	CMPI_LT(PR_6, PR_7, i1, r0);
+	CMPI_LT(PR_7, PR_6, i1 - 1, r0);
     else {
 	reg = jit_get_reg(jit_class_gpr);
 	movi(rn(reg), i1);
@@ -4417,7 +4422,7 @@ _blti_u(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, jit_word_t i1)
     jit_word_t		w;
     jit_int32_t		reg;
     if (i1 >= -128 && i1 <= 127)
-	CMPI_LTU(PR_6, PR_7, i1, r0);
+	CMPI_LTU(PR_7, PR_6, i1 - 1, r0);
     else {
 	reg = jit_get_reg(jit_class_gpr);
 	movi(rn(reg), i1);
@@ -4523,7 +4528,7 @@ _bgei(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, jit_word_t i1)
     jit_word_t		w;
     jit_int32_t		reg;
     if (i1 >= -128 && i1 <= 127)
-	CMPI_LT(PR_6, PR_7, i1, r0);
+	CMPI_LT(PR_7, PR_6, i1 - 1, r0);
     else {
 	reg = jit_get_reg(jit_class_gpr);
 	movi(rn(reg), i1);
@@ -4553,7 +4558,7 @@ _bgei_u(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, jit_word_t i1)
     jit_word_t		w;
     jit_int32_t		reg;
     if (i1 >= -128 && i1 <= 127)
-	CMPI_LTU(PR_6, PR_7, i1, r0);
+	CMPI_LTU(PR_7, PR_6, i1 - 1, r0);
     else {
 	reg = jit_get_reg(jit_class_gpr);
 	movi(rn(reg), i1);
diff --git a/lib/jit_ia64-fpu.c b/lib/jit_ia64-fpu.c
index 1c24df8fd..eb12ab57a 100644
--- a/lib/jit_ia64-fpu.c
+++ b/lib/jit_ia64-fpu.c
@@ -423,6 +423,15 @@ static void F16_(jit_state_t*,jit_word_t,
 #define movi_f(r0,i0)			ldi_f(r0,(jit_word_t)i0)
 #define movi_d(r0,i0)			ldi_d(r0,(jit_word_t)i0)
 
+#define movr_f_w(r0,r1)			_movr_f_w(_jit,r0,r1)
+static void _movr_f_w(jit_state_t*,jit_int32_t,jit_int32_t);
+#define movr_d_w(r0,r1)			_movr_d_w(_jit,r0,r1)
+static void _movr_d_w(jit_state_t*,jit_int32_t,jit_int32_t);
+#define movi_f_w(r0,i0)			_movi_f_w(_jit,r0,i0)
+static void _movi_f_w(jit_state_t*,jit_int32_t,jit_word_t);
+#define movi_d_w(r0,i0)			_movi_d_w(_jit,r0,i0)
+static void _movi_d_w(jit_state_t*,jit_int32_t,jit_word_t);
+
 #define absr_f(r0,r1)			absr_d(r0,r1)
 #define absr_d(r0,r1)			FABS(r0,r1)
 #define negr_f(r0,r1)			negr_d(r0,r1)
@@ -978,6 +987,42 @@ F16_(jit_state_t* _jit, jit_word_t _p,
     inst((((im>>20)&1L)<<36)|(y<<27)|(1L<<26)|((im&0xffffL)<<6)|_p, INST_F);
 }
 
+static void
+_movr_f_w(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
+{
+    /* Should actually be used only in this case (with out0 == 120) */
+    if (r0 >= 120)
+	r0 = _jitc->rout + (r0 - 120);
+    GETF_S(r0, r1);
+}
+
+static void
+_movi_f_w(jit_state_t *_jit, jit_int32_t r0, jit_word_t i0)
+{
+    /* Should actually be used only in this case (with out0 == 120) */
+    if (r0 >= 120)
+	r0 = _jitc->rout + (r0 - 120);
+    ldi_i(r0, i0);
+}
+
+static void
+_movr_d_w(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
+{
+    /* Should actually be used only in this case (with out0 == 120) */
+    if (r0 >= 120)
+	r0 = _jitc->rout + (r0 - 120);
+    GETF_D(r0, r1);
+}
+
+static void
+_movi_d_w(jit_state_t *_jit, jit_int32_t r0, jit_word_t i0)
+{
+    /* Should actually be used only in this case (with out0 == 120) */
+    if (r0 >= 120)
+	r0 = _jitc->rout + (r0 - 120);
+    ldi_l(r0, i0);
+}
+
 #define fpr_opi(name, type, size)					\
 static void								\
 _##name##i_##type(jit_state_t *_jit,					\
@@ -1333,7 +1378,7 @@ _stxi_d(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, jit_int32_t r1)
 static void
 _sqrtr_f(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
 {
-    //MOV(_jitc->rout, r1);
+    movr_f(GR_8, r1);
     calli((jit_word_t)sqrtf);
     MOVF(r0, GR_8);
 }
@@ -1341,7 +1386,7 @@ _sqrtr_f(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
 static void
 _sqrtr_d(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
 {
-    //MOV(_jitc->rout, r1);
+    movr_d(GR_8, r1);
     calli((jit_word_t)sqrt);
     MOVF(r0, GR_8);
 }
diff --git a/lib/jit_ia64.c b/lib/jit_ia64.c
index 659320fd8..d574d24ce 100644
--- a/lib/jit_ia64.c
+++ b/lib/jit_ia64.c
@@ -133,12 +133,12 @@ jit_register_t		_rvs[] = {
     /* Scratch */
     { rc(fpr)|6,	 "f6"  },	{ rc(fpr)|7,		"f7"  },
     /* Do not allocate for the sake of simplification */
-    { 8,		 "f8"  },
+    { rc(arg)|8,	 "f8"  },
     /* Scratch - Argument/return registers */
-    { rc(fpr)|9,	 "f9"  },
-    { rc(fpr)|10,	 "f10" },	{ rc(fpr)|11,		"f11" },
-    { rc(fpr)|12,	 "f12" },	{ rc(fpr)|13,		"f13" },
-    { rc(fpr)|14,	 "f14" },	{ rc(fpr)|15,		"f15" },
+    { rc(arg)|9,	 "f9"  },
+    { rc(arg)|10,	 "f10" },	{ rc(arg)|11,		"f11" },
+    { rc(arg)|12,	 "f12" },	{ rc(arg)|13,		"f13" },
+    { rc(arg)|14,	 "f14" },	{ rc(arg)|15,		"f15" },
     /* Do not touch callee save registers not automatically spill/reloaded */
     { rc(sav)|16,	 "f16" },	{ rc(sav)|17,		"f17" },
     { rc(sav)|18,	 "f18" },	{ rc(sav)|19,		"f19" },
@@ -456,23 +456,19 @@ _jit_getarg_l(jit_state_t *_jit, jit_int32_t u, jit_node_t *v)
 void
 _jit_getarg_f(jit_state_t *_jit, jit_int32_t u, jit_node_t *v)
 {
-/*
     if (v->u.w < 8)
-	jit_movr_f(u, _R32 + v->u.w);
+	jit_movr_f(u, _F8 + v->u.w);
     else
 	jit_ldxi_f(u, JIT_FP, v->u.w);
-*/
 }
 
 void
 _jit_getarg_d(jit_state_t *_jit, jit_int32_t u, jit_node_t *v)
 {
-/*
     if (v->u.w < 8)
-	jit_movr_d(u, _R32 + v->u.w);
+	jit_movr_d(u, _F8 + v->u.w);
     else
 	jit_ldxi_d(u, JIT_FP, v->u.w);
-*/
 }
 
 void
@@ -511,6 +507,17 @@ void
 _jit_pushargr_f(jit_state_t *_jit, jit_int32_t u)
 {
     assert(_jitc->function);
+    if (_jitc->function->call.argi < 8) {
+	if (!(_jitc->function->call.call & jit_call_varargs))
+	    jit_movr_f(_F8 + _jitc->function->call.argi, u);
+	else
+	    jit_movr_f_w(_OUT0 + _jitc->function->call.argi, u);
+	++_jitc->function->call.argi;
+    }
+    else {
+	jit_stxr_f(_jitc->function->call.size, JIT_SP, u);
+	_jitc->function->call.size += sizeof(jit_word_t);
+    }
 }
 
 void
@@ -518,16 +525,37 @@ _jit_pushargi_f(jit_state_t *_jit, jit_float32_t u)
 {
     jit_int32_t		 regno;
     assert(_jitc->function);
-    /* FIXME move to OUTn or stack */
-    regno = jit_get_reg(jit_class_fpr);
-    jit_movi_f(regno, u);
-    jit_unget_reg(regno);
+    if (_jitc->function->call.argi < 8) {
+	if (!(_jitc->function->call.call & jit_call_varargs))
+	    jit_movi_f(_F8 + _jitc->function->call.argi, u);
+	else
+	    jit_movi_f_w(_OUT0 + _jitc->function->call.argi, u);
+	++_jitc->function->call.argi;
+    }
+    else {
+	regno = jit_get_reg(jit_class_fpr);
+	jit_movi_f(regno, u);
+	jit_stxi_f(_jitc->function->call.size, JIT_SP, regno);
+	_jitc->function->call.size += sizeof(jit_word_t);
+	jit_unget_reg(regno);
+    }
 }
 
 void
 _jit_pushargr_d(jit_state_t *_jit, jit_int32_t u)
 {
     assert(_jitc->function);
+    if (_jitc->function->call.argi < 8) {
+	if (!(_jitc->function->call.call & jit_call_varargs))
+	    jit_movr_d(_F8 + _jitc->function->call.argi, u);
+	else
+	    jit_movr_d_w(_OUT0 + _jitc->function->call.argi, u);
+	++_jitc->function->call.argi;
+    }
+    else {
+	jit_stxr_d(_jitc->function->call.size, JIT_SP, u);
+	_jitc->function->call.size += sizeof(jit_word_t);
+    }
 }
 
 void
@@ -535,10 +563,20 @@ _jit_pushargi_d(jit_state_t *_jit, jit_float64_t u)
 {
     jit_int32_t		 regno;
     assert(_jitc->function);
-    /* FIXME move to OUTn or stack */
-    regno = jit_get_reg(jit_class_fpr);
-    jit_movi_d(regno, u);
-    jit_unget_reg(regno);
+    if (_jitc->function->call.argi < 8) {
+	if (!(_jitc->function->call.call & jit_call_varargs))
+	    jit_movi_d(_F8 + _jitc->function->call.argi, u);
+	else
+	    jit_movi_d_w(_OUT0 + _jitc->function->call.argi, u);
+	++_jitc->function->call.argi;
+    }
+    else {
+	regno = jit_get_reg(jit_class_fpr);
+	jit_movi_d(regno, u);
+	jit_stxi_d(_jitc->function->call.size, JIT_SP, regno);
+	_jitc->function->call.size += sizeof(jit_word_t);
+	jit_unget_reg(regno);
+    }
 }
 
 jit_bool_t
@@ -633,6 +671,7 @@ _jit_retval_d(jit_state_t *_jit, jit_int32_t r0)
 jit_pointer_t
 _emit_code(jit_state_t *_jit)
 {
+    jit_uint8_t		*end;
     jit_node_t		*node;
     jit_node_t		*temp;
     jit_word_t		 word;
@@ -745,8 +784,11 @@ _emit_code(jit_state_t *_jit)
 		    patch(word, node);					\
 		}							\
 		break
+    /* default of 64 bytes is too low for some possible
+     * quite long code generation sequences, e.g. qdivi */
+    end = _jit->code.ptr + _jit->code.length - 4096;
     for (node = _jitc->head; node; node = node->next) {
-	if (_jit->pc.uc >= _jitc->code.end && !jit_remap())
+	if (_jit->pc.uc >= end && !jit_remap())
 	    return (NULL);
 
 	value = jit_classify(node->code);
@@ -1195,6 +1237,18 @@ _emit_code(jit_state_t *_jit)
 	    case jit_code_arg:
 	    case jit_code_arg_f:		case jit_code_arg_d:
 		break;
+	    case jit_code_movr_f_w:
+		movr_f_w(rn(node->u.w), rn(node->v.w));
+		break;
+	    case jit_code_movr_d_w:
+		movr_d_w(rn(node->u.w), rn(node->v.w));
+		break;
+	    case jit_code_movi_f_w:
+		movi_f_w(rn(node->u.w), node->v.n->u.w);
+		break;
+	    case jit_code_movi_d_w:
+		movi_d_w(rn(node->u.w), node->v.n->u.w);
+		break;
 	    default:
 		abort();
 	}
diff --git a/lib/jit_print.c b/lib/jit_print.c
index f692e3c1f..fa3cee805 100644
--- a/lib/jit_print.c
+++ b/lib/jit_print.c
@@ -209,8 +209,10 @@ static char *code_name[] = {
     "bordr_d",		"bordi_d",
     "bunordr_d",	"bunordi_d",
     "movr_w_f",		"movr_ww_d",
+    "movr_w_d",
     "movr_f_w",		"movi_f_w",
     "movr_d_ww",	"movi_d_ww",
+    "movr_d_w",		"movi_d_w",
     "x86_retval_f",	"x86_retval_d",
 };
 
diff --git a/lib/lightning.c b/lib/lightning.c
index f3e0d9b77..6525b12b7 100644
--- a/lib/lightning.c
+++ b/lib/lightning.c
@@ -1153,7 +1153,7 @@ _jit_classify(jit_state_t *_jit, jit_code_t code)
 	case jit_code_movi_f:	case jit_code_movi_f_w:
 	    mask = jit_cc_a0_reg|jit_cc_a0_chg|jit_cc_a1_flt;
 	    break;
-	case jit_code_movi_d:
+	case jit_code_movi_d:	case jit_code_movi_d_w:
 	    mask = jit_cc_a0_reg|jit_cc_a0_chg|jit_cc_a1_dbl;
 	    break;
 	case jit_code_movi_d_ww:
@@ -1174,6 +1174,7 @@ _jit_classify(jit_state_t *_jit, jit_code_t code)
 	case jit_code_movr_d:	case jit_code_extr_d:	case jit_code_extr_f_d:
 	case jit_code_ldr_d:
 	case jit_code_movr_w_f:	case jit_code_movr_f_w:
+	case jit_code_movr_w_d:	case jit_code_movr_d_w:
 	    mask = jit_cc_a0_reg|jit_cc_a0_chg|jit_cc_a1_reg;
 	    break;
 	case jit_code_movr_d_ww: