Adapt PowerPC port to work in Darwin 32 bit and Linux 64 bit.

* include/lightning.h: Add check for __powerpc__ defined
	in Linux, while Darwin defines __ppc__.

	* include/lightning/jit_ppc.h: Adjust register definitions
	for Darwin 32 bit and Linux 64 bit ppc usage and/or ABI.

	* include/lightning/jit_private.h: Add proper check for
	Linux __powerpc__ and an data definition for an workaround
	to properly handle code that starts with a jump to a "main"
	label.

	* lib/jit_disasm.c: Add extra disassembler initialization
	for __powerpc64__.

	* lib/jit_ppc-cpu.c: Add extra macros and functions, and
	correct/adapt previous ones to handle powerpc64.

	* lib/jit_ppc-fpu.c: Adapt for 64 bit wordsize. Basically
	add conversion from/to int32/int64 and proper handling of
	load/store offsets too large for 32 bit.

	* lib/jit_ppc.c: Add calls to 64 bit codes and adaptation
	for the PowerPC 64 bit Linux ABI.

	* lib/jit_arm.c, lib/jit_mips.c, lib/jit_sparc, lib/jit_x86.c,
	lib/lightning.c: Correct off by one error when restarting jit
	of a function due to finding too late that needs to spill/reload
	some register. Problem was found by accident on a very special
	condition during PowerPC 64 code adaptation.

ChangeLog

@@ -1,3 +1,35 @@
+2013-03-10 Paulo Andrade <pcpa@gnu.org>
+
+	* include/lightning.h: Add check for __powerpc__ defined
+	in Linux, while Darwin defines __ppc__.
+
+	* include/lightning/jit_ppc.h: Adjust register definitions
+	for Darwin 32 bit and Linux 64 bit ppc usage and/or ABI.
+
+	* include/lightning/jit_private.h: Add proper check for
+	Linux __powerpc__ and an data definition for an workaround
+	to properly handle code that starts with a jump to a "main"
+	label.
+
+	* lib/jit_disasm.c: Add extra disassembler initialization
+	for __powerpc64__.
+
+	* lib/jit_ppc-cpu.c: Add extra macros and functions, and
+	correct/adapt previous ones to handle powerpc64.
+
+	* lib/jit_ppc-fpu.c: Adapt for 64 bit wordsize. Basically
+	add conversion from/to int32/int64 and proper handling of
+	load/store offsets too large for 32 bit.
+
+	* lib/jit_ppc.c: Add calls to 64 bit codes and adaptation
+	for the PowerPC 64 bit Linux ABI.
+
+	* lib/jit_arm.c, lib/jit_mips.c, lib/jit_sparc, lib/jit_x86.c,
+	lib/lightning.c: Correct off by one error when restarting jit
+	of a function due to finding too late that needs to spill/reload
+	some register. Problem was found by accident on a very special
+	condition during PowerPC 64 code adaptation.
+
 2013-03-08 Paulo Andrade <pcpa@gnu.org>
 
 	* check/lightning.c: Add missing ppc preprocessor definition.

include/lightning.h

@@ -70,7 +70,7 @@ typedef jit_int32_t		jit_fpr_t;
 #  include <lightning/jit_mips.h>
 #elif defined(__arm__)
 #  include <lightning/jit_arm.h>
-#elif defined(__ppc__)
+#elif defined(__ppc__) || defined(__powerpc__)
 #  include <lightning/jit_ppc.h>
 #elif defined(__sparc__)
 #  include <lightning/jit_sparc.h>

include/lightning/jit_ppc.h

@@ -28,34 +28,46 @@ typedef enum {
 #define jit_arg_reg_p(i)	((i) >= 0 && (i) < 8)
 #define jit_r(i)		(_R11 + (i))
 #define jit_r_num()		3
+#if __WORDSIZE == 32
 #  define jit_v(i)		(_R30 - (i))
 #  define jit_v_num()		17
+#else
+#  define jit_v(i)		(_R27 - (i))
+#  define jit_v_num()		14
+#endif
 #define jit_arg_f_reg_p(i)	((i) >= 0 && (i) < 13)
 #define jit_f(i)		(_F14 + (i))
 #define jit_f_num()		6
     _R0,
+#if __WORDSIZE == 32
 #  define JIT_R0		_R11
 #  define JIT_R1		_R12
 #  define JIT_R2		_R13
-#define JIT_R3			_R2
+#else
+#  define JIT_R0		_R28
+#  define JIT_R1		_R29
+#  define JIT_R2		_R30
+#endif
     _R11,	_R12,	_R13,	_R2,
-#define JIT_V0			_R30
+#define JIT_V0			jit_v(0)
-#define JIT_V1			_R29
+#define JIT_V1			jit_v(1)
-#define JIT_V2			_R28
+#define JIT_V2			jit_v(2)
-#define JIT_V3			_R28
+#define JIT_V3			jit_v(3)
-#define JIT_V4			_R26
+#define JIT_V4			jit_v(4)
-#define JIT_V5			_R25
+#define JIT_V5			jit_v(5)
-#define JIT_V6			_R24
+#define JIT_V6			jit_v(6)
-#define JIT_V7			_R23
+#define JIT_V7			jit_v(7)
-#define JIT_V8			_R22
+#define JIT_V8			jit_v(8)
-#define JIT_V9			_R21
+#define JIT_V9			jit_v(9)
-#define JIT_V10			_R20
+#define JIT_V10			jit_v(10)
-#define JIT_V11			_R19
+#define JIT_V11			jit_v(11)
-#define JIT_V12			_R18
+#define JIT_V12			jit_v(12)
-#define JIT_V13			_R17
+#define JIT_V13			jit_v(13)
-#define JIT_V14			_R16
+#if __WORDSIZE == 32
-#define JIT_V15			_R15
+#  define JIT_V14		jit_v(14)
-#define JIT_V16			_R14
+#  define JIT_V15		jit_v(15)
+#  define JIT_V16		jit_v(16)
+#endif
     _R14,	_R15,	_R16,	_R17,	_R18,	_R19,	_R20,	_R21,
     _R22,	_R23,	_R24,	_R25,	_R26,	_R27,	_R28,	_R29,
     _R30,

include/lightning/jit_private.h

@@ -66,7 +66,7 @@
 #  else
 #    define JIT_FRET		_R0
 #  endif
-#elif defined(__ppc__)
+#elif defined(__ppc__) || defined(__powerpc__)
 #  define JIT_SP		_R1
 #  define JIT_RET		_R3
 #  define JIT_FRET		_F1
@@ -347,6 +347,16 @@ struct jit_compiler {
 	jit_int32_t	  values[1024];	/* pending constants */
 	jit_word_t	  patches[2048];
     } consts;
+#elif __powerpc64__
+    /* Keep track of prolog addresses, just for the sake of making
+     * jit that starts with a jump to a "main" label work like other
+     * backends. */
+    struct {
+	jit_word_t	 *ptr;
+	jit_word_t	  offset;
+	jit_word_t	  length;
+    } prolog;
+    jit_bool_t		  jump;
 #endif
 };
 

lib/jit_arm.c

@@ -1437,6 +1437,7 @@ _emit_code(jit_state_t *_jit)
 			    temp->code == jit_code_epilog)
 			    temp->flag &= ~jit_flag_patch;
 		    }
+		    temp->flag &= ~jit_flag_patch;
 		    node = undo.node;
 		    _jit->pc.w = undo.word;
 		    invalidate_consts();

lib/jit_disasm.c

@@ -81,6 +81,12 @@ jit_init_debug(void)
     if (jit_cpu.thumb)
 	disasm_info.disassembler_options = "force-thumb";
 #  endif
+#  if defined(__powerpc64__)
+    disasm_info.arch = bfd_arch_powerpc;
+    disasm_info.mach = bfd_mach_ppc64;
+    disasm_info.disassembler_options = "64";
+    disassemble_init_powerpc(&disasm_info);
+#  endif
 #  if defined(__sparc__)
     disasm_info.endian = disasm_info.display_endian = BFD_ENDIAN_BIG;
 #endif

lib/jit_mips.c

@@ -1186,6 +1186,7 @@ _emit_code(jit_state_t *_jit)
 			    temp->code == jit_code_epilog)
 			    temp->flag &= ~jit_flag_patch;
 		    }
+		    temp->flag &= ~jit_flag_patch;
 		    node = undo.node;
 		    _jit->pc.w = undo.word;
 		    _jitc->patches.offset = undo.patch_offset;

lib/jit_ppc-fpu.c

@@ -36,6 +36,10 @@ static void _FXFL(jit_state_t*,int,int,int,int,int);
 #  define FCTIW_(d,b)			FX_(63,d,0,b,14)
 #  define FCTIWZ(d,b)			FX(63,d,0,b,15)
 #  define FCTIWZ_(d,b)			FX_(63,d,0,b,15)
+#  define FCTID(d,b)			FX(63,d,0,b,814)
+#  define FCTID_(d,b)			FX_(63,d,0,b,814)
+#  define FCTIDZ(d,b)			FX(63,d,0,b,815)
+#  define FCTIDZ_(d,b)			FX_(63,d,0,b,815)
 #  define FDIV(d,a,b)			FA(63,d,a,b,0,18)
 #  define FDIV_(d,a,b)			FA_(63,d,a,b,0,18)
 #  define FDIVS(d,a,b)			FA(59,d,a,b,0,18)
@@ -121,10 +125,20 @@ static void _movi_d(jit_state_t*,jit_int32_t,jit_float64_t*);
 #  define extr_f(r0,r1)			extr_d(r0,r1)
 #  define extr_d(r0,r1)			_extr_d(_jit,r0,r1)
 static void _extr_d(jit_state_t*,jit_int32_t,jit_int32_t);
-#  define truncr_f_i(r0,r1)		truncr_d(r0,r1)
+
-#  define truncr_d_i(r0,r1)		truncr_d(r0,r1)
+#  define truncr_f(r0,r1)		truncr_d(r0,r1)
-#  define truncr_d(r0,r1)		_truncr_d(_jit,r0,r1)
+#  define truncr_f_i(r0,r1)		truncr_d_i(r0,r1)
-static void _truncr_d(jit_state_t*,jit_int32_t,jit_int32_t);
+#  define truncr_d_i(r0,r1)		_truncr_d_i(_jit,r0,r1)
+static void _truncr_d_i(jit_state_t*,jit_int32_t,jit_int32_t);
+#  if __WORDSIZE == 32
+#    define truncr_d(r0,r1)		truncr_d_i(r0,r1)
+#  else
+#    define truncr_d(r0,r1)		truncr_d_l(r0,r1)
+#    define truncr_f_l(r0,r1)		truncr_d_l(r0,r1)
+#    define truncr_d_l(r0,r1)		_truncr_d_l(_jit,r0,r1)
+static void _truncr_d_l(jit_state_t*,jit_int32_t,jit_int32_t);
+#  endif
+
 #  define extr_d_f(r0,r1)		FRSP(r0,r1)
 #  define extr_f_d(r0,r1)		movr_d(r0,r1)
 
@@ -436,6 +450,7 @@ _movi_d(jit_state_t *_jit, jit_int32_t r0, jit_float64_t *i0)
 static void
 _extr_d(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
 {
+#  if __WORDSIZE == 32
     jit_int32_t		reg;
     reg = jit_get_reg(jit_class_gpr);
     rshi(rn(reg), r1, 31);
@@ -443,22 +458,39 @@ _extr_d(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
     stxi(alloca_offset - 4, _FP_REGNO, r1);
     stxi(alloca_offset - 8, _FP_REGNO, rn(reg));
     jit_unget_reg(reg);
+#  else
+    stxi(alloca_offset - 8, _FP_REGNO, r1);
+#  endif
     ldxi_d(r0, _FP_REGNO, alloca_offset - 8);
     FCFID(r0, r0);
 }
 
 static void
-_truncr_d(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
+_truncr_d_i(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
 {
     jit_int32_t		reg;
     reg = jit_get_reg(jit_class_fpr);
     FCTIWZ(rn(reg), r1);
     /* use reserved 8 bytes area */
     stxi_d(alloca_offset - 8, _FP_REGNO, rn(reg));
-    ldxi(r0, _FP_REGNO, alloca_offset - 4);
+    ldxi_i(r0, _FP_REGNO, alloca_offset - 4);
     jit_unget_reg(reg);
 }
 
+#  if __WORDSIZE == 64
+static void
+_truncr_d_l(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
+{
+    jit_int32_t		reg;
+    reg = jit_get_reg(jit_class_fpr);
+    FCTIDZ(rn(reg), r1);
+    /* use reserved 8 bytes area */
+    stxi_d(alloca_offset - 8, _FP_REGNO, rn(reg));
+    ldxi(r0, _FP_REGNO, alloca_offset - 8);
+    jit_unget_reg(reg);
+}
+#  endif
+
 #  define fpr_opi(name, type, size)					\
 static void								\
 _##name##i_##type(jit_state_t *_jit,					\
@@ -843,8 +875,8 @@ _ldi_f(jit_state_t *_jit, jit_int32_t r0, jit_word_t i0)
     jit_word_t		lo, hi;
     if (can_sign_extend_short_p(i0))
 	LFS(r0, _R0_REGNO, i0);
-    else {
+    else if (can_sign_extend_int_p(i0)) {
-	hi = (i0 >> 16) + ((jit_uint16_t)i0 >> 15);
+	hi = (jit_int16_t)((i0 >> 16) + ((jit_uint16_t)i0 >> 15));
 	lo = (jit_int16_t)(i0 - (hi << 16));
 	reg = jit_get_reg(jit_class_gpr);
 	if ((inv = reg == _R0))		reg = jit_get_reg(jit_class_gpr);
@@ -853,6 +885,12 @@ _ldi_f(jit_state_t *_jit, jit_int32_t r0, jit_word_t i0)
 	jit_unget_reg(reg);
 	if (inv)			jit_unget_reg(_R0);
     }
+    else {
+	reg = jit_get_reg(jit_class_gpr);
+	movi(rn(reg), i0);
+	ldr_f(r0, rn(reg));
+	jit_unget_reg(reg);
+    }
 }
 
 static void
@@ -863,8 +901,8 @@ _ldi_d(jit_state_t *_jit, jit_int32_t r0, jit_word_t i0)
     jit_word_t		lo, hi;
     if (can_sign_extend_short_p(i0))
 	LFD(r0, _R0_REGNO, i0);
-    else {
+    else if (can_sign_extend_int_p(i0)) {
-	hi = (i0 >> 16) + ((jit_uint16_t)i0 >> 15);
+	hi = (jit_int16_t)((i0 >> 16) + ((jit_uint16_t)i0 >> 15));
 	lo = (jit_int16_t)(i0 - (hi << 16));
 	reg = jit_get_reg(jit_class_gpr);
 	if ((inv = reg == _R0))		reg = jit_get_reg(jit_class_gpr);
@@ -873,6 +911,12 @@ _ldi_d(jit_state_t *_jit, jit_int32_t r0, jit_word_t i0)
 	jit_unget_reg(reg);
 	if (inv)			jit_unget_reg(_R0);
     }
+    else {
+	reg = jit_get_reg(jit_class_gpr);
+	movi(rn(reg), i0);
+	ldr_d(r0, rn(reg));
+	jit_unget_reg(reg);
+    }
 }
 
 static void
@@ -967,8 +1011,8 @@ _sti_f(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0)
     jit_word_t		lo, hi;
     if (can_sign_extend_short_p(i0))
 	STFS(r0, _R0_REGNO, i0);
-    else {
+    else if (can_sign_extend_int_p(i0)) {
-	hi = (i0 >> 16) + ((jit_uint16_t)i0 >> 15);
+	hi = (jit_int16_t)((i0 >> 16) + ((jit_uint16_t)i0 >> 15));
 	lo = (jit_int16_t)(i0 - (hi << 16));
 	reg = jit_get_reg(jit_class_gpr);
 	if ((inv = reg == _R0))		reg = jit_get_reg(jit_class_gpr);
@@ -977,6 +1021,12 @@ _sti_f(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0)
 	jit_unget_reg(reg);
 	if (inv)			jit_unget_reg(_R0);
     }
+    else {
+	reg = jit_get_reg(jit_class_gpr);
+	movi(rn(reg), i0);
+	str_f(rn(reg), r0);
+	jit_unget_reg(reg);
+    }
 }
 
 static void
@@ -987,8 +1037,8 @@ _sti_d(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0)
     jit_word_t		lo, hi;
     if (can_sign_extend_short_p(i0))
 	STFD(r0, _R0_REGNO, i0);
-    else {
+    else if (can_sign_extend_int_p(i0)) {
-	hi = (i0 >> 16) + ((jit_uint16_t)i0 >> 15);
+	hi = (jit_int16_t)((i0 >> 16) + ((jit_uint16_t)i0 >> 15));
 	lo = (jit_int16_t)(i0 - (hi << 16));
 	reg = jit_get_reg(jit_class_gpr);
 	if ((inv = reg == _R0))		reg = jit_get_reg(jit_class_gpr);
@@ -997,6 +1047,12 @@ _sti_d(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0)
 	jit_unget_reg(reg);
 	if (inv)			jit_unget_reg(_R0);
     }
+    else {
+	reg = jit_get_reg(jit_class_gpr);
+	movi(rn(reg), i0);
+	str_d(rn(reg), r0);
+	jit_unget_reg(reg);
+    }
 }
 
 static void

lib/jit_ppc.c

@@ -37,10 +37,17 @@ extern void __clear_cache(void *, void *);
  */
 jit_register_t		_rvs[] = {
     { rc(gpr) | 0,			"r0" },
+#if __WORDSIZE == 32
     { rc(gpr) | 11,			"r11" },
     { rc(gpr) | 12,			"r12" },
     { rc(gpr) | 13,			"r13" },
     { rc(gpr) | 2,			"r2" },
+#else
+    { rc(sav) | 11,			"r11" },	/* env */
+    { rc(sav) | 12,			"r12" },	/* exception */
+    { rc(sav) | 13,			"r13" },	/* thread */
+    { rc(sav) | 2,			"r2" },		/* toc */
+#endif
     { rc(sav) | rc(gpr) | 14,		"r14" },
     { rc(sav) | rc(gpr) | 15,		"r15" },
     { rc(sav) | rc(gpr) | 16,		"r16" },
@@ -270,16 +277,21 @@ _jit_arg_f(jit_state_t *_jit)
     assert(_jitc->function);
     if (_jitc->function->self.argf < 13)
 	offset = _jitc->function->self.argf++;
-    else
+    else {
+#if __WORDSIZE == 32
 	offset = _jitc->function->self.size;
-    _jitc->function->self.size += sizeof(jit_float32_t);
+#else
+	offset = _jitc->function->self.size + 4;
+#endif
+    }
+    _jitc->function->self.size += sizeof(jit_word_t);
     return (jit_new_node_w(jit_code_arg_f, offset));
 }
 
 jit_bool_t
 _jit_arg_f_reg_p(jit_state_t *_jit, jit_int32_t offset)
 {
-    return (jit_arg_d_reg_p(offset));
+    return (offset >= 0 && offset < 13);
 }
 
 jit_node_t *
@@ -298,7 +310,7 @@ _jit_arg_d(jit_state_t *_jit)
 jit_bool_t
 _jit_arg_d_reg_p(jit_state_t *_jit, jit_int32_t offset)
 {
-    return (offset >= 0 && offset < 13);
+    return (jit_arg_f_reg_p(offset));
 }
 
 void
@@ -364,8 +376,13 @@ _jit_getarg_us(jit_state_t *_jit, jit_int32_t u, jit_node_t *v)
 void
 _jit_getarg_i(jit_state_t *_jit, jit_int32_t u, jit_node_t *v)
 {
-    if (v->u.w < 8)
+    if (v->u.w < 8) {
+#if __WORDSIZE == 32
 	jit_movr(u, JIT_RA0 - v->u.w);
+#else
+	jit_extr_i(u, JIT_RA0 - v->u.w);
+#endif
+    }
     else {
 #if __BYTE_ORDER == __LITTLE_ENDIAN
 	jit_ldxi_i(u, JIT_FP, v->u.w);
@@ -460,9 +477,14 @@ _jit_pushargr_f(jit_state_t *_jit, jit_int32_t u)
 	++_jitc->function->call.argf;
 	if (!(_jitc->function->call.call & jit_call_varargs)) {
 	    /* in case of excess arguments */
-	    if (_jitc->function->call.argi < 8)
+	    if (_jitc->function->call.argi < 8) {
+#if __WORDSIZE == 32
 		_jitc->function->call.argi += 2;
-	    _jitc->function->call.size += sizeof(jit_float32_t);
+#else
+		_jitc->function->call.argi++;
+#endif
+	    }
+	    _jitc->function->call.size += sizeof(jit_word_t);
 	    return;
 	}
     }
@@ -472,13 +494,21 @@ _jit_pushargr_f(jit_state_t *_jit, jit_int32_t u)
 	jit_ldxi(JIT_RA0 - _jitc->function->call.argi, JIT_FP,
 		 alloca_offset - 8);
 	_jitc->function->call.argi++;
+#if __WORDSIZE == 32
 	jit_ldxi(JIT_RA0 - _jitc->function->call.argi, JIT_FP,
 		 alloca_offset - 4);
 	_jitc->function->call.argi++;
+#endif
     }
-    else
+    else {
+#if __WORDSIZE == 32
 	jit_stxi_f(_jitc->function->call.size + params_offset, JIT_SP, u);
-    _jitc->function->call.size += sizeof(jit_float32_t);
+#else
+	jit_stxi_f(_jitc->function->call.size + params_offset + 4,
+		   JIT_SP, u);
+#endif
+    }
+    _jitc->function->call.size += sizeof(jit_word_t);
 }
 
 void
@@ -492,9 +522,14 @@ _jit_pushargi_f(jit_state_t *_jit, jit_float32_t u)
 	++_jitc->function->call.argf;
 	if (!(_jitc->function->call.call & jit_call_varargs)) {
 	    /* in case of excess arguments */
-	    if (_jitc->function->call.argi < 8)
+	    if (_jitc->function->call.argi < 8) {
+#if __WORDSIZE == 32
 		_jitc->function->call.argi += 2;
-	    _jitc->function->call.size += sizeof(jit_float32_t);
+#else
+		_jitc->function->call.argi++;
+#endif
+	    }
+	    _jitc->function->call.size += sizeof(jit_word_t);
 	    return;
 	}
     }
@@ -506,13 +541,21 @@ _jit_pushargi_f(jit_state_t *_jit, jit_float32_t u)
 	jit_ldxi(JIT_RA0 - _jitc->function->call.argi, JIT_FP,
 		 alloca_offset - 8);
 	_jitc->function->call.argi++;
+#if __WORDSIZE == 32
 	jit_ldxi(JIT_RA0 - _jitc->function->call.argi, JIT_FP,
 		 alloca_offset - 4);
 	_jitc->function->call.argi++;
+#endif
     }
-    else
+    else {
+#if __WORDSIZE == 32
 	jit_stxi_f(_jitc->function->call.size + params_offset, JIT_SP, regno);
-    _jitc->function->call.size += sizeof(jit_float32_t);
+#else
+	jit_stxi_f(_jitc->function->call.size + params_offset + 4,
+		   JIT_SP, regno);
+#endif
+    }
+    _jitc->function->call.size += sizeof(jit_word_t);
     jit_unget_reg(regno);
 }
 
@@ -525,8 +568,13 @@ _jit_pushargr_d(jit_state_t *_jit, jit_int32_t u)
 	++_jitc->function->call.argf;
 	if (!(_jitc->function->call.call & jit_call_varargs)) {
 	    /* in case of excess arguments */
-	    if (_jitc->function->call.argi < 8)
+	    if (_jitc->function->call.argi < 8) {
+#if __WORDSIZE == 32
 		_jitc->function->call.argi += 2;
+#else
+		_jitc->function->call.argi++;
+#endif
+	    }
 	    _jitc->function->call.size += sizeof(jit_float64_t);
 	    return;
 	}
@@ -537,9 +585,11 @@ _jit_pushargr_d(jit_state_t *_jit, jit_int32_t u)
 	jit_ldxi(JIT_RA0 - _jitc->function->call.argi, JIT_FP,
 		 alloca_offset - 8);
 	_jitc->function->call.argi++;
+#if __WORDSIZE == 32
 	jit_ldxi(JIT_RA0 - _jitc->function->call.argi, JIT_FP,
 		 alloca_offset - 4);
 	_jitc->function->call.argi++;
+#endif
     }
     else
 	jit_stxi_d(_jitc->function->call.size + params_offset, JIT_SP, u);
@@ -557,8 +607,13 @@ _jit_pushargi_d(jit_state_t *_jit, jit_float64_t u)
 	++_jitc->function->call.argf;
 	if (!(_jitc->function->call.call & jit_call_varargs)) {
 	    /* in case of excess arguments */
-	    if (_jitc->function->call.argi < 8)
+	    if (_jitc->function->call.argi < 8) {
+#if __WORDSIZE == 32
 		_jitc->function->call.argi += 2;
+#else
+		_jitc->function->call.argi++;
+#endif
+	    }
 	    _jitc->function->call.size += sizeof(jit_float64_t);
 	    return;
 	}
@@ -571,9 +626,11 @@ _jit_pushargi_d(jit_state_t *_jit, jit_float64_t u)
 	jit_ldxi(JIT_RA0 - _jitc->function->call.argi, JIT_FP,
 		 alloca_offset - 8);
 	_jitc->function->call.argi++;
+#if __WORDSIZE == 32
 	jit_ldxi(JIT_RA0 - _jitc->function->call.argi, JIT_FP,
 		 alloca_offset - 4);
 	_jitc->function->call.argi++;
+#endif
     }
     else
 	jit_stxi_d(_jitc->function->call.size + params_offset, JIT_SP, regno);
@@ -704,7 +761,10 @@ _emit_code(jit_state_t *_jit)
     struct {
 	jit_node_t	*node;
 	jit_word_t	 word;
-	jit_int32_t	 patch_offset;
+	jit_word_t	 patch_offset;
+#if __powerpc64__
+	jit_word_t	 prolog_offset;
+#endif
     } undo;
 
     _jitc->function = NULL;
@@ -714,6 +774,14 @@ _emit_code(jit_state_t *_jit)
     undo.word = 0;
     undo.node = NULL;
     undo.patch_offset = 0;
+#if __powerpc64__
+    undo.prolog_offset = 0;
+
+    /* code may start with a jump so add an initial function descriptor */
+    il(_jit->pc.w + 24);	/* addr */
+    il(0);			/* toc */
+    il(0);			/* env */
+#endif
 
 #define case_rr(name, type)						\
 	    case jit_code_##name##r##type:				\
@@ -859,6 +927,10 @@ _emit_code(jit_state_t *_jit)
 		case_rr(ext, _uc);
 		case_rr(ext, _s);
 		case_rr(ext, _us);
+#  if __WORDSIZE == 64
+		case_rr(ext, _i);
+		case_rr(ext, _ui);
+#  endif
 		case_rr(hton,);
 		case_rr(neg,);
 		case_rr(com,);
@@ -882,6 +954,10 @@ _emit_code(jit_state_t *_jit)
 		break;
 		case_rr(trunc, _f_i);
 		case_rr(trunc, _d_i);
+#  if __WORDSIZE == 64
+		case_rr(trunc, _f_l);
+		case_rr(trunc, _d_l);
+#  endif
 		case_rrr(lt,);
 		case_rrw(lt,);
 		case_rrr(lt, _u);
@@ -962,6 +1038,16 @@ _emit_code(jit_state_t *_jit)
 		case_rw(ld, _i);
 		case_rrr(ldx, _i);
 		case_rrw(ldx, _i);
+#if __WORDSIZE == 64
+		case_rr(ld, _ui);
+		case_rw(ld, _ui);
+		case_rrr(ldx, _ui);
+		case_rrw(ldx, _ui);
+		case_rr(ld, _l);
+		case_rw(ld, _l);
+		case_rrr(ldx, _l);
+		case_rrw(ldx, _l);
+#endif
 		case_rr(st, _c);
 		case_wr(st, _c);
 		case_rrr(stx, _c);
@@ -974,6 +1060,12 @@ _emit_code(jit_state_t *_jit)
 		case_wr(st, _i);
 		case_rrr(stx, _i);
 		case_wrr(stx, _i);
+#if __WORDSIZE == 64
+		case_rr(st, _l);
+		case_wr(st, _l);
+		case_rrr(stx, _l);
+		case_wrr(stx, _l);
+#endif
 		case_rr(mov, _f);
 	    case jit_code_movi_f:
 		assert(node->flag & jit_flag_data);
@@ -1142,6 +1234,10 @@ _emit_code(jit_state_t *_jit)
 		jmpr(rn(node->u.w));
 		break;
 	    case jit_code_jmpi:
+#if __powerpc64__
+		if (_jit->pc.uc == _jit->code.ptr + 24)
+		    _jitc->jump = 1;
+#endif
 		temp = node->u.n;
 		assert(temp->code == jit_code_label ||
 		       temp->code == jit_code_epilog);
@@ -1173,8 +1269,29 @@ _emit_code(jit_state_t *_jit)
 		undo.node = node;
 		undo.word = _jit->pc.w;
 		undo.patch_offset = _jitc->patches.offset;
+#if __powerpc64__
+		undo.prolog_offset = _jitc->prolog.offset;
+#endif
 	    restart_function:
 		_jitc->again = 0;
+#if __powerpc64__
+		if (_jitc->jump) {
+		    /* remember prolog to hide offset adjustment for a jump
+		     * to the start of a function, what is expected to be
+		     * a common practice as first jit instruction */
+		    if (_jitc->prolog.offset >= _jitc->prolog.length) {
+			_jitc->prolog.length += 16;
+			_jitc->prolog.ptr = realloc(_jitc->prolog.ptr,
+						    _jitc->prolog.length *
+						    sizeof(jit_word_t));
+		    }
+		    _jitc->prolog.ptr[_jitc->prolog.offset++] = _jit->pc.w;
+		    /* function descriptor */
+		    il(_jit->pc.w + 24);	/* addr */
+		    il(0);			/* toc */
+		    il(0);			/* env */
+		}
+#endif
 		prolog(node);
 		break;
 	    case jit_code_epilog:
@@ -1186,9 +1303,13 @@ _emit_code(jit_state_t *_jit)
 			    temp->code == jit_code_epilog)
 			    temp->flag &= ~jit_flag_patch;
 		    }
+		    temp->flag &= ~jit_flag_patch;
 		    node = undo.node;
 		    _jit->pc.w = undo.word;
 		    _jitc->patches.offset = undo.patch_offset;
+#if __powerpc64__
+		    _jitc->prolog.offset = undo.prolog_offset;
+#endif
 		    goto restart_function;
 		}
 		/* remember label is defined */

lib/jit_sparc.c

@@ -1076,6 +1076,7 @@ _emit_code(jit_state_t *_jit)
 			    temp->code == jit_code_epilog)
 			    temp->flag &= ~jit_flag_patch;
 		    }
+		    temp->flag &= ~jit_flag_patch;
 		    node = undo.node;
 		    _jit->pc.w = undo.word;
 		    _jitc->patches.offset = undo.patch_offset;

lib/jit_x86.c

@@ -1575,6 +1575,7 @@ _emit_code(jit_state_t *_jit)
 			    temp->code == jit_code_epilog)
 			    temp->flag &= ~jit_flag_patch;
 		    }
+		    temp->flag &= ~jit_flag_patch;
 		    node = undo.node;
 		    _jit->pc.w = undo.word;
 		    _jitc->patches.offset = undo.patch_offset;

lib/lightning.c

@@ -587,6 +587,11 @@ _jit_clear_state(jit_state_t *_jit)
     _jitc->data_info.ptr = NULL;
 #endif
 
+#if __powerpc64__
+    free(_jitc->prolog.ptr);
+    _jitc->prolog.ptr = NULL;
+#endif
+
     free(_jitc);
 }
 
@@ -1402,7 +1407,9 @@ _jit_emit(jit_state_t *_jit)
     for (;;) {
 	if ((code = emit_code()) == NULL) {
 	    for (node = _jitc->head; node; node = node->next) {
-		if (node->code == jit_code_label && node->link)
+		if (node->link &&
+		    (node->code == jit_code_label ||
+		     node->code == jit_code_epilog))
 		    node->flag &= ~jit_flag_patch;
 	    }
 	    ++mult;
@@ -1439,7 +1446,7 @@ _jit_emit(jit_state_t *_jit)
     result = mprotect(_jit->code.ptr, _jit->code.length, PROT_READ | PROT_EXEC);
     assert(result == 0);
 
-    return (code);
+    return (_jit->code.ptr);
 }
 
 /*   Compute initial reglive and regmask set values of a basic block.
@@ -2605,7 +2612,7 @@ _patch_register(jit_state_t *_jit, jit_node_t *node, jit_node_t *link,
 #  include "jit_mips.c"
 #elif defined(__arm__)
 #  include "jit_arm.c"
-#elif defined(__ppc__)
+#elif defined(__ppc__) || defined(__powerpc__)
 #  include "jit_ppc.c"
 #elif defined(__sparc__)
 #  include "jit_sparc.c"