[Guile-commits] 313/437: Implement the jit_align interface

[Andy Wingo]

wingo pushed a commit to branch lightning
in repository guile.

commit afae5407f67646254c868a9fde6da97609d5f69d
Author: pcpa <address@hidden>
Date:   Tue Oct 14 17:05:25 2014 -0300

    Implement the jit_align interface
    
        * lib/jit_aarch64-cpu.c, lib/jit_alpha-cpu.c, lib/jit_arm-cpu.c,
        lib/jit_hppa-cpu.c, lib/jit_mips-cpu.c, lib/jit_ppc-cpu.c,
        lib/jit_sparc-cpu.c: Implement or correct the internal
        nop(count) call that receives an argument that tells the
        modulo bytes to align the code for the next instruction.
    
        * include/lightning.h, lib/lightning.c, lib/jit_aarch64.c,
        lib/jit_alpha.c, lib/jit_arm.c, lib/jit_hppa.c, lib/jit_ia64.c,
        lib/jit_mips.c, lib/jit_ppc.c, lib/jit_s390x.c, lib/jit_sparc.c,
        lib/jit_x86.c: Implement the new jit_align() call that receive
        an argument, that tells the modulo, in bytes, to align the
        next instruction. In most backends the only value that makes
        a difference is a value that matches sizeof(void*), as all
        other values usually are already automatically aligned in
        labels, but not guaranteed to be aligned at word size bytes.
    
        * check/align.ok, check/align.tst: New files, implementing
        a simple test for the new jit_align() interface.
    
        * check/Makefile.am, check/lightning.c, lib/jit_aarch64-sz.c,
        lib/jit_alpha-sz.c, lib/jit_arm-sz.c, lib/jit_hppa-sz.c,
        lib/jit_ia64-sz.c, lib/jit_mips-sz.c, lib/jit_ppc-sz.c,
        lib/jit_print.c, lib/jit_s390x-sz.c, lib/jit_sparc-sz.c,
        lib/jit_x86-sz.c: Update for the new jit_code_align code and
        the jit_align() interface.
---
 ChangeLog             | 28 ++++++++++++++++++++++++++++
 check/Makefile.am     |  5 +++--
 check/lightning.c     |  5 +++--
 include/lightning.h   |  3 ++-
 lib/jit_aarch64-cpu.c | 11 +++++++++++
 lib/jit_aarch64-sz.c  |  1 +
 lib/jit_aarch64.c     |  7 +++++++
 lib/jit_alpha-cpu.c   | 10 ++++++++++
 lib/jit_alpha-sz.c    |  1 +
 lib/jit_alpha.c       |  7 +++++++
 lib/jit_arm-cpu.c     |  6 +++---
 lib/jit_arm-sz.c      |  2 ++
 lib/jit_arm.c         |  7 +++++++
 lib/jit_hppa-cpu.c    | 10 ++++++++++
 lib/jit_hppa-sz.c     |  1 +
 lib/jit_hppa.c        |  7 +++++++
 lib/jit_ia64-sz.c     |  1 +
 lib/jit_ia64.c        |  5 +++++
 lib/jit_mips-cpu.c    |  8 +++++---
 lib/jit_mips-sz.c     |  3 +++
 lib/jit_mips.c        |  7 +++++++
 lib/jit_ppc-cpu.c     | 10 ++++++++++
 lib/jit_ppc-sz.c      |  3 +++
 lib/jit_ppc.c         |  7 +++++++
 lib/jit_print.c       |  2 +-
 lib/jit_s390x-sz.c    |  1 +
 lib/jit_s390x.c       | 10 +++++++++-
 lib/jit_sparc-cpu.c   |  3 ++-
 lib/jit_sparc-sz.c    |  1 +
 lib/jit_sparc.c       |  9 ++++++++-
 lib/jit_x86-sz.c      |  2 ++
 lib/jit_x86.c         |  8 +++++++-
 lib/lightning.c       |  9 +++++----
 33 files changed, 180 insertions(+), 20 deletions(-)

diff --git a/ChangeLog b/ChangeLog
index 35e26a1..cba8922 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,31 @@
+2014-10-14 Paulo Andrade <address@hidden>
+
+       * lib/jit_aarch64-cpu.c, lib/jit_alpha-cpu.c, lib/jit_arm-cpu.c,
+       lib/jit_hppa-cpu.c, lib/jit_mips-cpu.c, lib/jit_ppc-cpu.c,
+       lib/jit_sparc-cpu.c: Implement or correct the internal
+       nop(count) call that receives an argument that tells the
+       modulo bytes to align the code for the next instruction.
+
+       * include/lightning.h, lib/lightning.c, lib/jit_aarch64.c,
+       lib/jit_alpha.c, lib/jit_arm.c, lib/jit_hppa.c, lib/jit_ia64.c,
+       lib/jit_mips.c, lib/jit_ppc.c, lib/jit_s390x.c, lib/jit_sparc.c,
+       lib/jit_x86.c: Implement the new jit_align() call that receive
+       an argument, that tells the modulo, in bytes, to align the
+       next instruction. In most backends the only value that makes
+       a difference is a value that matches sizeof(void*), as all
+       other values usually are already automatically aligned in
+       labels, but not guaranteed to be aligned at word size bytes.
+
+       * check/align.ok, check/align.tst: New files, implementing
+       a simple test for the new jit_align() interface.
+
+       * check/Makefile.am, check/lightning.c, lib/jit_aarch64-sz.c,
+       lib/jit_alpha-sz.c, lib/jit_arm-sz.c, lib/jit_hppa-sz.c,
+       lib/jit_ia64-sz.c, lib/jit_mips-sz.c, lib/jit_ppc-sz.c,
+       lib/jit_print.c, lib/jit_s390x-sz.c, lib/jit_sparc-sz.c,
+       lib/jit_x86-sz.c: Update for the new jit_code_align code and
+       the jit_align() interface.
+
 2014-10-13 Paulo Andrade <address@hidden>
 
        * include/lightning.h, lib/jit_size.c, size.c: Use a
diff --git a/check/Makefile.am b/check/Makefile.am
index c65622b..5e0b21b 100644
--- a/check/Makefile.am
+++ b/check/Makefile.am
@@ -39,6 +39,7 @@ $(top_builddir)/lib/liblightning.la:
 EXTRA_DIST =                           \
        3to2.tst        3to2.ok         \
        add.tst         add.ok          \
+       align.tst       align.ok        \
        allocai.tst     allocai.ok      \
        bp.tst          bp.ok           \
        divi.tst        divi.ok         \
@@ -91,7 +92,7 @@ EXTRA_DIST =                          \
        run-test        all.tst
 
 base_TESTS =                           \
-       3to2 add allocai                \
+       3to2 add align allocai          \
        bp divi fib rpn                 \
        ldstr ldsti                     \
        ldstxr ldstxi                   \
@@ -164,7 +165,7 @@ endif
 if test_arm_arm
 #arm_TESTS = $(addsuffix .arm, $(base_TESTS))
 arm_TESTS =                                    \
-       3to2.arm add.arm allocai.arm            \
+       3to2.arm add.arm align.arm allocai.arm  \
        bp.arm divi.arm fib.arm rpn.arm         \
        ldstr.arm ldsti.arm                     \
        ldstxr.arm ldstxi.arm                   \
diff --git a/check/lightning.c b/check/lightning.c
index 60b75bc..cf08dea 100644
--- a/check/lightning.c
+++ b/check/lightning.c
@@ -267,7 +267,7 @@ static void call_forward(void *value, label_t *label);
 static void make_arg(void *value);
 static jit_pointer_t get_arg(void);
 static long get_imm(void);
-static void name(void);
+static void align(void);       static void name(void);
 static void prolog(void);
 static void frame(void);       static void tramp(void);
 static void ellipsis(void);
@@ -567,7 +567,7 @@ static char          *data;
 static size_t            data_offset, data_length;
 static instr_t           instr_vector[] = {
 #define entry(value)   { NULL, #value, value }
-    entry(name),
+    entry(align),      entry(name),
     entry(prolog),
     entry(frame),      entry(tramp),
     entry(ellipsis),
@@ -1324,6 +1324,7 @@ name(void) {
     (void)identifier(ch);
     jit_name(parser.string);
 }
+entry_im(align)
 entry(prolog)
 entry_im(frame)                        entry_im(tramp)
 entry(ellipsis)
diff --git a/include/lightning.h b/include/lightning.h
index 0179919..2e87905 100644
--- a/include/lightning.h
+++ b/include/lightning.h
@@ -177,7 +177,8 @@ typedef enum {
 #define jit_data(u,v,w)                _jit_data(_jit,u,v,w)
     jit_code_data,
 #define jit_live(u)            jit_new_node_w(jit_code_live, u)
-    jit_code_live,
+#define jit_align(u)           jit_new_node_w(jit_code_align, u)
+    jit_code_live,             jit_code_align,
     jit_code_save,             jit_code_load,
 #define jit_name(u)            _jit_name(_jit,u)
     jit_code_name,
diff --git a/lib/jit_aarch64-cpu.c b/lib/jit_aarch64-cpu.c
index a2d2ce0..fd25f7f 100644
--- a/lib/jit_aarch64-cpu.c
+++ b/lib/jit_aarch64-cpu.c
@@ -467,6 +467,7 @@ typedef union {
 #  define B_C(Cc,Simm19)               oc19(A64_B_C,Cc,Simm19)
 #  define CBZ(Rd,Simm19)               ox19(A64_CBZ|XS,Rd,Simm19)
 #  define CBNZ(Rd,Simm19)              ox19(A64_CBNZ|XS,Rd,Simm19)
+#  define NOP()                                ii(0xd503201f)
 static jit_int32_t logical_immediate(jit_word_t);
 #  define oxxx(Op,Rd,Rn,Rm)            _oxxx(_jit,Op,Rd,Rn,Rm)
 static void 
_oxxx(jit_state_t*,jit_int32_t,jit_int32_t,jit_int32_t,jit_int32_t);
@@ -499,6 +500,8 @@ static void _oxxxc(jit_state_t*,jit_int32_t,jit_int32_t,
 #  define oxxx7(Op,Rt,Rt2,Rn,Simm7)    _oxxx7(_jit,Op,Rt,Rt2,Rn,Simm7)
 static void _oxxx7(jit_state_t*,jit_int32_t,
                   jit_int32_t,jit_int32_t,jit_int32_t,jit_int32_t);
+#  define nop(i0)                      _nop(_jit,i0)
+static void _nop(jit_state_t*,jit_int32_t);
 #  define addr(r0,r1,r2)               ADD(r0,r1,r2)
 #  define addi(r0,r1,i0)               _addi(_jit,r0,r1,i0)
 static void _addi(jit_state_t*,jit_int32_t,jit_int32_t,jit_word_t);
@@ -1011,6 +1014,14 @@ _oxxx7(jit_state_t *_jit, jit_int32_t Op,
 }
 
 static void
+_nop(jit_state_t *_jit, jit_int32_t i0)
+{
+    for (; i0 > 0; i0 -= 4)
+       NOP();
+    assert(i0 == 0);
+}
+
+static void
 _addi(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_word_t i0)
 {
     jit_int32_t                reg;
diff --git a/lib/jit_aarch64-sz.c b/lib/jit_aarch64-sz.c
index 58efa4e..01b18f2 100644
--- a/lib/jit_aarch64-sz.c
+++ b/lib/jit_aarch64-sz.c
@@ -8,6 +8,7 @@
     0,
     0,
     0,
+    0,
     64,
     0,
     4,
diff --git a/lib/jit_aarch64.c b/lib/jit_aarch64.c
index 94c52b7..6de141c 100644
--- a/lib/jit_aarch64.c
+++ b/lib/jit_aarch64.c
@@ -717,6 +717,13 @@ _emit_code(jit_state_t *_jit)
        value = jit_classify(node->code);
        jit_regarg_set(node, value);
        switch (node->code) {
+           case jit_code_align:
+               assert(!(node->u.w & (node->u.w - 1)) &&
+                      node->u.w <= sizeof(jit_word_t));
+               if (node->u.w == sizeof(jit_word_t) &&
+                   (word = _jit->pc.w & (sizeof(jit_word_t) - 1)))
+                   nop(sizeof(jit_word_t) - word);
+               break;
            case jit_code_note:         case jit_code_name:
                node->u.w = _jit->pc.w;
                break;
diff --git a/lib/jit_alpha-cpu.c b/lib/jit_alpha-cpu.c
index 8fe4338..8787b0f 100644
--- a/lib/jit_alpha-cpu.c
+++ b/lib/jit_alpha-cpu.c
@@ -303,6 +303,8 @@ static void _Opi(jit_state_t*,int,int,unsigned int,unsigned 
int,int);
 #  define NEGL(ra,rc)                  SUBL(_R31_REGNO,ra,rc)
 #  define NEGQ(ra,rc)                  SUBQ(_R31_REGNO,ra,rc)
 #  define NOT(ra,rc)                   ORNOT(_R31_REGNO,ra,rc)
+#  define nop(i0)                      _nop(_jit,i0)
+static void _nop(jit_state_t*,jit_int32_t);
 #  define movr(r0,r1)                  _movr(_jit,r0,r1)
 static void _movr(jit_state_t*,jit_int32_t,jit_int32_t);
 #  define movi(r0,i0)                  _movi(_jit,r0,i0)
@@ -704,6 +706,14 @@ _Opi(jit_state_t *_jit, int o, int ra, unsigned int i, 
unsigned int f, int rc)
 }
 
 static void
+_nop(jit_state_t *_jit, jit_int32_t i0)
+{
+    for (; i0 > 0; i0 -= 4)
+       NOP();
+    assert(i0 == 0);
+}
+
+static void
 _movr(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
 {
     if (r0 != r1)
diff --git a/lib/jit_alpha-sz.c b/lib/jit_alpha-sz.c
index b2544fe..bf914d0 100644
--- a/lib/jit_alpha-sz.c
+++ b/lib/jit_alpha-sz.c
@@ -8,6 +8,7 @@
     0,
     0,
     0,
+    0,
     76,
     0,
     4,
diff --git a/lib/jit_alpha.c b/lib/jit_alpha.c
index 84b055c..6d78c70 100644
--- a/lib/jit_alpha.c
+++ b/lib/jit_alpha.c
@@ -744,6 +744,13 @@ _emit_code(jit_state_t *_jit)
        value = jit_classify(node->code);
        jit_regarg_set(node, value);
        switch (node->code) {
+           case jit_code_align:
+               assert(!(node->u.w & (node->u.w - 1)) &&
+                      node->u.w <= sizeof(jit_word_t));
+               if (node->u.w == sizeof(jit_word_t) &&
+                   (word = _jit->pc.w & (sizeof(jit_word_t) - 1)))
+                   nop(sizeof(jit_word_t) - word);
+               break;
            case jit_code_note:         case jit_code_name:
                node->u.w = _jit->pc.w;
                break;
diff --git a/lib/jit_arm-cpu.c b/lib/jit_arm-cpu.c
index 02edeed..3f7db05 100644
--- a/lib/jit_arm-cpu.c
+++ b/lib/jit_arm-cpu.c
@@ -833,7 +833,7 @@ static void _torl(jit_state_t*,int,int,int) maybe_unused;
        jit_unget_reg(_R0);                                             \
     } while (0)
 #  define nop(i0)                      _nop(_jit,i0)
-static void _nop(jit_state_t*,jit_word_t) maybe_unused;
+static void _nop(jit_state_t*,jit_int32_t);
 #  define movr(r0,r1)                  _movr(_jit,r0,r1)
 static void _movr(jit_state_t*,jit_int32_t,jit_int32_t);
 #  define movi(r0,i0)                  _movi(_jit,r0,i0)
@@ -1498,9 +1498,8 @@ _torl(jit_state_t *_jit, int o, int rn, int im)
 }
 
 static void
-_nop(jit_state_t *_jit, jit_word_t i0)
+_nop(jit_state_t *_jit, jit_int32_t i0)
 {
-    assert(i0 >= 0);
     if (jit_thumb_p()) {
        for (; i0 > 0; i0 -= 2)
            T1_NOP();
@@ -1509,6 +1508,7 @@ _nop(jit_state_t *_jit, jit_word_t i0)
        for (; i0 > 0; i0 -= 4)
            NOP();
     }
+    assert(i0 == 0);
 }
 
 static void
diff --git a/lib/jit_arm-sz.c b/lib/jit_arm-sz.c
index db7751e..a9f217b 100644
--- a/lib/jit_arm-sz.c
+++ b/lib/jit_arm-sz.c
@@ -9,6 +9,7 @@
     0,
     0,
     0,
+    0,
     26,
     0,
     4,
@@ -361,6 +362,7 @@
     0,
     0,
     0,
+    0,
     18,
     0,
     4,
diff --git a/lib/jit_arm.c b/lib/jit_arm.c
index 7dbf78a..179bf68 100644
--- a/lib/jit_arm.c
+++ b/lib/jit_arm.c
@@ -1043,6 +1043,13 @@ _emit_code(jit_state_t *_jit)
        value = jit_classify(node->code);
        jit_regarg_set(node, value);
        switch (node->code) {
+           case jit_code_align:
+               assert(!(node->u.w & (node->u.w - 1)) &&
+                      node->u.w <= sizeof(jit_word_t));
+               if (node->u.w == sizeof(jit_word_t) &&
+                   (word = _jit->pc.w & (sizeof(jit_word_t) - 1)))
+                   nop(sizeof(jit_word_t) - word);
+               break;
            case jit_code_note:         case jit_code_name:
                if (must_align_p(node->next))
                    nop(2);
diff --git a/lib/jit_hppa-cpu.c b/lib/jit_hppa-cpu.c
index 479d6cf..3fec623 100644
--- a/lib/jit_hppa-cpu.c
+++ b/lib/jit_hppa-cpu.c
@@ -640,6 +640,8 @@ static void _f38(jit_state_t*,jit_int32_t,
 #define UADDCM(r1,r2,t)                f8(0x2,r2,r1,ADD_CF_NV,2,0,1,2,0,t)
 #define UXOR(r1,r2,t)          f8(0x2,r2,r1,LOG_CC_NV,0,1,1,2,0,t)
 #define XOR(r1,r2,t)           f8(0x2,r2,r1,LOG_CC_NV,0,1,0,2,0,t)
+#  define nop(c)                       _nop(_jit,c)
+static void _nop(jit_state_t*,jit_int32_t);
 #define movr(r0,r1)            _movr(_jit,r0,r1)
 static void _movr(jit_state_t*,jit_int32_t,jit_int32_t);
 #define movi(r0,i0)            _movi(_jit,r0,i0)
@@ -1586,6 +1588,14 @@ _f38(jit_state_t *_jit, jit_int32_t o,
 }
 
 static void
+_nop(jit_state_t *_jit, jit_int32_t i0)
+{
+    for (; i0 > 0; i0 -= 4)
+       NOP();
+    assert(i0 == 0);
+}
+
+static void
 _movr(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
 {
     if (r0 != r1)
diff --git a/lib/jit_hppa-sz.c b/lib/jit_hppa-sz.c
index ab8c77d..f2b1918 100644
--- a/lib/jit_hppa-sz.c
+++ b/lib/jit_hppa-sz.c
@@ -8,6 +8,7 @@
     0,
     0,
     0,
+    0,
     64,
     0,
     4,
diff --git a/lib/jit_hppa.c b/lib/jit_hppa.c
index 289542f..636f364 100644
--- a/lib/jit_hppa.c
+++ b/lib/jit_hppa.c
@@ -722,6 +722,13 @@ _emit_code(jit_state_t *_jit)
        value = jit_classify(node->code);
        jit_regarg_set(node, value);
        switch (node->code) {
+           case jit_code_align:
+               assert(!(node->u.w & (node->u.w - 1)) &&
+                      node->u.w <= sizeof(jit_word_t));
+               if (node->u.w == sizeof(jit_word_t) &&
+                   (word = _jit->pc.w & (sizeof(jit_word_t) - 1)))
+                   nop(sizeof(jit_word_t) - word);
+               break;
            case jit_code_note:         case jit_code_name:
                node->u.w = _jit->pc.w;
                break;
diff --git a/lib/jit_ia64-sz.c b/lib/jit_ia64-sz.c
index 1185ada..5d88500 100644
--- a/lib/jit_ia64-sz.c
+++ b/lib/jit_ia64-sz.c
@@ -5,6 +5,7 @@
     0,
     0,
     0,
+    0,
     16,
     16,
     48,
diff --git a/lib/jit_ia64.c b/lib/jit_ia64.c
index 0a9eccd..e8eee8a 100644
--- a/lib/jit_ia64.c
+++ b/lib/jit_ia64.c
@@ -859,6 +859,11 @@ _emit_code(jit_state_t *_jit)
 #endif
        jit_regarg_set(node, value);
        switch (node->code) {
+           case jit_code_align:
+               assert(!(node->u.w & (node->u.w - 1)) &&
+                      node->u.w <= sizeof(jit_word_t));
+               /* nothing done */
+               break;
            case jit_code_note:         case jit_code_name:
                sync();
                node->u.w = _jit->pc.w;
diff --git a/lib/jit_mips-cpu.c b/lib/jit_mips-cpu.c
index 8976b31..9e9778f 100644
--- a/lib/jit_mips-cpu.c
+++ b/lib/jit_mips-cpu.c
@@ -303,7 +303,8 @@ 
_hrrrit(jit_state_t*,jit_int32_t,jit_int32_t,jit_int32_t,jit_int32_t,
 static void 
_hrri(jit_state_t*,jit_int32_t,jit_int32_t,jit_int32_t,jit_int32_t);
 #  define hi(hc,im)                    _hi(_jit,hc,im)
 static void _hi(jit_state_t*,jit_int32_t,jit_int32_t);
-#  define NOP(i0)                      _nop(_jit,i0)
+#  define NOP(i0)                      ii(0)
+#  define nop(i0)                      _nop(_jit,i0)
 static void _nop(jit_state_t*,jit_int32_t);
 #  define h_ri(hc,rt,im)               _hrri(_jit,hc,0,rt,im)
 #  define rrit(rt,rd,im,tc)            _hrrrit(_jit,0,0,rt,rd,im,tc)
@@ -769,8 +770,9 @@ _hi(jit_state_t *_jit, jit_int32_t hc, jit_int32_t im)
 static void
 _nop(jit_state_t *_jit, jit_int32_t i0)
 {
-    while (i0--)
-       ii(0);
+    for (; i0 > 0; i0 -= 4)
+       NOP();
+    assert(i0 == 0);
 }
 
 static void
diff --git a/lib/jit_mips-sz.c b/lib/jit_mips-sz.c
index 69f6dcf..9b5b57b 100644
--- a/lib/jit_mips-sz.c
+++ b/lib/jit_mips-sz.c
@@ -9,6 +9,7 @@
     0,
     0,
     0,
+    0,
     44,
     0,
     4,
@@ -361,6 +362,7 @@
     0,
     0,
     0,
+    0,
     96,
     0,
     4,
@@ -712,6 +714,7 @@
     0,
     0,
     0,
+    0,
     44,
     0,
     4,
diff --git a/lib/jit_mips.c b/lib/jit_mips.c
index 3eec489..c31d43b 100644
--- a/lib/jit_mips.c
+++ b/lib/jit_mips.c
@@ -978,6 +978,13 @@ _emit_code(jit_state_t *_jit)
        value = jit_classify(node->code);
        jit_regarg_set(node, value);
        switch (node->code) {
+           case jit_code_align:
+               assert(!(node->u.w & (node->u.w - 1)) &&
+                      node->u.w <= sizeof(jit_word_t));
+               if (node->u.w == sizeof(jit_word_t) &&
+                   (word = _jit->pc.w & (sizeof(jit_word_t) - 1)))
+                   nop(sizeof(jit_word_t) - word);
+               break;
            case jit_code_note:         case jit_code_name:
                node->u.w = _jit->pc.w;
                break;
diff --git a/lib/jit_ppc-cpu.c b/lib/jit_ppc-cpu.c
index ed93065..844c8b1 100644
--- a/lib/jit_ppc-cpu.c
+++ b/lib/jit_ppc-cpu.c
@@ -451,6 +451,8 @@ static void _FXS(jit_state_t*,int,int,int,int,int,int,int);
 #  define XOR_(d,a,b)                  FXO_(31,a,d,b,0,316)
 #  define XORI(s,a,u)                  FDu(26,s,a,u)
 #  define XORIS(s,a,u)                 FDu(27,s,a,u)
+#  define nop(c)                       _nop(_jit,c)
+static void _nop(jit_state_t*,jit_int32_t);
 #  define movr(r0,r1)                  _movr(_jit,r0,r1)
 static void _movr(jit_state_t*,jit_int32_t,jit_int32_t);
 #  define movi(r0,i0)                  _movi(_jit,r0,i0)
@@ -977,6 +979,14 @@ _FXS(jit_state_t *_jit, int o, int s, int a, int h, int x, 
int i, int r)
 #endif
 
 static void
+_nop(jit_state_t *_jit, jit_int32_t i0)
+{
+    for (; i0 > 0; i0 -= 4)
+       NOP();
+    assert(i0 == 0);
+}
+
+static void
 _movr(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)
 {
     if (r0 != r1)
diff --git a/lib/jit_ppc-sz.c b/lib/jit_ppc-sz.c
index e0c53b9..6025ffc 100644
--- a/lib/jit_ppc-sz.c
+++ b/lib/jit_ppc-sz.c
@@ -9,6 +9,7 @@
     0,
     0,
     0,
+    0,
     44,
     0,
     4,
@@ -361,6 +362,7 @@
     0,
     0,
     0,
+    0,
     72,
     0,
     4,
@@ -713,6 +715,7 @@
     0,
     0,
     0,
+    0,
     72,
     0,
     4,
diff --git a/lib/jit_ppc.c b/lib/jit_ppc.c
index 08da19f..88488f0 100644
--- a/lib/jit_ppc.c
+++ b/lib/jit_ppc.c
@@ -875,6 +875,13 @@ _emit_code(jit_state_t *_jit)
        value = jit_classify(node->code);
        jit_regarg_set(node, value);
        switch (node->code) {
+           case jit_code_align:
+               assert(!(node->u.w & (node->u.w - 1)) &&
+                      node->u.w <= sizeof(jit_word_t));
+               if (node->u.w == sizeof(jit_word_t) &&
+                   (word = _jit->pc.w & (sizeof(jit_word_t) - 1)))
+                   nop(sizeof(jit_word_t) - word);
+               break;
            case jit_code_note:         case jit_code_name:
                node->u.w = _jit->pc.w;
                break;
diff --git a/lib/jit_print.c b/lib/jit_print.c
index 17d2559..49a5536 100644
--- a/lib/jit_print.c
+++ b/lib/jit_print.c
@@ -38,7 +38,7 @@
  */
 static char *code_name[] = {
     "data",
-    "live",
+    "live",            "align",
     "save",            "load",
     "#name",           "#note",
     "label",
diff --git a/lib/jit_s390x-sz.c b/lib/jit_s390x-sz.c
index e7b8ea2..d3df0c6 100644
--- a/lib/jit_s390x-sz.c
+++ b/lib/jit_s390x-sz.c
@@ -7,6 +7,7 @@
     0,
     0,
     0,
+    0,
     2,
     38,
     0,
diff --git a/lib/jit_s390x.c b/lib/jit_s390x.c
index 99802a5..4c45db0 100644
--- a/lib/jit_s390x.c
+++ b/lib/jit_s390x.c
@@ -697,11 +697,19 @@ _emit_code(jit_state_t *_jit)
        value = jit_classify(node->code);
        jit_regarg_set(node, value);
        switch (node->code) {
+           case jit_code_align:
+               assert(!(node->u.w & (node->u.w - 1)) &&
+                      node->u.w <= sizeof(jit_word_t));
+               if (node->u.w == sizeof(jit_word_t) &&
+                   (word = _jit->pc.w & (sizeof(jit_word_t) - 1)))
+                   nop(sizeof(jit_word_t) - word);
+               break;
            case jit_code_note:         case jit_code_name:
                node->u.w = _jit->pc.w;
                break;
            case jit_code_label:
-               if (node->link && (word = _jit->pc.w & 3))
+               if ((node->link || (node->flag & jit_flag_use)) &&
+                   (word = _jit->pc.w & 3))
                    nop(4 - word);
                /* remember label is defined */
                node->flag |= jit_flag_patch;
diff --git a/lib/jit_sparc-cpu.c b/lib/jit_sparc-cpu.c
index ba56c61..095c364 100644
--- a/lib/jit_sparc-cpu.c
+++ b/lib/jit_sparc-cpu.c
@@ -809,8 +809,9 @@ _f1(jit_state_t *_jit, jit_int32_t op, jit_int32_t disp30)
 static void
 _nop(jit_state_t *_jit, jit_int32_t i0)
 {
-    while (i0--)
+    for (; i0 > 0; i0 -= 4)
        NOP();
+    assert(i0 == 0);
 }
 
 static void
diff --git a/lib/jit_sparc-sz.c b/lib/jit_sparc-sz.c
index 6056af2..730a16a 100644
--- a/lib/jit_sparc-sz.c
+++ b/lib/jit_sparc-sz.c
@@ -8,6 +8,7 @@
     0,
     0,
     0,
+    0,
     16,
     0,
     4,
diff --git a/lib/jit_sparc.c b/lib/jit_sparc.c
index 81a9f5c..a8488e2 100644
--- a/lib/jit_sparc.c
+++ b/lib/jit_sparc.c
@@ -715,11 +715,18 @@ _emit_code(jit_state_t *_jit)
        value = jit_classify(node->code);
        jit_regarg_set(node, value);
        switch (node->code) {
+           case jit_code_align:
+               assert(!(node->u.w & (node->u.w - 1)) &&
+                      node->u.w <= sizeof(jit_word_t));
+               if (node->u.w == sizeof(jit_word_t) &&
+                   (word = _jit->pc.w & (sizeof(jit_word_t) - 1)))
+                   nop(sizeof(jit_word_t) - word);
+               break;
            case jit_code_note:         case jit_code_name:
                node->u.w = _jit->pc.w;
                break;
            case jit_code_label:
-               if (node->link &&
+               if ((node->link || (node->flag & jit_flag_use)) &&
                    (word = _jit->pc.w & (sizeof(jit_word_t) - 1)))
                    nop(sizeof(jit_word_t) - word);
                /* remember label is defined */
diff --git a/lib/jit_x86-sz.c b/lib/jit_x86-sz.c
index b8c9af3..52101e4 100644
--- a/lib/jit_x86-sz.c
+++ b/lib/jit_x86-sz.c
@@ -7,6 +7,7 @@
     0,
     0,
     0,
+    0,
     3,
     26,
     0,
@@ -707,6 +708,7 @@
     0,
     0,
     0,
+    0,
     7,
     43,
     0,
diff --git a/lib/jit_x86.c b/lib/jit_x86.c
index 747741d..20dcc21 100644
--- a/lib/jit_x86.c
+++ b/lib/jit_x86.c
@@ -1268,11 +1268,17 @@ _emit_code(jit_state_t *_jit)
        value = jit_classify(node->code);
        jit_regarg_set(node, value);
        switch (node->code) {
+           case jit_code_align:
+               assert(!(node->u.w & (node->u.w - 1)) &&
+                      node->u.w <= sizeof(jit_word_t));
+               if ((word = _jit->pc.w & (node->u.w - 1)))
+                   nop(node->u.w - word);
+               break;
            case jit_code_note:         case jit_code_name:
                node->u.w = _jit->pc.w;
                break;
            case jit_code_label:
-               if (node->link &&
+               if ((node->link || (node->flag & jit_flag_use)) &&
                    (word = _jit->pc.w & (sizeof(jit_word_t) - 1)))
                    nop(sizeof(jit_word_t) - word);
                /* remember label is defined */
diff --git a/lib/lightning.c b/lib/lightning.c
index 42c3a45..9ee261d 100644
--- a/lib/lightning.c
+++ b/lib/lightning.c
@@ -1177,9 +1177,9 @@ _jit_classify(jit_state_t *_jit, jit_code_t code)
     jit_int32_t                mask;
 
     switch (code) {
-       case jit_code_data:     case jit_code_save:     case jit_code_load:
-       case jit_code_name:     case jit_code_label:    case jit_code_note:
-       case jit_code_prolog:   case jit_code_epilog:
+       case jit_code_data:     case jit_code_align:    case jit_code_save:
+       case jit_code_load:     case jit_code_name:     case jit_code_label:
+       case jit_code_note:     case jit_code_prolog:   case jit_code_epilog:
            mask = 0;
            break;
        case jit_code_live:
@@ -2395,7 +2395,8 @@ _redundant_jump(jit_state_t *_jit, jit_node_t *prev, 
jit_node_t *node)
                    return (1);
                }
                break;
-           case jit_code_note:
+           case jit_code_name:         case jit_code_note:
+           case jit_code_align:
                break;
            default:
                return (0);