lightning.git - Portable just-in-time compiler library

diff --git a/check/Makefile.am b/check/Makefile.am
index c77f5cd..2a96f8a 100644
--- a/check/Makefile.am
+++ b/check/Makefile.am

@@ -124,7 +124,8 @@ EXTRA_DIST = \

check.nodata.sh \

check.x87.nodata.sh \

run-test all.tst \

- collatz.tst factorial.tst

+ collatz.tst factorial.tst \

+ rbit.tst

base_TESTS = \

3to2 add align allocai \

diff --git a/check/lightning.c b/check/lightning.c
index 80ea081..52a313b 100644
--- a/check/lightning.c
+++ b/check/lightning.c

@@ -323,6 +323,7 @@ static void rshr_u(void); static void rshi_u(void);

static void negr(void); static void comr(void);

static void clor(void); static void clzr(void);

static void ctor(void); static void ctzr(void);

+static void rbitr(void);

static void ltr(void); static void lti(void);

static void ltr_u(void); static void lti_u(void);

static void ler(void); static void lei(void);

@@ -683,6 +684,7 @@ static instr_t instr_vector[] = {

entry(negr), entry(comr),

entry(clor), entry(clzr),

entry(ctor), entry(ctzr),

+ entry(rbitr),

entry(ltr), entry(lti),

entry(ltr_u), entry(lti_u),

entry(ler), entry(lei),

@@ -1536,6 +1538,7 @@ entry_ir_ir_ir(rshr_u) entry_ir_ir_im(rshi_u)

entry_ir_ir(negr) entry_ir_ir(comr)

entry_ir_ir(clor) entry_ir_ir(clzr)

entry_ir_ir(ctor) entry_ir_ir(ctzr)

+entry_ir_ir(rbitr)

entry_ir_ir_ir(ltr) entry_ir_ir_im(lti)

entry_ir_ir_ir(ltr_u) entry_ir_ir_im(lti_u)

entry_ir_ir_ir(ler) entry_ir_ir_im(lei)

diff --git a/check/rbit.tst b/check/rbit.tst
new file mode 100644
index 0000000..bc8bce3
--- /dev/null
+++ b/check/rbit.tst

@@ -0,0 +1,184 @@

+.data 4096

+swap_tab:

+.c 0 128 64 192 32 160 96 224 16 144 80 208 48 176 112 240 8 136 72 200 40 168 104 232 24 152 88 216 56 184 120 248 4 132 68 196 36 164 100 228 20 148 84 212 52 180 116 244 12 140 76 204 44 172 108 236 28 156 92 220 60 188 124 252 2 130 66 194 34 162 98 226 18 146 82 210 50 178 114 242 10 138 74 202 42 170 106 234 26 154 90 218 58 186 122 250 6 134 70 198 38 166 102 230 22 150 86 214 54 182 118 246 14 142 78 206 46 174 110 238 30 158 94 222 62 190 126 254 1 129 65 193 33 161 97 225 17 145 81 209 49 177 113 241 9 137 73 201 41 169 105 233 25 153 89 217 57 185 121 249 5 133 69 197 37 165 101 229 21 149 85 213 53 181 117 245 13 141 77 205 45 173 109 237 29 157 93 221 61 189 125 253 3 131 67 195 35 163 99 227 19 147 83 211 51 179 115 243 11 139 75 203 43 171 107 235 27 155 91 219 59 187 123 251 7 135 71 199 39 167 103 231 23 151 87 215 55 183 119 247 15 143 79 207 47 175 111 239 31 159 95 223 63 191 127 255

+fmt:

+#if __WORDSIZE == 32

+.c "0x%08lx = 0x%08lx\n"

+#else

+.c "0x%016lx = 0x%016lx\n"

+#endif

+

+.code

+ jmpi main

+ name rbit_table

+rbit_table:

+ prolog

+ arg $in

+ getarg %r1 $in

+ extr_uc %r2 %r1

+ movi %v0 swap_tab

+ ldxr_uc %r0 %v0 %r2

+ movi %v1 8

+rbit_table_loop:

+ rshr %r2 %r1 %v1

+ extr_uc %r2 %r2

+ lshi %r0 %r0 8

+ ldxr_uc %r2 %v0 %r2

+ orr %r0 %r0 %r2

+ addi %v1 %v1 8

+ blti rbit_table_loop %v1 __WORDSIZE

+ retr %r0

+ epilog

+

+ name rbit_unrolled

+rbit_unrolled:

+ prolog

+ arg $in

+ getarg %r0 $in

+#if __WORDSIZE == 32

+ movi %r1 0x55555555

+#else

+ movi %r1 0x5555555555555555

+#endif

+ rshi_u %r2 %r0 1 // r2 = r0 >> 1

+ andr %r2 %r2 %r1 // r2 &= r1

+ andr %v0 %r0 %r1 // v0 = r0 & r1

+ lshi %v0 %v0 1 // v0 <<= 1

+ orr %r0 %r2 %v0 // r0 = r2 | v0

+#if __WORDSIZE == 32

+ movi %r1 0x33333333

+#else

+ movi %r1 0x3333333333333333

+#endif

+ rshi_u %r2 %r0 2 // r2 = r0 >> 2

+ andr %r2 %r2 %r1 // r2 &= r1

+ andr %v0 %r0 %r1 // v0 = r0 & r1

+ lshi %v0 %v0 2 // v0 <<= 2

+ orr %r0 %r2 %v0 // r0 = r2 | v0

+#if __WORDSIZE == 32

+ movi %r1 0x0f0f0f0f

+#else

+ movi %r1 0x0f0f0f0f0f0f0f0f

+#endif

+ rshi_u %r2 %r0 4 // r2 = r0 >> 4

+ andr %r2 %r2 %r1 // r2 &= r1

+ andr %v0 %r0 %r1 // v0 = r0 & r1

+ lshi %v0 %v0 4 // v0 <<= 4

+ orr %r0 %r2 %v0 // r0 = r2 | v0

+#if __WORDSIZE == 32

+ movi %r1 0x00ff00ff

+#else

+ movi %r1 0x00ff00ff00ff00ff

+#endif

+ rshi_u %r2 %r0 8 // r2 = r0 >> 8

+ andr %r2 %r2 %r1 // r2 &= r1

+ andr %v0 %r0 %r1 // v0 = r0 & r1

+ lshi %v0 %v0 8 // v0 <<= 8

+ orr %r0 %r2 %v0 // r0 = r2 | v0

+#if __WORDSIZE == 32

+ rshi_u %r2 %r0 16 // r2 = r0 >> 16

+ lshi %v0 %r0 16 // v0 = r0 << 16

+ orr %r0 %r2 %v0 // r0 = r2 | v0

+#else

+ movi %r1 0x0000ffff0000ffff

+ rshi_u %r2 %r0 16 // r2 = r0 >> 16

+ andr %r2 %r2 %r1 // r2 &= r1

+ andr %v0 %r0 %r1 // v0 = r0 & r1

+ lshi %v0 %v0 16 // v0 <<= 16

+ orr %r0 %r2 %v0 // r0 = r2 | v0

+ rshi_u %r2 %r0 32 // r2 = r0 >> 32

+ lshi %v0 %r0 32 // v0 = r0 << 32

+ orr %r0 %r2 %v0 // r0 = r2 | v0

+#endif

+ retr %r0

+ epilog

+

+ name rbit_loop

+rbit_loop:

+ prolog

+ arg $in

+ getarg %r0 $in

+ movi %r1 __WORDSIZE

+ movi %r2 $(~0)

+rbit_loop_loop: // while (%r1 >>= 1) > 0

+ rshi %r1 %r1 1 // %r1 >>= 1

+ blei rbit_loop_done %r1 0 // no loop if %r1 <= 0

+ lshr %v0 %r2 %r1 // %v0 = %r2 << %r1

+ xorr %r2 %r2 %v0 // %r2 ^= %v0

+ rshr %v0 %r0 %r1 // %v0 = %r0 >> %r1

+ andr %v0 %v0 %r2 // %r2 = %v0 & %r2

+ lshr %v1 %r0 %r1 // %v1 = %r0 << %r1

+ comr %r0 %r2 // %r0 = ~%r2

+ andr %v1 %r0 %v1 // %v1 &= %r0

+ orr %r0 %v0 %v1 // %r0 = %v0 | %v1

+ jmpi rbit_loop_loop

+rbit_loop_done:

+ retr %r0

+ epilog

+

+ name main

+main:

+ prolog

+ arg $argc

+ arg $argv

+ getarg %r0 $argc

+ bnei default %r0 2

+ getarg %v0 $argv

+ ldxi %r0 %v0 $(__WORDSIZE >> 3)

+ prepare

+ pushargr %r0

+ pushargi 0

+ pushargi 0

+ finishi @strtoul

+ retval %v0

+ jmpi main_do

+default:

+#if __WORDSIZE == 32

+ movi %v0 0x8a13c851

+#else

+ movi %v0 0x984a137ffec85219

+#endif

+main_do:

+ prepare

+ pushargr %v0

+ finishi rbit_table

+ retval %r0

+ prepare

+ pushargi fmt

+ ellipsis

+ pushargr %v0

+ pushargr %r0

+ finishi @printf

+

+ prepare

+ pushargr %v0

+ finishi rbit_unrolled

+ retval %r0

+ prepare

+ pushargi fmt

+ ellipsis

+ pushargr %v0

+ pushargr %r0

+ finishi @printf

+

+ prepare

+ pushargr %v0

+ finishi rbit_loop

+ retval %r0

+ prepare

+ pushargi fmt

+ ellipsis

+ pushargr %v0

+ pushargr %r0

+ finishi @printf

+

+ rbitr %r0 %v0

+ prepare

+ pushargi fmt

+ ellipsis

+ pushargr %v0

+ pushargr %r0

+ finishi @printf

+

+ ret

+ epilog

diff --git a/doc/body.texi b/doc/body.texi
index 1bd3f67..e7b090d 100644
--- a/doc/body.texi
+++ b/doc/body.texi

@@ -291,10 +291,11 @@ These accept two operands, both of which must be registers.

@example

negr _f _d O1 = -O2

comr O1 = ~O2

-clor O1 = number of leading one bits

-clzr O1 = number of leading zero bits

-ctor O1 = number of trailing one bits

-ctzr O1 = number of trailing zero bits

+clor O1 = number of leading one bits in O2

+clzr O1 = number of leading zero bits in O2

+ctor O1 = number of trailing one bits in O2

+ctzr O1 = number of trailing zero bits in O2

+rbitr O1 = bits of O2 reversed

@end example

Note that @code{ctzr} is basically equivalent of a @code{C} call

diff --git a/include/lightning.h.in b/include/lightning.h.in
index 7aa654c..66e3068 100644
--- a/include/lightning.h.in
+++ b/include/lightning.h.in

@@ -1056,6 +1056,9 @@ typedef enum {

#define jit_ctzr(u,v) jit_new_node_ww(jit_code_ctzr,u,v)

jit_code_ctor, jit_code_ctzr,

+#define jit_rbitr(u,v) jit_new_node_ww(jit_code_rbitr,u,v)

+ jit_code_rbitr,

+

jit_code_last_code

} jit_code_t;

diff --git a/lib/jit_aarch64-cpu.c b/lib/jit_aarch64-cpu.c
index d5e64ad..99d8756 100644
--- a/lib/jit_aarch64-cpu.c
+++ b/lib/jit_aarch64-cpu.c

@@ -598,6 +598,7 @@ static void _clzr(jit_state_t*, jit_int32_t, jit_int32_t);

static void _ctor(jit_state_t*, jit_int32_t, jit_int32_t);

# define ctzr(r0, r1) _ctzr(_jit, r0, r1)

static void _ctzr(jit_state_t*, jit_int32_t, jit_int32_t);

+# define rbitr(r0, r1) RBIT(r0, r1)

# define andr(r0,r1,r2) AND(r0,r1,r2)

# define andi(r0,r1,i0) _andi(_jit,r0,r1,i0)

static void _andi(jit_state_t*,jit_int32_t,jit_int32_t,jit_word_t);

diff --git a/lib/jit_aarch64.c b/lib/jit_aarch64.c
index 243e677..8106a31 100644
--- a/lib/jit_aarch64.c
+++ b/lib/jit_aarch64.c

@@ -1448,6 +1448,7 @@ _emit_code(jit_state_t *_jit)

case_rr(clz,);

case_rr(cto,);

case_rr(ctz,);

+ case_rr(rbit,);

case_rrr(and,);

case_rrw(and,);

case_rrr(or,);

diff --git a/lib/jit_alpha.c b/lib/jit_alpha.c
index 25566f4..ef41c37 100644
--- a/lib/jit_alpha.c
+++ b/lib/jit_alpha.c

@@ -64,6 +64,7 @@ static void _patch(jit_state_t*,jit_word_t,jit_node_t*);

#define PROTO 1

# include "jit_alpha-cpu.c"

# include "jit_alpha-fpu.c"

+# include "jit_fallback.c"

#undef PROTO

/*

@@ -1134,6 +1135,8 @@ _emit_code(jit_state_t *_jit)

case_rr(clz,);

case_rr(cto,);

case_rr(ctz,);

+#define rbitr(r0, r1) fallback_rbit(r0, r1)

+ case_rr(rbit,);

case_rrr(lt,);

case_rrw(lt,);

case_rrr(lt, _u);

@@ -1555,6 +1558,7 @@ _emit_code(jit_state_t *_jit)

#define CODE 1

# include "jit_alpha-cpu.c"

# include "jit_alpha-fpu.c"

+# include "jit_fallback.c"

#undef CODE

void

diff --git a/lib/jit_arm-cpu.c b/lib/jit_arm-cpu.c
index a0852a2..78c5814 100644
--- a/lib/jit_arm-cpu.c
+++ b/lib/jit_arm-cpu.c

@@ -914,6 +914,8 @@ static void _clzr(jit_state_t*, jit_int32_t, jit_int32_t);

static void _ctor(jit_state_t*, jit_int32_t, jit_int32_t);

# define ctzr(r0, r1) _ctzr(_jit, r0, r1)

static void _ctzr(jit_state_t*, jit_int32_t, jit_int32_t);

+# define rbitr(r0, r1) _rbitr(_jit, r0, r1)

+static void _rbitr(jit_state_t*, jit_int32_t, jit_int32_t);

# define addr(r0,r1,r2) _addr(_jit,r0,r1,r2)

static void _addr(jit_state_t*,jit_int32_t,jit_int32_t,jit_int32_t);

# define addi(r0,r1,i0) _addi(_jit,r0,r1,i0)

@@ -1814,6 +1816,19 @@ _ctzr(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)

}

static void

+_rbitr(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)

+{

+ if (jit_armv7_p()) { /* armv6t2 actually */

+ if (jit_thumb_p())

+ T2_RBIT(r0, r1);

+ else

+ RBIT(r0, r1);

+ }

+ else

+ fallback_bitswap(r0, r1);

+}

+

+static void

_addr(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_int32_t r2)

{

if (jit_thumb_p()) {

diff --git a/lib/jit_fallback.c b/lib/jit_fallback.c
index 21e2f42..935f3e4 100644
--- a/lib/jit_fallback.c
+++ b/lib/jit_fallback.c

@@ -672,7 +672,7 @@ while ((s >>= 1) > 0)

comr(v, rn(mask)); /* v = ~mask */

andr(rn(t1), v, rn(t1)); /* t1 = t1 & v */

orr(v, rn(t0), rn(t1)); /* v = t0 | t1 */

- jmpi(loop, 0);

+ jmpi(loop);

flush();

patch_at(done, _jit->pc.w);

jit_unget_reg(t1);

diff --git a/lib/jit_hppa.c b/lib/jit_hppa.c
index d3c5ef7..d683dc1 100644
--- a/lib/jit_hppa.c
+++ b/lib/jit_hppa.c

@@ -1072,10 +1072,12 @@ _emit_code(jit_state_t *_jit)

#define clzr(r0, r1) fallback_clz(r0, r1)

#define ctor(r0, r1) fallback_cto(r0, r1)

#define ctzr(r0, r1) fallback_ctz(r0, r1)

+#define rbitr(r0, r1) fallback_bitswap(r0, r1)

case_rr(clo,);

case_rr(clz,);

case_rr(cto,);

case_rr(ctz,);

+ case_rr(rbit,);

case_rr(ext, _c);

case_rr(ext, _uc);

case_rr(ext, _s);

diff --git a/lib/jit_ia64.c b/lib/jit_ia64.c
index 2968278..c45f784 100644
--- a/lib/jit_ia64.c
+++ b/lib/jit_ia64.c

@@ -1197,6 +1197,8 @@ _emit_code(jit_state_t *_jit)

case_rr(clz,);

case_rr(cto,);

case_rr(ctz,);

+#define rbitr(r0, r1) fallback_bitswap(r0, r1)

+ case_rr(rbit,);

case jit_code_casr:

casr(rn(node->u.w), rn(node->v.w),

rn(node->w.q.l), rn(node->w.q.h));

diff --git a/lib/jit_loongarch-cpu.c b/lib/jit_loongarch-cpu.c
index ab05852..2de50c4 100644
--- a/lib/jit_loongarch-cpu.c
+++ b/lib/jit_loongarch-cpu.c

@@ -338,6 +338,7 @@ static void _oj26(jit_state_t*, jit_int32_t,jit_int32_t);

# define clzr(r0, r1) CLZ_D(r0, r1)

# define ctor(r0, r1) CTO_D(r0, r1)

# define ctzr(r0, r1) CTZ_D(r0, r1)

+# define rbitr(r0, r1) BITREV_D(r0, r1)

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);

diff --git a/lib/jit_mips-cpu.c b/lib/jit_mips-cpu.c
index 5d4137c..19d34a2 100644
--- a/lib/jit_mips-cpu.c
+++ b/lib/jit_mips-cpu.c

@@ -501,6 +501,8 @@ static void _clzr(jit_state_t*, jit_int32_t, jit_int32_t);

static void _ctor(jit_state_t*, jit_int32_t, jit_int32_t);

# define ctzr(r0, r1) _ctzr(_jit, r0, r1)

static void _ctzr(jit_state_t*, jit_int32_t, jit_int32_t);

+# define rbitr(r0, r1) _rbitr(_jit, r0, r1)

+static void _rbitr(jit_state_t*, jit_int32_t, jit_int32_t);

# if __WORDSIZE == 32

# define addr(rd,rs,rt) ADDU(rd,rs,rt)

# define addiu(r0,r1,i0) ADDIU(r0,r1,i0)

@@ -1710,6 +1712,22 @@ _ctzr(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)

}

static void

+_rbitr(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)

+{

+ if (jit_mips6_p()) {

+#if __WORDSIZE == 32

+ BITSWAP(r0, r1);

+ bswapr_ui(r0, r0);

+#else

+ DBITSWAP(r0, r1);

+ bswapr_ul(r0, r0);

+#endif

+ }

+ else

+ fallback_bitswap(r0, r1);

+}

+

+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_mips.c b/lib/jit_mips.c
index b3a8078..07d4867 100644
--- a/lib/jit_mips.c
+++ b/lib/jit_mips.c

@@ -1676,6 +1676,7 @@ _emit_code(jit_state_t *_jit)

case_rr(clz,);

case_rr(cto,);

case_rr(ctz,);

+ case_rr(rbit,);

case_rrr(lt,);

case_rrw(lt,);

case_rrr(lt, _u);

diff --git a/lib/jit_names.c b/lib/jit_names.c
index e5985a3..29f5906 100644
--- a/lib/jit_names.c
+++ b/lib/jit_names.c

@@ -255,4 +255,5 @@ static char *code_name[] = {

"movr_d_w", "movi_d_w",

"clo", "clz",

"cto", "ctz",

+ "rbit",

};

diff --git a/lib/jit_ppc.c b/lib/jit_ppc.c
index 0ad4ae8..2410deb 100644
--- a/lib/jit_ppc.c
+++ b/lib/jit_ppc.c

@@ -1379,6 +1379,8 @@ _emit_code(jit_state_t *_jit)

case_rr(clz,);

case_rr(cto,);

case_rr(ctz,);

+#define rbitr(r0, r1) fallback_bitswap(r0, r1)

+ case_rr(rbit,);

case jit_code_casr:

casr(rn(node->u.w), rn(node->v.w),

rn(node->w.q.l), rn(node->w.q.h));

diff --git a/lib/jit_riscv.c b/lib/jit_riscv.c
index 63a5cd9..8afe32e 100644
--- a/lib/jit_riscv.c
+++ b/lib/jit_riscv.c

@@ -1168,10 +1168,12 @@ _emit_code(jit_state_t *_jit)

#define clzr(r0, r1) fallback_clz(r0, r1)

#define ctor(r0, r1) fallback_cto(r0, r1)

#define ctzr(r0, r1) fallback_ctz(r0, r1)

+#define rbitr(r0, r1) fallback_bitswap(r0, r1)

case_rr(clo,);

case_rr(clz,);

case_rr(cto,);

case_rr(ctz,);

+ case_rr(rbit,);

case_rrr(and,);

case_rrw(and,);

case_rrr(or,);

diff --git a/lib/jit_s390.c b/lib/jit_s390.c
index 25c6421..24869a7 100644
--- a/lib/jit_s390.c
+++ b/lib/jit_s390.c

@@ -1137,6 +1137,8 @@ _emit_code(jit_state_t *_jit)

case_rr(clz,);

case_rr(cto,);

case_rr(ctz,);

+#define rbitr(r0, r1) fallback_bitswap(r0, r1)

+ case_rr(rbit,);

case_rrr(and,);

case_rrw(and,);

case_rrr(or,);

diff --git a/lib/jit_sparc.c b/lib/jit_sparc.c
index 9e837d8..34cfd68 100644
--- a/lib/jit_sparc.c
+++ b/lib/jit_sparc.c

@@ -1561,6 +1561,8 @@ _emit_code(jit_state_t *_jit)

case_rr(clz,);

case_rr(cto,);

case_rr(ctz,);

+#define rbitr(r0, r1) fallback_bitswap(r0, r1)

+ case_rr(rbit,);

case_brr(blt,);

case_brw(blt,);

case_brr(blt, _u);

diff --git a/lib/jit_x86.c b/lib/jit_x86.c
index b409457..10a6ec9 100644
--- a/lib/jit_x86.c
+++ b/lib/jit_x86.c

@@ -140,6 +140,7 @@ static void _x87_from_sse_d(jit_state_t*,jit_int32_t,jit_int32_t);

# include "jit_x86-cpu.c"

# include "jit_x86-sse.c"

# include "jit_x86-x87.c"

+# include "jit_fallback.c"

#undef PROTO

/*

@@ -1809,6 +1810,8 @@ _emit_code(jit_state_t *_jit)

case_rr(clz,);

case_rr(cto,);

case_rr(ctz,);

+#define rbitr(r0, r1) fallback_bitswap(r0, r1)

+ case_rr(rbit,);

case_rrr(lt,);

case_rrw(lt,);

case_rrr(lt, _u);

@@ -2426,6 +2429,7 @@ _emit_code(jit_state_t *_jit)

# include "jit_x86-cpu.c"

# include "jit_x86-sse.c"

# include "jit_x86-x87.c"

+# include "jit_fallback.c"

#undef CODE

void

diff --git a/lib/lightning.c b/lib/lightning.c
index b0b0ef7..1376773 100644
--- a/lib/lightning.c
+++ b/lib/lightning.c

@@ -1469,7 +1469,7 @@ _jit_classify(jit_state_t *_jit, jit_code_t code)

case jit_code_extr_f: case jit_code_extr_d_f: case jit_code_ldr_f:

case jit_code_negr_d: case jit_code_absr_d: case jit_code_sqrtr_d:

case jit_code_movr_d: case jit_code_extr_d: case jit_code_extr_f_d:

- case jit_code_ldr_d:

+ case jit_code_ldr_d: case jit_code_rbitr:

case jit_code_clor: case jit_code_clzr:

case jit_code_ctor: case jit_code_ctzr:

case jit_code_movr_w_f: case jit_code_movr_f_w: