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

diff --git a/include/lightning/jit_loongarch.h b/include/lightning/jit_loongarch.h
index 89b1a86..72e9f1f 100644
--- a/include/lightning/jit_loongarch.h
+++ b/include/lightning/jit_loongarch.h

@@ -74,4 +74,15 @@ typedef enum {

_NOREG,

} jit_reg_t;

+typedef struct {

+ /* generate special instructions for unaligned load/store? */

+ /* It is not guaranteed unaligned memory access is supported. */

+ jit_uint32_t unaligned : 1;

+} jit_cpu_t;

+

+/*

+ * Initialization

+ */

+extern jit_cpu_t jit_cpu;

+

#endif /* _jit_loongarch_h */

diff --git a/lib/jit_loongarch-cpu.c b/lib/jit_loongarch-cpu.c
index 3c9d93f..ef8261f 100644
--- a/lib/jit_loongarch-cpu.c
+++ b/lib/jit_loongarch-cpu.c

@@ -18,6 +18,7 @@

*/

#if PROTO

+# define jit_unaligned_p() (jit_cpu.unaligned)

# define ii(i) *_jit->pc.ui++ = (i)

# define can_sign_extend_si12_p(s12) ((s12) <= 2047 && (s12) >= -2048)

# define can_zero_extend_u12_p(u12) ((u12) <= 4095 && (u12) >= 0)

@@ -476,6 +477,14 @@ static void _ldxi_ui(jit_state_t*, jit_int32_t, jit_int32_t, jit_word_t);

# define ldxr_l(r0, r1, r2) LDX_D(r0, r1, r2)

# define ldxi_l(r0, r1, i0) _ldxi_l(_jit, r0, r1, i0)

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

+# define unldr(r0, r1, i0) _unldr(_jit, r0, r1, i0)

+static void _unldr(jit_state_t*, jit_int32_t, jit_int32_t, jit_word_t);

+# define unldi(r0, i0, i1) _unldi(_jit, r0, i0, i1)

+static void _unldi(jit_state_t*, jit_int32_t, jit_word_t, jit_word_t);

+# define unldr_u(r0, r1, i0) _unldr_u(_jit, r0, r1, i0)

+static void _unldr_u(jit_state_t*, jit_int32_t, jit_int32_t, jit_word_t);

+# define unldi_u(r0, i0, i1) _unldi_u(_jit, r0, i0, i1)

+static void _unldi_u(jit_state_t*, jit_int32_t, jit_word_t, jit_word_t);

# define str_c(r0, r1) ST_B(r1, r0, 0)

# define sti_c(i0, r0) _sti_c(_jit, i0, r0)

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

@@ -500,6 +509,10 @@ static void _stxi_i(jit_state_t*, jit_word_t, jit_int32_t, jit_int32_t);

# define stxr_l(r0, r1, r2) STX_D(r2, r1, r0)

# define stxi_l(i0, r0, r1) _stxi_l(_jit, i0, r0, r1)

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

+#define unstr(r0, r1, i0) _unstr(_jit, r0, r1, i0)

+static void _unstr(jit_state_t*, jit_int32_t, jit_int32_t, jit_word_t);

+#define unsti(i0, r0, i1) _unsti(_jit, i0, r0, i1)

+static void _unsti(jit_state_t*, jit_word_t, jit_int32_t, jit_word_t);

# define bswapr_us(r0,r1) _bswapr_us(_jit,r0,r1)

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

# define bswapr_ui(r0,r1) _bswapr_ui(_jit,r0,r1)

@@ -1594,6 +1607,44 @@ _ldxi_l(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_word_t i0)

}

static void

+_unldr(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_word_t i0)

+{

+ if (jit_unaligned_p())

+ fallback_unldr(r0, r1, i0);

+ else

+ generic_unldr(r0, r1, i0);

+}

+

+static void

+_unldi(jit_state_t *_jit, jit_int32_t r0, jit_word_t i0, jit_word_t i1)

+{

+ jit_int32_t t0, r2;

+ if (jit_unaligned_p())

+ fallback_unldi(r0, i0, i1);

+ else

+ generic_unldi(r0, i0, i1);

+}

+

+static void

+_unldr_u(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_word_t i0)

+{

+ if (jit_unaligned_p())

+ fallback_unldr_u(r0, r1, i0);

+ else

+ generic_unldr_u(r0, r1, i0);

+}

+

+static void

+_unldi_u(jit_state_t *_jit, jit_int32_t r0, jit_word_t i0, jit_word_t i1)

+{

+ jit_int32_t t0, r2;

+ if (jit_unaligned_p())

+ fallback_unldi_u(r0, i0, i1);

+ else

+ generic_unldi_u(r0, i0, i1);

+}

+

+static void

_sti_c(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0)

{

jit_int32_t reg;

@@ -1706,6 +1757,24 @@ _stxi_l(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, jit_int32_t r1)

}

static void

+_unstr(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_word_t i0)

+{

+ if (jit_unaligned_p())

+ fallback_unstr(r0, r1, i0);

+ else

+ generic_unstr(r0, r1, i0);

+}

+

+static void

+_unsti(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, jit_word_t i1)

+{

+ if (jit_unaligned_p())

+ fallback_unsti(i0, r0, i1);

+ else

+ generic_unsti(i0, r0, i1);

+}

+

+static void

_bswapr_us(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1)

{

REVB_2H(r0, r1);

diff --git a/lib/jit_loongarch-fpu.c b/lib/jit_loongarch-fpu.c
index 2871de3..9e019d2 100644
--- a/lib/jit_loongarch-fpu.c
+++ b/lib/jit_loongarch-fpu.c

@@ -223,18 +223,24 @@ static void _ldi_f(jit_state_t*, jit_int32_t, jit_word_t);

# define ldxr_f(r0, r1, r2) FLDX_S(r0, r1, r2)

# define ldxi_f(r0, r1, i0) _ldxi_f(_jit, r0, r1, i0)

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

+# define unldr_x(r0, r1, i0) _unldr_x(_jit, r0, r1, i0)

+static void _unldr_x(jit_state_t*, jit_int32_t, jit_int32_t, jit_word_t);

+# define unldi_x(r0, i0, i1) _unldi_x(_jit, r0, i0, i1)

+static void _unldi_x(jit_state_t*, jit_int32_t, jit_word_t, jit_word_t);

# define str_f(r0, r1) FST_S(r1, r0, 0)

# define sti_f(i0, r0) _sti_f(_jit, i0, r0)

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

# define stxr_f(r0, r1, r2) FSTX_S(r2, r1, r0)

# define stxi_f(i0, r0, r1) _stxi_f(_jit, i0, r0, r1)

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

+#define unstr_x(r0, r1, i0) _unstr_x(_jit, r0, r1, i0)

+static void _unstr_x(jit_state_t*, jit_int32_t, jit_int32_t, jit_word_t);

+#define unsti_x(i0, r0, i1) _unsti_x(_jit, i0, r0, i1)

+static void _unsti_x(jit_state_t*, jit_word_t, jit_int32_t, jit_word_t);

# define movr_f(r0, r1) FMOV_S(r0, r1)

# define movi_f(r0, i0) _movi_f(_jit, r0, i0)

static void _movi_f(jit_state_t*, jit_int32_t, jit_float32_t);

# define movr_f_w(r0, r1) MOVFR2GR_S(r0, r1)

-# define movi_f_w(r0, im) _movi_f_w(_jit, r0, im)

-static void _movi_f_w(jit_state_t*, jit_int32_t, jit_float32_t);

# define movr_w_f(r0, r1) MOVGR2FR_W(r0, r1)

# define extr_d_f(r0, r1) FCVT_S_D(r0, r1)

# define ltr_f(r0, r1, r2) _ltr_f(_jit, r0, r1, r2)

@@ -385,8 +391,6 @@ static void _stxi_d(jit_state_t*, jit_word_t, jit_int32_t, jit_int32_t);

# define movi_d(r0, i0) _movi_d(_jit, r0, i0)

static void _movi_d(jit_state_t*, jit_int32_t, jit_float64_t);

# define movr_d_w(r0, r1) MOVFR2GR_D(r0, r1)

-# define movi_d_w(r0, im) _movi_d_w(_jit, r0, im)

-static void _movi_d_w(jit_state_t*, jit_int32_t, jit_float64_t);

# define movr_w_d(r0, r1) MOVGR2FR_D(r0, r1)

# define extr_f_d(r0, r1) FCVT_D_S(r0, r1)

# define ltr_d(r0, r1, r2) _ltr_d(_jit, r0, r1, r2)

@@ -615,6 +619,24 @@ _ldxi_f(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_word_t i0)

}

static void

+_unldr_x(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_word_t i0)

+{

+ if (jit_unaligned_p())

+ fallback_unldr_x(r0, r1, i0);

+ else

+ generic_unldr_x(r0, r1, i0);

+}

+

+static void

+_unldi_x(jit_state_t *_jit, jit_int32_t r0, jit_word_t i0, jit_word_t i1)

+{

+ if (jit_unaligned_p())

+ fallback_unldi_x(r0, i0, i1);

+ else

+ generic_unldi_x(r0, i0, i1);

+}

+

+static void

_sti_f(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0)

{

jit_int32_t t0;

@@ -643,6 +665,24 @@ _stxi_f(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, jit_int32_t r1)

}

static void

+_unstr_x(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_word_t i0)

+{

+ if (jit_unaligned_p())

+ fallback_unstr_x(r0, r1, i0);

+ else

+ generic_unstr_x(r0, r1, i0);

+}

+

+static void

+_unsti_x(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, jit_word_t i1)

+{

+ if (jit_unaligned_p())

+ fallback_unsti_x(i0, r0, i1);

+ else

+ fallback_unsti_x(i0, r0, i1);

+}

+

+static void

_movi_f(jit_state_t *_jit, jit_int32_t r0, jit_float32_t i0)

{

union {

@@ -662,17 +702,6 @@ _movi_f(jit_state_t *_jit, jit_int32_t r0, jit_float32_t i0)

}

static void

-_movi_f_w(jit_state_t *_jit, jit_int32_t r0, jit_float32_t i0)

-{

- union {

- jit_int32_t i;

- jit_float32_t f;

- } data;

- data.f = i0;

- movi(r0, data.i);

-}

-

-static void

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

{

FCMP_SLT_S(0, r1, r2);

@@ -1028,17 +1057,6 @@ _movi_d(jit_state_t *_jit, jit_int32_t r0, jit_float64_t i0)

}

static void

-_movi_d_w(jit_state_t *_jit, jit_int32_t r0, jit_float64_t i0)

-{

- union {

- jit_int64_t l;

- jit_float64_t d;

- } data;

- data.d = i0;

- movi(r0, data.l);

-}

-

-static void

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

{

FCMP_SLT_D(0, r1, r2);

diff --git a/lib/jit_loongarch.c b/lib/jit_loongarch.c
index f8853fe..1ee828a 100644
--- a/lib/jit_loongarch.c
+++ b/lib/jit_loongarch.c

@@ -46,6 +46,7 @@ static void _patch(jit_state_t*,jit_word_t,jit_node_t*);

/*

* Initialization

*/

+jit_cpu_t jit_cpu;

jit_register_t _rvs[] = {

{ rc(gpr) | 0x14, "$t8" },

{ rc(gpr) | 0x13, "$t7" },

@@ -135,6 +136,8 @@ _jit_init(jit_state_t *_jit)

{

_jitc->reglen = jit_size(_rvs) - 1;

jit_carry = _NOREG;

+ /* By default assume kernel or cpu will handle unaligned load/store */

+ jit_cpu.unaligned = 0;

}

void

@@ -1178,6 +1181,18 @@ _emit_code(jit_state_t *_jit)

case_rrw(ldx, _ui);

case_rrr(ldx, _l);

case_rrw(ldx, _l);

+ case jit_code_unldr:

+ unldr(rn(node->u.w), rn(node->v.w), node->w.w);

+ break;

+ case jit_code_unldi:

+ unldi(rn(node->u.w), node->v.w, node->w.w);

+ break;

+ case jit_code_unldr_u:

+ unldr_u(rn(node->u.w), rn(node->v.w), node->w.w);

+ break;

+ case jit_code_unldi_u:

+ unldi_u(rn(node->u.w), node->v.w, node->w.w);

+ break;

case_rr(st, _c);

case_wr(st, _c);

case_rr(st, _s);

@@ -1194,6 +1209,12 @@ _emit_code(jit_state_t *_jit)

case_wrr(stx, _i);

case_rrr(stx, _l);

case_wrr(stx, _l);

+ case jit_code_unstr:

+ unstr(rn(node->u.w), rn(node->v.w), node->w.w);

+ break;

+ case jit_code_unsti:

+ unsti(node->u.w, rn(node->v.w), node->w.w);

+ break;

case_rr(hton, _us);

case_rr(hton, _ui);

case_rr(hton, _ul);

@@ -1323,10 +1344,22 @@ _emit_code(jit_state_t *_jit)

case_rw(ld, _f);

case_rrr(ldx, _f);

case_rrw(ldx, _f);

+ case jit_code_unldr_x:

+ unldr_x(rn(node->u.w), rn(node->v.w), node->w.w);

+ break;

+ case jit_code_unldi_x:

+ unldi_x(rn(node->u.w), node->v.w, node->w.w);

+ break;

case_rr(st, _f);

case_wr(st, _f);

case_rrr(stx, _f);

case_wrr(stx, _f);

+ case jit_code_unstr_x:

+ unstr_x(rn(node->u.w), rn(node->v.w), node->w.w);

+ break;

+ case jit_code_unsti_x:

+ unsti_x(node->u.w, rn(node->v.w), node->w.w);

+ break;

case_rr(mov, _f);

case jit_code_movi_f:

assert_data(node);