op_neon.h.svn-base

我们自己开发的一个OSEK操作系统！不知道可不可以？

    uint32_t low;
    uint32_t high;

    DO_MULL(low, T0, T1, uint8_t, uint16_t);
    DO_MULL(tmp, T0 >> 8, T1 >> 8, uint8_t, uint16_t);
    low |= tmp << 16;
    DO_MULL(high, T0 >> 16, T1 >> 16, uint8_t, uint16_t);
    DO_MULL(tmp, T0 >> 24, T1 >> 24, uint8_t, uint16_t);
    high |= tmp << 16;
    T0 = low;
    T1 = high;
    FORCE_RET();
}

NEON_OP(mull_s8)
{
    uint32_t tmp;
    uint32_t low;
    uint32_t high;

    DO_MULL(low, T0, T1, int8_t, uint16_t);
    DO_MULL(tmp, T0 >> 8, T1 >> 8, int8_t, uint16_t);
    low |= tmp << 16;
    DO_MULL(high, T0 >> 16, T1 >> 16, int8_t, uint16_t);
    DO_MULL(tmp, T0 >> 24, T1 >> 24, int8_t, uint16_t);
    high |= tmp << 16;
    T0 = low;
    T1 = high;
    FORCE_RET();
}

NEON_OP(mull_u16)
{
    uint32_t low;
    uint32_t high;

    DO_MULL(low, T0, T1, uint16_t, uint32_t);
    DO_MULL(high, T0 >> 16, T1 >> 16, uint16_t, uint32_t);
    T0 = low;
    T1 = high;
    FORCE_RET();
}

NEON_OP(mull_s16)
{
    uint32_t low;
    uint32_t high;

    DO_MULL(low, T0, T1, int16_t, uint32_t);
    DO_MULL(high, T0 >> 16, T1 >> 16, int16_t, uint32_t);
    T0 = low;
    T1 = high;
    FORCE_RET();
}

NEON_OP(addl_saturate_s32)
{
    uint32_t tmp;
    uint32_t res;

    tmp = env->vfp.scratch[0];
    res = T0 + tmp;
    if (((res ^ T0) & SIGNBIT) && !((T0 ^ tmp) & SIGNBIT)) {
        env->QF = 1;
        T0 = (T0 >> 31) ^ 0x7fffffff;
    } else {
      T0 = res;
    }
    tmp = env->vfp.scratch[1];
    res = T1 + tmp;
    if (((res ^ T1) & SIGNBIT) && !((T1 ^ tmp) & SIGNBIT)) {
        env->QF = 1;
        T1 = (T1 >> 31) ^ 0x7fffffff;
    } else {
      T1 = res;
    }
    FORCE_RET();
}

NEON_OP(addl_saturate_s64)
{
    uint64_t src1;
    uint64_t src2;
    uint64_t res;

    src1 = T0 + ((uint64_t)T1 << 32);
    src2 = env->vfp.scratch[0] + ((uint64_t)env->vfp.scratch[1] << 32);
    res = src1 + src2;
    if (((res ^ src1) & SIGNBIT64) && !((src1 ^ src2) & SIGNBIT64)) {
        env->QF = 1;
        T0 = ~(int64_t)src1 >> 63;
        T1 = T0 ^ 0x80000000;
    } else {
      T0 = res;
      T1 = res >> 32;
    }
    FORCE_RET();
}

NEON_OP(addl_saturate_u64)
{
    uint64_t src1;
    uint64_t src2;
    uint64_t res;

    src1 = T0 + ((uint64_t)T1 << 32);
    src2 = env->vfp.scratch[0] + ((uint64_t)env->vfp.scratch[1] << 32);
    res = src1 + src2;
    if (res < src1) {
        env->QF = 1;
        T0 = 0xffffffff;
        T1 = 0xffffffff;
    } else {
      T0 = res;
      T1 = res >> 32;
    }
    FORCE_RET();
}

NEON_OP(subl_saturate_s64)
{
    uint64_t src1;
    uint64_t src2;
    uint64_t res;

    src1 = T0 + ((uint64_t)T1 << 32);
    src2 = env->vfp.scratch[0] + ((uint64_t)env->vfp.scratch[1] << 32);
    res = src1 - src2;
    if (((res ^ src1) & SIGNBIT64) && ((src1 ^ src2) & SIGNBIT64)) {
        env->QF = 1;
        T0 = ~(int64_t)src1 >> 63;
        T1 = T0 ^ 0x80000000;
    } else {
      T0 = res;
      T1 = res >> 32;
    }
    FORCE_RET();
}

NEON_OP(subl_saturate_u64)
{
    uint64_t src1;
    uint64_t src2;
    uint64_t res;

    src1 = T0 + ((uint64_t)T1 << 32);
    src2 = env->vfp.scratch[0] + ((uint64_t)env->vfp.scratch[1] << 32);
    if (src1 < src2) {
        env->QF = 1;
        T0 = 0;
        T1 = 0;
    } else {
      res = src1 - src2;
      T0 = res;
      T1 = res >> 32;
    }
    FORCE_RET();
}

NEON_OP(negl_u16)
{
    uint32_t tmp;
    tmp = T0 >> 16;
    tmp = -tmp;
    T0 = (-T0 & 0xffff) | (tmp << 16);
    tmp = T1 >> 16;
    tmp = -tmp;
    T1 = (-T1 & 0xffff) | (tmp << 16);
    FORCE_RET();
}

NEON_OP(negl_u32)
{
    T0 = -T0;
    T1 = -T1;
    FORCE_RET();
}

NEON_OP(negl_u64)
{
    uint64_t val;

    val = T0 | ((uint64_t)T1 << 32);
    val = -val;
    T0 = val;
    T1 = val >> 32;
    FORCE_RET();
}

/* Scalar operations.  */
NEON_OP(dup_low16)
{
    T0 = (T0 & 0xffff) | (T0 << 16);
    FORCE_RET();
}

NEON_OP(dup_high16)
{
    T0 = (T0 >> 16) | (T0 & 0xffff0000);
    FORCE_RET();
}

/* Helper for VEXT */
NEON_OP(extract)
{
    int shift = PARAM1;
    T0 = (T0 >> shift) | (T1 << (32 - shift));
    FORCE_RET();
}

/* Pairwise add long.  Named type is source type.  */
NEON_OP(paddl_s8)
{
    int8_t src1;
    int8_t src2;
    uint16_t result;
    src1 = T0 >> 24;
    src2 = T0 >> 16;
    result = (uint16_t)src1 + src2;
    src1 = T0 >> 8;
    src2 = T0;
    T0 = (uint16_t)((uint16_t)src1 + src2) | ((uint32_t)result << 16);
    FORCE_RET();
}

NEON_OP(paddl_u8)
{
    uint8_t src1;
    uint8_t src2;
    uint16_t result;
    src1 = T0 >> 24;
    src2 = T0 >> 16;
    result = (uint16_t)src1 + src2;
    src1 = T0 >> 8;
    src2 = T0;
    T0 = (uint16_t)((uint16_t)src1 + src2) | ((uint32_t)result << 16);
    FORCE_RET();
}

NEON_OP(paddl_s16)
{
    T0 = (uint32_t)(int16_t)T0 + (uint32_t)(int16_t)(T0 >> 16);
    FORCE_RET();
}

NEON_OP(paddl_u16)
{
    T0 = (uint32_t)(uint16_t)T0 + (uint32_t)(uint16_t)(T0 >> 16);
    FORCE_RET();
}

NEON_OP(paddl_s32)
{
    int64_t tmp;
    tmp = (int64_t)(int32_t)T0 + (int64_t)(int32_t)T1;
    T0 = tmp;
    T1 = tmp >> 32;
    FORCE_RET();
}

NEON_OP(paddl_u32)
{
    uint64_t tmp;
    tmp = (uint64_t)T0 + (uint64_t)T1;
    T0 = tmp;
    T1 = tmp >> 32;
    FORCE_RET();
}

/* Count Leading Sign/Zero Bits.  */
static inline int do_clz8(uint8_t x)
{
    int n;
    for (n = 8; x; n--)
        x >>= 1;
    return n;
}

static inline int do_clz16(uint16_t x)
{
    int n;
    for (n = 16; x; n--)
        x >>= 1;
    return n;
}

NEON_OP(clz_u8)
{
    uint32_t result;
    uint32_t tmp;

    tmp = T0;
    result = do_clz8(tmp);
    result |= do_clz8(tmp >> 8) << 8;
    result |= do_clz8(tmp >> 16) << 16;
    result |= do_clz8(tmp >> 24) << 24;
    T0 = result;
    FORCE_RET();
}

NEON_OP(clz_u16)
{
    uint32_t result;
    uint32_t tmp;
    tmp = T0;
    result = do_clz16(tmp);
    result |= do_clz16(tmp >> 16) << 16;
    T0 = result;
    FORCE_RET();
}

NEON_OP(cls_s8)
{
    uint32_t result;
    int8_t tmp;
    tmp = T0;
    result = do_clz8((tmp < 0) ? ~tmp : tmp) - 1;
    tmp = T0 >> 8;
    result |= (do_clz8((tmp < 0) ? ~tmp : tmp) - 1) << 8;
    tmp = T0 >> 16;
    result |= (do_clz8((tmp < 0) ? ~tmp : tmp) - 1) << 16;
    tmp = T0 >> 24;
    result |= (do_clz8((tmp < 0) ? ~tmp : tmp) - 1) << 24;
    T0 = result;
    FORCE_RET();
}

NEON_OP(cls_s16)
{
    uint32_t result;
    int16_t tmp;
    tmp = T0;
    result = do_clz16((tmp < 0) ? ~tmp : tmp) - 1;
    tmp = T0 >> 16;
    result |= (do_clz16((tmp < 0) ? ~tmp : tmp) - 1) << 16;
    T0 = result;
    FORCE_RET();
}

NEON_OP(cls_s32)
{
    int count;
    if ((int32_t)T0 < 0)
        T0 = ~T0;
    for (count = 32; T0 > 0; count--)
        T0 = T0 >> 1;
    T0 = count - 1;
    FORCE_RET();
}

/* Bit count.  */
NEON_OP(cnt_u8)
{
    T0 = (T0 & 0x55555555) + ((T0 >>  1) & 0x55555555);
    T0 = (T0 & 0x33333333) + ((T0 >>  2) & 0x33333333);
    T0 = (T0 & 0x0f0f0f0f) + ((T0 >>  4) & 0x0f0f0f0f);
    FORCE_RET();
}

/* Saturnating negation.  */
/* ??? Make these use NEON_VOP1 */
#define DO_QABS8(x) do { \
    if (x == (int8_t)0x80) { \
        x = 0x7f; \
        env->QF = 1; \
    } else if (x < 0) { \
        x = -x; \
    }} while (0)
NEON_OP(qabs_s8)
{
    neon_s8 vec;
    NEON_UNPACK(neon_s8, vec, T0);
    DO_QABS8(vec.v1);
    DO_QABS8(vec.v2);
    DO_QABS8(vec.v3);
    DO_QABS8(vec.v4);
    NEON_PACK(neon_s8, T0, vec);
    FORCE_RET();
}
#undef DO_QABS8

#define DO_QNEG8(x) do { \
    if (x == (int8_t)0x80) { \
        x = 0x7f; \
        env->QF = 1; \
    } else { \
        x = -x; \
    }} while (0)
NEON_OP(qneg_s8)
{
    neon_s8 vec;
    NEON_UNPACK(neon_s8, vec, T0);
    DO_QNEG8(vec.v1);
    DO_QNEG8(vec.v2);
    DO_QNEG8(vec.v3);
    DO_QNEG8(vec.v4);
    NEON_PACK(neon_s8, T0, vec);
    FORCE_RET();
}
#undef DO_QNEG8

#define DO_QABS16(x) do { \
    if (x == (int16_t)0x8000) { \
        x = 0x7fff; \
        env->QF = 1; \
    } else if (x < 0) { \
        x = -x; \
    }} while (0)
NEON_OP(qabs_s16)
{
    neon_s16 vec;
    NEON_UNPACK(neon_s16, vec, T0);
    DO_QABS16(vec.v1);
    DO_QABS16(vec.v2);
    NEON_PACK(neon_s16, T0, vec);
    FORCE_RET();
}
#undef DO_QABS16

#define DO_QNEG16(x) do { \
    if (x == (int16_t)0x8000) { \
        x = 0x7fff; \
        env->QF = 1; \
    } else { \
        x = -x; \
    }} while (0)
NEON_OP(qneg_s16)
{
    neon_s16 vec;
    NEON_UNPACK(neon_s16, vec, T0);
    DO_QNEG16(vec.v1);
    DO_QNEG16(vec.v2);
    NEON_PACK(neon_s16, T0, vec);
    FORCE_RET();
}
#undef DO_QNEG16

NEON_OP(qabs_s32)
{
    if (T0 == 0x80000000) {
        T0 = 0x7fffffff;
        env->QF = 1;
    } else if ((int32_t)T0 < 0) {
        T0 = -T0;
    }
    FORCE_RET();
}

NEON_OP(qneg_s32)
{
    if (T0 == 0x80000000) {
        T0 = 0x7fffffff;
        env->QF = 1;
    } else {
        T0 = -T0;
    }
    FORCE_RET();
}

/* Unary opperations */
#define NEON_FN(dest, src, dummy) dest = (src < 0) ? -src : src
NEON_VOP1(abs_s8, neon_s8, 4)
NEON_VOP1(abs_s16, neon_s16, 2)
NEON_OP(abs_s32)
{
    if ((int32_t)T0 < 0)
        T0 = -T0;
    FORCE_RET();
}
#undef NEON_FN

/* Transpose.  Argument order is rather strange to avoid special casing
   the tranlation code.
   On input T0 = rm, T1 = rd.  On output T0 = rd, T1 = rm  */
NEON_OP(trn_u8)
{
    uint32_t rd;
    uint32_t rm;
    rd = ((T0 & 0x00ff00ff) << 8) | (T1 & 0x00ff00ff);
    rm = ((T1 & 0xff00ff00) >> 8) | (T0 & 0xff00ff00);
    T0 = rd;
    T1 = rm;
    FORCE_RET();
}

NEON_OP(trn_u16)
{
    uint32_t rd;
    uint32_t rm;
    rd = (T0 << 16) | (T1 & 0xffff);
    rm = (T1 >> 16) | (T0 & 0xffff0000);
    T0 = rd;
    T1 = rm;
    FORCE_RET();
}

/* Worker routines for zip and unzip.  */
NEON_OP(unzip_u8)
{
    uint32_t rd;
    uint32_t rm;
    rd = (T0 & 0xff) | ((T0 >> 8) & 0xff00)
         | ((T1 << 16) & 0xff0000) | ((T1 << 8) & 0xff000000);
    rm = ((T0 >> 8) & 0xff) | ((T0 >> 16) & 0xff00)
         | ((T1 << 8) & 0xff0000) | (T1 & 0xff000000);
    T0 = rd;
    T1 = rm;
    FORCE_RET();
}

NEON_OP(zip_u8)
{
    uint32_t rd;
    uint32_t rm;
    rd = (T0 & 0xff) | ((T1 << 8) & 0xff00)
         | ((T0 << 16) & 0xff0000) | ((T1 << 24) & 0xff000000);
    rm = ((T0 >> 16) & 0xff) | ((T1 >> 8) & 0xff00)
         | ((T0 >> 8) & 0xff0000) | (T1 & 0xff000000);
    T0 = rd;
    T1 = rm;
    FORCE_RET();
}

NEON_OP(zip_u16)
{
    uint32_t tmp;

    tmp = (T0 & 0xffff) | (T1 << 16);
    T1 = (T1 & 0xffff0000) | (T0 >> 16);
    T0 = tmp;
    FORCE_RET();
}

/* Reciprocal/root estimate.  */
NEON_OP(recpe_u32)
{
    T0 = helper_recpe_u32(T0);
}

NEON_OP(rsqrte_u32)
{
    T0 = helper_rsqrte_u32(T0);
}

NEON_OP(recpe_f32)
{
    FT0s = helper_recpe_f32(FT0s);
}

NEON_OP(rsqrte_f32)
{
    FT0s = helper_rsqrte_f32(FT0s);
}

/* Table lookup.  This accessed the register file directly.  */
NEON_OP(tbl)
{
    helper_neon_tbl(PARAM1, PARAM2);
}

NEON_OP(dup_u8)
{
    T0 = (T0 >> PARAM1) & 0xff;
    T0 |= T0 << 8;
    T0 |= T0 << 16;
    FORCE_RET();
}

/* Helpers for element load/store.  */
NEON_OP(insert_elt)
{
    int shift = PARAM1;
    uint32_t mask = PARAM2;
    T2 = (T2 & mask) | (T0 << shift);
    FORCE_RET();
}

NEON_OP(extract_elt)
{
    int shift = PARAM1;
    uint32_t mask = PARAM2;
    T0 = (T2 & mask) >> shift;
    FORCE_RET();
}