| 1 |
/* { dg-do compile } */ |
| 2 |
/* { dg-options "-std=c++17 -O2 -mfloat-abi=hard -mcpu=generic-armv7-a" } */ |
| 3 |
// { dg-require-effective-target arm_hard_ok } |
| 4 |
// { dg-require-effective-target arm_neon_ok } |
| 5 |
// { dg-add-options arm_neon } |
| 6 |
|
| 7 |
#include <arm_neon.h> |
| 8 |
|
| 9 |
typedef uint16x4_t e; |
| 10 |
typedef int16x4_t f; |
| 11 |
typedef int32x4_t g; |
| 12 |
typedef float32x4_t h; |
| 13 |
typedef uint32x4_t i; |
| 14 |
g j, p; |
| 15 |
g k(int l) { return vdupq_n_s32(l); } |
| 16 |
i n(f l) { return (i)vmovl_u16((e)l); } |
| 17 |
template <int, typename> struct q; |
| 18 |
template <int r, typename aa> q<r, aa> operator<(aa s, q<r, aa> t) { |
| 19 |
return q<r, aa>(s) < t; |
| 20 |
} |
| 21 |
template <typename ab, typename ac, int r> q<r, ab> ad(const q<r, ac> &); |
| 22 |
typedef q<4, int> ae; |
| 23 |
template <> class q<4, float> { |
| 24 |
public: |
| 25 |
q(h af) : ag(af) {} |
| 26 |
q(float) {} |
| 27 |
static q ah(void *ai) { |
| 28 |
float *l = (float *)ai; |
| 29 |
return vld1q_f32(l); |
| 30 |
} |
| 31 |
q operator+(q o) { |
| 32 |
h l = ag, m = o.ag; |
| 33 |
return vaddq_f32(l, m); |
| 34 |
} |
| 35 |
q operator*(q) { |
| 36 |
h l = ag, m; |
| 37 |
return vmulq_f32(l, m); |
| 38 |
} |
| 39 |
h ag; |
| 40 |
}; |
| 41 |
template <> class q<4, unsigned short> { |
| 42 |
public: |
| 43 |
q(f af) : ag(af) {} |
| 44 |
static q ah(void *ai) { |
| 45 |
unsigned short *l = (unsigned short *)ai; |
| 46 |
return (f)vld1_s16((int16_t *)l); |
| 47 |
} |
| 48 |
void aj() { |
| 49 |
f m = ag; |
| 50 |
vst1_u16(0, (e)m); |
| 51 |
} |
| 52 |
f ag; |
| 53 |
}; |
| 54 |
template <> class q<4, int> { |
| 55 |
public: |
| 56 |
q(g af) : ag(af) {} |
| 57 |
q(int u) { ag = k(u); } |
| 58 |
static q ah(void *ai) { |
| 59 |
int32_t *l = (int32_t *)ai; |
| 60 |
return vld1q_s32(l); |
| 61 |
} |
| 62 |
q operator&(q o) { |
| 63 |
g v = ag & o.ag; |
| 64 |
return v; |
| 65 |
} |
| 66 |
q operator|(q o) { |
| 67 |
g w = ag | o.ag; |
| 68 |
return w; |
| 69 |
} |
| 70 |
q operator^(q) { |
| 71 |
g x = ag ^ p; |
| 72 |
return x; |
| 73 |
} |
| 74 |
q operator>>(int ak) { return ag >> q(ak).ag; } |
| 75 |
q operator<(q) { |
| 76 |
g y, z = j < ag; |
| 77 |
y = (g)z; |
| 78 |
return y; |
| 79 |
} |
| 80 |
g ag; |
| 81 |
}; |
| 82 |
template <> ae ad(const q<4, unsigned short> &al) { return g(n(al.ag)); } |
| 83 |
template <> q<4, unsigned short> ad(const ae &al) { |
| 84 |
i l(i(al.ag)); |
| 85 |
return (f)vmovn_s32((g)l); |
| 86 |
} |
| 87 |
q<4, float> am(long long an) { |
| 88 |
q ao = q<4, unsigned short>::ah(&an); |
| 89 |
ae ak = ad<int>(ao), ap = ak & 8000, aq = ak ^ ap, ar = 55 < aq, as(aq); |
| 90 |
q at = as & ar; |
| 91 |
ae au = ap | at; |
| 92 |
return q<4, float>::ah(&au); |
| 93 |
} |
| 94 |
q<4, unsigned short> av(q<4, float> aw) { |
| 95 |
ae ak = ae::ah(&aw), ap = ak & 80000000, aq = ap, ax = 5, as = aq >> 3, |
| 96 |
ay = 6; |
| 97 |
q az = ax & as; |
| 98 |
ae au = ay | az; |
| 99 |
return ad<unsigned short>(au); |
| 100 |
} |
| 101 |
struct ba { |
| 102 |
typedef int bb; |
| 103 |
static q<4, float> bc(int s) { return am(s); } |
| 104 |
}; |
| 105 |
q<4, float> bd(q<4, float> s) { return s * 0; } |
| 106 |
template <typename be> void bf(void *bg, void *al, int bh, int bi) { |
| 107 |
int bj; |
| 108 |
auto bk(static_cast<typename be::bb *>(al) + bh), |
| 109 |
d = static_cast<typename be::bb *>(bg), |
| 110 |
bl = be::bc(static_cast<typename be::bb *>(al)[0]), bm = be::bc(0), |
| 111 |
c = bm; |
| 112 |
for (; bi;) { |
| 113 |
auto a = c, bn = be::bc(static_cast<typename be::bb *>(al)[1]), |
| 114 |
bo = be::bc(1); |
| 115 |
q bp = bn; |
| 116 |
q bq = bp; |
| 117 |
auto b = bq + bo; |
| 118 |
bl = be::bc(static_cast<typename be::bb *>(al)[2]); |
| 119 |
bm = be::bc(bk[2]); |
| 120 |
c = bl + bm; |
| 121 |
q br = a + b; |
| 122 |
auto bs = br; |
| 123 |
q bt = bd(bs); |
| 124 |
av(bt).aj(); |
| 125 |
d[0] = bj; |
| 126 |
} |
| 127 |
} |
| 128 |
int bu; |
| 129 |
void bv() { bf<ba>(0, 0, 0, bu); } |