27 #if defined(__aarch64__)
31 void a64_transpose_interleave_16(uint32_t *out,
const uint32_t *in,
size_t width,
size_t in_stride,
size_t height)
33 size_t out_stride = 4 * height *
sizeof(uint32_t);
36 "cmp %x[height], #0x4\n"
40 "mov x24, %x[width]\n"
42 "sub %x[height], %x[height], #0x4\n"
43 "add x22, x25, %x[in_stride]\n"
44 "add x21, x22, %x[in_stride]\n"
45 "add x20, x21, %x[in_stride]\n"
47 "add %x[in], x20, %x[in_stride]\n"
50 "ldr q19, [x25], #0x10\n"
51 "ldr q18, [x22], #0x10\n"
52 "sub x24, x24, #0x4\n"
53 "ldr q17, [x21], #0x10\n"
54 "ldr q16, [x20], #0x10\n"
56 "str q19, [x23, #0x0]\n"
57 "str q18, [x23, #0x10]\n"
58 "str q17, [x23, #0x20]\n"
59 "str q16, [x23, #0x30]\n"
60 "add x23, x23, %x[out_stride]\n"
65 "str q16, [x23, #0x0]\n"
66 "str q16, [x23, #0x10]\n"
67 "str q16, [x23, #0x20]\n"
68 "str q16, [x23, #0x30]\n"
70 "ldr s19, [x25], #0x4\n"
71 "ldr s18, [x22], #0x4\n"
72 "sub x24, x24, #0x1\n"
73 "ldr s17, [x21], #0x4\n"
74 "ldr s16, [x20], #0x4\n"
76 "str s19, [x23, #0x0]\n"
77 "str s18, [x23, #0x10]\n"
78 "str s17, [x23, #0x20]\n"
79 "str s16, [x23, #0x30]\n"
80 "add x23, x23, #0x4\n"
83 "cmp %x[height], #0x4\n"
84 "add %x[out], %x[out], #0x40\n"
86 "cbz %x[height], 12f\n"
89 "mov x20, %x[width]\n"
92 "sub %x[height], %x[height], #0x1\n"
94 "add %x[in], x25, %x[in_stride]\n"
97 "ldr q16, [x25], #0x10\n"
98 "sub x20, x20, #0x4\n"
100 "str q16, [x23, #0x0]\n"
101 "add x23, x23, %x[out_stride]\n"
105 "movi v16.4s, #0x0\n"
106 "str q16, [x23, #0x0]\n"
108 "ldr s16, [x25], #0x4\n"
109 "sub x20, x20, #0x1\n"
111 "str s16, [x23, #0x0]\n"
112 "add x23, x23, #0x4\n"
115 "cmp %x[height], #0x1\n"
116 "add %x[out], %x[out], #0x10\n"
119 : [height]
"+&r" (height), [in]
"+&r" (in), [out]
"+&r" (out)
120 : [in_stride]
"r" (in_stride), [out_stride]
"r" (out_stride), [width]
"r" (width)
121 :
"cc",
"memory",
"v16",
"v17",
"v18",
"v19",
"x20",
"x21",
"x22",
"x23",
"x24",
"x25"
128 void Transform<4, 1, true, VLType::None>(
129 float *out,
const float *in,
int stride,
int x0,
int xmax,
int k0,
int kmax)
131 a64_transpose_interleave_16(
132 reinterpret_cast<uint32_t *
>(out),
133 reinterpret_cast<const uint32_t *
>(in + k0 * stride + x0),
134 (xmax-x0) *
sizeof(
float) / 4,
135 stride *
sizeof(
float),
141 #endif // defined(__aarch64__)