27 #if defined(__aarch64__)
31 void a64_transpose_interleave_96(uint32_t *out,
const uint32_t *in,
size_t width,
size_t in_stride,
size_t height)
33 size_t out_stride = 24 * height *
sizeof(uint32_t);
36 "cmp %x[height], #0x4\n"
40 "mov x24, %x[width]\n"
41 "add x23, x25, %x[in_stride]\n"
42 "add x22, x23, %x[in_stride]\n"
43 "add x20, x22, %x[in_stride]\n"
45 "add %x[in], x20, %x[in_stride]\n"
47 "sub %x[height], %x[height], #0x4\n"
50 "ldr q7, [x25], #0x10\n"
51 "ldr q6, [x23], #0x10\n"
52 "sub x24, x24, #0x18\n"
54 "ldr q5, [x22], #0x10\n"
55 "ldr q4, [x20], #0x10\n"
56 "ldr q3, [x25], #0x10\n"
57 "ldr q2, [x23], #0x10\n"
58 "ldr q1, [x22], #0x10\n"
59 "ldr q0, [x20], #0x10\n"
60 "ldr q31, [x25], #0x10\n"
61 "ldr q30, [x23], #0x10\n"
62 "ldr q29, [x22], #0x10\n"
63 "ldr q28, [x20], #0x10\n"
64 "ldr q27, [x25], #0x10\n"
65 "ldr q26, [x23], #0x10\n"
66 "ldr q25, [x22], #0x10\n"
67 "ldr q24, [x20], #0x10\n"
68 "ldr q23, [x25], #0x10\n"
69 "ldr q22, [x23], #0x10\n"
70 "ldr q21, [x22], #0x10\n"
71 "ldr q20, [x20], #0x10\n"
72 "ldr q19, [x25], #0x10\n"
73 "ldr q18, [x23], #0x10\n"
74 "ldr q17, [x22], #0x10\n"
75 "ldr q16, [x20], #0x10\n"
76 "str q7, [x21, #0x0]\n"
77 "str q3, [x21, #0x10]\n"
78 "str q31, [x21, #0x20]\n"
79 "str q27, [x21, #0x30]\n"
80 "str q23, [x21, #0x40]\n"
81 "str q19, [x21, #0x50]\n"
82 "str q6, [x21, #0x60]\n"
83 "str q2, [x21, #0x70]\n"
84 "str q30, [x21, #0x80]\n"
85 "str q26, [x21, #0x90]\n"
86 "str q22, [x21, #0xa0]\n"
87 "str q18, [x21, #0xb0]\n"
88 "str q5, [x21, #0xc0]\n"
89 "str q1, [x21, #0xd0]\n"
90 "str q29, [x21, #0xe0]\n"
91 "str q25, [x21, #0xf0]\n"
92 "str q21, [x21, #0x100]\n"
93 "str q17, [x21, #0x110]\n"
94 "str q4, [x21, #0x120]\n"
95 "str q0, [x21, #0x130]\n"
96 "str q28, [x21, #0x140]\n"
97 "str q24, [x21, #0x150]\n"
98 "str q20, [x21, #0x160]\n"
99 "str q16, [x21, #0x170]\n"
100 "add x21, x21, %x[out_stride]\n"
106 "ldr q31, [x25], #0x10\n"
107 "ldr q30, [x23], #0x10\n"
108 "sub x24, x24, #0x10\n"
110 "ldr q29, [x22], #0x10\n"
111 "ldr q28, [x20], #0x10\n"
112 "ldr q27, [x25], #0x10\n"
113 "ldr q26, [x23], #0x10\n"
114 "ldr q25, [x22], #0x10\n"
115 "ldr q24, [x20], #0x10\n"
116 "ldr q23, [x25], #0x10\n"
117 "ldr q22, [x23], #0x10\n"
118 "ldr q21, [x22], #0x10\n"
119 "ldr q20, [x20], #0x10\n"
120 "ldr q19, [x25], #0x10\n"
121 "ldr q18, [x23], #0x10\n"
122 "ldr q17, [x22], #0x10\n"
123 "ldr q16, [x20], #0x10\n"
124 "str q31, [x21, #0x0]\n"
125 "str q27, [x21, #0x10]\n"
126 "str q23, [x21, #0x20]\n"
127 "str q19, [x21, #0x30]\n"
128 "str q30, [x21, #0x60]\n"
129 "str q26, [x21, #0x70]\n"
130 "str q22, [x21, #0x80]\n"
131 "str q18, [x21, #0x90]\n"
132 "str q29, [x21, #0xc0]\n"
133 "str q25, [x21, #0xd0]\n"
134 "str q21, [x21, #0xe0]\n"
135 "str q17, [x21, #0xf0]\n"
136 "str q28, [x21, #0x120]\n"
137 "str q24, [x21, #0x130]\n"
138 "str q20, [x21, #0x140]\n"
139 "str q16, [x21, #0x150]\n"
140 "add x21, x21, #0x40\n"
146 "ldr q19, [x25], #0x10\n"
147 "ldr q18, [x23], #0x10\n"
148 "sub x24, x24, #0x4\n"
150 "ldr q17, [x22], #0x10\n"
151 "ldr q16, [x20], #0x10\n"
152 "str q19, [x21, #0x0]\n"
153 "str q18, [x21, #0x60]\n"
154 "str q17, [x21, #0xc0]\n"
155 "str q16, [x21, #0x120]\n"
156 "add x21, x21, #0x10\n"
162 "ldr s19, [x25], #0x4\n"
163 "ldr s18, [x23], #0x4\n"
164 "sub x24, x24, #0x1\n"
166 "ldr s17, [x22], #0x4\n"
167 "ldr s16, [x20], #0x4\n"
168 "str s19, [x21, #0x0]\n"
169 "str s18, [x21, #0x60]\n"
170 "str s17, [x21, #0xc0]\n"
171 "str s16, [x21, #0x120]\n"
172 "add x21, x21, #0x4\n"
175 "cmp %x[height], #0x4\n"
176 "add %x[out], %x[out], #0x180\n"
178 "cbz %x[height], 20f\n"
181 "mov x20, %x[width]\n"
184 "add %x[in], x25, %x[in_stride]\n"
186 "sub %x[height], %x[height], #0x1\n"
189 "ldr q21, [x25], #0x10\n"
190 "ldr q20, [x25], #0x10\n"
191 "sub x20, x20, #0x18\n"
193 "ldr q19, [x25], #0x10\n"
194 "ldr q18, [x25], #0x10\n"
195 "ldr q17, [x25], #0x10\n"
196 "ldr q16, [x25], #0x10\n"
197 "str q21, [x21, #0x0]\n"
198 "str q20, [x21, #0x10]\n"
199 "str q19, [x21, #0x20]\n"
200 "str q18, [x21, #0x30]\n"
201 "str q17, [x21, #0x40]\n"
202 "str q16, [x21, #0x50]\n"
203 "add x21, x21, %x[out_stride]\n"
209 "ldr q19, [x25], #0x10\n"
210 "ldr q18, [x25], #0x10\n"
211 "sub x20, x20, #0x10\n"
213 "ldr q17, [x25], #0x10\n"
214 "ldr q16, [x25], #0x10\n"
215 "str q19, [x21, #0x0]\n"
216 "str q18, [x21, #0x10]\n"
217 "str q17, [x21, #0x20]\n"
218 "str q16, [x21, #0x30]\n"
219 "add x21, x21, #0x40\n"
225 "ldr q16, [x25], #0x10\n"
226 "sub x20, x20, #0x4\n"
228 "str q16, [x21, #0x0]\n"
229 "add x21, x21, #0x10\n"
235 "ldr s16, [x25], #0x4\n"
236 "sub x20, x20, #0x1\n"
238 "str s16, [x21, #0x0]\n"
239 "add x21, x21, #0x4\n"
242 "cmp %x[height], #0x1\n"
243 "add %x[out], %x[out], #0x60\n"
246 : [height]
"+&r" (height), [in]
"+&r" (in), [out]
"+&r" (out)
247 : [in_stride]
"r" (in_stride), [out_stride]
"r" (out_stride), [width]
"r" (width)
248 :
"cc",
"memory",
"v0",
"v1",
"v2",
"v3",
"v4",
"v5",
"v6",
"v7",
"v16",
"v17",
"v18",
"v19",
"v20",
"v21",
"v22",
"v23",
"v24",
"v25",
"v26",
"v27",
"v28",
"v29",
"v30",
"v31",
"x20",
"x21",
"x22",
"x23",
"x24",
"x25"
255 void Transform<24, 1, true, VLType::None>(
256 float *out,
const float *in,
int stride,
int x0,
int xmax,
int k0,
int kmax)
258 a64_transpose_interleave_96(
259 reinterpret_cast<uint32_t *
>(out),
260 reinterpret_cast<const uint32_t *
>(in + k0 * stride + x0),
261 (xmax-x0) *
sizeof(
float) / 4,
262 stride *
sizeof(
float),
268 #endif // defined(__aarch64__)