27 #if defined(ARM_COMPUTE_ENABLE_SVE)
31 void sve_transpose_interleave_1VL(uint32_t *out,
const uint32_t *in,
size_t width,
size_t in_stride,
size_t height)
33 size_t out_stride = 1 * height * get_vector_length<uint8_t>();
36 "cmp %x[height], #0x4\n"
41 "mov x25, %x[width]\n"
42 "cntw x24, ALL, MUL #2\n"
43 "add x23, x26, %x[in_stride]\n"
44 "add x21, x23, %x[in_stride]\n"
45 "add x20, x21, %x[in_stride]\n"
47 "add %x[in], x20, %x[in_stride]\n"
49 "sub %x[height], %x[height], #0x4\n"
53 "ld1w { z23.s }, p1/Z, [x26]\n"
54 "ld1w { z22.s }, p1/Z, [x26, #1, MUL VL]\n"
56 "ld1w { z21.s }, p1/Z, [x23]\n"
57 "ld1w { z20.s }, p1/Z, [x23, #1, MUL VL]\n"
58 "addvl x26, x26, #2\n"
59 "addvl x23, x23, #2\n"
60 "ld1w { z19.s }, p1/Z, [x21]\n"
61 "ld1w { z18.s }, p1/Z, [x21, #1, MUL VL]\n"
62 "addvl x21, x21, #2\n"
63 "ld1w { z17.s }, p1/Z, [x20]\n"
64 "ld1w { z16.s }, p1/Z, [x20, #1, MUL VL]\n"
65 "st1w { z23.s }, p1, [x22]\n"
66 "addvl x20, x20, #2\n"
67 "st1w { z21.s }, p1, [x22, #1, MUL VL]\n"
68 "st1w { z19.s }, p1, [x22, #2, MUL VL]\n"
69 "st1w { z17.s }, p1, [x22, #3, MUL VL]\n"
70 "add x22, x22, %x[out_stride]\n"
71 "st1w { z22.s }, p1, [x22]\n"
72 "st1w { z20.s }, p1, [x22, #1, MUL VL]\n"
73 "st1w { z18.s }, p1, [x22, #2, MUL VL]\n"
74 "st1w { z16.s }, p1, [x22, #3, MUL VL]\n"
75 "add x22, x22, %x[out_stride]\n"
80 "whilelt p0.s, XZR, x25\n"
82 "ld1w { z19.s }, p0/Z, [x26]\n"
83 "ld1w { z18.s }, p0/Z, [x23]\n"
85 "addvl x26, x26, #1\n"
86 "ld1w { z17.s }, p0/Z, [x21]\n"
87 "ld1w { z16.s }, p0/Z, [x20]\n"
88 "addvl x23, x23, #1\n"
89 "addvl x21, x21, #1\n"
90 "st1w { z19.s }, p1, [x22]\n"
91 "addvl x20, x20, #1\n"
92 "st1w { z18.s }, p1, [x22, #1, MUL VL]\n"
93 "st1w { z17.s }, p1, [x22, #2, MUL VL]\n"
94 "st1w { z16.s }, p1, [x22, #3, MUL VL]\n"
95 "add x22, x22, %x[out_stride]\n"
98 "cmp %x[height], #0x4\n"
99 "addvl %x[out], %x[out], #4\n"
101 "cbz %x[height], 12f\n"
104 "mov x21, %x[width]\n"
105 "cntw x20, ALL, MUL #2\n"
108 "add %x[in], x26, %x[in_stride]\n"
110 "sub %x[height], %x[height], #0x1\n"
113 "sub x21, x21, x20\n"
114 "ld1w { z17.s }, p1/Z, [x26]\n"
115 "ld1w { z16.s }, p1/Z, [x26, #1, MUL VL]\n"
116 "st1w { z17.s }, p1, [x22]\n"
117 "add x22, x22, %x[out_stride]\n"
119 "st1w { z16.s }, p1, [x22]\n"
120 "addvl x26, x26, #2\n"
121 "add x22, x22, %x[out_stride]\n"
126 "whilelt p0.s, XZR, x21\n"
128 "ld1w { z16.s }, p0/Z, [x26]\n"
129 "st1w { z16.s }, p1, [x22]\n"
131 "addvl x26, x26, #1\n"
132 "add x22, x22, %x[out_stride]\n"
135 "cmp %x[height], #0x1\n"
136 "addvl %x[out], %x[out], #1\n"
139 : [height]
"+&r" (height), [in]
"+&r" (in), [out]
"+&r" (out)
140 : [in_stride]
"r" (in_stride), [out_stride]
"r" (out_stride), [width]
"r" (width)
141 :
"cc",
"memory",
"p0",
"p1",
"x20",
"x21",
"x22",
"x23",
"x24",
"x25",
"x26",
"z16",
"z17",
"z18",
"z19",
"z20",
"z21",
"z22",
"z23"
148 void Transform<1, 1, true, VLType::SVE>(
149 float *out,
const float *in,
int stride,
int x0,
int xmax,
int k0,
int kmax)
151 sve_transpose_interleave_1VL(
152 reinterpret_cast<uint32_t *
>(out),
153 reinterpret_cast<const uint32_t *
>(in + k0 * stride + x0),
154 (xmax-x0) *
sizeof(
float) / 4,
155 stride *
sizeof(
float),
161 #endif // defined(ARM_COMPUTE_ENABLE_SVE)