29 #if defined(ARM_COMPUTE_ENABLE_SVE)
34 void sve_u8_nhwc_max_2x2_s1_output2x2_depthfirst_impl(
35 const unsigned int n_channels,
36 const uint8_t *
const *
const inptrs,
37 uint8_t *
const *
const outptrs,
38 const bool exclude_padding,
39 const unsigned int pad_left,
40 const unsigned int pad_top,
41 const unsigned int pad_right,
42 const unsigned int pad_bottom
47 const uint64_t n_channels;
48 const uint8_t *
const *
const inptrs;
49 uint8_t *
const *
const outptrs;
51 unsigned int channels,
52 const uint8_t *
const *input_ptrs,
53 uint8_t *
const * output_ptrs,
54 bool,
unsigned int,
unsigned int,
unsigned int,
unsigned int
55 ) : n_channels(channels),
62 const KernelArgs
args(n_channels, inptrs, outptrs, exclude_padding,
63 pad_left, pad_top, pad_right, pad_bottom);
66 "ldr x15, [%x[args], %[offsetof_n_channels]]\n"
67 "ldr x21, [%x[args], %[offsetof_outptrs]]\n"
69 "whilelt p0.b, x14, x15\n"
70 "ldr x20, [%x[args], %[offsetof_inptrs]]\n"
71 "ldp x13, x12, [x21, #0x0]\n"
74 "ldp x10, x9, [x21, #0x10]\n"
75 "ldp x28, x27, [x20, #0x0]\n"
76 "ldp x26, x25, [x20, #0x10]\n"
77 "ldp x24, x23, [x20, #0x20]\n"
78 "ldp x22, x21, [x20, #0x30]\n"
79 "ldr x20, [x20, #0x40]\n"
80 "ld1b { z31.b }, p0/Z, [x27, x14]\n"
81 "ld1b { z30.b }, p0/Z, [x24, x14]\n"
82 "ld1b { z29.b }, p0/Z, [x21, x14]\n"
83 "ld1b { z28.b }, p0/Z, [x25, x14]\n"
84 "ld1b { z27.b }, p0/Z, [x28, x14]\n"
85 "ld1b { z26.b }, p0/Z, [x26, x14]\n"
86 "ld1b { z25.b }, p0/Z, [x23, x14]\n"
87 "ld1b { z24.b }, p0/Z, [x22, x14]\n"
88 "ld1b { z23.b }, p0/Z, [x20, x14]\n"
90 "whilelt p1.b, x14, x15\n"
93 "movprfx z22, z31\n umax z22.b, p2/M, z22.b, z30.b\n"
94 "movprfx z21, z30\n umax z21.b, p2/M, z21.b, z29.b\n"
95 "ld1b { z31.b }, p1/Z, [x27, x14]\n"
96 "ld1b { z30.b }, p1/Z, [x24, x14]\n"
97 "movprfx z20, z28\n umax z20.b, p2/M, z20.b, z27.b\n"
98 "movprfx z19, z26\n umax z19.b, p2/M, z19.b, z25.b\n"
99 "ld1b { z29.b }, p1/Z, [x21, x14]\n"
100 "ld1b { z27.b }, p1/Z, [x28, x14]\n"
101 "movprfx z17, z28\n umax z17.b, p2/M, z17.b, z24.b\n"
102 "movprfx z18, z25\n umax z18.b, p2/M, z18.b, z23.b\n"
103 "ld1b { z28.b }, p1/Z, [x25, x14]\n"
104 "ld1b { z26.b }, p1/Z, [x26, x14]\n"
105 "ld1b { z25.b }, p1/Z, [x23, x14]\n"
106 "ld1b { z24.b }, p1/Z, [x22, x14]\n"
107 "whilelt p0.b, x11, x15\n"
108 "movprfx z16, z22\n umax z16.b, p2/M, z16.b, z20.b\n"
109 "ld1b { z23.b }, p1/Z, [x20, x14]\n"
111 "whilelt p1.b, x14, x15\n"
112 "st1b { z16.b }, p0, [x13, x11]\n"
113 "movprfx z16, z19\n umax z16.b, p2/M, z16.b, z22.b\n"
114 "umax z17.b, p2/M, z17.b, z21.b\n"
115 "st1b { z16.b }, p0, [x12, x11]\n"
116 "movprfx z16, z21\n umax z16.b, p2/M, z16.b, z18.b\n"
117 "st1b { z17.b }, p0, [x10, x11]\n"
118 "st1b { z16.b }, p0, [x9, x11]\n"
122 "movprfx z22, z31\n umax z22.b, p2/M, z22.b, z30.b\n"
123 "movprfx z21, z30\n umax z21.b, p2/M, z21.b, z29.b\n"
124 "movprfx z20, z28\n umax z20.b, p2/M, z20.b, z27.b\n"
125 "movprfx z19, z26\n umax z19.b, p2/M, z19.b, z25.b\n"
126 "movprfx z17, z28\n umax z17.b, p2/M, z17.b, z24.b\n"
127 "movprfx z18, z25\n umax z18.b, p2/M, z18.b, z23.b\n"
128 "whilelt p0.b, x11, x15\n"
129 "movprfx z16, z22\n umax z16.b, p2/M, z16.b, z20.b\n"
130 "st1b { z16.b }, p0, [x13, x11]\n"
131 "movprfx z16, z19\n umax z16.b, p2/M, z16.b, z22.b\n"
132 "umax z17.b, p2/M, z17.b, z21.b\n"
133 "st1b { z16.b }, p0, [x12, x11]\n"
134 "movprfx z16, z21\n umax z16.b, p2/M, z16.b, z18.b\n"
135 "st1b { z17.b }, p0, [x10, x11]\n"
136 "st1b { z16.b }, p0, [x9, x11]\n"
138 : [
args]
"r" (&
args), [offsetof_inptrs]
"I" (offsetof(KernelArgs, inptrs)), [offsetof_n_channels]
"I" (offsetof(KernelArgs, n_channels)), [offsetof_outptrs]
"I" (offsetof(KernelArgs, outptrs))
139 :
"cc",
"memory",
"p0",
"p1",
"p2",
"x9",
"x10",
"x11",
"x12",
"x13",
"x14",
"x15",
"x20",
"x21",
"x22",
"x23",
"x24",
"x25",
"x26",
"x27",
"x28",
"z16",
"z17",
"z18",
"z19",
"z20",
"z21",
"z22",
"z23",
"z24",
"z25",
"z26",
"z27",
"z28",
"z29",
"z30",
"z31"
146 #endif // defined(ARM_COMPUTE_ENABLE_SVE)