1 /*
2 * Copyright (c) 2022 Arm Limited.
3 *
4 * SPDX-License-Identifier: MIT
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to
8 * deal in the Software without restriction, including without limitation the
9 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
10 * sell copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
22 * IN THE SOFTWARE.
23 */
24
25 #if defined(__ARM_FEATURE_SVE)
26
27 template <>
interleave_block(bfloat16 * & out,const float * const * in,size_t width,size_t height,size_t row_offset,bool first)28 void interleave_block<4, 2, VLType::SME, false>(
29 bfloat16 * &out, const float * const *in,
30 size_t width, size_t height, size_t row_offset, bool first
31 )
32 {
33 ARM_COMPUTE_UNUSED(first);
34
35 __asm__ __volatile__(
36 ".inst 0xd503477f // SMSTART ZA\n"
37 "cntw x22, ALL, MUL #2\n"
38 "cntw x9\n"
39 "cntw x21, ALL, MUL #2\n"
40 "cntw x19, ALL, MUL #3\n"
41 "sub x20, x22, #0x1\n"
42 ".inst 0x25207817 // ptrue pn15.b\n"
43 "whilelt p1.s, XZR, %x[height]\n"
44 "whilelt p14.s, x9, %x[height]\n"
45 "whilelt p13.s, x21, %x[height]\n"
46 "whilelt p12.s, x19, %x[height]\n"
47 "sub x28, %x[width], #0x1\n"
48 "cntw x19, ALL, MUL #2\n"
49 "ands x27, %x[width], x20\n"
50 "mov x26, %x[in]\n"
51 "add x28, x28, x19\n"
52 "csel x27, x27, x22, NE\n"
53 "add x25, x26, x9, LSL #3\n"
54 "mov x24, #0x0\n"
55 "udiv x28, x28, x19\n"
56 "add x27, x27, #0x1\n"
57 "mov x19, %x[width]\n"
58 "add x23, x25, x9, LSL #3\n"
59 "ptrue p0.b\n"
60 "mov x22, %x[outptr_raw]\n"
61 "mov x21, %x[row_offset]\n"
62 "sub x28, x28, #0x1\n"
63 "lsr x27, x27, #0x1\n"
64 "mov x12, #0x0\n"
65 ".inst 0x25b34713 // whilelt pn11.s, x24, x19, VLx2\n"
66 "add x20, x23, x9, LSL #3\n"
67 "1:" // Width loop: Preamble: Loop
68 "ldr x19, [x26], #0x8\n"
69 ".inst 0x25306c28 // psel p8.s, p11.s/Z, p1.s[w12]\n"
70 ".inst 0x25306dca // psel p10.s, p11.s/Z, p14.s[w12]\n"
71 ".inst 0xa0154278 // ld1w { z24.s-z25.s }, pn8.s/Z, [x19, x21, LSL #2]\n"
72 "ldr x19, [x25], #0x8\n"
73 ".inst 0x25306da9 // psel p9.s, p11.s/Z, p13.s[w12]\n"
74 ".inst 0x25306d88 // psel p8.s, p11.s/Z, p12.s[w12]\n"
75 ".inst 0xa0154a62 // ld1w { z2.s-z3.s }, pn10.s/Z, [x19, x21, LSL #2]\n"
76 "ldr x19, [x23], #0x8\n"
77 ".inst 0xa015466a // ld1w { z10.s-z11.s }, pn9.s/Z, [x19, x21, LSL #2]\n"
78 ".inst 0xc160e318 // bfcvt z24.h, { z24.s-z25.s }\n"
79 ".inst 0xc160e042 // bfcvt z2.h, { z2.s-z3.s }\n"
80 "ldr x19, [x20], #0x8\n"
81 ".inst 0xa015426c // ld1w { z12.s-z13.s }, pn8.s/Z, [x19, x21, LSL #2]\n"
82 ".inst 0xc160e14a // bfcvt z10.h, { z10.s-z11.s }\n"
83 ".inst 0xc160e18c // bfcvt z12.h, { z12.s-z13.s }\n"
84 ".inst 0xc0800300 // mova za0h.s[x12], p0/M, z24.s\n"
85 ".inst 0xc0800044 // mova za1h.s[x12], p0/M, z2.s\n"
86 ".inst 0xc0800148 // mova za2h.s[x12], p0/M, z10.s\n"
87 ".inst 0xc080018c // mova za3h.s[x12], p0/M, z12.s\n"
88 "add x12, x12, #0x1\n"
89 "cmp x12, x9\n"
90 "blt 1b\n"
91 "incw x21, ALL, MUL #2\n"
92 "incw x24, ALL, MUL #2\n"
93 "cbz x28, 5f\n"
94 "2:" // Width loop
95 "mov x12, #0x0\n"
96 "3:" // Width loop: Store: Loop
97 ".inst 0xc0828011 // mova z17.s, p0/M, za0v.s[x12]\n"
98 ".inst 0xc0828095 // mova z21.s, p0/M, za1v.s[x12]\n"
99 ".inst 0xc0828119 // mova z25.s, p0/M, za2v.s[x12]\n"
100 ".inst 0xc082819d // mova z29.s, p0/M, za3v.s[x12]\n"
101 "add x12, x12, #0x1\n"
102 "cmp x12, x9\n"
103 ".inst 0xa160ded1 // st1w { z17.s, z21.s, z25.s, z29.s }, pn15.b, [x22]\n"
104 "addvl x22, x22, #4\n"
105 "blt 3b\n"
106 "mov x26, %x[in]\n"
107 "add x25, x26, x9, LSL #3\n"
108 "mov x19, %x[width]\n"
109 "add x23, x25, x9, LSL #3\n"
110 "mov x12, #0x0\n"
111 ".inst 0x25b34713 // whilelt pn11.s, x24, x19, VLx2\n"
112 "add x20, x23, x9, LSL #3\n"
113 "4:" // Width loop: Load: Loop
114 "ldr x19, [x26], #0x8\n"
115 ".inst 0x25306c28 // psel p8.s, p11.s/Z, p1.s[w12]\n"
116 ".inst 0x25306dca // psel p10.s, p11.s/Z, p14.s[w12]\n"
117 ".inst 0xa015426c // ld1w { z12.s-z13.s }, pn8.s/Z, [x19, x21, LSL #2]\n"
118 "ldr x19, [x25], #0x8\n"
119 ".inst 0x25306da9 // psel p9.s, p11.s/Z, p13.s[w12]\n"
120 ".inst 0x25306d88 // psel p8.s, p11.s/Z, p12.s[w12]\n"
121 ".inst 0xa0154a6e // ld1w { z14.s-z15.s }, pn10.s/Z, [x19, x21, LSL #2]\n"
122 "ldr x19, [x23], #0x8\n"
123 ".inst 0xa0154672 // ld1w { z18.s-z19.s }, pn9.s/Z, [x19, x21, LSL #2]\n"
124 ".inst 0xc160e18c // bfcvt z12.h, { z12.s-z13.s }\n"
125 ".inst 0xc160e1ce // bfcvt z14.h, { z14.s-z15.s }\n"
126 "ldr x19, [x20], #0x8\n"
127 ".inst 0xa015427e // ld1w { z30.s-z31.s }, pn8.s/Z, [x19, x21, LSL #2]\n"
128 ".inst 0xc160e252 // bfcvt z18.h, { z18.s-z19.s }\n"
129 ".inst 0xc160e3de // bfcvt z30.h, { z30.s-z31.s }\n"
130 ".inst 0xc0800180 // mova za0h.s[x12], p0/M, z12.s\n"
131 ".inst 0xc08001c4 // mova za1h.s[x12], p0/M, z14.s\n"
132 ".inst 0xc0800248 // mova za2h.s[x12], p0/M, z18.s\n"
133 ".inst 0xc08003cc // mova za3h.s[x12], p0/M, z30.s\n"
134 "add x12, x12, #0x1\n"
135 "cmp x12, x9\n"
136 "blt 4b\n"
137 "subs x28, x28, #0x1\n"
138 "incw x21, ALL, MUL #2\n"
139 "incw x24, ALL, MUL #2\n"
140 "bgt 2b\n"
141 "5:" // Width loop: Tails
142 "mov x12, #0x0\n"
143 "6:" // Width loop: Tails: Loop
144 ".inst 0xc0828011 // mova z17.s, p0/M, za0v.s[x12]\n"
145 ".inst 0xc0828095 // mova z21.s, p0/M, za1v.s[x12]\n"
146 ".inst 0xc0828119 // mova z25.s, p0/M, za2v.s[x12]\n"
147 ".inst 0xc082819d // mova z29.s, p0/M, za3v.s[x12]\n"
148 "add x12, x12, #0x1\n"
149 "cmp x12, x27\n"
150 ".inst 0xa160ded1 // st1w { z17.s, z21.s, z25.s, z29.s }, pn15.b, [x22]\n"
151 "addvl x22, x22, #4\n"
152 "blt 6b\n"
153 "7:" // End
154 "mov %x[outptr_raw], x22\n"
155 ".inst 0xd503467f // SMSTOP\n"
156 : [outptr_raw] "+&r" (out)
157 : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
158 : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x12", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
159 );
160 }
161
162 #endif // defined(__ARM_FEATURE_SVE)
163