File size: 11,513 Bytes
8b7c501 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 |
// Copyright 2019 Google LLC
//
// This source code is licensed under the BSD-style license found in the
// LICENSE file in the root directory of this source tree.
#include <xnnpack/assembly.h>
# void xnn_f16_igemm_minmax_ukernel_6x16__asm_aarch64_neonfp16arith_ld32(
# size_t mr, x0
# size_t nc, x1
# size_t kc, x2 / x0
# size_t ks, x3 / x9
# const void** restrict a, x4
# const void* restrict w, x5
# uint8_t* restrict c, x6
# size_t cm_stride, x7
# size_t cn_stride, [sp] -> x8
# size_t a_offset, [sp + 8] -> x11
# const void* zero, [sp + 16] -> x12
# const xnn_f16_minmax_params params [sp + 24] -> (x8)
# d8-d15, x19-x30 need to be preserved if used. x18 is reserved by the OS.
// Register usage
// A0 x14 v0
// A1 x15 v1
// A2 x20 v2
// A3 x21 v3
// A4 x22 v4
// A5 x23 v5
// B x5 v16 v17 v18 v19
// C0 x6 v20 v21
// C1 x16 v22 v23
// C2 x17 v24 v25
// C3 x10 v26 v27
// C4 x13 v28 v29
// C5 x7 v30 v31
// clamp v6, (v4), (v5)
// unused v7
// unused A v8 v9 v10 v11
// unused B v12 v13 v14 v15
BEGIN_FUNCTION xnn_f16_igemm_minmax_ukernel_6x16__asm_aarch64_neonfp16arith_ld32
# Load zero, params pointer
LDP x12, x8, [sp, 16]
# Clamp C pointers
CMP x0, 2 // if mr < 2
ADD x16, x6, x7 // c1 = c0 + cm_stride
CSEL x16, x6, x16, LO // c1 = c0
ADD x17, x16, x7 // c2 = c1 + cm_stride
// if mr <= 2
CSEL x17, x16, x17, LS // c2 = c1
# Load params
LDR s6, [x8]
CMP x0, 4 // if mr < 4
ADD x10, x17, x7 // c3 = c2 + cm_stride
CSEL x10, x17, x10, LO // c3 = c2
ADD x13, x10, x7 // c4 = c3 + cm_stride
// if mr <= 4
CSEL x13, x10, x13, LS // c4 = c3
CMP x0, 6 // if mr < 6
ADD x7, x13, x7 // c5 = c4 + cm_stride
CSEL x7, x13, x7, LO // c5 = c4
LDP x8, x11, [sp] // load cn_stride, a_offset
# Save x20-x23 on stack
STP x20, x21, [sp, -32]!
STP x22, x23, [sp, 16]
0:
# Load initial bias from w into accumulators
LDP q20, q21, [x5], 32
MOV x9, x3 // p = ks
MOV v22.16b, v20.16b
PRFM PLDL1KEEP, [x5, 0] // Prefetch B
MOV v23.16b, v21.16b
PRFM PLDL1KEEP, [x5, 64]
MOV v24.16b, v20.16b
PRFM PLDL1KEEP, [x5, 128]
MOV v25.16b, v21.16b
PRFM PLDL1KEEP, [x5, 192]
MOV v26.16b, v20.16b
PRFM PLDL1KEEP, [x5, 256]
MOV v27.16b, v21.16b
PRFM PLDL1KEEP, [x5, 320]
MOV v28.16b, v20.16b
MOV v29.16b, v21.16b
MOV v30.16b, v20.16b
MOV v31.16b, v21.16b
1:
# Load next 6 A pointers
LDP x14, x15, [x4], 16
LDP x20, x21, [x4], 16
LDP x22, x23, [x4], 16
CMP x14, x12 // if a0 == zero
ADD x14, x14, x11 // a0 += a_offset
CSEL x14, x12, x14, EQ // a0 = zero, else += a0 + a_offset
CMP x15, x12 // if a1 == zero
ADD x15, x15, x11 // a1 += a_offset
CSEL x15, x12, x15, EQ // a1 = zero, else += a1 + a_offset
CMP x20, x12 // if a2 == zero
ADD x20, x20, x11 // a2 += a_offset
CSEL x20, x12, x20, EQ // a2 = zero, else += a2 + a_offset
CMP x21, x12 // if a3 == zero
ADD x21, x21, x11 // a3 += a_offset
CSEL x21, x12, x21, EQ // a3 = zero, else += a3 + a_offset
CMP x22, x12 // if a4 == zero
ADD x22, x22, x11 // a4 += a_offset
CSEL x22, x12, x22, EQ // a4 = zero, else += a4 + a_offset
CMP x23, x12 // if a5 == zero
ADD x23, x23, x11 // a5 += a_offset
CSEL x23, x12, x23, EQ // a5 = zero, else += a5 + a_offset
# Is there at least 2 halffloats (4 bytes)?
SUBS x0, x2, 4 // k = kc - 4
B.LO 4f
.p2align 3
# Main loop - 2 halffloats of A (4 bytes)
# 24 FMA + 6 ld32 A + 4 LDR B
2:
LDR s0, [x14], 4 // A0
LDR q16, [x5], 16 // B
LDR q17, [x5], 16 // B
LDR s1, [x15], 4 // A1
LDR s2, [x20], 4 // A2
LDR s3, [x21], 4 // A3
LDR s4, [x22], 4 // A4
LDR s5, [x23], 4 // A5
LDR q18, [x5], 16 // B
LDR q19, [x5], 16 // B
SUBS x0, x0, 4
FMLA v20.8h, v16.8h, v0.h[0]
FMLA v21.8h, v17.8h, v0.h[0]
FMLA v22.8h, v16.8h, v1.h[0]
FMLA v23.8h, v17.8h, v1.h[0]
FMLA v24.8h, v16.8h, v2.h[0]
FMLA v25.8h, v17.8h, v2.h[0]
FMLA v26.8h, v16.8h, v3.h[0]
FMLA v27.8h, v17.8h, v3.h[0]
FMLA v28.8h, v16.8h, v4.h[0]
FMLA v29.8h, v17.8h, v4.h[0]
FMLA v30.8h, v16.8h, v5.h[0]
FMLA v31.8h, v17.8h, v5.h[0]
FMLA v20.8h, v18.8h, v0.h[1]
FMLA v21.8h, v19.8h, v0.h[1]
FMLA v22.8h, v18.8h, v1.h[1]
FMLA v23.8h, v19.8h, v1.h[1]
FMLA v24.8h, v18.8h, v2.h[1]
FMLA v25.8h, v19.8h, v2.h[1]
FMLA v26.8h, v18.8h, v3.h[1]
FMLA v27.8h, v19.8h, v3.h[1]
FMLA v28.8h, v18.8h, v4.h[1]
FMLA v29.8h, v19.8h, v4.h[1]
FMLA v30.8h, v18.8h, v5.h[1]
FMLA v31.8h, v19.8h, v5.h[1]
B.HS 2b
# Is there a remainder?- 1 halffloat of A (2 bytes)
TBNZ x0, 1, 4f
3:
# ks loop
SUBS x9, x9, 48 // ks -= MR * sizeof(void*)
B.HI 1b
# Clamp
DUP v4.8h, v6.h[0]
DUP v5.8h, v6.h[1]
FMAX v20.8h, v20.8h, v4.8h
FMAX v21.8h, v21.8h, v4.8h
FMAX v22.8h, v22.8h, v4.8h
FMAX v23.8h, v23.8h, v4.8h
FMAX v24.8h, v24.8h, v4.8h
FMAX v25.8h, v25.8h, v4.8h
FMAX v26.8h, v26.8h, v4.8h
FMAX v27.8h, v27.8h, v4.8h
FMAX v28.8h, v28.8h, v4.8h
FMAX v29.8h, v29.8h, v4.8h
FMAX v30.8h, v30.8h, v4.8h
FMAX v31.8h, v31.8h, v4.8h
SUBS x1, x1, 16
FMIN v20.8h, v20.8h, v5.8h
FMIN v21.8h, v21.8h, v5.8h
FMIN v22.8h, v22.8h, v5.8h
FMIN v23.8h, v23.8h, v5.8h
FMIN v24.8h, v24.8h, v5.8h
FMIN v25.8h, v25.8h, v5.8h
FMIN v26.8h, v26.8h, v5.8h
FMIN v27.8h, v27.8h, v5.8h
FMIN v28.8h, v28.8h, v5.8h
FMIN v29.8h, v29.8h, v5.8h
FMIN v30.8h, v30.8h, v5.8h
FMIN v31.8h, v31.8h, v5.8h
# Store full 6 x 16
B.LO 5f
ST1 {v30.16b, v31.16b}, [x7], x8
ST1 {v28.16b, v29.16b}, [x13], x8
ST1 {v26.16b, v27.16b}, [x10], x8
ST1 {v24.16b, v25.16b}, [x17], x8
ST1 {v22.16b, v23.16b}, [x16], x8
ST1 {v20.16b, v21.16b}, [x6], x8
SUB x4, x4, x3 // a -= ks
# nc loop
B.HI 0b
# Restore x20-x23 from stack
LDP x22, x23, [sp, 16]
LDP x20, x21, [sp], 32
RET
4:
# Remainder- 1 halffloat of A (2 bytes)
LDR h0, [x14], 2 // A0
LDR q16, [x5], 16 // B
LDR q17, [x5], 16 // B
LDR h1, [x15], 2 // A1
LDR h2, [x20], 2 // A2
LDR h3, [x21], 2 // A3
LDR h4, [x22], 2 // A4
LDR h5, [x23], 2 // A5
FMLA v20.8h, v16.8h, v0.h[0]
FMLA v21.8h, v17.8h, v0.h[0]
FMLA v22.8h, v16.8h, v1.h[0]
FMLA v23.8h, v17.8h, v1.h[0]
FMLA v24.8h, v16.8h, v2.h[0]
FMLA v25.8h, v17.8h, v2.h[0]
FMLA v26.8h, v16.8h, v3.h[0]
FMLA v27.8h, v17.8h, v3.h[0]
FMLA v28.8h, v16.8h, v4.h[0]
FMLA v29.8h, v17.8h, v4.h[0]
FMLA v30.8h, v16.8h, v5.h[0]
FMLA v31.8h, v17.8h, v5.h[0]
B 3b
# Store odd width
5:
TBZ x1, 3, 6f
STR q30, [x7], 16
MOV v30.16b, v31.16b
STR q28, [x13], 16
MOV v28.16b, v29.16b
STR q26, [x10], 16
MOV v26.16b, v27.16b
STR q24, [x17], 16
MOV v24.16b, v25.16b
STR q22, [x16], 16
MOV v22.16b, v23.16b
STR q20, [x6], 16
MOV v20.16b, v21.16b
6:
TBZ x1, 2, 7f
STR d30, [x7], 8
STR d28, [x13], 8
DUP d30, v30.d[1]
DUP d28, v28.d[1]
STR d26, [x10], 8
STR d24, [x17], 8
DUP d26, v26.d[1]
DUP d24, v24.d[1]
STR d22, [x16], 8
STR d20, [x6], 8
DUP d22, v22.d[1]
DUP d20, v20.d[1]
7:
TBZ x1, 1, 8f
STR s30, [x7], 4
STR s28, [x13], 4
DUP s30, v30.s[1]
DUP s28, v28.s[1]
STR s26, [x10], 4
STR s24, [x17], 4
DUP s26, v26.s[1]
DUP s24, v24.s[1]
STR s22, [x16], 4
STR s20, [x6], 4
DUP s22, v22.s[1]
DUP s20, v20.s[1]
8:
TBZ x1, 0, 9f
STR h30, [x7]
STR h28, [x13]
STR h26, [x10]
STR h24, [x17]
STR h22, [x16]
STR h20, [x6]
9:
# Restore x20-x23 from stack
LDP x22, x23, [sp, 16]
LDP x20, x21, [sp], 32
RET
END_FUNCTION xnn_f16_igemm_minmax_ukernel_6x16__asm_aarch64_neonfp16arith_ld32
#ifdef __ELF__
.section ".note.GNU-stack","",%progbits
#endif
|