1 /* Optimized version of the standard bzero() function.
2 This file is part of the GNU C Library.
3 Copyright (C) 2000, 2001, 2002 Free Software Foundation, Inc.
4 Contributed by Dan Pop for Itanium <Dan.Pop@cern.ch>.
5 Rewritten for McKinley by Sverre Jarp, HP Labs/CERN <Sverre.Jarp@cern.ch>
7 The GNU C Library is free software; you can redistribute it and/or
8 modify it under the terms of the GNU Lesser General Public
9 License as published by the Free Software Foundation; either
10 version 2.1 of the License, or (at your option) any later version.
12 The GNU C Library is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 Lesser General Public License for more details.
17 You should have received a copy of the GNU Lesser General Public
18 License along with the GNU C Library; if not, write to the Free
19 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
28 The algorithm is fairly straightforward: set byte by byte until we
29 we get to a 16B-aligned address, then loop on 128 B chunks using an
30 early store as prefetching, then loop on 32B chucks, then clear remaining
31 words, finally clear remaining bytes.
32 Since a stf.spill f0 can store 16B in one go, we use this instruction
52 // This routine uses only scratch predicate registers (p6 - p15)
53 #define p_scr p6 // default register for same-cycle branches
65 #define LSIZE_SH 7 // shift amount
72 #elif defined(USE_FLP)
81 alloc tmp = ar.pfs, 2, 0, 0, 0
87 mov ret0 = dest // return value
89 cmp.eq p_scr, p0 = cnt, r0
92 and ptr2 = -(MIN1+1), dest // aligned address
93 and tmp = MIN1, dest // prepare to check for alignment
94 tbit.nz p_y, p_n = dest, 0 // Do we have an odd address? (M_B_U)
98 (p_scr) br.ret.dpnt.many rp // return immediately if count = 0
101 cmp.ne p_unalgn, p0 = tmp, r0
102 } { .mib // NB: # of bytes to move is 1
103 sub bytecnt = (MIN1+1), tmp // higher than loopcnt
104 cmp.gt p_scr, p0 = 16, cnt // is it a minimalistic task?
105 (p_scr) br.cond.dptk.many .move_bytes_unaligned // go move just a few (M_B_U)
108 (p_unalgn) add ptr1 = (MIN1+1), ptr2 // after alignment
109 (p_unalgn) add ptr2 = MIN1P1HALF, ptr2 // after alignment
110 (p_unalgn) tbit.nz.unc p_y, p_n = bytecnt, 3 // should we do a st8 ?
113 (p_y) add cnt = -8, cnt
114 (p_unalgn) tbit.nz.unc p_yy, p_nn = bytecnt, 2 // should we do a st4 ?
116 (p_y) st8 [ptr2] = r0,-4
117 (p_n) add ptr2 = 4, ptr2
120 (p_yy) add cnt = -4, cnt
121 (p_unalgn) tbit.nz.unc p_y, p_n = bytecnt, 1 // should we do a st2 ?
123 (p_yy) st4 [ptr2] = r0,-2
124 (p_nn) add ptr2 = 2, ptr2
127 mov tmp = LINE_SIZE+1 // for compare
128 (p_y) add cnt = -2, cnt
129 (p_unalgn) tbit.nz.unc p_yy, p_nn = bytecnt, 0 // should we do a st1 ?
132 (p_y) st2 [ptr2] = r0,-1
133 (p_n) add ptr2 = 1, ptr2
137 (p_yy) st1 [ptr2] = r0
138 cmp.gt p_scr, p0 = tmp, cnt // is it a minimalistic task?
140 (p_yy) add cnt = -1, cnt
141 (p_scr) br.cond.dpnt.many .fraction_of_line // go move just a few
145 shr.u linecnt = cnt, LSIZE_SH
150 .l1b: // ------------------// L1B: store ahead into cache lines; fill later
152 and tmp = -(LINE_SIZE), cnt // compute end of range
153 mov ptr9 = ptr1 // used for prefetching
154 and cnt = (LINE_SIZE-1), cnt // remainder
156 mov loopcnt = PREF_AHEAD-1 // default prefetch loop
157 cmp.gt p_scr, p0 = PREF_AHEAD, linecnt // check against actual value
160 (p_scr) add loopcnt = -1, linecnt
161 add ptr2 = 16, ptr1 // start of stores (beyond prefetch stores)
162 add ptr1 = tmp, ptr1 // first address beyond total range
165 add tmp = -1, linecnt // next loop count
166 movi0 ar.lc = loopcnt
170 stf.spill [ptr9] = f0, 128 // Do stores one cache line apart
172 br.cloop.dptk.few .pref_l1b
175 add ptr0 = 16, ptr2 // Two stores in parallel
180 stf.spill [ptr2] = f0, 32
181 stf.spill [ptr0] = f0, 32
184 stf.spill [ptr2] = f0, 32
185 stf.spill [ptr0] = f0, 32
188 stf.spill [ptr2] = f0, 32
189 stf.spill [ptr0] = f0, 64
190 cmp.lt p_scr, p0 = ptr9, ptr1 // do we need more prefetching?
193 stf.spill [ptr2] = f0, 32
194 (p_scr) stf.spill [ptr9] = f0, 128
195 br.cloop.dptk.few .l1bx
198 cmp.gt p_scr, p0 = 8, cnt // just a few bytes left ?
199 (p_scr) br.cond.dpnt.many .move_bytes_from_alignment
205 shr.u loopcnt = cnt, 5 // loopcnt = cnt / 32
208 cmp.eq p_scr, p0 = loopcnt, r0
209 add loopcnt = -1, loopcnt
210 (p_scr) br.cond.dpnt.many .store_words
213 and cnt = 0x1f, cnt // compute the remaining cnt
214 movi0 ar.lc = loopcnt
217 .l2: // -----------------------------// L2A: store 32B in 2 cycles
219 store [ptr1] = myval, 8
220 store [ptr2] = myval, 8
222 store [ptr1] = myval, 24
223 store [ptr2] = myval, 24
224 br.cloop.dptk.many .l2
228 cmp.gt p_scr, p0 = 8, cnt // just a few bytes left ?
229 (p_scr) br.cond.dpnt.many .move_bytes_from_alignment // Branch
233 store [ptr1] = myval, 8 // store
234 cmp.le p_y, p_n = 16, cnt //
235 add cnt = -8, cnt // subtract
238 (p_y) store [ptr1] = myval, 8 // store
239 (p_y) cmp.le.unc p_yy, p_nn = 16, cnt
240 (p_y) add cnt = -8, cnt // subtract
243 (p_yy) store [ptr1] = myval, 8
244 (p_yy) add cnt = -8, cnt // subtract
247 .move_bytes_from_alignment:
249 cmp.eq p_scr, p0 = cnt, r0
250 tbit.nz.unc p_y, p0 = cnt, 2 // should we terminate with a st4 ?
251 (p_scr) br.cond.dpnt.few .restore_and_exit
254 (p_y) st4 [ptr1] = r0,4
255 tbit.nz.unc p_yy, p0 = cnt, 1 // should we terminate with a st2 ?
258 (p_yy) st2 [ptr1] = r0,2
259 tbit.nz.unc p_y, p0 = cnt, 0 // should we terminate with a st1 ?
263 (p_y) st1 [ptr1] = r0
268 movi0 ar.lc = save_lc
272 .move_bytes_unaligned:
274 .pred.rel "mutex",p_y, p_n
275 .pred.rel "mutex",p_yy, p_nn
276 (p_n) cmp.le p_yy, p_nn = 4, cnt
277 (p_y) cmp.le p_yy, p_nn = 5, cnt
278 (p_n) add ptr2 = 2, ptr1
280 (p_y) add ptr2 = 3, ptr1
281 (p_y) st1 [ptr1] = r0, 1 // fill 1 (odd-aligned) byte
282 (p_y) add cnt = -1, cnt // [15, 14 (or less) left]
285 (p_yy) cmp.le.unc p_y, p0 = 8, cnt
286 add ptr3 = ptr1, cnt // prepare last store
287 movi0 ar.lc = save_lc
289 (p_yy) st2 [ptr1] = r0, 4 // fill 2 (aligned) bytes
290 (p_yy) st2 [ptr2] = r0, 4 // fill 2 (aligned) bytes
291 (p_yy) add cnt = -4, cnt // [11, 10 (o less) left]
294 (p_y) cmp.le.unc p_yy, p0 = 8, cnt
295 add ptr3 = -1, ptr3 // last store
296 tbit.nz p_scr, p0 = cnt, 1 // will there be a st2 at the end ?
298 (p_y) st2 [ptr1] = r0, 4 // fill 2 (aligned) bytes
299 (p_y) st2 [ptr2] = r0, 4 // fill 2 (aligned) bytes
300 (p_y) add cnt = -4, cnt // [7, 6 (or less) left]
303 (p_yy) st2 [ptr1] = r0, 4 // fill 2 (aligned) bytes
304 (p_yy) st2 [ptr2] = r0, 4 // fill 2 (aligned) bytes
305 // [3, 2 (or less) left]
306 tbit.nz p_y, p0 = cnt, 0 // will there be a st1 at the end ?
308 (p_yy) add cnt = -4, cnt
311 (p_scr) st2 [ptr1] = r0 // fill 2 (aligned) bytes
312 (p_y) st1 [ptr3] = r0 // fill last byte (using ptr3)