1 /* Copyright (C) 2000, 2003 Free Software Foundation, Inc.
2 Contributed by Richard Henderson (rth@tamu.edu)
3 EV6 optimized by Rick Gorton <rick.gorton@alpha-processor.com>.
4 This file is part of the GNU C Library.
6 The GNU C Library is free software; you can redistribute it and/or
7 modify it under the terms of the GNU Lesser General Public
8 License as published by the Free Software Foundation; either
9 version 2.1 of the License, or (at your option) any later version.
11 The GNU C Library is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public
17 License along with the GNU C Library; if not, write to the Free
18 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
38 * Serious stalling happens. The only way to mitigate this is to
39 * undertake a major re-write to interleave the constant materialization
40 * with other parts of the fall-through code. This is important, even
41 * though it makes maintenance tougher.
44 and $17, 255, $1 # E : 00000000000000ch
45 insbl $17, 1, $2 # U : 000000000000ch00
46 mov $16, $0 # E : return value
47 ble $18, $end # U : zero length requested?
49 addq $18, $16, $6 # E : max address to write to
50 or $1, $2, $17 # E : 000000000000chch
51 insbl $1, 2, $3 # U : 0000000000ch0000
52 insbl $1, 3, $4 # U : 00000000ch000000
54 or $3, $4, $3 # E : 00000000chch0000
55 inswl $17, 4, $5 # U : 0000chch00000000
56 xor $16, $6, $1 # E : will complete write be within one quadword?
57 inswl $17, 6, $2 # U : chch000000000000
59 or $17, $3, $17 # E : 00000000chchchch
60 or $2, $5, $2 # E : chchchch00000000
61 bic $1, 7, $1 # E : fit within a single quadword?
62 and $16, 7, $3 # E : Target addr misalignment
64 or $17, $2, $17 # E : chchchchchchchch
65 beq $1, $within_quad # U :
67 beq $3, $aligned # U : target is 0mod8
70 * Target address is misaligned, and won't fit within a quadword.
72 ldq_u $4, 0($16) # L : Fetch first partial
73 mov $16, $5 # E : Save the address
74 insql $17, $16, $2 # U : Insert new bytes
75 subq $3, 8, $3 # E : Invert (for addressing uses)
77 addq $18, $3, $18 # E : $18 is new count ($3 is negative)
78 mskql $4, $16, $4 # U : clear relevant parts of the quad
79 subq $16, $3, $16 # E : $16 is new aligned destination
80 or $2, $4, $1 # E : Final bytes
83 stq_u $1,0($5) # L : Store result
90 * We are now guaranteed to be quad aligned, with at least
91 * one partial quad to write.
94 sra $18, 3, $3 # U : Number of remaining quads to write
95 and $18, 7, $18 # E : Number of trailing bytes to write
96 mov $16, $5 # E : Save dest address
97 beq $3, $no_quad # U : tail stuff only
100 * It's worth the effort to unroll this and use wh64 if possible.
101 * At this point, entry values are:
102 * $16 Current destination address
104 * $6 The max quadword address to write to
105 * $18 Number trailer bytes
106 * $3 Number quads to write
109 and $16, 0x3f, $2 # E : Forward work (only useful for unrolled loop)
110 subq $3, 16, $4 # E : Only try to unroll if > 128 bytes
111 subq $2, 0x40, $1 # E : bias counter (aligning stuff 0mod64)
115 * We know we've got at least 16 quads, minimum of one trip
116 * through unrolled loop. Do a quad at a time to get us 0mod64
123 beq $1, $bigalign # U :
127 subq $3, 1, $3 # E : For consistency later
128 addq $1, 8, $1 # E : Increment towards zero for alignment
129 addq $5, 8, $4 # E : Initial wh64 address (filler instruction)
133 addq $5, 8, $5 # E : Inc address
134 blt $1, $alignmod64 # U :
138 * $3 - number quads left to go
139 * $5 - target address (aligned 0mod64)
140 * $17 - mask of stuff to store
141 * Scratch registers available: $7, $2, $4, $1
142 * We know that we'll be taking a minimum of one trip through.
143 * CWG Section 3.7.6: do not expect a sustained store rate of > 1/cycle
144 * Assumes the wh64 needs to be for 2 trips through the loop in the future.
145 * The wh64 is issued on for the starting destination address for trip +2
146 * through the loop, and if there are less than two trips left, the target
147 * address will be for the current trip.
151 wh64 ($4) # L1 : memory subsystem write hint
152 subq $3, 24, $2 # E : For determining future wh64 addresses
156 addq $5, 128, $4 # E : speculative target of next wh64
158 stq $17, 16($5) # L :
159 addq $5, 64, $7 # E : Fallback address for wh64 (== next trip addr)
161 stq $17, 24($5) # L :
162 stq $17, 32($5) # L :
163 cmovlt $2, $7, $4 # E : Latency 2, extra mapping cycle
166 stq $17, 40($5) # L :
167 stq $17, 48($5) # L :
168 subq $3, 16, $2 # E : Repeat the loop at least once more?
171 stq $17, 56($5) # L :
172 addq $5, 64, $5 # E :
174 bge $2, $do_wh64 # U :
179 beq $3, $no_quad # U : Might have finished already
183 * Simple loop for trailing quadwords, or for small amounts
184 * of data (where we can't use an unrolled loop and wh64)
188 subq $3, 1, $3 # E : Decrement number quads left
189 addq $5, 8, $5 # E : Inc address
190 bne $3, $loop # U : more?
194 * Write 0..7 trailing bytes.
197 beq $18, $end # U : All done?
199 mskqh $7, $6, $2 # U : Mask final quad
201 insqh $17, $6, $4 # U : New bits
202 or $2, $4, $1 # E : Put it all together
203 stq $1, 0($5) # L : And back to memory
204 ret $31,($26),1 # L0 :
207 ldq_u $1, 0($16) # L :
208 insql $17, $16, $2 # U : New bits
209 mskql $1, $16, $4 # U : Clear old
210 or $2, $4, $2 # E : New result
212 mskql $2, $6, $4 # U :
213 mskqh $1, $6, $2 # U :
215 stq_u $1, 0($16) # L :
221 ret $31,($26),1 # L0 :
224 libc_hidden_builtin_def (memset)