Attachment #207525 for bug #289642




       * sysdeps/alpha/memchr.S: Use prefetch load.
       * sysdeps/alpha/alphaev6/memchr.S: Likewise.
++ b/ports/sysdeps/alpha/alphaev6/memchr.S

        cmpbge  $31, $1, $2	# E :
        bne     $2, $found_it	# U :
	# At least one byte left to process.
	ldq	$31, 8($0)	# L :
	subq	$5, 1, $18	# E : U L U L

	addq	$0, 8, $0	# E :

	and	$4, 8, $4	# E : odd number of quads?
	bne	$4, $odd_quad_count # U :
	# At least three quads remain to be accessed
	nop			# E : L U L U : move prefetched value to correct reg

	.align	4
$unrolled_loop:
	ldq	$1, 0($0)	# L : load quad
	xor	$17, $1, $2	# E :
	ldq	$31, 8($0)	# L : prefetch next quad
	cmpbge	$31, $2, $2	# E : U L U L

	bne	$2, $found_it	# U :
	addq	$0, 8, $0	# E :
	nop			# E :
	nop			# E :
	nop			# E :

$odd_quad_count:
	ldq	$1, 0($0)	# L : load quad
	xor	$17, $1, $2	# E :
	ldq	$31, 8($0)	# L : prefetch $4
	cmpbge	$31, $2, $2	# E :
	addq	$0, 8, $6	# E :

	addq	$0, 8, $6	# E :
	bne	$2, $found_it	# U :
	cmpult	$6, $18, $6	# E :
	addq	$0, 8, $0	# E :
	nop			# E :

	bne	$6, $unrolled_loop # U :
	mov	$4, $1		# E : move prefetched value into $1
	nop			# E :
	nop			# E :

$final:	subq	$5, $0, $18	# E : $18 <- number of bytes left to do
	nop			# E :

$final:	ldq	$1, 0($0)	# L : load last quad
	subq	$5, $0, $18	# E : $18 <- number of bytes left to do
	nop			# E :
	bne	$18, $last_quad	# U :

++ b/ports/sysdeps/alpha/memchr.S


	# At least one byte left to process.

	ldq	zero, 8(v0)	# e0	: prefetch next quad
	subq	t4, 1, a2	# .. e1 :
	addq	v0, 8, v0	#-e0	:



	# At least three quads remain to be accessed

	mov	t0, t3		# e0	: move prefetched value to correct reg

	.align	4
$unrolled_loop:
	ldq	t0, 0(v0)	# e0	: load quad
	xor	a1, t0, t1	# .. e1 :
	ldq	zero, 8(v0)	# e0	: prefetch next quad
	cmpbge	zero, t1, t1	# .. e1:
	bne	t1, $found_it	# e0    :

	addq	v0, 8, v0	#    e1	:
$odd_quad_count:
	ldq	t0, 0(v0)	# e0	: load quad
	xor	a1, t0, t1	# .. e1 :
	ldq	zero, 8(v0)	# e0	: prefetch next quad
	cmpbge	zero, t1, t1	# .. e1 :
	addq	v0, 8, t5	#-e0	:
	bne	t1, $found_it	# .. e1	:

	addq	v0, 8, v0	# .. e1 :
	bne	t5, $unrolled_loop #-e1 :

$final:	ldq	t0, 0(v0)	# e0	: load last quad
	subq	t4, v0, a2	# .. e1	: a2 <- number of bytes left to do
	bne	a2, $last_quad	# e1	:

$not_found:

Return to bug 289642

Line Link Here

(-)file_not_specified_in_diff (-26 / +26 lines)
	* sysdeps/alpha/memchr.S: Use prefetch load.		* sysdeps/alpha/memchr.S: Use prefetch load.
1	* sysdeps/alpha/alphaev6/memchr.S: Likewise.	1	* sysdeps/alpha/alphaev6/memchr.S: Likewise.
2	-- a/ports/sysdeps/alpha/alphaev6/memchr.S	2	++ b/ports/sysdeps/alpha/alphaev6/memchr.S
Lines 127-133 $first_quad: Link Here
127	cmpbge $31, $1, $2 # E :	127	cmpbge $31, $1, $2 # E :
128	bne $2, $found_it # U :	128	bne $2, $found_it # U :
129	# At least one byte left to process.	129	# At least one byte left to process.
130	ldq $1, 8($0) # L :	130	ldq $31, 8($0) # L :
131	subq $5, 1, $18 # E : U L U L	131	subq $5, 1, $18 # E : U L U L
132		132
133	addq $0, 8, $0 # E :	133	addq $0, 8, $0 # E :
Lines 143-180 $first_quad: Link Here
143	and $4, 8, $4 # E : odd number of quads?	143	and $4, 8, $4 # E : odd number of quads?
144	bne $4, $odd_quad_count # U :	144	bne $4, $odd_quad_count # U :
145	# At least three quads remain to be accessed	145	# At least three quads remain to be accessed
146	mov $1, $4 # E : L U L U : move prefetched value to correct reg	146	nop # E : L U L U : move prefetched value to correct reg
147		147
148	.align 4	148	.align 4
149	$unrolled_loop:	149	$unrolled_loop:
150	ldq $1, 8($0) # L : prefetch $1	150	ldq $1, 0($0) # L : load quad
151	xor $17, $4, $2 # E :	151	xor $17, $1, $2 # E :
152	cmpbge $31, $2, $2 # E :	152	ldq $31, 8($0) # L : prefetch next quad
153	bne $2, $found_it # U : U L U L	153	cmpbge $31, $2, $2 # E : U L U L
154		154
		155	bne $2, $found_it # U :
155	addq $0, 8, $0 # E :	156	addq $0, 8, $0 # E :
156	nop # E :	157	nop # E :
157	nop # E :	158	nop # E :
158	nop # E :
159		159
160	$odd_quad_count:	160	$odd_quad_count:
		161	ldq $1, 0($0) # L : load quad
161	xor $17, $1, $2 # E :	162	xor $17, $1, $2 # E :
162	ldq $4, 8($0) # L : prefetch $4	163	ldq $31, 8($0) # L : prefetch $4
163	cmpbge $31, $2, $2 # E :	164	cmpbge $31, $2, $2 # E :
164	addq $0, 8, $6 # E :
165		165
		166	addq $0, 8, $6 # E :
166	bne $2, $found_it # U :	167	bne $2, $found_it # U :
167	cmpult $6, $18, $6 # E :	168	cmpult $6, $18, $6 # E :
168	addq $0, 8, $0 # E :	169	addq $0, 8, $0 # E :
169	nop # E :
170		170
171	bne $6, $unrolled_loop # U :	171	bne $6, $unrolled_loop # U :
172	mov $4, $1 # E : move prefetched value into $1
173	nop # E :	172	nop # E :
174	nop # E :	173	nop # E :
175
176	$final: subq $5, $0, $18 # E : $18 <- number of bytes left to do
177	nop # E :	174	nop # E :
		175
		176	$final: ldq $1, 0($0) # L : load last quad
		177	subq $5, $0, $18 # E : $18 <- number of bytes left to do
178	nop # E :	178	nop # E :
179	bne $18, $last_quad # U :	179	bne $18, $last_quad # U :
180		180
181	-- a/ports/sysdeps/alpha/memchr.S	181	++ b/ports/sysdeps/alpha/memchr.S
Lines 119-125 $first_quad: Link Here
119		119
120	# At least one byte left to process.	120	# At least one byte left to process.
121		121
122	ldq t0, 8(v0) # e0 :	122	ldq zero, 8(v0) # e0 : prefetch next quad
123	subq t4, 1, a2 # .. e1 :	123	subq t4, 1, a2 # .. e1 :
124	addq v0, 8, v0 #-e0 :	124	addq v0, 8, v0 #-e0 :
125		125
Lines 138-156 $first_quad: Link Here
138		138
139	# At least three quads remain to be accessed	139	# At least three quads remain to be accessed
140		140
141	mov t0, t3 # e0 : move prefetched value to correct reg
142
143	.align 4	141	.align 4
144	$unrolled_loop:	142	$unrolled_loop:
145	ldq t0, 8(v0) #-e0 : prefetch t0	143	ldq t0, 0(v0) # e0 : load quad
146	xor a1, t3, t1 # .. e1 :	144	xor a1, t0, t1 # .. e1 :
147	cmpbge zero, t1, t1 # e0 :	145	ldq zero, 8(v0) # e0 : prefetch next quad
148	bne t1, $found_it # .. e1 :	146	cmpbge zero, t1, t1 # .. e1:
		147	bne t1, $found_it # e0 :
149		148
150	addq v0, 8, v0 #-e0 :	149	addq v0, 8, v0 # e1 :
151	$odd_quad_count:	150	$odd_quad_count:
		151	ldq t0, 0(v0) # e0 : load quad
152	xor a1, t0, t1 # .. e1 :	152	xor a1, t0, t1 # .. e1 :
153	ldq t3, 8(v0) # e0 : prefetch t3	153	ldq zero, 8(v0) # e0 : prefetch next quad
154	cmpbge zero, t1, t1 # .. e1 :	154	cmpbge zero, t1, t1 # .. e1 :
155	addq v0, 8, t5 #-e0 :	155	addq v0, 8, t5 #-e0 :
156	bne t1, $found_it # .. e1 :	156	bne t1, $found_it # .. e1 :
Lines 159-166 $odd_quad_count: Link Here
159	addq v0, 8, v0 # .. e1 :	159	addq v0, 8, v0 # .. e1 :
160	bne t5, $unrolled_loop #-e1 :	160	bne t5, $unrolled_loop #-e1 :
161		161
162	mov t3, t0 # e0 : move prefetched value into t0	162	$final: ldq t0, 0(v0) # e0 : load last quad
163	$final: subq t4, v0, a2 # .. e1 : a2 <- number of bytes left to do	163	subq t4, v0, a2 # .. e1 : a2 <- number of bytes left to do
164	bne a2, $last_quad # e1 :	164	bne a2, $last_quad # e1 :
165		165
166	$not_found:	166	$not_found: