Orange Pi5 kernel

^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300   1) /* SPDX-License-Identifier: GPL-2.0 */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300   2) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300   3)  *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300   4)  * Optimized version of the standard memcpy() function
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300   5)  *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300   6)  * Inputs:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300   7)  * 	in0:	destination address
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300   8)  *	in1:	source address
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300   9)  *	in2:	number of bytes to copy
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  10)  * Output:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  11)  * 	no return value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  12)  *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  13)  * Copyright (C) 2000-2001 Hewlett-Packard Co
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  14)  *	Stephane Eranian <eranian@hpl.hp.com>
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  15)  *	David Mosberger-Tang <davidm@hpl.hp.com>
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  16)  */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  17) #include <asm/asmmacro.h>
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  18) #include <asm/export.h>
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  19) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  20) GLOBAL_ENTRY(memcpy)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  21) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  22) #	define MEM_LAT	21		/* latency to memory */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  23) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  24) #	define dst	r2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  25) #	define src	r3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  26) #	define retval	r8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  27) #	define saved_pfs r9
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  28) #	define saved_lc	r10
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  29) #	define saved_pr	r11
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  30) #	define cnt	r16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  31) #	define src2	r17
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  32) #	define t0	r18
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  33) #	define t1	r19
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  34) #	define t2	r20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  35) #	define t3	r21
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  36) #	define t4	r22
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  37) #	define src_end	r23
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  38) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  39) #	define N	(MEM_LAT + 4)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  40) #	define Nrot	((N + 7) & ~7)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  41) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  42) 	/*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  43) 	 * First, check if everything (src, dst, len) is a multiple of eight.  If
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  44) 	 * so, we handle everything with no taken branches (other than the loop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  45) 	 * itself) and a small icache footprint.  Otherwise, we jump off to
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  46) 	 * the more general copy routine handling arbitrary
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  47) 	 * sizes/alignment etc.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  48) 	 */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  49) 	.prologue
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  50) 	.save ar.pfs, saved_pfs
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  51) 	alloc saved_pfs=ar.pfs,3,Nrot,0,Nrot
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  52) 	.save ar.lc, saved_lc
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  53) 	mov saved_lc=ar.lc
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  54) 	or t0=in0,in1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  55) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  56) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  57) 	or t0=t0,in2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  58) 	.save pr, saved_pr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  59) 	mov saved_pr=pr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  60) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  61) 	.body
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  62) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  63) 	cmp.eq p6,p0=in2,r0	// zero length?
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  64) 	mov retval=in0		// return dst
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  65) (p6)	br.ret.spnt.many rp	// zero length, return immediately
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  66) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  67) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  68) 	mov dst=in0		// copy because of rotation
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  69) 	shr.u cnt=in2,3		// number of 8-byte words to copy
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  70) 	mov pr.rot=1<<16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  71) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  72) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  73) 	adds cnt=-1,cnt		// br.ctop is repeat/until
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  74) 	cmp.gtu p7,p0=16,in2	// copying less than 16 bytes?
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  75) 	mov ar.ec=N
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  76) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  77) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  78) 	and t0=0x7,t0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  79) 	mov ar.lc=cnt
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  80) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  81) 	cmp.ne p6,p0=t0,r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  82) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  83) 	mov src=in1		// copy because of rotation
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  84) (p7)	br.cond.spnt.few .memcpy_short
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  85) (p6)	br.cond.spnt.few .memcpy_long
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  86) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  87) 	nop.m	0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  88) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  89) 	nop.m	0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  90) 	nop.i	0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  91) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  92) 	nop.m	0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  93) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  94) 	.rotr val[N]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  95) 	.rotp p[N]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  96) 	.align 32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  97) 1: { .mib
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  98) (p[0])	ld8 val[0]=[src],8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300  99) 	nop.i 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 100) 	brp.loop.imp 1b, 2f
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 101) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 102) 2: { .mfb
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 103) (p[N-1])st8 [dst]=val[N-1],8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 104) 	nop.f 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 105) 	br.ctop.dptk.few 1b
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 106) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 107) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 108) 	mov ar.lc=saved_lc
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 109) 	mov pr=saved_pr,-1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 110) 	mov ar.pfs=saved_pfs
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 111) 	br.ret.sptk.many rp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 112) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 113) 	/*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 114) 	 * Small (<16 bytes) unaligned copying is done via a simple byte-at-the-time
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 115) 	 * copy loop.  This performs relatively poorly on Itanium, but it doesn't
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 116) 	 * get used very often (gcc inlines small copies) and due to atomicity
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 117) 	 * issues, we want to avoid read-modify-write of entire words.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 118) 	 */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 119) 	.align 32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 120) .memcpy_short:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 121) 	adds cnt=-1,in2		// br.ctop is repeat/until
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 122) 	mov ar.ec=MEM_LAT
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 123) 	brp.loop.imp 1f, 2f
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 124) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 125) 	mov ar.lc=cnt
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 126) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 127) 	nop.m	0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 128) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 129) 	nop.m	0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 130) 	nop.i	0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 131) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 132) 	nop.m	0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 133) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 134) 	nop.m	0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 135) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 136) 	/*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 137) 	 * It is faster to put a stop bit in the loop here because it makes
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 138) 	 * the pipeline shorter (and latency is what matters on short copies).
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 139) 	 */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 140) 	.align 32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 141) 1: { .mib
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 142) (p[0])	ld1 val[0]=[src],1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 143) 	nop.i 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 144) 	brp.loop.imp 1b, 2f
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 145) } ;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 146) 2: { .mfb
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 147) (p[MEM_LAT-1])st1 [dst]=val[MEM_LAT-1],1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 148) 	nop.f 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 149) 	br.ctop.dptk.few 1b
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 150) } ;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 151) 	mov ar.lc=saved_lc
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 152) 	mov pr=saved_pr,-1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 153) 	mov ar.pfs=saved_pfs
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 154) 	br.ret.sptk.many rp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 155) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 156) 	/*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 157) 	 * Large (>= 16 bytes) copying is done in a fancy way.  Latency isn't
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 158) 	 * an overriding concern here, but throughput is.  We first do
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 159) 	 * sub-word copying until the destination is aligned, then we check
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 160) 	 * if the source is also aligned.  If so, we do a simple load/store-loop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 161) 	 * until there are less than 8 bytes left over and then we do the tail,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 162) 	 * by storing the last few bytes using sub-word copying.  If the source
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 163) 	 * is not aligned, we branch off to the non-congruent loop.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 164) 	 *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 165) 	 *   stage:   op:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 166) 	 *         0  ld
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 167) 	 *	   :
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 168) 	 * MEM_LAT+3  shrp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 169) 	 * MEM_LAT+4  st
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 170) 	 *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 171) 	 * On Itanium, the pipeline itself runs without stalls.  However,  br.ctop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 172) 	 * seems to introduce an unavoidable bubble in the pipeline so the overall
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 173) 	 * latency is 2 cycles/iteration.  This gives us a _copy_ throughput
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 174) 	 * of 4 byte/cycle.  Still not bad.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 175) 	 */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 176) #	undef N
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 177) #	undef Nrot
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 178) #	define N	(MEM_LAT + 5)		/* number of stages */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 179) #	define Nrot	((N+1 + 2 + 7) & ~7)	/* number of rotating regs */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 180) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 181) #define LOG_LOOP_SIZE	6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 182) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 183) .memcpy_long:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 184) 	alloc t3=ar.pfs,3,Nrot,0,Nrot	// resize register frame
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 185) 	and t0=-8,src		// t0 = src & ~7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 186) 	and t2=7,src		// t2 = src & 7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 187) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 188) 	ld8 t0=[t0]		// t0 = 1st source word
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 189) 	adds src2=7,src		// src2 = (src + 7)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 190) 	sub t4=r0,dst		// t4 = -dst
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 191) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 192) 	and src2=-8,src2	// src2 = (src + 7) & ~7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 193) 	shl t2=t2,3		// t2 = 8*(src & 7)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 194) 	shl t4=t4,3		// t4 = 8*(dst & 7)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 195) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 196) 	ld8 t1=[src2]		// t1 = 1st source word if src is 8-byte aligned, 2nd otherwise
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 197) 	sub t3=64,t2		// t3 = 64-8*(src & 7)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 198) 	shr.u t0=t0,t2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 199) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 200) 	add src_end=src,in2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 201) 	shl t1=t1,t3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 202) 	mov pr=t4,0x38		// (p5,p4,p3)=(dst & 7)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 203) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 204) 	or t0=t0,t1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 205) 	mov cnt=r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 206) 	adds src_end=-1,src_end
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 207) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 208) (p3)	st1 [dst]=t0,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 209) (p3)	shr.u t0=t0,8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 210) (p3)	adds cnt=1,cnt
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 211) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 212) (p4)	st2 [dst]=t0,2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 213) (p4)	shr.u t0=t0,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 214) (p4)	adds cnt=2,cnt
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 215) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 216) (p5)	st4 [dst]=t0,4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 217) (p5)	adds cnt=4,cnt
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 218) 	and src_end=-8,src_end	// src_end = last word of source buffer
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 219) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 220) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 221) 	// At this point, dst is aligned to 8 bytes and there at least 16-7=9 bytes left to copy:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 222) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 223) 1:{	add src=cnt,src			// make src point to remainder of source buffer
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 224) 	sub cnt=in2,cnt			// cnt = number of bytes left to copy
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 225) 	mov t4=ip
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 226)   }	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 227) 	and src2=-8,src			// align source pointer
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 228) 	adds t4=.memcpy_loops-1b,t4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 229) 	mov ar.ec=N
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 230) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 231) 	and t0=7,src			// t0 = src & 7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 232) 	shr.u t2=cnt,3			// t2 = number of 8-byte words left to copy
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 233) 	shl cnt=cnt,3			// move bits 0-2 to 3-5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 234) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 235) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 236) 	.rotr val[N+1], w[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 237) 	.rotp p[N]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 238) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 239) 	cmp.ne p6,p0=t0,r0		// is src aligned, too?
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 240) 	shl t0=t0,LOG_LOOP_SIZE		// t0 = 8*(src & 7)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 241) 	adds t2=-1,t2			// br.ctop is repeat/until
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 242) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 243) 	add t4=t0,t4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 244) 	mov pr=cnt,0x38			// set (p5,p4,p3) to # of bytes last-word bytes to copy
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 245) 	mov ar.lc=t2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 246) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 247) 	nop.m	0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 248) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 249) 	nop.m	0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 250) 	nop.i	0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 251) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 252) 	nop.m	0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 253) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 254) (p6)	ld8 val[1]=[src2],8		// prime the pump...
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 255) 	mov b6=t4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 256) 	br.sptk.few b6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 257) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 258) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 259) .memcpy_tail:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 260) 	// At this point, (p5,p4,p3) are set to the number of bytes left to copy (which is
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 261) 	// less than 8) and t0 contains the last few bytes of the src buffer:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 262) (p5)	st4 [dst]=t0,4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 263) (p5)	shr.u t0=t0,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 264) 	mov ar.lc=saved_lc
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 265) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 266) (p4)	st2 [dst]=t0,2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 267) (p4)	shr.u t0=t0,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 268) 	mov ar.pfs=saved_pfs
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 269) 	;;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 270) (p3)	st1 [dst]=t0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 271) 	mov pr=saved_pr,-1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 272) 	br.ret.sptk.many rp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 273) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 274) ///////////////////////////////////////////////////////
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 275) 	.align 64
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 276) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 277) #define COPY(shift,index)									\
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 278)  1: { .mib											\
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 279) 	(p[0])		ld8 val[0]=[src2],8;							\
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 280) 	(p[MEM_LAT+3])	shrp w[0]=val[MEM_LAT+3],val[MEM_LAT+4-index],shift;			\
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 281) 			brp.loop.imp 1b, 2f							\
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 282)     };												\
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 283)  2: { .mfb											\
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 284) 	(p[MEM_LAT+4])	st8 [dst]=w[1],8;							\
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 285) 			nop.f 0;								\
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 286) 			br.ctop.dptk.few 1b;							\
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 287)     };												\
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 288) 			;;									\
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 289) 			ld8 val[N-1]=[src_end];	/* load last word (may be same as val[N]) */	\
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 290) 			;;									\
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 291) 			shrp t0=val[N-1],val[N-index],shift;					\
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 292) 			br .memcpy_tail
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 293) .memcpy_loops:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 294) 	COPY(0, 1) /* no point special casing this---it doesn't go any faster without shrp */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 295) 	COPY(8, 0)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 296) 	COPY(16, 0)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 297) 	COPY(24, 0)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 298) 	COPY(32, 0)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 299) 	COPY(40, 0)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 300) 	COPY(48, 0)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 301) 	COPY(56, 0)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 302) 
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 303) END(memcpy)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 304) EXPORT_SYMBOL(memcpy)