Blame view
arch/alpha/lib/ev67-strchr.S
2.73 KB
1da177e4c
Linux-2.6.12-rc2
|
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 |
/* * arch/alpha/lib/ev67-strchr.S * 21264 version contributed by Rick Gorton <rick.gorton@alpha-processor.com> * * Return the address of a given character within a null-terminated * string, or null if it is not found. * * Much of the information about 21264 scheduling/coding comes from: * Compiler Writer's Guide for the Alpha 21264 * abbreviated as 'CWG' in other comments here * ftp.digital.com/pub/Digital/info/semiconductor/literature/dsc-library.html * Scheduling notation: * E - either cluster * U - upper subcluster; U0 - subcluster U0; U1 - subcluster U1 * L - lower subcluster; L0 - subcluster L0; L1 - subcluster L1 * Try not to change the actual algorithm if possible for consistency. */ #include <asm/regdef.h> .set noreorder .set noat .align 4 .globl strchr .ent strchr strchr: .frame sp, 0, ra .prologue 0 ldq_u t0, 0(a0) # L : load first quadword Latency=3 and a1, 0xff, t3 # E : 00000000000000ch insbl a1, 1, t5 # U : 000000000000ch00 insbl a1, 7, a2 # U : ch00000000000000 insbl t3, 6, a3 # U : 00ch000000000000 or t5, t3, a1 # E : 000000000000chch andnot a0, 7, v0 # E : align our loop pointer lda t4, -1 # E : build garbage mask mskqh t4, a0, t4 # U : only want relevant part of first quad or a2, a3, a2 # E : chch000000000000 inswl a1, 2, t5 # E : 00000000chch0000 inswl a1, 4, a3 # E : 0000chch00000000 or a1, a2, a1 # E : chch00000000chch or a3, t5, t5 # E : 0000chchchch0000 cmpbge zero, t0, t2 # E : bits set iff byte == zero cmpbge zero, t4, t4 # E : bits set iff byte is garbage /* This quad is _very_ serialized. Lots of stalling happens */ or t5, a1, a1 # E : chchchchchchchch xor t0, a1, t1 # E : make bytes == c zero cmpbge zero, t1, t3 # E : bits set iff byte == c or t2, t3, t0 # E : bits set iff char match or zero match andnot t0, t4, t0 # E : clear garbage bits cttz t0, a2 # U0 : speculative (in case we get a match) nop # E : bne t0, $found # U : /* * Yuk. This loop is going to stall like crazy waiting for the * data to be loaded. Not much can be done about it unless it's * unrolled multiple times - is that safe to do in kernel space? * Or would exception handling recovery code do the trick here? */ $loop: ldq t0, 8(v0) # L : Latency=3 addq v0, 8, v0 # E : xor t0, a1, t1 # E : cmpbge zero, t0, t2 # E : bits set iff byte == 0 cmpbge zero, t1, t3 # E : bits set iff byte == c or t2, t3, t0 # E : cttz t3, a2 # U0 : speculative (in case we get a match) beq t0, $loop # U : $found: negq t0, t1 # E : clear all but least set bit and t0, t1, t0 # E : and t0, t3, t1 # E : bit set iff byte was the char addq v0, a2, v0 # E : Add in the bit number from above cmoveq t1, $31, v0 # E : Two mapping slots, latency = 2 nop nop ret # L0 : .end strchr |