/* * Copyright (c) 2002 Apple Computer, Inc. All rights reserved. * * @APPLE_LICENSE_HEADER_START@ * * This file contains Original Code and/or Modifications of Original Code * as defined in and that are subject to the Apple Public Source License * Version 2.0 (the 'License'). You may not use this file except in * compliance with the License. Please obtain a copy of the License at * http://www.opensource.apple.com/apsl/ and read it before using this * file. * * The Original Code and all software distributed under the License are * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. * Please see the License for the specific language governing rights and * limitations under the License. * * @APPLE_LICENSE_HEADER_END@ */ #define ASSEMBLER #include #undef ASSEMBLER #define __APPLE_API_PRIVATE #include #undef __APPLE_API_PRIVATE /* We use mode-independent "g" opcodes such as "srgi". These expand * into word operations when targeting __ppc__, and into doubleword * operations when targeting __ppc64__. */ #include // *************** // * S T R C M P * // *************** // // int strcmp(const char *s1, const char *s2); // // We optimize the compare by doing it word parallel. This introduces // a complication: if we blindly did word loads from both sides until // finding a difference (or 0), we might get a spurious page fault by // reading bytes past the difference. To avoid this, we never do a "lwz" // that crosses a page boundary. // // The test for 0s relies on the following inobvious but very efficient // word-parallel test: // x = dataWord + 0xFEFEFEFF // y = ~dataWord & 0x80808080 // if (x & y) == 0 then no zero found // The test maps any non-zero byte to zero, and any zero byte to 0x80, // with one exception: 0x01 bytes preceeding the first zero are also // mapped to 0x80. // // In 64-bit mode, the algorithm is doubleword parallel. .text .globl EXT(strcmp) .align 5 LEXT(strcmp) // int strcmp(const char *s1, const char *s2); andi. r0,r3,GPR_BYTES-1 // is LHS aligned? #if defined(__ppc__) lis r5,hi16(0xFEFEFEFF) // start to generate 32-bit magic constants lis r6,hi16(0x80808080) ori r5,r5,lo16(0xFEFEFEFF) ori r6,r6,lo16(0x80808080) #else ld r5,_COMM_PAGE_MAGIC_FE(0) // get 0xFEFEFEFE FEFEFEFF from commpage ld r6,_COMM_PAGE_MAGIC_80(0) // get 0x80808080 80808080 from commpage #endif subi r3,r3,GPR_BYTES // we use "lgu" in the inner loops subi r4,r4,GPR_BYTES beq Laligned // LHS is aligned subfic r0,r0,GPR_BYTES // r0 <- #bytes to align LHS mtctr r0 // Loop over bytes. Lbyteloop: lbz r7,GPR_BYTES(r3) // r7 <- next LHS byte addi r3,r3,1 lbz r8,GPR_BYTES(r4) // r8 <- next RHS byte addi r4,r4,1 cntlzw r9,r7 // is r7 zero? sub r0,r7,r8 // different? srwi r9,r9,5 // r9 <- (r7==0) ? 1 : 0 or. r9,r9,r0 // r9 is nonzero if either different or 0 bdnzt eq,Lbyteloop // loop until different, 0, or buf end bne Ldone // done if different or 0 // LHS is aligned. If RHS also is, we need not worry about page // crossing. Otherwise, we must stop the loop before page is crossed. Laligned: andi. r0,r4,GPR_BYTES-1 // is RHS now aligned too? addi r9,r4,GPR_BYTES // restore true address of next RHS byte rlwinm r9,r9,0,0xFFF // get RHS offset in page beq Lalignedloop // RHS also aligned, use simple loop subfic r9,r9,4096 // get #bytes left in RHS page srwi. r0,r9,LOG2_GPR_BYTES// get #words or doublewords left in RHS page mtctr r0 // set up loop count bne++ Lunalignedloop // at least one word left in RHS page li r0,GPR_BYTES // must check GPR_BYTES, a byte at a time... mtctr r0 // ...in order to keep LHS aligned b Lbyteloop // go cross RHS page // Unaligned inner loop: compare a word or doubleword at a time, until one of // three conditions: // - a difference is found // - a zero byte is found // - end of RHS page (we dare not touch next page until we must) // At this point, registers are as follows: // r3 = LHS ptr - GPR_BYTES (aligned) // r4 = RHS ptr - GPR_BYTES (not aligned) // r5 = 0xFEFEFEFF // r6 = 0x80808080 // ctr = words or doublewords left in RHS page .align 5 // align inner loop, which is 8 words long Lunalignedloop: lgu r7,GPR_BYTES(r3) // r7 <- next LHS bytes lgu r8,GPR_BYTES(r4) // r8 <- next RHS bytes add r10,r7,r5 // r10 <- LHS + 0xFEFEFEFF andc r12,r6,r7 // r12 <- ~LHS & 0x80808080 xor r11,r7,r8 // r11 <- compare the words and r0,r10,r12 // r0 <- nonzero iff LHS has a 0-byte or. r12,r0,r11 // combine difference and 0-test vectors bdnzt eq,Lunalignedloop // loop if ctr!=0 and cr0_eq bne++ Ldifferent // done if we found a 0 or difference li r0,GPR_BYTES // must check GPR_BYTES, a byte at a time... mtctr r0 // ...in order to keep LHS word aligned b Lbyteloop // cross RHS page, then resume word loop // Aligned inner loop: compare a word at a time, until one of two conditions: // - a difference is found // - a zero byte is found // At this point, registers are as follows: // r3 = LHS ptr - 4 (word aligned) // r4 = RHS ptr - 4 (word aligned) // r5 = 0xFEFEFEFF // r6 = 0x80808080 .align 5 // align inner loop, which is 8 words ling Lalignedloop: lgu r7,GPR_BYTES(r3) // r7 <- next LHS bytes lgu r8,GPR_BYTES(r4) // r8 <- next RHS bytes add r10,r7,r5 // r10 <- LHS + 0xFEFEFEFF andc r12,r6,r7 // r12 <- ~LHS & 0x80808080 xor r11,r7,r8 // r11 <- compare the words and r0,r10,r12 // r0 <- nonzero iff LHS has a 0-byte or. r12,r0,r11 // combine difference and 0-test vectors beq Lalignedloop // loop if neither found // Found differing bytes and/or a 0-byte. Determine which comes first, and // subtract the bytes to compute the return value. We also need to mask out the // false hits in the 0-byte test, which consist of 0x01 bytes that preceed // the 0-byte. Ldifferent: // r0 == 0-test vector (with 0x01 false hits) slgi r9,r7,7 // move 0x01 bits in LHS into position 0x80 andc r0,r0,r9 // mask out the false 0-hits from 0x01 bytes or r11,r11,r0 // recompute difference vector cntlzg r9,r11 // find 1st difference (r9 = 0..31 or 63) rlwinm r9,r9,0,0x38 // byte align bit offset (now, r9 = 0,8,16, or 24 etc) addi r0,r9,8 // now, r0 = 8, 16, 24, or 32 #if defined(__ppc__) rlwnm r7,r7,r0,24,31 // right justify differing bytes and mask off rest rlwnm r8,r8,r0,24,31 #else rldcl r7,r7,r0,56 // right justify differing bytes and mask off rest rldcl r8,r8,r0,56 #endif Ldone: // differing bytes are in r7 and r8 sub r3,r7,r8 // compute difference (0, +, or -) blr