/* * Copyright (c) 2003 Apple Computer, Inc. All rights reserved. * * @APPLE_LICENSE_HEADER_START@ * * The contents of this file constitute Original Code as defined in and * are subject to the Apple Public Source License Version 1.1 (the * "License"). You may not use this file except in compliance with the * License. Please obtain a copy of the License at * http://www.apple.com/publicsource and read it before using this file. * * This 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 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 #include #include #include .text .align 2 /* * WARNING: this code is written for 32-bit mode, and ported by the kernel if necessary * to 64-bit mode for use in the 64-bit commpage. This "port" consists of the following * simple transformations: * - all word compares are changed to doubleword * - all "srwi[.]" opcodes are changed to "srdi[.]" * Nothing else is done. For this to work, the following rules must be * carefully followed: * - do not use carry or overflow * - only use record mode if you are sure the results are mode-invariant * for example, all "andi." and almost all "rlwinm." are fine * - do not use "slwi", "slw", or "srw" * An imaginative programmer could break the porting model in other ways, but the above * are the most likely problem areas. It is perhaps surprising how well in practice * this simple method works. */ /* ********************* * * M E M S E T _ G 5 * * ********************* * * This is a subroutine called by Libc memset and memset_pattern for large nonzero * operands (zero operands are funneled into bzero.) This version is for * 64-bit processors with a 128-byte cache line and Altivec. * * Registers at entry: * r4 = count of bytes to store (must be >= 32) * r8 = ptr to the 1st byte to store (16-byte aligned) * r9 = ptr to 16-byte pattern to store (16-byte aligned) * When we return: * r3 = not changed, since memset returns it * r4 = bytes remaining to store (will be <32) * r7 = not changed * r8 = ptr to next byte to store (still 16-byte aligned) * r12 = not changed (holds return value for memset) */ #define kBig (3*128) // big enough to warrant using dcbz (NB: must be >= 3*128) .align 5 memset_g5: cmplwi cr1,r4,kBig // big enough to warrant using dcbz? neg r10,r8 // start to align ptr mfspr r2,vrsave // we'll be using VRs andi. r10,r10,0x70 // get #bytes to cache line align oris r0,r2,0x8000 // we use vr0 mtspr vrsave,r0 li r5,16 // get offsets for "stvx" lvx v0,0,r9 // load the pattern into v0 li r6,32 blt cr1,LShort // not big enough to bother with dcbz li r9,48 // cache line align beq 2f // already aligned 1: subic. r10,r10,16 // more to go? stvx v0,0,r8 addi r8,r8,16 subi r4,r4,16 bne 1b // Loop over cache lines. This code uses a private protocol with the kernel: // when the kernel emulates an alignment exception on a DCBZ that occurs in the // commpage, it zeroes CR7. We use this to detect the case where we are operating on // uncached memory, and do not use DCBZ again in this code. We assume that either // all the operand is cacheable or none of it is, so we only check the first DCBZ. 2: cmpw cr7,r3,r3 // set cr7_eq (kernel will clear if DCBZ faults) dcbzl 0,r8 // zero first cache line (clearing cr7 if alignment exception) srwi r0,r4,7 // get #cache lines (>=2) rlwinm r4,r4,0,0x7F // mask down to residual count (0..127) bne-- cr7,LNoDcbz // exit if we took alignment exception on the first DCBZ subic r0,r0,1 // loop 1-too-few times li r11,128 // set DCBZ look-ahead mtctr r0 b 3f // use loop that DCBZs // Loop over cache lines. We DCBZ one line ahead, which is a little faster. .align 5 3: dcbzl r11,r8 // zero one line ahead addi r10,r8,64 stvx v0,0,r8 stvx v0,r5,r8 stvx v0,r6,r8 stvx v0,r9,r8 addi r8,r8,128 stvx v0,0,r10 stvx v0,r5,r10 stvx v0,r6,r10 stvx v0,r9,r10 bdnz++ 3b li r0,1 // we've already DCBZ'd the last line LNoDcbz: // r0: loop count mtctr r0 // Loop which does not DCBZ. Normally this is only used for last cache line, // because we've already zeroed it. 4: addi r10,r8,64 stvx v0,0,r8 stvx v0,r5,r8 stvx v0,r6,r8 stvx v0,r9,r8 addi r8,r8,128 stvx v0,0,r10 stvx v0,r5,r10 stvx v0,r6,r10 stvx v0,r9,r10 bdnz-- 4b // optimize for the cacheable case // loop over 32-byte chunks LShort: srwi. r0,r4,5 // get count of 32-byte chunks rlwinm r4,r4,0,0x1F // mask down to residual count (0..31) beq 7f // no chunks so done mtctr r0 6: stvx v0,0,r8 stvx v0,r5,r8 addi r8,r8,32 bdnz++ 6b 7: mtspr vrsave,r2 // restore caller's vrsave blr COMMPAGE_DESCRIPTOR(memset_g5,_COMM_PAGE_MEMSET_PATTERN,kCache128+k64Bit+kHasAltivec,0, \ kCommPageBoth+kPort32to64)