/* Copy SIZE bytes from SRC to DEST. For UltraSPARC-III. Copyright (C) 2001, 2003 Free Software Foundation, Inc. This file is part of the GNU C Library. Contributed by David S. Miller (davem@redhat.com) The GNU C Library is free software; you can redistribute it and/or modify it under the terms of the GNU Lesser General Public License as published by the Free Software Foundation; either version 2.1 of the License, or (at your option) any later version. The GNU C Library is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more details. You should have received a copy of the GNU Lesser General Public License along with the GNU C Library; if not, write to the Free Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA. */ #include <sysdep.h> #define ASI_BLK_P 0xf0 #define FPRS_FEF 0x04 #define VISEntryHalf rd %fprs, %o5; wr %g0, FPRS_FEF, %fprs #define VISExitHalf and %o5, FPRS_FEF, %o5; wr %o5, 0x0, %fprs #ifndef XCC #define USE_BPR #define XCC xcc #endif .register %g2,#scratch .register %g3,#scratch .register %g6,#scratch .text .align 32 ENTRY(bcopy) sub %o1, %o0, %o4 mov %o0, %g4 cmp %o4, %o2 mov %o1, %o0 bgeu,pt %XCC, 100f mov %g4, %o1 #ifndef USE_BPR srl %o2, 0, %o2 #endif brnz,pn %o2, 220f add %o0, %o2, %o0 retl nop END(bcopy) /* Special/non-trivial issues of this code: * * 1) %o5 is preserved from VISEntryHalf to VISExitHalf * 2) Only low 32 FPU registers are used so that only the * lower half of the FPU register set is dirtied by this * code. This is especially important in the kernel. * 3) This code never prefetches cachelines past the end * of the source buffer. * * The cheetah's flexible spine, oversized liver, enlarged heart, * slender muscular body, and claws make it the swiftest hunter * in Africa and the fastest animal on land. Can reach speeds * of up to 2.4GB per second. */ .align 32 ENTRY(memcpy) 100: /* %o0=dst, %o1=src, %o2=len */ mov %o0, %g5 cmp %o2, 0 be,pn %XCC, out 218: or %o0, %o1, %o3 cmp %o2, 16 bleu,a,pn %XCC, small_copy or %o3, %o2, %o3 cmp %o2, 256 blu,pt %XCC, medium_copy andcc %o3, 0x7, %g0 ba,pt %xcc, enter andcc %o0, 0x3f, %g2 /* Here len >= 256 and condition codes reflect execution * of "andcc %o0, 0x7, %g2", done by caller. */ .align 64 enter: /* Is 'dst' already aligned on an 64-byte boundary? */ be,pt %XCC, 2f /* Compute abs((dst & 0x3f) - 0x40) into %g2. This is the number * of bytes to copy to make 'dst' 64-byte aligned. We pre- * subtract this from 'len'. */ sub %g2, 0x40, %g2 sub %g0, %g2, %g2 sub %o2, %g2, %o2 /* Copy %g2 bytes from src to dst, one byte at a time. */ 1: ldub [%o1 + 0x00], %o3 add %o1, 0x1, %o1 add %o0, 0x1, %o0 subcc %g2, 0x1, %g2 bg,pt %XCC, 1b stb %o3, [%o0 + -1] 2: VISEntryHalf and %o1, 0x7, %g1 ba,pt %xcc, begin alignaddr %o1, %g0, %o1 .align 64 begin: prefetch [%o1 + 0x000], #one_read prefetch [%o1 + 0x040], #one_read andn %o2, (0x40 - 1), %o4 prefetch [%o1 + 0x080], #one_read prefetch [%o1 + 0x0c0], #one_read ldd [%o1 + 0x000], %f0 prefetch [%o1 + 0x100], #one_read ldd [%o1 + 0x008], %f2 prefetch [%o1 + 0x140], #one_read ldd [%o1 + 0x010], %f4 prefetch [%o1 + 0x180], #one_read faligndata %f0, %f2, %f16 ldd [%o1 + 0x018], %f6 faligndata %f2, %f4, %f18 ldd [%o1 + 0x020], %f8 faligndata %f4, %f6, %f20 ldd [%o1 + 0x028], %f10 faligndata %f6, %f8, %f22 ldd [%o1 + 0x030], %f12 faligndata %f8, %f10, %f24 ldd [%o1 + 0x038], %f14 faligndata %f10, %f12, %f26 ldd [%o1 + 0x040], %f0 sub %o4, 0x80, %o4 add %o1, 0x40, %o1 ba,pt %xcc, loop srl %o4, 6, %o3 .align 64 loop: ldd [%o1 + 0x008], %f2 faligndata %f12, %f14, %f28 ldd [%o1 + 0x010], %f4 faligndata %f14, %f0, %f30 stda %f16, [%o0] ASI_BLK_P ldd [%o1 + 0x018], %f6 faligndata %f0, %f2, %f16 ldd [%o1 + 0x020], %f8 faligndata %f2, %f4, %f18 ldd [%o1 + 0x028], %f10 faligndata %f4, %f6, %f20 ldd [%o1 + 0x030], %f12 faligndata %f6, %f8, %f22 ldd [%o1 + 0x038], %f14 faligndata %f8, %f10, %f24 ldd [%o1 + 0x040], %f0 prefetch [%o1 + 0x180], #one_read faligndata %f10, %f12, %f26 subcc %o3, 0x01, %o3 add %o1, 0x40, %o1 bg,pt %XCC, loop add %o0, 0x40, %o0 /* Finally we copy the last full 64-byte block. */ loopfini: ldd [%o1 + 0x008], %f2 faligndata %f12, %f14, %f28 ldd [%o1 + 0x010], %f4 faligndata %f14, %f0, %f30 stda %f16, [%o0] ASI_BLK_P ldd [%o1 + 0x018], %f6 faligndata %f0, %f2, %f16 ldd [%o1 + 0x020], %f8 faligndata %f2, %f4, %f18 ldd [%o1 + 0x028], %f10 faligndata %f4, %f6, %f20 ldd [%o1 + 0x030], %f12 faligndata %f6, %f8, %f22 ldd [%o1 + 0x038], %f14 faligndata %f8, %f10, %f24 cmp %g1, 0 be,pt %XCC, 1f add %o0, 0x40, %o0 ldd [%o1 + 0x040], %f0 1: faligndata %f10, %f12, %f26 faligndata %f12, %f14, %f28 faligndata %f14, %f0, %f30 stda %f16, [%o0] ASI_BLK_P add %o0, 0x40, %o0 add %o1, 0x40, %o1 membar #Sync /* Now we copy the (len modulo 64) bytes at the end. * Note how we borrow the %f0 loaded above. * * Also notice how this code is careful not to perform a * load past the end of the src buffer. */ loopend: and %o2, 0x3f, %o2 andcc %o2, 0x38, %g2 be,pn %XCC, endcruft subcc %g2, 0x8, %g2 be,pn %XCC, endcruft cmp %g1, 0 be,a,pt %XCC, 1f ldd [%o1 + 0x00], %f0 1: ldd [%o1 + 0x08], %f2 add %o1, 0x8, %o1 sub %o2, 0x8, %o2 subcc %g2, 0x8, %g2 faligndata %f0, %f2, %f8 std %f8, [%o0 + 0x00] be,pn %XCC, endcruft add %o0, 0x8, %o0 ldd [%o1 + 0x08], %f0 add %o1, 0x8, %o1 sub %o2, 0x8, %o2 subcc %g2, 0x8, %g2 faligndata %f2, %f0, %f8 std %f8, [%o0 + 0x00] bne,pn %XCC, 1b add %o0, 0x8, %o0 /* If anything is left, we copy it one byte at a time. * Note that %g1 is (src & 0x3) saved above before the * alignaddr was performed. */ endcruft: cmp %o2, 0 add %o1, %g1, %o1 VISExitHalf be,pn %XCC, out sub %o0, %o1, %o3 andcc %g1, 0x7, %g0 bne,pn %icc, small_copy_unaligned andcc %o2, 0x8, %g0 be,pt %icc, 1f nop ldx [%o1], %o5 stx %o5, [%o1 + %o3] add %o1, 0x8, %o1 1: andcc %o2, 0x4, %g0 be,pt %icc, 1f nop lduw [%o1], %o5 stw %o5, [%o1 + %o3] add %o1, 0x4, %o1 1: andcc %o2, 0x2, %g0 be,pt %icc, 1f nop lduh [%o1], %o5 sth %o5, [%o1 + %o3] add %o1, 0x2, %o1 1: andcc %o2, 0x1, %g0 be,pt %icc, out nop ldub [%o1], %o5 ba,pt %xcc, out stb %o5, [%o1 + %o3] medium_copy: /* 16 < len <= 64 */ bne,pn %XCC, small_copy_unaligned sub %o0, %o1, %o3 medium_copy_aligned: andn %o2, 0x7, %o4 and %o2, 0x7, %o2 1: subcc %o4, 0x8, %o4 ldx [%o1], %o5 stx %o5, [%o1 + %o3] bgu,pt %XCC, 1b add %o1, 0x8, %o1 andcc %o2, 0x4, %g0 be,pt %XCC, 1f nop sub %o2, 0x4, %o2 lduw [%o1], %o5 stw %o5, [%o1 + %o3] add %o1, 0x4, %o1 1: cmp %o2, 0 be,pt %XCC, out nop ba,pt %xcc, small_copy_unaligned nop small_copy: /* 0 < len <= 16 */ andcc %o3, 0x3, %g0 bne,pn %XCC, small_copy_unaligned sub %o0, %o1, %o3 small_copy_aligned: subcc %o2, 4, %o2 lduw [%o1], %g1 stw %g1, [%o1 + %o3] bgu,pt %XCC, small_copy_aligned add %o1, 4, %o1 out: retl mov %g5, %o0 .align 32 small_copy_unaligned: subcc %o2, 1, %o2 ldub [%o1], %g1 stb %g1, [%o1 + %o3] bgu,pt %XCC, small_copy_unaligned add %o1, 1, %o1 retl mov %g5, %o0 END(memcpy) #define RMOVE_BIGCHUNK(src, dst, offset, t0, t1, t2, t3) \ ldx [%src - offset - 0x20], %t0; \ ldx [%src - offset - 0x18], %t1; \ ldx [%src - offset - 0x10], %t2; \ ldx [%src - offset - 0x08], %t3; \ stw %t0, [%dst - offset - 0x1c]; \ srlx %t0, 32, %t0; \ stw %t0, [%dst - offset - 0x20]; \ stw %t1, [%dst - offset - 0x14]; \ srlx %t1, 32, %t1; \ stw %t1, [%dst - offset - 0x18]; \ stw %t2, [%dst - offset - 0x0c]; \ srlx %t2, 32, %t2; \ stw %t2, [%dst - offset - 0x10]; \ stw %t3, [%dst - offset - 0x04]; \ srlx %t3, 32, %t3; \ stw %t3, [%dst - offset - 0x08]; #define RMOVE_BIGALIGNCHUNK(src, dst, offset, t0, t1, t2, t3) \ ldx [%src - offset - 0x20], %t0; \ ldx [%src - offset - 0x18], %t1; \ ldx [%src - offset - 0x10], %t2; \ ldx [%src - offset - 0x08], %t3; \ stx %t0, [%dst - offset - 0x20]; \ stx %t1, [%dst - offset - 0x18]; \ stx %t2, [%dst - offset - 0x10]; \ stx %t3, [%dst - offset - 0x08]; \ ldx [%src - offset - 0x40], %t0; \ ldx [%src - offset - 0x38], %t1; \ ldx [%src - offset - 0x30], %t2; \ ldx [%src - offset - 0x28], %t3; \ stx %t0, [%dst - offset - 0x40]; \ stx %t1, [%dst - offset - 0x38]; \ stx %t2, [%dst - offset - 0x30]; \ stx %t3, [%dst - offset - 0x28]; #define RMOVE_LASTCHUNK(src, dst, offset, t0, t1, t2, t3) \ ldx [%src + offset + 0x00], %t0; \ ldx [%src + offset + 0x08], %t1; \ stw %t0, [%dst + offset + 0x04]; \ srlx %t0, 32, %t2; \ stw %t2, [%dst + offset + 0x00]; \ stw %t1, [%dst + offset + 0x0c]; \ srlx %t1, 32, %t3; \ stw %t3, [%dst + offset + 0x08]; #define RMOVE_LASTALIGNCHUNK(src, dst, offset, t0, t1) \ ldx [%src + offset + 0x00], %t0; \ ldx [%src + offset + 0x08], %t1; \ stx %t0, [%dst + offset + 0x00]; \ stx %t1, [%dst + offset + 0x08]; .align 32 228: andcc %o2, 1, %g0 /* IEU1 Group */ be,pt %icc, 2f+4 /* CTI */ 1: ldub [%o1 - 1], %o5 /* LOAD Group */ sub %o1, 1, %o1 /* IEU0 */ sub %o0, 1, %o0 /* IEU1 */ subcc %o2, 1, %o2 /* IEU1 Group */ be,pn %xcc, 229f /* CTI */ stb %o5, [%o0] /* Store */ 2: ldub [%o1 - 1], %o5 /* LOAD Group */ sub %o0, 2, %o0 /* IEU0 */ ldub [%o1 - 2], %g5 /* LOAD Group */ sub %o1, 2, %o1 /* IEU0 */ subcc %o2, 2, %o2 /* IEU1 Group */ stb %o5, [%o0 + 1] /* Store */ bne,pt %xcc, 2b /* CTI */ stb %g5, [%o0] /* Store */ 229: retl mov %g4, %o0 .align 32 ENTRY(memmove) mov %o0, %g5 #ifndef USE_BPR srl %o2, 0, %o2 /* IEU1 Group */ #endif brz,pn %o2, out /* CTI Group */ sub %o0, %o1, %o4 /* IEU0 */ cmp %o4, %o2 /* IEU1 Group */ bgeu,pt %XCC, 218b /* CTI */ mov %o0, %g4 /* IEU0 */ add %o0, %o2, %o0 /* IEU0 Group */ 220: add %o1, %o2, %o1 /* IEU1 */ cmp %o2, 15 /* IEU1 Group */ bleu,pn %xcc, 228b /* CTI */ andcc %o0, 7, %g2 /* IEU1 Group */ sub %o0, %o1, %g5 /* IEU0 */ andcc %g5, 3, %o5 /* IEU1 Group */ bne,pn %xcc, 232f /* CTI */ andcc %o1, 3, %g0 /* IEU1 Group */ be,a,pt %xcc, 236f /* CTI */ andcc %o1, 4, %g0 /* IEU1 Group */ andcc %o1, 1, %g0 /* IEU1 Group */ be,pn %xcc, 4f /* CTI */ andcc %o1, 2, %g0 /* IEU1 Group */ ldub [%o1 - 1], %g2 /* Load Group */ sub %o1, 1, %o1 /* IEU0 */ sub %o0, 1, %o0 /* IEU1 */ sub %o2, 1, %o2 /* IEU0 Group */ be,pn %xcc, 5f /* CTI Group */ stb %g2, [%o0] /* Store */ 4: lduh [%o1 - 2], %g2 /* Load Group */ sub %o1, 2, %o1 /* IEU0 */ sub %o0, 2, %o0 /* IEU1 */ sub %o2, 2, %o2 /* IEU0 */ sth %g2, [%o0] /* Store Group + bubble */ 5: andcc %o1, 4, %g0 /* IEU1 */ 236: be,a,pn %xcc, 2f /* CTI */ andcc %o2, -128, %g6 /* IEU1 Group */ lduw [%o1 - 4], %g5 /* Load Group */ sub %o1, 4, %o1 /* IEU0 */ sub %o0, 4, %o0 /* IEU1 */ sub %o2, 4, %o2 /* IEU0 Group */ stw %g5, [%o0] /* Store */ andcc %o2, -128, %g6 /* IEU1 Group */ 2: be,pn %xcc, 235f /* CTI */ andcc %o0, 4, %g0 /* IEU1 Group */ be,pn %xcc, 282f + 4 /* CTI Group */ 5: RMOVE_BIGCHUNK(o1, o0, 0x00, g1, g3, g5, o5) RMOVE_BIGCHUNK(o1, o0, 0x20, g1, g3, g5, o5) RMOVE_BIGCHUNK(o1, o0, 0x40, g1, g3, g5, o5) RMOVE_BIGCHUNK(o1, o0, 0x60, g1, g3, g5, o5) subcc %g6, 128, %g6 /* IEU1 Group */ sub %o1, 128, %o1 /* IEU0 */ bne,pt %xcc, 5b /* CTI */ sub %o0, 128, %o0 /* IEU0 Group */ 235: andcc %o2, 0x70, %g6 /* IEU1 Group */ 41: be,pn %xcc, 280f /* CTI */ andcc %o2, 8, %g0 /* IEU1 Group */ /* Clk1 8-( */ /* Clk2 8-( */ /* Clk3 8-( */ /* Clk4 8-( */ 279: rd %pc, %o5 /* PDU Group */ sll %g6, 1, %g5 /* IEU0 Group */ sub %o1, %g6, %o1 /* IEU1 */ sub %o5, %g5, %o5 /* IEU0 Group */ jmpl %o5 + %lo(280f - 279b), %g0 /* CTI Group brk forced*/ sub %o0, %g6, %o0 /* IEU0 Group */ RMOVE_LASTCHUNK(o1, o0, 0x60, g2, g3, g5, o5) RMOVE_LASTCHUNK(o1, o0, 0x50, g2, g3, g5, o5) RMOVE_LASTCHUNK(o1, o0, 0x40, g2, g3, g5, o5) RMOVE_LASTCHUNK(o1, o0, 0x30, g2, g3, g5, o5) RMOVE_LASTCHUNK(o1, o0, 0x20, g2, g3, g5, o5) RMOVE_LASTCHUNK(o1, o0, 0x10, g2, g3, g5, o5) RMOVE_LASTCHUNK(o1, o0, 0x00, g2, g3, g5, o5) 280: be,pt %xcc, 281f /* CTI */ andcc %o2, 4, %g0 /* IEU1 */ ldx [%o1 - 8], %g2 /* Load Group */ sub %o0, 8, %o0 /* IEU0 */ stw %g2, [%o0 + 4] /* Store Group */ sub %o1, 8, %o1 /* IEU1 */ srlx %g2, 32, %g2 /* IEU0 Group */ stw %g2, [%o0] /* Store */ 281: be,pt %xcc, 1f /* CTI */ andcc %o2, 2, %g0 /* IEU1 Group */ lduw [%o1 - 4], %g2 /* Load Group */ sub %o1, 4, %o1 /* IEU0 */ stw %g2, [%o0 - 4] /* Store Group */ sub %o0, 4, %o0 /* IEU0 */ 1: be,pt %xcc, 1f /* CTI */ andcc %o2, 1, %g0 /* IEU1 Group */ lduh [%o1 - 2], %g2 /* Load Group */ sub %o1, 2, %o1 /* IEU0 */ sth %g2, [%o0 - 2] /* Store Group */ sub %o0, 2, %o0 /* IEU0 */ 1: be,pt %xcc, 211f /* CTI */ nop /* IEU1 */ ldub [%o1 - 1], %g2 /* Load Group */ stb %g2, [%o0 - 1] /* Store Group + bubble */ 211: retl mov %g4, %o0 282: RMOVE_BIGALIGNCHUNK(o1, o0, 0x00, g1, g3, g5, o5) RMOVE_BIGALIGNCHUNK(o1, o0, 0x40, g1, g3, g5, o5) subcc %g6, 128, %g6 /* IEU1 Group */ sub %o1, 128, %o1 /* IEU0 */ bne,pt %xcc, 282b /* CTI */ sub %o0, 128, %o0 /* IEU0 Group */ andcc %o2, 0x70, %g6 /* IEU1 */ be,pn %xcc, 284f /* CTI */ andcc %o2, 8, %g0 /* IEU1 Group */ /* Clk1 8-( */ /* Clk2 8-( */ /* Clk3 8-( */ /* Clk4 8-( */ 283: rd %pc, %o5 /* PDU Group */ sub %o1, %g6, %o1 /* IEU0 Group */ sub %o5, %g6, %o5 /* IEU1 */ jmpl %o5 + %lo(284f - 283b), %g0 /* CTI Group brk forced*/ sub %o0, %g6, %o0 /* IEU0 Group */ RMOVE_LASTALIGNCHUNK(o1, o0, 0x60, g2, g3) RMOVE_LASTALIGNCHUNK(o1, o0, 0x50, g2, g3) RMOVE_LASTALIGNCHUNK(o1, o0, 0x40, g2, g3) RMOVE_LASTALIGNCHUNK(o1, o0, 0x30, g2, g3) RMOVE_LASTALIGNCHUNK(o1, o0, 0x20, g2, g3) RMOVE_LASTALIGNCHUNK(o1, o0, 0x10, g2, g3) RMOVE_LASTALIGNCHUNK(o1, o0, 0x00, g2, g3) 284: be,pt %xcc, 285f /* CTI Group */ andcc %o2, 4, %g0 /* IEU1 */ ldx [%o1 - 8], %g2 /* Load Group */ sub %o0, 8, %o0 /* IEU0 */ sub %o1, 8, %o1 /* IEU0 Group */ stx %g2, [%o0] /* Store */ 285: be,pt %xcc, 1f /* CTI */ andcc %o2, 2, %g0 /* IEU1 Group */ lduw [%o1 - 4], %g2 /* Load Group */ sub %o0, 4, %o0 /* IEU0 */ sub %o1, 4, %o1 /* IEU0 Group */ stw %g2, [%o0] /* Store */ 1: be,pt %xcc, 1f /* CTI */ andcc %o2, 1, %g0 /* IEU1 Group */ lduh [%o1 - 2], %g2 /* Load Group */ sub %o0, 2, %o0 /* IEU0 */ sub %o1, 2, %o1 /* IEU0 Group */ sth %g2, [%o0] /* Store */ 1: be,pt %xcc, 1f /* CTI */ nop /* IEU0 Group */ ldub [%o1 - 1], %g2 /* Load Group */ stb %g2, [%o0 - 1] /* Store Group + bubble */ 1: retl mov %g4, %o0 232: brz,pt %g2, 2f /* CTI Group */ sub %o2, %g2, %o2 /* IEU0 Group */ 1: ldub [%o1 - 1], %g5 /* Load Group */ sub %o1, 1, %o1 /* IEU0 */ sub %o0, 1, %o0 /* IEU1 */ subcc %g2, 1, %g2 /* IEU1 Group */ bne,pt %xcc, 1b /* CTI */ stb %g5, [%o0] /* Store */ 2: andn %o2, 7, %g5 /* IEU0 Group */ and %o2, 7, %o2 /* IEU1 */ fmovd %f0, %f2 /* FPU */ alignaddr %o1, %g0, %g1 /* GRU Group */ ldd [%g1], %f4 /* Load Group */ 1: ldd [%g1 - 8], %f6 /* Load Group */ sub %g1, 8, %g1 /* IEU0 Group */ subcc %g5, 8, %g5 /* IEU1 */ faligndata %f6, %f4, %f0 /* GRU Group */ std %f0, [%o0 - 8] /* Store */ sub %o1, 8, %o1 /* IEU0 Group */ be,pn %xcc, 233f /* CTI */ sub %o0, 8, %o0 /* IEU1 */ ldd [%g1 - 8], %f4 /* Load Group */ sub %g1, 8, %g1 /* IEU0 */ subcc %g5, 8, %g5 /* IEU1 */ faligndata %f4, %f6, %f0 /* GRU Group */ std %f0, [%o0 - 8] /* Store */ sub %o1, 8, %o1 /* IEU0 */ bne,pn %xcc, 1b /* CTI Group */ sub %o0, 8, %o0 /* IEU0 */ 233: brz,pn %o2, 234f /* CTI Group */ nop /* IEU0 */ 237: ldub [%o1 - 1], %g5 /* LOAD */ sub %o1, 1, %o1 /* IEU0 */ sub %o0, 1, %o0 /* IEU1 */ subcc %o2, 1, %o2 /* IEU1 */ bne,pt %xcc, 237b /* CTI */ stb %g5, [%o0] /* Store Group */ 234: wr %g0, FPRS_FEF, %fprs retl mov %g4, %o0 END(memmove) #ifdef USE_BPR weak_alias (memcpy, __align_cpy_1) weak_alias (memcpy, __align_cpy_2) weak_alias (memcpy, __align_cpy_4) weak_alias (memcpy, __align_cpy_8) weak_alias (memcpy, __align_cpy_16) #endif libc_hidden_builtin_def (memcpy) libc_hidden_builtin_def (memmove)