arch/ia64/lib/clear_page.S

   1 /*
   2  * Copyright (C) 1999-2002 Hewlett-Packard Co
   3  *      Stephane Eranian <eranian@hpl.hp.com>
   4  *      David Mosberger-Tang <davidm@hpl.hp.com>
   5  * Copyright (C) 2002 Ken Chen <kenneth.w.chen@intel.com>
   6  *
   7  * 1/06/01 davidm       Tuned for Itanium.
   8  * 2/12/02 kchen        Tuned for both Itanium and McKinley
   9  * 3/08/02 davidm       Some more tweaking
  10  */
  11 #include <linux/config.h>
  12
  13 #include <asm/asmmacro.h>
  14 #include <asm/page.h>
  15
  16 #ifdef CONFIG_ITANIUM
  17 # define L3_LINE_SIZE   64      // Itanium L3 line size
  18 # define PREFETCH_LINES 9       // magic number
  19 #else
  20 # define L3_LINE_SIZE   128     // McKinley L3 line size
  21 # define PREFETCH_LINES 12      // magic number
  22 #endif
  23
  24 #define saved_lc        r2
  25 #define dst_fetch       r3
  26 #define dst1            r8
  27 #define dst2            r9
  28 #define dst3            r10
  29 #define dst4            r11
  30
  31 #define dst_last        r31
  32
  33 GLOBAL_ENTRY(clear_page)
  34         .prologue
  35         .regstk 1,0,0,0
  36         mov r16 = PAGE_SIZE/L3_LINE_SIZE-1      // main loop count, -1=repeat/until
  37         .save ar.lc, saved_lc
  38         mov saved_lc = ar.lc
  39
  40         .body
  41         mov ar.lc = (PREFETCH_LINES - 1)
  42         mov dst_fetch = in0
  43         adds dst1 = 16, in0
  44         adds dst2 = 32, in0
  45         ;;
  46 .fetch: stf.spill.nta [dst_fetch] = f0, L3_LINE_SIZE
  47         adds dst3 = 48, in0             // executing this multiple times is harmless
  48         br.cloop.sptk.few .fetch
  49         ;;
  50         addl dst_last = (PAGE_SIZE - PREFETCH_LINES*L3_LINE_SIZE), dst_fetch
  51         mov ar.lc = r16                 // one L3 line per iteration
  52         adds dst4 = 64, in0
  53         ;;
  54 #ifdef CONFIG_ITANIUM
  55         // Optimized for Itanium
  56 1:      stf.spill.nta [dst1] = f0, 64
  57         stf.spill.nta [dst2] = f0, 64
  58         cmp.lt p8,p0=dst_fetch, dst_last
  59         ;;
  60 #else
  61         // Optimized for McKinley
  62 1:      stf.spill.nta [dst1] = f0, 64
  63         stf.spill.nta [dst2] = f0, 64
  64         stf.spill.nta [dst3] = f0, 64
  65         stf.spill.nta [dst4] = f0, 128
  66         cmp.lt p8,p0=dst_fetch, dst_last
  67         ;;
  68         stf.spill.nta [dst1] = f0, 64
  69         stf.spill.nta [dst2] = f0, 64
  70 #endif
  71         stf.spill.nta [dst3] = f0, 64
  72 (p8)    stf.spill.nta [dst_fetch] = f0, L3_LINE_SIZE
  73         br.cloop.sptk.few 1b
  74         ;;
  75         mov ar.lc = saved_lc            // restore lc
  76         br.ret.sptk.many rp
  77 END(clear_page)