| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | */ |
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 11 | |
| 12 | #include <asm/asmmacro.h> |
| 13 | #include <asm/page.h> |
| 14 | |
| 15 | #ifdef CONFIG_ITANIUM |
| 16 | # define L3_LINE_SIZE 64 // Itanium L3 line size |
| 17 | # define PREFETCH_LINES 9 // magic number |
| 18 | #else |
| 19 | # define L3_LINE_SIZE 128 // McKinley L3 line size |
| 20 | # define PREFETCH_LINES 12 // magic number |
| 21 | #endif |
| 22 | |
| 23 | #define saved_lc r2 |
| 24 | #define dst_fetch r3 |
| 25 | #define dst1 r8 |
| 26 | #define dst2 r9 |
| 27 | #define dst3 r10 |
| 28 | #define dst4 r11 |
| 29 | |
| 30 | #define dst_last r31 |
| 31 | |
| 32 | GLOBAL_ENTRY(clear_page) |
| 33 | .prologue |
| 34 | .regstk 1,0,0,0 |
| 35 | mov r16 = PAGE_SIZE/L3_LINE_SIZE-1 // main loop count, -1=repeat/until |
| 36 | .save ar.lc, saved_lc |
| 37 | mov saved_lc = ar.lc |
| 38 | |
| 39 | .body |
| 40 | mov ar.lc = (PREFETCH_LINES - 1) |
| 41 | mov dst_fetch = in0 |
| 42 | adds dst1 = 16, in0 |
| 43 | adds dst2 = 32, in0 |
| 44 | ;; |
| 45 | .fetch: stf.spill.nta [dst_fetch] = f0, L3_LINE_SIZE |
| 46 | adds dst3 = 48, in0 // executing this multiple times is harmless |
| 47 | br.cloop.sptk.few .fetch |
| 48 | ;; |
| 49 | addl dst_last = (PAGE_SIZE - PREFETCH_LINES*L3_LINE_SIZE), dst_fetch |
| 50 | mov ar.lc = r16 // one L3 line per iteration |
| 51 | adds dst4 = 64, in0 |
| 52 | ;; |
| 53 | #ifdef CONFIG_ITANIUM |
| 54 | // Optimized for Itanium |
| 55 | 1: stf.spill.nta [dst1] = f0, 64 |
| 56 | stf.spill.nta [dst2] = f0, 64 |
| 57 | cmp.lt p8,p0=dst_fetch, dst_last |
| 58 | ;; |
| 59 | #else |
| 60 | // Optimized for McKinley |
| 61 | 1: stf.spill.nta [dst1] = f0, 64 |
| 62 | stf.spill.nta [dst2] = f0, 64 |
| 63 | stf.spill.nta [dst3] = f0, 64 |
| 64 | stf.spill.nta [dst4] = f0, 128 |
| 65 | cmp.lt p8,p0=dst_fetch, dst_last |
| 66 | ;; |
| 67 | stf.spill.nta [dst1] = f0, 64 |
| 68 | stf.spill.nta [dst2] = f0, 64 |
| 69 | #endif |
| 70 | stf.spill.nta [dst3] = f0, 64 |
| 71 | (p8) stf.spill.nta [dst_fetch] = f0, L3_LINE_SIZE |
| 72 | br.cloop.sptk.few 1b |
| 73 | ;; |
| 74 | mov ar.lc = saved_lc // restore lc |
| 75 | br.ret.sptk.many rp |
| 76 | END(clear_page) |