Commit | Line | Data |
---|---|---|
35e5fb06 RD |
1 | /* |
2 | * Copyright (c) 2003 Romain Dolbeau <romain@dolbeau.org> | |
3 | * | |
4 | * This library is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU Lesser General Public | |
6 | * License as published by the Free Software Foundation; either | |
7 | * version 2 of the License, or (at your option) any later version. | |
8 | * | |
9 | * This library is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
12 | * Lesser General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU Lesser General Public | |
15 | * License along with this library; if not, write to the Free Software | |
16 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
17 | */ | |
18 | ||
19 | #ifndef _DSPUTIL_PPC_ | |
20 | #define _DSPUTIL_PPC_ | |
21 | ||
22 | #ifdef POWERPC_TBL_PERFORMANCE_REPORT | |
23 | void powerpc_display_perf_report(void); | |
24 | /* if you add to the enum below, also add to the perfname array | |
25 | in dsputil_ppc.c */ | |
26 | enum powerpc_perf_index { | |
27 | altivec_fft_num = 0, | |
28 | altivec_gmc1_num, | |
29 | altivec_dct_unquantize_h263_num, | |
30 | altivec_idct_add_num, | |
31 | altivec_idct_put_num, | |
35e5fb06 RD |
32 | altivec_put_pixels16_num, |
33 | altivec_avg_pixels16_num, | |
34 | altivec_avg_pixels8_num, | |
35 | altivec_put_pixels8_xy2_num, | |
fe50f385 RD |
36 | altivec_put_no_rnd_pixels8_xy2_num, |
37 | altivec_put_pixels16_xy2_num, | |
38 | altivec_put_no_rnd_pixels16_xy2_num, | |
35e5fb06 | 39 | powerpc_clear_blocks_dcbz32, |
a4adb608 | 40 | powerpc_clear_blocks_dcbz128, |
35e5fb06 RD |
41 | powerpc_perf_total |
42 | }; | |
43 | enum powerpc_data_index { | |
44 | powerpc_data_min = 0, | |
45 | powerpc_data_max, | |
46 | powerpc_data_sum, | |
47 | powerpc_data_num, | |
48 | powerpc_data_total | |
49 | }; | |
50 | extern unsigned long long perfdata[powerpc_perf_total][powerpc_data_total]; | |
51 | #ifdef POWERPC_PERF_USE_PMC | |
52 | extern unsigned long long perfdata_miss[powerpc_perf_total][powerpc_data_total]; | |
53 | #endif | |
54 | ||
55 | #ifndef POWERPC_PERF_USE_PMC | |
56 | #define POWERPC_GET_CYCLES(a) asm volatile("mftb %0" : "=r" (a)) | |
57 | #define POWERPC_TBL_DECLARE(a, cond) register unsigned long tbl_start, tbl_stop | |
58 | #define POWERPC_TBL_START_COUNT(a, cond) do { POWERPC_GET_CYCLES(tbl_start); } while (0) | |
59 | #define POWERPC_TBL_STOP_COUNT(a, cond) do { \ | |
60 | POWERPC_GET_CYCLES(tbl_stop); \ | |
61 | if (tbl_stop > tbl_start) \ | |
62 | { \ | |
63 | unsigned long diff = tbl_stop - tbl_start; \ | |
64 | if (cond) \ | |
65 | { \ | |
66 | if (diff < perfdata[a][powerpc_data_min]) \ | |
67 | perfdata[a][powerpc_data_min] = diff; \ | |
68 | if (diff > perfdata[a][powerpc_data_max]) \ | |
69 | perfdata[a][powerpc_data_max] = diff; \ | |
70 | perfdata[a][powerpc_data_sum] += diff; \ | |
71 | perfdata[a][powerpc_data_num] ++; \ | |
72 | } \ | |
73 | } \ | |
74 | } while (0) | |
75 | ||
76 | #else /* POWERPC_PERF_USE_PMC */ | |
77 | #define POWERPC_GET_CYCLES(a) asm volatile("mfspr %0, 937" : "=r" (a)) | |
78 | #define POWERPC_GET_MISS(a) asm volatile("mfspr %0, 938" : "=r" (a)) | |
475b46da | 79 | #define POWERPC_TBL_DECLARE(a, cond) register unsigned long cycles_start, cycles_stop, miss_start, miss_stop |
35e5fb06 RD |
80 | #define POWERPC_TBL_START_COUNT(a, cond) do { POWERPC_GET_MISS(miss_start); POWERPC_GET_CYCLES(cycles_start); } while (0) |
81 | #define POWERPC_TBL_STOP_COUNT(a, cond) do { \ | |
82 | POWERPC_GET_CYCLES(cycles_stop); \ | |
83 | POWERPC_GET_MISS(miss_stop); \ | |
84 | if (cycles_stop >= cycles_start) \ | |
85 | { \ | |
86 | unsigned long diff = \ | |
87 | cycles_stop - cycles_start; \ | |
88 | if (cond) \ | |
89 | { \ | |
90 | if (diff < perfdata[a][powerpc_data_min]) \ | |
91 | perfdata[a][powerpc_data_min] = diff; \ | |
92 | if (diff > perfdata[a][powerpc_data_max]) \ | |
93 | perfdata[a][powerpc_data_max] = diff; \ | |
94 | perfdata[a][powerpc_data_sum] += diff; \ | |
95 | perfdata[a][powerpc_data_num] ++; \ | |
96 | } \ | |
97 | } \ | |
98 | if (miss_stop >= miss_start) \ | |
99 | { \ | |
100 | unsigned long diff = \ | |
101 | miss_stop - miss_start; \ | |
102 | if (cond) \ | |
103 | { \ | |
104 | if (diff < perfdata_miss[a][powerpc_data_min]) \ | |
105 | perfdata_miss[a][powerpc_data_min] = diff; \ | |
106 | if (diff > perfdata_miss[a][powerpc_data_max]) \ | |
107 | perfdata_miss[a][powerpc_data_max] = diff; \ | |
108 | perfdata_miss[a][powerpc_data_sum] += diff; \ | |
109 | perfdata_miss[a][powerpc_data_num] ++; \ | |
110 | } \ | |
111 | } \ | |
112 | } while (0) | |
113 | ||
114 | #endif /* POWERPC_PERF_USE_PMC */ | |
115 | ||
116 | ||
117 | #else /* POWERPC_TBL_PERFORMANCE_REPORT */ | |
475b46da MN |
118 | // those are needed to avoid empty statements. |
119 | #define POWERPC_TBL_DECLARE(a, cond) int altivec_placeholder __attribute__ ((unused)) | |
120 | #define POWERPC_TBL_START_COUNT(a, cond) do {} while (0) | |
121 | #define POWERPC_TBL_STOP_COUNT(a, cond) do {} while (0) | |
35e5fb06 RD |
122 | #endif /* POWERPC_TBL_PERFORMANCE_REPORT */ |
123 | ||
124 | #endif /* _DSPUTIL_PPC_ */ |