libavcodec/ppc/dsputil_ppc.h

   1 /*
   2  * Copyright (c) 2003-2004 Romain Dolbeau <romain@dolbeau.org>
   3  *
   4  * This file is part of FFmpeg.
   5  *
   6  * FFmpeg is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2.1 of the License, or (at your option) any later version.
  10  *
  11  * FFmpeg is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * Lesser General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU Lesser General Public
  17  * License along with FFmpeg; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  19  */
  20
  21 #ifndef AVCODEC_PPC_DSPUTIL_PPC_H
  22 #define AVCODEC_PPC_DSPUTIL_PPC_H
  23
  24 #ifdef CONFIG_POWERPC_PERF
  25 void powerpc_display_perf_report(void);
  26 /* the 604* have 2, the G3* have 4, the G4s have 6,
  27    and the G5 are completely different (they MUST use
  28    HAVE_PPC64, and let's hope all future 64 bis PPC
  29    will use the same PMCs... */
  30 #define POWERPC_NUM_PMC_ENABLED 6
  31 /* if you add to the enum below, also add to the perfname array
  32    in dsputil_ppc.c */
  33 enum powerpc_perf_index {
  34     altivec_fft_num = 0,
  35     altivec_gmc1_num,
  36     altivec_dct_unquantize_h263_num,
  37     altivec_fdct,
  38     altivec_idct_add_num,
  39     altivec_idct_put_num,
  40     altivec_put_pixels16_num,
  41     altivec_avg_pixels16_num,
  42     altivec_avg_pixels8_num,
  43     altivec_put_pixels8_xy2_num,
  44     altivec_put_no_rnd_pixels8_xy2_num,
  45     altivec_put_pixels16_xy2_num,
  46     altivec_put_no_rnd_pixels16_xy2_num,
  47     altivec_hadamard8_diff8x8_num,
  48     altivec_hadamard8_diff16_num,
  49     altivec_avg_pixels8_xy2_num,
  50     powerpc_clear_blocks_dcbz32,
  51     powerpc_clear_blocks_dcbz128,
  52     altivec_put_h264_chroma_mc8_num,
  53     altivec_avg_h264_chroma_mc8_num,
  54     altivec_put_h264_qpel16_h_lowpass_num,
  55     altivec_avg_h264_qpel16_h_lowpass_num,
  56     altivec_put_h264_qpel16_v_lowpass_num,
  57     altivec_avg_h264_qpel16_v_lowpass_num,
  58     altivec_put_h264_qpel16_hv_lowpass_num,
  59     altivec_avg_h264_qpel16_hv_lowpass_num,
  60     powerpc_perf_total
  61 };
  62 enum powerpc_data_index {
  63     powerpc_data_min = 0,
  64     powerpc_data_max,
  65     powerpc_data_sum,
  66     powerpc_data_num,
  67     powerpc_data_total
  68 };
  69 extern unsigned long long perfdata[POWERPC_NUM_PMC_ENABLED][powerpc_perf_total][powerpc_data_total];
  70
  71 #ifndef HAVE_PPC64
  72 #define POWERP_PMC_DATATYPE unsigned long
  73 #define POWERPC_GET_PMC1(a) __asm__ volatile("mfspr %0, 937" : "=r" (a))
  74 #define POWERPC_GET_PMC2(a) __asm__ volatile("mfspr %0, 938" : "=r" (a))
  75 #if (POWERPC_NUM_PMC_ENABLED > 2)
  76 #define POWERPC_GET_PMC3(a) __asm__ volatile("mfspr %0, 941" : "=r" (a))
  77 #define POWERPC_GET_PMC4(a) __asm__ volatile("mfspr %0, 942" : "=r" (a))
  78 #else
  79 #define POWERPC_GET_PMC3(a) do {} while (0)
  80 #define POWERPC_GET_PMC4(a) do {} while (0)
  81 #endif
  82 #if (POWERPC_NUM_PMC_ENABLED > 4)
  83 #define POWERPC_GET_PMC5(a) __asm__ volatile("mfspr %0, 929" : "=r" (a))
  84 #define POWERPC_GET_PMC6(a) __asm__ volatile("mfspr %0, 930" : "=r" (a))
  85 #else
  86 #define POWERPC_GET_PMC5(a) do {} while (0)
  87 #define POWERPC_GET_PMC6(a) do {} while (0)
  88 #endif
  89 #else /* HAVE_PPC64 */
  90 #define POWERP_PMC_DATATYPE unsigned long long
  91 #define POWERPC_GET_PMC1(a) __asm__ volatile("mfspr %0, 771" : "=r" (a))
  92 #define POWERPC_GET_PMC2(a) __asm__ volatile("mfspr %0, 772" : "=r" (a))
  93 #if (POWERPC_NUM_PMC_ENABLED > 2)
  94 #define POWERPC_GET_PMC3(a) __asm__ volatile("mfspr %0, 773" : "=r" (a))
  95 #define POWERPC_GET_PMC4(a) __asm__ volatile("mfspr %0, 774" : "=r" (a))
  96 #else
  97 #define POWERPC_GET_PMC3(a) do {} while (0)
  98 #define POWERPC_GET_PMC4(a) do {} while (0)
  99 #endif
 100 #if (POWERPC_NUM_PMC_ENABLED > 4)
 101 #define POWERPC_GET_PMC5(a) __asm__ volatile("mfspr %0, 775" : "=r" (a))
 102 #define POWERPC_GET_PMC6(a) __asm__ volatile("mfspr %0, 776" : "=r" (a))
 103 #else
 104 #define POWERPC_GET_PMC5(a) do {} while (0)
 105 #define POWERPC_GET_PMC6(a) do {} while (0)
 106 #endif
 107 #endif /* HAVE_PPC64 */
 108 #define POWERPC_PERF_DECLARE(a, cond)       \
 109     POWERP_PMC_DATATYPE                     \
 110         pmc_start[POWERPC_NUM_PMC_ENABLED], \
 111         pmc_stop[POWERPC_NUM_PMC_ENABLED],  \
 112         pmc_loop_index;
 113 #define POWERPC_PERF_START_COUNT(a, cond) do { \
 114     POWERPC_GET_PMC6(pmc_start[5]); \
 115     POWERPC_GET_PMC5(pmc_start[4]); \
 116     POWERPC_GET_PMC4(pmc_start[3]); \
 117     POWERPC_GET_PMC3(pmc_start[2]); \
 118     POWERPC_GET_PMC2(pmc_start[1]); \
 119     POWERPC_GET_PMC1(pmc_start[0]); \
 120     } while (0)
 121 #define POWERPC_PERF_STOP_COUNT(a, cond) do { \
 122     POWERPC_GET_PMC1(pmc_stop[0]);            \
 123     POWERPC_GET_PMC2(pmc_stop[1]);            \
 124     POWERPC_GET_PMC3(pmc_stop[2]);            \
 125     POWERPC_GET_PMC4(pmc_stop[3]);            \
 126     POWERPC_GET_PMC5(pmc_stop[4]);            \
 127     POWERPC_GET_PMC6(pmc_stop[5]);            \
 128     if (cond) {                               \
 129         for(pmc_loop_index = 0;               \
 130             pmc_loop_index < POWERPC_NUM_PMC_ENABLED; \
 131             pmc_loop_index++) {               \
 132             if (pmc_stop[pmc_loop_index] >= pmc_start[pmc_loop_index]) {  \
 133                 POWERP_PMC_DATATYPE diff =                                \
 134                   pmc_stop[pmc_loop_index] - pmc_start[pmc_loop_index];   \
 135                 if (diff < perfdata[pmc_loop_index][a][powerpc_data_min]) \
 136                     perfdata[pmc_loop_index][a][powerpc_data_min] = diff; \
 137                 if (diff > perfdata[pmc_loop_index][a][powerpc_data_max]) \
 138                     perfdata[pmc_loop_index][a][powerpc_data_max] = diff; \
 139                 perfdata[pmc_loop_index][a][powerpc_data_sum] += diff;    \
 140                 perfdata[pmc_loop_index][a][powerpc_data_num] ++;         \
 141             }                                 \
 142         }                                     \
 143     }                                         \
 144 } while (0)
 145 #else /* CONFIG_POWERPC_PERF */
 146 // those are needed to avoid empty statements.
 147 #define POWERPC_PERF_DECLARE(a, cond)        int altivec_placeholder __attribute__ ((unused))
 148 #define POWERPC_PERF_START_COUNT(a, cond)    do {} while (0)
 149 #define POWERPC_PERF_STOP_COUNT(a, cond)     do {} while (0)
 150 #endif /* CONFIG_POWERPC_PERF */
 151
 152 #endif /*  AVCODEC_PPC_DSPUTIL_PPC_H */