Cinelerra: hvirtual/quicktime/ffmpeg/libavcodec/ppc/dsputil

00001 /*
00002  * Copyright (c) 2003-2004 Romain Dolbeau <romain@dolbeau.org>
00003  *
00004  * This library is free software; you can redistribute it and/or
00005  * modify it under the terms of the GNU Lesser General Public
00006  * License as published by the Free Software Foundation; either
00007  * version 2 of the License, or (at your option) any later version.
00008  *
00009  * This library is distributed in the hope that it will be useful,
00010  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00011  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00012  * Lesser General Public License for more details.
00013  *
00014  * You should have received a copy of the GNU Lesser General Public
00015  * License along with this library; if not, write to the Free Software
00016  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
00017  */
00018 
00019 #ifndef _DSPUTIL_PPC_
00020 #define _DSPUTIL_PPC_
00021 
00022 #ifdef CONFIG_DARWIN
00023 /* The Apple assembler shipped w/ gcc-3.3 knows about DCBZL, previous assemblers don't
00024    We assume here that the Darwin GCC is from Apple.... */
00025 #if (__GNUC__ * 100 + __GNUC_MINOR__ < 303)
00026 #define NO_DCBZL
00027 #endif
00028 #else /* CONFIG_DARWIN */
00029 /* I don't think any non-Apple assembler knows about DCBZL */
00030 #define NO_DCBZL
00031 #endif /* CONFIG_DARWIN */
00032 
00033 #ifdef POWERPC_PERFORMANCE_REPORT
00034 void powerpc_display_perf_report(void);
00035 /* the 604* have 2, the G3* have 4, the G4s have 6,
00036    and the G5 are completely different (they MUST use
00037    POWERPC_MODE_64BITS, and let's hope all future 64 bis PPC
00038    will use the same PMCs... */
00039 #define POWERPC_NUM_PMC_ENABLED 6
00040 /* if you add to the enum below, also add to the perfname array
00041    in dsputil_ppc.c */
00042 enum powerpc_perf_index {
00043   altivec_fft_num = 0,
00044   altivec_gmc1_num,
00045   altivec_dct_unquantize_h263_num,
00046   altivec_fdct,
00047   altivec_idct_add_num,
00048   altivec_idct_put_num,
00049   altivec_put_pixels16_num,
00050   altivec_avg_pixels16_num,
00051   altivec_avg_pixels8_num,
00052   altivec_put_pixels8_xy2_num,
00053   altivec_put_no_rnd_pixels8_xy2_num,
00054   altivec_put_pixels16_xy2_num,
00055   altivec_put_no_rnd_pixels16_xy2_num,
00056   altivec_hadamard8_diff8x8_num,
00057   altivec_hadamard8_diff16_num,
00058   altivec_avg_pixels8_xy2_num,
00059   powerpc_clear_blocks_dcbz32,
00060   powerpc_clear_blocks_dcbz128,
00061   altivec_put_h264_chroma_mc8_num,
00062   altivec_avg_h264_chroma_mc8_num,
00063   altivec_put_h264_qpel16_h_lowpass_num,
00064   altivec_avg_h264_qpel16_h_lowpass_num,
00065   altivec_put_h264_qpel16_v_lowpass_num,
00066   altivec_avg_h264_qpel16_v_lowpass_num,
00067   altivec_put_h264_qpel16_hv_lowpass_num,
00068   altivec_avg_h264_qpel16_hv_lowpass_num,
00069   powerpc_perf_total
00070 };
00071 enum powerpc_data_index {
00072   powerpc_data_min = 0,
00073   powerpc_data_max,
00074   powerpc_data_sum,
00075   powerpc_data_num,
00076   powerpc_data_total
00077 };
00078 extern unsigned long long perfdata[POWERPC_NUM_PMC_ENABLED][powerpc_perf_total][powerpc_data_total];
00079 
00080 #ifndef POWERPC_MODE_64BITS
00081 #define POWERP_PMC_DATATYPE unsigned long
00082 #define POWERPC_GET_PMC1(a) asm volatile("mfspr %0, 937" : "=r" (a))
00083 #define POWERPC_GET_PMC2(a) asm volatile("mfspr %0, 938" : "=r" (a))
00084 #if (POWERPC_NUM_PMC_ENABLED > 2)
00085 #define POWERPC_GET_PMC3(a) asm volatile("mfspr %0, 941" : "=r" (a))
00086 #define POWERPC_GET_PMC4(a) asm volatile("mfspr %0, 942" : "=r" (a))
00087 #else
00088 #define POWERPC_GET_PMC3(a) do {} while (0)
00089 #define POWERPC_GET_PMC4(a) do {} while (0)
00090 #endif
00091 #if (POWERPC_NUM_PMC_ENABLED > 4)
00092 #define POWERPC_GET_PMC5(a) asm volatile("mfspr %0, 929" : "=r" (a))
00093 #define POWERPC_GET_PMC6(a) asm volatile("mfspr %0, 930" : "=r" (a))
00094 #else
00095 #define POWERPC_GET_PMC5(a) do {} while (0)
00096 #define POWERPC_GET_PMC6(a) do {} while (0)
00097 #endif
00098 #else /* POWERPC_MODE_64BITS */
00099 #define POWERP_PMC_DATATYPE unsigned long long
00100 #define POWERPC_GET_PMC1(a) asm volatile("mfspr %0, 771" : "=r" (a))
00101 #define POWERPC_GET_PMC2(a) asm volatile("mfspr %0, 772" : "=r" (a))
00102 #if (POWERPC_NUM_PMC_ENABLED > 2)
00103 #define POWERPC_GET_PMC3(a) asm volatile("mfspr %0, 773" : "=r" (a))
00104 #define POWERPC_GET_PMC4(a) asm volatile("mfspr %0, 774" : "=r" (a))
00105 #else
00106 #define POWERPC_GET_PMC3(a) do {} while (0)
00107 #define POWERPC_GET_PMC4(a) do {} while (0)
00108 #endif
00109 #if (POWERPC_NUM_PMC_ENABLED > 4)
00110 #define POWERPC_GET_PMC5(a) asm volatile("mfspr %0, 775" : "=r" (a))
00111 #define POWERPC_GET_PMC6(a) asm volatile("mfspr %0, 776" : "=r" (a))
00112 #else
00113 #define POWERPC_GET_PMC5(a) do {} while (0)
00114 #define POWERPC_GET_PMC6(a) do {} while (0)
00115 #endif
00116 #endif /* POWERPC_MODE_64BITS */
00117 #define POWERPC_PERF_DECLARE(a, cond)                           \
00118   POWERP_PMC_DATATYPE                                           \
00119     pmc_start[POWERPC_NUM_PMC_ENABLED],                         \
00120     pmc_stop[POWERPC_NUM_PMC_ENABLED],                          \
00121     pmc_loop_index;
00122 #define POWERPC_PERF_START_COUNT(a, cond) do { \
00123   POWERPC_GET_PMC6(pmc_start[5]); \
00124   POWERPC_GET_PMC5(pmc_start[4]); \
00125   POWERPC_GET_PMC4(pmc_start[3]); \
00126   POWERPC_GET_PMC3(pmc_start[2]); \
00127   POWERPC_GET_PMC2(pmc_start[1]); \
00128   POWERPC_GET_PMC1(pmc_start[0]); \
00129   } while (0)
00130 #define POWERPC_PERF_STOP_COUNT(a, cond) do { \
00131   POWERPC_GET_PMC1(pmc_stop[0]); \
00132   POWERPC_GET_PMC2(pmc_stop[1]); \
00133   POWERPC_GET_PMC3(pmc_stop[2]); \
00134   POWERPC_GET_PMC4(pmc_stop[3]); \
00135   POWERPC_GET_PMC5(pmc_stop[4]); \
00136   POWERPC_GET_PMC6(pmc_stop[5]); \
00137   if (cond)                       \
00138   {                               \
00139     for(pmc_loop_index = 0;       \
00140         pmc_loop_index < POWERPC_NUM_PMC_ENABLED; \
00141         pmc_loop_index++)         \
00142     {                             \
00143       if (pmc_stop[pmc_loop_index] >= pmc_start[pmc_loop_index])  \
00144         {                                                         \
00145         POWERP_PMC_DATATYPE diff =                                \
00146           pmc_stop[pmc_loop_index] - pmc_start[pmc_loop_index];   \
00147         if (diff < perfdata[pmc_loop_index][a][powerpc_data_min]) \
00148           perfdata[pmc_loop_index][a][powerpc_data_min] = diff;   \
00149         if (diff > perfdata[pmc_loop_index][a][powerpc_data_max]) \
00150           perfdata[pmc_loop_index][a][powerpc_data_max] = diff;   \
00151         perfdata[pmc_loop_index][a][powerpc_data_sum] += diff;    \
00152         perfdata[pmc_loop_index][a][powerpc_data_num] ++;         \
00153       }                           \
00154     }                             \
00155   }                               \
00156 } while (0)
00157 #else /* POWERPC_PERFORMANCE_REPORT */
00158 // those are needed to avoid empty statements.
00159 #define POWERPC_PERF_DECLARE(a, cond)        int altivec_placeholder __attribute__ ((unused))
00160 #define POWERPC_PERF_START_COUNT(a, cond)    do {} while (0)
00161 #define POWERPC_PERF_STOP_COUNT(a, cond)     do {} while (0)
00162 #endif /* POWERPC_PERFORMANCE_REPORT */
00163 
00164 #endif /*  _DSPUTIL_PPC_ */
hvirtual/quicktime/ffmpeg/libavcodec/ppc/dsputil_ppc.h
