better hadamard8_diff16 in AltiVec, and more patch by (Romain Dolbeau <dolbeau at irisa dot fr>)

Originally committed as revision 3038 to svn://svn.ffmpeg.org/ffmpeg/trunk

configure

@@ -516,6 +516,7 @@ fi
 
 # Add processor-specific flags
 TUNECPU="generic"
+POWERPCMODE="32bits"
 if test $tune != "generic"; then
     case $tune in
 	601|ppc601|PowerPC601)
@@ -561,11 +562,12 @@ if test $tune != "generic"; then
 	    TUNECPU=ppc7400
 	;;
 	G5|g5|970|ppc970|PowerPC970|power4*|Power4*)
-	    CFLAGS="$CFLAGS -mcpu=970 -mtune=970 -mpowerpc64 -force_cpusubtype_ALL "
+	    CFLAGS="$CFLAGS -mcpu=970 -mtune=970 -mpowerpc-gfxopt -mpowerpc64"
 	    if test $altivec = "no"; then
 		echo "WARNING: tuning for PPC970 but altivec disabled !";
 	    fi
 	    TUNECPU=ppc970
+            POWERPCMODE="64bits"
 	;;
 	*)
 	echo "WARNING: unknown CPU "$tune", ignored"
@@ -1027,6 +1029,11 @@ elif test "$cpu" = "sparc64" ; then
 elif test "$cpu" = "powerpc" ; then
   echo "TARGET_ARCH_POWERPC=yes" >> config.mak
   echo "#define ARCH_POWERPC 1" >> $TMPH
+  if test $POWERPCMODE = "32bits"; then
+    echo "#define POWERPC_MODE_32BITS 1" >> $TMPH
+  else
+    echo "#define POWERPC_MODE_64BITS 1" >> $TMPH
+  fi
   if test "$powerpc_perf" = "yes"; then
     echo "#define POWERPC_PERFORMANCE_REPORT 1" >> $TMPH
   fi

doc/ffmpeg_powerpc_performance_evaluation_howto.txt

@@ -17,7 +17,7 @@ The firsts are always available, always active, but they're not very accurate :
 
 The PMC are much more useful : not only they can report cycle-accurate timing, but they can also be used to monitor many other parameters, such as the number of AltiVec stalls for every kind of instructions, or instruction cache misses. The downside is that not all processors support the PMC (all G3, all G4 and the 970 do support them), and they're inactive by default - you need to activate them with a dedicated tool. Also, the number of available PMC depend on the procesor : the various 604 have 2, the various 75x (aka. G3) have 4, anbd the various 74xx (aka G4) have 6.
 
-*WARNING*: The powerpc 970 is not very well documented, and it seems its PMC registers are 64bits wide. The current implementation in FFMpeg assume the register are 32bits wide, and will *not* work on a powerpc 970 (aka G5).
+*WARNING*: The powerpc 970 is not very well documented, and its PMC registers are 64bits wide. To properly notify the code, you *must* tune for the 970 (using --tune=970), or the code will assume 32bits registers.
 
 
 II - Enabling FFmpeg PowerPC performance support

libavcodec/ppc/dsputil_altivec.h

@@ -47,6 +47,7 @@ extern void put_no_rnd_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels
 extern void put_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, int line_size, int h);
 extern void put_no_rnd_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, int line_size, int h);
 extern int hadamard8_diff8x8_altivec(/*MpegEncContext*/ void *s, uint8_t *dst, uint8_t *src, int stride, int h);
+extern int hadamard8_diff16_altivec(/*MpegEncContext*/ void *s, uint8_t *dst, uint8_t *src, int stride, int h);
 
 extern void gmc1_altivec(uint8_t *dst, uint8_t *src, int stride, int h, int x16, int y16, int rounder);
 

libavcodec/ppc/dsputil_ppc.c

@@ -61,6 +61,7 @@ static unsigned char* perfname[] = {
   "put_pixels16_xy2_altivec",
   "put_no_rnd_pixels16_xy2_altivec",
   "hadamard8_diff8x8_altivec",
+  "hadamard8_diff16_altivec",
   "clear_blocks_dcbz32_ppc",
   "clear_blocks_dcbz128_ppc"
 };
@@ -226,12 +227,6 @@ long check_dcbzl_effect(void)
 }
 #endif
 
-#ifdef HAVE_ALTIVEC
-// can't put that in dsputil_altivec.c,
-// has WARPER8_16_SQ declare the function "static" ...
-WARPER8_16_SQ(hadamard8_diff8x8_altivec, hadamard8_diff16_altivec)
-#endif
-
 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx)
 {
     // Common optimizations whether Altivec is available or not

libavcodec/ppc/dsputil_ppc.h

@@ -51,6 +51,7 @@ enum powerpc_perf_index {
   altivec_put_pixels16_xy2_num,
   altivec_put_no_rnd_pixels16_xy2_num,
   altivec_hadamard8_diff8x8_num,
+  altivec_hadamard8_diff16_num,
   powerpc_clear_blocks_dcbz32,
   powerpc_clear_blocks_dcbz128,
   powerpc_perf_total
@@ -64,6 +65,8 @@ enum powerpc_data_index {
 };
 extern unsigned long long perfdata[POWERPC_NUM_PMC_ENABLED][powerpc_perf_total][powerpc_data_total];
 
+#ifndef POWERPC_MODE_64BITS
+#define POWERP_PMC_DATATYPE unsigned long
 #define POWERPC_GET_PMC1(a) asm volatile("mfspr %0, 937" : "=r" (a))
 #define POWERPC_GET_PMC2(a) asm volatile("mfspr %0, 938" : "=r" (a))
 #if (POWERPC_NUM_PMC_ENABLED > 2)
@@ -80,7 +83,30 @@ extern unsigned long long perfdata[POWERPC_NUM_PMC_ENABLED][powerpc_perf_total][
 #define POWERPC_GET_PMC5(a) do {} while (0)
 #define POWERPC_GET_PMC6(a) do {} while (0)
 #endif
-#define POWERPC_PERF_DECLARE(a, cond) unsigned long pmc_start[POWERPC_NUM_PMC_ENABLED], pmc_stop[POWERPC_NUM_PMC_ENABLED], pmc_loop_index;
+#else /* POWERPC_MODE_64BITS */
+#define POWERP_PMC_DATATYPE unsigned long long
+#define POWERPC_GET_PMC1(a) asm volatile("mfspr %0, 771" : "=r" (a))
+#define POWERPC_GET_PMC2(a) asm volatile("mfspr %0, 772" : "=r" (a))
+#if (POWERPC_NUM_PMC_ENABLED > 2)
+#define POWERPC_GET_PMC3(a) asm volatile("mfspr %0, 773" : "=r" (a))
+#define POWERPC_GET_PMC4(a) asm volatile("mfspr %0, 774" : "=r" (a))
+#else
+#define POWERPC_GET_PMC3(a) do {} while (0)
+#define POWERPC_GET_PMC4(a) do {} while (0)
+#endif
+#if (POWERPC_NUM_PMC_ENABLED > 4)
+#define POWERPC_GET_PMC5(a) asm volatile("mfspr %0, 775" : "=r" (a))
+#define POWERPC_GET_PMC6(a) asm volatile("mfspr %0, 776" : "=r" (a))
+#else
+#define POWERPC_GET_PMC5(a) do {} while (0)
+#define POWERPC_GET_PMC6(a) do {} while (0)
+#endif
+#endif /* POWERPC_MODE_64BITS */
+#define POWERPC_PERF_DECLARE(a, cond)				\
+  POWERP_PMC_DATATYPE						\
+    pmc_start[POWERPC_NUM_PMC_ENABLED],				\
+    pmc_stop[POWERPC_NUM_PMC_ENABLED],				\
+    pmc_loop_index;
 #define POWERPC_PERF_START_COUNT(a, cond) do { \
   POWERPC_GET_PMC6(pmc_start[5]); \
   POWERPC_GET_PMC5(pmc_start[4]); \
@@ -104,7 +130,7 @@ extern unsigned long long perfdata[POWERPC_NUM_PMC_ENABLED][powerpc_perf_total][
     {                             \
       if (pmc_stop[pmc_loop_index] >= pmc_start[pmc_loop_index])  \
 	{							  \
-        unsigned long diff =      \
+        POWERP_PMC_DATATYPE diff =				  \
           pmc_stop[pmc_loop_index] - pmc_start[pmc_loop_index];   \
         if (diff < perfdata[pmc_loop_index][a][powerpc_data_min]) \
           perfdata[pmc_loop_index][a][powerpc_data_min] = diff;   \