aboutsummaryrefslogtreecommitdiff
path: root/libavcodec/ppc
diff options
context:
space:
mode:
authorRomain Dolbeau2003-07-09 20:18:13 +0000
committerMichael Niedermayer2003-07-09 20:18:13 +0000
commite45a2872fafe631c14aee9f79d0963d68c4fc1fd (patch)
tree8c559f7a215414fb8207f7f524702d11d5b9dce0 /libavcodec/ppc
parent0273ceebbd01f9fd5238558e6151e0b9aa3305ab (diff)
1) remove TBL support in PPC performance. It's much more useful to use the
PMCs, and with Apple's CHUD it's fairly easy too. No reason to keep useless code around 2) make the PPC perf stuff a configure option 3) make put_pixels16_altivec a bit faster by unrolling the loop by 4 patch by (Romain Dolbeau <dolbeau at irisa dot fr>) Originally committed as revision 2022 to svn://svn.ffmpeg.org/ffmpeg/trunk
Diffstat (limited to 'libavcodec/ppc')
-rw-r--r--libavcodec/ppc/dsputil_altivec.c109
-rw-r--r--libavcodec/ppc/dsputil_ppc.c100
-rw-r--r--libavcodec/ppc/dsputil_ppc.h147
-rw-r--r--libavcodec/ppc/fft_altivec.c10
-rw-r--r--libavcodec/ppc/gmc_altivec.c10
-rw-r--r--libavcodec/ppc/idct_altivec.c20
-rw-r--r--libavcodec/ppc/mpegvideo_altivec.c6
7 files changed, 189 insertions, 213 deletions
diff --git a/libavcodec/ppc/dsputil_altivec.c b/libavcodec/ppc/dsputil_altivec.c
index 2c71d8e7bf..6354807844 100644
--- a/libavcodec/ppc/dsputil_altivec.c
+++ b/libavcodec/ppc/dsputil_altivec.c
@@ -655,11 +655,11 @@ void add_bytes_altivec(uint8_t *dst, uint8_t *src, int w) {
/* next one assumes that ((line_size % 16) == 0) */
void put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
-POWERPC_TBL_DECLARE(altivec_put_pixels16_num, 1);
+POWERPC_PERF_DECLARE(altivec_put_pixels16_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
int i;
-POWERPC_TBL_START_COUNT(altivec_put_pixels16_num, 1);
+POWERPC_PERF_START_COUNT(altivec_put_pixels16_num, 1);
for(i=0; i<h; i++) {
*((uint32_t*)(block )) = (((const struct unaligned_32 *) (pixels))->l);
@@ -670,15 +670,27 @@ POWERPC_TBL_START_COUNT(altivec_put_pixels16_num, 1);
block +=line_size;
}
-POWERPC_TBL_STOP_COUNT(altivec_put_pixels16_num, 1);
+POWERPC_PERF_STOP_COUNT(altivec_put_pixels16_num, 1);
#else /* ALTIVEC_USE_REFERENCE_C_CODE */
register vector unsigned char pixelsv1, pixelsv2;
+ register vector unsigned char pixelsv1B, pixelsv2B;
+ register vector unsigned char pixelsv1C, pixelsv2C;
+ register vector unsigned char pixelsv1D, pixelsv2D;
+
register vector unsigned char perm = vec_lvsl(0, pixels);
int i;
-
-POWERPC_TBL_START_COUNT(altivec_put_pixels16_num, 1);
-
+ register int line_size_2 = line_size << 1;
+ register int line_size_3 = line_size + line_size_2;
+ register int line_size_4 = line_size << 2;
+
+POWERPC_PERF_START_COUNT(altivec_put_pixels16_num, 1);
+// hand-unrolling the loop by 4 gains about 15%
+// mininum execution time goes from 74 to 60 cycles
+// it's faster than -funroll-loops, but using
+// -funroll-loops w/ this is bad - 74 cycles again.
+// all this is on a 7450, tuning for the 7450
+#if 0
for(i=0; i<h; i++) {
pixelsv1 = vec_ld(0, (unsigned char*)pixels);
pixelsv2 = vec_ld(16, (unsigned char*)pixels);
@@ -687,8 +699,29 @@ POWERPC_TBL_START_COUNT(altivec_put_pixels16_num, 1);
pixels+=line_size;
block +=line_size;
}
-
-POWERPC_TBL_STOP_COUNT(altivec_put_pixels16_num, 1);
+#else
+ for(i=0; i<h; i+=4) {
+ pixelsv1 = vec_ld(0, (unsigned char*)pixels);
+ pixelsv2 = vec_ld(16, (unsigned char*)pixels);
+ pixelsv1B = vec_ld(line_size, (unsigned char*)pixels);
+ pixelsv2B = vec_ld(16 + line_size, (unsigned char*)pixels);
+ pixelsv1C = vec_ld(line_size_2, (unsigned char*)pixels);
+ pixelsv2C = vec_ld(16 + line_size_2, (unsigned char*)pixels);
+ pixelsv1D = vec_ld(line_size_3, (unsigned char*)pixels);
+ pixelsv2D = vec_ld(16 + line_size_3, (unsigned char*)pixels);
+ vec_st(vec_perm(pixelsv1, pixelsv2, perm),
+ 0, (unsigned char*)block);
+ vec_st(vec_perm(pixelsv1B, pixelsv2B, perm),
+ line_size, (unsigned char*)block);
+ vec_st(vec_perm(pixelsv1C, pixelsv2C, perm),
+ line_size_2, (unsigned char*)block);
+ vec_st(vec_perm(pixelsv1D, pixelsv2D, perm),
+ line_size_3, (unsigned char*)block);
+ pixels+=line_size_4;
+ block +=line_size_4;
+ }
+#endif
+POWERPC_PERF_STOP_COUNT(altivec_put_pixels16_num, 1);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}
@@ -697,11 +730,11 @@ POWERPC_TBL_STOP_COUNT(altivec_put_pixels16_num, 1);
#define op_avg(a,b) a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEUL)>>1) )
void avg_pixels16_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
-POWERPC_TBL_DECLARE(altivec_avg_pixels16_num, 1);
+POWERPC_PERF_DECLARE(altivec_avg_pixels16_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
int i;
-POWERPC_TBL_START_COUNT(altivec_avg_pixels16_num, 1);
+POWERPC_PERF_START_COUNT(altivec_avg_pixels16_num, 1);
for(i=0; i<h; i++) {
op_avg(*((uint32_t*)(block)),(((const struct unaligned_32 *)(pixels))->l));
@@ -712,14 +745,14 @@ POWERPC_TBL_START_COUNT(altivec_avg_pixels16_num, 1);
block +=line_size;
}
-POWERPC_TBL_STOP_COUNT(altivec_avg_pixels16_num, 1);
+POWERPC_PERF_STOP_COUNT(altivec_avg_pixels16_num, 1);
#else /* ALTIVEC_USE_REFERENCE_C_CODE */
register vector unsigned char pixelsv1, pixelsv2, pixelsv, blockv;
register vector unsigned char perm = vec_lvsl(0, pixels);
int i;
-POWERPC_TBL_START_COUNT(altivec_avg_pixels16_num, 1);
+POWERPC_PERF_START_COUNT(altivec_avg_pixels16_num, 1);
for(i=0; i<h; i++) {
pixelsv1 = vec_ld(0, (unsigned char*)pixels);
@@ -732,7 +765,7 @@ POWERPC_TBL_START_COUNT(altivec_avg_pixels16_num, 1);
block +=line_size;
}
-POWERPC_TBL_STOP_COUNT(altivec_avg_pixels16_num, 1);
+POWERPC_PERF_STOP_COUNT(altivec_avg_pixels16_num, 1);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}
@@ -740,10 +773,10 @@ POWERPC_TBL_STOP_COUNT(altivec_avg_pixels16_num, 1);
/* next one assumes that ((line_size % 8) == 0) */
void avg_pixels8_altivec(uint8_t * block, const uint8_t * pixels, int line_size, int h)
{
-POWERPC_TBL_DECLARE(altivec_avg_pixels8_num, 1);
+POWERPC_PERF_DECLARE(altivec_avg_pixels8_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
int i;
-POWERPC_TBL_START_COUNT(altivec_avg_pixels8_num, 1);
+POWERPC_PERF_START_COUNT(altivec_avg_pixels8_num, 1);
for (i = 0; i < h; i++) {
*((uint32_t *) (block)) =
(((*((uint32_t *) (block))) |
@@ -761,13 +794,13 @@ POWERPC_TBL_START_COUNT(altivec_avg_pixels8_num, 1);
pixels += line_size;
block += line_size;
}
-POWERPC_TBL_STOP_COUNT(altivec_avg_pixels8_num, 1);
+POWERPC_PERF_STOP_COUNT(altivec_avg_pixels8_num, 1);
#else /* ALTIVEC_USE_REFERENCE_C_CODE */
register vector unsigned char pixelsv1, pixelsv2, pixelsv, blockv;
int i;
-POWERPC_TBL_START_COUNT(altivec_avg_pixels8_num, 1);
+POWERPC_PERF_START_COUNT(altivec_avg_pixels8_num, 1);
for (i = 0; i < h; i++) {
/*
@@ -798,7 +831,7 @@ POWERPC_TBL_START_COUNT(altivec_avg_pixels8_num, 1);
block += line_size;
}
-POWERPC_TBL_STOP_COUNT(altivec_avg_pixels8_num, 1);
+POWERPC_PERF_STOP_COUNT(altivec_avg_pixels8_num, 1);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}
@@ -806,10 +839,10 @@ POWERPC_TBL_STOP_COUNT(altivec_avg_pixels8_num, 1);
/* next one assumes that ((line_size % 8) == 0) */
void put_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
-POWERPC_TBL_DECLARE(altivec_put_pixels8_xy2_num, 1);
+POWERPC_PERF_DECLARE(altivec_put_pixels8_xy2_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
int j;
-POWERPC_TBL_START_COUNT(altivec_put_pixels8_xy2_num, 1);
+POWERPC_PERF_START_COUNT(altivec_put_pixels8_xy2_num, 1);
for (j = 0; j < 2; j++) {
int i;
const uint32_t a = (((const struct unaligned_32 *) (pixels))->l);
@@ -842,7 +875,7 @@ POWERPC_TBL_START_COUNT(altivec_put_pixels8_xy2_num, 1);
block += 4 - line_size * h;
}
-POWERPC_TBL_STOP_COUNT(altivec_put_pixels8_xy2_num, 1);
+POWERPC_PERF_STOP_COUNT(altivec_put_pixels8_xy2_num, 1);
#else /* ALTIVEC_USE_REFERENCE_C_CODE */
register int i;
@@ -873,7 +906,7 @@ POWERPC_TBL_STOP_COUNT(altivec_put_pixels8_xy2_num, 1);
(vector unsigned short)pixelsv2);
pixelssum1 = vec_add(pixelssum1, vctwo);
-POWERPC_TBL_START_COUNT(altivec_put_pixels8_xy2_num, 1);
+POWERPC_PERF_START_COUNT(altivec_put_pixels8_xy2_num, 1);
for (i = 0; i < h ; i++) {
int rightside = ((unsigned long)block & 0x0000000F);
blockv = vec_ld(0, block);
@@ -914,17 +947,17 @@ POWERPC_TBL_START_COUNT(altivec_put_pixels8_xy2_num, 1);
pixels += line_size;
}
-POWERPC_TBL_STOP_COUNT(altivec_put_pixels8_xy2_num, 1);
+POWERPC_PERF_STOP_COUNT(altivec_put_pixels8_xy2_num, 1);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}
/* next one assumes that ((line_size % 8) == 0) */
void put_no_rnd_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
-POWERPC_TBL_DECLARE(altivec_put_no_rnd_pixels8_xy2_num, 1);
+POWERPC_PERF_DECLARE(altivec_put_no_rnd_pixels8_xy2_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
int j;
-POWERPC_TBL_START_COUNT(altivec_put_no_rnd_pixels8_xy2_num, 1);
+POWERPC_PERF_START_COUNT(altivec_put_no_rnd_pixels8_xy2_num, 1);
for (j = 0; j < 2; j++) {
int i;
const uint32_t a = (((const struct unaligned_32 *) (pixels))->l);
@@ -957,7 +990,7 @@ POWERPC_TBL_START_COUNT(altivec_put_no_rnd_pixels8_xy2_num, 1);
block += 4 - line_size * h;
}
-POWERPC_TBL_STOP_COUNT(altivec_put_no_rnd_pixels8_xy2_num, 1);
+POWERPC_PERF_STOP_COUNT(altivec_put_no_rnd_pixels8_xy2_num, 1);
#else /* ALTIVEC_USE_REFERENCE_C_CODE */
register int i;
@@ -989,7 +1022,7 @@ POWERPC_TBL_STOP_COUNT(altivec_put_no_rnd_pixels8_xy2_num, 1);
(vector unsigned short)pixelsv2);
pixelssum1 = vec_add(pixelssum1, vcone);
-POWERPC_TBL_START_COUNT(altivec_put_no_rnd_pixels8_xy2_num, 1);
+POWERPC_PERF_START_COUNT(altivec_put_no_rnd_pixels8_xy2_num, 1);
for (i = 0; i < h ; i++) {
int rightside = ((unsigned long)block & 0x0000000F);
blockv = vec_ld(0, block);
@@ -1030,17 +1063,17 @@ POWERPC_TBL_START_COUNT(altivec_put_no_rnd_pixels8_xy2_num, 1);
pixels += line_size;
}
-POWERPC_TBL_STOP_COUNT(altivec_put_no_rnd_pixels8_xy2_num, 1);
+POWERPC_PERF_STOP_COUNT(altivec_put_no_rnd_pixels8_xy2_num, 1);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}
/* next one assumes that ((line_size % 16) == 0) */
void put_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, int line_size, int h)
{
-POWERPC_TBL_DECLARE(altivec_put_pixels16_xy2_num, 1);
+POWERPC_PERF_DECLARE(altivec_put_pixels16_xy2_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
int j;
-POWERPC_TBL_START_COUNT(altivec_put_pixels16_xy2_num, 1);
+POWERPC_PERF_START_COUNT(altivec_put_pixels16_xy2_num, 1);
for (j = 0; j < 4; j++) {
int i;
const uint32_t a = (((const struct unaligned_32 *) (pixels))->l);
@@ -1073,7 +1106,7 @@ POWERPC_TBL_START_COUNT(altivec_put_pixels16_xy2_num, 1);
block += 4 - line_size * h;
}
-POWERPC_TBL_STOP_COUNT(altivec_put_pixels16_xy2_num, 1);
+POWERPC_PERF_STOP_COUNT(altivec_put_pixels16_xy2_num, 1);
#else /* ALTIVEC_USE_REFERENCE_C_CODE */
register int i;
@@ -1087,7 +1120,7 @@ POWERPC_TBL_STOP_COUNT(altivec_put_pixels16_xy2_num, 1);
register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
-POWERPC_TBL_START_COUNT(altivec_put_pixels16_xy2_num, 1);
+POWERPC_PERF_START_COUNT(altivec_put_pixels16_xy2_num, 1);
temp1 = vec_ld(0, pixels);
temp2 = vec_ld(16, pixels);
@@ -1151,17 +1184,17 @@ POWERPC_TBL_START_COUNT(altivec_put_pixels16_xy2_num, 1);
pixels += line_size;
}
-POWERPC_TBL_STOP_COUNT(altivec_put_pixels16_xy2_num, 1);
+POWERPC_PERF_STOP_COUNT(altivec_put_pixels16_xy2_num, 1);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}
/* next one assumes that ((line_size % 16) == 0) */
void put_no_rnd_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, int line_size, int h)
{
-POWERPC_TBL_DECLARE(altivec_put_no_rnd_pixels16_xy2_num, 1);
+POWERPC_PERF_DECLARE(altivec_put_no_rnd_pixels16_xy2_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
int j;
-POWERPC_TBL_START_COUNT(altivec_put_no_rnd_pixels16_xy2_num, 1);
+POWERPC_PERF_START_COUNT(altivec_put_no_rnd_pixels16_xy2_num, 1);
for (j = 0; j < 4; j++) {
int i;
const uint32_t a = (((const struct unaligned_32 *) (pixels))->l);
@@ -1194,7 +1227,7 @@ POWERPC_TBL_START_COUNT(altivec_put_no_rnd_pixels16_xy2_num, 1);
block += 4 - line_size * h;
}
-POWERPC_TBL_STOP_COUNT(altivec_put_no_rnd_pixels16_xy2_num, 1);
+POWERPC_PERF_STOP_COUNT(altivec_put_no_rnd_pixels16_xy2_num, 1);
#else /* ALTIVEC_USE_REFERENCE_C_CODE */
register int i;
@@ -1209,7 +1242,7 @@ POWERPC_TBL_STOP_COUNT(altivec_put_no_rnd_pixels16_xy2_num, 1);
register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1);
register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
-POWERPC_TBL_START_COUNT(altivec_put_no_rnd_pixels16_xy2_num, 1);
+POWERPC_PERF_START_COUNT(altivec_put_no_rnd_pixels16_xy2_num, 1);
temp1 = vec_ld(0, pixels);
temp2 = vec_ld(16, pixels);
@@ -1273,7 +1306,7 @@ POWERPC_TBL_START_COUNT(altivec_put_no_rnd_pixels16_xy2_num, 1);
pixels += line_size;
}
-POWERPC_TBL_STOP_COUNT(altivec_put_no_rnd_pixels16_xy2_num, 1);
+POWERPC_PERF_STOP_COUNT(altivec_put_no_rnd_pixels16_xy2_num, 1);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}
diff --git a/libavcodec/ppc/dsputil_ppc.c b/libavcodec/ppc/dsputil_ppc.c
index 87772b4580..73ae8b4a73 100644
--- a/libavcodec/ppc/dsputil_ppc.c
+++ b/libavcodec/ppc/dsputil_ppc.c
@@ -41,8 +41,8 @@ int mm_support(void)
return result;
}
-#ifdef POWERPC_TBL_PERFORMANCE_REPORT
-unsigned long long perfdata[powerpc_perf_total][powerpc_data_total];
+#ifdef POWERPC_PERFORMANCE_REPORT
+unsigned long long perfdata[POWERPC_NUM_PMC_ENABLED][powerpc_perf_total][powerpc_data_total];
/* list below must match enum in dsputil_ppc.h */
static unsigned char* perfname[] = {
"fft_calc_altivec",
@@ -60,53 +60,32 @@ static unsigned char* perfname[] = {
"clear_blocks_dcbz32_ppc",
"clear_blocks_dcbz128_ppc"
};
-#ifdef POWERPC_PERF_USE_PMC
-unsigned long long perfdata_pmc2[powerpc_perf_total][powerpc_data_total];
-unsigned long long perfdata_pmc3[powerpc_perf_total][powerpc_data_total];
-#endif
#include <stdio.h>
#endif
-#ifdef POWERPC_TBL_PERFORMANCE_REPORT
+#ifdef POWERPC_PERFORMANCE_REPORT
void powerpc_display_perf_report(void)
{
- int i;
-#ifndef POWERPC_PERF_USE_PMC
- fprintf(stderr, "PowerPC performance report\n Values are from the Time Base register, and represent 4 bus cycles.\n");
-#else /* POWERPC_PERF_USE_PMC */
+ int i, j;
fprintf(stderr, "PowerPC performance report\n Values are from the PMC registers, and represent whatever the registers are set to record.\n");
-#endif /* POWERPC_PERF_USE_PMC */
for(i = 0 ; i < powerpc_perf_total ; i++)
{
- if (perfdata[i][powerpc_data_num] != (unsigned long long)0)
- fprintf(stderr, " Function \"%s\" (pmc1):\n\tmin: %llu\n\tmax: %llu\n\tavg: %1.2lf (%llu)\n",
- perfname[i],
- perfdata[i][powerpc_data_min],
- perfdata[i][powerpc_data_max],
- (double)perfdata[i][powerpc_data_sum] /
- (double)perfdata[i][powerpc_data_num],
- perfdata[i][powerpc_data_num]);
-#ifdef POWERPC_PERF_USE_PMC
- if (perfdata_pmc2[i][powerpc_data_num] != (unsigned long long)0)
- fprintf(stderr, " Function \"%s\" (pmc2):\n\tmin: %llu\n\tmax: %llu\n\tavg: %1.2lf (%llu)\n",
- perfname[i],
- perfdata_pmc2[i][powerpc_data_min],
- perfdata_pmc2[i][powerpc_data_max],
- (double)perfdata_pmc2[i][powerpc_data_sum] /
- (double)perfdata_pmc2[i][powerpc_data_num],
- perfdata_pmc2[i][powerpc_data_num]);
- if (perfdata_pmc3[i][powerpc_data_num] != (unsigned long long)0)
- fprintf(stderr, " Function \"%s\" (pmc3):\n\tmin: %llu\n\tmax: %llu\n\tavg: %1.2lf (%llu)\n",
- perfname[i],
- perfdata_pmc3[i][powerpc_data_min],
- perfdata_pmc3[i][powerpc_data_max],
- (double)perfdata_pmc3[i][powerpc_data_sum] /
- (double)perfdata_pmc3[i][powerpc_data_num],
- perfdata_pmc3[i][powerpc_data_num]);
-#endif
+ for (j = 0; j < POWERPC_NUM_PMC_ENABLED ; j++)
+ {
+ if (perfdata[j][i][powerpc_data_num] != (unsigned long long)0)
+ fprintf(stderr,
+ " Function \"%s\" (pmc%d):\n\tmin: %llu\n\tmax: %llu\n\tavg: %1.2lf (%llu)\n",
+ perfname[i],
+ j+1,
+ perfdata[j][i][powerpc_data_min],
+ perfdata[j][i][powerpc_data_max],
+ (double)perfdata[j][i][powerpc_data_sum] /
+ (double)perfdata[j][i][powerpc_data_num],
+ perfdata[j][i][powerpc_data_num]);
+ }
}
}
-#endif /* POWERPC_TBL_PERFORMANCE_REPORT */
+#endif /* POWERPC_PERFORMANCE_REPORT */
/* ***** WARNING ***** WARNING ***** WARNING ***** */
/*
@@ -135,10 +114,10 @@ void powerpc_display_perf_report(void)
*/
void clear_blocks_dcbz32_ppc(DCTELEM *blocks)
{
-POWERPC_TBL_DECLARE(powerpc_clear_blocks_dcbz32, 1);
+POWERPC_PERF_DECLARE(powerpc_clear_blocks_dcbz32, 1);
register int misal = ((unsigned long)blocks & 0x00000010);
register int i = 0;
-POWERPC_TBL_START_COUNT(powerpc_clear_blocks_dcbz32, 1);
+POWERPC_PERF_START_COUNT(powerpc_clear_blocks_dcbz32, 1);
#if 1
if (misal) {
((unsigned long*)blocks)[0] = 0L;
@@ -160,7 +139,7 @@ POWERPC_TBL_START_COUNT(powerpc_clear_blocks_dcbz32, 1);
#else
memset(blocks, 0, sizeof(DCTELEM)*6*64);
#endif
-POWERPC_TBL_STOP_COUNT(powerpc_clear_blocks_dcbz32, 1);
+POWERPC_PERF_STOP_COUNT(powerpc_clear_blocks_dcbz32, 1);
}
/* same as above, when dcbzl clear a whole 128B cache line
@@ -168,10 +147,10 @@ POWERPC_TBL_STOP_COUNT(powerpc_clear_blocks_dcbz32, 1);
#ifndef NO_DCBZL
void clear_blocks_dcbz128_ppc(DCTELEM *blocks)
{
-POWERPC_TBL_DECLARE(powerpc_clear_blocks_dcbz128, 1);
+POWERPC_PERF_DECLARE(powerpc_clear_blocks_dcbz128, 1);
register int misal = ((unsigned long)blocks & 0x0000007f);
register int i = 0;
-POWERPC_TBL_START_COUNT(powerpc_clear_blocks_dcbz128, 1);
+POWERPC_PERF_START_COUNT(powerpc_clear_blocks_dcbz128, 1);
#if 1
if (misal) {
// we could probably also optimize this case,
@@ -186,7 +165,7 @@ POWERPC_TBL_START_COUNT(powerpc_clear_blocks_dcbz128, 1);
#else
memset(blocks, 0, sizeof(DCTELEM)*6*64);
#endif
-POWERPC_TBL_STOP_COUNT(powerpc_clear_blocks_dcbz128, 1);
+POWERPC_PERF_STOP_COUNT(powerpc_clear_blocks_dcbz128, 1);
}
#else
void clear_blocks_dcbz128_ppc(DCTELEM *blocks)
@@ -277,6 +256,8 @@ void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx)
c->add_bytes= add_bytes_altivec;
#endif /* 0 */
c->put_pixels_tab[0][0] = put_pixels16_altivec;
+ /* the tow functions do the same thing, so use the same code */
+ c->put_no_rnd_pixels_tab[0][0] = put_pixels16_altivec;
c->avg_pixels_tab[0][0] = avg_pixels16_altivec;
// next one disabled as it's untested.
#if 0
@@ -301,28 +282,21 @@ void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx)
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}
-#ifdef POWERPC_TBL_PERFORMANCE_REPORT
+#ifdef POWERPC_PERFORMANCE_REPORT
{
- int i;
+ int i, j;
for (i = 0 ; i < powerpc_perf_total ; i++)
{
- perfdata[i][powerpc_data_min] = 0xFFFFFFFFFFFFFFFF;
- perfdata[i][powerpc_data_max] = 0x0000000000000000;
- perfdata[i][powerpc_data_sum] = 0x0000000000000000;
- perfdata[i][powerpc_data_num] = 0x0000000000000000;
-#ifdef POWERPC_PERF_USE_PMC
- perfdata_pmc2[i][powerpc_data_min] = 0xFFFFFFFFFFFFFFFF;
- perfdata_pmc2[i][powerpc_data_max] = 0x0000000000000000;
- perfdata_pmc2[i][powerpc_data_sum] = 0x0000000000000000;
- perfdata_pmc2[i][powerpc_data_num] = 0x0000000000000000;
- perfdata_pmc3[i][powerpc_data_min] = 0xFFFFFFFFFFFFFFFF;
- perfdata_pmc3[i][powerpc_data_max] = 0x0000000000000000;
- perfdata_pmc3[i][powerpc_data_sum] = 0x0000000000000000;
- perfdata_pmc3[i][powerpc_data_num] = 0x0000000000000000;
-#endif /* POWERPC_PERF_USE_PMC */
- }
+ for (j = 0; j < POWERPC_NUM_PMC_ENABLED ; j++)
+ {
+ perfdata[j][i][powerpc_data_min] = (unsigned long long)0xFFFFFFFFFFFFFFFF;
+ perfdata[j][i][powerpc_data_max] = (unsigned long long)0x0000000000000000;
+ perfdata[j][i][powerpc_data_sum] = (unsigned long long)0x0000000000000000;
+ perfdata[j][i][powerpc_data_num] = (unsigned long long)0x0000000000000000;
+ }
+ }
}
-#endif /* POWERPC_TBL_PERFORMANCE_REPORT */
+#endif /* POWERPC_PERFORMANCE_REPORT */
} else
#endif /* HAVE_ALTIVEC */
{
diff --git a/libavcodec/ppc/dsputil_ppc.h b/libavcodec/ppc/dsputil_ppc.h
index 8c325fbc75..4cb299dd9b 100644
--- a/libavcodec/ppc/dsputil_ppc.h
+++ b/libavcodec/ppc/dsputil_ppc.h
@@ -30,8 +30,10 @@
#define NO_DCBZL
#endif /* CONFIG_DARWIN */
-#ifdef POWERPC_TBL_PERFORMANCE_REPORT
+#ifdef POWERPC_PERFORMANCE_REPORT
void powerpc_display_perf_report(void);
+/* the 604* have 2, the G3* have 4, the G4s have 6 */
+#define POWERPC_NUM_PMC_ENABLED 4
/* if you add to the enum below, also add to the perfname array
in dsputil_ppc.c */
enum powerpc_perf_index {
@@ -58,98 +60,65 @@ enum powerpc_data_index {
powerpc_data_num,
powerpc_data_total
};
-extern unsigned long long perfdata[powerpc_perf_total][powerpc_data_total];
-#ifdef POWERPC_PERF_USE_PMC
-extern unsigned long long perfdata_pmc2[powerpc_perf_total][powerpc_data_total];
-extern unsigned long long perfdata_pmc3[powerpc_perf_total][powerpc_data_total];
-#endif
-
-#ifndef POWERPC_PERF_USE_PMC
-#define POWERPC_GET_CYCLES(a) asm volatile("mftb %0" : "=r" (a))
-#define POWERPC_TBL_DECLARE(a, cond) register unsigned long tbl_start, tbl_stop
-#define POWERPC_TBL_START_COUNT(a, cond) do { POWERPC_GET_CYCLES(tbl_start); } while (0)
-#define POWERPC_TBL_STOP_COUNT(a, cond) do { \
- POWERPC_GET_CYCLES(tbl_stop); \
- if (tbl_stop > tbl_start) \
- { \
- unsigned long diff = tbl_stop - tbl_start; \
- if (cond) \
- { \
- if (diff < perfdata[a][powerpc_data_min]) \
- perfdata[a][powerpc_data_min] = diff; \
- if (diff > perfdata[a][powerpc_data_max]) \
- perfdata[a][powerpc_data_max] = diff; \
- perfdata[a][powerpc_data_sum] += diff; \
- perfdata[a][powerpc_data_num] ++; \
- } \
- } \
-} while (0)
+extern unsigned long long perfdata[POWERPC_NUM_PMC_ENABLED][powerpc_perf_total][powerpc_data_total];
-#else /* POWERPC_PERF_USE_PMC */
-#define POWERPC_GET_CYCLES(a) asm volatile("mfspr %0, 937" : "=r" (a))
+#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)
#define POWERPC_GET_PMC3(a) asm volatile("mfspr %0, 941" : "=r" (a))
-#define POWERPC_TBL_DECLARE(a, cond) register unsigned long cycles_start, cycles_stop, pmc2_start, pmc2_stop, pmc3_start, pmc3_stop
-#define POWERPC_TBL_START_COUNT(a, cond) do { \
- POWERPC_GET_PMC3(pmc3_start); \
- POWERPC_GET_PMC2(pmc2_start); \
- POWERPC_GET_CYCLES(cycles_start); } while (0)
-#define POWERPC_TBL_STOP_COUNT(a, cond) do { \
- POWERPC_GET_CYCLES(cycles_stop); \
- POWERPC_GET_PMC2(pmc2_stop); \
- POWERPC_GET_PMC3(pmc3_stop); \
- if (cycles_stop >= cycles_start) \
- { \
- unsigned long diff = \
- cycles_stop - cycles_start; \
- if (cond) \
- { \
- if (diff < perfdata[a][powerpc_data_min]) \
- perfdata[a][powerpc_data_min] = diff; \
- if (diff > perfdata[a][powerpc_data_max]) \
- perfdata[a][powerpc_data_max] = diff; \
- perfdata[a][powerpc_data_sum] += diff; \
- perfdata[a][powerpc_data_num] ++; \
- } \
- } \
- if (pmc2_stop >= pmc2_start) \
- { \
- unsigned long diff = \
- pmc2_stop - pmc2_start; \
- if (cond) \
- { \
- if (diff < perfdata_pmc2[a][powerpc_data_min]) \
- perfdata_pmc2[a][powerpc_data_min] = diff; \
- if (diff > perfdata_pmc2[a][powerpc_data_max]) \
- perfdata_pmc2[a][powerpc_data_max] = diff; \
- perfdata_pmc2[a][powerpc_data_sum] += diff; \
- perfdata_pmc2[a][powerpc_data_num] ++; \
- } \
- } \
- if (pmc3_stop >= pmc3_start) \
- { \
- unsigned long diff = \
- pmc3_stop - pmc3_start; \
- if (cond) \
- { \
- if (diff < perfdata_pmc3[a][powerpc_data_min]) \
- perfdata_pmc3[a][powerpc_data_min] = diff; \
- if (diff > perfdata_pmc3[a][powerpc_data_max]) \
- perfdata_pmc3[a][powerpc_data_max] = diff; \
- perfdata_pmc3[a][powerpc_data_sum] += diff; \
- perfdata_pmc3[a][powerpc_data_num] ++; \
- } \
- } \
+#define POWERPC_GET_PMC4(a) asm volatile("mfspr %0, 942" : "=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, 929" : "=r" (a))
+#define POWERPC_GET_PMC6(a) asm volatile("mfspr %0, 930" : "=r" (a))
+#else
+#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;
+#define POWERPC_PERF_START_COUNT(a, cond) do { \
+ POWERPC_GET_PMC6(pmc_start[5]); \
+ POWERPC_GET_PMC5(pmc_start[4]); \
+ POWERPC_GET_PMC4(pmc_start[3]); \
+ POWERPC_GET_PMC3(pmc_start[2]); \
+ POWERPC_GET_PMC2(pmc_start[1]); \
+ POWERPC_GET_PMC1(pmc_start[0]); \
+ } while (0)
+#define POWERPC_PERF_STOP_COUNT(a, cond) do { \
+ POWERPC_GET_PMC1(pmc_stop[0]); \
+ POWERPC_GET_PMC2(pmc_stop[1]); \
+ POWERPC_GET_PMC3(pmc_stop[2]); \
+ POWERPC_GET_PMC4(pmc_stop[3]); \
+ POWERPC_GET_PMC5(pmc_stop[4]); \
+ POWERPC_GET_PMC6(pmc_stop[5]); \
+ if (cond) \
+ { \
+ for(pmc_loop_index = 0; \
+ pmc_loop_index < POWERPC_NUM_PMC_ENABLED; \
+ pmc_loop_index++) \
+ { \
+ if (pmc_stop[pmc_loop_index] >= pmc_start[pmc_loop_index]) \
+ { \
+ unsigned long 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; \
+ if (diff > perfdata[pmc_loop_index][a][powerpc_data_max]) \
+ perfdata[pmc_loop_index][a][powerpc_data_max] = diff; \
+ perfdata[pmc_loop_index][a][powerpc_data_sum] += diff; \
+ perfdata[pmc_loop_index][a][powerpc_data_num] ++; \
+ } \
+ } \
+ } \
} while (0)
-
-#endif /* POWERPC_PERF_USE_PMC */
-
-
-#else /* POWERPC_TBL_PERFORMANCE_REPORT */
+#else /* POWERPC_PERFORMANCE_REPORT */
// those are needed to avoid empty statements.
-#define POWERPC_TBL_DECLARE(a, cond) int altivec_placeholder __attribute__ ((unused))
-#define POWERPC_TBL_START_COUNT(a, cond) do {} while (0)
-#define POWERPC_TBL_STOP_COUNT(a, cond) do {} while (0)
-#endif /* POWERPC_TBL_PERFORMANCE_REPORT */
+#define POWERPC_PERF_DECLARE(a, cond) int altivec_placeholder __attribute__ ((unused))
+#define POWERPC_PERF_START_COUNT(a, cond) do {} while (0)
+#define POWERPC_PERF_STOP_COUNT(a, cond) do {} while (0)
+#endif /* POWERPC_PERFORMANCE_REPORT */
#endif /* _DSPUTIL_PPC_ */
diff --git a/libavcodec/ppc/fft_altivec.c b/libavcodec/ppc/fft_altivec.c
index 75c95bb879..e39c9dbb73 100644
--- a/libavcodec/ppc/fft_altivec.c
+++ b/libavcodec/ppc/fft_altivec.c
@@ -62,7 +62,7 @@
*/
void fft_calc_altivec(FFTContext *s, FFTComplex *z)
{
-POWERPC_TBL_DECLARE(altivec_fft_num, s->nbits >= 6);
+POWERPC_PERF_DECLARE(altivec_fft_num, s->nbits >= 6);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
int ln = s->nbits;
int j, np, np2;
@@ -72,7 +72,7 @@ POWERPC_TBL_DECLARE(altivec_fft_num, s->nbits >= 6);
int l;
FFTSample tmp_re, tmp_im;
-POWERPC_TBL_START_COUNT(altivec_fft_num, s->nbits >= 6);
+POWERPC_PERF_START_COUNT(altivec_fft_num, s->nbits >= 6);
np = 1 << ln;
@@ -137,7 +137,7 @@ POWERPC_TBL_START_COUNT(altivec_fft_num, s->nbits >= 6);
nloops = nloops << 1;
} while (nblocks != 0);
-POWERPC_TBL_STOP_COUNT(altivec_fft_num, s->nbits >= 6);
+POWERPC_PERF_STOP_COUNT(altivec_fft_num, s->nbits >= 6);
#else /* ALTIVEC_USE_REFERENCE_C_CODE */
#ifdef CONFIG_DARWIN
@@ -153,7 +153,7 @@ POWERPC_TBL_STOP_COUNT(altivec_fft_num, s->nbits >= 6);
FFTComplex *cptr, *cptr1;
int k;
-POWERPC_TBL_START_COUNT(altivec_fft_num, s->nbits >= 6);
+POWERPC_PERF_START_COUNT(altivec_fft_num, s->nbits >= 6);
np = 1 << ln;
@@ -241,7 +241,7 @@ POWERPC_TBL_START_COUNT(altivec_fft_num, s->nbits >= 6);
nloops = nloops << 1;
} while (nblocks != 0);
-POWERPC_TBL_STOP_COUNT(altivec_fft_num, s->nbits >= 6);
+POWERPC_PERF_STOP_COUNT(altivec_fft_num, s->nbits >= 6);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}
diff --git a/libavcodec/ppc/gmc_altivec.c b/libavcodec/ppc/gmc_altivec.c
index 9b141078bb..671ee110aa 100644
--- a/libavcodec/ppc/gmc_altivec.c
+++ b/libavcodec/ppc/gmc_altivec.c
@@ -31,7 +31,7 @@
#define GMC1_PERF_COND (h==8)
void gmc1_altivec(uint8_t *dst /* align 8 */, uint8_t *src /* align1 */, int stride, int h, int x16, int y16, int rounder)
{
-POWERPC_TBL_DECLARE(altivec_gmc1_num, GMC1_PERF_COND);
+POWERPC_PERF_DECLARE(altivec_gmc1_num, GMC1_PERF_COND);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
const int A=(16-x16)*(16-y16);
const int B=( x16)*(16-y16);
@@ -39,7 +39,7 @@ POWERPC_TBL_DECLARE(altivec_gmc1_num, GMC1_PERF_COND);
const int D=( x16)*( y16);
int i;
-POWERPC_TBL_START_COUNT(altivec_gmc1_num, GMC1_PERF_COND);
+POWERPC_PERF_START_COUNT(altivec_gmc1_num, GMC1_PERF_COND);
for(i=0; i<h; i++)
{
@@ -55,7 +55,7 @@ POWERPC_TBL_START_COUNT(altivec_gmc1_num, GMC1_PERF_COND);
src+= stride;
}
-POWERPC_TBL_STOP_COUNT(altivec_gmc1_num, GMC1_PERF_COND);
+POWERPC_PERF_STOP_COUNT(altivec_gmc1_num, GMC1_PERF_COND);
#else /* ALTIVEC_USE_REFERENCE_C_CODE */
const unsigned short __attribute__ ((aligned(16))) rounder_a[8] =
@@ -78,7 +78,7 @@ POWERPC_TBL_STOP_COUNT(altivec_gmc1_num, GMC1_PERF_COND);
unsigned long src_really_odd = (unsigned long)src & 0x0000000F;
-POWERPC_TBL_START_COUNT(altivec_gmc1_num, GMC1_PERF_COND);
+POWERPC_PERF_START_COUNT(altivec_gmc1_num, GMC1_PERF_COND);
tempA = vec_ld(0, (unsigned short*)ABCD);
Av = vec_splat(tempA, 0);
@@ -166,7 +166,7 @@ POWERPC_TBL_START_COUNT(altivec_gmc1_num, GMC1_PERF_COND);
src += stride;
}
-POWERPC_TBL_STOP_COUNT(altivec_gmc1_num, GMC1_PERF_COND);
+POWERPC_PERF_STOP_COUNT(altivec_gmc1_num, GMC1_PERF_COND);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}
diff --git a/libavcodec/ppc/idct_altivec.c b/libavcodec/ppc/idct_altivec.c
index f8a8aa6787..d821ecd223 100644
--- a/libavcodec/ppc/idct_altivec.c
+++ b/libavcodec/ppc/idct_altivec.c
@@ -165,16 +165,16 @@ static const vector_s16_t constants[5] = {
void idct_put_altivec(uint8_t* dest, int stride, vector_s16_t* block)
{
-POWERPC_TBL_DECLARE(altivec_idct_put_num, 1);
+POWERPC_PERF_DECLARE(altivec_idct_put_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
-POWERPC_TBL_START_COUNT(altivec_idct_put_num, 1);
+POWERPC_PERF_START_COUNT(altivec_idct_put_num, 1);
void simple_idct_put(uint8_t *dest, int line_size, int16_t *block);
simple_idct_put(dest, stride, (int16_t*)block);
-POWERPC_TBL_STOP_COUNT(altivec_idct_put_num, 1);
+POWERPC_PERF_STOP_COUNT(altivec_idct_put_num, 1);
#else /* ALTIVEC_USE_REFERENCE_C_CODE */
vector_u8_t tmp;
-POWERPC_TBL_START_COUNT(altivec_idct_put_num, 1);
+POWERPC_PERF_START_COUNT(altivec_idct_put_num, 1);
IDCT
@@ -192,18 +192,18 @@ POWERPC_TBL_START_COUNT(altivec_idct_put_num, 1);
COPY (dest, vx6) dest += stride;
COPY (dest, vx7)
-POWERPC_TBL_STOP_COUNT(altivec_idct_put_num, 1);
+POWERPC_PERF_STOP_COUNT(altivec_idct_put_num, 1);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}
void idct_add_altivec(uint8_t* dest, int stride, vector_s16_t* block)
{
-POWERPC_TBL_DECLARE(altivec_idct_add_num, 1);
+POWERPC_PERF_DECLARE(altivec_idct_add_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
-POWERPC_TBL_START_COUNT(altivec_idct_add_num, 1);
+POWERPC_PERF_START_COUNT(altivec_idct_add_num, 1);
void simple_idct_add(uint8_t *dest, int line_size, int16_t *block);
simple_idct_add(dest, stride, (int16_t*)block);
-POWERPC_TBL_STOP_COUNT(altivec_idct_add_num, 1);
+POWERPC_PERF_STOP_COUNT(altivec_idct_add_num, 1);
#else /* ALTIVEC_USE_REFERENCE_C_CODE */
vector_u8_t tmp;
vector_s16_t tmp2, tmp3;
@@ -211,7 +211,7 @@ POWERPC_TBL_STOP_COUNT(altivec_idct_add_num, 1);
vector_u8_t perm1;
vector_u8_t p0, p1, p;
-POWERPC_TBL_START_COUNT(altivec_idct_add_num, 1);
+POWERPC_PERF_START_COUNT(altivec_idct_add_num, 1);
IDCT
@@ -239,7 +239,7 @@ POWERPC_TBL_START_COUNT(altivec_idct_add_num, 1);
ADD (dest, vx6, perm0) dest += stride;
ADD (dest, vx7, perm1)
-POWERPC_TBL_STOP_COUNT(altivec_idct_add_num, 1);
+POWERPC_PERF_STOP_COUNT(altivec_idct_add_num, 1);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}
diff --git a/libavcodec/ppc/mpegvideo_altivec.c b/libavcodec/ppc/mpegvideo_altivec.c
index bbf9c44335..ae3170d91a 100644
--- a/libavcodec/ppc/mpegvideo_altivec.c
+++ b/libavcodec/ppc/mpegvideo_altivec.c
@@ -522,13 +522,13 @@ int dct_quantize_altivec(MpegEncContext* s,
void dct_unquantize_h263_altivec(MpegEncContext *s,
DCTELEM *block, int n, int qscale)
{
-POWERPC_TBL_DECLARE(altivec_dct_unquantize_h263_num, 1);
+POWERPC_PERF_DECLARE(altivec_dct_unquantize_h263_num, 1);
int i, level, qmul, qadd;
int nCoeffs;
assert(s->block_last_index[n]>=0);
-POWERPC_TBL_START_COUNT(altivec_dct_unquantize_h263_num, 1);
+POWERPC_PERF_START_COUNT(altivec_dct_unquantize_h263_num, 1);
qadd = (qscale - 1) | 1;
qmul = qscale << 1;
@@ -641,5 +641,5 @@ POWERPC_TBL_START_COUNT(altivec_dct_unquantize_h263_num, 1);
}
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
-POWERPC_TBL_STOP_COUNT(altivec_dct_unquantize_h263_num, nCoeffs == 63);
+POWERPC_PERF_STOP_COUNT(altivec_dct_unquantize_h263_num, nCoeffs == 63);
}