aboutsummaryrefslogtreecommitdiffstats
path: root/libavcodec/ppc
diff options
context:
space:
mode:
authorRomain Dolbeau <dolbeau@irisa.fr>2003-01-19 19:00:45 +0000
committerMichael Niedermayer <michaelni@gmx.at>2003-01-19 19:00:45 +0000
commit35e5fb067e0099fde6c3429131f11433eb92f22b (patch)
tree50480aa9498173e6b2f12f977d6e948b8f90fa53 /libavcodec/ppc
parent8c55915b14897fd9bdaae9241f2dd54c00a869ef (diff)
downloadffmpeg-35e5fb067e0099fde6c3429131f11433eb92f22b.tar.gz
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
Originally committed as revision 1476 to svn://svn.ffmpeg.org/ffmpeg/trunk
Diffstat (limited to 'libavcodec/ppc')
-rw-r--r--libavcodec/ppc/dsputil_altivec.c254
-rw-r--r--libavcodec/ppc/dsputil_altivec.h67
-rw-r--r--libavcodec/ppc/dsputil_ppc.c167
-rw-r--r--libavcodec/ppc/dsputil_ppc.h120
-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
8 files changed, 513 insertions, 141 deletions
diff --git a/libavcodec/ppc/dsputil_altivec.c b/libavcodec/ppc/dsputil_altivec.c
index e7f34ee16c..fcae7c1180 100644
--- a/libavcodec/ppc/dsputil_altivec.c
+++ b/libavcodec/ppc/dsputil_altivec.c
@@ -24,22 +24,6 @@
#include <sys/sysctl.h>
#endif
-#ifdef ALTIVEC_TBL_PERFORMANCE_REPORT
-unsigned long long perfdata[altivec_perf_total][altivec_data_total];
-/* list below must match enum in dsputil_altivec.h */
-static unsigned char* perfname[] = {
- "fft_calc",
- "gmc1",
- "dct_unquantize_h263",
- "idct_add",
- "idct_put",
- "put_pixels_clamped",
- "put_pixels16",
- "avg_pixels16"
-};
-#include <stdio.h>
-#endif
-
int pix_abs16x16_x2_altivec(uint8_t *pix1, uint8_t *pix2, int line_size)
{
int i;
@@ -648,12 +632,12 @@ extern UINT8 cropTbl[];
void put_pixels_clamped_altivec(const DCTELEM *block, UINT8 *restrict pixels,
int line_size)
{
-ALTIVEC_TBL_DECLARE(altivec_put_pixels_clamped_num, 1);
+POWERPC_TBL_DECLARE(altivec_put_pixels_clamped_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
int i;
UINT8 *cm = cropTbl + MAX_NEG_CROP;
-ALTIVEC_TBL_START_COUNT(altivec_put_pixels_clamped_num, 1);
+POWERPC_TBL_START_COUNT(altivec_put_pixels_clamped_num, 1);
/* read the pixels */
for(i=0;i<8;i++) {
@@ -670,7 +654,7 @@ ALTIVEC_TBL_START_COUNT(altivec_put_pixels_clamped_num, 1);
block += 8;
}
-ALTIVEC_TBL_STOP_COUNT(altivec_put_pixels_clamped_num, 1);
+POWERPC_TBL_STOP_COUNT(altivec_put_pixels_clamped_num, 1);
#else /* ALTIVEC_USE_REFERENCE_C_CODE */
register const vector short vczero = (const vector short)(0);
@@ -681,7 +665,7 @@ ALTIVEC_TBL_STOP_COUNT(altivec_put_pixels_clamped_num, 1);
pixelsv0, pixelsv1, pixelsv2, pixelsv3, pixelsv4,
pixelsv0old, pixelsv4old;
-ALTIVEC_TBL_START_COUNT(altivec_put_pixels_clamped_num, 1);
+POWERPC_TBL_START_COUNT(altivec_put_pixels_clamped_num, 1);
blockv0 = vec_ld(0, block);
blockv1 = vec_ld(16, block);
@@ -720,17 +704,17 @@ ALTIVEC_TBL_START_COUNT(altivec_put_pixels_clamped_num, 1);
vec_st(pixelsv3, 48, pixels);
}
-ALTIVEC_TBL_STOP_COUNT(altivec_put_pixels_clamped_num, 1);
+POWERPC_TBL_STOP_COUNT(altivec_put_pixels_clamped_num, 1);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}
void put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
-ALTIVEC_TBL_DECLARE(altivec_put_pixels16_num, 1);
+POWERPC_TBL_DECLARE(altivec_put_pixels16_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
int i;
-ALTIVEC_TBL_START_COUNT(altivec_put_pixels16_num, 1);
+POWERPC_TBL_START_COUNT(altivec_put_pixels16_num, 1);
for(i=0; i<h; i++) {
*((uint32_t*)(block )) = (((const struct unaligned_32 *) (pixels))->l);
@@ -741,25 +725,24 @@ ALTIVEC_TBL_START_COUNT(altivec_put_pixels16_num, 1);
block +=line_size;
}
-ALTIVEC_TBL_STOP_COUNT(altivec_put_pixels16_num, 1);
+POWERPC_TBL_STOP_COUNT(altivec_put_pixels16_num, 1);
#else /* ALTIVEC_USE_REFERENCE_C_CODE */
-
- register vector unsigned char perm = vec_lvsl(0, pixels);
register vector unsigned char pixelsv1, pixelsv2;
int i;
-ALTIVEC_TBL_START_COUNT(altivec_put_pixels16_num, 1);
+POWERPC_TBL_START_COUNT(altivec_put_pixels16_num, 1);
for(i=0; i<h; i++) {
pixelsv1 = vec_ld(0, (unsigned char*)pixels);
pixelsv2 = vec_ld(16, (unsigned char*)pixels);
- vec_st(vec_perm(pixelsv1, pixelsv2, perm), 0, (unsigned char*)block);
+ vec_st(vec_perm(pixelsv1, pixelsv2, vec_lvsl(0, pixels)),
+ 0, (unsigned char*)block);
pixels+=line_size;
block +=line_size;
}
-ALTIVEC_TBL_STOP_COUNT(altivec_put_pixels16_num, 1);
+POWERPC_TBL_STOP_COUNT(altivec_put_pixels16_num, 1);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}
@@ -767,11 +750,11 @@ ALTIVEC_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)
{
-ALTIVEC_TBL_DECLARE(altivec_avg_pixels16_num, 1);
+POWERPC_TBL_DECLARE(altivec_avg_pixels16_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
int i;
-ALTIVEC_TBL_START_COUNT(altivec_avg_pixels16_num, 1);
+POWERPC_TBL_START_COUNT(altivec_avg_pixels16_num, 1);
for(i=0; i<h; i++) {
op_avg(*((uint32_t*)(block)),(((const struct unaligned_32 *)(pixels))->l));
@@ -782,32 +765,210 @@ ALTIVEC_TBL_START_COUNT(altivec_avg_pixels16_num, 1);
block +=line_size;
}
-ALTIVEC_TBL_STOP_COUNT(altivec_avg_pixels16_num, 1);
+POWERPC_TBL_STOP_COUNT(altivec_avg_pixels16_num, 1);
#else /* ALTIVEC_USE_REFERENCE_C_CODE */
-
- register vector unsigned char perm = vec_lvsl(0, pixels);
register vector unsigned char pixelsv1, pixelsv2, pixelsv, blockv;
int i;
-ALTIVEC_TBL_START_COUNT(altivec_avg_pixels16_num, 1);
+POWERPC_TBL_START_COUNT(altivec_avg_pixels16_num, 1);
for(i=0; i<h; i++) {
pixelsv1 = vec_ld(0, (unsigned char*)pixels);
pixelsv2 = vec_ld(16, (unsigned char*)pixels);
blockv = vec_ld(0, block);
- pixelsv = vec_perm(pixelsv1, pixelsv2, perm);
+ pixelsv = vec_perm(pixelsv1, pixelsv2, vec_lvsl(0, pixels));
blockv = vec_avg(blockv,pixelsv);
vec_st(blockv, 0, (unsigned char*)block);
pixels+=line_size;
block +=line_size;
}
-ALTIVEC_TBL_STOP_COUNT(altivec_avg_pixels16_num, 1);
+POWERPC_TBL_STOP_COUNT(altivec_avg_pixels16_num, 1);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}
+void avg_pixels8_altivec(uint8_t * block, const uint8_t * pixels,
+ int line_size, int h)
+{
+POWERPC_TBL_DECLARE(altivec_avg_pixels8_num, 1);
+#ifdef ALTIVEC_USE_REFERENCE_C_CODE
+ int i;
+POWERPC_TBL_START_COUNT(altivec_avg_pixels8_num, 1);
+ for (i = 0; i < h; i++) {
+ *((uint32_t *) (block)) =
+ (((*((uint32_t *) (block))) |
+ ((((const struct unaligned_32 *) (pixels))->l))) -
+ ((((*((uint32_t *) (block))) ^
+ ((((const struct unaligned_32 *) (pixels))->
+ l))) & 0xFEFEFEFEUL) >> 1));
+ *((uint32_t *) (block + 4)) =
+ (((*((uint32_t *) (block + 4))) |
+ ((((const struct unaligned_32 *) (pixels + 4))->l))) -
+ ((((*((uint32_t *) (block + 4))) ^
+ ((((const struct unaligned_32 *) (pixels +
+ 4))->
+ l))) & 0xFEFEFEFEUL) >> 1));
+ pixels += line_size;
+ block += line_size;
+ }
+POWERPC_TBL_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);
+
+ for (i = 0; i < h; i++) {
+ /*
+ block is 8 bytes-aligned, so we're either in the
+ left block (16 bytes-aligned) or in the right block (not)
+ */
+ int rightside = ((unsigned long)block & 0x0000000F);
+
+ blockv = vec_ld(0, block);
+ pixelsv1 = vec_ld(0, (unsigned char*)pixels);
+ pixelsv2 = vec_ld(16, (unsigned char*)pixels);
+ pixelsv = vec_perm(pixelsv1, pixelsv2, vec_lvsl(0, pixels));
+
+ if (rightside)
+ {
+ pixelsv = vec_perm(blockv, pixelsv, vcprm(0,1,s0,s1));
+ }
+ else
+ {
+ pixelsv = vec_perm(blockv, pixelsv, vcprm(s0,s1,2,3));
+ }
+
+ blockv = vec_avg(blockv, pixelsv);
+
+ vec_st(blockv, 0, block);
+
+ pixels += line_size;
+ block += line_size;
+ }
+
+POWERPC_TBL_STOP_COUNT(altivec_avg_pixels8_num, 1);
+
+#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
+}
+
+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);
+#ifdef ALTIVEC_USE_REFERENCE_C_CODE
+ int j;
+POWERPC_TBL_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);
+ const uint32_t b =
+ (((const struct unaligned_32 *) (pixels + 1))->l);
+ uint32_t l0 =
+ (a & 0x03030303UL) + (b & 0x03030303UL) + 0x02020202UL;
+ uint32_t h0 =
+ ((a & 0xFCFCFCFCUL) >> 2) + ((b & 0xFCFCFCFCUL) >> 2);
+ uint32_t l1, h1;
+ pixels += line_size;
+ for (i = 0; i < h; i += 2) {
+ uint32_t a = (((const struct unaligned_32 *) (pixels))->l);
+ uint32_t b = (((const struct unaligned_32 *) (pixels + 1))->l);
+ l1 = (a & 0x03030303UL) + (b & 0x03030303UL);
+ h1 = ((a & 0xFCFCFCFCUL) >> 2) + ((b & 0xFCFCFCFCUL) >> 2);
+ *((uint32_t *) block) =
+ h0 + h1 + (((l0 + l1) >> 2) & 0x0F0F0F0FUL);
+ pixels += line_size;
+ block += line_size;
+ a = (((const struct unaligned_32 *) (pixels))->l);
+ b = (((const struct unaligned_32 *) (pixels + 1))->l);
+ l0 = (a & 0x03030303UL) + (b & 0x03030303UL) + 0x02020202UL;
+ h0 = ((a & 0xFCFCFCFCUL) >> 2) + ((b & 0xFCFCFCFCUL) >> 2);
+ *((uint32_t *) block) =
+ h0 + h1 + (((l0 + l1) >> 2) & 0x0F0F0F0FUL);
+ pixels += line_size;
+ block += line_size;
+ } pixels += 4 - line_size * (h + 1);
+ block += 4 - line_size * h;
+ }
+
+POWERPC_TBL_STOP_COUNT(altivec_put_pixels8_xy2_num, 1);
+
+#else /* ALTIVEC_USE_REFERENCE_C_CODE */
+ register int i;
+ register vector unsigned char
+ pixelsv1, pixelsv2,
+ pixelsavg;
+ register vector unsigned char
+ blockv, temp1, temp2;
+ register vector unsigned short
+ pixelssum1, pixelssum2, temp3;
+ register const vector unsigned char vczero = (const vector unsigned char)(0);
+ register const vector unsigned short vctwo = (const vector unsigned short)(2);
+
+ temp1 = vec_ld(0, pixels);
+ temp2 = vec_ld(16, pixels);
+ pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
+ if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F)
+ {
+ pixelsv2 = temp2;
+ }
+ else
+ {
+ pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
+ }
+ pixelsv1 = vec_mergeh(vczero, pixelsv1);
+ pixelsv2 = vec_mergeh(vczero, pixelsv2);
+ pixelssum1 = vec_add((vector unsigned short)pixelsv1,
+ (vector unsigned short)pixelsv2);
+ pixelssum1 = vec_add(pixelssum1, vctwo);
+
+POWERPC_TBL_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);
+
+ temp1 = vec_ld(line_size, pixels);
+ temp2 = vec_ld(line_size + 16, pixels);
+ pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
+ if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F)
+ {
+ pixelsv2 = temp2;
+ }
+ else
+ {
+ pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
+ }
+
+ pixelsv1 = vec_mergeh(vczero, pixelsv1);
+ pixelsv2 = vec_mergeh(vczero, pixelsv2);
+ pixelssum2 = vec_add((vector unsigned short)pixelsv1,
+ (vector unsigned short)pixelsv2);
+ temp3 = vec_add(pixelssum1, pixelssum2);
+ temp3 = vec_sra(temp3, vctwo);
+ pixelssum1 = vec_add(pixelssum2, vctwo);
+ pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);
+
+ if (rightside)
+ {
+ blockv = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
+ }
+ else
+ {
+ blockv = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
+ }
+
+ vec_st(blockv, 0, block);
+
+ block += line_size;
+ pixels += line_size;
+ }
+
+POWERPC_TBL_STOP_COUNT(altivec_put_pixels8_xy2_num, 1);
+#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
+}
+
int has_altivec(void)
{
#if CONFIG_DARWIN
@@ -822,22 +983,3 @@ int has_altivec(void)
#endif
return 0;
}
-
-#ifdef ALTIVEC_TBL_PERFORMANCE_REPORT
-void altivec_display_perf_report(void)
-{
- int i;
- fprintf(stderr, "AltiVec performance report\n Values are from the Time Base register, and represent 4 bus cycles.\n");
- for(i = 0 ; i < altivec_perf_total ; i++)
- {
- if (perfdata[i][altivec_data_num] != (unsigned long long)0)
- fprintf(stderr, " Function \"%s\":\n\tmin: %llu\n\tmax: %llu\n\tavg: %1.2lf (%llu)\n",
- perfname[i],
- perfdata[i][altivec_data_min],
- perfdata[i][altivec_data_max],
- (double)perfdata[i][altivec_data_sum] /
- (double)perfdata[i][altivec_data_num],
- perfdata[i][altivec_data_num]);
- }
-}
-#endif /* ALTIVEC_TBL_PERFORMANCE_REPORT */
diff --git a/libavcodec/ppc/dsputil_altivec.h b/libavcodec/ppc/dsputil_altivec.h
index b7b4238db9..1f66017270 100644
--- a/libavcodec/ppc/dsputil_altivec.h
+++ b/libavcodec/ppc/dsputil_altivec.h
@@ -17,6 +17,13 @@
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
+#ifndef _DSPUTIL_ALTIVEC_
+#define _DSPUTIL_ALTIVEC_
+
+#include "dsputil_ppc.h"
+
+#ifdef HAVE_ALTIVEC
+
extern int pix_abs16x16_x2_altivec(uint8_t *pix1, uint8_t *pix2, int line_size);
extern int pix_abs16x16_y2_altivec(uint8_t *pix1, uint8_t *pix2, int line_size);
extern int pix_abs16x16_xy2_altivec(uint8_t *pix1, uint8_t *pix2, int line_size);
@@ -33,15 +40,15 @@ extern void get_pixels_altivec(DCTELEM* block, const UINT8 * pixels, int line_si
extern void add_bytes_altivec(uint8_t *dst, uint8_t *src, int w);
extern void put_pixels_clamped_altivec(const DCTELEM *block, UINT8 *restrict pixels, int line_size);
-void put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h);
-void avg_pixels16_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h);
+extern void put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h);
+extern void avg_pixels16_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h);
+extern void avg_pixels8_altivec(uint8_t * block, const uint8_t * pixels, int line_size, int h);
+extern void put_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h);
extern void gmc1_altivec(UINT8 *dst, UINT8 *src, int stride, int h, int x16, int y16, int rounder);
extern int has_altivec(void);
-#ifdef HAVE_ALTIVEC
-
// used to build registers permutation vectors (vcprm)
// the 's' are for words in the _s_econd vector
#define WORD_0 0x00,0x01,0x02,0x03
@@ -67,58 +74,10 @@ extern int has_altivec(void);
#define vcii(a,b,c,d) (const vector float)(FLOAT_ ## a, FLOAT_ ## b, FLOAT_ ## c, FLOAT_ ## d)
-#ifdef ALTIVEC_TBL_PERFORMANCE_REPORT
-void altivec_display_perf_report(void);
-/* if you add to the enum below, also add to the perfname array
- in dsputil_altivec.c */
-enum altivec_perf_index {
- altivec_fft_num = 0,
- altivec_gmc1_num,
- altivec_dct_unquantize_h263_num,
- altivec_idct_add_num,
- altivec_idct_put_num,
- altivec_put_pixels_clamped_num,
- altivec_put_pixels16_num,
- altivec_avg_pixels16_num,
- altivec_perf_total
-};
-enum altivec_data_index {
- altivec_data_min = 0,
- altivec_data_max,
- altivec_data_sum,
- altivec_data_num,
- altivec_data_total
-};
-extern unsigned long long perfdata[altivec_perf_total][altivec_data_total];
-#define ALTIVEC_TBL_DECLARE(a, cond) register unsigned long tbl_start, tbl_stop
-#define ALTIVEC_TBL_START_COUNT(a, cond) do { asm("mftb %0" : "=r" (tbl_start)); } while (0)
-#define ALTIVEC_TBL_STOP_COUNT(a, cond) do { \
- asm volatile("mftb %0" : "=r" (tbl_stop)); \
- if (tbl_stop > tbl_start) \
- { \
- unsigned long diff = tbl_stop - tbl_start; \
- if (cond) \
- { \
- if (diff < perfdata[a][altivec_data_min]) \
- perfdata[a][altivec_data_min] = diff; \
- if (diff > perfdata[a][altivec_data_max]) \
- perfdata[a][altivec_data_max] = diff; \
- perfdata[a][altivec_data_sum] += diff; \
- perfdata[a][altivec_data_num] ++; \
- } \
- } \
-} while (0)
-#else /* ALTIVEC_TBL_PERFORMANCE_REPORT */
-#define ALTIVEC_TBL_DECLARE(a, cond)
-#define ALTIVEC_TBL_START_COUNT(a, cond)
-#define ALTIVEC_TBL_STOP_COUNT(a, cond)
-#endif /* ALTIVEC_TBL_PERFORMANCE_REPORT */
-
#else /* HAVE_ALTIVEC */
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
#error "I can't use ALTIVEC_USE_REFERENCE_C_CODE if I don't use HAVE_ALTIVEC"
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
-#ifdef ALTIVEC_TBL_PERFORMANCE_REPORT
-#error "I can't use ALTIVEC_TBL_PERFORMANCE_REPORT if I don't use HAVE_ALTIVEC"
-#endif /* ALTIVEC_TBL_PERFORMANCE_REPORT */
#endif /* HAVE_ALTIVEC */
+
+#endif /* _DSPUTIL_ALTIVEC_ */
diff --git a/libavcodec/ppc/dsputil_ppc.c b/libavcodec/ppc/dsputil_ppc.c
index ec2dda4690..7dcca85a7c 100644
--- a/libavcodec/ppc/dsputil_ppc.c
+++ b/libavcodec/ppc/dsputil_ppc.c
@@ -19,6 +19,8 @@
#include "../dsputil.h"
+#include "dsputil_ppc.h"
+
#ifdef HAVE_ALTIVEC
#include "dsputil_altivec.h"
#endif
@@ -36,12 +38,149 @@ int mm_support(void)
return result;
}
+#ifdef POWERPC_TBL_PERFORMANCE_REPORT
+unsigned long long perfdata[powerpc_perf_total][powerpc_data_total];
+/* list below must match enum in dsputil_altivec.h */
+static unsigned char* perfname[] = {
+ "fft_calc_altivec",
+ "gmc1_altivec",
+ "dct_unquantize_h263_altivec",
+ "idct_add_altivec",
+ "idct_put_altivec",
+ "put_pixels_clamped_altivec",
+ "put_pixels16_altivec",
+ "avg_pixels16_altivec",
+ "avg_pixels8_altivec",
+ "put_pixels8_xy2_altivec",
+ "clear_blocks_dcbz32_ppc"
+};
+#ifdef POWERPC_PERF_USE_PMC
+unsigned long long perfdata_miss[powerpc_perf_total][powerpc_data_total];
+#endif
+#include <stdio.h>
+#endif
+
+#ifdef POWERPC_TBL_PERFORMANCE_REPORT
+void powerpc_display_perf_report(void)
+{
+ int i;
+#ifndef POWERPC_PERF_USE_PMC
+ fprintf(stderr, "AltiVec performance report\n Values are from the Time Base register, and represent 4 bus cycles.\n");
+#else /* POWERPC_PERF_USE_PMC */
+ fprintf(stderr, "AltiVec 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_miss[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_miss[i][powerpc_data_min],
+ perfdata_miss[i][powerpc_data_max],
+ (double)perfdata_miss[i][powerpc_data_sum] /
+ (double)perfdata_miss[i][powerpc_data_num],
+ perfdata_miss[i][powerpc_data_num]);
+#endif
+ }
+}
+#endif /* POWERPC_TBL_PERFORMANCE_REPORT */
+
+/* ***** WARNING ***** WARNING ***** WARNING ***** */
+/*
+ clear_blocks_dcbz32_ppc will not work properly
+ on PowerPC processors with a cache line size
+ not equal to 32 bytes.
+ Fortunately all processor used by Apple up to
+ at least the 7450 (aka second generation G4)
+ use 32 bytes cache line.
+ This is due to the use of the 'dcbz' instruction.
+ It simply clear to zero a single cache line,
+ so you need to know the cache line size to use it !
+ It's absurd, but it's fast...
+*/
+void clear_blocks_dcbz32_ppc(DCTELEM *blocks)
+{
+POWERPC_TBL_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);
+#if 1
+ if (misal) {
+ ((unsigned long*)blocks)[0] = 0L;
+ ((unsigned long*)blocks)[1] = 0L;
+ ((unsigned long*)blocks)[2] = 0L;
+ ((unsigned long*)blocks)[3] = 0L;
+ vec_st((vector short)(0), 0, blocks);
+ i += 16;
+ }
+ for ( ; i < sizeof(DCTELEM)*6*64 ; i += 32) {
+ asm volatile("dcbz %0,%1" : : "r" (blocks), "r" (i) : "memory");
+ }
+ if (misal) {
+ ((unsigned long*)blocks)[188] = 0L;
+ ((unsigned long*)blocks)[189] = 0L;
+ ((unsigned long*)blocks)[190] = 0L;
+ ((unsigned long*)blocks)[191] = 0L;
+ i += 16;
+ }
+#else
+ memset(blocks, 0, sizeof(DCTELEM)*6*64);
+#endif
+POWERPC_TBL_STOP_COUNT(powerpc_clear_blocks_dcbz32, 1);
+}
+
+/* check dcbz report how many bytes are set to 0 by dcbz */
+long check_dcbz_effect(void)
+{
+ register char *fakedata = (char*)malloc(1024);
+ register char *fakedata_middle;
+ register long zero = 0;
+ register long i = 0;
+ long count = 0;
+
+ if (fakedata == NULL)
+ {
+ return 0L;
+ }
+
+
+ fakedata_middle = (fakedata + 512);
+
+ memset(fakedata, 0xFF, 1024);
+
+ asm volatile("dcbz %0, %1" : : "r" (fakedata_middle), "r" (zero));
+
+ for (i = 0; i < 1024 ; i ++)
+ {
+ if (fakedata[i] == (char)0)
+ count++;
+ }
+
+ free(fakedata);
+
+ return count;
+}
+
void dsputil_init_ppc(DSPContext* c, unsigned mask)
{
// Common optimisations whether Altivec or not
- // ... pending ...
-
+ switch (check_dcbz_effect()) {
+ case 32:
+ c->clear_blocks = clear_blocks_dcbz32_ppc;
+ break;
+ default:
+ break;
+ }
+
#if HAVE_ALTIVEC
if (has_altivec()) {
mm_flags |= MM_ALTIVEC;
@@ -67,17 +206,29 @@ void dsputil_init_ppc(DSPContext* c, unsigned mask)
#endif
c->put_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
+ c->avg_pixels_tab[1][0] = avg_pixels8_altivec;
+#endif
+ c->put_pixels_tab[1][3] = put_pixels8_xy2_altivec;
+
c->gmc1 = gmc1_altivec;
-#ifdef ALTIVEC_TBL_PERFORMANCE_REPORT
+#ifdef POWERPC_TBL_PERFORMANCE_REPORT
{
int i;
- for (i = 0 ; i < altivec_perf_total ; i++)
+ for (i = 0 ; i < powerpc_perf_total ; i++)
{
- perfdata[i][altivec_data_min] = 0xFFFFFFFFFFFFFFFF;
- perfdata[i][altivec_data_max] = 0x0000000000000000;
- perfdata[i][altivec_data_sum] = 0x0000000000000000;
- perfdata[i][altivec_data_num] = 0x0000000000000000;
+ 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_miss[i][powerpc_data_min] = 0xFFFFFFFFFFFFFFFF;
+ perfdata_miss[i][powerpc_data_max] = 0x0000000000000000;
+ perfdata_miss[i][powerpc_data_sum] = 0x0000000000000000;
+ perfdata_miss[i][powerpc_data_num] = 0x0000000000000000;
+#endif
}
}
#endif
diff --git a/libavcodec/ppc/dsputil_ppc.h b/libavcodec/ppc/dsputil_ppc.h
new file mode 100644
index 0000000000..e0611ff886
--- /dev/null
+++ b/libavcodec/ppc/dsputil_ppc.h
@@ -0,0 +1,120 @@
+/*
+ * Copyright (c) 2003 Romain Dolbeau <romain@dolbeau.org>
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+
+#ifndef _DSPUTIL_PPC_
+#define _DSPUTIL_PPC_
+
+#ifdef POWERPC_TBL_PERFORMANCE_REPORT
+void powerpc_display_perf_report(void);
+/* if you add to the enum below, also add to the perfname array
+ in dsputil_ppc.c */
+enum powerpc_perf_index {
+ altivec_fft_num = 0,
+ altivec_gmc1_num,
+ altivec_dct_unquantize_h263_num,
+ altivec_idct_add_num,
+ altivec_idct_put_num,
+ altivec_put_pixels_clamped_num,
+ altivec_put_pixels16_num,
+ altivec_avg_pixels16_num,
+ altivec_avg_pixels8_num,
+ altivec_put_pixels8_xy2_num,
+ powerpc_clear_blocks_dcbz32,
+ powerpc_perf_total
+};
+enum powerpc_data_index {
+ powerpc_data_min = 0,
+ powerpc_data_max,
+ powerpc_data_sum,
+ 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_miss[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)
+
+#else /* POWERPC_PERF_USE_PMC */
+#define POWERPC_GET_CYCLES(a) asm volatile("mfspr %0, 937" : "=r" (a))
+#define POWERPC_GET_MISS(a) asm volatile("mfspr %0, 938" : "=r" (a))
+#define POWERPC_TBL_DECLARE(a, cond) register unsigned long cycles_start, cycles_stop, miss_start, miss_stop;
+#define POWERPC_TBL_START_COUNT(a, cond) do { POWERPC_GET_MISS(miss_start); POWERPC_GET_CYCLES(cycles_start); } while (0)
+#define POWERPC_TBL_STOP_COUNT(a, cond) do { \
+ POWERPC_GET_CYCLES(cycles_stop); \
+ POWERPC_GET_MISS(miss_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 (miss_stop >= miss_start) \
+ { \
+ unsigned long diff = \
+ miss_stop - miss_start; \
+ if (cond) \
+ { \
+ if (diff < perfdata_miss[a][powerpc_data_min]) \
+ perfdata_miss[a][powerpc_data_min] = diff; \
+ if (diff > perfdata_miss[a][powerpc_data_max]) \
+ perfdata_miss[a][powerpc_data_max] = diff; \
+ perfdata_miss[a][powerpc_data_sum] += diff; \
+ perfdata_miss[a][powerpc_data_num] ++; \
+ } \
+ } \
+} while (0)
+
+#endif /* POWERPC_PERF_USE_PMC */
+
+
+#else /* POWERPC_TBL_PERFORMANCE_REPORT */
+#define POWERPC_TBL_DECLARE(a, cond)
+#define POWERPC_TBL_START_COUNT(a, cond)
+#define POWERPC_TBL_STOP_COUNT(a, cond)
+#endif /* POWERPC_TBL_PERFORMANCE_REPORT */
+
+#endif /* _DSPUTIL_PPC_ */
diff --git a/libavcodec/ppc/fft_altivec.c b/libavcodec/ppc/fft_altivec.c
index 55bfcdf4d2..db32acd23a 100644
--- a/libavcodec/ppc/fft_altivec.c
+++ b/libavcodec/ppc/fft_altivec.c
@@ -60,7 +60,7 @@
*/
void fft_calc_altivec(FFTContext *s, FFTComplex *z)
{
-ALTIVEC_TBL_DECLARE(altivec_fft_num, s->nbits >= 6);
+POWERPC_TBL_DECLARE(altivec_fft_num, s->nbits >= 6);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
int ln = s->nbits;
int j, np, np2;
@@ -70,7 +70,7 @@ ALTIVEC_TBL_DECLARE(altivec_fft_num, s->nbits >= 6);
int l;
FFTSample tmp_re, tmp_im;
-ALTIVEC_TBL_START_COUNT(altivec_fft_num, s->nbits >= 6);
+POWERPC_TBL_START_COUNT(altivec_fft_num, s->nbits >= 6);
np = 1 << ln;
@@ -135,7 +135,7 @@ ALTIVEC_TBL_START_COUNT(altivec_fft_num, s->nbits >= 6);
nloops = nloops << 1;
} while (nblocks != 0);
-ALTIVEC_TBL_STOP_COUNT(altivec_fft_num, s->nbits >= 6);
+POWERPC_TBL_STOP_COUNT(altivec_fft_num, s->nbits >= 6);
#else /* ALTIVEC_USE_REFERENCE_C_CODE */
register const vector float vczero = (const vector float)(0.);
@@ -147,7 +147,7 @@ ALTIVEC_TBL_STOP_COUNT(altivec_fft_num, s->nbits >= 6);
FFTComplex *cptr, *cptr1;
int k;
-ALTIVEC_TBL_START_COUNT(altivec_fft_num, s->nbits >= 6);
+POWERPC_TBL_START_COUNT(altivec_fft_num, s->nbits >= 6);
np = 1 << ln;
@@ -235,7 +235,7 @@ ALTIVEC_TBL_START_COUNT(altivec_fft_num, s->nbits >= 6);
nloops = nloops << 1;
} while (nblocks != 0);
-ALTIVEC_TBL_STOP_COUNT(altivec_fft_num, s->nbits >= 6);
+POWERPC_TBL_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 c2d908d039..810a237fe7 100644
--- a/libavcodec/ppc/gmc_altivec.c
+++ b/libavcodec/ppc/gmc_altivec.c
@@ -28,7 +28,7 @@
*/
void gmc1_altivec(UINT8 *dst /* align 8 */, UINT8 *src /* align1 */, int stride, int h, int x16, int y16, int rounder)
{
-ALTIVEC_TBL_DECLARE(altivec_gmc1_num, h == 8);
+POWERPC_TBL_DECLARE(altivec_gmc1_num, h == 8);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
const int A=(16-x16)*(16-y16);
const int B=( x16)*(16-y16);
@@ -36,7 +36,7 @@ ALTIVEC_TBL_DECLARE(altivec_gmc1_num, h == 8);
const int D=( x16)*( y16);
int i;
-ALTIVEC_TBL_START_COUNT(altivec_gmc1_num, h == 8);
+POWERPC_TBL_START_COUNT(altivec_gmc1_num, h == 8);
for(i=0; i<h; i++)
{
@@ -52,7 +52,7 @@ ALTIVEC_TBL_START_COUNT(altivec_gmc1_num, h == 8);
src+= stride;
}
-ALTIVEC_TBL_STOP_COUNT(altivec_gmc1_num, h == 8);
+POWERPC_TBL_STOP_COUNT(altivec_gmc1_num, h == 8);
#else /* ALTIVEC_USE_REFERENCE_C_CODE */
const unsigned short __attribute__ ((aligned(16))) rounder_a[8] =
@@ -75,7 +75,7 @@ ALTIVEC_TBL_STOP_COUNT(altivec_gmc1_num, h == 8);
unsigned long src_really_odd = (unsigned long)src & 0x0000000F;
-ALTIVEC_TBL_START_COUNT(altivec_gmc1_num, h == 8);
+POWERPC_TBL_START_COUNT(altivec_gmc1_num, h == 8);
tempA = vec_ld(0, (unsigned short*)ABCD);
Av = vec_splat(tempA, 0);
@@ -163,7 +163,7 @@ ALTIVEC_TBL_START_COUNT(altivec_gmc1_num, h == 8);
src += stride;
}
-ALTIVEC_TBL_STOP_COUNT(altivec_gmc1_num, h == 8);
+POWERPC_TBL_STOP_COUNT(altivec_gmc1_num, h == 8);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}
diff --git a/libavcodec/ppc/idct_altivec.c b/libavcodec/ppc/idct_altivec.c
index 533ab3e4c8..41db01f832 100644
--- a/libavcodec/ppc/idct_altivec.c
+++ b/libavcodec/ppc/idct_altivec.c
@@ -161,16 +161,16 @@ static const vector_s16_t constants[5] = {
void idct_put_altivec(uint8_t* dest, int stride, vector_s16_t* block)
{
-ALTIVEC_TBL_DECLARE(altivec_idct_put_num, 1);
+POWERPC_TBL_DECLARE(altivec_idct_put_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
-ALTIVEC_TBL_START_COUNT(altivec_idct_put_num, 1);
+POWERPC_TBL_START_COUNT(altivec_idct_put_num, 1);
void simple_idct_put(UINT8 *dest, int line_size, INT16 *block);
simple_idct_put(dest, stride, (INT16*)block);
-ALTIVEC_TBL_STOP_COUNT(altivec_idct_put_num, 1);
+POWERPC_TBL_STOP_COUNT(altivec_idct_put_num, 1);
#else /* ALTIVEC_USE_REFERENCE_C_CODE */
vector_u8_t tmp;
-ALTIVEC_TBL_START_COUNT(altivec_idct_put_num, 1);
+POWERPC_TBL_START_COUNT(altivec_idct_put_num, 1);
IDCT
@@ -188,18 +188,18 @@ ALTIVEC_TBL_START_COUNT(altivec_idct_put_num, 1);
COPY (dest, vx6) dest += stride;
COPY (dest, vx7)
-ALTIVEC_TBL_STOP_COUNT(altivec_idct_put_num, 1);
+POWERPC_TBL_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)
{
-ALTIVEC_TBL_DECLARE(altivec_idct_add_num, 1);
+POWERPC_TBL_DECLARE(altivec_idct_add_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
-ALTIVEC_TBL_START_COUNT(altivec_idct_add_num, 1);
+POWERPC_TBL_START_COUNT(altivec_idct_add_num, 1);
void simple_idct_add(UINT8 *dest, int line_size, INT16 *block);
simple_idct_add(dest, stride, (INT16*)block);
-ALTIVEC_TBL_STOP_COUNT(altivec_idct_add_num, 1);
+POWERPC_TBL_STOP_COUNT(altivec_idct_add_num, 1);
#else /* ALTIVEC_USE_REFERENCE_C_CODE */
vector_u8_t tmp;
vector_s16_t tmp2, tmp3;
@@ -207,7 +207,7 @@ ALTIVEC_TBL_STOP_COUNT(altivec_idct_add_num, 1);
vector_u8_t perm1;
vector_u8_t p0, p1, p;
-ALTIVEC_TBL_START_COUNT(altivec_idct_add_num, 1);
+POWERPC_TBL_START_COUNT(altivec_idct_add_num, 1);
IDCT
@@ -235,7 +235,7 @@ ALTIVEC_TBL_START_COUNT(altivec_idct_add_num, 1);
ADD (dest, vx6, perm0) dest += stride;
ADD (dest, vx7, perm1)
-ALTIVEC_TBL_STOP_COUNT(altivec_idct_add_num, 1);
+POWERPC_TBL_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 dfa4c0229a..268aef334f 100644
--- a/libavcodec/ppc/mpegvideo_altivec.c
+++ b/libavcodec/ppc/mpegvideo_altivec.c
@@ -511,13 +511,13 @@ int dct_quantize_altivec(MpegEncContext* s,
void dct_unquantize_h263_altivec(MpegEncContext *s,
DCTELEM *block, int n, int qscale)
{
-ALTIVEC_TBL_DECLARE(altivec_dct_unquantize_h263_num, 1);
+POWERPC_TBL_DECLARE(altivec_dct_unquantize_h263_num, 1);
int i, level, qmul, qadd;
int nCoeffs;
assert(s->block_last_index[n]>=0);
-ALTIVEC_TBL_START_COUNT(altivec_dct_unquantize_h263_num, 1);
+POWERPC_TBL_START_COUNT(altivec_dct_unquantize_h263_num, 1);
qadd = (qscale - 1) | 1;
qmul = qscale << 1;
@@ -630,5 +630,5 @@ ALTIVEC_TBL_START_COUNT(altivec_dct_unquantize_h263_num, 1);
}
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
-ALTIVEC_TBL_STOP_COUNT(altivec_dct_unquantize_h263_num, nCoeffs == 63);
+POWERPC_TBL_STOP_COUNT(altivec_dct_unquantize_h263_num, nCoeffs == 63);
}