diff options
| -rw-r--r-- | libavcodec/Makefile | 3 | ||||
| -rw-r--r-- | libavcodec/dsputil.h | 7 | ||||
| -rw-r--r-- | libavcodec/i386/dsputil_mmx.c | 12 | ||||
| -rw-r--r-- | libavcodec/vp3.c | 8 |
4 files changed, 23 insertions, 7 deletions
diff --git a/libavcodec/Makefile b/libavcodec/Makefile index e3a2429ec7..3b8f14b338 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -116,7 +116,8 @@ ifeq ($(TARGET_MMX),yes) OBJS += i386/fdct_mmx.o i386/cputest.o \ i386/dsputil_mmx.o i386/mpegvideo_mmx.o \ i386/idct_mmx.o i386/motion_est_mmx.o \ - i386/simple_idct_mmx.o i386/fft_sse.o i386/vp3dsp_mmx.o + i386/simple_idct_mmx.o i386/fft_sse.o i386/vp3dsp_mmx.o \ + i386/vp3dsp_sse2.o ifdef TARGET_BUILTIN_VECTOR i386/fft_sse.o: CFLAGS+= -msse depend: CFLAGS+= -msse diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h index d9266581ea..9ad407db8e 100644 --- a/libavcodec/dsputil.h +++ b/libavcodec/dsputil.h @@ -73,6 +73,12 @@ void vp3_idct_put_mmx(int16_t *input_data, int16_t *dequant_matrix, void vp3_idct_add_mmx(int16_t *input_data, int16_t *dequant_matrix, int coeff_count, uint8_t *dest, int stride); +void vp3_dsp_init_sse2(void); +void vp3_idct_put_sse2(int16_t *input_data, int16_t *dequant_matrix, + int coeff_count, uint8_t *dest, int stride); +void vp3_idct_add_sse2(int16_t *input_data, int16_t *dequant_matrix, + int coeff_count, uint8_t *dest, int stride); + /* minimum alignment rules ;) if u notice errors in the align stuff, need more alignment for some asm code for some cpu @@ -403,6 +409,7 @@ static inline void emms(void) } #define __align8 __attribute__ ((aligned (8))) +#define __align16 __attribute__ ((aligned (16))) void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx); void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx); diff --git a/libavcodec/i386/dsputil_mmx.c b/libavcodec/i386/dsputil_mmx.c index 15dc8eec2e..772c9c1f03 100644 --- a/libavcodec/i386/dsputil_mmx.c +++ b/libavcodec/i386/dsputil_mmx.c @@ -2147,9 +2147,15 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx) } /* VP3 optimized DSP functions */ - c->vp3_dsp_init = vp3_dsp_init_mmx; - c->vp3_idct_put = vp3_idct_put_mmx; - c->vp3_idct_add = vp3_idct_add_mmx; + if (mm_flags & MM_SSE2) { + c->vp3_dsp_init = vp3_dsp_init_sse2; + c->vp3_idct_put = vp3_idct_put_sse2; + c->vp3_idct_add = vp3_idct_add_sse2; + } else { + c->vp3_dsp_init = vp3_dsp_init_mmx; + c->vp3_idct_put = vp3_idct_put_mmx; + c->vp3_idct_add = vp3_idct_add_mmx; + } #ifdef CONFIG_ENCODERS c->get_pixels = get_pixels_mmx; diff --git a/libavcodec/vp3.c b/libavcodec/vp3.c index eadfd39b95..0667d99eb8 100644 --- a/libavcodec/vp3.c +++ b/libavcodec/vp3.c @@ -268,9 +268,11 @@ typedef struct Vp3DecodeContext { VLC ac_vlc_3[16]; VLC ac_vlc_4[16]; - int16_t intra_y_dequant[64]; - int16_t intra_c_dequant[64]; - int16_t inter_dequant[64]; + /* these arrays need to be on 16-byte boundaries since SSE2 operations + * index into them */ + int16_t __align16 intra_y_dequant[64]; + int16_t __align16 intra_c_dequant[64]; + int16_t __align16 inter_dequant[64]; /* This table contains superblock_count * 16 entries. Each set of 16 * numbers corresponds to the fragment indices 0..15 of the superblock. |