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
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
}
#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);
}
/* 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;
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.