comparison x86/dsputil_mmx.c @ 8590:7a463923ecd1 libavcodec

Change semantic of CONFIG_*, HAVE_* and ARCH_*. They are now always defined to either 0 or 1.

8589:a29b5b5c3c9d

        "pabsw     %%mm3, %%mm3 \n"\
        "pabsw     %%mm4, %%mm4 \n"\
        "pabsw     %%mm5, %%mm5 \n"

PAETH(mmx2, ABS3_MMX2)
#ifdef HAVE_SSSE3
PAETH(ssse3, ABS3_SSSE3)
#endif

#define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\
        "paddw " #m4 ", " #m3 "           \n\t" /* x1 */\

void ff_mmx_idct(DCTELEM *block);
void ff_mmxext_idct(DCTELEM *block);

/* XXX: those functions should be suppressed ASAP when all IDCTs are
   converted */
#ifdef CONFIG_GPL
static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
{
    ff_mmx_idct (block);
    put_pixels_clamped_mmx(block, dest, line_size);
}

        ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step);
}

static void vector_fmul_window_3dnow2(float *dst, const float *src0, const float *src1,
                                      const float *win, float add_bias, int len){
#ifdef HAVE_6REGS
    if(add_bias == 0){
        x86_reg i = -len*4;
        x86_reg j = len*4-8;
        __asm__ volatile(
            "1: \n"

        ff_vector_fmul_window_c(dst, src0, src1, win, add_bias, len);
}

static void vector_fmul_window_sse(float *dst, const float *src0, const float *src1,
                                   const float *win, float add_bias, int len){
#ifdef HAVE_6REGS
    if(add_bias == 0){
        x86_reg i = -len*4;
        x86_reg j = len*4-16;
        __asm__ volatile(
            "1: \n"

        " js 1b                             \n\t"
        :"+r"(reglen), "+r"(dst), "+r"(src)
    );
}

#ifdef HAVE_YASM
void ff_float_to_int16_interleave6_sse(int16_t *dst, const float **src, int len);
void ff_float_to_int16_interleave6_3dnow(int16_t *dst, const float **src, int len);
void ff_float_to_int16_interleave6_3dn2(int16_t *dst, const float **src, int len);
void ff_x264_deblock_v_luma_sse2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
void ff_x264_deblock_h_luma_sse2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
void ff_x264_deblock_v8_luma_intra_mmxext(uint8_t *pix, int stride, int alpha, int beta);
void ff_x264_deblock_h_luma_intra_mmxext(uint8_t *pix, int stride, int alpha, int beta);
#ifdef ARCH_X86_32
static void ff_x264_deblock_v_luma_intra_mmxext(uint8_t *pix, int stride, int alpha, int beta)
{
    ff_x264_deblock_v8_luma_intra_mmxext(pix+0, stride, alpha, beta);
    ff_x264_deblock_v8_luma_intra_mmxext(pix+8, stride, alpha, beta);
}

            if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){
                c->idct_put= ff_simple_idct_put_mmx;
                c->idct_add= ff_simple_idct_add_mmx;
                c->idct    = ff_simple_idct_mmx;
                c->idct_permutation_type= FF_SIMPLE_IDCT_PERM;
#ifdef CONFIG_GPL
            }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){
                if(mm_flags & FF_MM_MMXEXT){
                    c->idct_put= ff_libmpeg2mmx2_idct_put;
                    c->idct_add= ff_libmpeg2mmx2_idct_add;
                    c->idct    = ff_mmxext_idct;

            H264_QPEL_FUNCS(2, 3, sse2);
            H264_QPEL_FUNCS(3, 1, sse2);
            H264_QPEL_FUNCS(3, 2, sse2);
            H264_QPEL_FUNCS(3, 3, sse2);
        }
#ifdef HAVE_SSSE3
        if(mm_flags & FF_MM_SSSE3){
            H264_QPEL_FUNCS(1, 0, ssse3);
            H264_QPEL_FUNCS(1, 1, ssse3);
            H264_QPEL_FUNCS(1, 2, ssse3);
            H264_QPEL_FUNCS(1, 3, ssse3);

            c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_ssse3;
            c->add_png_paeth_prediction= add_png_paeth_prediction_ssse3;
        }
#endif

#if defined(CONFIG_GPL) && defined(HAVE_YASM)
        if( mm_flags&FF_MM_MMXEXT ){
#ifdef ARCH_X86_32
            c->h264_v_loop_filter_luma_intra = ff_x264_deblock_v_luma_intra_mmxext;
            c->h264_h_loop_filter_luma_intra = ff_x264_deblock_h_luma_intra_mmxext;
#endif
            if( mm_flags&FF_MM_SSE2 ){
#if defined(ARCH_X86_64) || !defined(__ICC) || __ICC > 1100
                c->h264_v_loop_filter_luma = ff_x264_deblock_v_luma_sse2;
                c->h264_h_loop_filter_luma = ff_x264_deblock_h_luma_sse2;
                c->h264_v_loop_filter_luma_intra = ff_x264_deblock_v_luma_intra_sse2;
                c->h264_h_loop_filter_luma_intra = ff_x264_deblock_h_luma_intra_sse2;
#endif

                c->h264_idct_add16intra = ff_h264_idct_add16intra_sse2;
            }
        }
#endif

#ifdef CONFIG_SNOW_DECODER
        if(mm_flags & FF_MM_SSE2 & 0){
            c->horizontal_compose97i = ff_snow_horizontal_compose97i_sse2;
#ifdef HAVE_7REGS
            c->vertical_compose97i = ff_snow_vertical_compose97i_sse2;
#endif
            c->inner_add_yblock = ff_snow_inner_add_yblock_sse2;
        }
        else{
            if(mm_flags & FF_MM_MMXEXT){
            c->horizontal_compose97i = ff_snow_horizontal_compose97i_mmx;
#ifdef HAVE_7REGS
            c->vertical_compose97i = ff_snow_vertical_compose97i_mmx;
#endif
            }
            c->inner_add_yblock = ff_snow_inner_add_yblock_mmx;
        }

8590:7a463923ecd1

        "pabsw     %%mm3, %%mm3 \n"\
        "pabsw     %%mm4, %%mm4 \n"\
        "pabsw     %%mm5, %%mm5 \n"

PAETH(mmx2, ABS3_MMX2)
#if HAVE_SSSE3
PAETH(ssse3, ABS3_SSSE3)
#endif

#define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\
        "paddw " #m4 ", " #m3 "           \n\t" /* x1 */\

void ff_mmx_idct(DCTELEM *block);
void ff_mmxext_idct(DCTELEM *block);

/* XXX: those functions should be suppressed ASAP when all IDCTs are
   converted */
#if CONFIG_GPL
static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
{
    ff_mmx_idct (block);
    put_pixels_clamped_mmx(block, dest, line_size);
}

        ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step);
}

static void vector_fmul_window_3dnow2(float *dst, const float *src0, const float *src1,
                                      const float *win, float add_bias, int len){
#if HAVE_6REGS
    if(add_bias == 0){
        x86_reg i = -len*4;
        x86_reg j = len*4-8;
        __asm__ volatile(
            "1: \n"

        ff_vector_fmul_window_c(dst, src0, src1, win, add_bias, len);
}

static void vector_fmul_window_sse(float *dst, const float *src0, const float *src1,
                                   const float *win, float add_bias, int len){
#if HAVE_6REGS
    if(add_bias == 0){
        x86_reg i = -len*4;
        x86_reg j = len*4-16;
        __asm__ volatile(
            "1: \n"

        " js 1b                             \n\t"
        :"+r"(reglen), "+r"(dst), "+r"(src)
    );
}

#if HAVE_YASM
void ff_float_to_int16_interleave6_sse(int16_t *dst, const float **src, int len);
void ff_float_to_int16_interleave6_3dnow(int16_t *dst, const float **src, int len);
void ff_float_to_int16_interleave6_3dn2(int16_t *dst, const float **src, int len);
void ff_x264_deblock_v_luma_sse2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
void ff_x264_deblock_h_luma_sse2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
void ff_x264_deblock_v8_luma_intra_mmxext(uint8_t *pix, int stride, int alpha, int beta);
void ff_x264_deblock_h_luma_intra_mmxext(uint8_t *pix, int stride, int alpha, int beta);
#if ARCH_X86_32
static void ff_x264_deblock_v_luma_intra_mmxext(uint8_t *pix, int stride, int alpha, int beta)
{
    ff_x264_deblock_v8_luma_intra_mmxext(pix+0, stride, alpha, beta);
    ff_x264_deblock_v8_luma_intra_mmxext(pix+8, stride, alpha, beta);
}

            if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){
                c->idct_put= ff_simple_idct_put_mmx;
                c->idct_add= ff_simple_idct_add_mmx;
                c->idct    = ff_simple_idct_mmx;
                c->idct_permutation_type= FF_SIMPLE_IDCT_PERM;
#if CONFIG_GPL
            }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){
                if(mm_flags & FF_MM_MMXEXT){
                    c->idct_put= ff_libmpeg2mmx2_idct_put;
                    c->idct_add= ff_libmpeg2mmx2_idct_add;
                    c->idct    = ff_mmxext_idct;

            H264_QPEL_FUNCS(2, 3, sse2);
            H264_QPEL_FUNCS(3, 1, sse2);
            H264_QPEL_FUNCS(3, 2, sse2);
            H264_QPEL_FUNCS(3, 3, sse2);
        }
#if HAVE_SSSE3
        if(mm_flags & FF_MM_SSSE3){
            H264_QPEL_FUNCS(1, 0, ssse3);
            H264_QPEL_FUNCS(1, 1, ssse3);
            H264_QPEL_FUNCS(1, 2, ssse3);
            H264_QPEL_FUNCS(1, 3, ssse3);

            c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_ssse3;
            c->add_png_paeth_prediction= add_png_paeth_prediction_ssse3;
        }
#endif

#if CONFIG_GPL && HAVE_YASM
        if( mm_flags&FF_MM_MMXEXT ){
#if ARCH_X86_32
            c->h264_v_loop_filter_luma_intra = ff_x264_deblock_v_luma_intra_mmxext;
            c->h264_h_loop_filter_luma_intra = ff_x264_deblock_h_luma_intra_mmxext;
#endif
            if( mm_flags&FF_MM_SSE2 ){
#if ARCH_X86_64 || !defined(__ICC) || __ICC > 1100
                c->h264_v_loop_filter_luma = ff_x264_deblock_v_luma_sse2;
                c->h264_h_loop_filter_luma = ff_x264_deblock_h_luma_sse2;
                c->h264_v_loop_filter_luma_intra = ff_x264_deblock_v_luma_intra_sse2;
                c->h264_h_loop_filter_luma_intra = ff_x264_deblock_h_luma_intra_sse2;
#endif

                c->h264_idct_add16intra = ff_h264_idct_add16intra_sse2;
            }
        }
#endif

#if CONFIG_SNOW_DECODER
        if(mm_flags & FF_MM_SSE2 & 0){
            c->horizontal_compose97i = ff_snow_horizontal_compose97i_sse2;
#if HAVE_7REGS
            c->vertical_compose97i = ff_snow_vertical_compose97i_sse2;
#endif
            c->inner_add_yblock = ff_snow_inner_add_yblock_sse2;
        }
        else{
            if(mm_flags & FF_MM_MMXEXT){
            c->horizontal_compose97i = ff_snow_horizontal_compose97i_mmx;
#if HAVE_7REGS
            c->vertical_compose97i = ff_snow_vertical_compose97i_mmx;
#endif
            }
            c->inner_add_yblock = ff_snow_inner_add_yblock_mmx;
        }