comparison i386/dsputil_mmx.c @ 7564:7cf793954871 libavcodec

simd int->float 20% faster ac3 if downmixing, 15% if not

7563:8390efaa0c03

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

static void float_to_int16_3dnow(int16_t *dst, const float *src, long len){
    // not bit-exact: pf2id uses different rounding than C and SSE
    asm volatile(
        "add        %0          , %0        \n\t"
        "lea         (%2,%0,2)  , %2        \n\t"

            c->ac3_downmix = ac3_downmix_sse;
            c->vector_fmul = vector_fmul_sse;
            c->vector_fmul_reverse = vector_fmul_reverse_sse;
            c->vector_fmul_add_add = vector_fmul_add_add_sse;
            c->vector_fmul_window = vector_fmul_window_sse;
            c->float_to_int16 = float_to_int16_sse;
            c->float_to_int16_interleave = float_to_int16_interleave_sse;
        }
        if(mm_flags & MM_3DNOW)
            c->vector_fmul_add_add = vector_fmul_add_add_3dnow; // faster than sse
        if(mm_flags & MM_SSE2){
            c->float_to_int16 = float_to_int16_sse2;
            c->float_to_int16_interleave = float_to_int16_interleave_sse2;
            c->add_int16 = add_int16_sse2;
            c->sub_int16 = sub_int16_sse2;
            c->scalarproduct_int16 = scalarproduct_int16_sse2;

7564:7cf793954871

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

static void int32_to_float_fmul_scalar_sse(float *dst, const int *src, float mul, int len)
{
    x86_reg i = -4*len;
    asm volatile(
        "movss  %3, %%xmm4 \n"
        "shufps $0, %%xmm4, %%xmm4 \n"
        "1: \n"
        "cvtpi2ps   (%2,%0), %%xmm0 \n"
        "cvtpi2ps  8(%2,%0), %%xmm1 \n"
        "cvtpi2ps 16(%2,%0), %%xmm2 \n"
        "cvtpi2ps 24(%2,%0), %%xmm3 \n"
        "movlhps  %%xmm1,    %%xmm0 \n"
        "movlhps  %%xmm3,    %%xmm2 \n"
        "mulps    %%xmm4,    %%xmm0 \n"
        "mulps    %%xmm4,    %%xmm2 \n"
        "movaps   %%xmm0,   (%1,%0) \n"
        "movaps   %%xmm2, 16(%1,%0) \n"
        "add $32, %0 \n"
        "jl 1b \n"
        :"+r"(i)
        :"r"(dst+len), "r"(src+len), "xm"(mul)
    );
}

static void int32_to_float_fmul_scalar_sse2(float *dst, const int *src, float mul, int len)
{
    x86_reg i = -4*len;
    asm volatile(
        "movss  %3, %%xmm4 \n"
        "shufps $0, %%xmm4, %%xmm4 \n"
        "1: \n"
        "cvtdq2ps   (%2,%0), %%xmm0 \n"
        "cvtdq2ps 16(%2,%0), %%xmm1 \n"
        "mulps    %%xmm4,    %%xmm0 \n"
        "mulps    %%xmm4,    %%xmm1 \n"
        "movaps   %%xmm0,   (%1,%0) \n"
        "movaps   %%xmm1, 16(%1,%0) \n"
        "add $32, %0 \n"
        "jl 1b \n"
        :"+r"(i)
        :"r"(dst+len), "r"(src+len), "xm"(mul)
    );
}

static void float_to_int16_3dnow(int16_t *dst, const float *src, long len){
    // not bit-exact: pf2id uses different rounding than C and SSE
    asm volatile(
        "add        %0          , %0        \n\t"
        "lea         (%2,%0,2)  , %2        \n\t"

            c->ac3_downmix = ac3_downmix_sse;
            c->vector_fmul = vector_fmul_sse;
            c->vector_fmul_reverse = vector_fmul_reverse_sse;
            c->vector_fmul_add_add = vector_fmul_add_add_sse;
            c->vector_fmul_window = vector_fmul_window_sse;
            c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_sse;
            c->float_to_int16 = float_to_int16_sse;
            c->float_to_int16_interleave = float_to_int16_interleave_sse;
        }
        if(mm_flags & MM_3DNOW)
            c->vector_fmul_add_add = vector_fmul_add_add_3dnow; // faster than sse
        if(mm_flags & MM_SSE2){
            c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_sse2;
            c->float_to_int16 = float_to_int16_sse2;
            c->float_to_int16_interleave = float_to_int16_interleave_sse2;
            c->add_int16 = add_int16_sse2;
            c->sub_int16 = sub_int16_sse2;
            c->scalarproduct_int16 = scalarproduct_int16_sse2;