snowdsp_mmx.c

mediastreamer2是开源的网络传输媒体流的库

                "movq    8(%1), %%mm4        \n\t"
                "movq    2(%1), %%mm1        \n\t"
                "movq   10(%1), %%mm5        \n\t"
                "paddw   %%mm6, %%mm0        \n\t"
                "paddw   %%mm6, %%mm4        \n\t"
                "paddw   %%mm7, %%mm1        \n\t"
                "paddw   %%mm7, %%mm5        \n\t"
                "pavgw   %%mm1, %%mm0        \n\t"
                "pavgw   %%mm5, %%mm4        \n\t"
                "psubw   %%mm7, %%mm0        \n\t"
                "psubw   %%mm7, %%mm4        \n\t"
                "psraw      $1, %%mm0        \n\t"
                "psraw      $1, %%mm4        \n\t"
                "movq     (%0), %%mm1        \n\t"
                "movq    8(%0), %%mm5        \n\t"
                "paddw   %%mm1, %%mm0        \n\t"
                "paddw   %%mm5, %%mm4        \n\t"
                "psraw      $2, %%mm0        \n\t"
                "psraw      $2, %%mm4        \n\t"
                "paddw   %%mm1, %%mm0        \n\t"
                "paddw   %%mm5, %%mm4        \n\t"
                "movq    %%mm0, (%0)         \n\t"
                "movq    %%mm4, 8(%0)        \n\t"
                :: "r"(&b[i]), "r"(&ref[i])
                 : "memory"
               );
        }
        snow_horizontal_compose_liftS_lead_out(i, b, b, ref, width, w_l);
    }

    { // Lift 3
        IDWTELEM * const src = b+w2;
        i = 0;

        for(; i<w_r-7; i+=8){
            asm volatile(
                "movq    2(%1), %%mm2        \n\t"
                "movq   10(%1), %%mm6        \n\t"
                "paddw    (%1), %%mm2        \n\t"
                "paddw   8(%1), %%mm6        \n\t"
                "movq     (%0), %%mm0        \n\t"
                "movq    8(%0), %%mm4        \n\t"
                "paddw   %%mm2, %%mm0        \n\t"
                "paddw   %%mm6, %%mm4        \n\t"
                "psraw      $1, %%mm2        \n\t"
                "psraw      $1, %%mm6        \n\t"
                "paddw   %%mm0, %%mm2        \n\t"
                "paddw   %%mm4, %%mm6        \n\t"
                "movq    %%mm2, (%2)         \n\t"
                "movq    %%mm6, 8(%2)        \n\t"
                :: "r"(&src[i]), "r"(&b[i]), "r"(&temp[i])
                 : "memory"
               );
        }
        snow_horizontal_compose_lift_lead_out(i, temp, src, b, width, w_r, 1, -W_AM, W_AO+1, W_AS);
    }

    {
        snow_interleave_line_header(&i, width, b, temp);

        for (; (i & 0x1E) != 0x1E; i-=2){
            b[i+1] = temp[i>>1];
            b[i] = b[i>>1];
        }
        for (i-=30; i>=0; i-=32){
            asm volatile(
                "movq        (%1), %%mm0       \n\t"
                "movq       8(%1), %%mm2       \n\t"
                "movq      16(%1), %%mm4       \n\t"
                "movq      24(%1), %%mm6       \n\t"
                "movq        (%1), %%mm1       \n\t"
                "movq       8(%1), %%mm3       \n\t"
                "movq      16(%1), %%mm5       \n\t"
                "movq      24(%1), %%mm7       \n\t"
                "punpcklwd   (%2), %%mm0       \n\t"
                "punpcklwd  8(%2), %%mm2       \n\t"
                "punpcklwd 16(%2), %%mm4       \n\t"
                "punpcklwd 24(%2), %%mm6       \n\t"
                "movq       %%mm0, (%0)        \n\t"
                "movq       %%mm2, 16(%0)      \n\t"
                "movq       %%mm4, 32(%0)      \n\t"
                "movq       %%mm6, 48(%0)      \n\t"
                "punpckhwd   (%2), %%mm1       \n\t"
                "punpckhwd  8(%2), %%mm3       \n\t"
                "punpckhwd 16(%2), %%mm5       \n\t"
                "punpckhwd 24(%2), %%mm7       \n\t"
                "movq       %%mm1, 8(%0)       \n\t"
                "movq       %%mm3, 24(%0)      \n\t"
                "movq       %%mm5, 40(%0)      \n\t"
                "movq       %%mm7, 56(%0)      \n\t"
                :: "r"(&b[i]), "r"(&b[i>>1]), "r"(&temp[i>>1])
                 : "memory"
               );
        }
    }
}

#ifdef HAVE_7REGS
#define snow_vertical_compose_sse2_load_add(op,r,t0,t1,t2,t3)\
        ""op" ("r",%%"REG_d"), %%"t0"      \n\t"\
        ""op" 16("r",%%"REG_d"), %%"t1"    \n\t"\
        ""op" 32("r",%%"REG_d"), %%"t2"    \n\t"\
        ""op" 48("r",%%"REG_d"), %%"t3"    \n\t"

#define snow_vertical_compose_sse2_load(r,t0,t1,t2,t3)\
        snow_vertical_compose_sse2_load_add("movdqa",r,t0,t1,t2,t3)

#define snow_vertical_compose_sse2_add(r,t0,t1,t2,t3)\
        snow_vertical_compose_sse2_load_add("paddw",r,t0,t1,t2,t3)

#define snow_vertical_compose_r2r_sub(s0,s1,s2,s3,t0,t1,t2,t3)\
        "psubw %%"s0", %%"t0" \n\t"\
        "psubw %%"s1", %%"t1" \n\t"\
        "psubw %%"s2", %%"t2" \n\t"\
        "psubw %%"s3", %%"t3" \n\t"

#define snow_vertical_compose_sse2_store(w,s0,s1,s2,s3)\
        "movdqa %%"s0", ("w",%%"REG_d")      \n\t"\
        "movdqa %%"s1", 16("w",%%"REG_d")    \n\t"\
        "movdqa %%"s2", 32("w",%%"REG_d")    \n\t"\
        "movdqa %%"s3", 48("w",%%"REG_d")    \n\t"

#define snow_vertical_compose_sra(n,t0,t1,t2,t3)\
        "psraw $"n", %%"t0" \n\t"\
        "psraw $"n", %%"t1" \n\t"\
        "psraw $"n", %%"t2" \n\t"\
        "psraw $"n", %%"t3" \n\t"

#define snow_vertical_compose_r2r_add(s0,s1,s2,s3,t0,t1,t2,t3)\
        "paddw %%"s0", %%"t0" \n\t"\
        "paddw %%"s1", %%"t1" \n\t"\
        "paddw %%"s2", %%"t2" \n\t"\
        "paddw %%"s3", %%"t3" \n\t"

#define snow_vertical_compose_r2r_pmulhw(s0,s1,s2,s3,t0,t1,t2,t3)\
        "pmulhw %%"s0", %%"t0" \n\t"\
        "pmulhw %%"s1", %%"t1" \n\t"\
        "pmulhw %%"s2", %%"t2" \n\t"\
        "pmulhw %%"s3", %%"t3" \n\t"

#define snow_vertical_compose_sse2_move(s0,s1,s2,s3,t0,t1,t2,t3)\
        "movdqa %%"s0", %%"t0" \n\t"\
        "movdqa %%"s1", %%"t1" \n\t"\
        "movdqa %%"s2", %%"t2" \n\t"\
        "movdqa %%"s3", %%"t3" \n\t"

void ff_snow_vertical_compose97i_sse2(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){
    long i = width;

    while(i & 0x1F)
    {
        i--;
        b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;
        b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS;
        b2[i] += (W_BM*(b1[i] + b3[i])+4*b2[i]+W_BO)>>W_BS;
        b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
    }
    i+=i;

         asm volatile (
        "jmp 2f                                      \n\t"
        "1:                                          \n\t"
        snow_vertical_compose_sse2_load("%4","xmm0","xmm2","xmm4","xmm6")
        snow_vertical_compose_sse2_add("%6","xmm0","xmm2","xmm4","xmm6")


        "pcmpeqw    %%xmm0, %%xmm0                   \n\t"
        "pcmpeqw    %%xmm2, %%xmm2                   \n\t"
        "paddw      %%xmm2, %%xmm2                   \n\t"
        "paddw      %%xmm0, %%xmm2                   \n\t"
        "psllw         $13, %%xmm2                   \n\t"
        snow_vertical_compose_r2r_add("xmm0","xmm0","xmm0","xmm0","xmm1","xmm3","xmm5","xmm7")
        snow_vertical_compose_r2r_pmulhw("xmm2","xmm2","xmm2","xmm2","xmm1","xmm3","xmm5","xmm7")
        snow_vertical_compose_sse2_add("%5","xmm1","xmm3","xmm5","xmm7")
        snow_vertical_compose_sse2_store("%5","xmm1","xmm3","xmm5","xmm7")
        snow_vertical_compose_sse2_load("%4","xmm0","xmm2","xmm4","xmm6")
        snow_vertical_compose_sse2_add("%3","xmm1","xmm3","xmm5","xmm7")
        snow_vertical_compose_r2r_sub("xmm1","xmm3","xmm5","xmm7","xmm0","xmm2","xmm4","xmm6")
        snow_vertical_compose_sse2_store("%4","xmm0","xmm2","xmm4","xmm6")

        "pcmpeqw %%xmm7, %%xmm7                      \n\t"
        "pcmpeqw %%xmm5, %%xmm5                      \n\t"
        "psllw $15, %%xmm7                           \n\t"
        "psrlw $13, %%xmm5                           \n\t"
        "paddw %%xmm7, %%xmm5                        \n\t"
        snow_vertical_compose_r2r_add("xmm5","xmm5","xmm5","xmm5","xmm0","xmm2","xmm4","xmm6")
        "movq   (%2,%%"REG_d"), %%xmm1        \n\t"
        "movq  8(%2,%%"REG_d"), %%xmm3        \n\t"
        "paddw %%xmm7, %%xmm1                        \n\t"
        "paddw %%xmm7, %%xmm3                        \n\t"
        "pavgw %%xmm1, %%xmm0                        \n\t"
        "pavgw %%xmm3, %%xmm2                        \n\t"
        "movq 16(%2,%%"REG_d"), %%xmm1        \n\t"
        "movq 24(%2,%%"REG_d"), %%xmm3        \n\t"
        "paddw %%xmm7, %%xmm1                        \n\t"
        "paddw %%xmm7, %%xmm3                        \n\t"
        "pavgw %%xmm1, %%xmm4                        \n\t"
        "pavgw %%xmm3, %%xmm6                        \n\t"
        snow_vertical_compose_r2r_sub("xmm7","xmm7","xmm7","xmm7","xmm0","xmm2","xmm4","xmm6")
        snow_vertical_compose_sra("1","xmm0","xmm2","xmm4","xmm6")
        snow_vertical_compose_sse2_add("%3","xmm0","xmm2","xmm4","xmm6")

        snow_vertical_compose_sra("2","xmm0","xmm2","xmm4","xmm6")
        snow_vertical_compose_sse2_add("%3","xmm0","xmm2","xmm4","xmm6")
        snow_vertical_compose_sse2_store("%3","xmm0","xmm2","xmm4","xmm6")
        snow_vertical_compose_sse2_add("%1","xmm0","xmm2","xmm4","xmm6")
        snow_vertical_compose_sse2_move("xmm0","xmm2","xmm4","xmm6","xmm1","xmm3","xmm5","xmm7")
        snow_vertical_compose_sra("1","xmm0","xmm2","xmm4","xmm6")
        snow_vertical_compose_r2r_add("xmm1","xmm3","xmm5","xmm7","xmm0","xmm2","xmm4","xmm6")
        snow_vertical_compose_sse2_add("%2","xmm0","xmm2","xmm4","xmm6")
        snow_vertical_compose_sse2_store("%2","xmm0","xmm2","xmm4","xmm6")

        "2:                                          \n\t"
        "sub $64, %%"REG_d"                          \n\t"
        "jge 1b                                      \n\t"
        :"+d"(i)
        :"r"(b0),"r"(b1),"r"(b2),"r"(b3),"r"(b4),"r"(b5));
}

#define snow_vertical_compose_mmx_load_add(op,r,t0,t1,t2,t3)\
        ""op" ("r",%%"REG_d"), %%"t0"   \n\t"\
        ""op" 8("r",%%"REG_d"), %%"t1"  \n\t"\
        ""op" 16("r",%%"REG_d"), %%"t2" \n\t"\
        ""op" 24("r",%%"REG_d"), %%"t3" \n\t"

#define snow_vertical_compose_mmx_load(r,t0,t1,t2,t3)\
        snow_vertical_compose_mmx_load_add("movq",r,t0,t1,t2,t3)

#define snow_vertical_compose_mmx_add(r,t0,t1,t2,t3)\
        snow_vertical_compose_mmx_load_add("paddw",r,t0,t1,t2,t3)

#define snow_vertical_compose_mmx_store(w,s0,s1,s2,s3)\
        "movq %%"s0", ("w",%%"REG_d")   \n\t"\
        "movq %%"s1", 8("w",%%"REG_d")  \n\t"\
        "movq %%"s2", 16("w",%%"REG_d") \n\t"\
        "movq %%"s3", 24("w",%%"REG_d") \n\t"

#define snow_vertical_compose_mmx_move(s0,s1,s2,s3,t0,t1,t2,t3)\
        "movq %%"s0", %%"t0" \n\t"\
        "movq %%"s1", %%"t1" \n\t"\
        "movq %%"s2", %%"t2" \n\t"\
        "movq %%"s3", %%"t3" \n\t"


void ff_snow_vertical_compose97i_mmx(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){
    long i = width;
    while(i & 15)
    {
        i--;
        b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;
        b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS;
        b2[i] += (W_BM*(b1[i] + b3[i])+4*b2[i]+W_BO)>>W_BS;
        b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
    }
    i+=i;
    asm volatile(
        "jmp 2f                                      \n\t"
        "1:                                          \n\t"

        snow_vertical_compose_mmx_load("%4","mm1","mm3","mm5","mm7")
        snow_vertical_compose_mmx_add("%6","mm1","mm3","mm5","mm7")
        "pcmpeqw    %%mm0, %%mm0                     \n\t"
        "pcmpeqw    %%mm2, %%mm2                     \n\t"
        "paddw      %%mm2, %%mm2                     \n\t"
        "paddw      %%mm0, %%mm2                     \n\t"
        "psllw        $13, %%mm2                     \n\t"
        snow_vertical_compose_r2r_add("mm0","mm0","mm0","mm0","mm1","mm3","mm5","mm7")
        snow_vertical_compose_r2r_pmulhw("mm2","mm2","mm2","mm2","mm1","mm3","mm5","mm7")
        snow_vertical_compose_mmx_add("%5","mm1","mm3","mm5","mm7")
        snow_vertical_compose_mmx_store("%5","mm1","mm3","mm5","mm7")
        snow_vertical_compose_mmx_load("%4","mm0","mm2","mm4","mm6")
        snow_vertical_compose_mmx_add("%3","mm1","mm3","mm5","mm7")
        snow_vertical_compose_r2r_sub("mm1","mm3","mm5","mm7","mm0","mm2","mm4","mm6")
        snow_vertical_compose_mmx_store("%4","mm0","mm2","mm4","mm6")
        "pcmpeqw %%mm7, %%mm7                        \n\t"
        "pcmpeqw %%mm5, %%mm5                        \n\t"
        "psllw $15, %%mm7                            \n\t"
        "psrlw $13, %%mm5                            \n\t"
        "paddw %%mm7, %%mm5                          \n\t"
        snow_vertical_compose_r2r_add("mm5","mm5","mm5","mm5","mm0","mm2","mm4","mm6")
        "movq   (%2,%%"REG_d"), %%mm1         \n\t"
        "movq  8(%2,%%"REG_d"), %%mm3         \n\t"
        "paddw %%mm7, %%mm1                          \n\t"
        "paddw %%mm7, %%mm3                          \n\t"
        "pavgw %%mm1, %%mm0                          \n\t"
        "pavgw %%mm3, %%mm2                          \n\t"
        "movq 16(%2,%%"REG_d"), %%mm1         \n\t"
        "movq 24(%2,%%"REG_d"), %%mm3         \n\t"
        "paddw %%mm7, %%mm1                          \n\t"
        "paddw %%mm7, %%mm3                          \n\t"
        "pavgw %%mm1, %%mm4                          \n\t"