Convert ff_imdct_half_sse() to yasm.

This is to avoid split asm sections that attempt to preserve some
registers between sections.

Originally committed as revision 24869 to svn://svn.ffmpeg.org/ffmpeg/trunk

1 files changed, 0 insertions, 101 deletions

diff --git a/libavcodec/x86/fft_sse.c b/libavcodec/x86/fft_sse.c
index a4cce69db1..92648b5ddb 100644
--- a/libavcodec/x86/fft_sse.c
+++ b/libavcodec/x86/fft_sse.c
@@ -71,107 +71,6 @@ void ff_fft_permute_sse(FFTContext *s, FFTComplex *z)
     memcpy(z, s->tmp_buf, n*sizeof(FFTComplex));
 }
 
-void ff_imdct_half_sse(FFTContext *s, FFTSample *output, const FFTSample *input)
-{
-    av_unused x86_reg i, j, k, l;
-    long n = 1 << s->mdct_bits;
-    long n2 = n >> 1;
-    long n4 = n >> 2;
-    long n8 = n >> 3;
-    const uint16_t *revtab = s->revtab + n8;
-    const FFTSample *tcos = s->tcos;
-    const FFTSample *tsin = s->tsin;
-    FFTComplex *z = (FFTComplex *)output;
-
-    /* pre rotation */
-    for(k=n8-2; k>=0; k-=2) {
-        __asm__ volatile(
-            "movaps     (%2,%1,2), %%xmm0 \n" // { z[k].re,    z[k].im,    z[k+1].re,  z[k+1].im  }
-            "movaps  -16(%2,%0,2), %%xmm1 \n" // { z[-k-2].re, z[-k-2].im, z[-k-1].re, z[-k-1].im }
-            "movaps        %%xmm0, %%xmm2 \n"
-            "shufps $0x88, %%xmm1, %%xmm0 \n" // { z[k].re,    z[k+1].re,  z[-k-2].re, z[-k-1].re }
-            "shufps $0x77, %%xmm2, %%xmm1 \n" // { z[-k-1].im, z[-k-2].im, z[k+1].im,  z[k].im    }
-            "movlps       (%3,%1), %%xmm4 \n"
-            "movlps       (%4,%1), %%xmm5 \n"
-            "movhps     -8(%3,%0), %%xmm4 \n" // { cos[k],     cos[k+1],   cos[-k-2],  cos[-k-1]  }
-            "movhps     -8(%4,%0), %%xmm5 \n" // { sin[k],     sin[k+1],   sin[-k-2],  sin[-k-1]  }
-            "movaps        %%xmm0, %%xmm2 \n"
-            "movaps        %%xmm1, %%xmm3 \n"
-            "mulps         %%xmm5, %%xmm0 \n" // re*sin
-            "mulps         %%xmm4, %%xmm1 \n" // im*cos
-            "mulps         %%xmm4, %%xmm2 \n" // re*cos
-            "mulps         %%xmm5, %%xmm3 \n" // im*sin
-            "subps         %%xmm0, %%xmm1 \n" // -> re
-            "addps         %%xmm3, %%xmm2 \n" // -> im
-            "movaps        %%xmm1, %%xmm0 \n"
-            "unpcklps      %%xmm2, %%xmm1 \n" // { z[k],    z[k+1]  }
-            "unpckhps      %%xmm2, %%xmm0 \n" // { z[-k-2], z[-k-1] }
-            ::"r"(-4*k), "r"(4*k),
-              "r"(input+n4), "r"(tcos+n8), "r"(tsin+n8)
-        );
-#if ARCH_X86_64
-        // if we have enough regs, don't let gcc make the luts latency-bound
-        // but if not, latency is faster than spilling
-        __asm__("movlps %%xmm0, %0 \n"
-            "movhps %%xmm0, %1 \n"
-            "movlps %%xmm1, %2 \n"
-            "movhps %%xmm1, %3 \n"
-            :"=m"(z[revtab[-k-2]]),
-             "=m"(z[revtab[-k-1]]),
-             "=m"(z[revtab[ k  ]]),
-             "=m"(z[revtab[ k+1]])
-        );
-#else
-        __asm__("movlps %%xmm0, %0" :"=m"(z[revtab[-k-2]]));
-        __asm__("movhps %%xmm0, %0" :"=m"(z[revtab[-k-1]]));
-        __asm__("movlps %%xmm1, %0" :"=m"(z[revtab[ k  ]]));
-        __asm__("movhps %%xmm1, %0" :"=m"(z[revtab[ k+1]]));
-#endif
-    }
-
-    ff_fft_dispatch_sse(z, s->nbits);
-
-    /* post rotation + reinterleave + reorder */
-
-#define CMUL(j,xmm0,xmm1)\
-        "movaps   (%2,"#j",2), %%xmm6 \n"\
-        "movaps 16(%2,"#j",2), "#xmm0"\n"\
-        "movaps        %%xmm6, "#xmm1"\n"\
-        "movaps        "#xmm0",%%xmm7 \n"\
-        "mulps      (%3,"#j"), %%xmm6 \n"\
-        "mulps      (%4,"#j"), "#xmm0"\n"\
-        "mulps      (%4,"#j"), "#xmm1"\n"\
-        "mulps      (%3,"#j"), %%xmm7 \n"\
-        "subps         %%xmm6, "#xmm0"\n"\
-        "addps         %%xmm7, "#xmm1"\n"
-
-    j = -n2;
-    k = n2-16;
-    __asm__ volatile(
-        "1: \n"
-        CMUL(%0, %%xmm0, %%xmm1)
-        CMUL(%1, %%xmm4, %%xmm5)
-        "shufps    $0x1b, %%xmm1, %%xmm1 \n"
-        "shufps    $0x1b, %%xmm5, %%xmm5 \n"
-        "movaps   %%xmm4, %%xmm6 \n"
-        "unpckhps %%xmm1, %%xmm4 \n"
-        "unpcklps %%xmm1, %%xmm6 \n"
-        "movaps   %%xmm0, %%xmm2 \n"
-        "unpcklps %%xmm5, %%xmm0 \n"
-        "unpckhps %%xmm5, %%xmm2 \n"
-        "movaps   %%xmm6,   (%2,%1,2) \n"
-        "movaps   %%xmm4, 16(%2,%1,2) \n"
-        "movaps   %%xmm0,   (%2,%0,2) \n"
-        "movaps   %%xmm2, 16(%2,%0,2) \n"
-        "sub $16, %1 \n"
-        "add $16, %0 \n"
-        "jl 1b \n"
-        :"+&r"(j), "+&r"(k)
-        :"r"(z+n8), "r"(tcos+n8), "r"(tsin+n8)
-        :"memory"
-    );
-}
-
 void ff_imdct_calc_sse(FFTContext *s, FFTSample *output, const FFTSample *input)
 {
     x86_reg j, k;