fmtconvert: port float_to_int16_interleave() 2-channel x86 inline asm to yasm

2 files changed, 61 insertions, 60 deletions

diff --git a/libavcodec/x86/fmtconvert.asm b/libavcodec/x86/fmtconvert.asm
index e3eb5d2286..854954835c 100644
--- a/libavcodec/x86/fmtconvert.asm
+++ b/libavcodec/x86/fmtconvert.asm
@@ -112,6 +112,58 @@ FLOAT_TO_INT16 3dnow, 0
 %undef cvtps2pi
 
 
+;-------------------------------------------------------------------------------
+; void ff_float_to_int16_interleave2(int16_t *dst, const float **src, long len);
+;-------------------------------------------------------------------------------
+%macro FLOAT_TO_INT16_INTERLEAVE2 1
+cglobal float_to_int16_interleave2_%1, 3,4,2, dst, src0, src1, len
+    lea      lenq, [4*r2q]
+    mov     src1q, [src0q+gprsize]
+    mov     src0q, [src0q]
+    add      dstq, lenq
+    add     src0q, lenq
+    add     src1q, lenq
+    neg      lenq
+.loop:
+%ifidn %1, sse2
+    cvtps2dq   m0, [src0q+lenq]
+    cvtps2dq   m1, [src1q+lenq]
+    packssdw   m0, m1
+    movhlps    m1, m0
+    punpcklwd  m0, m1
+    mova  [dstq+lenq], m0
+%else
+    cvtps2pi   m0, [src0q+lenq  ]
+    cvtps2pi   m1, [src0q+lenq+8]
+    cvtps2pi   m2, [src1q+lenq  ]
+    cvtps2pi   m3, [src1q+lenq+8]
+    packssdw   m0, m1
+    packssdw   m2, m3
+    mova       m1, m0
+    punpcklwd  m0, m2
+    punpckhwd  m1, m2
+    mova  [dstq+lenq  ], m0
+    mova  [dstq+lenq+8], m1
+%endif
+    add      lenq, 16
+    js .loop
+%ifnidn %1, sse2
+    emms
+%endif
+    REP_RET
+%endmacro
+
+INIT_MMX
+%define cvtps2pi pf2id
+FLOAT_TO_INT16_INTERLEAVE2 3dnow
+%undef cvtps2pi
+%define movdqa movaps
+FLOAT_TO_INT16_INTERLEAVE2 sse
+%undef movdqa
+INIT_XMM
+FLOAT_TO_INT16_INTERLEAVE2 sse2
+
+
 %macro PSWAPD_SSE 2
     pshufw %1, %2, 0x4e
 %endmacro
diff --git a/libavcodec/x86/fmtconvert_mmx.c b/libavcodec/x86/fmtconvert_mmx.c
index 86957b45ff..17079d3c82 100644
--- a/libavcodec/x86/fmtconvert_mmx.c
+++ b/libavcodec/x86/fmtconvert_mmx.c
@@ -35,13 +35,17 @@ void ff_float_to_int16_3dnow(int16_t *dst, const float *src, long len);
 void ff_float_to_int16_sse  (int16_t *dst, const float *src, long len);
 void ff_float_to_int16_sse2 (int16_t *dst, const float *src, long len);
 
+void ff_float_to_int16_interleave2_3dnow(int16_t *dst, const float **src, long len);
+void ff_float_to_int16_interleave2_sse  (int16_t *dst, const float **src, long len);
+void ff_float_to_int16_interleave2_sse2 (int16_t *dst, const float **src, long len);
+
 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);
 
 #define ff_float_to_int16_interleave6_sse2 ff_float_to_int16_interleave6_sse
 
-#define FLOAT_TO_INT16_INTERLEAVE(cpu, body) \
+#define FLOAT_TO_INT16_INTERLEAVE(cpu) \
 /* gcc pessimizes register allocation if this is in the same function as float_to_int16_interleave_sse2*/\
 static av_noinline void float_to_int16_interleave_misc_##cpu(int16_t *dst, const float **src, long len, int channels){\
     DECLARE_ALIGNED(16, int16_t, tmp)[len];\
@@ -57,71 +61,16 @@ static void float_to_int16_interleave_##cpu(int16_t *dst, const float **src, lon
     if(channels==1)\
         ff_float_to_int16_##cpu(dst, src[0], len);\
     else if(channels==2){\
-        x86_reg reglen = len; \
-        const float *src0 = src[0];\
-        const float *src1 = src[1];\
-        __asm__ volatile(\
-            "shl $2, %0 \n"\
-            "add %0, %1 \n"\
-            "add %0, %2 \n"\
-            "add %0, %3 \n"\
-            "neg %0 \n"\
-            body\
-            :"+r"(reglen), "+r"(dst), "+r"(src0), "+r"(src1)\
-        );\
+        ff_float_to_int16_interleave2_##cpu(dst, src, len);\
     }else if(channels==6){\
         ff_float_to_int16_interleave6_##cpu(dst, src, len);\
     }else\
         float_to_int16_interleave_misc_##cpu(dst, src, len, channels);\
 }
 
-FLOAT_TO_INT16_INTERLEAVE(3dnow,
-    "1:                         \n"
-    "pf2id     (%2,%0), %%mm0   \n"
-    "pf2id    8(%2,%0), %%mm1   \n"
-    "pf2id     (%3,%0), %%mm2   \n"
-    "pf2id    8(%3,%0), %%mm3   \n"
-    "packssdw    %%mm1, %%mm0   \n"
-    "packssdw    %%mm3, %%mm2   \n"
-    "movq        %%mm0, %%mm1   \n"
-    "punpcklwd   %%mm2, %%mm0   \n"
-    "punpckhwd   %%mm2, %%mm1   \n"
-    "movq        %%mm0,  (%1,%0)\n"
-    "movq        %%mm1, 8(%1,%0)\n"
-    "add $16, %0                \n"
-    "js 1b                      \n"
-    "femms                      \n"
-)
-
-FLOAT_TO_INT16_INTERLEAVE(sse,
-    "1:                         \n"
-    "cvtps2pi  (%2,%0), %%mm0   \n"
-    "cvtps2pi 8(%2,%0), %%mm1   \n"
-    "cvtps2pi  (%3,%0), %%mm2   \n"
-    "cvtps2pi 8(%3,%0), %%mm3   \n"
-    "packssdw    %%mm1, %%mm0   \n"
-    "packssdw    %%mm3, %%mm2   \n"
-    "movq        %%mm0, %%mm1   \n"
-    "punpcklwd   %%mm2, %%mm0   \n"
-    "punpckhwd   %%mm2, %%mm1   \n"
-    "movq        %%mm0,  (%1,%0)\n"
-    "movq        %%mm1, 8(%1,%0)\n"
-    "add $16, %0                \n"
-    "js 1b                      \n"
-    "emms                       \n"
-)
-
-FLOAT_TO_INT16_INTERLEAVE(sse2,
-    "1:                         \n"
-    "cvtps2dq  (%2,%0), %%xmm0  \n"
-    "cvtps2dq  (%3,%0), %%xmm1  \n"
-    "packssdw   %%xmm1, %%xmm0  \n"
-    "movhlps    %%xmm0, %%xmm1  \n"
-    "punpcklwd  %%xmm1, %%xmm0  \n"
-    "movdqa     %%xmm0, (%1,%0) \n"
-    "add $16, %0                \n"
-    "js 1b                      \n"
-)
+FLOAT_TO_INT16_INTERLEAVE(3dnow)
+FLOAT_TO_INT16_INTERLEAVE(sse)
+FLOAT_TO_INT16_INTERLEAVE(sse2)
 
 static void float_to_int16_interleave_3dn2(int16_t *dst, const float **src, long len, int channels){
     if(channels==6)