x86/dsputil: port ff_vector_clipf_sse to yasm

Signed-off-by: James Almer <jamrial@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

libavcodec/x86/dsputil.asm

@@ -625,3 +625,47 @@ INIT_MMX mmx
 PUT_SIGNED_PIXELS_CLAMPED 0
 INIT_XMM sse2
 PUT_SIGNED_PIXELS_CLAMPED 3
+
+;-----------------------------------------------------
+;void ff_vector_clipf(float *dst, const float *src,
+;                     float min, float max, int len)
+;-----------------------------------------------------
+INIT_XMM sse
+%if ARCH_X86_32
+cglobal vector_clipf, 5,5,6, dst, src, min, max, len
+%else
+cglobal vector_clipf, 3,3,6, dst, src, len
+%endif
+%if WIN64
+    SWAP 0, 2
+    SWAP 1, 3
+%elif ARCH_X86_32
+    movss   m0, minm
+    movss   m1, maxm
+%endif
+    SPLATD  m0
+    SPLATD  m1
+        shl lenq, 2
+        add srcq, lenq
+        add dstq, lenq
+        neg lenq
+.loop:
+    mova    m2,  [srcq+lenq+mmsize*0]
+    mova    m3,  [srcq+lenq+mmsize*1]
+    mova    m4,  [srcq+lenq+mmsize*2]
+    mova    m5,  [srcq+lenq+mmsize*3]
+    maxps   m2, m0
+    maxps   m3, m0
+    maxps   m4, m0
+    maxps   m5, m0
+    minps   m2, m1
+    minps   m3, m1
+    minps   m4, m1
+    minps   m5, m1
+    mova    [dstq+lenq+mmsize*0], m2
+    mova    [dstq+lenq+mmsize*1], m3
+    mova    [dstq+lenq+mmsize*2], m4
+    mova    [dstq+lenq+mmsize*3], m5
+    add     lenq, mmsize*4
+    jl .loop
+    REP_RET

libavcodec/x86/dsputil_init.c

@@ -585,12 +585,10 @@ static av_cold void dsputil_init_mmxext(DSPContext *c, AVCodecContext *avctx,
 static av_cold void dsputil_init_sse(DSPContext *c, AVCodecContext *avctx,
                                      int cpu_flags, unsigned high_bit_depth)
 {
-#if HAVE_SSE_INLINE
-    c->vector_clipf = ff_vector_clipf_sse;
-#endif /* HAVE_SSE_INLINE */
-
 #if HAVE_YASM
 #if HAVE_SSE_EXTERNAL
+    c->vector_clipf = ff_vector_clipf_sse;
+
     /* XvMCCreateBlocks() may not allocate 16-byte aligned blocks */
     if (CONFIG_XVMC && avctx->hwaccel && avctx->hwaccel->decode_mb)
         return;

libavcodec/x86/dsputil_mmx.c

@@ -506,37 +506,4 @@ void ff_avg_dirac_pixels32_sse2(uint8_t *dst, const uint8_t *src[5], int stride,
 #endif
 #endif
 
-void ff_vector_clipf_sse(float *dst, const float *src,
-                         float min, float max, int len)
-{
-    x86_reg i = (len - 16) * 4;
-    __asm__ volatile (
-        "movss          %3, %%xmm4      \n\t"
-        "movss          %4, %%xmm5      \n\t"
-        "shufps $0, %%xmm4, %%xmm4      \n\t"
-        "shufps $0, %%xmm5, %%xmm5      \n\t"
-        "1:                             \n\t"
-        "movaps   (%2, %0), %%xmm0      \n\t" // 3/1 on intel
-        "movaps 16(%2, %0), %%xmm1      \n\t"
-        "movaps 32(%2, %0), %%xmm2      \n\t"
-        "movaps 48(%2, %0), %%xmm3      \n\t"
-        "maxps      %%xmm4, %%xmm0      \n\t"
-        "maxps      %%xmm4, %%xmm1      \n\t"
-        "maxps      %%xmm4, %%xmm2      \n\t"
-        "maxps      %%xmm4, %%xmm3      \n\t"
-        "minps      %%xmm5, %%xmm0      \n\t"
-        "minps      %%xmm5, %%xmm1      \n\t"
-        "minps      %%xmm5, %%xmm2      \n\t"
-        "minps      %%xmm5, %%xmm3      \n\t"
-        "movaps     %%xmm0,   (%1, %0)  \n\t"
-        "movaps     %%xmm1, 16(%1, %0)  \n\t"
-        "movaps     %%xmm2, 32(%1, %0)  \n\t"
-        "movaps     %%xmm3, 48(%1, %0)  \n\t"
-        "sub           $64, %0          \n\t"
-        "jge            1b              \n\t"
-        : "+&r" (i)
-        : "r" (dst), "r" (src), "m" (min), "m" (max)
-        : "memory");
-}
-
 #endif /* HAVE_INLINE_ASM */