dsputil: use cpuflags in x86 versions of vector_clip_int32()

2 files changed, 27 insertions, 21 deletions

diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c
index 959a2c2e7e..dd6cbf51ad 100644
--- a/libavcodec/x86/dsputil_mmx.c
+++ b/libavcodec/x86/dsputil_mmx.c
@@ -2419,9 +2419,9 @@ void ff_vector_clip_int32_mmx     (int32_t *dst, const int32_t *src, int32_t min
                                    int32_t max, unsigned int len);
 void ff_vector_clip_int32_sse2    (int32_t *dst, const int32_t *src, int32_t min,
                                    int32_t max, unsigned int len);
-void ff_vector_clip_int32_sse2_int(int32_t *dst, const int32_t *src, int32_t min,
+void ff_vector_clip_int32_int_sse2(int32_t *dst, const int32_t *src, int32_t min,
                                    int32_t max, unsigned int len);
-void ff_vector_clip_int32_sse41   (int32_t *dst, const int32_t *src, int32_t min,
+void ff_vector_clip_int32_sse4    (int32_t *dst, const int32_t *src, int32_t min,
                                    int32_t max, unsigned int len);
 
 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
@@ -2877,7 +2877,7 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
             c->scalarproduct_int16 = ff_scalarproduct_int16_sse2;
             c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_sse2;
             if (mm_flags & AV_CPU_FLAG_ATOM) {
-                c->vector_clip_int32 = ff_vector_clip_int32_sse2_int;
+                c->vector_clip_int32 = ff_vector_clip_int32_int_sse2;
             } else {
                 c->vector_clip_int32 = ff_vector_clip_int32_sse2;
             }
@@ -2909,7 +2909,7 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
 
         if (mm_flags & AV_CPU_FLAG_SSE4 && HAVE_SSE) {
 #if HAVE_YASM
-            c->vector_clip_int32 = ff_vector_clip_int32_sse41;
+            c->vector_clip_int32 = ff_vector_clip_int32_sse4;
 #endif
         }
 
diff --git a/libavcodec/x86/dsputil_yasm.asm b/libavcodec/x86/dsputil_yasm.asm
index fe96d8b12b..8e3cbdc03d 100644
--- a/libavcodec/x86/dsputil_yasm.asm
+++ b/libavcodec/x86/dsputil_yasm.asm
@@ -1055,9 +1055,14 @@ emu_edge mmx
 ;                           int32_t max, unsigned int len)
 ;-----------------------------------------------------------------------------
 
-%macro VECTOR_CLIP_INT32 4
-cglobal vector_clip_int32_%1, 5,5,%2, dst, src, min, max, len
-%ifidn %1, sse2
+; %1 = number of xmm registers used
+; %2 = number of inline load/process/store loops per asm loop
+; %3 = process 4*mmsize (%3=0) or 8*mmsize (%3=1) bytes per loop
+; %4 = CLIPD function takes min/max as float instead of int (CLIPD_SSE2)
+; %5 = suffix
+%macro VECTOR_CLIP_INT32 4-5
+cglobal vector_clip_int32%5, 5,5,%2, dst, src, min, max, len
+%if %4
     cvtsi2ss  m4, minm
     cvtsi2ss  m5, maxm
 %else
@@ -1068,12 +1073,12 @@ cglobal vector_clip_int32_%1, 5,5,%2, dst, src, min, max, len
     SPLATD    m5
 .loop:
 %assign %%i 1
-%rep %3
+%rep %2
     mova      m0,  [srcq+mmsize*0*%%i]
     mova      m1,  [srcq+mmsize*1*%%i]
     mova      m2,  [srcq+mmsize*2*%%i]
     mova      m3,  [srcq+mmsize*3*%%i]
-%if %4
+%if %3
     mova      m7,  [srcq+mmsize*4*%%i]
     mova      m8,  [srcq+mmsize*5*%%i]
     mova      m9,  [srcq+mmsize*6*%%i]
@@ -1083,7 +1088,7 @@ cglobal vector_clip_int32_%1, 5,5,%2, dst, src, min, max, len
     CLIPD  m1,  m4, m5, m6
     CLIPD  m2,  m4, m5, m6
     CLIPD  m3,  m4, m5, m6
-%if %4
+%if %3
     CLIPD  m7,  m4, m5, m6
     CLIPD  m8,  m4, m5, m6
     CLIPD  m9,  m4, m5, m6
@@ -1093,7 +1098,7 @@ cglobal vector_clip_int32_%1, 5,5,%2, dst, src, min, max, len
     mova  [dstq+mmsize*1*%%i], m1
     mova  [dstq+mmsize*2*%%i], m2
     mova  [dstq+mmsize*3*%%i], m3
-%if %4
+%if %3
     mova  [dstq+mmsize*4*%%i], m7
     mova  [dstq+mmsize*5*%%i], m8
     mova  [dstq+mmsize*6*%%i], m9
@@ -1101,25 +1106,26 @@ cglobal vector_clip_int32_%1, 5,5,%2, dst, src, min, max, len
 %endif
 %assign %%i %%i+1
 %endrep
-    add     srcq, mmsize*4*(%3+%4)
-    add     dstq, mmsize*4*(%3+%4)
-    sub     lend, mmsize*(%3+%4)
+    add     srcq, mmsize*4*(%2+%3)
+    add     dstq, mmsize*4*(%2+%3)
+    sub     lend, mmsize*(%2+%3)
     jg .loop
     REP_RET
 %endmacro
 
-INIT_MMX
+INIT_MMX mmx
 %define SPLATD SPLATD_MMX
 %define CLIPD CLIPD_MMX
-VECTOR_CLIP_INT32 mmx, 0, 1, 0
-INIT_XMM
+VECTOR_CLIP_INT32 0, 1, 0, 0
+INIT_XMM sse2
 %define SPLATD SPLATD_SSE2
-VECTOR_CLIP_INT32 sse2_int, 6, 1, 0
+VECTOR_CLIP_INT32 6, 1, 0, 0, _int
 %define CLIPD CLIPD_SSE2
-VECTOR_CLIP_INT32 sse2, 6, 2, 0
+VECTOR_CLIP_INT32 6, 2, 0, 1
+INIT_XMM sse4
 %define CLIPD CLIPD_SSE41
 %ifdef m8
-VECTOR_CLIP_INT32 sse41, 11, 1, 1
+VECTOR_CLIP_INT32 11, 1, 1, 0
 %else
-VECTOR_CLIP_INT32 sse41, 6, 1, 0
+VECTOR_CLIP_INT32 6, 1, 0, 0
 %endif