x86/dsputil: port ff_put_signed_pixels_clamped_mmx to yasm

Also add an SSE2 version

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

4 files changed, 56 insertions, 37 deletions

diff --git a/libavcodec/x86/dsputil.asm b/libavcodec/x86/dsputil.asm
index c91dd8eb69..747c645666 100644
--- a/libavcodec/x86/dsputil.asm
+++ b/libavcodec/x86/dsputil.asm
@@ -31,6 +31,8 @@ pb_zzzz3333zzzzbbbb: db -1,-1,-1,-1,3,3,3,3,-1,-1,-1,-1,11,11,11,11
 pb_zz11zz55zz99zzdd: db -1,-1,1,1,-1,-1,5,5,-1,-1,9,9,-1,-1,13,13
 pb_bswap32: db 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12
 
+cextern pb_80
+
 SECTION_TEXT
 
 %macro SCALARPRODUCT 0
@@ -573,3 +575,53 @@ CLEAR_BLOCKS 0
 INIT_XMM sse
 %define ZERO xorps
 CLEAR_BLOCKS 1
+
+;--------------------------------------------------------------------------
+;void ff_put_signed_pixels_clamped(const int16_t *block, uint8_t *pixels,
+;                                  int line_size)
+;--------------------------------------------------------------------------
+
+%macro PUT_SIGNED_PIXELS_CLAMPED_HALF 1
+    mova     m1, [blockq+mmsize*0+%1]
+    mova     m2, [blockq+mmsize*2+%1]
+%if mmsize == 8
+    mova     m3, [blockq+mmsize*4+%1]
+    mova     m4, [blockq+mmsize*6+%1]
+%endif
+    packsswb m1, [blockq+mmsize*1+%1]
+    packsswb m2, [blockq+mmsize*3+%1]
+%if mmsize == 8
+    packsswb m3, [blockq+mmsize*5+%1]
+    packsswb m4, [blockq+mmsize*7+%1]
+%endif
+    paddb    m1, m0
+    paddb    m2, m0
+%if mmsize == 8
+    paddb    m3, m0
+    paddb    m4, m0
+    movq     [pixelsq+lsizeq*0], m1
+    movq     [pixelsq+lsizeq*1], m2
+    movq     [pixelsq+lsizeq*2], m3
+    movq     [pixelsq+lsize3q ], m4
+%else
+    movq     [pixelsq+lsizeq*0], m1
+    movhps   [pixelsq+lsizeq*1], m1
+    movq     [pixelsq+lsizeq*2], m2
+    movhps   [pixelsq+lsize3q ], m2
+%endif
+%endmacro
+
+%macro PUT_SIGNED_PIXELS_CLAMPED 1
+cglobal put_signed_pixels_clamped, 3, 4, %1, block, pixels, lsize, lsize3
+    mova     m0, [pb_80]
+    lea      lsize3q, [lsizeq*3]
+    PUT_SIGNED_PIXELS_CLAMPED_HALF 0
+    lea      pixelsq, [pixelsq+lsizeq*4]
+    PUT_SIGNED_PIXELS_CLAMPED_HALF 64
+    RET
+%endmacro
+
+INIT_MMX mmx
+PUT_SIGNED_PIXELS_CLAMPED 0
+INIT_XMM sse2
+PUT_SIGNED_PIXELS_CLAMPED 3
diff --git a/libavcodec/x86/dsputil_init.c b/libavcodec/x86/dsputil_init.c
index 4461ae464f..e274e671d7 100644
--- a/libavcodec/x86/dsputil_init.c
+++ b/libavcodec/x86/dsputil_init.c
@@ -530,7 +530,6 @@ static av_cold void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx,
 {
 #if HAVE_MMX_INLINE
     c->put_pixels_clamped        = ff_put_pixels_clamped_mmx;
-    c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmx;
     c->add_pixels_clamped        = ff_add_pixels_clamped_mmx;
 
     if (!high_bit_depth) {
@@ -550,6 +549,7 @@ static av_cold void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx,
         c->clear_blocks = ff_clear_blocks_mmx;
     }
     c->vector_clip_int32 = ff_vector_clip_int32_mmx;
+    c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmx;
 #endif /* HAVE_MMX_EXTERNAL */
 }
 
@@ -627,6 +627,7 @@ static av_cold void dsputil_init_sse2(DSPContext *c, AVCodecContext *avctx,
         c->vector_clip_int32 = ff_vector_clip_int32_sse2;
     }
     c->bswap_buf = ff_bswap32_buf_sse2;
+    c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_sse2;
 #endif /* HAVE_SSE2_EXTERNAL */
 }
 
diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c
index a9c584d88a..fa77a5c938 100644
--- a/libavcodec/x86/dsputil_mmx.c
+++ b/libavcodec/x86/dsputil_mmx.c
@@ -94,42 +94,6 @@ void ff_put_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels,
         : "memory");
 }
 
-#define put_signed_pixels_clamped_mmx_half(off)             \
-    "movq          "#off"(%2), %%mm1        \n\t"           \
-    "movq     16 + "#off"(%2), %%mm2        \n\t"           \
-    "movq     32 + "#off"(%2), %%mm3        \n\t"           \
-    "movq     48 + "#off"(%2), %%mm4        \n\t"           \
-    "packsswb  8 + "#off"(%2), %%mm1        \n\t"           \
-    "packsswb 24 + "#off"(%2), %%mm2        \n\t"           \
-    "packsswb 40 + "#off"(%2), %%mm3        \n\t"           \
-    "packsswb 56 + "#off"(%2), %%mm4        \n\t"           \
-    "paddb              %%mm0, %%mm1        \n\t"           \
-    "paddb              %%mm0, %%mm2        \n\t"           \
-    "paddb              %%mm0, %%mm3        \n\t"           \
-    "paddb              %%mm0, %%mm4        \n\t"           \
-    "movq               %%mm1, (%0)         \n\t"           \
-    "movq               %%mm2, (%0, %3)     \n\t"           \
-    "movq               %%mm3, (%0, %3, 2)  \n\t"           \
-    "movq               %%mm4, (%0, %1)     \n\t"
-
-void ff_put_signed_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels,
-                                      int line_size)
-{
-    x86_reg line_skip = line_size;
-    x86_reg line_skip3;
-
-    __asm__ volatile (
-        "movq "MANGLE(ff_pb_80)", %%mm0     \n\t"
-        "lea         (%3, %3, 2), %1        \n\t"
-        put_signed_pixels_clamped_mmx_half(0)
-        "lea         (%0, %3, 4), %0        \n\t"
-        put_signed_pixels_clamped_mmx_half(64)
-        : "+&r" (pixels), "=&r" (line_skip3)
-        : "r" (block), "r" (line_skip)
-          NAMED_CONSTRAINTS_ADD(ff_pb_80)
-        : "memory");
-}
-
 void ff_add_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels,
                                int line_size)
 {
diff --git a/libavcodec/x86/dsputil_x86.h b/libavcodec/x86/dsputil_x86.h
index 6a50a09e5d..1f4711dd2d 100644
--- a/libavcodec/x86/dsputil_x86.h
+++ b/libavcodec/x86/dsputil_x86.h
@@ -37,6 +37,8 @@ void ff_put_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels,
                                int line_size);
 void ff_put_signed_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels,
                                       int line_size);
+void ff_put_signed_pixels_clamped_sse2(const int16_t *block, uint8_t *pixels,
+                                       int line_size);
 
 void ff_clear_block_mmx(int16_t *block);
 void ff_clear_block_sse(int16_t *block);