[FFmpeg-cvslog] x86/vp8dsp: add ff_vp8_idct_dc_add_sse2

[James Almer]

ffmpeg | branch: master | James Almer <jamrial at gmail.com> | Thu Feb  2 17:18:58 2017 -0300| [ab5c4d006da0b3e7aea6fddcebdf5d3d9ad7fd30] | committer: James Almer

x86/vp8dsp: add ff_vp8_idct_dc_add_sse2

Also disable ff_vp8_idct_dc_add_mmx on x86_64 as the presence of sse2
is guaranteed in such builds.

Signed-off-by: James Almer <jamrial at gmail.com>

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=ab5c4d006da0b3e7aea6fddcebdf5d3d9ad7fd30
---

 libavcodec/x86/vp8dsp.asm    | 19 ++++++++++++++++++-
 libavcodec/x86/vp8dsp_init.c |  5 ++++-
 2 files changed, 22 insertions(+), 2 deletions(-)

diff --git a/libavcodec/x86/vp8dsp.asm b/libavcodec/x86/vp8dsp.asm
index 538b3f4..ea6dcf6 100644
--- a/libavcodec/x86/vp8dsp.asm
+++ b/libavcodec/x86/vp8dsp.asm
@@ -906,6 +906,7 @@ cglobal put_vp8_pixels16, 5, 5, 2, dst, dststride, src, srcstride, height
     %4 [dst2q+strideq+%3], m5
 %endmacro
 
+%if ARCH_X86_32
 INIT_MMX mmx
 cglobal vp8_idct_dc_add, 3, 3, 0, dst, block, stride
     ; load data
@@ -929,8 +930,9 @@ cglobal vp8_idct_dc_add, 3, 3, 0, dst, block, stride
     lea     dst2q, [dst1q+strideq*2]
     ADD_DC     m0, m1, 0, movh
     RET
+%endif
 
-INIT_XMM sse4
+%macro VP8_IDCT_DC_ADD 0
 cglobal vp8_idct_dc_add, 3, 3, 6, dst, block, stride
     ; load data
     movd       m0, [blockq]
@@ -956,10 +958,25 @@ cglobal vp8_idct_dc_add, 3, 3, 6, dst, block, stride
     paddw      m4, m0
     packuswb   m2, m4
     movd   [dst1q], m2
+%if cpuflag(sse4)
     pextrd [dst1q+strideq], m2, 1
     pextrd [dst2q], m2, 2
     pextrd [dst2q+strideq], m2, 3
+%else
+    psrldq     m2, 4
+    movd [dst1q+strideq], m2
+    psrldq     m2, 4
+    movd [dst2q], m2
+    psrldq     m2, 4
+    movd [dst2q+strideq], m2
+%endif
     RET
+%endmacro
+
+INIT_XMM sse2
+VP8_IDCT_DC_ADD
+INIT_XMM sse4
+VP8_IDCT_DC_ADD
 
 ;-----------------------------------------------------------------------------
 ; void ff_vp8_idct_dc_add4y_<opt>(uint8_t *dst, int16_t block[4][16], int stride);
diff --git a/libavcodec/x86/vp8dsp_init.c b/libavcodec/x86/vp8dsp_init.c
index 20c5fac..814bf69 100644
--- a/libavcodec/x86/vp8dsp_init.c
+++ b/libavcodec/x86/vp8dsp_init.c
@@ -233,6 +233,8 @@ HVBILIN(ssse3, 8, 16, 16)
 
 void ff_vp8_idct_dc_add_mmx(uint8_t *dst, int16_t block[16],
                             ptrdiff_t stride);
+void ff_vp8_idct_dc_add_sse2(uint8_t *dst, int16_t block[16],
+                             ptrdiff_t stride);
 void ff_vp8_idct_dc_add_sse4(uint8_t *dst, int16_t block[16],
                              ptrdiff_t stride);
 void ff_vp8_idct_dc_add4y_mmx(uint8_t *dst, int16_t block[4][16],
@@ -370,9 +372,9 @@ av_cold void ff_vp8dsp_init_x86(VP8DSPContext *c)
     int cpu_flags = av_get_cpu_flags();
 
     if (EXTERNAL_MMX(cpu_flags)) {
-        c->vp8_idct_dc_add    = ff_vp8_idct_dc_add_mmx;
         c->vp8_idct_dc_add4uv = ff_vp8_idct_dc_add4uv_mmx;
 #if ARCH_X86_32
+        c->vp8_idct_dc_add    = ff_vp8_idct_dc_add_mmx;
         c->vp8_idct_dc_add4y  = ff_vp8_idct_dc_add4y_mmx;
         c->vp8_idct_add       = ff_vp8_idct_add_mmx;
         c->vp8_luma_dc_wht    = ff_vp8_luma_dc_wht_mmx;
@@ -427,6 +429,7 @@ av_cold void ff_vp8dsp_init_x86(VP8DSPContext *c)
     }
 
     if (EXTERNAL_SSE2(cpu_flags)) {
+        c->vp8_idct_dc_add            = ff_vp8_idct_dc_add_sse2;
         c->vp8_idct_dc_add4y          = ff_vp8_idct_dc_add4y_sse2;
 
         c->vp8_h_loop_filter_simple   = ff_vp8_h_loop_filter_simple_sse2;