[FFmpeg-cvslog] x86/rv34dsp: add ff_rv34_idct_dc_add_sse2

[James Almer]

ffmpeg | branch: master | James Almer <jamrial at gmail.com> | Thu Feb  2 17:51:21 2017 -0300| [c8467abbadab424757ea23f71a1036abfb7f14b4] | committer: James Almer

x86/rv34dsp: add ff_rv34_idct_dc_add_sse2

Also disable ff_rv34_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=c8467abbadab424757ea23f71a1036abfb7f14b4
---

 libavcodec/x86/rv34dsp.asm    | 19 ++++++++++++++++++-
 libavcodec/x86/rv34dsp_init.c |  5 ++++-
 2 files changed, 22 insertions(+), 2 deletions(-)

diff --git a/libavcodec/x86/rv34dsp.asm b/libavcodec/x86/rv34dsp.asm
index 7732d65..692b4ac 100644
--- a/libavcodec/x86/rv34dsp.asm
+++ b/libavcodec/x86/rv34dsp.asm
@@ -64,6 +64,7 @@ rv34_idct dc
 rv34_idct dc_noround
 
 ; ff_rv34_idct_dc_add_mmx(uint8_t *dst, int stride, int dc);
+%if ARCH_X86_32
 INIT_MMX mmx
 cglobal rv34_idct_dc_add, 3, 3
     ; calculate DC
@@ -97,6 +98,7 @@ cglobal rv34_idct_dc_add, 3, 3
     movh       [r2], m4
     movh       [r2+r1], m5
     RET
+%endif
 
 ; Load coeffs and perform row transform
 ; Output: coeffs in mm[0467], rounder in mm5
@@ -167,7 +169,7 @@ cglobal rv34_idct_add, 3,3,0, d, s, b
     ret
 
 ; ff_rv34_idct_dc_add_sse4(uint8_t *dst, int stride, int dc);
-INIT_XMM sse4
+%macro RV34_IDCT_DC_ADD 0
 cglobal rv34_idct_dc_add, 3, 3, 6
     ; load data
     IDCT_DC_ROUND r2
@@ -190,7 +192,22 @@ cglobal rv34_idct_dc_add, 3, 3, 6
     paddw      m4, m0
     packuswb   m2, m4
     movd      [r0], m2
+%if cpuflag(sse4)
     pextrd [r0+r1], m2, 1
     pextrd    [r2], m2, 2
     pextrd [r2+r1], m2, 3
+%else
+    psrldq     m2, 4
+    movd   [r0+r1], m2
+    psrldq     m2, 4
+    movd      [r2], m2
+    psrldq     m2, 4
+    movd   [r2+r1], m2
+%endif
     RET
+%endmacro
+
+INIT_XMM sse2
+RV34_IDCT_DC_ADD
+INIT_XMM sse4
+RV34_IDCT_DC_ADD
diff --git a/libavcodec/x86/rv34dsp_init.c b/libavcodec/x86/rv34dsp_init.c
index d417dac..7310122 100644
--- a/libavcodec/x86/rv34dsp_init.c
+++ b/libavcodec/x86/rv34dsp_init.c
@@ -27,6 +27,7 @@
 void ff_rv34_idct_dc_mmxext(int16_t *block);
 void ff_rv34_idct_dc_noround_mmxext(int16_t *block);
 void ff_rv34_idct_dc_add_mmx(uint8_t *dst, ptrdiff_t stride, int dc);
+void ff_rv34_idct_dc_add_sse2(uint8_t *dst, ptrdiff_t stride, int dc);
 void ff_rv34_idct_dc_add_sse4(uint8_t *dst, ptrdiff_t stride, int dc);
 void ff_rv34_idct_add_mmxext(uint8_t *dst, ptrdiff_t stride, int16_t *block);
 
@@ -34,12 +35,14 @@ av_cold void ff_rv34dsp_init_x86(RV34DSPContext* c)
 {
     int cpu_flags = av_get_cpu_flags();
 
-    if (EXTERNAL_MMX(cpu_flags))
+    if (ARCH_X86_32 && EXTERNAL_MMX(cpu_flags))
         c->rv34_idct_dc_add = ff_rv34_idct_dc_add_mmx;
     if (EXTERNAL_MMXEXT(cpu_flags)) {
         c->rv34_inv_transform_dc = ff_rv34_idct_dc_noround_mmxext;
         c->rv34_idct_add         = ff_rv34_idct_add_mmxext;
     }
+    if (EXTERNAL_SSE2(cpu_flags))
+        c->rv34_idct_dc_add = ff_rv34_idct_dc_add_sse2;
     if (EXTERNAL_SSE4(cpu_flags))
         c->rv34_idct_dc_add = ff_rv34_idct_dc_add_sse4;
 }