rv34: Intra 16x16 handling

Extract processing of intra 16x16 blocks from intra macroblock
processing.
Also implement a function performing inverse transform and block
reconstruction for DC-only blocks in 1 pass instead of 2.

2 files changed, 83 insertions, 14 deletions

diff --git a/libavcodec/x86/rv34dsp.asm b/libavcodec/x86/rv34dsp.asm
index 58f1af0495..c8eeebbfeb 100644
--- a/libavcodec/x86/rv34dsp.asm
+++ b/libavcodec/x86/rv34dsp.asm
@@ -35,21 +35,84 @@ SECTION .text
     sar    %1, 10
 %endmacro
 
-%macro rv34_idct_dequant4x4_dc 1
-cglobal rv34_idct_dequant4x4_%1_mmx2, 1, 2, 0
+%macro rv34_idct 1
+cglobal rv34_idct_%1_mmx2, 1, 2, 0
     movsx   r1, word [r0]
     IDCT_DC r1
-    movd    mm0, r1
-    pshufw  mm0, mm0, 0
-    movq    [r0+ 0], mm0
-    movq    [r0+16], mm0
-    movq    [r0+32], mm0
-    movq    [r0+48], mm0
+    movd    m0, r1
+    pshufw  m0, m0, 0
+    movq    [r0+ 0], m0
+    movq    [r0+16], m0
+    movq    [r0+32], m0
+    movq    [r0+48], m0
     REP_RET
 %endmacro
 
 INIT_MMX
 %define IDCT_DC IDCT_DC_ROUND
-rv34_idct_dequant4x4_dc dc
+rv34_idct dc
 %define IDCT_DC IDCT_DC_NOROUND
-rv34_idct_dequant4x4_dc dc_noround
+rv34_idct dc_noround
+
+; ff_rv34_idct_dc_add_mmx(uint8_t *dst, int stride, int dc);
+cglobal rv34_idct_dc_add_mmx, 3, 3
+    ; calculate DC
+    IDCT_DC_ROUND r2
+    pxor       m1, m1
+    movd       m0, r2
+    psubw      m1, m0
+    packuswb   m0, m0
+    packuswb   m1, m1
+    punpcklbw  m0, m0
+    punpcklbw  m1, m1
+    punpcklwd  m0, m0
+    punpcklwd  m1, m1
+
+    ; add DC
+    lea        r2, [r0+r1*2]
+    movh       m2, [r0]
+    movh       m3, [r0+r1]
+    movh       m4, [r2]
+    movh       m5, [r2+r1]
+    paddusb    m2, m0
+    paddusb    m3, m0
+    paddusb    m4, m0
+    paddusb    m5, m0
+    psubusb    m2, m1
+    psubusb    m3, m1
+    psubusb    m4, m1
+    psubusb    m5, m1
+    movh       [r0], m2
+    movh       [r0+r1], m3
+    movh       [r2], m4
+    movh       [r2+r1], m5
+    RET
+
+; ff_rv34_idct_dc_add_sse4(uint8_t *dst, int stride, int dc);
+INIT_XMM
+cglobal rv34_idct_dc_add_sse4, 3, 3, 6
+    ; load data
+    IDCT_DC_ROUND r2
+    pxor       m1, m1
+
+    ; calculate DC
+    movd       m0, r2
+    lea        r2, [r0+r1*2]
+    movd       m2, [r0]
+    movd       m3, [r0+r1]
+    pshuflw    m0, m0, 0
+    movd       m4, [r2]
+    movd       m5, [r2+r1]
+    punpcklqdq m0, m0
+    punpckldq  m2, m3
+    punpckldq  m4, m5
+    punpcklbw  m2, m1
+    punpcklbw  m4, m1
+    paddw      m2, m0
+    paddw      m4, m0
+    packuswb   m2, m4
+    movd      [r0], m2
+    pextrd [r0+r1], m2, 1
+    pextrd    [r2], m2, 2
+    pextrd [r2+r1], m2, 3
+    RET
diff --git a/libavcodec/x86/rv34dsp_init.c b/libavcodec/x86/rv34dsp_init.c
index 4317e9b23b..c10ae4ee96 100644
--- a/libavcodec/x86/rv34dsp_init.c
+++ b/libavcodec/x86/rv34dsp_init.c
@@ -24,17 +24,23 @@
 #include "libavcodec/dsputil.h"
 #include "libavcodec/rv34dsp.h"
 
-void ff_rv34_idct_dequant4x4_dc_mmx2(DCTELEM *block);
-void ff_rv34_idct_dequant4x4_dc_noround_mmx2(DCTELEM *block);
+void ff_rv34_idct_dc_mmx2(DCTELEM *block);
+void ff_rv34_idct_dc_noround_mmx2(DCTELEM *block);
+void ff_rv34_idct_dc_add_mmx(uint8_t *dst, int stride, int dc);
+void ff_rv34_idct_dc_add_sse4(uint8_t *dst, int stride, int dc);
 
 av_cold void ff_rv34dsp_init_x86(RV34DSPContext* c, DSPContext *dsp)
 {
 #if HAVE_YASM
     int mm_flags = av_get_cpu_flags();
 
+    if (mm_flags & AV_CPU_FLAG_MMX)
+        c->rv34_idct_dc_add = ff_rv34_idct_dc_add_mmx;
     if (mm_flags & AV_CPU_FLAG_MMX2) {
-        c->rv34_inv_transform_dc_tab[0] = ff_rv34_idct_dequant4x4_dc_mmx2;
-        c->rv34_inv_transform_dc_tab[1] = ff_rv34_idct_dequant4x4_dc_noround_mmx2;
+        c->rv34_inv_transform_dc_tab[0] = ff_rv34_idct_dc_mmx2;
+        c->rv34_inv_transform_dc_tab[1] = ff_rv34_idct_dc_noround_mmx2;
     }
+    if (mm_flags & AV_CPU_FLAG_SSE4)
+        c->rv34_idct_dc_add = ff_rv34_idct_dc_add_sse4;
 #endif
 }