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.

libavcodec/rv34dsp.c

@@ -32,7 +32,7 @@
  * @{
  */
 
-static av_always_inline void rv34_row_transform(int temp[16], DCTELEM *block)
+static av_always_inline void rv34_row_transform(int temp[16], const DCTELEM *block)
 {
     int i;
 
@@ -72,6 +72,32 @@ static void rv34_inv_transform_c(DCTELEM *block){
     }
 }
 
+/**
+ * Real Video 3.0/4.0 inverse transform + sample reconstruction
+ * Code is almost the same as in SVQ3, only scaling is different.
+ */
+static void rv34_idct_add_c(uint8_t *dst, int stride, const DCTELEM *block){
+    int      temp[16];
+    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
+    int      i;
+
+    rv34_row_transform(temp, block);
+
+    for(i = 0; i < 4; i++){
+        const int z0 = 13*(temp[4*0+i] +    temp[4*2+i]) + 0x200;
+        const int z1 = 13*(temp[4*0+i] -    temp[4*2+i]) + 0x200;
+        const int z2 =  7* temp[4*1+i] - 17*temp[4*3+i];
+        const int z3 = 17* temp[4*1+i] +  7*temp[4*3+i];
+
+        dst[0] = cm[ dst[0] + ( (z0 + z3) >> 10 ) ];
+        dst[1] = cm[ dst[1] + ( (z1 + z2) >> 10 ) ];
+        dst[2] = cm[ dst[2] + ( (z1 - z2) >> 10 ) ];
+        dst[3] = cm[ dst[3] + ( (z0 - z3) >> 10 ) ];
+
+        dst  += stride;
+    }
+}
+
 /**
  * RealVideo 3.0/4.0 inverse transform for DC block
  *
@@ -97,6 +123,22 @@ static void rv34_inv_transform_noround_c(DCTELEM *block){
     }
 }
 
+static void rv34_idct_dc_add_c(uint8_t *dst, int stride, int dc)
+{
+    const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
+    int i, j;
+
+    cm += (13*13*dc + 0x200) >> 10;
+
+    for (i = 0; i < 4; i++)
+    {
+        for (j = 0; j < 4; j++)
+            dst[j] = cm[ dst[j] ];
+
+        dst += stride;
+    }
+}
+
 static void rv34_inv_transform_dc_c(DCTELEM *block)
 {
     DCTELEM dc = (13 * 13 * block[0] + 0x200) >> 10;
@@ -126,6 +168,9 @@ av_cold void ff_rv34dsp_init(RV34DSPContext *c, DSPContext* dsp) {
     c->rv34_inv_transform_dc_tab[0]  = rv34_inv_transform_dc_c;
     c->rv34_inv_transform_dc_tab[1]  = rv34_inv_transform_dc_noround_c;
 
+    c->rv34_idct_add    = rv34_idct_add_c;
+    c->rv34_idct_dc_add = rv34_idct_dc_add_c;
+
     if (HAVE_NEON)
         ff_rv34dsp_init_neon(c, dsp);
     if (HAVE_MMX)

libavcodec/rv34dsp.h

@@ -36,6 +36,11 @@ typedef void (*rv40_weight_func)(uint8_t *dst/*align width (8 or 16)*/,
 
 typedef void (*rv34_inv_transform_func)(DCTELEM *block);
 
+typedef void (*rv34_idct_add_func)(uint8_t *dst, int stride,
+                                   const DCTELEM *block);
+typedef void (*rv34_idct_dc_add_func)(uint8_t *dst, int stride,
+                                      int   dc);
+
 typedef void (*rv40_weak_loop_filter_func)(uint8_t *src, int stride,
                                            int filter_p1, int filter_q1,
                                            int alpha, int beta,
@@ -57,6 +62,8 @@ typedef struct RV34DSPContext {
     rv40_weight_func rv40_weight_pixels_tab[2];
     rv34_inv_transform_func rv34_inv_transform_tab[2];
     void (*rv34_inv_transform_dc_tab[2])(DCTELEM *block);
+    rv34_idct_add_func rv34_idct_add;
+    rv34_idct_dc_add_func rv34_idct_dc_add;
     rv40_weak_loop_filter_func rv40_weak_loop_filter[2];
     rv40_strong_loop_filter_func rv40_strong_loop_filter[2];
     rv40_loop_filter_strength_func rv40_loop_filter_strength[2];

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

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
 }