rv34: DC-only inverse transform

When decoding coefficients, detect whether the block is DC-only, and take
advantage of this knowledge to perform DC-only inverse transform.

This is achieved by:
- first, changing the 108x4 element modulo_three_table into a 108 element
  table (kind of base4), and accessing each value using mask and shifts.
- then, checking low bits for 0 (as they represent the presence of higher
  frequency coefficients)

Also provide x86 SIMD code for the DC-only inverse transform.
Signed-off-by: Kostya Shishkov <kostya.shishkov@gmail.com>

libavcodec/rv34.c

@@ -204,7 +204,7 @@ static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
     }
 
     for(i = 0; i < 4; i++){
-        t = modulo_three_table[code][i];
+        t = (modulo_three_table[code] >> (6 - 2*i)) & 3;
         if(t == 1)
             cbp |= cbp_masks[get_bits1(gb)] << i;
         if(t == 2)
@@ -238,41 +238,42 @@ static inline void decode_coeff(DCTELEM *dst, int coef, int esc, GetBitContext *
  */
 static inline void decode_subblock(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc, int q)
 {
-    int coeffs[4];
+    int flags = modulo_three_table[code];
 
-    coeffs[0] = modulo_three_table[code][0];
-    coeffs[1] = modulo_three_table[code][1];
-    coeffs[2] = modulo_three_table[code][2];
-    coeffs[3] = modulo_three_table[code][3];
-    decode_coeff(dst  , coeffs[0], 3, gb, vlc, q);
+    decode_coeff(    dst+0, (flags >> 6)    , 3, gb, vlc, q);
     if(is_block2){
-        decode_coeff(dst+8, coeffs[1], 2, gb, vlc, q);
-        decode_coeff(dst+1, coeffs[2], 2, gb, vlc, q);
+        decode_coeff(dst+8, (flags >> 4) & 3, 2, gb, vlc, q);
+        decode_coeff(dst+1, (flags >> 2) & 3, 2, gb, vlc, q);
     }else{
-        decode_coeff(dst+1, coeffs[1], 2, gb, vlc, q);
-        decode_coeff(dst+8, coeffs[2], 2, gb, vlc, q);
+        decode_coeff(dst+1, (flags >> 4) & 3, 2, gb, vlc, q);
+        decode_coeff(dst+8, (flags >> 2) & 3, 2, gb, vlc, q);
     }
-    decode_coeff(dst+9, coeffs[3], 2, gb, vlc, q);
+    decode_coeff(    dst+9, (flags >> 0) & 3, 2, gb, vlc, q);
+}
+
+/**
+ * Decode a single coefficient.
+ */
+static inline void decode_subblock1(DCTELEM *dst, int code, GetBitContext *gb, VLC *vlc, int q)
+{
+    int coeff = modulo_three_table[code] >> 6;
+    decode_coeff(dst, coeff, 3, gb, vlc, q);
 }
 
 static inline void decode_subblock3(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc,
                                     int q_dc, int q_ac1, int q_ac2)
 {
-    int coeffs[4];
+    int flags = modulo_three_table[code];
 
-    coeffs[0] = modulo_three_table[code][0];
-    coeffs[1] = modulo_three_table[code][1];
-    coeffs[2] = modulo_three_table[code][2];
-    coeffs[3] = modulo_three_table[code][3];
-    decode_coeff(dst  , coeffs[0], 3, gb, vlc, q_dc);
+    decode_coeff(    dst+0, (flags >> 6)    , 3, gb, vlc, q_dc);
     if(is_block2){
-        decode_coeff(dst+8, coeffs[1], 2, gb, vlc, q_ac1);
-        decode_coeff(dst+1, coeffs[2], 2, gb, vlc, q_ac1);
+        decode_coeff(dst+8, (flags >> 4) & 3, 2, gb, vlc, q_ac1);
+        decode_coeff(dst+1, (flags >> 2) & 3, 2, gb, vlc, q_ac1);
     }else{
-        decode_coeff(dst+1, coeffs[1], 2, gb, vlc, q_ac1);
-        decode_coeff(dst+8, coeffs[2], 2, gb, vlc, q_ac1);
+        decode_coeff(dst+1, (flags >> 4) & 3, 2, gb, vlc, q_ac1);
+        decode_coeff(dst+8, (flags >> 2) & 3, 2, gb, vlc, q_ac1);
     }
-    decode_coeff(dst+9, coeffs[3], 2, gb, vlc, q_ac2);
+    decode_coeff(    dst+9, (flags >> 0) & 3, 2, gb, vlc, q_ac2);
 }
 
 /**
@@ -286,16 +287,24 @@ static inline void decode_subblock3(DCTELEM *dst, int code, const int is_block2,
  *  o--o
  */
 
-static inline void rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc, int q_dc, int q_ac1, int q_ac2)
+static inline int rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc, int q_dc, int q_ac1, int q_ac2)
 {
-    int code, pattern;
+    int code, pattern, has_ac = 1;
 
     code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
 
     pattern = code & 0x7;
 
     code >>= 3;
+
+    if (modulo_three_table[code] & 0x3F) {
         decode_subblock3(dst, code, 0, gb, &rvlc->coefficient, q_dc, q_ac1, q_ac2);
+    } else {
+        decode_subblock1(dst, code, gb, &rvlc->coefficient, q_dc);
+        if (!pattern)
+            return 0;
+        has_ac = 0;
+    }
 
     if(pattern & 4){
         code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
@@ -309,7 +318,7 @@ static inline void rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *r
         code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
         decode_subblock(dst + 8*2+2, code, 0, gb, &rvlc->coefficient, q_ac2);
     }
-
+    return has_ac || pattern;
 }
 
 /**
@@ -1118,7 +1127,7 @@ static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types)
     MpegEncContext *s = &r->s;
     GetBitContext *gb = &s->gb;
     int cbp, cbp2;
-    int q_dc, q_ac;
+    int q_dc, q_ac, has_ac;
     int i, blknum, blkoff;
     LOCAL_ALIGNED_16(DCTELEM, block16, [64]);
     int luma_dc_quant;
@@ -1157,33 +1166,45 @@ static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types)
     if(r->is16){
         q_dc = rv34_qscale_tab[luma_dc_quant];
         q_ac = rv34_qscale_tab[s->qscale];
-        memset(block16, 0, 64 * sizeof(*block16));
-        rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac);
+        s->dsp.clear_block(block16);
+        if (rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac))
             r->rdsp.rv34_inv_transform_tab[1](block16);
+        else
+            r->rdsp.rv34_inv_transform_dc_tab[1](block16);
     }
 
     q_ac = rv34_qscale_tab[s->qscale];
     for(i = 0; i < 16; i++, cbp >>= 1){
+        DCTELEM *ptr;
         if(!r->is16 && !(cbp & 1)) continue;
         blknum = ((i & 2) >> 1) + ((i & 8) >> 2);
         blkoff = ((i & 1) << 2) + ((i & 4) << 3);
+        ptr    = s->block[blknum] + blkoff;
         if(cbp & 1)
-            rv34_decode_block(s->block[blknum] + blkoff, gb,
-                              r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
+            has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
+        else
+            has_ac = 0;
         if(r->is16) //FIXME: optimize
-            s->block[blknum][blkoff] = block16[(i & 3) | ((i & 0xC) << 1)];
-        r->rdsp.rv34_inv_transform_tab[0](s->block[blknum] + blkoff);
+            ptr[0] = block16[(i & 3) | ((i & 0xC) << 1)];
+        if(has_ac)
+            r->rdsp.rv34_inv_transform_tab[0](ptr);
+        else
+            r->rdsp.rv34_inv_transform_dc_tab[0](ptr);
     }
     if(r->block_type == RV34_MB_P_MIX16x16)
         r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
     q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
     q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
     for(; i < 24; i++, cbp >>= 1){
+        DCTELEM *ptr;
         if(!(cbp & 1)) continue;
         blknum = ((i & 4) >> 2) + 4;
         blkoff = ((i & 1) << 2) + ((i & 2) << 4);
-        rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->chroma_vlc, 1, q_dc, q_ac, q_ac);
-        r->rdsp.rv34_inv_transform_tab[0](s->block[blknum] + blkoff);
+        ptr    = s->block[blknum] + blkoff;
+        if (rv34_decode_block(ptr, gb, r->cur_vlcs, r->chroma_vlc, 1, q_dc, q_ac, q_ac))
+            r->rdsp.rv34_inv_transform_tab[0](ptr);
+        else
+            r->rdsp.rv34_inv_transform_dc_tab[0](ptr);
     }
     if (IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos]))
         rv34_output_macroblock(r, intra_types, cbp2, r->is16);

libavcodec/rv34data.h

@@ -50,34 +50,22 @@ static const uint8_t rv34_cbp_code[16] = {
  * A lot of four-tuples in RV40 are represented as c0*27+c1*9+c2*3+c3.
  * This table allows conversion from a value back to a vector.
  */
-static const uint8_t modulo_three_table[108][4] = {
- { 0, 0, 0, 0 }, { 0, 0, 0, 1 }, { 0, 0, 0, 2 }, { 0, 0, 1, 0 },
- { 0, 0, 1, 1 }, { 0, 0, 1, 2 }, { 0, 0, 2, 0 }, { 0, 0, 2, 1 },
- { 0, 0, 2, 2 }, { 0, 1, 0, 0 }, { 0, 1, 0, 1 }, { 0, 1, 0, 2 },
- { 0, 1, 1, 0 }, { 0, 1, 1, 1 }, { 0, 1, 1, 2 }, { 0, 1, 2, 0 },
- { 0, 1, 2, 1 }, { 0, 1, 2, 2 }, { 0, 2, 0, 0 }, { 0, 2, 0, 1 },
- { 0, 2, 0, 2 }, { 0, 2, 1, 0 }, { 0, 2, 1, 1 }, { 0, 2, 1, 2 },
- { 0, 2, 2, 0 }, { 0, 2, 2, 1 }, { 0, 2, 2, 2 }, { 1, 0, 0, 0 },
- { 1, 0, 0, 1 }, { 1, 0, 0, 2 }, { 1, 0, 1, 0 }, { 1, 0, 1, 1 },
- { 1, 0, 1, 2 }, { 1, 0, 2, 0 }, { 1, 0, 2, 1 }, { 1, 0, 2, 2 },
- { 1, 1, 0, 0 }, { 1, 1, 0, 1 }, { 1, 1, 0, 2 }, { 1, 1, 1, 0 },
- { 1, 1, 1, 1 }, { 1, 1, 1, 2 }, { 1, 1, 2, 0 }, { 1, 1, 2, 1 },
- { 1, 1, 2, 2 }, { 1, 2, 0, 0 }, { 1, 2, 0, 1 }, { 1, 2, 0, 2 },
- { 1, 2, 1, 0 }, { 1, 2, 1, 1 }, { 1, 2, 1, 2 }, { 1, 2, 2, 0 },
- { 1, 2, 2, 1 }, { 1, 2, 2, 2 }, { 2, 0, 0, 0 }, { 2, 0, 0, 1 },
- { 2, 0, 0, 2 }, { 2, 0, 1, 0 }, { 2, 0, 1, 1 }, { 2, 0, 1, 2 },
- { 2, 0, 2, 0 }, { 2, 0, 2, 1 }, { 2, 0, 2, 2 }, { 2, 1, 0, 0 },
- { 2, 1, 0, 1 }, { 2, 1, 0, 2 }, { 2, 1, 1, 0 }, { 2, 1, 1, 1 },
- { 2, 1, 1, 2 }, { 2, 1, 2, 0 }, { 2, 1, 2, 1 }, { 2, 1, 2, 2 },
- { 2, 2, 0, 0 }, { 2, 2, 0, 1 }, { 2, 2, 0, 2 }, { 2, 2, 1, 0 },
- { 2, 2, 1, 1 }, { 2, 2, 1, 2 }, { 2, 2, 2, 0 }, { 2, 2, 2, 1 },
- { 2, 2, 2, 2 }, { 3, 0, 0, 0 }, { 3, 0, 0, 1 }, { 3, 0, 0, 2 },
- { 3, 0, 1, 0 }, { 3, 0, 1, 1 }, { 3, 0, 1, 2 }, { 3, 0, 2, 0 },
- { 3, 0, 2, 1 }, { 3, 0, 2, 2 }, { 3, 1, 0, 0 }, { 3, 1, 0, 1 },
- { 3, 1, 0, 2 }, { 3, 1, 1, 0 }, { 3, 1, 1, 1 }, { 3, 1, 1, 2 },
- { 3, 1, 2, 0 }, { 3, 1, 2, 1 }, { 3, 1, 2, 2 }, { 3, 2, 0, 0 },
- { 3, 2, 0, 1 }, { 3, 2, 0, 2 }, { 3, 2, 1, 0 }, { 3, 2, 1, 1 },
- { 3, 2, 1, 2 }, { 3, 2, 2, 0 }, { 3, 2, 2, 1 }, { 3, 2, 2, 2 },
+static const uint8_t modulo_three_table[108] = {
+    0x00, 0x01, 0x02, 0x04, 0x05, 0x06, 0x08, 0x09, 0x0A,
+    0x10, 0x11, 0x12, 0x14, 0x15, 0x16, 0x18, 0x19, 0x1A,
+    0x20, 0x21, 0x22, 0x24, 0x25, 0x26, 0x28, 0x29, 0x2A,
+
+    0x40, 0x41, 0x42, 0x44, 0x45, 0x46, 0x48, 0x49, 0x4A,
+    0x50, 0x51, 0x52, 0x54, 0x55, 0x56, 0x58, 0x59, 0x5A,
+    0x60, 0x61, 0x62, 0x64, 0x65, 0x66, 0x68, 0x69, 0x6A,
+
+    0x80, 0x81, 0x82, 0x84, 0x85, 0x86, 0x88, 0x89, 0x8A,
+    0x90, 0x91, 0x92, 0x94, 0x95, 0x96, 0x98, 0x99, 0x9A,
+    0xA0, 0xA1, 0xA2, 0xA4, 0xA5, 0xA6, 0xA8, 0xA9, 0xAA,
+
+    0xC0, 0xC1, 0xC2, 0xC4, 0xC5, 0xC6, 0xC8, 0xC9, 0xCA,
+    0xD0, 0xD1, 0xD2, 0xD4, 0xD5, 0xD6, 0xD8, 0xD9, 0xDA,
+    0xE0, 0xE1, 0xE2, 0xE4, 0xE5, 0xE6, 0xE8, 0xE9, 0xEA,
 };
 
 /**

libavcodec/rv34dsp.c

@@ -97,13 +97,37 @@ static void rv34_inv_transform_noround_c(DCTELEM *block){
     }
 }
 
+static void rv34_inv_transform_dc_c(DCTELEM *block)
+{
+    DCTELEM dc = (13 * 13 * block[0] + 0x200) >> 10;
+    int i, j;
+
+    for (i = 0; i < 4; i++, block += 8)
+        for (j = 0; j < 4; j++)
+            block[j] = dc;
+}
+
+static void rv34_inv_transform_dc_noround_c(DCTELEM *block)
+{
+    DCTELEM dc = (13 * 13 * 3 * block[0]) >> 11;
+    int i, j;
+
+    for (i = 0; i < 4; i++, block += 8)
+        for (j = 0; j < 4; j++)
+            block[j] = dc;
+}
+
 /** @} */ // transform
 
 
 av_cold void ff_rv34dsp_init(RV34DSPContext *c, DSPContext* dsp) {
     c->rv34_inv_transform_tab[0] = rv34_inv_transform_c;
     c->rv34_inv_transform_tab[1] = rv34_inv_transform_noround_c;
+    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;
 
     if (HAVE_NEON)
         ff_rv34dsp_init_neon(c, dsp);
+    if (HAVE_MMX)
+        ff_rv34dsp_init_x86(c, dsp);
 }

libavcodec/rv34dsp.h

@@ -56,6 +56,7 @@ typedef struct RV34DSPContext {
     h264_chroma_mc_func avg_chroma_pixels_tab[3];
     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);
     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];
@@ -66,6 +67,7 @@ void ff_rv34dsp_init(RV34DSPContext *c, DSPContext* dsp);
 void ff_rv40dsp_init(RV34DSPContext *c, DSPContext* dsp);
 
 void ff_rv34dsp_init_neon(RV34DSPContext *c, DSPContext *dsp);
+void ff_rv34dsp_init_x86(RV34DSPContext *c, DSPContext *dsp);
 
 void ff_rv40dsp_init_x86(RV34DSPContext *c, DSPContext *dsp);
 void ff_rv40dsp_init_neon(RV34DSPContext *c, DSPContext *dsp);

libavcodec/x86/Makefile

@@ -24,7 +24,11 @@ YASM-OBJS-$(CONFIG_H264PRED)           += x86/h264_intrapred.o          \
                                           x86/h264_intrapred_10bit.o
 MMX-OBJS-$(CONFIG_H264PRED)            += x86/h264_intrapred_init.o
 
-MMX-OBJS-$(CONFIG_RV40_DECODER)        += x86/rv40dsp.o                 \
+MMX-OBJS-$(CONFIG_RV30_DECODER)        += x86/rv34dsp_init.o
+YASM-OBJS-$(CONFIG_RV30_DECODER)       += x86/rv34dsp.o
+MMX-OBJS-$(CONFIG_RV40_DECODER)        += x86/rv34dsp_init.o            \
+                                          x86/rv40dsp.o
+YASM-OBJS-$(CONFIG_RV40_DECODER)       += x86/rv34dsp.o
 
 YASM-OBJS-$(CONFIG_VC1_DECODER)        += x86/vc1dsp_yasm.o
 

libavcodec/x86/rv34dsp.asm

+;******************************************************************************
+;* MMX/SSE2-optimized functions for the RV30 and RV40 decoders
+;* Copyright (C) 2012 Christophe Gisquet <christophe.gisquet@gmail.com>
+;*
+;* This file is part of Libav.
+;*
+;* Libav is free software; you can redistribute it and/or
+;* modify it under the terms of the GNU Lesser General Public
+;* License as published by the Free Software Foundation; either
+;* version 2.1 of the License, or (at your option) any later version.
+;*
+;* Libav is distributed in the hope that it will be useful,
+;* but WITHOUT ANY WARRANTY; without even the implied warranty of
+;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+;* Lesser General Public License for more details.
+;*
+;* You should have received a copy of the GNU Lesser General Public
+;* License along with Libav; if not, write to the Free Software
+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+;******************************************************************************
+
+%include "x86inc.asm"
+%include "x86util.asm"
+
+SECTION .text
+
+%macro IDCT_DC_NOROUND 1
+    imul   %1, 13*13*3
+    sar    %1, 11
+%endmacro
+
+%macro IDCT_DC_ROUND 1
+    imul   %1, 13*13
+    add    %1, 0x200
+    sar    %1, 10
+%endmacro
+
+%macro rv34_idct_dequant4x4_dc 1
+cglobal rv34_idct_dequant4x4_%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
+    REP_RET
+%endmacro
+
+INIT_MMX
+%define IDCT_DC IDCT_DC_ROUND
+rv34_idct_dequant4x4_dc dc
+%define IDCT_DC IDCT_DC_NOROUND
+rv34_idct_dequant4x4_dc dc_noround

libavcodec/x86/rv34dsp_init.c

+/*
+ * RV30/40 MMX/SSE2 optimizations
+ * Copyright (C) 2012 Christophe Gisquet <christophe.gisquet@gmail.com>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/cpu.h"
+#include "libavutil/x86_cpu.h"
+#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);
+
+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_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;
+    }
+#endif
+}