hpeldsp: Add half-pel functions (currently copies of dsputil)

Signed-off-by: Martin Storsjö <martin@martin.st>

configure

@@ -1368,6 +1368,7 @@ CONFIG_EXTRA="
     h264dsp
     h264pred
     h264qpel
+    hpeldsp
     huffman
     lgplv3
     lpc

libavcodec/Makefile

@@ -44,6 +44,7 @@ OBJS-$(CONFIG_H264CHROMA)              += h264chroma.o
 OBJS-$(CONFIG_H264DSP)                 += h264dsp.o h264idct.o
 OBJS-$(CONFIG_H264PRED)                += h264pred.o
 OBJS-$(CONFIG_H264QPEL)                += h264qpel.o
+OBJS-$(CONFIG_HPELDSP)                 += hpeldsp.o
 OBJS-$(CONFIG_HUFFMAN)                 += huffman.o
 OBJS-$(CONFIG_LIBXVID)                 += libxvid_rc.o
 OBJS-$(CONFIG_LPC)                     += lpc.o

libavcodec/dsputil.h

@@ -32,6 +32,7 @@
 
 #include "libavutil/intreadwrite.h"
 #include "avcodec.h"
+#include "rnd_avg.h"
 
 
 //#define DEBUG
@@ -103,12 +104,6 @@ DEF_OLD_QPEL(qpel8_mc32_old_c)
 DEF_OLD_QPEL(qpel8_mc13_old_c)
 DEF_OLD_QPEL(qpel8_mc33_old_c)
 
-#define CALL_2X_PIXELS(a, b, n)\
-static void a(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h){\
-    b(block  , pixels  , line_size, h);\
-    b(block+n, pixels+n, line_size, h);\
-}
-
 /* motion estimation */
 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller than 2
 // although currently h<4 is not used as functions with width <8 are neither used nor implemented

libavcodec/hpeldsp.c

+/*
+ * Half-pel DSP functions.
+ * Copyright (c) 2000, 2001 Fabrice Bellard
+ * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
+ *
+ * gmc & q-pel & 32/64 bit based MC by Michael Niedermayer <michaelni@gmx.at>
+ *
+ * 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
+ */
+
+/**
+ * @file
+ * Half-pel DSP functions.
+ */
+
+#include "libavutil/attributes.h"
+#include "libavutil/intreadwrite.h"
+#include "hpeldsp.h"
+
+#define BIT_DEPTH 8
+#include "hpeldsp_template.c"
+
+av_cold void ff_hpeldsp_init(HpelDSPContext *c, int flags)
+{
+#define hpel_funcs(prefix, idx, num) \
+    c->prefix ## _pixels_tab idx [0] = prefix ## _pixels ## num ## _8_c; \
+    c->prefix ## _pixels_tab idx [1] = prefix ## _pixels ## num ## _x2_8_c; \
+    c->prefix ## _pixels_tab idx [2] = prefix ## _pixels ## num ## _y2_8_c; \
+    c->prefix ## _pixels_tab idx [3] = prefix ## _pixels ## num ## _xy2_8_c
+
+    hpel_funcs(put, [0], 16);
+    hpel_funcs(put, [1],  8);
+    hpel_funcs(put, [2],  4);
+    hpel_funcs(put, [3],  2);
+    hpel_funcs(put_no_rnd, [0], 16);
+    hpel_funcs(put_no_rnd, [1],  8);
+    hpel_funcs(avg, [0], 16);
+    hpel_funcs(avg, [1],  8);
+    hpel_funcs(avg, [2],  4);
+    hpel_funcs(avg, [3],  2);
+    hpel_funcs(avg_no_rnd,, 16);
+}

libavcodec/hpeldsp.h

+/*
+ * Half-pel DSP functions.
+ * Copyright (c) 2000, 2001, 2002 Fabrice Bellard
+ * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
+ *
+ * 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
+ */
+
+/**
+ * @file
+ * Half-pel DSP functions.
+ */
+
+#ifndef AVCODEC_HPELDSP_H
+#define AVCODEC_HPELDSP_H
+
+#include <stdint.h>
+#include <stddef.h>
+
+/* add and put pixel (decoding) */
+// blocksizes for hpel_pixels_func are 8x4,8x8 16x8 16x16
+// h for hpel_pixels_func is limited to {width/2, width} but never larger
+// than 16 and never smaller than 4
+typedef void (*hpel_pixels_func)(uint8_t *block /*align width (8 or 16)*/,
+                                 const uint8_t *pixels /*align 1*/,
+                                 ptrdiff_t line_size, int h);
+
+/**
+ * Half-pel DSP context.
+ */
+typedef struct HpelDSPContext {
+    /**
+     * Halfpel motion compensation with rounding (a+b+1)>>1.
+     * this is an array[4][4] of motion compensation functions for 4
+     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
+     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
+     * @param block destination where the result is stored
+     * @param pixels source
+     * @param line_size number of bytes in a horizontal line of block
+     * @param h height
+     */
+    hpel_pixels_func put_pixels_tab[4][4];
+
+    /**
+     * Halfpel motion compensation with rounding (a+b+1)>>1.
+     * This is an array[4][4] of motion compensation functions for 4
+     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
+     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
+     * @param block destination into which the result is averaged (a+b+1)>>1
+     * @param pixels source
+     * @param line_size number of bytes in a horizontal line of block
+     * @param h height
+     */
+    hpel_pixels_func avg_pixels_tab[4][4];
+
+    /**
+     * Halfpel motion compensation with no rounding (a+b)>>1.
+     * this is an array[2][4] of motion compensation functions for 2
+     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
+     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
+     * @param block destination where the result is stored
+     * @param pixels source
+     * @param line_size number of bytes in a horizontal line of block
+     * @param h height
+     */
+    hpel_pixels_func put_no_rnd_pixels_tab[2][4];
+
+    /**
+     * Halfpel motion compensation with no rounding (a+b)>>1.
+     * this is an array[4] of motion compensation functions for 1
+     * horizontal blocksize (16) and the 4 halfpel positions<br>
+     * *pixels_tab[0][ xhalfpel + 2*yhalfpel ]
+     * @param block destination into which the result is averaged (a+b)>>1
+     * @param pixels source
+     * @param line_size number of bytes in a horizontal line of block
+     * @param h height
+     */
+    hpel_pixels_func avg_no_rnd_pixels_tab[4];
+} HpelDSPContext;
+
+void ff_hpeldsp_init(HpelDSPContext *c, int flags);
+
+#endif /* AVCODEC_HPELDSP_H */

libavcodec/hpeldsp_template.c

+/*
+ * Half-pel DSP functions.
+ * Copyright (c) 2000, 2001 Fabrice Bellard
+ * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
+ *
+ * gmc & q-pel & 32/64 bit based MC by Michael Niedermayer <michaelni@gmx.at>
+ *
+ * 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
+ */
+
+/**
+ * @file
+ * Half-pel DSP functions.
+ */
+
+#include "bit_depth_template.c"
+
+#include "hpel_template.c"
+
+#define PIXOP2(OPNAME, OP) \
+static inline void FUNC(OPNAME ## _no_rnd_pixels8_l2)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \
+                                                int src_stride1, int src_stride2, int h){\
+    int i;\
+    for(i=0; i<h; i++){\
+        pixel4 a,b;\
+        a= AV_RN4P(&src1[i*src_stride1  ]);\
+        b= AV_RN4P(&src2[i*src_stride2  ]);\
+        OP(*((pixel4*)&dst[i*dst_stride  ]), no_rnd_avg_pixel4(a, b));\
+        a= AV_RN4P(&src1[i*src_stride1+4*sizeof(pixel)]);\
+        b= AV_RN4P(&src2[i*src_stride2+4*sizeof(pixel)]);\
+        OP(*((pixel4*)&dst[i*dst_stride+4*sizeof(pixel)]), no_rnd_avg_pixel4(a, b));\
+    }\
+}\
+\
+static inline void FUNCC(OPNAME ## _no_rnd_pixels8_x2)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h){\
+    FUNC(OPNAME ## _no_rnd_pixels8_l2)(block, pixels, pixels+sizeof(pixel), line_size, line_size, line_size, h);\
+}\
+\
+static inline void FUNCC(OPNAME ## _pixels8_x2)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h){\
+    FUNC(OPNAME ## _pixels8_l2)(block, pixels, pixels+sizeof(pixel), line_size, line_size, line_size, h);\
+}\
+\
+static inline void FUNCC(OPNAME ## _no_rnd_pixels8_y2)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h){\
+    FUNC(OPNAME ## _no_rnd_pixels8_l2)(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\
+}\
+\
+static inline void FUNCC(OPNAME ## _pixels8_y2)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h){\
+    FUNC(OPNAME ## _pixels8_l2)(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\
+}\
+\
+static inline void FUNCC(OPNAME ## _pixels4_x2)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h){\
+    FUNC(OPNAME ## _pixels4_l2)(block, pixels, pixels+sizeof(pixel), line_size, line_size, line_size, h);\
+}\
+\
+static inline void FUNCC(OPNAME ## _pixels4_y2)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h){\
+    FUNC(OPNAME ## _pixels4_l2)(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\
+}\
+\
+static inline void FUNCC(OPNAME ## _pixels2_x2)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h){\
+    FUNC(OPNAME ## _pixels2_l2)(block, pixels, pixels+sizeof(pixel), line_size, line_size, line_size, h);\
+}\
+\
+static inline void FUNCC(OPNAME ## _pixels2_y2)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h){\
+    FUNC(OPNAME ## _pixels2_l2)(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\
+}\
+\
+static inline void FUNCC(OPNAME ## _pixels2_xy2)(uint8_t *_block, const uint8_t *_pixels, ptrdiff_t line_size, int h)\
+{\
+        int i, a0, b0, a1, b1;\
+        pixel *block = (pixel*)_block;\
+        const pixel *pixels = (const pixel*)_pixels;\
+        line_size /= sizeof(pixel);\
+        a0= pixels[0];\
+        b0= pixels[1] + 2;\
+        a0 += b0;\
+        b0 += pixels[2];\
+\
+        pixels+=line_size;\
+        for(i=0; i<h; i+=2){\
+            a1= pixels[0];\
+            b1= pixels[1];\
+            a1 += b1;\
+            b1 += pixels[2];\
+\
+            block[0]= (a1+a0)>>2; /* FIXME non put */\
+            block[1]= (b1+b0)>>2;\
+\
+            pixels+=line_size;\
+            block +=line_size;\
+\
+            a0= pixels[0];\
+            b0= pixels[1] + 2;\
+            a0 += b0;\
+            b0 += pixels[2];\
+\
+            block[0]= (a1+a0)>>2;\
+            block[1]= (b1+b0)>>2;\
+            pixels+=line_size;\
+            block +=line_size;\
+        }\
+}\
+\
+static inline void FUNCC(OPNAME ## _pixels4_xy2)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)\
+{\
+        /* FIXME HIGH BIT DEPTH */\
+        int i;\
+        const uint32_t a= AV_RN32(pixels  );\
+        const uint32_t b= AV_RN32(pixels+1);\
+        uint32_t l0=  (a&0x03030303UL)\
+                    + (b&0x03030303UL)\
+                    + 0x02020202UL;\
+        uint32_t h0= ((a&0xFCFCFCFCUL)>>2)\
+                   + ((b&0xFCFCFCFCUL)>>2);\
+        uint32_t l1,h1;\
+\
+        pixels+=line_size;\
+        for(i=0; i<h; i+=2){\
+            uint32_t a= AV_RN32(pixels  );\
+            uint32_t b= AV_RN32(pixels+1);\
+            l1=  (a&0x03030303UL)\
+               + (b&0x03030303UL);\
+            h1= ((a&0xFCFCFCFCUL)>>2)\
+              + ((b&0xFCFCFCFCUL)>>2);\
+            OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\
+            pixels+=line_size;\
+            block +=line_size;\
+            a= AV_RN32(pixels  );\
+            b= AV_RN32(pixels+1);\
+            l0=  (a&0x03030303UL)\
+               + (b&0x03030303UL)\
+               + 0x02020202UL;\
+            h0= ((a&0xFCFCFCFCUL)>>2)\
+              + ((b&0xFCFCFCFCUL)>>2);\
+            OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\
+            pixels+=line_size;\
+            block +=line_size;\
+        }\
+}\
+\
+static inline void FUNCC(OPNAME ## _pixels8_xy2)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)\
+{\
+    /* FIXME HIGH BIT DEPTH */\
+    int j;\
+    for(j=0; j<2; j++){\
+        int i;\
+        const uint32_t a= AV_RN32(pixels  );\
+        const uint32_t b= AV_RN32(pixels+1);\
+        uint32_t l0=  (a&0x03030303UL)\
+                    + (b&0x03030303UL)\
+                    + 0x02020202UL;\
+        uint32_t h0= ((a&0xFCFCFCFCUL)>>2)\
+                   + ((b&0xFCFCFCFCUL)>>2);\
+        uint32_t l1,h1;\
+\
+        pixels+=line_size;\
+        for(i=0; i<h; i+=2){\
+            uint32_t a= AV_RN32(pixels  );\
+            uint32_t b= AV_RN32(pixels+1);\
+            l1=  (a&0x03030303UL)\
+               + (b&0x03030303UL);\
+            h1= ((a&0xFCFCFCFCUL)>>2)\
+              + ((b&0xFCFCFCFCUL)>>2);\
+            OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\
+            pixels+=line_size;\
+            block +=line_size;\
+            a= AV_RN32(pixels  );\
+            b= AV_RN32(pixels+1);\
+            l0=  (a&0x03030303UL)\
+               + (b&0x03030303UL)\
+               + 0x02020202UL;\
+            h0= ((a&0xFCFCFCFCUL)>>2)\
+              + ((b&0xFCFCFCFCUL)>>2);\
+            OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\
+            pixels+=line_size;\
+            block +=line_size;\
+        }\
+        pixels+=4-line_size*(h+1);\
+        block +=4-line_size*h;\
+    }\
+}\
+\
+static inline void FUNCC(OPNAME ## _no_rnd_pixels8_xy2)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)\
+{\
+    /* FIXME HIGH BIT DEPTH */\
+    int j;\
+    for(j=0; j<2; j++){\
+        int i;\
+        const uint32_t a= AV_RN32(pixels  );\
+        const uint32_t b= AV_RN32(pixels+1);\
+        uint32_t l0=  (a&0x03030303UL)\
+                    + (b&0x03030303UL)\
+                    + 0x01010101UL;\
+        uint32_t h0= ((a&0xFCFCFCFCUL)>>2)\
+                   + ((b&0xFCFCFCFCUL)>>2);\
+        uint32_t l1,h1;\
+\
+        pixels+=line_size;\
+        for(i=0; i<h; i+=2){\
+            uint32_t a= AV_RN32(pixels  );\
+            uint32_t b= AV_RN32(pixels+1);\
+            l1=  (a&0x03030303UL)\
+               + (b&0x03030303UL);\
+            h1= ((a&0xFCFCFCFCUL)>>2)\
+              + ((b&0xFCFCFCFCUL)>>2);\
+            OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\
+            pixels+=line_size;\
+            block +=line_size;\
+            a= AV_RN32(pixels  );\
+            b= AV_RN32(pixels+1);\
+            l0=  (a&0x03030303UL)\
+               + (b&0x03030303UL)\
+               + 0x01010101UL;\
+            h0= ((a&0xFCFCFCFCUL)>>2)\
+              + ((b&0xFCFCFCFCUL)>>2);\
+            OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\
+            pixels+=line_size;\
+            block +=line_size;\
+        }\
+        pixels+=4-line_size*(h+1);\
+        block +=4-line_size*h;\
+    }\
+}\
+\
+CALL_2X_PIXELS(FUNCC(OPNAME ## _pixels16_x2) , FUNCC(OPNAME ## _pixels8_x2) , 8*sizeof(pixel))\
+CALL_2X_PIXELS(FUNCC(OPNAME ## _pixels16_y2) , FUNCC(OPNAME ## _pixels8_y2) , 8*sizeof(pixel))\
+CALL_2X_PIXELS(FUNCC(OPNAME ## _pixels16_xy2), FUNCC(OPNAME ## _pixels8_xy2), 8*sizeof(pixel))\
+av_unused CALL_2X_PIXELS(FUNCC(OPNAME ## _no_rnd_pixels16)    , FUNCC(OPNAME ## _pixels8) , 8*sizeof(pixel))\
+CALL_2X_PIXELS(FUNCC(OPNAME ## _no_rnd_pixels16_x2) , FUNCC(OPNAME ## _no_rnd_pixels8_x2) , 8*sizeof(pixel))\
+CALL_2X_PIXELS(FUNCC(OPNAME ## _no_rnd_pixels16_y2) , FUNCC(OPNAME ## _no_rnd_pixels8_y2) , 8*sizeof(pixel))\
+CALL_2X_PIXELS(FUNCC(OPNAME ## _no_rnd_pixels16_xy2), FUNCC(OPNAME ## _no_rnd_pixels8_xy2), 8*sizeof(pixel))\
+
+#define op_avg(a, b) a = rnd_avg_pixel4(a, b)
+#define op_put(a, b) a = b
+#if BIT_DEPTH == 8
+#define put_no_rnd_pixels8_8_c put_pixels8_8_c
+PIXOP2(avg, op_avg)
+PIXOP2(put, op_put)
+#endif
+#undef op_avg
+#undef op_put

libavcodec/rnd_avg.h

@@ -19,8 +19,15 @@
 #ifndef AVCODEC_RND_AVG_H
 #define AVCODEC_RND_AVG_H
 
+#include <stddef.h>
 #include <stdint.h>
 
+#define CALL_2X_PIXELS(a, b, n)\
+static void a(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h){\
+    b(block  , pixels  , line_size, h);\
+    b(block+n, pixels+n, line_size, h);\
+}
+
 #define         BYTE_VEC32(c)   ((c)*0x01010101UL)
 #define         BYTE_VEC64(c)   ((c)*0x0001000100010001UL)