seperated out the C-based VP3 DSP functions into a different file; also

ported the MMX-optimized versions of those functions


git-svn-id: file:///var/local/repositories/ffmpeg/trunk@2855 9553f0bf-9b14-0410-a0b8-cfaf0461ba5b

libavcodec/Makefile

@@ -20,7 +20,7 @@ OBJS= common.o utils.o mem.o allcodecs.o \
       roqvideo.o dpcm.o interplayvideo.o xan.o rpza.o cinepak.o msrle.o \
       msvideo1.o vqavideo.o idcinvideo.o adx.o rational.o faandct.o 8bps.o \
       smc.o parser.o flicvideo.o truemotion1.o vmdav.o lcl.o qtrle.o g726.o \
-      flac.o
+      flac.o vp3dsp.o
 
 ifeq ($(AMR_NB),yes)
 ifeq ($(AMR_NB_FIXED),yes)
@@ -116,7 +116,7 @@ ifeq ($(TARGET_MMX),yes)
 OBJS += i386/fdct_mmx.o i386/cputest.o \
 	i386/dsputil_mmx.o i386/mpegvideo_mmx.o \
 	i386/idct_mmx.o i386/motion_est_mmx.o \
-	i386/simple_idct_mmx.o i386/fft_sse.o
+	i386/simple_idct_mmx.o i386/fft_sse.o i386/vp3dsp_mmx.o
 ifdef TARGET_BUILTIN_VECTOR
 i386/fft_sse.o: CFLAGS+= -msse
 endif

libavcodec/dsputil.c

@@ -3124,6 +3124,11 @@ void dsputil_init(DSPContext* c, AVCodecContext *avctx)
         c->idct_permutation_type= FF_NO_IDCT_PERM;
     }
 
+    /* VP3 DSP support */
+    c->vp3_dsp_init = vp3_dsp_init_c;
+    c->vp3_idct_put = vp3_idct_put_c;
+    c->vp3_idct_add = vp3_idct_add_c;
+
     c->get_pixels = get_pixels_c;
     c->diff_pixels = diff_pixels_c;
     c->put_pixels_clamped = put_pixels_clamped_c;

libavcodec/dsputil.h

@@ -60,6 +60,19 @@ extern const uint8_t ff_zigzag248_direct[64];
 extern uint32_t squareTbl[512];
 extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
 
+/* VP3 DSP functions */
+void vp3_dsp_init_c(void);
+void vp3_idct_put_c(int16_t *input_data, int16_t *dequant_matrix,
+    int coeff_count, uint8_t *dest, int stride);
+void vp3_idct_add_c(int16_t *input_data, int16_t *dequant_matrix,
+    int coeff_count, uint8_t *dest, int stride);
+
+void vp3_dsp_init_mmx(void);
+void vp3_idct_put_mmx(int16_t *input_data, int16_t *dequant_matrix,
+    int coeff_count, uint8_t *dest, int stride);
+void vp3_idct_add_mmx(int16_t *input_data, int16_t *dequant_matrix,
+    int coeff_count, uint8_t *dest, int stride);
+
 
 /* minimum alignment rules ;)
 if u notice errors in the align stuff, need more alignment for some asm code for some cpu
@@ -292,6 +305,40 @@ typedef struct DSPContext {
 #define BASIS_SHIFT 16
 #define RECON_SHIFT 6
 
+    /**
+     * This function handles any initialization for the VP3 DSP functions.
+     */
+    void (*vp3_dsp_init)(void);
+
+    /** 
+     * This function is responsible for taking a block of zigzag'd,
+     * quantized DCT coefficients, reconstructing the original block of
+     * samples, and placing it into the output.
+     * @param input_data 64 zigzag'd, quantized DCT coefficients
+     * @param dequant_matrix 64 zigzag'd quantizer coefficients
+     * @param coeff_count index of the last coefficient
+     * @param dest the final output location where the transformed samples
+     * are to be placed
+     * @param stride the width in 8-bit samples of a line on this plane
+     */
+    void (*vp3_idct_put)(int16_t *input_data, int16_t *dequant_matrix,
+        int coeff_count, uint8_t *dest, int stride);
+
+    /** 
+     * This function is responsible for taking a block of zigzag'd,
+     * quantized DCT coefficients, reconstructing the original block of
+     * samples, and adding the transformed samples to an existing block of
+     * samples in the output.
+     * @param input_data 64 zigzag'd, quantized DCT coefficients
+     * @param dequant_matrix 64 zigzag'd quantizer coefficients
+     * @param coeff_count index of the last coefficient
+     * @param dest the final output location where the transformed samples
+     * are to be placed
+     * @param stride the width in 8-bit samples of a line on this plane
+     */
+    void (*vp3_idct_add)(int16_t *input_data, int16_t *dequant_matrix,
+        int coeff_count, uint8_t *dest, int stride);
+
 } DSPContext;
 
 void dsputil_static_init(void);

libavcodec/vp3dsp.c

+/*
+ * Copyright (C) 2004 the ffmpeg project
+ *
+ * This library 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 of the License, or (at your option) any later version.
+ *
+ * This library 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 this library; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+/**
+ * @file vp3dsp.c
+ * Standard C DSP-oriented functions cribbed from the original VP3 
+ * source code.
+ */
+
+#include "common.h"
+#include "avcodec.h"
+#include "vp3data.h"
+
+#define IdctAdjustBeforeShift 8
+#define xC1S7 64277
+#define xC2S6 60547
+#define xC3S5 54491
+#define xC4S4 46341
+#define xC5S3 36410
+#define xC6S2 25080
+#define xC7S1 12785
+
+void vp3_dsp_init_c(void)
+{
+    /* nop */
+}
+
+static void vp3_idct_c(int32_t *dequantized_data, int16_t *output_data)
+{
+    int32_t *ip = dequantized_data;
+    int16_t *op = output_data;
+
+    int32_t A_, B_, C_, D_, _Ad, _Bd, _Cd, _Dd, E_, F_, G_, H_;
+    int32_t _Ed, _Gd, _Add, _Bdd, _Fd, _Hd;
+    int32_t t1, t2;
+
+    int i;
+
+    /* Inverse DCT on the rows now */
+    for (i = 0; i < 8; i++) {
+        /* Check for non-zero values */
+        if ( ip[0] | ip[1] | ip[2] | ip[3] | ip[4] | ip[5] | ip[6] | ip[7] ) {
+            t1 = (int32_t)(xC1S7 * ip[1]);
+            t2 = (int32_t)(xC7S1 * ip[7]);
+            t1 >>= 16;
+            t2 >>= 16;
+            A_ = t1 + t2;
+
+            t1 = (int32_t)(xC7S1 * ip[1]);
+            t2 = (int32_t)(xC1S7 * ip[7]);
+            t1 >>= 16;
+            t2 >>= 16;
+            B_ = t1 - t2;
+
+            t1 = (int32_t)(xC3S5 * ip[3]);
+            t2 = (int32_t)(xC5S3 * ip[5]);
+            t1 >>= 16;
+            t2 >>= 16;
+            C_ = t1 + t2;
+
+            t1 = (int32_t)(xC3S5 * ip[5]);
+            t2 = (int32_t)(xC5S3 * ip[3]);
+            t1 >>= 16;
+            t2 >>= 16;
+            D_ = t1 - t2;
+
+
+            t1 = (int32_t)(xC4S4 * (A_ - C_));
+            t1 >>= 16;
+            _Ad = t1;
+
+            t1 = (int32_t)(xC4S4 * (B_ - D_));
+            t1 >>= 16;
+            _Bd = t1;
+
+
+            _Cd = A_ + C_;
+            _Dd = B_ + D_;
+
+            t1 = (int32_t)(xC4S4 * (ip[0] + ip[4]));
+            t1 >>= 16;
+            E_ = t1;
+
+            t1 = (int32_t)(xC4S4 * (ip[0] - ip[4]));
+            t1 >>= 16;
+            F_ = t1;
+
+            t1 = (int32_t)(xC2S6 * ip[2]);
+            t2 = (int32_t)(xC6S2 * ip[6]);
+            t1 >>= 16;
+            t2 >>= 16;
+            G_ = t1 + t2;
+
+            t1 = (int32_t)(xC6S2 * ip[2]);
+            t2 = (int32_t)(xC2S6 * ip[6]);
+            t1 >>= 16;
+            t2 >>= 16;
+            H_ = t1 - t2;
+
+
+            _Ed = E_ - G_;
+            _Gd = E_ + G_;
+
+            _Add = F_ + _Ad;
+            _Bdd = _Bd - H_;
+
+            _Fd = F_ - _Ad;
+            _Hd = _Bd + H_;
+
+            /*  Final sequence of operations over-write original inputs. */
+            ip[0] = (int16_t)((_Gd + _Cd )   >> 0);
+            ip[7] = (int16_t)((_Gd - _Cd )   >> 0);
+
+            ip[1] = (int16_t)((_Add + _Hd )  >> 0);
+            ip[2] = (int16_t)((_Add - _Hd )  >> 0);
+
+            ip[3] = (int16_t)((_Ed + _Dd )   >> 0);
+            ip[4] = (int16_t)((_Ed - _Dd )   >> 0);
+
+            ip[5] = (int16_t)((_Fd + _Bdd )  >> 0);
+            ip[6] = (int16_t)((_Fd - _Bdd )  >> 0);
+
+        }
+
+        ip += 8;            /* next row */
+    }
+
+    ip = dequantized_data;
+
+    for ( i = 0; i < 8; i++) {
+        /* Check for non-zero values (bitwise or faster than ||) */
+        if ( ip[0 * 8] | ip[1 * 8] | ip[2 * 8] | ip[3 * 8] |
+             ip[4 * 8] | ip[5 * 8] | ip[6 * 8] | ip[7 * 8] ) {
+
+            t1 = (int32_t)(xC1S7 * ip[1*8]);
+            t2 = (int32_t)(xC7S1 * ip[7*8]);
+            t1 >>= 16;
+            t2 >>= 16;
+            A_ = t1 + t2;
+
+            t1 = (int32_t)(xC7S1 * ip[1*8]);
+            t2 = (int32_t)(xC1S7 * ip[7*8]);
+            t1 >>= 16;
+            t2 >>= 16;
+            B_ = t1 - t2;
+
+            t1 = (int32_t)(xC3S5 * ip[3*8]);
+            t2 = (int32_t)(xC5S3 * ip[5*8]);
+            t1 >>= 16;
+            t2 >>= 16;
+            C_ = t1 + t2;
+
+            t1 = (int32_t)(xC3S5 * ip[5*8]);
+            t2 = (int32_t)(xC5S3 * ip[3*8]);
+            t1 >>= 16;
+            t2 >>= 16;
+            D_ = t1 - t2;
+
+
+            t1 = (int32_t)(xC4S4 * (A_ - C_));
+            t1 >>= 16;
+            _Ad = t1;
+
+            t1 = (int32_t)(xC4S4 * (B_ - D_));
+            t1 >>= 16;
+            _Bd = t1;
+
+
+            _Cd = A_ + C_;
+            _Dd = B_ + D_;
+
+            t1 = (int32_t)(xC4S4 * (ip[0*8] + ip[4*8]));
+            t1 >>= 16;
+            E_ = t1;
+
+            t1 = (int32_t)(xC4S4 * (ip[0*8] - ip[4*8]));
+            t1 >>= 16;
+            F_ = t1;
+
+            t1 = (int32_t)(xC2S6 * ip[2*8]);
+            t2 = (int32_t)(xC6S2 * ip[6*8]);
+            t1 >>= 16;
+            t2 >>= 16;
+            G_ = t1 + t2;
+
+            t1 = (int32_t)(xC6S2 * ip[2*8]);
+            t2 = (int32_t)(xC2S6 * ip[6*8]);
+            t1 >>= 16;
+            t2 >>= 16;
+            H_ = t1 - t2;
+
+
+            _Ed = E_ - G_;
+            _Gd = E_ + G_;
+
+            _Add = F_ + _Ad;
+            _Bdd = _Bd - H_;
+
+            _Fd = F_ - _Ad;
+            _Hd = _Bd + H_;
+
+            _Gd += IdctAdjustBeforeShift;
+            _Add += IdctAdjustBeforeShift;
+            _Ed += IdctAdjustBeforeShift;
+            _Fd += IdctAdjustBeforeShift;
+
+            /* Final sequence of operations over-write original inputs. */
+            op[0*8] = (int16_t)((_Gd + _Cd )   >> 4);
+            op[7*8] = (int16_t)((_Gd - _Cd )   >> 4);
+
+            op[1*8] = (int16_t)((_Add + _Hd )  >> 4);
+            op[2*8] = (int16_t)((_Add - _Hd )  >> 4);
+
+            op[3*8] = (int16_t)((_Ed + _Dd )   >> 4);
+            op[4*8] = (int16_t)((_Ed - _Dd )   >> 4);
+
+            op[5*8] = (int16_t)((_Fd + _Bdd )  >> 4);
+            op[6*8] = (int16_t)((_Fd - _Bdd )  >> 4);
+
+        } else {
+
+            op[0*8] = 0;
+            op[7*8] = 0;
+            op[1*8] = 0;
+            op[2*8] = 0;
+            op[3*8] = 0;
+            op[4*8] = 0;
+            op[5*8] = 0;
+            op[6*8] = 0;
+        }
+
+        ip++;            /* next column */
+        op++;
+    }
+}
+
+void vp3_idct_put_c(int16_t *input_data, int16_t *dequant_matrix,
+    int coeff_count, uint8_t *dest, int stride)
+{
+    int32_t dequantized_data[64];
+    int16_t transformed_data[64];
+    int16_t *op;
+    int i, j;
+
+    /* de-zigzag and dequantize */
+    for (i = 0; i < coeff_count; i++) {
+        j = dezigzag_index[i];
+        dequantized_data[j] = dequant_matrix[i] * input_data[i];
+    }
+
+    vp3_idct_c(dequantized_data, transformed_data);
+
+    /* place in final output */
+    op = transformed_data;
+    for (i = 0; i < 8; i++) {
+        for (j = 0; j < 8; j++) {
+            if (*op < -128)
+                *dest = 0;
+            else if (*op > 127)
+                *dest = 255;
+            else
+                *dest = (uint8_t)(*op + 128);
+            op++;
+            dest++;
+        }
+        dest += (stride - 8);
+    }
+}
+
+void vp3_idct_add_c(int16_t *input_data, int16_t *dequant_matrix,
+    int coeff_count, uint8_t *dest, int stride)
+{
+    int32_t dequantized_data[64];
+    int16_t transformed_data[64];
+    int16_t *op;
+    int i, j;
+    int16_t sample;
+
+    /* de-zigzag and dequantize */
+    for (i = 0; i < coeff_count; i++) {
+        j = dezigzag_index[i];
+        dequantized_data[j] = dequant_matrix[i] * input_data[i];
+    }
+
+    vp3_idct_c(dequantized_data, transformed_data);
+
+    /* place in final output */
+    op = transformed_data;
+    for (i = 0; i < 8; i++) {
+        for (j = 0; j < 8; j++) {
+            sample = *dest + *op;
+            if (sample < 0)
+                *dest = 0;
+            else if (sample > 255)
+                *dest = 255;
+            else
+                *dest = (uint8_t)(sample & 0xFF);
+            op++;
+            dest++;
+        }
+        dest += (stride - 8);
+    }
+}