From 412ba501b166ce794cc5b1b6a4a09d7ddb82d01d Mon Sep 17 00:00:00 2001 From: Fabrice Bellard Date: Wed, 5 Jun 2002 14:25:06 +0000 Subject: [PATCH] removed unused code - began to merge alpha specific stuff - added mac macros for suitable CPUs Originally committed as revision 667 to svn://svn.ffmpeg.org/ffmpeg/trunk --- libavcodec/simple_idct.c | 505 +++++++++++++-------------------------- 1 file changed, 160 insertions(+), 345 deletions(-) diff --git a/libavcodec/simple_idct.c b/libavcodec/simple_idct.c index dc8bf3a003..a4d4486af5 100644 --- a/libavcodec/simple_idct.c +++ b/libavcodec/simple_idct.c @@ -47,61 +47,29 @@ #define COL_SHIFT 20 // 6 #endif -/* 8x8 Matrix used to do a trivial (slow) 8 point IDCT */ -static int coeff[64]={ - W4, W4, W4, W4, W4, W4, W4, W4, - W1, W3, W5, W7,-W7,-W5,-W3,-W1, - W2, W6,-W6,-W2,-W2,-W6, W6, W2, - W3,-W7,-W1,-W5, W5, W1, W7,-W3, - W4,-W4,-W4, W4, W4,-W4,-W4, W4, - W5,-W1, W7, W3,-W3,-W7, W1,-W5, - W6,-W2, W2,-W6,-W6, W2,-W2, W6, - W7,-W5, W3,-W1, W1,-W3, W5,-W7 -}; - -static inline int idctRowCondZ (int16_t * row) -{ - int a0, a1, a2, a3, b0, b1, b2, b3; +#ifdef ARCH_ALPHA +#define FAST_64BIT +#endif - if( !( ((uint32_t*)row)[0]|((uint32_t*)row)[1] |((uint32_t*)row)[2] |((uint32_t*)row)[3])) { -/* row[0] = row[1] = row[2] = row[3] = row[4] = - row[5] = row[6] = row[7] = 0;*/ - return 0; - } +#if defined(ARCH_POWERPC_405) - if(!( ((uint32_t*)row)[2] |((uint32_t*)row)[3] )){ - a0 = W4*row[0] + W2*row[2] + (1<<(ROW_SHIFT-1)); - a1 = W4*row[0] + W6*row[2] + (1<<(ROW_SHIFT-1)); - a2 = W4*row[0] - W6*row[2] + (1<<(ROW_SHIFT-1)); - a3 = W4*row[0] - W2*row[2] + (1<<(ROW_SHIFT-1)); - - b0 = W1*row[1] + W3*row[3]; - b1 = W3*row[1] - W7*row[3]; - b2 = W5*row[1] - W1*row[3]; - b3 = W7*row[1] - W5*row[3]; - }else{ - a0 = W4*row[0] + W2*row[2] + W4*row[4] + W6*row[6] + (1<<(ROW_SHIFT-1)); - a1 = W4*row[0] + W6*row[2] - W4*row[4] - W2*row[6] + (1<<(ROW_SHIFT-1)); - a2 = W4*row[0] - W6*row[2] - W4*row[4] + W2*row[6] + (1<<(ROW_SHIFT-1)); - a3 = W4*row[0] - W2*row[2] + W4*row[4] - W6*row[6] + (1<<(ROW_SHIFT-1)); - - b0 = W1*row[1] + W3*row[3] + W5*row[5] + W7*row[7]; - b1 = W3*row[1] - W7*row[3] - W1*row[5] - W5*row[7]; - b2 = W5*row[1] - W1*row[3] + W7*row[5] + W3*row[7]; - b3 = W7*row[1] - W5*row[3] + W3*row[5] - W1*row[7]; - } +/* signed 16x16 -> 32 multiply add accumulate */ +#define MAC16(rt, ra, rb) \ + asm ("maclhw %0, %2, %3" : "=r" (rt) : "0" (rt), "r" (ra), "r" (rb)); - row[0] = (a0 + b0) >> ROW_SHIFT; - row[1] = (a1 + b1) >> ROW_SHIFT; - row[2] = (a2 + b2) >> ROW_SHIFT; - row[3] = (a3 + b3) >> ROW_SHIFT; - row[4] = (a3 - b3) >> ROW_SHIFT; - row[5] = (a2 - b2) >> ROW_SHIFT; - row[6] = (a1 - b1) >> ROW_SHIFT; - row[7] = (a0 - b0) >> ROW_SHIFT; - - return 1; -} +/* signed 16x16 -> 32 multiply */ +#define MUL16(rt, ra, rb) \ + asm ("mullhw %0, %1, %2" : "=r" (rt) : "r" (ra), "r" (rb)); + +#else + +/* signed 16x16 -> 32 multiply add accumulate */ +#define MAC16(rt, ra, rb) rt += (ra) * (rb) + +/* signed 16x16 -> 32 multiply */ +#define MUL16(rt, ra, rb) rt = (ra) * (rb) + +#endif #ifdef ARCH_ALPHA /* 0: all entries 0, 1: only first entry nonzero, 2: otherwise */ @@ -129,10 +97,10 @@ static inline int idctRowCondDC(int16_t *row) } } - a0 = W4 * row[0]; - a1 = W4 * row[0]; - a2 = W4 * row[0]; - a3 = W4 * row[0]; + a0 = (W4 * row[0]) + (1 << (ROW_SHIFT - 1)); + a1 = a0; + a2 = a0; + a3 = a0; if (row[2]) { a0 += W2 * row[2]; @@ -155,11 +123,6 @@ static inline int idctRowCondDC(int16_t *row) a3 -= W6 * row[6]; } - a0 += 1 << (ROW_SHIFT - 1); - a1 += 1 << (ROW_SHIFT - 1); - a2 += 1 << (ROW_SHIFT - 1); - a3 += 1 << (ROW_SHIFT - 1); - if (row[1]) { b0 = W1 * row[1]; b1 = W3 * row[1]; @@ -205,38 +168,86 @@ static inline int idctRowCondDC(int16_t *row) return 2; } #else /* not ARCH_ALPHA */ -static inline int idctRowCondDC (int16_t * row) + +static inline void idctRowCondDC (int16_t * row) { int a0, a1, a2, a3, b0, b1, b2, b3; +#ifdef FAST_64BIT + uint64_t temp; +#else + uint32_t temp; +#endif - if( !( ((uint32_t*)row)[1] |((uint32_t*)row)[2] |((uint32_t*)row)[3]| row[1])) { -// row[0] = row[1] = row[2] = row[3] = row[4] = row[5] = row[6] = row[7] = row[0]<<3; - uint16_t temp= row[0]<<3; - ((uint32_t*)row)[0]=((uint32_t*)row)[1]= - ((uint32_t*)row)[2]=((uint32_t*)row)[3]= temp + (temp<<16); - return 0; +#ifdef FAST_64BIT +#ifdef WORDS_BIGENDIAN +#define ROW0_MASK 0xffff000000000000LL +#else +#define ROW0_MASK 0xffffLL +#endif + if ( ((((uint64_t *)row)[0] & ~ROW0_MASK) | + ((uint64_t *)row)[1]) == 0) { + temp = (row[0] << 3) & 0xffff; + temp += temp << 16; + temp += temp << 32; + ((uint64_t *)row)[0] = temp; + ((uint64_t *)row)[1] = temp; + return; } +#else + if (!(((uint32_t*)row)[1] | + ((uint32_t*)row)[2] | + ((uint32_t*)row)[3] | + row[1])) { + temp = (row[0] << 3) & 0xffff; + temp += temp << 16; + ((uint32_t*)row)[0]=((uint32_t*)row)[1] = + ((uint32_t*)row)[2]=((uint32_t*)row)[3] = temp; + return; + } +#endif - if(!( ((uint32_t*)row)[2] |((uint32_t*)row)[3] )){ - a0 = W4*row[0] + W2*row[2] + (1<<(ROW_SHIFT-1)); - a1 = W4*row[0] + W6*row[2] + (1<<(ROW_SHIFT-1)); - a2 = W4*row[0] - W6*row[2] + (1<<(ROW_SHIFT-1)); - a3 = W4*row[0] - W2*row[2] + (1<<(ROW_SHIFT-1)); - - b0 = W1*row[1] + W3*row[3]; - b1 = W3*row[1] - W7*row[3]; - b2 = W5*row[1] - W1*row[3]; - b3 = W7*row[1] - W5*row[3]; - }else{ - a0 = W4*row[0] + W2*row[2] + W4*row[4] + W6*row[6] + (1<<(ROW_SHIFT-1)); - a1 = W4*row[0] + W6*row[2] - W4*row[4] - W2*row[6] + (1<<(ROW_SHIFT-1)); - a2 = W4*row[0] - W6*row[2] - W4*row[4] + W2*row[6] + (1<<(ROW_SHIFT-1)); - a3 = W4*row[0] - W2*row[2] + W4*row[4] - W6*row[6] + (1<<(ROW_SHIFT-1)); - - b0 = W1*row[1] + W3*row[3] + W5*row[5] + W7*row[7]; - b1 = W3*row[1] - W7*row[3] - W1*row[5] - W5*row[7]; - b2 = W5*row[1] - W1*row[3] + W7*row[5] + W3*row[7]; - b3 = W7*row[1] - W5*row[3] + W3*row[5] - W1*row[7]; + a0 = (W4 * row[0]) + (1 << (ROW_SHIFT - 1)); + a1 = a0; + a2 = a0; + a3 = a0; + + /* no need to optimize : gcc does it */ + a0 += W2 * row[2]; + a1 += W6 * row[2]; + a2 -= W6 * row[2]; + a3 -= W2 * row[2]; + + MUL16(b0, W1, row[1]); + MAC16(b0, W3, row[3]); + MUL16(b1, W3, row[1]); + MAC16(b1, -W7, row[3]); + MUL16(b2, W5, row[1]); + MAC16(b2, -W1, row[3]); + MUL16(b3, W7, row[1]); + MAC16(b3, -W5, row[3]); + +#ifdef FAST_64BIT + temp = ((uint64_t*)row)[1]; +#else + temp = ((uint32_t*)row)[2] | ((uint32_t*)row)[3]; +#endif + if (temp != 0) { + a0 += W4*row[4] + W6*row[6]; + a1 += - W4*row[4] - W2*row[6]; + a2 += - W4*row[4] + W2*row[6]; + a3 += W4*row[4] - W6*row[6]; + + MAC16(b0, W5, row[5]); + MAC16(b0, W7, row[7]); + + MAC16(b1, -W1, row[5]); + MAC16(b1, -W5, row[7]); + + MAC16(b2, W7, row[5]); + MAC16(b2, W3, row[7]); + + MAC16(b3, W3, row[5]); + MAC16(b3, -W1, row[7]); } row[0] = (a0 + b0) >> ROW_SHIFT; @@ -247,202 +258,70 @@ static inline int idctRowCondDC (int16_t * row) row[5] = (a2 - b2) >> ROW_SHIFT; row[3] = (a3 + b3) >> ROW_SHIFT; row[4] = (a3 - b3) >> ROW_SHIFT; - - return 1; } #endif /* not ARCH_ALPHA */ -static inline void idctCol (int16_t * col) -{ - -/* - if( !(col[8*1] | col[8*2] |col[8*3] |col[8*4] |col[8*5] |col[8*6] | col[8*7])) { - col[8*0] = col[8*1] = col[8*2] = col[8*3] = col[8*4] = - col[8*5] = col[8*6] = col[8*7] = col[8*0]<<3; - return; - }*/ - - int a0, a1, a2, a3, b0, b1, b2, b3; - col[0] += (1<<(COL_SHIFT-1))/W4; - a0 = W4*col[8*0] + W2*col[8*2] + W4*col[8*4] + W6*col[8*6]; - a1 = W4*col[8*0] + W6*col[8*2] - W4*col[8*4] - W2*col[8*6]; - a2 = W4*col[8*0] - W6*col[8*2] - W4*col[8*4] + W2*col[8*6]; - a3 = W4*col[8*0] - W2*col[8*2] + W4*col[8*4] - W6*col[8*6]; - - b0 = W1*col[8*1] + W3*col[8*3] + W5*col[8*5] + W7*col[8*7]; - b1 = W3*col[8*1] - W7*col[8*3] - W1*col[8*5] - W5*col[8*7]; - b2 = W5*col[8*1] - W1*col[8*3] + W7*col[8*5] + W3*col[8*7]; - b3 = W7*col[8*1] - W5*col[8*3] + W3*col[8*5] - W1*col[8*7]; - - col[8*0] = (a0 + b0) >> COL_SHIFT; - col[8*7] = (a0 - b0) >> COL_SHIFT; - col[8*1] = (a1 + b1) >> COL_SHIFT; - col[8*6] = (a1 - b1) >> COL_SHIFT; - col[8*2] = (a2 + b2) >> COL_SHIFT; - col[8*5] = (a2 - b2) >> COL_SHIFT; - col[8*3] = (a3 + b3) >> COL_SHIFT; - col[8*4] = (a3 - b3) >> COL_SHIFT; -} - static inline void idctSparseCol (int16_t * col) { int a0, a1, a2, a3, b0, b1, b2, b3; - col[0] += (1<<(COL_SHIFT-1))/W4; - a0 = W4*col[8*0]; - a1 = W4*col[8*0]; - a2 = W4*col[8*0]; - a3 = W4*col[8*0]; - - if(col[8*2]){ - a0 += + W2*col[8*2]; - a1 += + W6*col[8*2]; - a2 += - W6*col[8*2]; - a3 += - W2*col[8*2]; - } - if(col[8*4]){ - a0 += + W4*col[8*4]; - a1 += - W4*col[8*4]; - a2 += - W4*col[8*4]; - a3 += + W4*col[8*4]; - } + /* XXX: I did that only to give same values as previous code */ + a0 = W4 * (col[8*0] + ((1<<(COL_SHIFT-1))/W4)); + a1 = a0; + a2 = a0; + a3 = a0; - if(col[8*6]){ - a0 += + W6*col[8*6]; - a1 += - W2*col[8*6]; - a2 += + W2*col[8*6]; - a3 += - W6*col[8*6]; - } - - if(col[8*1]){ - b0 = W1*col[8*1]; - b1 = W3*col[8*1]; - b2 = W5*col[8*1]; - b3 = W7*col[8*1]; - }else{ - b0 = - b1 = - b2 = - b3 = 0; - } - - if(col[8*3]){ - b0 += + W3*col[8*3]; - b1 += - W7*col[8*3]; - b2 += - W1*col[8*3]; - b3 += - W5*col[8*3]; - } - - if(col[8*5]){ - b0 += + W5*col[8*5]; - b1 += - W1*col[8*5]; - b2 += + W7*col[8*5]; - b3 += + W3*col[8*5]; - } - - if(col[8*7]){ - b0 += + W7*col[8*7]; - b1 += - W5*col[8*7]; - b2 += + W3*col[8*7]; - b3 += - W1*col[8*7]; - } + a0 += + W2*col[8*2]; + a1 += + W6*col[8*2]; + a2 += - W6*col[8*2]; + a3 += - W2*col[8*2]; -#ifndef ARCH_ALPHA - if(!(b0|b1|b2|b3)){ - col[8*0] = (a0) >> COL_SHIFT; - col[8*7] = (a0) >> COL_SHIFT; - col[8*1] = (a1) >> COL_SHIFT; - col[8*6] = (a1) >> COL_SHIFT; - col[8*2] = (a2) >> COL_SHIFT; - col[8*5] = (a2) >> COL_SHIFT; - col[8*3] = (a3) >> COL_SHIFT; - col[8*4] = (a3) >> COL_SHIFT; - }else{ -#endif - col[8*0] = (a0 + b0) >> COL_SHIFT; - col[8*7] = (a0 - b0) >> COL_SHIFT; - col[8*1] = (a1 + b1) >> COL_SHIFT; - col[8*6] = (a1 - b1) >> COL_SHIFT; - col[8*2] = (a2 + b2) >> COL_SHIFT; - col[8*5] = (a2 - b2) >> COL_SHIFT; - col[8*3] = (a3 + b3) >> COL_SHIFT; - col[8*4] = (a3 - b3) >> COL_SHIFT; -#ifndef ARCH_ALPHA - } -#endif -} + MUL16(b0, W1, col[8*1]); + MUL16(b1, W3, col[8*1]); + MUL16(b2, W5, col[8*1]); + MUL16(b3, W7, col[8*1]); -static inline void idctSparse2Col (int16_t * col) -{ - int a0, a1, a2, a3, b0, b1, b2, b3; - col[0] += (1<<(COL_SHIFT-1))/W4; - a0 = W4*col[8*0]; - a1 = W4*col[8*0]; - a2 = W4*col[8*0]; - a3 = W4*col[8*0]; - - if(col[8*2]){ - a0 += + W2*col[8*2]; - a1 += + W6*col[8*2]; - a2 += - W6*col[8*2]; - a3 += - W2*col[8*2]; - } + MAC16(b0, + W3, col[8*3]); + MAC16(b1, - W7, col[8*3]); + MAC16(b2, - W1, col[8*3]); + MAC16(b3, - W5, col[8*3]); if(col[8*4]){ - a0 += + W4*col[8*4]; - a1 += - W4*col[8*4]; - a2 += - W4*col[8*4]; - a3 += + W4*col[8*4]; + a0 += + W4*col[8*4]; + a1 += - W4*col[8*4]; + a2 += - W4*col[8*4]; + a3 += + W4*col[8*4]; } - if(col[8*6]){ - a0 += + W6*col[8*6]; - a1 += - W2*col[8*6]; - a2 += + W2*col[8*6]; - a3 += - W6*col[8*6]; - } - - if(col[8*1] || 1){ - b0 = W1*col[8*1]; - b1 = W3*col[8*1]; - b2 = W5*col[8*1]; - b3 = W7*col[8*1]; - }else{ - b0 = - b1 = - b2 = - b3 = 0; + if (col[8*5]) { + MAC16(b0, + W5, col[8*5]); + MAC16(b1, - W1, col[8*5]); + MAC16(b2, + W7, col[8*5]); + MAC16(b3, + W3, col[8*5]); } - if(col[8*3]){ - b0 += + W3*col[8*3]; - b1 += - W7*col[8*3]; - b2 += - W1*col[8*3]; - b3 += - W5*col[8*3]; - } - - if(col[8*5]){ - b0 += + W5*col[8*5]; - b1 += - W1*col[8*5]; - b2 += + W7*col[8*5]; - b3 += + W3*col[8*5]; - } - - if(col[8*7]){ - b0 += + W7*col[8*7]; - b1 += - W5*col[8*7]; - b2 += + W3*col[8*7]; - b3 += - W1*col[8*7]; - } - - col[8*0] = (a0 + b0) >> COL_SHIFT; - col[8*7] = (a0 - b0) >> COL_SHIFT; - col[8*1] = (a1 + b1) >> COL_SHIFT; - col[8*6] = (a1 - b1) >> COL_SHIFT; - col[8*2] = (a2 + b2) >> COL_SHIFT; - col[8*5] = (a2 - b2) >> COL_SHIFT; - col[8*3] = (a3 + b3) >> COL_SHIFT; - col[8*4] = (a3 - b3) >> COL_SHIFT; + if(col[8*6]){ + a0 += + W6*col[8*6]; + a1 += - W2*col[8*6]; + a2 += + W2*col[8*6]; + a3 += - W6*col[8*6]; + } + + if (col[8*7]) { + MAC16(b0, + W7, col[8*7]); + MAC16(b1, - W5, col[8*7]); + MAC16(b2, + W3, col[8*7]); + MAC16(b3, - W1, col[8*7]); + } + + col[8*0] = (a0 + b0) >> COL_SHIFT; + col[8*7] = (a0 - b0) >> COL_SHIFT; + col[8*1] = (a1 + b1) >> COL_SHIFT; + col[8*6] = (a1 - b1) >> COL_SHIFT; + col[8*2] = (a2 + b2) >> COL_SHIFT; + col[8*5] = (a2 - b2) >> COL_SHIFT; + col[8*3] = (a3 + b3) >> COL_SHIFT; + col[8*4] = (a3 - b3) >> COL_SHIFT; } #ifdef ARCH_ALPHA @@ -472,82 +351,11 @@ static inline void idctCol2(int16_t *col) lcol[12] = l; lcol[13] = r; lcol[14] = l; lcol[15] = r; } -#endif void simple_idct (short *block) { int i; - -#if 0 - int nonZero[8]; - int buffer[64]; - int nNonZero=0; - - idctRowCondDC(block); - - for(i=1; i<8; i++) - { - nonZero[nNonZero]=i; - nNonZero+= idctRowCondZ(block + i*8); - } - - if(nNonZero==0) - { - for(i=0; i<8; i++) - { - block[i ]= - block[i+8 ]= - block[i+16]= - block[i+24]= - block[i+32]= - block[i+40]= - block[i+48]= - block[i+56]= (W4*block[i] + (1<<(COL_SHIFT-1))) >> COL_SHIFT; - } - } - else if(nNonZero==1) - { - int index= nonZero[0]*8; - for(i=0; i<8; i++) - { - int bias= W4*block[i] + (1<<(COL_SHIFT-1)); - int c= block[i + index]; - block[i ]= (c*coeff[index ] + bias) >> COL_SHIFT; - block[i+8 ]= (c*coeff[index+1] + bias) >> COL_SHIFT; - block[i+16]= (c*coeff[index+2] + bias) >> COL_SHIFT; - block[i+24]= (c*coeff[index+3] + bias) >> COL_SHIFT; - block[i+32]= (c*coeff[index+4] + bias) >> COL_SHIFT; - block[i+40]= (c*coeff[index+5] + bias) >> COL_SHIFT; - block[i+48]= (c*coeff[index+6] + bias) >> COL_SHIFT; - block[i+56]= (c*coeff[index+7] + bias) >> COL_SHIFT; - } - } -/* else if(nNonZero==2) - { - int index1= nonZero[0]*8; - int index2= nonZero[1]*8; - for(i=0; i<8; i++) - { - int bias= W4*block[i] + (1<<(COL_SHIFT-1)); - int c1= block[i + index1]; - int c2= block[i + index2]; - block[i ]= (c1*coeff[index1 ] + c2*coeff[index2 ] + bias) >> COL_SHIFT; - block[i+8 ]= (c1*coeff[index1+1] + c2*coeff[index2+1] + bias) >> COL_SHIFT; - block[i+16]= (c1*coeff[index1+2] + c2*coeff[index2+2] + bias) >> COL_SHIFT; - block[i+24]= (c1*coeff[index1+3] + c2*coeff[index2+3] + bias) >> COL_SHIFT; - block[i+32]= (c1*coeff[index1+4] + c2*coeff[index2+4] + bias) >> COL_SHIFT; - block[i+40]= (c1*coeff[index1+5] + c2*coeff[index2+5] + bias) >> COL_SHIFT; - block[i+48]= (c1*coeff[index1+6] + c2*coeff[index2+6] + bias) >> COL_SHIFT; - block[i+56]= (c1*coeff[index1+7] + c2*coeff[index2+7] + bias) >> COL_SHIFT; - } - }*/ - else - { - for(i=0; i<8; i++) - idctSparse2Col(block + i); - } -#elif defined(ARCH_ALPHA) int rowsZero = 1; /* all rows except row 0 zero */ int rowsConstant = 1; /* all rows consist of a constant value */ @@ -579,13 +387,20 @@ void simple_idct (short *block) for (i = 0; i < 8; i++) idctSparseCol(block + i); } +} + #else - for(i=0; i<8; i++) - idctRowCondDC(block + i*8); - - for(i=0; i<8; i++) - idctSparseCol(block + i); -#endif + +void simple_idct (short *block) +{ + int i; + for(i=0; i<8; i++) + idctRowCondDC(block + i*8); + + for(i=0; i<8; i++) + idctSparseCol(block + i); } +#endif + #undef COL_SHIFT