From 1cca8d24232c9f81d71364c4725b3bc43eab637e Mon Sep 17 00:00:00 2001 From: Loren Merritt Date: Thu, 25 Dec 2008 01:20:37 +0000 Subject: flatten an array, since gcc fails at optimizing multidimensional arrays h264_idct8_add_c: 780 -> 735 cycles on conroe Originally committed as revision 16307 to svn://svn.ffmpeg.org/ffmpeg/trunk --- libavcodec/h264idct.c | 49 ++++++++++++++++++++++++------------------------- 1 file changed, 24 insertions(+), 25 deletions(-) (limited to 'libavcodec/h264idct.c') diff --git a/libavcodec/h264idct.c b/libavcodec/h264idct.c index 7ef7a2a4d5..57d1b75452 100644 --- a/libavcodec/h264idct.c +++ b/libavcodec/h264idct.c @@ -72,58 +72,57 @@ void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block){ void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride){ int i; - DCTELEM (*src)[8] = (DCTELEM(*)[8])block; uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; block[0] += 32; for( i = 0; i < 8; i++ ) { - const int a0 = src[i][0] + src[i][4]; - const int a2 = src[i][0] - src[i][4]; - const int a4 = (src[i][2]>>1) - src[i][6]; - const int a6 = (src[i][6]>>1) + src[i][2]; + const int a0 = block[0+i*8] + block[4+i*8]; + const int a2 = block[0+i*8] - block[4+i*8]; + const int a4 = (block[2+i*8]>>1) - block[6+i*8]; + const int a6 = (block[6+i*8]>>1) + block[2+i*8]; const int b0 = a0 + a6; const int b2 = a2 + a4; const int b4 = a2 - a4; const int b6 = a0 - a6; - const int a1 = -src[i][3] + src[i][5] - src[i][7] - (src[i][7]>>1); - const int a3 = src[i][1] + src[i][7] - src[i][3] - (src[i][3]>>1); - const int a5 = -src[i][1] + src[i][7] + src[i][5] + (src[i][5]>>1); - const int a7 = src[i][3] + src[i][5] + src[i][1] + (src[i][1]>>1); + const int a1 = -block[3+i*8] + block[5+i*8] - block[7+i*8] - (block[7+i*8]>>1); + const int a3 = block[1+i*8] + block[7+i*8] - block[3+i*8] - (block[3+i*8]>>1); + const int a5 = -block[1+i*8] + block[7+i*8] + block[5+i*8] + (block[5+i*8]>>1); + const int a7 = block[3+i*8] + block[5+i*8] + block[1+i*8] + (block[1+i*8]>>1); const int b1 = (a7>>2) + a1; const int b3 = a3 + (a5>>2); const int b5 = (a3>>2) - a5; const int b7 = a7 - (a1>>2); - src[i][0] = b0 + b7; - src[i][7] = b0 - b7; - src[i][1] = b2 + b5; - src[i][6] = b2 - b5; - src[i][2] = b4 + b3; - src[i][5] = b4 - b3; - src[i][3] = b6 + b1; - src[i][4] = b6 - b1; + block[0+i*8] = b0 + b7; + block[7+i*8] = b0 - b7; + block[1+i*8] = b2 + b5; + block[6+i*8] = b2 - b5; + block[2+i*8] = b4 + b3; + block[5+i*8] = b4 - b3; + block[3+i*8] = b6 + b1; + block[4+i*8] = b6 - b1; } for( i = 0; i < 8; i++ ) { - const int a0 = src[0][i] + src[4][i]; - const int a2 = src[0][i] - src[4][i]; - const int a4 = (src[2][i]>>1) - src[6][i]; - const int a6 = (src[6][i]>>1) + src[2][i]; + const int a0 = block[i+0*8] + block[i+4*8]; + const int a2 = block[i+0*8] - block[i+4*8]; + const int a4 = (block[i+2*8]>>1) - block[i+6*8]; + const int a6 = (block[i+6*8]>>1) + block[i+2*8]; const int b0 = a0 + a6; const int b2 = a2 + a4; const int b4 = a2 - a4; const int b6 = a0 - a6; - const int a1 = -src[3][i] + src[5][i] - src[7][i] - (src[7][i]>>1); - const int a3 = src[1][i] + src[7][i] - src[3][i] - (src[3][i]>>1); - const int a5 = -src[1][i] + src[7][i] + src[5][i] + (src[5][i]>>1); - const int a7 = src[3][i] + src[5][i] + src[1][i] + (src[1][i]>>1); + const int a1 = -block[i+3*8] + block[i+5*8] - block[i+7*8] - (block[i+7*8]>>1); + const int a3 = block[i+1*8] + block[i+7*8] - block[i+3*8] - (block[i+3*8]>>1); + const int a5 = -block[i+1*8] + block[i+7*8] + block[i+5*8] + (block[i+5*8]>>1); + const int a7 = block[i+3*8] + block[i+5*8] + block[i+1*8] + (block[i+1*8]>>1); const int b1 = (a7>>2) + a1; const int b3 = a3 + (a5>>2); -- cgit v1.2.3