6  * 8x8 Fast Walsh Hadamard Transform in sequency order based on the paper:
20  * be guaranteed that the magic 8 byte sequence (see below) can
33 	1,  8,
56 	s16 block[8 * 8];  in rlc()
66 	for (y = 0; y < 8; y++) {  in rlc()
67 		for (x = 0; x < 8; x++) {  in rlc()
68 			*wp = in[x + y * 8];  in rlc()
80 	to_encode = 8 * 8 - (lastzero_run > 14 ? lastzero_run : 0);  in rlc()
111  * one s16 value for the header and 8 * 8 coefficients of type s16.
120 	s16 block[8 * 8 + 16];  in derlc()
133 	 * So block has to be 8 * 8 + 16 bytes, the '+ 16' is to  in derlc()
136 	while (dec_count < 8 * 8) {  in derlc()
164 		int y = pos / 8;  in derlc()
165 		int x = pos % 8;  in derlc()
167 		dwht_out[x + y * 8] = *wp++;  in derlc()
181 	2, 2, 3, 6, 6, 6,  6,  8,
200 	for (j = 0; j < 8; j++) {  in quantize_intra()
201 		for (i = 0; i < 8; i++, quant++, coeff++, de_coeff++) {  in quantize_intra()
216 	for (j = 0; j < 8; j++)  in dequantize_intra()
217 		for (i = 0; i < 8; i++, quant++, coeff++)  in dequantize_intra()
226 	for (j = 0; j < 8; j++) {  in quantize_inter()
227 		for (i = 0; i < 8; i++, quant++, coeff++, de_coeff++) {  in quantize_inter()
242 	for (j = 0; j < 8; j++)  in dequantize_inter()
243 		for (i = 0; i < 8; i++, quant++, coeff++)  in dequantize_inter()
251 	/* we'll need more than 8 bits for the transformed coefficients */  in fwht()
252 	s32 workspace1[8], workspace2[8];  in fwht()
259 	for (i = 0; i < 8; i++, tmp += stride, out += 8) {  in fwht()
281 			workspace1[4]  = tmp[8] + tmp[10] - add;  in fwht()
282 			workspace1[5]  = tmp[8] - tmp[10];  in fwht()
304 			workspace1[2]  = tmp[8] + tmp[12] - add;  in fwht()
305 			workspace1[3]  = tmp[8] - tmp[12];  in fwht()
339 	for (i = 0; i < 8; i++, out++) {  in fwht()
341 		workspace1[0]  = out[0] + out[1 * 8];  in fwht()
342 		workspace1[1]  = out[0] - out[1 * 8];  in fwht()
344 		workspace1[2]  = out[2 * 8] + out[3 * 8];  in fwht()
345 		workspace1[3]  = out[2 * 8] - out[3 * 8];  in fwht()
347 		workspace1[4]  = out[4 * 8] + out[5 * 8];  in fwht()
348 		workspace1[5]  = out[4 * 8] - out[5 * 8];  in fwht()
350 		workspace1[6]  = out[6 * 8] + out[7 * 8];  in fwht()
351 		workspace1[7]  = out[6 * 8] - out[7 * 8];  in fwht()
364 		out[0 * 8] = workspace2[0] + workspace2[4];  in fwht()
365 		out[1 * 8] = workspace2[0] - workspace2[4];  in fwht()
366 		out[2 * 8] = workspace2[1] - workspace2[5];  in fwht()
367 		out[3 * 8] = workspace2[1] + workspace2[5];  in fwht()
368 		out[4 * 8] = workspace2[2] + workspace2[6];  in fwht()
369 		out[5 * 8] = workspace2[2] - workspace2[6];  in fwht()
370 		out[6 * 8] = workspace2[3] - workspace2[7];  in fwht()
371 		out[7 * 8] = workspace2[3] + workspace2[7];  in fwht()
377  * that of the I-blocks. Therefore we need a type bigger than 8 bits.
384 	/* we'll need more than 8 bits for the transformed coefficients */  in fwht16()
385 	s32 workspace1[8], workspace2[8];  in fwht16()
390 	for (i = 0; i < 8; i++, tmp += stride, out += 8) {  in fwht16()
428 	for (i = 0; i < 8; i++, out++) {  in fwht16()
430 		workspace1[0]  = out[0] + out[1*8];  in fwht16()
431 		workspace1[1]  = out[0] - out[1*8];  in fwht16()
433 		workspace1[2]  = out[2*8] + out[3*8];  in fwht16()
434 		workspace1[3]  = out[2*8] - out[3*8];  in fwht16()
436 		workspace1[4]  = out[4*8] + out[5*8];  in fwht16()
437 		workspace1[5]  = out[4*8] - out[5*8];  in fwht16()
439 		workspace1[6]  = out[6*8] + out[7*8];  in fwht16()
440 		workspace1[7]  = out[6*8] - out[7*8];  in fwht16()
454 		out[0*8] = workspace2[0] + workspace2[4];  in fwht16()
455 		out[1*8] = workspace2[0] - workspace2[4];  in fwht16()
456 		out[2*8] = workspace2[1] - workspace2[5];  in fwht16()
457 		out[3*8] = workspace2[1] + workspace2[5];  in fwht16()
458 		out[4*8] = workspace2[2] + workspace2[6];  in fwht16()
459 		out[5*8] = workspace2[2] - workspace2[6];  in fwht16()
460 		out[6*8] = workspace2[3] - workspace2[7];  in fwht16()
461 		out[7*8] = workspace2[3] + workspace2[7];  in fwht16()
469 	 * we'll need more than 8 bits for the transformed coefficients  in ifwht()
472 	int workspace1[8], workspace2[8];  in ifwht()
478 	for (i = 0; i < 8; i++, tmp += 8, out += 8) {  in ifwht()
516 	for (i = 0; i < 8; i++, out++) {  in ifwht()
518 		workspace1[0]  = out[0] + out[1 * 8];  in ifwht()
519 		workspace1[1]  = out[0] - out[1 * 8];  in ifwht()
521 		workspace1[2]  = out[2 * 8] + out[3 * 8];  in ifwht()
522 		workspace1[3]  = out[2 * 8] - out[3 * 8];  in ifwht()
524 		workspace1[4]  = out[4 * 8] + out[5 * 8];  in ifwht()
525 		workspace1[5]  = out[4 * 8] - out[5 * 8];  in ifwht()
527 		workspace1[6]  = out[6 * 8] + out[7 * 8];  in ifwht()
528 		workspace1[7]  = out[6 * 8] - out[7 * 8];  in ifwht()
545 			out[0 * 8] = workspace2[0] + workspace2[4];  in ifwht()
546 			out[1 * 8] = workspace2[0] - workspace2[4];  in ifwht()
547 			out[2 * 8] = workspace2[1] - workspace2[5];  in ifwht()
548 			out[3 * 8] = workspace2[1] + workspace2[5];  in ifwht()
549 			out[4 * 8] = workspace2[2] + workspace2[6];  in ifwht()
550 			out[5 * 8] = workspace2[2] - workspace2[6];  in ifwht()
551 			out[6 * 8] = workspace2[3] - workspace2[7];  in ifwht()
552 			out[7 * 8] = workspace2[3] + workspace2[7];  in ifwht()
554 			for (d = 0; d < 8; d++)  in ifwht()
555 				out[8 * d] >>= 6;  in ifwht()
559 			out[0 * 8] = workspace2[0] + workspace2[4];  in ifwht()
560 			out[1 * 8] = workspace2[0] - workspace2[4];  in ifwht()
561 			out[2 * 8] = workspace2[1] - workspace2[5];  in ifwht()
562 			out[3 * 8] = workspace2[1] + workspace2[5];  in ifwht()
563 			out[4 * 8] = workspace2[2] + workspace2[6];  in ifwht()
564 			out[5 * 8] = workspace2[2] - workspace2[6];  in ifwht()
565 			out[6 * 8] = workspace2[3] - workspace2[7];  in ifwht()
566 			out[7 * 8] = workspace2[3] + workspace2[7];  in ifwht()
568 			for (d = 0; d < 8; d++) {  in ifwht()
569 				out[8 * d] >>= 6;  in ifwht()
570 				out[8 * d] += 128;  in ifwht()
581 	for (i = 0; i < 8; i++) {  in fill_encoder_block()
582 		for (j = 0; j < 8; j++, input += input_step)  in fill_encoder_block()
584 		input += stride - 8 * input_step;  in fill_encoder_block()
595 	for (i = 0; i < 8 * 8; i++, tmp++)  in var_intra()
599 	for (i = 0; i < 8 * 8; i++, tmp++)  in var_intra()
609 	for (i = 0; i < 8 * 8; i++, old++, new++)  in var_inter()
626 	fill_encoder_block(reference, old, 8, 1);  in decide_blocktype()
629 	for (k = 0; k < 8; k++) {  in decide_blocktype()
630 		for (l = 0; l < 8; l++) {  in decide_blocktype()
647 	for (i = 0; i < 8; i++) {  in fill_decoder_block()
648 		for (j = 0; j < 8; j++, input++, dst += dst_step) {  in fill_decoder_block()
656 		dst += stride - (8 * dst_step);  in fill_decoder_block()
665 	for (k = 0; k < 8; k++) {  in add_deltas()
666 		for (l = 0; l < 8; l++) {  in add_deltas()
679 		ref += stride - (8 * ref_step);  in add_deltas()
696 	width = round_up(width, 8);  in encode_plane()
697 	height = round_up(height, 8);  in encode_plane()
699 	for (j = 0; j < height / 8; j++) {  in encode_plane()
700 		input = input_start + j * 8 * stride;  in encode_plane()
701 		for (i = 0; i < width / 8; i++) {  in encode_plane()
716 				fwht16(deltablock, cf->coeffs, 8, 0);  in encode_plane()
724 					add_deltas(cf->de_fwht, refp, 8, 1);  in encode_plane()
725 				fill_decoder_block(refp, cf->de_fwht, 8, 1);  in encode_plane()
728 			input += 8 * input_step;  in encode_plane()
729 			refp += 8 * 8;  in encode_plane()
841 	s16 copy[8 * 8];  in decode_plane()
846 	width = round_up(width, 8);  in decode_plane()
847 	height = round_up(height, 8);  in decode_plane()
868 	for (j = 0; j < height / 8; j++) {  in decode_plane()
869 		for (i = 0; i < width / 8; i++) {  in decode_plane()
870 			const u8 *refp = ref + j * 8 * ref_stride +  in decode_plane()
871 				i * 8 * ref_step;  in decode_plane()
872 			u8 *dstp = dst + j * 8 * dst_stride + i * 8 * dst_step;  in decode_plane()