smith_waterman_altivec.c

序列对齐 Compare a protein sequence to a protein sequence database or a DNA sequence to a DNA sequenc

        tmp = vec_subs(H,v_gapopen);
        E   = vec_max(E,tmp);
        
        // update F value
        F   = vec_subs(Fup,v_gapextend);
        tmp = vec_subs(Hup2,v_gapopen);
        F   = vec_max(F,tmp);
        
        v_score_load1 = vec_perm(v_score_load1,v_zero,merge_score_load);
        
        // add score to H
        H   = vec_adds(Hup1,v_score);
        H   = vec_subs(H,v_bias);
        
        // set H to max of H,E,F
        H   = vec_max(H,E);
        H   = vec_max(H,F);
        
        // Update highest score encountered this far
        v_maxscore = vec_max(v_maxscore,H);
        
        
        
        // PROLOGUE 7
        // prefetch next residue
        k                = db_sequence[7];
        
        v_score     = vec_perm(v_score_q1,  v_score_load1,  queue1_to_score);
        v_score_q1  = vec_perm(v_score_q2,  v_score_load1,  queue2_to_queue1);
        v_score_q2  = vec_perm(v_score_q3,  v_score_load1,  queue3_to_queue2);
        v_score_q3  = vec_perm(v_score_q4,  v_score_load1,  queue4_to_queue3);
        v_score_q4  = vec_perm(v_score_q5,  v_score_load1,  queue5_to_queue4);
        v_score_q5  = vec_perm(v_score_q5,  v_score_load1,  queue5_with_load);
        

        // prefetch score for next step 
        v_score_load1 = vec_ld(16*k,query_profile_byte);            
        
        // load values of F and H from previous row (one unit up)
        Fup    = vec_ld(0,  p);
        Hup1   = vec_ld(16, p);
        p += 32; // move ahead 32 bytes
        
        // shift into place so we have complete F and H vectors
        // that refer to the values one unit up from each cell
        // that we are currently working on.
        Fup    = vec_sld(Fup,F,15);
        Hup1    = vec_sld(Hup1,H,15);            
        
        // do the dynamic programming 
        
        // update E value
        E   = vec_subs(E,v_gapextend);
        tmp = vec_subs(H,v_gapopen);
        E   = vec_max(E,tmp);
        
        // update F value
        F   = vec_subs(Fup,v_gapextend);
        tmp = vec_subs(Hup1,v_gapopen);
        F   = vec_max(F,tmp);
        
        v_score_load1 = vec_perm(v_score_load1,v_zero,merge_score_load);
        
        // add score to H
        H   = vec_adds(Hup2,v_score);
        H   = vec_subs(H,v_bias);
        
        // set H to max of H,E,F
        H   = vec_max(H,E);
        H   = vec_max(H,F);
        
        // Update highest score encountered this far
        v_maxscore = vec_max(v_maxscore,H);
        
        
        
        // PROLOGUE 8
        // prefetch next residue
        k                = db_sequence[8];
        
        v_score     = vec_perm(v_score_q1,  v_score_load1,  queue1_to_score);
        v_score_q1  = vec_perm(v_score_q2,  v_score_load1,  queue2_to_queue1);
        v_score_q2  = vec_perm(v_score_q3,  v_score_load1,  queue3_to_queue2);
        v_score_q3  = vec_perm(v_score_q4,  v_score_load1,  queue4_to_queue3);
        v_score_q4  = vec_perm(v_score_q5,  v_score_load1,  queue5_to_queue4);
        v_score_q5  = vec_perm(v_score_q5,  v_score_load1,  queue5_with_load);
        

        // prefetch score for next step 
        v_score_load1 = vec_ld(16*k,query_profile_byte);            
        
        // load values of F and H from previous row (one unit up)
        Fup    = vec_ld(0,  p);
        Hup2   = vec_ld(16, p);
        p += 32; // move ahead 32 bytes
        
        // shift into place so we have complete F and H vectors
        // that refer to the values one unit up from each cell
        // that we are currently working on.
        Fup    = vec_sld(Fup,F,15);
        Hup2   = vec_sld(Hup2,H,15);            
        
        // do the dynamic programming 
        
        // update E value
        E   = vec_subs(E,v_gapextend);
        tmp = vec_subs(H,v_gapopen);
        E   = vec_max(E,tmp);
        
        // update F value
        F   = vec_subs(Fup,v_gapextend);
        tmp = vec_subs(Hup2,v_gapopen);
        F   = vec_max(F,tmp);
        
        v_score_load1 = vec_perm(v_score_load1,v_score_load2,merge_score_load);
        
        // add score to H
        H   = vec_adds(Hup1,v_score);
        H   = vec_subs(H,v_bias);
        
        // set H to max of H,E,F
        H   = vec_max(H,E);
        H   = vec_max(H,F);
        
        // Update highest score encountered this far
        v_maxscore = vec_max(v_maxscore,H);
        
        
        
        
        // PROLOGUE 9
        // prefetch next residue
        k                = db_sequence[9];
        k8               = db_sequence[1];
        
        v_score     = vec_perm(v_score_q1,  v_score_load1,  queue1_to_score);
        v_score_q1  = vec_perm(v_score_q2,  v_score_load1,  queue2_to_queue1);
        v_score_q2  = vec_perm(v_score_q3,  v_score_load1,  queue3_to_queue2);
        v_score_q3  = vec_perm(v_score_q4,  v_score_load1,  queue4_to_queue3);
        v_score_q4  = vec_perm(v_score_q5,  v_score_load1,  queue5_to_queue4);
        v_score_q5  = vec_perm(v_score_q5,  v_score_load1,  queue5_with_load);
        

        // prefetch score for next step 
        v_score_load1 = vec_ld(16*k,query_profile_byte);            
        v_score_load2 = vec_ld(16*k8,query_profile_byte);
        
        // load values of F and H from previous row (one unit up)
        Fup    = vec_ld(0,  p);
        Hup1    = vec_ld(16, p);
        p += 32; // move ahead 32 bytes
        
        // shift into place so we have complete F and H vectors
        // that refer to the values one unit up from each cell
        // that we are currently working on.
        Fup    = vec_sld(Fup,F,15);
        Hup1    = vec_sld(Hup1,H,15);            
        
        // do the dynamic programming 
        
        // update E value
        E   = vec_subs(E,v_gapextend);
        tmp = vec_subs(H,v_gapopen);
        E   = vec_max(E,tmp);
        
        // update F value
        F   = vec_subs(Fup,v_gapextend);
        tmp = vec_subs(Hup1,v_gapopen);
        F   = vec_max(F,tmp);
        
        v_score_load1 = vec_perm(v_score_load1,v_score_load2,merge_score_load);
        
        // add score to H
        H   = vec_adds(Hup2,v_score);
        H   = vec_subs(H,v_bias);
        
        // set H to max of H,E,F
        H   = vec_max(H,E);
        H   = vec_max(H,F);
        
        // Update highest score encountered this far
        v_maxscore = vec_max(v_maxscore,H);
        
        
        
        // PROLOGUE 10
        // prefetch next residue
        k                = db_sequence[10];
        k8               = db_sequence[2];
        
        v_score     = vec_perm(v_score_q1,  v_score_load1,  queue1_to_score);
        v_score_q1  = vec_perm(v_score_q2,  v_score_load1,  queue2_to_queue1);
        v_score_q2  = vec_perm(v_score_q3,  v_score_load1,  queue3_to_queue2);
        v_score_q3  = vec_perm(v_score_q4,  v_score_load1,  queue4_to_queue3);
        v_score_q4  = vec_perm(v_score_q5,  v_score_load1,  queue5_to_queue4);
        v_score_q5  = vec_perm(v_score_q5,  v_score_load1,  queue5_with_load);
        

        // prefetch score for next step 
        v_score_load1 = vec_ld(16*k,query_profile_byte);            
        v_score_load2 = vec_ld(16*k8,query_profile_byte);
        
        // load values of F and H from previous row (one unit up)
        Fup    = vec_ld(0,  p);
        Hup2   = vec_ld(16, p);
        p += 32; // move ahead 32 bytes
        
        // shift into place so we have complete F and H vectors
        // that refer to the values one unit up from each cell
        // that we are currently working on.
        Fup    = vec_sld(Fup,F,15);
        Hup2   = vec_sld(Hup2,H,15);            
        
        // do the dynamic programming 
        
        // update E value
        E   = vec_subs(E,v_gapextend);
        tmp = vec_subs(H,v_gapopen);
        E   = vec_max(E,tmp);
        
        // update F value
        F   = vec_subs(Fup,v_gapextend);
        tmp = vec_subs(Hup2,v_gapopen);
        F   = vec_max(F,tmp);
        
        v_score_load1 = vec_perm(v_score_load1,v_score_load2,merge_score_load);
        
        // add score to H
        H   = vec_adds(Hup1,v_score);
        H   = vec_subs(H,v_bias);
        
        // set H to max of H,E,F
        H   = vec_max(H,E);
        H   = vec_max(H,F);
        
        // Update highest score encountered this far
        v_maxscore = vec_max(v_maxscore,H);
        
        
        
        
        // PROLOGUE 11
        // prefetch next residue
        k                = db_sequence[11];
        k8               = db_sequence[3];
        
        v_score     = vec_perm(v_score_q1,  v_score_load1,  queue1_to_score);
        v_score_q1  = vec_perm(v_score_q2,  v_score_load1,  queue2_to_queue1);
        v_score_q2  = vec_perm(v_score_q3,  v_score_load1,  queue3_to_queue2);
        v_score_q3  = vec_perm(v_score_q4,  v_score_load1,  queue4_to_queue3);
        v_score_q4  = vec_perm(v_score_q5,  v_score_load1,  queue5_to_queue4);
        v_score_q5  = vec_perm(v_score_q5,  v_score_load1,  queue5_with_load);
        

        // prefetch score for next step 
        v_score_load1 = vec_ld(16*k,query_profile_byte);            
        v_score_load2 = vec_ld(16*k8,query_profile_byte);
        
        // load values of F and H from previous row (one unit up)
        Fup    = vec_ld(0,  p);
        Hup1    = vec_ld(16, p);
        p += 32; // move ahead 32 bytes
        
        // shift into place so we have complete F and H vectors
        // that refer to the values one unit up from each cell
        // that we are currently working on.
        Fup    = vec_sld(Fup,F,15);
        Hup1    = vec_sld(Hup1,H,15);            
        
        // do the dynamic programming 
        
        // update E value
        E   = vec_subs(E,v_gapextend);
        tmp = vec_subs(H,v_gapopen);
        E   = vec_max(E,tmp);
        
        // update F value
        F   = vec_subs(Fup,v_gapextend);
        tmp = vec_subs(Hup1,v_gapopen);
        F   = vec_max(F,tmp);
        
        v_score_load1 = vec_perm(v_score_load1,v_score_load2,merge_score_load);
        
        // add score to H
        H   = vec_adds(Hup2,v_score);
        H   = vec_subs(H,v_bias);
        
        // set H to max of H,E,F
        H   = vec_max(H,E);
        H   = vec_max(H,F);
        
        // Update highest score encountered this far
        v_maxscore = vec_max(v_maxscore,H);
        
        
        
        // PROLOGUE 12
        // prefetch next residue
        k                = db_sequence[12];
        k8               = db_sequence[4];
        
        v_score     = vec_perm(v_score_q1,  v_score_load1,  queue1_to_score);
        v_score_q1  = vec_perm(v_score_q2,  v_score_load1,  queue2_to_queue1);
        v_score_q2  = vec_perm(v_score_q3,  v_score_load1,  queue3_to_queue2);
        v_score_q3  = vec_perm(v_score_q4,  v_score_load1,  queue4_to_queue3);
        v_score_q4  = vec_perm(v_score_q5,  v_score_load1,  queue5_to_queue4);
        v_score_q5  = vec_perm(v_score_q5,  v_score_load1,  queue5_with_load);
        

        // prefetch score for next step 
        v_score_load1 = vec_ld(16*k,query_profile_byte);            
        v_score_load2 = vec_ld(16*k8,query_profile_byte);
        
        // load values of F and H from previous row (one unit up)
        Fup    = vec_ld(0,  p);
        Hup2   = vec_ld(16, p);
        p += 32; // move ahead 32 bytes
        
        // shift into place so we have complete F and H vectors
        // that refer to the values one unit up from each cell
        // that we are currently working on.
        Fup    = vec_sld(Fup,F,15);
        Hup2   = vec_sld(Hup2,H,15);            
        
        // do the dynamic programming 
        
        // update E value
        E   = vec_subs(E,v_gapextend);
        tmp = vec_subs(H,v_gapopen);
        E   = vec_max(E,tmp);
        
        // update F value
        F   = vec_subs(Fup,v_gapextend);
        tmp = vec_subs(Hup2,v_gapopen);
        F   = vec_max(F,tmp);
        
        v_score_load1 = vec_perm(v_score_load1,v_score_load2,merge_score_load);
        
        // add score to H
        H   = vec_adds(Hup1,v_score);
        H   = vec_subs(H,v_bias);
        
        // set H to max of H,E,F
        H   = vec_max(H,E);
        H   = vec_max(H,F);
        
        // Update highest score encountered this far
        v_maxscore = vec_max(v_maxscore,H);
        
        
        
        
        // PROLOGUE 13
        // prefetch next residue
        k                = db_sequence[13];
        k8               = db_sequence[5];
        
        v_score     = vec_perm(v_score_q1,  v_score_load1,  queue1_to_score);
        v_score_q1  = vec_perm(v_score_q2,  v_score_load1,  queue2_to_queue1);
        v_score_q2  = vec_perm(v_score_q3,  v_score_load1,  queue3_to_queue2);
        v_score_q3  = vec_perm(