r23964 - /branches/disp_spin_speed/lib/dispersion/ns_cpmg_2site_3d.py -- June 15, 2014 - 15:57

Author: tlinnet
Date: Sun Jun 15 15:57:33 2014
New Revision: 23964

URL: http://svn.gna.org/viewcvs/relax?rev=23964&view=rev
Log:
Implemented the BLAS method of dot product, which should be faster.

I cannot get the "out" argument to work.

Task #7807 (https://gna.org/task/index.php?7807): Speed-up of dispersion 
models for Clustered analysis.

Modified:
    branches/disp_spin_speed/lib/dispersion/ns_cpmg_2site_3d.py

Modified: branches/disp_spin_speed/lib/dispersion/ns_cpmg_2site_3d.py
URL: 
http://svn.gna.org/viewcvs/relax/branches/disp_spin_speed/lib/dispersion/ns_cpmg_2site_3d.py?rev=23964&r1=23963&r2=23964&view=diff
==============================================================================
--- branches/disp_spin_speed/lib/dispersion/ns_cpmg_2site_3d.py (original)
+++ branches/disp_spin_speed/lib/dispersion/ns_cpmg_2site_3d.py Sun Jun 15 
15:57:33 2014
@@ -56,6 +56,7 @@
 # Python module imports.
 from numpy import asarray, dot, fabs, isfinite, log, min, sum
 from numpy.ma import fix_invalid, masked_where
+from scipy.linalg.blas import dgemm as blas_dot
 
 
 # relax module imports.
@@ -158,14 +159,31 @@
                 # This matrix is a propagator that will evolve the 
magnetization with the matrix R for a delay tcp.
                 Rexpo = matrix_exponential(R*tcp_si_mi_di)
 
+                # The numpy way. Give dot two matrices that are both 
C_CONTIGUOUS, then the performance is better:
                 # The essential evolution matrix.
                 # This is the first round.
-                dot(Rexpo, r180x, evolution_matrix)
-                evolution_matrix = dot(evolution_matrix, Rexpo)
+                #dot(Rexpo.T, r180x.T, evolution_matrix)
+                #evolution_matrix = dot(evolution_matrix, Rexpo)
                 # The second round.
-                evolution_matrix = dot(evolution_matrix, evolution_matrix)
+                #evolution_matrix = dot(evolution_matrix, evolution_matrix)
                 # The third and fourth round,
-                evolution_matrix = dot(evolution_matrix, evolution_matrix)
+                #evolution_matrix = dot(evolution_matrix, evolution_matrix)
+
+                # Give dot two matrices that are both F_CONTIGUOUS, we can 
use BLAS through the method: 
+                # The become F_CONTIGUOUS by transposing them.
+                # See by: print Rexpo.flags.c_contiguous, 
Rexpo.T.flags.c_contiguous
+                # http://wiki.scipy.org/PerformanceTips
+                # The FORTRAN code.
+                # 
tchar=s,d,c,z>gemm(m,n,k,alpha,a,b,beta,c,trans_a,trans_b,lda,ka,ldb,kb)
+                #   ! c = 
gemm(alpha,a,b,beta=0,c=0,trans_a=0,trans_b=0,overwrite_c=0)
+                #   ! Calculate C <- alpha * op(A) * op(B) + beta * C
+                # This is the first round.
+                evolution_matrix[:] = blas_dot(alpha=1.0, a=Rexpo.T, 
b=r180x.T, trans_a=True, trans_b=True)
+                evolution_matrix[:] = blas_dot(alpha=1.0, 
a=evolution_matrix.T, b=Rexpo.T, trans_a=True, trans_b=True)
+                # The second round.
+                evolution_matrix[:] = blas_dot(alpha=1.0, 
a=evolution_matrix.T, b=evolution_matrix.T, trans_a=True, trans_b=True)
+                # The third and fourth round.
+                evolution_matrix[:] = blas_dot(alpha=1.0, 
a=evolution_matrix.T, b=evolution_matrix.T, trans_a=True, trans_b=True)
 
                 # Loop over the CPMG elements, propagating the magnetisation.
                 for j in range(power_si_mi_di/2):