Added analysis of CUDA threads mapping in Light CSR to Python script for SpMV benchmark processing.

   for f in formats:

      if not f in ['CSR Light Best']:

         sort_df.drop( labels=f, axis='columns', level=0, inplace=True )

   sort_df.to_html( f"LightSpMV-Threads-per-row.html" )

   sort_df.to_html( f"LightSpMV-Threads-per-row-best.html" )

   size = len(sort_df[('nonzeros per row', '','','')].index)

   t = np.arange( size )

   fig, axs = plt.subplots( 1, 1 )

   plt.savefig( f"LightSpMV-threads-mapping.pdf" )

   plt.close(fig)

   profiles = {}

   sort_df = df.sort_values(by=[('nonzeros per row','','','')],inplace=False,ascending=True)

   formats_list = ['CSR< Light > 1', 'CSR< Light > 2', 'CSR< Light > 4', 'CSR< Light > 8', 'CSR< Light > 16', 'CSR< Light > 32']

   for format in formats_list:

      sort_df.drop( labels=format, axis='columns', level=0, inplace=True )

      print( f'{format}')

      profiles[format] = df[(format,'GPU','bandwidth','')].copy()

   sort_df.to_html( f"LightSpMV-Threads-per-row.html" )

   draw_profiles( formats_list, profiles, 'non-zeros per row', 'BW', "nonzeros-bw.pdf", 'lower right', "none" )

def write_colormap( file, max_bw, size, x_position, y_position, standalone = False ):

   if standalone:

      file.write( '\\documentclass{standalone}\n' )

   df.to_html( f'output.html' )

   # Generate tables and figures

   effective_bw_profile( df, formats, head_size )

   cusparse_comparison( df, formats, head_size )

   csr_comparison( df, formats, head_size )

   legacy_formats_comparison( df, formats, head_size )

   csr_speedup_comparison( df, formats, head_size )

   cusparse_speedup_comparison( df, formats, head_size )

   binary_matrices_comparison( df, formats, head_size )

   symmetric_matrices_comparison( df, formats, head_size )

   csr_light_speedup_comparison( df, head_size )

   #effective_bw_profile( df, formats, head_size )

   #cusparse_comparison( df, formats, head_size )

   #csr_comparison( df, formats, head_size )

   #legacy_formats_comparison( df, formats, head_size )

   #csr_speedup_comparison( df, formats, head_size )

   #cusparse_speedup_comparison( df, formats, head_size )

   #binary_matrices_comparison( df, formats, head_size )

   #symmetric_matrices_comparison( df, formats, head_size )

   #csr_light_speedup_comparison( df, head_size )

   best = df[('TNL Best','GPU','format')].tolist()

   best_formats = list(set(best))

multicolumns, df_data = get_multiindex( input_df, formats )

print( "Converting data..." )

result = convert_data_frame( input_df, multicolumns, df_data, 0, 20000 )

result = convert_data_frame( input_df, multicolumns, df_data, 0, 2000 )

compute_speedup( result, formats )

result.replace( to_replace=' ',value=np.nan,inplace=True)