def create_packm_functions(myfile, coeffs):
''' Generate all of the custom add functions.
myfile is the file to which we are writing
coeffs is the set of all coefficients
'''
def all_adds(coeffs, name):
for i, coeff_set in enumerate(coeffs):
if len(coeff_set) > 0:
write_packm_func(myfile, coeff_set, i, name)
write_break(myfile)
# S matrices formed from A subblocks
all_adds(transpose(coeffs[0]), 'A')
# T matrices formed from B subblocks
all_adds(transpose(coeffs[1]), 'B')
def create_M_add_functions(myfile, coeffs):
def all_adds(coeffs, name):
for i, coeff_set in enumerate(coeffs):
if len(coeff_set) > 0:
write_M_add_func(myfile, coeff_set, i, name)
write_break(myfile)
# Output C formed from multiplications
#all_adds( coeffs[2], 'M' )
all_adds(transpose(coeffs[2]), 'M')
def create_kernel_header( myfile, coeffs ):
write_break( myfile )
abc_micro_kernel_gen.write_header_start( myfile )
for i, coeff_set in enumerate( transpose( coeffs[2] ) ):
if len( coeff_set ) > 0:
nonzero_coeffs = [coeff for coeff in coeff_set if is_nonzero(coeff)]
nnz = len( nonzero_coeffs )
abc_micro_kernel_gen.generate_kernel_header( myfile, nonzero_coeffs, i )
write_break( myfile )
abc_micro_kernel_gen.write_header_end( myfile )
def create_micro_functions( myfile, coeffs, kernel_header_filename ):
write_line( myfile, 0, '#include "%s"' % kernel_header_filename )
write_break( myfile )
abc_micro_kernel_gen.write_common_rankk_macro_assembly( myfile )
write_break( myfile )
abc_micro_kernel_gen.macro_initialize_assembly( myfile )
write_break( myfile )
for i, coeff_set in enumerate( transpose( coeffs[2] ) ):
if len( coeff_set ) > 0:
nonzero_coeffs = [coeff for coeff in coeff_set if is_nonzero(coeff)]
nnz = len( nonzero_coeffs )
if nnz <= 23:
abc_micro_kernel_gen.generate_micro_kernel( myfile, nonzero_coeffs, i )
write_break( myfile )
def gen_model_coefficient( coeff_filename_mix, level_mix ):
coeffs_mix = []
idx = 0
for coeff_file in coeff_filename_mix:
coeffs = read_coeffs( coeff_file )
level = level_mix[idx]
for level_id in range( level ):
coeffs_mix.append( coeffs )
idx += 1
cur_coeffs = generateCoeffs( coeffs_mix )
#N_A_mul = 0
#N_B_mul = 0
#N_C_mul = 0
#N_A_add = 0
#N_B_add = 0
#N_C_add = 0
abc_counter = [ 0 for i in range(6) ]
ab_counter = [ 0 for i in range(6) ]
naive_counter = [ 0 for i in range(6) ]
#N_mul = 0
#N_A_add = 0
#N_B_add = 0
#N_C_add = 0
comp_counter = [ 0 for i in range(4) ]
for i, coeff_set in enumerate( transpose( cur_coeffs[0] ) ):
nonzero_coeffs = [coeff for coeff in coeff_set if is_nonzero(coeff)]
nnz = len( nonzero_coeffs )
#if ( nnz == 1 ):
abc_counter[0] += nnz
ab_counter[0] += nnz
naive_counter[0] += 1
naive_counter[3] += nnz + 1 # if nnz == 1, naive_counter[3] += 0
comp_counter[1] += nnz - 1
for i, coeff_set in enumerate( transpose( cur_coeffs[1] ) ):
nonzero_coeffs = [coeff for coeff in coeff_set if is_nonzero(coeff)]
nnz = len( nonzero_coeffs )
#if ( nnz == 1 ):
abc_counter[1] += nnz
ab_counter[1] += nnz
naive_counter[1] += 1
naive_counter[4] += nnz + 1 # if nnz == 1, naive_counter[4] += 0
comp_counter[2] += nnz - 1
for i, coeff_set in enumerate( transpose( cur_coeffs[2] ) ):
nonzero_coeffs = [coeff for coeff in coeff_set if is_nonzero(coeff)]
nnz = len( nonzero_coeffs )
#if ( nnz == 1 ):
abc_counter[2] += nnz
ab_counter[2] += 1
ab_counter[5] += 3 * nnz
naive_counter[2] += 1
naive_counter[5] += 3 * nnz
comp_counter[0] += 1
comp_counter[3] += nnz
return [ comp_counter, abc_counter, ab_counter, naive_counter ]
def create_macro_functions( myfile, coeffs ):
for i, coeff_set in enumerate( transpose( coeffs[2] ) ):
if len( coeff_set ) > 0:
write_macro_func( myfile, coeff_set, i, 'C' )
write_break( myfile )