def gen_naive_fmm(coeff_filename_mix, dims_mix, level_mix, outfile):
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
dims_level_mix = []
idx = 0
for dims in dims_mix:
level = level_mix[idx]
for level_id in range(level):
dims_level_mix.append(dims)
idx += 1
with open(outfile, 'w') as myfile:
write_line(myfile, 0, '#include "bl_dgemm.h"')
write_break(myfile)
cur_coeffs = generateCoeffs(coeffs_mix)
num_multiplies = len(cur_coeffs[0][0])
create_add_functions(myfile, cur_coeffs)
create_straprim_naive_functions(myfile, cur_coeffs, dims_level_mix,
num_multiplies)
write_line(
myfile, 0,
'void bl_dgemm_strassen_naive( int m, int n, int k, double *XA, int lda, double *XB, int ldb, double *XC, int ldc )'
)
write_line(myfile, 0, '{')
write_naive_strassen_header(myfile)
writePartition(myfile, dims_level_mix)
write_break(myfile)
create_straprim_caller(myfile, cur_coeffs, dims_level_mix,
num_multiplies)
write_break(myfile)
level_dim = exp_dim_mix(dims_level_mix)
write_line(
myfile, 1,
'bl_dynamic_peeling( m, n, k, XA, lda, XB, ldb, XC, ldc, %d * DGEMM_MR, %d, %d * DGEMM_NR );'
% (level_dim[0], level_dim[1], level_dim[2]))
write_line(myfile, 0, '}')
def gen_abc_fmm( coeff_filename_mix, dims_mix, level_mix, outfilename, micro_kernel_filename, kernel_header_filename ):
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
dims_level_mix = []
idx = 0
for dims in dims_mix:
level = level_mix[idx]
for level_id in range( level ):
dims_level_mix.append( dims )
idx += 1
with open( outfilename, 'w' ) as myfile:
write_line( myfile, 0, '#include "%s"' % kernel_header_filename[10:] )
write_line( myfile, 0, '#include "bl_dgemm.h"' )
write_break( myfile )
cur_coeffs = generateCoeffs( coeffs_mix )
num_multiplies = len(cur_coeffs[0][0])
create_packm_functions( myfile, cur_coeffs )
my_micro_file = open( micro_kernel_filename, 'w' )
create_micro_functions( my_micro_file, cur_coeffs, kernel_header_filename[10:] )
my_kernel_header = open ( kernel_header_filename, 'w' )
create_kernel_header( my_kernel_header, cur_coeffs )
create_macro_functions( myfile, cur_coeffs )
create_straprim_abc_functions( myfile, cur_coeffs, dims_level_mix )
write_line( myfile, 0, 'void bl_dgemm_strassen_abc( int m, int n, int k, double *XA, int lda, double *XB, int ldb, double *XC, int ldc )' )
write_line( myfile, 0, '{' )
write_abc_strassen_header( myfile )
writePartition( myfile, dims_level_mix )
write_break( myfile )
write_line( myfile, 0, '#ifdef _PARALLEL_')
write_line( myfile, 1, '#pragma omp parallel num_threads( bl_ic_nt )' )
write_line( myfile, 0, '#endif')
write_line( myfile, 1, '{' )
create_straprim_caller( myfile, cur_coeffs, dims_level_mix, num_multiplies )
write_line( myfile, 1, '}' )
write_break( myfile )
level_dim = exp_dim_mix( dims_level_mix )
write_line( myfile, 1, 'bl_dynamic_peeling( m, n, k, XA, lda, XB, ldb, XC, ldc, %d * DGEMM_MR, %d, %d * DGEMM_NR );' % ( level_dim[0], level_dim[1], level_dim[2] ) )
write_break( myfile )
write_line( myfile, 1, '//free( packA );' )
write_line( myfile, 1, '//free( packB );' )
write_line( myfile, 0, '}' )