Changeset f734e7d in sasmodels


Ignore:
Timestamp:
Feb 22, 2015 1:44:54 AM (10 years ago)
Author:
pkienzle
Branches:
master, core_shell_microgels, costrafo411, magnetic_model, release_v0.94, release_v0.95, ticket-1257-vesicle-product, ticket_1156, ticket_1265_superball, ticket_822_more_unit_tests
Children:
6137124
Parents:
711d8e2
Message:

restructure c code generation for maintainability; extend test harness to allow opencl and ctypes tests

Location:
sasmodels
Files:
2 added
9 edited

Legend:

Unmodified
Added
Removed

  • sasmodels/direct_model.py

    r16bc3fc rf734e7d  
    33import numpy as np 
    44 
    5 from . import models 
    6 from . import weights 
    7  
    8 try: 
    9     from .kernelcl import load_model 
    10 except ImportError,exc: 
    11     warnings.warn(str(exc)) 
    12     warnings.warn("using ctypes instead") 
    13     from .kerneldll import load_model 
    14  
    15 def load_model_definition(model_name): 
    16     __import__('sasmodels.models.'+model_name) 
    17     model_definition = getattr(models, model_name, None) 
    18     return model_definition 
    19  
    20 # load_model is imported above.  It looks like the following 
    21 #def load_model(model_definition, dtype='single): 
    22 #    if kerneldll: 
    23 #        if source is newer than compiled: compile 
    24 #        load dll 
    25 #        return kernel 
    26 #    elif kernelcl: 
    27 #        compile source on context 
    28 #        return kernel 
    29  
    30  
    31 def make_kernel(model, q_vectors): 
    32     """ 
    33     Return a computation kernel from the model definition and the q input. 
    34     """ 
    35     input = model.make_input(q_vectors) 
    36     return model(input) 
    37  
    38 def get_weights(kernel, pars, name): 
    39     """ 
    40     Generate the distribution for parameter *name* given the parameter values 
    41     in *pars*. 
    42  
    43     Searches for "name", "name_pd", "name_pd_type", "name_pd_n", "name_pd_sigma" 
    44     """ 
    45     relative = name in kernel.info['partype']['pd-rel'] 
    46     limits = kernel.info['limits'] 
    47     disperser = pars.get(name+'_pd_type', 'gaussian') 
    48     value = pars.get(name, kernel.info['defaults'][name]) 
    49     npts = pars.get(name+'_pd_n', 0) 
    50     width = pars.get(name+'_pd', 0.0) 
    51     nsigma = pars.get(name+'_pd_nsigma', 3.0) 
    52     v,w = weights.get_weights( 
    53         disperser, npts, width, nsigma, 
    54         value, limits[name], relative) 
    55     return v,w/np.sum(w) 
    56  
    57 def call_kernel(kernel, pars): 
    58     fixed_pars = [pars.get(name, kernel.info['defaults'][name]) 
    59                   for name in kernel.fixed_pars] 
    60     pd_pars = [get_weights(kernel, pars, name) for name in kernel.pd_pars] 
    61     return kernel(fixed_pars, pd_pars) 
     5from .core import load_model_definition, make_kernel, call_kernel 
     6from .core import load_model_cl as load_model 
     7if load_model is None: 
     8    warnings.warn("unable to load opencl; using ctypes instead") 
     9    from .core import load_model_dll as load_model 
    6210 
    6311class DirectModel: 

  • sasmodels/generate.py

    rf1ecfa92 rf734e7d  
    190190 
    191191import sys 
    192 import os 
    193 import os.path 
     192from os.path import abspath, dirname, join as joinpath, exists 
    194193import re 
    195194 
    196195import numpy as np 
     196C_KERNEL_TEMPLATE_PATH = joinpath(dirname(__file__), 'kernel_template.c') 
    197197 
    198198F64 = np.dtype('float64') 
    … …  
    229229PARTABLE_VALUE_WIDTH = 10 
    230230 
    231 # Header included before every kernel. 
    232 # This makes sure that the appropriate math constants are defined, and does 
    233 # whatever is required to make the kernel compile as pure C rather than 
    234 # as an OpenCL kernel. 
    235 KERNEL_HEADER = """\ 
    236 // GENERATED CODE --- DO NOT EDIT --- 
    237 // Code is produced by sasmodels.gen from sasmodels/models/MODEL.c 
    238  
    239 #ifdef __OPENCL_VERSION__ 
    240 # define USE_OPENCL 
    241 #endif 
    242  
    243 // If opencl is not available, then we are compiling a C function 
    244 // Note: if using a C++ compiler, then define kernel as extern "C" 
    245 #ifndef USE_OPENCL 
    246 #  ifdef __cplusplus 
    247      #include <cmath> 
    248      #if defined(_MSC_VER) 
    249      #define kernel extern "C" __declspec( dllexport ) 
    250      #else 
    251      #define kernel extern "C" 
    252      #endif 
    253      using namespace std; 
    254      inline void SINCOS(double angle, double &svar, double &cvar) 
    255        { svar=sin(angle); cvar=cos(angle); } 
    256 #  else 
    257      #include <math.h> 
    258      #if defined(_MSC_VER) 
    259      #define kernel __declspec( dllexport ) 
    260      #else 
    261      #define kernel 
    262      #endif 
    263      #define SINCOS(angle,svar,cvar) do {svar=sin(angle);cvar=cos(angle);} while (0) 
    264 #  endif 
    265 #  define global 
    266 #  define local 
    267 #  define constant const 
    268 #  define powr(a,b) pow(a,b) 
    269 #  define pown(a,b) pow(a,b) 
    270 #else 
    271 #  ifdef USE_SINCOS 
    272 #    define SINCOS(angle,svar,cvar) svar=sincos(angle,&cvar) 
    273 #  else 
    274 #    define SINCOS(angle,svar,cvar) do {svar=sin(angle);cvar=cos(angle);} while (0) 
    275 #  endif 
    276 #endif 
    277  
    278 // Standard mathematical constants: 
    279 //   M_E, M_LOG2E, M_LOG10E, M_LN2, M_LN10, M_PI, M_PI_2=pi/2, M_PI_4=pi/4, 
    280 //   M_1_PI=1/pi, M_2_PI=2/pi, M_2_SQRTPI=2/sqrt(pi), SQRT2, SQRT1_2=sqrt(1/2) 
    281 // OpenCL defines M_constant_F for float constants, and nothing if double 
    282 // is not enabled on the card, which is why these constants may be missing 
    283 #ifndef M_PI 
    284 #  define M_PI 3.141592653589793 
    285 #endif 
    286 #ifndef M_PI_2 
    287 #  define M_PI_2 1.570796326794897 
    288 #endif 
    289 #ifndef M_PI_4 
    290 #  define M_PI_4 0.7853981633974483 
    291 #endif 
    292  
    293 // Non-standard pi/180, used for converting between degrees and radians 
    294 #ifndef M_PI_180 
    295 #  define M_PI_180 0.017453292519943295 
    296 #endif 
    297 """ 
    298  
    299  
    300 # The I(q) kernel and the I(qx, qy) kernel have one and two q parameters 
    301 # respectively, so the template builder will need to do extra work to 
    302 # declare, initialize and pass the q parameters. 
    303 KERNEL_1D = { 
    304     'fn': "Iq", 
    305     'q_par_decl': "global const double *q,", 
    306     'qinit': "const double qi = q[i];", 
    307     'qcall': "qi", 
    308     'qwork': ["q"], 
    309     } 
    310  
    311 KERNEL_2D = { 
    312     'fn': "Iqxy", 
    313     'q_par_decl': "global const double *qx,\n    global const double *qy,", 
    314     'qinit': "const double qxi = qx[i];\n    const double qyi = qy[i];", 
    315     'qcall': "qxi, qyi", 
    316     'qwork': ["qx", "qy"], 
    317     } 
    318  
    319 # Generic kernel template for the polydispersity loop. 
    320 # This defines the opencl kernel that is available to the host.  The same 
    321 # structure is used for Iq and Iqxy kernels, so extra flexibility is needed 
    322 # for q parameters.  The polydispersity loop is built elsewhere and 
    323 # substituted into this template. 
    324 KERNEL_TEMPLATE = """\ 
    325 kernel void %(name)s( 
    326     %(q_par_decl)s 
    327     global double *result, 
    328 #ifdef USE_OPENCL 
    329     global double *loops_g, 
    330 #else 
    331     const int Nq, 
    332 #endif 
    333     local double *loops, 
    334     const double cutoff, 
    335     %(par_decl)s 
    336     ) 
    337 { 
    338 #ifdef USE_OPENCL 
    339   // copy loops info to local memory 
    340   event_t e = async_work_group_copy(loops, loops_g, (%(pd_length)s)*2, 0); 
    341   wait_group_events(1, &e); 
    342  
    343   int i = get_global_id(0); 
    344   int Nq = get_global_size(0); 
    345 #endif 
    346  
    347 #ifdef USE_OPENCL 
    348   if (i < Nq) 
    349 #else 
    350   #pragma omp parallel for 
    351   for (int i=0; i < Nq; i++) 
    352 #endif 
    353   { 
    354     %(qinit)s 
    355     double ret=0.0, norm=0.0; 
    356     double vol=0.0, norm_vol=0.0; 
    357 %(loops)s 
    358     if (vol*norm_vol != 0.0) { 
    359       ret *= norm_vol/vol; 
    360     } 
    361     result[i] = scale*ret/norm+background; 
    362   } 
    363 } 
    364 """ 
    365  
    366 # Polydispersity loop level. 
    367 # This pulls the parameter value and weight from the looping vector in order 
    368 # in preperation for a nested loop. 
    369 LOOP_OPEN="""\ 
    370 for (int %(name)s_i=0; %(name)s_i < N%(name)s; %(name)s_i++) { 
    371   const double %(name)s = loops[2*(%(name)s_i%(offset)s)]; 
    372   const double %(name)s_w = loops[2*(%(name)s_i%(offset)s)+1];\ 
    373 """ 
    374  
    375  
    376  
    377 ########################################################## 
    378 #                                                        # 
    379 #   !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!   # 
    380 #   !!                                              !!   # 
    381 #   !!  KEEP THIS CODE CONSISTENT WITH PYKERNEL.PY  !!   # 
    382 #   !!                                              !!   # 
    383 #   !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!   # 
    384 #                                                        # 
    385 ########################################################## 
    386  
    387  
    388 # Polydispersity loop body. 
    389 # This computes the weight, and if it is sufficient, calls the scattering 
    390 # function and adds it to the total.  If there is a volume normalization, 
    391 # it will also be added here. 
    392 LOOP_BODY="""\ 
    393 const double weight = %(weight_product)s; 
    394 if (weight > cutoff) { 
    395   const double I = %(fn)s(%(qcall)s, %(pcall)s); 
    396   if (I>=0.0) { // scattering cannot be negative 
    397     ret += weight*I%(sasview_spherical)s; 
    398     norm += weight; 
    399     %(volume_norm)s 
    400   } 
    401   //else { printf("exclude qx,qy,I:%%g,%%g,%%g\\n",%(qcall)s,I); } 
    402 } 
    403 //else { printf("exclude weight:%%g\\n",weight); }\ 
    404 """ 
    405  
    406 # Use this when integrating over orientation 
    407 SPHERICAL_CORRECTION="""\ 
    408 // Correction factor for spherical integration p(theta) I(q) sin(theta) dtheta 
    409 double spherical_correction = (Ntheta>1 ? fabs(sin(M_PI_180*theta)) : 1.0);\ 
    410 """ 
    411 # Use this to reproduce sasview behaviour 
    412 SASVIEW_SPHERICAL_CORRECTION="""\ 
    413 // Correction factor for spherical integration p(theta) I(q) sin(theta) dtheta 
    414 double spherical_correction = (Ntheta>1 ? fabs(cos(M_PI_180*theta))*M_PI_2 : 1.0);\ 
    415 """ 
    416  
    417 # Volume normalization. 
    418 # If there are "volume" polydispersity parameters, then these will be used 
    419 # to call the form_volume function from the user supplied kernel, and accumulate 
    420 # a normalized weight. 
    421 VOLUME_NORM="""const double vol_weight = %(vol_weight)s; 
    422     vol += vol_weight*form_volume(%(vol_pars)s); 
    423     norm_vol += vol_weight;\ 
    424 """ 
    425  
    426 # functions defined as strings in the .py module 
    427 WORK_FUNCTION="""\ 
    428 double %(name)s(%(pars)s); 
    429 double %(name)s(%(pars)s) 
    430 { 
    431 %(body)s 
    432 }\ 
    433 """ 
    434  
    435231# Documentation header for the module, giving the model name, its short 
    436232# description and its parameter table.  The remainder of the doc comes 
    … …  
    449245%(docs)s 
    450246""" 
    451  
    452 def indent(s, depth): 
    453     """ 
    454     Indent a string of text with *depth* additional spaces on each line. 
    455     """ 
    456     spaces = " "*depth 
    457     sep = "\n"+spaces 
    458     return spaces + sep.join(s.split("\n")) 
    459  
    460  
    461 def kernel_name(info, is_2D): 
    462     """ 
    463     Name of the exported kernel symbol. 
    464     """ 
    465     return info['name'] + "_" + ("Iqxy" if is_2D else "Iq") 
    466  
    467  
    468 def use_single(source): 
    469     """ 
    470     Convert code from double precision to single precision. 
    471     """ 
    472     # Convert double keyword to float.  Accept an 'n' parameter for vector 
    473     # values, where n is 2, 4, 8 or 16. Assume complex numbers are represented 
    474     # as cdouble which is typedef'd to double2. 
    475     source = re.sub(r'(^|[^a-zA-Z0-9_]c?)double(([248]|16)?($|[^a-zA-Z0-9_]))', 
    476                     r'\1float\2', source) 
    477     # Convert floating point constants to single by adding 'f' to the end. 
    478     # OS/X driver complains if you don't do this. 
    479     source = re.sub(r'[^a-zA-Z_](\d*[.]\d+|\d+[.]\d*)([eE][+-]?\d+)?', 
    480                     r'\g<0>f', source) 
    481     return source 
    482  
    483  
    484 def make_kernel(info, is_2D): 
    485     """ 
    486     Build a kernel call from metadata supplied by the user. 
    487  
    488     *info* is the json object defined in the kernel file. 
    489  
    490     *form* is either "Iq" or "Iqxy". 
    491  
    492     This does not create a complete OpenCL kernel source, only the top 
    493     level kernel call with polydispersity and a call to the appropriate 
    494     Iq or Iqxy function. 
    495     """ 
    496  
    497     # If we are building the Iqxy kernel, we need to propagate qx,qy 
    498     # parameters, otherwise we can 
    499     dim = "2d" if is_2D else "1d" 
    500     fixed_pars = info['partype']['fixed-'+dim] 
    501     pd_pars = info['partype']['pd-'+dim] 
    502     vol_pars = info['partype']['volume'] 
    503     q_pars = KERNEL_2D if is_2D else KERNEL_1D 
    504     fn = q_pars['fn'] 
    505  
    506  
    507     # Build polydispersity loops 
    508     depth = 4 
    509     offset = "" 
    510     loop_head = [] 
    511     loop_end = [] 
    512     for name in pd_pars: 
    513         subst = { 'name': name, 'offset': offset } 
    514         loop_head.append(indent(LOOP_OPEN%subst, depth)) 
    515         loop_end.insert(0, (" "*depth) + "}") 
    516         offset += '+N'+name 
    517         depth += 2 
    518  
    519     # The volume parameters in the inner loop are used to call the volume() 
    520     # function in the kernel, with the parameters defined in vol_pars and the 
    521     # weight product defined in weight.  If there are no volume parameters, 
    522     # then there will be no volume normalization. 
    523     if vol_pars: 
    524         subst = { 
    525             'vol_weight': "*".join(p+"_w" for p in vol_pars), 
    526             'vol_pars': ", ".join(vol_pars), 
    527             } 
    528         volume_norm = VOLUME_NORM%subst 
    529     else: 
    530         volume_norm = "" 
    531  
    532     # Define the inner loop function call 
    533     # The parameters to the f(q,p1,p2...) call should occur in the same 
    534     # order as given in the parameter info structure.  This may be different 
    535     # from the parameter order in the call to the kernel since the kernel 
    536     # call places all fixed parameters before all polydisperse parameters. 
    537     fq_pars = [p[0] for p in info['parameters'][len(COMMON_PARAMETERS):] 
    538                if p[0] in set(fixed_pars+pd_pars)] 
    539     if False and "theta" in pd_pars: 
    540         spherical_correction = [indent(SPHERICAL_CORRECTION, depth)] 
    541         weights = [p+"_w" for p in pd_pars]+['spherical_correction'] 
    542         sasview_spherical = "" 
    543     elif True and "theta" in pd_pars: 
    544         spherical_correction = [indent(SASVIEW_SPHERICAL_CORRECTION,depth)] 
    545         weights = [p+"_w" for p in pd_pars] 
    546         sasview_spherical = "*spherical_correction" 
    547     else: 
    548         spherical_correction = [] 
    549         weights = [p+"_w" for p in pd_pars] 
    550         sasview_spherical = "" 
    551     weight_product = "*".join(weights) if len(weights) > 1 else "1.0" 
    552     subst = { 
    553         'weight_product': weight_product, 
    554         'volume_norm': volume_norm, 
    555         'fn': fn, 
    556         'qcall': q_pars['qcall'], 
    557         'pcall': ", ".join(fq_pars), # skip scale and background 
    558         'sasview_spherical': sasview_spherical, 
    559         } 
    560     loop_body = [indent(LOOP_BODY%subst, depth)] 
    561     loops = "\n".join(loop_head+spherical_correction+loop_body+loop_end) 
    562  
    563     # declarations for non-pd followed by pd pars 
    564     # e.g., 
    565     #     const double sld, 
    566     #     const int Nradius 
    567     fixed_par_decl = ",\n    ".join("const double %s"%p for p in fixed_pars) 
    568     pd_par_decl = ",\n    ".join("const int N%s"%p for p in pd_pars) 
    569     if fixed_par_decl and pd_par_decl: 
    570         par_decl = ",\n    ".join((fixed_par_decl, pd_par_decl)) 
    571     elif fixed_par_decl: 
    572         par_decl = fixed_par_decl 
    573     else: 
    574         par_decl = pd_par_decl 
    575  
    576     # Finally, put the pieces together in the kernel. 
    577     pd_length = "+".join('N'+p for p in pd_pars) if len(pd_pars) > 0 else "0" 
    578     subst = { 
    579         # kernel name is, e.g., cylinder_Iq 
    580         'name': kernel_name(info, is_2D), 
    581         # to declare, e.g., global double q[], 
    582         'q_par_decl': q_pars['q_par_decl'], 
    583         # to declare, e.g., double sld, int Nradius, int Nlength 
    584         'par_decl': par_decl, 
    585         # to copy global to local pd pars we need, e.g., Nradius+Nlength 
    586         'pd_length': pd_length, 
    587         # the q initializers, e.g., double qi = q[i]; 
    588         'qinit': q_pars['qinit'], 
    589         # the actual polydispersity loop 
    590         'loops': loops, 
    591         } 
    592     kernel = KERNEL_TEMPLATE%subst 
    593  
    594     # If the working function is defined in the kernel metadata as a 
    595     # string, translate the string to an actual function definition 
    596     # and put it before the kernel. 
    597     if info[fn]: 
    598         subst = { 
    599             'name': fn, 
    600             'pars': ", ".join("double "+p for p in q_pars['qwork']+fq_pars), 
    601             'body': info[fn], 
    602             } 
    603         kernel = "\n".join((WORK_FUNCTION%subst, kernel)) 
    604     return kernel 
    605247 
    606248def make_partable(pars): 
    … …  
    641283    """ 
    642284    for path in search_path: 
    643         target = os.path.join(path, filename) 
    644         if os.path.exists(target): 
     285        target = joinpath(path, filename) 
     286        if exists(target): 
    645287            return target 
    646288    raise ValueError("%r not found in %s"%(filename, search_path)) 
    … …  
    650292    Return a list of the sources file paths for the module. 
    651293    """ 
    652     from os.path import abspath, dirname, join as joinpath 
    653294    search_path = [ dirname(info['filename']), 
    654295                    abspath(joinpath(dirname(__file__),'models')) ] 
    655296    return [_search(search_path, f) for f in info['source']] 
    656297 
    657 def make_model(info): 
    658     """ 
    659     Generate the code for the kernel defined by info, using source files 
    660     found in the given search path. 
    661     """ 
    662     source = [open(f).read() for f in sources(info)] 
    663     # If the form volume is defined as a string, then wrap it in a 
    664     # function definition and place it after the external sources but 
    665     # before the kernel functions.  If the kernel functions are strings, 
    666     # they will be translated in the make_kernel call. 
    667     if info['form_volume']: 
    668         subst = { 
    669             'name': "form_volume", 
    670             'pars': ", ".join("double "+p for p in info['partype']['volume']), 
    671             'body': info['form_volume'], 
    672             } 
    673         source.append(WORK_FUNCTION%subst) 
    674     kernel_Iq = make_kernel(info, is_2D=False) if not callable(info['Iq']) else "" 
    675     kernel_Iqxy = make_kernel(info, is_2D=True) if not callable(info['Iqxy']) else "" 
    676     kernel = "\n\n".join([KERNEL_HEADER]+source+[kernel_Iq, kernel_Iqxy]) 
    677     return kernel 
     298def use_single(source): 
     299    """ 
     300    Convert code from double precision to single precision. 
     301    """ 
     302    # Convert double keyword to float.  Accept an 'n' parameter for vector 
     303    # values, where n is 2, 4, 8 or 16. Assume complex numbers are represented 
     304    # as cdouble which is typedef'd to double2. 
     305    source = re.sub(r'(^|[^a-zA-Z0-9_]c?)double(([248]|16)?($|[^a-zA-Z0-9_]))', 
     306                    r'\1float\2', source) 
     307    # Convert floating point constants to single by adding 'f' to the end. 
     308    # OS/X driver complains if you don't do this. 
     309    source = re.sub(r'[^a-zA-Z_](\d*[.]\d+|\d+[.]\d*)([eE][+-]?\d+)?', 
     310                    r'\g<0>f', source) 
     311    return source 
     312 
     313 
     314def kernel_name(info, is_2D): 
     315    """ 
     316    Name of the exported kernel symbol. 
     317    """ 
     318    return info['name'] + "_" + ("Iqxy" if is_2D else "Iq") 
     319 
    678320 
    679321def categorize_parameters(pars): 
    … …  
    708350    return partype 
    709351 
     352def indent(s, depth): 
     353    """ 
     354    Indent a string of text with *depth* additional spaces on each line. 
     355    """ 
     356    spaces = " "*depth 
     357    sep = "\n"+spaces 
     358    return spaces + sep.join(s.split("\n")) 
     359 
     360 
     361def build_polydispersity_loops(pd_pars): 
     362    """ 
     363    Build polydispersity loops 
     364 
     365    Returns loop opening and loop closing 
     366    """ 
     367    LOOP_OPEN="""\ 
     368for (int %(name)s_i=0; %(name)s_i < N%(name)s; %(name)s_i++) { 
     369  const double %(name)s = loops[2*(%(name)s_i%(offset)s)]; 
     370  const double %(name)s_w = loops[2*(%(name)s_i%(offset)s)+1];\ 
     371""" 
     372    depth = 4 
     373    offset = "" 
     374    loop_head = [] 
     375    loop_end = [] 
     376    for name in pd_pars: 
     377        subst = { 'name': name, 'offset': offset } 
     378        loop_head.append(indent(LOOP_OPEN%subst, depth)) 
     379        loop_end.insert(0, (" "*depth) + "}") 
     380        offset += '+N'+name 
     381        depth += 2 
     382    return "\n".join(loop_head), "\n".join(loop_end) 
     383 
     384C_KERNEL_TEMPLATE=None 
     385def make_model(info): 
     386    """ 
     387    Generate the code for the kernel defined by info, using source files 
     388    found in the given search path. 
     389    """ 
     390    # TODO: need something other than volume to indicate dispersion parameters 
     391    # No volume normalization despite having a volume parameter. 
     392    # Thickness is labelled a volume in order to trigger polydispersity. 
     393    # May want a separate dispersion flag, or perhaps a separate category for 
     394    # disperse, but not volume.  Volume parameters also use relative values 
     395    # for the distribution rather than the absolute values used by angular 
     396    # dispersion.  Need to be careful that necessary parameters are available 
     397    # for computing volume even if we allow non-disperse volume parameters. 
     398 
     399    # Load template 
     400    global C_KERNEL_TEMPLATE 
     401    if C_KERNEL_TEMPLATE is None: 
     402        with open(C_KERNEL_TEMPLATE_PATH) as fid: 
     403            C_KERNEL_TEMPLATE = fid.read() 
     404 
     405    # Load additional sources 
     406    source = [open(f).read() for f in sources(info)] 
     407 
     408    # Prepare defines 
     409    defines = [] 
     410    partype = info['partype'] 
     411    pd_1d = partype['pd-1d'] 
     412    pd_2d = partype['pd-2d'] 
     413    fixed_1d = partype['fixed-1d'] 
     414    fixed_2d = partype['fixed-1d'] 
     415 
     416    iq_parameters = [p[0] 
     417        for p in info['parameters'][2:] # skip scale, background 
     418        if p[0] in set(fixed_1d+pd_1d)] 
     419    iqxy_parameters = [p[0] 
     420        for p in info['parameters'][2:] # skip scale, background 
     421        if p[0] in set(fixed_2d+pd_2d)] 
     422    volume_parameters = [p[0] 
     423        for p in info['parameters'] 
     424        if p[4]=='volume'] 
     425 
     426    # Fill in defintions for volume parameters 
     427    if volume_parameters: 
     428        defines.append(('VOLUME_PARAMETERS', 
     429                        ','.join(volume_parameters))) 
     430        defines.append(('VOLUME_WEIGHT_PRODUCT', 
     431                        '*'.join(p+'_w' for p in volume_parameters))) 
     432 
     433    # Generate form_volume function from body only 
     434    if info['form_volume'] is not None: 
     435        defines.append(('VOLUME_PARAMETER_DECLARATIONS', 
     436                        ', '.join('double '+p for p in volume_parameters))) 
     437        fn = """\ 
     438double form_volume(VOLUME_PARAMETER_DECLARATIONS); 
     439double form_volume(VOLUME_PARAMETER_DECLARATIONS) { 
     440    %(body)s 
     441} 
     442"""%{'body':info['form_volume']} 
     443        source.append(fn) 
     444 
     445    # Fill in definitions for Iq parameters 
     446    defines.append(('IQ_KERNEL_NAME', info['name']+'_Iq')) 
     447    defines.append(('IQ_PARAMETERS', ', '.join(iq_parameters))) 
     448    if fixed_1d: 
     449        defines.append(('IQ_FIXED_PARAMETER_DECLARATIONS', 
     450                        ', \\\n    '.join('const double %s'%p for p in fixed_1d))) 
     451    if pd_1d: 
     452        defines.append(('IQ_WEIGHT_PRODUCT', 
     453                        '*'.join(p+'_w' for p in pd_1d))) 
     454        defines.append(('IQ_DISPERSION_LENGTH_DECLARATIONS', 
     455                        ', \\\n    '.join('const int N%s'%p for p in pd_1d))) 
     456        defines.append(('IQ_DISPERSION_LENGTH_SUM', 
     457                        '+'.join('N'+p for p in pd_1d))) 
     458        open_loops, close_loops = build_polydispersity_loops(pd_1d) 
     459        defines.append(('IQ_OPEN_LOOPS', 
     460                        open_loops.replace('\n',' \\\n'))) 
     461        defines.append(('IQ_CLOSE_LOOPS', 
     462                        close_loops.replace('\n',' \\\n'))) 
     463    if info['Iq'] is not None: 
     464        defines.append(('IQ_PARAMETER_DECLARATIONS', 
     465                       ', '.join('double '+p for p in iq_parameters))) 
     466        fn = """\ 
     467double Iq(double q, IQ_PARAMETER_DECLARATIONS); 
     468double Iq(double q, IQ_PARAMETER_DECLARATIONS) { 
     469    %(body)s 
     470} 
     471"""%{'body':info['Iq']} 
     472        source.append(fn) 
     473 
     474    # Fill in definitions for Iqxy parameters 
     475    defines.append(('IQXY_KERNEL_NAME', info['name']+'_Iqxy')) 
     476    defines.append(('IQXY_PARAMETERS', ', '.join(iqxy_parameters))) 
     477    if fixed_2d: 
     478        defines.append(('IQXY_FIXED_PARAMETER_DECLARATIONS', 
     479                        ', \\\n    '.join('const double %s'%p for p in fixed_2d))) 
     480    if pd_2d: 
     481        defines.append(('IQXY_WEIGHT_PRODUCT', 
     482                        '*'.join(p+'_w' for p in pd_2d))) 
     483        defines.append(('IQXY_DISPERSION_LENGTH_DECLARATIONS', 
     484                        ', \\\n    '.join('const int N%s'%p for p in pd_2d))) 
     485        defines.append(('IQXY_DISPERSION_LENGTH_SUM', 
     486                        '+'.join('N'+p for p in pd_2d))) 
     487        open_loops, close_loops = build_polydispersity_loops(pd_2d) 
     488        defines.append(('IQXY_OPEN_LOOPS', 
     489                        open_loops.replace('\n',' \\\n'))) 
     490        defines.append(('IQXY_CLOSE_LOOPS', 
     491                        close_loops.replace('\n',' \\\n'))) 
     492    if info['Iqxy'] is not None: 
     493        defines.append(('IQXY_PARAMETER_DECLARATIONS', 
     494                       ', '.join('double '+p for p in iqxy_parameters))) 
     495        fn = """\ 
     496double Iqxy(double qx, double qy, IQXY_PARAMETER_DECLARATIONS); 
     497double Iqxy(double qx, double qy, IQXY_PARAMETER_DECLARATIONS) { 
     498    %(body)s 
     499} 
     500"""%{'body':info['Iqxy']} 
     501        source.append(fn) 
     502 
     503    # Need to know if we have a theta parameter for Iqxy; it is not there 
     504    # for the magnetic sphere model, for example, which has a magnetic 
     505    # orientation but no shape orientation. 
     506    if 'theta' in pd_2d: 
     507        defines.append(('IQXY_HAS_THETA', '1')) 
     508 
     509    #for d in defines: print d 
     510    DEFINES='\n'.join('#define %s %s'%(k,v) for k,v in defines) 
     511    SOURCES='\n\n'.join(source) 
     512    return C_KERNEL_TEMPLATE%{ 
     513        'DEFINES':DEFINES, 
     514        'SOURCES':SOURCES, 
     515        } 
     516 
    710517def make(kernel_module): 
    711518    """ 
    … …  
    717524    """ 
    718525    # TODO: allow Iq and Iqxy to be defined in python 
    719     from os.path import abspath 
    720526    #print kernelfile 
    721527    info = dict( 
    … …  
    737543    info['defaults'] = dict((p[0],p[2]) for p in info['parameters']) 
    738544 
    739     source = make_model(info) 
    740  
     545    # Assume if one part of the kernel is python then all parts are. 
     546    source = make_model(info) if not callable(info['Iq']) else None 
    741547    return source, info 
    742548 

  • sasmodels/kernelcl.py

    rf1ecfa92 rf734e7d  
    4444 
    4545from . import generate 
    46 from .kernelpy import PyInput, PyKernel 
     46from .kernelpy import PyInput, PyModel 
    4747 
    4848F64_DEFS = """\ 
    … …  
    6868    """ 
    6969    source, info = generate.make(kernel_module) 
     70    if callable(info.get('Iq',None)): 
     71        return PyModel(info) 
    7072    ## for debugging, save source to a .cl file, edit it, and reload as model 
    7173    #open(info['name']+'.cl','w').write(source) 
    … …  
    234236 
    235237    def __call__(self, input): 
    236         # Support pure python kernel call 
    237         if input.is_2D and callable(self.info['Iqxy']): 
    238             return PyKernel(self.info['Iqxy'], self.info, input) 
    239         elif not input.is_2D and callable(self.info['Iq']): 
    240             return PyKernel(self.info['Iq'], self.info, input) 
    241  
    242238        if self.dtype != input.dtype: 
    243239            raise TypeError("data and kernel have different types") 
    … …  
    261257        ctypes and some may be pure python. 
    262258        """ 
    263         # Support pure python kernel call 
    264         if len(q_vectors) == 1 and callable(self.info['Iq']): 
    265             return PyInput(q_vectors, dtype=self.dtype) 
    266         elif callable(self.info['Iqxy']): 
    267             return PyInput(q_vectors, dtype=self.dtype) 
    268         else: 
    269             return GpuInput(q_vectors, dtype=self.dtype) 
     259        return GpuInput(q_vectors, dtype=self.dtype) 
    270260 
    271261# TODO: check that we don't need a destructor for buffers which go out of scope 
    … …  
    349339 
    350340 
    351     def __call__(self, pars, pd_pars, cutoff=1e-5): 
     341    def __call__(self, fixed_pars, pd_pars, cutoff=1e-5): 
    352342        real = np.float32 if self.input.dtype == generate.F32 else np.float64 
    353         fixed = [real(p) for p in pars] 
    354         cutoff = real(cutoff) 
    355         loops = np.hstack(pd_pars) if pd_pars else np.empty(0,dtype=self.input.dtype) 
    356         loops = np.ascontiguousarray(loops.T, self.input.dtype).flatten() 
    357         Nloops = [np.uint32(len(p[0])) for p in pd_pars] 
    358         #print "loops",Nloops, loops 
    359  
    360         #import sys; print >>sys.stderr,"opencl eval",pars 
    361         #print "opencl eval",pars 
    362         if len(loops) > 2*MAX_LOOPS: 
    363             raise ValueError("too many polydispersity points") 
     343 
    364344        device_num = 0 
     345        queuei = environment().queues[device_num] 
    365346        res_bi = self.res_b[device_num] 
    366         queuei = environment().queues[device_num] 
    367         loops_bi = self.loops_b[device_num] 
    368         loops_l = cl.LocalMemory(len(loops.data)) 
    369         cl.enqueue_copy(queuei, loops_bi, loops) 
    370         #ctx = environment().context 
    371         #loops_bi = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=loops) 
    372         args = self.input.q_buffers + [res_bi,loops_bi,loops_l,cutoff] + fixed + Nloops 
     347        nq = np.uint32(self.input.nq) 
     348        if pd_pars: 
     349            cutoff = real(cutoff) 
     350            loops_N = [np.uint32(len(p[0])) for p in pd_pars] 
     351            loops = np.hstack(pd_pars) if pd_pars else np.empty(0,dtype=self.input.dtype) 
     352            loops = np.ascontiguousarray(loops.T, self.input.dtype).flatten() 
     353            #print "loops",Nloops, loops 
     354 
     355            #import sys; print >>sys.stderr,"opencl eval",pars 
     356            #print "opencl eval",pars 
     357            if len(loops) > 2*MAX_LOOPS: 
     358                raise ValueError("too many polydispersity points") 
     359 
     360            loops_bi = self.loops_b[device_num] 
     361            cl.enqueue_copy(queuei, loops_bi, loops) 
     362            loops_l = cl.LocalMemory(len(loops.data)) 
     363            #ctx = environment().context 
     364            #loops_bi = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=loops) 
     365            dispersed = [loops_bi, loops_l, cutoff] + loops_N 
     366        else: 
     367            dispersed = [] 
     368        fixed = [real(p) for p in fixed_pars] 
     369        args = self.input.q_buffers + [res_bi, nq] + dispersed + fixed 
    373370        self.kernel(queuei, self.input.global_size, None, *args) 
    374371        cl.enqueue_copy(queuei, self.res, res_bi) 

  • sasmodels/kerneldll.py

    r68d3c1b rf734e7d  
    1111 
    1212from . import generate 
    13 from .kernelpy import PyInput, PyKernel 
     13from .kernelpy import PyInput, PyModel 
    1414 
    1515from .generate import F32, F64 
    … …  
    2222    if "VCINSTALLDIR" in os.environ: 
    2323        # MSVC compiler is available, so use it. 
     24        # TODO: remove intermediate OBJ file created in the directory 
     25        # TODO: maybe don't use randomized name for the c file 
    2426        COMPILE = "cl /nologo /Ox /MD /W3 /GS- /DNDEBUG /Tp%(source)s /openmp /link /DLL /INCREMENTAL:NO /MANIFEST /OUT:%(output)s" 
    2527        # Can't find VCOMP90.DLL (don't know why), so remove openmp support from windows compiler build 
    … …  
    5456    be defined without using too many resources. 
    5557    """ 
    56     import tempfile 
    57  
    5858    source, info = generate.make(kernel_module) 
     59    if callable(info.get('Iq',None)): 
     60        return PyModel(info) 
    5961    source_files = generate.sources(info) + [info['filename']] 
    6062    newest = max(os.path.getmtime(f) for f in source_files) 
    … …  
    6870        status = os.system(command) 
    6971        if status != 0: 
    70             print "compile failed.  File is in %r"%filename 
     72            raise RuntimeError("compile failed.  File is in %r"%filename) 
    7173        else: 
    7274            ## uncomment the following to keep the generated c file 
    … …  
    7678 
    7779 
    78 IQ_ARGS = [c_void_p, c_void_p, c_int, c_void_p, c_double] 
    79 IQXY_ARGS = [c_void_p, c_void_p, c_void_p, c_int, c_void_p, c_double] 
     80IQ_ARGS = [c_void_p, c_void_p, c_int] 
     81IQXY_ARGS = [c_void_p, c_void_p, c_void_p, c_int] 
    8082 
    8183class DllModel(object): 
    … …  
    107109        self.dll = ct.CDLL(self.dllpath) 
    108110 
     111        pd_args_1d = [c_void_p, c_double] + [c_int]*Npd1d if Npd1d else [] 
     112        pd_args_2d= [c_void_p, c_double] + [c_int]*Npd2d if Npd2d else [] 
    109113        self.Iq = self.dll[generate.kernel_name(self.info, False)] 
    110         self.Iq.argtypes = IQ_ARGS + [c_double]*Nfixed1d + [c_int]*Npd1d 
     114        self.Iq.argtypes = IQ_ARGS + pd_args_1d + [c_double]*Nfixed1d 
    111115 
    112116        self.Iqxy = self.dll[generate.kernel_name(self.info, True)] 
    113         self.Iqxy.argtypes = IQXY_ARGS + [c_double]*Nfixed2d + [c_int]*Npd2d 
     117        self.Iqxy.argtypes = IQXY_ARGS + pd_args_2d + [c_double]*Nfixed2d 
    114118 
    115119    def __getstate__(self): 
    … …  
    120124 
    121125    def __call__(self, input): 
    122         # Support pure python kernel call 
    123         if input.is_2D and callable(self.info['Iqxy']): 
    124             return PyKernel(self.info['Iqxy'], self.info, input) 
    125         elif not input.is_2D and callable(self.info['Iq']): 
    126             return PyKernel(self.info['Iq'], self.info, input) 
    127  
    128126        if self.dll is None: self._load_dll() 
    129127        kernel = self.Iqxy if input.is_2D else self.Iq 
    … …  
    175173        self.p_res = self.res.ctypes.data 
    176174 
    177     def __call__(self, pars, pd_pars, cutoff): 
     175    def __call__(self, fixed_pars, pd_pars, cutoff): 
    178176        real = np.float32 if self.input.dtype == F32 else np.float64 
    179         fixed = [real(p) for p in pars] 
    180         cutoff = real(cutoff) 
    181         loops = np.hstack(pd_pars) 
    182         loops = np.ascontiguousarray(loops.T, self.input.dtype).flatten() 
    183         loops_N = [np.uint32(len(p[0])) for p in pd_pars] 
    184177 
    185178        nq = c_int(self.input.nq) 
    186         p_loops = loops.ctypes.data 
    187         args = self.input.q_pointers + [self.p_res, nq, p_loops, cutoff] + fixed + loops_N 
     179        if pd_pars: 
     180            cutoff = real(cutoff) 
     181            loops_N = [np.uint32(len(p[0])) for p in pd_pars] 
     182            loops = np.hstack(pd_pars) 
     183            loops = np.ascontiguousarray(loops.T, self.input.dtype).flatten() 
     184            p_loops = loops.ctypes.data 
     185            dispersed = [p_loops, cutoff] + loops_N 
     186        else: 
     187            dispersed = [] 
     188        fixed = [real(p) for p in fixed_pars] 
     189        args = self.input.q_pointers + [self.p_res, nq] + dispersed + fixed 
    188190        #print pars 
    189191        self.kernel(*args) 

  • sasmodels/kernelpy.py

    r6edb74a rf734e7d  
    33 
    44from .generate import F32, F64 
     5 
     6class PyModel(object): 
     7    def __init__(self, info): 
     8        self.info = info 
     9    def __call__(self, input): 
     10        kernel = self.info['Iqxy'] if input.is_2D else self.info['Iq'] 
     11        return PyKernel(kernel, self.info, input) 
     12    def make_input(self, q_vectors): 
     13        return PyInput(q_vectors, dtype=F64) 
     14    def release(self): 
     15        pass 
    516 
    617class PyInput(object): 
    … …  
    134145    """ 
    135146 
    136     ########################################################## 
    137     #                                                        # 
    138     #   !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!   # 
    139     #   !!                                              !!   # 
    140     #   !!  KEEP THIS CODE CONSISTENT WITH GENERATE.PY  !!   # 
    141     #   !!                                              !!   # 
    142     #   !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!   # 
    143     #                                                        # 
    144     ########################################################## 
     147    ################################################################ 
     148    #                                                              # 
     149    #   !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!   # 
     150    #   !!                                                    !!   # 
     151    #   !!  KEEP THIS CODE CONSISTENT WITH KERNEL_TEMPLATE.C  !!   # 
     152    #   !!                                                    !!   # 
     153    #   !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!   # 
     154    #                                                              # 
     155    ################################################################ 
    145156 
    146157    weight = np.empty(len(pd), 'd') 
    … …  
    164175        stride = np.array([1]) 
    165176        vol_weight_index = slice(None, None) 
     177        # keep lint happy 
     178        fast_value = [None] 
     179        fast_weight = [None] 
    166180 
    167181    ret = np.zeros_like(args[0]) 
    … …  
    193207            # Correction factor for spherical integration p(theta) I(q) sin(theta) dtheta 
    194208            #spherical_correction = abs(sin(pi*args[theta_index])) if theta_index>=0 else 1.0 
    195             #spherical_correction = abs(cos(pi*args[theta_index]))*pi/2 if theta_index>=0 else 1.0 
    196             spherical_correction = 1.0 
     209            spherical_correction = abs(cos(pi*args[theta_index]))*pi/2 if theta_index>=0 else 1.0 
     210            #spherical_correction = 1.0 
    197211            ret += w*I*spherical_correction*positive 
    198212            norm += w*positive 

  • sasmodels/model_test.py

    r5ca9762 rf734e7d  
    66""" 
    77 
     8import sys 
    89import unittest 
    9 import warnings 
     10 
    1011import numpy as np 
    1112 
    12 from os.path import basename, dirname, join as joinpath 
    13 from glob import glob 
    14  
    15 try: 
    16     from .kernelcl import load_model 
    17 except ImportError,exc: 
    18     warnings.warn(str(exc)) 
    19     warnings.warn("using ctypes instead") 
    20     from .kerneldll import load_model 
    21  
    22 def load_kernel(model, dtype='single'):    
    23     kernel = load_model(model, dtype=dtype) 
    24     kernel.info['defaults'] = dict((p[0],p[2]) for p in kernel.info['parameters']) 
    25     return kernel 
    26  
    27 def get_weights(model, pars, name): 
    28     from . import weights 
    29      
    30     relative = name in model.info['partype']['pd-rel'] 
    31     disperser = pars.get(name+"_pd_type", "gaussian") 
    32     value = pars.get(name, model.info['defaults'][name]) 
    33     width = pars.get(name+"_pd", 0.0) 
    34     npts = pars.get(name+"_pd_n", 30) 
    35     nsigma = pars.get(name+"_pd_nsigma", 3.0) 
    36     v,w = weights.get_weights( 
    37             disperser, npts, width, nsigma, 
    38             value, model.info['limits'][name], relative) 
    39     return v,w/np.sum(w) 
    40  
    41 def eval_kernel(kernel, q, pars, cutoff=1e-5): 
    42     input = kernel.make_input(q) 
    43     finput = kernel(input) 
    44  
    45     fixed_pars = [pars.get(name, finput.info['defaults'][name]) 
    46                   for name in finput.fixed_pars] 
    47     pd_pars = [get_weights(finput, pars, p) for p in finput.pd_pars] 
    48     return finput(fixed_pars, pd_pars, cutoff) 
     13from .core import list_models, load_model_definition 
     14from .core import load_model_cl, load_model_dll 
     15from .core import make_kernel, call_kernel 
    4916 
    5017def annotate_exception(exc, msg): 
    … …  
    7441            exc.args = (" ".join((str(exc),msg)),) 
    7542     
    76 def suite(): 
    77     root = dirname(__file__) 
    78     files = sorted(glob(joinpath(root, 'models', "[a-zA-Z]*.py"))) 
    79     models_names = [basename(f)[:-3] for f in files] 
    80      
     43def suite(loaders, models): 
     44 
    8145    suite = unittest.TestSuite() 
    82      
    83     for model_name in models_names: 
    84         module = __import__('sasmodels.models.' + model_name) 
    85         module = getattr(module, 'models', None) 
    8646 
    87         model = getattr(module, model_name, None) 
     47    if models[0] == 'all': 
     48        skip = models[1:] 
     49        models = list_models() 
     50    else: 
     51        skip = [] 
     52    for model_name in models: 
     53        if model_name in skip: continue 
     54        model_definition = load_model_definition(model_name) 
     55 
    8856        smoke_tests = [[{},0.1,None],[{},(0.1,0.1),None]] 
    89         tests = smoke_tests + getattr(model, 'tests', []) 
     57        tests = smoke_tests + getattr(model_definition, 'tests', []) 
    9058         
    91         if tests: 
     59        if tests: # in case there are no smoke tests... 
    9260            #print '------' 
    9361            #print 'found tests in', model_name 
    9462            #print '------' 
    95      
    96             kernel = load_kernel(model) 
    97             suite.addTest(ModelTestCase(model_name, kernel, tests)) 
     63 
     64            # if ispy then use the dll loader to call pykernel 
     65            # don't try to call cl kernel since it will not be 
     66            # available in some environmentes. 
     67            ispy = callable(getattr(model_definition,'Iq', None)) 
     68 
     69            # test using opencl if desired 
     70            if not ispy and ('opencl' in loaders and load_model_cl): 
     71                test_name = "Model: %s, Kernel: OpenCL"%model_name 
     72                test = ModelTestCase(test_name, model_definition, 
     73                                     load_model_cl, tests) 
     74                print "defining", test_name 
     75                suite.addTest(test) 
     76 
     77            # test using dll if desired 
     78            if ispy or ('dll' in loaders and load_model_dll): 
     79                test_name = "Model: %s, Kernel: dll"%model_name 
     80                test = ModelTestCase(test_name, model_definition, 
     81                                     load_model_dll, tests) 
     82                print "defining", test_name 
     83                suite.addTest(test) 
    9884 
    9985    return suite 
    … …  
    10187class ModelTestCase(unittest.TestCase): 
    10288     
    103     def __init__(self, model_name, kernel, tests): 
     89    def __init__(self, test_name, definition, loader, tests): 
    10490        unittest.TestCase.__init__(self) 
    10591         
    106         self.model_name = model_name 
    107         self.kernel = kernel 
     92        self.test_name = test_name 
     93        self.definition = definition 
     94        self.loader = loader 
    10895        self.tests = tests 
    10996 
    11097    def runTest(self): 
    111         #print '------' 
    112         #print self.model_name 
    113         #print '------' 
     98        print "running", self.test_name 
    11499        try: 
     100            model = self.loader(self.definition) 
    115101            for test in self.tests: 
    116                 params = test[0] 
    117                 Q = test[1] 
    118                 I = test[2] 
    119                        
     102                pars, Q, I = test 
     103 
    120104                if not isinstance(Q, list): 
    121105                    Q = [Q] 
    … …  
    130114 
    131115                self.assertEqual(len(I), len(Q)) 
    132              
    133                 Iq = eval_kernel(self.kernel, Q_vectors, params) 
     116 
     117                kernel = make_kernel(model, Q_vectors) 
     118                Iq = call_kernel(kernel, pars) 
    134119             
    135120                self.assertGreater(len(Iq), 0)     
    … …  
    138123                for q, i, iq in zip(Q, I, Iq): 
    139124                    if i is None: continue # smoke test --- make sure it runs 
    140                     err = np.abs(i - iq) 
    141                     nrm = np.abs(i) 
     125                    err = abs(i - iq) 
     126                    nrm = abs(i) 
    142127             
    143128                    self.assertLess(err * 10**5, nrm, 'q:%s; expected:%s; actual:%s' % (q, i, iq)) 
    144129                     
    145130        except Exception,exc:  
    146             annotate_exception(exc, '\r\nModel: %s' % self.model_name) 
     131            annotate_exception(exc, self.test_name) 
    147132            raise 
    148133 
    149134def main(): 
    150     #unittest.main() 
    151     runner = unittest.TextTestRunner() 
    152     runner.run(suite()) 
     135 
     136    models = sys.argv[1:] 
     137    if models and models[0] == 'opencl': 
     138        if load_model_cl is None: 
     139            print >>sys.stderr, "opencl is not available" 
     140            sys.exit(1) 
     141        loaders = ['opencl'] 
     142        models = models[1:] 
     143    elif models and models[0] == 'dll': 
     144        # TODO: test if compiler is available? 
     145        loaders = ['dll'] 
     146        models = models[1:] 
     147    else: 
     148        loaders = ['opencl', 'dll'] 
     149    if models: 
     150        runner = unittest.TextTestRunner() 
     151        runner.run(suite(loaders, models)) 
     152    else: 
     153        print >>sys.stderr, "usage: python -m sasmodels.model_test [opencl|dll] model1 model2 ..." 
     154        print >>sys.stderr, "if model1 is 'all', then all except the remaining models will be tested" 
    153155 
    154156if __name__ == "__main__": 

  • sasmodels/models/broad_peak.py

    rf57d123 rf734e7d  
    9191 
    9292 
    93 def form_volume(): 
    94     return 1 
     93#def form_volume(): 
     94#    return 1 
    9595 
    9696def Iq(q, porod_scale, porod_exp, lorentz_scale, lorentz_length, peak_pos, lorentz_exp): 
    … …  
    102102Iq.vectorized = True 
    103103 
    104 def Iqxy(qx, qy, porod_scale, porod_exp, lorentz_scale, lorentz_length, peak_pos, lorentz_exp): 
    105     return Iq(sqrt(qx**2 + qy**2), porod_scale, porod_exp, lorentz_scale, lorentz_length, peak_pos, lorentz_exp) 
     104def Iqxy(qx, qy, *args): 
     105    return Iq(sqrt(qx**2 + qy**2), *args) 
    106106 
    107107# FOR VECTORIZED VERSION, UNCOMMENT THE NEXT LINE 

  • sasmodels/models/lamellarPC.py

    rdc02af0 rf734e7d  
    9292source = [ "lamellarPC_kernel.c"] 
    9393 
    94 # No volume normalization despite having a volume parameter 
    95 # This should perhaps be volume normalized? 
    9694form_volume = """ 
    9795    return 1.0; 
    … …  
    9997 
    10098Iqxy = """ 
    101     // never called since no orientation or magnetic parameters. 
    102     return -1.0; 
     99    return Iq(sqrt(qx*qx+qy*qy), IQ_PARAMETERS); 
    103100    """ 
    104101 

  • sasmodels/models/lamellarPC_kernel.c

    rdc02af0 rf734e7d  
    2828        //get the fractional part of Nlayers, to determine the "mixing" of N's 
    2929         
    30         n1 = trunc(Nlayers);            //rounds towards zero 
     30        n1 = (long)trunc(Nlayers);              //rounds towards zero 
    3131        n2 = n1 + 1; 
    3232        xn = (double)n2 - Nlayers;                      //fractional contribution of n1 
Note: See TracChangeset for help on using the changeset viewer.