tidying up pyclblast override_parameters api, and added example

src/pyclblast/samples/sgemm.py

@@ -10,6 +10,7 @@ import numpy as np
 import pyopencl as cl
 from pyopencl.array import Array
 import pyclblast
+from datetime import datetime
 
 # Settings for this sample
 dtype = 'float32'
@@ -19,7 +20,7 @@ ctx = cl.create_some_context()
 queue = cl.CommandQueue(ctx)
 
 print("# Setting up Numpy arrays")
-m, n, k = 2, 3, 4
+m, n, k = 128, 256, 512
 a = np.random.rand(m, k).astype(dtype=dtype)
 b = np.random.rand(k, n).astype(dtype=dtype)
 c = np.random.rand(m, n).astype(dtype=dtype)
@@ -34,5 +35,17 @@ clc.set(c)
 
 print("# Example level-3 operation: GEMM")
 pyclblast.gemm(queue, m, n, k, cla, clb, clc, a_ld=k, b_ld=n, c_ld=n)
-print("# Matrix C result: %s" % clc.get())
-print("# Expected result: %s" % (np.dot(a, b)))
+print("# PyCLBlast matrix result is correct?:", np.allclose(clc.get(), np.dot(a, b)))
+
+print("# GFLOPS when tuned with different values of MWG:")
+params = { "KWG": 32, "KWI": 2, "MDIMA": 8, "MDIMC": 8, "MWG": 64, "NDIMB": 8, "NDIMC": 8, "NWG": 64, "SA": 0, "SB": 0, "STRM": 0, "STRN": 0, "VWM": 4, "VWN": 1 }
+mwg = 1
+while mwg <= 256:
+    params["MWG"] = mwg
+    pyclblast.override_parameters(ctx.devices[0], 'Xgemm', 32, params)
+    for i in range(100):
+        if i == 10:
+            t0 = datetime.now()
+        pyclblast.gemm(queue, m, n, k, cla, clb, clc, a_ld=k, b_ld=n, c_ld=n)
+    print("#\tMWG = %-3d : %4d" % (mwg,  int(2 * m * n * k / ((datetime.now() - t0).total_seconds() / 100) / 1024 ** 3)))
+    mwg *= 4

src/pyclblast/src/pyclblast.pyx

@@ -15,6 +15,8 @@ import numpy as np
 import pyopencl as cl
 from pyopencl.array import Array
 from libcpp cimport bool
+from cpython.mem cimport PyMem_Malloc, PyMem_Free
+from libc.string cimport strdup
 
 ####################################################################################################
 # CLBlast and OpenCL data-types
@@ -2086,38 +2088,36 @@ def trsm(queue, m, n, a, b, a_ld, b_ld, alpha = 1.0, right_side = False, lower_t
 # Overrides the parameters
 ####################################################################################################
 
-from libc.stdlib cimport malloc, free
-from libc.string cimport strdup
-
 cdef extern from "clblast_c.h":
     ctypedef struct _cl_device_id:
         pass
     ctypedef _cl_device_id* cl_device_id
-
     CLBlastStatusCode CLBlastOverrideParameters(const cl_device_id device, const char* kernel_name, const CLBlastPrecision precision, const size_t num_parameters, const char** parameters_names, const size_t* parameters_values)
 
 def override_parameters(device, kernel_name, precision, parameters):
     """
-    precision = 16, 32, 64, 3232, 6464
-    kernel name = unicode string
-    parameters = 
+    Override the current parameters for the given kernel, on this device, with this precision.
     """
  
     cdef cl_device_id device_id = <cl_device_id><size_t>device.int_ptr
+
+    # read the parameters dictionary into names/values arrays, for use in CLBlastOverrideParameters
     cdef size_t n = len(parameters)
-    cdef const char **parameter_names = <const char**> malloc(n * sizeof(char*))
-    cdef size_t *parameter_values = <size_t*> malloc(n * sizeof(size_t))
-
-    # TODO: check mallocs
-
+    cdef const char **parameter_names = <const char**> PyMem_Malloc(n * sizeof(char*))
+    cdef size_t *parameter_values = <size_t*> PyMem_Malloc(n * sizeof(size_t))
+    if not (parameter_names or parameter_values):
+        raise MemoryError()
     for i, (k, v) in enumerate(parameters.items()):
         parameter_names[i] = strdup(k.encode('ascii'))
         parameter_values[i] = v
     
+    # call the underlying API
     err = CLBlastOverrideParameters(device_id, kernel_name.encode('ascii'), precision, n, parameter_names, parameter_values)
     if err != CLBlastSuccess:
         raise RuntimeError("PyCLBlast: 'OverrideParameters' failed: %s" % get_status_message(err))
 
-    # TODO: free etc.
+    # tidy up:
+    PyMem_Free(parameter_names)
+    PyMem_Free(parameter_values)
 
 ####################################################################################################