Merge pull request #255 from kodonnell/py_override

Adding override parameters to pyclblast

scripts/generator/generator.py

@@ -49,8 +49,8 @@ FILES = [
     "/src/clblast_cuda.cpp",
     "/src/pyclblast/src/pyclblast.pyx"
 ]
-HEADER_LINES = [123, 21, 127, 24, 29, 41, 29, 65, 32, 95, 21, 288]
-FOOTER_LINES = [98, 56, 112, 275, 6, 6, 6, 9, 2, 41, 55, 1]
+HEADER_LINES = [123, 21, 127, 24, 29, 41, 29, 65, 32, 95, 21, 290]
+FOOTER_LINES = [98, 56, 112, 275, 6, 6, 6, 9, 2, 41, 55, 37]
 HEADER_LINES_DOC = 0
 FOOTER_LINES_DOC = 232
 

src/pyclblast/samples/override_parameters.py

@@ -0,0 +1,42 @@
+#!/usr/bin/env python
+
+# This file is part of the CLBlast project. The project is licensed under Apache Version 2.0.
+# This file follows the PEP8 Python style guide and uses a max-width of 100 characters per line.
+#
+# Author(s):
+#   Cedric Nugteren <www.cedricnugteren.nl>
+
+import numpy as np
+import pyopencl as cl
+from pyopencl.array import Array
+import pyclblast
+from datetime import datetime
+
+if __name__ == "__main__":
+
+    # Set up pyopencl:
+    ctx = cl.create_some_context()
+    queue = cl.CommandQueue(ctx)
+
+    # Set up a basic sgemm example:
+    m, n, k = 2, 3, 4
+    a = np.random.rand(m, k).astype(dtype=np.float32)
+    b = np.random.rand(k, n).astype(dtype=np.float32)
+    c = np.empty((m, n), np.float32)
+    cla = Array(queue, a.shape, a.dtype)
+    clb = Array(queue, b.shape, b.dtype)
+    clc = Array(queue, c.shape, c.dtype)
+    cla.set(a)
+    clb.set(b)
+    clc.set(c)
+
+    # Perform sgemm on these matrices, overriding the CLBlast parameters. In this example, we'll
+    # just change the 'MWG' parameter a couple of times:
+    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 }
+    for mwg in (32, 64, 256):
+        print("Running sgemm tuned with MWG = %d" % mwg)
+        params["MWG"] = mwg
+        pyclblast.override_parameters(ctx.devices[0], 'Xgemm', 32, params)
+        pyclblast.gemm(queue, m, n, k, cla, clb, clc, a_ld=k, b_ld=n, c_ld=n)
+        assert np.allclose(clc.get(), a.dot(b)), "uh-oh, xgemm isn't behaving correctly"

src/pyclblast/src/pyclblast.pyx

@@ -1,3 +1,4 @@
+#distutils: language = c++
 #cython: binding=True
 ####################################################################################################
 # This file is part of the CLBlast project. The project is licensed under Apache Version 2.0.
@@ -13,8 +14,9 @@
 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
@@ -2083,3 +2085,39 @@ def trsm(queue, m, n, a, b, a_ld, b_ld, alpha = 1.0, right_side = False, lower_t
     return cl.Event.from_int_ptr(<size_t>event)
 
 ####################################################################################################
+# Overrides the parameters
+####################################################################################################
+
+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):
+    """
+    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**> 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))
+
+    # tidy up:
+    PyMem_Free(parameter_names)
+    PyMem_Free(parameter_values)
+
+####################################################################################################