Kaushik Kulkarni kaushikcfd

Make sure about the assertions.
Merging ArrayCallInstruction with CallInstruction(or Bye Bye ArrayCall;)
Remove the _underscores
Coefficient collector for the strides
Handle temporaries
Stride for all types
Negotiation for inames??? i.e. dont have mulitple l.0 along the interface
read about it: TYpe inference: check from user
Learned about callable
One line after the mvim folding marker and title with lower case

Introduction

After the discussion with the Firedrake group on 8th Februray, we started working towards function calls on arrays in Loopy.
In order to tackle the generalized problem we are working towards calling kernels from kernels within Loopy, so that the intended operation on arrays could be packed in a separate callee kernel.
WIP Merge Request: https://gitlab.tiker.net/inducer/loopy/merge_requests/232

The expected behavior is:

callee_knl = lp.make_kernel(
 "{i, j}: 0&lt;=i, j&lt;16",

Context: argmin/argmax are one of the last bits whose tests are needed to be handled. I am having the following troubles with intergrating these reductions into the new function interface.

Problems about `argmin/argmax`

Outline about how ArgReductionOp works:

During the creation phase -- gets identified and tagged as a reduction operation.
During the type inference, gets interpreted as a single instruction(which is a good thing). For the type inference redirects to the result_dtypes method of the class ArgReductionOp.
At the realize_reduction step in the preprocess part of the pipeline it gets converted to 3 instructions and one of the instructions would point be a call to Call(<class ArgExtOp>, (*parameters))
In the end, we append the inline dtype1 loopy_argmin_dtype1_dtype2_op(...) function to the set of preambles of the device code.

Do a local install of petsc.
the configuration options of petsc are:

CUDA:

./configure --download-eigen=/home/kgk2/pack/eigen-3.3.3.tgz --with-fortran-bindings=0 --download-chaco --download-metis --download-parmetis --download-scalapack --download-hypre --download-mumps --download-netcdf --download-hdf5=https://support.hdfgroup.org/ftp/HDF5/releases/hdf5-1.10/hdf5-1.10.6/src/hdf5-1.10.6.tar.bz2 --download-pnetcdf --download-exodusii --download-fblaslapack --with-cuda=1 --with-cuda-dir=/home/kgk2/pack/cuda --download-zlib --with-cudac=nvcc

ViennaCL:

./configure --download-eigen=/home/kgk2/pack/eigen-3.3.3.tgz --with-fortran-bindings=0 --download-chaco --download-metis --download-parmetis --download-scalapack --download-hypre --download-mumps --download-netcdf --download-hdf5=https://support.hdfgroup.org/ftp/HDF5/releases/hdf5-1.10/hdf5-1.10.6/src/hdf5-1.10.6.tar.bz2 --download-pnetcdf --download-exodusii --download-fblaslapack --download-zlib --with-viennacl=1 --download-viennacl

	#include <math.h>
	#include <petsc.h>


	static double const form_t0[3 * 3] = { 0.6666666666666669, 0.16666666666666663, 0.16666666666666666, 0.16666666666666674, 0.16666666666666663, 0.6666666666666665, 0.16666666666666669, 0.6666666666666666, 0.16666666666666663 };
	static double const form_t4[3] = { 0.16666666666666666, 0.16666666666666666, 0.16666666666666666 };

	void wrap_form00_cell_integral_otherwise(int const start, int const end, Mat const mat0, double const __restrict__ dat0, double const __restrict__ glob0, int const *__restrict__ map0)
	{
	double form_t1;

	import loopy as lp
	from loopy.isl_helpers import simplify_via_aff
	from pymbolic.primitives import CallWithKwargs
	from loopy.kernel.function_interface import (get_kw_pos_association,
	register_pymbolic_calls_to_knl_callables)
	from loopy.symbolic import IdentityMapper


	class DimChanger(IdentityMapper):
	def __init__(self, caller_arg_dict, callee_arg_dict, callee_to_caller_args):

	import loopy as lp
	import pyopencl as cl
	import pyopencl.clrandom as cl_random
	import numpy as np
	import numpy.linalg as la

	from loopy.transform.register_callable import _match_caller_callee_argument_dimension

	ctx = cl.create_some_context()
	queue = cl.CommandQueue(ctx)

	from petsc4py import PETSc
	import numpy as np
	import coffee.system
	from pyop2 import compilation
	import ctypes


	n = 10

	a_petsc = PETSc.Vec().create(PETSc.COMM_WORLD)

	from petsc4py import PETSc
	import loopy as lp
	import numpy as np
	import pyopencl as cl
	import coffee.system
	from pyop2 import compilation
	import ctypes
	from mako.template import Template
	import re

Kaushik Kulkarni kaushikcfd

Introduction

Problems about argmin/argmax

Problems about `argmin/argmax`