Add database files for arch and hardware; remove Sparse and DenseMatr…

…ix files

benchmarks/double-gemm/generate.py

@@ -1,5 +1,6 @@
-from kernelforge import DenseMatrix, GenerationError, GemmGenerator
+from kernelforge import GenerationError, GemmGenerator
 from kernelforge.common.vm.vm import vm_factory
+from kernelforge.common.matrix.tensor import Tensor
 import os
 import yaml
 import argparse
@@ -22,8 +23,7 @@
 
 
 def produce_matrix(spec):
-    return DenseMatrix(num_rows=spec['num_rows'],
-                       num_cols=spec['num_cols'],
+    return Tensor(shape = [spec['num_rows'], spec['num_cols']],
                        addressing=spec['addressing'],
                        bbox=spec['bbox'])
 

benchmarks/fused/internals/frontend/aux.py

@@ -1,13 +1,13 @@
-from kernelforge.common import DenseMatrix, Addressing
+from kernelforge.common import Addressing
+from kernelforge.common.matrix.tensor import Tensor
 
 
 class VarFactory:
   @classmethod
   def produce_matrix(cls, text_descr):
     addrs = Addressing.str2addr(text_descr['addressing'])
 
-    descr = DenseMatrix(num_rows=text_descr['num_rows'],
-                        num_cols=text_descr['num_cols'],
+    descr = Tensor(shape=[text_descr['num_rows'], text_descr['num_cols']],
                         addressing=addrs,
                         bbox=text_descr['bbox'],
                         alias=text_descr['name'],

benchmarks/fused/internals/frontend/symbol_table.py

@@ -1,6 +1,7 @@
 import enum
 from typing import Dict, Union
-from kernelforge.common import DenseMatrix, Addressing
+from kernelforge.common import Addressing
+from kernelforge.common.matrix.tensor import Tensor
 
 
 class ObjType(enum.Enum):
@@ -12,7 +13,7 @@ class Attribute:
   def __init__(self, descr):
     self.descr = descr
     self.obj_type = None
-    if isinstance(descr, DenseMatrix):
+    if isinstance(descr, Tensor):
       self.obj_type = ObjType.MATRIX
     elif isinstance(descr, float):
       self.obj_type = ObjType.SCALAR

benchmarks/gemm-chain/generate.py

@@ -1,4 +1,5 @@
-from kernelforge import DenseMatrix, GenerationError, GemmGenerator
+from kernelforge import GenerationError, GemmGenerator
+from kernelforge.common.matrix.tensor import Tensor
 from kernelforge.common.vm.vm import vm_factory
 from jinja2 import Environment, FileSystemLoader
 import os
@@ -25,8 +26,7 @@
 
 
 def produce_matrix(spec):
-    return DenseMatrix(num_rows=spec['num_rows'],
-                       num_cols=spec['num_cols'],
+    return Tensor(shape = [spec['num_rows'], spec['num_cols']],
                        addressing=spec['addressing'],
                        bbox=spec['bbox'])
 

benchmarks/simple-gemm/generate.py

@@ -1,4 +1,5 @@
-from kernelforge import DenseMatrix, GenerationError
+from kernelforge import GenerationError
+from kernelforge.common.matrix.tensor import Tensor
 from kernelforge import GemmGenerator, GemmKernelType
 from kernelforge.common.vm.vm import vm_factory
 import os
@@ -22,8 +23,7 @@
 args = parser.parse_args()
 
 def produce_matrix(spec):
-    return DenseMatrix(num_rows=spec['num_rows'],
-                       num_cols=spec['num_cols'],
+    return Tensor(shape = [spec['num_rows'], spec['num_cols']],
                        addressing=spec['addressing'],
                        bbox=spec['bbox'])
 

example/abb_trans.py

@@ -1,4 +1,4 @@
-from kernelforge.common import DenseMatrix
+from kernelforge.common.matrix.tensor import Tensor
 from kernelforge.common.context import Context
 from kernelforge.common.aux import generate_tmp_matrix
 from kernelforge.generators.descriptions import GemmDescr, FloatingPointType, Addressing

example/csa.py

@@ -1,4 +1,5 @@
-from kernelforge import DenseMatrix, GenerationError, CsaGenerator
+from kernelforge import GenerationError, CsaGenerator
+from kernelforge.common.matrix.tensor import Tensor
 from kernelforge.common.vm.vm import vm_factory
 import argparse
 from kernelforge.backend.instructions.csa import CSA
@@ -18,13 +19,11 @@
 
 args = parser.parse_args()
 
-mat_a = DenseMatrix(num_rows=9,
-                    num_cols=9,
+mat_a = Tensor(shape=[9, 9],
                     addressing='strided',
                     bbox=[0, 0, 9, 9])
 
-mat_b = DenseMatrix(num_rows=9,
-                    num_cols=9,
+mat_b = Tensor(shape=[9, 9],
                     addressing='strided',
                     bbox=[0, 0, 9, 9])
 

example/four_matrices.py

@@ -1,4 +1,3 @@
-from kernelforge.common.matrix.dense import DenseMatrix
 from kernelforge.common.matrix.boundingbox import BoundingBox
 from kernelforge.common.matrix.tensor import Tensor
 from kernelforge.common.context import Context

example/gemm.py

@@ -1,4 +1,3 @@
-from kernelforge.common.matrix.dense import DenseMatrix
 from kernelforge.common.context import Context
 from kernelforge.generators.descriptions import GemmDescr
 from kernelforge.common.basic_types import FloatingPointType, Addressing

example/trace.py

@@ -1,4 +1,3 @@
-from kernelforge.common.matrix.dense import DenseMatrix
 from kernelforge.common.matrix.boundingbox import BoundingBox
 from kernelforge.common.matrix.tensor import Tensor
 from kernelforge.common.context import Context

examples/hosvd.py

@@ -15,4 +15,4 @@ def add(g):
 
   Alist = [A[i][alph[13+i] + alph[i]] for i in range(N)]
   hosvd = G[alph[0:N]] <= X[alph[13:13+N]] * reduce(lambda x, y: x * y, Alist)
-  g.add('hosvd', hosvd)
+  g.add('hosvd', hosvd, target = "gpu")

examples/minimal.py

@@ -10,4 +10,4 @@ def add(g):
   C = Tensor('C', (N, N))
 
   kernel = C['ij'] <= 2.0 * C['ij'] + A['lj'] * B['ikl'] * w['k']
-  g.add('kernel', kernel)
+  g.add('kernel', kernel, target = "gpu")

examples/optimal_ind.py

@@ -28,7 +28,7 @@ def add(g):
             kernel = [  tmp1[i1]  <= B['befl'] * D['cdel'],
                         tmp2[i2]  <= tmp1[i1]  * C['dfjk'],
                         S['abij'] <= tmp2[i2]  * A['acik'] ]
-            g.add('kernel_{}_{}'.format(i1,i2), kernel)
+            g.add('kernel_{}_{}'.format(i1,i2), kernel, target = "gpu")
 
 
 

examples/seissol_eqspp.py

@@ -13,7 +13,7 @@ def add(g):
 
   Q = Tensor('Q', (8, 20, 15))
   I = Tensor('I', (8, 20, 15))
-  g.add('seissol_stiffness', Q['skp'] <= db.kXiTDivM['lk'] * I['slq'] * db.star['qp'])
+  g.add('seissol_stiffness', Q['skp'] <= db.kXiTDivM['lk'] * I['slq'] * db.star['qp'], target = "gpu")
 
   # Reproduces recursive generation of zero blocks in Cauchy-Kowalevski prodedure,
   # described in "Sustained Petascale Performance of Seismic Simulations with SeisSol on SuperMUC",
@@ -29,5 +29,5 @@ def add(g):
     derivativeSum = DeduceIndices('kp').visit(derivativeSum)
     derivativeSum = EquivalentSparsityPattern().visit(derivativeSum)
     dQ = Tensor('dQ({})'.format(i), dQ_shape, spp=derivativeSum.eqspp())
-    g.add('derivative({})'.format(i), dQ['kp'] <= derivativeSum)
+    g.add('derivative({})'.format(i), dQ['kp'] <= derivativeSum, target = "gpu")
     dQ_prev = dQ

examples/springer.py

@@ -25,7 +25,7 @@ def add_bench(g, descr, sizes):
   B = add_tensor('B', Bind, size)
   C = add_tensor('C', Cind, size)
 
-  g.add(descr.replace('-','_'), C[Cind] <= A[Aind] * B[Bind])
+  g.add(descr.replace('-','_'), C[Cind] <= A[Aind] * B[Bind], target = "gpu")
   _bench_no = _bench_no + 1
 
 def add(g):

examples/stock.py

@@ -34,11 +34,11 @@ def add(g):
       ZR = Tensor('ZR' + pqx, (q,p))
 
       stock = R['ijk'] <= S['xyz'] * XL['xl'] * XR['li'] * YL['ym'] * YR['mj'] * ZL['zn'] * ZR['nk']
-      g.add('stock' + pqx, stock)
+      g.add('stock' + pqx, stock, target = "gpu")
 
       stock = R['ijk'] <= S['xyz'] * XLT['lx'] * XRT['il'] * YL['ym'] * YR['mj'] * ZL['zn'] * ZR['nk']
-      g.add('stock{}_trans'.format(pqx), stock)
+      g.add('stock{}_trans'.format(pqx), stock, target = "gpu")
 
       stock = R['ijk'] <= S['xyz'] * XLTP['lx'] * XRTP['il'] * YL['ym'] * YR['mj'] * ZL['zn'] * ZR['nk']
-      g.add('stock{}_trans_pad'.format(pqx), stock)
+      g.add('stock{}_trans_pad'.format(pqx), stock, target = "gpu")
 

examples/tce.py

@@ -25,10 +25,10 @@ def add(g):
   B2 = Tensor('T', (V, V, V, V))
 
   kernel = S['abij'] <= A['acik'] * B['befl'] * C['dfjk'] * D['cdel']
-  g.add('tce1', kernel)
+  g.add('tce1', kernel, target = "gpu")
 
   kernel = B2['abcd'] <= C1['sd'] * C2['rc'] * C3['qb'] * C4['pa'] * A2['pqrs']
-  g.add('tce2', kernel)
+  g.add('tce2', kernel, target = "gpu")
 
   kernel = B2['abcd'] <= C1['sd'] * C2['rc'] * C3['qb'] * C4T['ap'] * A2['pqrs']
-  g.add('tce2_trans', kernel)
+  g.add('tce2_trans', kernel, target = "gpu")