Merge branch 'develop' of github.com:SeisSol/TensorForge into develop

AUTHORS

@@ -1,4 +1,4 @@
 Ravil Dorozhinskii <[email protected]>
 Yakup Budanaz <[email protected]>
 David Schneller <[email protected]>
-
+Elias Reutelsterz <[email protected]>

example/a_atrans_bc.py

@@ -1,34 +1,23 @@
-from kernelforge.common import DenseMatrix
 from kernelforge.common.context import Context
 from kernelforge.common.aux import generate_tmp_matrix
 from kernelforge.generators.descriptions import GemmDescr, FloatingPointType, Addressing
 from kernelforge.generators.generator import Generator
+from kernelforge.common.matrix.boundingbox import BoundingBox
+from kernelforge.common.matrix.tensor import Tensor
 
 
 # Q += A x ((A^T x B) x C)
 
 variants = {'v0': Addressing.STRIDED,
             'v1': Addressing.NONE}
 
-mat_q = DenseMatrix(num_rows=56,
-                    num_cols=56,
-                    addressing=Addressing.STRIDED,
-                    bbox=[0, 0, 20, 9])
+mat_q = Tensor([56, 56], Addressing.STRIDED, BoundingBox([0,0], [20,9]))
 
-mat_a = DenseMatrix(num_rows=56,
-                    num_cols=56,
-                    addressing=variants['v0'],
-                    bbox=[0, 0, 20, 9])
+mat_a = Tensor([56, 56], addressing=variants['v0'], bbox=BoundingBox([0,0], [20,9]))
 
-mat_b = DenseMatrix(num_rows=56,
-                    num_cols=56,
-                    addressing=Addressing.STRIDED,
-                    bbox=[0, 0, 20, 9])
+mat_b = Tensor([56, 56], Addressing.STRIDED, BoundingBox([0,0], [20,9]))
 
-mat_c = DenseMatrix(num_rows=56,
-                    num_cols=9,
-                    bbox=[0, 0, 9, 9],
-                    addressing=Addressing.STRIDED)
+mat_c = Tensor([56, 9], Addressing.STRIDED, BoundingBox([0,0], [9,9]))
 
 
 tmp1 = generate_tmp_matrix(mat_a, mat_b, True, False)
@@ -53,8 +42,11 @@
 generator = Generator(gemm_list, context)
 generator.generate()
 
-print(generator.get_launcher())
-print()
-print(generator.get_header())
-print()
-print(generator.get_kernel())
+with_output = True
+if with_output:
+  print(generator.get_header())
+  print(generator.default_generate_call_site())
+  print()
+  print(generator.get_launcher())
+  print()
+  print(generator.get_kernel())

example/ab_transb.py

@@ -1,35 +1,29 @@
-from kernelforge.common import DenseMatrix
 from kernelforge.common.context import Context
 from kernelforge.common.aux import generate_tmp_matrix
 from kernelforge.generators.descriptions import GemmDescr, FloatingPointType, Addressing
 from kernelforge.generators.generator import Generator
+from kernelforge.common.matrix.boundingbox import BoundingBox
+from kernelforge.common.matrix.tensor import Tensor
 
 
 # Q += (A x B^T) x B
 
-mat_q = DenseMatrix(num_rows=56,
-                    num_cols=56,
-                    addressing=Addressing.PTR_BASED,
-                    bbox=[0, 0, 20, 9])
+mat_q = Tensor([56, 56], Addressing.PTR_BASED, BoundingBox([0,0], [20,9]))
 
-mat_a = DenseMatrix(num_rows=56,
-                    num_cols=56,
-                    addressing=Addressing.NONE,
-                    bbox=[0, 0, 20, 9])
+mat_a = Tensor([56, 56], Addressing.NONE, BoundingBox([0,0], [20,9]))
 
-mat_b = DenseMatrix(num_rows=56,
-                    num_cols=56,
-                    addressing=Addressing.STRIDED,
-                    bbox=[0, 0, 20, 9])
+mat_b = Tensor([56, 56], Addressing.STRIDED, BoundingBox([0,0], [20,9]))
 
-tmp1 = generate_tmp_matrix(mat_a, mat_b, trans_op1=False, trans_op2=True)
+tmp1 = generate_tmp_matrix(mat_a, mat_b, trans_a=False, trans_b=True)
 
 gemm_list = [GemmDescr(trans_a=False,
                        trans_b=True,
                        a=mat_a, b=mat_b, c=tmp1),
              GemmDescr(trans_a=False,
                        trans_b=False,
-                       a=tmp1, b=mat_b, c=mat_q)]
+                       a=tmp1, b=mat_b, c=mat_q,
+                       alpha=1.0,
+                       beta=1.0)]
 
 context = Context(arch='sm_60',
                   backend='cuda',
@@ -38,8 +32,11 @@
 generator = Generator(gemm_list, context)
 generator.generate()
 
-print(generator.get_launcher())
-print()
-print(generator.get_header())
-print()
-print(generator.get_kernel())
+with_output = True
+if with_output:
+  print(generator.get_header())
+  print(generator.default_generate_call_site())
+  print()
+  print(generator.get_launcher())
+  print()
+  print(generator.get_kernel())

example/abb_trans.py

@@ -3,23 +3,18 @@
 from kernelforge.common.aux import generate_tmp_matrix
 from kernelforge.generators.descriptions import GemmDescr, FloatingPointType, Addressing
 from kernelforge.generators.generator import Generator
+from kernelforge.common.matrix.boundingbox import BoundingBox
+from kernelforge.common.matrix.tensor import Tensor
 
 
 # Q += (A x B) x B^T
-mat_q = DenseMatrix(num_rows=56,
-                    num_cols=56,
-                    addressing=Addressing.PTR_BASED,
-                    bbox=[0, 0, 20, 9])
 
-mat_a = DenseMatrix(num_rows=56,
-                    num_cols=56,
-                    addressing=Addressing.NONE,
-                    bbox=[0, 0, 20, 9])
+mat_q = Tensor([56, 56], Addressing.PTR_BASED, BoundingBox([0,0], [20,9]))
+
+mat_a = Tensor([56, 56], Addressing.NONE, BoundingBox([0,0], [20,9]))
+
+mat_b = Tensor([56, 56], Addressing.STRIDED, BoundingBox([0,0], [9,20]))
 
-mat_b = DenseMatrix(num_rows=56,
-                    num_cols=56,
-                    addressing=Addressing.STRIDED,
-                    bbox=[0, 0, 9, 20])
 
 tmp1 = generate_tmp_matrix(mat_a, mat_b)
 
@@ -28,7 +23,9 @@
                        a=mat_a, b=mat_b, c=tmp1),
              GemmDescr(trans_a=False,
                        trans_b=True,
-                       a=tmp1, b=mat_b, c=mat_q)]
+                       a=tmp1, b=mat_b, c=mat_q,
+                       alpha=1.0,
+                       beta=1.0)]
 
 context = Context(arch='sm_60',
                   backend='cuda',
@@ -37,8 +34,11 @@
 generator = Generator(gemm_list, context)
 generator.generate()
 
-print(generator.get_launcher())
-print()
-print(generator.get_header())
-print()
-print(generator.get_kernel())
+with_output = True
+if with_output:
+  print(generator.get_header())
+  print(generator.default_generate_call_site())
+  print()
+  print(generator.get_launcher())
+  print()
+  print(generator.get_kernel())

example/csa.py

@@ -1,6 +1,7 @@
 from kernelforge import DenseMatrix, GenerationError, CsaGenerator
 from kernelforge.common.vm.vm import vm_factory
 import argparse
+from kernelforge.backend.instructions.csa import CSA
 
 
 parser = argparse.ArgumentParser(description='Specify Backend and Arch of the GPU')

example/five_multiplies.py

@@ -1,49 +1,37 @@
-from kernelforge.common.matrix.dense import DenseMatrix
 from kernelforge.common.context import Context
 from kernelforge.common.aux import generate_tmp_matrix
 from kernelforge.generators.descriptions import GemmDescr
 from kernelforge.common.basic_types import FloatingPointType, Addressing
 from kernelforge.generators.generator import Generator
+from kernelforge.common.matrix.boundingbox import BoundingBox
+from kernelforge.common.matrix.tensor import Tensor
 
 
 # Q = (((A x B) x (C x B)) x D)
-mat_q = DenseMatrix(num_rows=56,
-                    num_cols=9,
-                    addressing=Addressing.STRIDED,
-                    bbox=[0, 0, 56, 9],)
 
-mat_a = DenseMatrix(num_rows=56,
-                    num_cols=56,
-                    addressing=Addressing.STRIDED,
-                    bbox=[0, 0, 56, 56])
+mat_q = Tensor([9, 9], Addressing.STRIDED, BoundingBox([0,0], [9,9]))
 
-mat_b = DenseMatrix(num_rows=56,
-                    num_cols=9,
-                    addressing=Addressing.STRIDED,
-                    bbox=[0, 0, 56, 9])
+mat_a = Tensor([56, 56], Addressing.STRIDED, BoundingBox([0,0], [56,56]))
 
-mat_c = DenseMatrix(num_rows=56,
-                    num_cols=56,
-                    bbox=[0, 0, 56, 56],
-                    addressing=Addressing.STRIDED)
+mat_b = Tensor([56, 9], Addressing.STRIDED, BoundingBox([0,0], [56,9]))
+
+mat_c = Tensor([56, 56], Addressing.STRIDED, BoundingBox([0,0], [56,56]))
+
+mat_d = Tensor([9, 9], Addressing.STRIDED, BoundingBox([0,0], [9,9]))
 
-mat_d = DenseMatrix(num_rows=9,
-                    num_cols=9,
-                    bbox=[0, 0, 9, 9],
-                    addressing=Addressing.STRIDED)
 
 
 tmp0 = generate_tmp_matrix(mat_a, mat_b)
 tmp1 = generate_tmp_matrix(mat_c, mat_b)
-tmp2 = generate_tmp_matrix(tmp0, tmp1)
+tmp2 = generate_tmp_matrix(tmp0, tmp1, trans_a=True)
 
 gemm_list = [GemmDescr(trans_a=False,
                        trans_b=False,
                        a=mat_a, b=mat_b, c=tmp0),
              GemmDescr(trans_a=False,
                        trans_b=False,
                        a=mat_c, b=mat_b, c=tmp1),
-             GemmDescr(trans_a=False, trans_b=False,
+             GemmDescr(trans_a=True, trans_b=False,
                        a=tmp0, b=tmp1, c=tmp2),
              GemmDescr(trans_a=False, trans_b=False,
                        a=tmp2, b=mat_d, c=mat_q,

example/gemm.py

@@ -1,11 +1,12 @@
 from kernelforge.common.matrix.dense import DenseMatrix
 from kernelforge.common.context import Context
-from kernelforge.common.aux import generate_tmp_matrix
 from kernelforge.generators.descriptions import GemmDescr
 from kernelforge.common.basic_types import FloatingPointType, Addressing
 from kernelforge.generators.generator import Generator
 from kernelforge.common.exceptions import GenerationError
 import argparse
+from kernelforge.common.matrix.boundingbox import BoundingBox
+from kernelforge.common.matrix.tensor import Tensor
 
 
 parser = argparse.ArgumentParser(description="Specify Backend and Arch of the GPU")
@@ -23,20 +24,12 @@
 
 args = parser.parse_args()
 
-mat_a = DenseMatrix(num_rows=56,
-                    num_cols=18,
-                    addressing=Addressing.STRIDED,
-                    bbox=[0, 0, 56, 18])
+mat_a = Tensor([56, 18], Addressing.STRIDED, BoundingBox([0,0], [56,18]))
 
-mat_b = DenseMatrix(num_rows=18,
-                    num_cols=18,
-                    addressing=Addressing.STRIDED,
-                    bbox=[0, 0, 18, 18])
+mat_b = Tensor([18, 18], Addressing.STRIDED, BoundingBox([0,0], [18,18]))
+
+mat_c = Tensor([56, 18], Addressing.STRIDED, BoundingBox([0,0], [56,18]))
 
-mat_c = DenseMatrix(num_rows=56,
-                    num_cols=18,
-                    bbox=[0, 0, 56, 18],
-                    addressing=Addressing.STRIDED)
 
 try:
   vm = Context(arch=args.arch, backend=args.backend, fp_type=FloatingPointType.FLOAT)