add snitch support

CMakeLists.txt

@@ -15,8 +15,8 @@ if(TOOLCHAIN STREQUAL GCC)
   set(CMAKE_INTERPROCEDURAL_OPTIMIZATION TRUE)
 endif()
 
-set(platform MemPool CACHE STRING "Platform (MemPool, QEMU, Siracusa, Siracusa_w_neureka, PULP-Open, Generic)")
-set_property(CACHE platform PROPERTY STRINGS MemPool QEMU Siracusa Siracusa_w_neureka PULP-Open Generic)
+set(platform MemPool CACHE STRING "Platform (MemPool, QEMU, Siracusa, Siracusa_w_neureka, PULP-Open, Generic, Snitch)")
+set_property(CACHE platform PROPERTY STRINGS MemPool QEMU Siracusa Siracusa_w_neureka PULP-Open Generic Snitch)
 
 if(platform STREQUAL MemPool)
   message(STATUS "Building for platform 'MemPool'")
@@ -30,6 +30,8 @@ elseif(platform STREQUAL PULPOpen)
   message(STATUS "Building for platform 'PULP-Open'")
 elseif(platform STREQUAL Generic)
   message(STATUS "Building for platform 'Generic'")
+elseif(platform STREQUAL Snitch)
+  message(STATUS "Building for platform 'Snitch'")
 else()
   message(FATAL_ERROR "Invalid platform '${platform}' specified!")
 endif()
@@ -181,4 +183,33 @@ if(platform STREQUAL Siracusa OR platform STREQUAL Siracusa_w_neureka OR platfor
 
 endif()
 
+if(platform STREQUAL Snitch)
+
+  if(TOOLCHAIN STREQUAL LLVM)
+    set(CMAKE_TOOLCHAIN_FILE ${CMAKE_CURRENT_LIST_DIR}/cmake/snitch/toolchain_llvm.cmake)
+  else()
+    set(CMAKE_TOOLCHAIN_FILE ${CMAKE_CURRENT_LIST_DIR}/cmake/snitch/toolchain_gcc.cmake)
+  endif()
+
+  include(${CMAKE_CURRENT_LIST_DIR}/cmake/snitch/snitch.cmake)
+
+  include(${CMAKE_CURRENT_LIST_DIR}/cmake/snitch/snitch_cluster/snitch_cluster.cmake)
+
+  project(deeploy LANGUAGES C ASM)
+
+  message(STATUS "============================= ${platform} Configuration ============================")
+  message(STATUS "[cMake  ]   Number of total cores  = " ${NUM_CORES})
+  message(STATUS "================================================================================")
+  message(STATUS "")
+
+  add_subdirectory(TargetLibraries/Generic)
+  add_subdirectory(TargetLibraries/Snitch)
+  target_include_directories(deeploysnitch PUBLIC TargetLibraries/Generic/inc)
+
+  add_subdirectory(DeeployTest)
+  target_link_libraries(deeploylib INTERFACE deeploybasic deeploysnitch)
+
+endif()
+
+
 print_simulation_config()

Deeploy/Targets/Snitch/Deployer.py

@@ -0,0 +1,77 @@
+# ----------------------------------------------------------------------
+#
+# File: SnitchDeployer.py
+#
+# Last edited: 23.04.2024
+#
+# Copyright (C) 2024, ETH Zurich and University of Bologna.
+#
+# Authors:
+# - Philip Wiese ([email protected]), ETH Zurich
+#
+# ----------------------------------------------------------------------
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the License); you may
+# not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an AS IS BASIS, WITHOUT
+# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Callable, Dict, Type
+
+import onnx_graphsurgeon as gs
+
+from Deeploy.AbstractDataTypes import Pointer
+from Deeploy.DeeployTypes import DeploymentPlatform, TopologyOptimizer
+
+# from Deeploy.NetworkDeployers.SignPropDeployer import SignPropDeployer
+from Deeploy.CommonExtensions.NetworkDeployers.SignPropDeployer import SignPropDeployer
+
+# from Deeploy.OptimizationPasses.TopologyOptimizationPasses.BasicPasses import ReshapeConstOptPass, \
+#    TransposeConstOptPass, TransposeMergePass, TransposeSplitPass
+from Deeploy.Targets.Generic.TopologyOptimizationPasses.Passes import ReshapeConstOptPass, TransposeConstOptPass, \
+    TransposeMergePass, TransposeSplitPass
+
+# from Deeploy.OptimizationPasses.TopologyOptimizationPasses.LoweringOptimizationPasses import NCHWtoNHWCPass, \
+#    RemoveGlobalOutputReshapePass, TransposeMatmulInputsPass
+from Deeploy.CommonExtensions.OptimizationPasses.TopologyOptimizationPasses.LoweringOptimizationPasses import \
+    NCHWtoNHWCPass, RemoveGlobalOutputReshapePass, TransposeMatmulInputsPass
+
+class SnitchDeployer(SignPropDeployer):
+
+    def __init__(self,
+                 graph: gs.Graph,
+                 deploymentPlatform: DeploymentPlatform,
+                 inputTypes: Dict[str, Type[Pointer]],
+                 loweringOptimizer: TopologyOptimizer,
+                 scheduler: Callable = lambda x: x,
+                 name: str = 'DeeployNetwork',
+                 default_channels_first = False,
+                 deeployStateDir: str = "DeeployStateDir",
+                 inputOffsets = {}):
+        super().__init__(graph,
+                         deploymentPlatform,
+                         inputTypes,
+                         loweringOptimizer,
+                         scheduler,
+                         name,
+                         default_channels_first = default_channels_first,
+                         deeployStateDir = deeployStateDir,
+                         inputOffsets = inputOffsets)
+
+        self.loweringOptimizer.passes += [
+            TransposeMatmulInputsPass(),
+            NCHWtoNHWCPass(self.default_channels_first),
+            TransposeSplitPass(),
+            TransposeMergePass(),
+            TransposeConstOptPass(),
+            ReshapeConstOptPass(),
+            RemoveGlobalOutputReshapePass(),
+        ]

Deeploy/Targets/Snitch/Platform.py

@@ -0,0 +1,179 @@
+# ----------------------------------------------------------------------
+#
+# File: SnitchPlatform.py
+#
+# Last edited: 23.04.2024
+#
+# Copyright (C) 2024, ETH Zurich and University of Bologna.
+#
+# Authors:
+# - Philip Wiese ([email protected]), ETH Zurich
+#
+# ----------------------------------------------------------------------
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the License); you may
+# not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an AS IS BASIS, WITHOUT
+# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import List
+
+import numpy as np
+
+# from Deeploy.Bindings.BasicBindings import BasicGatherBindings, BasicMatMulBinding, BasicPad1DBindings, \
+#    BasicPad2DBindings, BasicReshapeBindings, BasicRQIntegerDivBinding
+
+from Deeploy.Targets.Generic.Bindings import BasicGatherBindings,BasicMatMulBinding, BasicPad1DBindings, \
+    BasicPad2DBindings, BasicReshapeBindings, BasicRQIntegerDivBinding
+
+from Deeploy.DeeployTypes import ConstantBuffer, DeploymentEngine, DeploymentPlatform, NodeMapper, NodeTemplate, \
+    StructBuffer, TopologyOptimizer, TransientBuffer, VariableBuffer
+
+# from Deeploy.Layers.BasicLayers import GatherLayer, MatMulLayer, PadLayer, ReshapeLayer, RQIntegerDivLayer
+from Deeploy.Targets.Generic.Layers import GatherLayer, MatMulLayer, PadLayer, ReshapeLayer, RQIntegerDivLayer
+
+# from Deeploy.OptimizationPasses.TopologyOptimizationPasses.BasicPasses import IntegerDivRequantMergePass, \
+#    MergeConstAddAndRequantPass, MergeTrueIntegerDivRequantShiftPass, RQSSplitPass, SkipEmptyConcatPass, \
+#    SkipUnityRequantPass, iGELURequantMergePass, iHardswishRequantMergePass
+
+from Deeploy.Targets.Generic.TopologyOptimizationPasses.Passes import IntegerDivRequantMergePass, \
+    MergeConstAddAndRequantPass, MergeTrueIntegerDivRequantShiftPass, RQSSplitPass, SkipEmptyConcatPass, \
+    SkipUnityRequantPass, iGELURequantMergePass, iHardswishRequantMergePass
+
+# from Deeploy.Parsers.BasicParsers import GatherParser, MatMulParser, Pad1DParser, Pad2DParser, RQIntegerDivParser, \
+#    UnsqueezeParser
+from Deeploy.Targets.Generic.Parsers import GatherParser, MatMulParser, Pad1DParser, Pad2DParser, RQIntegerDivParser, \
+    UnsqueezeParser
+
+# from Deeploy.Templates.BasicTemplates import AllocateTemplate as BasicAllocateTemplate
+from Deeploy.Targets.Generic.Templates import AllocateTemplate as BasicAllocateTemplate
+
+from Deeploy.Targets.Snitch.Templates import AllocateTemplate, FreeTemplate
+
+GatherMapper = NodeMapper(GatherParser(), BasicGatherBindings)
+Pad1DMapper = NodeMapper(Pad1DParser(), BasicPad1DBindings)
+Pad2DMapper = NodeMapper(Pad2DParser(), BasicPad2DBindings)
+UnsqueezeMapper = NodeMapper(UnsqueezeParser(), BasicReshapeBindings)
+
+RQIntegerDivMapper = NodeMapper(RQIntegerDivParser(), [BasicRQIntegerDivBinding])
+
+MatMulMapper = NodeMapper(MatMulParser(), [BasicMatMulBinding])
+
+SnitchMapping = {
+    'RQIntegerDiv': RQIntegerDivLayer([RQIntegerDivMapper]),
+    'Gather': GatherLayer([GatherMapper]),
+    'Pad': PadLayer([Pad1DMapper, Pad2DMapper]),
+    'Unsqueeze': ReshapeLayer([UnsqueezeMapper]),
+    'MatMul': MatMulLayer([MatMulMapper]),
+}
+
+
+class SnitchVariableBuffer(VariableBuffer):
+
+    initTemplate = AllocateTemplate.snitchL2InitTemplate
+    allocTemplate = AllocateTemplate.snitchGenericAllocate
+    deallocTemplate = FreeTemplate.snitchGenericFree
+
+    def _bufferRepresentation(self):
+
+        if hasattr(self, "_memoryLevel"):
+            memoryLevel = self._memoryLevel
+        else:
+            memoryLevel = None
+
+        return {
+            "type": self._instance,
+            "name": self.name,
+            "size": int(np.prod(self.shape)),
+            "_memoryLevel": memoryLevel
+        }
+
+
+class SnitchTransientBuffer(TransientBuffer):
+
+    initTemplate = AllocateTemplate.snitchL2InitTemplate
+    allocTemplate = AllocateTemplate.snitchGenericAllocate
+    deallocTemplate = FreeTemplate.snitchGenericFree
+
+    # allocTemplate = AllocateTemplate.snitchL2AllocateTemplate
+    # deallocTemplate = FreeTemplate.snitchL2GlobalTemplate
+
+    def _bufferRepresentation(self):
+
+        if hasattr(self, "_memoryLevel"):
+            memoryLevel = self._memoryLevel
+        else:
+            memoryLevel = None
+
+        return {"type": self._type, "name": self.name, "size": self.size, "_memoryLevel": memoryLevel}
+
+
+class SnitchConstantBuffer(ConstantBuffer):
+
+    initTemplate = AllocateTemplate.snitchGenericGlobalInitTemplate
+    allocTemplate = AllocateTemplate.snitchL2GlobalAllocateTemplate
+    deallocTemplate = FreeTemplate.snitchL2GlobalTemplate
+
+    def _bufferRepresentation(self):
+        nodeRep = super()._bufferRepresentation()
+
+        if hasattr(self, "_memoryLevel"):
+            memoryLevel = self._memoryLevel
+        else:
+            memoryLevel = None
+
+        nodeRep["_memoryLevel"] = memoryLevel
+
+        return nodeRep
+
+
+class SnitchStructBuffer(StructBuffer):
+
+    initTemplate = BasicAllocateTemplate.referenceStructInitTemplate
+    allocTemplate = BasicAllocateTemplate.referenceStructAllocateTemplate
+    deallocTemplate = NodeTemplate("")
+
+
+SnitchOptimizer = TopologyOptimizer([
+    SkipEmptyConcatPass(),
+    SkipUnityRequantPass(previous_op_regex = "Concat", num_inputs = 2),
+    SkipUnityRequantPass(previous_op_regex = "Reshape|Transpose", num_inputs = 1),
+    SkipUnityRequantPass(previous_op_regex = "Reshape|Transpose", num_inputs = 1),
+    RQSSplitPass(),
+    MergeTrueIntegerDivRequantShiftPass(),
+    IntegerDivRequantMergePass(),
+    iGELURequantMergePass(),
+    iHardswishRequantMergePass(),
+    MergeConstAddAndRequantPass(),
+])
+
+_includeList = [
+    "snrt.h",
+    "DeeploySnitchMath.h",
+]
+
+
+class SnitchClusterEngine(DeploymentEngine):
+
+    def __init__(self, name: str, Mapping = SnitchMapping, initCode = "", includeList = _includeList) -> None:
+        super().__init__(name, Mapping, initCode, includeList)
+
+
+class SnitchPlatform(DeploymentPlatform):
+
+    def __init__(self,
+                 engines = [SnitchClusterEngine("SnitchCluster")],
+                 variableBuffer = SnitchVariableBuffer,
+                 constantBuffer = SnitchConstantBuffer,
+                 structBuffer = SnitchStructBuffer,
+                 transientBuffer = SnitchTransientBuffer,
+                 includeList: List[str] = _includeList):
+        super().__init__(engines, variableBuffer, constantBuffer, structBuffer, transientBuffer)

Deeploy/Targets/Snitch/Templates/AllocateTemplate.py

@@ -0,0 +1,79 @@
+# ----------------------------------------------------------------------
+#
+# File: AllocateTemplate.py
+#
+# Last edited: 23.04.2024
+#
+# Copyright (C) 2024, ETH Zurich and University of Bologna.
+#
+# Authors:
+# - Philip Wiese ([email protected]), ETH Zurich
+#
+# ----------------------------------------------------------------------
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the License); you may
+# not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an AS IS BASIS, WITHOUT
+# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from Deeploy.DeeployTypes import NodeTemplate
+
+snitchL2InitTemplate = NodeTemplate("${type.typeName} ${name};\n")
+
+snitchL1InitTemplate = NodeTemplate("${type.typeName} ${name};\n")
+
+snitchL2AllocateTemplate = NodeTemplate(
+    "${name} = (${type.typeName}) snrt_l3alloc(sizeof(${type.referencedType.typeName}) * ${size});\n")
+
+snitchL1AllocateTemplate = NodeTemplate(
+    "${name} = (${type.typeName}) snrt_l1alloc(sizeof(${type.referencedType.typeName}) * ${size});\n")
+
+snitchL2GlobalInitTemplate = NodeTemplate("static ${type.referencedType.typeName} ${name}[${size}] = {${values}};\n")
+
+snitchL1GlobalInitTemplate = NodeTemplate("static ${type.referencedType.typeName} ${name}[${size}] = {${values}};\n")
+
+snitchL2GlobalAllocateTemplate = NodeTemplate("")
+
+snitchL1GlobalAllocateTemplate = NodeTemplate("")
+
+snitchL2StructInitTemplate = NodeTemplate("""static ${type.typeName} ${name};
+""")
+
+snitchL2StructAllocateTemplate = NodeTemplate(""" % for key, value in structDict.items():
+    ${name}.${key} = ${value};
+% endfor """)
+
+snitchGenericStructInitTemplate = NodeTemplate("""
+% if _memoryLevel == "L1":
+static ${type.typeName} ${name};\n
+% elif _memoryLevel == "L2" or _memoryLevel is None:
+static ${type.typeName} ${name};\n
+% endif
+""")
+
+snitchGenericGlobalInitTemplate = NodeTemplate("""
+% if _memoryLevel == "L1":
+static ${type.referencedType.typeName} ${name}[${size}] = {${values}};\n
+% elif _memoryLevel == "L2" or _memoryLevel is None:
+static ${type.referencedType.typeName} ${name}[${size}] = {${values}};\n
+% endif
+""")
+
+snitchGenericAllocate = NodeTemplate("""
+% if _memoryLevel == "L1":
+${name} = (${type.typeName}) snrt_l1alloc(sizeof(${type.referencedType.typeName}) * ${size});\n
+% elif _memoryLevel == "L2" or _memoryLevel is None:
+${name} = (${type.typeName}) snrt_l3alloc(sizeof(${type.referencedType.typeName}) * ${size});\n% else:
+//COMPILER BLOCK - MEMORYLEVEL ${_memoryLevel} NOT FOUND \n
+${name} = (${type.typeName}) snrt_l3alloc(sizeof(${type.referencedType.typeName}) * ${size});\n
+// ${name} with size ${size} allocated in L2!
+% endif
+""")