Remove memory-aware node bindings#123
Conversation
📝 WalkthroughSummary by CodeRabbit
WalkthroughRemoved memory-aware per-node parsing and memory-level binding wrappers; eliminated Wmem binding/tiling variants across Neureka target; added requantization helpers and refactored Neureka tile-constraint classes to branch on weight SRAM vs non‑SRAM and produce requant-aware load schedules. Memory-level per-node checks removed. Changes
Sequence Diagram(s)sequenceDiagram
participant Parser as Parser
participant Engine as Engine/Mapper
participant Tiler as Tiler
participant Binding as TilingBinding
Note over Parser,Engine: Before — parser → (Wmem + non‑Wmem bindings)
Parser->>Engine: map node to (WmemBinding + Binding)
Engine->>Tiler: provide Wmem + non‑Wmem tiling bindings
Tiler->>Binding: use Wmem-aware constraints
Note over Parser,Engine: After — parser → Binding (direct)
Parser->>Engine: map node to Binding
Engine->>Tiler: provide non‑Wmem tiling bindings
Tiler->>Binding: use unified / requant-aware constraints
sequenceDiagram
participant Constraint as TileConstraint
participant Ctxt as NetworkContext
participant Requant as RequantHelpers
participant Schedule as TilingSchedule
Constraint->>Ctxt: lookup weight buffer (VariableBuffer)
alt weight memory == SRAM
Ctxt-->>Constraint: SRAM
Constraint->>Schedule: compute per-cube weight_addr_offset (SRAM path)
else weight memory != SRAM
Ctxt-->>Constraint: non‑SRAM
Constraint->>Requant: requantLoadSchedule(absoluteOutputCubes, ctxt, op)
Requant-->>Schedule: mul/add load schedule per cube
Constraint->>Schedule: merge weight info into input/load schedule
end
Schedule-->>Constraint: final tiling load schedule
Estimated code review effort🎯 4 (Complex) | ⏱️ ~60 minutes Suggested labels
Suggested reviewers
Pre-merge checks and finishing touches❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✨ Finishing touches
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
Deeploy/CommonExtensions/NetworkDeployers/NetworkDeployerWrapper.py (1)
5-64: RestoreNetworkContextimport to avoid runtime failure
NetworkContextis still used in the type hints (e.g., Line 51) but is no longer imported. Because these annotations are evaluated at module import time (nofrom __future__ import annotationssafeguard), the module will now raise aNameErroras soon as it loads. Please reintroduce the import (or switch to a quoted forward reference).
🧹 Nitpick comments (12)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py (2)
227-227: Addstrict=Truetozip()for safety.Python 3.10+ supports
strict=Trueto catch length mismatches between iterables. This prevents silent bugs ifoutputCubesandinputLoadSchedulediverge in length.Apply this diff:
- for cube, load in zip(outputCubes, inputLoadSchedule): + for cube, load in zip(outputCubes, inputLoadSchedule, strict=True):
257-261: Addstrict=Truetozip()for safety.Similar to line 227, this
zip()should includestrict=Trueto catch length mismatches.Apply this diff:
newInputLoadSchedule = [{ **load, **rqLoad - } for load, rqLoad in zip(tilingSchedule.inputLoadSchedule, requantSchedule)] + } for load, rqLoad in zip(tilingSchedule.inputLoadSchedule, requantSchedule, strict=True)]Deeploy/Targets/Neureka/TileConstraints/RequantHelpers.py (1)
64-68: Addstrict=Truetozip()for safety.For consistency and to catch potential length mismatches, add
strict=True.Apply this diff:
newInputLoadSchedule = [{ **schedule, "mul": mul, "add": add, - } for schedule, mul, add in zip(tilingSchedule.inputLoadSchedule, inputMulCubes, inputAddCubes)] + } for schedule, mul, add in zip(tilingSchedule.inputLoadSchedule, inputMulCubes, inputAddCubes, strict=True)]Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py (4)
161-162: Consider using_prefix for unused unpacked variables.Static analysis flags
BatchOffset,HOffset,WOffset, andBatchSizeas unused. While the current code works, prefixing with_makes the intent explicit.Apply this diff:
- (BatchOffset, HOffset, WOffset, COffset) = cube.offset - (BatchSize, HSize, WSize, CSize) = cube.dims + (_BatchOffset, _HOffset, _WOffset, COffset) = cube.offset + (_BatchSize, HSize, WSize, CSize) = cube.dims
212-232: Code duplication with NeurekaDepthwiseConstraint.This weight handling logic is nearly identical to lines 210-230 in NeurekaDepthwiseConstraint.py. The only difference is the weight cube construction (different dimension ordering).
Consider extracting this pattern into a helper function in RequantHelpers.py to reduce duplication across Dense, Depthwise, and Pointwise constraint classes.
229-229: Addstrict=Truetozip()for safety.For consistency with best practices in Python 3.10+, add
strict=True.Apply this diff:
- for cube, load in zip(outputCubes, inputLoadSchedule): + for cube, load in zip(outputCubes, inputLoadSchedule, strict=True):
259-263: Addstrict=Truetozip()for safety.Apply this diff:
newInputLoadSchedule = [{ **load, **rqLoad - } for load, rqLoad in zip(tilingSchedule.inputLoadSchedule, requantSchedule)] + } for load, rqLoad in zip(tilingSchedule.inputLoadSchedule, requantSchedule, strict=True)]Deeploy/Targets/Neureka/TileConstraints/NeurekaPointwiseConstraint.py (5)
48-52: Document why SRAM weights are constrained to Max().Unlike the Dense and Depthwise variants, the Pointwise constraint explicitly constrains
weightOutChannelVar == weightOutChannelVar.Max()for SRAM weights. This difference suggests Pointwise has specific requirements for SRAM weight tiling.Consider adding a comment explaining why Pointwise forces full channel tiling for SRAM weights while Dense/Depthwise don't.
191-192: Consider using_prefix for unused unpacked variables.Static analysis flags
BatchOffset,HOffset,WOffset, andBatchSizeas unused.Apply this diff:
- (BatchOffset, HOffset, WOffset, COffset) = cube.offset - (BatchSize, HSize, WSize, CSize) = cube.dims + (_BatchOffset, _HOffset, _WOffset, COffset) = cube.offset + (_BatchSize, HSize, WSize, CSize) = cube.dims
242-262: Code duplication with Dense and Depthwise constraints.This weight handling logic is duplicated across NeurekaDenseConstraint.py (lines 212-232), NeurekaDepthwiseConstraint.py (lines 210-230), and here. Consider consolidating this pattern into a shared helper function.
259-259: Addstrict=Truetozip()for safety.Apply this diff:
- for cube, load in zip(outputCubes, inputLoadSchedule): + for cube, load in zip(outputCubes, inputLoadSchedule, strict=True):
289-293: Addstrict=Truetozip()for safety.Apply this diff:
newInputLoadSchedule = [{ **load, **rqLoad - } for load, rqLoad in zip(tilingSchedule.inputLoadSchedule, requantSchedule)] + } for load, rqLoad in zip(tilingSchedule.inputLoadSchedule, requantSchedule, strict=True)]
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (10)
Deeploy/CommonExtensions/NetworkDeployers/NetworkDeployerWrapper.py(1 hunks)Deeploy/MemoryLevelExtension/MemoryLevels.py(1 hunks)Deeploy/MemoryLevelExtension/NetworkDeployers/MemoryLevelDeployer.py(1 hunks)Deeploy/Targets/Neureka/Bindings.py(0 hunks)Deeploy/Targets/Neureka/Engine.py(1 hunks)Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py(7 hunks)Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py(6 hunks)Deeploy/Targets/Neureka/TileConstraints/NeurekaPointwiseConstraint.py(5 hunks)Deeploy/Targets/Neureka/TileConstraints/RequantHelpers.py(1 hunks)Deeploy/Targets/Neureka/Tiler.py(1 hunks)
💤 Files with no reviewable changes (1)
- Deeploy/Targets/Neureka/Bindings.py
🧰 Additional context used
🧬 Code graph analysis (8)
Deeploy/Targets/Neureka/Engine.py (2)
Deeploy/DeeployTypes.py (1)
NodeMapper(1716-1872)Deeploy/Targets/Neureka/Parsers.py (6)
NeurekaRQSPWConv2DParser(140-162)NeurekaPWConv2DParser(125-137)NeurekaRQSDWConv2DParser(99-122)NeurekaDWConv2DParser(80-96)NeurekaRQSDenseConv2DParser(180-202)NeurekaDenseConv2DParser(165-177)
Deeploy/Targets/Neureka/TileConstraints/RequantHelpers.py (6)
Deeploy/DeeployTypes.py (2)
NetworkContext(564-1076)add(737-774)Deeploy/TilingExtension/MemoryConstraints.py (1)
NodeMemoryConstraint(95-167)Deeploy/TilingExtension/TilerModel.py (3)
TilerModel(34-402)addTensorDimToModel(143-157)getTensorDimVar(131-135)Deeploy/TilingExtension/TilingCodegen.py (4)
AbsoluteHyperRectangle(39-49)HyperRectangle(24-35)TilingSchedule(53-122)VariableReplacementScheme(126-158)Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py (2)
serializeTilingSolution(92-236)serializeTilingSolution(247-268)Deeploy/TilingExtension/TileConstraint.py (1)
extractBaseAddr(56-74)
Deeploy/CommonExtensions/NetworkDeployers/NetworkDeployerWrapper.py (1)
Deeploy/DeeployTypes.py (3)
CodeGenVerbosity(49-55)NetworkDeployer(3231-3596)ONNXLayer(1875-2203)
Deeploy/MemoryLevelExtension/NetworkDeployers/MemoryLevelDeployer.py (1)
Deeploy/DeeployTypes.py (3)
NetworkContext(564-1076)NetworkOptimizationPass(2263-2283)NetworkOptimizer(2286-2309)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py (5)
Deeploy/DeeployTypes.py (3)
NetworkContext(564-1076)VariableBuffer(232-391)lookup(776-808)Deeploy/Targets/Neureka/TileConstraints/RequantHelpers.py (2)
requantAddGeometricalConstraint(14-34)requantLoadSchedule(76-94)Deeploy/TilingExtension/TilerModel.py (2)
getTensorDimVar(131-135)TilerModel(34-402)Deeploy/TilingExtension/TilingCodegen.py (4)
HyperRectangle(24-35)calculateFlatOffsetInBytes(239-242)TilingSchedule(53-122)VariableReplacementScheme(126-158)Deeploy/TilingExtension/TileConstraint.py (1)
extractBaseAddr(56-74)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py (5)
Deeploy/DeeployTypes.py (3)
NetworkContext(564-1076)VariableBuffer(232-391)lookup(776-808)Deeploy/Targets/Neureka/TileConstraints/RequantHelpers.py (2)
requantAddGeometricalConstraint(14-34)requantLoadSchedule(76-94)Deeploy/TilingExtension/TilerModel.py (2)
getTensorDimVar(131-135)TilerModel(34-402)Deeploy/TilingExtension/TilingCodegen.py (4)
HyperRectangle(24-35)calculateFlatOffsetInBytes(239-242)TilingSchedule(53-122)VariableReplacementScheme(126-158)Deeploy/TilingExtension/TileConstraint.py (1)
extractBaseAddr(56-74)
Deeploy/Targets/Neureka/Tiler.py (3)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py (2)
NeurekaDenseConv2DTileConstraint(21-236)NeurekaRQSDenseConv2DTileConstraint(239-268)Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py (2)
NeurekaDWConv2DTileConstraint(21-234)NeurekaRQSDWConv2DTileConstraint(237-266)Deeploy/Targets/Neureka/TileConstraints/NeurekaPointwiseConstraint.py (2)
NeurekaPWConv2DTileConstraint(21-266)NeurekaRQSPWConv2DTileConstraint(269-298)
Deeploy/Targets/Neureka/TileConstraints/NeurekaPointwiseConstraint.py (5)
Deeploy/DeeployTypes.py (3)
NetworkContext(564-1076)VariableBuffer(232-391)lookup(776-808)Deeploy/Targets/Neureka/TileConstraints/RequantHelpers.py (2)
requantAddGeometricalConstraint(14-34)requantLoadSchedule(76-94)Deeploy/Targets/PULPOpen/TileConstraints/ConvTileConstraint.py (1)
computeInputCube(320-347)Deeploy/TilingExtension/TilingCodegen.py (4)
HyperRectangle(24-35)calculateFlatOffsetInBytes(239-242)TilingSchedule(53-122)VariableReplacementScheme(126-158)Deeploy/TilingExtension/TileConstraint.py (1)
extractBaseAddr(56-74)
🪛 Ruff (0.14.0)
Deeploy/Targets/Neureka/TileConstraints/RequantHelpers.py
48-48: Undefined name cls
(F821)
68-68: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py
227-227: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
261-261: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py
161-161: Unpacked variable BatchOffset is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
161-161: Unpacked variable HOffset is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
161-161: Unpacked variable WOffset is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
162-162: Unpacked variable BatchSize is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
229-229: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
263-263: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
Deeploy/Targets/Neureka/TileConstraints/NeurekaPointwiseConstraint.py
191-191: Unpacked variable BatchOffset is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
191-191: Unpacked variable HOffset is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
191-191: Unpacked variable WOffset is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
192-192: Unpacked variable BatchSize is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
259-259: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
293-293: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
🔇 Additional comments (7)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py (2)
210-230: LGTM! Clean SRAM-aware weight handling.The conditional weight handling properly separates the SRAM path (computing offsets per tile) from the non-SRAM path (merging base offsets and adding weight to load schedule). This aligns with the PR's goal to simplify by removing memory-aware bindings while still supporting SRAM weights.
237-266: LGTM! Requantization subclass correctly delegates.The
NeurekaRQSDWConv2DTileConstraintclass properly delegates to the base class and then applies requantization-specific constraints viarequantAddGeometricalConstraintandrequantLoadSchedule. This composition pattern is clean and maintainable.Deeploy/Targets/Neureka/TileConstraints/RequantHelpers.py (2)
14-34: LGTM! Clean constraint addition for requantization.This helper correctly adds dimension variables for mul/add buffers and constrains their channel dimensions to match the output channel. The logic is straightforward and well-commented.
76-94: LGTM! Clean load schedule generation.This helper correctly constructs HyperRectangles for mul/add buffers based on output channel offsets and sizes. The logic matches the pattern used in the tile constraint classes.
Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py (2)
54-58: LGTM! Clean conditional weight constraint.Unlike the depthwise variant, this implementation doesn't have a TODO and cleanly handles the SRAM case with a
passstatement, making the intent clear.
239-268: LGTM! Requantization integration follows established pattern.The
NeurekaRQSDenseConv2DTileConstraintclass correctly implements the same pattern as the depthwise variant, delegating to the base class and then applying requantization constraints.Deeploy/Targets/Neureka/TileConstraints/NeurekaPointwiseConstraint.py (1)
269-298: LGTM! Consistent requantization pattern.The
NeurekaRQSPWConv2DTileConstraintclass follows the same clean pattern as Dense and Depthwise variants.
lukamac
force-pushed
the
remove-memory-aware-binding
branch
2 times, most recently
from
f0d4de7 to
28faade
Compare
lukamac
changed the title
There was a problem hiding this comment.
Actionable comments posted: 0
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py (1)
86-88: Bug: kernel dims swapped (height vs width).Height should compare to dim_kernel_y, width to dim_kernel_x.
- tilerModel.addConstraint(inputHeightVar >= parseDict['dim_kernel_x']) - tilerModel.addConstraint(inputWidthVar >= parseDict['dim_kernel_y']) + tilerModel.addConstraint(inputHeightVar >= parseDict['dim_kernel_y']) + tilerModel.addConstraint(inputWidthVar >= parseDict['dim_kernel_x'])
♻️ Duplicate comments (1)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py (1)
52-58: Clarify the TODO comment or remove it.The TODO at line 54 suggests uncertainty about whether to uncomment the constraint
weightOutChannelVar == weightOutChannelVar.Max()for SRAM weights. This conditional logic is critical for handling weights in dedicated weight memory.If skipping the constraint is intentional for SRAM weights, document why. Otherwise, either uncomment it or remove the TODO.
Could you clarify the intended behavior for SRAM weights? Should the constraint be:
- Skipped entirely (current behavior)
- Uncommented to enforce full channel loading
- Something else?
🧹 Nitpick comments (7)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py (2)
210-230: Consider addingstrict=Trueto zip() call.The conditional weight handling logic correctly separates SRAM and non-SRAM paths:
- SRAM weights: pre-calculate offsets for direct access
- Non-SRAM weights: include in load schedule like other inputs
However, static analysis flags the
zip()call at line 227 for missingstrict=parameter. While the lists should have matching lengths by construction, addingstrict=Truemakes this assumption explicit and will fail fast if violated.Consider applying this change:
- for cube, load in zip(outputCubes, inputLoadSchedule): + for cube, load in zip(outputCubes, inputLoadSchedule, strict=True):
237-266: LGTM: Requantized DW conv constraint correctly extends base class.The
NeurekaRQSDWConv2DTileConstraintclass properly layers requantization support onto the base DW convolution constraints by:
- Calling base
addGeometricalConstraintthen applyingrequantAddGeometricalConstraint- Calling base
serializeTilingSolutionthen merging in requant load schedulesThis design allows requantization to be added without duplicating the base convolution tiling logic.
Consider adding
strict=Trueto thezip()call at line 261 for the same reason as mentioned in the previous comment:- } for load, rqLoad in zip(tilingSchedule.inputLoadSchedule, requantSchedule)] + } for load, rqLoad in zip(tilingSchedule.inputLoadSchedule, requantSchedule, strict=True)]Deeploy/Targets/Neureka/TileConstraints/NeurekaPointwiseConstraint.py (2)
190-195: Silence unused tuple-unpack vars.Prefix unused items with underscores to satisfy linters and reduce noise.
- (BatchOffset, HOffset, WOffset, COffset) = cube.offset - (BatchSize, HSize, WSize, CSize) = cube.dims + (_BatchOffset, _HOffset, _WOffset, _COffset) = cube.offset + (_BatchSize, HSize, WSize, CSize) = cube.dims
259-262: Use zip(..., strict=True) to guard against length mismatches
Since the project requires Python >=3.10, update your loops as follows:- for cube, load in zip(outputCubes, inputLoadSchedule): + for cube, load in zip(outputCubes, inputLoadSchedule, strict=True):- } for load, rqLoad in zip(tilingSchedule.inputLoadSchedule, requantSchedule)] + } for load, rqLoad in zip(tilingSchedule.inputLoadSchedule, requantSchedule, strict=True)]Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py (3)
65-66: Avoid hard-coded 3x3; use parsed kernel dims (optional).Use dim_kernel_y/x to generalize and reduce magic numbers.
- tilerModel.addConstraint((outputHeightVar == (effectiveHeight - (3 - 1) - 1) // strides[0] + 1)) - tilerModel.addConstraint((outputWidthVar == (effectiveWidth - (3 - 1) - 1) // strides[1] + 1)) + kH, kW = parseDict["dim_kernel_y"], parseDict["dim_kernel_x"] + tilerModel.addConstraint((outputHeightVar == (effectiveHeight - (kH - 1) - 1) // strides[0] + 1)) + tilerModel.addConstraint((outputWidthVar == (effectiveWidth - (kW - 1) - 1) // strides[1] + 1))
161-163: Silence unused tuple-unpack vars.Prefix unused items to satisfy linters.
- (BatchOffset, HOffset, WOffset, COffset) = cube.offset - (BatchSize, HSize, WSize, CSize) = cube.dims + (_BatchOffset, _HOffset, _WOffset, _COffset) = cube.offset + (_BatchSize, HSize, WSize, CSize) = cube.dims
229-231: Addstrict=Trueto zip() calls (optional defensive programming).Since the project requires Python >= 3.10 (per
pyproject.toml), usestrict=Trueunconditionally rather than conditional logic. This ensures the iterables always have matching lengths:
- Line 229:
for cube, load in zip(outputCubes, inputLoadSchedule, strict=True):- Line 263:
for load, rqLoad in zip(tilingSchedule.inputLoadSchedule, requantSchedule, strict=True)]
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (11)
CHANGELOG.md(4 hunks)Deeploy/CommonExtensions/NetworkDeployers/NetworkDeployerWrapper.py(1 hunks)Deeploy/MemoryLevelExtension/MemoryLevels.py(1 hunks)Deeploy/MemoryLevelExtension/NetworkDeployers/MemoryLevelDeployer.py(1 hunks)Deeploy/Targets/Neureka/Bindings.py(0 hunks)Deeploy/Targets/Neureka/Engine.py(1 hunks)Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py(7 hunks)Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py(6 hunks)Deeploy/Targets/Neureka/TileConstraints/NeurekaPointwiseConstraint.py(5 hunks)Deeploy/Targets/Neureka/TileConstraints/RequantHelpers.py(1 hunks)Deeploy/Targets/Neureka/Tiler.py(1 hunks)
💤 Files with no reviewable changes (1)
- Deeploy/Targets/Neureka/Bindings.py
🚧 Files skipped from review as they are similar to previous changes (2)
- Deeploy/MemoryLevelExtension/MemoryLevels.py
- Deeploy/Targets/Neureka/Engine.py
🧰 Additional context used
🧬 Code graph analysis (6)
Deeploy/Targets/Neureka/TileConstraints/RequantHelpers.py (3)
Deeploy/DeeployTypes.py (1)
NetworkContext(564-1076)Deeploy/TilingExtension/TilerModel.py (3)
TilerModel(34-402)addTensorDimToModel(143-157)getTensorDimVar(131-135)Deeploy/TilingExtension/TilingCodegen.py (2)
AbsoluteHyperRectangle(39-49)HyperRectangle(24-35)
Deeploy/MemoryLevelExtension/NetworkDeployers/MemoryLevelDeployer.py (1)
Deeploy/DeeployTypes.py (4)
NetworkContext(564-1076)NetworkOptimizationPass(2263-2283)NetworkOptimizer(2286-2309)StructBuffer(489-509)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py (4)
Deeploy/DeeployTypes.py (3)
NetworkContext(564-1076)VariableBuffer(232-391)lookup(776-808)Deeploy/Targets/Neureka/TileConstraints/RequantHelpers.py (2)
requantAddGeometricalConstraint(12-32)requantLoadSchedule(35-53)Deeploy/TilingExtension/TilerModel.py (2)
getTensorDimVar(131-135)TilerModel(34-402)Deeploy/TilingExtension/TilingCodegen.py (2)
HyperRectangle(24-35)calculateFlatOffsetInBytes(239-242)
Deeploy/Targets/Neureka/TileConstraints/NeurekaPointwiseConstraint.py (3)
Deeploy/DeeployTypes.py (3)
NetworkContext(564-1076)VariableBuffer(232-391)lookup(776-808)Deeploy/Targets/Neureka/TileConstraints/RequantHelpers.py (2)
requantAddGeometricalConstraint(12-32)requantLoadSchedule(35-53)Deeploy/TilingExtension/TilingCodegen.py (2)
HyperRectangle(24-35)calculateFlatOffsetInBytes(239-242)
Deeploy/Targets/Neureka/Tiler.py (3)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py (1)
NeurekaDenseConv2DTileConstraint(21-236)Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py (1)
NeurekaDWConv2DTileConstraint(21-234)Deeploy/Targets/Neureka/TileConstraints/NeurekaPointwiseConstraint.py (1)
NeurekaPWConv2DTileConstraint(21-266)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py (6)
Deeploy/DeeployTypes.py (3)
NetworkContext(564-1076)VariableBuffer(232-391)lookup(776-808)Deeploy/Targets/Neureka/TileConstraints/RequantHelpers.py (2)
requantAddGeometricalConstraint(12-32)requantLoadSchedule(35-53)Deeploy/TilingExtension/TilerModel.py (2)
getTensorDimVar(131-135)TilerModel(34-402)Deeploy/Targets/PULPOpen/TileConstraints/ConvTileConstraint.py (1)
computeInputCube(320-347)Deeploy/TilingExtension/TilingCodegen.py (4)
HyperRectangle(24-35)calculateFlatOffsetInBytes(239-242)TilingSchedule(53-122)VariableReplacementScheme(126-158)Deeploy/TilingExtension/TileConstraint.py (1)
extractBaseAddr(56-74)
🪛 Ruff (0.14.0)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py
227-227: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
261-261: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
Deeploy/Targets/Neureka/TileConstraints/NeurekaPointwiseConstraint.py
191-191: Unpacked variable BatchOffset is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
191-191: Unpacked variable HOffset is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
191-191: Unpacked variable WOffset is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
192-192: Unpacked variable BatchSize is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
259-259: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
293-293: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py
161-161: Unpacked variable BatchOffset is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
161-161: Unpacked variable HOffset is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
161-161: Unpacked variable WOffset is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
162-162: Unpacked variable BatchSize is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
229-229: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
263-263: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (150)
- GitHub Check: mempool-models / test-runner-mempool
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (simpleRegression, 45000, 30000, 16000, 8, L3) / test-runner-siracusa-tiled (30000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (Attention, 60000, 10000, 5000, 2500, 8, L3) / test-runner-siracusa-tiled (60000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (60000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (6000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (Attention, 60000, 10000, 5000, 2500, 8, L3) / test-runner-siracusa-tiled (5000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (12000)
- GitHub Check: siracusa-kernels-tiled-doublebuffer-L2 / test-runner-siracusa-tiled
- GitHub Check: siracusa-kernels / test-runner-siracusa
- GitHub Check: siracusa-kernels-tiled-singlebuffer-L2 / test-runner-siracusa-tiled
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (Transformer, 15000, 8, L3) / test-runner-siracusa-neureka-tiled (15000)
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (microLlama/microLlama1, 10000, 8, L3) / test-runner-siracusa-neureka-tiled (10000)
- GitHub Check: generic-kernels / test-runner-generic
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (miniMobileNet, 2000, 8, L3) / test-runner-siracusa-neureka-tiled (2000)
- GitHub Check: siracusa-neureka-models-tiled-doublebuffer-L3-wmem (microLlama/microLlama1, 10000, 8, true, L3, t... / test-runner-siracusa-neureka-tiled (10000)
- GitHub Check: mempool-models / test-runner-mempool
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (simpleRegression, 45000, 30000, 16000, 8, L3) / test-runner-siracusa-tiled (30000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (Attention, 60000, 10000, 5000, 2500, 8, L3) / test-runner-siracusa-tiled (60000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (60000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (6000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (Attention, 60000, 10000, 5000, 2500, 8, L3) / test-runner-siracusa-tiled (5000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (12000)
- GitHub Check: siracusa-kernels-tiled-doublebuffer-L2 / test-runner-siracusa-tiled
- GitHub Check: siracusa-kernels / test-runner-siracusa
- GitHub Check: siracusa-kernels-tiled-singlebuffer-L2 / test-runner-siracusa-tiled
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (Transformer, 15000, 8, L3) / test-runner-siracusa-neureka-tiled (15000)
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (microLlama/microLlama1, 10000, 8, L3) / test-runner-siracusa-neureka-tiled (10000)
- GitHub Check: generic-kernels / test-runner-generic
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (miniMobileNet, 2000, 8, L3) / test-runner-siracusa-neureka-tiled (2000)
- GitHub Check: siracusa-neureka-models-tiled-doublebuffer-L3-wmem (microLlama/microLlama1, 10000, 8, true, L3, t... / test-runner-siracusa-neureka-tiled (10000)
- GitHub Check: mempool-models / test-runner-mempool
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (simpleRegression, 45000, 30000, 16000, 8, L3) / test-runner-siracusa-tiled (30000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (Attention, 60000, 10000, 5000, 2500, 8, L3) / test-runner-siracusa-tiled (60000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (60000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (6000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (Attention, 60000, 10000, 5000, 2500, 8, L3) / test-runner-siracusa-tiled (5000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (12000)
- GitHub Check: siracusa-kernels-tiled-doublebuffer-L2 / test-runner-siracusa-tiled
- GitHub Check: siracusa-kernels / test-runner-siracusa
- GitHub Check: siracusa-kernels-tiled-singlebuffer-L2 / test-runner-siracusa-tiled
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (Transformer, 15000, 8, L3) / test-runner-siracusa-neureka-tiled (15000)
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (microLlama/microLlama1, 10000, 8, L3) / test-runner-siracusa-neureka-tiled (10000)
- GitHub Check: generic-kernels / test-runner-generic
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (miniMobileNet, 2000, 8, L3) / test-runner-siracusa-neureka-tiled (2000)
- GitHub Check: siracusa-neureka-models-tiled-doublebuffer-L3-wmem (microLlama/microLlama1, 10000, 8, true, L3, t... / test-runner-siracusa-neureka-tiled (10000)
- GitHub Check: mempool-models / test-runner-mempool
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (simpleRegression, 45000, 30000, 16000, 8, L3) / test-runner-siracusa-tiled (30000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (Attention, 60000, 10000, 5000, 2500, 8, L3) / test-runner-siracusa-tiled (60000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (60000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (6000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (Attention, 60000, 10000, 5000, 2500, 8, L3) / test-runner-siracusa-tiled (5000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (12000)
- GitHub Check: siracusa-kernels-tiled-doublebuffer-L2 / test-runner-siracusa-tiled
- GitHub Check: siracusa-kernels / test-runner-siracusa
- GitHub Check: siracusa-kernels-tiled-singlebuffer-L2 / test-runner-siracusa-tiled
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (Transformer, 15000, 8, L3) / test-runner-siracusa-neureka-tiled (15000)
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (microLlama/microLlama1, 10000, 8, L3) / test-runner-siracusa-neureka-tiled (10000)
- GitHub Check: generic-kernels / test-runner-generic
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (miniMobileNet, 2000, 8, L3) / test-runner-siracusa-neureka-tiled (2000)
- GitHub Check: siracusa-neureka-models-tiled-doublebuffer-L3-wmem (microLlama/microLlama1, 10000, 8, true, L3, t... / test-runner-siracusa-neureka-tiled (10000)
- GitHub Check: mempool-models / test-runner-mempool
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (simpleRegression, 45000, 30000, 16000, 8, L3) / test-runner-siracusa-tiled (30000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (Attention, 60000, 10000, 5000, 2500, 8, L3) / test-runner-siracusa-tiled (60000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (60000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (6000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (Attention, 60000, 10000, 5000, 2500, 8, L3) / test-runner-siracusa-tiled (5000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (12000)
- GitHub Check: siracusa-kernels-tiled-doublebuffer-L2 / test-runner-siracusa-tiled
- GitHub Check: siracusa-kernels / test-runner-siracusa
- GitHub Check: siracusa-kernels-tiled-singlebuffer-L2 / test-runner-siracusa-tiled
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (Transformer, 15000, 8, L3) / test-runner-siracusa-neureka-tiled (15000)
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (microLlama/microLlama1, 10000, 8, L3) / test-runner-siracusa-neureka-tiled (10000)
- GitHub Check: generic-kernels / test-runner-generic
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (miniMobileNet, 2000, 8, L3) / test-runner-siracusa-neureka-tiled (2000)
- GitHub Check: siracusa-neureka-models-tiled-doublebuffer-L3-wmem (microLlama/microLlama1, 10000, 8, true, L3, t... / test-runner-siracusa-neureka-tiled (10000)
- GitHub Check: mempool-models / test-runner-mempool
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (simpleRegression, 45000, 30000, 16000, 8, L3) / test-runner-siracusa-tiled (30000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (Attention, 60000, 10000, 5000, 2500, 8, L3) / test-runner-siracusa-tiled (60000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (60000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (6000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (Attention, 60000, 10000, 5000, 2500, 8, L3) / test-runner-siracusa-tiled (5000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (12000)
- GitHub Check: siracusa-kernels-tiled-doublebuffer-L2 / test-runner-siracusa-tiled
- GitHub Check: siracusa-kernels / test-runner-siracusa
- GitHub Check: siracusa-kernels-tiled-singlebuffer-L2 / test-runner-siracusa-tiled
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (Transformer, 15000, 8, L3) / test-runner-siracusa-neureka-tiled (15000)
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (microLlama/microLlama1, 10000, 8, L3) / test-runner-siracusa-neureka-tiled (10000)
- GitHub Check: generic-kernels / test-runner-generic
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (miniMobileNet, 2000, 8, L3) / test-runner-siracusa-neureka-tiled (2000)
- GitHub Check: siracusa-neureka-models-tiled-doublebuffer-L3-wmem (microLlama/microLlama1, 10000, 8, true, L3, t... / test-runner-siracusa-neureka-tiled (10000)
- GitHub Check: mempool-models / test-runner-mempool
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (simpleRegression, 45000, 30000, 16000, 8, L3) / test-runner-siracusa-tiled (30000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (Attention, 60000, 10000, 5000, 2500, 8, L3) / test-runner-siracusa-tiled (60000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (60000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (6000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (Attention, 60000, 10000, 5000, 2500, 8, L3) / test-runner-siracusa-tiled (5000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (12000)
- GitHub Check: siracusa-kernels-tiled-doublebuffer-L2 / test-runner-siracusa-tiled
- GitHub Check: siracusa-kernels / test-runner-siracusa
- GitHub Check: siracusa-kernels-tiled-singlebuffer-L2 / test-runner-siracusa-tiled
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (Transformer, 15000, 8, L3) / test-runner-siracusa-neureka-tiled (15000)
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (microLlama/microLlama1, 10000, 8, L3) / test-runner-siracusa-neureka-tiled (10000)
- GitHub Check: generic-kernels / test-runner-generic
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (miniMobileNet, 2000, 8, L3) / test-runner-siracusa-neureka-tiled (2000)
- GitHub Check: siracusa-neureka-models-tiled-doublebuffer-L3-wmem (microLlama/microLlama1, 10000, 8, true, L3, t... / test-runner-siracusa-neureka-tiled (10000)
- GitHub Check: mempool-models / test-runner-mempool
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (simpleRegression, 45000, 30000, 16000, 8, L3) / test-runner-siracusa-tiled (30000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (Attention, 60000, 10000, 5000, 2500, 8, L3) / test-runner-siracusa-tiled (60000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (60000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (6000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (Attention, 60000, 10000, 5000, 2500, 8, L3) / test-runner-siracusa-tiled (5000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (12000)
- GitHub Check: siracusa-kernels-tiled-doublebuffer-L2 / test-runner-siracusa-tiled
- GitHub Check: siracusa-kernels / test-runner-siracusa
- GitHub Check: siracusa-kernels-tiled-singlebuffer-L2 / test-runner-siracusa-tiled
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (Transformer, 15000, 8, L3) / test-runner-siracusa-neureka-tiled (15000)
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (microLlama/microLlama1, 10000, 8, L3) / test-runner-siracusa-neureka-tiled (10000)
- GitHub Check: generic-kernels / test-runner-generic
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (miniMobileNet, 2000, 8, L3) / test-runner-siracusa-neureka-tiled (2000)
- GitHub Check: siracusa-neureka-models-tiled-doublebuffer-L3-wmem (microLlama/microLlama1, 10000, 8, true, L3, t... / test-runner-siracusa-neureka-tiled (10000)
- GitHub Check: mempool-models / test-runner-mempool
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (simpleRegression, 45000, 30000, 16000, 8, L3) / test-runner-siracusa-tiled (30000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (Attention, 60000, 10000, 5000, 2500, 8, L3) / test-runner-siracusa-tiled (60000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (60000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (6000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (Attention, 60000, 10000, 5000, 2500, 8, L3) / test-runner-siracusa-tiled (5000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (12000)
- GitHub Check: siracusa-kernels-tiled-doublebuffer-L2 / test-runner-siracusa-tiled
- GitHub Check: siracusa-kernels / test-runner-siracusa
- GitHub Check: siracusa-kernels-tiled-singlebuffer-L2 / test-runner-siracusa-tiled
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (Transformer, 15000, 8, L3) / test-runner-siracusa-neureka-tiled (15000)
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (microLlama/microLlama1, 10000, 8, L3) / test-runner-siracusa-neureka-tiled (10000)
- GitHub Check: generic-kernels / test-runner-generic
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (miniMobileNet, 2000, 8, L3) / test-runner-siracusa-neureka-tiled (2000)
- GitHub Check: siracusa-neureka-models-tiled-doublebuffer-L3-wmem (microLlama/microLlama1, 10000, 8, true, L3, t... / test-runner-siracusa-neureka-tiled (10000)
- GitHub Check: mempool-models / test-runner-mempool
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (simpleRegression, 45000, 30000, 16000, 8, L3) / test-runner-siracusa-tiled (30000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (Attention, 60000, 10000, 5000, 2500, 8, L3) / test-runner-siracusa-tiled (60000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (60000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (6000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (Attention, 60000, 10000, 5000, 2500, 8, L3) / test-runner-siracusa-tiled (5000)
- GitHub Check: siracusa-models-tiled-singlebuffer-L3 (miniMobileNet, 60000, 12000, 6000, 8, L3) / test-runner-siracusa-tiled (12000)
- GitHub Check: siracusa-kernels-tiled-doublebuffer-L2 / test-runner-siracusa-tiled
- GitHub Check: siracusa-kernels / test-runner-siracusa
- GitHub Check: siracusa-kernels-tiled-singlebuffer-L2 / test-runner-siracusa-tiled
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (Transformer, 15000, 8, L3) / test-runner-siracusa-neureka-tiled (15000)
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (microLlama/microLlama1, 10000, 8, L3) / test-runner-siracusa-neureka-tiled (10000)
- GitHub Check: generic-kernels / test-runner-generic
- GitHub Check: siracusa-neureka-models-tiled-singlebuffer-L3 (miniMobileNet, 2000, 8, L3) / test-runner-siracusa-neureka-tiled (2000)
- GitHub Check: siracusa-neureka-models-tiled-doublebuffer-L3-wmem (microLlama/microLlama1, 10000, 8, true, L3, t... / test-runner-siracusa-neureka-tiled (10000)
🔇 Additional comments (5)
Deeploy/Targets/Neureka/TileConstraints/RequantHelpers.py (2)
12-32: LGTM: Requantization constraint logic is correct.The function properly adds mul and add tensor dimensions to the tiler model and enforces channel equality constraints. The implementation correctly uses the TilerModel API to retrieve dimension variables and add constraints.
35-53: LGTM: Load schedule generation is correct.The function correctly generates per-cube load schedules for mul and add buffers by:
- Extracting channel offset and size from each output cube
- Creating HyperRectangles with zeros for all dimensions except the last (channel) dimension
- Properly preserving the rank of the mul/add tensors as mentioned in the PR objectives
This implementation aligns with the NHWC layout convention used by Neureka.
Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py (1)
9-9: LGTM: Required imports added.The imports for
VariableBufferand requantization helpers (requantAddGeometricalConstraint,requantLoadSchedule) are necessary for the new memory-level checking and requantization functionality introduced in this PR.Also applies to: 12-12
Deeploy/Targets/Neureka/TileConstraints/NeurekaPointwiseConstraint.py (1)
48-52: SRAM-aware weight constraint is correct.Using weightBuffer._memoryLevel to switch channel constraint is the right replacement for Wmem variants.
Deeploy/Targets/Neureka/Tiler.py (1)
6-13: Wmem bindings removal verified as complete.Ripgrep confirms zero stale references to removed Wmem variants, MemoryAwareNodeBinding, or NodeMemoryLevelChecker across the codebase. The CHANGELOG documents the intentional removal at lines 78–80, and the import cleanup in Tiler.py is consistent with the broader cleanup. Code changes are sound.
lukamac
force-pushed
the
remove-memory-aware-binding
branch
from
e9f8bef to
b27c5ac
Compare
lukamac
changed the title
Xeratec
added
Feature
Xeratec
left a comment
Xeratec
left a comment
There was a problem hiding this comment.
Please check the open TODO comment, otherwise it look good.
Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py
Outdated
Show resolved
Hide resolved
There was a problem hiding this comment.
Actionable comments posted: 0
🧹 Nitpick comments (5)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py (2)
225-227: Addstrict=Trueto zip() for safety.The zip on line 225 iterates over
outputCubesandinputLoadSchedule, which should have matching lengths by construction. Addingstrict=Truewill catch bugs if they ever fall out of sync.Apply this diff:
- for cube, load in zip(outputCubes, inputLoadSchedule): + for cube, load in zip(outputCubes, inputLoadSchedule, strict=True): COffset, CSize = cube.offset[-1], cube.dims[-1] load['weight'] = HyperRectangle((COffset, 0, 0), (CSize, weightShape[-2], weightShape[-1]))
255-263: Addstrict=Trueto zip() for safety.The zip on line 259 pairs each input load with its corresponding requant load. Both collections are derived from the same output cubes and should have matching lengths. Adding
strict=Truewill catch any mismatch bugs.Apply this diff:
requantSchedule = requantLoadSchedule(absoluteOutputCubes, ctxt, operatorRepresentation) newInputLoadSchedule = [{ **load, **rqLoad - } for load, rqLoad in zip(tilingSchedule.inputLoadSchedule, requantSchedule)] + } for load, rqLoad in zip(tilingSchedule.inputLoadSchedule, requantSchedule, strict=True)]Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py (3)
161-162: Consider prefixing unused unpacked variables with underscore.Variables
BatchOffset,HOffset,WOffset, andBatchSizeare unpacked but never used. While explicit unpacking aids readability, prefixing unused variables with_follows Python conventions and silences linter warnings.Apply this diff:
- (BatchOffset, HOffset, WOffset, COffset) = cube.offset - (BatchSize, HSize, WSize, CSize) = cube.dims + (_BatchOffset, _HOffset, _WOffset, COffset) = cube.offset + (_BatchSize, HSize, WSize, CSize) = cube.dimsAlternatively, use direct indexing if only a few values are needed:
COffset = cube.offset[-1] HSize, WSize, CSize = cube.dims[1:]
229-231: Addstrict=Trueto zip() for safety.The zip on line 229 iterates over
outputCubesandinputLoadSchedule, which should have matching lengths by construction. Addingstrict=Truewill catch bugs if they ever fall out of sync.Apply this diff:
- for cube, load in zip(outputCubes, inputLoadSchedule): + for cube, load in zip(outputCubes, inputLoadSchedule, strict=True): COffset, CSize = cube.offset[-1], cube.dims[-1] load['weight'] = HyperRectangle((COffset, 0, 0), (CSize, weightShape[-2], weightShape[-1]))
259-263: Addstrict=Trueto zip() for safety.The zip on line 263 pairs each input load with its corresponding requant load. Both collections are derived from the same output cubes and should have matching lengths. Adding
strict=Truewill catch any mismatch bugs.Apply this diff:
requantSchedule = requantLoadSchedule(absoluteOutputCubes, ctxt, operatorRepresentation) newInputLoadSchedule = [{ **load, **rqLoad - } for load, rqLoad in zip(tilingSchedule.inputLoadSchedule, requantSchedule)] + } for load, rqLoad in zip(tilingSchedule.inputLoadSchedule, requantSchedule, strict=True)]
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (2)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py(7 hunks)Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py(6 hunks)
🧰 Additional context used
🧬 Code graph analysis (2)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py (5)
Deeploy/DeeployTypes.py (3)
NetworkContext(564-1076)VariableBuffer(232-391)lookup(776-808)Deeploy/Targets/Neureka/TileConstraints/RequantHelpers.py (2)
requantAddGeometricalConstraint(12-32)requantLoadSchedule(35-53)Deeploy/TilingExtension/TilerModel.py (2)
getTensorDimVar(131-135)TilerModel(34-402)Deeploy/TilingExtension/TilingCodegen.py (4)
HyperRectangle(24-35)calculateFlatOffsetInBytes(239-242)TilingSchedule(53-122)VariableReplacementScheme(126-158)Deeploy/TilingExtension/TileConstraint.py (1)
extractBaseAddr(56-74)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py (4)
Deeploy/DeeployTypes.py (3)
NetworkContext(564-1076)VariableBuffer(232-391)lookup(776-808)Deeploy/Targets/Neureka/TileConstraints/RequantHelpers.py (2)
requantAddGeometricalConstraint(12-32)requantLoadSchedule(35-53)Deeploy/TilingExtension/TilingCodegen.py (2)
HyperRectangle(24-35)calculateFlatOffsetInBytes(239-242)Deeploy/TilingExtension/TileConstraint.py (1)
extractBaseAddr(56-74)
🪛 Ruff (0.14.1)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py
225-225: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
259-259: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py
161-161: Unpacked variable BatchOffset is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
161-161: Unpacked variable HOffset is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
161-161: Unpacked variable WOffset is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
162-162: Unpacked variable BatchSize is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
229-229: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
263-263: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
There was a problem hiding this comment.
Actionable comments posted: 0
🧹 Nitpick comments (1)
CHANGELOG.md (1)
78-80: Move removal entries from "Changed" to "Removed" section.These entries describe deletions, not modifications. Per this changelog's own structure and standard conventions, they belong in the "Removed" section (which starts at line 93).
### Changed - ... Deployer workflow now uses `prepare(...)` instead of `generateFunction(...)`. - Removed NodeMemoryLevelChecker, MemoryAwareNodeBinding - Removed _parseNode from MemoryNetworkDeployer since we don't need the annotations before typeChecking anymore - Removed Wmem variants of bindings and tile constraints from Neureka ### FixedApply this diff to relocate them:
### Removed + - NodeMemoryLevelChecker, MemoryAwareNodeBinding + - _parseNode from MemoryNetworkDeployer since we don't need the annotations before typeChecking anymore + - Wmem variants of bindings and tile constraints from Neureka - Delete outdated and unused `.gitlab-ci.yml` file - dory_dma.c and dory_dma.h
…ve then unused memory-aware bindings
…de bindings anymore
…eady expects annotations
…ated weight memory since they should not be tiled
Xeratec
force-pushed
the
remove-memory-aware-binding
branch
from
25cd83a to
82323f8
Compare
Xeratec
left a comment
Xeratec
left a comment
There was a problem hiding this comment.
I will merge it once the CI is through.
There was a problem hiding this comment.
Actionable comments posted: 1
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (3)
Deeploy/MemoryLevelExtension/NetworkDeployers/MemoryLevelDeployer.py (2)
110-125: Fix mutable default arguments in constructors.
memoryLevelAnnotationPasses: List[NetworkOptimizationPass] = []shares state across instances. UseNoneand create a new list per call.- memoryLevelAnnotationPasses: List[NetworkOptimizationPass] = []): + memoryLevelAnnotationPasses: Optional[List[NetworkOptimizationPass]] = None): @@ - if len(memoryLevelAnnotationPasses) == 0: - memoryLevelAnnotationPasses.append(AnnotateDefaultMemoryLevel(self.Platform.memoryHierarchy)) - self.memoryLevelAnnotationOptimizer = NetworkOptimizer(memoryLevelAnnotationPasses) + passes = list(memoryLevelAnnotationPasses) if memoryLevelAnnotationPasses is not None else [] + if not passes: + passes.append(AnnotateDefaultMemoryLevel(self.Platform.memoryHierarchy)) + self.memoryLevelAnnotationOptimizer = NetworkOptimizer(passes)Apply the same pattern to:
- Lines 153-169 in
MemoryLevelAwareSignPropDeployer.__init__- Lines 197-204 in
MemoryDeployerWrapper.__init__Also applies to: 151-169, 197-204
80-106: Guard against division by zero in memory summary.
memLevel.sizecan be 0 (default) → division by zero. Treat non-positive capacity as unknown and print N/A.- if memLevel is None or getattr(memLevel, "size", None) is None: + if memLevel is None or getattr(memLevel, "size", None) is None or memLevel.size <= 0: log.info(f" {str(level):<20} {'N/A':>10} {total:10,d} " f"({staticSize:10,d} + {dynamicSize:10,d}) " f"({'N/A':>5})") else: capacity = memLevel.size log.info(f" {str(level):<20} {capacity:10,} {total:10,d} " f"({staticSize:10,d} + {dynamicSize:10,d}) " f"({total / capacity * 100:5.1f}%)")Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py (1)
31-31: Remove unuseddilation.Variable is assigned but never used.
- dilation = parseDict["dilations"]
🧹 Nitpick comments (14)
Deeploy/Targets/Neureka/TileConstraints/NeurekaPointwiseConstraint.py (5)
191-193: Silence unused tuple items (Ruff RUF059).Only COffset/CSize are used. Replace unused bindings with
_to avoid warnings and keep intent clear.- (BatchOffset, HOffset, WOffset, COffset) = cube.offset - (BatchSize, HSize, WSize, CSize) = cube.dims + (_, _, _, COffset) = cube.offset + (_, HSize, WSize, CSize) = cube.dims
187-196: Use looked-up output buffer for computeInputCube (good); minor simplification.You already looked up
outputBuffer. Reuse it to avoid a second lookup and keep shapes consistent.- InCube, padding_tuple = Conv2DTileConstraint.computeInputCube((weightH, weightW), pads, strides, weightC, - cube, outputBuffer.shape) + InCube, padding_tuple = Conv2DTileConstraint.computeInputCube( + (weightH, weightW), pads, strides, weightC, cube, outputBuffer.shape + )
48-53: SRAM gating logic looks consistent with Dense/DW; consider a shared helper.The
_memoryLevel == "WeightMemory_SRAM"pattern is duplicated across tile constraints. Extract a tiny helper (e.g.,force_full_weight_channels_if_in_sram(tilerModel, weightBufferName, outVar)) to centralize behavior and avoid drift.
289-294: Addstrict=Truewhen zipping requant load schedules.Prevents silent truncation if schedules diverge.
- } for load, rqLoad in zip(tilingSchedule.inputLoadSchedule, requantSchedule)] + } for load, rqLoad in zip(tilingSchedule.inputLoadSchedule, requantSchedule, strict=True)]If Python < 3.10, assert equal lengths before zipping.
Also applies to: 289-293
246-262: Usestrict=Trueonzip()to catch schedule-length mismatches.The project supports Python 3.10, 3.11, and 3.12, so zip()'s
strictparameter is available to raise ValueError when iterables have different lengths. Update line 259 to:for cube, load in zip(outputCubes, inputLoadSchedule, strict=True):Since the error occurs when iteration stops, this catches accidental schedule-length drifts at runtime without needing explicit length checks.
Deeploy/MemoryLevelExtension/MemoryLevels.py (1)
22-28: Harden__eq__against non-MemoryLevel comparisons.Return
NotImplementedwhenotherisn’tMemoryLevel; avoids attribute access on foreign types and follows Python equality conventions.- def __eq__(self, other): - - ret = [neighbour_name in other.neighbourNames for neighbour_name in self.neighbourNames] + def __eq__(self, other): + if not isinstance(other, MemoryLevel): + return NotImplemented + ret = [neighbour_name in other.neighbourNames for neighbour_name in self.neighbourNames] ret += [neighbour_name in self.neighbourNames for neighbour_name in other.neighbourNames] ret += [self.name == other.name, self.size == other.size] return all(ret)Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py (6)
29-33: SRAM gating pattern: consistent; consider helper extraction.Same duplication as PW/Dense. A shared helper would reduce drift across ops.
Also applies to: 52-57
73-75: Use canonical tensor name via lookup when fetching dim vars.To avoid alias/name mismatches, resolve the buffer first and pass
buffer.nametogetTensorDimVar(as done elsewhere).- inputHeightVar = tilerModel.getTensorDimVar(tensorName = parseDict['data_in'], dimIdx = 1) - inputWidthVar = tilerModel.getTensorDimVar(tensorName = parseDict['data_in'], dimIdx = 2) + _in = ctxt.lookup(parseDict['data_in']) + inputHeightVar = tilerModel.getTensorDimVar(tensorName = _in.name, dimIdx = 1) + inputWidthVar = tilerModel.getTensorDimVar(tensorName = _in.name, dimIdx = 2)
152-162: Reuse looked-upoutputBufferto compute input cube (micro‑cleanup).- outputBuffer = ctxt.lookup(varOut) - assert isinstance(outputBuffer, VariableBuffer) - for cube in outputCubes: + outputBuffer = ctxt.lookup(varOut) + assert isinstance(outputBuffer, VariableBuffer) + for cube in outputCubes: @@ - InCube, padding_tuple = Conv2DTileConstraint.computeInputCube((weightH, weightW), pads, strides, weightC, - cube, - ctxt.lookup(varOut).shape) + InCube, padding_tuple = Conv2DTileConstraint.computeInputCube( + (weightH, weightW), pads, strides, weightC, cube, outputBuffer.shape + )
156-158: Silence unused tuple items (Ruff RUF059).- (BatchOffset, HOffset, WOffset, COffset) = cube.offset - (BatchSize, HSize, WSize, CSize) = cube.dims + (_, _, _, COffset) = cube.offset + (_, HSize, WSize, CSize) = cube.dims
225-227: Addstrict=Trueto zip when merging weight tiles.- for cube, load in zip(outputCubes, inputLoadSchedule): + for cube, load in zip(outputCubes, inputLoadSchedule, strict=True): COffset, CSize = cube.offset[-1], cube.dims[-1] load['weight'] = HyperRectangle((COffset, 0, 0), (CSize, weightShape[-2], weightShape[-1]))Fallback for Python < 3.10: assert equal lengths before zipping.
255-259: Addstrict=Truewhen zipping requant schedules.- } for load, rqLoad in zip(tilingSchedule.inputLoadSchedule, requantSchedule)] + } for load, rqLoad in zip(tilingSchedule.inputLoadSchedule, requantSchedule, strict=True)]Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py (2)
223-231: Add explicitstrict=parameter tozip().The
zip()call should includestrict=Trueto catch length mismatches betweenoutputCubesandinputLoadSchedule, ensuring all cubes are paired correctly.Apply this diff:
- for cube, load in zip(outputCubes, inputLoadSchedule): + for cube, load in zip(outputCubes, inputLoadSchedule, strict=True):
259-263: Add explicitstrict=parameter tozip().The
zip()call merging load schedules should includestrict=Trueto catch any length mismatches betweentilingSchedule.inputLoadScheduleandrequantSchedule.Apply this diff:
newInputLoadSchedule = [{ **load, **rqLoad - } for load, rqLoad in zip(tilingSchedule.inputLoadSchedule, requantSchedule)] + } for load, rqLoad in zip(tilingSchedule.inputLoadSchedule, requantSchedule, strict=True)]
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (11)
CHANGELOG.md(4 hunks)Deeploy/CommonExtensions/NetworkDeployers/NetworkDeployerWrapper.py(1 hunks)Deeploy/MemoryLevelExtension/MemoryLevels.py(1 hunks)Deeploy/MemoryLevelExtension/NetworkDeployers/MemoryLevelDeployer.py(4 hunks)Deeploy/Targets/Neureka/Bindings.py(0 hunks)Deeploy/Targets/Neureka/Engine.py(1 hunks)Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py(7 hunks)Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py(6 hunks)Deeploy/Targets/Neureka/TileConstraints/NeurekaPointwiseConstraint.py(5 hunks)Deeploy/Targets/Neureka/TileConstraints/RequantHelpers.py(1 hunks)Deeploy/Targets/Neureka/Tiler.py(1 hunks)
💤 Files with no reviewable changes (1)
- Deeploy/Targets/Neureka/Bindings.py
🚧 Files skipped from review as they are similar to previous changes (3)
- Deeploy/CommonExtensions/NetworkDeployers/NetworkDeployerWrapper.py
- Deeploy/Targets/Neureka/TileConstraints/RequantHelpers.py
- CHANGELOG.md
🧰 Additional context used
🧬 Code graph analysis (6)
Deeploy/Targets/Neureka/Tiler.py (3)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py (2)
NeurekaDenseConv2DTileConstraint(21-236)NeurekaRQSDenseConv2DTileConstraint(239-268)Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py (2)
NeurekaDWConv2DTileConstraint(21-232)NeurekaRQSDWConv2DTileConstraint(235-264)Deeploy/Targets/Neureka/TileConstraints/NeurekaPointwiseConstraint.py (2)
NeurekaPWConv2DTileConstraint(21-266)NeurekaRQSPWConv2DTileConstraint(269-298)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py (4)
Deeploy/DeeployTypes.py (3)
NetworkContext(508-1020)VariableBuffer(232-360)lookup(720-752)Deeploy/Targets/Neureka/TileConstraints/RequantHelpers.py (2)
requantAddGeometricalConstraint(12-32)requantLoadSchedule(35-53)Deeploy/TilingExtension/TilingCodegen.py (4)
HyperRectangle(24-35)calculateFlatOffsetInBytes(239-242)TilingSchedule(53-122)VariableReplacementScheme(126-158)Deeploy/TilingExtension/TileConstraint.py (1)
extractBaseAddr(56-74)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py (4)
Deeploy/DeeployTypes.py (3)
NetworkContext(508-1020)VariableBuffer(232-360)lookup(720-752)Deeploy/Targets/Neureka/TileConstraints/RequantHelpers.py (2)
requantAddGeometricalConstraint(12-32)requantLoadSchedule(35-53)Deeploy/TilingExtension/TilerModel.py (2)
getTensorDimVar(131-135)TilerModel(34-402)Deeploy/TilingExtension/TilingCodegen.py (4)
HyperRectangle(24-35)calculateFlatOffsetInBytes(239-242)TilingSchedule(53-122)VariableReplacementScheme(126-158)
Deeploy/MemoryLevelExtension/NetworkDeployers/MemoryLevelDeployer.py (1)
Deeploy/DeeployTypes.py (5)
NetworkContext(508-1020)NetworkOptimizationPass(2207-2227)NetworkOptimizer(2230-2253)optimize(2184-2204)optimize(2234-2253)
Deeploy/Targets/Neureka/TileConstraints/NeurekaPointwiseConstraint.py (4)
Deeploy/DeeployTypes.py (3)
NetworkContext(508-1020)VariableBuffer(232-360)lookup(720-752)Deeploy/Targets/Neureka/TileConstraints/RequantHelpers.py (2)
requantAddGeometricalConstraint(12-32)requantLoadSchedule(35-53)Deeploy/TilingExtension/TilingCodegen.py (4)
HyperRectangle(24-35)calculateFlatOffsetInBytes(239-242)TilingSchedule(53-122)VariableReplacementScheme(126-158)Deeploy/TilingExtension/TileConstraint.py (1)
extractBaseAddr(56-74)
Deeploy/Targets/Neureka/Engine.py (2)
Deeploy/DeeployTypes.py (1)
NodeMapper(1660-1816)Deeploy/Targets/Neureka/Parsers.py (6)
NeurekaRQSPWConv2DParser(140-162)NeurekaPWConv2DParser(125-137)NeurekaRQSDWConv2DParser(99-122)NeurekaDWConv2DParser(80-96)NeurekaRQSDenseConv2DParser(180-202)NeurekaDenseConv2DParser(165-177)
🪛 Ruff (0.14.1)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDepthwiseConstraint.py
225-225: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
259-259: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py
161-161: Unpacked variable BatchOffset is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
161-161: Unpacked variable HOffset is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
161-161: Unpacked variable WOffset is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
162-162: Unpacked variable BatchSize is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
229-229: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
263-263: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
Deeploy/Targets/Neureka/TileConstraints/NeurekaPointwiseConstraint.py
191-191: Unpacked variable BatchOffset is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
191-191: Unpacked variable HOffset is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
191-191: Unpacked variable WOffset is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
192-192: Unpacked variable BatchSize is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
259-259: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
293-293: zip() without an explicit strict= parameter
Add explicit value for parameter strict=
(B905)
🔇 Additional comments (9)
Deeploy/MemoryLevelExtension/NetworkDeployers/MemoryLevelDeployer.py (1)
131-145: Annotation-before-bind flow: LGTM.Running the memory-level annotation optimizer before and after bind/codeTransform is consistent with hoisting behavior. No issues.
If performance becomes a concern, consider logging per-pass timing to confirm negligible overhead.
Also applies to: 175-189, 210-224
Deeploy/Targets/Neureka/Tiler.py (1)
6-13: Imports and tiling-ready bindings consolidation: LGTM.Wmem variants removed; non-Wmem bindings wired to corresponding tile constraints. Consistent with Engine/Bindings changes.
Please confirm Engine.py/Bindings.py no longer reference removed Wmem symbols.
Also applies to: 16-29
Deeploy/Targets/Neureka/TileConstraints/NeurekaDenseConstraint.py (5)
9-9: LGTM: New imports support runtime memory-level branching.The imports of
VariableBufferand the requant helpers properly support the refactored architecture where bindings check weight memory location at runtime rather than at parse time.Also applies to: 12-12
54-58: Well-structured runtime branching for weight memory levels.The refactored logic correctly differentiates SRAM-resident weights (which must use full output channels) from non-SRAM weights (which can be tiled by output channel). The
hasattrcheck provides appropriate fallback for weights without an explicit memory-level attribute.
157-159: Good defensive type checking.The explicit
isinstanceassertion ensuresoutputBufferis aVariableBufferbefore accessing shape and other attributes, preventing runtime errors.
212-222: Clean SRAM weight handling.The SRAM branch correctly computes per-tile weight offsets as flat byte addresses, leveraging the fact that SRAM-resident weights are fully loaded and accessed via computed offsets rather than scheduled load operations.
239-244: Well-designed requantized tile constraint class.The
NeurekaRQSDenseConv2DTileConstraintclass properly extends the base class and integrates requantization helpers by calling the base implementation and then augmenting it with requant-specific constraints.Deeploy/Targets/Neureka/Engine.py (2)
15-15: Import correctly updated to remove Wmem variants.The import statement now excludes Wmem tiling bindings, consistent with the PR objective of removing memory-aware binding variants.
18-25: Simplified mapper definitions align with refactoring goals.The removal of Wmem binding concatenation successfully moves memory-level decisions from parse-time binding selection to runtime branching within the tile constraints. This simplification eliminates the dependency on hoisted buffers during parsing, as stated in the PR objectives.
This release includes improvements to the tiling and DMA code generation, new networks and operators, improved CI workflows, migration to PyTest, and support for PyPi package releases. Note: Since the release tag references the Docker container tagged with the release tag (ghcr.io/pulp-platform/deeploy:v0.2.1), the CI will initially fail. The Deeploy Docker image must be built after the release PR is merged and the CI restarted. ### List of Pull Requests - PyPi Package Deployment + Remove Banshee Dept [#154](#154) - PyTest Migration [#144](#144) - Update submodule `pulp-nn-mixed` [#145](#145) - Improve Profiling [#138](#138) - FP32 ReduceMean operator improvement [#137](#137) - Support for RMSNorm (Pow and Sqrt operators) [#136](#136) - Demo TinyViT compatibility with tiled Siracusa [#124](#124) - TinyViT on non-tiled Siracusa [#117](#117) - Support Fully Asynchronous DMAs [#114](#114) - Disallow shape inference [#128](#128) - Remove memory-aware node bindings [#123](#123) - Fix missing const's layout transformation and refactor NCHWtoNHWC passes [#122](#122) - Fix aliasing [#125](#125) - Support for 1D Autoencoder [#98](#98) - Refactor Logging for Improved Debugging [#115](#115) - Add reuse-tool as an SPDX license header linter [#113](#113) - Bug fixes, API Cleanup and Reduce Compiler Warning on PULP [#112](#112) - Fix PULP GEMM `batch` serialization [#109](#109) - Split CI Workflows by Platform and Task, Improve Formatting and Linting Reliability [#108](#108) - Refactor tiling code generation [#105](#105) - Change order of typeMatching entries [#68](#68) - Node Mangling to avoid duplication [#93](#93) - Prepare Post v0.2.0 Release [#104](#104) - Use Docker digests instead of arch-specific tags [#106](#106) - Fix `Unsqueeze` Op. when using ONNX opset 13 or higher (from attribute to input) [#119](#119) - Fix bias hoisting in generic GEMM with no bias [#126](#126)
2 participants
Remove memory-aware node bindings because it makes the great parser refactor easier.
The memory-aware node bindings exist only for Neureka to be able to separate bindings that use the dedicated weight memory vs. those who don't.
But those bindings can simply be rewritten to check whether the weights reside in weight memory and change behavior accordingly.
By removing the memory-aware node bindings, we remove another dependency on having hoisted buffers in the middle of parsing.
The RequantHelpers are a bonus that fixes the requantization mul and add hyperrectangles to keep the rank of the original tensors.
Added
Changed
Fixed
PR Merge Checklist
develcommit and pointing todevel.CHANGELOG.mdfile has been updated.