Fix/tiling stack scoping and tiling information corruption (#177)

* [Deeploy PR] put the tiling information into layer code as well

* [Deeploy PR] Fix the tiling information corruption. Add bracket before and after L3 code for each layer to reduce stack usage

- Previosuly the tiling information was corrupted after each run,  because the generated code put the next element in the tiling information array to current location. So after one runnetwork run, we will see that the last element in the tiling array goes to the first location, corrupting the tiling information. The fix in tilingvariablereplacement fix this by pointer the reference to the new location, instead of assigning value into the reference.

- The C code generated by Deeploy has been causing big stack usage. This is because all variables defined in Runnetwork lives in stack, including all the tiling pointers and call arguements. By adding bracket before and after each layer in RunNetwork, this makes the call args only live for one layer, thus significantly reduce the stack usage. The tiling pointers still live in stack, they need to be moved as well. But this require more changes

* Update CHANGELOG.md

Author: pauloohaha

CHANGELOG.md

@@ -46,6 +46,8 @@ This file contains the changelog for the Deeploy project. The changelog is divid
 - im2col buffer size in Conv1d template
 - Fix missing dependency in pre-commit-config
 - Fix test paths in Deeploy 101 tutorial
+- Fix tiling variable replacement corrupting static arrays by changing pointer update from value copy to address reassignment
+- Reduce RunNetwork stack usage by scoping per-layer variables with braces and moving tileIdxPtr allocation into per-layer execution blocks
 
 ### Removed
 - `testDMA.py` was an old test; we now have `test_dmas.py` instead.

Deeploy/DeeployTypes.py

@@ -2800,7 +2800,8 @@ def generateInferenceCode(self) -> str:
             self.ctxt, code = node.generate(self.ctxt)
 
             sections = reduce(lambda a, b: a + b, code, [])
-            callStack += reduce(lambda a, b: a + b, sections, "")
+            layerCode = reduce(lambda a, b: a + b, sections, "")
+            callStack += "{\n" + layerCode + "\n}\n"
 
         return callStack
 

Deeploy/TilingExtension/CodeTransformationPasses/TilingHoistingMixIn.py

@@ -92,11 +92,11 @@ def _hoistTileNumAndIdxPtr(self, ctxt: NetworkContext,
         # LMACAN: Intentionally don't annotate memory level so it gets allocated
         # outside of the tiling loops
 
-        tileIdxPtr.allocTemplate = NodeTemplate("")
-        tileIdxPtr.deallocTemplate = NodeTemplate("")
-        tileIdxPtr.initTemplate = NodeTemplate("""
+        tileIdxPtr.allocTemplate = NodeTemplate("""
         ${type.referencedType.typeName} bu_${name} = 0;
         ${type.referencedType.typeName}* ${name} = &bu_${name};""")
+        tileIdxPtr.deallocTemplate = NodeTemplate("")
+        tileIdxPtr.initTemplate = NodeTemplate("")
 
         return (tileNum, tileIdxPtr)
 

Deeploy/TilingExtension/CodeTransformationPasses/TilingVariableReplacement.py

@@ -177,7 +177,7 @@ class TilingVariableReplacementUpdate(CodeTransformationPass, IntrospectiveCodeT
 
     _updateReferenceTemplate = NodeTemplate("""
     // UPDATE VARIABLE ${reference}
-    *${reference} = ${baseReference}[${tileIdxVar}];
+    ${reference} = &${baseReference}[${tileIdxVar}];
     """)
 
     def __init__(self, targetMemLevel: str, tileIdxVar: str = "TILING_I"):