fix interned string problem with -profiling

compiler/res/prog8lib/math.p8

@@ -576,6 +576,7 @@ log2_tab
         ; update the "streaming" crc16 with next byte value
         ; note: tracks the crc16 checksum in cx16.r15!
         ;       if your code uses that, it must save/restore it before calling this routine
+        ; note: this routine doesn't use a lookup table to speed up the calculation, so it is not particularly fast.
         %asm {{
             eor  cx16.r15H
             sta  cx16.r15H

compiler/src/prog8/compiler/simpleastprocessing/Instrumentation.kt

@@ -1,11 +1,13 @@
 package prog8.compiler.simpleastprocessing
 
+import prog8.code.INTERNED_STRINGS_MODULENAME
 import prog8.code.StExtSub
 import prog8.code.SymbolTable
 import prog8.code.ast.*
 import prog8.code.core.DataType
 import prog8.code.core.Encoding
 import prog8.code.core.IErrorReporter
+import prog8.code.source.SourceCode
 
 
 fun profilingInstrumentation(program: PtProgram, st: SymbolTable, errors: IErrorReporter) {
@@ -13,6 +15,19 @@ fun profilingInstrumentation(program: PtProgram, st: SymbolTable, errors: IError
 
         val instrumentationsToAdd = mutableListOf<Pair<PtNode, PtNode>>()
 
+        // make sure there is a block to put interned strings in
+        val newInternededStringBlock: Boolean
+        var internedStringsBlock: PtBlock? = program.children.firstOrNull { it is PtBlock && it.name == INTERNED_STRINGS_MODULENAME } as? PtBlock
+        if(internedStringsBlock==null) {
+            // there are no interned strings (anymore) so re-insert the block that we need
+            val options = PtBlock.Options(noSymbolPrefixing = true)
+            internedStringsBlock = PtBlock(INTERNED_STRINGS_MODULENAME, false, SourceCode.Generated(INTERNED_STRINGS_MODULENAME), options, program.position)
+            program.add(internedStringsBlock)
+            newInternededStringBlock = true
+        } else {
+            newInternededStringBlock = false
+        }
+
         walkAst(program) { node, depth ->
             when(node) {
                 is PtSub -> {
@@ -70,6 +85,13 @@ fun profilingInstrumentation(program: PtProgram, st: SymbolTable, errors: IError
         }
 
         st.resetCachedFlat()
+
+        if(newInternededStringBlock) {
+            if(internedStringsBlock.children.isEmpty()) {
+                // no interned strings, remove the block again
+                program.children.remove(internedStringsBlock)
+            }
+        }
     }
 }
 

docs/source/profiling.rst

@@ -94,7 +94,7 @@ To use subroutine profiling:
    - **Timestamp** (hex): The emulator cycle count when the call was made
    - **Routine name**: The fully qualified name of the subroutine (e.g., ``main.start``, ``galaxy.init``)
 
-   The file may contain compiler output headers before the actual CSV data starts. The data section begins after a line of dashes (``---``).
+   The file may contain compiler output headers before the actual CSV data starts. The data section begins after a line of dashes.
 
 Using the parse_profile_csv.py script
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
@@ -168,6 +168,9 @@ Here's a typical workflow for profiling a Prog8 program::
 The output will show you which routines consume the most time. For example, if a routine shows 49% of total time,
 optimizing that routine would have the biggest impact on overall performance.
 
+Make sure to eventually get rid of the profiling code and recompile the program normally without the ``-profiling`` option,
+because the logic takes up space and does slow down the actual program a bit.
+
 Tips for effective profiling
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
@@ -177,6 +180,7 @@ Tips for effective profiling
 - A routine with high average time per call may need algorithmic improvements
 - Use the flame graph to quickly identify hot paths in the call stack
 - Compare profiles before and after optimizations to measure improvement
+= **Recompile the program WITHOUT profiling mode after you're done** -- because the profiling logic takes up space and slows down the actual program
 
 Limitations
 ^^^^^^^^^^^