diff --git a/.idea/kotlinc.xml b/.idea/kotlinc.xml
index 51f6d408e..f44d15ce9 100644
--- a/.idea/kotlinc.xml
+++ b/.idea/kotlinc.xml
@@ -4,6 +4,6 @@
     <option name="jvmTarget" value="11" />
   </component>
   <component name="KotlinJpsPluginSettings">
-    <option name="version" value="1.8.20" />
+    <option name="version" value="1.8.0-release-345" />
   </component>
 </project>
\ No newline at end of file
diff --git a/codeCore/src/prog8/code/core/MemoryRegions.kt b/codeCore/src/prog8/code/core/MemoryRegions.kt
index dab32a398..2726eee4b 100644
--- a/codeCore/src/prog8/code/core/MemoryRegions.kt
+++ b/codeCore/src/prog8/code/core/MemoryRegions.kt
@@ -131,6 +131,7 @@ abstract class Zeropage(options: CompilationOptions): MemoryAllocator(options) {
 }
 
 
+// TODO: this class is not yet used
 class GoldenRam(options: CompilationOptions, val region: UIntRange): MemoryAllocator(options) {
     private var nextLocation: UInt = region.first
 
diff --git a/codeCore/src/prog8/code/target/cx16/CX16MachineDefinition.kt b/codeCore/src/prog8/code/target/cx16/CX16MachineDefinition.kt
index 2ed4ea7c2..34b63c252 100644
--- a/codeCore/src/prog8/code/target/cx16/CX16MachineDefinition.kt
+++ b/codeCore/src/prog8/code/target/cx16/CX16MachineDefinition.kt
@@ -64,7 +64,7 @@ class CX16MachineDefinition: IMachineDefinition {
 
     override fun initializeMemoryAreas(compilerOptions: CompilationOptions) {
         zeropage = CX16Zeropage(compilerOptions)
-        golden = GoldenRam(compilerOptions, 0x0600u until 0x0800u)
+        golden = GoldenRam(compilerOptions, 0x0400u until ESTACK_LO)
     }
 
 }
diff --git a/codeGenIntermediate/src/prog8/codegen/intermediate/IRCodeGen.kt b/codeGenIntermediate/src/prog8/codegen/intermediate/IRCodeGen.kt
index ac3430db8..3051c496f 100644
--- a/codeGenIntermediate/src/prog8/codegen/intermediate/IRCodeGen.kt
+++ b/codeGenIntermediate/src/prog8/codegen/intermediate/IRCodeGen.kt
@@ -1465,9 +1465,12 @@ class IRCodeGen(
                         }
                     } else {
                         // regular asmsub
-                        val assemblyChild = child.children.single() as PtInlineAssembly
+                        if(child.children.map { (it as PtInlineAssembly).isIR }.toSet().size>1)
+                            errors.err("asmsub mixes IR and non-IR assembly code (could be compiler-generated)", child.position)
+                        val asmblocks = child.children.map { (it as PtInlineAssembly).assembly.trimEnd() }
+                        val assembly = asmblocks.joinToString("\n")
                         val asmChunk = IRInlineAsmChunk(
-                            child.name, assemblyChild.assembly, assemblyChild.isIR, null
+                            child.name, assembly, (child.children[0] as PtInlineAssembly).isIR , null
                         )
                         irBlock += IRAsmSubroutine(
                             child.name,
diff --git a/codeGenIntermediate/test/TestVmCodeGen.kt b/codeGenIntermediate/test/TestVmCodeGen.kt
index cac64f2a4..c75e06057 100644
--- a/codeGenIntermediate/test/TestVmCodeGen.kt
+++ b/codeGenIntermediate/test/TestVmCodeGen.kt
@@ -8,6 +8,7 @@ import prog8.code.target.VMTarget
 import prog8.codegen.vm.VmAssemblyProgram
 import prog8.codegen.vm.VmCodeGen
 import prog8.intermediate.IRSubroutine
+import prog8.intermediate.Opcode
 
 class TestVmCodeGen: FunSpec({
 
@@ -440,4 +441,33 @@ class TestVmCodeGen: FunSpec({
         irChunks.size shouldBe 1
     }
 
+    test("romsub allowed in codegen") {
+//main {
+//    romsub $5000 = routine()
+//
+//    sub start()  {
+//        routine()
+//    }
+//}
+        val codegen = VmCodeGen()
+        val program = PtProgram("test", DummyMemsizer, DummyStringEncoder)
+        val block = PtBlock("main", null, false, false, PtBlock.BlockAlignment.NONE, SourceCode.Generated("test"), Position.DUMMY)
+        val romsub = PtAsmSub("routine", 0x5000u, emptySet(), emptyList(), emptyList(), false, Position.DUMMY)
+        block.add(romsub)
+        val sub = PtSub("start", emptyList(), null, Position.DUMMY)
+        val call = PtFunctionCall("main.routine", true, DataType.UNDEFINED, Position.DUMMY)
+        sub.add(call)
+        block.add(sub)
+        program.add(block)
+
+        val options = getTestOptions()
+        val st = SymbolTableMaker(program, options).make()
+        val errors = ErrorReporterForTests()
+        val result = codegen.generate(program, st, options, errors) as VmAssemblyProgram
+        val irChunks = (result.irProgram.blocks.first().children.single() as IRSubroutine).chunks
+        irChunks.size shouldBe 1
+        val callInstr = irChunks.single().instructions.single()
+        callInstr.opcode shouldBe Opcode.CALL
+        callInstr.value shouldBe 0x5000
+    }
 })
\ No newline at end of file
diff --git a/compiler/src/prog8/compiler/astprocessing/AstChecker.kt b/compiler/src/prog8/compiler/astprocessing/AstChecker.kt
index 1d4ee27eb..55f3a708c 100644
--- a/compiler/src/prog8/compiler/astprocessing/AstChecker.kt
+++ b/compiler/src/prog8/compiler/astprocessing/AstChecker.kt
@@ -326,8 +326,6 @@ internal class AstChecker(private val program: Program,
             err("subroutines can only be defined in the scope of a block or within another subroutine")
 
         if(subroutine.isAsmSubroutine) {
-            if(compilerOptions.compTarget.name==VMTarget.NAME)
-                err("cannot use asmsub for vm target, use regular subs")
             if(subroutine.asmParameterRegisters.size != subroutine.parameters.size)
                 err("number of asm parameter registers is not the isSameAs as number of parameters")
             if(subroutine.asmReturnvaluesRegisters.size != subroutine.returntypes.size)
diff --git a/compiler/src/prog8/compiler/astprocessing/IntermediateAstMaker.kt b/compiler/src/prog8/compiler/astprocessing/IntermediateAstMaker.kt
index 152ef229e..ca7bfe266 100644
--- a/compiler/src/prog8/compiler/astprocessing/IntermediateAstMaker.kt
+++ b/compiler/src/prog8/compiler/astprocessing/IntermediateAstMaker.kt
@@ -355,9 +355,9 @@ class IntermediateAstMaker(private val program: Program, private val options: Co
             for (asm in srcSub.statements) {
                 asm as InlineAssembly
                 if(asm.isIR)
-                    combinedIrAsm += asm.assembly + "\n"
+                    combinedIrAsm += asm.assembly.trimEnd() + "\n"
                 else
-                    combinedTrueAsm += asm.assembly + "\n"
+                    combinedTrueAsm += asm.assembly.trimEnd() + "\n"
             }
 
             if(combinedTrueAsm.isNotEmpty()) {
diff --git a/compiler/test/vm/TestCompilerVirtual.kt b/compiler/test/vm/TestCompilerVirtual.kt
index 4cdb4d93b..11b83ca78 100644
--- a/compiler/test/vm/TestCompilerVirtual.kt
+++ b/compiler/test/vm/TestCompilerVirtual.kt
@@ -14,7 +14,6 @@ import prog8.intermediate.IRFileReader
 import prog8.intermediate.IRSubroutine
 import prog8.intermediate.Opcode
 import prog8.vm.VmRunner
-import prog8tests.helpers.ErrorReporterForTests
 import prog8tests.helpers.compileText
 import kotlin.io.path.readText
 
@@ -219,54 +218,6 @@ main {
         }
     }
 
-    test("asmsub for virtual target not supported") {
-        val src = """
-main {
-  sub start() {
-    void test(42)
-  }
-
-  asmsub test(ubyte xx @A) -> ubyte @Y {
-    %asm {{
-        lda #99
-        tay
-        rts
-    }}
-  }
-}"""
-        val othertarget = Cx16Target()
-        compileText(othertarget, true, src, writeAssembly = true) shouldNotBe null
-
-        val target = VMTarget()
-        val errors = ErrorReporterForTests()
-        compileText(target, false, src, writeAssembly = false, errors = errors) shouldBe null
-        errors.errors.size shouldBe 1
-        errors.errors[0] shouldContain "cannot use asmsub for vm target"
-    }
-
-    test("asmsub for virtual target not supported even with IR") {
-        val src = """
-main {
-  sub start() {
-    void test(42)
-  }
-
-  asmsub test(ubyte xx @A) -> ubyte @Y {
-    %ir {{
-        return
-    }}
-  }
-}"""
-        val othertarget = Cx16Target()
-        compileText(othertarget, true, src, writeAssembly = true) shouldNotBe null
-
-        val target = VMTarget()
-        val errors = ErrorReporterForTests()
-        compileText(target, false, src, writeAssembly = false, errors = errors) shouldBe null
-        errors.errors.size shouldBe 1
-        errors.errors[0] shouldContain "cannot use asmsub for vm target"
-    }
-
     test("inline asm for virtual target should be IR") {
         val src = """
 main {
diff --git a/docs/source/todo.rst b/docs/source/todo.rst
index 1c58e74c0..d3373db59 100644
--- a/docs/source/todo.rst
+++ b/docs/source/todo.rst
@@ -40,6 +40,9 @@ Compiler:
 - ir: peephole opt: reuse registers in chunks (but keep result registers in mind that pass values out! and don't renumber registers above SyscallRegisterBase!)
 - ir: add more optimizations in IRPeepholeOptimizer
 - ir: for expressions with array indexes that occur multiple times, can we avoid loading them into new virtualregs everytime and just reuse a single virtualreg as indexer? (simple form of common subexpression elimination)
+- ir: write arguments for subroutine calls in IR differently?
+  don't load them explicitly into the variables but use a new special instruction to do it transparently.
+  (vm should take care of it based on the subroutine's parameter list)
 - PtAst/IR: more complex common subexpression eliminations
 - vm: somehow be able to load a label address as value? (VmProgramLoader) this may require storing the program as bytecodes in actual memory though...
 - 6502 codegen: see if we can let for loops skip the loop if startvar>endvar, without adding a lot of code size/duplicating the loop condition.
diff --git a/intermediate/src/prog8/intermediate/IRProgram.kt b/intermediate/src/prog8/intermediate/IRProgram.kt
index 2bd682483..47b33c650 100644
--- a/intermediate/src/prog8/intermediate/IRProgram.kt
+++ b/intermediate/src/prog8/intermediate/IRProgram.kt
@@ -414,11 +414,14 @@ private fun registersUsedInAssembly(isIR: Boolean, assembly: String): RegistersU
 
     if(isIR) {
         assembly.lineSequence().forEach { line ->
-            val result = parseIRCodeLine(line.trim(), null, mutableMapOf())
-            result.fold(
-                ifLeft = { it.addUsedRegistersCounts(readRegsCounts, writeRegsCounts,readFpRegsCounts, writeFpRegsCounts, regsTypes) },
-                ifRight = { /* labels can be skipped */ }
-            )
+            val t = line.trim()
+            if(t.isNotEmpty()) {
+                val result = parseIRCodeLine(t, null, mutableMapOf())
+                result.fold(
+                    ifLeft = { it.addUsedRegistersCounts(readRegsCounts, writeRegsCounts,readFpRegsCounts, writeFpRegsCounts, regsTypes) },
+                    ifRight = { /* labels can be skipped */ }
+                )
+            }
         }
     }
     return RegistersUsed(readRegsCounts, writeRegsCounts, readFpRegsCounts, writeFpRegsCounts, regsTypes)
diff --git a/virtualmachine/src/prog8/vm/VirtualMachine.kt b/virtualmachine/src/prog8/vm/VirtualMachine.kt
index 2e0771a8c..58e9685dd 100644
--- a/virtualmachine/src/prog8/vm/VirtualMachine.kt
+++ b/virtualmachine/src/prog8/vm/VirtualMachine.kt
@@ -1,6 +1,7 @@
 package prog8.vm
 
 import prog8.code.StMemVar
+import prog8.code.core.toHex
 import prog8.code.target.virtual.IVirtualMachineRunner
 import prog8.intermediate.*
 import java.awt.Color
@@ -147,7 +148,12 @@ class VirtualMachine(irProgram: IRProgram) {
                 pcChunk = target
                 pcIndex = 0
             }
-            null -> throw IllegalArgumentException("no branchtarget in $i")
+            null -> {
+                if(i.value!=null)
+                    throw IllegalArgumentException("vm program can't jump to system memory address (${i.opcode} ${i.value!!.toHex()})")
+                else
+                    throw IllegalArgumentException("no branchtarget in $i")
+            }
             else -> throw IllegalArgumentException("VM can't execute code in a non-codechunk: $target")
         }
     }
diff --git a/virtualmachine/src/prog8/vm/VmProgramLoader.kt b/virtualmachine/src/prog8/vm/VmProgramLoader.kt
index b7d2b55c0..205a284e2 100644
--- a/virtualmachine/src/prog8/vm/VmProgramLoader.kt
+++ b/virtualmachine/src/prog8/vm/VmProgramLoader.kt
@@ -69,7 +69,7 @@ class VmProgramLoader {
 
         programChunks.forEach {
             it.instructions.forEach { ins ->
-                if (ins.labelSymbol != null && ins.opcode !in OpcodesThatBranch)
+                if (ins.labelSymbol != null && ins.opcode !in OpcodesThatBranch && ins.opcode !in OpcodesForCpuRegisters)
                     require(ins.value != null) { "instruction with labelSymbol for a var should have value set to var's memory address" }
             }
         }
@@ -153,8 +153,11 @@ class VmProgramLoader {
                     // placeholder is not a variable, so it must be a label of a code chunk instead
                     val target: IRCodeChunk? = chunks.firstOrNull { it.label==label }
                     val opcode = chunk.instructions[line].opcode
-                    if(target==null)
-                        throw IRParseException("placeholder not found in variables nor labels: $label")
+                    if(target==null) {
+                        // exception allowed: storecpu/loadcpu instructions that refer to CPU registers
+                        if(opcode !in OpcodesForCpuRegisters)
+                            throw IRParseException("placeholder not found in variables nor labels: $label")
+                    }
                     else if(opcode in OpcodesThatBranch) {
                         chunk.instructions[line] = chunk.instructions[line].copy(branchTarget = target, value = null)
                     } else if(opcode in OpcodesWithMemoryAddressAsValue) {