[RFC] Dynamic Vector Support (#4200)

Introduce a new concept in the IR that allows a producer to create
dynamic vector types. An IR function can now contain global value(s)
that represent a dynamic scaling factor, for a given fixed-width
vector type. A dynamic type is then created by 'multiplying' the
corresponding global value with a fixed-width type. These new types
can be used just like the existing types and the type system has a
set of hard-coded dynamic types, such as I32X4XN, which the user
defined types map onto. The dynamic types are also used explicitly
to create dynamic stack slots, which have no set size like their
existing counterparts. New IR instructions are added to access these
new stack entities.

Currently, during codegen, the dynamic scaling factor has to be
lowered to a constant so the dynamic slots do eventually have a
compile-time known size, as do spill slots.

The current lowering for aarch64 just targets Neon, using a dynamic
scale of 1.

Copyright (c) 2022, Arm Limited.

cranelift/codegen/src/legalizer/globalvalue.rs

@@ -28,9 +28,23 @@ pub fn expand_global_value(
             readonly,
         } => load_addr(inst, func, base, offset, global_type, readonly, isa),
         ir::GlobalValueData::Symbol { tls, .. } => symbol(inst, func, global_value, isa, tls),
+        ir::GlobalValueData::DynScaleTargetConst { vector_type } => {
+            const_vector_scale(inst, func, vector_type, isa)
+        }
     }
 }
 
+fn const_vector_scale(inst: ir::Inst, func: &mut ir::Function, ty: ir::Type, isa: &dyn TargetIsa) {
+    assert!(ty.bytes() <= 16);
+
+    // Use a minimum of 128-bits for the base type.
+    let base_bytes = std::cmp::max(ty.bytes(), 16);
+    let scale = (isa.dynamic_vector_bytes(ty) / base_bytes) as i64;
+    assert!(scale > 0);
+    let pos = FuncCursor::new(func).at_inst(inst);
+    pos.func.dfg.replace(inst).iconst(isa.pointer_type(), scale);
+}
+
 /// Expand a `global_value` instruction for a vmctx global.
 fn vmctx_addr(inst: ir::Inst, func: &mut ir::Function) {
     // Get the value representing the `vmctx` argument.

cranelift/codegen/src/legalizer/mod.rs

@@ -115,6 +115,41 @@ pub fn simple_legalize(func: &mut ir::Function, cfg: &mut ControlFlowGraph, isa:
                     mflags.set_aligned();
                     pos.func.dfg.replace(inst).store(mflags, arg, addr, 0);
                 }
+                InstructionData::DynamicStackLoad {
+                    opcode: ir::Opcode::DynamicStackLoad,
+                    dynamic_stack_slot,
+                } => {
+                    let ty = pos.func.dfg.value_type(pos.func.dfg.first_result(inst));
+                    assert!(ty.is_dynamic_vector());
+                    let addr_ty = isa.pointer_type();
+
+                    let mut pos = FuncCursor::new(pos.func).at_inst(inst);
+                    pos.use_srcloc(inst);
+
+                    let addr = pos.ins().dynamic_stack_addr(addr_ty, dynamic_stack_slot);
+
+                    // Stack slots are required to be accessible and aligned.
+                    let mflags = MemFlags::trusted();
+                    pos.func.dfg.replace(inst).load(ty, mflags, addr, 0);
+                }
+                InstructionData::DynamicStackStore {
+                    opcode: ir::Opcode::DynamicStackStore,
+                    arg,
+                    dynamic_stack_slot,
+                } => {
+                    pos.use_srcloc(inst);
+                    let addr_ty = isa.pointer_type();
+                    let vector_ty = pos.func.dfg.value_type(arg);
+                    assert!(vector_ty.is_dynamic_vector());
+
+                    let addr = pos.ins().dynamic_stack_addr(addr_ty, dynamic_stack_slot);
+
+                    let mut mflags = MemFlags::new();
+                    // Stack slots are required to be accessible and aligned.
+                    mflags.set_notrap();
+                    mflags.set_aligned();
+                    pos.func.dfg.replace(inst).store(mflags, arg, addr, 0);
+                }
                 InstructionData::TableAddr {
                     opcode: ir::Opcode::TableAddr,
                     table,