AArch64: Migrate calls and returns to ISLE. (#4788)

cranelift/codegen/src/isa/aarch64/abi.rs

@@ -225,12 +225,12 @@ impl ABIMachineSpec for AArch64MachineDeps {
                         slots: smallvec![
                             ABIArgSlot::Reg {
                                 reg: lower_reg.to_real_reg().unwrap(),
-                                ty: param.value_type,
+                                ty: reg_types[0],
                                 extension: param.extension,
                             },
                             ABIArgSlot::Reg {
                                 reg: upper_reg.to_real_reg().unwrap(),
-                                ty: param.value_type,
+                                ty: reg_types[1],
                                 extension: param.extension,
                             },
                         ],

cranelift/codegen/src/isa/aarch64/inst.isle

@@ -2818,3 +2818,11 @@
       (let ((dst WritableReg (temp_writable_reg $I8X16))
             (_ Unit (emit (MInst.IntToFpu op dst src))))
        dst))
+
+;;;; Helpers for Emitting Calls ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+(decl gen_call (SigRef ExternalName RelocDistance ValueSlice) InstOutput)
+(extern constructor gen_call gen_call)
+
+(decl gen_call_indirect (SigRef Value ValueSlice) InstOutput)
+(extern constructor gen_call_indirect gen_call_indirect)

cranelift/codegen/src/isa/aarch64/lower.isle

@@ -2016,3 +2016,17 @@
 
 (rule (lower (get_return_address))
       (aarch64_link))
+
+;;;; Rules for calls ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+(rule (lower (call (func_ref_data sig_ref extname dist) inputs))
+      (gen_call sig_ref extname dist inputs))
+
+(rule (lower (call_indirect sig_ref val inputs))
+      (gen_call_indirect sig_ref val inputs))
+
+;;;; Rules for `return` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+;; N.B.: the Ret itself is generated by the ABI.
+(rule (lower (return args))
+      (lower_return (range 0 (value_slice_len args)) args))

cranelift/codegen/src/isa/aarch64/lower/isle.rs

@@ -2,6 +2,7 @@
 
 // Pull in the ISLE generated code.
 pub mod generated_code;
+use generated_code::Context;
 
 // Types that the generated ISLE code uses via `use super::*`.
 use super::{
@@ -14,6 +15,7 @@ use super::{
 use crate::isa::aarch64::inst::{FPULeftShiftImm, FPURightShiftImm};
 use crate::isa::aarch64::lower::{lower_address, lower_splat_const};
 use crate::isa::aarch64::settings::Flags as IsaFlags;
+use crate::machinst::valueregs;
 use crate::machinst::{isle::*, InputSourceInst};
 use crate::settings::Flags;
 use crate::{
@@ -22,10 +24,11 @@ use crate::{
         immediates::*, types::*, AtomicRmwOp, ExternalName, Inst, InstructionData, MemFlags,
         TrapCode, Value, ValueList,
     },
+    isa::aarch64::abi::{AArch64Caller, AArch64MachineDeps},
     isa::aarch64::inst::args::{ShiftOp, ShiftOpShiftImm},
     isa::aarch64::lower::{writable_vreg, writable_xreg, xreg},
     isa::unwind::UnwindInst,
-    machinst::{ty_bits, InsnOutput, Lower, VCodeConstant, VCodeConstantData},
+    machinst::{ty_bits, InsnOutput, Lower, MachInst, VCodeConstant, VCodeConstantData},
 };
 use regalloc2::PReg;
 use std::boxed::Box;
@@ -69,8 +72,13 @@ pub struct SinkableAtomicLoad {
     atomic_addr: Value,
 }
 
-impl generated_code::Context for IsleContext<'_, '_, MInst, Flags, IsaFlags, 6> {
+impl IsleContext<'_, '_, MInst, Flags, IsaFlags, 6> {
+    isle_prelude_method_helpers!(AArch64Caller);
+}
+
+impl Context for IsleContext<'_, '_, MInst, Flags, IsaFlags, 6> {
     isle_prelude_methods!();
+    isle_prelude_caller_methods!(AArch64MachineDeps, AArch64Caller);
 
     fn sign_return_address_disabled(&mut self) -> Option<()> {
         if self.isa_flags.sign_return_address() {

cranelift/codegen/src/isa/aarch64/lower_inst.rs

@@ -5,7 +5,6 @@ use crate::binemit::CodeOffset;
 use crate::ir::types::*;
 use crate::ir::Inst as IRInst;
 use crate::ir::{InstructionData, Opcode};
-use crate::isa::aarch64::abi::*;
 use crate::isa::aarch64::inst::*;
 use crate::isa::aarch64::settings as aarch64_settings;
 use crate::machinst::lower::*;
@@ -469,29 +468,7 @@ pub(crate) fn lower_insn_to_regs(
             }
         }
 
-        Opcode::Return => {
-            for (i, input) in inputs.iter().enumerate() {
-                // N.B.: according to the AArch64 ABI, the top bits of a register
-                // (above the bits for the value's type) are undefined, so we
-                // need not extend the return values.
-                let src_regs = put_input_in_regs(ctx, *input);
-                let retval_regs = ctx.retval(i);
-
-                assert_eq!(src_regs.len(), retval_regs.len());
-                let ty = ctx.input_ty(insn, i);
-                let (_, tys) = Inst::rc_for_type(ty)?;
-
-                src_regs
-                    .regs()
-                    .iter()
-                    .zip(retval_regs.regs().iter())
-                    .zip(tys.iter())
-                    .for_each(|((&src, &dst), &ty)| {
-                        ctx.emit(Inst::gen_move(dst, src, ty));
-                    });
-            }
-            // N.B.: the Ret itself is generated by the ABI.
-        }
+        Opcode::Return => implemented_in_isle(ctx),
 
         Opcode::Ifcmp | Opcode::Ffcmp => {
             // An Ifcmp/Ffcmp must always be seen as a use of a brif/brff or trueif/trueff
@@ -577,52 +554,7 @@ pub(crate) fn lower_insn_to_regs(
 
         Opcode::SymbolValue => implemented_in_isle(ctx),
 
-        Opcode::Call | Opcode::CallIndirect => {
-            let caller_conv = ctx.abi().call_conv();
-            let (mut abi, inputs) = match op {
-                Opcode::Call => {
-                    let (extname, dist) = ctx.call_target(insn).unwrap();
-                    let extname = extname.clone();
-                    let sig = ctx.call_sig(insn).unwrap();
-                    assert!(inputs.len() == sig.params.len());
-                    assert!(outputs.len() == sig.returns.len());
-                    (
-                        AArch64Caller::from_func(sig, &extname, dist, caller_conv, flags)?,
-                        &inputs[..],
-                    )
-                }
-                Opcode::CallIndirect => {
-                    let ptr = put_input_in_reg(ctx, inputs[0], NarrowValueMode::ZeroExtend64);
-                    let sig = ctx.call_sig(insn).unwrap();
-                    assert!(inputs.len() - 1 == sig.params.len());
-                    assert!(outputs.len() == sig.returns.len());
-                    (
-                        AArch64Caller::from_ptr(sig, ptr, op, caller_conv, flags)?,
-                        &inputs[1..],
-                    )
-                }
-                _ => unreachable!(),
-            };
-
-            abi.emit_stack_pre_adjust(ctx);
-            assert!(inputs.len() == abi.num_args());
-            let mut arg_regs = vec![];
-            for input in inputs {
-                arg_regs.push(put_input_in_regs(ctx, *input))
-            }
-            for (i, arg_regs) in arg_regs.iter().enumerate() {
-                abi.emit_copy_regs_to_buffer(ctx, i, *arg_regs);
-            }
-            for (i, arg_regs) in arg_regs.iter().enumerate() {
-                abi.emit_copy_regs_to_arg(ctx, i, *arg_regs);
-            }
-            abi.emit_call(ctx);
-            for (i, output) in outputs.iter().enumerate() {
-                let retval_regs = get_output_reg(ctx, *output);
-                abi.emit_copy_retval_to_regs(ctx, i, retval_regs);
-            }
-            abi.emit_stack_post_adjust(ctx);
-        }
+        Opcode::Call | Opcode::CallIndirect => implemented_in_isle(ctx),
 
         Opcode::GetPinnedReg => {
             let rd = get_output_reg(ctx, outputs[0]).only_reg().unwrap();

cranelift/codegen/src/isa/s390x/inst.isle

@@ -2940,8 +2940,8 @@
 (rule (abi_ext_ty (ArgumentExtension.Sext) _) $I64)
 
 ;; Copy a return value to a set of registers.
-(decl copy_to_regs (WritableValueRegs Value) Unit)
-(rule (copy_to_regs (only_writable_reg reg) val @ (value_type ty))
+(decl s390x_copy_to_regs (WritableValueRegs Value) Unit)
+(rule (s390x_copy_to_regs (only_writable_reg reg) val @ (value_type ty))
       (emit_mov ty reg val))
 
 

cranelift/codegen/src/isa/s390x/lower.isle

@@ -4047,13 +4047,13 @@
 ;;;; Rules for `return` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 
 (rule (lower (return args))
-      (lower_return (range 0 (value_slice_len args)) args))
+      (s390x_lower_return (range 0 (value_slice_len args)) args))
 
-(decl lower_return (Range ValueSlice) InstOutput)
-(rule (lower_return (range_empty) _) (output_none))
-(rule (lower_return (range_unwrap head tail) args)
-      (let ((_ Unit (copy_to_regs (retval head) (value_slice_get args head))))
-         (lower_return tail args)))
+(decl s390x_lower_return (Range ValueSlice) InstOutput)
+(rule (s390x_lower_return (range_empty) _) (output_none))
+(rule (s390x_lower_return (range_unwrap head tail) args)
+      (let ((_ Unit (s390x_copy_to_regs (retval head) (value_slice_get args head))))
+         (s390x_lower_return tail args)))
 
 
 ;;;; Rules for `call` and `call_indirect` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;

cranelift/codegen/src/isa/s390x/lower/isle.rs

@@ -22,7 +22,7 @@ use crate::{
     isa::unwind::UnwindInst,
     isa::CallConv,
     machinst::abi_impl::ABIMachineSpec,
-    machinst::{InsnOutput, Lower, VCodeConstant, VCodeConstantData},
+    machinst::{InsnOutput, Lower, MachInst, VCodeConstant, VCodeConstantData},
 };
 use regalloc2::PReg;
 use smallvec::{smallvec, SmallVec};

cranelift/codegen/src/isa/x64/inst.isle

@@ -1596,31 +1596,6 @@
 
 ;;;; Helpers for Emitting Loads ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 
-;; Generate a move between two registers.
-(decl gen_move (Type WritableReg Reg) MInst)
-(extern constructor gen_move gen_move)
-
-;; Copy a return value to a set of registers.
-(decl copy_to_regs (WritableValueRegs Value) Unit)
-(rule (copy_to_regs dsts val @ (value_type ty))
-      (let ((srcs ValueRegs (put_in_regs val)))
-        (copy_to_regs_range ty (value_regs_range srcs) dsts srcs)))
-
-;; Helper for `copy_to_regs` that uses a range to index into the reg/value
-;; vectors. Fails for the empty range.
-(decl copy_to_regs_range (Type Range WritableValueRegs ValueRegs) Unit)
-
-(rule (copy_to_regs_range ty (range_singleton idx) dsts srcs)
-      (let ((dst WritableReg (writable_regs_get dsts idx))
-            (src Reg (value_regs_get srcs idx)))
-        (emit (gen_move ty dst src))))
-
-(rule (copy_to_regs_range ty (range_unwrap head tail) dsts srcs)
-      (let ((dst WritableReg (writable_regs_get dsts head))
-            (src Reg (value_regs_get srcs head))
-            (_ Unit (emit (gen_move ty dst src))))
-        (copy_to_regs_range ty tail dsts srcs)))
-
 ;; Helper for constructing a LoadExtName instruction.
 (decl load_ext_name (ExternalName i64) Reg)
 (rule (load_ext_name extname offset)

cranelift/codegen/src/isa/x64/lower.isle

@@ -1441,13 +1441,6 @@
 (rule (lower (return args))
       (lower_return (range 0 (value_slice_len args)) args))
 
-(decl lower_return (Range ValueSlice) InstOutput)
-(rule (lower_return (range_empty) _) (output_none))
-(rule (lower_return (range_unwrap head tail) args)
-      (let ((_ Unit (copy_to_regs (retval head) (value_slice_get args head))))
-        (lower_return tail args)))
-
-
 ;;;; Rules for `icmp` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 
 (rule (lower (icmp cc a @ (value_type (fits_in_64 ty)) b))

cranelift/codegen/src/isa/x64/lower/isle.rs

@@ -92,6 +92,7 @@ pub(crate) fn lower_branch(
 
 impl Context for IsleContext<'_, '_, MInst, Flags, IsaFlags, 6> {
     isle_prelude_methods!();
+    isle_prelude_caller_methods!(X64ABIMachineSpec, X64Caller);
 
     #[inline]
     fn operand_size_of_type_32_64(&mut self, ty: Type) -> OperandSize {
@@ -708,54 +709,6 @@ impl Context for IsleContext<'_, '_, MInst, Flags, IsaFlags, 6> {
         MachAtomicRmwOp::from(*op)
     }
 
-    #[inline]
-    fn gen_move(&mut self, ty: Type, dst: WritableReg, src: Reg) -> MInst {
-        MInst::gen_move(dst, src, ty)
-    }
-
-    fn gen_call(
-        &mut self,
-        sig_ref: SigRef,
-        extname: ExternalName,
-        dist: RelocDistance,
-        args @ (inputs, off): ValueSlice,
-    ) -> InstOutput {
-        let caller_conv = self.lower_ctx.abi().call_conv();
-        let sig = &self.lower_ctx.dfg().signatures[sig_ref];
-        let num_rets = sig.returns.len();
-        let abi = ABISig::from_func_sig::<X64ABIMachineSpec>(sig, self.flags).unwrap();
-        let caller = X64Caller::from_func(sig, &extname, dist, caller_conv, self.flags).unwrap();
-
-        assert_eq!(
-            inputs.len(&self.lower_ctx.dfg().value_lists) - off,
-            sig.params.len()
-        );
-
-        self.gen_call_common(abi, num_rets, caller, args)
-    }
-
-    fn gen_call_indirect(
-        &mut self,
-        sig_ref: SigRef,
-        val: Value,
-        args @ (inputs, off): ValueSlice,
-    ) -> InstOutput {
-        let caller_conv = self.lower_ctx.abi().call_conv();
-        let ptr = self.put_in_reg(val);
-        let sig = &self.lower_ctx.dfg().signatures[sig_ref];
-        let num_rets = sig.returns.len();
-        let abi = ABISig::from_func_sig::<X64ABIMachineSpec>(sig, self.flags).unwrap();
-        let caller =
-            X64Caller::from_ptr(sig, ptr, Opcode::CallIndirect, caller_conv, self.flags).unwrap();
-
-        assert_eq!(
-            inputs.len(&self.lower_ctx.dfg().value_lists) - off,
-            sig.params.len()
-        );
-
-        self.gen_call_common(abi, num_rets, caller, args)
-    }
-
     #[inline]
     fn preg_rbp(&mut self) -> PReg {
         regs::rbp().to_real_reg().unwrap().into()
@@ -1062,63 +1015,7 @@ impl Context for IsleContext<'_, '_, MInst, Flags, IsaFlags, 6> {
 }
 
 impl IsleContext<'_, '_, MInst, Flags, IsaFlags, 6> {
-    fn abi_arg_slot_regs(&mut self, arg: &ABIArg) -> Option<WritableValueRegs> {
-        match arg {
-            &ABIArg::Slots { ref slots, .. } => match slots.len() {
-                1 => {
-                    let a = self.temp_writable_reg(slots[0].get_type());
-                    Some(WritableValueRegs::one(a))
-                }
-                2 => {
-                    let a = self.temp_writable_reg(slots[0].get_type());
-                    let b = self.temp_writable_reg(slots[1].get_type());
-                    Some(WritableValueRegs::two(a, b))
-                }
-                _ => panic!("Expected to see one or two slots only from {:?}", arg),
-            },
-            _ => None,
-        }
-    }
-
-    fn gen_call_common(
-        &mut self,
-        abi: ABISig,
-        num_rets: usize,
-        mut caller: X64Caller,
-        (inputs, off): ValueSlice,
-    ) -> InstOutput {
-        caller.emit_stack_pre_adjust(self.lower_ctx);
-
-        assert_eq!(
-            inputs.len(&self.lower_ctx.dfg().value_lists) - off,
-            abi.num_args()
-        );
-        let mut arg_regs = vec![];
-        for i in 0..abi.num_args() {
-            let input = inputs
-                .get(off + i, &self.lower_ctx.dfg().value_lists)
-                .unwrap();
-            arg_regs.push(self.lower_ctx.put_value_in_regs(input));
-        }
-        for (i, arg_regs) in arg_regs.iter().enumerate() {
-            caller.emit_copy_regs_to_buffer(self.lower_ctx, i, *arg_regs);
-        }
-        for (i, arg_regs) in arg_regs.iter().enumerate() {
-            caller.emit_copy_regs_to_arg(self.lower_ctx, i, *arg_regs);
-        }
-        caller.emit_call(self.lower_ctx);
-
-        let mut outputs = InstOutput::new();
-        for i in 0..num_rets {
-            let ret = abi.get_ret(i);
-            let retval_regs = self.abi_arg_slot_regs(&ret).unwrap();
-            caller.emit_copy_retval_to_regs(self.lower_ctx, i, retval_regs.clone());
-            outputs.push(valueregs::non_writable_value_regs(retval_regs));
-        }
-        caller.emit_stack_post_adjust(self.lower_ctx);
-
-        outputs
-    }
+    isle_prelude_method_helpers!(X64Caller);
 }
 
 // Since x64 doesn't have 8x16 shifts and we must use a 16x8 shift instead, we