Merge pull request #1825 from cfallin/spidermonkey-fixes

Three fixes to various SpiderMonkey-related issues

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

@@ -212,9 +212,14 @@ pub(crate) fn input_to_reg<C: LowerCtx<I = Inst>>(
     let from_bits = ty_bits(ty) as u8;
     let inputs = ctx.get_input(input.insn, input.input);
     let in_reg = if let Some(c) = inputs.constant {
+        let masked = if from_bits < 64 {
+            c & ((1u64 << from_bits) - 1)
+        } else {
+            c
+        };
         // Generate constants fresh at each use to minimize long-range register pressure.
         let to_reg = ctx.alloc_tmp(Inst::rc_for_type(ty).unwrap(), ty);
-        for inst in Inst::gen_constant(to_reg, c, ty).into_iter() {
+        for inst in Inst::gen_constant(to_reg, masked, ty).into_iter() {
             ctx.emit(inst);
         }
         to_reg.to_reg()

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

@@ -1252,7 +1252,15 @@ pub(crate) fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
         Opcode::Trapif | Opcode::Trapff => {
             let trap_info = (ctx.srcloc(insn), inst_trapcode(ctx.data(insn)).unwrap());
 
-            let cond = if op == Opcode::Trapif {
+            let cond = if maybe_input_insn(ctx, inputs[0], Opcode::IaddIfcout).is_some() {
+                let condcode = inst_condcode(ctx.data(insn)).unwrap();
+                let cond = lower_condcode(condcode);
+                // The flags must not have been clobbered by any other
+                // instruction between the iadd_ifcout and this instruction, as
+                // verified by the CLIF validator; so we can simply use the
+                // flags here.
+                cond
+            } else if op == Opcode::Trapif {
                 let condcode = inst_condcode(ctx.data(insn)).unwrap();
                 let cond = lower_condcode(condcode);
                 let is_signed = condcode_is_signed(condcode);
@@ -1852,6 +1860,35 @@ pub(crate) fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
             });
         }
 
+        Opcode::IaddIfcout => {
+            // This is a two-output instruction that is needed for the
+            // legalizer's explicit heap-check sequence, among possible other
+            // uses. Its second output is a flags output only ever meant to
+            // check for overflow using the
+            // `backend.unsigned_add_overflow_condition()` condition.
+            //
+            // Note that the CLIF validation will ensure that no flag-setting
+            // operation comes between this IaddIfcout and its use (e.g., a
+            // Trapif). Thus, we can rely on implicit communication through the
+            // processor flags rather than explicitly generating flags into a
+            // register. We simply use the variant of the add instruction that
+            // sets flags (`adds`) here.
+
+            // Ensure that the second output isn't directly called for: it
+            // should only be used by a flags-consuming op, which will directly
+            // understand this instruction and merge the comparison.
+            assert!(!ctx.is_reg_needed(insn, ctx.get_output(insn, 1).to_reg()));
+
+            // Now handle the iadd as above, except use an AddS opcode that sets
+            // flags.
+            let rd = output_to_reg(ctx, outputs[0]);
+            let rn = input_to_reg(ctx, inputs[0], NarrowValueMode::None);
+            let rm = input_to_rse_imm12(ctx, inputs[1], NarrowValueMode::None);
+            let ty = ty.unwrap();
+            let alu_op = choose_32_64(ty, ALUOp::AddS32, ALUOp::AddS64);
+            ctx.emit(alu_inst_imm12(alu_op, rd, rn, rm));
+        }
+
         Opcode::IaddImm
         | Opcode::ImulImm
         | Opcode::UdivImm
@@ -1862,7 +1899,6 @@ pub(crate) fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
         | Opcode::IaddCin
         | Opcode::IaddIfcin
         | Opcode::IaddCout
-        | Opcode::IaddIfcout
         | Opcode::IaddCarry
         | Opcode::IaddIfcarry
         | Opcode::IsubBin

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

@@ -1,5 +1,6 @@
 //! ARM 64-bit Instruction Set Architecture.
 
+use crate::ir::condcodes::IntCC;
 use crate::ir::Function;
 use crate::isa::Builder as IsaBuilder;
 use crate::machinst::{
@@ -92,6 +93,19 @@ impl MachBackend for AArch64Backend {
     fn reg_universe(&self) -> &RealRegUniverse {
         &self.reg_universe
     }
+
+    fn unsigned_add_overflow_condition(&self) -> IntCC {
+        // Unsigned `>=`; this corresponds to the carry flag set on aarch64, which happens on
+        // overflow of an add.
+        IntCC::UnsignedGreaterThanOrEqual
+    }
+
+    fn unsigned_sub_overflow_condition(&self) -> IntCC {
+        // unsigned `<`; this corresponds to the carry flag cleared on aarch64, which happens on
+        // underflow of a subtract (aarch64 follows a carry-cleared-on-borrow convention, the
+        // opposite of x86).
+        IntCC::UnsignedLessThan
+    }
 }
 
 /// Create a new `isa::Builder`.

cranelift/codegen/src/isa/x64/mod.rs

@@ -5,6 +5,7 @@ use alloc::boxed::Box;
 use regalloc::RealRegUniverse;
 use target_lexicon::Triple;
 
+use crate::ir::condcodes::IntCC;
 use crate::ir::Function;
 use crate::isa::Builder as IsaBuilder;
 use crate::machinst::pretty_print::ShowWithRRU;
@@ -84,6 +85,18 @@ impl MachBackend for X64Backend {
     fn reg_universe(&self) -> &RealRegUniverse {
         &self.reg_universe
     }
+
+    fn unsigned_add_overflow_condition(&self) -> IntCC {
+        // Unsigned `>=`; this corresponds to the carry flag set on x86, which happens on
+        // overflow of an add.
+        IntCC::UnsignedGreaterThanOrEqual
+    }
+
+    fn unsigned_sub_overflow_condition(&self) -> IntCC {
+        // unsigned `>=`; this corresponds to the carry flag set on x86, which happens on
+        // underflow of a subtract (carry is borrow for subtract).
+        IntCC::UnsignedGreaterThanOrEqual
+    }
 }
 
 /// Create a new `isa::Builder`.

cranelift/codegen/src/machinst/lower.rs

@@ -133,7 +133,7 @@ pub trait LowerCtx {
     /// Get the `idx`th output register of the given IR instruction. When
     /// `backend.lower_inst_to_regs(ctx, inst)` is called, it is expected that
     /// the backend will write results to these output register(s).
-    fn get_output(&mut self, ir_inst: Inst, idx: usize) -> Writable<Reg>;
+    fn get_output(&self, ir_inst: Inst, idx: usize) -> Writable<Reg>;
 
     // Codegen primitives: allocate temps, emit instructions, set result registers,
     // ask for an input to be gen'd into a register.
@@ -146,6 +146,10 @@ pub trait LowerCtx {
     /// `get_input()`. Codegen may not happen otherwise for the producing
     /// instruction if it has no side effects and no uses.
     fn use_input_reg(&mut self, input: LowerInput);
+    /// Is the given register output needed after the given instruction? Allows
+    /// instructions with multiple outputs to make fine-grained decisions on
+    /// which outputs to actually generate.
+    fn is_reg_needed(&self, ir_inst: Inst, reg: Reg) -> bool;
     /// Retrieve constant data given a handle.
     fn get_constant_data(&self, constant_handle: Constant) -> &ConstantData;
 }
@@ -906,7 +910,7 @@ impl<'func, I: VCodeInst> LowerCtx for Lower<'func, I> {
         self.get_input_for_val(ir_inst, val)
     }
 
-    fn get_output(&mut self, ir_inst: Inst, idx: usize) -> Writable<Reg> {
+    fn get_output(&self, ir_inst: Inst, idx: usize) -> Writable<Reg> {
         let val = self.f.dfg.inst_results(ir_inst)[idx];
         Writable::from_reg(self.value_regs[val])
     }
@@ -928,6 +932,10 @@ impl<'func, I: VCodeInst> LowerCtx for Lower<'func, I> {
         self.vreg_needed[input.reg.get_index()] = true;
     }
 
+    fn is_reg_needed(&self, ir_inst: Inst, reg: Reg) -> bool {
+        self.inst_needed[ir_inst] || self.vreg_needed[reg.get_index()]
+    }
+
     fn get_constant_data(&self, constant_handle: Constant) -> &ConstantData {
         self.f.dfg.constants.get(constant_handle)
     }

cranelift/codegen/src/machinst/mod.rs

@@ -308,14 +308,10 @@ pub trait MachBackend {
     fn reg_universe(&self) -> &RealRegUniverse;
 
     /// Machine-specific condcode info needed by TargetIsa.
-    fn unsigned_add_overflow_condition(&self) -> IntCC {
-        // TODO: this is what x86 specifies. Is this right for arm64?
-        IntCC::UnsignedLessThan
-    }
+    /// Condition that will be true when an IaddIfcout overflows.
+    fn unsigned_add_overflow_condition(&self) -> IntCC;
 
     /// Machine-specific condcode info needed by TargetIsa.
-    fn unsigned_sub_overflow_condition(&self) -> IntCC {
-        // TODO: this is what x86 specifies. Is this right for arm64?
-        IntCC::UnsignedLessThan
-    }
+    /// Condition that will be true when an IsubIfcout overflows.
+    fn unsigned_sub_overflow_condition(&self) -> IntCC;
 }

cranelift/codegen/src/value_label.rs

@@ -91,6 +91,11 @@ pub fn build_value_labels_ranges<T>(
 where
     T: From<SourceLoc> + Deref<Target = SourceLoc> + Ord + Copy,
 {
+    // FIXME(#1523): New-style backend does not yet have debug info.
+    if isa.get_mach_backend().is_some() {
+        return HashMap::new();
+    }
+
     let values_labels = build_value_labels_index::<T>(func);
 
     let mut blocks = func.layout.blocks().collect::<Vec<_>>();