changes

Cargo.lock

@@ -607,7 +607,7 @@ dependencies = [
  "cranelift-isle",
  "criterion",
  "gimli",
- "hashbrown 0.13.2",
+ "hashbrown 0.13.1",
  "log",
  "regalloc2",
  "serde",
@@ -677,7 +677,7 @@ name = "cranelift-frontend"
 version = "0.96.0"
 dependencies = [
  "cranelift-codegen",
- "hashbrown 0.13.2",
+ "hashbrown 0.13.1",
  "log",
  "similar",
  "smallvec",
@@ -747,7 +747,7 @@ dependencies = [
  "anyhow",
  "cranelift-codegen",
  "cranelift-control",
- "hashbrown 0.13.2",
+ "hashbrown 0.13.1",
 ]
 
 [[package]]
@@ -837,7 +837,7 @@ dependencies = [
  "cranelift-codegen",
  "cranelift-entity",
  "cranelift-frontend",
- "hashbrown 0.13.2",
+ "hashbrown 0.13.1",
  "itertools",
  "log",
  "serde",
@@ -1498,9 +1498,9 @@ checksum = "8a9ee70c43aaf417c914396645a0fa852624801b24ebb7ae78fe8272889ac888"
 
 [[package]]
 name = "hashbrown"
-version = "0.13.2"
+version = "0.13.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "43a3c133739dddd0d2990f9a4bdf8eb4b21ef50e4851ca85ab661199821d510e"
+checksum = "33ff8ae62cd3a9102e5637afc8452c55acf3844001bd5374e0b0bd7b6616c038"
 dependencies = [
  "ahash",
 ]
@@ -2049,7 +2049,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "ea86265d3d3dcb6a27fc51bd29a4bf387fae9d2986b823079d4986af253eb439"
 dependencies = [
  "crc32fast",
- "hashbrown 0.13.2",
+ "hashbrown 0.13.1",
  "indexmap",
  "memchr",
 ]
@@ -2578,10 +2578,11 @@ dependencies = [
 [[package]]
 name = "regalloc2"
 version = "0.6.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "80535183cae11b149d618fbd3c37e38d7cda589d82d7769e196ca9a9042d7621"
 dependencies = [
- "hashbrown 0.13.2",
+ "fxhash",
  "log",
- "rustc-hash",
  "serde",
  "slice-group-by",
  "smallvec",
@@ -2673,12 +2674,6 @@ version = "0.1.21"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "7ef03e0a2b150c7a90d01faf6254c9c48a41e95fb2a8c2ac1c6f0d2b9aefc342"
 
-[[package]]
-name = "rustc-hash"
-version = "1.1.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "08d43f7aa6b08d49f382cde6a7982047c3426db949b1424bc4b7ec9ae12c6ce2"
-
 [[package]]
 name = "rustix"
 version = "0.36.8"

Cargo.toml

@@ -84,9 +84,6 @@ anyhow = { workspace = true }
 [profile.release.build-override]
 opt-level = 0
 
-[profile.release]
-debug = true
-
 [workspace]
 resolver = '2'
 members = [

cranelift/codegen/Cargo.toml

@@ -27,7 +27,7 @@ serde = { version = "1.0.94", features = ["derive"], optional = true }
 bincode = { version = "1.2.1", optional = true }
 gimli = { workspace = true, features = ["write"], optional = true }
 smallvec = { workspace = true }
-regalloc2 = { path = "../../../regalloc2/", features = ["checker", "trace-log"] }
+regalloc2 = { version = "0.6.1", features = ["checker"] }
 souper-ir = { version = "2.1.0", optional = true }
 sha2 = { version = "0.10.2", optional = true }
 # It is a goal of the cranelift-codegen crate to have minimal external dependencies.

cranelift/codegen/src/isa/aarch64/inst/regs.rs

@@ -230,7 +230,6 @@ pub fn create_reg_env(flags: &settings::Flags) -> MachineEnv {
             ],
         ],
         fixed_stack_slots: vec![],
-        scratch_by_class: [None, None],
     };
 
     if !flags.enable_pinned_reg() {

cranelift/codegen/src/isa/riscv64/inst/regs.rs

@@ -182,7 +182,6 @@ pub fn crate_reg_eviroment(_flags: &settings::Flags) -> MachineEnv {
         preferred_regs_by_class,
         non_preferred_regs_by_class,
         fixed_stack_slots: vec![],
-        scratch_by_class: [None, None],
     }
 }
 

cranelift/codegen/src/isa/s390x/inst/regs.rs

@@ -151,7 +151,6 @@ pub fn create_machine_env(_flags: &settings::Flags) -> MachineEnv {
             ],
         ],
         fixed_stack_slots: vec![],
-        scratch_by_class: [None, None],
     }
 }
 

cranelift/codegen/src/isa/x64/inst/regs.rs

@@ -203,7 +203,6 @@ pub(crate) fn create_reg_env_systemv(flags: &settings::Flags) -> MachineEnv {
             vec![],
         ],
         fixed_stack_slots: vec![],
-        scratch_by_class: [None, None]
     };
 
     debug_assert_eq!(r15(), pinned_reg());

cranelift/codegen/src/machinst/blockorder.rs

@@ -81,15 +81,6 @@ pub struct BlockLoweringOrder {
     /// Ranges in `lowered_succ_indices` giving the successor lists for each lowered
     /// block. Indexed by lowering-order index (`BlockIndex`).
     lowered_succ_ranges: Vec<(Option<Inst>, std::ops::Range<usize>)>,
-    /// Cold blocks. These blocks are not reordered in the
-    /// `lowered_order` above; the lowered order must respect RPO
-    /// (uses after defs) in order for lowering to be
-    /// correct. Instead, this set is used to provide `is_cold()`,
-    /// which is used by VCode emission to sink the blocks at the last
-    /// moment (when we actually emit bytes into the MachBuffer).
-    cold_blocks: FxHashSet<BlockIndex>,
-    /// Lowered blocks that are indirect branch targets.
-    indirect_branch_targets: FxHashSet<BlockIndex>,
 }
 
 #[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
@@ -98,6 +89,15 @@ pub enum LoweredBlock {
     Orig {
         /// Original CLIF block.
         block: Block,
+        /// Cold block. These blocks are not reordered in the
+        /// `lowered_order` above; the lowered order must respect RPO
+        /// (uses after defs) in order for lowering to be
+        /// correct. Instead, this bool is used to provide `is_cold()`,
+        /// which is used by VCode emission to sink the blocks at the last
+        /// moment (when we actually emit bytes into the MachBuffer).
+        cold: bool,
+        /// Target of indirect branch
+        indirect_branch_target: bool,
     },
 
     /// Critical edge between two CLIF blocks.
@@ -112,6 +112,11 @@ pub enum LoweredBlock {
         /// indexing order as `inst_predicates::visit_block_succs`. This is used to distinguish
         /// multiple edges between the same CLIF blocks.
         succ_idx: u32,
+
+        /// See [LoweredBlock::orig]
+        cold: bool,
+        /// See [LoweredBlock::orig]
+        indirect_branch_target: bool,
     },
 }
 
@@ -119,7 +124,7 @@ impl LoweredBlock {
     /// Unwrap an `Orig` block.
     pub fn orig_block(&self) -> Option<Block> {
         match self {
-            &LoweredBlock::Orig { block } => Some(block),
+            &LoweredBlock::Orig { block, .. } => Some(block),
             &LoweredBlock::CriticalEdge { .. } => None,
         }
     }
@@ -141,6 +146,20 @@ impl LoweredBlock {
             &LoweredBlock::Orig { .. } => None,
         }
     }
+
+    pub fn is_cold(&self) -> bool {
+        match self {
+            &LoweredBlock::CriticalEdge { cold, .. } => cold,
+            &LoweredBlock::Orig { cold, .. } => cold
+        }
+    }
+
+    pub fn is_indirect_branch_target(&self) -> bool {
+        match self {
+            &LoweredBlock::CriticalEdge { indirect_branch_target, .. } => indirect_branch_target,
+            &LoweredBlock::Orig { indirect_branch_target, .. } => indirect_branch_target
+        }
+    }
 }
 
 impl BlockLoweringOrder {
@@ -165,7 +184,7 @@ impl BlockLoweringOrder {
             visit_block_succs(f, block, |_, succ, from_table| {
                 block_out_count[block] += 1;
                 block_in_count[succ] += 1;
-                block_succs.push(LoweredBlock::Orig { block: succ });
+                block_succs.push(LoweredBlock::Orig { block: succ, cold: f.layout.is_cold(succ), indirect_branch_target: false });
 
                 if from_table {
                     indirect_branch_target_clif_blocks.insert(succ);
@@ -192,7 +211,7 @@ impl BlockLoweringOrder {
         let mut lowered_order = Vec::new();
 
         for &block in domtree.cfg_postorder().iter().rev() {
-            let lb = LoweredBlock::Orig { block };
+            let lb = LoweredBlock::Orig { block, cold: f.layout.is_cold(block), indirect_branch_target: indirect_branch_target_clif_blocks.contains(&block) };
             let bindex = BlockIndex::new(lowered_order.len());
             lb_to_bindex.insert(lb.clone(), bindex);
             lowered_order.push(lb);
@@ -208,6 +227,10 @@ impl BlockLoweringOrder {
                             pred: block,
                             succ,
                             succ_idx: succ_ix as u32,
+                            // Edges inherit indirect branch and cold block metadata from their
+                            // successor.
+                            cold: f.layout.is_cold(succ),
+                            indirect_branch_target: indirect_branch_target_clif_blocks.contains(&succ)
                         };
                         let bindex = BlockIndex::new(lowered_order.len());
                         lb_to_bindex.insert(*lb, bindex);
@@ -221,8 +244,6 @@ impl BlockLoweringOrder {
         // during the creation of `lowering_order`, as we need `lb_to_bindex` to be fully populated
         // first.
         let mut lowered_succ_indices = Vec::new();
-        let mut cold_blocks = FxHashSet::default();
-        let mut indirect_branch_targets = FxHashSet::default();
         let lowered_succ_ranges =
             Vec::from_iter(lowered_order.iter().enumerate().map(|(ix, lb)| {
                 let bindex = BlockIndex::new(ix);
@@ -230,18 +251,24 @@ impl BlockLoweringOrder {
                 let opt_inst = match lb {
                     // Block successors are pulled directly over, as they'll have been mutated when
                     // determining the block order already.
-                    &LoweredBlock::Orig { block } => {
+                    &LoweredBlock::Orig { block, .. } => {
                         let range = block_succ_range[block].clone();
                         lowered_succ_indices
-                            .extend(block_succs[range].iter().map(|lb| lb_to_bindex[lb]));
-
-                        if f.layout.is_cold(block) {
-                            cold_blocks.insert(bindex);
-                        }
-
-                        if indirect_branch_target_clif_blocks.contains(&block) {
-                            indirect_branch_targets.insert(bindex);
-                        }
+                            .extend(block_succs[range].iter().map(|lb| {
+                                match lb {
+                                    LoweredBlock::Orig { block, cold, .. } => {
+                                        match lb_to_bindex.get(&LoweredBlock::Orig { block: *block, cold: *cold, indirect_branch_target: indirect_branch_target_clif_blocks.contains(block) } ) {
+                                            Some(idx) => *idx,
+                                            None => {
+                                                panic!("Trying to hashlookup {:?} in HashMap: {:?}", lb, lb_to_bindex);
+                                            }
+                                        }
+                                    },
+                                    LoweredBlock::CriticalEdge { .. } => {
+                                        lb_to_bindex[&lb]
+                                    }
+                                }
+                            }));
 
                         let last = f.layout.last_inst(block).unwrap();
                         let opcode = f.dfg.insts[last].opcode();
@@ -253,21 +280,10 @@ impl BlockLoweringOrder {
 
                     // Critical edges won't have successor information in block_succ_range, but
                     // they only have a single known successor to record anyway.
-                    &LoweredBlock::CriticalEdge { succ, .. } => {
-                        let succ_index = lb_to_bindex[&LoweredBlock::Orig { block: succ }];
+                    &LoweredBlock::CriticalEdge { succ, cold, indirect_branch_target, .. } => {
+                        let succ_index = lb_to_bindex[&LoweredBlock::Orig { block: succ, cold, indirect_branch_target }];
                         lowered_succ_indices.push(succ_index);
 
-                        // Edges inherit indirect branch and cold block metadata from their
-                        // successor.
-
-                        if f.layout.is_cold(succ) {
-                            cold_blocks.insert(bindex);
-                        }
-
-                        if indirect_branch_target_clif_blocks.contains(&succ) {
-                            indirect_branch_targets.insert(bindex);
-                        }
-
                         None
                     }
                 };
@@ -278,9 +294,7 @@ impl BlockLoweringOrder {
         let result = BlockLoweringOrder {
             lowered_order,
             lowered_succ_indices,
-            lowered_succ_ranges,
-            cold_blocks,
-            indirect_branch_targets,
+            lowered_succ_ranges
         };
 
         trace!("BlockLoweringOrder: {:#?}", result);
@@ -300,13 +314,13 @@ impl BlockLoweringOrder {
 
     /// Determine whether the given lowered-block index is cold.
     pub fn is_cold(&self, block: BlockIndex) -> bool {
-        self.cold_blocks.contains(&block)
+        self.lowered_order[block.index()].is_cold()
     }
 
     /// Determine whether the given lowered block index is an indirect branch
     /// target.
     pub fn is_indirect_branch_target(&self, block: BlockIndex) -> bool {
-        self.indirect_branch_targets.contains(&block)
+        self.lowered_order[block.index()].is_indirect_branch_target()
     }
 }
 

cranelift/codegen/src/machinst/compile.rs

@@ -51,7 +51,6 @@ pub fn compile<B: LowerBackend + TargetIsa>(
         let _tt = timing::regalloc();
         let mut options = RegallocOptions::default();
         options.verbose_log = b.flags().regalloc_verbose_logs();
-        options.fast_alloc = true;
 
         if cfg!(debug_assertions) {
             options.validate_ssa = true;

cranelift/codegen/src/machinst/vcode.rs

@@ -771,7 +771,6 @@ impl<I: VCodeInst> VCode<I> {
     {
         // To write into disasm string.
         use core::fmt::Write;
-        let want_disasm = true;
 
         let _tt = timing::vcode_emit();
         let mut buffer = MachBuffer::new();
@@ -909,7 +908,6 @@ impl<I: VCodeInst> VCode<I> {
             for inst_or_edit in regalloc.block_insts_and_edits(&self, block) {
                 match inst_or_edit {
                     InstOrEdit::Inst(iix) => {
-                        trace!("Emitting inst {}", iix.index());
                         if !self.debug_value_labels.is_empty() {
                             // If we need to produce debug info,
                             // record the offset of each instruction
@@ -935,7 +933,7 @@ impl<I: VCodeInst> VCode<I> {
                             // and they come out the other end, if
                             // still needed (not elided), as
                             // regalloc-inserted moves.
-                            //continue;
+                            continue;
                         }
 
                         // Update the srcloc at this point in the buffer.
@@ -1011,7 +1009,6 @@ impl<I: VCodeInst> VCode<I> {
                         // immediately emit it.
                         match (from.as_reg(), to.as_reg()) {
                             (Some(from), Some(to)) => {
-                                trace!("Emitting move from {} to {}", from, to);
                                 // Reg-to-reg move.
                                 let from_rreg = Reg::from(from);
                                 let to_rreg = Writable::from_reg(Reg::from(to));
@@ -1021,7 +1018,6 @@ impl<I: VCodeInst> VCode<I> {
                                 do_emit(&mv, &[], &mut disasm, &mut buffer, &mut state);
                             }
                             (Some(from), None) => {
-                                trace!("Emitting move from {} to slot {}", from, to.as_stack().unwrap().index());
                                 // Spill from register to spillslot.
                                 let to = to.as_stack().unwrap();
                                 let from_rreg = RealReg::from(from);
@@ -1029,7 +1025,6 @@ impl<I: VCodeInst> VCode<I> {
                                 do_emit(&spill, &[], &mut disasm, &mut buffer, &mut state);
                             }
                             (None, Some(to)) => {
-                                trace!("Emitting move from slot {} to {}", from.as_stack().unwrap().index(), to);
                                 // Load from spillslot to register.
                                 let from = from.as_stack().unwrap();
                                 let to_rreg = Writable::from_reg(RealReg::from(to));
@@ -1105,10 +1100,6 @@ impl<I: VCodeInst> VCode<I> {
             self.compute_value_labels_ranges(regalloc, &inst_offsets[..], func_body_len);
         let frame_size = self.abi.frame_size();
 
-        if want_disasm {
-            trace!("Disassembly:\n{}", disasm);
-        }
-
         EmitResult {
             buffer,
             bb_offsets,
@@ -1291,6 +1282,14 @@ impl<I: VCodeInst> RegallocFunction for VCode<I> {
         self.insts[insn.index()].is_safepoint()
     }
 
+    fn is_move(&self, insn: InsnIndex) -> Option<(Operand, Operand)> {
+        let (a, b) = self.is_move.get(&insn)?;
+        Some((
+            self.assert_operand_not_vreg_alias(*a),
+            self.assert_operand_not_vreg_alias(*b),
+        ))
+    }
+
     fn inst_operands(&self, insn: InsnIndex) -> &[Operand] {
         let (start, end) = self.operand_ranges[insn.index()];
         let ret = &self.operands[start as usize..end as usize];
@@ -1350,39 +1349,12 @@ impl<I: VCodeInst> fmt::Debug for VCode<I> {
 
         for block in 0..self.num_blocks() {
             let block = BlockIndex::new(block);
-            write!(f, "Block {}(", block.index())?;
-            {
-                let mut first = true;
-                for vreg in self.block_params(block) {
-                    if !first {
-                        write!(f, ", ")?;
-                    } else {
-                        first = false;
-                    }
-                    write!(f, "{}", vreg)?;
-                }
-            }
-            writeln!(f, "):")?;
+            writeln!(f, "Block {}:", block.index())?;
             if let Some(bb) = self.bindex_to_bb(block) {
                 writeln!(f, "    (original IR block: {})", bb)?;
             }
-            for (i, succ) in self.succs(block).iter().enumerate() {
-                let last_inst = self.block_insns(block).last();
-                if self.is_branch(last_inst) {
-                    write!(f, "    (successor: Block {}(", succ.index())?;
-                    let mut first = true;
-                    for vreg in self.branch_blockparams(block, last_inst, i) {
-                        if !first {
-                            write!(f, ", ")?;
-                        } else {
-                            first = false;
-                        }
-                        write!(f, "{}", vreg)?;
-                    }
-                    writeln!(f, "))")?;
-                } else {
-                    write!(f, "    (successor: Block {})", succ.index())?;
-                }
+            for succ in self.succs(block) {
+                writeln!(f, "    (successor: Block {})", succ.index())?;
             }
             let (start, end) = self.block_ranges[block.index()];
             writeln!(

cranelift/filetests/src/function_runner.rs

@@ -17,7 +17,6 @@ use std::cmp::max;
 use std::collections::hash_map::Entry;
 use std::collections::HashMap;
 use thiserror::Error;
-use log::trace;
 
 const TESTFILE_NAMESPACE: u32 = 0;
 
@@ -192,7 +191,6 @@ impl TestFileCompiler {
         // First, rename the function
         let func_original_name = func.name;
         func.name = UserFuncName::User(defined_func.new_name.clone());
-        trace!("Rename func from '{}' to '{}'", func_original_name, func.name);
 
         // Rename any functions that it references
         // Do this in stages to appease the borrow checker
@@ -235,10 +233,6 @@ impl TestFileCompiler {
             .get(&func.name)
             .ok_or(anyhow!("Undeclared function {} found!", &func.name))?;
 
-        if log::log_enabled!(log::Level::Trace) {
-            self.ctx.set_disasm(true);
-        }
-
         self.ctx.func = self.apply_func_rename(func, defined_func)?;
         self.module.define_function_with_control_plane(
             defined_func.func_id,
@@ -283,11 +277,6 @@ impl TestFileCompiler {
         // available).
         self.module.finalize_definitions()?;
 
-        for (name, func) in self.defined_functions.iter() {
-            let ptr = self.module.get_finalized_function(func.func_id);
-            trace!("Function '{}' at {:#X}", name, ptr as u64);
-        }
-
         Ok(CompiledTestFile {
             module: Some(self.module),
             defined_functions: self.defined_functions,

winch/codegen/Cargo.toml

@@ -17,7 +17,7 @@ target-lexicon = { workspace = true, features = ["std"] }
 # In the next iteration we'll factor out the common bits so that they can be consumed
 # by Cranelift and Winch.
 cranelift-codegen = { workspace = true }
-regalloc2 = { path = "../../../regalloc2" }
+regalloc2 = "0.6.0"
 gimli = { workspace = true }
 
 [features]