Add timing for several new-backend stages.

This PR adds a bit more granularity to the output of e.g. `clif-util
compile -T`, indicating how much time is spent in VCode lowering and
various other new-backend-specific tasks.

cranelift/codegen/src/machinst/buffer.rs

@@ -143,6 +143,7 @@
 use crate::binemit::{Addend, CodeOffset, CodeSink, Reloc, Stackmap};
 use crate::ir::{ExternalName, Opcode, SourceLoc, TrapCode};
 use crate::machinst::{BlockIndex, MachInstLabelUse, VCodeInst};
+use crate::timing;
 
 use log::trace;
 use smallvec::SmallVec;
@@ -1074,6 +1075,8 @@ impl<I: VCodeInst> MachBuffer<I> {
 
     /// Finish any deferred emissions and/or fixups.
     pub fn finish(mut self) -> MachBufferFinalized {
+        let _tt = timing::vcode_emit_finish();
+
         // Ensure that all labels are defined. This is a full (release-mode)
         // assert because we must avoid looping indefinitely below; an
         // unresolved label will prevent the fixup_records vec from emptying.

cranelift/codegen/src/machinst/compile.rs

@@ -23,7 +23,10 @@ where
     // Build the lowering context.
     let lower = Lower::new(f, abi, block_order)?;
     // Lower the IR.
-    let (mut vcode, stackmap_request_info) = lower.lower(b)?;
+    let (mut vcode, stackmap_request_info) = {
+        let _tt = timing::vcode_lower();
+        lower.lower(b)?
+    };
 
     debug!(
         "vcode from lowering: \n{}",
@@ -92,7 +95,10 @@ where
 
     // Reorder vcode into final order and copy out final instruction sequence
     // all at once. This also inserts prologues/epilogues.
+    {
+        let _tt = timing::vcode_post_ra();
         vcode.replace_insns_from_regalloc(result);
+    }
 
     debug!(
         "vcode after regalloc: final version:\n{}",

cranelift/codegen/src/machinst/vcode.rs

@@ -20,6 +20,7 @@
 use crate::ir::{self, types, SourceLoc};
 use crate::machinst::*;
 use crate::settings;
+use crate::timing;
 
 use regalloc::Function as RegallocFunction;
 use regalloc::Set as RegallocSet;
@@ -424,6 +425,7 @@ impl<I: VCodeInst> VCode<I> {
     where
         I: MachInstEmit,
     {
+        let _tt = timing::vcode_emit();
         let mut buffer = MachBuffer::new();
         let mut state = I::State::new(&*self.abi);
 

cranelift/codegen/src/timing.rs

@@ -64,6 +64,11 @@ define_passes! {
     unreachable_code: "Remove unreachable blocks",
     remove_constant_phis: "Remove constant phi-nodes",
 
+    vcode_lower: "VCode lowering",
+    vcode_post_ra: "VCode post-register allocation finalization",
+    vcode_emit: "VCode emission",
+    vcode_emit_finish: "VCode emission finalization",
+
     regalloc: "Register allocation",
     ra_liveness: "RA liveness analysis",
     ra_cssa: "RA coalescing CSSA",