cranelift_codegen/
context.rs

1//! Cranelift compilation context and main entry point.
2//!
3//! When compiling many small functions, it is important to avoid repeatedly allocating and
4//! deallocating the data structures needed for compilation. The `Context` struct is used to hold
5//! on to memory allocations between function compilations.
6//!
7//! The context does not hold a `TargetIsa` instance which has to be provided as an argument
8//! instead. This is because an ISA instance is immutable and can be used by multiple compilation
9//! contexts concurrently. Typically, you would have one context per compilation thread and only a
10//! single ISA instance.
11
12use crate::alias_analysis::AliasAnalysis;
13use crate::dominator_tree::DominatorTree;
14use crate::egraph::EgraphPass;
15use crate::flowgraph::ControlFlowGraph;
16use crate::ir::Function;
17use crate::isa::TargetIsa;
18use crate::legalizer::simple_legalize;
19use crate::loop_analysis::LoopAnalysis;
20use crate::machinst::{CompiledCode, CompiledCodeStencil};
21use crate::nan_canonicalization::do_nan_canonicalization;
22use crate::remove_constant_phis::do_remove_constant_phis;
23use crate::result::{CodegenResult, CompileResult};
24use crate::settings::{FlagsOrIsa, OptLevel};
25use crate::trace;
26use crate::unreachable_code::eliminate_unreachable_code;
27use crate::verifier::{verify_context, VerifierErrors, VerifierResult};
28use crate::{timing, CompileError};
29#[cfg(feature = "souper-harvest")]
30use alloc::string::String;
31use alloc::vec::Vec;
32use cranelift_control::ControlPlane;
33use target_lexicon::Architecture;
34
35#[cfg(feature = "souper-harvest")]
36use crate::souper_harvest::do_souper_harvest;
37
38/// Persistent data structures and compilation pipeline.
39pub struct Context {
40    /// The function we're compiling.
41    pub func: Function,
42
43    /// The control flow graph of `func`.
44    pub cfg: ControlFlowGraph,
45
46    /// Dominator tree for `func`.
47    pub domtree: DominatorTree,
48
49    /// Loop analysis of `func`.
50    pub loop_analysis: LoopAnalysis,
51
52    /// Result of MachBackend compilation, if computed.
53    pub(crate) compiled_code: Option<CompiledCode>,
54
55    /// Flag: do we want a disassembly with the CompiledCode?
56    pub want_disasm: bool,
57}
58
59impl Context {
60    /// Allocate a new compilation context.
61    ///
62    /// The returned instance should be reused for compiling multiple functions in order to avoid
63    /// needless allocator thrashing.
64    pub fn new() -> Self {
65        Self::for_function(Function::new())
66    }
67
68    /// Allocate a new compilation context with an existing Function.
69    ///
70    /// The returned instance should be reused for compiling multiple functions in order to avoid
71    /// needless allocator thrashing.
72    pub fn for_function(func: Function) -> Self {
73        Self {
74            func,
75            cfg: ControlFlowGraph::new(),
76            domtree: DominatorTree::new(),
77            loop_analysis: LoopAnalysis::new(),
78            compiled_code: None,
79            want_disasm: false,
80        }
81    }
82
83    /// Clear all data structures in this context.
84    pub fn clear(&mut self) {
85        self.func.clear();
86        self.cfg.clear();
87        self.domtree.clear();
88        self.loop_analysis.clear();
89        self.compiled_code = None;
90        self.want_disasm = false;
91    }
92
93    /// Returns the compilation result for this function, available after any `compile` function
94    /// has been called.
95    pub fn compiled_code(&self) -> Option<&CompiledCode> {
96        self.compiled_code.as_ref()
97    }
98
99    /// Returns the compilation result for this function, available after any `compile` function
100    /// has been called.
101    pub fn take_compiled_code(&mut self) -> Option<CompiledCode> {
102        self.compiled_code.take()
103    }
104
105    /// Set the flag to request a disassembly when compiling with a
106    /// `MachBackend` backend.
107    pub fn set_disasm(&mut self, val: bool) {
108        self.want_disasm = val;
109    }
110
111    /// Compile the function, and emit machine code into a `Vec<u8>`.
112    ///
113    /// Run the function through all the passes necessary to generate
114    /// code for the target ISA represented by `isa`, as well as the
115    /// final step of emitting machine code into a `Vec<u8>`. The
116    /// machine code is not relocated. Instead, any relocations can be
117    /// obtained from `compiled_code()`.
118    ///
119    /// Performs any optimizations that are enabled, unless
120    /// `optimize()` was already invoked.
121    ///
122    /// This function calls `compile`, taking care to resize `mem` as
123    /// needed.
124    ///
125    /// Returns information about the function's code and read-only
126    /// data.
127    pub fn compile_and_emit(
128        &mut self,
129        isa: &dyn TargetIsa,
130        mem: &mut Vec<u8>,
131        ctrl_plane: &mut ControlPlane,
132    ) -> CompileResult<&CompiledCode> {
133        let compiled_code = self.compile(isa, ctrl_plane)?;
134        mem.extend_from_slice(compiled_code.code_buffer());
135        Ok(compiled_code)
136    }
137
138    /// Internally compiles the function into a stencil.
139    ///
140    /// Public only for testing and fuzzing purposes.
141    pub fn compile_stencil(
142        &mut self,
143        isa: &dyn TargetIsa,
144        ctrl_plane: &mut ControlPlane,
145    ) -> CodegenResult<CompiledCodeStencil> {
146        let _tt = timing::compile();
147
148        self.verify_if(isa)?;
149
150        self.optimize(isa, ctrl_plane)?;
151
152        isa.compile_function(&self.func, &self.domtree, self.want_disasm, ctrl_plane)
153    }
154
155    /// Optimize the function, performing all compilation steps up to
156    /// but not including machine-code lowering and register
157    /// allocation.
158    ///
159    /// Public only for testing purposes.
160    pub fn optimize(
161        &mut self,
162        isa: &dyn TargetIsa,
163        ctrl_plane: &mut ControlPlane,
164    ) -> CodegenResult<()> {
165        log::debug!(
166            "Number of CLIF instructions to optimize: {}",
167            self.func.dfg.num_insts()
168        );
169        log::debug!(
170            "Number of CLIF blocks to optimize: {}",
171            self.func.dfg.num_blocks()
172        );
173
174        let opt_level = isa.flags().opt_level();
175        crate::trace!(
176            "Optimizing (opt level {:?}):\n{}",
177            opt_level,
178            self.func.display()
179        );
180
181        self.compute_cfg();
182        if isa.flags().enable_nan_canonicalization() {
183            self.canonicalize_nans(isa)?;
184        }
185
186        self.legalize(isa)?;
187
188        self.compute_domtree();
189        self.eliminate_unreachable_code(isa)?;
190        self.remove_constant_phis(isa)?;
191
192        self.func.dfg.resolve_all_aliases();
193
194        if opt_level != OptLevel::None {
195            self.egraph_pass(isa, ctrl_plane)?;
196        }
197
198        Ok(())
199    }
200
201    /// Compile the function.
202    ///
203    /// Run the function through all the passes necessary to generate code for the target ISA
204    /// represented by `isa`. This does not include the final step of emitting machine code into a
205    /// code sink.
206    ///
207    /// Returns information about the function's code and read-only data.
208    pub fn compile(
209        &mut self,
210        isa: &dyn TargetIsa,
211        ctrl_plane: &mut ControlPlane,
212    ) -> CompileResult<&CompiledCode> {
213        let stencil = self
214            .compile_stencil(isa, ctrl_plane)
215            .map_err(|error| CompileError {
216                inner: error,
217                func: &self.func,
218            })?;
219        Ok(self
220            .compiled_code
221            .insert(stencil.apply_params(&self.func.params)))
222    }
223
224    /// If available, return information about the code layout in the
225    /// final machine code: the offsets (in bytes) of each basic-block
226    /// start, and all basic-block edges.
227    #[deprecated = "use CompiledCode::get_code_bb_layout"]
228    pub fn get_code_bb_layout(&self) -> Option<(Vec<usize>, Vec<(usize, usize)>)> {
229        self.compiled_code().map(CompiledCode::get_code_bb_layout)
230    }
231
232    /// Creates unwind information for the function.
233    ///
234    /// Returns `None` if the function has no unwind information.
235    #[cfg(feature = "unwind")]
236    #[deprecated = "use CompiledCode::create_unwind_info"]
237    pub fn create_unwind_info(
238        &self,
239        isa: &dyn TargetIsa,
240    ) -> CodegenResult<Option<crate::isa::unwind::UnwindInfo>> {
241        self.compiled_code().unwrap().create_unwind_info(isa)
242    }
243
244    /// Run the verifier on the function.
245    ///
246    /// Also check that the dominator tree and control flow graph are consistent with the function.
247    ///
248    /// TODO: rename to "CLIF validate" or similar.
249    pub fn verify<'a, FOI: Into<FlagsOrIsa<'a>>>(&self, fisa: FOI) -> VerifierResult<()> {
250        let mut errors = VerifierErrors::default();
251        let _ = verify_context(&self.func, &self.cfg, &self.domtree, fisa, &mut errors);
252
253        if errors.is_empty() {
254            Ok(())
255        } else {
256            Err(errors)
257        }
258    }
259
260    /// Run the verifier only if the `enable_verifier` setting is true.
261    pub fn verify_if<'a, FOI: Into<FlagsOrIsa<'a>>>(&self, fisa: FOI) -> CodegenResult<()> {
262        let fisa = fisa.into();
263        if fisa.flags.enable_verifier() {
264            self.verify(fisa)?;
265        }
266        Ok(())
267    }
268
269    /// Perform constant-phi removal on the function.
270    pub fn remove_constant_phis<'a, FOI: Into<FlagsOrIsa<'a>>>(
271        &mut self,
272        fisa: FOI,
273    ) -> CodegenResult<()> {
274        do_remove_constant_phis(&mut self.func, &mut self.domtree);
275        self.verify_if(fisa)?;
276        Ok(())
277    }
278
279    /// Perform NaN canonicalizing rewrites on the function.
280    pub fn canonicalize_nans(&mut self, isa: &dyn TargetIsa) -> CodegenResult<()> {
281        // Currently only RiscV64 is the only arch that may not have vector support.
282        let has_vector_support = match isa.triple().architecture {
283            Architecture::Riscv64(_) => match isa.isa_flags().iter().find(|f| f.name == "has_v") {
284                Some(value) => value.as_bool().unwrap_or(false),
285                None => false,
286            },
287            _ => true,
288        };
289        do_nan_canonicalization(&mut self.func, has_vector_support);
290        self.verify_if(isa)
291    }
292
293    /// Run the legalizer for `isa` on the function.
294    pub fn legalize(&mut self, isa: &dyn TargetIsa) -> CodegenResult<()> {
295        // Legalization invalidates the domtree and loop_analysis by mutating the CFG.
296        // TODO: Avoid doing this when legalization doesn't actually mutate the CFG.
297        self.domtree.clear();
298        self.loop_analysis.clear();
299
300        // Run some specific legalizations only.
301        simple_legalize(&mut self.func, &mut self.cfg, isa);
302        self.verify_if(isa)
303    }
304
305    /// Compute the control flow graph.
306    pub fn compute_cfg(&mut self) {
307        self.cfg.compute(&self.func)
308    }
309
310    /// Compute dominator tree.
311    pub fn compute_domtree(&mut self) {
312        self.domtree.compute(&self.func, &self.cfg)
313    }
314
315    /// Compute the loop analysis.
316    pub fn compute_loop_analysis(&mut self) {
317        self.loop_analysis
318            .compute(&self.func, &self.cfg, &self.domtree)
319    }
320
321    /// Compute the control flow graph and dominator tree.
322    pub fn flowgraph(&mut self) {
323        self.compute_cfg();
324        self.compute_domtree()
325    }
326
327    /// Perform unreachable code elimination.
328    pub fn eliminate_unreachable_code<'a, FOI>(&mut self, fisa: FOI) -> CodegenResult<()>
329    where
330        FOI: Into<FlagsOrIsa<'a>>,
331    {
332        eliminate_unreachable_code(&mut self.func, &mut self.cfg, &self.domtree);
333        self.verify_if(fisa)
334    }
335
336    /// Replace all redundant loads with the known values in
337    /// memory. These are loads whose values were already loaded by
338    /// other loads earlier, as well as loads whose values were stored
339    /// by a store instruction to the same instruction (so-called
340    /// "store-to-load forwarding").
341    pub fn replace_redundant_loads(&mut self) -> CodegenResult<()> {
342        let mut analysis = AliasAnalysis::new(&self.func, &self.domtree);
343        analysis.compute_and_update_aliases(&mut self.func);
344        Ok(())
345    }
346
347    /// Harvest candidate left-hand sides for superoptimization with Souper.
348    #[cfg(feature = "souper-harvest")]
349    pub fn souper_harvest(
350        &mut self,
351        out: &mut std::sync::mpsc::Sender<String>,
352    ) -> CodegenResult<()> {
353        do_souper_harvest(&self.func, out);
354        Ok(())
355    }
356
357    /// Run optimizations via the egraph infrastructure.
358    pub fn egraph_pass<'a, FOI>(
359        &mut self,
360        fisa: FOI,
361        ctrl_plane: &mut ControlPlane,
362    ) -> CodegenResult<()>
363    where
364        FOI: Into<FlagsOrIsa<'a>>,
365    {
366        let _tt = timing::egraph();
367
368        trace!(
369            "About to optimize with egraph phase:\n{}",
370            self.func.display()
371        );
372        let fisa = fisa.into();
373        self.compute_loop_analysis();
374        let mut alias_analysis = AliasAnalysis::new(&self.func, &self.domtree);
375        let mut pass = EgraphPass::new(
376            &mut self.func,
377            &self.domtree,
378            &self.loop_analysis,
379            &mut alias_analysis,
380            &fisa.flags,
381            ctrl_plane,
382        );
383        pass.run();
384        log::debug!("egraph stats: {:?}", pass.stats);
385        trace!("pinned_union_count: {}", pass.eclasses.pinned_union_count);
386        trace!("After egraph optimization:\n{}", self.func.display());
387
388        self.verify_if(fisa)
389    }
390}
cranelift_codegen/context.rs

cranelift_codegen/
context.rs
