rust/compiler/rustc_mir/src/transform/coverage/query.rs

192 lines
8.2 KiB
Rust
Raw Normal View History

use super::*;
use rustc_middle::mir::coverage::*;
use rustc_middle::mir::{self, Body, Coverage, CoverageInfo};
use rustc_middle::ty::query::Providers;
use rustc_middle::ty::{self, TyCtxt};
use rustc_span::def_id::DefId;
coverage bug fixes and optimization support Adjusted LLVM codegen for code compiled with `-Zinstrument-coverage` to address multiple, somewhat related issues. Fixed a significant flaw in prior coverage solution: Every counter generated a new counter variable, but there should have only been one counter variable per function. This appears to have bloated .profraw files significantly. (For a small program, it increased the size by about 40%. I have not tested large programs, but there is anecdotal evidence that profraw files were way too large. This is a good fix, regardless, but hopefully it also addresses related issues. Fixes: #82144 Invalid LLVM coverage data produced when compiled with -C opt-level=1 Existing tests now work up to at least `opt-level=3`. This required a detailed analysis of the LLVM IR, comparisons with Clang C++ LLVM IR when compiled with coverage, and a lot of trial and error with codegen adjustments. The biggest hurdle was figuring out how to continue to support coverage results for unused functions and generics. Rust's coverage results have three advantages over Clang's coverage results: 1. Rust's coverage map does not include any overlapping code regions, making coverage counting unambiguous. 2. Rust generates coverage results (showing zero counts) for all unused functions, including generics. (Clang does not generate coverage for uninstantiated template functions.) 3. Rust's unused functions produce minimal stubbed functions in LLVM IR, sufficient for including in the coverage results; while Clang must generate the complete LLVM IR for each unused function, even though it will never be called. This PR removes the previous hack of attempting to inject coverage into some other existing function instance, and generates dedicated instances for each unused function. This change, and a few other adjustments (similar to what is required for `-C link-dead-code`, but with lower impact), makes it possible to support LLVM optimizations. Fixes: #79651 Coverage report: "Unexecuted instantiation:..." for a generic function from multiple crates Fixed by removing the aforementioned hack. Some "Unexecuted instantiation" notices are unavoidable, as explained in the `used_crate.rs` test, but `-Zinstrument-coverage` has new options to back off support for either unused generics, or all unused functions, which avoids the notice, at the cost of less coverage of unused functions. Fixes: #82875 Invalid LLVM coverage data produced with crate brotli_decompressor Fixed by disabling the LLVM function attribute that forces inlining, if `-Z instrument-coverage` is enabled. This attribute is applied to Rust functions with `#[inline(always)], and in some cases, the forced inlining breaks coverage instrumentation and reports.
2021-03-15 23:32:45 +00:00
/// A `query` provider for retrieving coverage information injected into MIR.
pub(crate) fn provide(providers: &mut Providers) {
coverage bug fixes and optimization support Adjusted LLVM codegen for code compiled with `-Zinstrument-coverage` to address multiple, somewhat related issues. Fixed a significant flaw in prior coverage solution: Every counter generated a new counter variable, but there should have only been one counter variable per function. This appears to have bloated .profraw files significantly. (For a small program, it increased the size by about 40%. I have not tested large programs, but there is anecdotal evidence that profraw files were way too large. This is a good fix, regardless, but hopefully it also addresses related issues. Fixes: #82144 Invalid LLVM coverage data produced when compiled with -C opt-level=1 Existing tests now work up to at least `opt-level=3`. This required a detailed analysis of the LLVM IR, comparisons with Clang C++ LLVM IR when compiled with coverage, and a lot of trial and error with codegen adjustments. The biggest hurdle was figuring out how to continue to support coverage results for unused functions and generics. Rust's coverage results have three advantages over Clang's coverage results: 1. Rust's coverage map does not include any overlapping code regions, making coverage counting unambiguous. 2. Rust generates coverage results (showing zero counts) for all unused functions, including generics. (Clang does not generate coverage for uninstantiated template functions.) 3. Rust's unused functions produce minimal stubbed functions in LLVM IR, sufficient for including in the coverage results; while Clang must generate the complete LLVM IR for each unused function, even though it will never be called. This PR removes the previous hack of attempting to inject coverage into some other existing function instance, and generates dedicated instances for each unused function. This change, and a few other adjustments (similar to what is required for `-C link-dead-code`, but with lower impact), makes it possible to support LLVM optimizations. Fixes: #79651 Coverage report: "Unexecuted instantiation:..." for a generic function from multiple crates Fixed by removing the aforementioned hack. Some "Unexecuted instantiation" notices are unavoidable, as explained in the `used_crate.rs` test, but `-Zinstrument-coverage` has new options to back off support for either unused generics, or all unused functions, which avoids the notice, at the cost of less coverage of unused functions. Fixes: #82875 Invalid LLVM coverage data produced with crate brotli_decompressor Fixed by disabling the LLVM function attribute that forces inlining, if `-Z instrument-coverage` is enabled. This attribute is applied to Rust functions with `#[inline(always)], and in some cases, the forced inlining breaks coverage instrumentation and reports.
2021-03-15 23:32:45 +00:00
providers.coverageinfo = |tcx, def_id| coverageinfo(tcx, def_id);
providers.covered_file_name = |tcx, def_id| covered_file_name(tcx, def_id);
providers.covered_code_regions = |tcx, def_id| covered_code_regions(tcx, def_id);
}
/// The `num_counters` argument to `llvm.instrprof.increment` is the max counter_id + 1, or in
/// other words, the number of counter value references injected into the MIR (plus 1 for the
/// reserved `ZERO` counter, which uses counter ID `0` when included in an expression). Injected
/// counters have a counter ID from `1..num_counters-1`.
///
/// `num_expressions` is the number of counter expressions added to the MIR body.
///
/// Both `num_counters` and `num_expressions` are used to initialize new vectors, during backend
/// code generate, to lookup counters and expressions by simple u32 indexes.
///
/// MIR optimization may split and duplicate some BasicBlock sequences, or optimize out some code
/// including injected counters. (It is OK if some counters are optimized out, but those counters
/// are still included in the total `num_counters` or `num_expressions`.) Simply counting the
/// calls may not work; but computing the number of counters or expressions by adding `1` to the
/// highest ID (for a given instrumented function) is valid.
///
/// This visitor runs twice, first with `add_missing_operands` set to `false`, to find the maximum
/// counter ID and maximum expression ID based on their enum variant `id` fields; then, as a
/// safeguard, with `add_missing_operands` set to `true`, to find any other counter or expression
/// IDs referenced by expression operands, if not already seen.
///
2020-10-25 18:13:16 +00:00
/// Ideally, each operand ID in a MIR `CoverageKind::Expression` will have a separate MIR `Coverage`
/// statement for the `Counter` or `Expression` with the referenced ID. but since current or future
/// MIR optimizations can theoretically optimize out segments of a MIR, it may not be possible to
/// guarantee this, so the second pass ensures the `CoverageInfo` counts include all referenced IDs.
struct CoverageVisitor {
info: CoverageInfo,
add_missing_operands: bool,
}
impl CoverageVisitor {
/// Updates `num_counters` to the maximum encountered zero-based counter_id plus 1. Note the
/// final computed number of counters should be the number of all `CoverageKind::Counter`
/// statements in the MIR *plus one* for the implicit `ZERO` counter.
#[inline(always)]
fn update_num_counters(&mut self, counter_id: u32) {
self.info.num_counters = std::cmp::max(self.info.num_counters, counter_id + 1);
}
/// Computes an expression index for each expression ID, and updates `num_expressions` to the
/// maximum encountered index plus 1.
#[inline(always)]
fn update_num_expressions(&mut self, expression_id: u32) {
let expression_index = u32::MAX - expression_id;
self.info.num_expressions = std::cmp::max(self.info.num_expressions, expression_index + 1);
}
fn update_from_expression_operand(&mut self, operand_id: u32) {
if operand_id >= self.info.num_counters {
let operand_as_expression_index = u32::MAX - operand_id;
if operand_as_expression_index >= self.info.num_expressions {
// The operand ID is outside the known range of counter IDs and also outside the
// known range of expression IDs. In either case, the result of a missing operand
// (if and when used in an expression) will be zero, so from a computation
// perspective, it doesn't matter whether it is interepretted as a counter or an
// expression.
//
// However, the `num_counters` and `num_expressions` query results are used to
// allocate arrays when generating the coverage map (during codegen), so choose
// the type that grows either `num_counters` or `num_expressions` the least.
if operand_id - self.info.num_counters
< operand_as_expression_index - self.info.num_expressions
{
self.update_num_counters(operand_id)
} else {
self.update_num_expressions(operand_id)
}
}
}
}
fn visit_body(&mut self, body: &Body<'_>) {
for bb_data in body.basic_blocks().iter() {
for statement in bb_data.statements.iter() {
if let StatementKind::Coverage(box ref coverage) = statement.kind {
if is_inlined(body, statement) {
continue;
}
self.visit_coverage(coverage);
}
}
}
}
fn visit_coverage(&mut self, coverage: &Coverage) {
if self.add_missing_operands {
match coverage.kind {
CoverageKind::Expression { lhs, rhs, .. } => {
self.update_from_expression_operand(u32::from(lhs));
self.update_from_expression_operand(u32::from(rhs));
}
_ => {}
}
} else {
match coverage.kind {
CoverageKind::Counter { id, .. } => {
self.update_num_counters(u32::from(id));
}
CoverageKind::Expression { id, .. } => {
self.update_num_expressions(u32::from(id));
}
_ => {}
}
}
}
}
coverage bug fixes and optimization support Adjusted LLVM codegen for code compiled with `-Zinstrument-coverage` to address multiple, somewhat related issues. Fixed a significant flaw in prior coverage solution: Every counter generated a new counter variable, but there should have only been one counter variable per function. This appears to have bloated .profraw files significantly. (For a small program, it increased the size by about 40%. I have not tested large programs, but there is anecdotal evidence that profraw files were way too large. This is a good fix, regardless, but hopefully it also addresses related issues. Fixes: #82144 Invalid LLVM coverage data produced when compiled with -C opt-level=1 Existing tests now work up to at least `opt-level=3`. This required a detailed analysis of the LLVM IR, comparisons with Clang C++ LLVM IR when compiled with coverage, and a lot of trial and error with codegen adjustments. The biggest hurdle was figuring out how to continue to support coverage results for unused functions and generics. Rust's coverage results have three advantages over Clang's coverage results: 1. Rust's coverage map does not include any overlapping code regions, making coverage counting unambiguous. 2. Rust generates coverage results (showing zero counts) for all unused functions, including generics. (Clang does not generate coverage for uninstantiated template functions.) 3. Rust's unused functions produce minimal stubbed functions in LLVM IR, sufficient for including in the coverage results; while Clang must generate the complete LLVM IR for each unused function, even though it will never be called. This PR removes the previous hack of attempting to inject coverage into some other existing function instance, and generates dedicated instances for each unused function. This change, and a few other adjustments (similar to what is required for `-C link-dead-code`, but with lower impact), makes it possible to support LLVM optimizations. Fixes: #79651 Coverage report: "Unexecuted instantiation:..." for a generic function from multiple crates Fixed by removing the aforementioned hack. Some "Unexecuted instantiation" notices are unavoidable, as explained in the `used_crate.rs` test, but `-Zinstrument-coverage` has new options to back off support for either unused generics, or all unused functions, which avoids the notice, at the cost of less coverage of unused functions. Fixes: #82875 Invalid LLVM coverage data produced with crate brotli_decompressor Fixed by disabling the LLVM function attribute that forces inlining, if `-Z instrument-coverage` is enabled. This attribute is applied to Rust functions with `#[inline(always)], and in some cases, the forced inlining breaks coverage instrumentation and reports.
2021-03-15 23:32:45 +00:00
fn coverageinfo<'tcx>(tcx: TyCtxt<'tcx>, def_id: DefId) -> CoverageInfo {
let mir_body = mir_body(tcx, def_id);
let mut coverage_visitor = CoverageVisitor {
// num_counters always has at least the `ZERO` counter.
info: CoverageInfo { num_counters: 1, num_expressions: 0 },
add_missing_operands: false,
};
coverage_visitor.visit_body(mir_body);
coverage_visitor.add_missing_operands = true;
coverage_visitor.visit_body(mir_body);
coverage_visitor.info
}
fn covered_file_name<'tcx>(tcx: TyCtxt<'tcx>, def_id: DefId) -> Option<Symbol> {
coverage bug fixes and optimization support Adjusted LLVM codegen for code compiled with `-Zinstrument-coverage` to address multiple, somewhat related issues. Fixed a significant flaw in prior coverage solution: Every counter generated a new counter variable, but there should have only been one counter variable per function. This appears to have bloated .profraw files significantly. (For a small program, it increased the size by about 40%. I have not tested large programs, but there is anecdotal evidence that profraw files were way too large. This is a good fix, regardless, but hopefully it also addresses related issues. Fixes: #82144 Invalid LLVM coverage data produced when compiled with -C opt-level=1 Existing tests now work up to at least `opt-level=3`. This required a detailed analysis of the LLVM IR, comparisons with Clang C++ LLVM IR when compiled with coverage, and a lot of trial and error with codegen adjustments. The biggest hurdle was figuring out how to continue to support coverage results for unused functions and generics. Rust's coverage results have three advantages over Clang's coverage results: 1. Rust's coverage map does not include any overlapping code regions, making coverage counting unambiguous. 2. Rust generates coverage results (showing zero counts) for all unused functions, including generics. (Clang does not generate coverage for uninstantiated template functions.) 3. Rust's unused functions produce minimal stubbed functions in LLVM IR, sufficient for including in the coverage results; while Clang must generate the complete LLVM IR for each unused function, even though it will never be called. This PR removes the previous hack of attempting to inject coverage into some other existing function instance, and generates dedicated instances for each unused function. This change, and a few other adjustments (similar to what is required for `-C link-dead-code`, but with lower impact), makes it possible to support LLVM optimizations. Fixes: #79651 Coverage report: "Unexecuted instantiation:..." for a generic function from multiple crates Fixed by removing the aforementioned hack. Some "Unexecuted instantiation" notices are unavoidable, as explained in the `used_crate.rs` test, but `-Zinstrument-coverage` has new options to back off support for either unused generics, or all unused functions, which avoids the notice, at the cost of less coverage of unused functions. Fixes: #82875 Invalid LLVM coverage data produced with crate brotli_decompressor Fixed by disabling the LLVM function attribute that forces inlining, if `-Z instrument-coverage` is enabled. This attribute is applied to Rust functions with `#[inline(always)], and in some cases, the forced inlining breaks coverage instrumentation and reports.
2021-03-15 23:32:45 +00:00
if tcx.is_mir_available(def_id) {
let body = mir_body(tcx, def_id);
for bb_data in body.basic_blocks().iter() {
for statement in bb_data.statements.iter() {
if let StatementKind::Coverage(box ref coverage) = statement.kind {
if let Some(code_region) = coverage.code_region.as_ref() {
if is_inlined(body, statement) {
continue;
}
return Some(code_region.file_name);
}
}
}
}
}
coverage bug fixes and optimization support Adjusted LLVM codegen for code compiled with `-Zinstrument-coverage` to address multiple, somewhat related issues. Fixed a significant flaw in prior coverage solution: Every counter generated a new counter variable, but there should have only been one counter variable per function. This appears to have bloated .profraw files significantly. (For a small program, it increased the size by about 40%. I have not tested large programs, but there is anecdotal evidence that profraw files were way too large. This is a good fix, regardless, but hopefully it also addresses related issues. Fixes: #82144 Invalid LLVM coverage data produced when compiled with -C opt-level=1 Existing tests now work up to at least `opt-level=3`. This required a detailed analysis of the LLVM IR, comparisons with Clang C++ LLVM IR when compiled with coverage, and a lot of trial and error with codegen adjustments. The biggest hurdle was figuring out how to continue to support coverage results for unused functions and generics. Rust's coverage results have three advantages over Clang's coverage results: 1. Rust's coverage map does not include any overlapping code regions, making coverage counting unambiguous. 2. Rust generates coverage results (showing zero counts) for all unused functions, including generics. (Clang does not generate coverage for uninstantiated template functions.) 3. Rust's unused functions produce minimal stubbed functions in LLVM IR, sufficient for including in the coverage results; while Clang must generate the complete LLVM IR for each unused function, even though it will never be called. This PR removes the previous hack of attempting to inject coverage into some other existing function instance, and generates dedicated instances for each unused function. This change, and a few other adjustments (similar to what is required for `-C link-dead-code`, but with lower impact), makes it possible to support LLVM optimizations. Fixes: #79651 Coverage report: "Unexecuted instantiation:..." for a generic function from multiple crates Fixed by removing the aforementioned hack. Some "Unexecuted instantiation" notices are unavoidable, as explained in the `used_crate.rs` test, but `-Zinstrument-coverage` has new options to back off support for either unused generics, or all unused functions, which avoids the notice, at the cost of less coverage of unused functions. Fixes: #82875 Invalid LLVM coverage data produced with crate brotli_decompressor Fixed by disabling the LLVM function attribute that forces inlining, if `-Z instrument-coverage` is enabled. This attribute is applied to Rust functions with `#[inline(always)], and in some cases, the forced inlining breaks coverage instrumentation and reports.
2021-03-15 23:32:45 +00:00
return None;
}
fn covered_code_regions<'tcx>(tcx: TyCtxt<'tcx>, def_id: DefId) -> Vec<&'tcx CodeRegion> {
let body = mir_body(tcx, def_id);
body.basic_blocks()
.iter()
.map(|data| {
data.statements.iter().filter_map(|statement| match statement.kind {
StatementKind::Coverage(box ref coverage) => {
if is_inlined(body, statement) {
None
} else {
coverage.code_region.as_ref() // may be None
}
}
_ => None,
})
})
.flatten()
.collect()
}
fn is_inlined(body: &Body<'_>, statement: &Statement<'_>) -> bool {
let scope_data = &body.source_scopes[statement.source_info.scope];
scope_data.inlined.is_some() || scope_data.inlined_parent_scope.is_some()
}
coverage bug fixes and optimization support Adjusted LLVM codegen for code compiled with `-Zinstrument-coverage` to address multiple, somewhat related issues. Fixed a significant flaw in prior coverage solution: Every counter generated a new counter variable, but there should have only been one counter variable per function. This appears to have bloated .profraw files significantly. (For a small program, it increased the size by about 40%. I have not tested large programs, but there is anecdotal evidence that profraw files were way too large. This is a good fix, regardless, but hopefully it also addresses related issues. Fixes: #82144 Invalid LLVM coverage data produced when compiled with -C opt-level=1 Existing tests now work up to at least `opt-level=3`. This required a detailed analysis of the LLVM IR, comparisons with Clang C++ LLVM IR when compiled with coverage, and a lot of trial and error with codegen adjustments. The biggest hurdle was figuring out how to continue to support coverage results for unused functions and generics. Rust's coverage results have three advantages over Clang's coverage results: 1. Rust's coverage map does not include any overlapping code regions, making coverage counting unambiguous. 2. Rust generates coverage results (showing zero counts) for all unused functions, including generics. (Clang does not generate coverage for uninstantiated template functions.) 3. Rust's unused functions produce minimal stubbed functions in LLVM IR, sufficient for including in the coverage results; while Clang must generate the complete LLVM IR for each unused function, even though it will never be called. This PR removes the previous hack of attempting to inject coverage into some other existing function instance, and generates dedicated instances for each unused function. This change, and a few other adjustments (similar to what is required for `-C link-dead-code`, but with lower impact), makes it possible to support LLVM optimizations. Fixes: #79651 Coverage report: "Unexecuted instantiation:..." for a generic function from multiple crates Fixed by removing the aforementioned hack. Some "Unexecuted instantiation" notices are unavoidable, as explained in the `used_crate.rs` test, but `-Zinstrument-coverage` has new options to back off support for either unused generics, or all unused functions, which avoids the notice, at the cost of less coverage of unused functions. Fixes: #82875 Invalid LLVM coverage data produced with crate brotli_decompressor Fixed by disabling the LLVM function attribute that forces inlining, if `-Z instrument-coverage` is enabled. This attribute is applied to Rust functions with `#[inline(always)], and in some cases, the forced inlining breaks coverage instrumentation and reports.
2021-03-15 23:32:45 +00:00
/// This function ensures we obtain the correct MIR for the given item irrespective of
/// whether that means const mir or runtime mir. For `const fn` this opts for runtime
/// mir.
fn mir_body<'tcx>(tcx: TyCtxt<'tcx>, def_id: DefId) -> &'tcx mir::Body<'tcx> {
let id = ty::WithOptConstParam::unknown(def_id);
let def = ty::InstanceDef::Item(id);
tcx.instance_mir(def)
}