Merge branch 'main' into users/evelez7/clang-doc-json-generator

Author: evelez7

.ci/metrics/metrics.py

@@ -1,12 +1,9 @@
 import collections
 import datetime
-import dateutil
 import github
-import json
 import logging
 import os
 import requests
-import sys
 import time
 
 from dataclasses import dataclass
@@ -55,18 +52,6 @@
 # by trial and error).
 GRAFANA_METRIC_MAX_AGE_MN = 120
 
-# Lists the BuildKite jobs we want to track. Maps the BuildKite job name to
-# the metric name in Grafana. This is important not to lose metrics history
-# if the workflow name changes.
-BUILDKITE_WORKFLOW_TO_TRACK = {
-    ":linux: Linux x64": "buildkite_linux",
-    ":windows: Windows x64": "buildkite_windows",
-}
-
-# Number of builds to fetch per page. Since we scrape regularly, this can
-# remain small.
-BUILDKITE_GRAPHQL_BUILDS_PER_PAGE = 50
-
 
 @dataclass
 class JobMetrics:
@@ -86,181 +71,6 @@ class GaugeMetric:
     time_ns: int
 
 
-def buildkite_fetch_page_build_list(
-    buildkite_token: str, after_cursor: str = None
-) -> list[dict[str, str]]:
-    """Fetches a page of the build list using the GraphQL BuildKite API.
-
-    Returns the BUILDKITE_GRAPHQL_BUILDS_PER_PAGE last running/queued builds,
-    or the BUILDKITE_GRAPHQL_BUILDS_PER_PAGE running/queued builds
-    older than the one pointer by |after_cursor| if provided.
-    The |after_cursor| value is taken from the previous page returned by the
-    API.
-
-    Args:
-      buildkite_token: the secret token to authenticate GraphQL requests.
-      after_cursor: cursor after which to start the page fetch.
-
-    Returns:
-      The most recent builds after cursor (if set) with the following format:
-      [
-        {
-            "cursor": <value>,
-            "number": <build-number>,
-        }
-      ]
-    """
-
-    BUILDKITE_GRAPHQL_QUERY = """
-    query OrganizationShowQuery {{
-      organization(slug: "llvm-project") {{
-        pipelines(search: "Github pull requests", first: 1) {{
-          edges {{
-            node {{
-              builds (state: [CANCELING, CREATING, FAILING, RUNNING], first: {PAGE_SIZE}, after: {AFTER}) {{
-                edges {{
-                  cursor
-                  node {{
-                    number
-                  }}
-                }}
-              }}
-            }}
-          }}
-        }}
-      }}
-    }}
-    """
-    query = BUILDKITE_GRAPHQL_QUERY.format(
-        PAGE_SIZE=BUILDKITE_GRAPHQL_BUILDS_PER_PAGE,
-        AFTER="null" if after_cursor is None else '"{}"'.format(after_cursor),
-    )
-    query = json.dumps({"query": query})
-    url = "https://graphql.buildkite.com/v1"
-    headers = {
-        "Authorization": "Bearer " + buildkite_token,
-        "Content-Type": "application/json",
-    }
-    data = requests.post(url, data=query, headers=headers).json()
-    # De-nest the build list.
-    if "errors" in data:
-        logging.info("Failed to fetch BuildKite jobs: {}".format(data["errors"]))
-        return []
-    builds = data["data"]["organization"]["pipelines"]["edges"][0]["node"]["builds"][
-        "edges"
-    ]
-    # Fold cursor info into the node dictionnary.
-    return [{**x["node"], "cursor": x["cursor"]} for x in builds]
-
-
-def buildkite_get_build_info(build_number: str) -> dict:
-    """Returns all the info associated with the provided build number.
-
-    Note: for unknown reasons, graphql returns no jobs for a given build,
-    while this endpoint does, hence why this uses this API instead of graphql.
-
-    Args:
-      build_number: which build number to fetch info for.
-
-    Returns:
-      The info for the target build, a JSON dictionnary.
-    """
-
-    URL = "https://buildkite.com/llvm-project/github-pull-requests/builds/{}.json"
-    return requests.get(URL.format(build_number)).json()
-
-
-def buildkite_get_incomplete_tasks(buildkite_token: str) -> list:
-    """Returns all the running/pending BuildKite builds.
-
-    Args:
-     buildkite_token: the secret token to authenticate GraphQL requests.
-     last_cursor: the cursor to stop at if set. If None, a full page is fetched.
-    """
-    output = []
-    cursor = None
-    while True:
-        page = buildkite_fetch_page_build_list(buildkite_token, cursor)
-        if len(page) == 0:
-            break
-        cursor = page[-1]["cursor"]
-        output += page
-    return output
-
-
-def buildkite_get_metrics(
-    buildkite_token: str, previously_incomplete: set[int]
-) -> (list[JobMetrics], set[int]):
-    """Returns a tuple with:
-
-    - the metrics recorded for newly completed workflow jobs.
-    - the set of workflow still running now.
-
-    Args:
-      buildkite_token: the secret token to authenticate GraphQL requests.
-        previously_incomplete: the set of running workflows the last time this
-        function was called.
-    """
-
-    running_builds = buildkite_get_incomplete_tasks(buildkite_token)
-    incomplete_now = set([x["number"] for x in running_builds])
-    output = []
-
-    for build_id in previously_incomplete:
-        if build_id in incomplete_now:
-            continue
-
-        info = buildkite_get_build_info(build_id)
-        metric_timestamp = dateutil.parser.isoparse(info["finished_at"])
-        for job in info["jobs"]:
-            # This workflow is not interesting to us.
-            if job["name"] not in BUILDKITE_WORKFLOW_TO_TRACK:
-                continue
-
-            # Don't count canceled jobs.
-            if job["canceled_at"]:
-                continue
-
-            created_at = dateutil.parser.isoparse(job["created_at"])
-            scheduled_at = dateutil.parser.isoparse(job["scheduled_at"])
-            started_at = dateutil.parser.isoparse(job["started_at"])
-            finished_at = dateutil.parser.isoparse(job["finished_at"])
-
-            job_name = BUILDKITE_WORKFLOW_TO_TRACK[job["name"]]
-            queue_time = (started_at - scheduled_at).seconds
-            run_time = (finished_at - started_at).seconds
-            status = bool(job["passed"])
-
-            # Grafana will refuse to ingest metrics older than ~2 hours, so we
-            # should avoid sending historical data.
-            metric_age_mn = (
-                datetime.datetime.now(datetime.timezone.utc) - metric_timestamp
-            ).total_seconds() / 60
-            if metric_age_mn > GRAFANA_METRIC_MAX_AGE_MN:
-                logging.warning(
-                    f"Job {job['name']} from workflow {build_id} dropped due"
-                    + f" to staleness: {metric_age_mn}mn old."
-                )
-                continue
-
-            metric_timestamp_ns = int(metric_timestamp.timestamp()) * 10**9
-            workflow_id = build_id
-            workflow_name = "Github pull requests"
-            output.append(
-                JobMetrics(
-                    job_name,
-                    queue_time,
-                    run_time,
-                    status,
-                    metric_timestamp_ns,
-                    workflow_id,
-                    workflow_name,
-                )
-            )
-
-    return output, incomplete_now
-
-
 def github_get_metrics(
     github_repo: github.Repository, last_workflows_seen_as_completed: set[int]
 ) -> tuple[list[JobMetrics], int]:
@@ -478,17 +288,13 @@ def upload_metrics(workflow_metrics, metrics_userid, api_key):
 def main():
     # Authenticate with Github
     github_auth = Auth.Token(os.environ["GITHUB_TOKEN"])
-    buildkite_token = os.environ["BUILDKITE_TOKEN"]
     grafana_api_key = os.environ["GRAFANA_API_KEY"]
     grafana_metrics_userid = os.environ["GRAFANA_METRICS_USERID"]
 
     # The last workflow this script processed.
     # Because the Github queries are broken, we'll simply log a 'processed'
     # bit for the last COUNT_TO_PROCESS workflows.
     gh_last_workflows_seen_as_completed = set()
-    # Stores the list of pending/running builds in BuildKite we need to check
-    # at the next iteration.
-    bk_incomplete = set()
 
     # Enter the main loop. Every five minutes we wake up and dump metrics for
     # the relevant jobs.
@@ -500,13 +306,8 @@ def main():
             github_repo, gh_last_workflows_seen_as_completed
         )
 
-        bk_metrics, bk_incomplete = buildkite_get_metrics(
-            buildkite_token, bk_incomplete
-        )
-
-        metrics = gh_metrics + bk_metrics
-        upload_metrics(metrics, grafana_metrics_userid, grafana_api_key)
-        logging.info(f"Uploaded {len(metrics)} metrics")
+        upload_metrics(gh_metrics, grafana_metrics_userid, grafana_api_key)
+        logging.info(f"Uploaded {len(gh_metrics)} metrics")
 
         time.sleep(SCRAPE_INTERVAL_SECONDS)
 

.github/new-prs-labeler.yml

@@ -554,6 +554,12 @@ flang:fir-hlfir:
 flang:codegen:
   - flang/**/CodeGen/**
 
+llvm:codegen:
+  - llvm/lib/CodeGen/*
+  - llvm/lib/CodeGen/MIRParser/*
+  - llvm/lib/CodeGen/LiveDebugValues/*
+  - llvm/lib/CodeGen/AsmPrinter/*
+
 llvm:globalisel:
   - llvm/**/GlobalISel/**
   - llvm/utils/TableGen/GlobalISel*

.github/workflows/containers/github-action-ci-windows/Dockerfile

@@ -85,7 +85,7 @@ RUN powershell -Command \
 RUN git config --system core.longpaths true & \
     git config --global core.autocrlf false
 
-ARG RUNNER_VERSION=2.324.0
+ARG RUNNER_VERSION=2.325.0
 ENV RUNNER_VERSION=$RUNNER_VERSION
 
 RUN powershell -Command \

.github/workflows/containers/github-action-ci/Dockerfile

@@ -86,7 +86,7 @@ WORKDIR /home/gha
 
 FROM ci-container as ci-container-agent
 
-ENV GITHUB_RUNNER_VERSION=2.324.0
+ENV GITHUB_RUNNER_VERSION=2.325.0
 
 RUN mkdir actions-runner && \
     cd actions-runner && \

.github/workflows/docs.yml

@@ -205,7 +205,7 @@ jobs:
           steps.docs-changed-subprojects.outputs.workflow_any_changed == 'true'
         run: |
           cmake -B flang-build -GNinja -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_PROJECTS="clang;mlir;flang" -DLLVM_ENABLE_SPHINX=ON ./llvm
-          TZ=UTC ninja -C flang-build docs-flang-html
+          TZ=UTC ninja -C flang-build docs-flang-html docs-flang-man
           mkdir built-docs/flang
           cp -r flang-build/docs/* built-docs/flang/
       - name: Upload docs

bolt/include/bolt/Core/BinaryFunction.h

@@ -2348,6 +2348,7 @@ class BinaryFunction {
       releaseCFG();
       CurrentState = State::Emitted;
     }
+    clearList(Relocations);
   }
 
   /// Process LSDA information for the function.

bolt/lib/Core/BinaryContext.cpp

@@ -1032,10 +1032,8 @@ void BinaryContext::adjustCodePadding() {
 
     if (!hasValidCodePadding(BF)) {
       if (HasRelocations) {
-        if (opts::Verbosity >= 1) {
-          this->outs() << "BOLT-INFO: function " << BF
-                       << " has invalid padding. Ignoring the function.\n";
-        }
+        this->errs() << "BOLT-WARNING: function " << BF
+                     << " has invalid padding. Ignoring the function\n";
         BF.setIgnored();
       } else {
         BF.setMaxSize(BF.getSize());

bolt/lib/Core/BinaryFunction.cpp

@@ -1526,8 +1526,6 @@ Error BinaryFunction::disassemble() {
   if (uint64_t Offset = getFirstInstructionOffset())
     Labels[Offset] = BC.Ctx->createNamedTempSymbol();
 
-  clearList(Relocations);
-
   if (!IsSimple) {
     clearList(Instructions);
     return createNonFatalBOLTError("");
@@ -3244,16 +3242,26 @@ void BinaryFunction::setTrapOnEntry() {
 }
 
 void BinaryFunction::setIgnored() {
+  IsIgnored = true;
+
   if (opts::processAllFunctions()) {
     // We can accept ignored functions before they've been disassembled.
-    // In that case, they would still get disassembled and emited, but not
+    // In that case, they would still get disassembled and emitted, but not
     // optimized.
-    assert(CurrentState == State::Empty &&
-           "cannot ignore non-empty functions in current mode");
-    IsIgnored = true;
+    if (CurrentState != State::Empty) {
+      BC.errs() << "BOLT-ERROR: cannot ignore non-empty function " << *this
+                << " in current mode\n";
+      exit(1);
+    }
     return;
   }
 
+  IsSimple = false;
+  LLVM_DEBUG(dbgs() << "Ignoring " << getPrintName() << '\n');
+
+  if (CurrentState == State::Empty)
+    return;
+
   clearDisasmState();
 
   // Clear CFG state too.
@@ -3273,9 +3281,11 @@ void BinaryFunction::setIgnored() {
 
   CurrentState = State::Empty;
 
-  IsIgnored = true;
-  IsSimple = false;
-  LLVM_DEBUG(dbgs() << "Ignoring " << getPrintName() << '\n');
+  // Fix external references in the original function body.
+  if (BC.HasRelocations) {
+    LLVM_DEBUG(dbgs() << "Scanning refs in " << *this << '\n');
+    scanExternalRefs();
+  }
 }
 
 void BinaryFunction::duplicateConstantIslands() {
@@ -3764,7 +3774,6 @@ void BinaryFunction::postProcessBranches() {
 
 MCSymbol *BinaryFunction::addEntryPointAtOffset(uint64_t Offset) {
   assert(Offset && "cannot add primary entry point");
-  assert(CurrentState == State::Empty || CurrentState == State::Disassembled);
 
   const uint64_t EntryPointAddress = getAddress() + Offset;
   MCSymbol *LocalSymbol = getOrCreateLocalLabel(EntryPointAddress);
@@ -3773,6 +3782,8 @@ MCSymbol *BinaryFunction::addEntryPointAtOffset(uint64_t Offset) {
   if (EntrySymbol)
     return EntrySymbol;
 
+  assert(CurrentState == State::Empty || CurrentState == State::Disassembled);
+
   if (BinaryData *EntryBD = BC.getBinaryDataAtAddress(EntryPointAddress)) {
     EntrySymbol = EntryBD->getSymbol();
   } else {