Merge pull request #517 from Xilinx/bump_to_0195ec45

[AutoBump] Merge with 0195ec4 (Jan 15) (46)

Author: jorickert

.ci/metrics/metrics.py

@@ -26,7 +26,67 @@ class JobMetrics:
     workflow_id: int
 
 
-def get_metrics(github_repo: github.Repository, workflows_to_track: dict[str, int]):
+@dataclass
+class GaugeMetric:
+    name: str
+    value: int
+    time_ns: int
+
+
+def get_sampled_workflow_metrics(github_repo: github.Repository):
+    """Gets global statistics about the Github workflow queue
+
+    Args:
+      github_repo: A github repo object to use to query the relevant information.
+
+    Returns:
+      Returns a list of GaugeMetric objects, containing the relevant metrics about
+      the workflow
+    """
+
+    # Other states are available (pending, waiting, etc), but the meaning
+    # is not documented (See #70540).
+    # "queued" seems to be the info we want.
+    queued_workflow_count = len(
+        [
+            x
+            for x in github_repo.get_workflow_runs(status="queued")
+            if x.name in WORKFLOWS_TO_TRACK
+        ]
+    )
+    running_workflow_count = len(
+        [
+            x
+            for x in github_repo.get_workflow_runs(status="in_progress")
+            if x.name in WORKFLOWS_TO_TRACK
+        ]
+    )
+
+    workflow_metrics = []
+    workflow_metrics.append(
+        GaugeMetric(
+            "workflow_queue_size",
+            queued_workflow_count,
+            time.time_ns(),
+        )
+    )
+    workflow_metrics.append(
+        GaugeMetric(
+            "running_workflow_count",
+            running_workflow_count,
+            time.time_ns(),
+        )
+    )
+    # Always send a hearbeat metric so we can monitor is this container is still able to log to Grafana.
+    workflow_metrics.append(
+        GaugeMetric("metrics_container_heartbeat", 1, time.time_ns())
+    )
+    return workflow_metrics
+
+
+def get_per_workflow_metrics(
+    github_repo: github.Repository, workflows_to_track: dict[str, int]
+):
     """Gets the metrics for specified Github workflows.
 
     This function takes in a list of workflows to track, and optionally the
@@ -43,14 +103,14 @@ def get_metrics(github_repo: github.Repository, workflows_to_track: dict[str, in
       Returns a list of JobMetrics objects, containing the relevant metrics about
       the workflow.
     """
-    workflow_runs = iter(github_repo.get_workflow_runs())
-
     workflow_metrics = []
 
     workflows_to_include = set(workflows_to_track.keys())
 
-    while len(workflows_to_include) > 0:
-        workflow_run = next(workflow_runs)
+    for workflow_run in iter(github_repo.get_workflow_runs()):
+        if len(workflows_to_include) == 0:
+            break
+
         if workflow_run.status != "completed":
             continue
 
@@ -139,12 +199,27 @@ def upload_metrics(workflow_metrics, metrics_userid, api_key):
       metrics_userid: The userid to use for the upload.
       api_key: The API key to use for the upload.
     """
+
+    if len(workflow_metrics) == 0:
+        print("No metrics found to upload.", file=sys.stderr)
+        return
+
     metrics_batch = []
     for workflow_metric in workflow_metrics:
-        workflow_formatted_name = workflow_metric.job_name.lower().replace(" ", "_")
-        metrics_batch.append(
-            f"{workflow_formatted_name} queue_time={workflow_metric.queue_time},run_time={workflow_metric.run_time},status={workflow_metric.status} {workflow_metric.created_at_ns}"
-        )
+        if isinstance(workflow_metric, GaugeMetric):
+            name = workflow_metric.name.lower().replace(" ", "_")
+            metrics_batch.append(
+                f"{name} value={workflow_metric.value} {workflow_metric.time_ns}"
+            )
+        elif isinstance(workflow_metric, JobMetrics):
+            name = workflow_metric.job_name.lower().replace(" ", "_")
+            metrics_batch.append(
+                f"{name} queue_time={workflow_metric.queue_time},run_time={workflow_metric.run_time},status={workflow_metric.status} {workflow_metric.created_at_ns}"
+            )
+        else:
+            raise ValueError(
+                f"Unsupported object type {type(workflow_metric)}: {str(workflow_metric)}"
+            )
 
     request_data = "\n".join(metrics_batch)
     response = requests.post(
@@ -176,16 +251,21 @@ def main():
     # Enter the main loop. Every five minutes we wake up and dump metrics for
     # the relevant jobs.
     while True:
-        current_metrics = get_metrics(github_repo, workflows_to_track)
-        if len(current_metrics) == 0:
-            print("No metrics found to upload.", file=sys.stderr)
-            continue
+        current_metrics = get_per_workflow_metrics(github_repo, workflows_to_track)
+        current_metrics += get_sampled_workflow_metrics(github_repo)
+        # Always send a hearbeat metric so we can monitor is this container is still able to log to Grafana.
+        current_metrics.append(
+            GaugeMetric("metrics_container_heartbeat", 1, time.time_ns())
+        )
 
         upload_metrics(current_metrics, grafana_metrics_userid, grafana_api_key)
         print(f"Uploaded {len(current_metrics)} metrics", file=sys.stderr)
 
         for workflow_metric in reversed(current_metrics):
-            workflows_to_track[workflow_metric.job_name] = workflow_metric.workflow_id
+            if isinstance(workflow_metric, JobMetrics):
+                workflows_to_track[
+                    workflow_metric.job_name
+                ] = workflow_metric.workflow_id
 
         time.sleep(SCRAPE_INTERVAL_SECONDS)
 

bolt/CMakeLists.txt

@@ -163,8 +163,8 @@ if (BOLT_ENABLE_RUNTIME)
   add_llvm_install_targets(install-bolt_rt
     DEPENDS bolt_rt bolt
     COMPONENT bolt)
-  set(LIBBOLT_RT_INSTR "${CMAKE_CURRENT_BINARY_DIR}/bolt_rt-bins/lib/libbolt_rt_instr.a")
-  set(LIBBOLT_RT_HUGIFY "${CMAKE_CURRENT_BINARY_DIR}/bolt_rt-bins/lib/libbolt_rt_hugify.a")
+  set(LIBBOLT_RT_INSTR "${CMAKE_CURRENT_BINARY_DIR}/bolt_rt-bins/lib${LLVM_LIBDIR_SUFFIX}/libbolt_rt_instr.a")
+  set(LIBBOLT_RT_HUGIFY "${CMAKE_CURRENT_BINARY_DIR}/bolt_rt-bins/lib${LLVM_LIBDIR_SUFFIX}/libbolt_rt_hugify.a")
 endif()
 
 find_program(GNU_LD_EXECUTABLE NAMES ${LLVM_DEFAULT_TARGET_TRIPLE}-ld.bfd ld.bfd DOC "GNU ld")

bolt/include/bolt/Core/BinaryContext.h

@@ -1363,6 +1363,12 @@ class BinaryContext {
     if (std::optional<uint32_t> Size = MIB->getSize(Inst))
       return *Size;
 
+    if (MIB->isPseudo(Inst))
+      return 0;
+
+    if (std::optional<uint32_t> Size = MIB->getInstructionSize(Inst))
+      return *Size;
+
     if (!Emitter)
       Emitter = this->MCE.get();
     SmallString<256> Code;

bolt/include/bolt/Core/MCPlusBuilder.h

@@ -1204,6 +1204,11 @@ class MCPlusBuilder {
   /// Get instruction size specified via annotation.
   std::optional<uint32_t> getSize(const MCInst &Inst) const;
 
+  /// Get target-specific instruction size.
+  virtual std::optional<uint32_t> getInstructionSize(const MCInst &Inst) const {
+    return std::nullopt;
+  }
+
   /// Set instruction size.
   void setSize(MCInst &Inst, uint32_t Size) const;
 

bolt/lib/Passes/Inliner.cpp

@@ -310,13 +310,13 @@ Inliner::inlineCall(BinaryBasicBlock &CallerBB,
       if (MIB.isPseudo(Inst))
         continue;
 
-      MIB.stripAnnotations(Inst, /*KeepTC=*/BC.isX86());
+      MIB.stripAnnotations(Inst, /*KeepTC=*/BC.isX86() || BC.isAArch64());
 
       // Fix branch target. Strictly speaking, we don't have to do this as
       // targets of direct branches will be fixed later and don't matter
       // in the CFG state. However, disassembly may look misleading, and
       // hence we do the fixing.
-      if (MIB.isBranch(Inst)) {
+      if (MIB.isBranch(Inst) && !MIB.isTailCall(Inst)) {
         assert(!MIB.isIndirectBranch(Inst) &&
                "unexpected indirect branch in callee");
         const BinaryBasicBlock *TargetBB =

bolt/lib/Target/AArch64/AArch64MCPlusBuilder.cpp

@@ -133,6 +133,36 @@ class AArch64MCPlusBuilder : public MCPlusBuilder {
 public:
   using MCPlusBuilder::MCPlusBuilder;
 
+  MCPhysReg getStackPointer() const override { return AArch64::SP; }
+
+  bool isPush(const MCInst &Inst) const override { return false; }
+
+  bool isPop(const MCInst &Inst) const override { return false; }
+
+  void createCall(MCInst &Inst, const MCSymbol *Target,
+                  MCContext *Ctx) override {
+    createDirectCall(Inst, Target, Ctx, false);
+  }
+
+  bool convertTailCallToCall(MCInst &Inst) override {
+    int NewOpcode;
+    switch (Inst.getOpcode()) {
+    default:
+      return false;
+    case AArch64::B:
+      NewOpcode = AArch64::BL;
+      break;
+    case AArch64::BR:
+      NewOpcode = AArch64::BLR;
+      break;
+    }
+
+    Inst.setOpcode(NewOpcode);
+    removeAnnotation(Inst, MCPlus::MCAnnotation::kTailCall);
+    clearOffset(Inst);
+    return true;
+  }
+
   bool equals(const MCTargetExpr &A, const MCTargetExpr &B,
               CompFuncTy Comp) const override {
     const auto &AArch64ExprA = cast<AArch64MCExpr>(A);
@@ -1792,6 +1822,11 @@ class AArch64MCPlusBuilder : public MCPlusBuilder {
   }
 
   uint16_t getMinFunctionAlignment() const override { return 4; }
+
+  std::optional<uint32_t>
+  getInstructionSize(const MCInst &Inst) const override {
+    return 4;
+  }
 };
 
 } // end anonymous namespace

bolt/runtime/CMakeLists.txt

@@ -16,18 +16,18 @@ add_library(bolt_rt_instr STATIC
   instr.cpp
   ${CMAKE_CURRENT_BINARY_DIR}/config.h
   )
-set_target_properties(bolt_rt_instr PROPERTIES ARCHIVE_OUTPUT_DIRECTORY "${CMAKE_INSTALL_LIBDIR}${LLVM_LIBDIR_SUFFIX}")
+set_target_properties(bolt_rt_instr PROPERTIES ARCHIVE_OUTPUT_DIRECTORY "lib${LLVM_LIBDIR_SUFFIX}")
 add_library(bolt_rt_hugify STATIC
   hugify.cpp
   ${CMAKE_CURRENT_BINARY_DIR}/config.h
   )
-set_target_properties(bolt_rt_hugify PROPERTIES ARCHIVE_OUTPUT_DIRECTORY "${CMAKE_INSTALL_LIBDIR}${LLVM_LIBDIR_SUFFIX}")
+set_target_properties(bolt_rt_hugify PROPERTIES ARCHIVE_OUTPUT_DIRECTORY "lib${LLVM_LIBDIR_SUFFIX}")
 
 if(NOT BOLT_BUILT_STANDALONE)
   add_custom_command(TARGET bolt_rt_instr POST_BUILD
-    COMMAND ${CMAKE_COMMAND} -E copy "${CMAKE_CURRENT_BINARY_DIR}/lib/libbolt_rt_instr.a" "${LLVM_LIBRARY_DIR}")
+    COMMAND ${CMAKE_COMMAND} -E copy "${CMAKE_CURRENT_BINARY_DIR}/lib${LLVM_LIBDIR_SUFFIX}/libbolt_rt_instr.a" "${LLVM_LIBRARY_DIR}")
   add_custom_command(TARGET bolt_rt_hugify POST_BUILD
-    COMMAND ${CMAKE_COMMAND} -E copy "${CMAKE_CURRENT_BINARY_DIR}/lib/libbolt_rt_hugify.a" "${LLVM_LIBRARY_DIR}")
+    COMMAND ${CMAKE_COMMAND} -E copy "${CMAKE_CURRENT_BINARY_DIR}/lib${LLVM_LIBDIR_SUFFIX}/libbolt_rt_hugify.a" "${LLVM_LIBRARY_DIR}")
 endif()
 
 set(BOLT_RT_FLAGS
@@ -53,23 +53,23 @@ target_include_directories(bolt_rt_instr PRIVATE ${CMAKE_CURRENT_BINARY_DIR})
 target_compile_options(bolt_rt_hugify PRIVATE ${BOLT_RT_FLAGS})
 target_include_directories(bolt_rt_hugify PRIVATE ${CMAKE_CURRENT_BINARY_DIR})
 
-install(TARGETS bolt_rt_instr DESTINATION "${CMAKE_INSTALL_LIBDIR}${LLVM_LIBDIR_SUFFIX}")
-install(TARGETS bolt_rt_hugify DESTINATION "${CMAKE_INSTALL_LIBDIR}${LLVM_LIBDIR_SUFFIX}")
+install(TARGETS bolt_rt_instr DESTINATION "lib${LLVM_LIBDIR_SUFFIX}")
+install(TARGETS bolt_rt_hugify DESTINATION "lib${LLVM_LIBDIR_SUFFIX}")
 
 if (CMAKE_CXX_COMPILER_ID MATCHES ".*Clang.*" AND CMAKE_SYSTEM_NAME STREQUAL "Darwin")
   add_library(bolt_rt_instr_osx STATIC
     instr.cpp
     ${CMAKE_CURRENT_BINARY_DIR}/config.h
   )
-  set_target_properties(bolt_rt_instr_osx PROPERTIES ARCHIVE_OUTPUT_DIRECTORY "${CMAKE_INSTALL_LIBDIR}${LLVM_LIBDIR_SUFFIX}")
+  set_target_properties(bolt_rt_instr_osx PROPERTIES ARCHIVE_OUTPUT_DIRECTORY "lib${LLVM_LIBDIR_SUFFIX}")
   target_include_directories(bolt_rt_instr_osx PRIVATE ${CMAKE_CURRENT_BINARY_DIR})
   target_compile_options(bolt_rt_instr_osx PRIVATE
     -target x86_64-apple-darwin19.6.0
     ${BOLT_RT_FLAGS})
-  install(TARGETS bolt_rt_instr_osx DESTINATION "${CMAKE_INSTALL_LIBDIR}${LLVM_LIBDIR_SUFFIX}")
+  install(TARGETS bolt_rt_instr_osx DESTINATION "lib${LLVM_LIBDIR_SUFFIX}")
 
   if(NOT BOLT_BUILT_STANDALONE)
     add_custom_command(TARGET bolt_rt_instr_osx POST_BUILD
-      COMMAND ${CMAKE_COMMAND} -E copy "${CMAKE_CURRENT_BINARY_DIR}/lib/libbolt_rt_instr_osx.a" "${LLVM_LIBRARY_DIR}")
+      COMMAND ${CMAKE_COMMAND} -E copy "${CMAKE_CURRENT_BINARY_DIR}/lib${LLVM_LIBDIR_SUFFIX}/libbolt_rt_instr_osx.a" "${LLVM_LIBRARY_DIR}")
   endif()
 endif()

bolt/test/AArch64/inline-small-function-1.s

@@ -0,0 +1,42 @@
+## This test checks that inline is properly handled by BOLT on aarch64.
+
+# REQUIRES: system-linux
+
+# RUN: llvm-mc -filetype=obj -triple aarch64-unknown-unknown %s -o %t.o
+# RUN: %clang %cflags -O0 %t.o -o %t.exe -Wl,-q
+# RUN: llvm-bolt --inline-small-functions --print-inline  --print-only=_Z3barP1A  \
+# RUN: %t.exe -o %t.bolt  | FileCheck %s
+
+# CHECK: BOLT-INFO: inlined 0 calls at 1 call sites in 2 iteration(s). Change in binary size: 4 bytes.
+# CHECK: Binary Function "_Z3barP1A" after inlining {
+# CHECK-NOT: bl	_Z3fooP1A
+# CHECK: ldr	x8, [x0]
+# CHECK-NEXT: ldr	w0, [x8]
+ 
+	.text
+	.globl	_Z3fooP1A                      
+	.type	_Z3fooP1A,@function
+_Z3fooP1A:                              
+	ldr	x8, [x0]
+	ldr	w0, [x8]
+	ret
+	.size	_Z3fooP1A, .-_Z3fooP1A
+
+	.globl	_Z3barP1A                       
+	.type	_Z3barP1A,@function
+_Z3barP1A:                              
+	stp	x29, x30, [sp, #-16]!           
+	mov	x29, sp
+	bl	_Z3fooP1A
+	mul	w0, w0, w0
+	ldp	x29, x30, [sp], #16             
+	ret
+	.size	_Z3barP1A, .-_Z3barP1A
+
+	.globl	main                            
+	.p2align	2
+	.type	main,@function
+main:                                   
+	mov	w0, wzr
+	ret
+	.size	main, .-main

bolt/test/AArch64/inline-small-function-2.s

@@ -0,0 +1,48 @@
+## This test checks that inline is properly handled by BOLT on aarch64.
+
+# REQUIRES: system-linux
+
+# RUN: llvm-mc -filetype=obj -triple aarch64-unknown-unknown %s -o %t.o
+# RUN: %clang %cflags -O0 %t.o -o %t.exe -Wl,-q
+# RUN: llvm-bolt --inline-small-functions --print-inline  --print-only=test  \
+# RUN: %t.exe -o %t.bolt | FileCheck %s
+
+#CHECK: BOLT-INFO: inlined 0 calls at 1 call sites in 2 iteration(s). Change in binary size: 4 bytes.
+#CHECK: Binary Function "test" after inlining {
+#CHECK-NOT: bl	indirect
+#CHECK: add	w0, w1, w0
+#CHECK-NEXT: blr	x2
+ 
+	.text
+	.globl	indirect                       
+	.type	indirect,@function
+indirect:                               
+	add	w0, w1, w0
+	br	x2
+	.size	indirect, .-indirect
+
+	.globl	test                           
+	.type	test,@function
+test:                                   
+	stp	x29, x30, [sp, #-32]!          
+	stp	x20, x19, [sp, #16]            
+	mov	x29, sp
+	mov	w19, w1
+	mov	w20, w0
+	bl	indirect
+	add	w8, w19, w20
+	cmp	w0, #0
+	csinc	w0, w8, wzr, eq
+	ldp	x20, x19, [sp, #16]             
+	ldp	x29, x30, [sp], #32            
+	ret
+	.size	test, .-test
+
+	.globl	main                            
+	.type	main,@function
+main:                                   
+	mov	w0, wzr
+	ret
+	.size	main, .-main
+
+ 

bolt/test/lit.local.cfg

@@ -1,5 +1,5 @@
 host_linux_triple = config.target_triple.split("-")[0] + "-unknown-linux-gnu"
-common_linker_flags = "-fuse-ld=lld -Wl,--unresolved-symbols=ignore-all -pie"
+common_linker_flags = "-fuse-ld=lld -Wl,--unresolved-symbols=ignore-all -Wl,--build-id=none -pie"
 flags = f"--target={host_linux_triple} -fPIE {common_linker_flags}"
 
 config.substitutions.insert(0, ("%cflags", f"%cflags {flags}"))