Merge from 'main' to 'sycl-web' (126 commits)

Author: iclsrc

.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*

bolt/lib/Passes/ADRRelaxationPass.cpp

@@ -81,17 +81,15 @@ void ADRRelaxationPass::runOnFunction(BinaryFunction &BF) {
         It = BB.eraseInstruction(std::prev(It));
       } else if (std::next(It) != BB.end() && BC.MIB->isNoop(*std::next(It))) {
         BB.eraseInstruction(std::next(It));
-      } else if (!opts::StrictMode && !BF.isSimple()) {
+      } else if (!BF.isSimple()) {
         // If the function is not simple, it may contain a jump table undetected
         // by us. This jump table may use an offset from the branch instruction
         // to land in the desired place. If we add new instructions, we
         // invalidate this offset, so we have to rely on linker-inserted NOP to
         // replace it with ADRP, and abort if it is not present.
         auto L = BC.scopeLock();
-        BC.errs() << formatv(
-            "BOLT-ERROR: Cannot relax adr in non-simple function "
-            "{0}. Use --strict option to override\n",
-            BF.getOneName());
+        BC.errs() << "BOLT-ERROR: cannot relax ADR in non-simple function "
+                  << BF << '\n';
         PassFailed = true;
         return;
       }

bolt/test/AArch64/adr-relaxation-fail.s

@@ -0,0 +1,33 @@
+## Check that llvm-bolt generates a proper error message when ADR instruction
+## cannot be relaxed.
+
+# RUN: llvm-mc -filetype=obj -triple aarch64-unknown-unknown %s -o %t.o
+# RUN: %clang %cflags %t.o -o %t.exe -Wl,-q -static
+# RUN: not llvm-bolt %t.exe -o %t.bolt 2>&1 | FileCheck %s
+# RUN: not llvm-bolt %t.exe -o %t.bolt --strict 2>&1 | FileCheck %s
+
+# CHECK: BOLT-ERROR: cannot relax ADR in non-simple function _start
+
+## The function contains unknown control flow and llvm-bolt fails to recover
+## CFG. As BOLT has to preserve the function layout, the ADR instruction cannot
+## be relaxed into ADRP+ADD.
+  .text
+  .globl _start
+  .type _start, %function
+_start:
+  .cfi_startproc
+  adr x1, foo
+  adr x2, .L1
+.L1:
+  br x0
+  ret  x0
+  .cfi_endproc
+  .size _start, .-_start
+
+  .globl foo
+  .type foo, %function
+foo:
+  .cfi_startproc
+  ret  x0
+  .cfi_endproc
+  .size foo, .-foo

bolt/test/AArch64/adr-relaxation.s

@@ -3,7 +3,7 @@
 # RUN: llvm-mc -filetype=obj -triple aarch64-unknown-unknown %s -o %t.o
 # RUN: %clang %cflags %t.o -o %t.exe -Wl,-q -static
 # RUN: llvm-bolt %t.exe -o %t.bolt --split-functions --split-strategy=random2
-# RUN: llvm-objdump -d --disassemble-symbols=_start %t.bolt | FileCheck %s
+# RUN: llvm-objdump -d -j .text %t.bolt | FileCheck %s
 # RUN: llvm-objdump -d --disassemble-symbols=foo.cold.0 %t.bolt \
 # RUN:   | FileCheck --check-prefix=CHECK-FOO %s
 
@@ -13,11 +13,10 @@
   .globl _start
   .type _start, %function
 _start:
-# CHECK: <_start>:
   .cfi_startproc
+# CHECK: <_start>:
+# CHECK-NEXT: adr
   adr x1, _start
-# CHECK-NOT: adrp
-# CHECK: adr
   cmp  x1, x11
   b.hi  .L1
   mov  x0, #0x0
@@ -38,12 +37,27 @@ foo:
   mov  x0, #0x0
 .L2:
 # CHECK-FOO: <foo.cold.0>:
-  adr x1, foo
-# CHECK-FOO: adrp
+# CHECK-FOO-NEXT: adrp
 # CHECK-FOO-NEXT: add
+  adr x1, foo
   ret  x30
   .cfi_endproc
 .size foo, .-foo
 
-## Force relocation mode.
-  .reloc 0, R_AARCH64_NONE
+## bar is a non-simple function. We can still relax ADR, because it has a
+## preceding NOP.
+  .globl bar
+  .type bar, %function
+bar:
+  .cfi_startproc
+# CHECK-LABEL: <bar>:
+# CHECK-NEXT: adrp
+# CHECK-NEXT: add
+  nop
+  adr x0, foo
+  adr x1, .L3
+  br x1
+.L3:
+  ret  x0
+  .cfi_endproc
+  .size bar, .-bar

bolt/test/AArch64/r_aarch64_prelxx.s

@@ -12,16 +12,14 @@
 // CHECKPREL-NEXT:  R_AARCH64_PREL32      {{.*}} _start + 4
 // CHECKPREL-NEXT:  R_AARCH64_PREL64      {{.*}} _start + 8
 
-// RUN: llvm-bolt %t.exe -o %t.bolt --strict
+// RUN: llvm-bolt %t.exe -o %t.bolt
 // RUN: llvm-objdump -D %t.bolt | FileCheck %s --check-prefix=CHECKPREL32
 
 // CHECKPREL32: [[#%x,DATATABLEADDR:]] <datatable>:
 // CHECKPREL32-NEXT: 00:
 // CHECKPREL32-NEXT: 04: {{.*}} .word 0x[[#%x,VALUE:]]
 
-// 4 is offset in datatable
-// 8 is addend
-// CHECKPREL32: [[#DATATABLEADDR + 4 - 8 + VALUE]] <_start>:
+// CHECKPREL32: [[#DATATABLEADDR + VALUE]] <_start>:
 
 // RUN: llvm-objdump -D %t.bolt | FileCheck %s --check-prefix=CHECKPREL64
 // CHECKPREL64: [[#%x,DATATABLEADDR:]] <datatable>:
@@ -30,16 +28,15 @@
 // CHECKPREL64-NEXT: 08: {{.*}} .word 0x[[#%x,VALUE:]]
 // CHECKPREL64-NEXT: 0c: {{.*}} .word 0x00000000
 
-// 8 is offset in datatable
-// 12 is addend
-// CHECKPREL64: [[#DATATABLEADDR + 8 - 12 + VALUE]] <_start>:
+// CHECKPREL64: [[#DATATABLEADDR + VALUE]] <_start>:
 
   .section .text
   .align 4
   .globl _start
   .type _start, %function
 _start:
-  adr x0, datatable
+  adrp x0, datatable
+  add x0, x0, :lo12:datable
   mov x0, #0
   ret 
 

bolt/test/AArch64/test-indirect-branch.s

@@ -7,8 +7,8 @@
 
 // RUN: llvm-mc -filetype=obj -triple aarch64-unknown-unknown -mattr=+pauth %s -o %t.o
 // RUN: %clang %cflags --target=aarch64-unknown-linux %t.o -o %t.exe -Wl,-q
-// RUN: llvm-bolt %t.exe -o %t.bolt --print-cfg --strict --debug-only=mcplus \
-// RUN:  -v=1 2>&1 | FileCheck %s
+// RUN: llvm-bolt %t.exe -o %t.bolt --print-cfg --debug-only=mcplus -v=1 2>&1 \
+// RUN:   | FileCheck %s
 
 // Pattern 1: there is no shift amount after the 'add' instruction.
 //
@@ -45,6 +45,7 @@ _start:
   .type  test1, %function
 test1:
   mov     x1, #0
+  nop
   adr     x3, datatable
   add     x3, x3, x1, lsl #2
   ldr     w2, [x3]
@@ -56,6 +57,11 @@ test1_1:
    ret
 test1_2:
    ret
+   .size test1, .-test1
+
+// Temporary workaround for PC-relative relocations from datatable leaking into
+// test2 function and creating phantom entry points.
+.skip 0x100
 
 // Pattern 2
 // CHECK: BOLT-DEBUG: failed to match indirect branch: nop/adr instead of adrp/add
@@ -65,6 +71,7 @@ test2:
   nop
   adr     x3, jump_table
   ldrh    w3, [x3, x1, lsl #1]
+  nop
   adr     x1, test2_0
   add     x3, x1, w3, sxth #2
   br      x3

bolt/test/runtime/AArch64/adrrelaxationpass.s

@@ -9,6 +9,7 @@
 #include "BitcodeReader.h"
 #include "llvm/ADT/IndexedMap.h"
 #include "llvm/Support/Error.h"
+#include "llvm/Support/TimeProfiler.h"
 #include "llvm/Support/raw_ostream.h"
 #include <optional>
 
@@ -672,6 +673,7 @@ llvm::Error ClangDocBitcodeReader::readRecord(unsigned ID, T I) {
 
 template <>
 llvm::Error ClangDocBitcodeReader::readRecord(unsigned ID, Reference *I) {
+  llvm::TimeTraceScope("Reducing infos", "readRecord");
   Record R;
   llvm::StringRef Blob;
   llvm::Expected<unsigned> MaybeRecID = Stream.readRecord(ID, R, &Blob);
@@ -683,6 +685,7 @@ llvm::Error ClangDocBitcodeReader::readRecord(unsigned ID, Reference *I) {
 // Read a block of records into a single info.
 template <typename T>
 llvm::Error ClangDocBitcodeReader::readBlock(unsigned ID, T I) {
+  llvm::TimeTraceScope("Reducing infos", "readBlock");
   if (llvm::Error Err = Stream.EnterSubBlock(ID))
     return Err;
 
@@ -713,6 +716,7 @@ llvm::Error ClangDocBitcodeReader::readBlock(unsigned ID, T I) {
 
 template <typename T>
 llvm::Error ClangDocBitcodeReader::readSubBlock(unsigned ID, T I) {
+  llvm::TimeTraceScope("Reducing infos", "readSubBlock");
   switch (ID) {
   // Blocks can only have certain types of sub blocks.
   case BI_COMMENT_BLOCK_ID: {
@@ -819,6 +823,7 @@ llvm::Error ClangDocBitcodeReader::readSubBlock(unsigned ID, T I) {
 
 ClangDocBitcodeReader::Cursor
 ClangDocBitcodeReader::skipUntilRecordOrBlock(unsigned &BlockOrRecordID) {
+  llvm::TimeTraceScope("Reducing infos", "skipUntilRecordOrBlock");
   BlockOrRecordID = 0;
 
   while (!Stream.AtEndOfStream()) {
@@ -880,6 +885,7 @@ llvm::Error ClangDocBitcodeReader::validateStream() {
 }
 
 llvm::Error ClangDocBitcodeReader::readBlockInfoBlock() {
+  llvm::TimeTraceScope("Reducing infos", "readBlockInfoBlock");
   Expected<std::optional<llvm::BitstreamBlockInfo>> MaybeBlockInfo =
       Stream.ReadBlockInfoBlock();
   if (!MaybeBlockInfo)
@@ -895,6 +901,7 @@ llvm::Error ClangDocBitcodeReader::readBlockInfoBlock() {
 template <typename T>
 llvm::Expected<std::unique_ptr<Info>>
 ClangDocBitcodeReader::createInfo(unsigned ID) {
+  llvm::TimeTraceScope("Reducing infos", "createInfo");
   std::unique_ptr<Info> I = std::make_unique<T>();
   if (auto Err = readBlock(ID, static_cast<T *>(I.get())))
     return std::move(Err);
@@ -903,6 +910,7 @@ ClangDocBitcodeReader::createInfo(unsigned ID) {
 
 llvm::Expected<std::unique_ptr<Info>>
 ClangDocBitcodeReader::readBlockToInfo(unsigned ID) {
+  llvm::TimeTraceScope("Reducing infos", "readBlockToInfo");
   switch (ID) {
   case BI_NAMESPACE_BLOCK_ID:
     return createInfo<NamespaceInfo>(ID);

clang-tools-extra/clang-doc/BitcodeReader.cpp

@@ -19,6 +19,7 @@
 #include "llvm/Support/MemoryBuffer.h"
 #include "llvm/Support/Mustache.h"
 #include "llvm/Support/Path.h"
+#include "llvm/Support/TimeProfiler.h"
 
 using namespace llvm;
 using namespace llvm::json;
@@ -125,13 +126,18 @@ static Error setupTemplateFiles(const clang::doc::ClangDocContext &CDCtx) {
 Error MustacheHTMLGenerator::generateDocs(
     StringRef RootDir, StringMap<std::unique_ptr<doc::Info>> Infos,
     const clang::doc::ClangDocContext &CDCtx) {
-  if (auto Err = setupTemplateFiles(CDCtx))
-    return Err;
+  {
+    llvm::TimeTraceScope TS("Setup Templates");
+    if (auto Err = setupTemplateFiles(CDCtx))
+      return Err;
+  }
+
   // Track which directories we already tried to create.
   StringSet<> CreatedDirs;
   // Collect all output by file name and create the necessary directories.
   StringMap<std::vector<doc::Info *>> FileToInfos;
   for (const auto &Group : Infos) {
+    llvm::TimeTraceScope TS("setup directories");
     doc::Info *Info = Group.getValue().get();
 
     SmallString<128> Path;
@@ -148,15 +154,19 @@ Error MustacheHTMLGenerator::generateDocs(
     FileToInfos[Path].push_back(Info);
   }
 
-  for (const auto &Group : FileToInfos) {
-    std::error_code FileErr;
-    raw_fd_ostream InfoOS(Group.getKey(), FileErr, sys::fs::OF_None);
-    if (FileErr)
-      return createFileOpenError(Group.getKey(), FileErr);
-
-    for (const auto &Info : Group.getValue())
-      if (Error Err = generateDocForInfo(Info, InfoOS, CDCtx))
-        return Err;
+  {
+    llvm::TimeTraceScope TS("Generate Docs");
+    for (const auto &Group : FileToInfos) {
+      llvm::TimeTraceScope TS("Info to Doc");
+      std::error_code FileErr;
+      raw_fd_ostream InfoOS(Group.getKey(), FileErr, sys::fs::OF_None);
+      if (FileErr)
+        return createFileOpenError(Group.getKey(), FileErr);
+
+      for (const auto &Info : Group.getValue())
+        if (Error Err = generateDocForInfo(Info, InfoOS, CDCtx))
+          return Err;
+    }
   }
   return Error::success();
 }