llvm: Cleanups and improvements

Changes:
- Shrink the 32bit build significantly by limiting the target support to X86 and AMDGPU, and also only building LLVM. Also disable several additional features not needed for the 32bit build
- And ARM and WebAssembly target support to the 64bit build for users cross-compiling
- Get ccache working properly, and add debugging information so that it's possible to see how effective it is.
- Disable building with PGO (this is temporary and will be re-enabled post LLVM 19)
- Add openmp support (libomp)
- Add clang-tools-extra (clang-tidy plus many other tools) so that we can build a more generally useful LLVM distribution for users.
- Create `clang-tools` subpackage and move all of the build artifacts of `clang-tools-extra` there as well as clang-analyzer and other tools that were in `clang` that were not strictly necessary for compilation
- Ensure that subpackage rundeps effectively pull in the packages they need. For instance `clang-devel` now depends on `llvm-devel`
- Update various package descriptions
- Build lld, clang, and llvm-ar so that they use the mimalloc memory allocator (improves performance by ~7%)
- Patch LLD so that it always behaves as though `--build-id=sha1` was passed (this matches binutils/mold)
- Patch LLD so that it always behaves as though `--compress-debug-symbols=zstd` was passed (this matches binutils/mold)

Signed-off-by: Reilly Brogan <[email protected]>

l/llvm/pkg/defs.h.in

@@ -0,0 +1,9 @@
+#include <bits/wordsize.h>
+
+#if __WORDSIZE == 32
+#include "32/@@REPLACE@@"
+#elif __WORDSIZE == 64
+#include "64/@@REPLACE@@"
+#else
+#error "Unknown word size"
+#endif

l/llvm/pkg/llvm-config.h

@@ -1,9 +1,9 @@
 #include <bits/wordsize.h>
 
 #if __WORDSIZE == 32
-#include "llvm-config-32.h"
+#include "llvm-config32.h"
 #elif __WORDSIZE == 64
-#include "llvm-config-64.h"
+#include "llvm-config64.h"
 #else
 #error "Unknown word size"
 #endif

l/llvm/pkg/serpent/0001-Build-lld-libraries-shared.patch → l/llvm/pkg/patches/build/0001-Build-lld-libraries-shared.patch

@@ -3,6 +3,11 @@ From: Reilly Brogan <[email protected]>
 Date: Fri, 2 Aug 2024 15:55:23 -0500
 Subject: [PATCH] Build lld libraries shared
 
+We don't want to enable `ENABLE_SHARED` since it causes every LLVM library to be built shared even if they're intended to be private.
+
+Origin: vendor
+Forwarded: not-needed
+Last-Update: 2025-01-10
 ---
  lld/cmake/modules/AddLLD.cmake | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)

l/llvm/pkg/serpent/0001-Build-polly-libraries-shared.patch → l/llvm/pkg/patches/build/0001-Build-polly-libraries-shared.patch

@@ -3,6 +3,9 @@ From: Reilly Brogan <[email protected]>
 Date: Fri, 2 Aug 2024 17:24:39 -0500
 Subject: [PATCH] Build polly libraries shared
 
+Origin: vendor
+Forwarded: not-needed
+Last-Update: 2025-01-10
 ---
  polly/cmake/polly_macros.cmake | 4 +---
  2 files changed, 1 insertion(+), 5 deletions(-)

l/llvm/pkg/serpent/0001-Emit-relocations-for-BOLT.patch → l/llvm/pkg/patches/build/0001-Emit-relocations-for-BOLT.patch

@@ -3,6 +3,11 @@ From: Reilly Brogan <[email protected]>
 Date: Sun, 26 May 2024 16:08:54 -0500
 Subject: [PATCH] Emit relocations for BOLT
 
+Ensure that relocations are always emitted for the libraries/binaries that we want to BOLT. This allows us to not have them emitted for all other files.
+
+Origin: vendor
+Forwarded: not-needed
+Last-Update: 2025-01-10
 ---
  clang/tools/clang-shlib/CMakeLists.txt | 2 ++
  clang/tools/driver/CMakeLists.txt      | 2 ++

l/llvm/pkg/patches/build/0001-downstream-Suppress-runpath-in-openmp.patch

@@ -0,0 +1,27 @@
+From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
+From: Reilly Brogan <[email protected]>
+Date: Thu, 9 Jan 2025 23:27:13 -0600
+Subject: [PATCH] downstream: Suppress runpath in openmp
+
+The libraries for openmp are installed into libdir, so RUNPATH is unnecessary. If this argument is not set then a relative RUNPATH is added to `llvm-omp-*` binaries which Boulder can't currently handle.
+
+Origin: vendor
+Forwarded: not-needed
+Last-Update: 2025-01-10
+---
+ openmp/libomptarget/tools/CMakeLists.txt | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+
+diff --git a/openmp/libomptarget/tools/CMakeLists.txt b/openmp/libomptarget/tools/CMakeLists.txt
+index a850647fbd58..1eaa784e5f50 100644
+--- a/openmp/libomptarget/tools/CMakeLists.txt
++++ b/openmp/libomptarget/tools/CMakeLists.txt
+@@ -17,7 +17,7 @@ mark_as_advanced(OPENMP_TOOLS_INSTALL_DIR)
+ # Move these macros to AddOpenMP if such a CMake module is ever created.
+
+ macro(add_openmp_tool name)
+-  llvm_add_tool(OPENMP ${ARGV})
++  llvm_add_tool(OPENMP ${ARGV} NO_INSTALL_RPATH)
+ endmacro()
+
+ macro(add_openmp_tool_symlink name)

l/llvm/pkg/serpent/0003-PATCH-clang-Don-t-install-static-libraries.patch → l/llvm/pkg/patches/build/0003-PATCH-clang-Don-t-install-static-libraries.patch

@@ -3,6 +3,9 @@ From: Tom Stellard <[email protected]>
 Date: Fri, 31 Jan 2020 11:04:57 -0800
 Subject: [PATCH][clang] Don't install static libraries
 
+Origin: vendor
+Forwarded: not-needed
+Last-Update: 2025-01-10
 ---
  clang/cmake/modules/AddClang.cmake | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)

l/llvm/pkg/serpent/0001-Make-gnu-hash-the-default-for-lld-and-clang.patch → l/llvm/pkg/patches/config/0001-Make-gnu-hash-the-default-for-lld-and-clang.patch

@@ -3,7 +3,9 @@ From: Peter O'Connor <[email protected]>
 Date: Mon, 5 Jul 2021 14:22:14 +1000
 Subject: [PATCH] Make gnu hash the default for lld and clang
 
-Signed-off-by: Peter O'Connor <[email protected]>
+Origin: vendor
+Forwarded: not-needed
+Last-Update: 2025-01-10
 ---
  lld/ELF/Driver.cpp | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)

l/llvm/pkg/patches/config/0001-lld-Always-enable-build-id-and-use-20-byte-hashes.patch

@@ -0,0 +1,54 @@
+From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
+From: Reilly Brogan <[email protected]>
+Date: Thu, 12 Dec 2024 13:14:41 -0600
+Subject: [PATCH] lld: Always enable build-id and use 20-byte hashes
+
+Always ensure that build-id is used, and ensure that is is 20bytes in length. If the user does not want this behavior they can use `--build-id=none` explicitly.
+
+Note that though this patch uses "sha1" LLVM actually uses the BLAKE3 algorithm and truncates the output after the first 20 bytes.
+
+It would be nice if llvm supported configuring this at build time like binutils does.
+
+Origin: vendor
+Forwarded: not-needed
+Last-Update: 2025-01-10
+---
+ lld/ELF/Config.h   | 2 +-
+ lld/ELF/Driver.cpp | 4 ++--
+ 2 files changed, 3 insertions(+), 3 deletions(-)
+
+diff --git a/lld/ELF/Config.h b/lld/ELF/Config.h
+index 28726d48e428..5b2a1541e529 100644
+--- a/lld/ELF/Config.h
++++ b/lld/ELF/Config.h
+@@ -351,7 +351,7 @@ struct Config {
+   GcsPolicy zGcs;
+   bool power10Stubs;
+   ARMVFPArgKind armVFPArgs = ARMVFPArgKind::Default;
+-  BuildIdKind buildId = BuildIdKind::None;
++  BuildIdKind buildId = BuildIdKind::Sha1;
+   SeparateSegmentKind zSeparate;
+   ELFKind ekind = ELFNoneKind;
+   uint16_t emachine = llvm::ELF::EM_NONE;
+diff --git a/lld/ELF/Driver.cpp b/lld/ELF/Driver.cpp
+index bcc7971042c6..7a0a4006a099 100644
+--- a/lld/ELF/Driver.cpp
++++ b/lld/ELF/Driver.cpp
+@@ -900,7 +900,7 @@ static std::pair<BuildIdKind, SmallVector<uint8_t, 0>>
+ getBuildId(opt::InputArgList &args) {
+   auto *arg = args.getLastArg(OPT_build_id);
+   if (!arg)
+-    return {BuildIdKind::None, {}};
++    return {BuildIdKind::Sha1, {}};
+
+   StringRef s = arg->getValue();
+   if (s == "fast")
+@@ -916,7 +916,7 @@ getBuildId(opt::InputArgList &args) {
+
+   if (s != "none")
+     error("unknown --build-id style: " + s);
+-  return {BuildIdKind::None, {}};
++  return {BuildIdKind::Sha1, {}};
+ }
+
+ static std::pair<bool, bool> getPackDynRelocs(opt::InputArgList &args) {

l/llvm/pkg/patches/config/0001-lld-Use-zstd-compression-for-debug-symbols-by-defa.patch

@@ -0,0 +1,120 @@
+From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
+From: Reilly Brogan <[email protected]>
+Date: Thu, 12 Dec 2024 20:50:14 -0600
+Subject: [PATCH] lld: Use zstd compression for debug symbols by default
+
+This causes LLVM to always behave as though `--compress-debug-sections=zstd` is present.
+
+It would be nice if llvm supported configuring this at build time like binutils does.
+
+Origin: vendor
+Forwarded: not-needed
+Last-Update: 2025-01-10
+---
+ clang/include/clang/Driver/Options.td  | 4 ++--
+ clang/tools/driver/cc1as_main.cpp      | 4 ++--
+ lld/ELF/Config.h                       | 2 +-
+ lld/ELF/Driver.cpp                     | 8 ++++----
+ lld/ELF/Options.td                     | 2 +-
+ llvm/tools/llvm-objcopy/ObjcopyOpts.td | 4 ++--
+ 6 files changed, 12 insertions(+), 12 deletions(-)
+
+diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
+index b8d1608a1453..b517f4aadb51 100644
+--- a/clang/include/clang/Driver/Options.td
++++ b/clang/include/clang/Driver/Options.td
+@@ -6781,9 +6781,9 @@ def record_command_line : Separate<["-"], "record-command-line">,
+ def compress_debug_sections_EQ : Joined<["-", "--"], "compress-debug-sections=">,
+     HelpText<"DWARF debug sections compression type">, Values<"none,zlib,zstd">,
+     NormalizedValuesScope<"llvm::DebugCompressionType">, NormalizedValues<["None", "Zlib", "Zstd"]>,
+-    MarshallingInfoEnum<CodeGenOpts<"CompressDebugSections">, "None">;
++    MarshallingInfoEnum<CodeGenOpts<"CompressDebugSections">, "Zstd">;
+ def compress_debug_sections : Flag<["-", "--"], "compress-debug-sections">,
+-  Alias<compress_debug_sections_EQ>, AliasArgs<["zlib"]>;
++  Alias<compress_debug_sections_EQ>, AliasArgs<["zstd"]>;
+ def mno_exec_stack : Flag<["-"], "mnoexecstack">,
+   HelpText<"Mark the file as not needing an executable stack">,
+   MarshallingInfoFlag<CodeGenOpts<"NoExecStack">>;
+diff --git a/clang/tools/driver/cc1as_main.cpp b/clang/tools/driver/cc1as_main.cpp
+index bc398fa0731f..62bdf73939da 100644
+--- a/clang/tools/driver/cc1as_main.cpp
++++ b/clang/tools/driver/cc1as_main.cpp
+@@ -100,7 +100,7 @@ struct AssemblerInvocation {
+   std::string DebugCompilationDir;
+   llvm::SmallVector<std::pair<std::string, std::string>, 0> DebugPrefixMap;
+   llvm::DebugCompressionType CompressDebugSections =
+-      llvm::DebugCompressionType::None;
++      llvm::DebugCompressionType::Zstd;
+   std::string MainFileName;
+   std::string SplitDwarfOutput;
+
+@@ -263,7 +263,7 @@ bool AssemblerInvocation::CreateFromArgs(AssemblerInvocation &Opts,
+             .Case("none", llvm::DebugCompressionType::None)
+             .Case("zlib", llvm::DebugCompressionType::Zlib)
+             .Case("zstd", llvm::DebugCompressionType::Zstd)
+-            .Default(llvm::DebugCompressionType::None);
++            .Default(llvm::DebugCompressionType::Zstd);
+   }
+
+   Opts.RelaxELFRelocations = !Args.hasArg(OPT_mrelax_relocations_no);
+diff --git a/lld/ELF/Config.h b/lld/ELF/Config.h
+index 3a4a47d900fe..0ec8ea5b144c 100644
+--- a/lld/ELF/Config.h
++++ b/lld/ELF/Config.h
+@@ -221,7 +221,7 @@ struct Config {
+   CGProfileSortKind callGraphProfileSort;
+   bool checkSections;
+   bool checkDynamicRelocs;
+-  llvm::DebugCompressionType compressDebugSections;
++  llvm::DebugCompressionType compressDebugSections = llvm::DebugCompressionType::Zstd;
+   bool cref;
+   llvm::SmallVector<std::pair<llvm::GlobPattern, uint64_t>, 0>
+       deadRelocInNonAlloc;
+diff --git a/lld/ELF/Driver.cpp b/lld/ELF/Driver.cpp
+index 8079be691436..3fa84a7cb4d3 100644
+--- a/lld/ELF/Driver.cpp
++++ b/lld/ELF/Driver.cpp
+@@ -1086,10 +1086,10 @@ static CGProfileSortKind getCGProfileSortKind(opt::InputArgList &args) {
+ static DebugCompressionType getCompressionType(StringRef s, StringRef option) {
+   DebugCompressionType type = StringSwitch<DebugCompressionType>(s)
+                                   .Case("zlib", DebugCompressionType::Zlib)
+-                                  .Case("zstd", DebugCompressionType::Zstd)
+-                                  .Default(DebugCompressionType::None);
+-  if (type == DebugCompressionType::None) {
+-    if (s != "none")
++                                  .Case("none", DebugCompressionType::None)
++                                  .Default(DebugCompressionType::Zstd);
++  if (type == DebugCompressionType::Zstd) {
++    if (s != "zstd")
+       error("unknown " + option + " value: " + s);
+   } else if (const char *reason = compression::getReasonIfUnsupported(
+                  compression::formatFor(type))) {
+diff --git a/lld/ELF/Options.td b/lld/ELF/Options.td
+index 35d9c5ffa29b..2cc5c296c0ca 100644
+--- a/lld/ELF/Options.td
++++ b/lld/ELF/Options.td
+@@ -64,7 +64,7 @@ defm check_sections: B<"check-sections",
+     "Do not check section addresses for overlaps">;
+
+ defm compress_debug_sections:
+-  Eq<"compress-debug-sections", "Compress DWARF debug sections">,
++  Eq<"compress-debug-sections", "Compress DWARF debug sections (default: zstd)">,
+   MetaVarName<"[none,zlib,zstd]">;
+
+ defm defsym: Eq<"defsym", "Define a symbol alias">, MetaVarName<"<symbol>=<value>">;
+diff --git a/llvm/tools/llvm-objcopy/ObjcopyOpts.td b/llvm/tools/llvm-objcopy/ObjcopyOpts.td
+index ead8cd28d387..16b3dd8830ae 100644
+--- a/llvm/tools/llvm-objcopy/ObjcopyOpts.td
++++ b/llvm/tools/llvm-objcopy/ObjcopyOpts.td
+@@ -33,9 +33,9 @@ def compress_debug_sections
+     : Joined<["--"], "compress-debug-sections=">,
+       MetaVarName<"format">,
+       HelpText<"Compress DWARF debug sections using specified format. Supported "
+-               "formats: zlib, zstd. Select zlib if <format> is omitted">;
++               "formats: zlib, zstd. Select zstd if <format> is omitted.">;
+ def : Flag<["--"], "compress-debug-sections">, Alias<compress_debug_sections>,
+-      AliasArgs<["zlib"]>;
++      AliasArgs<["zstd"]>;
+ def decompress_debug_sections : Flag<["--"], "decompress-debug-sections">,
+                                 HelpText<"Decompress DWARF debug sections">;
+ defm split_dwo