[llvm] add zstd to `llvm::compression` namespace

[ckissane]

Summary

• add zstd to llvm::compression namespace
• add a CMake option LLVM_ENABLE_ZSTD with behavior mirroring that of LLVM_ENABLE_ZLIB
• add tests for zstd to llvm/unittests/Support/CompressionTest.cpp
• debian users should install libzstd when using LLVM_ENABLE_ZSTD=FORCE_ON from source due to this bug https://bugs.launchpad.net/ubuntu/+source/libzstd/+bug/1941956

[ckissane]

Fix missing cmake findZstd, remove unused crc32 from compression namespaces and simplify some code

[ckissane]

fix type for FindZstd cmake usage

[phosek]

I think this patch should be broken into at least two:

1. Refactor llvm/include/llvm/Support/Compression.h and llvm/lib/Support/Compression.cpp to introduce a generic interface and use it throughout the codebase.
2. zstd support in llvm/include/llvm/Support/Compression.h including the CMake bits.

When uploading future changes, please also make sure to include full context.

llvm/include/llvm/Support/Compression.h

67–87

I think think that this should be a runtime rather than compile time selection. Many (if not most) toolchain vendors would likely want to enable both zstd and zlib and choose the algorithm based an option or file content.

For example, in the case of profiles, we want to be able to read existing profiles that use zlib, but for newly created profiles we may want to use zstd if available.

To make it possible, I suspect we may want to wrap the existing functions into a class that implements a common interface. I expect we would also need a function to detect the compression type based on the buffer content.

[ckissane]

more context

[MaskRay]

In D128465#3612948, @phosek wrote:

I think this patch should be broken into at least two:

1. Refactor llvm/include/llvm/Support/Compression.h and llvm/lib/Support/Compression.cpp to introduce a generic interface and use it throughout the codebase.
2. zstd support in llvm/include/llvm/Support/Compression.h including the CMake bits.

When uploading future changes, please also make sure to include full context.

Agree. As soon as the namespace refactoring is in a good enough shape, I think you may land the refactoring part before the rest of zstd patches.

Note: if you have a deep stacked patches, it may be useful to have a branch somewhere (e.g. your llvm-project fork on Github) so that interested folks can get the whole picture more easily.
(arc patch Dxxxxx can technically apply a deep stack, but it often fails to apply changes cleanly.)

[ckissane]

format

[ckissane]

make it the second stage

[ckissane]

add AlgorithmName to compression namespaces

[ckissane]

• feat: add zstd compression
• Merge branch 'ckissane.refactor-compression-namespace' into ckissane.add-zstd-compression
• Merge branch 'ckissane.refactor-compression-namespace' into ckissane.add-zstd-compression

[MaskRay]

llvm/cmake/modules/FindZSTD.cmake

2

How did you derive this?

The file seems contributed by you (I don't think facebook/zstd has such a file). It should not have a Meta Platforms, Inc notice.

4

Search SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception from other .cmake files

[ckissane]

start breaking this up further, this is now the add cmake config patch

[ckissane]

mark outdated comments done

[leonardchan]

LGTM although I think I'd be more comfortable with Petr's +2 for cmake changes.

Would it be better instead to expose LLVM_ENABLE_ZSTD to subrepos like clangd or flang until they're needed there? Not sure in this case if this will need +2s from their respective owners.

[MaskRay]

Please check both LLVM_ENABLE_ZSTD={on,off} work. Ideally, check that when zstd' cmake (libzstd-dev on Debian) is unavailable, LLVM_ENABLE_ZSTD=on works (there is no zstd functionality but the cmake invocation should succeed).

llvm/cmake/modules/FindZSTD.cmake

4

Delete //

[MaskRay]

Consider adding have_zstd to compiler-rt/test/lit.common.cfg.py and several other lit.site.cfg.py.in files

[phosek]

llvm/cmake/modules/FindZSTD.cmake

2

We usually don't include the copyright header in .cmake files at all.

[ckissane]

• Merge remote-tracking branch 'origin/main' into ckissane.add-zstd.0-cmake
• added have_zstd to compiler-rt/test/lit.common.cfg.py, clang-tools-extra/clangd/test/lit.cfg.py and several lit.site.cfg.py.in files

[ckissane]

marked fixed comments as done

[ckissane]

• Merge remote-tracking branch 'origin/main' into ckissane.add-zstd.0-cmake

[ldionne]

llvm/CMakeLists.txt

441

This broke builds that do not specify LLVM_ENABLE_ZSTD=OFF explicitly on macOS, and I suspect on other platforms as well:

ld: library not found for -lzstd
clang: error: linker command failed with exit code 1 (use -v to see invocation)

It looks to me like the build should not require zstd to be installed by default.

[MaskRay]

lld/test/lit.site.cfg.py.in

21

This needs a change in lld/test/CMakeLists.txt

Looks like you haven't run ninja check-lld as I mentioned.

21

Actually I am preparing a lld patch and you probably should drop have_zstd from lld/ for this patch.

[MaskRay]

This patch has changed a lot from what I have reviewed. The CMake change should be added along with llvm::compression::zstd::* functions.
Otherwise the change just introduces some CMake variables which cannot be tested.

Since you haven't touched flang, compiler-rt, etc. The patch should not updated their CMake files.

For lld/ELF, I've created D129406 to add the support. It will need to wait until you have landed these zstd changes.

[ckissane]

clean a space and make accurate

[ckissane]

lld/test/lit.site.cfg.py.in

21

I get this when I try to run that:
ninja: error: unknown target 'check-lld', did you mean 'check-lit'?

21

Got it

[MaskRay]

As I mentioned, the proper approach is to add zstd functionality along with the CMake change, instead of adding CMake to all llvm-project components without a way to test them.

lld/test/lit.site.cfg.py.in

21

During cmake, ensure that LLVM_ENABLE_PROJECTS contains lld

I posted the patch link to another patch of yours: https://reviews.llvm.org/D129406

[ckissane]

• make sure LLVM_ENABLE_ZSTD OFF turns into 0
• add zstd compression namespace

[ckissane]

In D128465#3642997, @MaskRay wrote:

As I mentioned, the proper approach is to add zstd functionality along with the CMake change, instead of adding CMake to all llvm-project components without a way to test them.

@MaskRay, I have now done this and ran the ldd tests as requested:

With LLVM_ENABLE_ZSTD=ON
$ ninja check-lld
Testing Time: 8.98s
  Unsupported      :   17
  Passed           : 2638
  Expectedly Failed:    1
With LLVM_ENABLE_ZSTD=OFF
$ ninja check-lld
Testing Time: 8.95s
  Unsupported      :   17
  Passed           : 2638
  Expectedly Failed:    1

[MaskRay]

In D128465#3646203, @ckissane wrote:

In D128465#3642997, @MaskRay wrote:

As I mentioned, the proper approach is to add zstd functionality along with the CMake change, instead of adding CMake to all llvm-project components without a way to test them.

@MaskRay, I have now done this and ran the ldd tests as requested:

With LLVM_ENABLE_ZSTD=ON
$ ninja check-lld
Testing Time: 8.98s
  Unsupported      :   17
  Passed           : 2638
  Expectedly Failed:    1
With LLVM_ENABLE_ZSTD=OFF
$ ninja check-lld
Testing Time: 8.95s
  Unsupported      :   17
  Passed           : 2638
  Expectedly Failed:    1

I request that you abandon this patch and incorporate the llvm cmake part into the llvm patch which you actually use zstd.
It is not appropriate to add zstd cmake to llvm-project components which don't use zstd.

[ckissane]

• add zstd tests to CompressionTest.cpp

[ckissane]

In D128465#3646258, @MaskRay wrote:

In D128465#3646203, @ckissane wrote:

In D128465#3642997, @MaskRay wrote:

As I mentioned, the proper approach is to add zstd functionality along with the CMake change, instead of adding CMake to all llvm-project components without a way to test them.

@MaskRay, I have now done this and ran the ldd tests as requested:

With LLVM_ENABLE_ZSTD=ON
$ ninja check-lld
Testing Time: 8.98s
  Unsupported      :   17
  Passed           : 2638
  Expectedly Failed:    1
With LLVM_ENABLE_ZSTD=OFF
$ ninja check-lld
Testing Time: 8.95s
  Unsupported      :   17
  Passed           : 2638
  Expectedly Failed:    1

I request that you abandon this patch and incorporate the llvm cmake part into the llvm patch which you actually use zstd.
It is not appropriate to add zstd cmake to llvm-project components which don't use zstd.

Let me see if I understand this correctly:
Are you saying that I should abandon this patch, and create a new patch that:

• adds findZSTD.cmake
• adds zstd to compression namespace
• adds tests to CompressionTest.cpp
• and does the minimal amount of cmake work to have the flags and test work
  • thus differing per component cmake config to patches like how you are doing in https://reviews.llvm.org/D129406

Is this correct?

I realize though that then https://reviews.llvm.org/D129406 or similar would be "the [first] llvm patch which actually use[s] zstd"

[MaskRay]

In D128465#3646561, @ckissane wrote:

In D128465#3646258, @MaskRay wrote:

In D128465#3646203, @ckissane wrote:

In D128465#3642997, @MaskRay wrote:

As I mentioned, the proper approach is to add zstd functionality along with the CMake change, instead of adding CMake to all llvm-project components without a way to test them.

@MaskRay, I have now done this and ran the ldd tests as requested:

With LLVM_ENABLE_ZSTD=ON
$ ninja check-lld
Testing Time: 8.98s
  Unsupported      :   17
  Passed           : 2638
  Expectedly Failed:    1
With LLVM_ENABLE_ZSTD=OFF
$ ninja check-lld
Testing Time: 8.95s
  Unsupported      :   17
  Passed           : 2638
  Expectedly Failed:    1

I request that you abandon this patch and incorporate the llvm cmake part into the llvm patch which you actually use zstd.
It is not appropriate to add zstd cmake to llvm-project components which don't use zstd.

Let me see if I understand this correctly:
Are you saying that I should abandon this patch, and create a new patch that:

• adds findZSTD.cmake
• adds zstd to compression namespace
• adds tests to CompressionTest.cpp
• and does the minimal amount of cmake work to have the flags and test work
  • thus differing per component cmake config to patches like how you are doing in https://reviews.llvm.org/D129406

Is this correct?

I realize though that then https://reviews.llvm.org/D129406 or similar would be "the [first] llvm patch which actually use[s] zstd"

You can find a component in llvm which uses zstd (e.g. lib/Support). Add the functionality with tests (e.g. a CompressionTest.cpp unittest) along with the CMake/(optional Bazel, gn) changes.
Keep clang, clang-tools-extra, lld, etc unchanged. These components don't yet use zstd and the CMake changes in this patch will be untestable.

[ckissane]

In D128465#3646601, @MaskRay wrote:

In D128465#3646561, @ckissane wrote:

In D128465#3646258, @MaskRay wrote:

In D128465#3646203, @ckissane wrote:

In D128465#3642997, @MaskRay wrote:

As I mentioned, the proper approach is to add zstd functionality along with the CMake change, instead of adding CMake to all llvm-project components without a way to test them.

@MaskRay, I have now done this and ran the ldd tests as requested:

With LLVM_ENABLE_ZSTD=ON
$ ninja check-lld
Testing Time: 8.98s
  Unsupported      :   17
  Passed           : 2638
  Expectedly Failed:    1
With LLVM_ENABLE_ZSTD=OFF
$ ninja check-lld
Testing Time: 8.95s
  Unsupported      :   17
  Passed           : 2638
  Expectedly Failed:    1

I request that you abandon this patch and incorporate the llvm cmake part into the llvm patch which you actually use zstd.
It is not appropriate to add zstd cmake to llvm-project components which don't use zstd.

Let me see if I understand this correctly:
Are you saying that I should abandon this patch, and create a new patch that:

• adds findZSTD.cmake
• adds zstd to compression namespace
• adds tests to CompressionTest.cpp
• and does the minimal amount of cmake work to have the flags and test work
  • thus differing per component cmake config to patches like how you are doing in https://reviews.llvm.org/D129406

Is this correct?

I realize though that then https://reviews.llvm.org/D129406 or similar would be "the [first] llvm patch which actually use[s] zstd"

You can find a component in llvm which uses zstd (e.g. lib/Support). Add the functionality with tests (e.g. a CompressionTest.cpp unittest) along with the CMake/(optional Bazel, gn) changes.
Keep clang, clang-tools-extra, lld, etc unchanged. These components don't yet use zstd and the CMake changes in this patch will be untestable.

Got it!, this patch already has such tests (CompressionTest.cpp unittest) in lib/Support, so I should just remove the untestable cmake config.

[ckissane]

• Trim CMAKE modifications to minimum needed

[MaskRay]

The patch looks mostly good but please fix the subject and the summary.

Note: if you fix the local commit message, you can use arc diff --head=HEAD 'HEAD^' --verbatim to update the subject and the summary.

llvm/lib/Support/Compression.cpp

85

https://llvm.org/docs/CodingStandards.html#don-t-use-braces-on-simple-single-statement-bodies-of-if-else-loop-statements

ditto below

[MaskRay]

llvm/lib/Support/Compression.cpp

129

const size_t

llvm/unittests/Support/CompressionTest.cpp

69 ↗

(On Diff #444374)

delete blank line

70 ↗

(On Diff #444374)

static void testZstd...

73 ↗

(On Diff #444374)

delete blank line

99 ↗

(On Diff #444374)

https://llvm.org/docs/CodingStandards.html#don-t-use-braces-on-simple-single-statement-bodies-of-if-else-loop-statements

[ckissane]

tidy some code

[ckissane]

Mark handled comments from @MaskRay as done

[phosek]

llvm/lib/Support/Compression.cpp

63

I'd make changes to zlib separately.

85

Ditto here.

[ckissane]

llvm/lib/Support/Compression.cpp

63

Here https://reviews.llvm.org/D129698

[ckissane]

• Merge branch main

[MaskRay]

llvm/include/llvm/Support/Compression.h

54

I changed the zlib functions to use uint8_t * and ArrayRef<uint8_t> instead. This patch needs to switch to uint8_t too.

llvm/lib/Support/Compression.cpp

153

uint8_t

[ckissane]

• Merge remote-tracking branch 'origin/main' into ckissane.add-zstd
• [Support] update zstd interface to use uint8_t * and ArrayRef<uint8_t>

[MaskRay]

Why was this reverted? Please make extensive tests especially for something dealing with the build system. When reverting a commit, briefly describe what happened.

[sebastian-ne]

This breaks builds when LLVM is included with CMake’s add_subdirectory.
I think the zstd target needs to be marked as imported.
The following patch fixes the problem, although I’m not familiar enough with CMake to know if this is the right way to go:

diff --git a/llvm/cmake/modules/FindZSTD.cmake b/llvm/cmake/modules/FindZSTD.cmake
index 43ccf7232138..8e6ec6e8a988 100644
--- a/llvm/cmake/modules/FindZSTD.cmake
+++ b/llvm/cmake/modules/FindZSTD.cmake
@@ -12,6 +12,10 @@ find_package_handle_standard_args(
     ZSTD_LIBRARY ZSTD_INCLUDE_DIR)

 if(ZSTD_FOUND)
+    if(NOT TARGET Zstd::zstd)
+      add_library(Zstd::zstd UNKNOWN IMPORTED)
+      set_target_properties(Zstd::zstd PROPERTIES IMPORTED_LOCATION "${ZSTD_LIBRARY}")
+    endif()
     set(ZSTD_LIBRARIES ${ZSTD_LIBRARY})
     set(ZSTD_INCLUDE_DIRS ${ZSTD_INCLUDE_DIR})
 endif()
diff --git a/llvm/lib/Support/CMakeLists.txt b/llvm/lib/Support/CMakeLists.txt
index 52b95c5377d3..031adfa33ba8 100644
--- a/llvm/lib/Support/CMakeLists.txt
+++ b/llvm/lib/Support/CMakeLists.txt
@@ -26,7 +26,7 @@ if(LLVM_ENABLE_ZLIB)
 endif()

 if(LLVM_ENABLE_ZSTD)
-  list(APPEND imported_libs zstd)
+  list(APPEND imported_libs Zstd::zstd)
 endif()

 if( MSVC OR MINGW )

[aemerson]

I just reverted this in 6e6be5f9504d because it seems to have broken macOS builds:

llvm/lib/Support/Compression.cpp:24:10: fatal error: 'zstd.h' file not found
#include <zstd.h>
         ^~~~~~~~

[ckissane]

In D128465#3651025, @aemerson wrote:

I just reverted this in 6e6be5f9504d because it seems to have broken macOS builds:

llvm/lib/Support/Compression.cpp:24:10: fatal error: 'zstd.h' file not found
#include <zstd.h>
         ^~~~~~~~

@aemerson Could you provide the output of your cmake command?
(I can't easily reproduce because I don't have a mac on hand)

[ckissane]

@aemerson can you let me know if https://reviews.llvm.org/D129786 works on MacOS?
I am continuing work on that revision because this one has a lot of outdated unrelated history.

[jasonmolenda]

In D128465#3652525, @ckissane wrote:

In D128465#3651025, @aemerson wrote:

I just reverted this in 6e6be5f9504d because it seems to have broken macOS builds:

llvm/lib/Support/Compression.cpp:24:10: fatal error: 'zstd.h' file not found
#include <zstd.h>
         ^~~~~~~~

@aemerson Could you provide the output of your cmake command?
(I can't easily reproduce because I don't have a mac on hand)

FWIW I hit this last night on my mac desktop - it looked like I had zstd installed by homebrew (as a dependency on something I installed), and cmake was able to find it in /opt/homebrew (somehow) but when Compression.cpp was built, nothing had added -I/opt/homebrew/include so the header wasn't found. I hacked the FindZSTD.cmake to not find it, to finish what I was working on before bedtime. This might not be related to what @aemerson saw though.

[ckissane]

In D128465#3653051, @jasonmolenda wrote:

In D128465#3652525, @ckissane wrote:

In D128465#3651025, @aemerson wrote:

I just reverted this in 6e6be5f9504d because it seems to have broken macOS builds:

llvm/lib/Support/Compression.cpp:24:10: fatal error: 'zstd.h' file not found
#include <zstd.h>
         ^~~~~~~~

@aemerson Could you provide the output of your cmake command?
(I can't easily reproduce because I don't have a mac on hand)

FWIW I hit this last night on my mac desktop - it looked like I had zstd installed by homebrew (as a dependency on something I installed), and cmake was able to find it in /opt/homebrew (somehow) but when Compression.cpp was built, nothing had added -I/opt/homebrew/include so the header wasn't found. I hacked the FindZSTD.cmake to not find it, to finish what I was working on before bedtime. This might not be related to what @aemerson saw though.

@jasonmolenda Thank you for the info, I believe it is similar, could you try: https://reviews.llvm.org/D129786 for me?

[jasonmolenda]

In D128465#3653216, @ckissane wrote:

In D128465#3653051, @jasonmolenda wrote:

In D128465#3652525, @ckissane wrote:

In D128465#3651025, @aemerson wrote:

I just reverted this in 6e6be5f9504d because it seems to have broken macOS builds:

llvm/lib/Support/Compression.cpp:24:10: fatal error: 'zstd.h' file not found
#include <zstd.h>
         ^~~~~~~~

@aemerson Could you provide the output of your cmake command?
(I can't easily reproduce because I don't have a mac on hand)

FWIW I hit this last night on my mac desktop - it looked like I had zstd installed by homebrew (as a dependency on something I installed), and cmake was able to find it in /opt/homebrew (somehow) but when Compression.cpp was built, nothing had added -I/opt/homebrew/include so the header wasn't found. I hacked the FindZSTD.cmake to not find it, to finish what I was working on before bedtime. This might not be related to what @aemerson saw though.

@jasonmolenda Thank you for the info, I believe it is similar, could you try: https://reviews.llvm.org/D129786 for me?

I applied the patch from https://reviews.llvm.org/D129786 on a clean checkout, cmake ../llvm -G Ninja -DCMAKE_BUILD_TYPE=Debug -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_ENABLE_PROJECTS='llvm;lldb;clang' '-DLLVM_ENABLE_RUNTIMES=libcxx;libcxxabi' -DLLDB_ENABLE_PYTHON=1 (my normal cmake line) and got

-- Found ZSTD: /opt/homebrew/lib/libzstd.dylib  
-- Looking for ZSTD_compress
-- Looking for ZSTD_compress - found

[...]

FAILED: lib/Support/CMakeFiles/LLVMSupport.dir/Compression.cpp.o 
/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/c++ -DGTEST_HAS_RTTI=0 -D_DEBUG -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS -I/Users/jason/k/llvm/llvm-project/build/lib/Support -I/Users/jason/k/llvm/llvm-project/llvm/lib/Support -I/Users/jason/k/llvm/llvm-project/build/include -I/Users/jason/k/llvm/llvm-project/llvm/include -fPIC -fvisibility-inlines-hidden -Werror=date-time -Werror=unguarded-availability-new -Wall -Wextra -Wno-unused-parameter -Wwrite-strings -Wcast-qual -Wmissing-field-initializers -pedantic -Wno-long-long -Wc++98-compat-extra-semi -Wimplicit-fallthrough -Wcovered-switch-default -Wno-noexcept-type -Wnon-virtual-dtor -Wdelete-non-virtual-dtor -Wsuggest-override -Wstring-conversion -Wmisleading-indentation -fdiagnostics-color  -g -arch arm64 -isysroot /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX12.3.sdk -std=c++14  -fno-exceptions -fno-rtti -MD -MT lib/Support/CMakeFiles/LLVMSupport.dir/Compression.cpp.o -MF lib/Support/CMakeFiles/LLVMSupport.dir/Compression.cpp.o.d -o lib/Support/CMakeFiles/LLVMSupport.dir/Compression.cpp.o -c /Users/jason/k/llvm/llvm-project/llvm/lib/Support/Compression.cpp
/Users/jason/k/llvm/llvm-project/llvm/lib/Support/Compression.cpp:24:10: fatal error: 'zstd.h' file not found
#include <zstd.h>
         ^~~~~~~~
1 error generated.

% file /opt/homebrew/lib/libzstd.dylib 
/opt/homebrew/lib/libzstd.dylib: Mach-O 64-bit dynamically linked shared library arm64
% head -2 /opt/homebrew/include/zstd.h 
/*
 * Copyright (c) Yann Collet, Facebook, Inc.
%

% grep -i zstd CMakeCache.txt 
//Use zstd for compression/decompression if available. Can be ON,
LLVM_ENABLE_ZSTD:STRING=ON
ZSTD_INCLUDE_DIR:PATH=/opt/homebrew/include
ZSTD_LIBRARY:FILEPATH=/opt/homebrew/lib/libzstd.dylib
//Details about finding ZSTD
FIND_PACKAGE_MESSAGE_DETAILS_ZSTD:INTERNAL=[/opt/homebrew/lib/libzstd.dylib][/opt/homebrew/include][v()]
//Have symbol ZSTD_compress
HAVE_ZSTD:INTERNAL=1
//ADVANCED property for variable: ZSTD_INCLUDE_DIR
ZSTD_INCLUDE_DIR-ADVANCED:INTERNAL=1
//ADVANCED property for variable: ZSTD_LIBRARY
ZSTD_LIBRARY-ADVANCED:INTERNAL=1

[MaskRay]

Instead of marking a differential [OLD] (which may confuse readers whether this is to be abandoned), you can mark a differential Request Reviews/Request Changes. Then it will not be in the green state.

I clicked "Reopen" to make it clear the patch hasn't relanded.

[ckissane]

In D128465#3660714, @MaskRay wrote:

Instead of marking a differential [OLD] (which may confuse readers whether this is to be abandoned), you can mark a differential Request Reviews/Request Changes. Then it will not be in the green state.

I clicked "Reopen" to make it clear the patch hasn't relanded.

Oh, sorry about that!
I had actually created https://reviews.llvm.org/D129786 instead of reopening this one because at the time I tried, reopening didn't seem to work.

[MaskRay]

In D128465#3660733, @ckissane wrote:

In D128465#3660714, @MaskRay wrote:

Instead of marking a differential [OLD] (which may confuse readers whether this is to be abandoned), you can mark a differential Request Reviews/Request Changes. Then it will not be in the green state.

I clicked "Reopen" to make it clear the patch hasn't relanded.

Oh, sorry about that!

That's fine :) We need to learn the web UI.

I had actually created https://reviews.llvm.org/D129786 instead of reopening this one because at the time I tried, reopening didn't seem to work.

A reverted patch by default is still in the "Closed" state. You can "reopen" it so that people know it hasn't relanded.
When repairing a differential, just reuse the existing patch. It keeps all discussions (including past suggestions and issues a previous commit did not address) in one thread.
If the new patch is very similar to this one, you probably need to abandon it in favor of this one.

---

If you revert B because you haven't done A, and A and B are somewhat not so related (e.g. B addresses an existing brittle build system issue which happens to affect some platform when A is added), then having A as a separate differential is fine.

[ckissane]

• move try_find_dependency into it's own file
• add newline to end of TryFindDependencyMacro.cmake
• clarify comment in TryFindDependencyMacro.cmake

[ckissane]

• remove extra cmake info

[MaskRay]

Does this address the macOS build failure?

[ckissane]

In D128465#3661049, @MaskRay wrote:

Does this address the macOS build failure?

I believe so! @leonardchan is double checking

[leonardchan]

LGTM tested on mac on my end

[barannikov88]

@ckissane
I took fresh sources of LLVM and when I run cmake I get the following warning:

-- Looking for compress2
-- Looking for compress2 - found
CMake Warning at cmake/config-ix.cmake:149 (find_package):
  By not providing "Findzstd.cmake" in CMAKE_MODULE_PATH this project has
  asked CMake to find a package configuration file provided by "zstd", but
  CMake did not find one.

  Could not find a package configuration file provided by "zstd" with any of
  the following names:

    zstdConfig.cmake
    zstd-config.cmake

  Add the installation prefix of "zstd" to CMAKE_PREFIX_PATH or set
  "zstd_DIR" to a directory containing one of the above files.  If "zstd"
  provides a separate development package or SDK, be sure it has been
  installed.
Call Stack (most recent call first):
  CMakeLists.txt:768 (include)

Ubuntu 22.04 LTS
cmake version 3.22.3

[phosek]

llvm/cmake/config-ix.cmake

147

Since there's no Findzstd.cmake module shipped with CMake, and we don't provide in LLVM, we should use the config mode.

[MaskRay]

I am still seeing the

zstdConfig.cmake
zstd-config.cmake

warning. @ckissane @phosek will you fix it? :)

[abrachet]

In D128465#3672515, @MaskRay wrote:

I am still seeing the

zstdConfig.cmake
zstd-config.cmake

warning. @ckissane @phosek will you fix it? :)

I've quieted the warnings in https://github.com/llvm/llvm-project/commit/a71a01bc10d509bd4e746e7d277c4613ef886875, though I'm just now seeing @phosek's suggestion to use CONFIG I'm not sure what that does

[MaskRay]

llvm/include/llvm/Support/Compression.h

47

I missed the values here. Why is -5 picked for NoCompression? What does it mean?

zstd.h says ZSTD_maxCLevel() is currently 22. The CLI program mentions 19. Why is BestSizeCompression 12?

ZSTD_CLEVEL_DEFAULT/the CLI CLI uses 3 for the default level. Why is DefaultCompression 5?

[MaskRay]

llvm/include/llvm/Support/Compression.h

47

Ping @ckissane about the choice.

[andrewrk]

if(LLVM_ENABLE_ZSTD)
  list(APPEND imported_libs zstd::libzstd_shared)
endif()

This hard codes shared linking which breaks the use case of static linking LLVM.

Also LLVM needs to now include a Findzstd.cmake file or else we get this error:

CMake Error at cmake/config-ix.cmake:144 (find_package):
  By not providing "Findzstd.cmake" in CMAKE_MODULE_PATH this project has
  asked CMake to find a package configuration file provided by "zstd", but
  CMake did not find one.

  Could not find a package configuration file provided by "zstd" with any of
  the following names:

    zstdConfig.cmake
    zstd-config.cmake

  Add the installation prefix of "zstd" to CMAKE_PREFIX_PATH or set
  "zstd_DIR" to a directory containing one of the above files.  If "zstd"
  provides a separate development package or SDK, be sure it has been
  installed.
Call Stack (most recent call first):
  CMakeLists.txt:774 (include)

It is impossible to satisfy this dependency when bootstrapping a static build of zig without patching LLVM.

[MaskRay]

zstd provides GNU Makefile, CMake, and Meson. The CMake files are only installed when configured with CMake. Debian and Ubuntu lack them.
The pkg-config file libzstd.pc probably provides the best portability on at least *NIX systems. (I have also used it for a binutils-gdb patch.)

I think we can then avoid the

zstdConfig.cmake
zstd-config.cmake

issue.

[andrewrk]

Compiler infrastructure should not assume the existence of a Linux distribution. The portable way is simple and can easily be supported. One should be able to install $prefix/lib/libzstd.a and $prefix/include/zstd.h, then have that prefix searched as part of the standard CMAKE_PREFIX_PATH.

If LLVM will not carry Findzstd.cmake then it should not use FIND_PACKAGE in the case of -DLLVM_ENABLE_ZSTD=FORCE_ON; it should just throw -lzstd on the linker line instead of throwing rakes in my path for no reason.

[phosek]

In D128465#3800479, @andrewrk wrote:

if(LLVM_ENABLE_ZSTD)
  list(APPEND imported_libs zstd::libzstd_shared)
endif()

This hard codes shared linking which breaks the use case of static linking LLVM.

This was addressed by D132870 and D133222.

Also LLVM needs to now include a Findzstd.cmake file or else we get this error:

CMake Error at cmake/config-ix.cmake:144 (find_package):
  By not providing "Findzstd.cmake" in CMAKE_MODULE_PATH this project has
  asked CMake to find a package configuration file provided by "zstd", but
  CMake did not find one.

  Could not find a package configuration file provided by "zstd" with any of
  the following names:

    zstdConfig.cmake
    zstd-config.cmake

  Add the installation prefix of "zstd" to CMAKE_PREFIX_PATH or set
  "zstd_DIR" to a directory containing one of the above files.  If "zstd"
  provides a separate development package or SDK, be sure it has been
  installed.
Call Stack (most recent call first):
  CMakeLists.txt:774 (include)

It is impossible to satisfy this dependency when bootstrapping a static build of zig without patching LLVM.

It should be possible to use CMAKE_FIND_PACKAGE_PREFER_CONFIG=ON to pick up config file that's part of the zstd installation, but I'd be also fine including Findzstd.cmake in LLVM to avoid needing that.

[mgorny]

Is anyone working on a solution that would work for people using a zstd install method that's actually supported upstream?

[phosek]

In D128465#3826150, @mgorny wrote:

Is anyone working on a solution that would work for people using a zstd install method that's actually supported upstream?

I have created D134990, it'd be great if others could test it on their system.

[aaronmondal]

Porting to bazel in https://reviews.llvm.org/D143344

clang-tools-extra/clangd/CMakeLists.txt

 
 llvm_canonicalize_cmake_booleans(
   CLANGD_BUILD_XPC
   CLANGD_ENABLE_REMOTE
   ENABLE_GRPC_REFLECTION
   CLANGD_MALLOC_TRIM
   CLANGD_TIDY_CHECKS
   LLVM_ENABLE_ZLIB
+  LLVM_ENABLE_ZSTD
 )
 
 configure_file(
   ${CMAKE_CURRENT_SOURCE_DIR}/Features.inc.in
   ${CMAKE_CURRENT_BINARY_DIR}/Features.inc
 )
 
 set(LLVM_LINK_COMPONENTS

clang-tools-extra/clangd/test/lit.cfg.py

 if config.clangd_enable_remote:
   config.available_features.add('clangd-remote-index')
 
 if config.clangd_tidy_checks:
   config.available_features.add('clangd-tidy-checks')
 
 if config.have_zlib:
   config.available_features.add('zlib')
+
+if config.have_zstd:
+  config.available_features.add('zstd')

clang-tools-extra/clangd/test/lit.site.cfg.py.in

 config.llvm_libs_dir = lit_config.substitute("@LLVM_LIBS_DIR@")
 
 config.clangd_source_dir = "@CMAKE_CURRENT_SOURCE_DIR@/.."
 config.clangd_binary_dir = "@CMAKE_CURRENT_BINARY_DIR@/.."
 config.clangd_build_xpc = @CLANGD_BUILD_XPC@
 config.clangd_enable_remote = @CLANGD_ENABLE_REMOTE@
 config.clangd_tidy_checks = @CLANGD_TIDY_CHECKS@
 config.have_zlib = @LLVM_ENABLE_ZLIB@
+config.have_zstd = @LLVM_ENABLE_ZSTD@
 
 # Delegate logic to lit.cfg.py.
 lit_config.load_config(config, "@CMAKE_CURRENT_SOURCE_DIR@/lit.cfg.py")

clang/test/CMakeLists.txt

 # Test runner infrastructure for Clang. This configures the Clang test trees
 # for use by Lit, and delegates to LLVM's lit test handlers.
 
 llvm_canonicalize_cmake_booleans(
   CLANG_BUILD_EXAMPLES
   CLANG_DEFAULT_PIE_ON_LINUX
   CLANG_ENABLE_OPAQUE_POINTERS_INTERNAL
   CLANG_ENABLE_ARCMT
   CLANG_ENABLE_STATIC_ANALYZER
   CLANG_PLUGIN_SUPPORT
   CLANG_SPAWN_CC1
   ENABLE_BACKTRACES
   LLVM_ENABLE_ZLIB
+  LLVM_ENABLE_ZSTD
   LLVM_ENABLE_PER_TARGET_RUNTIME_DIR
   LLVM_ENABLE_THREADS
   LLVM_WITH_Z3
   )
 
 configure_lit_site_cfg(
   ${CMAKE_CURRENT_SOURCE_DIR}/lit.site.cfg.py.in
   ${CMAKE_CURRENT_BINARY_DIR}/lit.site.cfg.py

clang/test/lit.site.cfg.py.in

 config.clang_tools_dir = lit_config.substitute(path(r"@CURRENT_TOOLS_DIR@"))
 config.clang_lib_dir = path(r"@CMAKE_LIBRARY_OUTPUT_DIRECTORY@")
 config.host_triple = "@LLVM_HOST_TRIPLE@"
 config.target_triple = "@LLVM_TARGET_TRIPLE@"
 config.host_cc = "@CMAKE_C_COMPILER@"
 config.host_cxx = "@CMAKE_CXX_COMPILER@"
 config.llvm_use_sanitizer = "@LLVM_USE_SANITIZER@"
 config.have_zlib = @LLVM_ENABLE_ZLIB@
+config.have_zstd = @LLVM_ENABLE_ZSTD@
 config.clang_arcmt = @CLANG_ENABLE_ARCMT@
 config.clang_default_pie_on_linux = @CLANG_DEFAULT_PIE_ON_LINUX@
 config.clang_enable_opaque_pointers = @CLANG_ENABLE_OPAQUE_POINTERS_INTERNAL@
 config.clang_default_cxx_stdlib = "@CLANG_DEFAULT_CXX_STDLIB@"
 config.clang_staticanalyzer = @CLANG_ENABLE_STATIC_ANALYZER@
 config.clang_staticanalyzer_z3 = @LLVM_WITH_Z3@
 config.clang_examples = @CLANG_BUILD_EXAMPLES@
 config.enable_shared = @ENABLE_SHARED@

compiler-rt/lib/sanitizer_common/symbolizer/scripts/build_symbolizer.sh

cmake -GNinja \
     -DCMAKE_BUILD_TYPE=Release \
     -DCMAKE_C_COMPILER=$CC \
     -DCMAKE_CXX_COMPILER=$CXX \
     -DCMAKE_C_FLAGS_RELEASE="${LLVM_FLAGS}" \
     -DCMAKE_CXX_FLAGS_RELEASE="${LLVM_FLAGS}" \
     -DLLVM_TABLEGEN=$TBLGEN \
     -DLLVM_DEFAULT_TARGET_TRIPLE="${TARGET_TRIPLE}" \
     -DLLVM_ENABLE_ZLIB=ON \
+    -DLLVM_ENABLE_ZSTD=ON \
     -DLLVM_ENABLE_TERMINFO=OFF \
     -DLLVM_ENABLE_THREADS=OFF \
   $LLVM_SRC
 fi
 cd ${LLVM_BUILD}
 ninja LLVMSymbolize LLVMObject LLVMBinaryFormat LLVMDebugInfoDWARF LLVMSupport LLVMDebugInfoPDB LLVMDebuginfod LLVMMC LLVMDemangle LLVMTextAPI
 
 cd ${BUILD_DIR}

compiler-rt/test/lit.common.cfg.py

 if re.match(r'^x86_64.*-linux', config.target_triple):
   config.available_features.add("x86_64-linux")
 
 config.available_features.add("host-byteorder-" + sys.byteorder + "-endian")
 
 if config.have_zlib == "1":
   config.available_features.add("zlib")
 
+if config.have_zstd == "1":
+  config.available_features.add("zstd")
+
 # Use ugly construction to explicitly prohibit "clang", "clang++" etc.
 # in RUN lines.
 config.substitutions.append(
     (' clang', """\n\n*** Do not use 'clangXXX' in tests,
      instead define '%clangXXX' substitution in lit config. ***\n\n""") )
 
 if config.host_os == 'NetBSD':
   nb_commands_dir = os.path.join(config.compiler_rt_src_root,

compiler-rt/test/lit.common.configured.in

elif config.android:
   if config.target_arch == "i386":
     set_default("target_suffix", "-i686-android")
   else:
     set_default("target_suffix", "-%s-android" % config.target_arch)
 else:
   set_default("target_suffix", "-%s" % config.target_arch)
 
 set_default("have_zlib", "@LLVM_ENABLE_ZLIB@")
+set_default("have_zstd", "@LLVM_ENABLE_ZSTD@")
 set_default("libcxx_used", "@LLVM_LIBCXX_USED@")
 
 # LLVM tools dir can be passed in lit parameters, so try to
 # apply substitution.
 config.llvm_tools_dir = lit_config.substitute(config.llvm_tools_dir)
 config.clang = lit_config.substitute(config.clang)
 config.compiler_rt_libdir = lit_config.substitute(config.compiler_rt_libdir)
 
 if not os.path.exists(config.clang):
   lit_config.fatal("Can't find compiler on path %r" % config.clang)
 
 # Setup attributes common for all compiler-rt projects.
 lit_config.load_config(config, "@COMPILER_RT_SOURCE_DIR@/test/lit.common.cfg.py")

flang/CMakeLists.txt

if (FLANG_STANDALONE_BUILD)
   if (NOT Clang_FOUND)
     message(FATAL_ERROR "Failed to find Clang")
   endif()
 
   # If LLVM links to zlib we need the imported targets so we can too.
   if(LLVM_ENABLE_ZLIB)
     find_package(ZLIB REQUIRED)
   endif()
+  
+  # If LLVM links to zstd we need the imported targets so we can too.
+  if(LLVM_ENABLE_ZSTD)
+    find_package(ZSTD REQUIRED)
+  endif()
+
   option(LLVM_ENABLE_PEDANTIC "Compile with pedantic enabled." ON)
   if(CMAKE_COMPILER_IS_GNUCXX)
     set(USE_NO_MAYBE_UNINITIALIZED 1)
   endif()
 
   include(CMakeParseArguments)
   include(AddLLVM)
   include(HandleLLVMOptions)

lld/ELF/CMakeLists.txt

 set(LLVM_TARGET_DEFINITIONS Options.td)
 tablegen(LLVM Options.inc -gen-opt-parser-defs)
 add_public_tablegen_target(ELFOptionsTableGen)
 
 if(LLVM_ENABLE_ZLIB)
   set(imported_libs ZLIB::ZLIB)
 endif()
 
+if(LLVM_ENABLE_ZSTD)
+  list(APPEND imported_libs zstd)
+endif()
+
 add_lld_library(lldELF
   AArch64ErrataFix.cpp
   Arch/AArch64.cpp
   Arch/AMDGPU.cpp
   Arch/ARM.cpp
   Arch/AVR.cpp
   Arch/Hexagon.cpp
   Arch/Mips.cpp

lld/test/lit.site.cfg.py.in

 config.enable_backtrace = @ENABLE_BACKTRACES@
 config.errc_messages = "@LLVM_LIT_ERRC_MESSAGES@"
 config.lld_obj_root = "@LLD_BINARY_DIR@"
 config.lld_libs_dir = lit_config.substitute("@CURRENT_LIBS_DIR@")
 config.lld_tools_dir = lit_config.substitute("@CURRENT_TOOLS_DIR@")
 config.target_triple = "@LLVM_TARGET_TRIPLE@"
 config.python_executable = "@Python3_EXECUTABLE@"
 config.have_zlib = @LLVM_ENABLE_ZLIB@
+config.have_zstd = @LLVM_ENABLE_ZSTD@
 config.have_libxar = @LLVM_HAVE_LIBXAR@
 config.have_libxml2 = @LLVM_ENABLE_LIBXML2@
 config.sizeof_void_p = @CMAKE_SIZEOF_VOID_P@
 config.ld_lld_default_mingw = @LLD_DEFAULT_LD_LLD_IS_MINGW@
 config.build_examples = @LLVM_BUILD_EXAMPLES@
 config.has_plugins = @LLVM_ENABLE_PLUGINS@
 config.linked_bye_extension = @LLVM_BYE_LINK_INTO_TOOLS@
 
 import lit.llvm
 lit.llvm.initialize(lit_config, config)
 
 # Let the main config do the real work.
 lit_config.load_config(config, "@LLD_SOURCE_DIR@/test/lit.cfg.py")

lldb/source/Plugins/Process/gdb-remote/CMakeLists.txt

 if(HAVE_LIBCOMPRESSION)
   set(LIBCOMPRESSION compression)
 endif()
 
 if(LLVM_ENABLE_ZLIB)
   list(APPEND LLDB_SYSTEM_LIBS ZLIB::ZLIB)
 endif()
 
+if(LLVM_ENABLE_ZSTD)
+  list(APPEND LLDB_SYSTEM_LIBS zstd)
+endif()
+
 add_lldb_library(lldbPluginProcessGDBRemote PLUGIN
   GDBRemoteClientBase.cpp
   GDBRemoteCommunication.cpp
   GDBRemoteCommunicationClient.cpp
   GDBRemoteCommunicationHistory.cpp
   GDBRemoteCommunicationServer.cpp
   GDBRemoteCommunicationServerCommon.cpp
   GDBRemoteCommunicationServerLLGS.cpp

lldb/test/Shell/lit.site.cfg.py.in

 config.lldb_libs_dir = lit_config.substitute("@LLDB_LIBS_DIR@")
 config.lldb_tools_dir = lit_config.substitute("@LLDB_TOOLS_DIR@")
 # Since it comes from the command line, it may have backslashes which
 # should not need to be escaped.
 config.lldb_lit_tools_dir = lit_config.substitute(r"@LLDB_LIT_TOOLS_DIR@")
 config.target_triple = "@LLVM_TARGET_TRIPLE@"
 config.python_executable = "@Python3_EXECUTABLE@"
 config.have_zlib = @LLVM_ENABLE_ZLIB@
+config.have_zstd = @LLVM_ENABLE_ZSTD@
 config.lldb_enable_lzma = @LLDB_ENABLE_LZMA@
 config.host_triple = "@LLVM_HOST_TRIPLE@"
 config.lldb_bitness = 64 if @LLDB_IS_64_BITS@ else 32
 config.lldb_enable_python = @LLDB_ENABLE_PYTHON@
 config.lldb_enable_lua = @LLDB_ENABLE_LUA@
 config.lldb_build_directory = "@LLDB_TEST_BUILD_DIRECTORY@"
 config.lldb_system_debugserver = @LLDB_USE_SYSTEM_DEBUGSERVER@
 # The shell tests use their own module caches.
 config.lldb_module_cache = os.path.join("@LLDB_TEST_MODULE_CACHE_LLDB@", "lldb-shell")
 config.clang_module_cache = os.path.join("@LLDB_TEST_MODULE_CACHE_CLANG@", "lldb-shell")
 
 import lit.llvm
 lit.llvm.initialize(lit_config, config)
 
 # Let the main config do the real work.
 lit_config.load_config(config, os.path.join(config.lldb_src_root, "test", "Shell", "lit.cfg.py"))

llvm/CMakeLists.txt

 if (CMAKE_SYSTEM_NAME MATCHES "OS390")
   option(LLVM_ENABLE_THREADS "Use threads if available." OFF)
 else()
   option(LLVM_ENABLE_THREADS "Use threads if available." ON)
 endif()
 
 set(LLVM_ENABLE_ZLIB "ON" CACHE STRING "Use zlib for compression/decompression if available. Can be ON, OFF, or FORCE_ON")
 
+set(LLVM_ENABLE_ZSTD "ON" CACHE STRING "Use zstd for compression/decompression if available. Can be ON, OFF, or FORCE_ON")
+
 set(LLVM_ENABLE_CURL "OFF" CACHE STRING "Use libcurl for the HTTP client if available. Can be ON, OFF, or FORCE_ON")
 
 set(LLVM_ENABLE_HTTPLIB "OFF" CACHE STRING "Use cpp-httplib HTTP server library if available. Can be ON, OFF, or FORCE_ON")
 
 set(LLVM_Z3_INSTALL_DIR "" CACHE STRING "Install directory of the Z3 solver.")
 
 option(LLVM_ENABLE_Z3_SOLVER
   "Enable Support for the Z3 constraint solver in LLVM."

llvm/cmake/config-ix.cmake

if(ZLIB_FOUND)
     cmake_pop_check_state()
     if(LLVM_ENABLE_ZLIB STREQUAL FORCE_ON AND NOT HAVE_ZLIB)
       message(FATAL_ERROR "Failed to configure zlib")
     endif()
   endif()
   set(LLVM_ENABLE_ZLIB "${HAVE_ZLIB}")
 endif()
 
+if(LLVM_ENABLE_ZSTD)
+  if(LLVM_ENABLE_ZSTD STREQUAL FORCE_ON)
+    find_package(ZSTD REQUIRED)
+  elseif(NOT LLVM_USE_SANITIZER MATCHES "Memory.*")
+    find_package(ZSTD)
+  endif()
+  if(ZSTD_FOUND)
+    # Check if zstd we found is usable; for example, we may have found a 32-bit
+    # library on a 64-bit system which would result in a link-time failure.
+    cmake_push_check_state()
+    list(APPEND CMAKE_REQUIRED_INCLUDES ${ZSTD_INCLUDE_DIR})
+    list(APPEND CMAKE_REQUIRED_LIBRARIES ${ZSTD_LIBRARY})
+    check_symbol_exists(ZSTD_compress zstd.h HAVE_ZSTD)
+    cmake_pop_check_state()
+    if(LLVM_ENABLE_ZSTD STREQUAL FORCE_ON AND NOT HAVE_ZSTD)
+      message(FATAL_ERROR "Failed to configure zstd")
+    endif()
+  endif()
+  set(LLVM_ENABLE_ZSTD "${HAVE_ZSTD}")
+else()
+  set(LLVM_ENABLE_ZSTD 0)
+endif()
+
 if(LLVM_ENABLE_LIBXML2)
   if(LLVM_ENABLE_LIBXML2 STREQUAL FORCE_ON)
     find_package(LibXml2 REQUIRED)
   elseif(NOT LLVM_USE_SANITIZER MATCHES "Memory.*")
     find_package(LibXml2)
   endif()
   if(LibXml2_FOUND)
     # Check if libxml2 we found is usable; for example, we may have found a 32-bit

llvm/cmake/modules/FindZSTD.cmake

+find_path(ZSTD_INCLUDE_DIR
+  NAMES zstd.h
+  HINTS ${ZSTD_ROOT_DIR}/include)
+
+find_library(ZSTD_LIBRARY
+  NAMES zstd
+  HINTS ${ZSTD_ROOT_DIR}/lib)
+
+include(FindPackageHandleStandardArgs)
+find_package_handle_standard_args(
+    ZSTD DEFAULT_MSG
+    ZSTD_LIBRARY ZSTD_INCLUDE_DIR)
+
+if(ZSTD_FOUND)
+    set(ZSTD_LIBRARIES ${ZSTD_LIBRARY})
+    set(ZSTD_INCLUDE_DIRS ${ZSTD_INCLUDE_DIR})
+endif()
+
+mark_as_advanced(
+  ZSTD_INCLUDE_DIR
+  ZSTD_LIBRARY)

llvm/cmake/modules/LLVMConfig.cmake.in

 set(LLVM_ENABLE_UNWIND_TABLES @LLVM_ENABLE_UNWIND_TABLES@)
 
 set(LLVM_ENABLE_ZLIB @LLVM_ENABLE_ZLIB@)
 if(LLVM_ENABLE_ZLIB)
   set(ZLIB_ROOT @ZLIB_ROOT@)
   find_package(ZLIB)
 endif()
 
+set(LLVM_ENABLE_ZSTD @LLVM_ENABLE_ZSTD@)
+if(LLVM_ENABLE_ZSTD)
+  set(ZSTD_ROOT @ZSTD_ROOT@)
+  find_package(ZSTD)
+endif()
+
 set(LLVM_ENABLE_LIBXML2 @LLVM_ENABLE_LIBXML2@)
 if(LLVM_ENABLE_LIBXML2)
   find_package(LibXml2)
 endif()
 
 set(LLVM_WITH_Z3 @LLVM_WITH_Z3@)
 
 set(LLVM_ENABLE_DIA_SDK @LLVM_ENABLE_DIA_SDK@)

llvm/include/llvm/Config/llvm-config.h.cmake

 #cmakedefine LLVM_ENABLE_CURL ${LLVM_ENABLE_CURL}
 
 /* Define if we have cpp-httplib and want to use it */
 #cmakedefine LLVM_ENABLE_HTTPLIB ${LLVM_ENABLE_HTTPLIB}
 
 /* Define if zlib compression is available */
 #cmakedefine01 LLVM_ENABLE_ZLIB
 
+/* Define if zstd compression is available */
+#cmakedefine01 LLVM_ENABLE_ZSTD
+
 /* Define if LLVM was built with a dependency to the libtensorflow dynamic library */
 #cmakedefine LLVM_HAVE_TF_API
 
 /* Define to 1 if you have the <sysexits.h> header file. */
 #cmakedefine HAVE_SYSEXITS_H ${HAVE_SYSEXITS_H}
 
 /* Define if the xar_open() function is supported on this platform. */
 #cmakedefine LLVM_HAVE_LIBXAR ${LLVM_HAVE_LIBXAR}

llvm/include/llvm/Support/Compression.h

Error uncompress(StringRef InputBuffer, char *UncompressedBuffer,
                  size_t &UncompressedSize);
 
 Error uncompress(StringRef InputBuffer,
                  SmallVectorImpl<char> &UncompressedBuffer,
                  size_t UncompressedSize);
 
 } // End of namespace zlib
 
+namespace zstd {
+
+constexpr int NoCompression = -5;
+constexpr int BestSpeedCompression = 1;
+constexpr int DefaultCompression = 5;
+constexpr int BestSizeCompression = 12;
+
+bool isAvailable();
+
+void compress(StringRef InputBuffer, SmallVectorImpl<char> &CompressedBuffer,
+              int Level = DefaultCompression);
+
+Error uncompress(StringRef InputBuffer, char *UncompressedBuffer,
+                 size_t &UncompressedSize);
+
+Error uncompress(StringRef InputBuffer,
+                 SmallVectorImpl<char> &UncompressedBuffer,
+                 size_t UncompressedSize);
+
+} // End of namespace zstd
+
 } // End of namespace compression
 
 } // End of namespace llvm
 
 #endif

llvm/lib/Support/CMakeLists.txt

if (NOT LLVM_HAS_NOGLOBAL_CTOR_MUTEX)
     string(REPLACE "-Werror=global-constructors" "" CMAKE_CXX_FLAGS ${CMAKE_CXX_FLAGS})
   endif()
 endif()
 
 if(LLVM_ENABLE_ZLIB)
   set(imported_libs ZLIB::ZLIB)
 endif()
 
+if(LLVM_ENABLE_ZSTD)
+  list(APPEND imported_libs zstd)
+endif()
+
 if( MSVC OR MINGW )
   # libuuid required for FOLDERID_Profile usage in lib/Support/Windows/Path.inc.
   # advapi32 required for CryptAcquireContextW in lib/Support/Windows/Path.inc.
   set(system_libs ${system_libs} psapi shell32 ole32 uuid advapi32)
 elseif( CMAKE_HOST_UNIX )
   if( HAVE_LIBRT )
     set(system_libs ${system_libs} rt)
   endif()

llvm/lib/Support/Compression.cpp

 #include "llvm/ADT/StringRef.h"
 #include "llvm/Config/config.h"
 #include "llvm/Support/Compiler.h"
 #include "llvm/Support/Error.h"
 #include "llvm/Support/ErrorHandling.h"
 #if LLVM_ENABLE_ZLIB
 #include <zlib.h>
 #endif
+#if LLVM_ENABLE_ZSTD
+#include <zstd.h>
+#endif
 
 using namespace llvm;
 using namespace llvm::compression;
 
 #if LLVM_ENABLE_ZLIB
 
 static StringRef convertZlibCodeToString(int Code) {
   switch (Code) {
int Res =
       ::compress2((Bytef *)CompressedBuffer.data(), &CompressedSize,
                   (const Bytef *)InputBuffer.data(), InputBuffer.size(), Level);
   if (Res == Z_MEM_ERROR)
     report_bad_alloc_error("Allocation failed");
   assert(Res == Z_OK);
   // Tell MemorySanitizer that zlib output buffer is fully initialized.
   // This avoids a false report when running LLVM with uninstrumented ZLib.
   __msan_unpoison(CompressedBuffer.data(), CompressedSize);
   CompressedBuffer.truncate(CompressedSize);
 }
 
 Error zlib::uncompress(StringRef InputBuffer, char *UncompressedBuffer,
                        size_t &UncompressedSize) {
   int Res =
       ::uncompress((Bytef *)UncompressedBuffer, (uLongf *)&UncompressedSize,
                    (const Bytef *)InputBuffer.data(), InputBuffer.size());
   // Tell MemorySanitizer that zlib output buffer is fully initialized.
   // This avoids a false report when running LLVM with uninstrumented ZLib.
   __msan_unpoison(UncompressedBuffer, UncompressedSize);
   return Res ? make_error<StringError>(convertZlibCodeToString(Res),
                                        inconvertibleErrorCode())
              : Error::success();
 }
 
 Error zlib::uncompress(StringRef InputBuffer,
                        SmallVectorImpl<char> &UncompressedBuffer,
                        size_t UncompressedSize) {
   UncompressedBuffer.resize_for_overwrite(UncompressedSize);
   Error E = zlib::uncompress(InputBuffer, UncompressedBuffer.data(),
                              UncompressedSize);
   UncompressedBuffer.truncate(UncompressedSize);
   return E;
 }
 
 #else
 bool zlib::isAvailable() { return false; }
 void zlib::compress(StringRef InputBuffer,
                     SmallVectorImpl<char> &CompressedBuffer, int Level) {
   llvm_unreachable("zlib::compress is unavailable");
 }
 Error zlib::uncompress(StringRef InputBuffer, char *UncompressedBuffer,
                        size_t &UncompressedSize) {
   llvm_unreachable("zlib::uncompress is unavailable");
 }
 Error zlib::uncompress(StringRef InputBuffer,
                        SmallVectorImpl<char> &UncompressedBuffer,
                        size_t UncompressedSize) {
   llvm_unreachable("zlib::uncompress is unavailable");
 }
 #endif
+
+#if LLVM_ENABLE_ZSTD
+
+bool zstd::isAvailable() { return true; }
+
+void zstd::compress(StringRef InputBuffer,
+                    SmallVectorImpl<char> &CompressedBuffer, int Level) {
+  unsigned long CompressedBufferSize = ::ZSTD_compressBound(InputBuffer.size());
+  CompressedBuffer.resize_for_overwrite(CompressedBufferSize);
+  unsigned long CompressedSize = ::ZSTD_compress(
+      (char *)CompressedBuffer.data(), CompressedBufferSize,
+      (const char *)InputBuffer.data(), InputBuffer.size(), Level);
+  if (ZSTD_isError(CompressedSize))
+    report_bad_alloc_error("Allocation failed");
+  // Tell MemorySanitizer that zstd output buffer is fully initialized.
+  // This avoids a false report when running LLVM with uninstrumented ZLib.
+  __msan_unpoison(CompressedBuffer.data(), CompressedSize);
+  CompressedBuffer.truncate(CompressedSize);
+}
+
+Error zstd::uncompress(StringRef InputBuffer, char *UncompressedBuffer,
+                       size_t &UncompressedSize) {
+  unsigned long long const FrameContentSize = ZSTD_getFrameContentSize(
+      (const char *)InputBuffer.data(), InputBuffer.size());
+  size_t const Res =
+      ::ZSTD_decompress((char *)UncompressedBuffer, FrameContentSize,
+                        (const char *)InputBuffer.data(), InputBuffer.size());
+  UncompressedSize = Res;
+  // Tell MemorySanitizer that zstd output buffer is fully initialized.
+  // This avoids a false report when running LLVM with uninstrumented ZLib.
+  __msan_unpoison(UncompressedBuffer, UncompressedSize);
+  return Res != FrameContentSize
+             ? make_error<StringError>(ZSTD_getErrorName(Res),
+                                       inconvertibleErrorCode())
+             : Error::success();
+}
+
+Error zstd::uncompress(StringRef InputBuffer,
+                       SmallVectorImpl<char> &UncompressedBuffer,
+                       size_t UncompressedSize) {
+  UncompressedBuffer.resize_for_overwrite(UncompressedSize);
+  Error E = zstd::uncompress(InputBuffer, UncompressedBuffer.data(),
+                             UncompressedSize);
+  UncompressedBuffer.truncate(UncompressedSize);
+  return E;
+}
+
+#else
+bool zstd::isAvailable() { return false; }
+void zstd::compress(StringRef InputBuffer,
+                    SmallVectorImpl<char> &CompressedBuffer, int Level) {
+  llvm_unreachable("zstd::compress is unavailable");
+}
+Error zstd::uncompress(StringRef InputBuffer, char *UncompressedBuffer,
+                       size_t &UncompressedSize) {
+  llvm_unreachable("zstd::uncompress is unavailable");
+}
+Error zstd::uncompress(StringRef InputBuffer,
+                       SmallVectorImpl<char> &UncompressedBuffer,
+                       size_t UncompressedSize) {
+  llvm_unreachable("zstd::uncompress is unavailable");
+}
+#endif

llvm/test/lit.site.cfg.py.in

 config.host_os = "@HOST_OS@"
 config.host_cc = "@HOST_CC@"
 config.host_cxx = "@HOST_CXX@"
 # Note: ldflags can contain double-quoted paths, so must use single quotes here.
 config.host_ldflags = '@HOST_LDFLAGS@'
 config.llvm_use_intel_jitevents = @LLVM_USE_INTEL_JITEVENTS@
 config.llvm_use_sanitizer = "@LLVM_USE_SANITIZER@"
 config.have_zlib = @LLVM_ENABLE_ZLIB@
+config.have_zstd = @LLVM_ENABLE_ZSTD@
 config.have_libxar = @LLVM_HAVE_LIBXAR@
 config.have_libxml2 = @LLVM_ENABLE_LIBXML2@
 config.have_curl = @LLVM_ENABLE_CURL@
 config.have_httplib = @LLVM_ENABLE_HTTPLIB@
 config.have_dia_sdk = @LLVM_ENABLE_DIA_SDK@
 config.enable_ffi = @LLVM_ENABLE_FFI@
 config.build_examples = @LLVM_BUILD_EXAMPLES@
 config.enable_threads = @LLVM_ENABLE_THREADS@

utils/bazel/llvm_configs/llvm-config.h.cmake

 #cmakedefine LLVM_ENABLE_CURL ${LLVM_ENABLE_CURL}
 
 /* Define if we have cpp-httplib and want to use it */
 #cmakedefine LLVM_ENABLE_HTTPLIB ${LLVM_ENABLE_HTTPLIB}
 
 /* Define if zlib compression is available */
 #cmakedefine01 LLVM_ENABLE_ZLIB
 
+/* Define if zstd compression is available */
+#cmakedefine01 LLVM_ENABLE_ZSTD
+
 /* Define if LLVM was built with a dependency to the libtensorflow dynamic library */
 #cmakedefine LLVM_HAVE_TF_API
 
 /* Define to 1 if you have the <sysexits.h> header file. */
 #cmakedefine HAVE_SYSEXITS_H ${HAVE_SYSEXITS_H}
 
 /* Define if the xar_open() function is supported on this platform. */
 #cmakedefine LLVM_HAVE_LIBXAR ${LLVM_HAVE_LIBXAR}