Let clang-cl support CUDA/HIP #68921

yxsamliu · 2023-10-12T18:38:00Z

clang-cl is a driver mode that accepts options of MSVC cl.exe as a drop-in replacement for cl.exe. Currently clang-cl accepts mixed clang style options and cl style options. To let clang-cl accept a clang-style option, just need to add visibility CLOption to that option.

Currently nvcc can pass cl style options to cl.exe, which allows nvcc to compile C++ and CUDA programs with mixed nvcc and cl style options. On the other hand, clang cannot use mixed clang and cl style options to compile CUDA/HIP programs.

This patch add visibility CLOption to options needed to compile CUDA/HIP programs. This allows clang-cl to compile CUDA/HIP programs with mixed clang and cl style options.

clang-cl is a driver mode that accepts options of MSVC cl.exe as a drop-in replacement for cl.exe. Currently clang-cl accepts mixed clang style options and cl style options. To let clang-cl accept a clang-style option, just need to add visibility CLOption to that option. Currently nvcc can pass cl style options to cl.exe, which allows nvcc to compile C++ and CUDA programs with mixed nvcc and cl style options. On the other hand, clang cannot use mixed clang and cl style options to compile CUDA/HIP programs. This patch add visibility CLOption to options needed to compile CUDA/HIP programs. This allows clang-cl to compile CUDA/HIP programs with mixed clang and cl style options.

llvmbot · 2023-10-12T18:39:07Z

@llvm/pr-subscribers-clang-driver

@llvm/pr-subscribers-clang

Author: Yaxun (Sam) Liu (yxsamliu)

Changes

clang-cl is a driver mode that accepts options of MSVC cl.exe as a drop-in replacement for cl.exe. Currently clang-cl accepts mixed clang style options and cl style options. To let clang-cl accept a clang-style option, just need to add visibility CLOption to that option.

Currently nvcc can pass cl style options to cl.exe, which allows nvcc to compile C++ and CUDA programs with mixed nvcc and cl style options. On the other hand, clang cannot use mixed clang and cl style options to compile CUDA/HIP programs.

This patch add visibility CLOption to options needed to compile CUDA/HIP programs. This allows clang-cl to compile CUDA/HIP programs with mixed clang and cl style options.

Full diff: https://github.com/llvm/llvm-project/pull/68921.diff

3 Files Affected:

(modified) clang/include/clang/Driver/Options.td (+30-17)
(modified) clang/lib/Driver/Driver.cpp (+4-1)
(added) clang/test/Driver/cl-offload.cu (+27)

diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index 3f2058a5d4650ca..ceeb4c5e7c424cf 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -148,7 +148,8 @@ def pedantic_Group : OptionGroup<"<pedantic group>">, Group<f_Group>,
                      DocFlatten;
 
 def offload_Group : OptionGroup<"<offload group>">, Group<f_Group>,
-                   DocName<"Common Offloading options">;
+                   DocName<"Common Offloading options">,
+                   Visibility<[ClangOption, CLOption]>;
 
 def opencl_Group : OptionGroup<"<opencl group>">, Group<f_Group>,
                    DocName<"OpenCL options">;
@@ -157,13 +158,16 @@ def sycl_Group : OptionGroup<"<SYCL group>">, Group<f_Group>,
                  DocName<"SYCL options">;
 
 def cuda_Group : OptionGroup<"<CUDA group>">, Group<f_Group>,
-                   DocName<"CUDA options">;
+                   DocName<"CUDA options">,
+                   Visibility<[ClangOption, CLOption]>;
 
 def hip_Group : OptionGroup<"<HIP group>">, Group<f_Group>,
-                   DocName<"HIP options">;
+                   DocName<"HIP options">,
+                   Visibility<[ClangOption, CLOption]>;
 
 def m_Group : OptionGroup<"<m group>">, Group<CompileOnly_Group>,
-              DocName<"Target-dependent compilation options">;
+              DocName<"Target-dependent compilation options">,
+              Visibility<[ClangOption, CLOption]>;
 
 // Feature groups - these take command line options that correspond directly to
 // target specific features and can be translated directly from command line
@@ -5164,14 +5168,16 @@ def prebind__all__twolevel__modules : Flag<["-"], "prebind_all_twolevel_modules"
 def prebind : Flag<["-"], "prebind">;
 def preload : Flag<["-"], "preload">;
 def print_file_name_EQ : Joined<["-", "--"], "print-file-name=">,
-  HelpText<"Print the full library path of <file>">, MetaVarName<"<file>">;
+  HelpText<"Print the full library path of <file>">, MetaVarName<"<file>">,
+  Visibility<[ClangOption, CLOption]>;
 def print_ivar_layout : Flag<["-"], "print-ivar-layout">,
   Visibility<[ClangOption, CC1Option]>,
   HelpText<"Enable Objective-C Ivar layout bitmap print trace">,
   MarshallingInfoFlag<LangOpts<"ObjCGCBitmapPrint">>;
 def print_libgcc_file_name : Flag<["-", "--"], "print-libgcc-file-name">,
   HelpText<"Print the library path for the currently used compiler runtime "
-           "library (\"libgcc.a\" or \"libclang_rt.builtins.*.a\")">;
+           "library (\"libgcc.a\" or \"libclang_rt.builtins.*.a\")">,
+  Visibility<[ClangOption, CLOption]>;
 def print_multi_directory : Flag<["-", "--"], "print-multi-directory">;
 def print_multi_lib : Flag<["-", "--"], "print-multi-lib">;
 def print_multi_flags : Flag<["-", "--"], "print-multi-flags-experimental">,
@@ -5180,27 +5186,34 @@ def print_multi_os_directory : Flag<["-", "--"], "print-multi-os-directory">,
   Flags<[Unsupported]>;
 def print_target_triple : Flag<["-", "--"], "print-target-triple">,
   HelpText<"Print the normalized target triple">,
-  Visibility<[ClangOption, FlangOption]>;
+  Visibility<[ClangOption, FlangOption, CLOption]>;
 def print_effective_triple : Flag<["-", "--"], "print-effective-triple">,
   HelpText<"Print the effective target triple">,
-  Visibility<[ClangOption, FlangOption]>;
+  Visibility<[ClangOption, FlangOption, CLOption]>;
 // GCC --disable-multiarch, GCC --enable-multiarch (upstream and Debian
 // specific) have different behaviors. We choose not to support the option.
 def : Flag<["-", "--"], "print-multiarch">, Flags<[Unsupported]>;
 def print_prog_name_EQ : Joined<["-", "--"], "print-prog-name=">,
-  HelpText<"Print the full program path of <name>">, MetaVarName<"<name>">;
+  HelpText<"Print the full program path of <name>">, MetaVarName<"<name>">,
+  Visibility<[ClangOption, CLOption]>;
 def print_resource_dir : Flag<["-", "--"], "print-resource-dir">,
-  HelpText<"Print the resource directory pathname">;
+  HelpText<"Print the resource directory pathname">,
+  Visibility<[ClangOption, CLOption]>;
 def print_search_dirs : Flag<["-", "--"], "print-search-dirs">,
-  HelpText<"Print the paths used for finding libraries and programs">;
+  HelpText<"Print the paths used for finding libraries and programs">,
+  Visibility<[ClangOption, CLOption]>;
 def print_targets : Flag<["-", "--"], "print-targets">,
-  HelpText<"Print the registered targets">;
+  HelpText<"Print the registered targets">,
+  Visibility<[ClangOption, CLOption]>;
 def print_rocm_search_dirs : Flag<["-", "--"], "print-rocm-search-dirs">,
-  HelpText<"Print the paths used for finding ROCm installation">;
+  HelpText<"Print the paths used for finding ROCm installation">,
+  Visibility<[ClangOption, CLOption]>;
 def print_runtime_dir : Flag<["-", "--"], "print-runtime-dir">,
-  HelpText<"Print the directory pathname containing clangs runtime libraries">;
+  HelpText<"Print the directory pathname containing clangs runtime libraries">,
+  Visibility<[ClangOption, CLOption]>;
 def print_diagnostic_options : Flag<["-", "--"], "print-diagnostic-options">,
-  HelpText<"Print all of Clang's warning options">;
+  HelpText<"Print all of Clang's warning options">,
+  Visibility<[ClangOption, CLOption]>;
 def private__bundle : Flag<["-"], "private_bundle">;
 def pthreads : Flag<["-"], "pthreads">;
 defm pthread : BoolOption<"", "pthread",
@@ -5227,7 +5240,7 @@ def resource_dir_EQ : Joined<["-"], "resource-dir=">, Flags<[NoXarchOption]>,
   Visibility<[ClangOption, CLOption, DXCOption]>,
   Alias<resource_dir>;
 def rpath : Separate<["-"], "rpath">, Flags<[LinkerInput]>, Group<Link_Group>;
-def rtlib_EQ : Joined<["-", "--"], "rtlib=">,
+def rtlib_EQ : Joined<["-", "--"], "rtlib=">, Visibility<[ClangOption, CLOption]>,
   HelpText<"Compiler runtime library to use">;
 def frtlib_add_rpath: Flag<["-"], "frtlib-add-rpath">, Flags<[NoArgumentUnused]>,
   HelpText<"Add -rpath with architecture-specific resource directory to the linker flags. "
@@ -5393,7 +5406,7 @@ def w : Flag<["-"], "w">, HelpText<"Suppress all warnings">,
   MarshallingInfoFlag<DiagnosticOpts<"IgnoreWarnings">>;
 def x : JoinedOrSeparate<["-"], "x">,
 Flags<[NoXarchOption]>,
-  Visibility<[ClangOption, CC1Option, FlangOption, FC1Option]>,
+  Visibility<[ClangOption, CC1Option, FlangOption, FC1Option, CLOption]>,
   HelpText<"Treat subsequent input files as having type <language>">,
   MetaVarName<"<language>">;
 def y : Joined<["-"], "y">;
diff --git a/clang/lib/Driver/Driver.cpp b/clang/lib/Driver/Driver.cpp
index 77328e1f99e5021..f5fd900a6447fe3 100644
--- a/clang/lib/Driver/Driver.cpp
+++ b/clang/lib/Driver/Driver.cpp
@@ -2589,8 +2589,11 @@ void Driver::BuildInputs(const ToolChain &TC, DerivedArgList &Args,
       Diag(clang::diag::note_drv_t_option_is_global);
   }
 
+  // CUDA/HIP and their preprocessor expansions can be accepted by CL mode.
   // Warn -x after last input file has no effect
-  if (!IsCLMode()) {
+  auto LastXArg = Args.getLastArgValue(options::OPT_x);
+  const llvm::StringSet<> ValidXArgs = {"cuda", "hip", "cui", "hipi"};
+  if (!IsCLMode() || ValidXArgs.find(LastXArg) != ValidXArgs.end()) {
     Arg *LastXArg = Args.getLastArgNoClaim(options::OPT_x);
     Arg *LastInputArg = Args.getLastArgNoClaim(options::OPT_INPUT);
     if (LastXArg && LastInputArg &&
diff --git a/clang/test/Driver/cl-offload.cu b/clang/test/Driver/cl-offload.cu
new file mode 100644
index 000000000000000..aa5c096338110b0
--- /dev/null
+++ b/clang/test/Driver/cl-offload.cu
@@ -0,0 +1,27 @@
+// RUN: %clang_cl -### --offload-arch=sm_35 -fgpu-rdc \
+// RUN:   --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
+// RUN:   /Wall -x cuda %s 2>&1 \
+// RUN:   | FileCheck %s -check-prefix=CUDA
+
+// RUN: %clang_cl -### --offload-arch=gfx1010 -fgpu-rdc --hip-link \
+// RUN:   --rocm-path=%S/Inputs/rocm /Wall -x hip %s 2>&1 \
+// RUN:   | FileCheck %s -check-prefix=HIP
+
+// CUDA: "-cc1" "-triple" "nvptx64-nvidia-cuda" "-aux-triple" "x86_64-pc-windows-msvc"
+// CUDA-SAME: "-Weverything"
+// CUDA: ptxas
+// CUDA: "-cc1" "-triple" "x86_64-pc-windows-msvc{{.*}}" "-aux-triple" "nvptx64-nvidia-cuda"
+// CUDA-SAME: "-Weverything"
+// CUDA: link
+
+// HIP: "-cc1" "-triple" "x86_64-pc-windows-msvc{{.*}}" "-aux-triple" "amdgcn-amd-amdhsa"
+// HIP-SAME: "-Weverything"
+// HIP: "-cc1" "-triple" "amdgcn-amd-amdhsa" "-aux-triple" "x86_64-pc-windows-msvc"
+// HIP-SAME: "-Weverything"
+// HIP: {{lld.* "-flavor" "gnu" "-m" "elf64_amdgpu"}}
+// HIP: {{link.* "amdhip64.lib"}}
+
+// CMake uses this option when finding packages for HIP, so
+// make sure it does not cause error.
+
+// RUN: %clang_cl --print-libgcc-file-name

Artem-B · 2023-10-13T00:41:00Z

@rnk -- would that be acceptable for clang-cl on windows?

yxsamliu · 2023-10-13T13:52:10Z

The main objective is to make porting cmake files easier. Without it, you have to use clang-cl.exe to compile C++ and clang.exe to compile HIP since the options are not compatible. Now you can use clang-cl.exe to compile both, which accepts both MSVC options and clang options.

yxsamliu · 2023-10-19T14:43:19Z

ping. we tested it on Windows and it is able to compile both C++ and HIP without issues.

rnk

Yep, we are fairly liberal about adding new clang options to the clang-cl mode, especially if they are in the feature -f flag namespace, which we believe will not conflict with MSVC flags.

nico · 2023-10-20T00:03:02Z

Hello, it looks like this breaks check-clang on mac: http://45.33.8.238/macm1/71368/step_7.txt

Please take a look and revert for now if it takes a while to fix.

yxsamliu · 2023-10-20T01:42:44Z

[AMD Official Use Only - General] Will fix it now. Thanks. Sam From: Nico Weber ***@***.***> Sent: Thursday, October 19, 2023 8:03 PM To: llvm/llvm-project ***@***.***> Cc: Liu, Yaxun (Sam) ***@***.***>; State change ***@***.***> Subject: Re: [llvm/llvm-project] Let clang-cl support CUDA/HIP (PR #68921) Caution: This message originated from an External Source. Use proper caution when opening attachments, clicking links, or responding. Hello, it looks like this breaks check-clang on mac: http://45.33.8.238/macm1/71368/step_7.txt Please take a look and revert for now if it takes a while to fix. — Reply to this email directly, view it on GitHub<#68921 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ABO4LZ3O6C4EVLAX6NBK52LYAG5UFAVCNFSM6AAAAAA56BAQB6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTONZRHA3DGOJZHE>. You are receiving this because you modified the open/close state.Message ID: ***@***.******@***.***>>

yxsamliu · 2023-10-20T01:50:01Z

[AMD Official Use Only - General] Fixed by 9d10fbb From: Nico Weber ***@***.***> Sent: Thursday, October 19, 2023 8:03 PM To: llvm/llvm-project ***@***.***> Cc: Liu, Yaxun (Sam) ***@***.***>; State change ***@***.***> Subject: Re: [llvm/llvm-project] Let clang-cl support CUDA/HIP (PR #68921) Caution: This message originated from an External Source. Use proper caution when opening attachments, clicking links, or responding. Hello, it looks like this breaks check-clang on mac: http://45.33.8.238/macm1/71368/step_7.txt Please take a look and revert for now if it takes a while to fix. — Reply to this email directly, view it on GitHub<#68921 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ABO4LZ3O6C4EVLAX6NBK52LYAG5UFAVCNFSM6AAAAAA56BAQB6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTONZRHA3DGOJZHE>. You are receiving this because you modified the open/close state.Message ID: ***@***.******@***.***>>

rorth · 2023-10-23T21:44:42Z

This patch also broke the Solaris/amd64 buildbot. I suspect this happens because that one is configured with -DLLVM_TARGETS_TO_BUILD=X86, thus lacking cuda support. In fact, when manually building with support for all targets, the test PASSes. Please fix.

yxsamliu · 2023-10-23T22:15:52Z

[AMD Official Use Only - General] Sorry I missed this failure. Will fix it now. Sam From: Rainer Orth ***@***.***> Sent: Monday, October 23, 2023 5:45 PM To: llvm/llvm-project ***@***.***> Cc: Liu, Yaxun (Sam) ***@***.***>; State change ***@***.***> Subject: Re: [llvm/llvm-project] Let clang-cl support CUDA/HIP (PR #68921) Caution: This message originated from an External Source. Use proper caution when opening attachments, clicking links, or responding. This patch also broke the Solaris/amd64 buildbot<https://lab.llvm.org/staging/#/builders/8/builds/484>. I suspect this happens because that one is configured with -DLLVM_TARGETS_TO_BUILD=X86, thus lacking cuda support. In fact, when manually building with support for all targets, the test PASSes. Please fix. — Reply to this email directly, view it on GitHub<#68921 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ABO4LZ7WVOQA43EQZHLLFHLYA3QNPAVCNFSM6AAAAAA56BAQB6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTONZWGA3TANZSGA>. You are receiving this because you modified the open/close state.Message ID: ***@***.******@***.***>>

statham-arm · 2023-10-24T15:39:51Z

@yxsamliu I've just raised #70055 which fixes an issue with the new test here. Perhaps it might also allow you to remove the exclusion for system-darwin?

yxsamliu · 2023-10-24T15:43:19Z

@yxsamliu I've just raised #70055 which fixes an issue with the new test here. Perhaps it might also allow you to remove the exclusion for system-darwin?

Yes it should allow it to remove exclusion of system-darwin. Thanks

After llvm#68921, clang-cl gained option `-x` but only for CUDA/HIP. This commit simply removes the restriction on parameters to `-x`. Especially, it is able to use `-x c++-module` and `-x c++-user-header` to build C++20 modules and header units with clang-cl. This effectively reverts commit fe08212. Closes llvm#88006.

After llvm#68921, clang-cl gained option `-x` but only for CUDA/HIP. This commit simply removes the restriction on parameters to `-x`. Especially, it is able to use `-x c++-module` and `-x c++-system-header` to build C++20 modules and header units with clang-cl. This effectively reverts commit fe08212. Closes llvm#88006.

After #68921, clang-cl gained option `-x` but only for CUDA/HIP. This commit simply removes the restriction on parameters to `-x`. Especially, it is able to use `-x c++-module` and `-x c++-system-header` to build C++20 modules and header units with clang-cl. This effectively reverts commit fe08212. Closes #88006.

yxsamliu requested review from MaskRay and Artem-B October 12, 2023 18:38

llvmbot added clang Clang issues not falling into any other category clang:driver 'clang' and 'clang++' user-facing binaries. Not 'clang-cl' labels Oct 12, 2023

Artem-B requested a review from rnk October 13, 2023 00:40

rnk approved these changes Oct 19, 2023

View reviewed changes

yxsamliu merged commit e880e8a into llvm:main Oct 19, 2023

madhur13490 mentioned this pull request Oct 20, 2023

Revert commit ba8565fbcb975e2d067ce3ae5a7dbaae4953edd3 madhur13490/llvm-project#3

Closed

banach-space mentioned this pull request Oct 24, 2023

[mlir][vector] Add scalable vectors to tests for vector.contract #70039

Merged

This was referenced Apr 8, 2024

Support -x for clang-cl #88006

Closed

Clang-CL 15.x reporting warning when using /clang:-xc #59307

Closed

huangqinjin mentioned this pull request Apr 23, 2024

[clang][driver] Support -x for all languages in CL mode #89772

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Let clang-cl support CUDA/HIP #68921

Let clang-cl support CUDA/HIP #68921

Uh oh!

yxsamliu commented Oct 12, 2023

Uh oh!

llvmbot commented Oct 12, 2023 •

edited

Loading

Uh oh!

Artem-B commented Oct 13, 2023

Uh oh!

yxsamliu commented Oct 13, 2023

Uh oh!

yxsamliu commented Oct 19, 2023

Uh oh!

rnk left a comment

Uh oh!

nico commented Oct 20, 2023

Uh oh!

yxsamliu commented Oct 20, 2023 via email

Uh oh!

yxsamliu commented Oct 20, 2023 via email

Uh oh!

rorth commented Oct 23, 2023

Uh oh!

yxsamliu commented Oct 23, 2023 via email

Uh oh!

statham-arm commented Oct 24, 2023

Uh oh!

yxsamliu commented Oct 24, 2023

Uh oh!

Uh oh!

Let clang-cl support CUDA/HIP #68921

Let clang-cl support CUDA/HIP #68921

Uh oh!

Conversation

yxsamliu commented Oct 12, 2023

Uh oh!

llvmbot commented Oct 12, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Artem-B commented Oct 13, 2023

Uh oh!

yxsamliu commented Oct 13, 2023

Uh oh!

yxsamliu commented Oct 19, 2023

Uh oh!

rnk left a comment

Choose a reason for hiding this comment

Uh oh!

nico commented Oct 20, 2023

Uh oh!

yxsamliu commented Oct 20, 2023 via email

Uh oh!

yxsamliu commented Oct 20, 2023 via email

Uh oh!

rorth commented Oct 23, 2023

Uh oh!

yxsamliu commented Oct 23, 2023 via email

Uh oh!

statham-arm commented Oct 24, 2023

Uh oh!

yxsamliu commented Oct 24, 2023

Uh oh!

Uh oh!

llvmbot commented Oct 12, 2023 •

edited

Loading