[X86][clang-cl] Add CL option /vlen #166375
Merged
+55
−0
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
llvmbot
added
clang
Member
|
@llvm/pr-subscribers-clang-driver @llvm/pr-subscribers-clang Author: Phoebe Wang (phoebewang) ChangesRef: https://learn.microsoft.com/en-us/cpp/build/reference/vlen?view=msvc-170 Full diff: https://github.com/llvm/llvm-project/pull/166375.diff 4 Files Affected:
diff --git a/clang/include/clang/Basic/DiagnosticDriverKinds.td b/clang/include/clang/Basic/DiagnosticDriverKinds.td
index 83980e3ac35b7..e657abb86d95f 100644
--- a/clang/include/clang/Basic/DiagnosticDriverKinds.td
+++ b/clang/include/clang/Basic/DiagnosticDriverKinds.td
@@ -207,6 +207,9 @@ def err_drv_amdgpu_ieee_without_no_honor_nans : Error<
"invalid argument '-mno-amdgpu-ieee' only allowed with relaxed NaN handling">;
def err_drv_argument_not_allowed_with : Error<
"invalid argument '%0' not allowed with '%1'">;
+def warn_drv_argument_not_allowed_with : Warning<
+ "invalid argument '%0' not allowed with '%1'">,
+ InGroup<OptionIgnored>;
def err_drv_cannot_open_randomize_layout_seed_file : Error<
"cannot read randomize layout seed file '%0'">;
def err_drv_invalid_version_number : Error<
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index 11e81e032d5fc..e271444184e44 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -9218,6 +9218,10 @@ def : CLFlag<"Qscatter-">, Alias<mno_scatter>,
def _SLASH_arch : CLCompileJoined<"arch:">,
HelpText<"Set architecture for code generation">;
+def _SLASH_vlen : CLCompileJoined<"vlen">,
+ HelpText<"Set vector length for autovectorization and other optimizations">;
+def _SLASH_vlen_default : CLFlag<"vlen">,
+ HelpText<"Set default vector length for autovectorization and other optimizations">;
def _SLASH_M_Group : OptionGroup<"</M group>">, Group<cl_compile_Group>;
def _SLASH_volatile_Group : OptionGroup<"</volatile group>">,
diff --git a/clang/lib/Driver/ToolChains/Clang.cpp b/clang/lib/Driver/ToolChains/Clang.cpp
index 30d3e5293a31b..add34c63d9c24 100644
--- a/clang/lib/Driver/ToolChains/Clang.cpp
+++ b/clang/lib/Driver/ToolChains/Clang.cpp
@@ -8266,6 +8266,37 @@ void Clang::AddClangCLArgs(const ArgList &Args, types::ID InputType,
<< "/kernel";
}
+ if (Args.hasArg(options::OPT__SLASH_vlen_default)) {
+ // Override /vlen= options.
+ for (const Arg *A : Args.filtered(options::OPT__SLASH_vlen))
+ A->claim();
+ // Nothing to do for AVX512F/AVX512.
+ } else if (const Arg *A = Args.getLastArg(options::OPT__SLASH_vlen)) {
+ StringRef VLen = A->getValue();
+ llvm::SmallSet<std::string, 4> Arch512 = {"AVX512F", "AVX512"};
+ llvm::SmallSet<std::string, 4> Arch256 = {"AVX", "AVX2"};
+
+ StringRef Arch = Args.getLastArgValue(options::OPT__SLASH_arch);
+ if (Arch != "") {
+ if (VLen == "=512") {
+ if (Arch512.contains(Arch.str()))
+ CmdArgs.push_back("-mprefer-vector-width=512");
+ else
+ D.Diag(diag::warn_drv_argument_not_allowed_with)
+ << "/vlen=512" << std::string("/arch:").append(Arch);
+ } else if (VLen == "=256") {
+ if (Arch512.contains(Arch.str()))
+ CmdArgs.push_back("-mprefer-vector-width=256");
+ else if (!Arch256.contains(Arch.str()))
+ D.Diag(diag::warn_drv_argument_not_allowed_with)
+ << "/vlen=256" << std::string("/arch:").append(Arch);
+ } else {
+ D.Diag(diag::warn_drv_unknown_argument_clang_cl)
+ << std::string("/vlen").append(VLen);
+ }
+ }
+ }
+
Arg *MostGeneralArg = Args.getLastArg(options::OPT__SLASH_vmg);
Arg *BestCaseArg = Args.getLastArg(options::OPT__SLASH_vmb);
if (MostGeneralArg && BestCaseArg)
diff --git a/clang/test/Driver/cl-x86-flags.c b/clang/test/Driver/cl-x86-flags.c
index 89526744c0a49..56cfcb420f4ad 100644
--- a/clang/test/Driver/cl-x86-flags.c
+++ b/clang/test/Driver/cl-x86-flags.c
@@ -133,6 +133,27 @@
// tune: "-target-cpu" "sandybridge"
// tune-SAME: "-tune-cpu" "haswell"
+// RUN: %clang_cl -m64 -arch:AVX512 -vlen=512 --target=x86_64-pc-windows -### -- 2>&1 %s | FileCheck -check-prefix=vlen512 %s
+// vlen512: "-mprefer-vector-width=512"
+
+// RUN: %clang_cl -m64 -arch:AVX512 -vlen=256 --target=x86_64-pc-windows -### -- 2>&1 %s | FileCheck -check-prefix=vlen256 %s
+// vlen256: "-mprefer-vector-width=256"
+
+// RUN: %clang_cl -m64 -arch:AVX512 -vlen=512 -vlen --target=x86_64-pc-windows -### -- 2>&1 %s | FileCheck -check-prefix=noprefer %s
+// noprefer-NOT: -mprefer-vector-width
+
+// RUN: %clang_cl -m64 -arch:AVX2 -vlen=512 --target=x86_64-pc-windows -### -- 2>&1 %s | FileCheck -check-prefix=avx2vlen512 %s
+// avx2vlen512: invalid argument '/vlen=512' not allowed with '/arch:AVX2'
+
+// RUN: %clang_cl -m64 -arch:AVX2 -vlen=256 --target=x86_64-pc-windows -### -- 2>&1 %s | FileCheck -check-prefix=avx2vlen256 %s
+// avx2vlen256-NOT: invalid argument
+
+// RUN: %clang_cl -m32 -arch:SSE2 -vlen=256 --target=i386-pc-windows -### -- 2>&1 %s | FileCheck -check-prefix=sse2vlen256 %s
+// sse2vlen256: invalid argument '/vlen=256' not allowed with '/arch:SSE2'
+
+// RUN: %clang_cl -m64 -arch:AVX -vlen256 --target=x86_64-pc-windows -### -- 2>&1 %s | FileCheck -check-prefix=avxvlen256 %s
+// avxvlen256: unknown argument ignored in clang-cl: '/vlen256'
+
void f(void) {
}
|
You can test this locally with the following command:git-clang-format --diff origin/main HEAD --extensions cpp,c -- clang/lib/Driver/ToolChains/Clang.cpp clang/test/Driver/cl-x86-flags.c --diff_from_common_commit
View the diff from clang-format here.diff --git a/clang/lib/Driver/ToolChains/Clang.cpp b/clang/lib/Driver/ToolChains/Clang.cpp
index e20963ac2..5d3d4e741 100644
--- a/clang/lib/Driver/ToolChains/Clang.cpp
+++ b/clang/lib/Driver/ToolChains/Clang.cpp
@@ -8266,29 +8266,29 @@ void Clang::AddClangCLArgs(const ArgList &Args, types::ID InputType,
<< "/kernel";
}
- if (const Arg *A = Args.getLastArg(options::OPT__SLASH_vlen,
- options::OPT__SLASH_vlen_EQ_256,
- options::OPT__SLASH_vlen_EQ_512)) {
- llvm::Triple::ArchType AT = getToolChain().getArch();
- StringRef Default = AT == llvm::Triple::x86 ? "IA32" : "SSE2";
- StringRef Arch = Args.getLastArgValue(options::OPT__SLASH_arch, Default);
-
- if (A->getOption().matches(options::OPT__SLASH_vlen_EQ_512)) {
- if (Arch == "AVX512F" || Arch == "AVX512")
- CmdArgs.push_back("-mprefer-vector-width=512");
- else
- D.Diag(diag::warn_drv_argument_not_allowed_with)
- << "/vlen=512" << std::string("/arch:").append(Arch);
- }
+ if (const Arg *A = Args.getLastArg(options::OPT__SLASH_vlen,
+ options::OPT__SLASH_vlen_EQ_256,
+ options::OPT__SLASH_vlen_EQ_512)) {
+ llvm::Triple::ArchType AT = getToolChain().getArch();
+ StringRef Default = AT == llvm::Triple::x86 ? "IA32" : "SSE2";
+ StringRef Arch = Args.getLastArgValue(options::OPT__SLASH_arch, Default);
+
+ if (A->getOption().matches(options::OPT__SLASH_vlen_EQ_512)) {
+ if (Arch == "AVX512F" || Arch == "AVX512")
+ CmdArgs.push_back("-mprefer-vector-width=512");
+ else
+ D.Diag(diag::warn_drv_argument_not_allowed_with)
+ << "/vlen=512" << std::string("/arch:").append(Arch);
+ }
- if (A->getOption().matches(options::OPT__SLASH_vlen_EQ_256)) {
- if (Arch == "AVX512F" || Arch == "AVX512")
- CmdArgs.push_back("-mprefer-vector-width=256");
- else if (Arch != "AVX" && Arch != "AVX2")
- D.Diag(diag::warn_drv_argument_not_allowed_with)
- << "/vlen=256" << std::string("/arch:").append(Arch);
- }
- }
+ if (A->getOption().matches(options::OPT__SLASH_vlen_EQ_256)) {
+ if (Arch == "AVX512F" || Arch == "AVX512")
+ CmdArgs.push_back("-mprefer-vector-width=256");
+ else if (Arch != "AVX" && Arch != "AVX2")
+ D.Diag(diag::warn_drv_argument_not_allowed_with)
+ << "/vlen=256" << std::string("/arch:").append(Arch);
+ }
+ }
Arg *MostGeneralArg = Args.getLastArg(options::OPT__SLASH_vmg);
Arg *BestCaseArg = Args.getLastArg(options::OPT__SLASH_vmb);
|
Collaborator
Doesn't this add /vlen ? clang-cl already handles /arch |
phoebewang
changed the title
zmodem
reviewed
Nov 6, 2025
Comment on lines
9221
to
9224
| def _SLASH_vlen : CLCompileJoined<"vlen">, | ||
| HelpText<"Set vector length for autovectorization and other optimizations">; | ||
| def _SLASH_vlen_default : CLFlag<"vlen">, | ||
| HelpText<"Set default vector length for autovectorization and other optimizations">; |
Collaborator
There was a problem hiding this comment.
I think these should be _SLASH_vlen (for the default) and _SLASH_vlen_EQ for the one ending in = (and that should be included in the flag spelling.
But, since only two values are allowed, it might be simpler to do something like:
def _SLASH_vlen : CLFLag<"vlen"> ...
def _SLASH_vlen_EQ_256 : CLFLag<"vlen=256"> ...
def _SLASH_vlen_EQ_512 : CLFlag<"vlen=512"> ...
Does Clang provide good diagnostics when passing an -mprefer-vector-width that doesn't fit the target architecture?
If so, could we make _SLASH_vlen_EQ_256 an alias for -mprefer-vector-width=256 and so on, and not have to add any new driver logic at all?
Contributor
Author
There was a problem hiding this comment.
Good idea! No, unfortunately: https://godbolt.org/z/hhvjP8W34
| } | ||
|
|
||
| if (Args.hasArg(options::OPT__SLASH_vlen_default)) { | ||
| // Override /vlen= options. |
Collaborator
There was a problem hiding this comment.
Shouldn't the last flag take precedence? For example, in /arch:AVX512 /vlen /vlen=256, shouldn't 256 win? What does MSVC do?
Contributor
Author
There was a problem hiding this comment.
Yes, MSVC also takes precedence of the last flag: https://godbolt.org/z/43Ga39WhE
zmodem
reviewed
Nov 7, 2025
| StringRef Arch = Args.getLastArgValue(options::OPT__SLASH_arch); | ||
|
|
||
| if (A->getOption().matches(options::OPT__SLASH_vlen_EQ_512)) { | ||
| if (Arch512.contains(Arch.str())) |
Collaborator
There was a problem hiding this comment.
Since the set is so small, and only used here, I think just open-coding the logic would be easier:
if (Arch == "AVX512" || Arch == "AVX512F")
Same for the Arch256 set below.
Contributor
Author
There was a problem hiding this comment.
We will add AVX10.1, AVX10.2 soon, but I can change it for now.
| llvm::SmallSet<std::string, 4> Arch512 = {"AVX512F", "AVX512"}; | ||
| llvm::SmallSet<std::string, 4> Arch256 = {"AVX", "AVX2"}; | ||
|
|
||
| StringRef Arch = Args.getLastArgValue(options::OPT__SLASH_arch); |
Collaborator
There was a problem hiding this comment.
What if the user didn't pass /arch?
What if they passed -mavx512?
Contributor
Author
There was a problem hiding this comment.
Good question. MSVC has a default value for /arch. We should match it here. -mavx512xxx is a single feature, which is orthogonal to /arch. We don't need to check them.
clang/test/Driver/cl-x86-flags.c
Outdated
| // RUN: %clang_cl -m64 -arch:AVX512 -vlen=256 --target=x86_64-pc-windows -### -- 2>&1 %s | FileCheck -check-prefix=vlen256 %s | ||
| // vlen256: "-mprefer-vector-width=256" | ||
|
|
||
| // RUN: %clang_cl -m64 -arch:AVX512 -vlen=512 -vlen --target=x86_64-pc-windows -### -- 2>&1 %s | FileCheck -check-prefix=noprefer %s |
Collaborator
There was a problem hiding this comment.
how about novlen instead of noprefer?
Collaborator
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/27/builds/18645 Here is the relevant piece of the build log for the reference |
vinay-deshmukh
pushed a commit
to vinay-deshmukh/llvm-project
that referenced
this pull request
Nov 8, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Ref: https://learn.microsoft.com/en-us/cpp/build/reference/vlen?view=msvc-170