[Flang][AArch64][VecLib] Add libmvec support for Flang/AArch64 #146453
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
`-fveclib=libmvec` for AArch64 (NEON and SVE) in Clang was supported by llvm#143696. This patch does the same for Flang. Vector functions defined in `libmvec` are used for the following Fortran operator and functions currently. - Power operator (`**`) - Fortran intrinsic functions listed below for `real(kind=4)` and `real(kind=8)` (including their coresponding specific intrinsic functions) - Fortran intrinsic functions which are expanded using functions listed below (for example, `sin` for `complex(kind=8)`) ``` sin tan cos asin acos atan (both atan(x) and atan(y, x)) atan2 cosh tanh asinh acosh atanh erf erfc exp log log10 ``` As with Clang/AArch64, glibc 2.40 or higher is required to use all these functions.
|
@llvm/pr-subscribers-clang-driver @llvm/pr-subscribers-backend-aarch64 Author: KAWASHIMA Takahiro (kawashima-fj) Changes
Vector functions defined in
As with Clang/AArch64, glibc 2.40 or higher is required to use all these functions. Full diff: https://github.com/llvm/llvm-project/pull/146453.diff 4 Files Affected:
diff --git a/clang/lib/Driver/ToolChains/Flang.cpp b/clang/lib/Driver/ToolChains/Flang.cpp
index e4e321ba1e195..75ba2af543c7a 100644
--- a/clang/lib/Driver/ToolChains/Flang.cpp
+++ b/clang/lib/Driver/ToolChains/Flang.cpp
@@ -484,11 +484,18 @@ void Flang::addTargetOptions(const ArgList &Args,
Triple.getArch() != llvm::Triple::x86_64)
D.Diag(diag::err_drv_unsupported_opt_for_target)
<< Name << Triple.getArchName();
- } else if (Name == "libmvec" || Name == "AMDLIBM") {
+ } else if (Name == "AMDLIBM") {
if (Triple.getArch() != llvm::Triple::x86 &&
Triple.getArch() != llvm::Triple::x86_64)
D.Diag(diag::err_drv_unsupported_opt_for_target)
<< Name << Triple.getArchName();
+ } else if (Name == "libmvec") {
+ if (Triple.getArch() != llvm::Triple::x86 &&
+ Triple.getArch() != llvm::Triple::x86_64 &&
+ Triple.getArch() != llvm::Triple::aarch64 &&
+ Triple.getArch() != llvm::Triple::aarch64_be)
+ D.Diag(diag::err_drv_unsupported_opt_for_target)
+ << Name << Triple.getArchName();
} else if (Name == "SLEEF" || Name == "ArmPL") {
if (Triple.getArch() != llvm::Triple::aarch64 &&
Triple.getArch() != llvm::Triple::aarch64_be)
diff --git a/flang/docs/ReleaseNotes.md b/flang/docs/ReleaseNotes.md
index 35da8323e0a10..bd3a215323de2 100644
--- a/flang/docs/ReleaseNotes.md
+++ b/flang/docs/ReleaseNotes.md
@@ -34,6 +34,7 @@ page](https://llvm.org/releases/).
* -floop-interchange is now recognized by flang.
* -floop-interchange is enabled by default at -O2 and above.
+* -fveclib=libmvec is supported for AArch64 (same as Flang/x86 and Clang/AArch64)
## Windows Support
diff --git a/flang/test/Driver/fveclib-codegen.f90 b/flang/test/Driver/fveclib-codegen.f90
index 4cbb1e284f18e..88b31da40167e 100644
--- a/flang/test/Driver/fveclib-codegen.f90
+++ b/flang/test/Driver/fveclib-codegen.f90
@@ -1,6 +1,8 @@
! test that -fveclib= is passed to the backend
! RUN: %if aarch64-registered-target %{ %flang -S -Ofast -target aarch64-unknown-linux-gnu -fveclib=SLEEF -o - %s | FileCheck %s --check-prefix=SLEEF %}
! RUN: %if x86-registered-target %{ %flang -S -Ofast -target x86_64-unknown-linux-gnu -fveclib=libmvec -o - %s | FileCheck %s %}
+! RUN: %if aarch64-registered-target %{ %flang -S -Ofast -target aarch64-unknown-linux-gnu -fveclib=libmvec -march=armv8.2-a+nosve -o - %s | FileCheck %s --check-prefix=LIBMVEC-AARCH64-NEON %}
+! RUN: %if aarch64-registered-target %{ %flang -S -Ofast -target aarch64-unknown-linux-gnu -fveclib=libmvec -march=armv8.2-a+sve -o - %s | FileCheck %s --check-prefix=LIBMVEC-AARCH64-SVE %}
! RUN: %if x86-registered-target %{ %flang -S -O3 -ffast-math -target x86_64-unknown-linux-gnu -fveclib=AMDLIBM -o - %s | FileCheck %s --check-prefix=AMDLIBM %}
! RUN: %flang -S -Ofast -fveclib=NoLibrary -o - %s | FileCheck %s --check-prefix=NOLIB
@@ -11,6 +13,8 @@ subroutine sb(a, b)
! check that we used a vectorized call to powf()
! CHECK: _ZGVbN4vv_powf
! SLEEF: _ZGVnN4vv_powf
+! LIBMVEC-AARCH64-NEON: _ZGVnN4vv_powf
+! LIBMVEC-AARCH64-SVE: _ZGVsMxvv_powf
! AMDLIBM: amd_vrs4_powf
! NOLIB: powf
a(i) = a(i) ** b(i)
diff --git a/flang/test/Driver/fveclib.f90 b/flang/test/Driver/fveclib.f90
index 431a4bfc02522..d21e85e486f8d 100644
--- a/flang/test/Driver/fveclib.f90
+++ b/flang/test/Driver/fveclib.f90
@@ -1,6 +1,7 @@
! RUN: %flang -### -c -fveclib=none %s 2>&1 | FileCheck -check-prefix CHECK-NOLIB %s
! RUN: %flang -### -c -fveclib=Accelerate %s 2>&1 | FileCheck -check-prefix CHECK-ACCELERATE %s
! RUN: %flang -### -c --target=x86_64-unknown-linux-gnu -fveclib=libmvec %s 2>&1 | FileCheck -check-prefix CHECK-libmvec %s
+! RUN: %flang -### -c --target=aarch64-unknown-linux-gnu -fveclib=libmvec %s 2>&1 | FileCheck -check-prefix CHECK-libmvec %s
! RUN: %flang -### -c -fveclib=MASSV %s 2>&1 | FileCheck -check-prefix CHECK-MASSV %s
! RUN: %flang -### -c -fveclib=Darwin_libsystem_m %s 2>&1 | FileCheck -check-prefix CHECK-DARWIN_LIBSYSTEM_M %s
! RUN: %flang -### -c --target=aarch64-none-none -fveclib=SLEEF %s 2>&1 | FileCheck -check-prefix CHECK-SLEEF %s
@@ -23,7 +24,7 @@
! RUN: not %flang --target=x86-none-none -c -fveclib=SLEEF %s 2>&1 | FileCheck -check-prefix CHECK-ERROR %s
! RUN: not %flang --target=x86-none-none -c -fveclib=ArmPL %s 2>&1 | FileCheck -check-prefix CHECK-ERROR %s
-! RUN: not %flang --target=aarch64-none-none -c -fveclib=libmvec %s 2>&1 | FileCheck -check-prefix CHECK-ERROR %s
+! RUN: not %flang --target=riscv64-none-none -c -fveclib=libmvec %s 2>&1 | FileCheck -check-prefix CHECK-ERROR %s
! RUN: not %flang --target=aarch64-none-none -c -fveclib=SVML %s 2>&1 | FileCheck -check-prefix CHECK-ERROR %s
! RUN: not %flang --target=aarch64-none-none -c -fveclib=AMDLIBM %s 2>&1 | FileCheck -check-prefix CHECK-ERROR %s
! CHECK-ERROR: unsupported option {{.*}} for target
|
|
@llvm/pr-subscribers-flang-driver Author: KAWASHIMA Takahiro (kawashima-fj) Changes
Vector functions defined in
As with Clang/AArch64, glibc 2.40 or higher is required to use all these functions. Full diff: https://github.com/llvm/llvm-project/pull/146453.diff 4 Files Affected:
diff --git a/clang/lib/Driver/ToolChains/Flang.cpp b/clang/lib/Driver/ToolChains/Flang.cpp
index e4e321ba1e195..75ba2af543c7a 100644
--- a/clang/lib/Driver/ToolChains/Flang.cpp
+++ b/clang/lib/Driver/ToolChains/Flang.cpp
@@ -484,11 +484,18 @@ void Flang::addTargetOptions(const ArgList &Args,
Triple.getArch() != llvm::Triple::x86_64)
D.Diag(diag::err_drv_unsupported_opt_for_target)
<< Name << Triple.getArchName();
- } else if (Name == "libmvec" || Name == "AMDLIBM") {
+ } else if (Name == "AMDLIBM") {
if (Triple.getArch() != llvm::Triple::x86 &&
Triple.getArch() != llvm::Triple::x86_64)
D.Diag(diag::err_drv_unsupported_opt_for_target)
<< Name << Triple.getArchName();
+ } else if (Name == "libmvec") {
+ if (Triple.getArch() != llvm::Triple::x86 &&
+ Triple.getArch() != llvm::Triple::x86_64 &&
+ Triple.getArch() != llvm::Triple::aarch64 &&
+ Triple.getArch() != llvm::Triple::aarch64_be)
+ D.Diag(diag::err_drv_unsupported_opt_for_target)
+ << Name << Triple.getArchName();
} else if (Name == "SLEEF" || Name == "ArmPL") {
if (Triple.getArch() != llvm::Triple::aarch64 &&
Triple.getArch() != llvm::Triple::aarch64_be)
diff --git a/flang/docs/ReleaseNotes.md b/flang/docs/ReleaseNotes.md
index 35da8323e0a10..bd3a215323de2 100644
--- a/flang/docs/ReleaseNotes.md
+++ b/flang/docs/ReleaseNotes.md
@@ -34,6 +34,7 @@ page](https://llvm.org/releases/).
* -floop-interchange is now recognized by flang.
* -floop-interchange is enabled by default at -O2 and above.
+* -fveclib=libmvec is supported for AArch64 (same as Flang/x86 and Clang/AArch64)
## Windows Support
diff --git a/flang/test/Driver/fveclib-codegen.f90 b/flang/test/Driver/fveclib-codegen.f90
index 4cbb1e284f18e..88b31da40167e 100644
--- a/flang/test/Driver/fveclib-codegen.f90
+++ b/flang/test/Driver/fveclib-codegen.f90
@@ -1,6 +1,8 @@
! test that -fveclib= is passed to the backend
! RUN: %if aarch64-registered-target %{ %flang -S -Ofast -target aarch64-unknown-linux-gnu -fveclib=SLEEF -o - %s | FileCheck %s --check-prefix=SLEEF %}
! RUN: %if x86-registered-target %{ %flang -S -Ofast -target x86_64-unknown-linux-gnu -fveclib=libmvec -o - %s | FileCheck %s %}
+! RUN: %if aarch64-registered-target %{ %flang -S -Ofast -target aarch64-unknown-linux-gnu -fveclib=libmvec -march=armv8.2-a+nosve -o - %s | FileCheck %s --check-prefix=LIBMVEC-AARCH64-NEON %}
+! RUN: %if aarch64-registered-target %{ %flang -S -Ofast -target aarch64-unknown-linux-gnu -fveclib=libmvec -march=armv8.2-a+sve -o - %s | FileCheck %s --check-prefix=LIBMVEC-AARCH64-SVE %}
! RUN: %if x86-registered-target %{ %flang -S -O3 -ffast-math -target x86_64-unknown-linux-gnu -fveclib=AMDLIBM -o - %s | FileCheck %s --check-prefix=AMDLIBM %}
! RUN: %flang -S -Ofast -fveclib=NoLibrary -o - %s | FileCheck %s --check-prefix=NOLIB
@@ -11,6 +13,8 @@ subroutine sb(a, b)
! check that we used a vectorized call to powf()
! CHECK: _ZGVbN4vv_powf
! SLEEF: _ZGVnN4vv_powf
+! LIBMVEC-AARCH64-NEON: _ZGVnN4vv_powf
+! LIBMVEC-AARCH64-SVE: _ZGVsMxvv_powf
! AMDLIBM: amd_vrs4_powf
! NOLIB: powf
a(i) = a(i) ** b(i)
diff --git a/flang/test/Driver/fveclib.f90 b/flang/test/Driver/fveclib.f90
index 431a4bfc02522..d21e85e486f8d 100644
--- a/flang/test/Driver/fveclib.f90
+++ b/flang/test/Driver/fveclib.f90
@@ -1,6 +1,7 @@
! RUN: %flang -### -c -fveclib=none %s 2>&1 | FileCheck -check-prefix CHECK-NOLIB %s
! RUN: %flang -### -c -fveclib=Accelerate %s 2>&1 | FileCheck -check-prefix CHECK-ACCELERATE %s
! RUN: %flang -### -c --target=x86_64-unknown-linux-gnu -fveclib=libmvec %s 2>&1 | FileCheck -check-prefix CHECK-libmvec %s
+! RUN: %flang -### -c --target=aarch64-unknown-linux-gnu -fveclib=libmvec %s 2>&1 | FileCheck -check-prefix CHECK-libmvec %s
! RUN: %flang -### -c -fveclib=MASSV %s 2>&1 | FileCheck -check-prefix CHECK-MASSV %s
! RUN: %flang -### -c -fveclib=Darwin_libsystem_m %s 2>&1 | FileCheck -check-prefix CHECK-DARWIN_LIBSYSTEM_M %s
! RUN: %flang -### -c --target=aarch64-none-none -fveclib=SLEEF %s 2>&1 | FileCheck -check-prefix CHECK-SLEEF %s
@@ -23,7 +24,7 @@
! RUN: not %flang --target=x86-none-none -c -fveclib=SLEEF %s 2>&1 | FileCheck -check-prefix CHECK-ERROR %s
! RUN: not %flang --target=x86-none-none -c -fveclib=ArmPL %s 2>&1 | FileCheck -check-prefix CHECK-ERROR %s
-! RUN: not %flang --target=aarch64-none-none -c -fveclib=libmvec %s 2>&1 | FileCheck -check-prefix CHECK-ERROR %s
+! RUN: not %flang --target=riscv64-none-none -c -fveclib=libmvec %s 2>&1 | FileCheck -check-prefix CHECK-ERROR %s
! RUN: not %flang --target=aarch64-none-none -c -fveclib=SVML %s 2>&1 | FileCheck -check-prefix CHECK-ERROR %s
! RUN: not %flang --target=aarch64-none-none -c -fveclib=AMDLIBM %s 2>&1 | FileCheck -check-prefix CHECK-ERROR %s
! CHECK-ERROR: unsupported option {{.*}} for target
|
|
|
MacDue
reviewed
Jul 1, 2025
There was a problem hiding this comment.
You may also want to change Options.td. Currently, the note about needing GLIBC 2.40 is only displayed for Clang as it's under HelpTextForVariants<[ClangOption, CC1Option]:
llvm-project/clang/include/clang/Driver/Options.td
Lines 3491 to 3494 in cb80651
tblah
approved these changes
Jul 1, 2025
tarunprabhu
reviewed
Jul 1, 2025
| @@ -3487,7 +3487,8 @@ def fno_experimental_isel : Flag<["-"], "fno-experimental-isel">, Group<f_clang_ | |||
| Alias<fno_global_isel>; | |||
| def fveclib : Joined<["-"], "fveclib=">, Group<f_Group>, | |||
| Visibility<[ClangOption, CC1Option, FlangOption, FC1Option]>, | |||
| HelpText<"Use the given vector functions library">, | |||
| HelpText<"Use the given vector functions library.\n" | |||
| " Note: -fveclib=libmvec on AArch64 requires GLIBC 2.40 or newer.">, | |||
There was a problem hiding this comment.
Is the constraint on glibc >=2.40 true for clang as well? What happens if the user attempts to build with an older glibc? Will they see a configure-time error or will the build fail?
There was a problem hiding this comment.
Ach. Please ignore the question about the constraint on glibc. The second question about failure modes with older glibc still holds though.
There was a problem hiding this comment.
The toolchain user will see a link time failure when trying to resolve a call to a vector math routine. Note: the exact requirement depends on the function in question, but version 2.40 is what was used when deciding which math functions to add mappings for.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
-fveclib=libmvecfor AArch64 (NEON and SVE) in Clang was supported by #143696. This patch does the same for Flang.Vector functions defined in
libmvecare used for the following Fortran operator and functions currently.**)real(kind=4)andreal(kind=8)(including their coresponding specific intrinsic functions)sinforcomplex(kind=8))As with Clang/AArch64, glibc 2.40 or higher is required to use all these functions.