[RISCV] Add -march=unset to cancel and ignore a previous -march. #148321

topperc · 2025-07-12T00:57:37Z

-mcpu is used to determine the ISA string if an explicit -march is not present on the command line. If there is a -march present it always has priority over -mcpu regardless of where it appears in the command line.

This can cause issues if -march appears in a Makefile and a user wants to override it with an -mcpu on the command line. The user would need to provide a potentially long ISA string to -march that matches the -mcpu in order to override the MakeFile.

This issue can also be seen on Compiler Explorer where the rv64gc toolchain is passed -march=rv64gc before any user command line options are added. If you pass -mcpu=spacemit-x60, vectors will not be enabled due to the hidden -march.

This patch adds a new option for -march, "unset" that makes the -march ignored for purposes of prioritizing over -mcpu. Now a user can write -march=unset -mcpu=<cpu_name>. This is also implemented by gcc for ARM32.

An alternative would be to allow -march to take a cpu name, but that requires "-march=<cpu_name> -mcpu=<cpu_name>" or "-march=<cpu_name> -mtune=<cpu_name>" to ensure the tune cpu also gets updated. IMHO, needing to repeat the CPU name twice isn't friendly and invites mistakes.

More discussion here riscv-non-isa/riscv-toolchain-conventions#88 I won't merge this until we reach consensus there.

-mcpu is used to determine the ISA string if an explicit -march is not present on the command line. If there is a -march present it always has priority over -mcpu regardless of where it appears in the command line. This can cause issues if -march appears in a Makefile and a user wants to override it with an -mcpu on the command line. The user would need to provide a potentially long ISA string to -march that matches the -mcpu in order to override the MakeFile. This issue also shows up on Compiler Explorer where the rv64gc toolchain is passed -march=rv64gc before any user command line options are added. This patch adds a new option for -march, "unset" that makes the -march ignored for purposes of prioritizing over -mcpu. Now a user can write -march=unset -mcpu=<cpu_name>. This is also implemented by gcc for ARM32. An alternative would be to allow -march to take a cpu name, but that requires "-march=<cpu_name> -mcpu=<cpu_name>" or "-march=<cpu_name> -mcpu=<cpu_name>" to ensure the tune cpu also gets updated. IMHO, needing to repeat the CPU name twice isn't friendly and invites mistakes. More discussion here riscv-non-isa/riscv-toolchain-conventions#88 I won't merge this until we reach consensus there.

llvmbot · 2025-07-12T00:58:10Z

@llvm/pr-subscribers-backend-risc-v

@llvm/pr-subscribers-clang-driver

Author: Craig Topper (topperc)

Changes

-mcpu is used to determine the ISA string if an explicit -march is not present on the command line. If there is a -march present it always has priority over -mcpu regardless of where it appears in the command line.

This can cause issues if -march appears in a Makefile and a user wants to override it with an -mcpu on the command line. The user would need to provide a potentially long ISA string to -march that matches the -mcpu in order to override the MakeFile.

This issue can also be seen on Compiler Explorer where the rv64gc toolchain is passed -march=rv64gc before any user command line options are added. If you pass -mcpu=spacemit-x60, vectors will not be enabled due to the hidden -march.

This patch adds a new option for -march, "unset" that makes the -march ignored for purposes of prioritizing over -mcpu. Now a user can write -march=unset -mcpu=<cpu_name>. This is also implemented by gcc for ARM32.

An alternative would be to allow -march to take a cpu name, but that requires "-march=<cpu_name> -mcpu=<cpu_name>" or "-march=<cpu_name> -mcpu=<cpu_name>" to ensure the tune cpu also gets updated. IMHO, needing to repeat the CPU name twice isn't friendly and invites mistakes.

More discussion here riscv-non-isa/riscv-toolchain-conventions#88 I won't merge this until we reach consensus there.

Full diff: https://github.com/llvm/llvm-project/pull/148321.diff

2 Files Affected:

(modified) clang/lib/Driver/ToolChains/Arch/RISCV.cpp (+7-4)
(modified) clang/test/Driver/riscv-cpus.c (+5)

diff --git a/clang/lib/Driver/ToolChains/Arch/RISCV.cpp b/clang/lib/Driver/ToolChains/Arch/RISCV.cpp
index baa2c8c0bcfb2..76dde0da8e849 100644
--- a/clang/lib/Driver/ToolChains/Arch/RISCV.cpp
+++ b/clang/lib/Driver/ToolChains/Arch/RISCV.cpp
@@ -273,9 +273,12 @@ std::string riscv::getRISCVArch(const llvm::opt::ArgList &Args,
   // Clang does not yet support MULTILIB_REUSE, so we use `rv{XLEN}imafdc`
   // instead of `rv{XLEN}gc` though they are (currently) equivalent.
 
-  // 1. If `-march=` is specified, use it.
-  if (const Arg *A = Args.getLastArg(options::OPT_march_EQ))
-    return A->getValue();
+  // 1. If `-march=` is specified, use it unless the value is "unset".
+  if (const Arg *A = Args.getLastArg(options::OPT_march_EQ)) {
+    StringRef MArch = A->getValue();
+    if (MArch != "unset")
+      return MArch.str();
+  }
 
   // 2. Get march (isa string) based on `-mcpu=`
   if (const Arg *A = Args.getLastArg(options::OPT_mcpu_EQ)) {
@@ -300,7 +303,7 @@ std::string riscv::getRISCVArch(const llvm::opt::ArgList &Args,
 
     StringRef MArch = llvm::RISCV::getMArchFromMcpu(CPU);
     // Bypass if target cpu's default march is empty.
-    if (MArch != "")
+    if (!MArch.empty())
       return MArch.str();
   }
 
diff --git a/clang/test/Driver/riscv-cpus.c b/clang/test/Driver/riscv-cpus.c
index d32d1c1a8183f..661bacce5d22f 100644
--- a/clang/test/Driver/riscv-cpus.c
+++ b/clang/test/Driver/riscv-cpus.c
@@ -403,6 +403,11 @@
 // MCPU-MARCH: "-nostdsysteminc" "-target-cpu" "sifive-e31" "-target-feature" "+m" "-target-feature" "+c"
 // MCPU-MARCH: "-target-abi" "ilp32"
 
+// march=unset erases previous march
+// RUN: %clang --target=riscv32 -### -c %s 2>&1 -march=rv32imc -march=unset -mcpu=sifive-e31 | FileCheck -check-prefix=MARCH-UNSET %s
+// MARCH-UNSET: "-nostdsysteminc" "-target-cpu" "sifive-e31" "-target-feature" "+m" "-target-feature" "+a" "-target-feature" "+c"
+// MARCH-UNSET: "-target-abi" "ilp32"
+
 // Check interaction between mcpu and mtune, mtune won't affect arch related
 // target feature, but mcpu will.
 //

MaskRay

Thanks for implementing -mcpu=unset. This looks good. I'm glad to see -mcpu=unset gaining support, especially after feeling like the lone voice in the GCC thread: https://gcc.gnu.org/pipermail/gcc-patches/2025-June/685537.html

MaskRay · 2025-07-12T02:26:15Z

clang/test/Driver/riscv-cpus.c

+// march=unset erases previous march
+// RUN: %clang --target=riscv32 -### -c %s 2>&1 -march=rv32imc -march=unset -mcpu=sifive-e31 | FileCheck -check-prefix=MARCH-UNSET %s
+// MARCH-UNSET: "-nostdsysteminc" "-target-cpu" "sifive-e31" "-target-feature" "+m" "-target-feature" "+a" "-target-feature" "+c"
+// MARCH-UNSET: "-target-abi" "ilp32"


Add -SAME if applicable.

MaskRay · 2025-07-12T21:54:31Z

clang/test/Driver/riscv-cpus.c

 // MCPU-MARCH: "-nostdsysteminc" "-target-cpu" "sifive-e31" "-target-feature" "+m" "-target-feature" "+c"
 // MCPU-MARCH: "-target-abi" "ilp32"

+// march=unset erases previous march


/// -march=unset erases previous -march.

Always include - when referring to a - option.

MaskRay

Thanks! The reactions include a thumbs-up and a love. Setting my approval.

wangpc-pp

LGTM and I totally support this!

lenary

I don't recall running into this issue, but I'm happy with this solution over accepting cpu names in -march, it feels much clearer.

topperc · 2025-07-14T21:02:10Z

I forgot about tomorrow's LLVM 21 branch so maybe I should merge this based on the support I've received so far?

lenary · 2025-07-15T02:26:56Z

@topperc (thanks autocorrect) what's the status of this in GCC? If it's in there I think we should land/backport this, otherwise after the branch is probably ok.

MaskRay · 2025-07-15T02:49:23Z

I forgot about tomorrow's LLVM 21 branch so maybe I should merge this based on the support I've received so far?

Looks good!

https://github.com/toppers what's the status of this in GCC? If it's in there I think we should land/backport this, otherwise after the branch is probably ok.

https://gcc.gnu.org/releases.html GCC has a X.1 release every year. There should be plenty of time. @kito-cheng

asb

I like this solution too.

kito-cheng · 2025-08-07T02:23:46Z

GCC has landed -march=unset now :) gcc-mirror/gcc@aa1ae52

topperc requested review from asb, lenary, jrtc27, preames, kito-cheng and wangpc-pp July 12, 2025 00:57

llvmbot added clang Clang issues not falling into any other category backend:RISC-V clang:driver 'clang' and 'clang++' user-facing binaries. Not 'clang-cl' labels Jul 12, 2025

topperc mentioned this pull request Jul 12, 2025

Extend the -march syntax to accept CPU name riscv-non-isa/riscv-toolchain-conventions#88

Closed

topperc requested a review from MaskRay July 12, 2025 01:00

fixup! Add release note.

07f1362

MaskRay reviewed Jul 12, 2025

View reviewed changes

fixup! Address review comment

6db85e3

MaskRay reviewed Jul 12, 2025

View reviewed changes

MaskRay approved these changes Jul 12, 2025

View reviewed changes

wangpc-pp approved these changes Jul 14, 2025

View reviewed changes

fixup! address review comment

69b284c

lenary approved these changes Jul 14, 2025

View reviewed changes

asb approved these changes Jul 17, 2025

View reviewed changes

ppenzin approved these changes Jul 17, 2025

View reviewed changes

kito-cheng mentioned this pull request Jul 31, 2025

Document -march=unset option behavior and usage riscv-non-isa/riscv-toolchain-conventions#98

Merged

Merge remote-tracking branch 'origin/main' into pr/unset

96d3e7e

topperc merged commit 92a966b into llvm:main Aug 9, 2025
9 checks passed

topperc deleted the pr/unset branch August 9, 2025 14:23

davidstone mentioned this pull request Aug 16, 2025

[clang][modules] Virtual function with dependent return type included and imported causes compiler to hang #153933

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[RISCV] Add -march=unset to cancel and ignore a previous -march. #148321

[RISCV] Add -march=unset to cancel and ignore a previous -march. #148321

Uh oh!

topperc commented Jul 12, 2025 •

edited

Loading

Uh oh!

llvmbot commented Jul 12, 2025 •

edited

Loading

Uh oh!

MaskRay left a comment

Uh oh!

MaskRay Jul 12, 2025

Uh oh!

MaskRay Jul 12, 2025

Uh oh!

MaskRay left a comment

Uh oh!

wangpc-pp left a comment

Uh oh!

lenary left a comment

Uh oh!

topperc commented Jul 14, 2025

Uh oh!

lenary commented Jul 15, 2025 •

edited

Loading

Uh oh!

MaskRay commented Jul 15, 2025

Uh oh!

asb left a comment

Uh oh!

kito-cheng commented Aug 7, 2025

Uh oh!

Uh oh!

Uh oh!

[RISCV] Add -march=unset to cancel and ignore a previous -march. #148321

[RISCV] Add -march=unset to cancel and ignore a previous -march. #148321

Uh oh!

Conversation

topperc commented Jul 12, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Jul 12, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

MaskRay left a comment

Choose a reason for hiding this comment

Uh oh!

MaskRay Jul 12, 2025

Choose a reason for hiding this comment

Uh oh!

MaskRay Jul 12, 2025

Choose a reason for hiding this comment

Uh oh!

MaskRay left a comment

Choose a reason for hiding this comment

Uh oh!

wangpc-pp left a comment

Choose a reason for hiding this comment

Uh oh!

lenary left a comment

Choose a reason for hiding this comment

Uh oh!

topperc commented Jul 14, 2025

Uh oh!

lenary commented Jul 15, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

MaskRay commented Jul 15, 2025

Uh oh!

asb left a comment

Choose a reason for hiding this comment

Uh oh!

kito-cheng commented Aug 7, 2025

Uh oh!

Uh oh!

Uh oh!

topperc commented Jul 12, 2025 •

edited

Loading

llvmbot commented Jul 12, 2025 •

edited

Loading

lenary commented Jul 15, 2025 •

edited

Loading