Skip to content

[clang] Adjust -mlarge-data-threshold handling #77958

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Jan 12, 2024

Conversation

aeubanks
Copy link
Contributor

@aeubanks aeubanks commented Jan 12, 2024

Make it apply to x86-64 medium and large code models since that's what the backend does.

Limit logic to exclude x86-32.

Default to 0, let the driver set it to 65536 for the medium code model
if one is not passed. Set it to 0 for the large code model by default to
match gcc and since some users make assumptions about the large code
model that any small data will break.

Make it apply to x86-64 medium and large code models since that's what the backend does.

Warn if it's used for x86-32.

Default to 0, let the driver set it to 65536 for the medium code model
if one is not passed. Set it to 0 for the large code model by default to
match gcc and since some users make assumptions about the large code
model that any small data will break.
@llvmbot llvmbot added clang Clang issues not falling into any other category clang:driver 'clang' and 'clang++' user-facing binaries. Not 'clang-cl' clang:frontend Language frontend issues, e.g. anything involving "Sema" clang:codegen IR generation bugs: mangling, exceptions, etc. labels Jan 12, 2024
@aeubanks aeubanks requested review from rnk and MaskRay January 12, 2024 18:17
@llvmbot
Copy link
Member

llvmbot commented Jan 12, 2024

@llvm/pr-subscribers-clang-codegen
@llvm/pr-subscribers-clang

@llvm/pr-subscribers-clang-driver

Author: Arthur Eubanks (aeubanks)

Changes

Make it apply to x86-64 medium and large code models since that's what the backend does.

Warn if it's used for x86-32.

Default to 0, let the driver set it to 65536 for the medium code model
if one is not passed. Set it to 0 for the large code model by default to
match gcc and since some users make assumptions about the large code
model that any small data will break.


Full diff: https://github.com/llvm/llvm-project/pull/77958.diff

6 Files Affected:

  • (modified) clang/include/clang/Basic/DiagnosticDriverKinds.td (+1-1)
  • (modified) clang/include/clang/Driver/Options.td (+1-1)
  • (modified) clang/lib/CodeGen/CodeGenModule.cpp (+1-1)
  • (modified) clang/lib/Driver/ToolChains/Clang.cpp (+18-9)
  • (modified) clang/test/CodeGen/large-data-threshold.c (+4-1)
  • (modified) clang/test/Driver/large-data-threshold.c (+9-3)
diff --git a/clang/include/clang/Basic/DiagnosticDriverKinds.td b/clang/include/clang/Basic/DiagnosticDriverKinds.td
index 0a8a77fadbeb1b..8b5232a6df3958 100644
--- a/clang/include/clang/Basic/DiagnosticDriverKinds.td
+++ b/clang/include/clang/Basic/DiagnosticDriverKinds.td
@@ -472,7 +472,7 @@ def warn_unsupported_branch_protection: Warning <
 def err_sls_hardening_arm_not_supported : Error<
   "-mharden-sls is only supported on armv7-a or later">;
 def warn_drv_large_data_threshold_invalid_code_model: Warning<
-  "'%0' only applies to medium code model">,
+  "'%0' only applies to medium and large code models">,
   InGroup<UnusedCommandLineArgument>;
 
 def note_drv_command_failed_diag_msg : Note<
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index 7f97d6b6faa398..e8afa1ea4126d8 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -4393,7 +4393,7 @@ def mcmodel_EQ : Joined<["-"], "mcmodel=">, Group<m_Group>,
   MarshallingInfoString<TargetOpts<"CodeModel">, [{"default"}]>;
 def mlarge_data_threshold_EQ : Joined<["-"], "mlarge-data-threshold=">, Group<m_Group>,
   Visibility<[ClangOption, CC1Option]>,
-  MarshallingInfoInt<TargetOpts<"LargeDataThreshold">, "65535">;
+  MarshallingInfoInt<TargetOpts<"LargeDataThreshold">, "0">;
 def mtls_size_EQ : Joined<["-"], "mtls-size=">, Group<m_Group>,
   Visibility<[ClangOption, CC1Option]>,
   HelpText<"Specify bit size of immediate TLS offsets (AArch64 ELF only): "
diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index ad6fc71c1e5038..0cfe7a0133b7e3 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -1201,7 +1201,7 @@ void CodeGenModule::Release() {
       llvm::CodeModel::Model codeModel = static_cast<llvm::CodeModel::Model>(CM);
       getModule().setCodeModel(codeModel);
 
-      if (CM == llvm::CodeModel::Medium &&
+      if ((CM == llvm::CodeModel::Medium || CM == llvm::CodeModel::Large) &&
           Context.getTargetInfo().getTriple().getArch() ==
               llvm::Triple::x86_64) {
         getModule().setLargeDataThreshold(getCodeGenOpts().LargeDataThreshold);
diff --git a/clang/lib/Driver/ToolChains/Clang.cpp b/clang/lib/Driver/ToolChains/Clang.cpp
index 1ee7ae602f3ce5..27a50b2c32e94d 100644
--- a/clang/lib/Driver/ToolChains/Clang.cpp
+++ b/clang/lib/Driver/ToolChains/Clang.cpp
@@ -5752,20 +5752,29 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
     }
   }
 
-  if (Arg *A = Args.getLastArg(options::OPT_mlarge_data_threshold_EQ)) {
-    if (!Triple.isX86()) {
-      D.Diag(diag::err_drv_unsupported_opt_for_target)
-          << A->getOption().getName() << TripleStr;
-    } else {
-      bool IsMediumCM = false;
-      if (Arg *A = Args.getLastArg(options::OPT_mcmodel_EQ))
-        IsMediumCM = StringRef(A->getValue()) == "medium";
-      if (!IsMediumCM) {
+  if (Triple.getArch() == llvm::Triple::x86_64) {
+    bool IsMediumCM = false;
+    bool IsLargeCM = false;
+    if (Arg *A = Args.getLastArg(options::OPT_mcmodel_EQ)) {
+      IsMediumCM = StringRef(A->getValue()) == "medium";
+      IsLargeCM = StringRef(A->getValue()) == "large";
+    }
+    if (Arg *A = Args.getLastArg(options::OPT_mlarge_data_threshold_EQ)) {
+      if (!IsMediumCM && !IsLargeCM) {
         D.Diag(diag::warn_drv_large_data_threshold_invalid_code_model)
             << A->getOption().getRenderName();
       } else {
         A->render(Args, CmdArgs);
       }
+    } else if (IsMediumCM) {
+      CmdArgs.push_back("-mlarge-data-threshold=65536");
+    } else if (IsLargeCM) {
+      CmdArgs.push_back("-mlarge-data-threshold=0");
+    }
+  } else {
+    if (Arg *A = Args.getLastArg(options::OPT_mlarge_data_threshold_EQ)) {
+      D.Diag(diag::err_drv_unsupported_opt_for_target)
+          << A->getOption().getName() << TripleStr;
     }
   }
 
diff --git a/clang/test/CodeGen/large-data-threshold.c b/clang/test/CodeGen/large-data-threshold.c
index 29ae19e9b71899..d110ad2125c7bc 100644
--- a/clang/test/CodeGen/large-data-threshold.c
+++ b/clang/test/CodeGen/large-data-threshold.c
@@ -2,11 +2,14 @@
 
 // RUN: %clang_cc1 -triple x86_64-unknown-unknown -emit-llvm %s -o - -mcmodel=medium | FileCheck %s --check-prefix=IR-DEFAULT
 // RUN: %clang_cc1 -triple x86_64-unknown-unknown -emit-llvm %s -o - -mcmodel=medium -mlarge-data-threshold=200 | FileCheck %s --check-prefix=IR-CUSTOM
+// RUN: %clang_cc1 -triple x86_64-unknown-unknown -emit-llvm %s -o - -mcmodel=large -mlarge-data-threshold=200 | FileCheck %s --check-prefix=IR-CUSTOM
+// RUN: %clang_cc1 -triple x86_64-unknown-unknown -emit-llvm %s -o - -mcmodel=small -mlarge-data-threshold=200 | FileCheck %s --check-prefix=IR-SMALL
 // RUN: %clang_cc1 -triple x86_64-unknown-unknown -S %s -o - -mcmodel=medium -mlarge-data-threshold=200 | FileCheck %s --check-prefix=ASM-SMALL
 // RUN: %clang_cc1 -triple x86_64-unknown-unknown -S %s -o - -mcmodel=medium -mlarge-data-threshold=2 | FileCheck %s --check-prefix=ASM-LARGE
 
-// IR-DEFAULT: !{i32 1, !"Large Data Threshold", i64 65535}
+// IR-DEFAULT: !{i32 1, !"Large Data Threshold", i64 0}
 // IR-CUSTOM: !{i32 1, !"Large Data Threshold", i64 200}
+// IR-SMALL-NOT: !"Large Data Threshold"
 
 // ASM-SMALL-NOT: movabsq
 // ASM-LARGE: movabsq
diff --git a/clang/test/Driver/large-data-threshold.c b/clang/test/Driver/large-data-threshold.c
index 9914286701c6c1..dfc51349f7bf16 100644
--- a/clang/test/Driver/large-data-threshold.c
+++ b/clang/test/Driver/large-data-threshold.c
@@ -1,8 +1,14 @@
+// RUN: %clang --target=x86_64 -### -c -mcmodel=large -mlarge-data-threshold=200 %s 2>&1 | FileCheck --check-prefix=ARG %s
 // RUN: %clang --target=x86_64 -### -c -mcmodel=medium -mlarge-data-threshold=200 %s 2>&1 | FileCheck --check-prefix=ARG %s
-// RUN: %clang --target=x86_64 -### -c -mcmodel=small -mlarge-data-threshold=200 %s 2>&1 | FileCheck --check-prefix=SMALL %s
+// RUN: %clang --target=x86_64 -### -c -mcmodel=large %s 2>&1 | FileCheck --check-prefix=ARG-LARGE-DEFAULT %s
+// RUN: %clang --target=x86_64 -### -c -mcmodel=medium %s 2>&1 | FileCheck --check-prefix=ARG-MEDIUM-DEFAULT %s
+// RUN: %clang --target=x86_64 -### -c -mcmodel=small -mlarge-data-threshold=200 %s 2>&1 | FileCheck --check-prefix=SMALL --implicit-check-not=mlarge-data-threshold %s
+// RUN: not %clang --target=x86 -### -c -mcmodel=medium -mlarge-data-threshold=200 %s 2>&1 | FileCheck --check-prefix=ARCH %s
 // RUN: not %clang --target=riscv32 -### -c -mcmodel=medium -mlarge-data-threshold=200 %s 2>&1 | FileCheck --check-prefix=ARCH %s
 
 // ARG: "-mlarge-data-threshold=200"
+// ARG-MEDIUM-DEFAULT: "-mlarge-data-threshold=65536"
+// ARG-LARGE-DEFAULT: "-mlarge-data-threshold=0"
 
-// SMALL: 'mlarge-data-threshold=' only applies to medium code model
-// ARCH: unsupported option 'mlarge-data-threshold=' for target 'riscv32'
+// SMALL: 'mlarge-data-threshold=' only applies to medium and large code models
+// ARCH: unsupported option 'mlarge-data-threshold=' for target

CmdArgs.push_back("-mlarge-data-threshold=0");
}
} else {
if (Arg *A = Args.getLastArg(options::OPT_mlarge_data_threshold_EQ)) {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You can remove err_drv_unsupported_opt_for_target here in favor of TargetSpecific at def mlarge_data_threshold_EQ

`Flags<[TargetSpecific]>`
remove x86-32 test since the target doesn't accept mcmodel=medium anyway
@aeubanks aeubanks merged commit f05b081 into llvm:main Jan 12, 2024
@aeubanks aeubanks deleted the largedatathreshold branch January 12, 2024 20:23
justinfargnoli pushed a commit to justinfargnoli/llvm-project that referenced this pull request Jan 28, 2024
Make it apply to x86-64 medium and large code models since that's what
the backend does.

Limit logic to exclude x86-32.

Default to 0, let the driver set it to 65536 for the medium code model
if one is not passed. Set it to 0 for the large code model by default to
match gcc and since some users make assumptions about the large code
model that any small data will break.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
clang:codegen IR generation bugs: mangling, exceptions, etc. clang:driver 'clang' and 'clang++' user-facing binaries. Not 'clang-cl' clang:frontend Language frontend issues, e.g. anything involving "Sema" clang Clang issues not falling into any other category
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants