[clang][flang][Triple][llvm] Add isOffload function to LangOpts and isGPU function to Triple #126956

sarnex · 2025-02-12T19:04:30Z

I'm adding support for SPIR-V, so let's consolidate these checks.

sarnex · 2025-02-12T21:56:34Z

clang/lib/CodeGen/CGOpenMPRuntime.cpp

-  bool isGPUDistribute =
-      CGM.getLangOpts().OpenMPIsTargetDevice &&
-      (CGM.getTriple().isAMDGCN() || CGM.getTriple().isNVPTX());
+  bool isGPUDistribute = CGM.getLangOpts().OpenMPIsTargetDevice &&


here (and a few other places) we were checking for AMDGCN, but the new code checks for AMDGPU.

im not sure if theyre basically interchangeable in offloading code, or if i need to be careful, or if the existing code is wrong and we should be using one or the other everywhere

tThey are functionally equivalent given that r600 is essentially dead and doesn't support any offload languages. But probably should stick to amdgpu

Think we should do a sed s/isAMDGCN()/isAMDGPU()/g at some point?

thanks, so new code seems fine then

llvmbot · 2025-02-12T22:01:37Z

@llvm/pr-subscribers-flang-fir-hlfir
@llvm/pr-subscribers-flang-openmp
@llvm/pr-subscribers-pgo
@llvm/pr-subscribers-flang-driver
@llvm/pr-subscribers-clang
@llvm/pr-subscribers-clang-driver
@llvm/pr-subscribers-clang-codegen

@llvm/pr-subscribers-compiler-rt-sanitizer

Author: Nick Sarnie (sarnex)

Changes

I'm adding support for SPIR-V, so let's consolidate these checks.

Full diff: https://github.com/llvm/llvm-project/pull/126956.diff

11 Files Affected:

(modified) clang/lib/CodeGen/CGOpenMPRuntime.cpp (+3-4)
(modified) clang/lib/CodeGen/CodeGenModule.cpp (+1-1)
(modified) clang/lib/CodeGen/CodeGenModule.h (+1-2)
(modified) clang/lib/Driver/Driver.cpp (+3-4)
(modified) clang/lib/Driver/ToolChains/Clang.cpp (+2-4)
(modified) clang/lib/Driver/ToolChains/CommonArgs.cpp (+2-2)
(modified) clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp (+1-1)
(modified) llvm/include/llvm/TargetParser/Triple.h (+5)
(modified) llvm/include/llvm/Transforms/IPO/Attributor.h (+1-3)
(modified) llvm/lib/Transforms/IPO/Attributor.cpp (+2-2)
(modified) llvm/lib/Transforms/Instrumentation/SanitizerBinaryMetadata.cpp (+1-1)

diff --git a/clang/lib/CodeGen/CGOpenMPRuntime.cpp b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
index b679d63874b3b..32cba453b3d10 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntime.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -2624,9 +2624,8 @@ void CGOpenMPRuntime::emitDistributeStaticInit(
       emitUpdateLocation(CGF, Loc, OMP_IDENT_WORK_DISTRIBUTE);
   llvm::Value *ThreadId = getThreadID(CGF, Loc);
   llvm::FunctionCallee StaticInitFunction;
-  bool isGPUDistribute =
-      CGM.getLangOpts().OpenMPIsTargetDevice &&
-      (CGM.getTriple().isAMDGCN() || CGM.getTriple().isNVPTX());
+  bool isGPUDistribute = CGM.getLangOpts().OpenMPIsTargetDevice &&
+                         CGM.getTriple().isOffloadingTargetGPU();
   StaticInitFunction = OMPBuilder.createForStaticInitFunction(
       Values.IVSize, Values.IVSigned, isGPUDistribute);
 
@@ -2656,7 +2655,7 @@ void CGOpenMPRuntime::emitForStaticFinish(CodeGenFunction &CGF,
   auto DL = ApplyDebugLocation::CreateDefaultArtificial(CGF, Loc);
   if (isOpenMPDistributeDirective(DKind) &&
       CGM.getLangOpts().OpenMPIsTargetDevice &&
-      (CGM.getTriple().isAMDGCN() || CGM.getTriple().isNVPTX()))
+      CGM.getTriple().isOffloadingTargetGPU())
     CGF.EmitRuntimeCall(
         OMPBuilder.getOrCreateRuntimeFunction(
             CGM.getModule(), OMPRTL___kmpc_distribute_static_fini),
diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index 7924c32fcf633..99c8f6e547e87 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -842,7 +842,7 @@ static void setVisibilityFromDLLStorageClass(const clang::LangOptions &LO,
 static bool isStackProtectorOn(const LangOptions &LangOpts,
                                const llvm::Triple &Triple,
                                clang::LangOptions::StackProtectorMode Mode) {
-  if (Triple.isAMDGPU() || Triple.isNVPTX())
+  if (Triple.isOffloadingTargetGPU())
     return false;
   return LangOpts.getStackProtector() == Mode;
 }
diff --git a/clang/lib/CodeGen/CodeGenModule.h b/clang/lib/CodeGen/CodeGenModule.h
index c6f6fd5b9a7bd..f197c420f6d44 100644
--- a/clang/lib/CodeGen/CodeGenModule.h
+++ b/clang/lib/CodeGen/CodeGenModule.h
@@ -1067,8 +1067,7 @@ class CodeGenModule : public CodeGenTypeCache {
   bool shouldEmitRTTI(bool ForEH = false) {
     return (ForEH || getLangOpts().RTTI) && !getLangOpts().CUDAIsDevice &&
            !(getLangOpts().OpenMP && getLangOpts().OpenMPIsTargetDevice &&
-             (getTriple().isNVPTX() || getTriple().isAMDGPU() ||
-              getTriple().isSPIRV()));
+             getTriple().isOffloadingTargetGPU());
   }
 
   /// Get the address of the RTTI descriptor for the given type.
diff --git a/clang/lib/Driver/Driver.cpp b/clang/lib/Driver/Driver.cpp
index 50941d2aaa429..b1847322add16 100644
--- a/clang/lib/Driver/Driver.cpp
+++ b/clang/lib/Driver/Driver.cpp
@@ -3351,10 +3351,9 @@ class OffloadingActionBuilder final {
 
       const ToolChain *HostTC = C.getSingleOffloadToolChain<Action::OFK_Host>();
       assert(HostTC && "No toolchain for host compilation.");
-      if (HostTC->getTriple().isNVPTX() ||
-          HostTC->getTriple().getArch() == llvm::Triple::amdgcn) {
-        // We do not support targeting NVPTX/AMDGCN for host compilation. Throw
-        // an error and abort pipeline construction early so we don't trip
+      if (HostTC->getTriple().isOffloadingTargetGPU()) {
+        // We do not support targeting offloading GPUs for host compilation.
+        // Throw an error and abort pipeline construction early so we don't trip
         // asserts that assume device-side compilation.
         C.getDriver().Diag(diag::err_drv_cuda_host_arch)
             << HostTC->getTriple().getArchName();
diff --git a/clang/lib/Driver/ToolChains/Clang.cpp b/clang/lib/Driver/ToolChains/Clang.cpp
index 5deafa2ad0f4a..b25a15338d9ca 100644
--- a/clang/lib/Driver/ToolChains/Clang.cpp
+++ b/clang/lib/Driver/ToolChains/Clang.cpp
@@ -1133,8 +1133,7 @@ void Clang::AddPreprocessingOptions(Compilation &C, const JobAction &JA,
   if (JA.isDeviceOffloading(Action::OFK_OpenMP) &&
       !Args.hasArg(options::OPT_nostdinc) &&
       !Args.hasArg(options::OPT_nogpuinc) &&
-      (getToolChain().getTriple().isNVPTX() ||
-       getToolChain().getTriple().isAMDGCN())) {
+      getToolChain().getTriple().isOffloadingTargetGPU()) {
     if (!Args.hasArg(options::OPT_nobuiltininc)) {
       // Add openmp_wrappers/* to our system include path.  This lets us wrap
       // standard library headers.
@@ -1321,8 +1320,7 @@ void Clang::AddPreprocessingOptions(Compilation &C, const JobAction &JA,
     // Without an offloading language we will include these headers directly.
     // Offloading languages will instead only use the declarations stored in
     // the resource directory at clang/lib/Headers/llvm_libc_wrappers.
-    if ((getToolChain().getTriple().isNVPTX() ||
-         getToolChain().getTriple().isAMDGCN()) &&
+    if (getToolChain().getTriple().isOffloadingTargetGPU() &&
         C.getActiveOffloadKinds() == Action::OFK_None) {
       SmallString<128> P(llvm::sys::path::parent_path(D.Dir));
       llvm::sys::path::append(P, "include");
diff --git a/clang/lib/Driver/ToolChains/CommonArgs.cpp b/clang/lib/Driver/ToolChains/CommonArgs.cpp
index 9a4d3f55c911c..b73cd86579e20 100644
--- a/clang/lib/Driver/ToolChains/CommonArgs.cpp
+++ b/clang/lib/Driver/ToolChains/CommonArgs.cpp
@@ -2973,8 +2973,8 @@ void tools::addMCModel(const Driver &D, const llvm::opt::ArgList &Args,
     } else if (Triple.getArch() == llvm::Triple::x86_64) {
       Ok = llvm::is_contained({"small", "kernel", "medium", "large", "tiny"},
                               CM);
-    } else if (Triple.isNVPTX() || Triple.isAMDGPU() || Triple.isSPIRV()) {
-      // NVPTX/AMDGPU/SPIRV does not care about the code model and will accept
+    } else if (Triple.isOffloadingTargetGPU()) {
+      // Offloading GPU targets do not care about the code model and will accept
       // whatever works for the host.
       Ok = true;
     } else if (Triple.isSPARC64()) {
diff --git a/clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp b/clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp
index aa43b2f5f2a1b..99e2c99759a94 100644
--- a/clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp
+++ b/clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp
@@ -512,7 +512,7 @@ Expected<StringRef> clang(ArrayRef<StringRef> InputFiles, const ArgList &Args) {
     CmdArgs.push_back(InputFile);
 
   // If this is CPU offloading we copy the input libraries.
-  if (!Triple.isAMDGPU() && !Triple.isNVPTX() && !Triple.isSPIRV()) {
+  if (!Triple.isOffloadingTargetGPU()) {
     CmdArgs.push_back("-Wl,-Bsymbolic");
     CmdArgs.push_back("-shared");
     ArgStringList LinkerArgs;
diff --git a/llvm/include/llvm/TargetParser/Triple.h b/llvm/include/llvm/TargetParser/Triple.h
index 09c0d223d9b4d..0885d2a44245a 100644
--- a/llvm/include/llvm/TargetParser/Triple.h
+++ b/llvm/include/llvm/TargetParser/Triple.h
@@ -1109,6 +1109,11 @@ class Triple {
            Env == llvm::Triple::EABIHF;
   }
 
+  /// Tests if the target represents a GPU which can be offloaded to.
+  bool isOffloadingTargetGPU() const {
+    return isAMDGPU() || isNVPTX() || isSPIRV();
+  }
+
   /// Tests whether the target supports comdat
   bool supportsCOMDAT() const {
     return !(isOSBinFormatMachO() || isOSBinFormatXCOFF() ||
diff --git a/llvm/include/llvm/Transforms/IPO/Attributor.h b/llvm/include/llvm/Transforms/IPO/Attributor.h
index 995ee54a73ce4..5c2c707e24a14 100644
--- a/llvm/include/llvm/Transforms/IPO/Attributor.h
+++ b/llvm/include/llvm/Transforms/IPO/Attributor.h
@@ -1338,9 +1338,7 @@ struct InformationCache {
   bool stackIsAccessibleByOtherThreads() { return !targetIsGPU(); }
 
   /// Return true if the target is a GPU.
-  bool targetIsGPU() {
-    return TargetTriple.isAMDGPU() || TargetTriple.isNVPTX();
-  }
+  bool targetIsGPU() { return TargetTriple.isOffloadingTargetGPU(); }
 
   /// Return all functions that might be called indirectly, only valid for
   /// closed world modules (see isClosedWorldModule).
diff --git a/llvm/lib/Transforms/IPO/Attributor.cpp b/llvm/lib/Transforms/IPO/Attributor.cpp
index a1e1a51b201b0..ba9148d2aedc7 100644
--- a/llvm/lib/Transforms/IPO/Attributor.cpp
+++ b/llvm/lib/Transforms/IPO/Attributor.cpp
@@ -199,7 +199,7 @@ ChangeStatus &llvm::operator&=(ChangeStatus &L, ChangeStatus R) {
 
 bool AA::isGPU(const Module &M) {
   Triple T(M.getTargetTriple());
-  return T.isAMDGPU() || T.isNVPTX();
+  return T.isOffloadingTargetGPU();
 }
 
 bool AA::isNoSyncInst(Attributor &A, const Instruction &I,
@@ -3296,7 +3296,7 @@ InformationCache::getIndirectlyCallableFunctions(Attributor &A) const {
 }
 
 std::optional<unsigned> InformationCache::getFlatAddressSpace() const {
-  if (TargetTriple.isAMDGPU() || TargetTriple.isNVPTX())
+  if (TargetTriple.isOffloadingTargetGPU())
     return 0;
   return std::nullopt;
 }
diff --git a/llvm/lib/Transforms/Instrumentation/SanitizerBinaryMetadata.cpp b/llvm/lib/Transforms/Instrumentation/SanitizerBinaryMetadata.cpp
index 1759b95ddbc30..85cfff005cb53 100644
--- a/llvm/lib/Transforms/Instrumentation/SanitizerBinaryMetadata.cpp
+++ b/llvm/lib/Transforms/Instrumentation/SanitizerBinaryMetadata.cpp
@@ -133,7 +133,7 @@ class SanitizerBinaryMetadata {
         VersionStr(utostr(getVersion())), IRB(M.getContext()) {
     // FIXME: Make it work with other formats.
     assert(TargetTriple.isOSBinFormatELF() && "ELF only");
-    assert(!(TargetTriple.isNVPTX() || TargetTriple.isAMDGPU()) &&
+    assert(!(TargetTriple.isOffloadingTargetGPU()) &&
            "Device targets are not supported");
   }

llvmbot · 2025-02-12T22:01:37Z

@llvm/pr-subscribers-llvm-transforms

Author: Nick Sarnie (sarnex)

Changes

I'm adding support for SPIR-V, so let's consolidate these checks.

Full diff: https://github.com/llvm/llvm-project/pull/126956.diff

11 Files Affected:

(modified) clang/lib/CodeGen/CGOpenMPRuntime.cpp (+3-4)
(modified) clang/lib/CodeGen/CodeGenModule.cpp (+1-1)
(modified) clang/lib/CodeGen/CodeGenModule.h (+1-2)
(modified) clang/lib/Driver/Driver.cpp (+3-4)
(modified) clang/lib/Driver/ToolChains/Clang.cpp (+2-4)
(modified) clang/lib/Driver/ToolChains/CommonArgs.cpp (+2-2)
(modified) clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp (+1-1)
(modified) llvm/include/llvm/TargetParser/Triple.h (+5)
(modified) llvm/include/llvm/Transforms/IPO/Attributor.h (+1-3)
(modified) llvm/lib/Transforms/IPO/Attributor.cpp (+2-2)
(modified) llvm/lib/Transforms/Instrumentation/SanitizerBinaryMetadata.cpp (+1-1)

diff --git a/clang/lib/CodeGen/CGOpenMPRuntime.cpp b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
index b679d63874b3b..32cba453b3d10 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntime.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -2624,9 +2624,8 @@ void CGOpenMPRuntime::emitDistributeStaticInit(
       emitUpdateLocation(CGF, Loc, OMP_IDENT_WORK_DISTRIBUTE);
   llvm::Value *ThreadId = getThreadID(CGF, Loc);
   llvm::FunctionCallee StaticInitFunction;
-  bool isGPUDistribute =
-      CGM.getLangOpts().OpenMPIsTargetDevice &&
-      (CGM.getTriple().isAMDGCN() || CGM.getTriple().isNVPTX());
+  bool isGPUDistribute = CGM.getLangOpts().OpenMPIsTargetDevice &&
+                         CGM.getTriple().isOffloadingTargetGPU();
   StaticInitFunction = OMPBuilder.createForStaticInitFunction(
       Values.IVSize, Values.IVSigned, isGPUDistribute);
 
@@ -2656,7 +2655,7 @@ void CGOpenMPRuntime::emitForStaticFinish(CodeGenFunction &CGF,
   auto DL = ApplyDebugLocation::CreateDefaultArtificial(CGF, Loc);
   if (isOpenMPDistributeDirective(DKind) &&
       CGM.getLangOpts().OpenMPIsTargetDevice &&
-      (CGM.getTriple().isAMDGCN() || CGM.getTriple().isNVPTX()))
+      CGM.getTriple().isOffloadingTargetGPU())
     CGF.EmitRuntimeCall(
         OMPBuilder.getOrCreateRuntimeFunction(
             CGM.getModule(), OMPRTL___kmpc_distribute_static_fini),
diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index 7924c32fcf633..99c8f6e547e87 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -842,7 +842,7 @@ static void setVisibilityFromDLLStorageClass(const clang::LangOptions &LO,
 static bool isStackProtectorOn(const LangOptions &LangOpts,
                                const llvm::Triple &Triple,
                                clang::LangOptions::StackProtectorMode Mode) {
-  if (Triple.isAMDGPU() || Triple.isNVPTX())
+  if (Triple.isOffloadingTargetGPU())
     return false;
   return LangOpts.getStackProtector() == Mode;
 }
diff --git a/clang/lib/CodeGen/CodeGenModule.h b/clang/lib/CodeGen/CodeGenModule.h
index c6f6fd5b9a7bd..f197c420f6d44 100644
--- a/clang/lib/CodeGen/CodeGenModule.h
+++ b/clang/lib/CodeGen/CodeGenModule.h
@@ -1067,8 +1067,7 @@ class CodeGenModule : public CodeGenTypeCache {
   bool shouldEmitRTTI(bool ForEH = false) {
     return (ForEH || getLangOpts().RTTI) && !getLangOpts().CUDAIsDevice &&
            !(getLangOpts().OpenMP && getLangOpts().OpenMPIsTargetDevice &&
-             (getTriple().isNVPTX() || getTriple().isAMDGPU() ||
-              getTriple().isSPIRV()));
+             getTriple().isOffloadingTargetGPU());
   }
 
   /// Get the address of the RTTI descriptor for the given type.
diff --git a/clang/lib/Driver/Driver.cpp b/clang/lib/Driver/Driver.cpp
index 50941d2aaa429..b1847322add16 100644
--- a/clang/lib/Driver/Driver.cpp
+++ b/clang/lib/Driver/Driver.cpp
@@ -3351,10 +3351,9 @@ class OffloadingActionBuilder final {
 
       const ToolChain *HostTC = C.getSingleOffloadToolChain<Action::OFK_Host>();
       assert(HostTC && "No toolchain for host compilation.");
-      if (HostTC->getTriple().isNVPTX() ||
-          HostTC->getTriple().getArch() == llvm::Triple::amdgcn) {
-        // We do not support targeting NVPTX/AMDGCN for host compilation. Throw
-        // an error and abort pipeline construction early so we don't trip
+      if (HostTC->getTriple().isOffloadingTargetGPU()) {
+        // We do not support targeting offloading GPUs for host compilation.
+        // Throw an error and abort pipeline construction early so we don't trip
         // asserts that assume device-side compilation.
         C.getDriver().Diag(diag::err_drv_cuda_host_arch)
             << HostTC->getTriple().getArchName();
diff --git a/clang/lib/Driver/ToolChains/Clang.cpp b/clang/lib/Driver/ToolChains/Clang.cpp
index 5deafa2ad0f4a..b25a15338d9ca 100644
--- a/clang/lib/Driver/ToolChains/Clang.cpp
+++ b/clang/lib/Driver/ToolChains/Clang.cpp
@@ -1133,8 +1133,7 @@ void Clang::AddPreprocessingOptions(Compilation &C, const JobAction &JA,
   if (JA.isDeviceOffloading(Action::OFK_OpenMP) &&
       !Args.hasArg(options::OPT_nostdinc) &&
       !Args.hasArg(options::OPT_nogpuinc) &&
-      (getToolChain().getTriple().isNVPTX() ||
-       getToolChain().getTriple().isAMDGCN())) {
+      getToolChain().getTriple().isOffloadingTargetGPU()) {
     if (!Args.hasArg(options::OPT_nobuiltininc)) {
       // Add openmp_wrappers/* to our system include path.  This lets us wrap
       // standard library headers.
@@ -1321,8 +1320,7 @@ void Clang::AddPreprocessingOptions(Compilation &C, const JobAction &JA,
     // Without an offloading language we will include these headers directly.
     // Offloading languages will instead only use the declarations stored in
     // the resource directory at clang/lib/Headers/llvm_libc_wrappers.
-    if ((getToolChain().getTriple().isNVPTX() ||
-         getToolChain().getTriple().isAMDGCN()) &&
+    if (getToolChain().getTriple().isOffloadingTargetGPU() &&
         C.getActiveOffloadKinds() == Action::OFK_None) {
       SmallString<128> P(llvm::sys::path::parent_path(D.Dir));
       llvm::sys::path::append(P, "include");
diff --git a/clang/lib/Driver/ToolChains/CommonArgs.cpp b/clang/lib/Driver/ToolChains/CommonArgs.cpp
index 9a4d3f55c911c..b73cd86579e20 100644
--- a/clang/lib/Driver/ToolChains/CommonArgs.cpp
+++ b/clang/lib/Driver/ToolChains/CommonArgs.cpp
@@ -2973,8 +2973,8 @@ void tools::addMCModel(const Driver &D, const llvm::opt::ArgList &Args,
     } else if (Triple.getArch() == llvm::Triple::x86_64) {
       Ok = llvm::is_contained({"small", "kernel", "medium", "large", "tiny"},
                               CM);
-    } else if (Triple.isNVPTX() || Triple.isAMDGPU() || Triple.isSPIRV()) {
-      // NVPTX/AMDGPU/SPIRV does not care about the code model and will accept
+    } else if (Triple.isOffloadingTargetGPU()) {
+      // Offloading GPU targets do not care about the code model and will accept
       // whatever works for the host.
       Ok = true;
     } else if (Triple.isSPARC64()) {
diff --git a/clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp b/clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp
index aa43b2f5f2a1b..99e2c99759a94 100644
--- a/clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp
+++ b/clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp
@@ -512,7 +512,7 @@ Expected<StringRef> clang(ArrayRef<StringRef> InputFiles, const ArgList &Args) {
     CmdArgs.push_back(InputFile);
 
   // If this is CPU offloading we copy the input libraries.
-  if (!Triple.isAMDGPU() && !Triple.isNVPTX() && !Triple.isSPIRV()) {
+  if (!Triple.isOffloadingTargetGPU()) {
     CmdArgs.push_back("-Wl,-Bsymbolic");
     CmdArgs.push_back("-shared");
     ArgStringList LinkerArgs;
diff --git a/llvm/include/llvm/TargetParser/Triple.h b/llvm/include/llvm/TargetParser/Triple.h
index 09c0d223d9b4d..0885d2a44245a 100644
--- a/llvm/include/llvm/TargetParser/Triple.h
+++ b/llvm/include/llvm/TargetParser/Triple.h
@@ -1109,6 +1109,11 @@ class Triple {
            Env == llvm::Triple::EABIHF;
   }
 
+  /// Tests if the target represents a GPU which can be offloaded to.
+  bool isOffloadingTargetGPU() const {
+    return isAMDGPU() || isNVPTX() || isSPIRV();
+  }
+
   /// Tests whether the target supports comdat
   bool supportsCOMDAT() const {
     return !(isOSBinFormatMachO() || isOSBinFormatXCOFF() ||
diff --git a/llvm/include/llvm/Transforms/IPO/Attributor.h b/llvm/include/llvm/Transforms/IPO/Attributor.h
index 995ee54a73ce4..5c2c707e24a14 100644
--- a/llvm/include/llvm/Transforms/IPO/Attributor.h
+++ b/llvm/include/llvm/Transforms/IPO/Attributor.h
@@ -1338,9 +1338,7 @@ struct InformationCache {
   bool stackIsAccessibleByOtherThreads() { return !targetIsGPU(); }
 
   /// Return true if the target is a GPU.
-  bool targetIsGPU() {
-    return TargetTriple.isAMDGPU() || TargetTriple.isNVPTX();
-  }
+  bool targetIsGPU() { return TargetTriple.isOffloadingTargetGPU(); }
 
   /// Return all functions that might be called indirectly, only valid for
   /// closed world modules (see isClosedWorldModule).
diff --git a/llvm/lib/Transforms/IPO/Attributor.cpp b/llvm/lib/Transforms/IPO/Attributor.cpp
index a1e1a51b201b0..ba9148d2aedc7 100644
--- a/llvm/lib/Transforms/IPO/Attributor.cpp
+++ b/llvm/lib/Transforms/IPO/Attributor.cpp
@@ -199,7 +199,7 @@ ChangeStatus &llvm::operator&=(ChangeStatus &L, ChangeStatus R) {
 
 bool AA::isGPU(const Module &M) {
   Triple T(M.getTargetTriple());
-  return T.isAMDGPU() || T.isNVPTX();
+  return T.isOffloadingTargetGPU();
 }
 
 bool AA::isNoSyncInst(Attributor &A, const Instruction &I,
@@ -3296,7 +3296,7 @@ InformationCache::getIndirectlyCallableFunctions(Attributor &A) const {
 }
 
 std::optional<unsigned> InformationCache::getFlatAddressSpace() const {
-  if (TargetTriple.isAMDGPU() || TargetTriple.isNVPTX())
+  if (TargetTriple.isOffloadingTargetGPU())
     return 0;
   return std::nullopt;
 }
diff --git a/llvm/lib/Transforms/Instrumentation/SanitizerBinaryMetadata.cpp b/llvm/lib/Transforms/Instrumentation/SanitizerBinaryMetadata.cpp
index 1759b95ddbc30..85cfff005cb53 100644
--- a/llvm/lib/Transforms/Instrumentation/SanitizerBinaryMetadata.cpp
+++ b/llvm/lib/Transforms/Instrumentation/SanitizerBinaryMetadata.cpp
@@ -133,7 +133,7 @@ class SanitizerBinaryMetadata {
         VersionStr(utostr(getVersion())), IRB(M.getContext()) {
     // FIXME: Make it work with other formats.
     assert(TargetTriple.isOSBinFormatELF() && "ELF only");
-    assert(!(TargetTriple.isNVPTX() || TargetTriple.isAMDGPU()) &&
+    assert(!(TargetTriple.isOffloadingTargetGPU()) &&
            "Device targets are not supported");
   }

shiltian · 2025-02-12T22:17:12Z

llvm/include/llvm/TargetParser/Triple.h

  }

+  /// Tests if the target represents a GPU which can be offloaded to.
+  bool isOffloadingTargetGPU() const {


Is it better to use isOffloadingTarget?

sure, thats good point. i think doing that will resolve your other comment about SPIR-V being a gpu. ill implement this.

shiltian · 2025-02-12T22:17:35Z

llvm/include/llvm/Transforms/IPO/Attributor.h

-  bool targetIsGPU() {
-    return TargetTriple.isAMDGPU() || TargetTriple.isNVPTX();
-  }
+  bool targetIsGPU() { return TargetTriple.isOffloadingTargetGPU(); }


This also implies that SPIR-V is a GPU target triple, but is it true?

shiltian · 2025-02-12T22:19:03Z

clang/lib/CodeGen/CGOpenMPRuntime.cpp

-  bool isGPUDistribute =
-      CGM.getLangOpts().OpenMPIsTargetDevice &&
-      (CGM.getTriple().isAMDGCN() || CGM.getTriple().isNVPTX());
+  bool isGPUDistribute = CGM.getLangOpts().OpenMPIsTargetDevice &&


shiltian

LGTM but I'd like to get a second stamp on it.

jhuber6 · 2025-02-17T16:46:33Z

llvm/include/llvm/TargetParser/Triple.h

I don't think this belongs in regular triple handling. It could be a more generic 'isTargetGPU` but I think there were some objections given that SPIR-V can target other things?

My concern is that this is a triple which just handles a single target, while the concept of 'offloading' is only relevant with an auxiliary triple. I think this is better as a helper in Clang, we could possibly even combine all of these isCUDADevice, isOpenMPDevice checks as well which would make this more useful. (Thinking about this since #127439 has another one of these uses.)

Sorry, was OOO. Will work to address this soon.

Sorry for late response.

So for cases like in the linked PR, we want to know if we are offloading by checking the main/aux triple through the ASTContext. Once we are into the middle end that information won't be available as far as I can tell.

Do you have an idea on how we can detect offloading at all phases in the compiler without solely using the triple?

@jhuber6 Any recommendations on the above?

Offloading is a clang concept, it's usually available through the language info, which should be available all the way to codegen. For example CGM.getLangOpts().OpenMPIsTargetDevice.

Got it, makes sense, will do.

Maybe something like the latest commit is closer to what you had in mind?

Signed-off-by: Sarnie, Nick <[email protected]>

jhuber6

Thanks

Fznamznon

I appreciate the generalization, thanks!

arsenm

The target isn't part of the language, why is it in LangOpts?

sarnex · 2025-03-06T15:17:45Z

The target isn't part of the language, why is it in LangOpts?

If you have a better suggestion I'm all ears, we seem to already have similar stuff in LangOpts such as OMPTargetTriples, OMPHostIRFile, GPUDefaultStream, and CUID.

sarnex · 2025-03-10T18:03:30Z

@arsenm Any comments on the above? Thx

sarnex · 2025-03-17T15:15:48Z

@arsenm Ping :)

jhuber6 · 2025-03-17T16:20:00Z

clang/lib/Frontend/CompilerInvocation.cpp

  }

+  Opts.IsOffloadingTarget =
+      (Opts.OpenMPIsTargetDevice || Opts.SYCLIsDevice || Opts.CUDAIsDevice) &&


I guess the concern is that this isn't strictly a language option since it depends on the target. This is more like "isGPUOffloadingTarget." We could possibly split these into two. One goes in the TargetInfo and one goes in the LangOpts?

Sorry, what would the two flags be? IsOffloading and IsGPU?

Presumably, yeah.

Thanks, back into the depths.

Signed-off-by: Sarnie, Nick <[email protected]>

jhuber6

Seems reasonable so long as we know that SPIR-V is always a GPU. I'll defer to SPIR people.

Signed-off-by: Sarnie, Nick <[email protected]>

hvdijk · 2025-04-10T13:49:05Z

Seems reasonable so long as we know that SPIR-V is always a GPU. I'll defer to SPIR people.

It's not. SPIR-V is mostly-device-agnostic and implementations may run it on GPU, on CPU, or on any other device. I'm looking at this while trying to update 94229, I think isGPU() should return false for SPIR-V, and instead there should be checks for SYCL rather than for SPIR-V as non-SPIR-V SYCL targets need the same handling, does that sound reasonable to you?

arsenm · 2025-04-10T13:52:55Z

Seems reasonable so long as we know that sound reasonable to you?

No. This is a bikeshedding problem on the isGPU name. SPIRV is functionally a "GPU" target. The abstract future physical device it nay run on is unimportant

hvdijk · 2025-04-10T13:55:31Z

It's not a bikeshedding problem that it doesn't handle other SYCL targets consistently though. In some places there are checks for SYCL, in other places there are checks for SPIR-V.

arsenm · 2025-04-10T13:57:22Z

It's not a bikeshedding problem that it doesn't handle other SYCL targets consistently though. In some places there are checks for SYCL, in other places there are checks for SPIR-V.

Usage context wrong is a different problem than whether SPIRV should count as "isGPU"

hvdijk · 2025-04-10T14:00:26Z

I did include "as non-SPIR-V SYCL targets need the same handling" already in my first comment, so not sure why it gets that strong immediate reaction that ignores that part of it.

sarnex commented Feb 12, 2025

View reviewed changes

sarnex marked this pull request as ready for review February 12, 2025 22:01

sarnex requested a review from jhuber6 February 12, 2025 22:02

shiltian reviewed Feb 12, 2025

View reviewed changes

sarnex changed the title ~~[LLVM][Triple][NFCI] Add function to test for GPU offloading triples~~ [LLVM][Triple][NFCI] Add function to test for offloading triples Feb 13, 2025

sarnex requested review from arsenm and shiltian February 14, 2025 15:14

jhuber6 mentioned this pull request Feb 17, 2025

[OpenMP] Silently accept neon_vector_type for the offloading device #127439

Closed

shiltian approved these changes Feb 17, 2025

View reviewed changes

jhuber6 requested changes Feb 17, 2025

View reviewed changes

[clang] Add isOffloadingTarget function to LangOpts

044158c

Signed-off-by: Sarnie, Nick <[email protected]>

sarnex force-pushed the helper branch from 59b8195 to 044158c Compare March 5, 2025 21:09

llvmbot added the clang:frontend Language frontend issues, e.g. anything involving "Sema" label Mar 5, 2025

sarnex changed the title ~~[LLVM][Triple][NFCI] Add function to test for offloading triples~~ [clang] Add isOffloadingTarget function to LangOpts Mar 5, 2025

jhuber6 approved these changes Mar 5, 2025

View reviewed changes

sarnex requested a review from Fznamznon March 5, 2025 21:13

Fznamznon approved these changes Mar 6, 2025

View reviewed changes

arsenm reviewed Mar 6, 2025

View reviewed changes

jhuber6 reviewed Mar 17, 2025

View reviewed changes

add two flags

417a6b0

Signed-off-by: Sarnie, Nick <[email protected]>

llvmbot added flang:driver PGO Profile Guided Optimizations flang Flang issues not falling into any other category labels Mar 21, 2025

jhuber6 approved these changes Mar 21, 2025

View reviewed changes

sarnex added 2 commits March 21, 2025 11:07

Merge remote-tracking branch 'upstream/main'

7e9f6d1

remove newline

3167b3f

Signed-off-by: Sarnie, Nick <[email protected]>

sarnex changed the title ~~[clang] Add isOffloadingTarget function to LangOpts~~ [clang][flang][Triple][llvm] Add isOffload function to LangOpts and isGPU function to Triple Mar 21, 2025

flang fixes

3224283

Signed-off-by: Sarnie, Nick <[email protected]>

llvmbot added flang:fir-hlfir flang:openmp labels Mar 24, 2025

sarnex requested a review from arsenm March 25, 2025 16:19

sarnex merged commit 48b7530 into llvm:main Mar 28, 2025
13 checks passed

[clang][flang][Triple][llvm] Add isOffload function to LangOpts and isGPU function to Triple #126956

[clang][flang][Triple][llvm] Add isOffload function to LangOpts and isGPU function to Triple #126956

Uh oh!

Conversation

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

Uh oh!

Uh oh!

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

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

arsenm Feb 13, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

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

Uh oh!

Uh oh!

llvmbot commented Feb 12, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

shiltian left a comment

Choose a reason for hiding this comment

Uh oh!

jhuber6 Feb 17, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jhuber6 left a comment

Choose a reason for hiding this comment

Uh oh!

Fznamznon left a comment

Choose a reason for hiding this comment

Uh oh!

arsenm left a comment

Choose a reason for hiding this comment

Uh oh!

sarnex commented Mar 6, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

sarnex commented Mar 10, 2025

Uh oh!

sarnex commented Mar 17, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jhuber6 left a comment

sarnex commented Feb 12, 2025 •

edited

Loading

sarnex Feb 12, 2025 •

edited

Loading

arsenm Feb 13, 2025 •

edited

Loading

llvmbot commented Feb 12, 2025 •

edited

Loading

jhuber6 Feb 17, 2025 •

edited

Loading

sarnex commented Mar 6, 2025 •

edited

Loading