Merge branch 'sycl' into p2p-fix-query

Author: JackAKirk

.github/workflows/sycl-linux-build.yml

@@ -212,6 +212,12 @@ jobs:
       if: always() && !cancelled() && contains(inputs.changes, 'libdevice')
       run: |
         cmake --build $GITHUB_WORKSPACE/build --target check-libdevice
+    - name: Check E2E test requirements
+      if: always() && !cancelled() && !contains(inputs.changes, 'sycl')
+      run: |
+        # TODO consider moving this to Dockerfile.
+        export LD_LIBRARY_PATH=/usr/local/cuda/compat/:/usr/local/cuda/lib64:$LD_LIBRARY_PATH
+        LIT_OPTS="--allow-empty-runs" LIT_FILTER="e2e_test_requirements" cmake --build $GITHUB_WORKSPACE/build --target check-sycl
     - name: Install
       if: ${{ always() && !cancelled() && steps.build.conclusion == 'success' }}
       # TODO replace utility installation with a single CMake target

clang/lib/Driver/ToolChains/Clang.cpp

@@ -5590,7 +5590,7 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
     // only be used for SPIR/SPIR-V based targets.
     if (Triple.isSPIROrSPIRV())
       if (Args.hasFlag(options::OPT_fsycl_instrument_device_code,
-                       options::OPT_fno_sycl_instrument_device_code, true))
+                       options::OPT_fno_sycl_instrument_device_code, false))
         CmdArgs.push_back("-fsycl-instrument-device-code");
 
     if (!SYCLStdArg) {

clang/lib/Driver/ToolChains/HIPUtility.cpp

@@ -324,7 +324,10 @@ void HIP::constructHIPFatbinCommand(Compilation &C, const JobAction &JA,
       Args.MakeArgString(std::string("-output=").append(Output));
   BundlerArgs.push_back(BundlerOutputArg);
 
-  addOffloadCompressArgs(Args, BundlerArgs);
+  // For SYCL, the compression is occurring during the wrapping step, so we do
+  // not want to do additional compression here.
+  if (!JA.isDeviceOffloading(Action::OFK_SYCL))
+    addOffloadCompressArgs(Args, BundlerArgs);
 
   const char *Bundler = Args.MakeArgString(
       T.getToolChain().GetProgramPath("clang-offload-bundler"));

clang/lib/Driver/ToolChains/SYCL.cpp

@@ -508,7 +508,7 @@ SYCL::getDeviceLibraries(const Compilation &C, const llvm::Triple &TargetTriple,
   }
 
   if (Args.hasFlag(options::OPT_fsycl_instrument_device_code,
-                   options::OPT_fno_sycl_instrument_device_code, true))
+                   options::OPT_fno_sycl_instrument_device_code, false))
     addLibraries(SYCLDeviceAnnotationLibs);
 
 #if !defined(_WIN32)
@@ -1780,6 +1780,16 @@ void SYCLToolChain::AddImpliedTargetArgs(const llvm::Triple &Triple,
     ArgStringList TargArgs;
     Args.AddAllArgValues(TargArgs, options::OPT_Xs, options::OPT_Xs_separate);
     Args.AddAllArgValues(TargArgs, options::OPT_Xsycl_backend);
+    // For -Xsycl-target-backend=<triple> the triple value is used to push
+    // specific options to the matching device compilation using that triple.
+    // Scrutinize this to make sure we are only checking the values needed
+    // for the current device compilation.
+    for (auto *A : Args) {
+      if (!A->getOption().matches(options::OPT_Xsycl_backend_EQ))
+        continue;
+      if (getDriver().MakeSYCLDeviceTriple(A->getValue()) == Triple)
+        TargArgs.push_back(A->getValue(1));
+    }
     // Check for any -device settings.
     std::string DevArg;
     if (IsJIT || Device == "pvc" || hasPVCDevice(TargArgs, DevArg)) {

clang/lib/Sema/SemaSYCL.cpp

@@ -3167,7 +3167,7 @@ class SyclKernelDeclCreator : public SyclKernelFieldHandler {
 //        // code
 //      }
 //
-//      [[intel::reqd_sub_group_size(4)]] void operator()(sycl::id<1> id) const
+//      [[sycl::reqd_sub_group_size(4)]] void operator()(sycl::id<1> id) const
 //      {
 //        // code
 //      }

clang/lib/Sema/SemaSYCLDeclAttr.cpp

@@ -132,6 +132,13 @@ void SemaSYCL::checkDeprecatedSYCLAttributeSpelling(const ParsedAttr &A,
     return;
   }
 
+  // Additionally, diagnose deprecated [[intel::reqd_sub_group_size]] spelling
+  if (A.getKind() == ParsedAttr::AT_IntelReqdSubGroupSize && A.getScopeName() &&
+      A.getScopeName()->isStr("intel")) {
+    diagnoseDeprecatedAttribute(A, "sycl", "reqd_sub_group_size");
+    return;
+  }
+
   // Diagnose SYCL 2020 spellings in later SYCL modes.
   if (getLangOpts().getSYCLVersion() >= LangOptions::SYCL_2020) {
     // All attributes in the cl vendor namespace are deprecated in favor of a

clang/test/CodeGenSYCL/kernel-op-calls.cpp

@@ -11,7 +11,7 @@ class Functor1 {
 public:
  Functor1(){}
 
- [[intel::reqd_sub_group_size(4)]] void operator()(sycl::id<1> id) const {}
+ [[sycl::reqd_sub_group_size(4)]] void operator()(sycl::id<1> id) const {}
 
  [[sycl::work_group_size_hint(1, 2, 3)]] void operator()(sycl::id<2> id) const {}
 

clang/test/CodeGenSYCL/reqd-sub-group-size.cpp

@@ -7,25 +7,25 @@ queue q;
 
 class Functor16 {
 public:
-  [[intel::reqd_sub_group_size(16)]] void operator()() const {}
+  [[sycl::reqd_sub_group_size(16)]] void operator()() const {}
 };
 
 template <int SIZE>
 class Functor2 {
 public:
-  [[intel::reqd_sub_group_size(SIZE)]] void operator()() const {}
+  [[sycl::reqd_sub_group_size(SIZE)]] void operator()() const {}
 };
 
 template <int N>
-[[intel::reqd_sub_group_size(N)]] void func() {}
+[[sycl::reqd_sub_group_size(N)]] void func() {}
 
 int main() {
   q.submit([&](handler &h) {
     Functor16 f16;
     h.single_task<class kernel_name1>(f16);
 
     h.single_task<class kernel_name3>(
-        []() [[intel::reqd_sub_group_size(4)]]{});
+        []() [[sycl::reqd_sub_group_size(4)]]{});
 
     Functor2<2> f2;
     h.single_task<class kernel_name4>(f2);

clang/test/CodeGenSYCL/sycl-multi-kernel-attr.cpp

@@ -7,12 +7,12 @@ queue q;
 
 class Functor {
 public:
-  [[intel::reqd_sub_group_size(4), cl::reqd_work_group_size(32, 16, 16)]] void operator()() const {}
+  [[sycl::reqd_sub_group_size(4), cl::reqd_work_group_size(32, 16, 16)]] void operator()() const {}
 };
 
 class Functor1 {
 public:
-  [[intel::reqd_sub_group_size(2), sycl::reqd_work_group_size(64, 32, 32)]] void operator()() const {}
+  [[sycl::reqd_sub_group_size(2), sycl::reqd_work_group_size(64, 32, 32)]] void operator()() const {}
 };
 
 template <int SIZE, int SIZE1, int SIZE2>

clang/test/Driver/linker-wrapper-sycl-win.cpp

@@ -112,3 +112,7 @@
 // CHK-CMDS-AOT-AMD-NEXT: offload-wrapper: input: [[BUNDLEROUT]], output: [[WRAPPEROUT:.*]].bc
 // CHK-CMDS-AOT-AMD-NEXT: "{{.*}}clang.exe"{{.*}} -c -o [[LLCOUT:.*]].o [[WRAPPEROUT]].bc
 // CHK-CMDS-AOT-AMD-NEXT: "{{.*}}ld" -- HOST_LINKER_FLAGS -dynamic-linker HOST_DYN_LIB -o a.out [[LLCOUT]].o HOST_LIB_PATH HOST_STAT_LIB {{.*}}.o
+
+// Error handling when --linker-path is not provided for clang-linker-wrapper
+// RUN: not clang-linker-wrapper 2>&1 | FileCheck --check-prefix=LINKER-PATH-NOT-PROVIDED %s
+// LINKER-PATH-NOT-PROVIDED: linker path missing, must pass 'linker-path'