Merge from 'sycl' to 'sycl-web' (intel#11)

Author: iclsrc

buildbot/configure.py

@@ -26,13 +26,14 @@ def do_configure(args):
                  "-DLLVM_EXTERNAL_PROJECTS=sycl;llvm-spirv",
                  "-DLLVM_EXTERNAL_SYCL_SOURCE_DIR={}".format(sycl_dir),
                  "-DLLVM_EXTERNAL_LLVM_SPIRV_SOURCE_DIR={}".format(spirv_dir),
-                 "-DLLVM_ENABLE_PROJECTS=clang;llvm-spirv;sycl",
+                 "-DLLVM_ENABLE_PROJECTS=clang;sycl;llvm-spirv",
                  "-DOpenCL_INCLUDE_DIR={}".format(ocl_header_dir),
                  "-DOpenCL_LIBRARY={}".format(icd_loader_lib),
                  "-DLLVM_BUILD_TOOLS=OFF",
                  "-DSYCL_ENABLE_WERROR=ON",
                  "-DLLVM_ENABLE_ASSERTIONS=ON",
                  "-DCMAKE_INSTALL_PREFIX={}".format(install_dir),
+                 "-DSYCL_INCLUDE_TESTS=ON", # Explicitly include all kinds of SYCL tests.
                  llvm_dir]
 
     print(cmake_cmd)

clang/lib/CodeGen/CodeGenFunction.cpp

@@ -540,6 +540,11 @@ void CodeGenFunction::EmitOpenCLKernelMetadata(const FunctionDecl *FD,
   if (!FD->hasAttr<OpenCLKernelAttr>())
     return;
 
+  // TODO Module identifier is not reliable for this purpose since two modules
+  // can have the same ID, needs improvement
+  if (getLangOpts().SYCLIsDevice)
+    Fn->addFnAttr("sycl-module-id", Fn->getParent()->getModuleIdentifier());
+
   llvm::LLVMContext &Context = getLLVMContext();
 
   CGM.GenOpenCLArgMetadata(Fn, FD, this);

clang/lib/Driver/ToolChains/MSVC.cpp

@@ -324,7 +324,7 @@ void visualstudio::Linker::constructMSVCLibCommand(Compilation &C,
   CmdArgs.push_back(
       C.getArgs().MakeArgString(Twine("-OUT:") + Output.getFilename()));
 
-  SmallString<128> ExecPath(getToolChain().GetProgramPath("lib"));
+  SmallString<128> ExecPath(getToolChain().GetProgramPath("lib.exe"));
   const char *Exec = C.getArgs().MakeArgString(ExecPath);
   C.addCommand(std::make_unique<Command>(JA, *this, Exec, CmdArgs, None));
 }

clang/test/CodeGenSYCL/module-id.cpp

@@ -0,0 +1,14 @@
+// RUN: %clang_cc1 -triple spir64-unknown-linux-sycldevice -std=c++11 -fsycl-is-device -disable-llvm-passes -S -emit-llvm -x c++ %s -o - | FileCheck %s
+
+template <typename name, typename Func>
+__attribute__((sycl_kernel)) void kernel_single_task(Func kernelFunc) {
+  kernelFunc();
+}
+
+int main() {
+  kernel_single_task<class kernel>([]() {});
+  return 0;
+}
+// CHECK: define spir_kernel void @{{.*}}kernel{{.*}}() #[[KERN_ATTR:[0-9]+]]
+
+// CHECK: #[[KERN_ATTR]] = { {{.*}}"sycl-module-id"="{{.*}}module-id.cpp"{{.*}} }

clang/test/Driver/sycl-offload-intelfpga.cpp

@@ -30,6 +30,17 @@
 // CHK-FPGA-IMAGE: aoc{{.*}} "-o" "[[OUTPUT5:.+\.aocx]]" "[[OUTPUT4]]" "-sycl"
 // CHK-FPGA-LINK: {{lib|llvm-ar}}{{.*}}
 
+/// -fintelfpga -fsycl-link clang-cl specific
+// RUN:  touch %t.obj
+// RUN:  %clang_cl -### -clang:-target -clang:x86_64-pc-windows-msvc -fsycl -fintelfpga -fsycl-link %t.obj 2>&1 \
+// RUN:  | FileCheck -check-prefixes=CHK-FPGA-LINK-WIN %s
+// CHK-FPGA-LINK-WIN: clang-offload-bundler{{.*}} "-type=o" "-targets=host-x86_64-pc-windows-msvc,sycl-spir64_fpga-unknown-{{.*}}-sycldevice{{.*}}" "-inputs=[[INPUT:.+\.obj]]" "-outputs=[[OUTPUT1:.+\.obj]],[[OUTPUT2:.+\.obj]]" "-unbundle"
+// CHK-FPGA-LINK-WIN: llvm-link{{.*}} "[[OUTPUT2]]" "-o" "[[OUTPUT3:.+\.bc]]"
+// CHK-FPGA-LINK-WIN: llvm-spirv{{.*}} "-spirv-max-version=1.1" "-spirv-ext=+all" "-o" "[[OUTPUT4:.+\.spv]]" "[[OUTPUT3]]"
+// CHK-FPGA-LINK-WIN: aoc{{.*}} "-o" "[[OUTPUT5:.+\.aocr]]" "[[OUTPUT4]]" "-sycl" "-rtl"
+// CHK-FPGA-LINK-WIN: lib.exe{{.*}}
+
+
 /// Check -fintelfpga -fsycl-link with an FPGA archive
 // Create the dummy archive
 // RUN:  echo "Dummy AOCR image" > %t.aocr

sycl/CMakeLists.txt

@@ -159,9 +159,23 @@ add_custom_target( sycl-toolchain
   COMMENT "Building SYCL compiler toolchain..."
 )
 
-add_subdirectory( test )
-add_subdirectory( unittests )
-add_subdirectory( tools )
+if (NOT DEFINED LLVM_INCLUDE_TESTS)
+  set(LLVM_INCLUDE_TESTS ON)
+endif()
+
+option(SYCL_INCLUDE_TESTS
+  "Generate build targets for the SYCL unit tests."
+  ${LLVM_INCLUDE_TESTS})
+
+add_subdirectory(tools)
+
+if(SYCL_INCLUDE_TESTS)
+  if(EXISTS ${LLVM_MAIN_SRC_DIR}/utils/unittest/googletest/include/gtest/gtest.h)
+    add_subdirectory(unittests)
+    list(APPEND SYCL_TEST_DEPS SYCLUnitTests)
+  endif()
+  add_subdirectory(test)
+endif()
 
 # Package deploy support
 # Listed here are component names contributing the package

sycl/doc/extensions/USM/USM.adoc

@@ -528,9 +528,9 @@ float* a = static_cast<float*>(malloc_shared(10*sizeof(float), dev, ctxt));
 float* b = static_cast<float*>(malloc_shared(10*sizeof(float), dev, ctxt));
 float* c = static_cast<float*>(malloc_shared(10*sizeof(float), dev, ctxt));
 
-queue Q;
-auto e = Q.submit([&](handler& cgh) {
-  cgh.parallel_for<class vec_add>(range<1> {10}, [=](id<1> i) {
+auto e = q.submit([&](handler& cgh) {
+  cgh.parallel_for<class vec_add>(range<1> {10}, [=](id<1> ID) {
+    size_t i = ID[0];
     c[i] = a[i] + b[i];
   });
 });

sycl/include/CL/sycl/detail/pi.h

@@ -56,6 +56,7 @@ typedef enum {
   PI_INVALID_CONTEXT = CL_INVALID_CONTEXT,
   PI_INVALID_PLATFORM = CL_INVALID_PLATFORM,
   PI_INVALID_DEVICE = CL_INVALID_DEVICE,
+  PI_INVALID_BINARY = CL_INVALID_BINARY,
   PI_MISALIGNED_SUB_BUFFER_OFFSET = CL_MISALIGNED_SUB_BUFFER_OFFSET,
   PI_OUT_OF_HOST_MEMORY = CL_OUT_OF_HOST_MEMORY
 } _pi_result;
@@ -245,9 +246,20 @@ static const uint8_t PI_DEVICE_BINARY_OFFLOAD_KIND_SYCL = 4;
 /// Target identification strings for
 /// pi_device_binary_struct.DeviceTargetSpec
 ///
+/// A device type represented by a particular target
+/// triple requires specific binary images. We need
+/// to map the image type onto the device target triple
+///
 #define PI_DEVICE_BINARY_TARGET_UNKNOWN "<unknown>"
+/// SPIR-V 32-bit image <-> 32-bit OpenCL device
 #define PI_DEVICE_BINARY_TARGET_SPIRV32 "spir"
-#define PI_DEVICE_BINARY_TARGET_SPIRV64 "spir64";
+/// SPIR-V 64-bit image <-> 64-bit OpenCL device
+#define PI_DEVICE_BINARY_TARGET_SPIRV64 "spir64"
+/// Device-specific binary images produced from SPIR-V 64-bit <->
+/// various triples for specific 64-bit OpenCL devices
+#define PI_DEVICE_BINARY_TARGET_SPIRV64_X86_64 "spir64_x86_64"
+#define PI_DEVICE_BINARY_TARGET_SPIRV64_GEN "spir64_gen"
+#define PI_DEVICE_BINARY_TARGET_SPIRV64_FPGA "spir64_fpga"
 
 /// This struct is a record of the device binary information. If the Kind field
 /// denotes a portable binary type (SPIRV or LLVMIR), the DeviceTargetSpec field
@@ -264,6 +276,13 @@ struct pi_device_binary_struct {
   /// format of the binary data - SPIRV, LLVMIR bitcode,...
   uint8_t Format;
   /// null-terminated string representation of the device's target architecture
+  /// which holds one of:
+  /// PI_DEVICE_BINARY_TARGET_UNKNOWN - unknown
+  /// PI_DEVICE_BINARY_TARGET_SPIRV32 - general value for 32-bit OpenCL devices
+  /// PI_DEVICE_BINARY_TARGET_SPIRV64 - general value for 64-bit OpenCL devices
+  /// PI_DEVICE_BINARY_TARGET_SPIRV64_X86_64 - 64-bit OpenCL CPU device
+  /// PI_DEVICE_BINARY_TARGET_SPIRV64_GEN - GEN GPU device (64-bit OpenCL)
+  /// PI_DEVICE_BINARY_TARGET_SPIRV64_FPGA - 64-bit OpenCL FPGA device
   const char *DeviceTargetSpec;
   /// a null-terminated string; target- and compiler-specific options
   /// which are suggested to use to "build" program at runtime
@@ -405,7 +424,7 @@ pi_result piDevicePartition(
 /// and the IR characteristics.
 ///
 pi_result piextDeviceSelectBinary(
-  pi_device           device, // TODO: does this need to be context?
+  pi_device           device,
   pi_device_binary *  binaries,
   pi_uint32           num_binaries,
   pi_device_binary *  selected_binary);

sycl/include/CL/sycl/detail/scheduler/commands.hpp

@@ -66,7 +66,8 @@ class Command {
     RELEASE,
     MAP_MEM_OBJ,
     UNMAP_MEM_OBJ,
-    UPDATE_REQUIREMENT
+    UPDATE_REQUIREMENT,
+    EMPTY_TASK
   };
 
   Command(CommandType Type, QueueImplPtr Queue, bool UseExclusiveQueue = false);
@@ -123,6 +124,23 @@ class Command {
   std::atomic<bool> MEnqueued;
 };
 
+// The command does nothing during enqueue. The task can be used to implement
+// lock in the graph, or to merge several nodes into one.
+class EmptyCommand : public Command {
+public:
+  EmptyCommand(QueueImplPtr Queue, Requirement *Req)
+      : Command(CommandType::EMPTY_TASK, std::move(Queue)),
+        MStoredRequirement(*Req) {}
+
+  Requirement *getStoredRequirement() { return &MStoredRequirement; }
+
+private:
+  cl_int enqueueImp() override { return CL_SUCCESS; }
+  void printDot(std::ostream &Stream) const override;
+
+  Requirement MStoredRequirement;
+};
+
 // The command enqueues release instance of memory allocated on Host or
 // underlying framework.
 class ReleaseCommand : public Command {

sycl/include/CL/sycl/detail/usm_impl.hpp

@@ -19,8 +19,8 @@ namespace usm {
 void *alignedAlloc(size_t Alignment, size_t Bytes, const context &Ctxt,
                    const device &Dev, cl::sycl::usm::alloc Kind);
 
-void *alignedAlloc(size_t Alignment, size_t Bytes, const context &Ctxt,
-                   cl::sycl::usm::alloc Kind);
+void *alignedAllocHost(size_t Alignment, size_t Bytes, const context &Ctxt,
+                       cl::sycl::usm::alloc Kind);
 
 void free(void *Ptr, const context &Ctxt);