Merge branch 'sycl' into buildbot

bader · web-flow · commit 162703f52555 · 2020-04-04T13:28:43.000+03:00
diff --git a/clang/include/clang/Basic/DiagnosticDriverKinds.td b/clang/include/clang/Basic/DiagnosticDriverKinds.td
@@ -255,6 +255,8 @@ def err_drv_expecting_fopenmp_with_fopenmp_targets : Error<
   "The option -fopenmp-targets must be used in conjunction with a -fopenmp option compatible with offloading, please use -fopenmp=libomp or -fopenmp=libiomp5.">;
 def err_drv_expecting_fsycl_with_sycl_opt : Error<
   "The option %0 must be used in conjunction with -fsycl to enable offloading.">;
+def err_drv_fsycl_with_c_type : Error<
+  "The option %0%1 must not be used in conjunction with -fsycl which expects C++ source.">;
 def warn_drv_omp_offload_target_duplicate : Warning<
   "The OpenMP offloading target '%0' is similar to target '%1' already specified - will be ignored.">,
   InGroup<OpenMPTarget>;
diff --git a/clang/lib/Driver/Driver.cpp b/clang/lib/Driver/Driver.cpp
@@ -769,6 +769,15 @@ void Driver::CreateOffloadingDeviceToolChains(Compilation &C,
     }
     return SYCLArg;
   };
+
+  // Emit an error if c-compilation is forced in -fsycl mode
+  if (HasValidSYCLRuntime)
+    for (StringRef XValue : C.getInputArgs().getAllArgValues(options::OPT_x)) {
+      if (XValue == "c" || XValue == "c-header")
+        C.getDriver().Diag(clang::diag::err_drv_fsycl_with_c_type)
+            << "-x " << XValue;
+    }
+
   Arg *SYCLTargets = getArgRequiringSYCLRuntime(options::OPT_fsycl_targets_EQ);
   Arg *SYCLLinkTargets =
       getArgRequiringSYCLRuntime(options::OPT_fsycl_link_targets_EQ);
diff --git a/clang/test/Driver/sycl-offload.c b/clang/test/Driver/sycl-offload.c
@@ -795,5 +795,10 @@
 // LIB-NODEVICE: 1: linker, {0}, image, (host-sycl)
 // LIB-NODEVICE-NOT: linker, {{.*}}, spirv, (device-sycl)
 
+// Checking for an error if c-compilation is forced
+// RUN: %clangxx -### -c -fsycl -xc %s 2>&1 | FileCheck -check-prefixes=CHECK_XC_FSYCL %s
+// RUN: %clangxx -### -c -fsycl -xc-header %s 2>&1 | FileCheck -check-prefixes=CHECK_XC_FSYCL %s
+// CHECK_XC_FSYCL: The option -x c{{.*}} must not be used in conjunction with -fsycl{{.*}}
+
 // TODO: SYCL specific fail - analyze and enable
 // XFAIL: windows-msvc
diff --git a/libclc/generic/libspirv/async/async_work_group_strided_copy.inc b/libclc/generic/libspirv/async/async_work_group_strided_copy.inc
@@ -6,19 +6,17 @@
 //
 //===----------------------------------------------------------------------===//
 
-#define STRIDED_COPY(DST_AS, SRC_AS, DST_STRIDE, SRC_STRIDE)               \
-  size_t size = __spirv_LocalInvocationId_x() *                            \
-                __spirv_LocalInvocationId_y() *                            \
-                __spirv_LocalInvocationId_z();                             \
-  size_t id = (__spirv_WorkgroupSize_y() * __spirv_WorkgroupSize_z() *     \
-               __spirv_LocalInvocationId_x()) +                            \
-              (__spirv_WorkgroupSize_z() *                                 \
-               __spirv_LocalInvocationId_y()) +                            \
-              __spirv_LocalInvocationId_z();                               \
-  size_t i;                                                                \
-                                                                           \
-  for (i = id; i < num_gentypes; i += size) {                              \
-    dst[i * DST_STRIDE] = src[i * SRC_STRIDE];                             \
+#define STRIDED_COPY(DST_AS, SRC_AS, DST_STRIDE, SRC_STRIDE)                   \
+  size_t size = __spirv_WorkgroupSize_x() * __spirv_WorkgroupSize_y() *        \
+                __spirv_WorkgroupSize_z();                                     \
+  size_t id = (__spirv_WorkgroupSize_y() * __spirv_WorkgroupSize_x() *         \
+               __spirv_LocalInvocationId_z()) +                                \
+              (__spirv_WorkgroupSize_x() * __spirv_LocalInvocationId_y()) +    \
+              __spirv_LocalInvocationId_x();                                   \
+  size_t i;                                                                    \
+                                                                               \
+  for (i = id; i < num_gentypes; i += size) {                                  \
+    dst[i * DST_STRIDE] = src[i * SRC_STRIDE];                                 \
   }
 
 _CLC_OVERLOAD _CLC_DEF event_t
diff --git a/llvm/test/tools/sycl-post-link/sym_but_no_split.ll b/llvm/test/tools/sycl-post-link/sym_but_no_split.ll
@@ -0,0 +1,25 @@
+; This test checks that the post-link tool produces a correct resulting file
+; table and a symbol file for an input module with two kernels when no code
+; splitting is requested.
+;
+; RUN: sycl-post-link -symbols -spec-const=rt -S %s -o %t.files.table
+; RUN: FileCheck %s -input-file=%t.files.table --check-prefixes CHECK-TABLE
+; RUN: FileCheck %s -input-file=%t.files_0.sym --match-full-lines --check-prefixes CHECK-SYM
+
+define dso_local spir_kernel void @KERNEL_AAA() {
+; CHECK-SYM-NOT: {{[a-zA-Z0-9._@]+}}
+; CHECK-SYM: KERNEL_AAA
+entry:
+  ret void
+}
+
+define dso_local spir_kernel void @KERNEL_BBB() {
+; CHECK-SYM-NEXT: KERNEL_BBB
+; CHECK-SYM-EMPTY:
+entry:
+  ret void
+}
+
+; CHECK-TABLE: [Code|Properties|Symbols]
+; CHECK-TABLE-NEXT: {{.*}}files_0.sym
+; CHECK-TABLE-EMPTY:
diff --git a/llvm/tools/sycl-post-link/sycl-post-link.cpp b/llvm/tools/sycl-post-link/sycl-post-link.cpp
@@ -121,23 +121,51 @@ static void writeToFile(std::string Filename, std::string Content) {
   OS.close();
 }
 
-// Output parameter ResKernelModuleMap is a map containing groups of kernels
-// with same values of the sycl-module-id attribute.
-// The function fills ResKernelModuleMap using input module M.
+// Describes scope covered by each entry in the module-kernel map populated by
+// the collectKernelModuleMap function.
+enum KernelMapEntryScope {
+  Scope_PerKernel, // one entry per kernel
+  Scope_PerModule, // one entry per module
+  Scope_Global     // single entry in the map for all kernels
+};
+
+// This function decides how kernels of the input module M will be distributed
+// ("split") into multiple modules based on the command options and IR
+// attributes. The decision is recorded in the output map parameter
+// ResKernelModuleMap which maps some key to a group of kernels. Each such group
+// along with IR it depends on (globals, functions from its call graph,...) will
+// constitute a separate module.
 static void collectKernelModuleMap(
     Module &M, std::map<StringRef, std::vector<Function *>> &ResKernelModuleMap,
-    bool OneKernelPerModule) {
-
-  constexpr char ATTR_SYCL_MODULE_ID[] = "sycl-module-id";
+    KernelMapEntryScope EntryScope) {
 
   for (auto &F : M.functions()) {
     if (F.getCallingConv() == CallingConv::SPIR_KERNEL) {
-      if (OneKernelPerModule) {
+      switch (EntryScope) {
+      case Scope_PerKernel:
         ResKernelModuleMap[F.getName()].push_back(&F);
-      } else if (F.hasFnAttribute(ATTR_SYCL_MODULE_ID)) {
+        break;
+      case Scope_PerModule: {
+        constexpr char ATTR_SYCL_MODULE_ID[] = "sycl-module-id";
+
+        // TODO It may make sense to group all kernels w/o the attribute into
+        // a separate module rather than issuing an error. Should probably be
+        // controlled by an option.
+        if (!F.hasFnAttribute(ATTR_SYCL_MODULE_ID))
+          error("no '" + Twine(ATTR_SYCL_MODULE_ID) +
+                "' attribute in kernel '" + F.getName() +
+                "', per-module split not possible");
         Attribute Id = F.getFnAttribute(ATTR_SYCL_MODULE_ID);
         StringRef Val = Id.getValueAsString();
         ResKernelModuleMap[Val].push_back(&F);
+        break;
+      }
+      case Scope_Global:
+        // the map key is not significant here
+        ResKernelModuleMap["<GLOBAL>"].push_back(&F);
+        break;
+      default:
+        llvm_unreachable("unknown scope");
       }
     }
   }
@@ -375,8 +403,12 @@ int main(int argc, char **argv) {
 
   std::map<StringRef, std::vector<Function *>> GlobalsSet;
 
-  if (DoSplit || DoSymGen)
-    collectKernelModuleMap(*MPtr, GlobalsSet, SplitMode == SPLIT_PER_KERNEL);
+  if (DoSplit || DoSymGen) {
+    KernelMapEntryScope Scope = Scope_Global;
+    if (DoSplit)
+      Scope = SplitMode == SPLIT_PER_KERNEL ? Scope_PerKernel : Scope_PerModule;
+    collectKernelModuleMap(*MPtr, GlobalsSet, Scope);
+  }
 
   std::vector<std::unique_ptr<Module>> ResultModules;
   std::vector<SpecIDMapTy> ResultSpecIDMaps;
diff --git a/sycl/doc/GetStartedGuide.md b/sycl/doc/GetStartedGuide.md
@@ -24,6 +24,7 @@ and a wide range of compute accelerators such as GPU and FPGA.
 * `git` - https://git-scm.com/downloads
 * `cmake` version 3.2 or later - http://www.cmake.org/download/
 * `python` - https://www.python.org/downloads/release/python-2716/
+* `ninja` - https://github.com/ninja-build/ninja/wiki/Pre-built-Ninja-packages
 * C++ compiler
   * Linux: `GCC` version 5.1.0 or later (including libstdc++) -
     https://gcc.gnu.org/install/
diff --git a/sycl/source/CMakeLists.txt b/sycl/source/CMakeLists.txt
@@ -105,8 +105,6 @@ set(SYCL_SOURCES
     "detail/scheduler/graph_builder.cpp"
     "detail/spec_constant_impl.cpp"
     "detail/sycl_mem_obj_t.cpp"
-    "detail/usm/clusm.cpp"
-    "detail/usm/usm_dispatch.cpp"
     "detail/usm/usm_impl.cpp"
     "detail/util.cpp"
     "accessor.cpp"
diff --git a/sycl/source/detail/context_impl.hpp b/sycl/source/detail/context_impl.hpp
@@ -17,7 +17,6 @@
 #include <detail/kernel_program_cache.hpp>
 #include <detail/platform_impl.hpp>
 #include <detail/program_manager/program_manager.hpp>
-#include <detail/usm/usm_dispatch.hpp>
 
 #include <map>
 #include <memory>
@@ -152,7 +151,6 @@ class context_impl {
   bool MPluginInterop;
   bool MHostContext;
   bool MUseCUDAPrimaryContext;
-  std::shared_ptr<usm::USMDispatcher> MUSMDispatch;
   std::map<DeviceLibExt, RT::PiProgram> MCachedLibPrograms;
   mutable KernelProgramCache MKernelProgramCache;
 };
diff --git a/sycl/source/detail/memory_manager.cpp b/sycl/source/detail/memory_manager.cpp
@@ -10,7 +10,6 @@
 #include <detail/context_impl.hpp>
 #include <detail/event_impl.hpp>
 #include <detail/queue_impl.hpp>
-#include <detail/usm/usm_dispatch.hpp>
 
 #include <algorithm>
 #include <cassert>
diff --git a/sycl/source/detail/queue_impl.cpp b/sycl/source/detail/queue_impl.cpp
@@ -12,7 +12,6 @@
 #include <CL/sycl/detail/pi.hpp>
 #include <CL/sycl/device.hpp>
 #include <detail/queue_impl.hpp>
-#include <detail/usm/usm_dispatch.hpp>
 
 #include <cstring>
 
diff --git a/sycl/source/detail/usm/clusm.cpp b/sycl/source/detail/usm/clusm.cpp
diff --git a/sycl/source/detail/usm/usm_dispatch.cpp b/sycl/source/detail/usm/usm_dispatch.cpp
diff --git a/sycl/source/detail/usm/usm_dispatch.hpp b/sycl/source/detail/usm/usm_dispatch.hpp
diff --git a/sycl/source/device_selector.cpp b/sycl/source/device_selector.cpp
