Merge branch 'sycl' into oneapins

Author: jbrodman

buildbot/dependency.conf

@@ -4,15 +4,15 @@ ocl_cpu_rt_ver=2020.10.7.0.15
 # https://github.com/intel/llvm/releases/download/2020-WW31/win-oclcpuexp-2020.10.7.0.15_rel.zip
 ocl_cpu_rt_ver_win=2020.10.7.0.15
 # Same GPU driver supports Level Zero and OpenCL:
-# https://github.com/intel/compute-runtime/releases/tag/20.25.17111
+# https://github.com/intel/compute-runtime/releases/tag/20.29.17408
 ocl_gpu_rt_ver=20.29.17408
 # Same GPU driver supports Level Zero and OpenCL:
 # https://downloadmirror.intel.com/29674/a08/igfx_win10_100.8336.zip
 ocl_gpu_rt_ver_win=27.20.100.8336
 intel_sycl_ver=build
-# https://github.com/oneapi-src/oneTBB/releases/download/v2020.2/tbb-2020.2-lin.tgz
+# TODO provide URL for Linux TBB driver
 tbb_ver=2021.1.8.515
-# https://github.com/oneapi-src/oneTBB/releases/download/v2020.2/tbb-2020.2-win.zip
+# TODO provide URL for Windows TBB driver
 tbb_ver_win=2021.1.8.514
 # https://github.com/intel/llvm/releases/download/2020-WW31/fpgaemu-2020.10.7.0.15_rel.tar.gz
 ocl_fpga_emu_ver=2020.10.7.0.15

llvm-spirv/lib/SPIRV/SPIRVWriter.cpp

@@ -563,10 +563,11 @@ SPIRVFunction *LLVMToSPIRV::transFunctionDecl(Function *F) {
       // Order of integer numbers in MD node follows the order of function
       // parameters on which we shall attach the appropriate decoration. Add
       // decoration only if MD value is not negative.
-      BM->addCapability(CapabilityFPGABufferLocationINTEL);
       int LocID = getMDOperandAsInt(BufferLocation, ArgNo);
-      if (LocID >= 0)
+      if (LocID >= 0) {
+        BM->addCapability(CapabilityFPGABufferLocationINTEL);
         BA->addDecorate(DecorationBufferLocationINTEL, LocID);
+      }
     }
   }
   if (Attrs.hasAttribute(AttributeList::ReturnIndex, Attribute::ZExt))

sycl/CMakeLists.txt

@@ -107,6 +107,9 @@ if( NOT OpenCL_INCLUDE_DIRS )
     INSTALL_COMMAND   ""
     STEP_TARGETS      build
     COMMENT           "Downloading OpenCL headers."
+    LOG_DOWNLOAD 1
+    LOG_UPDATE 1
+    LOG_BUILD 1
   )
   add_definitions(-DCL_TARGET_OPENCL_VERSION=220)
 else()
@@ -154,6 +157,11 @@ if( NOT OpenCL_LIBRARIES )
     STEP_TARGETS      configure,build,install
     DEPENDS           ocl-headers
     BUILD_BYPRODUCTS ${OpenCL_LIBRARIES}
+    LOG_DOWNLOAD 1
+    LOG_UPDATE 1
+    LOG_CONFIGURE 1
+    LOG_BUILD 1
+    LOG_INSTALL 1
   )
   ExternalProject_Add_Step(ocl-icd llvminstall
     COMMAND ${CMAKE_COMMAND} -E copy_directory <INSTALL_DIR>/ ${LLVM_BINARY_DIR}

sycl/plugins/level_zero/CMakeLists.txt

@@ -33,6 +33,11 @@ if (NOT DEFINED LEVEL_ZERO_LIBRARY OR NOT DEFINED LEVEL_ZERO_INCLUDE_DIR)
                -DOpenCL_INCLUDE_DIR=${OpenCL_INCLUDE_DIRS}
                -DCMAKE_INSTALL_PREFIX=<INSTALL_DIR>
                -DCMAKE_INSTALL_LIBDIR:PATH=lib${LLVM_LIBDIR_SUFFIX}
+               LOG_DOWNLOAD 1
+               LOG_UPDATE 1
+               LOG_CONFIGURE 1
+               LOG_BUILD 1
+               LOG_INSTALL 1
                ${AUX_CMAKE_FLAGS}
     STEP_TARGETS      configure,build,install
     DEPENDS           ocl-headers

sycl/plugins/level_zero/pi_level_zero.cpp

@@ -114,6 +114,20 @@ pi_result getInfoArray(size_t array_length, size_t param_value_size,
                      array_length * sizeof(T), memcpy);
 }
 
+template <typename T, typename RetType>
+pi_result getInfoArray(size_t array_length, size_t param_value_size,
+                       void *param_value, size_t *param_value_size_ret,
+                       T *value) {
+  if (param_value) {
+    memset(param_value, 0, param_value_size);
+    for (uint32_t I = 0; I < array_length; I++)
+      ((RetType *)param_value)[I] = (RetType)value[I];
+  }
+  if (param_value_size_ret)
+    *param_value_size_ret = array_length * sizeof(RetType);
+  return PI_SUCCESS;
+}
+
 template <>
 pi_result getInfo<const char *>(size_t param_value_size, void *param_value,
                                 size_t *param_value_size_ret,
@@ -1061,9 +1075,10 @@ pi_result piDeviceGetInfo(pi_device Device, pi_device_info ParamName,
   case PI_DEVICE_INFO_SUB_GROUP_SIZES_INTEL: {
     // ze_device_compute_properties.subGroupSizes is in uint32_t whereas the
     // expected return is size_t datatype. size_t can be 8 bytes of data.
-    return getInfoArray(Device->ZeDeviceComputeProperties.numSubGroupSizes,
-                        ParamValueSize, ParamValue, ParamValueSizeRet,
-                        Device->ZeDeviceComputeProperties.subGroupSizes);
+    return getInfoArray<uint32_t, size_t>(
+        Device->ZeDeviceComputeProperties.numSubGroupSizes, ParamValueSize,
+        ParamValue, ParamValueSizeRet,
+        Device->ZeDeviceComputeProperties.subGroupSizes);
   }
   case PI_DEVICE_INFO_IL_VERSION: {
     // Set to a space separated list of IL version strings of the form
@@ -3108,7 +3123,8 @@ pi_result piEnqueueMemUnmap(pi_queue Queue, pi_mem MemObj, void *MappedPtr,
   // piEnqueueMemBufferMap, but can only do so after the above copy
   // is completed. Instead of waiting for It here (blocking), we shall
   // do so in piEventRelease called for the pi_event tracking the unmap.
-  (*Event)->CommandData = MemObj->MapHostPtr ? nullptr : MappedPtr;
+  if (Event)
+    (*Event)->CommandData = MemObj->MapHostPtr ? nullptr : MappedPtr;
 
   // Execute command list asynchronously, as the event will be used
   // to track down its completion.

sycl/test/regression/get_subgroup_sizes.cpp

@@ -0,0 +1,27 @@
+// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
+// RUN: %GPU_RUN_PLACEHOLDER %t.out
+
+//==-- get_subgroup_sizes.cpp - Test for bug fix in subgroup sizes query --==//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include <CL/sycl.hpp>
+
+using namespace cl::sycl;
+
+int main() {
+  queue Q;
+  auto Dev = Q.get_device();
+  if (Dev.has_extension("cl_intel_required_subgroup_size")) {
+    cl::sycl::vector_class<size_t> SubGroupSizes =
+        Dev.get_info<cl::sycl::info::device::sub_group_sizes>();
+    cl::sycl::vector_class<size_t>::const_iterator MaxIter =
+        std::max_element(SubGroupSizes.begin(), SubGroupSizes.end());
+    int MaxSubGroup_size = *MaxIter;
+  }
+  return 0;
+}