P4 to Git Change 1404293 by lmoriche@lmoriche_opencl_dev2 on 2017/05/01 16:43:07

SWDEV-102726 - [OCL-LC-ROCm] Open source OpenCL Runtime - Changes required to use the Khronos ICD sources instead of ours. Affected files ... ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/CMakeLists.txt#3 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_gl.cpp#54 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_icd.cpp#29 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/khronos/headers/opencl2.0/CL/cl_platform.h#7 edit ... //depot/stg/opencl/drivers/opencl/runtime/CMakeLists.txt#2 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpubinary.cpp#12 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpubinary.hpp#5 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.cpp#319 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.hpp#127 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuprogram.cpp#233 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuresource.cpp#239 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/object.hpp#18 edit [ROCm/clr commit: 0af2c7043c]
2017-05-01 15:54:18 -05:00
@@ -1,11 +1,12 @@
 set(CMAKE_CXX_FLAGS "-std=c++11")

-add_definitions(-DLINUX -D__x86_64__ -D__AMD64__ -DUNIX_OS -DqLittleEndian -DAMD_LIBELF -DOPENCL_MAJOR=2 -DOPENCL_MINOR=0 -DWITH_AQL -DWITH_ONLINE_COMPILER -DATI_OS_LINUX -DATI_ARCH_X86 -DLITTLEENDIAN_CPU -DATI_BITS_64 -DATI_COMP_GCC -DWITH_HSA_DEVICE -DWITH_TARGET_AMDGCN -DOPENCL_EXPORTS -DCL_USE_DEPRECATED_OPENCL_1_1_APIS -DCL_USE_DEPRECATED_OPENCL_1_2_APIS -DCL_USE_DEPRECATED_OPENCL_2_0_APIS -DWITH_VERSION_0_8=1 -DVEGA10_ONLY=false -DWITH_LIGHTNING_COMPILER -DBSD_LIBELF)
+add_definitions(-DLINUX -D__x86_64__ -D__AMD64__ -DUNIX_OS -DqLittleEndian -DOPENCL_MAJOR=2 -DOPENCL_MINOR=0 -DWITH_AQL -DWITH_ONLINE_COMPILER -DATI_OS_LINUX -DATI_ARCH_X86 -DLITTLEENDIAN_CPU -DATI_BITS_64 -DATI_COMP_GCC -DWITH_HSA_DEVICE -DWITH_TARGET_AMDGCN -DOPENCL_EXPORTS -DCL_USE_DEPRECATED_OPENCL_1_0_APIS -DCL_USE_DEPRECATED_OPENCL_1_1_APIS -DCL_USE_DEPRECATED_OPENCL_1_2_APIS -DCL_USE_DEPRECATED_OPENCL_2_0_APIS -DVEGA10_ONLY=false -DWITH_LIGHTNING_COMPILER)

 include_directories(/opt/rocm/include/hsa)

 include_directories(${CMAKE_SOURCE_DIR}/runtime)
 include_directories(${CMAKE_SOURCE_DIR}/api/opencl)
+include_directories(${CMAKE_SOURCE_DIR}/api/opencl/khronos)
 include_directories(${CMAKE_SOURCE_DIR}/api/opencl/khronos/headers)
 include_directories(${CMAKE_SOURCE_DIR}/api/opencl/khronos/headers/opencl2.0)
 include_directories(${CMAKE_SOURCE_DIR}/compiler/llvm/lib/Target/AMDGPU/MCTargetDesc)
@@ -13,11 +14,13 @@ include_directories(${CMAKE_SOURCE_DIR}/compiler/driver/src)
 include_directories(${CMAKE_SOURCE_DIR}/compiler/lib)
 include_directories(${CMAKE_SOURCE_DIR}/compiler/lib/include)
 include_directories(${CMAKE_SOURCE_DIR}/compiler/lib/backends/common)
-include_directories(${CMAKE_SOURCE_DIR}/compiler/lib/loaders/elf/utils/common)
-include_directories(${CMAKE_SOURCE_DIR}/compiler/lib/loaders/elf/utils/libelf)
 include_directories(${CMAKE_SOURCE_DIR}/compiler/lib/loaders)
 include_directories(${CMAKE_SOURCE_DIR}/compiler/tools)

+add_definitions(-DBSD_LIBELF)
+include_directories(${CMAKE_SOURCE_DIR}/compiler/lib/loaders/elf/utils/common)
+include_directories(${CMAKE_SOURCE_DIR}/compiler/lib/loaders/elf/utils/libelf)
+
 add_subdirectory(device/rocm)

 add_library(oclruntime OBJECT
@@ -27,7 +27,7 @@ ClBinary::FeatureCheckResult ClBinary::checkFeatures() {
  amd::OclElf::oclElfPlatform platform;
  if (!elfIn()->getTarget(elf_target, platform)) {
    LogError("Loading OCL CPU binary: incorrect format");
-    return ERROR;
+    return fcERROR;
  }
  uint64_t chip_options = 0x0;
  if (platform == amd::OclElf::COMPLIB_PLATFORM) {
@@ -35,7 +35,7 @@ ClBinary::FeatureCheckResult ClBinary::checkFeatures() {
    uint32_t flag;
    if (!elfIn()->getFlags(flag)) {
      LogError("Loading OCL CPU binary: incorrect format");
-      return ERROR;
+      return fcERROR;
    }
    aclTargetInfo tgtInfo = aclGetTargetInfoFromChipID(LP64_SWITCH("x86", "x86-64"), flag, NULL);
    chip_options = aclGetChipOptions(tgtInfo);
@@ -43,13 +43,13 @@ ClBinary::FeatureCheckResult ClBinary::checkFeatures() {
        ((elf_target == EM_386) && (strcmp(LP64_SWITCH("x86", "x86-64"), "x86") != 0)) ||
        ((elf_target == EM_X86_64) && (strcmp(LP64_SWITCH("x86", "x86-64"), "x86-64") != 0))) {
      LogError("Loading OCL CPU binary: different target");
-      return ERROR;
+      return fcERROR;
    }
  } else {
    // BIF 2.0
    if ((platform != amd::OclElf::CPU_PLATFORM) || ((target & elf_target) != elf_target)) {
      LogError("Loading OCL CPU binary: different target");
-      return ERROR;
+      return fcERROR;
    }
  }
  char* section;
@@ -64,11 +64,11 @@ ClBinary::FeatureCheckResult ClBinary::checkFeatures() {
    if (elfIn_->getSection(amd::OclElf::LLVMIR, &section, &sz)) {
      if ((section != NULL) && (sz > 0)) {
        // hasDLL being false to force recompiling
-        RECOMPILE;
+        fcRECOMPILE;
      }
    }
  }
-  return OK;
+  return fcOK;
 }

 bool ClBinary::loadX86(Program& program, std::string& dllName, bool& hasDLL) {
@@ -79,11 +79,11 @@ bool ClBinary::loadX86(Program& program, std::string& dllName, bool& hasDLL) {
  dllName = tempName + "." WINDOWS_SWITCH("dll", MACOS_SWITCH("dyld", "so"));

  switch (checkFeatures()) {
-    case ERROR:
+    case fcERROR:
      return false;
-    case RECOMPILE:
+    case fcRECOMPILE:
      return true;
-    case OK:
+    case fcOK:
      // Fallthrough
      break;
  }
@@ -143,11 +143,11 @@ bool ClBinary::loadX86JIT(Program& program, bool& hasJITBinary) {
  hasJITBinary = false;

  switch (checkFeatures()) {
-    case ERROR:
+    case fcERROR:
      return false;
-    case RECOMPILE:
+    case fcRECOMPILE:
      return true;
-    case OK:
+    case fcOK:
      // Fallthrough
      break;
  }
@@ -56,7 +56,7 @@ class ClBinary : public device::ClBinary {
  bool storeX86Asm(const char* buffer, size_t size);

 private:
-  enum FeatureCheckResult { ERROR, RECOMPILE, OK };
+  enum FeatureCheckResult { fcERROR, fcRECOMPILE, fcOK };

  FeatureCheckResult checkFeatures();

@@ -29,7 +29,7 @@ const MetaDataConst ArgState[ArgStateTotal] = {
    // Name                 Type                    Properties
    // Kernel description (special properties)
    {"memory:compilerwrite", KernelArg::PrivateFixed, {0, 0, 0, 0, 0, 0, 0}},
-    {"uniqueid:", KernelArg::None, {0, 0, 0, 0, 0, 0, 0}},
+    {"uniqueid:", KernelArg::NoType, {0, 0, 0, 0, 0, 0, 0}},
    {"memory:private:", KernelArg::PrivateSize, {0, 0, 0, 0, 0, 0, 0}},
    {"memory:local:", KernelArg::LocalSize, {0, 0, 0, 0, 0, 0, 0}},
    {"memory:hwprivate:", KernelArg::HwPrivateSize, {0, 0, 0, 0, 0, 0, 0}},
@@ -37,8 +37,8 @@ const MetaDataConst ArgState[ArgStateTotal] = {
    {"memory:hwlocal:", KernelArg::HwLocalSize, {0, 0, 0, 0, 0, 0, 0}},
    {"memory:64bitABI", KernelArg::ABI64Bit, {0, 0, 0, 0, 0, 0, 0}},
    {"limitgroupsize", KernelArg::Wavefront, {0, 0, 0, 0, 0, 0, 0}},
-    {"function:", KernelArg::None, {1, 1, 0, 0, 0, 0, 0}},
-    {"intrinsic:", KernelArg::None, {1, 0, 0, 0, 0, 0, 0}},
+    {"function:", KernelArg::NoType, {1, 1, 0, 0, 0, 0, 0}},
+    {"intrinsic:", KernelArg::NoType, {1, 0, 0, 0, 0, 0, 0}},
    {"error:", KernelArg::ErrorMessage, {0, 0, 0, 0, 0, 0, 0}},
    {"warning:", KernelArg::WarningMessage, {0, 0, 0, 0, 0, 0, 0}},
    {"printf_fmt:", KernelArg::PrintfFormatStr, {0, 0, 0, 0, 0, 0, 0}},
@@ -312,13 +312,13 @@ CalImageReference::~CalImageReference() {
 }

 KernelArg::KernelArg()
-    : type_(KernelArg::None),
+    : type_(KernelArg::NoType),
      size_(0),
      cbIdx_(0),
      cbPos_(0),
      index_(0),
      alignment_(1),
-      dataType_(KernelArg::None) {
+      dataType_(KernelArg::NoType) {
  name_ = "";
  buf_ = "";
  memory_.value_ = 0;
@@ -349,7 +349,7 @@ KernelArg& KernelArg::operator=(const KernelArg& data) {

 bool KernelArg::isCbNeeded() const {
  //! \note not a safe way
-  bool result = ((type_ > None) && (type_ < Sampler)) ? true : false;
+  bool result = ((type_ > NoType) && (type_ < Sampler)) ? true : false;
  if ((type_ == Sampler) && (location_ == 0)) {
    // Sampler is defined outside the kernel
    result = true;
@@ -359,7 +359,7 @@ bool KernelArg::isCbNeeded() const {

 size_t KernelArg::size(bool gpuLayer) const {
  switch (type_) {
-    case None:
+    case NoType:
      return 0;
    case PointerConst:
    case PointerHwConst:
@@ -507,7 +507,7 @@ clk_value_type_t KernelArg::type() const {
      return T_SAMPLER;
    case PointerPrivate:
    case PointerHwPrivate:
-    case None:
+    case NoType:
    default:
      return T_VOID;
  }
@@ -2188,7 +2188,7 @@ bool NullKernel::parseArguments(const std::string& metaData, uint* uavRefCount)
      break;
    }

-    arg.type_ = KernelArg::None;
+    arg.type_ = KernelArg::NoType;

    // Loop through all available metadata types
    for (uint i = 0; i < ArgStateTotal; ++i) {
@@ -2196,7 +2196,7 @@ bool NullKernel::parseArguments(const std::string& metaData, uint* uavRefCount)
      // Find the name tag
      if (expect(metaData, &pos, ArgState[i].typeName_)) {
        switch (ArgState[i].type_) {
-          case KernelArg::None:
+          case KernelArg::NoType:
            // Process next ...
            continue;
          case KernelArg::Reflection: {
@@ -128,7 +128,7 @@ struct KernelArg : public amd::HeapObject {
 public:
  //! \enum Kernel argument type
  enum ArgumentType {
-    None = 0,
+    NoType = 0,
    PointerGlobal,
    Value,
    Image,
@@ -951,7 +951,7 @@ bool NullProgram::initGlobalData(const std::string& source, size_t start) {
                    failed = true;
                  }
                } break;
-                case KernelArg::None:
+                case KernelArg::NoType:
                default:
                  break;
              }
@@ -1,6 +1,13 @@
 // Copyright (c) 2008 Advanced Micro Devices, Inc. All rights reserved.
 //

+#if defined(ATI_OS_WIN)
+#define WIN32_LEAN_AND_MEAN 1
+#include <Windows.h>
+#endif
+#include <GL/gl.h>
+#include "GL/glATIInternal.h"
+
 #include "platform/program.hpp"
 #include "platform/kernel.hpp"
 #include "os/os.hpp"
@@ -13,8 +20,6 @@
 #include "device/gpu/gputimestamp.hpp"
 #include "thread/atomic.hpp"
 #include "hsa_ext_image.h"
-#include <GL/gl.h>
-#include "GL/glATIInternal.h"

 #include <string>
 #include <fstream>
@@ -9,8 +9,10 @@
 #include "os/alloc.hpp"
 #include "thread/monitor.hpp"
 #include "utils/util.hpp"
+#include <icd/icd_dispatch.h>

-#define CL_TYPES_DO(F)                                                                             \
+
+#define KHR_CL_TYPES_DO(F)                                                                         \
  /* OpenCL type          Runtime type */                                                          \
  F(cl_context, Context)                                                                           \
  F(cl_event, Event)                                                                               \
@@ -19,21 +21,26 @@
  F(cl_program, Program)                                                                           \
  F(cl_device_id, Device)                                                                          \
  F(cl_mem, Memory)                                                                                \
-  F(cl_sampler, Sampler)                                                                           \
+  F(cl_sampler, Sampler)
+
+#define AMD_CL_TYPES_DO(F)                                                                         \
  F(cl_counter_amd, Counter)                                                                       \
  F(cl_perfcounter_amd, PerfCounter)                                                               \
  F(cl_threadtrace_amd, ThreadTrace)                                                               \
  F(cl_file_amd, LiquidFlashFile)

+
+#define CL_TYPES_DO(F)                                                                             \
+  KHR_CL_TYPES_DO(F)                                                                               \
+  AMD_CL_TYPES_DO(F)
+
 // Forward declare ::cl_* types and amd::Class types
 //

 #define DECLARE_CL_TYPES(CL, AMD)                                                                  \
  namespace amd {                                                                                  \
  class AMD;                                                                                       \
-  }                                                                                                \
-  typedef struct _##CL {                                                                           \
-  } * CL;
+  }

 CL_TYPES_DO(DECLARE_CL_TYPES);

@@ -41,6 +48,15 @@ CL_TYPES_DO(DECLARE_CL_TYPES);

 struct KHRicdVendorDispatchRec;

+#define DECLARE_CL_TYPES(CL, AMD)                                                                  \
+  typedef struct _##CL {                                                                           \
+    struct KHRicdVendorDispatchRec* dispatch;                                                      \
+  } * CL;
+
+AMD_CL_TYPES_DO(DECLARE_CL_TYPES);
+
+#undef DECLARE_CL_TYPES
+
 namespace amd {

 // Define the cl_*_type tokens for type checking.