Add support for clang offload bundles and <<<>>> kernel launch.

[ROCm/clr commit: 323caee8d2]
2018-04-04 13:32:33 -07:00
parent b9b4c2026e
commit f22fbe4271
@@ -175,6 +175,7 @@ if(HIP_PLATFORM STREQUAL "hcc")

    set(SOURCE_FILES_RUNTIME
        src/code_object_bundle.cpp
+        src/hip_clang.cpp
        src/hip_hcc.cpp
        src/hip_context.cpp
        src/hip_device.cpp
@@ -534,18 +534,6 @@ typedef int hipLaunchParm;

 #include <hip/hip_runtime_api.h>

-#if defined(__cplusplus)
-extern "C" {
-#endif /*__cplusplus*/
-
-hipError_t hipConfigureCall(dim3 gridDim, dim3 blockDim, size_t sharedMem, hipStream_t stream);
-hipError_t hipSetupArgument(const void* arg, size_t size, size_t offset);
-hipError_t hipLaunchByPtr(const void* func);
-
-#if defined(__cplusplus)
-}
-#endif /*__cplusplus*/
-
 #include <__clang_cuda_builtin_vars.h>

 #define hipThreadIdx_x threadIdx.x
@@ -2427,6 +2427,59 @@ hipError_t hipIpcCloseMemHandle(void* devPtr);
 // hipError_t hipIpcOpenMemHandle(void** devPtr, hipIpcMemHandle_t handle, unsigned int flags);


+/**
+ *-------------------------------------------------------------------------------------------------
+ *-------------------------------------------------------------------------------------------------
+ *  @defgroup Clang Launch API to support the triple-chevron syntax
+ *  @{
+ */
+
+/**
+ * @brief Configure a kernel launch.
+ *
+ * @param [in] gridDim   grid dimension specified as multiple of blockDim.
+ * @param [in] blockDim  block dimensions specified in work-items
+ * @param [in] sharedMem Amount of dynamic shared memory to allocate for this kernel.  The
+ * kernel can access this with HIP_DYNAMIC_SHARED.
+ * @param [in] stream    Stream where the kernel should be dispatched.  May be 0, in which case the
+ * default stream is used with associated synchronization rules.
+ *
+ * @returns hipSuccess, hipInvalidDevice, hipErrorNotInitialized, hipErrorInvalidValue
+ *
+ */
+hipError_t hipConfigureCall(dim3 gridDim, dim3 blockDim, size_t sharedMem, hipStream_t stream);
+
+
+/**
+ * @brief Set a kernel argument.
+ *
+ * @returns hipSuccess, hipInvalidDevice, hipErrorNotInitialized, hipErrorInvalidValue
+ *
+ * @param [in] arg    Pointer the argument in host memory.
+ * @param [in] size   Size of the argument.
+ * @param [in] offset Offset of the argument on the argument stack.
+ *
+ */
+hipError_t hipSetupArgument(const void* arg, size_t size, size_t offset);
+
+
+/**
+ * @brief Launch a kernel.
+ *
+ * @param [in] func Kernel to launch.
+ *
+ * @returns hipSuccess, hipInvalidDevice, hipErrorNotInitialized, hipErrorInvalidValue
+ *
+ */
+hipError_t hipLaunchByPtr(const void* func);
+
+
+
+/**
+ * @}
+ */
+
+
 #ifdef __cplusplus
 } /* extern "c" */
 #endif
@@ -0,0 +1,219 @@
+/*
+Copyright (c) 2018 - present Advanced Micro Devices, Inc. All rights reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in
+all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+THE SOFTWARE.
+*/
+
+#include <unordered_map>
+#include <string>
+
+#include "hip/hip_runtime.h"
+#include "hip_hcc_internal.h"
+#include "trace_helper.h"
+
+constexpr unsigned __cudaFatMAGIC2 = 0x466243b1;
+
+#define CLANG_OFFLOAD_BUNDLER_MAGIC "__CLANG_OFFLOAD_BUNDLE__"
+#define AMDGCN_AMDHSA_TRIPLE "openmp-amdgcn--amdhsa"
+
+struct __ClangOffloadBundleDesc {
+  uint64_t offset;
+  uint64_t size;
+  uint64_t tripleSize;
+  const char triple[1];
+};
+
+struct __ClangOffloadBundleHeader {
+  const char magic[sizeof(CLANG_OFFLOAD_BUNDLER_MAGIC) - 1];
+  uint64_t numBundles;
+  __ClangOffloadBundleDesc desc[1];
+};
+
+struct __CudaFatBinaryWrapper {
+  unsigned int                magic;
+  unsigned int                version;
+  __ClangOffloadBundleHeader* binary;
+  void*                       unused;
+};
+
+
+extern "C" std::unordered_map<std::string, hipModule_t>*
+__hipRegisterFatBinary(const void* data)
+{
+  HIP_INIT();
+
+  const __CudaFatBinaryWrapper* fbwrapper = reinterpret_cast<const __CudaFatBinaryWrapper*>(data);
+  if (fbwrapper->magic != __cudaFatMAGIC2 || fbwrapper->version != 1) {
+    return nullptr;
+  }
+
+  const __ClangOffloadBundleHeader* header = fbwrapper->binary;
+  std::string magic(reinterpret_cast<const char*>(header), sizeof(CLANG_OFFLOAD_BUNDLER_MAGIC) - 1);
+  if (magic.compare(CLANG_OFFLOAD_BUNDLER_MAGIC)) {
+    return nullptr;
+  }
+
+  auto modules = new std::unordered_map<std::string, hipModule_t>{};
+  if (!modules) {
+    return nullptr;
+  }
+
+  const __ClangOffloadBundleDesc* desc = &header->desc[0];
+  for (uint64_t i = 0; i < header->numBundles; ++i,
+       desc = reinterpret_cast<const __ClangOffloadBundleDesc*>(
+           reinterpret_cast<uintptr_t>(&desc->triple[0]) + desc->tripleSize)) {
+
+    std::string triple{&desc->triple[0], sizeof(AMDGCN_AMDHSA_TRIPLE) - 1};
+    if (triple.compare(AMDGCN_AMDHSA_TRIPLE))
+      continue;
+
+    hipModule_t module;
+    if (hipSuccess == hipModuleLoadData(&module, reinterpret_cast<const void*>(
+                                        reinterpret_cast<uintptr_t>(header) + desc->offset))) {
+      modules->emplace(std::string{&desc->triple[sizeof(AMDGCN_AMDHSA_TRIPLE)],
+                                   desc->tripleSize - sizeof(AMDGCN_AMDHSA_TRIPLE)},
+                       module);
+    }
+  }
+
+  return modules;
+}
+
+std::map<const void*, std::vector<hipFunction_t>> g_functions;
+
+extern "C" void __hipRegisterFunction(
+  std::unordered_map<std::string, hipModule_t>* modules,
+  const void*  hostFunction,
+  char*        deviceFunction,
+  const char*  deviceName,
+  unsigned int threadLimit,
+  uint3*       tid,
+  uint3*       bid,
+  dim3*        blockDim,
+  dim3*        gridDim,
+  int*         wSize)
+{
+  std::vector<hipFunction_t> functions{g_deviceCnt};
+
+  for (auto&& it : *modules) {
+    hipFunction_t function;
+    if (hipSuccess != hipModuleGetFunction(&function, it.second, deviceName)) {
+      continue;
+    }
+
+    for (int deviceId = 0; deviceId < g_deviceCnt; ++deviceId) {
+      char name[64] = {};
+      hsa_agent_get_info(g_allAgents[deviceId + 1], HSA_AGENT_INFO_NAME, name);
+      if (!it.first.compare(name)) {
+        functions[deviceId] = function;
+      }
+    }
+  }
+
+  g_functions.insert(std::make_pair(hostFunction, std::move(functions)));
+}
+
+extern "C" void __hipRegisterVar(
+  hipModule_t module,
+  char*       hostVar,
+  char*       deviceVar,
+  const char* deviceName,
+  int         ext,
+  int         size,
+  int         constant,
+  int         global)
+{
+}
+
+extern "C" void __hipUnregisterFatBinary(std::unordered_map<std::string, hipModule_t>* modules)
+{
+  for (auto&& it : *modules) {
+    delete it.second;
+  }
+  delete modules;
+}
+
+hipError_t hipConfigureCall(
+  dim3 gridDim,
+  dim3 blockDim,
+  size_t sharedMem,
+  hipStream_t stream)
+{
+  auto ctx = ihipGetTlsDefaultCtx();
+  LockedAccessor_CtxCrit_t crit(ctx->criticalData());
+
+  crit->_execStack.push(ihipExec_t{gridDim, blockDim, sharedMem, stream});
+  return hipSuccess;
+}
+
+hipError_t hipSetupArgument(
+  const void *arg,
+  size_t size,
+  size_t offset)
+{
+  auto ctx = ihipGetTlsDefaultCtx();
+  LockedAccessor_CtxCrit_t crit(ctx->criticalData());
+  auto& arguments = crit->_execStack.top()._arguments;
+
+  if (arguments.size() < offset + size) {
+    arguments.resize(offset + size);
+  }
+
+  ::memcpy(&arguments[offset], arg, size);
+  return hipSuccess;
+}
+
+hipError_t hipLaunchByPtr(const void *hostFunction)
+{
+  ihipExec_t exec;
+  {
+    auto ctx = ihipGetTlsDefaultCtx();
+    LockedAccessor_CtxCrit_t crit(ctx->criticalData());
+    exec = std::move(crit->_execStack.top());
+    crit->_execStack.pop();
+  }
+
+  int deviceId;
+  if (exec._hStream) {
+    deviceId = exec._hStream->getDevice()->_deviceId;
+  }
+  else if (ihipGetTlsDefaultCtx() && ihipGetTlsDefaultCtx()->getDevice()) {
+    deviceId = ihipGetTlsDefaultCtx()->getDevice()->_deviceId;
+  }
+  else {
+    deviceId = 0;
+  }
+
+  decltype(g_functions)::iterator it;
+  if ((it = g_functions.find(hostFunction)) == g_functions.end())
+    return hipErrorUnknown;
+
+  size_t size = exec._arguments.size();
+  void *extra[] = {
+      HIP_LAUNCH_PARAM_BUFFER_POINTER, &exec._arguments[0],
+      HIP_LAUNCH_PARAM_BUFFER_SIZE, &size,
+      HIP_LAUNCH_PARAM_END
+    };
+
+  return hipModuleLaunchKernel(it->second[deviceId],
+    exec._gridDim.x, exec._gridDim.y, exec._gridDim.z,
+    exec._blockDim.x, exec._blockDim.y, exec._blockDim.z,
+    exec._sharedMem, exec._hStream, nullptr, extra);
+}
+
@@ -789,6 +789,16 @@ class ihipDevice_t {
 //=============================================================================


+//---
+//
+struct ihipExec_t {
+  dim3 _gridDim;
+  dim3 _blockDim;
+  size_t _sharedMem;
+  hipStream_t _hStream;
+  std::vector<char> _arguments;
+};
+
 //=============================================================================
 // class ihipCtxCriticalBase_t
 template <typename MUTEX_TYPE>
@@ -827,6 +837,8 @@ class ihipCtxCriticalBase_t : LockedBase<MUTEX_TYPE> {

    // TODO - move private
    std::list<ihipCtx_t*> _peers;  // list of enabled peer devices.
+    //--- Execution stack:
+    std::stack<ihipExec_t> _execStack; // Execution stack for this device.

    friend class LockedAccessor<ihipCtxCriticalBase_t>;