[AMDGPU][Libomptarget] Delete g_atl_machine global

With uses of g_atl_machine gone, a significant portion of dead code has been removed. This patch depends on D104691 and D104695. Reviewed By: JonChesterfield Differential Revision: https://reviews.llvm.org/D104696
llvm · Aug 24, 2021 · 9b8b7c1 · 9b8b7c1
1 parent 1f8602e
commit 9b8b7c1
Show file tree

Hide file tree

Showing 10 changed files with 26 additions and 430 deletions.
diff --git a/openmp/libomptarget/plugins/amdgpu/CMakeLists.txt b/openmp/libomptarget/plugins/amdgpu/CMakeLists.txt
@@ -66,7 +66,6 @@ add_library(omptarget.rtl.amdgpu SHARED
       impl/atmi_interop_hsa.cpp
       impl/data.cpp
       impl/get_elf_mach_gfx_name.cpp
-      impl/machine.cpp
       impl/system.cpp
       impl/utils.cpp
       impl/msgpack.cpp

diff --git a/openmp/libomptarget/plugins/amdgpu/impl/atmi.cpp b/openmp/libomptarget/plugins/amdgpu/impl/atmi.cpp
@@ -50,7 +50,8 @@ struct atmiFreePtrDeletor {
 
 hsa_status_t atmi_memcpy_h2d(hsa_signal_t signal, void *deviceDest,
                              const void *hostSrc, size_t size,
-                             hsa_agent_t agent) {
+                             hsa_agent_t agent,
+                             hsa_amd_memory_pool_t MemoryPool) {
   hsa_status_t rc = hsa_memory_copy(deviceDest, hostSrc, size);
 
   // hsa_memory_copy sometimes fails in situations where
@@ -61,7 +62,7 @@ hsa_status_t atmi_memcpy_h2d(hsa_signal_t signal, void *deviceDest,
   }
 
   void *tempHostPtr;
-  hsa_status_t ret = core::Runtime::HostMalloc(&tempHostPtr, size);
+  hsa_status_t ret = core::Runtime::HostMalloc(&tempHostPtr, size, MemoryPool);
   if (ret != HSA_STATUS_SUCCESS) {
     DEBUG_PRINT("HostMalloc: Unable to alloc %zu bytes for temp scratch\n",
                 size);
@@ -79,7 +80,8 @@ hsa_status_t atmi_memcpy_h2d(hsa_signal_t signal, void *deviceDest,
 
 hsa_status_t atmi_memcpy_d2h(hsa_signal_t signal, void *dest,
                              const void *deviceSrc, size_t size,
-                             hsa_agent_t agent) {
+                             hsa_agent_t agent,
+                             hsa_amd_memory_pool_t MemoryPool) {
   hsa_status_t rc = hsa_memory_copy(dest, deviceSrc, size);
 
   // hsa_memory_copy sometimes fails in situations where
@@ -90,7 +92,7 @@ hsa_status_t atmi_memcpy_d2h(hsa_signal_t signal, void *dest,
   }
 
   void *tempHostPtr;
-  hsa_status_t ret = core::Runtime::HostMalloc(&tempHostPtr, size);
+  hsa_status_t ret = core::Runtime::HostMalloc(&tempHostPtr, size, MemoryPool);
   if (ret != HSA_STATUS_SUCCESS) {
     DEBUG_PRINT("HostMalloc: Unable to alloc %zu bytes for temp scratch\n",
                 size);

diff --git a/openmp/libomptarget/plugins/amdgpu/impl/atmi_interop_hsa.cpp b/openmp/libomptarget/plugins/amdgpu/impl/atmi_interop_hsa.cpp
@@ -7,11 +7,6 @@
 //===----------------------------------------------------------------------===//
 #include "atmi_interop_hsa.h"
 #include "internal.h"
-#include "machine.h"
-
-// TODO: need to get rid of this as well
-
-extern ATLMachine g_atl_machine;
 
 hsa_status_t atmi_interop_hsa_get_symbol_info(
     const std::map<std::string, atl_symbol_info_t> &SymbolInfoTable,

diff --git a/openmp/libomptarget/plugins/amdgpu/impl/atmi_runtime.h b/openmp/libomptarget/plugins/amdgpu/impl/atmi_runtime.h
@@ -55,11 +55,13 @@ hsa_status_t atmi_module_register_from_memory_to_place(
 
 hsa_status_t atmi_memcpy_h2d(hsa_signal_t signal, void *deviceDest,
                              const void *hostSrc, size_t size,
-                             hsa_agent_t agent);
+                             hsa_agent_t agent,
+                             hsa_amd_memory_pool_t MemoryPool);
 
 hsa_status_t atmi_memcpy_d2h(hsa_signal_t sig, void *hostDest,
                              const void *deviceSrc, size_t size,
-                             hsa_agent_t agent);
+                             hsa_agent_t agent,
+                             hsa_amd_memory_pool_t MemoryPool);
 
 /** @} */
 

diff --git a/openmp/libomptarget/plugins/amdgpu/impl/data.cpp b/openmp/libomptarget/plugins/amdgpu/impl/data.cpp
@@ -8,55 +8,24 @@
 #include "atmi_runtime.h"
 #include "hsa_api.h"
 #include "internal.h"
-#include "machine.h"
 #include "rt.h"
 #include <cassert>
 #include <stdio.h>
 #include <string.h>
 #include <vector>
 
 using core::TaskImpl;
-extern ATLMachine g_atl_machine;
 
 namespace core {
 
-namespace {
-ATLProcessor &get_processor_by_mem_place(int DeviceId,
-                                         atmi_devtype_t DeviceType) {
-  switch (DeviceType) {
-  case ATMI_DEVTYPE_CPU:
-    return g_atl_machine.processors<ATLCPUProcessor>()[DeviceId];
-  case ATMI_DEVTYPE_GPU:
-    return g_atl_machine.processors<ATLGPUProcessor>()[DeviceId];
-  }
-}
-
-hsa_amd_memory_pool_t get_memory_pool_by_mem_place(int DeviceId,
-                                                   atmi_devtype_t DeviceType) {
-  ATLProcessor &proc = get_processor_by_mem_place(DeviceId, DeviceType);
-  return get_memory_pool(proc, 0 /*Memory Type (always zero) */);
-}
-} // namespace
+hsa_status_t Runtime::HostMalloc(void **ptr, size_t size,
+                                 hsa_amd_memory_pool_t MemoryPool) {
+  hsa_status_t err = hsa_amd_memory_pool_allocate(MemoryPool, size, 0, ptr);
+  DEBUG_PRINT("Malloced [CPU %d] %p\n", DeviceId, *ptr);
 
-hsa_status_t Runtime::DeviceMalloc(void **ptr, size_t size, int DeviceId) {
-  return Runtime::Malloc(ptr, size, DeviceId, ATMI_DEVTYPE_GPU);
-}
-
-hsa_status_t Runtime::HostMalloc(void **ptr, size_t size) {
-  hsa_status_t Err = Runtime::Malloc(ptr, size, 0, ATMI_DEVTYPE_CPU);
-  if (Err == HSA_STATUS_SUCCESS) {
-    Err = core::allow_access_to_all_gpu_agents(*ptr);
+  if (err == HSA_STATUS_SUCCESS) {
+    err = core::allow_access_to_all_gpu_agents(*ptr);
   }
-  return Err;
-}
-
-hsa_status_t Runtime::Malloc(void **ptr, size_t size, int DeviceId,
-                             atmi_devtype_t DeviceType) {
-  hsa_amd_memory_pool_t pool =
-      get_memory_pool_by_mem_place(DeviceId, DeviceType);
-  hsa_status_t err = hsa_amd_memory_pool_allocate(pool, size, 0, ptr);
-  DEBUG_PRINT("Malloced [%s %d] %p\n",
-              DeviceType == ATMI_DEVTYPE_CPU ? "CPU" : "GPU", DeviceId, *ptr);
   return (err == HSA_STATUS_SUCCESS) ? HSA_STATUS_SUCCESS : HSA_STATUS_ERROR;
 }
 

diff --git a/openmp/libomptarget/plugins/amdgpu/impl/machine.cpp b/openmp/libomptarget/plugins/amdgpu/impl/machine.cpp
diff --git a/openmp/libomptarget/plugins/amdgpu/impl/machine.h b/openmp/libomptarget/plugins/amdgpu/impl/machine.h
diff --git a/openmp/libomptarget/plugins/amdgpu/impl/rt.h b/openmp/libomptarget/plugins/amdgpu/impl/rt.h
@@ -60,16 +60,12 @@ class Runtime final {
   // data
   static hsa_status_t Memcpy(hsa_signal_t, void *, const void *, size_t);
   static hsa_status_t Memfree(void *);
-  static hsa_status_t HostMalloc(void **ptr, size_t size);
-  static hsa_status_t DeviceMalloc(void **ptr, size_t size, int DeviceId);
+  static hsa_status_t HostMalloc(void **ptr, size_t size,
+                                 hsa_amd_memory_pool_t MemoryPool);
 
   int getMaxQueueSize() const { return env_.getMaxQueueSize(); }
   int getDebugMode() const { return env_.getDebugMode(); }
 
-private:
-  static hsa_status_t Malloc(void **ptr, size_t size, int DeviceId,
-                             atmi_devtype_t DeviceType);
-
 protected:
   Runtime() = default;
   ~Runtime() = default;