fix(ascend): adapt Memcpy/Memset arity, assert workspace alloc, remove missing include

- Wrap `aclrtMemcpy` (5-arg) and `aclrtMemset` (4-arg) in lambdas to
  match the generic 4-arg / 3-arg calling convention used by examples.
- Assert `aclrtMalloc` return value in `WorkspacePool::ensure()`.
- Remove `ascend/gemm/kernel.h` include from `runtime_api.h` (file
  does not exist until the kernels commit).

Author: zhangyue

examples/runtime_api.h

@@ -19,9 +19,6 @@
 #elif WITH_MOORE
 #include "moore/gemm/mublas.h"
 #include "moore/runtime_.h"
-#elif WITH_ASCEND
-#include "ascend/gemm/kernel.h"
-#include "ascend/runtime_.h"
 #elif WITH_CPU
 #include "cpu/gemm/gemm.h"
 #include "cpu/runtime_.h"
@@ -41,8 +38,6 @@ using DefaultRuntimeUtils = Runtime<Device::Type::kMetax>;
 using DefaultRuntimeUtils = Runtime<Device::Type::kCambricon>;
 #elif WITH_MOORE
 using DefaultRuntimeUtils = Runtime<Device::Type::kMoore>;
-#elif WITH_ASCEND
-using DefaultRuntimeUtils = Runtime<Device::Type::kAscend>;
 #elif WITH_CPU
 using DefaultRuntimeUtils = Runtime<Device::Type::kCpu>;
 #endif

src/ascend/runtime_.h

@@ -23,13 +23,18 @@ struct Runtime<Device::Type::kAscend>
 
   static constexpr auto Free = aclrtFree;
 
-  static constexpr auto Memcpy = aclrtMemcpy;
+  static constexpr auto Memcpy = [](void* dst, const void* src, size_t count,
+                                    aclrtMemcpyKind kind) {
+    return aclrtMemcpy(dst, count, src, count, kind);
+  };
 
   static constexpr auto MemcpyHostToDevice = ACL_MEMCPY_HOST_TO_DEVICE;
 
   static constexpr auto MemcpyDeviceToHost = ACL_MEMCPY_DEVICE_TO_HOST;
 
-  static constexpr auto Memset = aclrtMemset;
+  static constexpr auto Memset = [](void* ptr, int value, size_t count) {
+    return aclrtMemset(ptr, count, value, count);
+  };
 };
 
 static_assert(Runtime<Device::Type::kAscend>::Validate());

src/ascend/workspace_pool_.h

@@ -1,6 +1,7 @@
 #ifndef INFINI_OPS_ASCEND_WORKSPACE_POOL__H_
 #define INFINI_OPS_ASCEND_WORKSPACE_POOL__H_
 
+#include <cassert>
 #include <cstdint>
 #include <mutex>
 #include <unordered_map>
@@ -25,7 +26,9 @@ class WorkspacePool {
       aclrtFree(arena.buf);
     }
     if (needed > 0) {
-      aclrtMalloc(&arena.buf, needed, ACL_MEM_MALLOC_NORMAL_ONLY);
+      assert(aclrtMalloc(&arena.buf, needed, ACL_MEM_MALLOC_NORMAL_ONLY) ==
+                 ACL_SUCCESS &&
+             "`WorkspacePool`: `aclrtMalloc` failed");
     }
     arena.capacity = needed;
     return arena;