[CANN] get_rows and dup optimization (#12671)

* [CANN]get_rows and dup optimization.

Co-authored-by: hipudding <[email protected]>
Signed-off-by: noemotiovon <[email protected]>

* [CANN]GET_ROWS and CPY/DUP optimization

Co-authored-by: hipudding <[email protected]>
Signed-off-by: noemotiovon <[email protected]>

* [CANN]code style adjustment

Signed-off-by: noemotiovon <[email protected]>

* [CANN]code style adjustment

Signed-off-by: noemotiovon <[email protected]>

* [CANN]code style adjustment

Signed-off-by: noemotiovon <[email protected]>

* [CANN]code style adjustment

Signed-off-by: noemotiovon <[email protected]>

---------

Signed-off-by: noemotiovon <[email protected]>
Co-authored-by: noemotiovon <[email protected]>
Co-authored-by: hipudding <[email protected]>

Author: 3 people

ggml/src/ggml-cann/CMakeLists.txt

@@ -51,13 +51,11 @@ if (CANN_INSTALL_DIR)
         ${CANN_INSTALL_DIR}/acllib/include
     )
 
-    add_subdirectory(kernels)
     list(APPEND CANN_LIBRARIES
         ascendcl
         nnopbase
         opapi
         acl_op_compiler
-        ascendc_kernels
     )
 
     file(GLOB GGML_SOURCES_CANN "*.cpp")

ggml/src/ggml-cann/aclnn_ops.cpp

@@ -1704,24 +1704,28 @@ static bool ggml_backend_cann_supports_op(ggml_backend_dev_t dev,
             switch (op->src[0]->type) {
                 case GGML_TYPE_F32:
                 case GGML_TYPE_F16:
-                case GGML_TYPE_Q4_0:
                 case GGML_TYPE_Q8_0:
                     return true;
                 default:
                     return false;
             }
         } break;
         case GGML_OP_CPY: {
-            switch (op->type) {
-                case GGML_TYPE_F32:
-                case GGML_TYPE_F16:
-                case GGML_TYPE_Q8_0:
-                case GGML_TYPE_Q4_0:
-                    return true;
-                default:
-                    return false;
+            ggml_tensor *src = op->src[0];
+            if ((op->type != GGML_TYPE_F32 && op->type != GGML_TYPE_F16) ||
+                  (src->type != GGML_TYPE_F32 &&
+                    src->type != GGML_TYPE_F16)) {
+                // only support F32 and F16.
+                return false;
             }
-        }
+
+            if (!ggml_are_same_shape(op, src) && !ggml_is_contiguous(op)) {
+                // unsupport dst is not contiguous.
+                return false;
+            }
+
+            return true;
+        } break;
         case GGML_OP_CONT: {
             // TODO: support GGML_TYPE_BF16
             switch (op->src[0]->type) {
@@ -1762,9 +1766,9 @@ static bool ggml_backend_cann_supports_op(ggml_backend_dev_t dev,
             }
             return true;
         }
+        case GGML_OP_DUP:
         case GGML_OP_IM2COL:
         case GGML_OP_CONCAT:
-        case GGML_OP_DUP:
         case GGML_OP_REPEAT:
         case GGML_OP_NONE:
         case GGML_OP_RESHAPE: