mindspore 2.3.0__cp39-none-any.whl → 2.3.0rc2__cp39-none-any.whl

mindspore/lib/plugin/ascend/custom_ascendc_ops/op_impl/ai_core/tbe/config/ascend310p/aic-ascend310p-ops-info.json

@@ -1,5 +1,5 @@
 {
-    "AllFinite":{
+    "AddCustom":{
         "dynamicCompileStatic":{
             "flag":"true"
         },
@@ -13,209 +13,37 @@
             "flag":"true"
         },
         "input0":{
-            "dtype":"float16,float32,bfloat16",
+            "dtype":"float16,float32,int32",
             "format":"ND,ND,ND",
-            "name":"gradient",
+            "name":"x",
             "paramType":"required",
             "shape":"all",
             "unknownshape_format":"ND,ND,ND"
         },
-        "needCheckSupport":{
-            "flag":"false"
-        },
-        "opFile":{
-            "value":"all_finite"
-        },
-        "opInterface":{
-            "value":"all_finite"
-        },
-        "output0":{
-            "dtype":"bool,bool,bool",
+        "input1":{
+            "dtype":"float16,float32,int32",
             "format":"ND,ND,ND",
-            "name":"is_finite",
+            "name":"y",
             "paramType":"required",
             "shape":"all",
             "unknownshape_format":"ND,ND,ND"
         },
-        "precision_reduce":{
-            "flag":"true"
-        }
-    },
-    "DecoderKvCache":{
-        "dynamicCompileStatic":{
-            "flag":"true"
-        },
-        "dynamicFormat":{
-            "flag":"true"
-        },
-        "dynamicRankSupport":{
-            "flag":"true"
-        },
-        "dynamicShapeSupport":{
-            "flag":"true"
-        },
-        "input0":{
-            "dtype":"float16,float32,int8,int16,int32,uint8,uint16,bfloat16",
-            "format":"ND,ND,ND,ND,ND,ND,ND,ND",
-            "name":"cache",
-            "paramType":"required",
-            "shape":"all",
-            "unknownshape_format":"ND,ND,ND,ND,ND,ND,ND,ND"
-        },
-        "input1":{
-            "dtype":"float16,float32,int8,int16,int32,uint8,uint16,bfloat16",
-            "format":"ND,ND,ND,ND,ND,ND,ND,ND",
-            "name":"update",
-            "paramType":"required",
-            "shape":"all",
-            "unknownshape_format":"ND,ND,ND,ND,ND,ND,ND,ND"
-        },
-        "input2":{
-            "dtype":"int64,int64,int64,int64,int64,int64,int64,int64",
-            "format":"ND,ND,ND,ND,ND,ND,ND,ND",
-            "name":"valid_seq_len",
-            "paramType":"required",
-            "shape":"all",
-            "unknownshape_format":"ND,ND,ND,ND,ND,ND,ND,ND"
-        },
-        "input3":{
-            "dtype":"int64,int64,int64,int64,int64,int64,int64,int64",
-            "format":"ND,ND,ND,ND,ND,ND,ND,ND",
-            "name":"batch_index",
-            "paramType":"required",
-            "shape":"all",
-            "unknownshape_format":"ND,ND,ND,ND,ND,ND,ND,ND"
-        },
-        "input4":{
-            "dtype":"int64,int64,int64,int64,int64,int64,int64,int64",
-            "format":"ND,ND,ND,ND,ND,ND,ND,ND",
-            "name":"seq_len_axis",
-            "paramType":"required",
-            "shape":"all",
-            "unknownshape_format":"ND,ND,ND,ND,ND,ND,ND,ND"
-        },
-        "input5":{
-            "dtype":"int64,int64,int64,int64,int64,int64,int64,int64",
-            "format":"ND,ND,ND,ND,ND,ND,ND,ND",
-            "name":"new_max_seq_len",
-            "paramType":"required",
-            "shape":"all",
-            "unknownshape_format":"ND,ND,ND,ND,ND,ND,ND,ND"
-        },
-        "input6":{
-            "dtype":"int64,int64,int64,int64,int64,int64,int64,int64",
-            "format":"ND,ND,ND,ND,ND,ND,ND,ND",
-            "name":"cur_max_seq_len",
-            "paramType":"required",
-            "shape":"all",
-            "unknownshape_format":"ND,ND,ND,ND,ND,ND,ND,ND"
-        },
-        "needCheckSupport":{
-            "flag":"false"
-        },
-        "opFile":{
-            "value":"decoder_kv_cache"
-        },
-        "opInterface":{
-            "value":"decoder_kv_cache"
-        },
-        "output0":{
-            "dtype":"float16,float32,int8,int16,int32,uint8,uint16,bfloat16",
-            "format":"ND,ND,ND,ND,ND,ND,ND,ND",
-            "name":"out",
-            "paramType":"required",
-            "shape":"all",
-            "unknownshape_format":"ND,ND,ND,ND,ND,ND,ND,ND"
-        },
-        "precision_reduce":{
-            "flag":"true"
-        }
-    },
-    "PromptKvCache":{
-        "dynamicCompileStatic":{
-            "flag":"true"
-        },
-        "dynamicFormat":{
-            "flag":"true"
-        },
-        "dynamicRankSupport":{
-            "flag":"true"
-        },
-        "dynamicShapeSupport":{
-            "flag":"true"
-        },
-        "input0":{
-            "dtype":"float16,float32,int8,int16,int32,uint8,uint16,bfloat16",
-            "format":"ND,ND,ND,ND,ND,ND,ND,ND",
-            "name":"cache",
-            "paramType":"required",
-            "shape":"all",
-            "unknownshape_format":"ND,ND,ND,ND,ND,ND,ND,ND"
-        },
-        "input1":{
-            "dtype":"float16,float32,int8,int16,int32,uint8,uint16,bfloat16",
-            "format":"ND,ND,ND,ND,ND,ND,ND,ND",
-            "name":"update",
-            "paramType":"required",
-            "shape":"all",
-            "unknownshape_format":"ND,ND,ND,ND,ND,ND,ND,ND"
-        },
-        "input2":{
-            "dtype":"int64,int64,int64,int64,int64,int64,int64,int64",
-            "format":"ND,ND,ND,ND,ND,ND,ND,ND",
-            "name":"valid_seq_len",
-            "paramType":"required",
-            "shape":"all",
-            "unknownshape_format":"ND,ND,ND,ND,ND,ND,ND,ND"
-        },
-        "input3":{
-            "dtype":"int64,int64,int64,int64,int64,int64,int64,int64",
-            "format":"ND,ND,ND,ND,ND,ND,ND,ND",
-            "name":"batch_index",
-            "paramType":"required",
-            "shape":"all",
-            "unknownshape_format":"ND,ND,ND,ND,ND,ND,ND,ND"
-        },
-        "input4":{
-            "dtype":"int64,int64,int64,int64,int64,int64,int64,int64",
-            "format":"ND,ND,ND,ND,ND,ND,ND,ND",
-            "name":"seq_len_axis",
-            "paramType":"required",
-            "shape":"all",
-            "unknownshape_format":"ND,ND,ND,ND,ND,ND,ND,ND"
-        },
-        "input5":{
-            "dtype":"int64,int64,int64,int64,int64,int64,int64,int64",
-            "format":"ND,ND,ND,ND,ND,ND,ND,ND",
-            "name":"new_max_seq_len",
-            "paramType":"required",
-            "shape":"all",
-            "unknownshape_format":"ND,ND,ND,ND,ND,ND,ND,ND"
-        },
-        "input6":{
-            "dtype":"int64,int64,int64,int64,int64,int64,int64,int64",
-            "format":"ND,ND,ND,ND,ND,ND,ND,ND",
-            "name":"cur_max_seq_len",
-            "paramType":"required",
-            "shape":"all",
-            "unknownshape_format":"ND,ND,ND,ND,ND,ND,ND,ND"
-        },
         "needCheckSupport":{
             "flag":"false"
         },
         "opFile":{
-            "value":"prompt_kv_cache"
+            "value":"add_custom"
         },
         "opInterface":{
-            "value":"prompt_kv_cache"
+            "value":"add_custom"
         },
         "output0":{
-            "dtype":"float16,float32,int8,int16,int32,uint8,uint16,bfloat16",
-            "format":"ND,ND,ND,ND,ND,ND,ND,ND",
-            "name":"out",
+            "dtype":"float16,float32,int32",
+            "format":"ND,ND,ND",
+            "name":"z",
             "paramType":"required",
             "shape":"all",
-            "unknownshape_format":"ND,ND,ND,ND,ND,ND,ND,ND"
+            "unknownshape_format":"ND,ND,ND"
         },
         "precision_reduce":{
             "flag":"true"

mindspore/lib/plugin/ascend/custom_ascendc_ops/op_impl/ai_core/tbe/config/ascend910/aic-ascend910-ops-info.json

@@ -1,5 +1,5 @@
 {
-    "AllFinite":{
+    "AddCustom":{
         "dynamicCompileStatic":{
             "flag":"true"
         },
@@ -13,9 +13,17 @@
             "flag":"true"
         },
         "input0":{
-            "dtype":"float16,float32,bfloat16",
+            "dtype":"float16,float32,int32",
             "format":"ND,ND,ND",
-            "name":"gradient",
+            "name":"x",
+            "paramType":"required",
+            "shape":"all",
+            "unknownshape_format":"ND,ND,ND"
+        },
+        "input1":{
+            "dtype":"float16,float32,int32",
+            "format":"ND,ND,ND",
+            "name":"y",
             "paramType":"required",
             "shape":"all",
             "unknownshape_format":"ND,ND,ND"
@@ -24,15 +32,15 @@
             "flag":"false"
         },
         "opFile":{
-            "value":"all_finite"
+            "value":"add_custom"
         },
         "opInterface":{
-            "value":"all_finite"
+            "value":"add_custom"
         },
         "output0":{
-            "dtype":"bool,bool,bool",
+            "dtype":"float16,float32,int32",
             "format":"ND,ND,ND",
-            "name":"is_finite",
+            "name":"z",
             "paramType":"required",
             "shape":"all",
             "unknownshape_format":"ND,ND,ND"

mindspore/lib/plugin/ascend/custom_ascendc_ops/op_impl/ai_core/tbe/config/ascend910b/aic-ascend910b-ops-info.json

@@ -1,5 +1,5 @@
 {
-    "AllFinite":{
+    "AddCustom":{
         "dynamicCompileStatic":{
             "flag":"true"
         },
@@ -13,9 +13,17 @@
             "flag":"true"
         },
         "input0":{
-            "dtype":"float16,float32,bfloat16",
+            "dtype":"float16,float32,int32",
             "format":"ND,ND,ND",
-            "name":"gradient",
+            "name":"x",
+            "paramType":"required",
+            "shape":"all",
+            "unknownshape_format":"ND,ND,ND"
+        },
+        "input1":{
+            "dtype":"float16,float32,int32",
+            "format":"ND,ND,ND",
+            "name":"y",
             "paramType":"required",
             "shape":"all",
             "unknownshape_format":"ND,ND,ND"
@@ -24,15 +32,15 @@
             "flag":"false"
         },
         "opFile":{
-            "value":"all_finite"
+            "value":"add_custom"
         },
         "opInterface":{
-            "value":"all_finite"
+            "value":"add_custom"
         },
         "output0":{
-            "dtype":"bool,bool,bool",
+            "dtype":"float16,float32,int32",
             "format":"ND,ND,ND",
-            "name":"is_finite",
+            "name":"z",
             "paramType":"required",
             "shape":"all",
             "unknownshape_format":"ND,ND,ND"

mindspore/lib/plugin/ascend/custom_ascendc_ops/op_impl/ai_core/tbe/custom_ascendc_ops_impl/dynamic/add_custom.cpp

@@ -0,0 +1,81 @@
+/*
+ * Copyright (c) Huawei Technologies Co., Ltd. 2022-2023. All rights reserved.
+ *
+ * Function : z = x + y
+ * This sample is a very basic sample that implements vector add on Ascend platform.
+ */
+#include "kernel_operator.h"
+using namespace AscendC;
+
+constexpr int32_t BUFFER_NUM = 2;
+
+class KernelAdd {
+ public:
+  __aicore__ inline KernelAdd() {}
+  __aicore__ inline void Init(GM_ADDR x, GM_ADDR y, GM_ADDR z, uint32_t totalLength, uint32_t tileNum) {
+    ASSERT(GetBlockNum() != 0 && "block dim can not be zero!");
+    this->blockLength = totalLength / GetBlockNum();
+    this->tileNum = tileNum;
+    ASSERT(tileNum != 0 && "tile num can not be zero!");
+    this->tileLength = this->blockLength / tileNum / BUFFER_NUM;
+
+    xGm.SetGlobalBuffer((__gm__ DTYPE_X *)x + this->blockLength * GetBlockIdx(), this->blockLength);
+    yGm.SetGlobalBuffer((__gm__ DTYPE_Y *)y + this->blockLength * GetBlockIdx(), this->blockLength);
+    zGm.SetGlobalBuffer((__gm__ DTYPE_Z *)z + this->blockLength * GetBlockIdx(), this->blockLength);
+    pipe.InitBuffer(inQueueX, BUFFER_NUM, this->tileLength * sizeof(DTYPE_X));
+    pipe.InitBuffer(inQueueY, BUFFER_NUM, this->tileLength * sizeof(DTYPE_Y));
+    pipe.InitBuffer(outQueueZ, BUFFER_NUM, this->tileLength * sizeof(DTYPE_Z));
+  }
+  __aicore__ inline void Process() {
+    int32_t loopCount = this->tileNum * BUFFER_NUM;
+    for (int32_t i = 0; i < loopCount; i++) {
+      CopyIn(i);
+      Compute(i);
+      CopyOut(i);
+    }
+  }
+
+ private:
+  __aicore__ inline void CopyIn(int32_t progress) {
+    LocalTensor<DTYPE_X> xLocal = inQueueX.AllocTensor<DTYPE_X>();
+    LocalTensor<DTYPE_Y> yLocal = inQueueY.AllocTensor<DTYPE_Y>();
+    DataCopy(xLocal, xGm[progress * this->tileLength], this->tileLength);
+    DataCopy(yLocal, yGm[progress * this->tileLength], this->tileLength);
+    inQueueX.EnQue(xLocal);
+    inQueueY.EnQue(yLocal);
+  }
+  __aicore__ inline void Compute(int32_t progress) {
+    LocalTensor<DTYPE_X> xLocal = inQueueX.DeQue<DTYPE_X>();
+    LocalTensor<DTYPE_Y> yLocal = inQueueY.DeQue<DTYPE_Y>();
+    LocalTensor<DTYPE_Z> zLocal = outQueueZ.AllocTensor<DTYPE_Z>();
+    Add(zLocal, xLocal, yLocal, this->tileLength);
+    outQueueZ.EnQue<DTYPE_Z>(zLocal);
+    inQueueX.FreeTensor(xLocal);
+    inQueueY.FreeTensor(yLocal);
+  }
+  __aicore__ inline void CopyOut(int32_t progress) {
+    LocalTensor<DTYPE_Z> zLocal = outQueueZ.DeQue<DTYPE_Z>();
+    DataCopy(zGm[progress * this->tileLength], zLocal, this->tileLength);
+    outQueueZ.FreeTensor(zLocal);
+  }
+
+ private:
+  TPipe pipe;
+  TQue<QuePosition::VECIN, BUFFER_NUM> inQueueX, inQueueY;
+  TQue<QuePosition::VECOUT, BUFFER_NUM> outQueueZ;
+  GlobalTensor<DTYPE_X> xGm;
+  GlobalTensor<DTYPE_Y> yGm;
+  GlobalTensor<DTYPE_Z> zGm;
+  uint32_t blockLength = 0;
+  uint32_t tileNum = 0;
+  uint32_t tileLength = 0;
+};
+
+extern "C" __global__ __aicore__ void add_custom(GM_ADDR x, GM_ADDR y, GM_ADDR z, GM_ADDR workspace, GM_ADDR tiling) {
+  GET_TILING_DATA(tilingData, tiling);
+  KernelAdd op;
+  op.Init(x, y, z, tilingData.totalLength, tilingData.tileNum);
+  if (TILING_KEY_IS(1)) {
+    op.Process();
+  }
+}

mindspore/lib/plugin/ascend/custom_ascendc_ops/op_impl/ai_core/tbe/custom_ascendc_ops_impl/dynamic/add_custom.py

@@ -0,0 +1,134 @@
+
+import os, sys
+import ctypes
+import json
+import shutil
+from tbe.common.platform import get_soc_spec
+from tbe.common.utils import para_check
+from tbe.tikcpp import compile_op, replay_op, check_op_cap, generalize_op_params, get_code_channel, OpInfo
+from tbe.common.buildcfg import get_default_build_config
+from impl.util.platform_adapter import tbe_register
+from tbe.common.buildcfg import get_current_build_config
+PYF_PATH = os.path.dirname(os.path.realpath(__file__))
+
+DTYPE_MAP = {"float32": ["DT_FLOAT", "float"],
+    "float16": ["DT_FLOAT16", "half"],
+    "int8": ["DT_INT8", "int8_t"],
+    "int16": ["DT_INT16", "int16_t"],
+    "int32": ["DT_INT32", "int32_t"],
+    "int64": ["DT_INT64", "int64_t"],
+    "uint1": ["DT_UINT1", "uint8_t"],
+    "uint8": ["DT_UINT8", "uint8_t"],
+    "uint16": ["DT_UINT16", "uint16_t"],
+    "uint32": ["DT_UINT32", "uint32_t"],
+    "uint64": ["DT_UINT64", "uint64_t"],
+    "bool": ["DT_BOOL", "bool"],
+    "double": ["DT_DOUBLE", "double"],
+    "dual": ["DT_DUAL", "unknown"],
+    "dual_sub_int8": ["DT_DUAL_SUB_INT8", "unknown"],
+    "dual_sub_uint8": ["DT_DUAL_SUB_UINT8", "unknown"],
+    "string": ["DT_STRING", "unknown"],
+    "complex64": ["DT_COMPLEX64", "unknown"],
+    "complex128": ["DT_COMPLEX128", "unknown"],
+    "qint8": ["DT_QINT8", "unknown"],
+    "qint16": ["DT_QINT16", "unknown"],
+    "qint32": ["DT_QINT32", "unknown"],
+    "quint8": ["DT_QUINT8", "unknown"],
+    "quint16": ["DT_QUINT16", "unknown"],
+    "resource": ["DT_RESOURCE", "unknown"],
+    "string_ref": ["DT_STRING_REF", "unknown"],
+    "int4": ["DT_INT4", "int8_t"],
+    "bfloat16": ["DT_BF16", "bfloat16_t"]}
+
+def get_dtype_fmt_options(__inputs__, __outputs__):
+    options = []
+    for x in __inputs__ + __outputs__:
+        x_n = x.get("param_name").upper()
+        x_fmt = x.get("format")
+        x_dtype = x.get("dtype")
+        options.append("-DDTYPE_{n}={t}".format(n=x_n, t=DTYPE_MAP.get(x_dtype)[1]))
+        options.append("-DORIG_DTYPE_{n}={ot}".format(n=x_n, ot=DTYPE_MAP.get(x_dtype)[0]))
+        options.append("-DFORMAT_{n}=FORMAT_{f}".format(n=x_n, f=x_fmt))
+    return options
+
+def load_dso(so_path):
+    try:
+        ctypes.CDLL(so_path)
+    except OSError as error :
+        print(error)
+        raise RuntimeError("cannot open %s" %(so_path))
+    else:
+        print("load so succ ", so_path)
+
+def _build_args(x, y, z):
+    __inputs__ = []
+    for arg in [x, y]:
+        if arg != None:
+            if type(arg) is list:
+                if len(arg) == 0:
+                    continue
+                __inputs__.append(arg[0])
+            else:
+                __inputs__.append(arg)
+    __outputs__ = []
+    for arg in [z]:
+        if arg != None:
+            if type(arg) is list:
+                if len(arg) == 0:
+                    continue
+                __outputs__.append(arg[0])
+            else:
+                __outputs__.append(arg)
+    __attrs__ = []
+    return __inputs__, __outputs__, __attrs__
+
+@tbe_register.register_operator("AddCustom")
+@para_check.check_op_params(para_check.REQUIRED_INPUT, para_check.REQUIRED_INPUT, para_check.REQUIRED_OUTPUT, para_check.KERNEL_NAME)
+def add_custom(x, y, z, kernel_name="add_custom", impl_mode=""):
+    if get_current_build_config("enable_op_prebuild"):
+        return
+    __inputs__, __outputs__, __attrs__ = _build_args(x, y, z)
+    options = get_dtype_fmt_options(__inputs__, __outputs__)
+    options += ["-x", "cce"]
+    ccec = os.environ.get('CCEC_REAL_PATH')
+    if ccec is None:
+        ccec = shutil.which("ccec")
+    if ccec != None:
+        ccec_path = os.path.dirname(ccec)
+        tikcpp_path = os.path.realpath(os.path.join(ccec_path, "..", "..", "tikcpp"))
+    else:
+        tikcpp_path = os.path.realpath("/usr/local/Ascend/latest/compiler/tikcpp")
+    options.append("-I" + tikcpp_path)
+    options.append("-I" + os.path.join(tikcpp_path, "tikcfw"))
+    options.append("-I" + os.path.join(tikcpp_path, "tikcfw", "impl"))
+    options.append("-I" + os.path.join(tikcpp_path, "tikcfw", "interface"))
+    options.append("-I" + os.path.join(PYF_PATH, "..", "ascendc", "common"))
+    if impl_mode == "high_performance":
+        options.append("-DHIGH_PERFORMANCE=1")
+    elif impl_mode == "high_precision":
+        options.append("-DHIGH_PRECISION=1")
+    if get_default_build_config("enable_deterministic_mode") == 1:
+        options.append("-DDETEMINISTIC_MODE=1")
+    origin_func_name = "add_custom"
+    ascendc_src_dir = "add_custom"
+    ascendc_src_file = "add_custom.cpp"
+    src = os.path.join(PYF_PATH, "..", "ascendc", ascendc_src_dir, ascendc_src_file)
+    if not os.path.exists(src):
+        src = os.path.join(PYF_PATH, ascendc_src_file)
+
+    print("start compile Ascend C operator AddCustom. kernel name is add_custom")
+    op_type = "AddCustom"
+    code_channel = get_code_channel(src, kernel_name, op_type, options)
+    op_info = OpInfo(kernel_name = kernel_name, op_type = op_type, inputs = __inputs__, outputs = __outputs__,\
+        attrs = __attrs__, impl_mode = impl_mode, origin_inputs=[x, y], origin_outputs = [z])
+    compile_op(src, origin_func_name, op_info, options, code_channel, '{}')
+
+def op_select_format(x, y, z, impl_mode=""):
+    __inputs__, __outputs__, __attrs__ = _build_args(x, y, z)
+    result = check_op_cap("op_select_format", "AddCustom", __inputs__, __outputs__, __attrs__)
+    return result.decode("utf-8")
+
+def get_op_specific_info(x, y, z, impl_mode=""):
+    __inputs__, __outputs__, __attrs__ = _build_args(x, y, z)
+    result = check_op_cap("get_op_specific_info", "AddCustom", __inputs__, __outputs__, __attrs__)
+    return result.decode("utf-8")