deepxIR:指令文档

Author: miaobyte

doc/op.md

@@ -20,4 +20,73 @@
 融合算子
 神经网络层：Conv, ConvInteger, ConvTranspose, DeformConv, GRU, LSTM, RNN, QLinearConv, QLinearMatMul
 损失函数：NegativeLogLikelihoodLoss, SoftmaxCrossEntropyLoss
-其他融合操作：AffineGrid, CenterCropPad, Col2Im, Compress, DFT, ImageDecoder, Loop, NonMaxSuppression, Optional, OptionalGetElement, OptionalHasElement, RegexFullMatch, Scan, TfIdfVectorizer, Upsample
+其他融合操作：AffineGrid, CenterCropPad, Col2Im, Compress, DFT, ImageDecoder, Loop, NonMaxSuppression, Optional, OptionalGetElement, OptionalHasElement, RegexFullMatch, Scan, TfIdfVectorizer, Upsample
+
+## op通讯指令格式
+
+### 设置常量
++ int数组
+argset@int 1 2 3->v_1
++ int
+argset@int 10->int_10
++ float数组
+argset@float 1.0 2.0 3.0->v_3
++ float
+argset@float 10.0->float_10
+
+
+
+### 创建张量
+newtensor@int32 1 2 3->a
+
+### 填充张量
+
+constant@int32 int_10->a
+
+### 打印张量
+print a
+
+### 逐元素操作
+add@float32 a b->a requires_grad=true
+add_scalar@float32 a 10->a requires_grad=true
+mul@float32 a(a_grad) b(b_grad)->a(a_grad) requires_grad=true
+mul_scalar@float32 a 10->a requires_grad=true
+div@float32 a b->a requires_grad=true
+div_scalar@float32 a 10->a requires_grad=true
+sqrt@float32 a->a requires_grad=true
+exp@float32 a->a requires_grad=true
+
+
+### 广播操作
+broadcast@float32 a b->a requires_grad=true
+
+
+### 矩阵操作
+matmul@float32 a b->a requires_grad=true
+
+
+### 归约操作
+sum@float32 a->a requires_grad=true
+max@float32 a->a requires_grad=true
+min@float32 a->a requires_grad=true
+
+
+### 形状操作
+reshape@float32 a 1 2 3->a requires_grad=true
+flatten@float32 a->a requires_grad=true
+transpose@float32 a 1 2->a requires_grad=true
+unsqueeze@float32 a 1->a requires_grad=true
+squeeze@float32 a 1->a requires_grad=true
+
+### 类型操作
+cast@float32 a->a requires_grad=true
+cast_like@float32 a b->a requires_grad=true
+ 
+
+
+
+
+
+
+
+

doc/range.md

@@ -9,4 +9,6 @@ range函数是shape类中的一个函数，用于根据shape对tensor进行遍
 |      | N    |        | print      |
 | 函数 | 否   | 0      | 不需要并行 |
 | 函数 | 是   | 0      | 需要并行   |
-| 函数 | 否   | 0      | 不需要并行 |
+| 函数 | 否   | 0      | 不需要并行 |
+
+## 

excuter/op-mem-ompsimd/src/deepx/op/opfactory.cpp

@@ -86,7 +86,7 @@ namespace deepx::op
         opfactory.add_op(Exp<double>());
     }
     void register_elementwise_op(OpFactory &opfactory){
-         register_add(opfactory);
+        register_add(opfactory);
         register_add_scalar(opfactory);
         register_sub(opfactory);
         register_mul(opfactory);