[New Doc] add Triton 内存与数据搬运

README.md

@@ -57,6 +57,7 @@
 ### Triton 系列 💡
 
 + [Triton 编程范式入门](./docs/18_triton/01_triton_programming_paradigms/README.md)
++ [Triton 内存和数据传输](./docs/18_triton/02_triton_memory_and_data_movement/README.md)
 
 ### LLM 推理技术 🤖
 

docs/18_triton/01_triton_programming_paradigms/homework.ipynb

@@ -6,12 +6,7 @@
    "source": [
     "# Triton 编程范式 - 课后练习\n",
     "\n",
-    "本 Notebook 包含三个练习，帮助你巩固 Triton 的核心概念。\n",
-    "\n",
-    "**学习目标**：\n",
-    "- 掌握 Triton 的基本语法和向量化操作\n",
-    "- 理解 `BLOCK_SIZE` 对性能的影响\n",
-    "- 学会用向量化方式处理复杂的数据访问模式"
+    "本 Notebook 包含俩个练习，帮助你巩固 Triton 的核心概念"
    ]
   },
   {
@@ -212,15 +207,6 @@
     "    print(f\"Torch:  {y_torch[:5].cpu().numpy()}\")"
    ]
   },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "**思考题**（高级）：\n",
-    "1. 为什么这种方法效率不高？（提示：重复加载）\n",
-    "2. 如何优化？（提示：加载更大的块然后切片）"
-   ]
-  },
   {
    "cell_type": "markdown",
    "metadata": {},
@@ -229,9 +215,9 @@
     "\n",
     "## 总结\n",
     "\n",
-    "完成这三个练习后，你应该掌握了 Triton kernel 的基本写法\n",
+    "完成这两个练习后，你应该掌握了 Triton kernel 的基本写法\n",
     "\n",
-    "**下一步**：学习 Triton 的 Shared Memory 和 Block Reduction 操作！\n",
+    "**下一步**：学习 Triton 的内存与数据搬运\n",
     "\n",
     "## 课后答案\n",
     "\n",
@@ -277,7 +263,7 @@
     "    mask = offsets < n_elements\n",
     "    \n",
     "    x_center = tl.load(x_ptr + offsets, mask=mask, other=0.0)\n",
-    "    x_left = tl.load(x_ptr + offsets - 1, mask=offsets > 0, other=0.0)\n",
+    "    x_left = tl.load(x_ptr + offsets - 1, mask=mask & (offsets > 0), other=0.0)\n",
     "    x_right = tl.load(x_ptr + offsets + 1, mask=offsets < n_elements - 1, other=0.0)\n",
     "    \n",
     "    y = x_left + x_center + x_right\n",