Title: 在電腦圖學中記憶體模型與邏輯嵌入記憶體之應用
Memory Model and Logic-Enhanced Memory in Computer Graphics
Authors: 楊穎智
Yang, Ying-Chi
任建葳
Chein-Wei Jen
電子研究所
Keywords: 電腦圖學;記憶系統;邏輯嵌入記憶體;記憶體模型;Computer Graphics;Memory System;Logic-Enhanced Memory;Memory Model
Issue Date: 1995
Abstract: 隨著邏輯運算電路的進步,記憶體逐漸跟不上運算電路的速度。在這
個情形下,一個好的記憶體系統對於電腦系統的整體效率便顯得相當重要
。然而各種不同架構的記憶體以及多樣的存取方式使得設計一個好的記憶
體系統變得十分複雜。此時一個適當的記憶體模型與記憶體模擬器可以簡
化並幫助這個設計過程。不過現有的記憶體模型多半是應用於快取記憶分
析。這些模型大多著重於記憶系統中的階級關係,對於記憶體存取方式並
沒有比較細節的模擬與分析。因此對於某些應用系統,這些模型就無法提
供適當的資訊。在本篇論文中我們提出一個新的記憶體模型。這個模型是
架構在各類應用中常見的三種區域性:時間上、位址空間上以及順序上的
區域性。根據這三種區域性我們把記憶體存取方式分成四類並分別加入不
同的參數。在設計此記憶體模型的時候我們加入了非循序放置資料的功能
,這項功能在電腦繪圖的應用上十分重要。接著我們為這個記憶體模型設
計了一個三維電腦繪圖的模擬環境。利用這個模擬環境我們將可以簡化測
試程式的設計和使用真實的應用程式來測試所模擬的記憶系統。同時經由
上面的模擬環境我們分析了現在的各種記憶體技術在電腦繪圖上的表現。
分析的結果表示目前的記憶體技術不適合處理小的繪圖元素。晶片階層的
系統整合成為未來的趨勢。我們分析了各種系統整合的方式。分析的結果
顯示對於電腦繪圖應用而言,邏輯嵌入記憶體是個比較好的選擇。在文章
的最後我們提出兩個邏輯嵌入記憶體在電腦圖學上的應用。首先是以邏輯
嵌入記憶體來加快清除螢幕的動作。我們使用邏輯嵌入記憶體來製作虛擬
清除。在這項設計之中我們估計可以用不到百分之十的晶片面積代價來加
速這個動作。在第二項設計中我們使用邏輯嵌入記憶體來加速消除鋸齒以
及透明效果的處理。我們使用在像素層次的排序配合邏輯嵌入記憶體提供
的大量記憶體頻寬來加速這項動作。我們估計可以用大約百分之十的晶片
面積代價來達成快速而正確的消除鋸齒以及透明效果。
With the growing performance gap between the memory and
computing logic technologies, the memory system design has
become an important design problem. Anyway the dynamics of
memory technologies complicate the memory system design. To
help with the memory system design, a suitable memory model and
memory simulator are important tools. However most existing
memory models focus on the memory hierarchy and do not address
the detail of various memory technologies. For some
applications, such as computer graphics ones, these simplified
models are not appropriate. In this thesis we proposed a new
high level memory model based on the 3 locality properties
observed in most applications: the temporal, the spatial and the
sequential locality. With the locality properties we classified
memory accesses into four types and modeled the four access
types with different parameters. We have also designed
rectangle page capability into the memory model for graphics
applications. We built a 3-D graphics simulation environment
using the memory model. With the simulation environment we are
able to test memory system performance for computer graphics
applications. Using the simulation environment we have analyzed
the performance of current memory technologies. We observed
that memory technologies today do not yield satisfactory
performance for small primitives. To improve the performance
chip-level integration is necessary. We compared different
system integration approaches and found that logic-enhanced
memory technology is a better one for graphics applications.
Finally we proposed two logic-enhanced memory designs in
computer graphics. The first is to implement a virtual clear
logic onto the memory. In this design we placed a management
table on the memory using chip area overhead less than 10%. The
second application is to implement correct anti-aliasing and
transparency using pixel-level sorting. With logic-enhanced
memory we could trade the memory capacity for the correctness
and performance. The resulting logic area overhead is also less
than 10%.
URI: http://140.113.39.130/cdrfb3/record/nctu/#NT840430110
http://hdl.handle.net/11536/60718
Appears in Collections:Thesis