1<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"> 2<html> 3 4<head> 5<META http-equiv="Content-Type" content="text/html; charset=windows-1252"> 6<META name="GENERATOR" content="Microsoft FrontPage 5.0"> 7<META http-equiv="Content-Style-Type" content="text/css"> 8<LINK rel="stylesheet" type="text/css" href="../../CSS/revolution.css"> 9<title>LCStoreData</title> 10</head> 11 12<body> 13 14<h1>LCStoreData</h1> 15 16 17<h2>C Specification</h2> 18 19<dl> 20<dd><pre><code>#include <revolution/os.h></code></pre> 21 </dd> 22<dd><pre><code>u32 LCStoreData(void* <em>destAddr</em>, void* <em>srcAddr</em>, u32 <em>nBytes</em>);</code></pre> 23 </dd> 24</dl> 25<H2>Arguments</H2> 26<TABLE border="1" cellpadding="3" cellspacing="0.1"> 27 <TBODY> 28 <TR> 29<TD width="120" bgcolor="#ffffe8"><EM><STRONG><CODE>destAddr</CODE></STRONG></EM></TD> 30<TD width="520">Start address of the destination in memory. Must be 32-byte aligned.</TD> 31 </TR> 32 <TR> 33<TD width="120" bgcolor="#ffffe8"><EM><STRONG><CODE>srcAddr</CODE></STRONG></EM></TD> 34<TD width="520">Start address for the locked cache address to be copied to <CODE><EM><STRONG>destAddr</STRONG></EM></CODE>. Must be 32-byte aligned.</TD> 35 </TR> 36 <TR> 37<TD width="120" bgcolor="#ffffe8"><STRONG><EM><CODE><STRONG><EM><CODE>nBytes</CODE></EM></STRONG></CODE></EM></STRONG></TD> 38<TD width="520">Transfer size. Must be 32-byte aligned.</TD> 39 </TR> 40 </TBODY> 41</TABLE> 42<H2>Return Values</H2> 43<P>Returns the number of transactions added to the DMA queue.</P> 44<h2>Description</h2> 45 46<p>Enqueues DMA transactions to send data from the locked cache at <code><em><strong>srcAddr</strong></em></code> to main memory at <code><em><strong>destAddr</strong></em></code>. The range of valid source addresses is the 16KB region starting from the value returned by <a href="LCGetBase.html"><code>LCGetBase</code></a>. The number of transactions issued is returned.</p> 47 48<p>The only way to determine if the transactions are complete is to poll the length of the DMA queue with <code><a href="LCQueueLength.html">LCQueueLength</a></code>, or to wait until the queue length reaches a certain value with <code><a href="LCQueueWait.html">LCQueueWait</a></code>. See the example below. </p> 49 50<p>The largest allowed single DMA transfer is 128 cache blocks (4 KB). Thus, the number of transfers added (to DMA queue) is always rounded up <code>(nBytes/4 KB)</code> . <a href="LCStoreBlocks.html">LCStoreBlocks</a></code> is a more efficient function that creates a single DMA transfer. However, it does not provide error checking or argument restrictions.</p> 51 52<p>No more than 15 outstanding DMA transfers are allowed in the DMA queue. If the queue overflows, a machine check exception will occur.</p> 53 54<p>The locked cache must be enabled throughout the transfer; otherwise, a machine check exception will occur during the DMA.</p> 55 56<p>If the DMA finds the source address in the normal cache, a machine check exception occurs.</p> 57 58<h3>Example</h3> 59 60<p>The following code example splits the locked cache into 2 buffers, and ping pongs between them to process a large array in main memory. The code is structured so that each buffer has no more than one outstanding load and store at any given time. When a buffer finishes storing its old (freshly processed) data, this information can be used to ensure the next block of data to process loads correctly.</p> 61<BLOCKQUOTE><code>// define 2 8k buffers in locked cache region<br> // note that NUMBUFFERS * BUFFER_SIZE <= 16k<br> #define BUFFER_SIZE (8*1024)<br> #define NUM_BUFFERS (2)<br> <br> #define DATA_ELEMENTS (10*1024*1024)<br> :<br><br>// real mem loc of Buffers[i] is at BufAddr[i]<br> u8* Buffers[NUM_BUFFERS];<br> u8* BufAddr[NUM_BUFFERS]; <br> :<br><br> void main ()<br> {<br> u8* data;<br> u8* currDataPtr; // offset into data<br> u32 i;<br> void* arenaLo;<br> void* arenaHi;<br> u32 numTransactions;</code></BLOCKQUOTE> 62<BLOCKQUOTE><code> OSInit();<br> LCEnable();<br> <br> arenaLo = OSGetArenaLo();<br> arenaHi = OSGetArenaHi();<br> :<br> OSReport("Splitting locked cache into %d buffers\n", NUM_BUFFERS);<br> <br> for (i = 0; i < NUM_BUFFERS; i++)<br> {<br> Buffers[i] = (u8*) ((u32)LCGetBase() + BUFFER_SIZE*i);<br> OSReport("Locked Cache : Allocated %d bytes at 0x%x\n",<br> BUFFER_SIZE,<br> Buffers[i]);<br> }<br> <br> // Initialize source data<br> data = (u8*)OSAlloc(DATA_ELEMENTS * sizeof(u8));<br> :<br> DCFlushRange(data, DATA_ELEMENTS);<br> <br> OSReport(" Test 1 : using high level interface for DMA load/store \n");<br> <br> for (i = 0; i < NUM_BUFFERS; i++)<br> {<br> BufAddr[i] = data + BUFFER_SIZE*i;<br> numTransactions = LCLoadData(Buffers[i], BufAddr[i], BUFFER_SIZE);<br> }<br> <br> currDataPtr = data + BUFFER_SIZE * NUM_BUFFERS;<br> <br> LCQueueWait((NUM_BUFFERS-1) * 4);<br> <br> while (currDataPtr <= data+DATA_ELEMENTS)<br> { <br> for (i = 0; i < NUM_BUFFERS; i++)<br> {<br> LCQueueWait((NUM_BUFFERS-1)*numTransactions); // prevstore + prevload, each takes 2<br> ProcessBuf(Buffers[i]);<br> LCStoreData(BufAddr[i], Buffers[i], BUFFER_SIZE);<br> BufAddr[i] = currDataPtr; // move to next unprocessed buffer<br> LCLoadData(Buffers[i], BufAddr[i], BUFFER_SIZE);<br></code></BLOCKQUOTE> 63<BLOCKQUOTE><code> // advance the next block to be read<br> currDataPtr += BUFFER_SIZE;<br> }<br> }<br> LCQueueWait(numTransactions); // don't care about last dma's<br> : <br> <br> OSHalt("Test complete");<br> }<br> <br></code></BLOCKQUOTE> 64 65 66 67<h2>See Also</h2> 68 69<p><a href="../toc.html#Cache" target="contents">Cache Functions</a>, <br> <code><a href="LCEnable.html">LCEnable</a>, <a href="LCLoadData.html">LCLoadData</a>, <a href="LCQueueLength.html">LCQueueLength</a>, <a href="LCQueueWait.html">LCQueueWait</a>, <a href="LCStoreBlocks.html">LCStoreBlocks</a></code></p> 70 71<h2>Revision History</h2> 72<P>03/01/2006 Initial version.</P> 73</body> 74</html>