Copyright (c) 2012-2016 Hyperion Entertainment and contributors.

Difference between revisions of "DMA Engine"

From AmigaOS Documentation Wiki
Jump to: navigation, search
(Author)
(Example 1: "Blocking" mode)
(4 intermediate revisions by the same user not shown)
Line 45: Line 45:
 
=== The FslDMA API ===
 
=== The FslDMA API ===
  
The API provided by the fsldma.resource breaks down into three main parts:
+
The API provided by the fsldma.resource current consists of:
:* Memory management
+
 
:* Copy Memory functions
 
:* Copy Memory functions
:* Utility / Miscellaneous
 
  
==== Memory Management Functions ====
+
==== Copy Memory Functions ====
  
The FslDMA resource API provides convenience functions for the allocation and freeing of '''DMA Compliant''' blocks of memory. Two functions are provided for this purpose; DMAAllocPhysicalMemory() and DMAFreePhysicalMemory().
+
Three API calls make up the FslDMA API; CopyMemDMA(), CopyMemDMATagList() and CopyMemDMATags().
 
+
The memory allocation function also includes two Tags based versions of the call to allow for addition variables to be passed into the call using a variable set of Tags or fixed TagList. Currently the only supported Tag is ''FSLDMA_APM_ClearWithValue'', which is the equivalent to the IExec->AllocVecTagList() function's AVT_ClearWithValue tag. If the FSLDMA_APM_ClearWithValue tag is not provided then the requested memory block is cleared with zeroes by default before being returned. DMAAllocPhysicalMemory() returns the ''Physical'' Address pointer to the allocated memory or NULL if it is unable to allocate the requested memory.
+
  
 
<syntaxhighlight>
 
<syntaxhighlight>
   APTR DMAAllocPhysicalMemoryTagList( uint32 lSize, const struct TagItem *tags );
+
   BOOL CopyMemDMA( CONST_DMAPTR pSourceBuffer, DMAPTR pDestBuffer, uint32 lBufferSize );
   APTR DMAAllocPhysicalMemoryTags( uint32 lSize, uint32 Tag1, ... );
+
   BOOL CopyMemDMATags( CONST_DMAPTR pSourceBuffer, DMAPTR pDestBuffer, uint32 lBufferSize, uint32 TagItem1, ... );
   APTR DMAAllocPhysicalMemory( uint32 lSize );
+
   BOOL CopyMemDMATagList( CONST_DMAPTR pSourceBuffer, DMAPTR pDestBuffer, uint32 lBufferSize, const struct TagItem *tags );
 
</syntaxhighlight>
 
</syntaxhighlight>
  
The corresponding function to allocating a DMA Compliant memory block is DMAFreePhysicalMemory(). Any memory allocated using DMAAllocPhysicalMemory() must be eventually freed using DMAFreePhysicalMemory().
+
The first call, CopyMemDMA(), attempts to perform a "blocking" (does not return to the user until after the requested transfer has either succeeded or failed) operation. A value of TRUE is returned upon success and FALSE if an error occurred. The CopyMemDMA() call will automatically fall back to using an internal fast CPU copy if the requested size is too small to be efficient, or an error occurred in the DMA transaction.
  
<syntaxhighlight>
+
In theory any ''contiguous'' block of memory from 1 Byte up to 4GB in size may be transferred using any one of these calls. In practice the DMA hardware can directly accept memory blocks up to FSLDMA_MAXBLOCK_SIZE (or 64MB - 1 Byte). Therefore any data blocks greater than FSLDMA_MAXBLOCK_SIZE will automatically be sent to the DMA hardware in a series of smaller chunks. For maximum efficiency transfer sizes should be at least 256 Bytes in size and be an even multiple of 64 Bytes. Odd sizes are handled by the hardware but will degrade performance.
  void DMAFreePhysicalMemory( APTR pPhysicalMemoryBlock );
+
</syntaxhighlight>
+
  
==== Copy Memory Functions ====
+
If you are transferring many large blocks of data in series and wish to manually send them to the FslDMA API's memory copy functions one block at a time, then setting your block sizes to FSLDMA_OPTIMAL_BLKSIZE (or 64 MB - 64 Bytes) will provide the fastest transfer speeds will the least amount of CPU overhead.
  
Two API calls make up the core of the FslDMA API; DMAPhysicalCopyMem() and DMACopyMem().
+
==== Further reference - The fsldma.resource AutoDoc ====
Both of these calls will "block" (not return to the user) until after the requested transfer has either succeeded or failed. A value of TRUE is returned upon success and FALSE if an error occurred.
+
 
 +
See the [[http://wiki.amigaos.net/amiga/autodocs/fsldmares.doc.txt fsldma.resource AutoDoc]] file for more details on each API call.
 +
 
 +
=== Example 1: "Blocking" mode ===
  
 
<syntaxhighlight>
 
<syntaxhighlight>
  BOOL DMAPhysicalCopyMem( CONST_APTR pPhysicalSourceBuffer, APTR pPhysicalDestBuffer, uint32 lBufferSize );
+
#include <stdio.h>
  BOOL DMACopyMem( CONST_APTR pSourceBuffer, APTR pDestBuffer, uint32 lBufferSize );
+
#include <stdlib.h>
</syntaxhighlight>
+
#include <amiga_compiler.h>
 +
#include <exec/types.h>
 +
#include <proto/exec.h>
 +
#include <dos/dos.h>
 +
#include <interfaces/exec.h>
 +
#include <interfaces/fsldma.h>
  
As the name implies the first (and core version) of the copy memory functions, DMAPhysicalCopyMem(), accepts the Physical Addresses to the source and destination buffers (as returned by DMAAllocPhysicalMemory()), along with an unsigned 32-Bit value for the amount of bytes that should be copied (lBuffsize must be greater than zero(0) and no more than the maximum size of the source and destination buffers).
+
// fsldma.resource Example 1, "Blocking" DMA Copy
 +
// test using Virtual memory areas
 +
int main()
 +
{
 +
  struct fslDMAIFace *IfslDMA      = NULL;
 +
  uint32            lSize        = 0;
 +
  CONST_APTR        pSrc          = NULL;
 +
  APTR              pDest        = NULL;
 +
  BOOL              bGotResources = FALSE;
  
DMAPhysicalCopyMem() is the most direct an efficient means of using the DMA hardware to effect a single memory copy. The DMACopyMem() function is provided as an alternative way to request a DMA transfer by passing in the ''Virtual Addresses'' to the source and destination buffers instead of the Physical Addresses. DMACopyMem() will attempt to determine if the supplied memory buffers are DMA Compliant and if so it will internally use DMAPhysicalCopyMem() to handle the transaction. If DMACopyMem() determines that the supplied memory is '''not''' DMA Compliant it will return FALSE and no data copy will occur.
+
  // Obtain the fsldma.resource
 +
  IfslDMA = IExec->OpenResource(FSLDMA_NAME);
  
In theory any ''contiguous'' block of memory from 1 Byte up to 4GB in size may be transferred using either one of these calls. In practice the DMA hardware can directly accept memory blocks up to FSLDMA_MAXBLOCK_SIZE (or 64MB - 1 Byte). Therefore any data blocks greater than FSLDMA_MAXBLOCK_SIZE will automatically be feed to the DMA hardware in a series of smaller chunks. For maximum efficiency transfer sizes should be at least 256 Bytes in size and be an even multiple of 64 Bytes. Odd sizes are handled by the hardware but will degrade performance.
+
  // Set the size of our test buffers
 +
  lSize = FSLDMA_OPTIMAL_BLKSIZE; // About 64 MB
  
If you are transferring many large blocks of data in series and wish to manually send them to the FslDMA API's memory copy functions one block at a time, then setting your block sizes to FSLDMA_OPTIMAL_BLKSIZE (or 64 MB - 64 Bytes) and exclusively using DMAPhysicalCopyMem() will provide the fastest transfer speeds will the least amount of CPU overhead.
+
  // Allocate a test source buffer (fill with some data)
 +
  pSrc  = IExec->AllocVecTags(lSize,
 +
                              AVT_Type,              MEMF_SHARED,
 +
                              AVT_ClearWithValue,    0xB3,
 +
                              TAG_DONE);
  
==== Utility / Miscellaneous Functions ====
+
  // Allocate a test destination buffer (clear with zeroes)
 +
  pDest = IExec->AllocVecTags(lSize,
 +
                              AVT_Type,              MEMF_SHARED,
 +
                              AVT_ClearWithValue,    0,
 +
                              TAG_DONE);
  
The last section to the FslDMA API provides a single convenience function call; DMAGetVirtualAddress().
+
  // Verify we got all resources needed for the test
 +
  if ( (NULL != IfslDMA) && (NULL != pSrc) && (NULL != pDest) )
 +
  {
 +
    bGotResources = TRUE;
 +
  }
  
<syntaxhighlight>
+
   if ( TRUE == bGotResources )
   APTR DMAGetVirtualAddress( APTR pPhysicalAddress );
+
  {
</syntaxhighlight>
+
    // Call IfslDMA->CopyMemDMA() to perform the memory copy using the
 +
    // DMA hardware. Wait for the transaction to complete before continuing.
  
The DMAGetVirtualAddress() function returns the Virtual memory location for the Physical memory location that was itself allocated and returned by the DMAAllocPhysicalMemory() function. DMAGetVirtualAddress() will not work on physical addresses returned by IMMU->GetPhysicalAddress() on memory not allocated using DMAAllocPhysicalMemory().
+
    // We use full 64-Bit values to pass in the source and destination to
 +
    // IfslDMA->CopyMemDMA() so we need to use a "double cast" on the
 +
    // 32-Bit pointers returned by AllocVecTags() in order to properly pass
 +
    // in fully extended 64-Bit pointers.
 +
    // (see <resource/fsldma.h> for more details)
  
The main purpose of this function (other then debugging purposes) is to provide the "normal" Virtual Address of memory allocated by the FslDMA API for use by functions that expect "normal" Virtual address locations; for example IExec->CopyMemQuick().
+
    printf("Starting \"Blocking\" DMA Transaction"
 +
          " of %ld bytes from 0x%08lx to 0x%08lx...\n",lSize,pSrc,pDest);
 +
    fflush(stdout);
 +
    if ( TRUE == IfslDMA->CopyMemDMA((CONST_DMAPTR)(uint32)pSrc,
 +
                                    (DMAPTR)(uint32)pDest,
 +
                                    lSize) )
 +
    {
 +
      // Success - Do something with the copied data and quit
 +
      printf("Returned with Success.\n");
 +
      fflush(stdout);
 +
    }
 +
    else
 +
    {
 +
      // Report the error
 +
      printf("Received an Error!\n");
 +
      fflush(stdout);
 +
    }
 +
  }
 +
  else
 +
  {
 +
    printf("Failed to obtain resources, aborting test.\n");
 +
    fflush(stdout);
 +
  }
  
==== Further reference - The fsldma.resource AutoDoc ====
+
  // Free our test buffers (as needed)
 +
  if ( NULL != pSrc  ) IExec->FreeVec((APTR)pSrc);
 +
  if ( NULL != pDest ) IExec->FreeVec(pDest);
 +
}
 +
</syntaxhighlight>
  
See the [[http://wiki.amigaos.net/amiga/autodocs/fsldmares.doc.txt fsldma.resource AutoDoc]] file for more details on each API call.
+
=== Example 2: "Non-Blocking" mode ===
  
=== Example usage ===
 
 
<syntaxhighlight>
 
<syntaxhighlight>
 +
#include <stdio.h>
 +
#include <stdlib.h>
 +
#include <amiga_compiler.h>
 +
#include <exec/types.h>
 +
#include <proto/exec.h>
 +
#include <dos/dos.h>
 +
#include <interfaces/exec.h>
 
#include <interfaces/fsldma.h>
 
#include <interfaces/fsldma.h>
// Obtain the fsldma.resource
+
 
struct fslDMAIFace *IfslDMA = IExec->OpenResource(FSLDMA_NAME);
+
// fsldma.resource Example 2, "Non-blocking" DMA Copy
if ( NULL != IfslDMA )
+
// test using user notifications and Physical memory areas
 +
int main()
 
{
 
{
   uint32     lTestSize         = 1024;
+
  struct ExecBase    *ExecBase          = (*(struct ExecBase **)4);
   CONST_APTR pPhysicalSrcAddr  = NULL;
+
  struct MMUIFace    *IMMU              = NULL;
   APTR       pPhysicalDestAddr = NULL;
+
  struct fslDMAIFace *IfslDMA          = NULL;
   // Allocate the Source Buffer (for DMA)
+
   uint32             lSize              = 0;
   // and set the contents to 0xB3 (just an example value)
+
  CONST_APTR         pSrc              = NULL;
   pPhysicalSrcAddr = (CONST_APTR)IfslDMA->DMAAllocPhysicalMemoryTags(lTestSize,
+
  APTR              pDest              = NULL;
                                  FSLDMA_APM_ClearWithValue, 0xB3,
+
   CONST_APTR         pPhySrcAttr        = NULL;
                                  TAG_END);
+
   APTR               pPhyDstAttr        = NULL;
   if ( NULL != pPhysicalSrcAddr )
+
  int8              nSuccessSigNum    = -1;
 +
  int8              nInProgressSigNum  = -1;
 +
  int8              nErrorSigNum      = -1;
 +
  uint32            lSuccessSigMask    = 0;
 +
  uint32            lInProgressSigMask = 0;
 +
  uint32            lErrorSigMask      = 0;
 +
  uint32            lAllSigsMask      = 0;
 +
  BOOL              bGotResources      = FALSE;
 +
 
 +
   // Obtain the MMU Interface
 +
  IMMU = (struct MMUIFace *)IExec->GetInterface((struct Library *)ExecBase,
 +
                                                "MMU",1,NULL);
 +
   // Obtain the fsldma.resource
 +
  IfslDMA = IExec->OpenResource(FSLDMA_NAME);
 +
 
 +
   // Set our test buffer size large enough to generate some sub-block transfers
 +
  lSize = FSLDMA_OPTIMAL_BLKSIZE * 4; // About 256 MB
 +
 
 +
  // Allocate a test source buffer (fill with some data)
 +
  pSrc  = IExec->AllocVecTags(lSize,
 +
                              AVT_Type,             MEMF_SHARED,
 +
                              AVT_Contiguous,        TRUE,
 +
                              AVT_Lock,              TRUE,
 +
                              AVT_Alignment,        64,
 +
                              AVT_PhysicalAlignment, 64,
 +
                              AVT_ClearWithValue,    0xB3,
 +
                              TAG_DONE);
 +
 
 +
  // Allocate a test destination buffer (clear with zeroes)
 +
  pDest = IExec->AllocVecTags(lSize,
 +
                              AVT_Type,              MEMF_SHARED,
 +
                              AVT_Contiguous,        TRUE,
 +
                              AVT_Lock,              TRUE,
 +
                              AVT_Alignment,        64,
 +
                              AVT_PhysicalAlignment, 64,
 +
                              AVT_ClearWithValue,    0,
 +
                              TAG_DONE);
 +
 
 +
  // Allocate signals to wait on
 +
  nSuccessSigNum    = IExec->AllocSignal((int8)-1);
 +
  nInProgressSigNum = IExec->AllocSignal((int8)-1);
 +
  nErrorSigNum      = IExec->AllocSignal((int8)-1);
 +
 
 +
  // Construct the signal masks we need to Wait() on later.
 +
  // We are assuming these values are being built with
 +
  // allocated signals here, but we will verify it before use.
 +
  lSuccessSigMask    = (uint32)(1L << nSuccessSigNum);
 +
  lInProgressSigMask = (uint32)(1L << nInProgressSigNum);
 +
  lErrorSigMask      = (uint32)(1L << nErrorSigNum);
 +
  lAllSigsMask = (lSuccessSigMask | lInProgressSigMask | lErrorSigMask);
 +
 
 +
  // Obtain the pointer to ourselves (this Task)
 +
  struct Task *pThisTask = IExec->FindTask(NULL);
 +
 
 +
  // Verify we got all resources needed for the test
 +
   if ( (NULL != IMMU) && (NULL != IfslDMA) &&
 +
      (NULL != pSrc) && (NULL != pDest)  &&
 +
      (NULL != pThisTask )                &&
 +
      (-1 != nSuccessSigNum)              &&
 +
      (-1 != nInProgressSigNum)          &&
 +
      (-1 != nErrorSigNum)                )
 
   {
 
   {
     // Allocate the Destination Buffer (for DMA)
+
     bGotResources = TRUE;
    //  - contents will by cleared by default
+
    pPhysicalDestAddr = IfslDMA->DMAAllocPhysicalMemory(lTestSize);
+
    if ( NULL != pPhysicalDestAddr )
+
    {
+
      // Call IfslDMA->DMAPhysicalCopyMem()
+
      // to perform the memory copy using the DMA hardware
+
      if ( TRUE == IfslDMA->DMAPhysicalCopyMem(pPhysicalSrcAddr,
+
                                              pPhysicalDestAddr,lTestSize) )
+
      {
+
        // Success - Do something with the copied data
+
      }
+
      else
+
      {
+
        // Fallback and use CPU Copy instead (or do something else)
+
        // Since we allocated the memory buffers using the FslDMA API,
+
        // which returns the Physical address to that memory, and the
+
        // IExec->CopyMemQuick() function expects the Virtual address,
+
        // we first need to obtain the Virtual address for the Physical
+
        // ones (using the FslDMA API again) before we can call our
+
        // fallback copy function.
+
        CONST_APTR pVirtualSrcAddr  = NULL;
+
        APTR      pVirtualDestAddr = NULL;
+
        pVirtualSrcAddr  = IfslDMA->DMAGetVirtualAddress((APTR)pPhysicalSrcAddr);
+
        pVirtualDestAddr = IfslDMA->DMAGetVirtualAddress(pPhysicalDestAddr);
+
        IExec->CopyMemQuick(pVirtualSrcAddr,pVirtualDestAddr,lTestSize);
+
      }
+
      // We *must* use DMAFreePhysicalMemory() to free the memory
+
      // that we allocated with DMAAllocPhysicalMemory()
+
      IfslDMA->DMAFreePhysicalMemory(pPhysicalDestAddr);
+
    }
+
    // We *must* use DMAFreePhysicalMemory() to free the memory
+
    // that we allocated with DMAAllocPhysicalMemory()
+
    IfslDMA->DMAFreePhysicalMemory((APTR)pPhysicalSrcAddr);
+
 
   }
 
   }
}
 
</syntaxhighlight>
 
  
==== Obtaining the fsldma.resource ====
+
  if ( TRUE == bGotResources )
 +
  {
 +
    // In this test we are using Physical memory areas
 +
    // for both source and destination, so before we
 +
    // hand the transaction to the DMA hardware, we first
 +
    // need to set the flush the data cache, set the
 +
    // buffers to cache-inhibited and obtain the Physical
 +
    // addresses for both buffers
  
Breaking the above example down the first thing we do is include the interface header for the fsldma.resource and obtain the resource itself.
+
    // Enter Supervisor Mode
 +
    APTR pUserStack = IExec->SuperState();
  
<syntaxhighlight>
+
    // Flush out any cache for our two buffers
#include <interfaces/fsldma.h>
+
    IExec->CacheClearE((APTR)pSrc,lSize,CACRF_ClearD);
struct fslDMAIFace *IfslDMA = IExec->OpenResource(FSLDMA_NAME);
+
    IExec->CacheClearE(pDest,lSize,CACRF_ClearD);
</syntaxhighlight>
+
  
==== Allocating DMA Compliant Memory ====
+
    // Now set the memory attributes to prevent further cache operations
 +
    uint32 lSrcMemAttrs = IMMU->GetMemoryAttrs((APTR)pSrc,0);
 +
    uint32 lDstMemAttrs = IMMU->GetMemoryAttrs(pDest,0);
 +
    IMMU->SetMemoryAttrs((APTR)pSrc,lSize,(lSrcMemAttrs | FSLDMA_PHYMEM_ATTRS));
 +
    IMMU->SetMemoryAttrs(pDest,lSize,(lDstMemAttrs | FSLDMA_PHYMEM_ATTRS));
  
Once we have successfully obtained the DMA resource we can directly use its API. The next step we need to do is to allocate some memory '''which we know is DMA compliant''' for use by the resource. The easiest way to accomplish this is to use the IfslDMA->DMAAllocPhysicalMemory() function. This will automatically take care of ensuring the memory that is returned is properly aligned, contiguous, cache-inhibited and coherent.
+
    // Get the Physical addresses for our two buffers
 +
    pPhySrcAttr = IMMU->GetPhysicalAddress((APTR)pSrc);
 +
    pPhyDstAttr = IMMU->GetPhysicalAddress(pDest);
  
We use the Tags version of the allocate memory function first so we can allocate a block of memory for our source buffer and fill it with some test value (in this case the byte value 0xB3).
+
    // Return to User Mode
 +
    if ( NULL != pUserStack ) IExec->UserState(pUserStack);
  
<syntaxhighlight>
+
    // Call IfslDMA->CopyMemDMATags() to perform the memory copy using the
  pPhysicalSrcAddr = (CONST_APTR)IfslDMA->DMAAllocPhysicalMemoryTags(lTestSize,
+
    // DMA hardware. Start off the transaction then wait on completion signals.
                                  FSLDMA_APM_ClearWithValue, 0xB3,
+
                                  TAG_END);
+
</syntaxhighlight>
+
  
We also need a destination buffer for our test. Since it only needs to start out being cleared (filled with zeroes), we can use the simplest form of the allocate memory function here as the allocated memory is cleared by default.
+
    // We use full 64-Bit values to pass in the source and destination to
 +
    // IfslDMA->CopyMemDMATags() so we need to use a "double cast" on the
 +
    // 32-Bit pointers returned by AllocVecTags() in order to properly pass
 +
    // in fully extended 64-Bit pointers.
 +
    // (see <resource/fsldma.h> for more details)
  
<syntaxhighlight>
+
    printf("Starting \"Non-Blocking\" DMA Transaction"
  pPhysicalDestAddr = IfslDMA->DMAAllocPhysicalMemory(lTestSize);
+
          " of %ld bytes from 0x%08lx to 0x%08lx...\n",lSize,pPhySrcAttr,pPhyDstAttr);
</syntaxhighlight>
+
    fflush(stdout);
 +
    if ( TRUE == IfslDMA->CopyMemDMATags((CONST_DMAPTR)(uint32)pPhySrcAttr,
 +
                            (DMAPTR)(uint32)pPhyDstAttr,
 +
                            lSize,
 +
                            FSLDMA_CM_SourceIsPhysical,          TRUE,
 +
                            FSLDMA_CM_DestinationIsPhysical,    TRUE,
 +
                            FSLDMA_CM_DoNotWait,                TRUE,
 +
                            FSLDMA_CM_NotifyTask,                pThisTask,
 +
                            FSLDMA_CM_NotifySignalNumber,        nSuccessSigNum,
 +
                            FSLDMA_CM_NotifyProgessSignalNumber, nInProgressSigNum,
 +
                            FSLDMA_CM_NotifyErrorSignalNumber,  nErrorSigNum,
 +
                            TAG_DONE) )
 +
    {
 +
      // If the above call sets up correctly, it returns immeditately
 +
      // So do something here *while* the data is being copied...
 +
      printf("Doing other stuff before waiting...\n");
 +
      fflush(stdout);
  
==== The core function - Copy Physical Memory ====
+
      // Now we will Wait() for the completion signals from the DMA
 +
      BOOL  bStillRunning = TRUE;
 +
      uint32 lSigReceived  = 0;
 +
      do
 +
      {
 +
        // Wait for the DMA completion/status/error signals
 +
        printf("Waiting for signals...\n");
 +
        fflush(stdout);
 +
        lSigReceived = IExec->Wait( lAllSigsMask | SIGBREAKF_CTRL_C );
  
Now that we have two DMA Compliant memory buffers available we can get to the heart of the API usage and make the call to IfslDMA->DMAPhysicalCopyMem().
+
        // Test for the "In Progress" signal
 +
        if ( lInProgressSigMask == (lSigReceived & lInProgressSigMask) )
 +
        {
 +
          // Do something on the "In Process" signal and continue...
 +
          printf("Received an \"In Progress\" signal...\n");
 +
          fflush(stdout);
 +
        }
  
<syntaxhighlight>
+
        // Test for the "Success" signal
  IfslDMA->DMAPhysicalCopyMem(pPhysicalSrcAddr,pPhysicalDestAddr,lTestSize);
+
        if ( lSuccessSigMask == (lSigReceived & lSuccessSigMask) )
</syntaxhighlight>
+
        {
 +
          // Success - Do something with the copied data and quit
 +
          printf("Received an \"Completed Successfully\" signal.\n");
 +
          fflush(stdout);
 +
          bStillRunning = FALSE;
 +
        }
  
Looking at the full [[#Example usage|Example]] source above you can see that the DMAPhysicalCopyMem() call returns a Boolean value to indicate success or failure. Therefore, TRUE will be returned after the DMA hardware had completed copying the requested data (blocking call) and FALSE will be returned if a problem occurred.
+
        // Test for the "Error" signal
 +
        if ( lErrorSigMask == (lSigReceived & lErrorSigMask) )
 +
        {
 +
          // Report the error and quit
 +
          printf("Received an \"Error\" signal!\n");
 +
          fflush(stdout);
 +
          bStillRunning = FALSE;
 +
        }
  
Since the memory buffers we are using were allocated using the FslDMA API and our transfer size was greater than zero and less than or equal to the total size of either buffer, (in others words a legal copy request), there is '''very''' little chance that the DMAPhysicalCopyMem() function will fail. In fact, about the only reason the DMA hardware would fail to handle a legal transaction would be if the physical RAM installed in the system was faulty.
+
        // Finally, check if we got a Break signal from the user
 +
        // just in case we want to quit out early for some reason
 +
        if ( SIGBREAKF_CTRL_C == (lSigReceived & SIGBREAKF_CTRL_C) )
 +
        {
 +
          printf("Received an \"User break\" signal, ending.\n");
 +
          fflush(stdout);
 +
          bStillRunning = FALSE;
 +
        }
 +
      } while( TRUE == bStillRunning );
 +
    }
 +
  }
 +
  else
 +
  {
 +
    printf("Failed to obtain resources, aborting test.\n");
 +
    fflush(stdout);
 +
  }
  
So even though it is extremely unlikely for our DMA copy to have failed and returned FALSE, let's take a look at how we might handle the failure. In other words, falling back and using a CPU based copy function instead to complete the data move.
+
  // Free our test buffers (as needed)
 +
  if ( NULL != pSrc  ) IExec->FreeVec((APTR)pSrc);
 +
  if ( NULL != pDest ) IExec->FreeVec(pDest);
 +
  // Free any signals we (may have) obtained earlier
 +
  if ( -1 != nSuccessSigNum    ) IExec->FreeSignal(nSuccessSigNum);
 +
  if ( -1 != nInProgressSigNum ) IExec->FreeSignal(nInProgressSigNum);
 +
  if ( -1 != nErrorSigNum      ) IExec->FreeSignal(nErrorSigNum);
 +
  // Release the MMU Interface (as needed)
 +
  if ( NULL != IMMU ) IExec->DropInterface((struct Interface *)IMMU);
 +
}
 +
</syntaxhighlight>
  
Since we are reverting back to using a normal system copy memory function here, we first need to obtain the ''Virtual Addresses'' to both our source and destination buffers. This is accomplished by using the DMAGetVirtualAddress() function and passing in the Physical Address that was returned by DMAAllocPhysicalMemory().
+
==== Obtaining the fsldma.resource ====
 +
 
 +
Breaking the above example down the first thing we do is include the interface header for the fsldma.resource and obtain the resource itself.
  
 
<syntaxhighlight>
 
<syntaxhighlight>
  CONST_APTR pVirtualSrcAddr  = NULL;
+
#include <interfaces/fsldma.h>
  APTR      pVirtualDestAddr = NULL;
+
struct fslDMAIFace *IfslDMA = IExec->OpenResource(FSLDMA_NAME);
  pVirtualSrcAddr  = IfslDMA->DMAGetVirtualAddress((APTR)pPhysicalSrcAddr);
+
  pVirtualDestAddr = IfslDMA->DMAGetVirtualAddress(pPhysicalDestAddr);
+
 
</syntaxhighlight>
 
</syntaxhighlight>
  
Now that we have the Virtual Address equivalents to the Physical Addresses that were returned by DMAAllocPhysicalMemory(), we can proceed to call the Exec CopyMemQuick() function to complete the copy. Here we can only trust that the IExec->CopyMemQuick() call can not fail since it does not return a result code.
+
==== Allocating DMA Compliant Memory ====
 +
 
 +
Once we have successfully obtained the DMA resource we can directly use its API. The next step we need to do is to allocate some memory '''which we know is DMA compliant''' for use by the resource. This can be accomplished by using the IExec->AllocVecTags() function together with a couple of MMU functions. Together they allow you to ensuring the memory that is returned is properly aligned, contiguous, cache-inhibited and coherent.
 +
 
 +
We use the Tags version of the allocate memory function first so we can allocate a block of memory for our source buffer and fill it with some test value (in this case the byte value 0xB3).
  
 
<syntaxhighlight>
 
<syntaxhighlight>
  IExec->CopyMemQuick(pVirtualSrcAddr,pVirtualDestAddr,lTestSize);
 
 
</syntaxhighlight>
 
</syntaxhighlight>
  
==== Cleaning up - Freeing DMA Compliant memory ====
+
We also need a destination buffer for our test. Since it only needs to start out being cleared (filled with zeroes), we can use the simplest form of the allocate memory function here as the allocated memory is cleared by default.
 
+
It is essential that you free any memory that was allocated using the DMAAllocPhysicalMemory() function using the corresponding DMAFreePhysicalMemory() call. The reason for this is two-fold; first because additional resource tracking is maintained by the FslDMA API whenever it allocates memory which must itself be released, and second because the ''Physical Address'' to the memory is returned by the allocate call and not the ''Virtual Address'', so if you attempt to pass the address returned by DMAAllocPhysicalMemory() directly into IExec->FreeVec() it would likely result in a crash.
+
  
 
<syntaxhighlight>
 
<syntaxhighlight>
  IfslDMA->DMAFreePhysicalMemory(pPhysicalSrcAddr);
 
  IfslDMA->DMAFreePhysicalMemory(pPhysicalDestAddr);
 
 
</syntaxhighlight>
 
</syntaxhighlight>
  
Line 240: Line 408:
  
 
Additionally the memory blocks (source and destination) should be aligned to start on a 64 Byte boundary.
 
Additionally the memory blocks (source and destination) should be aligned to start on a 64 Byte boundary.
Alignment requirements are one of items taken care of for you by the DMAAllocPhysicalMemory() function.
 
  
 
The DMA Engine can and does handle misaligned and odd sized data blocks by first shifting the minimum
 
The DMA Engine can and does handle misaligned and odd sized data blocks by first shifting the minimum

Revision as of 00:37, 5 December 2019

Author

Jamie Krueger, BITbyBIT Software Group LLC
Copyright (c) 2019 Trevor Dickinson
Used by permission.

DMA Engine

Some hardware targets include a DMA engine which can be used for general purpose copying. This article describes the DMA engines available and how to use them.

Hardware Features

The Direct Memory Access (DMA) Engines found in the NXP/Freescale p5020, p5040 and p1022 System On a Chip (SoC)s, as found in the AmigaONE X5000/20, X5000/40 and A1222 respectively, are quite flexible and powerful. Each of these chips contains two distinct engines with four data channels each. This provides the ability to have a total of eight DMA Channels working at once, with up to two DMA transactions actually being executed at the same time (one on each of the two DMA Engines).

Further, each of the four DMA Channels found in a DMA Engine may be individually programmed to handle either; a single transaction, a Chain of transactions, or even Lists of Chains of transactions. The DMA Engines automatically arbitrate control between each DMA Channel following programmed bandwidth settings for each Channel (typically 1024 bytes).

This means that after completing a transfer of 1024 bytes (for example), the hardware will consider switching to the next Channel to allow it to move another block of data, and so on in a round-robin fashion. If all other DMA Channels on a given DMA Engine are idle when arbitration would take place, the hardware will not arbitrate control to another Channel, but simply continue processing the transaction(s) for the Channel it is on.

DMA Copy Memory - Execution Flow Diagram

DMACopyMem-Diagram.png

What a call to perform a DMA copy does internally

As shown in the above diagram, when the user makes a call to request a memory copy be performed by the DMA hardware, the next available DMA Channel is selected for use and a DMA Transaction (source, destination and size) is constructed. The DMA Transaction is then programmed into the DMA Engine which owns the available DMA Channel.

At this point the calling task will Wait() until it hears the transaction has been completed. It will then return to the caller with the result. This provides a basic blocking function, which only returns to the caller once that data has been copied. This single tasking behavior is the simplest to use and what is normally expected by most applications using a memory copy function.

Diagram of multitasking DMA Copies

DMACopyMem-Multitasking-Diagram.png

How multiple simultaneous DMA copies are handled

When multiple user calls requesting a DMA copy arrive at once, each one is handed to a dedicated DMA Channel handling task for processing. As the diagram above demonstrates, there are two separate DMA Engines available, each with four channels that may be programmed at the same time. The hardware will then arbitrate the actual data move across these channels according to their respective bandwidth settings (usually 1024 bytes).

In the diagram above, a separate color indicates a distinct data path from the caller through the DMA hardware to the system RAM. A dashed line of the matching color indicates an Interrupt line signaling the respective DMA Channel Handler with the completion of the transaction. The handler task then signals back to the original caller, which returns to the user with a success or failure result.

All eight DMA Channels can handle each a single block transaction or an entire chain of block transactions before it signals completion and returns to the original caller. If all eight DMA Channels are busy processing their requested transactions when further DMA copy requests arrive, they will each be assigned a DMA Channel to wait on (managed via a Mutex lock on each Channel) and will block until allowed to add their DMA transaction to the Channel's queue.

fsldma.resource

The fsldma.resource API is provided automatically in the kernel for all supported machines (Currently the AmigaONE X5000/20, X5000/40 and A1222).

The FslDMA API

The API provided by the fsldma.resource current consists of:

  • Copy Memory functions

Copy Memory Functions

Three API calls make up the FslDMA API; CopyMemDMA(), CopyMemDMATagList() and CopyMemDMATags().

  BOOL CopyMemDMA( CONST_DMAPTR pSourceBuffer, DMAPTR pDestBuffer, uint32 lBufferSize );
  BOOL CopyMemDMATags( CONST_DMAPTR pSourceBuffer, DMAPTR pDestBuffer, uint32 lBufferSize, uint32 TagItem1, ... );
  BOOL CopyMemDMATagList( CONST_DMAPTR pSourceBuffer, DMAPTR pDestBuffer, uint32 lBufferSize, const struct TagItem *tags );

The first call, CopyMemDMA(), attempts to perform a "blocking" (does not return to the user until after the requested transfer has either succeeded or failed) operation. A value of TRUE is returned upon success and FALSE if an error occurred. The CopyMemDMA() call will automatically fall back to using an internal fast CPU copy if the requested size is too small to be efficient, or an error occurred in the DMA transaction.

In theory any contiguous block of memory from 1 Byte up to 4GB in size may be transferred using any one of these calls. In practice the DMA hardware can directly accept memory blocks up to FSLDMA_MAXBLOCK_SIZE (or 64MB - 1 Byte). Therefore any data blocks greater than FSLDMA_MAXBLOCK_SIZE will automatically be sent to the DMA hardware in a series of smaller chunks. For maximum efficiency transfer sizes should be at least 256 Bytes in size and be an even multiple of 64 Bytes. Odd sizes are handled by the hardware but will degrade performance.

If you are transferring many large blocks of data in series and wish to manually send them to the FslDMA API's memory copy functions one block at a time, then setting your block sizes to FSLDMA_OPTIMAL_BLKSIZE (or 64 MB - 64 Bytes) will provide the fastest transfer speeds will the least amount of CPU overhead.

Further reference - The fsldma.resource AutoDoc

See the [fsldma.resource AutoDoc] file for more details on each API call.

Example 1: "Blocking" mode

#include <stdio.h>
#include <stdlib.h>
#include <amiga_compiler.h>
#include <exec/types.h>
#include <proto/exec.h>
#include <dos/dos.h>
#include <interfaces/exec.h>
#include <interfaces/fsldma.h>
 
// fsldma.resource Example 1, "Blocking" DMA Copy
// test using Virtual memory areas
int main()
{
  struct fslDMAIFace *IfslDMA      = NULL;
  uint32             lSize         = 0;
  CONST_APTR         pSrc          = NULL;
  APTR               pDest         = NULL;
  BOOL               bGotResources = FALSE;
 
  // Obtain the fsldma.resource
  IfslDMA = IExec->OpenResource(FSLDMA_NAME);
 
  // Set the size of our test buffers
  lSize = FSLDMA_OPTIMAL_BLKSIZE; // About 64 MB
 
  // Allocate a test source buffer (fill with some data)
  pSrc  = IExec->AllocVecTags(lSize,
                              AVT_Type,              MEMF_SHARED,
                              AVT_ClearWithValue,    0xB3,
                              TAG_DONE);
 
  // Allocate a test destination buffer (clear with zeroes)
  pDest = IExec->AllocVecTags(lSize,
                              AVT_Type,              MEMF_SHARED,
                              AVT_ClearWithValue,    0,
                              TAG_DONE);
 
  // Verify we got all resources needed for the test
  if ( (NULL != IfslDMA) && (NULL != pSrc) && (NULL != pDest) )
  {
    bGotResources = TRUE;
  }
 
  if ( TRUE == bGotResources )
  {
    // Call IfslDMA->CopyMemDMA() to perform the memory copy using the
    // DMA hardware. Wait for the transaction to complete before continuing.
 
    // We use full 64-Bit values to pass in the source and destination to
    // IfslDMA->CopyMemDMA() so we need to use a "double cast" on the
    // 32-Bit pointers returned by AllocVecTags() in order to properly pass
    // in fully extended 64-Bit pointers.
    // (see <resource/fsldma.h> for more details)
 
    printf("Starting \"Blocking\" DMA Transaction"
           " of %ld bytes from 0x%08lx to 0x%08lx...\n",lSize,pSrc,pDest);
    fflush(stdout);
    if ( TRUE == IfslDMA->CopyMemDMA((CONST_DMAPTR)(uint32)pSrc,
                                     (DMAPTR)(uint32)pDest,
                                     lSize) )
    {
      // Success - Do something with the copied data and quit
      printf("Returned with Success.\n");
      fflush(stdout);
    }
    else
    {
      // Report the error
      printf("Received an Error!\n");
      fflush(stdout);
    }
  }
  else
  {
    printf("Failed to obtain resources, aborting test.\n");
    fflush(stdout);
  }
 
  // Free our test buffers (as needed)
  if ( NULL != pSrc  ) IExec->FreeVec((APTR)pSrc);
  if ( NULL != pDest ) IExec->FreeVec(pDest);
}

Example 2: "Non-Blocking" mode

#include <stdio.h>
#include <stdlib.h>
#include <amiga_compiler.h>
#include <exec/types.h>
#include <proto/exec.h>
#include <dos/dos.h>
#include <interfaces/exec.h>
#include <interfaces/fsldma.h>
 
// fsldma.resource Example 2, "Non-blocking" DMA Copy
// test using user notifications and Physical memory areas
int main()
{
  struct ExecBase    *ExecBase          = (*(struct ExecBase **)4);
  struct MMUIFace    *IMMU              = NULL;
  struct fslDMAIFace *IfslDMA           = NULL;
  uint32             lSize              = 0;
  CONST_APTR         pSrc               = NULL;
  APTR               pDest              = NULL;
  CONST_APTR         pPhySrcAttr        = NULL;
  APTR               pPhyDstAttr        = NULL;
  int8               nSuccessSigNum     = -1;
  int8               nInProgressSigNum  = -1;
  int8               nErrorSigNum       = -1;
  uint32             lSuccessSigMask    = 0;
  uint32             lInProgressSigMask = 0;
  uint32             lErrorSigMask      = 0;
  uint32             lAllSigsMask       = 0;
  BOOL               bGotResources      = FALSE;
 
  // Obtain the MMU Interface
  IMMU = (struct MMUIFace *)IExec->GetInterface((struct Library *)ExecBase,
                                                "MMU",1,NULL);
  // Obtain the fsldma.resource
  IfslDMA = IExec->OpenResource(FSLDMA_NAME);
 
  // Set our test buffer size large enough to generate some sub-block transfers
  lSize = FSLDMA_OPTIMAL_BLKSIZE * 4; // About 256 MB
 
  // Allocate a test source buffer (fill with some data)
  pSrc  = IExec->AllocVecTags(lSize,
                              AVT_Type,              MEMF_SHARED,
                              AVT_Contiguous,        TRUE,
                              AVT_Lock,              TRUE,
                              AVT_Alignment,         64,
                              AVT_PhysicalAlignment, 64,
                              AVT_ClearWithValue,    0xB3,
                              TAG_DONE);
 
  // Allocate a test destination buffer (clear with zeroes)
  pDest = IExec->AllocVecTags(lSize,
                              AVT_Type,              MEMF_SHARED,
                              AVT_Contiguous,        TRUE,
                              AVT_Lock,              TRUE,
                              AVT_Alignment,         64,
                              AVT_PhysicalAlignment, 64,
                              AVT_ClearWithValue,    0,
                              TAG_DONE);
 
  // Allocate signals to wait on
  nSuccessSigNum    = IExec->AllocSignal((int8)-1);
  nInProgressSigNum = IExec->AllocSignal((int8)-1);
  nErrorSigNum      = IExec->AllocSignal((int8)-1);
 
  // Construct the signal masks we need to Wait() on later.
  // We are assuming these values are being built with
  // allocated signals here, but we will verify it before use.
  lSuccessSigMask    = (uint32)(1L << nSuccessSigNum);
  lInProgressSigMask = (uint32)(1L << nInProgressSigNum);
  lErrorSigMask      = (uint32)(1L << nErrorSigNum);
  lAllSigsMask = (lSuccessSigMask | lInProgressSigMask | lErrorSigMask);
 
  // Obtain the pointer to ourselves (this Task)
  struct Task *pThisTask = IExec->FindTask(NULL);
 
  // Verify we got all resources needed for the test
  if ( (NULL != IMMU) && (NULL != IfslDMA) &&
       (NULL != pSrc) && (NULL != pDest)   &&
       (NULL != pThisTask )                &&
       (-1 != nSuccessSigNum)              &&
       (-1 != nInProgressSigNum)           &&
       (-1 != nErrorSigNum)                 )
  {
    bGotResources = TRUE;
  }
 
  if ( TRUE == bGotResources )
  {
    // In this test we are using Physical memory areas
    // for both source and destination, so before we
    // hand the transaction to the DMA hardware, we first
    // need to set the flush the data cache, set the
    // buffers to cache-inhibited and obtain the Physical
    // addresses for both buffers
 
    // Enter Supervisor Mode
    APTR pUserStack = IExec->SuperState();
 
    // Flush out any cache for our two buffers
    IExec->CacheClearE((APTR)pSrc,lSize,CACRF_ClearD);
    IExec->CacheClearE(pDest,lSize,CACRF_ClearD);
 
    // Now set the memory attributes to prevent further cache operations
    uint32 lSrcMemAttrs = IMMU->GetMemoryAttrs((APTR)pSrc,0);
    uint32 lDstMemAttrs = IMMU->GetMemoryAttrs(pDest,0);
    IMMU->SetMemoryAttrs((APTR)pSrc,lSize,(lSrcMemAttrs | FSLDMA_PHYMEM_ATTRS));
    IMMU->SetMemoryAttrs(pDest,lSize,(lDstMemAttrs | FSLDMA_PHYMEM_ATTRS));
 
    // Get the Physical addresses for our two buffers
    pPhySrcAttr = IMMU->GetPhysicalAddress((APTR)pSrc);
    pPhyDstAttr = IMMU->GetPhysicalAddress(pDest);
 
    // Return to User Mode
    if ( NULL != pUserStack ) IExec->UserState(pUserStack);
 
    // Call IfslDMA->CopyMemDMATags() to perform the memory copy using the
    // DMA hardware. Start off the transaction then wait on completion signals.
 
    // We use full 64-Bit values to pass in the source and destination to
    // IfslDMA->CopyMemDMATags() so we need to use a "double cast" on the
    // 32-Bit pointers returned by AllocVecTags() in order to properly pass
    // in fully extended 64-Bit pointers.
    // (see <resource/fsldma.h> for more details)
 
    printf("Starting \"Non-Blocking\" DMA Transaction"
           " of %ld bytes from 0x%08lx to 0x%08lx...\n",lSize,pPhySrcAttr,pPhyDstAttr);
    fflush(stdout);
    if ( TRUE == IfslDMA->CopyMemDMATags((CONST_DMAPTR)(uint32)pPhySrcAttr,
                            (DMAPTR)(uint32)pPhyDstAttr,
                            lSize,
                            FSLDMA_CM_SourceIsPhysical,          TRUE,
                            FSLDMA_CM_DestinationIsPhysical,     TRUE,
                            FSLDMA_CM_DoNotWait,                 TRUE,
                            FSLDMA_CM_NotifyTask,                pThisTask,
                            FSLDMA_CM_NotifySignalNumber,        nSuccessSigNum,
                            FSLDMA_CM_NotifyProgessSignalNumber, nInProgressSigNum,
                            FSLDMA_CM_NotifyErrorSignalNumber,   nErrorSigNum,
                            TAG_DONE) )
    {
      // If the above call sets up correctly, it returns immeditately
      // So do something here *while* the data is being copied...
      printf("Doing other stuff before waiting...\n");
      fflush(stdout);
 
      // Now we will Wait() for the completion signals from the DMA
      BOOL   bStillRunning = TRUE;
      uint32 lSigReceived  = 0;
      do
      {
        // Wait for the DMA completion/status/error signals
        printf("Waiting for signals...\n");
        fflush(stdout);
        lSigReceived = IExec->Wait( lAllSigsMask | SIGBREAKF_CTRL_C );
 
        // Test for the "In Progress" signal
        if ( lInProgressSigMask == (lSigReceived & lInProgressSigMask) )
        {
          // Do something on the "In Process" signal and continue...
          printf("Received an \"In Progress\" signal...\n");
          fflush(stdout);
        }
 
        // Test for the "Success" signal
        if ( lSuccessSigMask == (lSigReceived & lSuccessSigMask) )
        {
          // Success - Do something with the copied data and quit
          printf("Received an \"Completed Successfully\" signal.\n");
          fflush(stdout);
          bStillRunning = FALSE;
        }
 
        // Test for the "Error" signal
        if ( lErrorSigMask == (lSigReceived & lErrorSigMask) )
        {
          // Report the error and quit
          printf("Received an \"Error\" signal!\n");
          fflush(stdout);
          bStillRunning = FALSE;
        }
 
        // Finally, check if we got a Break signal from the user
        // just in case we want to quit out early for some reason
        if ( SIGBREAKF_CTRL_C == (lSigReceived & SIGBREAKF_CTRL_C) )
        {
          printf("Received an \"User break\" signal, ending.\n");
          fflush(stdout);
          bStillRunning = FALSE;
        }
      } while( TRUE == bStillRunning );
    }
  }
  else
  {
    printf("Failed to obtain resources, aborting test.\n");
    fflush(stdout);
  }
 
  // Free our test buffers (as needed)
  if ( NULL != pSrc  ) IExec->FreeVec((APTR)pSrc);
  if ( NULL != pDest ) IExec->FreeVec(pDest);
  // Free any signals we (may have) obtained earlier
  if ( -1 != nSuccessSigNum    ) IExec->FreeSignal(nSuccessSigNum);
  if ( -1 != nInProgressSigNum ) IExec->FreeSignal(nInProgressSigNum);
  if ( -1 != nErrorSigNum      ) IExec->FreeSignal(nErrorSigNum);
  // Release the MMU Interface (as needed)
  if ( NULL != IMMU ) IExec->DropInterface((struct Interface *)IMMU);
}

Obtaining the fsldma.resource

Breaking the above example down the first thing we do is include the interface header for the fsldma.resource and obtain the resource itself.

#include <interfaces/fsldma.h>
struct fslDMAIFace *IfslDMA = IExec->OpenResource(FSLDMA_NAME);

Allocating DMA Compliant Memory

Once we have successfully obtained the DMA resource we can directly use its API. The next step we need to do is to allocate some memory which we know is DMA compliant for use by the resource. This can be accomplished by using the IExec->AllocVecTags() function together with a couple of MMU functions. Together they allow you to ensuring the memory that is returned is properly aligned, contiguous, cache-inhibited and coherent.

We use the Tags version of the allocate memory function first so we can allocate a block of memory for our source buffer and fill it with some test value (in this case the byte value 0xB3).

 

We also need a destination buffer for our test. Since it only needs to start out being cleared (filled with zeroes), we can use the simplest form of the allocate memory function here as the allocated memory is cleared by default.

 

Performance

Testing DMA memory copies vs their CPU based equivalent indicates that the DMA hardware on all three models (X5000/20, X5000/40 and A1222) move data approximately two to three times faster than the CPU does so alone.

It should also be noted that these tests were performed when the CPU was effectively idle. CPU memory copy operations scale down roughly equal with the CPU actively scaling up. So the more the CPU is doing, the longer it takes to complete the memory copy.

Conversely, DMA memory copy operation times are far more predictable as they use the same amount of (minimal) CPU overhead for each copy, and the actual time it takes the DMA hardware to complete the transaction can be calculated to range from all the DMA Channels being idle, to all DMA Channels being busy at once and data moves arbitrating between channels every 1024 bytes.

Optimal use of the DMA Engine

To gain the best performance from the DMA Engines, block sizes of 256 bytes or more should be used. Also, if possible the size should be an even multiple of at least 4 bytes.

Additionally the memory blocks (source and destination) should be aligned to start on a 64 Byte boundary.

The DMA Engine can and does handle misaligned and odd sized data blocks by first shifting the minimum required bytes to correct the alignment, then taking smaller chunks of data until the largest chunk of the block can be moved. It can also handle copies of as little as one byte (don't do this).

However, this does degrade performance.

In general, the larger the data block the better.