Aren
/
haiku
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

finalized DMA buffer overfill checking for optimum speed. On NV11 accelerated functions are running at 130% to 200% speed in DMA mode compared to the 'old' PIO mode. Without trouble now. 

Still only 32bit mode, and still missing invert_rect function though. Also still not explicitly adhering the engine constraints which needs to be rechecked/retuned compared to PIO mode (hence the non working 800x600x8 mode I expect..)


git-svn-id: file:///srv/svn/repos/haiku/trunk/current@11009 a95241bf-73f2-0310-859d-f6bbb57e9c96
This commit is contained in:
Rudolf Cornelissen 2005-01-24 20:11:59 +00:00
parent c586de1d62
commit ab26d7f837
1 changed files with 8 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
src/add-ons/accelerants/nvidia/engine/nv_acc_dma.c
View File

@ -825,11 +825,15 @@ static status_t nv_acc_fifofree_dma(uint16 cmd_size)
while ((si->engine.dma.free < cmd_size) && (cnt < 10000) && (err < 3))
{
/* see where the engine is currently fetching from the buffer */
/* note:
* read this only once in the code as accessing registers is relatively slow */
dmaget = ((NV_REG32(NVACC_FIFO + NV_GENERAL_DMAGET +
si->engine.fifo.handle[(si->engine.fifo.ch_ptr[NV_ROP5_SOLID])])) >> 2);
/* snooze a bit so I do not hammer the bus */
snooze (100);
/* update timeout counter: on NV11 on a Pentium4 2.8Ghz max reached count
* using BeRoMeter 1.2.6 was about 600; so counting 10000 before generating
* a timeout should definately do it. Snooze()-ing cannot be done without a
* serious speed penalty, even if done for only 1 microSecond. */
cnt++;
/* where's the engine fetching viewed from us issuing? */
@ -857,7 +861,7 @@ static status_t nv_acc_fifofree_dma(uint16 cmd_size)
/* note the updated current free space we have in the DMA buffer */
si->engine.dma.free = dmaget - si->engine.dma.current;
/* mind this pittfall:
/* mind this pittfall (? there might be another reason):
* as the engine is DMA triggered for fetching chunks every 128 bytes,
* we may not touch that much at the end of our free space!
* (confirmed NV11 when it's buffer is full for a longer period of
@ -875,7 +879,7 @@ static status_t nv_acc_fifofree_dma(uint16 cmd_size)
/* note the updated current free space we have in the DMA buffer */
si->engine.dma.free = dmaget - si->engine.dma.current;
/* mind this pittfall:
/* mind this pittfall (? there might be another reason):
* as the engine is DMA triggered for fetching chunks every 128 bytes,
* we may not touch that much at the end of our free space!
* (confirmed NV11 when it's buffer is full for a longer period of