r200: make use of DMA buffers for Elts a lot better.

This allows us to return the unused portion of the dma buffer
to the allocator instead of wasting nearly 16k a pop.

2 files changed, 14 insertions, 6 deletions

diff --git a/src/mesa/drivers/dri/radeon/radeon_dma.c b/src/mesa/drivers/dri/radeon/radeon_dma.c
index 48b0d63818..0b91434b2b 100644
--- a/src/mesa/drivers/dri/radeon/radeon_dma.c
+++ b/src/mesa/drivers/dri/radeon/radeon_dma.c
@@ -167,7 +167,7 @@ void radeonRefillCurrentDmaRegion(radeonContextPtr rmesa, int size)
 	size = MAX2(size, MAX_DMA_BUF_SZ);
 
 	if (RADEON_DEBUG & (DEBUG_IOCTL | DEBUG_DMA))
-		fprintf(stderr, "%s\n", __FUNCTION__);
+		fprintf(stderr, "%s %d\n", __FUNCTION__, rmesa->dma.nr_released_bufs);
 
 	if (rmesa->dma.flush) {
 		rmesa->dma.flush(rmesa->glCtx);
@@ -178,11 +178,7 @@ void radeonRefillCurrentDmaRegion(radeonContextPtr rmesa, int size)
 		rmesa->dma.nr_released_bufs = 0;
 	}
 
-	if (rmesa->dma.current) {
-		radeon_bo_unmap(rmesa->dma.current);
-		radeon_bo_unref(rmesa->dma.current);
-		rmesa->dma.current = 0;
-	}
+	radeonReleaseDmaRegion(rmesa);
 
 again_alloc:	
 	rmesa->dma.current = radeon_bo_open(rmesa->radeonScreen->bom,
@@ -244,6 +240,17 @@ void radeonAllocDmaRegion(radeonContextPtr rmesa,
 	assert(rmesa->dma.current_used <= rmesa->dma.current->size);
 }
 
+void radeonReturnDmaRegion(radeonContextPtr rmesa, int return_bytes)
+{
+	if (!rmesa->dma.current)
+		return;
+
+	if (RADEON_DEBUG & DEBUG_IOCTL)
+		fprintf(stderr, "%s %d\n", __FUNCTION__, return_bytes);
+	rmesa->dma.current_used -= return_bytes;
+	rmesa->dma.current_vertexptr = rmesa->dma.current_used;
+}
+
 void radeonReleaseDmaRegion(radeonContextPtr rmesa)
 {
 	if (RADEON_DEBUG & DEBUG_IOCTL)
diff --git a/src/mesa/drivers/dri/radeon/radeon_dma.h b/src/mesa/drivers/dri/radeon/radeon_dma.h
index 06e388fc1d..c6eabd3bc8 100644
--- a/src/mesa/drivers/dri/radeon/radeon_dma.h
+++ b/src/mesa/drivers/dri/radeon/radeon_dma.h
@@ -39,6 +39,7 @@ void radeonEmitVec12(uint32_t *out, GLvoid * data, int stride, int count);
 void rcommon_emit_vector(GLcontext * ctx, struct radeon_aos *aos,
 			 GLvoid * data, int size, int stride, int count);
 
+void radeonReturnDmaRegion(radeonContextPtr rmesa, int return_bytes);
 void radeonRefillCurrentDmaRegion(radeonContextPtr rmesa, int size);
 void radeonAllocDmaRegion(radeonContextPtr rmesa,
 			  struct radeon_bo **pbo, int *poffset,