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.

Cherry picked and ported to new code by Pauli.

3 files changed, 18 insertions, 11 deletions

diff --git a/src/mesa/drivers/dri/r200/r200_cmdbuf.c b/src/mesa/drivers/dri/r200/r200_cmdbuf.c
index 14d6bc19c9..e63935378e 100644
--- a/src/mesa/drivers/dri/r200/r200_cmdbuf.c
+++ b/src/mesa/drivers/dri/r200/r200_cmdbuf.c
@@ -219,6 +219,9 @@ void r200FlushElts(GLcontext *ctx)
    radeon_bo_unref(rmesa->radeon.tcl.elt_dma_bo);
    rmesa->radeon.tcl.elt_dma_bo = NULL;
 
+   if (R200_ELT_BUF_SZ > elt_used)
+     radeonReturnDmaRegion(&rmesa->radeon, R200_ELT_BUF_SZ - elt_used);
+
    if (R200_DEBUG & DEBUG_SYNC) {
       fprintf(stderr, "%s: Syncing\n", __FUNCTION__);
       radeonFinish( rmesa->radeon.glCtx );
@@ -240,22 +243,13 @@ GLushort *r200AllocEltsOpenEnded( r200ContextPtr rmesa,
    
    radeonEmitState(&rmesa->radeon);
 
-   rmesa->radeon.tcl.elt_dma_bo = radeon_bo_open(rmesa->radeon.radeonScreen->bom,
-					  0, R200_ELT_BUF_SZ, 4,
-					  RADEON_GEM_DOMAIN_GTT, 0);
-   rmesa->radeon.tcl.elt_dma_offset = 0;
+   radeonAllocDmaRegion(&rmesa->radeon, &rmesa->radeon.tcl.elt_dma_bo,
+			&rmesa->radeon.tcl.elt_dma_offset, R200_ELT_BUF_SZ, 4);
    rmesa->tcl.elt_used = min_nr * 2;
 
-   ret = radeon_cs_space_check_with_bo(rmesa->radeon.cmdbuf.cs, rmesa->radeon.tcl.elt_dma_bo,
-				 RADEON_GEM_DOMAIN_GTT, 0);
-   if (ret) {
-      fprintf(stderr,"failure to revalidate BOs - badness\n");
-   }
-
    radeon_bo_map(rmesa->radeon.tcl.elt_dma_bo, 1);
    retval = rmesa->radeon.tcl.elt_dma_bo->ptr + rmesa->radeon.tcl.elt_dma_offset;
    
-
    if (R200_DEBUG & DEBUG_PRIMS)
       fprintf(stderr, "%s: header prim %x \n",
 	      __FUNCTION__, primitive);
diff --git a/src/mesa/drivers/dri/radeon/radeon_dma.c b/src/mesa/drivers/dri/radeon/radeon_dma.c
index 93bcae2454..b2376ea28b 100644
--- a/src/mesa/drivers/dri/radeon/radeon_dma.c
+++ b/src/mesa/drivers/dri/radeon/radeon_dma.c
@@ -271,6 +271,17 @@ void radeonFreeDmaRegions(radeonContextPtr rmesa)
 	}
 }
 
+void radeonReturnDmaRegion(radeonContextPtr rmesa, int return_bytes)
+{
+	if (is_empty_list(&rmesa->dma.reserved))
+		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 radeonReleaseDmaRegions(radeonContextPtr rmesa)
 {
 	struct radeon_dma_bo *dma_bo;
diff --git a/src/mesa/drivers/dri/radeon/radeon_dma.h b/src/mesa/drivers/dri/radeon/radeon_dma.h
index 21f956e321..74e653fd18 100644
--- a/src/mesa/drivers/dri/radeon/radeon_dma.h
+++ b/src/mesa/drivers/dri/radeon/radeon_dma.h
@@ -41,8 +41,10 @@ void radeonEmitVec16(uint32_t *out, const GLvoid * data, int stride, int count);
 void rcommon_emit_vector(GLcontext * ctx, struct radeon_aos *aos,
 			 const GLvoid * data, int size, int stride, int count);
 
+void radeonReturnDmaRegion(radeonContextPtr rmesa, int return_bytes);
 void radeonRefillCurrentDmaRegion(radeonContextPtr rmesa, int size);
 void radeon_init_dma(radeonContextPtr rmesa);
+void radeonReturnDmaRegion(radeonContextPtr rmesa, int return_bytes);
 void radeonAllocDmaRegion(radeonContextPtr rmesa,
 			  struct radeon_bo **pbo, int *poffset,
 			  int bytes, int alignment);