r300: rework index buffer setup

Copy elements directly to DMA bo to get rid of one memcpy, and prepare for using VBOs for index buffer.

3 files changed, 126 insertions, 119 deletions

diff --git a/src/mesa/drivers/dri/r300/r300_context.h b/src/mesa/drivers/dri/r300/r300_context.h
index 09de898748..d620417422 100644
--- a/src/mesa/drivers/dri/r300/r300_context.h
+++ b/src/mesa/drivers/dri/r300/r300_context.h
@@ -498,9 +498,10 @@ struct r300_vertex_buffer {
 };
 
 struct r300_index_buffer {
-	GLvoid *ptr;
+	struct radeon_bo *bo;
+	int bo_offset;
+
 	GLboolean is_32bit;
-	GLboolean free_needed;
 	GLuint count;
 };
 
diff --git a/src/mesa/drivers/dri/r300/r300_draw.c b/src/mesa/drivers/dri/r300/r300_draw.c
index 99c73d27a2..1d6e6db773 100644
--- a/src/mesa/drivers/dri/r300/r300_draw.c
+++ b/src/mesa/drivers/dri/r300/r300_draw.c
@@ -47,32 +47,53 @@
 #include "swrast/swrast.h"
 #include "swrast_setup/swrast_setup.h"
 
+
+static int getTypeSize(GLenum type)
+{
+	switch (type) {
+		case GL_DOUBLE:
+			return sizeof(GLdouble);
+		case GL_FLOAT:
+			return sizeof(GLfloat);
+		case GL_INT:
+			return sizeof(GLint);
+		case GL_UNSIGNED_INT:
+			return sizeof(GLuint);
+		case GL_SHORT:
+			return sizeof(GLshort);
+		case GL_UNSIGNED_SHORT:
+			return sizeof(GLushort);
+		case GL_BYTE:
+			return sizeof(GLbyte);
+		case GL_UNSIGNED_BYTE:
+			return sizeof(GLubyte);
+		default:
+			assert(0);
+			return 0;
+	}
+}
+
 static void r300FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer *mesa_ind_buf)
 {
 	r300ContextPtr r300 = R300_CONTEXT(ctx);
-	struct r300_index_buffer *ind_buf = &r300->ind_buf;
 	GLvoid *src_ptr;
-	GLboolean mapped_bo = GL_FALSE;
+	GLuint *out;
+	int i;
 
-	if (!mesa_ind_buf) {
-		ind_buf->ptr = NULL;
-		return;
-	}
-
-	ind_buf->count = mesa_ind_buf->count;
 	if (mesa_ind_buf->obj->Name && !mesa_ind_buf->obj->Pointer) {
 		ctx->Driver.MapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER, GL_READ_ONLY_ARB, mesa_ind_buf->obj);
-		mapped_bo = GL_TRUE;
 		assert(mesa_ind_buf->obj->Pointer != NULL);
 	}
 	src_ptr = ADD_POINTERS(mesa_ind_buf->obj->Pointer, mesa_ind_buf->ptr);
 
 	if (mesa_ind_buf->type == GL_UNSIGNED_BYTE) {
+		GLuint size = sizeof(GLushort) * ((mesa_ind_buf->count + 1) & ~1);
 		GLubyte *in = (GLubyte *)src_ptr;
-		GLuint *out = _mesa_malloc(sizeof(GLushort) * ((mesa_ind_buf->count + 1) & ~1));
-		int i;
 
-		ind_buf->ptr = out;
+		radeonAllocDmaRegion(&r300->radeon, &r300->ind_buf.bo, &r300->ind_buf.bo_offset, size, 4);
+
+		assert(r300->ind_buf.bo->ptr != NULL);
+		out = (GLuint *)ADD_POINTERS(r300->ind_buf.bo->ptr, r300->ind_buf.bo_offset);
 
 		for (i = 0; i + 1 < mesa_ind_buf->count; i += 2) {
 			*out++ = in[i] | in[i + 1] << 16;
@@ -82,16 +103,15 @@ static void r300FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer
 			*out++ = in[i];
 		}
 
-		ind_buf->free_needed = GL_TRUE;
-		ind_buf->is_32bit = GL_FALSE;
-	} else if (mesa_ind_buf->type == GL_UNSIGNED_SHORT) {
 #if MESA_BIG_ENDIAN
+	} else { /* if (mesa_ind_buf->type == GL_UNSIGNED_SHORT) */
 		GLushort *in = (GLushort *)src_ptr;
-		GLuint *out = _mesa_malloc(sizeof(GLushort) *
-					   ((mesa_ind_buf->count + 1) & ~1));
-		int i;
+		size = sizeof(GLushort) * ((mesa_ind_buf->count + 1) & ~1);
+
+		radeonAllocDmaRegion(&r300->radeon, &r300->ind_buf.bo, &r300->ind_buf.bo_offet, size, 4);
 
-		ind_buf->ptr = out;
+		assert(r300->ind_buf.bo->ptr != NULL)
+		out = (GLuint *)ADD_POINTERS(r300->ind_buf.bo->ptr, r300->ind_buf.bo_offset);
 
 		for (i = 0; i + 1 < mesa_ind_buf->count; i += 2) {
 			*out++ = in[i] | in[i + 1] << 16;
@@ -100,46 +120,52 @@ static void r300FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer
 		if (i < mesa_ind_buf->count) {
 			*out++ = in[i];
 		}
-
-		ind_buf->free_needed = GL_TRUE;
-#else
-		ind_buf->ptr = src_ptr;
-		ind_buf->free_needed = GL_FALSE;
 #endif
-		ind_buf->is_32bit = GL_FALSE;
-	} else {
-		ind_buf->ptr = src_ptr;
-		ind_buf->free_needed = GL_FALSE;
-		ind_buf->is_32bit = GL_TRUE;
 	}
 
-	if (mapped_bo) {
-		ctx->Driver.UnmapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER, mesa_ind_buf->obj);
-	}
+	r300->ind_buf.is_32bit = GL_FALSE;
+	r300->ind_buf.count = mesa_ind_buf->count;
 }
 
-static int getTypeSize(GLenum type)
+
+static void r300SetupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer *mesa_ind_buf)
 {
-	switch (type) {
-		case GL_DOUBLE:
-			return sizeof(GLdouble);
-		case GL_FLOAT:
-			return sizeof(GLfloat);
-		case GL_INT:
-			return sizeof(GLint);
-		case GL_UNSIGNED_INT:
-			return sizeof(GLuint);
-		case GL_SHORT:
-			return sizeof(GLshort);
-		case GL_UNSIGNED_SHORT:
-			return sizeof(GLushort);
-		case GL_BYTE:
-			return sizeof(GLbyte);
-		case GL_UNSIGNED_BYTE:
-			return sizeof(GLubyte);
-		default:
-			assert(0);
-			return 0;
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	GLboolean mapped_named_bo = GL_FALSE;
+
+	if (!mesa_ind_buf) {
+		r300->ind_buf.bo = NULL;
+		return;
+	}
+
+#if MESA_BIG_ENDIAN
+	if (mesa_ind_buf->type == GL_UNSIGNED_INT) {
+#else
+	if (mesa_ind_buf->type != GL_UNSIGNED_BYTE) {
+#endif
+		const GLvoid *src_ptr;
+		GLvoid *dst_ptr;
+
+		if (mesa_ind_buf->obj->Name && !mesa_ind_buf->obj->Pointer) {
+			ctx->Driver.MapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER, GL_READ_ONLY_ARB, mesa_ind_buf->obj);
+			assert(mesa_ind_buf->obj->Pointer != NULL);
+			mapped_named_bo = GL_TRUE;
+		}
+
+		src_ptr = ADD_POINTERS(mesa_ind_buf->obj->Pointer, mesa_ind_buf->ptr);
+
+		const GLuint size = mesa_ind_buf->count * getTypeSize(mesa_ind_buf->type);
+
+		radeonAllocDmaRegion(&r300->radeon, &r300->ind_buf.bo, &r300->ind_buf.bo_offset, size, 4);
+
+		assert(r300->ind_buf.bo->ptr != NULL);
+		dst_ptr = ADD_POINTERS(r300->ind_buf.bo->ptr, r300->ind_buf.bo_offset);
+		_mesa_memcpy(dst_ptr, src_ptr, size);
+
+		r300->ind_buf.is_32bit = (mesa_ind_buf->type == GL_UNSIGNED_INT);
+		r300->ind_buf.count = mesa_ind_buf->count;
+	} else {
+		r300FixupIndexBuffer(ctx, mesa_ind_buf);
 	}
 }
 
@@ -473,13 +499,22 @@ static void r300SetVertexFormat(GLcontext *ctx, const struct gl_client_array *ar
 											  RADEON_GEM_DOMAIN_GTT, 0);
 			}
 		}
-
 		r300->radeon.tcl.aos_count = vbuf->num_attribs;
+
+		if (r300->ind_buf.bo) {
+			radeon_cs_space_check_with_bo(r300->radeon.cmdbuf.cs,
+										  r300->ind_buf.bo,
+										  RADEON_GEM_DOMAIN_GTT, 0);
+		}
 	}
 }
 
 static void r300FreeData(GLcontext *ctx)
 {
+	/* Need to zero tcl.aos[n].bo and tcl.elt_dma_bo
+	 * to prevent double unref in radeonReleaseArrays
+	 * called during context destroy
+	 */
 	r300ContextPtr r300 = R300_CONTEXT(ctx);
 	{
 		int i;
@@ -493,15 +528,9 @@ static void r300FreeData(GLcontext *ctx)
 	}
 
 	{
-		struct r300_index_buffer *ind_buf = &R300_CONTEXT(ctx)->ind_buf;
-		if (ind_buf->free_needed) {
-			_mesa_free(ind_buf->ptr);
-		}
-
-		if (r300->radeon.tcl.elt_dma_bo) {
-			radeon_bo_unref(r300->radeon.tcl.elt_dma_bo);
+		if (r300->ind_buf.bo != NULL) {
+			radeon_bo_unref(r300->ind_buf.bo);
 		}
-		r300->radeon.tcl.elt_dma_bo = NULL;
 	}
 }
 
@@ -526,7 +555,7 @@ static GLboolean r300TryDrawPrims(GLcontext *ctx,
 
 	r300SwitchFallback(ctx, R300_FALLBACK_INVALID_BUFFERS, !r300ValidateBuffers(ctx));
 
-	r300FixupIndexBuffer(ctx, ib);
+	r300SetupIndexBuffer(ctx, ib);
 
 	/* ensure we have the cmd buf space in advance to cover
  	 * the state + DMA AOS pointers */
diff --git a/src/mesa/drivers/dri/r300/r300_render.c b/src/mesa/drivers/dri/r300/r300_render.c
index 22b0d316cf..196cb47fef 100644
--- a/src/mesa/drivers/dri/r300/r300_render.c
+++ b/src/mesa/drivers/dri/r300/r300_render.c
@@ -172,64 +172,42 @@ int r300NumVerts(r300ContextPtr rmesa, int num_verts, int prim)
 	return num_verts - verts_off;
 }
 
-static void r300EmitElts(GLcontext * ctx, unsigned long n_elts)
-{
-	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	void *out;
-	GLuint size;
-
-	size = ((rmesa->ind_buf.is_32bit ? 4 : 2) * n_elts + 3) & ~3;
-
-	radeonAllocDmaRegion(&rmesa->radeon, &rmesa->radeon.tcl.elt_dma_bo,
-			     &rmesa->radeon.tcl.elt_dma_offset, size, 4);
-	radeon_bo_map(rmesa->radeon.tcl.elt_dma_bo, 1);
-	out = rmesa->radeon.tcl.elt_dma_bo->ptr + rmesa->radeon.tcl.elt_dma_offset;
-	memcpy(out, rmesa->ind_buf.ptr, size);
-	radeon_bo_unmap(rmesa->radeon.tcl.elt_dma_bo);
-}
-
 static void r300FireEB(r300ContextPtr rmesa, int vertex_count, int type)
 {
 	BATCH_LOCALS(&rmesa->radeon);
+	int size;
 
-    r300_emit_scissor(rmesa->radeon.glCtx);
-	if (vertex_count > 0) {
-		int size;
-
-		BEGIN_BATCH(10);
-		OUT_BATCH_PACKET3(R300_PACKET3_3D_DRAW_INDX_2, 0);
-		if (rmesa->ind_buf.is_32bit) {
-			size = vertex_count;
-			OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_INDICES |
-			  ((vertex_count + 0) << 16) | type |
-			  R300_VAP_VF_CNTL__INDEX_SIZE_32bit);
-		} else {
-			size = (vertex_count + 1) >> 1;
-			OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_INDICES |
-			   ((vertex_count + 0) << 16) | type);
-		}
+	r300_emit_scissor(rmesa->radeon.glCtx);
 
-		if (!rmesa->radeon.radeonScreen->kernel_mm) {
-			OUT_BATCH_PACKET3(R300_PACKET3_INDX_BUFFER, 2);
-			OUT_BATCH(R300_INDX_BUFFER_ONE_REG_WR | (0 << R300_INDX_BUFFER_SKIP_SHIFT) |
-	    			 (R300_VAP_PORT_IDX0 >> 2));
-			OUT_BATCH_RELOC(rmesa->radeon.tcl.elt_dma_offset,
-					rmesa->radeon.tcl.elt_dma_bo,
-					rmesa->radeon.tcl.elt_dma_offset,
-					RADEON_GEM_DOMAIN_GTT, 0, 0);
-			OUT_BATCH(size);
-		} else {
-			OUT_BATCH_PACKET3(R300_PACKET3_INDX_BUFFER, 2);
-			OUT_BATCH(R300_INDX_BUFFER_ONE_REG_WR | (0 << R300_INDX_BUFFER_SKIP_SHIFT) |
-	    			 (R300_VAP_PORT_IDX0 >> 2));
-			OUT_BATCH(rmesa->radeon.tcl.elt_dma_offset);
-			OUT_BATCH(size);
-			radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs,
-					      rmesa->radeon.tcl.elt_dma_bo,
-					      RADEON_GEM_DOMAIN_GTT, 0, 0);
-		}
-		END_BATCH();
+	BEGIN_BATCH(10);
+	OUT_BATCH_PACKET3(R300_PACKET3_3D_DRAW_INDX_2, 0);
+	if (rmesa->ind_buf.is_32bit) {
+		size = vertex_count;
+		OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_INDICES |
+		  (vertex_count << 16) | type |
+		  R300_VAP_VF_CNTL__INDEX_SIZE_32bit);
+	} else {
+		size = (vertex_count + 1) >> 1;
+		OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_INDICES |
+		   (vertex_count << 16) | type);
+	}
+
+	if (!rmesa->radeon.radeonScreen->kernel_mm) {
+		OUT_BATCH_PACKET3(R300_PACKET3_INDX_BUFFER, 2);
+		OUT_BATCH(R300_INDX_BUFFER_ONE_REG_WR | (0 << R300_INDX_BUFFER_SKIP_SHIFT) |
+				 (R300_VAP_PORT_IDX0 >> 2));
+		OUT_BATCH_RELOC(0, rmesa->ind_buf.bo, rmesa->ind_buf.bo_offset, RADEON_GEM_DOMAIN_GTT, 0, 0);
+		OUT_BATCH(size);
+	} else {
+		OUT_BATCH_PACKET3(R300_PACKET3_INDX_BUFFER, 2);
+		OUT_BATCH(R300_INDX_BUFFER_ONE_REG_WR | (0 << R300_INDX_BUFFER_SKIP_SHIFT) |
+				 (R300_VAP_PORT_IDX0 >> 2));
+		OUT_BATCH(rmesa->ind_buf.bo_offset);
+		OUT_BATCH(size);
+		radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs,
+				      rmesa->ind_buf.bo, RADEON_GEM_DOMAIN_GTT, 0, 0);
 	}
+	END_BATCH();
 }
 
 static void r300EmitAOS(r300ContextPtr rmesa, GLuint nr, GLuint offset)
@@ -365,8 +343,7 @@ void r300RunRenderPrimitive(GLcontext * ctx, int start, int end, int prim)
 	 */
 	rcommonEnsureCmdBufSpace(&rmesa->radeon, 128, __FUNCTION__);
 
-	if (rmesa->ind_buf.ptr) {
-		r300EmitElts(ctx, num_verts);
+	if (rmesa->ind_buf.bo) {
 		r300EmitAOS(rmesa, rmesa->radeon.tcl.aos_count, 0);
 		if (rmesa->radeon.radeonScreen->kernel_mm) {
 			BEGIN_BATCH_NO_AUTOSTATE(2);