intel: combine short memcpy using a temporary allocated buffer

Using a temporary buffer for large discontiguous uploads into the common
buffer and a single buffered upload is faster than performing the
discontiguous copies through a mapping into the GTT.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

3 files changed, 27 insertions, 38 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_draw_upload.c b/src/mesa/drivers/dri/i965/brw_draw_upload.c
index 8952e97674..62e43e83bb 100644
--- a/src/mesa/drivers/dri/i965/brw_draw_upload.c
+++ b/src/mesa/drivers/dri/i965/brw_draw_upload.c
@@ -253,8 +253,7 @@ copy_array_to_vbo_array( struct brw_context *brw,
 			&buffer->bo, &buffer->offset);
    } else {
       const unsigned char *src = element->glarray->Ptr;
-      char *dst = intel_upload_map(&brw->intel, size,
-				   &buffer->bo, &buffer->offset);
+      char *dst = intel_upload_map(&brw->intel, size);
       int i;
 
       for (i = 0; i < element->count; i++) {
@@ -262,6 +261,8 @@ copy_array_to_vbo_array( struct brw_context *brw,
 	 src += element->glarray->StrideB;
 	 dst += dst_stride;
       }
+      intel_upload_unmap(&brw->intel, dst, size,
+			 &buffer->bo, &buffer->offset);
    }
 }
 
@@ -421,9 +422,7 @@ static void brw_prepare_vertices(struct brw_context *brw)
 	 int count = upload[0]->count, offset;
 	 char *map;
 
-	 map = intel_upload_map(&brw->intel, total_size * count,
-				&buffer->bo, &buffer->offset);
-
+	 map = intel_upload_map(&brw->intel, total_size * count);
 	 for (i = offset = 0; i < nr_uploads; i++) {
 	    const unsigned char *src = upload[i]->glarray->Ptr;
 	    int size = upload[i]->element_size;
@@ -442,6 +441,8 @@ static void brw_prepare_vertices(struct brw_context *brw)
 
 	    offset += size;
 	 }
+	 intel_upload_unmap(&brw->intel, map, total_size * count,
+			    &buffer->bo, &buffer->offset);
 	 buffer->stride = offset;
 	 j++;
 
diff --git a/src/mesa/drivers/dri/intel/intel_buffer_objects.c b/src/mesa/drivers/dri/intel/intel_buffer_objects.c
index 4edf29ec75..9d54f47fea 100644
--- a/src/mesa/drivers/dri/intel/intel_buffer_objects.c
+++ b/src/mesa/drivers/dri/intel/intel_buffer_objects.c
@@ -619,7 +619,7 @@ void intel_upload_data(struct intel_context *intel,
    *return_offset = intel->upload.offset;
 
    if (intel->upload.buffer_len &&
-       intel->upload.buffer_len + asize > sizeof(intel->upload.buffer))
+       intel->upload.buffer_len + size > sizeof(intel->upload.buffer))
    {
       drm_intel_bo_subdata(intel->upload.bo,
 			   intel->upload.buffer_offset,
@@ -646,25 +646,12 @@ void intel_upload_data(struct intel_context *intel,
    intel->upload.offset += asize;
 }
 
-void *intel_upload_map(struct intel_context *intel,
-		       GLuint size,
-		       drm_intel_bo **return_bo,
-		       GLuint *return_offset)
+void *intel_upload_map(struct intel_context *intel, GLuint size)
 {
-   GLuint asize = ALIGN(size, 64);
    char *ptr;
 
-   if (intel->upload.bo == NULL ||
-       intel->upload.offset + size > intel->upload.bo->size) {
-      wrap_buffers(intel, size);
-   }
-
-   drm_intel_bo_reference(intel->upload.bo);
-   *return_bo = intel->upload.bo;
-   *return_offset = intel->upload.offset;
-
    if (intel->upload.buffer_len &&
-       intel->upload.buffer_len + asize > sizeof(intel->upload.buffer))
+       intel->upload.buffer_len + size > sizeof(intel->upload.buffer))
    {
       drm_intel_bo_subdata(intel->upload.bo,
 			   intel->upload.buffer_offset,
@@ -673,26 +660,25 @@ void *intel_upload_map(struct intel_context *intel,
       intel->upload.buffer_len = 0;
    }
 
-   if (size < sizeof(intel->upload.buffer))
-   {
-      if (intel->upload.buffer_len == 0)
-	 intel->upload.buffer_offset = intel->upload.offset;
-
+   if (size <= sizeof(intel->upload.buffer))
       ptr = intel->upload.buffer + intel->upload.buffer_len;
-      intel->upload.buffer_len += asize;
-   }
    else
-   {
-      drm_intel_gem_bo_map_gtt(intel->upload.bo);
-      ptr = intel->upload.bo->virtual;
-      ptr += intel->upload.offset;
-   }
-
-   intel->upload.offset += asize;
+      ptr = malloc(size);
 
    return ptr;
 }
 
+void intel_upload_unmap(struct intel_context *intel,
+			const void *ptr, GLuint size,
+			drm_intel_bo **return_bo,
+			GLuint *return_offset)
+{
+   intel_upload_data(intel, ptr, size, return_bo, return_offset);
+
+   if (size > sizeof(intel->upload.buffer))
+      free((void*)ptr);
+}
+
 drm_intel_bo *
 intel_bufferobj_source(struct intel_context *intel,
                        struct intel_buffer_object *intel_obj,
diff --git a/src/mesa/drivers/dri/intel/intel_buffer_objects.h b/src/mesa/drivers/dri/intel/intel_buffer_objects.h
index 4b66bf7d59..2d86becdf3 100644
--- a/src/mesa/drivers/dri/intel/intel_buffer_objects.h
+++ b/src/mesa/drivers/dri/intel/intel_buffer_objects.h
@@ -76,9 +76,11 @@ void intel_upload_data(struct intel_context *intel,
 		       GLuint *return_offset);
 
 void *intel_upload_map(struct intel_context *intel,
-		       GLuint size,
-		       drm_intel_bo **return_bo,
-		       GLuint *return_offset);
+		       GLuint size);
+void intel_upload_unmap(struct intel_context *intel,
+			const void *ptr, GLuint size,
+			drm_intel_bo **return_bo,
+			GLuint *return_offset);
 
 void intel_upload_finish(struct intel_context *intel);