intel: Use specified alignment for writes into the upload buffer

Signed-off-by: Chris Wilson <[email protected]>

3 files changed, 57 insertions, 30 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_draw_upload.c b/src/mesa/drivers/dri/i965/brw_draw_upload.c
index daae136a143..d4841226bfa 100644
--- a/src/mesa/drivers/dri/i965/brw_draw_upload.c
+++ b/src/mesa/drivers/dri/i965/brw_draw_upload.c
@@ -249,11 +249,11 @@ copy_array_to_vbo_array( struct brw_context *brw,
 
    buffer->stride = dst_stride;
    if (dst_stride == element->glarray->StrideB) {
-      intel_upload_data(&brw->intel, element->glarray->Ptr, size,
+      intel_upload_data(&brw->intel, element->glarray->Ptr, size, dst_stride,
 			&buffer->bo, &buffer->offset);
    } else {
       const unsigned char *src = element->glarray->Ptr;
-      char *dst = intel_upload_map(&brw->intel, size);
+      char *dst = intel_upload_map(&brw->intel, size, dst_stride);
       int i;
 
       for (i = 0; i < element->count; i++) {
@@ -261,7 +261,7 @@ copy_array_to_vbo_array( struct brw_context *brw,
 	 src += element->glarray->StrideB;
 	 dst += dst_stride;
       }
-      intel_upload_unmap(&brw->intel, dst, size,
+      intel_upload_unmap(&brw->intel, dst, size, dst_stride,
 			 &buffer->bo, &buffer->offset);
    }
 }
@@ -421,7 +421,7 @@ 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);
+	 map = intel_upload_map(&brw->intel, total_size * count, total_size);
 	 for (i = offset = 0; i < nr_uploads; i++) {
 	    const unsigned char *src = upload[i]->glarray->Ptr;
 	    int size = upload[i]->element_size;
@@ -440,7 +440,7 @@ static void brw_prepare_vertices(struct brw_context *brw)
 
 	    offset += size;
 	 }
-	 intel_upload_unmap(&brw->intel, map, total_size * count,
+	 intel_upload_unmap(&brw->intel, map, total_size * count, total_size,
 			    &buffer->bo, &buffer->offset);
 	 buffer->stride = offset;
 	 j++;
@@ -608,7 +608,8 @@ static void brw_prepare_indices(struct brw_context *brw)
 
       /* Get new bufferobj, offset:
        */
-      intel_upload_data(&brw->intel, index_buffer->ptr, ib_size, &bo, &offset);
+      intel_upload_data(&brw->intel, index_buffer->ptr, ib_size, ib_type_size,
+			&bo, &offset);
       brw->ib.start_vertex_offset = offset / ib_type_size;
       offset = 0;
    } else {
@@ -624,7 +625,8 @@ static void brw_prepare_indices(struct brw_context *brw)
                                                 bufferobj);
            map += offset;
 
-	   intel_upload_data(&brw->intel, map, ib_size, &bo, &offset);
+	   intel_upload_data(&brw->intel, map, ib_size, ib_type_size,
+			     &bo, &offset);
 	   brw->ib.start_vertex_offset = offset / ib_type_size;
 	   offset = 0;
 
diff --git a/src/mesa/drivers/dri/intel/intel_buffer_objects.c b/src/mesa/drivers/dri/intel/intel_buffer_objects.c
index 9d54f47fead..81afe170c1b 100644
--- a/src/mesa/drivers/dri/intel/intel_buffer_objects.c
+++ b/src/mesa/drivers/dri/intel/intel_buffer_objects.c
@@ -603,23 +603,25 @@ static void wrap_buffers(struct intel_context *intel, GLuint size)
 }
 
 void intel_upload_data(struct intel_context *intel,
-		       const void *ptr, GLuint size,
+		       const void *ptr, GLuint size, GLuint align,
 		       drm_intel_bo **return_bo,
 		       GLuint *return_offset)
 {
-   GLuint asize = ALIGN(size, 64);
+   GLuint base, delta;
 
-   if (intel->upload.bo == NULL ||
-       intel->upload.offset + size > intel->upload.bo->size) {
+   base = (intel->upload.offset + align - 1) / align * align;
+   if (intel->upload.bo == NULL || base + size > intel->upload.bo->size) {
       wrap_buffers(intel, size);
+      base = 0;
    }
 
    drm_intel_bo_reference(intel->upload.bo);
    *return_bo = intel->upload.bo;
-   *return_offset = intel->upload.offset;
+   *return_offset = base;
 
+   delta = base - intel->upload.offset;
    if (intel->upload.buffer_len &&
-       intel->upload.buffer_len + size > sizeof(intel->upload.buffer))
+       intel->upload.buffer_len + delta + size > sizeof(intel->upload.buffer))
    {
       drm_intel_bo_subdata(intel->upload.bo,
 			   intel->upload.buffer_offset,
@@ -631,27 +633,35 @@ void intel_upload_data(struct intel_context *intel,
    if (size < sizeof(intel->upload.buffer))
    {
       if (intel->upload.buffer_len == 0)
-	 intel->upload.buffer_offset = intel->upload.offset;
+	 intel->upload.buffer_offset = base;
+      else
+	 intel->upload.buffer_len += delta;
 
       memcpy(intel->upload.buffer + intel->upload.buffer_len, ptr, size);
-      intel->upload.buffer_len += asize;
+      intel->upload.buffer_len += size;
    }
    else
    {
-      drm_intel_bo_subdata(intel->upload.bo,
-			   intel->upload.offset,
-			   size, ptr);
+      drm_intel_bo_subdata(intel->upload.bo, base, size, ptr);
    }
 
-   intel->upload.offset += asize;
+   intel->upload.offset = base + size;
 }
 
-void *intel_upload_map(struct intel_context *intel, GLuint size)
+void *intel_upload_map(struct intel_context *intel, GLuint size, GLuint align)
 {
+   GLuint base, delta;
    char *ptr;
 
+   base = (intel->upload.offset + align - 1) / align * align;
+   if (intel->upload.bo == NULL || base + size > intel->upload.bo->size) {
+      wrap_buffers(intel, size);
+      base = 0;
+   }
+
+   delta = base - intel->upload.offset;
    if (intel->upload.buffer_len &&
-       intel->upload.buffer_len + size > sizeof(intel->upload.buffer))
+       intel->upload.buffer_len + delta + size > sizeof(intel->upload.buffer))
    {
       drm_intel_bo_subdata(intel->upload.bo,
 			   intel->upload.buffer_offset,
@@ -660,23 +670,38 @@ void *intel_upload_map(struct intel_context *intel, GLuint size)
       intel->upload.buffer_len = 0;
    }
 
-   if (size <= sizeof(intel->upload.buffer))
+   if (size <= sizeof(intel->upload.buffer)) {
+      if (intel->upload.buffer_len == 0)
+	 intel->upload.buffer_offset = base;
+      else
+	 intel->upload.buffer_len += delta;
+
       ptr = intel->upload.buffer + intel->upload.buffer_len;
-   else
+      intel->upload.buffer_len += size;
+   } else
       ptr = malloc(size);
 
    return ptr;
 }
 
 void intel_upload_unmap(struct intel_context *intel,
-			const void *ptr, GLuint size,
+			const void *ptr, GLuint size, GLuint align,
 			drm_intel_bo **return_bo,
 			GLuint *return_offset)
 {
-   intel_upload_data(intel, ptr, size, return_bo, return_offset);
+   GLuint base;
 
-   if (size > sizeof(intel->upload.buffer))
+   base = (intel->upload.offset + align - 1) / align * align;
+   if (size > sizeof(intel->upload.buffer)) {
+      drm_intel_bo_subdata(intel->upload.bo, base, size, ptr);
       free((void*)ptr);
+   }
+
+   drm_intel_bo_reference(intel->upload.bo);
+   *return_bo = intel->upload.bo;
+   *return_offset = base;
+
+   intel->upload.offset = base + size;
 }
 
 drm_intel_bo *
@@ -686,7 +711,7 @@ intel_bufferobj_source(struct intel_context *intel,
 {
    if (intel_obj->buffer == NULL) {
       intel_upload_data(intel,
-			intel_obj->sys_buffer, intel_obj->Base.Size,
+			intel_obj->sys_buffer, intel_obj->Base.Size, 64,
 			&intel_obj->buffer, &intel_obj->offset);
       intel_obj->source = 1;
    }
diff --git a/src/mesa/drivers/dri/intel/intel_buffer_objects.h b/src/mesa/drivers/dri/intel/intel_buffer_objects.h
index 2d86becdf3e..3ec3a521382 100644
--- a/src/mesa/drivers/dri/intel/intel_buffer_objects.h
+++ b/src/mesa/drivers/dri/intel/intel_buffer_objects.h
@@ -71,14 +71,14 @@ drm_intel_bo *intel_bufferobj_source(struct intel_context *intel,
 				     GLuint *offset);
 
 void intel_upload_data(struct intel_context *intel,
-		       const void *ptr, GLuint size,
+		       const void *ptr, GLuint size, GLuint align,
 		       drm_intel_bo **return_bo,
 		       GLuint *return_offset);
 
 void *intel_upload_map(struct intel_context *intel,
-		       GLuint size);
+		       GLuint size, GLuint align);
 void intel_upload_unmap(struct intel_context *intel,
-			const void *ptr, GLuint size,
+			const void *ptr, GLuint size, GLuint align,
 			drm_intel_bo **return_bo,
 			GLuint *return_offset);