[Mesa-dev] [PATCH 14/17] Modified i915 intel_bufferobj_data to use _mesa_align_malloc instead of malloc.

[Ian Romanick]

On 11/24/2013 11:17 PM, Chris Forbes wrote:
> I'm not convinced that just using _mesa_align_malloc() is sufficient
> here. Issue 2 from the ARB_map_buffer_alignment spec says:
> 
>     2) What alignment should be used for MapBufferRange when <offset> is
>     non-zero?
> 
>     RESOLVED:  The pointer returned should be consistent with the base of the
>     buffer being aligned.  For example, if a buffer was mapped with an
>     <offset> of 24 bytes on an implementation with a
>     MIN_MAP_BUFFER_ALIGNMENT of 64, the value of the pointer returned
>     would have a value of the form 64*<N>+24 for some integer <N>.
> 
> Cc'ing Ian, since I assume this is based on his instructions.

Yeah... Ken told me that Eric mentioned this issue when Ken sent out
some patches to enable this extension just for i965.  There are a couple
ways to solve this, I think.

One way would be to modify MapBuffer and friends to only request aligned
offsets from the driver and do some pointer fidgeting.  For drivers that
advertise small alignments, that should be fine... for drivers that
advertise large alignments this could be a problem.

Another way would be some similar sorts of fudging in each driver.  The
MapBufferRange handler in the driver would have to do an aligned
allocation but only copy the necessary bits.

I don't have a strong opinion either way about the implementation...
though I would like to see some tests that could tickle this bug, if it
existed.  AFAIK, we tend to return a direct mapping, so an obvious test
of glMapBufferRange(target, 7, size, access) probably won't hit it.

> -- Chris
> 
> 
> On Mon, Nov 25, 2013 at 6:36 PM, Siavash Eliasi <siavashserver at gmail.com> wrote:
>> ---
>>  src/mesa/drivers/dri/i915/intel_buffer_objects.c | 4 ++--
>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/src/mesa/drivers/dri/i915/intel_buffer_objects.c b/src/mesa/drivers/dri/i915/intel_buffer_objects.c
>> index bc58c70..4fbf954 100644
>> --- a/src/mesa/drivers/dri/i915/intel_buffer_objects.c
>> +++ b/src/mesa/drivers/dri/i915/intel_buffer_objects.c
>> @@ -137,7 +137,7 @@ intel_bufferobj_data(struct gl_context * ctx,
>>         * contents anyway.
>>         */
>>        if (target == GL_ARRAY_BUFFER || target == GL_ELEMENT_ARRAY_BUFFER) {
>> -        intel_obj->sys_buffer = malloc(size);
>> +        intel_obj->sys_buffer = _mesa_align_malloc(size, ctx->Const.MinMapBufferAlignment);
>>          if (intel_obj->sys_buffer != NULL) {
>>             if (data != NULL)
>>                memcpy(intel_obj->sys_buffer, data, size);
>> @@ -337,7 +337,7 @@ intel_bufferobj_map_range(struct gl_context * ctx,
>>     if ((access & GL_MAP_INVALIDATE_RANGE_BIT) &&
>>         drm_intel_bo_busy(intel_obj->buffer)) {
>>        if (access & GL_MAP_FLUSH_EXPLICIT_BIT) {
>> -        intel_obj->range_map_buffer = malloc(length);
>> +        intel_obj->range_map_buffer = _mesa_align_malloc(length, ctx->Const.MinMapBufferAlignment);
>>          obj->Pointer = intel_obj->range_map_buffer;
>>        } else {
>>          intel_obj->range_map_bo = drm_intel_bo_alloc(intel->bufmgr,
>> --
>> 1.8.4.2
>>
>> _______________________________________________
>> mesa-dev mailing list
>> mesa-dev at lists.freedesktop.org
>> http://lists.freedesktop.org/mailman/listinfo/mesa-dev