[Mesa-dev] [PATCH 2/5] i965: Initialize compaction tables once per process.

[Matt Turner]

On Wed, Nov 26, 2014 at 10:59 AM, Jason Ekstrand <jason at jlekstrand.net> wrote:
> On Wed, Nov 26, 2014 at 10:39 AM, Matt Turner <mattst88 at gmail.com> wrote:
>>
>> ---
>>  src/mesa/drivers/dri/i965/brw_eu_compact.c | 5 +++++
>>  1 file changed, 5 insertions(+)
>>
>> diff --git a/src/mesa/drivers/dri/i965/brw_eu_compact.c
>> b/src/mesa/drivers/dri/i965/brw_eu_compact.c
>> index 7117890..8e33bcb 100644
>> --- a/src/mesa/drivers/dri/i965/brw_eu_compact.c
>> +++ b/src/mesa/drivers/dri/i965/brw_eu_compact.c
>> @@ -75,6 +75,7 @@
>>  #include "brw_context.h"
>>  #include "brw_eu.h"
>>  #include "intel_asm_annotation.h"
>> +#include "util/u_atomic.h" /* for p_atomic_cmpxchg */
>>
>>  static const uint32_t g45_control_index_table[32] = {
>>     0b00000000000000000,
>> @@ -1247,6 +1248,10 @@ update_gen4_jump_count(struct brw_context *brw,
>> brw_inst *insn,
>>  void
>>  brw_init_compaction_tables(struct brw_context *brw)
>>  {
>> +   static bool initialized;
>> +   if (initialized || p_atomic_cmpxchg(&initialized, false, true) !=
>> false)
>> +      return;
>> +
>
>
> Sure, this protects the initialized flag, but what happens if a thread tries
> to use compaction after someone else starts initializing but before they've
> finished?  Same comment for the other two patches that do more-or-less the
> same thing.

Yes, that is theoretically possible.

But for it to happen two threads would have to arrive at
brw_init_compaction_tables() at the same time, one would get through
the initialized lock and the other wouldn't. The one that didn't would
then have to compile an entire shader with everything that entails,
run the generator, and then get to compacting the instructions before
the initializing thread executed the 27 instructions in this function.

Since the initializations in the other two patches are just single
variables (i.e., not a variable protecting a whole initialization
function) I don't think they even have the same theoretical problem.