[Mesa-stable] [Mesa-dev] [PATCH] i965: Fix GPU hangs on Broadwell in shaders with some control flow.

[Matt Turner]

On Thu, May 8, 2014 at 5:45 PM, Kenneth Graunke <kenneth at whitecape.org> wrote:
> On 05/08/2014 05:34 PM, Matt Turner wrote:
>> On Thu, May 8, 2014 at 4:44 PM, Kenneth Graunke <kenneth at whitecape.org> wrote:
>>> According to the documentation, we need to set the source 0 register
>>> type to IMM for flow control instructions that have both JIP and UIP.
>>>
>>> Fixes GPU hangs in approximately 10 Piglit tests, 5 es3conform tests,
>>> Unigine Crypt, a WebGL raytracer demo, and several Steam titles.
>>>
>>> Cc: "10.2" <mesa-stable at lists.freedesktop.org>
>>> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=75478
>>> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=75878
>>> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=76939
>>> Signed-off-by: Kenneth Graunke <kenneth at whitecape.org>
>>> ---
>>>  src/mesa/drivers/dri/i965/gen8_generator.cpp | 14 +++++++-------
>>>  1 file changed, 7 insertions(+), 7 deletions(-)
>>>
>>> diff --git a/src/mesa/drivers/dri/i965/gen8_generator.cpp b/src/mesa/drivers/dri/i965/gen8_generator.cpp
>>> index dd434a7..faca9c0 100644
>>> --- a/src/mesa/drivers/dri/i965/gen8_generator.cpp
>>> +++ b/src/mesa/drivers/dri/i965/gen8_generator.cpp
>>> @@ -422,6 +422,7 @@ gen8_generator::IF(unsigned predicate)
>>>  {
>>>     gen8_instruction *inst = next_inst(BRW_OPCODE_IF);
>>>     gen8_set_dst(brw, inst, vec1(retype(brw_null_reg(), BRW_REGISTER_TYPE_D)));
>>> +   gen8_set_src0(brw, inst, brw_imm_d(0));
>>>     gen8_set_exec_size(inst, default_state.exec_size);
>>>     gen8_set_pred_control(inst, predicate);
>>>     gen8_set_mask_control(inst, BRW_MASK_ENABLE);
>>> @@ -435,6 +436,7 @@ gen8_generator::ELSE()
>>>  {
>>>     gen8_instruction *inst = next_inst(BRW_OPCODE_ELSE);
>>>     gen8_set_dst(brw, inst, retype(brw_null_reg(), BRW_REGISTER_TYPE_D));
>>> +   gen8_set_src0(brw, inst, brw_imm_d(0));
>>>     gen8_set_mask_control(inst, BRW_MASK_ENABLE);
>>>     push_if_stack(inst);
>>>     return inst;
>>> @@ -456,6 +458,7 @@ gen8_generator::ENDIF()
>>>
>>>     gen8_instruction *endif_inst = next_inst(BRW_OPCODE_ENDIF);
>>>     gen8_set_mask_control(endif_inst, BRW_MASK_ENABLE);
>>> +   gen8_set_src0(brw, endif_inst, brw_imm_d(0));
>>
>> Else and endif don't have UIP, so they don't need this.
>>
>> With those hunks removed:
>>
>> Reviewed-by: Matt Turner <mattst88 at gmail.com>
>
> It's probably not necessary, but I figured consistency would be best.
> Would it be okay to leave it as is?

I think applying a work around when it's not needed is confusing.

But whatever.