Re: [Mesa-dev] [PATCH v2 3/5] anv: Add support for the PMA fix on Broadwell

[Nanley Chery]

On Thu, Feb 09, 2017 at 11:38:31AM -0800, Jason Ekstrand wrote:
> On Thu, Feb 9, 2017 at 10:33 AM, Nanley Chery  wrote:
> 
> > On Wed, Feb 08, 2017 at 06:27:52PM -0800, Jason Ekstrand wrote:
> > > On Wed, Feb 8, 2017 at 5:34 PM, Nanley Chery 
> > wrote:
> > >
> > > > On Thu, Feb 02, 2017 at 01:26:05PM -0800, Jason Ekstrand wrote:
> > > > > In order to get good performance numbers for this, I had to hack up
> > the
> > > > > driver to whack wm_prog_data::uses_kill to true to emulate a discard
> > and
> > > > > used the Sascha "shadowmapping" demo.  Setting uses_kill to true
> > dropped
> > > > > the framerate on the demo by 25-30%.  Enabling the PMA fix brought it
> > > > > back up to around 90% of the original framerate.  This doesn't seem
> > to
> > > > > really impact Dota 2;  probably because it doesn't use 16-bit depth.
> > > > >
> > > > > Reviewed-by: Lionel Landwerlin 
> > > > > ---
> > > > >  src/intel/vulkan/TODO  |   1 -
> > > > >  src/intel/vulkan/anv_cmd_buffer.c  |   2 +
> > > > >  src/intel/vulkan/anv_genX.h|   3 +
> > > > >  src/intel/vulkan/anv_private.h |  17 +
> > > > >  src/intel/vulkan/gen7_cmd_buffer.c |   7 ++
> > > > >  src/intel/vulkan/gen8_cmd_buffer.c | 133
> > ++
> > > > +++
> > > > >  src/intel/vulkan/genX_blorp_exec.c |   5 ++
> > > > >  src/intel/vulkan/genX_cmd_buffer.c |  15 -
> > > > >  src/intel/vulkan/genX_pipeline.c   |  38 +++
> > > > >  9 files changed, 219 insertions(+), 2 deletions(-)
> > > > >
> > > > > diff --git a/src/intel/vulkan/TODO b/src/intel/vulkan/TODO
> > > > > index 38acc0d..f8b73a1 100644
> > > > > --- a/src/intel/vulkan/TODO
> > > > > +++ b/src/intel/vulkan/TODO
> > > > > @@ -12,5 +12,4 @@ Performance:
> > > > >   - Compressed multisample support
> > > > >   - Pushing pieces of UBOs?
> > > > >   - Enable guardband clipping
> > > > > - - pma stall workaround
> > > > >   - Use soft-pin to avoid relocations
> > > > > diff --git a/src/intel/vulkan/anv_cmd_buffer.c
> > > > b/src/intel/vulkan/anv_cmd_buffer.c
> > > > > index 5886fa6..8c08f8d 100644
> > > > > --- a/src/intel/vulkan/anv_cmd_buffer.c
> > > > > +++ b/src/intel/vulkan/anv_cmd_buffer.c
> > > > > @@ -135,6 +135,8 @@ anv_cmd_state_reset(struct anv_cmd_buffer
> > > > *cmd_buffer)
> > > > > state->restart_index = UINT32_MAX;
> > > > > state->dynamic = default_dynamic_state;
> > > > > state->need_query_wa = true;
> > > > > +   state->pma_fix_enabled = false;
> > > > > +   state->hiz_enabled = false;
> > > > >
> > > > > if (state->attachments != NULL) {
> > > > >vk_free(_buffer->pool->alloc, state->attachments);
> > > > > diff --git a/src/intel/vulkan/anv_genX.h
> > b/src/intel/vulkan/anv_genX.h
> > > > > index d04fe38..67147b0 100644
> > > > > --- a/src/intel/vulkan/anv_genX.h
> > > > > +++ b/src/intel/vulkan/anv_genX.h
> > > > > @@ -55,6 +55,9 @@ void genX(cmd_buffer_flush_dynamic_state)(struct
> > > > anv_cmd_buffer *cmd_buffer);
> > > > >
> > > > >  void genX(cmd_buffer_flush_compute_state)(struct anv_cmd_buffer
> > > > *cmd_buffer);
> > > > >
> > > > > +void genX(cmd_buffer_enable_pma_fix)(struct anv_cmd_buffer
> > *cmd_buffer,
> > > > > + bool enable);
> > > > > +
> > > > >  void
> > > > >  genX(emit_urb_setup)(struct anv_device *device, struct anv_batch
> > *batch,
> > > > >   const struct gen_l3_config *l3_config,
> > > > > diff --git a/src/intel/vulkan/anv_private.h b/src/intel/vulkan/anv_
> > > > private.h
> > > > > index 4fe3ebc..6efe4ea 100644
> > > > > --- a/src/intel/vulkan/anv_private.h
> > > > > +++ b/src/intel/vulkan/anv_private.h
> > > > > @@ -1163,6 +1163,20 @@ struct anv_cmd_state {
> > > > > bool need_query_wa;
> > > > >
> > > > > /**
> > > > > +* Whether or not the gen8 PMA fix is enabled.  We ensure that,
> > at
> > > > the top
> > > > > +* of any command buffer it disabled by disabling it in
> > > > EndCommandBuffer
> > > >  ^
> > > >  is?
> > > >
> >
> > Fixed?
> >
> 
> done
> 
> 
> > > > > +* and before invoking the secondary in ExecuteCommands.
> > > > > +*/
> > > > > +   bool pma_fix_enabled;
> > > > > +
> > > > > +   /**
> > > > > +* Whether or not we now for certain that HiZ is enabled for the
> > > > current
> > > >^
> > > >know
> > > >
> >
> > Fixed?
> >
> 
> done
> 
> 
> > > > > +* subpass.  If, for whatever reason, we are unsure as to whether
> > > > HiZ is
> > > > > +* enabled or not, this will be false.
> > > > > +*/
> > > > > +   bool hiz_enabled;
> > > > > +
> > > > > +   /**
> > > > >  * Array length is anv_cmd_state::pass::attachment_count. Array
> > > > content 

Re: [Mesa-dev] [PATCH v2 3/5] anv: Add support for the PMA fix on Broadwell

[Jason Ekstrand]

On Thu, Feb 9, 2017 at 10:33 AM, Nanley Chery  wrote:

> On Wed, Feb 08, 2017 at 06:27:52PM -0800, Jason Ekstrand wrote:
> > On Wed, Feb 8, 2017 at 5:34 PM, Nanley Chery 
> wrote:
> >
> > > On Thu, Feb 02, 2017 at 01:26:05PM -0800, Jason Ekstrand wrote:
> > > > In order to get good performance numbers for this, I had to hack up
> the
> > > > driver to whack wm_prog_data::uses_kill to true to emulate a discard
> and
> > > > used the Sascha "shadowmapping" demo.  Setting uses_kill to true
> dropped
> > > > the framerate on the demo by 25-30%.  Enabling the PMA fix brought it
> > > > back up to around 90% of the original framerate.  This doesn't seem
> to
> > > > really impact Dota 2;  probably because it doesn't use 16-bit depth.
> > > >
> > > > Reviewed-by: Lionel Landwerlin 
> > > > ---
> > > >  src/intel/vulkan/TODO  |   1 -
> > > >  src/intel/vulkan/anv_cmd_buffer.c  |   2 +
> > > >  src/intel/vulkan/anv_genX.h|   3 +
> > > >  src/intel/vulkan/anv_private.h |  17 +
> > > >  src/intel/vulkan/gen7_cmd_buffer.c |   7 ++
> > > >  src/intel/vulkan/gen8_cmd_buffer.c | 133
> ++
> > > +++
> > > >  src/intel/vulkan/genX_blorp_exec.c |   5 ++
> > > >  src/intel/vulkan/genX_cmd_buffer.c |  15 -
> > > >  src/intel/vulkan/genX_pipeline.c   |  38 +++
> > > >  9 files changed, 219 insertions(+), 2 deletions(-)
> > > >
> > > > diff --git a/src/intel/vulkan/TODO b/src/intel/vulkan/TODO
> > > > index 38acc0d..f8b73a1 100644
> > > > --- a/src/intel/vulkan/TODO
> > > > +++ b/src/intel/vulkan/TODO
> > > > @@ -12,5 +12,4 @@ Performance:
> > > >   - Compressed multisample support
> > > >   - Pushing pieces of UBOs?
> > > >   - Enable guardband clipping
> > > > - - pma stall workaround
> > > >   - Use soft-pin to avoid relocations
> > > > diff --git a/src/intel/vulkan/anv_cmd_buffer.c
> > > b/src/intel/vulkan/anv_cmd_buffer.c
> > > > index 5886fa6..8c08f8d 100644
> > > > --- a/src/intel/vulkan/anv_cmd_buffer.c
> > > > +++ b/src/intel/vulkan/anv_cmd_buffer.c
> > > > @@ -135,6 +135,8 @@ anv_cmd_state_reset(struct anv_cmd_buffer
> > > *cmd_buffer)
> > > > state->restart_index = UINT32_MAX;
> > > > state->dynamic = default_dynamic_state;
> > > > state->need_query_wa = true;
> > > > +   state->pma_fix_enabled = false;
> > > > +   state->hiz_enabled = false;
> > > >
> > > > if (state->attachments != NULL) {
> > > >vk_free(_buffer->pool->alloc, state->attachments);
> > > > diff --git a/src/intel/vulkan/anv_genX.h
> b/src/intel/vulkan/anv_genX.h
> > > > index d04fe38..67147b0 100644
> > > > --- a/src/intel/vulkan/anv_genX.h
> > > > +++ b/src/intel/vulkan/anv_genX.h
> > > > @@ -55,6 +55,9 @@ void genX(cmd_buffer_flush_dynamic_state)(struct
> > > anv_cmd_buffer *cmd_buffer);
> > > >
> > > >  void genX(cmd_buffer_flush_compute_state)(struct anv_cmd_buffer
> > > *cmd_buffer);
> > > >
> > > > +void genX(cmd_buffer_enable_pma_fix)(struct anv_cmd_buffer
> *cmd_buffer,
> > > > + bool enable);
> > > > +
> > > >  void
> > > >  genX(emit_urb_setup)(struct anv_device *device, struct anv_batch
> *batch,
> > > >   const struct gen_l3_config *l3_config,
> > > > diff --git a/src/intel/vulkan/anv_private.h b/src/intel/vulkan/anv_
> > > private.h
> > > > index 4fe3ebc..6efe4ea 100644
> > > > --- a/src/intel/vulkan/anv_private.h
> > > > +++ b/src/intel/vulkan/anv_private.h
> > > > @@ -1163,6 +1163,20 @@ struct anv_cmd_state {
> > > > bool need_query_wa;
> > > >
> > > > /**
> > > > +* Whether or not the gen8 PMA fix is enabled.  We ensure that,
> at
> > > the top
> > > > +* of any command buffer it disabled by disabling it in
> > > EndCommandBuffer
> > >  ^
> > >  is?
> > >
>
> Fixed?
>

done


> > > > +* and before invoking the secondary in ExecuteCommands.
> > > > +*/
> > > > +   bool pma_fix_enabled;
> > > > +
> > > > +   /**
> > > > +* Whether or not we now for certain that HiZ is enabled for the
> > > current
> > >^
> > >know
> > >
>
> Fixed?
>

done


> > > > +* subpass.  If, for whatever reason, we are unsure as to whether
> > > HiZ is
> > > > +* enabled or not, this will be false.
> > > > +*/
> > > > +   bool hiz_enabled;
> > > > +
> > > > +   /**
> > > >  * Array length is anv_cmd_state::pass::attachment_count. Array
> > > content is
> > > >  * valid only when recording a render pass instance.
> > > >  */
> > > > @@ -1465,8 +1479,11 @@ struct anv_pipeline {
> > > >
> > > > uint32_t cs_right_mask;
> > > >
> > > > +   bool writes_depth;
> > > > +   bool  

Re: [Mesa-dev] [PATCH v2 3/5] anv: Add support for the PMA fix on Broadwell

[Nanley Chery]

On Wed, Feb 08, 2017 at 09:38:52PM -0800, Jason Ekstrand wrote:
> On Wed, Feb 8, 2017 at 8:11 PM, Jason Ekstrand  wrote:
> 
> > On Wed, Feb 8, 2017 at 6:27 PM, Jason Ekstrand 
> > wrote:
> >
> >> On Wed, Feb 8, 2017 at 5:34 PM, Nanley Chery 
> >> wrote:
> >>
> >>> On Thu, Feb 02, 2017 at 01:26:05PM -0800, Jason Ekstrand wrote:
> >>> > In order to get good performance numbers for this, I had to hack up the
> >>> > driver to whack wm_prog_data::uses_kill to true to emulate a discard
> >>> and
> >>> > used the Sascha "shadowmapping" demo.  Setting uses_kill to true
> >>> dropped
> >>> > the framerate on the demo by 25-30%.  Enabling the PMA fix brought it
> >>> > back up to around 90% of the original framerate.  This doesn't seem to
> >>> > really impact Dota 2;  probably because it doesn't use 16-bit depth.
> >>> >
> >>> > Reviewed-by: Lionel Landwerlin 
> >>> > ---
> >>> >  src/intel/vulkan/TODO  |   1 -
> >>> >  src/intel/vulkan/anv_cmd_buffer.c  |   2 +
> >>> >  src/intel/vulkan/anv_genX.h|   3 +
> >>> >  src/intel/vulkan/anv_private.h |  17 +
> >>> >  src/intel/vulkan/gen7_cmd_buffer.c |   7 ++
> >>> >  src/intel/vulkan/gen8_cmd_buffer.c | 133
> >>> +
> >>> >  src/intel/vulkan/genX_blorp_exec.c |   5 ++
> >>> >  src/intel/vulkan/genX_cmd_buffer.c |  15 -
> >>> >  src/intel/vulkan/genX_pipeline.c   |  38 +++
> >>> >  9 files changed, 219 insertions(+), 2 deletions(-)
> >>> >
> >>> > diff --git a/src/intel/vulkan/TODO b/src/intel/vulkan/TODO
> >>> > index 38acc0d..f8b73a1 100644
> >>> > --- a/src/intel/vulkan/TODO
> >>> > +++ b/src/intel/vulkan/TODO
> >>> > @@ -12,5 +12,4 @@ Performance:
> >>> >   - Compressed multisample support
> >>> >   - Pushing pieces of UBOs?
> >>> >   - Enable guardband clipping
> >>> > - - pma stall workaround
> >>> >   - Use soft-pin to avoid relocations
> >>> > diff --git a/src/intel/vulkan/anv_cmd_buffer.c
> >>> b/src/intel/vulkan/anv_cmd_buffer.c
> >>> > index 5886fa6..8c08f8d 100644
> >>> > --- a/src/intel/vulkan/anv_cmd_buffer.c
> >>> > +++ b/src/intel/vulkan/anv_cmd_buffer.c
> >>> > @@ -135,6 +135,8 @@ anv_cmd_state_reset(struct anv_cmd_buffer
> >>> *cmd_buffer)
> >>> > state->restart_index = UINT32_MAX;
> >>> > state->dynamic = default_dynamic_state;
> >>> > state->need_query_wa = true;
> >>> > +   state->pma_fix_enabled = false;
> >>> > +   state->hiz_enabled = false;
> >>> >
> >>> > if (state->attachments != NULL) {
> >>> >vk_free(_buffer->pool->alloc, state->attachments);
> >>> > diff --git a/src/intel/vulkan/anv_genX.h b/src/intel/vulkan/anv_genX.h
> >>> > index d04fe38..67147b0 100644
> >>> > --- a/src/intel/vulkan/anv_genX.h
> >>> > +++ b/src/intel/vulkan/anv_genX.h
> >>> > @@ -55,6 +55,9 @@ void genX(cmd_buffer_flush_dynamic_state)(struct
> >>> anv_cmd_buffer *cmd_buffer);
> >>> >
> >>> >  void genX(cmd_buffer_flush_compute_state)(struct anv_cmd_buffer
> >>> *cmd_buffer);
> >>> >
> >>> > +void genX(cmd_buffer_enable_pma_fix)(struct anv_cmd_buffer
> >>> *cmd_buffer,
> >>> > + bool enable);
> >>> > +
> >>> >  void
> >>> >  genX(emit_urb_setup)(struct anv_device *device, struct anv_batch
> >>> *batch,
> >>> >   const struct gen_l3_config *l3_config,
> >>> > diff --git a/src/intel/vulkan/anv_private.h
> >>> b/src/intel/vulkan/anv_private.h
> >>> > index 4fe3ebc..6efe4ea 100644
> >>> > --- a/src/intel/vulkan/anv_private.h
> >>> > +++ b/src/intel/vulkan/anv_private.h
> >>> > @@ -1163,6 +1163,20 @@ struct anv_cmd_state {
> >>> > bool need_query_wa;
> >>> >
> >>> > /**
> >>> > +* Whether or not the gen8 PMA fix is enabled.  We ensure that, at
> >>> the top
> >>> > +* of any command buffer it disabled by disabling it in
> >>> EndCommandBuffer
> >>>  ^
> >>>  is?
> >>>
> >>> > +* and before invoking the secondary in ExecuteCommands.
> >>> > +*/
> >>> > +   bool pma_fix_enabled;
> >>> > +
> >>> > +   /**
> >>> > +* Whether or not we now for certain that HiZ is enabled for the
> >>> current
> >>>^
> >>>know
> >>>
> >>> > +* subpass.  If, for whatever reason, we are unsure as to whether
> >>> HiZ is
> >>> > +* enabled or not, this will be false.
> >>> > +*/
> >>> > +   bool hiz_enabled;
> >>> > +
> >>> > +   /**
> >>> >  * Array length is anv_cmd_state::pass::attachment_count. Array
> >>> content is
> >>> >  * valid only when recording a render pass instance.
> >>> >  */
> >>> > @@ -1465,8 +1479,11 @@ struct anv_pipeline {
> >>> >
> >>> > uint32_t cs_right_mask;
> >>> >
> >>> > +   bool   

Re: [Mesa-dev] [PATCH v2 3/5] anv: Add support for the PMA fix on Broadwell

[Nanley Chery]

On Wed, Feb 08, 2017 at 06:27:52PM -0800, Jason Ekstrand wrote:
> On Wed, Feb 8, 2017 at 5:34 PM, Nanley Chery  wrote:
> 
> > On Thu, Feb 02, 2017 at 01:26:05PM -0800, Jason Ekstrand wrote:
> > > In order to get good performance numbers for this, I had to hack up the
> > > driver to whack wm_prog_data::uses_kill to true to emulate a discard and
> > > used the Sascha "shadowmapping" demo.  Setting uses_kill to true dropped
> > > the framerate on the demo by 25-30%.  Enabling the PMA fix brought it
> > > back up to around 90% of the original framerate.  This doesn't seem to
> > > really impact Dota 2;  probably because it doesn't use 16-bit depth.
> > >
> > > Reviewed-by: Lionel Landwerlin 
> > > ---
> > >  src/intel/vulkan/TODO  |   1 -
> > >  src/intel/vulkan/anv_cmd_buffer.c  |   2 +
> > >  src/intel/vulkan/anv_genX.h|   3 +
> > >  src/intel/vulkan/anv_private.h |  17 +
> > >  src/intel/vulkan/gen7_cmd_buffer.c |   7 ++
> > >  src/intel/vulkan/gen8_cmd_buffer.c | 133 ++
> > +++
> > >  src/intel/vulkan/genX_blorp_exec.c |   5 ++
> > >  src/intel/vulkan/genX_cmd_buffer.c |  15 -
> > >  src/intel/vulkan/genX_pipeline.c   |  38 +++
> > >  9 files changed, 219 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/src/intel/vulkan/TODO b/src/intel/vulkan/TODO
> > > index 38acc0d..f8b73a1 100644
> > > --- a/src/intel/vulkan/TODO
> > > +++ b/src/intel/vulkan/TODO
> > > @@ -12,5 +12,4 @@ Performance:
> > >   - Compressed multisample support
> > >   - Pushing pieces of UBOs?
> > >   - Enable guardband clipping
> > > - - pma stall workaround
> > >   - Use soft-pin to avoid relocations
> > > diff --git a/src/intel/vulkan/anv_cmd_buffer.c
> > b/src/intel/vulkan/anv_cmd_buffer.c
> > > index 5886fa6..8c08f8d 100644
> > > --- a/src/intel/vulkan/anv_cmd_buffer.c
> > > +++ b/src/intel/vulkan/anv_cmd_buffer.c
> > > @@ -135,6 +135,8 @@ anv_cmd_state_reset(struct anv_cmd_buffer
> > *cmd_buffer)
> > > state->restart_index = UINT32_MAX;
> > > state->dynamic = default_dynamic_state;
> > > state->need_query_wa = true;
> > > +   state->pma_fix_enabled = false;
> > > +   state->hiz_enabled = false;
> > >
> > > if (state->attachments != NULL) {
> > >vk_free(_buffer->pool->alloc, state->attachments);
> > > diff --git a/src/intel/vulkan/anv_genX.h b/src/intel/vulkan/anv_genX.h
> > > index d04fe38..67147b0 100644
> > > --- a/src/intel/vulkan/anv_genX.h
> > > +++ b/src/intel/vulkan/anv_genX.h
> > > @@ -55,6 +55,9 @@ void genX(cmd_buffer_flush_dynamic_state)(struct
> > anv_cmd_buffer *cmd_buffer);
> > >
> > >  void genX(cmd_buffer_flush_compute_state)(struct anv_cmd_buffer
> > *cmd_buffer);
> > >
> > > +void genX(cmd_buffer_enable_pma_fix)(struct anv_cmd_buffer *cmd_buffer,
> > > + bool enable);
> > > +
> > >  void
> > >  genX(emit_urb_setup)(struct anv_device *device, struct anv_batch *batch,
> > >   const struct gen_l3_config *l3_config,
> > > diff --git a/src/intel/vulkan/anv_private.h b/src/intel/vulkan/anv_
> > private.h
> > > index 4fe3ebc..6efe4ea 100644
> > > --- a/src/intel/vulkan/anv_private.h
> > > +++ b/src/intel/vulkan/anv_private.h
> > > @@ -1163,6 +1163,20 @@ struct anv_cmd_state {
> > > bool need_query_wa;
> > >
> > > /**
> > > +* Whether or not the gen8 PMA fix is enabled.  We ensure that, at
> > the top
> > > +* of any command buffer it disabled by disabling it in
> > EndCommandBuffer
> >  ^
> >  is?
> >

Fixed?

> > > +* and before invoking the secondary in ExecuteCommands.
> > > +*/
> > > +   bool pma_fix_enabled;
> > > +
> > > +   /**
> > > +* Whether or not we now for certain that HiZ is enabled for the
> > current
> >^
> >know
> >

Fixed?

> > > +* subpass.  If, for whatever reason, we are unsure as to whether
> > HiZ is
> > > +* enabled or not, this will be false.
> > > +*/
> > > +   bool hiz_enabled;
> > > +
> > > +   /**
> > >  * Array length is anv_cmd_state::pass::attachment_count. Array
> > content is
> > >  * valid only when recording a render pass instance.
> > >  */
> > > @@ -1465,8 +1479,11 @@ struct anv_pipeline {
> > >
> > > uint32_t cs_right_mask;
> > >
> > > +   bool writes_depth;
> > > +   bool depth_test_enable;
> > > bool writes_stencil;
> > > bool depth_clamp_enable;
> > > +   bool kill_pixel;
> > >
> > > struct {
> > >uint32_t  

Re: [Mesa-dev] [PATCH v2 3/5] anv: Add support for the PMA fix on Broadwell

[Jason Ekstrand]

On Wed, Feb 8, 2017 at 8:11 PM, Jason Ekstrand  wrote:

> On Wed, Feb 8, 2017 at 6:27 PM, Jason Ekstrand 
> wrote:
>
>> On Wed, Feb 8, 2017 at 5:34 PM, Nanley Chery 
>> wrote:
>>
>>> On Thu, Feb 02, 2017 at 01:26:05PM -0800, Jason Ekstrand wrote:
>>> > In order to get good performance numbers for this, I had to hack up the
>>> > driver to whack wm_prog_data::uses_kill to true to emulate a discard
>>> and
>>> > used the Sascha "shadowmapping" demo.  Setting uses_kill to true
>>> dropped
>>> > the framerate on the demo by 25-30%.  Enabling the PMA fix brought it
>>> > back up to around 90% of the original framerate.  This doesn't seem to
>>> > really impact Dota 2;  probably because it doesn't use 16-bit depth.
>>> >
>>> > Reviewed-by: Lionel Landwerlin 
>>> > ---
>>> >  src/intel/vulkan/TODO  |   1 -
>>> >  src/intel/vulkan/anv_cmd_buffer.c  |   2 +
>>> >  src/intel/vulkan/anv_genX.h|   3 +
>>> >  src/intel/vulkan/anv_private.h |  17 +
>>> >  src/intel/vulkan/gen7_cmd_buffer.c |   7 ++
>>> >  src/intel/vulkan/gen8_cmd_buffer.c | 133
>>> +
>>> >  src/intel/vulkan/genX_blorp_exec.c |   5 ++
>>> >  src/intel/vulkan/genX_cmd_buffer.c |  15 -
>>> >  src/intel/vulkan/genX_pipeline.c   |  38 +++
>>> >  9 files changed, 219 insertions(+), 2 deletions(-)
>>> >
>>> > diff --git a/src/intel/vulkan/TODO b/src/intel/vulkan/TODO
>>> > index 38acc0d..f8b73a1 100644
>>> > --- a/src/intel/vulkan/TODO
>>> > +++ b/src/intel/vulkan/TODO
>>> > @@ -12,5 +12,4 @@ Performance:
>>> >   - Compressed multisample support
>>> >   - Pushing pieces of UBOs?
>>> >   - Enable guardband clipping
>>> > - - pma stall workaround
>>> >   - Use soft-pin to avoid relocations
>>> > diff --git a/src/intel/vulkan/anv_cmd_buffer.c
>>> b/src/intel/vulkan/anv_cmd_buffer.c
>>> > index 5886fa6..8c08f8d 100644
>>> > --- a/src/intel/vulkan/anv_cmd_buffer.c
>>> > +++ b/src/intel/vulkan/anv_cmd_buffer.c
>>> > @@ -135,6 +135,8 @@ anv_cmd_state_reset(struct anv_cmd_buffer
>>> *cmd_buffer)
>>> > state->restart_index = UINT32_MAX;
>>> > state->dynamic = default_dynamic_state;
>>> > state->need_query_wa = true;
>>> > +   state->pma_fix_enabled = false;
>>> > +   state->hiz_enabled = false;
>>> >
>>> > if (state->attachments != NULL) {
>>> >vk_free(_buffer->pool->alloc, state->attachments);
>>> > diff --git a/src/intel/vulkan/anv_genX.h b/src/intel/vulkan/anv_genX.h
>>> > index d04fe38..67147b0 100644
>>> > --- a/src/intel/vulkan/anv_genX.h
>>> > +++ b/src/intel/vulkan/anv_genX.h
>>> > @@ -55,6 +55,9 @@ void genX(cmd_buffer_flush_dynamic_state)(struct
>>> anv_cmd_buffer *cmd_buffer);
>>> >
>>> >  void genX(cmd_buffer_flush_compute_state)(struct anv_cmd_buffer
>>> *cmd_buffer);
>>> >
>>> > +void genX(cmd_buffer_enable_pma_fix)(struct anv_cmd_buffer
>>> *cmd_buffer,
>>> > + bool enable);
>>> > +
>>> >  void
>>> >  genX(emit_urb_setup)(struct anv_device *device, struct anv_batch
>>> *batch,
>>> >   const struct gen_l3_config *l3_config,
>>> > diff --git a/src/intel/vulkan/anv_private.h
>>> b/src/intel/vulkan/anv_private.h
>>> > index 4fe3ebc..6efe4ea 100644
>>> > --- a/src/intel/vulkan/anv_private.h
>>> > +++ b/src/intel/vulkan/anv_private.h
>>> > @@ -1163,6 +1163,20 @@ struct anv_cmd_state {
>>> > bool need_query_wa;
>>> >
>>> > /**
>>> > +* Whether or not the gen8 PMA fix is enabled.  We ensure that, at
>>> the top
>>> > +* of any command buffer it disabled by disabling it in
>>> EndCommandBuffer
>>>  ^
>>>  is?
>>>
>>> > +* and before invoking the secondary in ExecuteCommands.
>>> > +*/
>>> > +   bool pma_fix_enabled;
>>> > +
>>> > +   /**
>>> > +* Whether or not we now for certain that HiZ is enabled for the
>>> current
>>>^
>>>know
>>>
>>> > +* subpass.  If, for whatever reason, we are unsure as to whether
>>> HiZ is
>>> > +* enabled or not, this will be false.
>>> > +*/
>>> > +   bool hiz_enabled;
>>> > +
>>> > +   /**
>>> >  * Array length is anv_cmd_state::pass::attachment_count. Array
>>> content is
>>> >  * valid only when recording a render pass instance.
>>> >  */
>>> > @@ -1465,8 +1479,11 @@ struct anv_pipeline {
>>> >
>>> > uint32_t cs_right_mask;
>>> >
>>> > +   bool writes_depth;
>>> > +   bool depth_test_enable;
>>> > bool writes_stencil;
>>> > bool depth_clamp_enable;
>>> > +   bool

Re: [Mesa-dev] [PATCH v2 3/5] anv: Add support for the PMA fix on Broadwell

[Jason Ekstrand]

On Wed, Feb 8, 2017 at 6:27 PM, Jason Ekstrand  wrote:

> On Wed, Feb 8, 2017 at 5:34 PM, Nanley Chery 
> wrote:
>
>> On Thu, Feb 02, 2017 at 01:26:05PM -0800, Jason Ekstrand wrote:
>> > In order to get good performance numbers for this, I had to hack up the
>> > driver to whack wm_prog_data::uses_kill to true to emulate a discard and
>> > used the Sascha "shadowmapping" demo.  Setting uses_kill to true dropped
>> > the framerate on the demo by 25-30%.  Enabling the PMA fix brought it
>> > back up to around 90% of the original framerate.  This doesn't seem to
>> > really impact Dota 2;  probably because it doesn't use 16-bit depth.
>> >
>> > Reviewed-by: Lionel Landwerlin 
>> > ---
>> >  src/intel/vulkan/TODO  |   1 -
>> >  src/intel/vulkan/anv_cmd_buffer.c  |   2 +
>> >  src/intel/vulkan/anv_genX.h|   3 +
>> >  src/intel/vulkan/anv_private.h |  17 +
>> >  src/intel/vulkan/gen7_cmd_buffer.c |   7 ++
>> >  src/intel/vulkan/gen8_cmd_buffer.c | 133
>> +
>> >  src/intel/vulkan/genX_blorp_exec.c |   5 ++
>> >  src/intel/vulkan/genX_cmd_buffer.c |  15 -
>> >  src/intel/vulkan/genX_pipeline.c   |  38 +++
>> >  9 files changed, 219 insertions(+), 2 deletions(-)
>> >
>> > diff --git a/src/intel/vulkan/TODO b/src/intel/vulkan/TODO
>> > index 38acc0d..f8b73a1 100644
>> > --- a/src/intel/vulkan/TODO
>> > +++ b/src/intel/vulkan/TODO
>> > @@ -12,5 +12,4 @@ Performance:
>> >   - Compressed multisample support
>> >   - Pushing pieces of UBOs?
>> >   - Enable guardband clipping
>> > - - pma stall workaround
>> >   - Use soft-pin to avoid relocations
>> > diff --git a/src/intel/vulkan/anv_cmd_buffer.c
>> b/src/intel/vulkan/anv_cmd_buffer.c
>> > index 5886fa6..8c08f8d 100644
>> > --- a/src/intel/vulkan/anv_cmd_buffer.c
>> > +++ b/src/intel/vulkan/anv_cmd_buffer.c
>> > @@ -135,6 +135,8 @@ anv_cmd_state_reset(struct anv_cmd_buffer
>> *cmd_buffer)
>> > state->restart_index = UINT32_MAX;
>> > state->dynamic = default_dynamic_state;
>> > state->need_query_wa = true;
>> > +   state->pma_fix_enabled = false;
>> > +   state->hiz_enabled = false;
>> >
>> > if (state->attachments != NULL) {
>> >vk_free(_buffer->pool->alloc, state->attachments);
>> > diff --git a/src/intel/vulkan/anv_genX.h b/src/intel/vulkan/anv_genX.h
>> > index d04fe38..67147b0 100644
>> > --- a/src/intel/vulkan/anv_genX.h
>> > +++ b/src/intel/vulkan/anv_genX.h
>> > @@ -55,6 +55,9 @@ void genX(cmd_buffer_flush_dynamic_state)(struct
>> anv_cmd_buffer *cmd_buffer);
>> >
>> >  void genX(cmd_buffer_flush_compute_state)(struct anv_cmd_buffer
>> *cmd_buffer);
>> >
>> > +void genX(cmd_buffer_enable_pma_fix)(struct anv_cmd_buffer
>> *cmd_buffer,
>> > + bool enable);
>> > +
>> >  void
>> >  genX(emit_urb_setup)(struct anv_device *device, struct anv_batch
>> *batch,
>> >   const struct gen_l3_config *l3_config,
>> > diff --git a/src/intel/vulkan/anv_private.h
>> b/src/intel/vulkan/anv_private.h
>> > index 4fe3ebc..6efe4ea 100644
>> > --- a/src/intel/vulkan/anv_private.h
>> > +++ b/src/intel/vulkan/anv_private.h
>> > @@ -1163,6 +1163,20 @@ struct anv_cmd_state {
>> > bool need_query_wa;
>> >
>> > /**
>> > +* Whether or not the gen8 PMA fix is enabled.  We ensure that, at
>> the top
>> > +* of any command buffer it disabled by disabling it in
>> EndCommandBuffer
>>  ^
>>  is?
>>
>> > +* and before invoking the secondary in ExecuteCommands.
>> > +*/
>> > +   bool pma_fix_enabled;
>> > +
>> > +   /**
>> > +* Whether or not we now for certain that HiZ is enabled for the
>> current
>>^
>>know
>>
>> > +* subpass.  If, for whatever reason, we are unsure as to whether
>> HiZ is
>> > +* enabled or not, this will be false.
>> > +*/
>> > +   bool hiz_enabled;
>> > +
>> > +   /**
>> >  * Array length is anv_cmd_state::pass::attachment_count. Array
>> content is
>> >  * valid only when recording a render pass instance.
>> >  */
>> > @@ -1465,8 +1479,11 @@ struct anv_pipeline {
>> >
>> > uint32_t cs_right_mask;
>> >
>> > +   bool writes_depth;
>> > +   bool depth_test_enable;
>> > bool writes_stencil;
>> > bool depth_clamp_enable;
>> > +   bool kill_pixel;
>> >
>> > struct {
>> >uint32_t  sf[7];
>> > diff --git a/src/intel/vulkan/gen7_cmd_buffer.c
>> b/src/intel/vulkan/gen7_cmd_buffer.c
>> 

Re: [Mesa-dev] [PATCH v2 3/5] anv: Add support for the PMA fix on Broadwell

[Jason Ekstrand]

On Wed, Feb 8, 2017 at 5:34 PM, Nanley Chery  wrote:

> On Thu, Feb 02, 2017 at 01:26:05PM -0800, Jason Ekstrand wrote:
> > In order to get good performance numbers for this, I had to hack up the
> > driver to whack wm_prog_data::uses_kill to true to emulate a discard and
> > used the Sascha "shadowmapping" demo.  Setting uses_kill to true dropped
> > the framerate on the demo by 25-30%.  Enabling the PMA fix brought it
> > back up to around 90% of the original framerate.  This doesn't seem to
> > really impact Dota 2;  probably because it doesn't use 16-bit depth.
> >
> > Reviewed-by: Lionel Landwerlin 
> > ---
> >  src/intel/vulkan/TODO  |   1 -
> >  src/intel/vulkan/anv_cmd_buffer.c  |   2 +
> >  src/intel/vulkan/anv_genX.h|   3 +
> >  src/intel/vulkan/anv_private.h |  17 +
> >  src/intel/vulkan/gen7_cmd_buffer.c |   7 ++
> >  src/intel/vulkan/gen8_cmd_buffer.c | 133 ++
> +++
> >  src/intel/vulkan/genX_blorp_exec.c |   5 ++
> >  src/intel/vulkan/genX_cmd_buffer.c |  15 -
> >  src/intel/vulkan/genX_pipeline.c   |  38 +++
> >  9 files changed, 219 insertions(+), 2 deletions(-)
> >
> > diff --git a/src/intel/vulkan/TODO b/src/intel/vulkan/TODO
> > index 38acc0d..f8b73a1 100644
> > --- a/src/intel/vulkan/TODO
> > +++ b/src/intel/vulkan/TODO
> > @@ -12,5 +12,4 @@ Performance:
> >   - Compressed multisample support
> >   - Pushing pieces of UBOs?
> >   - Enable guardband clipping
> > - - pma stall workaround
> >   - Use soft-pin to avoid relocations
> > diff --git a/src/intel/vulkan/anv_cmd_buffer.c
> b/src/intel/vulkan/anv_cmd_buffer.c
> > index 5886fa6..8c08f8d 100644
> > --- a/src/intel/vulkan/anv_cmd_buffer.c
> > +++ b/src/intel/vulkan/anv_cmd_buffer.c
> > @@ -135,6 +135,8 @@ anv_cmd_state_reset(struct anv_cmd_buffer
> *cmd_buffer)
> > state->restart_index = UINT32_MAX;
> > state->dynamic = default_dynamic_state;
> > state->need_query_wa = true;
> > +   state->pma_fix_enabled = false;
> > +   state->hiz_enabled = false;
> >
> > if (state->attachments != NULL) {
> >vk_free(_buffer->pool->alloc, state->attachments);
> > diff --git a/src/intel/vulkan/anv_genX.h b/src/intel/vulkan/anv_genX.h
> > index d04fe38..67147b0 100644
> > --- a/src/intel/vulkan/anv_genX.h
> > +++ b/src/intel/vulkan/anv_genX.h
> > @@ -55,6 +55,9 @@ void genX(cmd_buffer_flush_dynamic_state)(struct
> anv_cmd_buffer *cmd_buffer);
> >
> >  void genX(cmd_buffer_flush_compute_state)(struct anv_cmd_buffer
> *cmd_buffer);
> >
> > +void genX(cmd_buffer_enable_pma_fix)(struct anv_cmd_buffer *cmd_buffer,
> > + bool enable);
> > +
> >  void
> >  genX(emit_urb_setup)(struct anv_device *device, struct anv_batch *batch,
> >   const struct gen_l3_config *l3_config,
> > diff --git a/src/intel/vulkan/anv_private.h b/src/intel/vulkan/anv_
> private.h
> > index 4fe3ebc..6efe4ea 100644
> > --- a/src/intel/vulkan/anv_private.h
> > +++ b/src/intel/vulkan/anv_private.h
> > @@ -1163,6 +1163,20 @@ struct anv_cmd_state {
> > bool need_query_wa;
> >
> > /**
> > +* Whether or not the gen8 PMA fix is enabled.  We ensure that, at
> the top
> > +* of any command buffer it disabled by disabling it in
> EndCommandBuffer
>  ^
>  is?
>
> > +* and before invoking the secondary in ExecuteCommands.
> > +*/
> > +   bool pma_fix_enabled;
> > +
> > +   /**
> > +* Whether or not we now for certain that HiZ is enabled for the
> current
>^
>know
>
> > +* subpass.  If, for whatever reason, we are unsure as to whether
> HiZ is
> > +* enabled or not, this will be false.
> > +*/
> > +   bool hiz_enabled;
> > +
> > +   /**
> >  * Array length is anv_cmd_state::pass::attachment_count. Array
> content is
> >  * valid only when recording a render pass instance.
> >  */
> > @@ -1465,8 +1479,11 @@ struct anv_pipeline {
> >
> > uint32_t cs_right_mask;
> >
> > +   bool writes_depth;
> > +   bool depth_test_enable;
> > bool writes_stencil;
> > bool depth_clamp_enable;
> > +   bool kill_pixel;
> >
> > struct {
> >uint32_t  sf[7];
> > diff --git a/src/intel/vulkan/gen7_cmd_buffer.c
> b/src/intel/vulkan/gen7_cmd_buffer.c
> > index 013ed87..c1a25e8 100644
> > --- a/src/intel/vulkan/gen7_cmd_buffer.c
> > +++ b/src/intel/vulkan/gen7_cmd_buffer.c
> > @@ -260,6 +260,13 @@ genX(cmd_buffer_flush_dynamic_state)(struct
> 

Re: [Mesa-dev] [PATCH v2 3/5] anv: Add support for the PMA fix on Broadwell

[Nanley Chery]

On Thu, Feb 02, 2017 at 01:26:05PM -0800, Jason Ekstrand wrote:
> In order to get good performance numbers for this, I had to hack up the
> driver to whack wm_prog_data::uses_kill to true to emulate a discard and
> used the Sascha "shadowmapping" demo.  Setting uses_kill to true dropped
> the framerate on the demo by 25-30%.  Enabling the PMA fix brought it
> back up to around 90% of the original framerate.  This doesn't seem to
> really impact Dota 2;  probably because it doesn't use 16-bit depth.
> 
> Reviewed-by: Lionel Landwerlin 
> ---
>  src/intel/vulkan/TODO  |   1 -
>  src/intel/vulkan/anv_cmd_buffer.c  |   2 +
>  src/intel/vulkan/anv_genX.h|   3 +
>  src/intel/vulkan/anv_private.h |  17 +
>  src/intel/vulkan/gen7_cmd_buffer.c |   7 ++
>  src/intel/vulkan/gen8_cmd_buffer.c | 133 
> +
>  src/intel/vulkan/genX_blorp_exec.c |   5 ++
>  src/intel/vulkan/genX_cmd_buffer.c |  15 -
>  src/intel/vulkan/genX_pipeline.c   |  38 +++
>  9 files changed, 219 insertions(+), 2 deletions(-)
> 
> diff --git a/src/intel/vulkan/TODO b/src/intel/vulkan/TODO
> index 38acc0d..f8b73a1 100644
> --- a/src/intel/vulkan/TODO
> +++ b/src/intel/vulkan/TODO
> @@ -12,5 +12,4 @@ Performance:
>   - Compressed multisample support
>   - Pushing pieces of UBOs?
>   - Enable guardband clipping
> - - pma stall workaround
>   - Use soft-pin to avoid relocations
> diff --git a/src/intel/vulkan/anv_cmd_buffer.c 
> b/src/intel/vulkan/anv_cmd_buffer.c
> index 5886fa6..8c08f8d 100644
> --- a/src/intel/vulkan/anv_cmd_buffer.c
> +++ b/src/intel/vulkan/anv_cmd_buffer.c
> @@ -135,6 +135,8 @@ anv_cmd_state_reset(struct anv_cmd_buffer *cmd_buffer)
> state->restart_index = UINT32_MAX;
> state->dynamic = default_dynamic_state;
> state->need_query_wa = true;
> +   state->pma_fix_enabled = false;
> +   state->hiz_enabled = false;
>  
> if (state->attachments != NULL) {
>vk_free(_buffer->pool->alloc, state->attachments);
> diff --git a/src/intel/vulkan/anv_genX.h b/src/intel/vulkan/anv_genX.h
> index d04fe38..67147b0 100644
> --- a/src/intel/vulkan/anv_genX.h
> +++ b/src/intel/vulkan/anv_genX.h
> @@ -55,6 +55,9 @@ void genX(cmd_buffer_flush_dynamic_state)(struct 
> anv_cmd_buffer *cmd_buffer);
>  
>  void genX(cmd_buffer_flush_compute_state)(struct anv_cmd_buffer *cmd_buffer);
>  
> +void genX(cmd_buffer_enable_pma_fix)(struct anv_cmd_buffer *cmd_buffer,
> + bool enable);
> +
>  void
>  genX(emit_urb_setup)(struct anv_device *device, struct anv_batch *batch,
>   const struct gen_l3_config *l3_config,
> diff --git a/src/intel/vulkan/anv_private.h b/src/intel/vulkan/anv_private.h
> index 4fe3ebc..6efe4ea 100644
> --- a/src/intel/vulkan/anv_private.h
> +++ b/src/intel/vulkan/anv_private.h
> @@ -1163,6 +1163,20 @@ struct anv_cmd_state {
> bool need_query_wa;
>  
> /**
> +* Whether or not the gen8 PMA fix is enabled.  We ensure that, at the top
> +* of any command buffer it disabled by disabling it in EndCommandBuffer
 ^
 is?

> +* and before invoking the secondary in ExecuteCommands.
> +*/
> +   bool pma_fix_enabled;
> +
> +   /**
> +* Whether or not we now for certain that HiZ is enabled for the current
   ^
   know

> +* subpass.  If, for whatever reason, we are unsure as to whether HiZ is
> +* enabled or not, this will be false.
> +*/
> +   bool hiz_enabled;
> +
> +   /**
>  * Array length is anv_cmd_state::pass::attachment_count. Array content is
>  * valid only when recording a render pass instance.
>  */
> @@ -1465,8 +1479,11 @@ struct anv_pipeline {
>  
> uint32_t cs_right_mask;
>  
> +   bool writes_depth;
> +   bool depth_test_enable;
> bool writes_stencil;
> bool depth_clamp_enable;
> +   bool kill_pixel;
>  
> struct {
>uint32_t  sf[7];
> diff --git a/src/intel/vulkan/gen7_cmd_buffer.c 
> b/src/intel/vulkan/gen7_cmd_buffer.c
> index 013ed87..c1a25e8 100644
> --- a/src/intel/vulkan/gen7_cmd_buffer.c
> +++ b/src/intel/vulkan/gen7_cmd_buffer.c
> @@ -260,6 +260,13 @@ genX(cmd_buffer_flush_dynamic_state)(struct 
> anv_cmd_buffer *cmd_buffer)
> cmd_buffer->state.dirty = 0;
>  }
>  
> +void
> +genX(cmd_buffer_enable_pma_fix)(struct anv_cmd_buffer *cmd_buffer,
> +bool enable)
> +{
> +   /* The NP PMA fix doesn't exist on gen7 */
> +}
> +
>  void genX(CmdSetEvent)(
>

[Mesa-dev] [PATCH v2 3/5] anv: Add support for the PMA fix on Broadwell

[Jason Ekstrand]

In order to get good performance numbers for this, I had to hack up the
driver to whack wm_prog_data::uses_kill to true to emulate a discard and
used the Sascha "shadowmapping" demo.  Setting uses_kill to true dropped
the framerate on the demo by 25-30%.  Enabling the PMA fix brought it
back up to around 90% of the original framerate.  This doesn't seem to
really impact Dota 2;  probably because it doesn't use 16-bit depth.

Reviewed-by: Lionel Landwerlin 
---
 src/intel/vulkan/TODO  |   1 -
 src/intel/vulkan/anv_cmd_buffer.c  |   2 +
 src/intel/vulkan/anv_genX.h|   3 +
 src/intel/vulkan/anv_private.h |  17 +
 src/intel/vulkan/gen7_cmd_buffer.c |   7 ++
 src/intel/vulkan/gen8_cmd_buffer.c | 133 +
 src/intel/vulkan/genX_blorp_exec.c |   5 ++
 src/intel/vulkan/genX_cmd_buffer.c |  15 -
 src/intel/vulkan/genX_pipeline.c   |  38 +++
 9 files changed, 219 insertions(+), 2 deletions(-)

diff --git a/src/intel/vulkan/TODO b/src/intel/vulkan/TODO
index 38acc0d..f8b73a1 100644
--- a/src/intel/vulkan/TODO
+++ b/src/intel/vulkan/TODO
@@ -12,5 +12,4 @@ Performance:
  - Compressed multisample support
  - Pushing pieces of UBOs?
  - Enable guardband clipping
- - pma stall workaround
  - Use soft-pin to avoid relocations
diff --git a/src/intel/vulkan/anv_cmd_buffer.c 
b/src/intel/vulkan/anv_cmd_buffer.c
index 5886fa6..8c08f8d 100644
--- a/src/intel/vulkan/anv_cmd_buffer.c
+++ b/src/intel/vulkan/anv_cmd_buffer.c
@@ -135,6 +135,8 @@ anv_cmd_state_reset(struct anv_cmd_buffer *cmd_buffer)
state->restart_index = UINT32_MAX;
state->dynamic = default_dynamic_state;
state->need_query_wa = true;
+   state->pma_fix_enabled = false;
+   state->hiz_enabled = false;
 
if (state->attachments != NULL) {
   vk_free(_buffer->pool->alloc, state->attachments);
diff --git a/src/intel/vulkan/anv_genX.h b/src/intel/vulkan/anv_genX.h
index d04fe38..67147b0 100644
--- a/src/intel/vulkan/anv_genX.h
+++ b/src/intel/vulkan/anv_genX.h
@@ -55,6 +55,9 @@ void genX(cmd_buffer_flush_dynamic_state)(struct 
anv_cmd_buffer *cmd_buffer);
 
 void genX(cmd_buffer_flush_compute_state)(struct anv_cmd_buffer *cmd_buffer);
 
+void genX(cmd_buffer_enable_pma_fix)(struct anv_cmd_buffer *cmd_buffer,
+ bool enable);
+
 void
 genX(emit_urb_setup)(struct anv_device *device, struct anv_batch *batch,
  const struct gen_l3_config *l3_config,
diff --git a/src/intel/vulkan/anv_private.h b/src/intel/vulkan/anv_private.h
index 4fe3ebc..6efe4ea 100644
--- a/src/intel/vulkan/anv_private.h
+++ b/src/intel/vulkan/anv_private.h
@@ -1163,6 +1163,20 @@ struct anv_cmd_state {
bool need_query_wa;
 
/**
+* Whether or not the gen8 PMA fix is enabled.  We ensure that, at the top
+* of any command buffer it disabled by disabling it in EndCommandBuffer
+* and before invoking the secondary in ExecuteCommands.
+*/
+   bool pma_fix_enabled;
+
+   /**
+* Whether or not we now for certain that HiZ is enabled for the current
+* subpass.  If, for whatever reason, we are unsure as to whether HiZ is
+* enabled or not, this will be false.
+*/
+   bool hiz_enabled;
+
+   /**
 * Array length is anv_cmd_state::pass::attachment_count. Array content is
 * valid only when recording a render pass instance.
 */
@@ -1465,8 +1479,11 @@ struct anv_pipeline {
 
uint32_t cs_right_mask;
 
+   bool writes_depth;
+   bool depth_test_enable;
bool writes_stencil;
bool depth_clamp_enable;
+   bool kill_pixel;
 
struct {
   uint32_t  sf[7];
diff --git a/src/intel/vulkan/gen7_cmd_buffer.c 
b/src/intel/vulkan/gen7_cmd_buffer.c
index 013ed87..c1a25e8 100644
--- a/src/intel/vulkan/gen7_cmd_buffer.c
+++ b/src/intel/vulkan/gen7_cmd_buffer.c
@@ -260,6 +260,13 @@ genX(cmd_buffer_flush_dynamic_state)(struct anv_cmd_buffer 
*cmd_buffer)
cmd_buffer->state.dirty = 0;
 }
 
+void
+genX(cmd_buffer_enable_pma_fix)(struct anv_cmd_buffer *cmd_buffer,
+bool enable)
+{
+   /* The NP PMA fix doesn't exist on gen7 */
+}
+
 void genX(CmdSetEvent)(
 VkCommandBuffer commandBuffer,
 VkEvent event,
diff --git a/src/intel/vulkan/gen8_cmd_buffer.c 
b/src/intel/vulkan/gen8_cmd_buffer.c
index 8c8de62..271ab3f 100644
--- a/src/intel/vulkan/gen8_cmd_buffer.c
+++ b/src/intel/vulkan/gen8_cmd_buffer.c
@@ -155,6 +155,135 @@ __emit_sf_state(struct anv_cmd_buffer *cmd_buffer)
 #endif
 
 void