As a side-note, I pushed code on Friday that gets rid of two of these (the
create_pipeline calls).  I think the best way to solve
emit_state_base_address is by moving a bit of code into genX_cmd_buffer.c.
null surface states probably need to be moved into ISL.

In other words, while it's a nice cleanup, I think the correct answer for
almost all cases is to just stop doing the switches.

On Mon, Oct 17, 2016 at 8:51 AM, Kenneth Graunke <kenn...@whitecape.org>
wrote:

> On Wednesday, October 12, 2016 11:28:04 PM PDT Lionel Landwerlin wrote:
> > Signed-off-by: Lionel Landwerlin <lionel.g.landwer...@intel.com>
> > ---
> >  src/intel/vulkan/anv_cmd_buffer.c | 33 ++++++------------------------
> ---
> >  src/intel/vulkan/anv_device.c     | 19 +------------------
> >  src/intel/vulkan/anv_pipeline.c   | 30 ++++--------------------------
> >  src/intel/vulkan/anv_private.h    | 23 +++++++++++++++++++++++
> >  4 files changed, 34 insertions(+), 71 deletions(-)
> >
> > diff --git a/src/intel/vulkan/anv_cmd_buffer.c
> b/src/intel/vulkan/anv_cmd_buffer.c
> > index 5bcd5e0..b051489 100644
> > --- a/src/intel/vulkan/anv_cmd_buffer.c
> > +++ b/src/intel/vulkan/anv_cmd_buffer.c
> > @@ -356,19 +356,9 @@ VkResult anv_ResetCommandBuffer(
> >  void
> >  anv_cmd_buffer_emit_state_base_address(struct anv_cmd_buffer
> *cmd_buffer)
> >  {
> > -   switch (cmd_buffer->device->info.gen) {
> > -   case 7:
> > -      if (cmd_buffer->device->info.is_haswell)
> > -         return gen75_cmd_buffer_emit_state_base_address(cmd_buffer);
> > -      else
> > -         return gen7_cmd_buffer_emit_state_base_address(cmd_buffer);
> > -   case 8:
> > -      return gen8_cmd_buffer_emit_state_base_address(cmd_buffer);
> > -   case 9:
> > -      return gen9_cmd_buffer_emit_state_base_address(cmd_buffer);
> > -   default:
> > -      unreachable("unsupported gen\n");
> > -   }
> > +   ANV_GEN_DISPATCH(cmd_buffer->device,
> > +                    cmd_buffer_emit_state_base_address,
> > +                    cmd_buffer);
> >  }
> >
> >  VkResult anv_BeginCommandBuffer(
> > @@ -714,20 +704,9 @@ static struct anv_state
> >  anv_cmd_buffer_alloc_null_surface_state(struct anv_cmd_buffer
> *cmd_buffer,
> >                                          struct anv_framebuffer *fb)
> >  {
> > -   switch (cmd_buffer->device->info.gen) {
> > -   case 7:
> > -      if (cmd_buffer->device->info.is_haswell) {
> > -         return gen75_cmd_buffer_alloc_null_surface_state(cmd_buffer,
> fb);
> > -      } else {
> > -         return gen7_cmd_buffer_alloc_null_surface_state(cmd_buffer,
> fb);
> > -      }
> > -   case 8:
> > -      return gen8_cmd_buffer_alloc_null_surface_state(cmd_buffer, fb);
> > -   case 9:
> > -      return gen9_cmd_buffer_alloc_null_surface_state(cmd_buffer, fb);
> > -   default:
> > -      unreachable("Invalid hardware generation");
> > -   }
> > +   return ANV_GEN_DISPATCH(cmd_buffer->device,
> > +                           cmd_buffer_alloc_null_surface_state,
> > +                           cmd_buffer, fb);
> >  }
> >
> >  VkResult
> > diff --git a/src/intel/vulkan/anv_device.c
> b/src/intel/vulkan/anv_device.c
> > index 24f7227..6dfdfdb 100644
> > --- a/src/intel/vulkan/anv_device.c
> > +++ b/src/intel/vulkan/anv_device.c
> > @@ -921,24 +921,7 @@ VkResult anv_CreateDevice(
> >
> >     anv_queue_init(device, &device->queue);
> >
> > -   switch (device->info.gen) {
> > -   case 7:
> > -      if (!device->info.is_haswell)
> > -         result = gen7_init_device_state(device);
> > -      else
> > -         result = gen75_init_device_state(device);
> > -      break;
> > -   case 8:
> > -      result = gen8_init_device_state(device);
> > -      break;
> > -   case 9:
> > -      result = gen9_init_device_state(device);
> > -      break;
> > -   default:
> > -      /* Shouldn't get here as we don't create physical devices for any
> other
> > -       * gens. */
> > -      unreachable("unhandled gen");
> > -   }
> > +   result = ANV_GEN_DISPATCH(device, init_device_state, device);
> >     if (result != VK_SUCCESS)
> >        goto fail_fd;
> >
> > diff --git a/src/intel/vulkan/anv_pipeline.c b/src/intel/vulkan/anv_
> pipeline.c
> > index 6b393a6..f9f1cbf 100644
> > --- a/src/intel/vulkan/anv_pipeline.c
> > +++ b/src/intel/vulkan/anv_pipeline.c
> > @@ -1182,19 +1182,8 @@ anv_graphics_pipeline_create(
> >     ANV_FROM_HANDLE(anv_device, device, _device);
> >     ANV_FROM_HANDLE(anv_pipeline_cache, cache, _cache);
> >
> > -   switch (device->info.gen) {
> > -   case 7:
> > -      if (device->info.is_haswell)
> > -         return gen75_graphics_pipeline_create(_device, cache,
> pCreateInfo, extra, pAllocator, pPipeline);
> > -      else
> > -         return gen7_graphics_pipeline_create(_device, cache,
> pCreateInfo, extra, pAllocator, pPipeline);
> > -   case 8:
> > -      return gen8_graphics_pipeline_create(_device, cache,
> pCreateInfo, extra, pAllocator, pPipeline);
> > -   case 9:
> > -      return gen9_graphics_pipeline_create(_device, cache,
> pCreateInfo, extra, pAllocator, pPipeline);
> > -   default:
> > -      unreachable("unsupported gen\n");
> > -   }
> > +   return ANV_GEN_DISPATCH(device, graphics_pipeline_create,
> > +                           _device, cache, pCreateInfo, extra,
> pAllocator, pPipeline);
> >  }
> >
> >  VkResult anv_CreateGraphicsPipelines(
> > @@ -1235,19 +1224,8 @@ static VkResult anv_compute_pipeline_create(
> >     ANV_FROM_HANDLE(anv_device, device, _device);
> >     ANV_FROM_HANDLE(anv_pipeline_cache, cache, _cache);
> >
> > -   switch (device->info.gen) {
> > -   case 7:
> > -      if (device->info.is_haswell)
> > -         return gen75_compute_pipeline_create(_device, cache,
> pCreateInfo, pAllocator, pPipeline);
> > -      else
> > -         return gen7_compute_pipeline_create(_device, cache,
> pCreateInfo, pAllocator, pPipeline);
> > -   case 8:
> > -      return gen8_compute_pipeline_create(_device, cache, pCreateInfo,
> pAllocator, pPipeline);
> > -   case 9:
> > -      return gen9_compute_pipeline_create(_device, cache, pCreateInfo,
> pAllocator, pPipeline);
> > -   default:
> > -      unreachable("unsupported gen\n");
> > -   }
> > +   return ANV_GEN_DISPATCH(device, compute_pipeline_create,
> > +                           _device, cache, pCreateInfo, pAllocator,
> pPipeline);
> >  }
> >
> >  VkResult anv_CreateComputePipelines(
> > diff --git a/src/intel/vulkan/anv_private.h b/src/intel/vulkan/anv_
> private.h
> > index 0705263..27ae5e88 100644
> > --- a/src/intel/vulkan/anv_private.h
> > +++ b/src/intel/vulkan/anv_private.h
> > @@ -1994,6 +1994,29 @@ ANV_DEFINE_STRUCT_CASTS(anv_common,
> VkMemoryBarrier)
> >  ANV_DEFINE_STRUCT_CASTS(anv_common, VkBufferMemoryBarrier)
> >  ANV_DEFINE_STRUCT_CASTS(anv_common, VkImageMemoryBarrier)
> >
> > +#define ANV_GEN_DISPATCH(device, name, ...)      \
> > +   ({                                            \
> > +      __typeof(gen7_ ## name)* __func = NULL;    \
> > +      switch ((device)->info.gen) {              \
> > +      case 7:                                    \
> > +         if ((device)->info.is_haswell) {        \
> > +            __func = gen75_ ## name;             \
> > +         } else {                                \
> > +            __func = gen7_ ## name;              \
> > +         }                                       \
> > +         break;                                  \
> > +      case 8:                                    \
> > +         __func = gen8_ ## name;                 \
> > +         break;                                  \
> > +      case 9:                                    \
> > +         __func = gen9_ ## name;                 \
> > +         break;                                  \
> > +      default:                                   \
> > +         unreachable("unhandled gen");           \
> > +      };                                         \
> > +      __func( __VA_ARGS__);                      \
> > +   })
> > +
> >  /* Gen-specific function declarations */
> >  #ifdef genX
> >  #  include "anv_genX.h"
> >
>
> How does the generated assembly look now compared to before this patch?
> I don't have a good sense of whether the compiler can eliminate the
> function pointer or not.
>
> Assuming the compiler is able to generate sensible code for this,
> it looks great.
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>
>
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev

Reply via email to