[Mesa-dev] [PATCH 6/6] radv: partially enable VK_KHR_shader_float16_int8

[Samuel Pitoiset]

On 4/1/19 5:22 PM, Bas Nieuwenhuizen wrote:
> On Mon, Apr 1, 2019 at 4:15 PM Samuel Pitoiset
> <samuel.pitoiset at gmail.com> wrote:
>> Only 8-bit integers for now, float16 requires a bit more work.
>>
>> Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
>> ---
>>   src/amd/vulkan/radv_device.c      | 8 ++++++++
>>   src/amd/vulkan/radv_extensions.py | 1 +
>>   src/amd/vulkan/radv_shader.c      | 1 +
>>   3 files changed, 10 insertions(+)
>>
>> diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c
>> index d26557fd449..ffec7cfdb19 100644
>> --- a/src/amd/vulkan/radv_device.c
>> +++ b/src/amd/vulkan/radv_device.c
>> @@ -902,6 +902,14 @@ void radv_GetPhysicalDeviceFeatures2(
>>                          features->storagePushConstant8 = enabled;
>>                          break;
>>                  }
>> +               case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT16_INT8_FEATURES_KHR: {
>> +                       VkPhysicalDeviceFloat16Int8FeaturesKHR *features =
>> +                               (VkPhysicalDeviceFloat16Int8FeaturesKHR*)ext;
>> +                       bool enabled = pdevice->rad_info.chip_class >= VI;
>> +                       features->shaderFloat16 = VK_FALSE;
>> +                       features->shaderInt8 = enabled;
>> +                       break;
>> +               }
>>                  default:
>>                          break;
>>                  }
>> diff --git a/src/amd/vulkan/radv_extensions.py b/src/amd/vulkan/radv_extensions.py
>> index 79562ff6e97..4b118b4ba82 100644
>> --- a/src/amd/vulkan/radv_extensions.py
>> +++ b/src/amd/vulkan/radv_extensions.py
>> @@ -82,6 +82,7 @@ EXTENSIONS = [
>>       Extension('VK_KHR_relaxed_block_layout',              1, True),
>>       Extension('VK_KHR_sampler_mirror_clamp_to_edge',      1, True),
>>       Extension('VK_KHR_shader_draw_parameters',            1, True),
>> +    Extension('VK_KHR_shader_float16_int8',               1, 'device->rad_info.chip_class >= VI'),
> Why only VI+? float16 I understand, but int8 should be universal, no?

Because it's untested, and also because I plan to enable 16bit integer 
for SI/CIK at the same time.

>
>>       Extension('VK_KHR_storage_buffer_storage_class',      1, True),
>>       Extension('VK_KHR_surface',                          25, 'RADV_HAS_SURFACE'),
>>       Extension('VK_KHR_surface_protected_capabilities',    1, 'RADV_HAS_SURFACE'),
>> diff --git a/src/amd/vulkan/radv_shader.c b/src/amd/vulkan/radv_shader.c
>> index f905dfd9e5b..d3d073d1db8 100644
>> --- a/src/amd/vulkan/radv_shader.c
>> +++ b/src/amd/vulkan/radv_shader.c
>> @@ -251,6 +251,7 @@ radv_shader_compile_to_nir(struct radv_device *device,
>>                                  .trinary_minmax = true,
>>                                  .variable_pointers = true,
>>                                  .storage_8bit = true,
>> +                               .int8 = true,
>>                          },
>>                          .ubo_ptr_type = glsl_vector_type(GLSL_TYPE_UINT, 2),
>>                          .ssbo_ptr_type = glsl_vector_type(GLSL_TYPE_UINT, 2),
>> --
>> 2.21.0
>>
>> _______________________________________________
>> mesa-dev mailing list
>> mesa-dev at lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev