vkd3d: Only enable native FP16 codegen for RADV.
Regression in Deathloop on NV with native FP16 FSR. Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
This commit is contained in:
parent
46470017a3
commit
fd05839eb9
|
@ -6081,8 +6081,10 @@ static void vkd3d_init_shader_extensions(struct d3d12_device *device)
|
|||
VKD3D_SHADER_TARGET_EXTENSION_BARYCENTRIC_KHR;
|
||||
}
|
||||
|
||||
if (device->d3d12_caps.options4.Native16BitShaderOpsSupported)
|
||||
if (device->d3d12_caps.options4.Native16BitShaderOpsSupported &&
|
||||
device->device_info.driver_properties.driverID == VK_DRIVER_ID_MESA_RADV)
|
||||
{
|
||||
/* Native FP16 is buggy on NV for now. */
|
||||
device->vk_info.shader_extensions[device->vk_info.shader_extension_count++] =
|
||||
VKD3D_SHADER_TARGET_EXTENSION_MIN_PRECISION_IS_NATIVE_16BIT;
|
||||
}
|
||||
|
|
Loading…
Reference in New Issue