nvk: Implement VK_KHR_zero_initialize_workgroup_memory

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/9609
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27130>

src/nouveau/vulkan/nvk_physical_device.c

@@ -146,6 +146,7 @@ nvk_get_device_extensions(const struct nvk_instance *instance,
       .KHR_variable_pointers = true,
       .KHR_vulkan_memory_model = nvk_use_nak(info),
       .KHR_workgroup_memory_explicit_layout = true,
+      .KHR_zero_initialize_workgroup_memory = true,
       .EXT_4444_formats = true,
       .EXT_attachment_feedback_loop_layout = true,
       .EXT_border_color_swizzle = true,
@@ -334,6 +335,7 @@ nvk_get_device_features(const struct nv_device_info *info,
       .subgroupSizeControl = true,
       .computeFullSubgroups = true,
       .synchronization2 = true,
+      .shaderZeroInitializeWorkgroupMemory = true,
       .dynamicRendering = true,
       .shaderIntegerDotProduct = true,
       .maintenance4 = true,

src/nouveau/vulkan/nvk_shader.c

@@ -389,6 +389,15 @@ nvk_lower_nir(struct nvk_device *dev, nir_shader *nir,
    }
    NIR_PASS(_, nir, nir_lower_explicit_io, nir_var_mem_shared,
             nir_address_format_32bit_offset);
+
+   if (nir->info.zero_initialize_shared_memory && nir->info.shared_size > 0) {
+      /* QMD::SHARED_MEMORY_SIZE requires an alignment of 256B so it's safe to
+       * align everything up to 16B so we can write whole vec4s.
+       */
+      nir->info.shared_size = align(nir->info.shared_size, 16);
+      NIR_PASS(_, nir, nir_zero_initialize_shared_memory,
+               nir->info.shared_size, 16);
+   }
 }
 
 #ifndef NDEBUG