radv: implement VK_KHR_push_descriptor

Signed-off-by: Fredrik Höglund <[email protected]>
Reviewed-by: Bas Nieuwenhuizen <[email protected]>

6 files changed, 128 insertions, 2 deletions

diff --git a/src/amd/vulkan/radv_cmd_buffer.c b/src/amd/vulkan/radv_cmd_buffer.c
index b0a68548295..46c139a8fcb 100644
--- a/src/amd/vulkan/radv_cmd_buffer.c
+++ b/src/amd/vulkan/radv_cmd_buffer.c
@@ -202,6 +202,7 @@ radv_cmd_buffer_destroy(struct radv_cmd_buffer *cmd_buffer)
 	if (cmd_buffer->upload.upload_bo)
 		cmd_buffer->device->ws->buffer_destroy(cmd_buffer->upload.upload_bo);
 	cmd_buffer->device->ws->cs_destroy(cmd_buffer->cs);
+	free(cmd_buffer->push_descriptors.set.mapped_ptr);
 	vk_free(&cmd_buffer->pool->alloc, cmd_buffer);
 }
 
@@ -1298,6 +1299,24 @@ radv_emit_descriptor_set_userdata(struct radv_cmd_buffer *cmd_buffer,
 }
 
 static void
+radv_flush_push_descriptors(struct radv_cmd_buffer *cmd_buffer)
+{
+	struct radv_descriptor_set *set = &cmd_buffer->push_descriptors.set;
+	uint32_t *ptr = NULL;
+	unsigned bo_offset;
+
+	if (!radv_cmd_buffer_upload_alloc(cmd_buffer, set->size, 32,
+	                                  &bo_offset,
+	                                  (void**) &ptr))
+		return;
+
+	set->va = cmd_buffer->device->ws->buffer_get_va(cmd_buffer->upload.upload_bo);
+	set->va += bo_offset;
+
+	memcpy(ptr, set->mapped_ptr, set->size);
+}
+
+static void
 radv_flush_descriptors(struct radv_cmd_buffer *cmd_buffer,
 		       struct radv_pipeline *pipeline,
 		       VkShaderStageFlags stages)
@@ -1306,6 +1325,9 @@ radv_flush_descriptors(struct radv_cmd_buffer *cmd_buffer,
 	if (!cmd_buffer->state.descriptors_dirty)
 		return;
 
+	if (cmd_buffer->state.push_descriptors_dirty)
+		radv_flush_push_descriptors(cmd_buffer);
+
 	for (i = 0; i < MAX_SETS; i++) {
 		if (!(cmd_buffer->state.descriptors_dirty & (1 << i)))
 			continue;
@@ -1316,6 +1338,7 @@ radv_flush_descriptors(struct radv_cmd_buffer *cmd_buffer,
 		radv_emit_descriptor_set_userdata(cmd_buffer, pipeline, stages, set, i);
 	}
 	cmd_buffer->state.descriptors_dirty = 0;
+	cmd_buffer->state.push_descriptors_dirty = false;
 }
 
 static void
@@ -1912,6 +1935,59 @@ void radv_CmdBindDescriptorSets(
 	assert(cmd_buffer->cs->cdw <= cdw_max);
 }
 
+static bool radv_init_push_descriptor_set(struct radv_cmd_buffer *cmd_buffer,
+                                          struct radv_descriptor_set *set,
+                                          struct radv_descriptor_set_layout *layout)
+{
+	set->size = layout->size;
+	set->layout = layout;
+
+	if (cmd_buffer->push_descriptors.capacity < set->size) {
+		size_t new_size = MAX2(set->size, 1024);
+		new_size = MAX2(new_size, 2 * cmd_buffer->push_descriptors.capacity);
+		new_size = MIN2(new_size, 96 * MAX_PUSH_DESCRIPTORS);
+
+		free(set->mapped_ptr);
+		set->mapped_ptr = malloc(new_size);
+
+		if (!set->mapped_ptr) {
+			cmd_buffer->push_descriptors.capacity = 0;
+			cmd_buffer->record_fail = true;
+			return false;
+		}
+
+		cmd_buffer->push_descriptors.capacity = new_size;
+	}
+
+	return true;
+}
+
+void radv_CmdPushDescriptorSetKHR(
+	VkCommandBuffer                             commandBuffer,
+	VkPipelineBindPoint                         pipelineBindPoint,
+	VkPipelineLayout                            _layout,
+	uint32_t                                    set,
+	uint32_t                                    descriptorWriteCount,
+	const VkWriteDescriptorSet*                 pDescriptorWrites)
+{
+	RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
+	RADV_FROM_HANDLE(radv_pipeline_layout, layout, _layout);
+	struct radv_descriptor_set *push_set = &cmd_buffer->push_descriptors.set;
+
+	assert(layout->set[set].layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR);
+
+	if (!radv_init_push_descriptor_set(cmd_buffer, push_set, layout->set[set].layout))
+		return;
+
+	radv_update_descriptor_sets(cmd_buffer->device, cmd_buffer,
+	                            radv_descriptor_set_to_handle(push_set),
+	                            descriptorWriteCount, pDescriptorWrites, 0, NULL);
+
+	cmd_buffer->state.descriptors[set] = push_set;
+	cmd_buffer->state.descriptors_dirty |= (1 << set);
+	cmd_buffer->state.push_descriptors_dirty = true;
+}
+
 void radv_CmdPushConstants(VkCommandBuffer commandBuffer,
 			   VkPipelineLayout layout,
 			   VkShaderStageFlags stageFlags,
diff --git a/src/amd/vulkan/radv_descriptor_set.c b/src/amd/vulkan/radv_descriptor_set.c
index aefe0c38081..0ff9aa206d2 100644
--- a/src/amd/vulkan/radv_descriptor_set.c
+++ b/src/amd/vulkan/radv_descriptor_set.c
@@ -59,6 +59,8 @@ VkResult radv_CreateDescriptorSetLayout(
 	if (!set_layout)
 		return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 
+	set_layout->flags = pCreateInfo->flags;
+
 	/* We just allocate all the samplers at the end of the struct */
 	uint32_t *samplers = (uint32_t*)&set_layout->binding[max_binding + 1];
 
@@ -79,6 +81,7 @@ VkResult radv_CreateDescriptorSetLayout(
 		switch (binding->descriptorType) {
 		case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
 		case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
+			assert(!(pCreateInfo->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
 			set_layout->binding[b].dynamic_offset_count = 1;
 			set_layout->dynamic_shader_stages |= binding->stageFlags;
 			set_layout->binding[b].size = 0;
@@ -499,6 +502,8 @@ VkResult radv_AllocateDescriptorSets(
 		RADV_FROM_HANDLE(radv_descriptor_set_layout, layout,
 				 pAllocateInfo->pSetLayouts[i]);
 
+		assert(!(layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
+
 		result = radv_descriptor_set_create(device, pool, NULL, layout, &set);
 		if (result != VK_SUCCESS)
 			break;
@@ -656,6 +661,12 @@ void radv_update_descriptor_sets(
 			set->layout->binding + writeset->dstBinding;
 		uint32_t *ptr = set->mapped_ptr;
 		struct radeon_winsys_bo **buffer_list =  set->descriptors;
+		/* Immutable samplers are not copied into push descriptors when they are
+		 * allocated, so if we are writing push descriptors we have to copy the
+		 * immutable samplers into them now.
+		 */
+		const bool copy_immutable_samplers = cmd_buffer &&
+			binding_layout->immutable_samplers && !binding_layout->immutable_samplers_equal;
 
 		ptr += binding_layout->offset / 4;
 		ptr += binding_layout->size * writeset->dstArrayElement / 4;
@@ -667,6 +678,7 @@ void radv_update_descriptor_sets(
 			case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: {
 				unsigned idx = writeset->dstArrayElement + j;
 				idx += binding_layout->dynamic_offset_offset;
+				assert(!(set->layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
 				write_dynamic_buffer_descriptor(device, set->dynamic_descriptors + idx,
 								buffer_list, writeset->pBufferInfo + j);
 				break;
@@ -691,11 +703,19 @@ void radv_update_descriptor_sets(
 				write_combined_image_sampler_descriptor(device, cmd_buffer, ptr, buffer_list,
 									writeset->pImageInfo + j,
 									!binding_layout->immutable_samplers);
+				if (copy_immutable_samplers) {
+					const unsigned idx = writeset->dstArrayElement + j;
+					memcpy(ptr + 16, binding_layout->immutable_samplers + 4 * idx, 16);
+				}
 				break;
 			case VK_DESCRIPTOR_TYPE_SAMPLER:
-				if (!binding_layout->immutable_samplers)
+				if (!binding_layout->immutable_samplers) {
 					write_sampler_descriptor(device, ptr,
 					                         writeset->pImageInfo + j);
+				} else if (copy_immutable_samplers) {
+					unsigned idx = writeset->dstArrayElement + j;
+					memcpy(ptr, binding_layout->immutable_samplers + 4 * idx, 16);
+				}
 				break;
 			default:
 				unreachable("unimplemented descriptor type");
diff --git a/src/amd/vulkan/radv_descriptor_set.h b/src/amd/vulkan/radv_descriptor_set.h
index c373489ccb3..b3bbb193e80 100644
--- a/src/amd/vulkan/radv_descriptor_set.h
+++ b/src/amd/vulkan/radv_descriptor_set.h
@@ -49,6 +49,9 @@ struct radv_descriptor_set_binding_layout {
 };
 
 struct radv_descriptor_set_layout {
+   /* The create flags for this descriptor set layout */
+   VkDescriptorSetLayoutCreateFlags flags;
+
    /* Number of bindings in this descriptor set */
    uint16_t binding_count;
 
diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c
index 779e47dc9ea..059c013f359 100644
--- a/src/amd/vulkan/radv_device.c
+++ b/src/amd/vulkan/radv_device.c
@@ -100,6 +100,10 @@ static const VkExtensionProperties common_device_extensions[] = {
 		.specVersion = 1,
 	},
 	{
+		.extensionName = VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME,
+		.specVersion = 1,
+	},
+	{
 		.extensionName = VK_KHR_SAMPLER_MIRROR_CLAMP_TO_EDGE_EXTENSION_NAME,
 		.specVersion = 1,
 	},
@@ -665,7 +669,20 @@ void radv_GetPhysicalDeviceProperties2KHR(
 	VkPhysicalDevice                            physicalDevice,
 	VkPhysicalDeviceProperties2KHR             *pProperties)
 {
-	return radv_GetPhysicalDeviceProperties(physicalDevice, &pProperties->properties);
+	radv_GetPhysicalDeviceProperties(physicalDevice, &pProperties->properties);
+
+	vk_foreach_struct(ext, pProperties->pNext) {
+		switch (ext->sType) {
+		case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR: {
+			VkPhysicalDevicePushDescriptorPropertiesKHR *properties =
+				(VkPhysicalDevicePushDescriptorPropertiesKHR *) ext;
+			properties->maxPushDescriptors = MAX_PUSH_DESCRIPTORS;
+			break;
+		}
+		default:
+			break;
+		}
+	}
 }
 
 static void radv_get_physical_device_queue_family_properties(
diff --git a/src/amd/vulkan/radv_entrypoints_gen.py b/src/amd/vulkan/radv_entrypoints_gen.py
index bee29dc1202..4651c63dcf8 100644
--- a/src/amd/vulkan/radv_entrypoints_gen.py
+++ b/src/amd/vulkan/radv_entrypoints_gen.py
@@ -33,6 +33,7 @@ supported_extensions = [
    'VK_KHR_get_physical_device_properties2',
    'VK_KHR_incremental_present',
    'VK_KHR_maintenance1',
+   'VK_KHR_push_descriptor',
    'VK_KHR_sampler_mirror_clamp_to_edge',
    'VK_KHR_shader_draw_parameters',
    'VK_KHR_surface',
diff --git a/src/amd/vulkan/radv_private.h b/src/amd/vulkan/radv_private.h
index cba1891d1e5..e39819fff3e 100644
--- a/src/amd/vulkan/radv_private.h
+++ b/src/amd/vulkan/radv_private.h
@@ -79,6 +79,7 @@ typedef uint32_t xcb_window_t;
 #define MAX_VIEWPORTS   16
 #define MAX_SCISSORS    16
 #define MAX_PUSH_CONSTANTS_SIZE 128
+#define MAX_PUSH_DESCRIPTORS 32
 #define MAX_DYNAMIC_BUFFERS 16
 #define MAX_SAMPLES_LOG2 4
 #define NUM_META_FS_KEYS 11
@@ -549,6 +550,12 @@ struct radv_descriptor_set {
 	struct radeon_winsys_bo *descriptors[0];
 };
 
+struct radv_push_descriptor_set
+{
+	struct radv_descriptor_set set;
+	uint32_t capacity;
+};
+
 struct radv_descriptor_pool {
 	struct radeon_winsys_bo *bo;
 	uint8_t *mapped_ptr;
@@ -682,6 +689,7 @@ struct radv_cmd_state {
 	uint32_t                                      vb_dirty;
 	radv_cmd_dirty_mask_t                         dirty;
 	bool                                          vertex_descriptors_dirty;
+	bool                                          push_descriptors_dirty;
 
 	struct radv_pipeline *                        pipeline;
 	struct radv_pipeline *                        emitted_pipeline;
@@ -739,6 +747,7 @@ struct radv_cmd_buffer {
 	uint8_t push_constants[MAX_PUSH_CONSTANTS_SIZE];
 	uint32_t dynamic_buffers[4 * MAX_DYNAMIC_BUFFERS];
 	VkShaderStageFlags push_constant_stages;
+	struct radv_push_descriptor_set push_descriptors;
 
 	struct radv_cmd_buffer_upload upload;