radv: implement depth/stencil expand on compute

This works as long as the image is TC-compatible HTILE.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12452>
This commit is contained in:
Samuel Pitoiset 2021-08-18 17:02:09 +02:00 committed by Marge Bot
parent 966b780992
commit 0ac8731b6c
2 changed files with 243 additions and 2 deletions

View File

@ -33,6 +33,123 @@ enum radv_depth_op {
DEPTH_RESUMMARIZE,
};
static nir_shader *
build_expand_depth_stencil_compute_shader(struct radv_device *dev)
{
const struct glsl_type *img_type = glsl_image_type(GLSL_SAMPLER_DIM_2D, false, GLSL_TYPE_FLOAT);
nir_builder b =
nir_builder_init_simple_shader(MESA_SHADER_COMPUTE, NULL, "expand_depth_stencil_compute");
/* We need at least 8/8/1 to cover an entire HTILE block in a single workgroup. */
b.shader->info.workgroup_size[0] = 8;
b.shader->info.workgroup_size[1] = 8;
b.shader->info.workgroup_size[2] = 1;
nir_variable *input_img = nir_variable_create(b.shader, nir_var_uniform, img_type, "in_img");
input_img->data.descriptor_set = 0;
input_img->data.binding = 0;
nir_variable *output_img = nir_variable_create(b.shader, nir_var_uniform, img_type, "out_img");
output_img->data.descriptor_set = 0;
output_img->data.binding = 1;
nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b);
nir_ssa_def *wg_id = nir_load_workgroup_id(&b, 32);
nir_ssa_def *block_size =
nir_imm_ivec4(&b, b.shader->info.workgroup_size[0], b.shader->info.workgroup_size[1],
b.shader->info.workgroup_size[2], 0);
nir_ssa_def *global_id = nir_iadd(&b, nir_imul(&b, wg_id, block_size), invoc_id);
nir_ssa_def *data = nir_image_deref_load(
&b, 4, 32, &nir_build_deref_var(&b, input_img)->dest.ssa, global_id, nir_ssa_undef(&b, 1, 32),
nir_imm_int(&b, 0), .image_dim = GLSL_SAMPLER_DIM_2D);
/* We need a NIR_SCOPE_DEVICE memory_scope because ACO will avoid
* creating a vmcnt(0) because it expects the L1 cache to keep memory
* operations in-order for the same workgroup. The vmcnt(0) seems
* necessary however. */
nir_scoped_barrier(&b, .execution_scope = NIR_SCOPE_WORKGROUP, .memory_scope = NIR_SCOPE_DEVICE,
.memory_semantics = NIR_MEMORY_ACQ_REL, .memory_modes = nir_var_mem_ssbo);
nir_image_deref_store(&b, &nir_build_deref_var(&b, output_img)->dest.ssa, global_id,
nir_ssa_undef(&b, 1, 32), data, nir_imm_int(&b, 0),
.image_dim = GLSL_SAMPLER_DIM_2D);
return b.shader;
}
static VkResult
create_expand_depth_stencil_compute(struct radv_device *device)
{
VkResult result = VK_SUCCESS;
nir_shader *cs = build_expand_depth_stencil_compute_shader(device);
VkDescriptorSetLayoutCreateInfo ds_create_info = {
.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
.flags = VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR,
.bindingCount = 2,
.pBindings = (VkDescriptorSetLayoutBinding[]){
{.binding = 0,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = NULL},
{.binding = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = NULL},
}};
result = radv_CreateDescriptorSetLayout(
radv_device_to_handle(device), &ds_create_info, &device->meta_state.alloc,
&device->meta_state.expand_depth_stencil_compute_ds_layout);
if (result != VK_SUCCESS)
goto cleanup;
VkPipelineLayoutCreateInfo pl_create_info = {
.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
.setLayoutCount = 1,
.pSetLayouts = &device->meta_state.expand_depth_stencil_compute_ds_layout,
.pushConstantRangeCount = 0,
.pPushConstantRanges = NULL,
};
result = radv_CreatePipelineLayout(
radv_device_to_handle(device), &pl_create_info, &device->meta_state.alloc,
&device->meta_state.expand_depth_stencil_compute_p_layout);
if (result != VK_SUCCESS)
goto cleanup;
/* compute shader */
VkPipelineShaderStageCreateInfo pipeline_shader_stage = {
.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
.stage = VK_SHADER_STAGE_COMPUTE_BIT,
.module = vk_shader_module_handle_from_nir(cs),
.pName = "main",
.pSpecializationInfo = NULL,
};
VkComputePipelineCreateInfo vk_pipeline_info = {
.sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO,
.stage = pipeline_shader_stage,
.flags = 0,
.layout = device->meta_state.expand_depth_stencil_compute_p_layout,
};
result = radv_CreateComputePipelines(
radv_device_to_handle(device), radv_pipeline_cache_to_handle(&device->meta_state.cache), 1,
&vk_pipeline_info, NULL,
&device->meta_state.expand_depth_stencil_compute_pipeline);
if (result != VK_SUCCESS)
goto cleanup;
cleanup:
ralloc_free(cs);
return result;
}
static VkResult
create_pass(struct radv_device *device, uint32_t samples, VkRenderPass *pass)
{
@ -263,6 +380,13 @@ radv_device_finish_meta_depth_decomp_state(struct radv_device *device)
radv_DestroyPipeline(radv_device_to_handle(device),
state->depth_decomp[i].resummarize_pipeline, &state->alloc);
}
radv_DestroyPipeline(radv_device_to_handle(device),
state->expand_depth_stencil_compute_pipeline, &state->alloc);
radv_DestroyPipelineLayout(radv_device_to_handle(device),
state->expand_depth_stencil_compute_p_layout, &state->alloc);
radv_DestroyDescriptorSetLayout(radv_device_to_handle(device),
state->expand_depth_stencil_compute_ds_layout, &state->alloc);
}
VkResult
@ -298,6 +422,10 @@ radv_device_init_meta_depth_decomp_state(struct radv_device *device, bool on_dem
goto fail;
}
res = create_expand_depth_stencil_compute(device);
if (res != VK_SUCCESS)
goto fail;
return VK_SUCCESS;
fail:
@ -481,6 +609,112 @@ radv_process_depth_stencil(struct radv_cmd_buffer *cmd_buffer, struct radv_image
radv_meta_restore(&saved_state, cmd_buffer);
}
static void
radv_expand_depth_stencil_compute(struct radv_cmd_buffer *cmd_buffer, struct radv_image *image,
const VkImageSubresourceRange *subresourceRange)
{
struct radv_meta_saved_state saved_state;
struct radv_image_view load_iview = {0};
struct radv_image_view store_iview = {0};
struct radv_device *device = cmd_buffer->device;
assert(radv_image_is_tc_compat_htile(image));
cmd_buffer->state.flush_bits |=
radv_dst_access_flush(cmd_buffer, VK_ACCESS_SHADER_WRITE_BIT, image);
radv_meta_save(&saved_state, cmd_buffer,
RADV_META_SAVE_DESCRIPTORS | RADV_META_SAVE_COMPUTE_PIPELINE);
radv_CmdBindPipeline(radv_cmd_buffer_to_handle(cmd_buffer), VK_PIPELINE_BIND_POINT_COMPUTE,
device->meta_state.expand_depth_stencil_compute_pipeline);
for (uint32_t l = 0; l < radv_get_levelCount(image, subresourceRange); l++) {
uint32_t width, height;
/* Do not decompress levels without HTILE. */
if (!radv_htile_enabled(image, subresourceRange->baseMipLevel + l))
continue;
width = radv_minify(image->info.width, subresourceRange->baseMipLevel + l);
height = radv_minify(image->info.height, subresourceRange->baseMipLevel + l);
for (uint32_t s = 0; s < radv_get_layerCount(image, subresourceRange); s++) {
radv_image_view_init(
&load_iview, cmd_buffer->device,
&(VkImageViewCreateInfo){
.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
.image = radv_image_to_handle(image),
.viewType = VK_IMAGE_VIEW_TYPE_2D,
.format = image->vk_format,
.subresourceRange = {.aspectMask = subresourceRange->aspectMask,
.baseMipLevel = subresourceRange->baseMipLevel + l,
.levelCount = 1,
.baseArrayLayer = subresourceRange->baseArrayLayer + s,
.layerCount = 1},
},
&(struct radv_image_view_extra_create_info){.enable_compression = true});
radv_image_view_init(
&store_iview, cmd_buffer->device,
&(VkImageViewCreateInfo){
.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
.image = radv_image_to_handle(image),
.viewType = VK_IMAGE_VIEW_TYPE_2D,
.format = image->vk_format,
.subresourceRange = {.aspectMask = subresourceRange->aspectMask,
.baseMipLevel = subresourceRange->baseMipLevel + l,
.levelCount = 1,
.baseArrayLayer = subresourceRange->baseArrayLayer + s,
.layerCount = 1},
},
&(struct radv_image_view_extra_create_info){.disable_compression = true});
radv_meta_push_descriptor_set(
cmd_buffer, VK_PIPELINE_BIND_POINT_COMPUTE,
device->meta_state.expand_depth_stencil_compute_p_layout, 0, /* set */
2, /* descriptorWriteCount */
(VkWriteDescriptorSet[]){{.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
.dstBinding = 0,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
.pImageInfo =
(VkDescriptorImageInfo[]){
{
.sampler = VK_NULL_HANDLE,
.imageView = radv_image_view_to_handle(&load_iview),
.imageLayout = VK_IMAGE_LAYOUT_GENERAL,
},
}},
{.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
.dstBinding = 1,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
.pImageInfo = (VkDescriptorImageInfo[]){
{
.sampler = VK_NULL_HANDLE,
.imageView = radv_image_view_to_handle(&store_iview),
.imageLayout = VK_IMAGE_LAYOUT_GENERAL,
},
}}});
radv_unaligned_dispatch(cmd_buffer, width, height, 1);
}
}
radv_meta_restore(&saved_state, cmd_buffer);
cmd_buffer->state.flush_bits |=
RADV_CMD_FLAG_CS_PARTIAL_FLUSH | RADV_CMD_FLAG_INV_VCACHE |
radv_src_access_flush(cmd_buffer, VK_ACCESS_SHADER_WRITE_BIT, image);
/* Initialize the HTILE metadata as "fully expanded". */
uint32_t htile_value = radv_get_htile_initial_value(cmd_buffer->device, image);
cmd_buffer->state.flush_bits |= radv_clear_htile(cmd_buffer, image, subresourceRange, htile_value);
}
void
radv_expand_depth_stencil(struct radv_cmd_buffer *cmd_buffer, struct radv_image *image,
const VkImageSubresourceRange *subresourceRange,
@ -491,8 +725,11 @@ radv_expand_depth_stencil(struct radv_cmd_buffer *cmd_buffer, struct radv_image
barrier.layout_transitions.depth_stencil_expand = 1;
radv_describe_layout_transition(cmd_buffer, &barrier);
assert(cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL);
radv_process_depth_stencil(cmd_buffer, image, subresourceRange, sample_locs, DEPTH_DECOMPRESS);
if (cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL) {
radv_process_depth_stencil(cmd_buffer, image, subresourceRange, sample_locs, DEPTH_DECOMPRESS);
} else {
radv_expand_depth_stencil_compute(cmd_buffer, image, subresourceRange);
}
}
void

View File

@ -619,6 +619,10 @@ struct radv_meta_state {
VkRenderPass pass;
} depth_decomp[MAX_SAMPLES_LOG2];
VkDescriptorSetLayout expand_depth_stencil_compute_ds_layout;
VkPipelineLayout expand_depth_stencil_compute_p_layout;
VkPipeline expand_depth_stencil_compute_pipeline;
struct {
VkPipelineLayout p_layout;
VkPipeline cmask_eliminate_pipeline;