{
const struct isl_device *isl_dev = &cmd_buffer->device->isl_dev;
- anv_reloc_list_add(&cmd_buffer->surface_relocs, &cmd_buffer->pool->alloc,
- state.offset + isl_dev->ss.addr_offset, bo, offset);
+ VkResult result =
+ anv_reloc_list_add(&cmd_buffer->surface_relocs, &cmd_buffer->pool->alloc,
+ state.offset + isl_dev->ss.addr_offset, bo, offset);
+ if (result != VK_SUCCESS)
+ anv_batch_set_error(&cmd_buffer->batch, result);
}
static void
{
const struct isl_device *isl_dev = &cmd_buffer->device->isl_dev;
- anv_reloc_list_add(&cmd_buffer->surface_relocs, &cmd_buffer->pool->alloc,
- state.offset + isl_dev->ss.addr_offset,
- iview->bo, iview->offset);
+ add_surface_state_reloc(cmd_buffer, state, iview->bo, iview->offset);
if (aux_usage != ISL_AUX_USAGE_NONE) {
uint32_t aux_offset = iview->offset + iview->image->aux_surface.offset;
uint32_t *aux_addr_dw = state.map + isl_dev->ss.aux_addr_offset;
aux_offset += *aux_addr_dw & 0xfff;
- anv_reloc_list_add(&cmd_buffer->surface_relocs, &cmd_buffer->pool->alloc,
- state.offset + isl_dev->ss.aux_addr_offset,
- iview->bo, aux_offset);
+ VkResult result =
+ anv_reloc_list_add(&cmd_buffer->surface_relocs,
+ &cmd_buffer->pool->alloc,
+ state.offset + isl_dev->ss.aux_addr_offset,
+ iview->bo, aux_offset);
+ if (result != VK_SUCCESS)
+ anv_batch_set_error(&cmd_buffer->batch, result);
}
}
att_state->input_aux_usage = ISL_AUX_USAGE_NONE;
att_state->fast_clear = false;
return;
+ } else if (iview->image->aux_usage == ISL_AUX_USAGE_MCS) {
+ att_state->aux_usage = ISL_AUX_USAGE_MCS;
+ att_state->input_aux_usage = ISL_AUX_USAGE_MCS;
+ att_state->fast_clear = false;
+ return;
}
assert(iview->image->aux_surface.isl.usage & ISL_SURF_USAGE_CCS_BIT);
* doesn't also support color compression.
*/
att_state->input_aux_usage = ISL_AUX_USAGE_NONE;
- } else if (GEN_GEN == 8) {
- /* Broadwell can sample from fast-cleared images */
+ } else if (GEN_GEN >= 8) {
+ /* Broadwell/Skylake can sample from fast-cleared images */
att_state->input_aux_usage = ISL_AUX_USAGE_CCS_D;
} else {
/* Ivy Bridge and Haswell cannot */
return vk_format_is_color(att->format);
}
-static enum isl_aux_usage
-layout_to_hiz_usage(VkImageLayout layout, uint8_t samples)
-{
- switch (layout) {
- case VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL:
- return ISL_AUX_USAGE_HIZ;
- case VK_IMAGE_LAYOUT_DEPTH_STENCIL_READ_ONLY_OPTIMAL:
- case VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL:
- if (anv_can_sample_with_hiz(GEN_GEN, samples))
- return ISL_AUX_USAGE_HIZ;
- /* Fall-through */
- case VK_IMAGE_LAYOUT_GENERAL:
- /* This buffer could be used as a source or destination in a transfer
- * operation. Transfer operations current don't perform HiZ-enabled reads
- * and writes.
- */
- default:
- return ISL_AUX_USAGE_NONE;
- }
-}
-
/* Transitions a HiZ-enabled depth buffer from one layout to another. Unless
* the initial layout is undefined, the HiZ buffer and depth buffer will
* represent the same data at the end of this operation.
{
assert(image);
- if (image->aux_usage != ISL_AUX_USAGE_HIZ || final_layout == initial_layout)
+ /* A transition is a no-op if HiZ is not enabled, or if the initial and
+ * final layouts are equal.
+ *
+ * The undefined layout indicates that the user doesn't care about the data
+ * that's currently in the buffer. Therefore, a data-preserving resolve
+ * operation is not needed.
+ *
+ * The pre-initialized layout is equivalent to the undefined layout for
+ * optimally-tiled images. Anv only exposes support for optimally-tiled
+ * depth buffers.
+ */
+ if (image->aux_usage != ISL_AUX_USAGE_HIZ ||
+ initial_layout == final_layout ||
+ initial_layout == VK_IMAGE_LAYOUT_UNDEFINED ||
+ initial_layout == VK_IMAGE_LAYOUT_PREINITIALIZED)
return;
- const bool hiz_enabled = layout_to_hiz_usage(initial_layout, image->samples) ==
- ISL_AUX_USAGE_HIZ;
- const bool enable_hiz = layout_to_hiz_usage(final_layout, image->samples) ==
- ISL_AUX_USAGE_HIZ;
+ const bool hiz_enabled = ISL_AUX_USAGE_HIZ ==
+ anv_layout_to_aux_usage(&cmd_buffer->device->info, image, image->aspects,
+ initial_layout);
+ const bool enable_hiz = ISL_AUX_USAGE_HIZ ==
+ anv_layout_to_aux_usage(&cmd_buffer->device->info, image, image->aspects,
+ final_layout);
enum blorp_hiz_op hiz_op;
- if (initial_layout == VK_IMAGE_LAYOUT_UNDEFINED) {
- /* We've already initialized the aux HiZ buffer at BindImageMemory time,
- * so there's no need to perform a HIZ resolve or clear to avoid GPU hangs.
- * This initial layout indicates that the user doesn't care about the data
- * that's currently in the buffer, so resolves are not necessary except
- * for the special case noted below.
- */
- hiz_op = BLORP_HIZ_OP_NONE;
- } else if (hiz_enabled && !enable_hiz) {
+ if (hiz_enabled && !enable_hiz) {
hiz_op = BLORP_HIZ_OP_DEPTH_RESOLVE;
} else if (!hiz_enabled && enable_hiz) {
hiz_op = BLORP_HIZ_OP_HIZ_RESOLVE;
} else {
assert(hiz_enabled == enable_hiz);
- /* If the same buffer will be used, no resolves are necessary except for
- * the special case noted below.
- */
+ /* If the same buffer will be used, no resolves are necessary. */
hiz_op = BLORP_HIZ_OP_NONE;
}
if (hiz_op != BLORP_HIZ_OP_NONE)
anv_gen8_hiz_op_resolve(cmd_buffer, image, hiz_op);
-
- /* Images that have sampling with HiZ enabled cause all shader sampling to
- * load data with the HiZ buffer. Therefore, in the case of transitioning to
- * the general layout - which currently routes all writes to the depth
- * buffer - we must ensure that the HiZ buffer remains consistent with the
- * depth buffer by performing an additional HIZ resolve if the operation
- * required by this transition was not already a HiZ resolve.
- */
- if (final_layout == VK_IMAGE_LAYOUT_GENERAL &&
- anv_can_sample_with_hiz(GEN_GEN, image->samples) &&
- hiz_op != BLORP_HIZ_OP_HIZ_RESOLVE) {
- anv_gen8_hiz_op_resolve(cmd_buffer, image, BLORP_HIZ_OP_HIZ_RESOLVE);
- }
}
/**
* Setup anv_cmd_state::attachments for vkCmdBeginRenderPass.
*/
-static void
+static VkResult
genX(cmd_buffer_setup_attachments)(struct anv_cmd_buffer *cmd_buffer,
struct anv_render_pass *pass,
const VkRenderPassBeginInfo *begin)
if (pass->attachment_count == 0) {
state->attachments = NULL;
- return;
+ return VK_SUCCESS;
}
state->attachments = vk_alloc(&cmd_buffer->pool->alloc,
sizeof(state->attachments[0]),
8, VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
if (state->attachments == NULL) {
- /* FIXME: Propagate VK_ERROR_OUT_OF_HOST_MEMORY to vkEndCommandBuffer */
- abort();
+ /* Propagate VK_ERROR_OUT_OF_HOST_MEMORY to vkEndCommandBuffer */
+ return anv_batch_set_error(&cmd_buffer->batch,
+ VK_ERROR_OUT_OF_HOST_MEMORY);
}
bool need_null_state = false;
state->attachments[i].clear_value = begin->pClearValues[i];
struct anv_image_view *iview = framebuffer->attachments[i];
- assert(iview->vk_format == att->format);
+ anv_assert(iview->vk_format == att->format);
union isl_color_value clear_color = { .u32 = { 0, } };
if (att_aspects == VK_IMAGE_ASPECT_COLOR_BIT) {
state->attachments[i].aux_usage,
state->attachments[i].color_rt_state);
} else {
- if (iview->image->aux_usage == ISL_AUX_USAGE_HIZ) {
- state->attachments[i].aux_usage =
- layout_to_hiz_usage(att->initial_layout, iview->image->samples);
- } else {
- state->attachments[i].aux_usage = ISL_AUX_USAGE_NONE;
- }
+ /* This field will be initialized after the first subpass
+ * transition.
+ */
+ state->attachments[i].aux_usage = ISL_AUX_USAGE_NONE;
+
state->attachments[i].input_aux_usage = ISL_AUX_USAGE_NONE;
}
}
}
- if (!cmd_buffer->device->info.has_llc)
- anv_state_flush(state->render_pass_states);
+ anv_state_flush(cmd_buffer->device, state->render_pass_states);
}
+
+ return VK_SUCCESS;
}
VkResult
genX(cmd_buffer_emit_state_base_address)(cmd_buffer);
+ VkResult result = VK_SUCCESS;
if (cmd_buffer->usage_flags &
VK_COMMAND_BUFFER_USAGE_RENDER_PASS_CONTINUE_BIT) {
cmd_buffer->state.pass =
&cmd_buffer->state.pass->subpasses[pBeginInfo->pInheritanceInfo->subpass];
cmd_buffer->state.framebuffer = NULL;
- genX(cmd_buffer_setup_attachments)(cmd_buffer, cmd_buffer->state.pass,
- NULL);
+ result = genX(cmd_buffer_setup_attachments)(cmd_buffer,
+ cmd_buffer->state.pass, NULL);
cmd_buffer->state.dirty |= ANV_CMD_DIRTY_RENDER_TARGETS;
}
- return VK_SUCCESS;
+ return result;
}
VkResult
{
ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
+ if (anv_batch_has_error(&cmd_buffer->batch))
+ return cmd_buffer->batch.status;
+
/* We want every command buffer to start with the PMA fix in a known state,
* so we disable it at the end of the command buffer.
*/
assert(primary->level == VK_COMMAND_BUFFER_LEVEL_PRIMARY);
+ if (anv_batch_has_error(&primary->batch))
+ return;
+
/* The secondary command buffers will assume that the PMA fix is disabled
* when they begin executing. Make sure this is true.
*/
genX(cmd_buffer_enable_pma_fix)(primary, false);
+ /* The secondary command buffer doesn't know which textures etc. have been
+ * flushed prior to their execution. Apply those flushes now.
+ */
+ genX(cmd_buffer_apply_pipe_flushes)(primary);
+
for (uint32_t i = 0; i < commandBufferCount; i++) {
ANV_FROM_HANDLE(anv_cmd_buffer, secondary, pCmdBuffers[i]);
assert(secondary->level == VK_COMMAND_BUFFER_LEVEL_SECONDARY);
+ assert(!anv_batch_has_error(&secondary->batch));
if (secondary->usage_flags &
VK_COMMAND_BUFFER_USAGE_RENDER_PASS_CONTINUE_BIT) {
const VkImageMemoryBarrier* pImageMemoryBarriers)
{
ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
- uint32_t b;
/* XXX: Right now, we're really dumb and just flush whatever categories
* the app asks for. One of these days we may make this a bit better
}
}
- enum anv_pipe_bits pipe_bits = 0;
-
- for_each_bit(b, src_flags) {
- switch ((VkAccessFlagBits)(1 << b)) {
- case VK_ACCESS_SHADER_WRITE_BIT:
- pipe_bits |= ANV_PIPE_DATA_CACHE_FLUSH_BIT;
- break;
- case VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT:
- pipe_bits |= ANV_PIPE_RENDER_TARGET_CACHE_FLUSH_BIT;
- break;
- case VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT:
- pipe_bits |= ANV_PIPE_DEPTH_CACHE_FLUSH_BIT;
- break;
- case VK_ACCESS_TRANSFER_WRITE_BIT:
- pipe_bits |= ANV_PIPE_RENDER_TARGET_CACHE_FLUSH_BIT;
- pipe_bits |= ANV_PIPE_DEPTH_CACHE_FLUSH_BIT;
- break;
- default:
- break; /* Nothing to do */
- }
- }
-
- for_each_bit(b, dst_flags) {
- switch ((VkAccessFlagBits)(1 << b)) {
- case VK_ACCESS_INDIRECT_COMMAND_READ_BIT:
- case VK_ACCESS_INDEX_READ_BIT:
- case VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT:
- pipe_bits |= ANV_PIPE_VF_CACHE_INVALIDATE_BIT;
- break;
- case VK_ACCESS_UNIFORM_READ_BIT:
- pipe_bits |= ANV_PIPE_CONSTANT_CACHE_INVALIDATE_BIT;
- pipe_bits |= ANV_PIPE_TEXTURE_CACHE_INVALIDATE_BIT;
- break;
- case VK_ACCESS_SHADER_READ_BIT:
- case VK_ACCESS_INPUT_ATTACHMENT_READ_BIT:
- case VK_ACCESS_TRANSFER_READ_BIT:
- pipe_bits |= ANV_PIPE_TEXTURE_CACHE_INVALIDATE_BIT;
- break;
- default:
- break; /* Nothing to do */
- }
- }
-
- cmd_buffer->state.pending_pipe_bits |= pipe_bits;
+ cmd_buffer->state.pending_pipe_bits |=
+ anv_pipe_flush_bits_for_access_flags(src_flags) |
+ anv_pipe_invalidate_bits_for_access_flags(dst_flags);
}
static void
assert(stage == MESA_SHADER_FRAGMENT);
assert(binding->binding == 0);
if (binding->index < subpass->color_count) {
- const unsigned att = subpass->color_attachments[binding->index];
+ const unsigned att = subpass->color_attachments[binding->index].attachment;
surface_state = cmd_buffer->state.attachments[att].color_rt_state;
} else {
surface_state = cmd_buffer->state.null_surface_state;
case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
- surface_state = desc->image_view->sampler_surface_state;
+ surface_state = desc->aux_usage == ISL_AUX_USAGE_NONE ?
+ desc->image_view->no_aux_sampler_surface_state :
+ desc->image_view->sampler_surface_state;
assert(surface_state.alloc_size);
add_image_view_relocs(cmd_buffer, desc->image_view,
- desc->image_view->image->aux_usage,
- surface_state);
+ desc->aux_usage, surface_state);
break;
-
case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
assert(stage == MESA_SHADER_FRAGMENT);
if (desc->image_view->aspect_mask != VK_IMAGE_ASPECT_COLOR_BIT) {
/* For depth and stencil input attachments, we treat it like any
* old texture that a user may have bound.
*/
- surface_state = desc->image_view->sampler_surface_state;
+ surface_state = desc->aux_usage == ISL_AUX_USAGE_NONE ?
+ desc->image_view->no_aux_sampler_surface_state :
+ desc->image_view->sampler_surface_state;
assert(surface_state.alloc_size);
add_image_view_relocs(cmd_buffer, desc->image_view,
- desc->image_view->image->aux_usage,
- surface_state);
+ desc->aux_usage, surface_state);
} else {
/* For color input attachments, we create the surface state at
* vkBeginRenderPass time so that we can include aux and clear
*/
assert(binding->input_attachment_index < subpass->input_count);
const unsigned subpass_att = binding->input_attachment_index;
- const unsigned att = subpass->input_attachments[subpass_att];
+ const unsigned att = subpass->input_attachments[subpass_att].attachment;
surface_state = cmd_buffer->state.attachments[att].input_att_state;
}
break;
case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
- case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
- case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
surface_state = desc->buffer_view->surface_state;
assert(surface_state.alloc_size);
desc->buffer_view->offset);
break;
+ case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
+ case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: {
+ uint32_t dynamic_offset_idx =
+ pipeline->layout->set[binding->set].dynamic_offset_start +
+ set->layout->binding[binding->binding].dynamic_offset_index +
+ binding->index;
+
+ /* Compute the offset within the buffer */
+ uint64_t offset = desc->offset +
+ cmd_buffer->state.dynamic_offsets[dynamic_offset_idx];
+ /* Clamp to the buffer size */
+ offset = MIN2(offset, desc->buffer->size);
+ /* Clamp the range to the buffer size */
+ uint32_t range = MIN2(desc->range, desc->buffer->size - offset);
+
+ surface_state =
+ anv_state_stream_alloc(&cmd_buffer->surface_state_stream, 64, 64);
+ enum isl_format format =
+ anv_isl_format_for_descriptor_type(desc->type);
+
+ anv_fill_buffer_surface_state(cmd_buffer->device, surface_state,
+ format, offset, range, 1);
+ add_surface_state_reloc(cmd_buffer, surface_state,
+ desc->buffer->bo,
+ desc->buffer->offset + offset);
+ break;
+ }
+
case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
surface_state = (binding->write_only)
? desc->buffer_view->writeonly_storage_surface_state
assert(image == map->image_count);
out:
- if (!cmd_buffer->device->info.has_llc)
- anv_state_flush(*bt_state);
+ anv_state_flush(cmd_buffer->device, *bt_state);
return VK_SUCCESS;
}
sampler->state, sizeof(sampler->state));
}
- if (!cmd_buffer->device->info.has_llc)
- anv_state_flush(*state);
+ anv_state_flush(cmd_buffer->device, *state);
return VK_SUCCESS;
}
assert(result == VK_ERROR_OUT_OF_DEVICE_MEMORY);
result = anv_cmd_buffer_new_binding_table_block(cmd_buffer);
- assert(result == VK_SUCCESS);
+ if (result != VK_SUCCESS)
+ return 0;
/* Re-emit state base addresses so we get the new surface state base
* address before we start emitting binding tables etc.
dirty |= cmd_buffer->state.pipeline->active_stages;
anv_foreach_stage(s, dirty) {
result = emit_samplers(cmd_buffer, s, &cmd_buffer->state.samplers[s]);
- if (result != VK_SUCCESS)
- return result;
+ if (result != VK_SUCCESS) {
+ anv_batch_set_error(&cmd_buffer->batch, result);
+ return 0;
+ }
result = emit_binding_table(cmd_buffer, s,
&cmd_buffer->state.binding_tables[s]);
- if (result != VK_SUCCESS)
- return result;
+ if (result != VK_SUCCESS) {
+ anv_batch_set_error(&cmd_buffer->batch, result);
+ return 0;
+ }
}
}
((uint32_t *)id_state.map)[0] = base_vertex;
((uint32_t *)id_state.map)[1] = base_instance;
- if (!cmd_buffer->device->info.has_llc)
- anv_state_flush(id_state);
+ anv_state_flush(cmd_buffer->device, id_state);
emit_base_vertex_instance_bo(cmd_buffer,
&cmd_buffer->device->dynamic_state_block_pool.bo, id_state.offset);
((uint32_t *)state.map)[0] = draw_index;
- if (!cmd_buffer->device->info.has_llc)
- anv_state_flush(state);
+ anv_state_flush(cmd_buffer->device, state);
emit_vertex_bo(cmd_buffer,
&cmd_buffer->device->dynamic_state_block_pool.bo,
struct anv_pipeline *pipeline = cmd_buffer->state.pipeline;
const struct brw_vs_prog_data *vs_prog_data = get_vs_prog_data(pipeline);
+ if (anv_batch_has_error(&cmd_buffer->batch))
+ return;
+
genX(cmd_buffer_flush_state)(cmd_buffer);
if (vs_prog_data->uses_basevertex || vs_prog_data->uses_baseinstance)
struct anv_pipeline *pipeline = cmd_buffer->state.pipeline;
const struct brw_vs_prog_data *vs_prog_data = get_vs_prog_data(pipeline);
+ if (anv_batch_has_error(&cmd_buffer->batch))
+ return;
+
genX(cmd_buffer_flush_state)(cmd_buffer);
if (vs_prog_data->uses_basevertex || vs_prog_data->uses_baseinstance)
struct anv_bo *bo = buffer->bo;
uint32_t bo_offset = buffer->offset + offset;
+ if (anv_batch_has_error(&cmd_buffer->batch))
+ return;
+
genX(cmd_buffer_flush_state)(cmd_buffer);
if (vs_prog_data->uses_basevertex || vs_prog_data->uses_baseinstance)
struct anv_bo *bo = buffer->bo;
uint32_t bo_offset = buffer->offset + offset;
+ if (anv_batch_has_error(&cmd_buffer->batch))
+ return;
+
genX(cmd_buffer_flush_state)(cmd_buffer);
/* TODO: We need to stomp base vertex to 0 somehow */
result = emit_binding_table(cmd_buffer, MESA_SHADER_COMPUTE, &surfaces);
if (result != VK_SUCCESS) {
assert(result == VK_ERROR_OUT_OF_DEVICE_MEMORY);
+
result = anv_cmd_buffer_new_binding_table_block(cmd_buffer);
- assert(result == VK_SUCCESS);
+ if (result != VK_SUCCESS)
+ return result;
/* Re-emit state base addresses so we get the new surface state base
* address before we start emitting binding tables etc.
genX(cmd_buffer_emit_state_base_address)(cmd_buffer);
result = emit_binding_table(cmd_buffer, MESA_SHADER_COMPUTE, &surfaces);
- assert(result == VK_SUCCESS);
+ if (result != VK_SUCCESS) {
+ anv_batch_set_error(&cmd_buffer->batch, result);
+ return result;
+ }
}
result = emit_samplers(cmd_buffer, MESA_SHADER_COMPUTE, &samplers);
- assert(result == VK_SUCCESS);
+ if (result != VK_SUCCESS) {
+ anv_batch_set_error(&cmd_buffer->batch, result);
+ return result;
+ }
uint32_t iface_desc_data_dw[GENX(INTERFACE_DESCRIPTOR_DATA_length)];
struct GENX(INTERFACE_DESCRIPTOR_DATA) desc = {
(cmd_buffer->state.compute_dirty & ANV_CMD_DIRTY_PIPELINE)) {
/* FIXME: figure out descriptors for gen7 */
result = flush_compute_descriptor_set(cmd_buffer);
- assert(result == VK_SUCCESS);
+ if (result != VK_SUCCESS)
+ return;
+
cmd_buffer->state.descriptors_dirty &= ~VK_SHADER_STAGE_COMPUTE_BIT;
}
struct anv_pipeline *pipeline = cmd_buffer->state.compute_pipeline;
const struct brw_cs_prog_data *prog_data = get_cs_prog_data(pipeline);
+ if (anv_batch_has_error(&cmd_buffer->batch))
+ return;
+
if (prog_data->uses_num_work_groups) {
struct anv_state state =
anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, 12, 4);
sizes[0] = x;
sizes[1] = y;
sizes[2] = z;
- if (!cmd_buffer->device->info.has_llc)
- anv_state_flush(state);
+ anv_state_flush(cmd_buffer->device, state);
cmd_buffer->state.num_workgroups_offset = state.offset;
cmd_buffer->state.num_workgroups_bo =
&cmd_buffer->device->dynamic_state_block_pool.bo;
}
static void
-flush_pipeline_before_pipeline_select(struct anv_cmd_buffer *cmd_buffer,
- uint32_t pipeline)
+genX(flush_pipeline_select)(struct anv_cmd_buffer *cmd_buffer,
+ uint32_t pipeline)
{
+ if (cmd_buffer->state.current_pipeline == pipeline)
+ return;
+
#if GEN_GEN >= 8 && GEN_GEN < 10
/* From the Broadwell PRM, Volume 2a: Instructions, PIPELINE_SELECT:
*
*/
if (pipeline == GPGPU)
anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_CC_STATE_POINTERS), t);
-#elif GEN_GEN <= 7
- /* From "BXML » GT » MI » vol1a GPU Overview » [Instruction]
- * PIPELINE_SELECT [DevBWR+]":
- *
- * Project: DEVSNB+
- *
- * Software must ensure all the write caches are flushed through a
- * stalling PIPE_CONTROL command followed by another PIPE_CONTROL
- * command to invalidate read only caches prior to programming
- * MI_PIPELINE_SELECT command to change the Pipeline Select Mode.
- */
- anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
- pc.RenderTargetCacheFlushEnable = true;
- pc.DepthCacheFlushEnable = true;
- pc.DCFlushEnable = true;
- pc.PostSyncOperation = NoWrite;
- pc.CommandStreamerStallEnable = true;
- }
+#endif
- anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
- pc.TextureCacheInvalidationEnable = true;
- pc.ConstantCacheInvalidationEnable = true;
- pc.StateCacheInvalidationEnable = true;
- pc.InstructionCacheInvalidateEnable = true;
- pc.PostSyncOperation = NoWrite;
- }
+ /* From "BXML » GT » MI » vol1a GPU Overview » [Instruction]
+ * PIPELINE_SELECT [DevBWR+]":
+ *
+ * Project: DEVSNB+
+ *
+ * Software must ensure all the write caches are flushed through a
+ * stalling PIPE_CONTROL command followed by another PIPE_CONTROL
+ * command to invalidate read only caches prior to programming
+ * MI_PIPELINE_SELECT command to change the Pipeline Select Mode.
+ */
+ anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
+ pc.RenderTargetCacheFlushEnable = true;
+ pc.DepthCacheFlushEnable = true;
+ pc.DCFlushEnable = true;
+ pc.PostSyncOperation = NoWrite;
+ pc.CommandStreamerStallEnable = true;
+ }
+
+ anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
+ pc.TextureCacheInvalidationEnable = true;
+ pc.ConstantCacheInvalidationEnable = true;
+ pc.StateCacheInvalidationEnable = true;
+ pc.InstructionCacheInvalidateEnable = true;
+ pc.PostSyncOperation = NoWrite;
+ }
+
+ anv_batch_emit(&cmd_buffer->batch, GENX(PIPELINE_SELECT), ps) {
+#if GEN_GEN >= 9
+ ps.MaskBits = 3;
#endif
+ ps.PipelineSelection = pipeline;
+ }
+
+ cmd_buffer->state.current_pipeline = pipeline;
}
void
genX(flush_pipeline_select_3d)(struct anv_cmd_buffer *cmd_buffer)
{
- if (cmd_buffer->state.current_pipeline != _3D) {
- flush_pipeline_before_pipeline_select(cmd_buffer, _3D);
-
- anv_batch_emit(&cmd_buffer->batch, GENX(PIPELINE_SELECT), ps) {
-#if GEN_GEN >= 9
- ps.MaskBits = 3;
-#endif
- ps.PipelineSelection = _3D;
- }
-
- cmd_buffer->state.current_pipeline = _3D;
- }
+ genX(flush_pipeline_select)(cmd_buffer, _3D);
}
void
genX(flush_pipeline_select_gpgpu)(struct anv_cmd_buffer *cmd_buffer)
{
- if (cmd_buffer->state.current_pipeline != GPGPU) {
- flush_pipeline_before_pipeline_select(cmd_buffer, GPGPU);
-
- anv_batch_emit(&cmd_buffer->batch, GENX(PIPELINE_SELECT), ps) {
-#if GEN_GEN >= 9
- ps.MaskBits = 3;
-#endif
- ps.PipelineSelection = GPGPU;
- }
-
- cmd_buffer->state.current_pipeline = GPGPU;
- }
+ genX(flush_pipeline_select)(cmd_buffer, GPGPU);
}
void
anv_cmd_buffer_get_depth_stencil_view(cmd_buffer);
const struct anv_image *image = iview ? iview->image : NULL;
const bool has_depth = image && (image->aspects & VK_IMAGE_ASPECT_DEPTH_BIT);
- const uint32_t ds = cmd_buffer->state.subpass->depth_stencil_attachment;
+ const uint32_t ds = cmd_buffer->state.subpass->depth_stencil_attachment.attachment;
const bool has_hiz = image != NULL &&
cmd_buffer->state.attachments[ds].aux_usage == ISL_AUX_USAGE_HIZ;
const bool has_stencil =
assert(image->depth_surface.isl.dim != ISL_SURF_DIM_3D);
db.Depth =
- db.RenderTargetViewExtent =
- iview->isl.array_len - iview->isl.base_array_layer - 1;
+ db.RenderTargetViewExtent = iview->isl.array_len - 1;
#if GEN_GEN >= 8
db.SurfaceQPitch =
}
}
+
+/**
+ * @brief Perform any layout transitions required at the beginning and/or end
+ * of the current subpass for depth buffers.
+ *
+ * TODO: Consider preprocessing the attachment reference array at render pass
+ * create time to determine if no layout transition is needed at the
+ * beginning and/or end of each subpass.
+ *
+ * @param cmd_buffer The command buffer the transition is happening within.
+ * @param subpass_end If true, marks that the transition is happening at the
+ * end of the subpass.
+ */
+static void
+cmd_buffer_subpass_transition_layouts(struct anv_cmd_buffer * const cmd_buffer,
+ const bool subpass_end)
+{
+ /* We need a non-NULL command buffer. */
+ assert(cmd_buffer);
+
+ const struct anv_cmd_state * const cmd_state = &cmd_buffer->state;
+ const struct anv_subpass * const subpass = cmd_state->subpass;
+
+ /* This function must be called within a subpass. */
+ assert(subpass);
+
+ /* If there are attachment references, the array shouldn't be NULL.
+ */
+ if (subpass->attachment_count > 0)
+ assert(subpass->attachments);
+
+ /* Iterate over the array of attachment references. */
+ for (const VkAttachmentReference *att_ref = subpass->attachments;
+ att_ref < subpass->attachments + subpass->attachment_count; att_ref++) {
+
+ /* If the attachment is unused, we can't perform a layout transition. */
+ if (att_ref->attachment == VK_ATTACHMENT_UNUSED)
+ continue;
+
+ /* This attachment index shouldn't go out of bounds. */
+ assert(att_ref->attachment < cmd_state->pass->attachment_count);
+
+ const struct anv_render_pass_attachment * const att_desc =
+ &cmd_state->pass->attachments[att_ref->attachment];
+ struct anv_attachment_state * const att_state =
+ &cmd_buffer->state.attachments[att_ref->attachment];
+
+ /* The attachment should not be used in a subpass after its last. */
+ assert(att_desc->last_subpass_idx >= anv_get_subpass_id(cmd_state));
+
+ if (subpass_end && anv_get_subpass_id(cmd_state) <
+ att_desc->last_subpass_idx) {
+ /* We're calling this function on a buffer twice in one subpass and
+ * this is not the last use of the buffer. The layout should not have
+ * changed from the first call and no transition is necessary.
+ */
+ assert(att_ref->layout == att_state->current_layout);
+ continue;
+ }
+
+ /* Get the appropriate target layout for this attachment. */
+ const VkImageLayout target_layout = subpass_end ?
+ att_desc->final_layout : att_ref->layout;
+
+ /* The attachment index must be less than the number of attachments
+ * within the framebuffer.
+ */
+ assert(att_ref->attachment < cmd_state->framebuffer->attachment_count);
+
+ const struct anv_image * const image =
+ cmd_state->framebuffer->attachments[att_ref->attachment]->image;
+
+ /* Perform the layout transition. */
+ if (image->aspects & VK_IMAGE_ASPECT_DEPTH_BIT) {
+ transition_depth_buffer(cmd_buffer, image,
+ att_state->current_layout, target_layout);
+ att_state->aux_usage =
+ anv_layout_to_aux_usage(&cmd_buffer->device->info, image,
+ image->aspects, target_layout);
+ }
+
+ att_state->current_layout = target_layout;
+ }
+}
+
static void
genX(cmd_buffer_set_subpass)(struct anv_cmd_buffer *cmd_buffer,
struct anv_subpass *subpass)
cmd_buffer->state.dirty |= ANV_CMD_DIRTY_RENDER_TARGETS;
- const struct anv_image_view *iview =
- anv_cmd_buffer_get_depth_stencil_view(cmd_buffer);
-
- if (iview && iview->image->aux_usage == ISL_AUX_USAGE_HIZ) {
- const uint32_t ds = subpass->depth_stencil_attachment;
- transition_depth_buffer(cmd_buffer, iview->image,
- cmd_buffer->state.attachments[ds].current_layout,
- cmd_buffer->state.subpass->depth_stencil_layout);
- cmd_buffer->state.attachments[ds].current_layout =
- cmd_buffer->state.subpass->depth_stencil_layout;
- cmd_buffer->state.attachments[ds].aux_usage =
- layout_to_hiz_usage(cmd_buffer->state.subpass->depth_stencil_layout,
- iview->image->samples);
- }
+ /* Perform transitions to the subpass layout before any writes have
+ * occurred.
+ */
+ cmd_buffer_subpass_transition_layouts(cmd_buffer, false);
cmd_buffer_emit_depth_stencil(cmd_buffer);
cmd_buffer->state.framebuffer = framebuffer;
cmd_buffer->state.pass = pass;
cmd_buffer->state.render_area = pRenderPassBegin->renderArea;
- genX(cmd_buffer_setup_attachments)(cmd_buffer, pass, pRenderPassBegin);
+ VkResult result =
+ genX(cmd_buffer_setup_attachments)(cmd_buffer, pass, pRenderPassBegin);
+
+ /* If we failed to setup the attachments we should not try to go further */
+ if (result != VK_SUCCESS) {
+ assert(anv_batch_has_error(&cmd_buffer->batch));
+ return;
+ }
genX(flush_pipeline_select_3d)(cmd_buffer);
genX(cmd_buffer_set_subpass)(cmd_buffer, pass->subpasses);
+
+ cmd_buffer->state.pending_pipe_bits |=
+ cmd_buffer->state.pass->subpass_flushes[0];
}
void genX(CmdNextSubpass)(
{
ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
- assert(cmd_buffer->level == VK_COMMAND_BUFFER_LEVEL_PRIMARY);
+ if (anv_batch_has_error(&cmd_buffer->batch))
+ return;
- const struct anv_image_view *iview =
- anv_cmd_buffer_get_depth_stencil_view(cmd_buffer);
+ assert(cmd_buffer->level == VK_COMMAND_BUFFER_LEVEL_PRIMARY);
- if (iview && iview->image->aux_usage == ISL_AUX_USAGE_HIZ) {
- const uint32_t ds = cmd_buffer->state.subpass->depth_stencil_attachment;
+ anv_cmd_buffer_resolve_subpass(cmd_buffer);
- if (cmd_buffer->state.subpass - cmd_buffer->state.pass->subpasses ==
- cmd_buffer->state.pass->attachments[ds].last_subpass_idx) {
- transition_depth_buffer(cmd_buffer, iview->image,
- cmd_buffer->state.attachments[ds].current_layout,
- cmd_buffer->state.pass->attachments[ds].final_layout);
- }
- }
+ /* Perform transitions to the final layout after all writes have occurred.
+ */
+ cmd_buffer_subpass_transition_layouts(cmd_buffer, true);
- anv_cmd_buffer_resolve_subpass(cmd_buffer);
genX(cmd_buffer_set_subpass)(cmd_buffer, cmd_buffer->state.subpass + 1);
+
+ uint32_t subpass_id = anv_get_subpass_id(&cmd_buffer->state);
+ cmd_buffer->state.pending_pipe_bits |=
+ cmd_buffer->state.pass->subpass_flushes[subpass_id];
}
void genX(CmdEndRenderPass)(
{
ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
- const struct anv_image_view *iview =
- anv_cmd_buffer_get_depth_stencil_view(cmd_buffer);
+ if (anv_batch_has_error(&cmd_buffer->batch))
+ return;
- if (iview && iview->image->aux_usage == ISL_AUX_USAGE_HIZ) {
- const uint32_t ds = cmd_buffer->state.subpass->depth_stencil_attachment;
+ anv_cmd_buffer_resolve_subpass(cmd_buffer);
- if (cmd_buffer->state.subpass - cmd_buffer->state.pass->subpasses ==
- cmd_buffer->state.pass->attachments[ds].last_subpass_idx) {
- transition_depth_buffer(cmd_buffer, iview->image,
- cmd_buffer->state.attachments[ds].current_layout,
- cmd_buffer->state.pass->attachments[ds].final_layout);
- }
- }
+ /* Perform transitions to the final layout after all writes have occurred.
+ */
+ cmd_buffer_subpass_transition_layouts(cmd_buffer, true);
- anv_cmd_buffer_resolve_subpass(cmd_buffer);
+ cmd_buffer->state.pending_pipe_bits |=
+ cmd_buffer->state.pass->subpass_flushes[cmd_buffer->state.pass->subpass_count];
cmd_buffer->state.hiz_enabled = false;
#ifndef NDEBUG
anv_dump_add_framebuffer(cmd_buffer, cmd_buffer->state.framebuffer);
#endif
+
+ /* Remove references to render pass specific state. This enables us to
+ * detect whether or not we're in a renderpass.
+ */
+ cmd_buffer->state.framebuffer = NULL;
+ cmd_buffer->state.pass = NULL;
+ cmd_buffer->state.subpass = NULL;
}