Repository navigation
fix(agc): cover the console's DCC key extent for 1, 2, 8 and 16-byte elements - #7896
Closed
eljackjoestar-cmd wants to merge 1 commit into
Closed
eljackjoestar-cmd wants to merge 1 commit into
eljackjoestar-cmd wants to merge 1 commit into
Conversation
…elements DccKeyCount took the console's DCC extent only for 4-byte SW_64KB_R_X surfaces and left every other element size at one key per 256 bytes. addrlib's GFX10 meta blocks hold 4 KiB of keys over 2^(20 - log2 element bytes) texels, the odd bit going to the width, so a 2420x1364 8-byte target has 0x1e000 key bytes, not 0x1a200. A key fill of that whole extent over memory that other live images overlap was longer than the computed extent, so it was classified by its overlaps instead of as a fast clear of the target, and the target kept stale texels. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
github-actions
Bot
added
the
unblocks title
|
Some comparisons could not be completed because a pull request conflicts with the base branch. Review these shared files manually before concluding that there is no overlap.
|
Contributor
|
Host: Arch Linux, kernel 7.2.11, AMD Radeon RX 7900 GRE (RADV NAVI31), Mesa 26.2.5, GCC 16.2.1, Release,
The fixture ( The PR's own tests pass: |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
DccKeyCounttook the console's DCC key extent only for 4-byte SW_64KB_R_X surfaces: 512x512-texel meta blocks of 4 KiB of keys, measured in #788. Every other element size kept one key per 256 bytes.addrlib's GFX10 meta blocks hold 4 KiB of keys over 2^(20 - log2 element bytes) texels, the odd bit going to the width:
The extent now follows these blocks for 1, 2, 4, 8 and 16-byte elements. Other element sizes (12-byte) and other tile modes keep one key per 256 bytes. The 4-byte result is unchanged.
Effect on PPSA28099. Its start screen showed a black sky.
DccKeyCountgave 0x1a200, so the fill was longer than the image's key extent. The title's transient allocator also places other live images over that memory.StorageTexture::keysFillMatchestherefore classified the fill by its overlaps instead of as the target's key fill.scene * translucency.a + translucency.rgb), and the sky turned black.With the 8-byte extent, the fill is the target's key fill and the sky renders.
The debt entry "DCC key stores" is cut to what stays open: the 1, 2 and 16-byte extents follow addrlib and are not measured.
New
KeyCountTestscases (agc_display_buffer_dcc):Tested
Windows 11 Pro 26300, Ryzen 7 7800X3D, NVIDIA GeForce RTX 5070 (driver 617.14), WinLibs GCC 15.2 UCRT, Release.
Focused tests. Run on
main6762694 and on this branch, same machine, withVK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation ANYPS5_NO_SHADER_CACHE=1 ANYPS5_REQUIRE_VULKAN=1 ctest --test-dir build -R <pattern> -V. The tests includeagc_display_buffer_dcc,agc_dcc_key_sharing,agc_driver_fill_cover_keys,agc_driver_dcc_key_presync,agc_driver_dcc_keys_first_write,agc_driver_memory_value_fillandagc_driver_staged_clear_sampled_read.was not found but was requested by env var VK_INSTANCE_LAYERS.agc_driver_indirect_arguments_pending_producer:VUID-vkCmdWaitEvents-srcStageMask-01158andVUID-vkSetEvent-event-09543. This change adds none.The 63 focused tests (same set on main and the change)
agc_command_batch_fence_failure,agc_compute_descriptor_layout_flags,agc_copy_input_kept_bytes,agc_dcc_key_sharing,agc_depth_copy_kept_bytes,agc_descriptor_cache,agc_descriptor_pool_reset_failure,agc_display_buffer_dcc,agc_driver_address_space_teardown,agc_driver_alpha_to_coverage_state,agc_driver_buffer_descriptor_runtime,agc_driver_covered_centroid_state,agc_driver_dcc_key_presync,agc_driver_dcc_keys_first_write,agc_driver_fill_cover_keys,agc_driver_gpu_indirect_arguments_misaligned_writer_ahead,agc_driver_gpu_indirect_arguments_written_ahead,agc_driver_graphics,agc_driver_guarded_null_pointer,agc_driver_indirect_arguments_pending_producer,agc_driver_indirect_arguments_written_ahead,agc_driver_indirect_patched_stage,agc_driver_memory_value_fill,agc_driver_mesh,agc_driver_mesh_draw_relocation,agc_driver_mesh_index_cache,agc_driver_mesh_provoking_layer,agc_driver_mesh_provoking_vertex,agc_driver_msaa_dcc_state,agc_driver_rect_list_indirect,agc_driver_sample_location_grid_fallback,agc_driver_sparse_indirect_indices,agc_driver_sparse_indirect_indices_unreadable,agc_driver_staged_clear_sampled_read,agc_driver_undithered_alpha_to_coverage_state,agc_driver_unused_viewport_depth,agc_heap_mirror_run_unwatch,agc_heap_mirror_run_unwatch_no_space_cache,agc_heap_mirror_unwatch,agc_heap_mirror_unwatch_no_space_cache,agc_pipeline_journal,agc_pipeline_library_clear_waits_for_build,agc_pipeline_store_lazy_buffer_pool,agc_shader_bindless,agc_shader_entry_lane_proof_budget,agc_shader_frontend_pair,agc_shader_frontend_pair_off,agc_shader_layout_pair,agc_shader_layout_pair_off,agc_shader_memory,agc_shader_quad_mask_no_demote,agc_shader_selected_records_direct_image_limit,agc_shader_selected_records_direct_pair_limit,agc_shader_selected_records_image_limit,agc_shader_selected_records_pair_limit,agc_shader_srt_read_patching,agc_shader_table_snapshot_reuse,agc_shader_type_zero_image_materialization,agc_sparse_buffer_holes,agc_sparse_heap_mirror_unwatch,agc_sparse_heap_mirror_unwatch_no_space_cache,agc_write_tracking,agc_write_watch_coverageTitle. PPSA28099 start screen at 3840x2160.
maincannot run this title: at 6762694 it stops at boot withindirect register sentinel semantics are not implemented(Execution/src/Pm4.cpp:47). The check used the base the title runs on: main a62c724 with #6094 and #6160 (both merged since), #6146, #6366 and #7644.Overlap.
tools/pr_overlap.py --branchlists #7480, which adds a debt entry on the line next to the edited one (docs/dev/debt/libSceAgcDriver.md). The conflict is textual only. The other PRs it lists conflict withmainand shareDccMetadata.cppor the debt file; none of them changesDccKeyCount.Checklist
main; no other open PR implements the same functionsNotImplemented_nid_no_patch); silent stubs are listed under Silent stubs in their library's technical debt file.mdfiles or images added to the repository (attach them to this PR)