The .inc consumers and the C++ host namespace moved to d3d_common in24bacbf, but the HLSL sources and the regen tool kept their d3d12 tags for "historical / churn-avoidance reasons" - the commit message there explicitly flagged it as deferred cleanup ("data/shaders/d3d12 -> data/shaders/d3d_common + tool rename. Currently kept as d3d12 for historical/churn-avoidance reasons"). With the downsample foundation landed at251e457also putting its HLSL under data/shaders/d3d12/ (despite being consumed only by D3D11 today, with D3D12 consumption arriving in a later phase), the naming inconsistency is now actively confusing. This commit completes the rename. What changes (mechanically, via git mv + sweep): data/shaders/d3d12/ -> data/shaders/d3d_common/ Whole directory renames via git mv. 16 files move: README.md + 10 .hlsl source templates (5 batch + display + copy + vram_{copy,fill,read,update_depth,write} + fullscreen_quad + 4 downsample). Each .hlsl content is unchanged except where a comment string referenced the old path (3 files: box_sample_downsample.ps.hlsl, adaptive_downsample_composite.ps.hlsl, vram_fill.ps.hlsl). tools/regen_d3d12_dxbc.py -> tools/regen_d3d_common_dxbc.py Renamed via git mv. The HLSL_DIR constant repointed to the new HLSL directory; the .inc preamble template updated to emit the new paths; the top-of-file comment updated to drop the "kept as d3d12 for historical reasons" note now that it doesn't apply. Behaviour unchanged: same HLSL -> fxc /T ps_5_0 /E main /O3 -> same .inc bytes. src/common/d3d_common/embedded_dxbc/*.inc (182 files) Preamble line 1 updated: // Autogenerated from data/shaders/d3d12/<name>.hlsl. -> // Autogenerated from data/shaders/d3d_common/<name>.hlsl. Preamble line 2 updated: // DO NOT EDIT. Regenerate with tools/regen_d3d12_dxbc.py. -> // DO NOT EDIT. Regenerate with tools/regen_d3d_common_dxbc.py. The byte arrays themselves are bit-identical pre/post-commit (fxc emits identical DXBC from the same HLSL source). Only the two comment lines of each .inc change. src/common/d3d_common/embedded_shaders.h 14 doc comments updated: "// Source: data/shaders/d3d12/<name>.hlsl" -> "// Source: data/shaders/d3d_common/<name>.hlsl" Plus the top-of-file rationale block updated to drop the d3d12-tagged path reference. src/core/gpu_hw_d3d12.cpp Two comment lines referencing the old data/shaders/d3d12/ path and the old regen tool name. The pre-bake call sites themselves were already namespace-qualified (D3DCommon:: EmbeddedShaders::*) and need no change. data/shaders/d3d_common/README.md 8 doc references to the old paths updated. What doesn't change in this commit: - DXBC byte content. fxc is deterministic on identical source + flags; the post-rename HLSL is byte-identical to the pre-rename HLSL except for comment text, and comments don't survive into the compiled bytecode. Re-running tools/regen_d3d_common_dxbc.py produces .inc files that match this commit's preamble lines + the pre-existing byte arrays. - C++ namespaces. Already D3DCommon::EmbeddedShaders since24bacbf. Not touched here. - Header / library / TU layout. embedded_shaders.{h,cpp} already live at src/common/d3d_common/ since24bacbf; embedded_dxbc/ already lives at src/common/d3d_common/embedded_dxbc/. This commit completes the symmetric rename on the data-side and tools-side. - Build system files (CMakeLists.txt, Makefile.libretro): none of them references the data/shaders directory tree or the regen tool - the data is consumed at HLSL-source-edit time only, not at build time. Untouched. - Runtime behaviour: zero diff across D3D11 + D3D12. Same blobs, same pickers, same PSOs. Test: - git status post-rename shows 16 renames + 184 minor edits (preamble-only diffs in the .inc files + comment-only diffs in the regen tool + embedded_shaders.h + gpu_hw_d3d12.cpp). No moved file's content changed structurally. - Hand-compile of a representative variant (box_sample_downsample at RESOLUTION_SCALE=4) post-rename: 1044 bytes DXBC, identical to pre-rename. - String search across the tree: zero remaining references to "data/shaders/d3d12" or "regen_d3d12_dxbc" outside of .git history. Next: - Activation commit for the downsample pre-bake: add 22 extern pairs + 22 #include lines to embedded_shaders.{h,cpp}, add 4 picker functions, rewrite the 5 shadergen calls in GPU_HW_D3D11::CompileShaders. The d3d_common naming is now coherent across HLSL sources, regen tool, .inc files, host TU, and host namespace - new pre-bake additions don't have to thread the old d3d12 naming. - Phase 2 (separate effort): port the D3D11 downsample machinery to D3D12 - PSOs, explicit resource barriers, texture / mip-view creation - and set m_supports_adaptive_downsampling = true on the D3D12 caps path. The pre-baked blobs are already in place by then.
5.8 KiB
D3D12 shader sources (pre-baked DXBC)
This directory holds the HLSL sources for shaders used by the D3D12 backend
that have been moved off the runtime ShaderGen path. Each .hlsl file here
has a matching pre-compiled DXBC blob checked in to
src/common/d3d12/embedded_dxbc/.
Why this exists
Historically the D3D12 backend generated HLSL at runtime via
src/core/shadergen.cpp and src/core/gpu_hw_shadergen.cpp, then handed the
text to D3DCompile (the in-process HLSL -> DXBC compiler from d3dcompiler_47.dll)
to produce a DXBC blob for each PSO. On a cold shader cache, walking the
1164-cell batch pipeline matrix in precompile_sync mode takes on the order
of one minute, which is the worst entry in the post-dim-cache cost matrix
(see the cost-matrix comments in 10c53b8, 57ac62e, and the patches that
followed). The runtime cache hides this on warm boots, but cold boots and
shader cache invalidations re-pay the full per-cell D3DCompile cost every
time the user changes a setting that hits the full-flush branch.
Pre-baking the DXBC removes:
- per-launch HLSL compilation stall on shader cache misses,
- a class of "stale cache after
d3dcompiler_47.dllversion drift" failures, - the implicit dependency on
D3DCompilebeing available at runtime for the cells that have been pre-baked.
The OpenGL / OpenGL ES / D3D11 / Vulkan backends still use the runtime
ShaderGen path; only the D3D12 backend reads from the embedded DXBC. (The
Vulkan equivalent of this infrastructure lives at data/shaders/vulkan/ /
src/common/vulkan/embedded_spirv/ / tools/regen_vulkan_spirv.py; this
directory is the mirror.)
Build-time constraint
The core build (Makefile.libretro and src/common/CMakeLists.txt) must
never invoke fxc.exe or any other external shader compiler. The .inc
files under src/common/d3d12/embedded_dxbc/ are checked into the
repository and compiled like any other C++ source. The
tools/regen_d3d_common_dxbc.py helper exists for contributors who edit the
.hlsl sources in this directory; it is not part of the build.
Regenerating the DXBC blobs
After editing a .hlsl file here, run:
python3 tools/regen_d3d_common_dxbc.py
This invokes fxc.exe (Shader Model 5.0, /O3, /E main) on each .hlsl
file and rewrites the matching .inc under src/common/d3d12/embedded_dxbc/.
Both the .hlsl and the regenerated .inc must be committed together so the
two stay in sync.
Microsoft fxc.exe is required
The regen script will refuse to write .inc files produced by any compiler
other than Microsoft's fxc.exe. The check parses the DXBC RDEF chunk and
matches the creator string against a Microsoft-prefix allowlist.
The reason is empirical: Wine ships its own d3dcompiler_47.dll
implementation (vkd3d-shader), and that compiler's output, while a valid
DXBC container, doesn't pass the Microsoft D3D12 runtime's
CreateGraphicsPipelineState validation for non-trivial shaders. The
failure mode is E_FAIL (0x80004005) at PSO-creation time with no
useful diagnostic from Log_ErrorPrintf. Commit cd971cd shipped a
vkd3d-shader-compiled vram_copy_ps.inc and broke the entire D3D12
backend on launch; da20f5b reverts it. The script's creator-string
validator exists to prevent that mistake from recurring.
Installing Microsoft fxc.exe
Windows: Install the Windows 10 or 11 SDK from Microsoft. fxc.exe
ends up at
C:\Program Files (x86)\Windows Kits\10\bin\<version>\x64\fxc.exe (or
the same path under \Program Files\ if the 64-bit SDK is installed).
The regen script auto-detects it there.
Linux: Microsoft's fxc.exe is a Windows x86 PE binary and runs
fine under Wine - the trick is just to get the SDK installed inside a
Wine prefix. Two options:
- Run the Windows SDK installer (
winsdksetup.exefrom Microsoft) underwineand select the Windows SDK Signing Tools / SDK component that shipsfxc.exe. The default install puts it at<prefix>/drive_c/Program Files (x86)/Windows Kits/10/bin/<version>/x64/fxc.exe, which the regen script picks up automatically. - Or extract
fxc.exe+d3dcompiler_47.dlldirectly from the SDK MSI usingmsiextractfrommsitools, drop both in any directory, and pass--fxc PATH/TO/fxc.exeto the regen script. Both files must live in the same directory so Wine's loader finds Microsoft'sd3dcompiler_47.dllbefore falling back to its bundled vkd3d-shader version.
The script searches for fxc.exe in: $WINEPREFIX (if set), ~/.wine,
and any subdirectory under ~/.local/share/wineprefixes/ that has a
drive_c tree. Set WINEPREFIX to your prefix path if it lives
elsewhere.
The --wine PATH flag lets you point at a specific wine binary if
wine isn't on $PATH.
Filename convention
Stage is encoded in the suffix:
.vs.hlsl-> vertex shader (compiled asvs_5_0).ps.hlsl-> pixel shader (compiled asps_5_0).cs.hlsl-> compute shader (compiled ascs_5_0).gs.hlsl-> geometry shader (compiled asgs_5_0)
The _5_0 shader model targets D3D_FEATURE_LEVEL_11_0, matching the
runtime baseline of the D3D12 backend (see the shared compile path in
src/common/d3d11/shader_compiler.cpp which both D3D11 and D3D12 used
prior to the pre-bake migration).
Template variants
Shaders that need multiple pre-baked variants (e.g. one per BatchRenderMode
or per GPUTextureFilter) are listed in the TEMPLATE_VARIANTS dict at the
top of tools/regen_d3d_common_dxbc.py. Each entry maps an .hlsl filename to
the list of (variant_suffix, [/D defines]) tuples that the regen script
should produce.
If you add a template variant to the dict, add the matching extern
declaration to src/common/d3d12/embedded_shaders.h AND the matching slot
lookup to src/common/d3d12/embedded_shaders.cpp - the index ordering at
the runtime call site must match the order in the dict so the correct
blob gets bound at PSO-creation time.