scaffold: CMake + API skeleton + smoke test

First code on daedalus-decoder per the Phase 1 decisions merged 2026-05-24.
Repo skeleton only — no Vulkan pipeline yet, no shaders, no libavcodec
intercept.  Establishes the build shape so subsequent work has a place
to land.

Layout:

  LICENSE                          BSD-2-Clause (matches daedalus-fourier)
  .gitignore                       build/, CMake artefacts, *.spv
  CMakeLists.txt                   top-level — finds daedalus-fourier
                                   ≥0.1.0 via pkg-config (per §9.6
                                   decision: find_package, pinned to
                                   tagged release; .pc consumed via
                                   pkg_check_modules until we ship a
                                   CMake config), Vulkan via
                                   find_package, builds static lib
                                   + smoke test, GNUInstallDirs install
  include/daedalus_decoder.h       public API surface:
                                     - daedalus_decoder_{create,destroy,
                                                         version,has_qpu}
                                     - daedalus_decoder_set_output_format
                                       (NV12 default, RGBA opt-in per §5)
                                     - daedalus_decoder_append_mb +
                                       struct daedalus_decoder_mb_input
                                       (matches §3 per-MB descriptor)
                                     - daedalus_decoder_flush_frame
                                       (per-frame submit + wait)
                                     - daedalus_decoder_export_dmabuf
                                       (Vulkan-native VkImage export per
                                       §9.4 decision)
                                   Dimensions are CODED frame size
                                   (mod-16), not displayed — caller
                                   translates from SPS + crop offsets.
  src/internal.h                   internal mb_desc struct (matches
                                   shader std430 layout, to be nailed
                                   down once shaders exist) + per-ctx
                                   state
  src/daedalus_decoder.c           stub bodies:
                                     - create/destroy with proper resource
                                       lifecycle
                                     - append_mb validates + writes CPU
                                       staging buffers (no GPU yet)
                                     - flush_frame returns -2 (not
                                       implemented) — Phase 1 work
                                     - export_dmabuf returns -1
                                     - has_qpu / version diagnostics
  tests/test_smoke.c               link + lifecycle test: bad dims
                                   reject, OOB MB reject, null inputs
                                   reject, raster-order enforcement,
                                   mid-frame format-change reject,
                                   incomplete-frame flush reject.
                                   On hosts without V3D7 Vulkan,
                                   SKIPs gracefully (returns 0).

Verified on hertz (Pi 5 / V3D 7.1 / Mesa V3DV via daedalus-fourier
0.1.0):

  $ cmake -B build -G Ninja -DCMAKE_BUILD_TYPE=Release
  $ cmake --build build
  $ ctest --test-dir build --output-on-failure
  Test #1: smoke ... Passed

  $ ./build/test_smoke
  daedalus-decoder version: 0.0.1
  ctx created: 1920x1088, has_qpu=1
  smoke OK

Note the coded-vs-displayed dims trap: 1080p H.264 has coded height
1088 with 8 rows cropped via SPS frame_cropping_*.  Header docstring
on daedalus_decoder_create() spells this out so future callers don't
hit the multiple-of-16 reject (smoke test caught it during scaffold
write).

Next: Phase 1 implementation begins — IDCT 4×4 / 8×8 frame-scaled
dispatch (reusing daedalus-fourier shaders per Appendix A), intra
prediction wavefront, reconstruct stage, NV12 output via dmabuf
export.  Smoke test grows from "ctx lifecycle works" to
"I-frame-only Baseline decode bit-exact vs FFmpeg reference".
This commit is contained in:
2026-05-24 22:08:46 +02:00
parent 59885dd868
commit 08080f062c
7 changed files with 683 additions and 0 deletions
+182
View File
@@ -0,0 +1,182 @@
/* SPDX-License-Identifier: BSD-2-Clause */
/*
* daedalus-decoder — public C API.
*
* Frame-level GPU H.264 decoder targeting V3D7 (Raspberry Pi 5). Built
* on daedalus-fourier's V3D compute primitives at frame granularity —
* one Vulkan submit per frame, one fence wait per frame, encoded
* bitstream in (via libavcodec's per-MB intercept), NV12 frame out.
*
* Per the 2026-05-24 Phase 1 design decisions:
* - libavcodec intercept is at macroblock-level (substitution-arc
* evolution): the caller is expected to drive the per-MB CABAC /
* CAVLC entropy decode and feed each macroblock's descriptor +
* coefficients via daedalus_decoder_append_mb(). flush_frame()
* builds the per-frame VkCommandBuffer and submits.
* - DPB is Vulkan-native VkImage with VK_KHR_external_memory_dma_buf
* export. The caller can obtain the output frame's dmabuf fd
* via daedalus_decoder_export_dmabuf().
* - Daemon integration shape: this library is statically linked into
* daedalus_v4l2_daemon. No IPC.
*
* STATUS: scaffold. No GPU pipeline implemented yet; all functions
* are stubs that compile but do not decode anything. See DESIGN.md
* for the architecture.
*
* ABI: pre-0.1 — every signature here may change. Don't rely on
* stability yet.
*/
#ifndef DAEDALUS_DECODER_H
#define DAEDALUS_DECODER_H
#include <stddef.h>
#include <stdint.h>
#ifdef __cplusplus
extern "C" {
#endif
/* -------------------------------------------------------------------
* Opaque decoder context. One per concurrent stream.
* ----------------------------------------------------------------- */
typedef struct daedalus_decoder daedalus_decoder;
/* -------------------------------------------------------------------
* Per-macroblock input. Mirrors §3 of DESIGN.md. The caller's
* libavcodec intercept populates this from the H264SliceContext
* fields after ff_h264_decode_mb_cabac/cavlc returns and before
* ff_h264_hl_decode_mb is supposed to run (we replace the latter).
* ----------------------------------------------------------------- */
struct daedalus_decoder_mb_input {
/* Frame coordinates (macroblock units). */
uint16_t mb_x;
uint16_t mb_y;
/* Type + quantisation. */
uint8_t mb_type; /* H.264 spec table 7-13/7-14/7-17/7-18 enum */
uint8_t mb_qp_y;
uint8_t mb_qp_uv;
uint8_t cbp; /* coded block pattern, 0..47 */
/* Intra prediction (used iff mb_type == I_NxN or I_16x16). */
uint8_t intra_4x4_modes[16];
uint8_t intra_16x16_mode;
uint8_t intra_chroma_mode;
/* Inter motion / partitions (used iff P_* or B_*). */
uint8_t partition_mode; /* P_16x16 / P_16x8 / P_8x16 / P_8x8 / etc. */
int8_t ref_idx_l0[4]; /* per partition; -1 = not used */
int8_t ref_idx_l1[4]; /* B only */
int16_t mv_l0[4][2]; /* qpel precision (1/4 sample); (x, y) */
int16_t mv_l1[4][2];
/* Deblocking filter parameters. */
uint8_t deblock_disable; /* 0 = enabled */
int8_t deblock_alpha_c0;
int8_t deblock_beta;
/* Transform coefficients — 256 luma (4x4 x 16) + 64 cb + 64 cr,
* column-major within each 4x4 block (matches FFmpeg convention).
* Caller-owned; copied during append. */
const int16_t *coeffs; /* points at exactly 384 int16_t */
};
/* -------------------------------------------------------------------
* Output frame format selector.
* ----------------------------------------------------------------- */
typedef enum {
DAEDALUS_DECODER_OUTPUT_NV12 = 0, /* default; Stage 4 final */
DAEDALUS_DECODER_OUTPUT_RGBA = 1, /* Stage 5 opt-in */
} daedalus_decoder_output_format;
/* -------------------------------------------------------------------
* Lifecycle
* ----------------------------------------------------------------- */
/* Create a decoder context for the given **coded** frame dimensions.
*
* width, height: pixels of the H.264 coded picture, NOT the displayed
* picture. Both must be multiples of 16 (macroblock granularity).
* For displayed 1080p (1920×1080), the coded frame is 1920×1088 with
* the SPS's `frame_cropping_*` offsets cropping the bottom 8 rows.
* The caller is responsible for translating from SPS dims + crop
* rectangle to the values passed here; we decode the coded frame.
*
* Returns NULL on bad dimensions or allocation failure. Returns a
* usable context with daedalus_decoder_has_qpu() == 0 when Vulkan
* init fails — callers that need GPU work should check has_qpu
* before relying on it.
*/
daedalus_decoder *daedalus_decoder_create(int width, int height);
/* Free all resources. Safe with NULL. */
void daedalus_decoder_destroy(daedalus_decoder *dec);
/* Switch output format BEFORE the first append_mb call of a frame.
* Default is NV12. Returns 0 on success, -1 if called mid-frame
* (caller must flush first). */
int daedalus_decoder_set_output_format(daedalus_decoder *dec,
daedalus_decoder_output_format fmt);
/* -------------------------------------------------------------------
* Per-frame submission
* ----------------------------------------------------------------- */
/* Append one macroblock's data to the current frame's descriptor SSBO
* + coefficient SSBO. No GPU dispatch yet — just CPU-side writes.
*
* Must be called in raster order (mb_y * mb_width + mb_x) for the
* intra-prediction wavefront to work correctly in Phase 1.
*
* Returns 0 on success, negative on bounds violation or OOM.
*/
int daedalus_decoder_append_mb(daedalus_decoder *dec,
const struct daedalus_decoder_mb_input *mb);
/* End-of-frame flush: builds the per-frame VkCommandBuffer with all
* pipeline stages, submits once, waits on a single fence, copies the
* NV12 (or RGBA when opted in) output into the caller-provided
* planes.
*
* For NV12:
* out_y / y_stride: Y plane (W*H bytes minimum, at the given stride)
* out_uv / uv_stride: interleaved UV plane (W*(H/2) bytes minimum)
*
* For RGBA: out_y receives 4*W*H bytes at y_stride; out_uv ignored.
*
* Returns 0 on success, negative on Vulkan failure or undecodable
* frame. After return, the decoder is ready for the next frame's
* append calls.
*/
int daedalus_decoder_flush_frame(daedalus_decoder *dec,
uint8_t *out_y, size_t y_stride,
uint8_t *out_uv, size_t uv_stride);
/* Export the most-recently-decoded frame as a dma_buf fd. The fd is
* owned by the caller and must be closed when done. Lets V4L2
* consumers (daedalus_v4l2_daemon, libva-v4l2-request-fourier) attach
* the GPU-decoded surface directly to a CAPTURE plane without a CPU
* round-trip.
*
* Returns the dmabuf fd on success, -1 on failure. Must be called
* AFTER flush_frame returns for the relevant frame.
*/
int daedalus_decoder_export_dmabuf(daedalus_decoder *dec, int plane);
/* -------------------------------------------------------------------
* Diagnostics
* ----------------------------------------------------------------- */
/* daedalus-decoder build version (semver string, e.g. "0.0.1+g0a1b2c3"). */
const char *daedalus_decoder_version(void);
/* Whether the underlying daedalus-fourier context picked up a working
* V3D7 Vulkan instance. Returns 0 if Vulkan init failed and the
* decoder is operating in stub / failure mode. */
int daedalus_decoder_has_qpu(const daedalus_decoder *dec);
#ifdef __cplusplus
}
#endif
#endif /* DAEDALUS_DECODER_H */