h264: V3D shader for qpel mc22 (2D half-pel "j" position)

Cascaded H+V 6-tap filter per H.264 §8.4.2.2.1.  Highest per-frame
impact among missing qpel positions (PR #24 bench: 71.5 ns/block
NEON, 2.33 ms/frame worst-case all-mc22 at 1080p).

Per-lane structure: each lane runs the FULL cascade independently —
computes 6 horizontal lowpass int16 intermediates at rows r-2..r+3
of its column, then a vertical lowpass on those with +512 >> 10
final scale.  ~50 ALU ops per lane.

Design choice: NO shared memory / barriers.  Alternative was to
cache the h-lowpass intermediates in shared memory (13 rows × 8 cols
of int16 per WG), trading shared-memory bank pressure + a barrier
for ~6× less h-lowpass compute.  V3D L2 absorbs the redundant src
reads across lanes; the per-lane compute is cheap (multiply-add ALU
units idle anyway during dst write).  Simpler shader, fewer SPIR-V
ops, easier to extend to mc12/mc21/etc. later.

CANNOT just cascade mc20 → mc02 because the intermediate must be
int16 (no per-stage clip): the +512 >> 10 final scale assumes both
6-tap scalings preserved through the pipeline.  Dedicated kernel.

dispatch_h264_qpel_mc22_qpu mirrors the existing mc20/mc02 shape;
src_max = src_off + 10*stride + 11 covers both the V (rows -2..+10)
and H (cols -2..+10) read windows in one bound.

Recipe table flips DAEDALUS_KERNEL_H264_QPEL_MC22 from CPU to QPU.

Verified on hertz:

  $ ./build/test_api_h264 | grep qpel
    H.264 qpel mc20: 1024/1024 bytes bit-exact (100.0000%)
    H.264 qpel mc02: 2048/2048 bytes bit-exact (100.0000%)
    H.264 qpel mc22: 2048/2048 bytes bit-exact (100.0000%)

Qpel QPU coverage now: 3 anchors (mc20 H, mc02 V, mc22 HV) — these
are the half-pel "building blocks" the 12 other qpel positions
combine via L2 averaging.  Remaining variants (quarter-pel singles
mc01/03/10/30 and the 8 diagonals) can dispatch through the existing
shaders + a small L2-averaging compose step, or get dedicated kernels.

src/v3d_h264_qpel_mc22.comp

@@ -0,0 +1,86 @@
+// daedalus-fourier — H.264 luma qpel mc22 (8x8, 2D half-pel "j" position).
+// V3D 7.1.
+//
+// Cascaded H+V 6-tap per H.264 §8.4.2.2.1 / FFmpeg ff_put_h264_qpel8_mc22_neon:
+//
+//   tmp[r,c] = src[r,c-2] - 5*src[r,c-1] + 20*src[r,c] + 20*src[r,c+1]
+//              - 5*src[r,c+2] + src[r,c+3]                    (int16)
+//
+//   dst[r,c] = clip255((tmp[r-2,c] - 5*tmp[r-1,c] + 20*tmp[r,c]
+//                       + 20*tmp[r+1,c] - 5*tmp[r+2,c] + tmp[r+3,c]
+//                       + 512) >> 10)
+//
+// The +512 >> 10 final scale compensates for both 6-tap scalings.
+// CANNOT just cascade mc20→mc02 because intermediate must be int16
+// (no per-stage clip), so this is a dedicated kernel.
+//
+// Per-lane structure: each lane computes its own (r, c) output by
+// running the FULL cascade — 6 horizontal lowpass int16 values for
+// rows r-2..r+3, then a vertical lowpass on those.  ~50 ALU ops per
+// lane.  No shared memory / barriers needed; V3D L2 absorbs the
+// redundant src reads across lanes.
+//
+// WG layout: 64 lanes / 1 block-per-WG / 1 lane-per-output-pixel
+// (same as mc20 / mc02).
+//
+// License: BSD-2-Clause.
+
+#version 450
+#extension GL_EXT_shader_8bit_storage             : require
+#extension GL_EXT_shader_explicit_arithmetic_types : require
+
+layout(local_size_x = 64, local_size_y = 1, local_size_z = 1) in;
+
+layout(binding = 0) readonly buffer Src { uint8_t src[]; } u_src;
+layout(binding = 1) buffer Dst { uint8_t dst[]; } u_dst;
+layout(binding = 2) readonly buffer Meta { uvec4 meta[]; } u_meta;
+
+layout(push_constant) uniform PC {
+    uint n_blocks;
+    uint stride_u8;
+    uint _pad0, _pad1;
+} pc;
+
+// Horizontal 6-tap filter at (row_off, c) — reads src at cols c-2..c+3
+// of the row identified by row_off, returns int16 intermediate (NOT
+// scaled — the v-pass does the +512 >> 10 for both stages).
+int hpel_h(uint row_off, uint c)
+{
+    int s_m2 = int(u_src.src[row_off + c - 2u]);
+    int s_m1 = int(u_src.src[row_off + c - 1u]);
+    int s_0  = int(u_src.src[row_off + c       ]);
+    int s_p1 = int(u_src.src[row_off + c + 1u]);
+    int s_p2 = int(u_src.src[row_off + c + 2u]);
+    int s_p3 = int(u_src.src[row_off + c + 3u]);
+    return s_m2 - 5 * s_m1 + 20 * s_0 + 20 * s_p1 - 5 * s_p2 + s_p3;
+}
+
+void main()
+{
+    uint block_idx = gl_WorkGroupID.x;
+    if (block_idx >= pc.n_blocks) return;
+
+    uint lane = gl_LocalInvocationID.x;
+    uint r = lane >> 3;
+    uint c = lane & 7u;
+
+    uint dst_off = u_meta.meta[block_idx].x;
+    uint src_off = u_meta.meta[block_idx].y;
+    uint stride  = pc.stride_u8;
+
+    // Compute 6 horizontal lowpass values at rows r-2..r+3 (relative
+    // to the output row r) of column c.  src_off+r*stride+c is the
+    // output pixel position; we sample rows r-2..r+3.
+    // Unsigned-safe because src_off >= 2*stride per the caller contract.
+    int t0 = hpel_h(src_off + (r - 2u) * stride, c);
+    int t1 = hpel_h(src_off + (r - 1u) * stride, c);
+    int t2 = hpel_h(src_off +  r       * stride, c);
+    int t3 = hpel_h(src_off + (r + 1u) * stride, c);
+    int t4 = hpel_h(src_off + (r + 2u) * stride, c);
+    int t5 = hpel_h(src_off + (r + 3u) * stride, c);
+
+    int v = t0 - 5 * t1 + 20 * t2 + 20 * t3 - 5 * t4 + t5 + 512;
+    int p = clamp(v >> 10, 0, 255);
+
+    u_dst.dst[dst_off + r * stride + c] = uint8_t(p);
+}