Add M9-M14: UI redesign, plugins, macros, 3D viz, binary/segmentation ops, Metal GPU, scientific LUTs

M9: TailwindCSS + shadcn/ui frontend redesign M10: Dynamic plugin system with shared library loading M11: Macro recording, playback, and batch processing M12: 3D volume visualization, marching cubes, STL/OBJ export, rivol:// protocol M13: Additional GPU shaders, enhanced auto-threshold, merge channels M14: Binary image processing (EDT, watershed, skeleton, connected components, voronoi), segmentation & analysis (particles, colocalization, find maxima), math/noise/filter/transform ops (60+ total), scientific LUTs (Fire, Ice, Spectrum, Jet, Phase, HiLo + .lut file I/O), Apple Metal GPU optimization (metal-only feature, Apple Silicon detection, lower dispatch threshold), native Metal compute crate (ri-metal with MSL shaders via objc2-metal), 7 new WGSL GPU shaders (bilateral, variance, mean, math_ops, outline, minmax_filter, affine transform). 25 crates, 162 tests, 75+ IPC commands. Co-Authored-By: Claude Opus 4.6 <[email protected]>
2026-03-10 07:41:56 -07:00
parent 1889715f3c
commit 0fd39d7fd3
135 changed files with 19456 additions and 1905 deletions
@@ -0,0 +1,45 @@
+#include <metal_stdlib>
+using namespace metal;
+
+/// Per-threadgroup histogram using shared memory atomics, then reduced to global.
+/// Input:  buffer of f32 pixel values (expected range [0, 1]).
+/// Output: buffer of 256 uint32 bins.
+
+struct HistogramParams {
+    uint pixel_count;
+    uint _pad0;
+    uint _pad1;
+    uint _pad2;
+};
+
+kernel void histogram_compute(
+    device const float*   input       [[buffer(0)]],
+    device atomic_uint*   output      [[buffer(1)]],
+    constant HistogramParams& params  [[buffer(2)]],
+    threadgroup atomic_uint* local_hist [[threadgroup(0)]],
+    uint tid       [[thread_index_in_threadgroup]],
+    uint tg_size   [[threads_per_threadgroup]],
+    uint gid       [[thread_position_in_grid]]
+) {
+    // Initialize threadgroup histogram bins to zero
+    for (uint i = tid; i < 256; i += tg_size) {
+        atomic_store_explicit(&local_hist[i], 0, memory_order_relaxed);
+    }
+    threadgroup_barrier(mem_flags::mem_threadgroup);
+
+    // Each thread accumulates its pixel into the local histogram
+    if (gid < params.pixel_count) {
+        float val = input[gid];
+        uint bin = (uint)clamp(val * 255.0f, 0.0f, 255.0f);
+        atomic_fetch_add_explicit(&local_hist[bin], 1, memory_order_relaxed);
+    }
+    threadgroup_barrier(mem_flags::mem_threadgroup);
+
+    // First threads in group flush local histogram to global
+    for (uint i = tid; i < 256; i += tg_size) {
+        uint count = atomic_load_explicit(&local_hist[i], memory_order_relaxed);
+        if (count > 0) {
+            atomic_fetch_add_explicit(&output[i], count, memory_order_relaxed);
+        }
+    }
+}