Change computenbody shader to better match glsl

This enables much higher perf

Author: LegNeato

shaders/rust/computenbody/particle_calculate.comp.spv

@@ -1,8 +1,9 @@
 #![cfg_attr(target_arch = "spirv", no_std)]
 
 use spirv_std::{
-    glam::{vec3, vec4, Vec3, Vec4},
+    glam::{vec3, vec4, Vec4, Vec4Swizzles},
     spirv,
+    arch::workgroup_memory_barrier_with_group_sync,
     num_traits::Float,
 };
 
@@ -23,40 +24,62 @@ pub struct UBO {
     pub soften: f32,
 }
 
+const SHARED_DATA_SIZE: usize = 512;
+
 #[spirv(compute(threads(256)))]
 pub fn main_cs(
     #[spirv(global_invocation_id)] global_id: spirv_std::glam::UVec3,
+    #[spirv(local_invocation_id)] local_id: spirv_std::glam::UVec3,
+    #[spirv(workgroup)] shared_data: &mut [Vec4; SHARED_DATA_SIZE],
     #[spirv(storage_buffer, descriptor_set = 0, binding = 0)] particles: &mut [Particle],
     #[spirv(uniform, descriptor_set = 0, binding = 1)] ubo: &UBO,
 ) {
     let index = global_id.x as usize;
+    let local_index = local_id.x as usize;
 
     if index >= ubo.particle_count as usize {
         return;
     }
 
     let position = vec4(particles[index].pos[0], particles[index].pos[1], particles[index].pos[2], particles[index].pos[3]);
     let mut velocity = vec4(particles[index].vel[0], particles[index].vel[1], particles[index].vel[2], particles[index].vel[3]);
-    let mut acceleration = vec4(0.0, 0.0, 0.0, 0.0);
+    let mut acceleration = vec3(0.0, 0.0, 0.0);
 
-    // Calculate forces from all other particles (simplified O(N²) approach)
-    for i in 0..ubo.particle_count as usize {
-        if i == index {
-            continue; // Skip self-interaction
+    // Process particles in chunks of SHARED_DATA_SIZE
+    let mut i = 0u32;
+    while i < ubo.particle_count {
+        // Load particle data into shared memory
+        if i + (local_index as u32) < ubo.particle_count {
+            let particle_idx = i as usize + local_index;
+            shared_data[local_index] = vec4(
+                particles[particle_idx].pos[0],
+                particles[particle_idx].pos[1],
+                particles[particle_idx].pos[2],
+                particles[particle_idx].pos[3]
+            );
+        } else {
+            shared_data[local_index] = vec4(0.0, 0.0, 0.0, 0.0);
+        }
+        
+        // Ensure all threads have loaded their data
+        unsafe {
+            workgroup_memory_barrier_with_group_sync();
         }
 
-        let other = vec4(particles[i].pos[0], particles[i].pos[1], particles[i].pos[2], particles[i].pos[3]);
-        let len = vec3(other.x - position.x, other.y - position.y, other.z - position.z);
-        let distance_sq = len.dot(len) + ubo.soften;
-        let distance = distance_sq.sqrt();
-        let force_magnitude = ubo.gravity * other.w / distance_sq.powf(ubo.power / 2.0);
+        // Calculate forces from particles in shared memory
+        for j in 0..256 {  // gl_WorkGroupSize.x = 256
+            let other = shared_data[j];
+            let len = other.xyz() - position.xyz();
+            let distance_sq = len.dot(len) + ubo.soften;
+            acceleration += ubo.gravity * len * other.w / distance_sq.powf(ubo.power * 0.5);
+        }
+        
+        // Synchronize before next iteration
+        unsafe {
+            workgroup_memory_barrier_with_group_sync();
+        }
 
-        acceleration = acceleration + vec4(
-            len.x * force_magnitude, 
-            len.y * force_magnitude, 
-            len.z * force_magnitude, 
-            0.0
-        );
+        i += SHARED_DATA_SIZE as u32;
     }
 
     // Update velocity with acceleration