Change extracts of pressure byutes to a f32x4_make. Change andnot masking to a bitselect. Rename readme.md to readme.txt to hopefully fix gitea formatting

README.txt

@@ -9,6 +9,7 @@ Release:
         - Z-prepass fringe bug (also, when do we enable the prepass?)
         - Textured quads (pictures, code already written in older version)
         - Resize and move pictures (draw handles)
+        - SIMD for LOD?
     + Bugs
         + GC stalls!!!
         + Stroke previews get connected when drawn without panning on touch devices
@@ -21,10 +22,10 @@ Release:
         + Fix blinking own stroke inbetween SYN->server and SYN->client
         + Drag with mouse button 3
         + Investigate skipped inputs on mobile (panning, zooming) [Events were not actually getting skipped. The stroke previews were just not being drawn]
+        - Be able to have multiple "current" strokes per player. In case of bad internet this can happen!
         - Do NOT use session id as player id LUL
         - Save events to indexeddb (as some kind of a blob), restore on reconnect and page reload
         - Local prediction for tools!
-        - Be able to have multiple "current" strokes per player. In case of bad internet this can happen!
     * Missing features I do not consider bonus
         + Player pointers
         + Pretty player pointers
@@ -52,7 +53,7 @@ Bonus:
         + Add pressure data to quads
         + Account for pressure in quad/bbox calc
         + Adjust curve simplification to include pressure info
-        - Migrate old non-pressure desks
+        + Migrate old non-pressure desks
         - Check out e.pressure on touch devices
         - Send pressure in PREDRAW event
     - Curve modification

client/wasm/lod.c

@@ -62,6 +62,7 @@ rdp_find_max(float *xs, float *ys, unsigned char *pressures, float zoom, int coo
     float dist_ab = __builtin_sqrtf(dx * dx + dy * dy);
     float dir_nx = dy / dist_ab * 255.0f;
     float dir_ny = -dx / dist_ab * 255.0f;
+
 #if 0
         for (int i = segment_start + 1; i < segment_end; ++i) {
             float px = xs[coords_from + i];
@@ -81,7 +82,6 @@ rdp_find_max(float *xs, float *ys, unsigned char *pressures, float zoom, int coo
             }
         }
 #else
-    v128_t eps_x4 = wasm_f32x4_splat(EPS);
     v128_t ax_x4 = wasm_f32x4_splat(ax);
     v128_t ay_x4 = wasm_f32x4_splat(ay);
     v128_t ap_x4 = wasm_f32x4_splat(ap);
@@ -91,18 +91,19 @@ rdp_find_max(float *xs, float *ys, unsigned char *pressures, float zoom, int coo
 
     v128_t index_x4 = wasm_u32x4_make(segment_start + 1, segment_start + 2, segment_start + 3, segment_start + 4);
     v128_t four_x4 = wasm_u32x4_const_splat(4);
-    v128_t max_vals_x4 = wasm_f32x4_const_splat(0.0f);
+    v128_t max_dist_x4 = wasm_f32x4_splat(EPS);
     v128_t max_index_x4 = wasm_u32x4_const_splat(-1);
 
     for (int i = segment_start + 1; i < segment_end - 3; i += 4) {
         v128_t px_x4 = wasm_v128_load(xs + coords_from + i);
         v128_t py_x4 = wasm_v128_load(ys + coords_from + i);
-        v128_t pp_x16 = wasm_v128_load(pressures + coords_from / 2 + i);
 
-        // Take 4 highest bytes and convert to float
-        v128_t pp_x8 = wasm_u16x8_extend_high_u8x16(pp_x16);
-        v128_t pp_x4i = wasm_u32x4_extend_high_u16x8(pp_x8);
-        v128_t pp_x4 = wasm_f32x4_convert_i32x4(pp_x4i); // i version is 8 times faster on x64?
+        v128_t pp_x4 = wasm_f32x4_make(
+            pressures[coords_from / 2 + i + 0],
+            pressures[coords_from / 2 + i + 1],
+            pressures[coords_from / 2 + i + 2],
+            pressures[coords_from / 2 + i + 3]
+        );
 
         v128_t apx_x4 = wasm_f32x4_sub(px_x4, ax_x4);
         v128_t apy_x4 = wasm_f32x4_sub(py_x4, ay_x4);
@@ -120,19 +121,10 @@ rdp_find_max(float *xs, float *ys, unsigned char *pressures, float zoom, int coo
             )
         );
         
-        v128_t dist_mask = wasm_f32x4_gt(dist_x4, eps_x4);
-        v128_t max_mask = wasm_f32x4_gt(dist_x4, max_vals_x4);
-        v128_t mask = wasm_v128_and(dist_mask, max_mask);
-
-        max_index_x4 = wasm_v128_or(
-            wasm_v128_and(index_x4, mask),
-            wasm_v128_andnot(max_index_x4, mask)
-        );
+        v128_t mask = wasm_f32x4_gt(dist_x4, max_dist_x4);
 
-        max_vals_x4 = wasm_v128_or(
-            wasm_v128_and(dist_x4, mask),
-            wasm_v128_andnot(max_vals_x4, mask)
-        );
+        max_index_x4 = wasm_v128_bitselect(index_x4, max_index_x4, mask);
+        max_dist_x4 = wasm_v128_bitselect(dist_x4, max_dist_x4, mask);
         
         index_x4 = wasm_i32x4_add(index_x4, four_x4);
     }
@@ -141,7 +133,7 @@ rdp_find_max(float *xs, float *ys, unsigned char *pressures, float zoom, int coo
     float values[4];
 
     wasm_v128_store(indices, max_index_x4);
-    wasm_v128_store(values, max_vals_x4);
+    wasm_v128_store(values, max_dist_x4);
 
     for (int i = 0; i < 4; ++i) {
         if (indices[i] != -1) {
@@ -152,6 +144,11 @@ rdp_find_max(float *xs, float *ys, unsigned char *pressures, float zoom, int coo
         }
     }
 
+    if (max_dist == EPS) {
+        max_dist = 0.0f;
+        result = -1;
+    }
+
     int remainder = (segment_end - segment_start - 1) % 4;
 
     for (int i = segment_end - remainder; i < segment_end; ++i) {
@@ -171,7 +168,6 @@ rdp_find_max(float *xs, float *ys, unsigned char *pressures, float zoom, int coo
             max_dist = dist;
         }
     }
-
 #endif
 
     return(result);