(more) branchless serializing

JairusSW · Nov 25, 2024 · bec20c0 · bec20c0
1 parent 64f5797
commit bec20c0
Show file tree

Hide file tree

Showing 2 changed files with 46 additions and 42 deletions.
diff --git a/assembly/serialize/simd/string.ts b/assembly/serialize/simd/string.ts
@@ -1,14 +1,31 @@
 import { OBJECT, TOTAL_OVERHEAD } from "rt/common";
 
 const ESCAPE_TABLE = memory.data<u16>([
-    48,  48,    48,  49,   48,  50,    48,  51, // 0-3
-    48,  52,    48,  53,   48,  54,    48,  55, // 4-7
-    48,  56,    92, 116,   92, 110,    48,  98, // 8-11
-    92, 102,    92, 114,   48, 101,    48, 102, // 12-15
-    49,  48,    49,  49,   49,  50,    49,  51, // 16-19
-    49,  52,    49,  53,   49,  54,    49,  55, // 20-23
-    49,  56,    49,  57,   49,  97,    49,  98, // 24-27
-    49,  99,    49, 100,   49, 101,    49, 102, // 28-31
+    48, 48, 48, 49, 48, 50, 48, 51, // 0-3
+    48, 52, 48, 53, 48, 54, 48, 55, // 4-7
+    48, 56, 92, 116, 92, 110, 48, 98, // 8-11
+    92, 102, 92, 114, 48, 101, 48, 102, // 12-15
+    49, 48, 49, 49, 49, 50, 49, 51, // 16-19
+    49, 52, 49, 53, 49, 54, 49, 55, // 20-23
+    49, 56, 49, 57, 49, 97, 49, 98, // 24-27
+    49, 99, 49, 100, 49, 101, 49, 102, // 28-31
+    0, 0, 0, 0, 92, 34, 0, 0, // 32-35
+    0, 0, 0, 0, 0, 0, 0, 0, // 36-39
+    0, 0, 0, 0, 0, 0, 0, 0, // 40-43
+    0, 0, 0, 0, 0, 0, 0, 0, // 44-47
+    0, 0, 0, 0, 0, 0, 0, 0, // 48-51
+    0, 0, 0, 0, 0, 0, 0, 0, // 52-55
+    0, 0, 0, 0, 0, 0, 0, 0, // 56-59
+    0, 0, 0, 0, 0, 0, 0, 0, // 60-63
+    0, 0, 0, 0, 0, 0, 0, 0, // 64-67
+    0, 0, 0, 0, 0, 0, 0, 0, // 68-71
+    0, 0, 0, 0, 0, 0, 0, 0, // 72-75
+    0, 0, 0, 0, 0, 0, 0, 0, // 76-79
+    0, 0, 0, 0, 0, 0, 0, 0, // 80-83
+    0, 0, 0, 0, 0, 0, 0, 0, // 86-89
+    0, 0, 0, 0, 92, 92, 0, 0, // 90-93
+    0, 0, 0, 0, 0, 0, 0, 0, // 94-97
+    0, 0, 0, 0, 0, 0, 0, 0, // 98-101
 ]);
 
 const SPLAT_34 = i16x8.splat(34); /* " */
@@ -41,41 +58,25 @@ const SPLAT_32 = i16x8.splat(32); /* [ESC] */
         const escape_indices = i16x8.lt_u(block, SPLAT_32);
         const sieve = v128.or(char_indices, escape_indices);
 
-        if (v128.any_true(sieve)) {
-            let char_mask = i16x8.bitmask(char_indices);
-            let escape_mask = i16x8.bitmask(escape_indices);
+        v128.store(dst_ptr, block);
 
-            v128.store(dst_ptr, block);
+        if (v128.any_true(sieve)) {
+            let sieve_mask = i16x8.bitmask(sieve);
+            do {
+                const lane_index = ctz(sieve_mask) << 1;
+                const dst_offset = dst_ptr + lane_index;
+                const src_offset = src_ptr + lane_index;
+                const table_index = load<u16>(src_offset) << 2;
+                const escaped = load<u32>(ESCAPE_TABLE + table_index);
+                store<u32>(dst_offset, escaped);
+                v128.store(dst_offset, v128.load(src_offset, 2), 4);
+                sieve_mask &= sieve_mask - 1;
+                dst_ptr += 2;
+            } while (sieve_mask != 0);
 
-            if (char_mask != 0) {
-                do {
-                    const lane_index = ctz(char_mask) << 1;
-                    const dst_offset = dst_ptr + lane_index;
-                    store<u16>(dst_offset, 92); /* \ */
-                    v128.store(dst_offset, v128.load(src_ptr + lane_index), 2);
-                    char_mask &= char_mask - 1;
-                    dst_ptr += 2;
-                } while (char_mask != 0);
-                // dst_ptr += offset;
-            }
-            if (escape_mask != 0) {
-                do {
-                    const lane_index = ctz(escape_mask) << 1;
-                    const dst_offset = dst_ptr + lane_index;
-                    const src_offset = src_ptr + lane_index;
-                    const table_index = load<u16>(src_offset) << 2;
-                    store<u32>(dst_offset, load<u32>(ESCAPE_TABLE + table_index));
-                    v128.store(dst_offset, v128.load(src_offset, 2), 4);
-                    escape_mask &= escape_mask - 1;
-                    dst_ptr += 4;
-                } while (escape_mask != 0);
-            }
-        } else {
-            v128.store(dst_ptr, block);
+            src_ptr += 16;
+            dst_ptr += 16;
         }
-
-        src_ptr += 16;
-        dst_ptr += 16;
     }
 
     do {

diff --git a/assembly/test.ts b/assembly/test.ts
@@ -1,6 +1,8 @@
 import { JSON } from "."
 import { serializeString_SIMD } from "./serialize/simd/string";
-
+const a = memory.data<u16>([
+  92, 117, 48, 48
+])
 @json
 class Vec3 {
   public x: i32 = 0;
@@ -29,8 +31,9 @@ class Vec3 {
 //   // ^ this is not okay
 // }
 const out = new ArrayBuffer(128);
-const len = serializeString_SIMD("h\\e\tllo wor\"ld", changetype<usize>(out));
+const len = serializeString_SIMD("h\ne\tllo wor\"ld", changetype<usize>(out));
 const serialized = String.UTF16.decodeUnsafe(changetype<usize>(out), out.byteLength);
 console.log("Serialized: " + serialized);
+console.log(load<u64>(a).toString())
 // const deserialized = JSON.parseSafe<Vec3>(`{"x":1,"y":true,"z":3}`);
 // console.log("Deserialized: " + JSON.stringify(deserialized));