diff --git a/embed/bcw2/bcw2.wasm b/embed/bcw2/bcw2.wasm
index 837fa85..33157b2 100755
Binary files a/embed/bcw2/bcw2.wasm and b/embed/bcw2/bcw2.wasm differ
diff --git a/embed/bcw2/build.sh b/embed/bcw2/build.sh
index 16b7808..b96ac1a 100755
--- a/embed/bcw2/build.sh
+++ b/embed/bcw2/build.sh
@@ -43,8 +43,8 @@ cd ~-
 
 "$WASI_SDK/clang" --target=wasm32-wasi -std=c23 -g0 -O2 \
 	-Wall -Wextra -Wno-unused-parameter -Wno-unused-function \
-	-o bcw2.wasm "build/main.c" \
-	-I"build" \
+	-o bcw2.wasm build/main.c \
+	-I"$ROOT/sqlite3/libc" -I"build" \
 	-mexec-model=reactor \
 	-msimd128 -mmutable-globals -mmultivalue \
 	-mbulk-memory -mreference-types \
diff --git a/embed/sqlite3.wasm b/embed/sqlite3.wasm
index 88ebe99..b46a351 100755
Binary files a/embed/sqlite3.wasm and b/embed/sqlite3.wasm differ
diff --git a/go.mod b/go.mod
index 3417b16..24a494e 100644
--- a/go.mod
+++ b/go.mod
@@ -8,16 +8,16 @@ require (
 	github.com/ncruces/julianday v1.0.0
 	github.com/ncruces/sort v0.1.5
 	github.com/tetratelabs/wazero v1.9.0
-	golang.org/x/crypto v0.37.0
-	golang.org/x/sys v0.32.0
+	golang.org/x/crypto v0.38.0
+	golang.org/x/sys v0.33.0
 )
 
 require (
 	github.com/dchest/siphash v1.2.3 // ext/bloom
 	github.com/google/uuid v1.6.0 // ext/uuid
 	github.com/psanford/httpreadat v0.1.0 // example
-	golang.org/x/sync v0.13.0 // test
-	golang.org/x/text v0.24.0 // ext/unicode
+	golang.org/x/sync v0.14.0 // test
+	golang.org/x/text v0.25.0 // ext/unicode
 	lukechampine.com/adiantum v1.1.1 // vfs/adiantum
 )
 
diff --git a/go.sum b/go.sum
index 55dc895..d901f5d 100644
--- a/go.sum
+++ b/go.sum
@@ -10,13 +10,13 @@ github.com/psanford/httpreadat v0.1.0 h1:VleW1HS2zO7/4c7c7zNl33fO6oYACSagjJIyMIw
 github.com/psanford/httpreadat v0.1.0/go.mod h1:Zg7P+TlBm3bYbyHTKv/EdtSJZn3qwbPwpfZ/I9GKCRE=
 github.com/tetratelabs/wazero v1.9.0 h1:IcZ56OuxrtaEz8UYNRHBrUa9bYeX9oVY93KspZZBf/I=
 github.com/tetratelabs/wazero v1.9.0/go.mod h1:TSbcXCfFP0L2FGkRPxHphadXPjo1T6W+CseNNY7EkjM=
-golang.org/x/crypto v0.37.0 h1:kJNSjF/Xp7kU0iB2Z+9viTPMW4EqqsrywMXLJOOsXSE=
-golang.org/x/crypto v0.37.0/go.mod h1:vg+k43peMZ0pUMhYmVAWysMK35e6ioLh3wB8ZCAfbVc=
-golang.org/x/sync v0.13.0 h1:AauUjRAJ9OSnvULf/ARrrVywoJDy0YS2AwQ98I37610=
-golang.org/x/sync v0.13.0/go.mod h1:1dzgHSNfp02xaA81J2MS99Qcpr2w7fw1gpm99rleRqA=
-golang.org/x/sys v0.32.0 h1:s77OFDvIQeibCmezSnk/q6iAfkdiQaJi4VzroCFrN20=
-golang.org/x/sys v0.32.0/go.mod h1:BJP2sWEmIv4KK5OTEluFJCKSidICx8ciO85XgH3Ak8k=
-golang.org/x/text v0.24.0 h1:dd5Bzh4yt5KYA8f9CJHCP4FB4D51c2c6JvN37xJJkJ0=
-golang.org/x/text v0.24.0/go.mod h1:L8rBsPeo2pSS+xqN0d5u2ikmjtmoJbDBT1b7nHvFCdU=
+golang.org/x/crypto v0.38.0 h1:jt+WWG8IZlBnVbomuhg2Mdq0+BBQaHbtqHEFEigjUV8=
+golang.org/x/crypto v0.38.0/go.mod h1:MvrbAqul58NNYPKnOra203SB9vpuZW0e+RRZV+Ggqjw=
+golang.org/x/sync v0.14.0 h1:woo0S4Yywslg6hp4eUFjTVOyKt0RookbpAHG4c1HmhQ=
+golang.org/x/sync v0.14.0/go.mod h1:1dzgHSNfp02xaA81J2MS99Qcpr2w7fw1gpm99rleRqA=
+golang.org/x/sys v0.33.0 h1:q3i8TbbEz+JRD9ywIRlyRAQbM0qF7hu24q3teo2hbuw=
+golang.org/x/sys v0.33.0/go.mod h1:BJP2sWEmIv4KK5OTEluFJCKSidICx8ciO85XgH3Ak8k=
+golang.org/x/text v0.25.0 h1:qVyWApTSYLk/drJRO5mDlNYskwQznZmkpV2c8q9zls4=
+golang.org/x/text v0.25.0/go.mod h1:WEdwpYrmk1qmdHvhkSTNPm3app7v4rsT8F2UD6+VHIA=
 lukechampine.com/adiantum v1.1.1 h1:4fp6gTxWCqpEbLy40ExiYDDED3oUNWx5cTqBCtPdZqA=
 lukechampine.com/adiantum v1.1.1/go.mod h1:LrAYVnTYLnUtE/yMp5bQr0HstAf060YUF8nM0B6+rUw=
diff --git a/sqlite3/libc/build.sh b/sqlite3/libc/build.sh
index fd4661b..cfb717c 100755
--- a/sqlite3/libc/build.sh
+++ b/sqlite3/libc/build.sh
@@ -27,16 +27,24 @@ EOF
 	-Wl,--stack-first \
 	-Wl,--import-undefined \
 	-Wl,--initial-memory=16777216 \
+	-Wl,--export=memccpy \
 	-Wl,--export=memchr \
 	-Wl,--export=memcmp \
 	-Wl,--export=memcpy \
+	-Wl,--export=memmove \
+	-Wl,--export=memrchr \
 	-Wl,--export=memset \
+	-Wl,--export=stpcpy \
+	-Wl,--export=stpncpy \
 	-Wl,--export=strchr \
 	-Wl,--export=strchrnul \
 	-Wl,--export=strcmp \
+	-Wl,--export=strcpy \
 	-Wl,--export=strcspn \
 	-Wl,--export=strlen \
+	-Wl,--export=strncat \
 	-Wl,--export=strncmp \
+	-Wl,--export=strncpy \
 	-Wl,--export=strrchr \
 	-Wl,--export=strspn \
 	-Wl,--export=qsort
diff --git a/sqlite3/libc/libc.wasm b/sqlite3/libc/libc.wasm
index 1efb229..be51968 100755
Binary files a/sqlite3/libc/libc.wasm and b/sqlite3/libc/libc.wasm differ
diff --git a/sqlite3/libc/libc.wat b/sqlite3/libc/libc.wat
index e8e5d8f..d988fcd 100644
--- a/sqlite3/libc/libc.wat
+++ b/sqlite3/libc/libc.wat
@@ -2,15 +2,18 @@
  (type $0 (func (param i32 i32) (result i32)))
  (type $1 (func (param i32 i32 i32) (result i32)))
  (type $2 (func (param i32) (result i32)))
- (type $3 (func (param i32 i32 i32 i32)))
+ (type $3 (func (param i32 i32 i32 i32) (result i32)))
+ (type $4 (func (param i32 i32 i32 i32)))
  (memory $0 256)
  (data $0 (i32.const 1024) "\01")
  (table $0 1 1 funcref)
  (export "memory" (memory $0))
  (export "memset" (func $memset))
  (export "memcpy" (func $memcpy))
+ (export "memmove" (func $memcpy))
  (export "memcmp" (func $memcmp))
  (export "memchr" (func $memchr))
+ (export "memrchr" (func $memrchr))
  (export "strlen" (func $strlen))
  (export "strcmp" (func $strcmp))
  (export "strncmp" (func $strncmp))
@@ -19,6 +22,12 @@
  (export "strrchr" (func $strrchr))
  (export "strspn" (func $strspn))
  (export "strcspn" (func $strcspn))
+ (export "memccpy" (func $memccpy))
+ (export "strncat" (func $strncat))
+ (export "stpcpy" (func $stpcpy))
+ (export "strcpy" (func $strcpy))
+ (export "stpncpy" (func $stpncpy))
+ (export "strncpy" (func $strncpy))
  (export "qsort" (func $qsort))
  (func $memset (param $0 i32) (param $1 i32) (param $2 i32) (result i32)
   (memory.fill
@@ -37,130 +46,142 @@
   (local.get $0)
  )
  (func $memcmp (param $0 i32) (param $1 i32) (param $2 i32) (result i32)
-  (local $3 v128)
+  (local $3 i32)
   (local $4 i32)
-  (local $5 i32)
+  (local $5 v128)
   (block $block
-   (br_if $block
-    (i32.lt_u
+   (if
+    (i32.ge_u
      (local.get $2)
      (i32.const 16)
     )
-   )
-   (loop $label
-    (if
-     (i8x16.all_true
-      (local.tee $3
-       (i8x16.eq
-        (v128.load align=1
-         (local.get $0)
+    (then
+     (loop $label
+      (if
+       (i32.eqz
+        (i8x16.all_true
+         (local.tee $5
+          (i8x16.eq
+           (v128.load align=1
+            (local.get $0)
+           )
+           (v128.load align=1
+            (local.get $1)
+           )
+          )
+         )
         )
-        (v128.load align=1
-         (local.get $1)
+       )
+       (then
+        (return
+         (i32.sub
+          (i32.load8_u
+           (i32.add
+            (local.get $0)
+            (local.tee $2
+             (i32.ctz
+              (i32.xor
+               (i8x16.bitmask
+                (local.get $5)
+               )
+               (i32.const -1)
+              )
+             )
+            )
+           )
+          )
+          (i32.load8_u
+           (i32.add
+            (local.get $1)
+            (local.get $2)
+           )
+          )
+         )
         )
        )
       )
-     )
-     (then
       (local.set $1
        (i32.add
         (local.get $1)
-        (i32.const 16)
+        (local.tee $3
+         (i32.add
+          (i32.and
+           (i32.sub
+            (local.get $2)
+            (i32.const 1)
+           )
+           (i32.const 15)
+          )
+          (i32.const 1)
+         )
+        )
        )
       )
       (local.set $0
        (i32.add
         (local.get $0)
-        (i32.const 16)
+        (local.get $3)
        )
       )
       (br_if $label
-       (i32.gt_u
-        (local.tee $2
-         (i32.sub
-          (local.get $2)
-          (i32.const 16)
-         )
-        )
-        (i32.const 15)
-       )
-      )
-      (br $block)
-     )
-    )
-   )
-   (return
-    (i32.sub
-     (i32.load8_u
-      (i32.add
-       (local.get $0)
        (local.tee $2
-        (i32.ctz
-         (i32.xor
-          (i8x16.bitmask
-           (local.get $3)
-          )
-          (i32.const -1)
-         )
+        (i32.sub
+         (local.get $2)
+         (local.get $3)
         )
        )
       )
      )
-     (i32.load8_u
-      (i32.add
-       (local.get $1)
-       (local.get $2)
+     (br $block)
+    )
+   )
+   (br_if $block
+    (i32.eqz
+     (local.get $2)
+    )
+   )
+   (loop $label1
+    (if
+     (i32.ne
+      (local.tee $3
+       (i32.load8_u
+        (local.get $0)
+       )
+      )
+      (local.tee $4
+       (i32.load8_u
+        (local.get $1)
+       )
+      )
+     )
+     (then
+      (return
+       (i32.sub
+        (local.get $3)
+        (local.get $4)
+       )
       )
      )
     )
-   )
-  )
-  (if
-   (local.get $2)
-   (then
-    (loop $label1
-     (if
-      (i32.ne
-       (local.tee $4
-        (i32.load8_u
-         (local.get $0)
-        )
-       )
-       (local.tee $5
-        (i32.load8_u
-         (local.get $1)
-        )
-       )
-      )
-      (then
-       (return
-        (i32.sub
-         (local.get $4)
-         (local.get $5)
-        )
-       )
-      )
+    (local.set $1
+     (i32.add
+      (local.get $1)
+      (i32.const 1)
      )
-     (local.set $1
-      (i32.add
-       (local.get $1)
+    )
+    (local.set $0
+     (i32.add
+      (local.get $0)
+      (i32.const 1)
+     )
+    )
+    (br_if $label1
+     (local.tee $2
+      (i32.sub
+       (local.get $2)
        (i32.const 1)
       )
      )
-     (local.set $0
-      (i32.add
-       (local.get $0)
-       (i32.const 1)
-      )
-     )
-     (br_if $label1
-      (local.tee $2
-       (i32.sub
-        (local.get $2)
-        (i32.const 1)
-       )
-      )
-     )
     )
    )
   )
@@ -324,6 +345,118 @@
    )
   )
  )
+ (func $memrchr (param $0 i32) (param $1 i32) (param $2 i32) (result i32)
+  (local $3 v128)
+  (local $4 v128)
+  (block $block
+   (br_if $block
+    (i32.lt_u
+     (local.get $2)
+     (i32.const 16)
+    )
+   )
+   (local.set $3
+    (i8x16.splat
+     (local.get $1)
+    )
+   )
+   (loop $label
+    (if
+     (i32.eqz
+      (v128.any_true
+       (local.tee $4
+        (i8x16.eq
+         (v128.load align=1
+          (i32.sub
+           (i32.add
+            (local.get $0)
+            (local.get $2)
+           )
+           (i32.const 16)
+          )
+         )
+         (local.get $3)
+        )
+       )
+      )
+     )
+     (then
+      (br_if $label
+       (i32.gt_u
+        (local.tee $2
+         (i32.sub
+          (local.get $2)
+          (i32.const 16)
+         )
+        )
+        (i32.const 15)
+       )
+      )
+      (br $block)
+     )
+    )
+   )
+   (return
+    (i32.add
+     (i32.add
+      (i32.sub
+       (local.get $0)
+       (i32.clz
+        (i8x16.bitmask
+         (local.get $4)
+        )
+       )
+      )
+      (local.get $2)
+     )
+     (i32.const 15)
+    )
+   )
+  )
+  (local.set $0
+   (i32.add
+    (local.get $0)
+    (local.get $2)
+   )
+  )
+  (local.set $1
+   (i32.extend8_s
+    (local.get $1)
+   )
+  )
+  (loop $label1
+   (if
+    (i32.eqz
+     (local.get $2)
+    )
+    (then
+     (return
+      (i32.const 0)
+     )
+    )
+   )
+   (local.set $2
+    (i32.sub
+     (local.get $2)
+     (i32.const 1)
+    )
+   )
+   (br_if $label1
+    (i32.ne
+     (local.get $1)
+     (i32.load8_s
+      (local.tee $0
+       (i32.sub
+        (local.get $0)
+        (i32.const 1)
+       )
+      )
+     )
+    )
+   )
+  )
+  (local.get $0)
+ )
  (func $strlen (param $0 i32) (result i32)
   (local $1 i32)
   (local $2 i32)
@@ -912,123 +1045,15 @@
   )
  )
  (func $strrchr (param $0 i32) (param $1 i32) (result i32)
-  (local $2 i32)
-  (local $3 v128)
-  (local $4 v128)
-  (block $block1 (result i32)
-   (block $block
-    (br_if $block
-     (i32.lt_u
-      (local.tee $2
-       (i32.add
-        (call $strlen
-         (local.get $0)
-        )
-        (i32.const 1)
-       )
-      )
-      (i32.const 16)
-     )
-    )
-    (local.set $3
-     (i8x16.splat
-      (local.get $1)
-     )
-    )
-    (loop $label
-     (if
-      (i32.eqz
-       (v128.any_true
-        (local.tee $4
-         (i8x16.eq
-          (v128.load align=1
-           (i32.sub
-            (i32.add
-             (local.get $0)
-             (local.get $2)
-            )
-            (i32.const 16)
-           )
-          )
-          (local.get $3)
-         )
-        )
-       )
-      )
-      (then
-       (br_if $label
-        (i32.gt_u
-         (local.tee $2
-          (i32.sub
-           (local.get $2)
-           (i32.const 16)
-          )
-         )
-         (i32.const 15)
-        )
-       )
-       (br $block)
-      )
-     )
-    )
-    (br $block1
-     (i32.add
-      (i32.add
-       (i32.sub
-        (local.get $0)
-        (i32.clz
-         (i8x16.bitmask
-          (local.get $4)
-         )
-        )
-       )
-       (local.get $2)
-      )
-      (i32.const 15)
-     )
-    )
-   )
-   (local.set $0
-    (i32.add
-     (local.get $0)
-     (local.get $2)
-    )
-   )
-   (local.set $1
-    (i32.extend8_s
-     (local.get $1)
-    )
-   )
-   (loop $label1
-    (drop
-     (br_if $block1
-      (i32.const 0)
-      (i32.eqz
-       (local.get $2)
-      )
-     )
-    )
-    (local.set $2
-     (i32.sub
-      (local.get $2)
-      (i32.const 1)
-     )
-    )
-    (br_if $label1
-     (i32.ne
-      (local.get $1)
-      (i32.load8_s
-       (local.tee $0
-        (i32.sub
-         (local.get $0)
-         (i32.const 1)
-        )
-       )
-      )
-     )
-    )
-   )
+  (call $memrchr
    (local.get $0)
+   (local.get $1)
+   (i32.add
+    (call $strlen
+     (local.get $0)
+    )
+    (i32.const 1)
+   )
   )
  )
  (func $strspn (param $0 i32) (param $1 i32) (result i32)
@@ -1170,69 +1195,10 @@
      (br $block)
     )
    )
-   (v128.store
-    (i32.const 1280)
-    (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-   )
-   (v128.store
-    (i32.const 1264)
-    (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-   )
-   (v128.store
-    (i32.const 1248)
-    (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-   )
-   (v128.store
-    (i32.const 1232)
-    (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-   )
-   (v128.store
-    (i32.const 1216)
-    (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-   )
-   (v128.store
-    (i32.const 1200)
-    (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-   )
-   (v128.store
-    (i32.const 1184)
-    (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-   )
-   (v128.store
-    (i32.const 1168)
-    (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-   )
-   (v128.store
-    (i32.const 1152)
-    (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-   )
-   (v128.store
-    (i32.const 1136)
-    (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-   )
-   (v128.store
-    (i32.const 1120)
-    (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-   )
-   (v128.store
-    (i32.const 1104)
-    (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-   )
-   (v128.store
-    (i32.const 1088)
-    (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-   )
-   (v128.store
-    (i32.const 1072)
-    (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-   )
-   (v128.store
-    (i32.const 1056)
-    (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-   )
-   (v128.store
+   (memory.fill
     (i32.const 1040)
-    (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
+    (i32.const 0)
+    (i32.const 256)
    )
    (local.set $1
     (i32.add
@@ -1368,272 +1334,211 @@
   (local $3 v128)
   (local $4 v128)
   (local $scratch i32)
-  (block $block1
-   (block $block2
-    (block $block3
-     (block $block
-      (br_if $block
-       (i32.eqz
-        (local.tee $2
-         (i32.load8_u
-          (local.get $1)
-         )
-        )
-       )
+  (block $block
+   (if
+    (local.tee $2
+     (i32.load8_u
+      (local.get $1)
+     )
+    )
+    (then
+     (br_if $block
+      (i32.load8_u offset=1
+       (local.get $1)
       )
-      (br_if $block
-       (i32.eqz
-        (i32.load8_u offset=1
-         (local.get $1)
-        )
-       )
-      )
-      (v128.store
-       (i32.const 1536)
-       (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-      )
-      (v128.store
-       (i32.const 1520)
-       (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-      )
-      (v128.store
-       (i32.const 1504)
-       (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-      )
-      (v128.store
-       (i32.const 1488)
-       (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-      )
-      (v128.store
-       (i32.const 1472)
-       (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-      )
-      (v128.store
-       (i32.const 1456)
-       (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-      )
-      (v128.store
-       (i32.const 1440)
-       (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-      )
-      (v128.store
-       (i32.const 1424)
-       (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-      )
-      (v128.store
-       (i32.const 1408)
-       (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-      )
-      (v128.store
-       (i32.const 1392)
-       (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-      )
-      (v128.store
-       (i32.const 1376)
-       (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-      )
-      (v128.store
-       (i32.const 1360)
-       (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-      )
-      (v128.store
-       (i32.const 1344)
-       (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-      )
-      (v128.store
-       (i32.const 1328)
-       (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-      )
-      (v128.store
-       (i32.const 1312)
-       (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-      )
-      (v128.store
-       (i32.const 1296)
-       (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-      )
-      (loop $label
-       (i32.store8
-        (i32.add
-         (local.tee $2
-          (i32.load8_u
-           (local.get $1)
-          )
-         )
-         (i32.const 1296)
-        )
-        (i32.const 1)
-       )
-       (local.set $1
-        (i32.add
-         (local.get $1)
-         (i32.const 1)
-        )
-       )
-       (br_if $label
-        (local.get $2)
-       )
-      )
-      (local.set $1
-       (local.get $0)
-      )
-      (loop $label1
-       (br_if $block1
-        (i32.load8_u
-         (i32.add
-          (i32.load8_u
-           (local.get $1)
-          )
-          (i32.const 1296)
-         )
-        )
-       )
-       (br_if $block2
-        (i32.load8_u
-         (i32.add
-          (i32.load8_u offset=1
-           (local.get $1)
-          )
-          (i32.const 1296)
-         )
-        )
-       )
-       (br_if $block3
-        (i32.load8_u
-         (i32.add
-          (i32.load8_u offset=2
-           (local.get $1)
-          )
-          (i32.const 1296)
-         )
-        )
-       )
-       (br_if $label1
-        (i32.eqz
-         (i32.load8_u
-          (i32.add
-           (block (result i32)
-            (local.set $scratch
-             (i32.load8_u offset=3
-              (local.get $1)
-             )
+     )
+    )
+   )
+   (block $block1
+    (if
+     (v128.any_true
+      (local.tee $3
+       (v128.or
+        (i8x16.eq
+         (local.tee $3
+          (v128.load
+           (local.tee $1
+            (i32.and
+             (local.get $0)
+             (i32.const -16)
             )
-            (local.set $1
-             (i32.add
-              (local.get $1)
-              (i32.const 4)
-             )
-            )
-            (local.get $scratch)
            )
-           (i32.const 1296)
+          )
+         )
+         (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
+        )
+        (i8x16.eq
+         (local.get $3)
+         (local.tee $4
+          (i8x16.splat
+           (local.get $2)
           )
          )
         )
        )
       )
-      (return
-       (i32.sub
-        (i32.sub
-         (local.get $1)
-         (i32.const 1)
-        )
-        (local.get $0)
-       )
-      )
      )
-     (block $block4
-      (if
+     (then
+      (br_if $block1
+       (local.tee $2
+        (i32.and
+         (i8x16.bitmask
+          (local.get $3)
+         )
+         (i32.shl
+          (i32.const -1)
+          (i32.and
+           (local.get $0)
+           (i32.const 15)
+          )
+         )
+        )
+       )
+      )
+     )
+    )
+    (loop $label
+     (local.set $3
+      (v128.load offset=16
+       (local.get $1)
+      )
+     )
+     (local.set $1
+      (i32.add
+       (local.get $1)
+       (i32.const 16)
+      )
+     )
+     (br_if $label
+      (i32.eqz
        (v128.any_true
         (local.tee $3
          (v128.or
           (i8x16.eq
-           (local.tee $3
-            (v128.load
-             (local.tee $1
-              (i32.and
-               (local.get $0)
-               (i32.const -16)
-              )
-             )
-            )
-           )
+           (local.get $3)
            (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
           )
           (i8x16.eq
            (local.get $3)
-           (local.tee $4
-            (i8x16.splat
-             (local.get $2)
-            )
-           )
-          )
-         )
-        )
-       )
-       (then
-        (br_if $block4
-         (local.tee $2
-          (i32.and
-           (i8x16.bitmask
-            (local.get $3)
-           )
-           (i32.shl
-            (i32.const -1)
-            (i32.and
-             (local.get $0)
-             (i32.const 15)
-            )
-           )
+           (local.get $4)
           )
          )
         )
        )
       )
-      (loop $label2
-       (local.set $3
-        (v128.load offset=16
-         (local.get $1)
-        )
-       )
-       (local.set $1
+     )
+    )
+    (local.set $2
+     (i8x16.bitmask
+      (local.get $3)
+     )
+    )
+   )
+   (return
+    (i32.sub
+     (i32.add
+      (local.get $1)
+      (i32.ctz
+       (local.get $2)
+      )
+     )
+     (local.get $0)
+    )
+   )
+  )
+  (memory.fill
+   (i32.const 1296)
+   (i32.const 0)
+   (i32.const 256)
+  )
+  (loop $label1
+   (i32.store8
+    (i32.add
+     (local.tee $2
+      (i32.load8_u
+       (local.get $1)
+      )
+     )
+     (i32.const 1296)
+    )
+    (i32.const 1)
+   )
+   (local.set $1
+    (i32.add
+     (local.get $1)
+     (i32.const 1)
+    )
+   )
+   (br_if $label1
+    (local.get $2)
+   )
+  )
+  (local.set $1
+   (local.get $0)
+  )
+  (block $block2
+   (block $block3
+    (block $block4
+     (loop $label2
+      (br_if $block2
+       (i32.load8_u
         (i32.add
-         (local.get $1)
-         (i32.const 16)
-        )
-       )
-       (br_if $label2
-        (i32.eqz
-         (v128.any_true
-          (local.tee $3
-           (v128.or
-            (i8x16.eq
-             (local.get $3)
-             (v128.const i32x4 0x00000000 0x00000000 0x00000000 0x00000000)
-            )
-            (i8x16.eq
-             (local.get $3)
-             (local.get $4)
-            )
-           )
-          )
+         (i32.load8_u
+          (local.get $1)
          )
+         (i32.const 1296)
         )
        )
       )
-      (local.set $2
-       (i8x16.bitmask
-        (local.get $3)
+      (br_if $block3
+       (i32.load8_u
+        (i32.add
+         (i32.load8_u offset=1
+          (local.get $1)
+         )
+         (i32.const 1296)
+        )
+       )
+      )
+      (br_if $block4
+       (i32.load8_u
+        (i32.add
+         (i32.load8_u offset=2
+          (local.get $1)
+         )
+         (i32.const 1296)
+        )
+       )
+      )
+      (br_if $label2
+       (i32.eqz
+        (i32.load8_u
+         (i32.add
+          (block (result i32)
+           (local.set $scratch
+            (i32.load8_u offset=3
+             (local.get $1)
+            )
+           )
+           (local.set $1
+            (i32.add
+             (local.get $1)
+             (i32.const 4)
+            )
+           )
+           (local.get $scratch)
+          )
+          (i32.const 1296)
+         )
+        )
        )
       )
      )
      (return
       (i32.sub
-       (i32.add
+       (i32.sub
         (local.get $1)
-        (i32.ctz
-         (local.get $2)
-        )
+        (i32.const 1)
        )
        (local.get $0)
       )
@@ -1661,6 +1566,148 @@
    (local.get $0)
   )
  )
+ (func $memccpy (param $0 i32) (param $1 i32) (param $2 i32) (param $3 i32) (result i32)
+  (memory.copy
+   (local.get $0)
+   (local.get $1)
+   (select
+    (local.tee $1
+     (i32.add
+      (i32.sub
+       (local.tee $2
+        (call $memchr
+         (local.get $1)
+         (local.get $2)
+         (local.get $3)
+        )
+       )
+       (local.get $1)
+      )
+      (i32.const 1)
+     )
+    )
+    (local.get $3)
+    (local.get $2)
+   )
+  )
+  (select
+   (i32.add
+    (local.get $0)
+    (local.get $1)
+   )
+   (i32.const 0)
+   (local.get $2)
+  )
+ )
+ (func $strncat (param $0 i32) (param $1 i32) (param $2 i32) (result i32)
+  (local $3 i32)
+  (memory.copy
+   (local.tee $3
+    (i32.add
+     (call $strlen
+      (local.get $0)
+     )
+     (local.get $0)
+    )
+   )
+   (local.get $1)
+   (local.tee $1
+    (call $strnlen
+     (local.get $1)
+     (local.get $2)
+    )
+   )
+  )
+  (i32.store8
+   (i32.add
+    (local.get $1)
+    (local.get $3)
+   )
+   (i32.const 0)
+  )
+  (local.get $0)
+ )
+ (func $stpcpy (param $0 i32) (param $1 i32) (result i32)
+  (memory.copy
+   (local.get $0)
+   (local.get $1)
+   (i32.add
+    (local.tee $1
+     (call $strlen
+      (local.get $1)
+     )
+    )
+    (i32.const 1)
+   )
+  )
+  (i32.add
+   (local.get $0)
+   (local.get $1)
+  )
+ )
+ (func $strcpy (param $0 i32) (param $1 i32) (result i32)
+  (memory.copy
+   (local.get $0)
+   (local.get $1)
+   (i32.add
+    (call $strlen
+     (local.get $1)
+    )
+    (i32.const 1)
+   )
+  )
+  (local.get $0)
+ )
+ (func $stpncpy (param $0 i32) (param $1 i32) (param $2 i32) (result i32)
+  (memory.copy
+   (local.get $0)
+   (local.get $1)
+   (local.tee $1
+    (call $strnlen
+     (local.get $1)
+     (local.get $2)
+    )
+   )
+  )
+  (memory.fill
+   (local.tee $0
+    (i32.add
+     (local.get $0)
+     (local.get $1)
+    )
+   )
+   (i32.const 0)
+   (i32.sub
+    (local.get $2)
+    (local.get $1)
+   )
+  )
+  (local.get $0)
+ )
+ (func $strncpy (param $0 i32) (param $1 i32) (param $2 i32) (result i32)
+  (memory.copy
+   (local.get $0)
+   (local.get $1)
+   (local.tee $1
+    (call $strnlen
+     (local.get $1)
+     (local.get $2)
+    )
+   )
+  )
+  (memory.fill
+   (i32.add
+    (local.get $0)
+    (local.get $1)
+   )
+   (i32.const 0)
+   (i32.sub
+    (local.get $2)
+    (local.get $1)
+   )
+  )
+  (local.get $0)
+ )
  (func $qsort (param $0 i32) (param $1 i32) (param $2 i32) (param $3 i32)
   (local $4 i32)
   (local $5 i32)
@@ -2107,6 +2154,23 @@
    )
   )
  )
+ (func $strnlen (param $0 i32) (param $1 i32) (result i32)
+  (local $2 i32)
+  (select
+   (i32.sub
+    (local.tee $2
+     (call $memchr
+      (local.get $0)
+      (i32.const 0)
+      (local.get $1)
+     )
+    )
+    (local.get $0)
+   )
+   (local.get $1)
+   (local.get $2)
+  )
+ )
  ;; features section: mutable-globals, nontrapping-float-to-int, simd, bulk-memory, sign-ext, reference-types, multivalue, bulk-memory-opt
 )
 
diff --git a/sqlite3/libc/string.h b/sqlite3/libc/string.h
index a12bfbd..67cc2ea 100644
--- a/sqlite3/libc/string.h
+++ b/sqlite3/libc/string.h
@@ -42,13 +42,25 @@ void *memmove(void *dest, const void *src, size_t n) {
 
 __attribute__((weak))
 int memcmp(const void *v1, const void *v2, size_t n) {
+  // Baseline algorithm.
+  if (n < sizeof(v128_t)) {
+    const unsigned char *u1 = (unsigned char *)v1;
+    const unsigned char *u2 = (unsigned char *)v2;
+    while (n--) {
+      if (*u1 != *u2) return *u1 - *u2;
+      u1++;
+      u2++;
+    }
+    return 0;
+  }
+
   // memcmp is allowed to read up to n bytes from each object.
   // Find the first different character in the objects.
   // Unaligned loads handle the case where the objects
   // have mismatching alignments.
   const v128_t *w1 = (v128_t *)v1;
   const v128_t *w2 = (v128_t *)v2;
-  for (; n >= sizeof(v128_t); n -= sizeof(v128_t)) {
+  while (n) {
     const v128_t cmp = wasm_i8x16_eq(wasm_v128_load(w1), wasm_v128_load(w2));
     // Bitmask is slow on AArch64, all_true is much faster.
     if (!wasm_i8x16_all_true(cmp)) {
@@ -60,17 +72,12 @@ int memcmp(const void *v1, const void *v2, size_t n) {
       __builtin_assume(*u1 - *u2 != 0);
       return *u1 - *u2;
     }
-    w1++;
-    w2++;
-  }
-
-  // Baseline algorithm.
-  const unsigned char *u1 = (unsigned char *)w1;
-  const unsigned char *u2 = (unsigned char *)w2;
-  while (n--) {
-    if (*u1 != *u2) return *u1 - *u2;
-    u1++;
-    u2++;
+    // This makes n a multiple of sizeof(v128_t)
+    // for every iteration except the first.
+    size_t align = (n - 1) % sizeof(v128_t) + 1;
+    w1 = (v128_t *)((char *)w1 + align);
+    w2 = (v128_t *)((char *)w2 + align);
+    n -= align;
   }
   return 0;
 }
@@ -359,29 +366,13 @@ size_t strspn(const char *s, const char *c) {
     return s - a;
   }
 
-#if !__OPTIMIZE__ || __OPTIMIZE_SIZE__
-
-  // Unoptimized version.
   memset(byteset, 0, sizeof(byteset));
-  while (*c && (byteset[*(unsigned char *)c] = 1)) c++;
-  while (byteset[*(unsigned char *)s]) s++;
-
-#else  // __OPTIMIZE__
-
-  // This is faster than memset.
-  // Going backward helps bounds check elimination.
-  volatile v128_t *w = (v128_t *)byteset;
-  #pragma unroll
-  for (size_t i = sizeof(byteset) / sizeof(v128_t); i--;) w[i] = (v128_t){};
-  static_assert(sizeof(byteset) % sizeof(v128_t) == 0);
-
   // Keeping byteset[0] = 0 avoids the next loop needing that check.
   while (*c && (byteset[*(unsigned char *)c] = 1)) c++;
-  #pragma unroll 4
+#if __OPTIMIZE__ && !__OPTIMIZE_SIZE__
+#pragma unroll 4
+#endif
   while (byteset[*(unsigned char *)s]) s++;
-
-#endif  // __OPTIMIZE__
-
   return s - a;
 }
 
@@ -395,29 +386,13 @@ size_t strcspn(const char *s, const char *c) {
 
   if (!c[0] || !c[1]) return __strchrnul(s, *c) - s;
 
-#if !__OPTIMIZE__ || __OPTIMIZE_SIZE__
-
-  // Unoptimized version.
   memset(byteset, 0, sizeof(byteset));
-  while ((byteset[*(unsigned char *)c] = 1) && *c) c++;
-  while (!byteset[*(unsigned char *)s]) s++;
-
-#else  // __OPTIMIZE__
-
-  // This is faster than memset.
-  // Going backward helps bounds check elimination.
-  volatile v128_t *w = (v128_t *)byteset;
-  #pragma unroll
-  for (size_t i = sizeof(byteset) / sizeof(v128_t); i--;) w[i] = (v128_t){};
-  static_assert(sizeof(byteset) % sizeof(v128_t) == 0);
-
   // Setting byteset[0] = 1 avoids the next loop needing that check.
   while ((byteset[*(unsigned char *)c] = 1) && *c) c++;
-  #pragma unroll 4
+#if __OPTIMIZE__ && !__OPTIMIZE_SIZE__
+#pragma unroll 4
+#endif
   while (!byteset[*(unsigned char *)s]) s++;
-
-#endif  // __OPTIMIZE__
-
   return s - a;
 }
 
@@ -435,8 +410,9 @@ size_t strcspn(const char *s, const char *c) {
 //  - strsep
 //  - strtok
 
-__attribute__((weak, always_inline))
+__attribute__((weak))
 void *memccpy(void *__restrict dest, const void *__restrict src, int c, size_t n) {
+  void *memchr(const void *v, int c, size_t n);
   const void *m = memchr(src, c, n);
   if (m != NULL) {
     n = (char *)m - (char *)src + 1;
@@ -446,15 +422,23 @@ void *memccpy(void *__restrict dest, const void *__restrict src, int c, size_t n
   return (void *)m;
 }
 
-__attribute__((weak, always_inline))
-char *stpcpy(char *__restrict dest, const char *__restrict src) {
+__attribute__((weak))
+char *strncat(char *__restrict dest, const char *__restrict src, size_t n) {
+  size_t strnlen(const char *s, size_t n);
+  size_t dlen = strlen(dest);
+  size_t slen = strnlen(src, n);
+  memcpy(dest + dlen, src, slen);
+  dest[dlen + slen] = 0;
+  return dest;
+}
+
+static char *__stpcpy(char *__restrict dest, const char *__restrict src) {
   size_t slen = strlen(src);
   memcpy(dest, src, slen + 1);
   return dest + slen;
 }
 
-__attribute__((weak, always_inline))
-char *stpncpy(char *__restrict dest, const char *__restrict src, size_t n) {
+static char *__stpncpy(char *__restrict dest, const char *__restrict src, size_t n) {
   size_t strnlen(const char *s, size_t n);
   size_t slen = strnlen(src, n);
   memcpy(dest, src, slen);
@@ -463,24 +447,23 @@ char *stpncpy(char *__restrict dest, const char *__restrict src, size_t n) {
 }
 
 __attribute__((weak, always_inline))
+char *stpcpy(char *__restrict dest, const char *__restrict src) {
+  return __stpcpy(dest, src);
+}
+
 char *strcpy(char *__restrict dest, const char *__restrict src) {
-  stpcpy(dest, src);
+  __stpcpy(dest, src);
   return dest;
 }
 
+__attribute__((weak, always_inline))
+char *stpncpy(char *__restrict dest, const char *__restrict src, size_t n) {
+  return __stpncpy(dest, src, n);
+}
+
 __attribute__((weak, always_inline))
 char *strncpy(char *__restrict dest, const char *__restrict src, size_t n) {
-  stpncpy(dest, src, n);
-  return dest;
-}
-
-__attribute__((weak, always_inline))
-char *strncat(char *__restrict dest, const char *__restrict src, size_t n) {
-  size_t strnlen(const char *s, size_t n);
-  size_t dlen = strlen(dest);
-  size_t slen = strnlen(src, n);
-  memcpy(dest + dlen, src, slen);
-  dest[dlen + slen] = 0;
+  __stpncpy(dest, src, n);
   return dest;
 }
 
diff --git a/sqlite3/libc/strings.h b/sqlite3/libc/strings.h
index 6915aba..9d427fd 100644
--- a/sqlite3/libc/strings.h
+++ b/sqlite3/libc/strings.h
@@ -16,24 +16,34 @@ __attribute__((weak))
 int bcmp(const void *v1, const void *v2, size_t n) {
   // bcmp is the same as memcmp but only compares for equality.
 
+  // Baseline algorithm.
+  if (n < sizeof(v128_t)) {
+    const unsigned char *u1 = (unsigned char *)v1;
+    const unsigned char *u2 = (unsigned char *)v2;
+    while (n--) {
+      if (*u1 != *u2) return 1;
+      u1++;
+      u2++;
+    }
+    return 0;
+  }
+
+  // bcmp is allowed to read up to n bytes from each object.
+  // Unaligned loads handle the case where the objects
+  // have mismatching alignments.
   const v128_t *w1 = (v128_t *)v1;
   const v128_t *w2 = (v128_t *)v2;
-  for (; n >= sizeof(v128_t); n -= sizeof(v128_t)) {
+  while (n) {
     // Find any single bit difference.
     if (wasm_v128_any_true(wasm_v128_load(w1) ^ wasm_v128_load(w2))) {
       return 1;
     }
-    w1++;
-    w2++;
-  }
-
-  // Continue byte-by-byte.
-  const unsigned char *u1 = (unsigned char *)w1;
-  const unsigned char *u2 = (unsigned char *)w2;
-  while (n--) {
-    if (*u1 != *u2) return 1;
-    u1++;
-    u2++;
+    // This makes n a multiple of sizeof(v128_t)
+    // for every iteration except the first.
+    size_t align = (n - 1) % sizeof(v128_t) + 1;
+    w1 = (v128_t *)((char *)w1 + align);
+    w2 = (v128_t *)((char *)w2 + align);
+    n -= align;
   }
   return 0;
 }
diff --git a/vfs/tests/mptest/wasm/mptest.wasm b/vfs/tests/mptest/wasm/mptest.wasm
index 1655131..2517ff4 100644
Binary files a/vfs/tests/mptest/wasm/mptest.wasm and b/vfs/tests/mptest/wasm/mptest.wasm differ
diff --git a/vfs/tests/speedtest1/wasm/speedtest1.wasm b/vfs/tests/speedtest1/wasm/speedtest1.wasm
index 299df6f..c206243 100644
Binary files a/vfs/tests/speedtest1/wasm/speedtest1.wasm and b/vfs/tests/speedtest1/wasm/speedtest1.wasm differ