diff --git a/.gitignore b/.gitignore
index 2830032..866c978 100644
--- a/.gitignore
+++ b/.gitignore
@@ -12,4 +12,6 @@
 # rspec failure tracking
 .rspec_status
 
+*.bundle
+
 .DS_Store
diff --git a/Rakefile b/Rakefile
index b89b97c..b146588 100644
--- a/Rakefile
+++ b/Rakefile
@@ -1,5 +1,14 @@
 require "bundler/gem_tasks"
+require "rake/extensiontask"
 
 task :scream do
     puts "みくは自分を曲げないよ！"
 end
+
+task :build => :compile
+
+Rake::ExtensionTask.new('decoders/native') do |ext|
+    ext.lib_dir = 'lib/mikunyan'
+end
+
+task :default => [:clobber, :compile, :spec]
diff --git a/ext/decoders/native/astc.c b/ext/decoders/native/astc.c
new file mode 100644
index 0000000..5920ddf
--- /dev/null
+++ b/ext/decoders/native/astc.c
@@ -0,0 +1,765 @@
+#include <assert.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+#include "astc.h"
+
+static inline uint32_t color(uint8_t r, uint8_t g, uint8_t b, uint8_t a) {
+    return r | g << 8 | b << 16 | a << 24;
+}
+
+static inline int getbits(const uint8_t *buf, const int bit, const int len) {
+    return (*(int*)(buf + bit / 8) >> (bit % 8)) & ((1 << len) - 1);
+}
+
+static inline uint64_t getbits64(const uint8_t *buf, const int bit, const int len) {
+    uint64_t mask = len == 64 ? -1 : (1ull << len) - 1;
+    if (len < 1)
+        return 0;
+    else if (bit >= 64)
+        return ((uint64_t*)buf)[1] >> (bit - 64) & mask;
+    else if (bit <= 0)
+        return ((uint64_t*)buf)[0] << -bit & mask;
+    else if (bit + len <= 64)
+        return (*(uint64_t*)buf) >> bit & mask;
+    else
+        return ((*(uint64_t*)buf) >> bit | ((uint64_t*)buf)[1] << (64 - bit)) & mask;
+}
+
+static inline uint8_t clamp(const int n) {
+    return n < 0 ? 0 : n > 255 ? 255 : n;
+}
+
+static inline void bit_transfer_signed(int *a, int *b) {
+    *b = (*b >> 1) | (*a & 0x80);
+    *a = (*a >> 1) & 0x3f;
+    if (*a & 0x20)
+        *a -= 0x40;
+}
+
+static inline void set_endpoint(uint8_t endpoint[8], uint8_t r1, uint8_t g1, uint8_t b1, uint8_t a1, uint8_t r2, uint8_t g2, uint8_t b2, uint8_t a2) {
+    endpoint[0] = r1;
+    endpoint[1] = g1;
+    endpoint[2] = b1;
+    endpoint[3] = a1;
+    endpoint[4] = r2;
+    endpoint[5] = g2;
+    endpoint[6] = b2;
+    endpoint[7] = a2;
+}
+
+static inline void set_endpoint_clamp(uint8_t endpoint[8], int r1, int g1, int b1, int a1, int r2, int g2, int b2, int a2) {
+    endpoint[0] = clamp(r1);
+    endpoint[1] = clamp(g1);
+    endpoint[2] = clamp(b1);
+    endpoint[3] = clamp(a1);
+    endpoint[4] = clamp(r2);
+    endpoint[5] = clamp(g2);
+    endpoint[6] = clamp(b2);
+    endpoint[7] = clamp(a2);
+}
+
+static inline void set_endpoint_blue(uint8_t endpoint[8], int r1, int g1, int b1, int a1, int r2, int g2, int b2, int a2) {
+    endpoint[0] = (r1 + b1) >> 1;
+    endpoint[1] = (g1 + b1) >> 1;
+    endpoint[2] = b1;
+    endpoint[3] = a1;
+    endpoint[4] = (r2 + b2) >> 1;
+    endpoint[5] = (g2 + b2) >> 1;
+    endpoint[6] = b2;
+    endpoint[7] = a2;
+}
+
+static inline void set_endpoint_blue_clamp(uint8_t endpoint[8], int r1, int g1, int b1, int a1, int r2, int g2, int b2, int a2) {
+    endpoint[0] = clamp((r1 + b1) >> 1);
+    endpoint[1] = clamp((g1 + b1) >> 1);
+    endpoint[2] = clamp(b1);
+    endpoint[3] = clamp(a1);
+    endpoint[4] = clamp((r2 + b2) >> 1);
+    endpoint[5] = clamp((g2 + b2) >> 1);
+    endpoint[6] = clamp(b2);
+    endpoint[7] = clamp(a2);
+}
+
+static const uint8_t BitReverseTable[] = {
+    0x00, 0x80, 0x40, 0xC0, 0x20, 0xA0, 0x60, 0xE0, 0x10, 0x90, 0x50, 0xD0, 0x30, 0xB0, 0x70, 0xF0,
+    0x08, 0x88, 0x48, 0xC8, 0x28, 0xA8, 0x68, 0xE8, 0x18, 0x98, 0x58, 0xD8, 0x38, 0xB8, 0x78, 0xF8,
+    0x04, 0x84, 0x44, 0xC4, 0x24, 0xA4, 0x64, 0xE4, 0x14, 0x94, 0x54, 0xD4, 0x34, 0xB4, 0x74, 0xF4,
+    0x0C, 0x8C, 0x4C, 0xCC, 0x2C, 0xAC, 0x6C, 0xEC, 0x1C, 0x9C, 0x5C, 0xDC, 0x3C, 0xBC, 0x7C, 0xFC,
+    0x02, 0x82, 0x42, 0xC2, 0x22, 0xA2, 0x62, 0xE2, 0x12, 0x92, 0x52, 0xD2, 0x32, 0xB2, 0x72, 0xF2,
+    0x0A, 0x8A, 0x4A, 0xCA, 0x2A, 0xAA, 0x6A, 0xEA, 0x1A, 0x9A, 0x5A, 0xDA, 0x3A, 0xBA, 0x7A, 0xFA,
+    0x06, 0x86, 0x46, 0xC6, 0x26, 0xA6, 0x66, 0xE6, 0x16, 0x96, 0x56, 0xD6, 0x36, 0xB6, 0x76, 0xF6,
+    0x0E, 0x8E, 0x4E, 0xCE, 0x2E, 0xAE, 0x6E, 0xEE, 0x1E, 0x9E, 0x5E, 0xDE, 0x3E, 0xBE, 0x7E, 0xFE,
+    0x01, 0x81, 0x41, 0xC1, 0x21, 0xA1, 0x61, 0xE1, 0x11, 0x91, 0x51, 0xD1, 0x31, 0xB1, 0x71, 0xF1,
+    0x09, 0x89, 0x49, 0xC9, 0x29, 0xA9, 0x69, 0xE9, 0x19, 0x99, 0x59, 0xD9, 0x39, 0xB9, 0x79, 0xF9,
+    0x05, 0x85, 0x45, 0xC5, 0x25, 0xA5, 0x65, 0xE5, 0x15, 0x95, 0x55, 0xD5, 0x35, 0xB5, 0x75, 0xF5,
+    0x0D, 0x8D, 0x4D, 0xCD, 0x2D, 0xAD, 0x6D, 0xED, 0x1D, 0x9D, 0x5D, 0xDD, 0x3D, 0xBD, 0x7D, 0xFD,
+    0x03, 0x83, 0x43, 0xC3, 0x23, 0xA3, 0x63, 0xE3, 0x13, 0x93, 0x53, 0xD3, 0x33, 0xB3, 0x73, 0xF3,
+    0x0B, 0x8B, 0x4B, 0xCB, 0x2B, 0xAB, 0x6B, 0xEB, 0x1B, 0x9B, 0x5B, 0xDB, 0x3B, 0xBB, 0x7B, 0xFB,
+    0x07, 0x87, 0x47, 0xC7, 0x27, 0xA7, 0x67, 0xE7, 0x17, 0x97, 0x57, 0xD7, 0x37, 0xB7, 0x77, 0xF7,
+    0x0F, 0x8F, 0x4F, 0xCF, 0x2F, 0xAF, 0x6F, 0xEF, 0x1F, 0x9F, 0x5F, 0xDF, 0x3F, 0xBF, 0x7F, 0xFF
+};
+
+static inline uint8_t bit_reverse_u8(const uint8_t c, const int bits) {
+    assert(bits <= 8);
+    return BitReverseTable[c] >> (8 - bits);
+}
+
+static inline uint64_t bit_reverse_u64(const uint64_t d, const int bits) {
+    assert(bits <= 64);
+
+    uint64_t ret =
+        (uint64_t)BitReverseTable[d & 0xff] << 56 |
+        (uint64_t)BitReverseTable[d >> 8 & 0xff] << 48 |
+        (uint64_t)BitReverseTable[d >> 16 & 0xff] << 40 |
+        (uint64_t)BitReverseTable[d >> 24 & 0xff] << 32 |
+        (uint64_t)BitReverseTable[d >> 32 & 0xff] << 24 |
+        (uint64_t)BitReverseTable[d >> 40 & 0xff] << 16 |
+        (uint64_t)BitReverseTable[d >> 48 & 0xff] << 8 | BitReverseTable[d >> 56 & 0xff];
+    return ret >> (64 - bits);
+}
+
+static int WeightPrecTableA[] = {0, 0, 0, 3, 0, 5, 3, 0, 0, 0, 5, 3, 0, 5, 3, 0};
+static int WeightPrecTableB[] = {0, 0, 1, 0, 2, 0, 1, 3, 0, 0, 1, 2, 4, 2, 3, 5};
+
+static int CemTableA[] = {0, 3, 5, 0, 3, 5, 0, 3, 5, 0, 3, 5, 0, 3, 5, 0, 3, 0, 0};
+static int CemTableB[] = {8, 6, 5, 7, 5, 4, 6, 4, 3, 5, 3, 2, 4, 2, 1, 3, 1, 2, 1};
+
+typedef struct {
+    int bw;
+    int bh;
+    int width;
+    int height;
+    int part_num;
+    int dual_plane;
+    int plane_selector;
+    int weight_range;
+    int weight_num; // max: 120
+    int cem[4];
+    int cem_range;
+    int endpoint_value_num; // max: 32
+    uint8_t endpoints[4][8];
+    uint8_t weights[144][2];
+    uint8_t partition[144];
+} BlockData;
+
+typedef struct {
+    int bits;
+    int nonbits;
+} IntSeqData;
+
+void decode_intseq(const uint8_t *buf, int offset, const int a, const int b, const int count, const int reverse, IntSeqData *out) {
+    static int mt[] = { 0, 2, 4, 5, 7 };
+    static int mq[] = { 0, 3, 5 };
+    static int TritsTable[5][256] = {
+        {0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 1, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 1, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2},
+        {0, 0, 0, 0, 1, 1, 1, 0, 2, 2, 2, 0, 2, 2, 2, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 2, 2, 1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 2, 2, 2, 0, 2, 2, 2, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 2, 2, 1, 1, 1, 1, 0, 0, 0, 0, 0, 1, 1, 1, 0, 2, 2, 2, 0, 2, 2, 2, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 2, 2, 1, 2, 2, 2, 0, 0, 0, 0, 0, 1, 1, 1, 0, 2, 2, 2, 0, 2, 2, 2, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 2, 2, 1, 2, 2, 2, 0, 0, 0, 0, 0, 1, 1, 1, 0, 2, 2, 2, 0, 2, 2, 2, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 2, 2, 1, 0, 0, 0, 1, 0, 0, 0, 0, 1, 1, 1, 0, 2, 2, 2, 0, 2, 2, 2, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 2, 2, 1, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 0, 2, 2, 2, 0, 2, 2, 2, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 2, 2, 1, 2, 2, 2, 1, 0, 0, 0, 0, 1, 1, 1, 0, 2, 2, 2, 0, 2, 2, 2, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 2, 2, 1, 2, 2, 2, 1},
+        {0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 2, 2, 2, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 2, 2, 2, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 2, 2, 2, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 2, 2, 2, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 2, 2, 2, 2, 0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 2, 2, 2, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 2, 2, 2, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 2, 2, 2, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 2, 2, 2, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 2, 2, 2, 2},
+        {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2},
+        {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2}
+    };
+    static int QuintsTable[3][128] = {
+        {0, 1, 2, 3, 4, 0, 4, 4, 0, 1, 2, 3, 4, 1, 4, 4, 0, 1, 2, 3, 4, 2, 4, 4, 0, 1, 2, 3, 4, 3, 4, 4, 0, 1, 2, 3, 4, 0, 4, 0, 0, 1, 2, 3, 4, 1, 4, 1, 0, 1, 2, 3, 4, 2, 4, 2, 0, 1, 2, 3, 4, 3, 4, 3, 0, 1, 2, 3, 4, 0, 2, 3, 0, 1, 2, 3, 4, 1, 2, 3, 0, 1, 2, 3, 4, 2, 2, 3, 0, 1, 2, 3, 4, 3, 2, 3, 0, 1, 2, 3, 4, 0, 0, 1, 0, 1, 2, 3, 4, 1, 0, 1, 0, 1, 2, 3, 4, 2, 0, 1, 0, 1, 2, 3, 4, 3, 0, 1},
+        {0, 0, 0, 0, 0, 4, 4, 4, 1, 1, 1, 1, 1, 4, 4, 4, 2, 2, 2, 2, 2, 4, 4, 4, 3, 3, 3, 3, 3, 4, 4, 4, 0, 0, 0, 0, 0, 4, 0, 4, 1, 1, 1, 1, 1, 4, 1, 4, 2, 2, 2, 2, 2, 4, 2, 4, 3, 3, 3, 3, 3, 4, 3, 4, 0, 0, 0, 0, 0, 4, 0, 0, 1, 1, 1, 1, 1, 4, 1, 1, 2, 2, 2, 2, 2, 4, 2, 2, 3, 3, 3, 3, 3, 4, 3, 3, 0, 0, 0, 0, 0, 4, 0, 0, 1, 1, 1, 1, 1, 4, 1, 1, 2, 2, 2, 2, 2, 4, 2, 2, 3, 3, 3, 3, 3, 4, 3, 3},
+        {0, 0, 0, 0, 0, 0, 0, 4, 0, 0, 0, 0, 0, 0, 1, 4, 0, 0, 0, 0, 0, 0, 2, 4, 0, 0, 0, 0, 0, 0, 3, 4, 1, 1, 1, 1, 1, 1, 4, 4, 1, 1, 1, 1, 1, 1, 4, 4, 1, 1, 1, 1, 1, 1, 4, 4, 1, 1, 1, 1, 1, 1, 4, 4, 2, 2, 2, 2, 2, 2, 4, 4, 2, 2, 2, 2, 2, 2, 4, 4, 2, 2, 2, 2, 2, 2, 4, 4, 2, 2, 2, 2, 2, 2, 4, 4, 3, 3, 3, 3, 3, 3, 4, 4, 3, 3, 3, 3, 3, 3, 4, 4, 3, 3, 3, 3, 3, 3, 4, 4, 3, 3, 3, 3, 3, 3, 4, 4}
+    };
+
+    if (count <= 0)
+        return;
+
+    int n = 0;
+
+    if (a == 3) {
+        int mask = (1 << b) - 1;
+        int block_count = (count + 4) / 5;
+        int last_block_count = (count + 4) % 5 + 1;
+        int block_size = 8 + 5 * b;
+        int last_block_size = (block_size * last_block_count + 4) / 5;
+
+        if (reverse) {
+            for (int i = 0, p = offset; i < block_count; i++, p -= block_size) {
+                int now_size = (i < block_count - 1) ? block_size : last_block_size;
+                uint64_t d = bit_reverse_u64(getbits64(buf, p - now_size, now_size), now_size);
+                int x = (d >> b & 3) | (d >> b * 2 & 0xc) | (d >> b * 3 & 0x10) | (d >> b * 4 & 0x60) | (d >> b * 5 & 0x80);
+                for (int j = 0; j < 5 && n < count; j++, n++)
+                    out[n] = (IntSeqData){ d >> (mt[j] + b * j) & mask, TritsTable[j][x] };
+            }
+        } else {
+            for (int i = 0, p = offset; i < block_count; i++, p += block_size) {
+                uint64_t d = getbits64(buf, p, (i < block_count - 1) ? block_size : last_block_size);
+                int x = (d >> b & 3) | (d >> b * 2 & 0xc) | (d >> b * 3 & 0x10) | (d >> b * 4 & 0x60) | (d >> b * 5 & 0x80);
+                for (int j = 0; j < 5 && n < count; j++, n++)
+                    out[n] = (IntSeqData){ d >> (mt[j] + b * j) & mask, TritsTable[j][x] };
+            }
+        }
+    } else if (a == 5) {
+        int mask = (1 << b) - 1;
+        int block_count = (count + 2) / 3;
+        int last_block_count = (count + 2) % 3 + 1;
+        int block_size = 7 + 3 * b;
+        int last_block_size = (block_size * last_block_count + 2) / 3;
+
+        if (reverse) {
+            for (int i = 0, p = offset; i < block_count; i++, p -= block_size) {
+                int now_size = (i < block_count - 1) ? block_size : last_block_size;
+                uint64_t d = bit_reverse_u64(getbits64(buf, p - now_size, now_size), now_size);
+                int x = (d >> b & 7) | (d >> b * 2 & 0x18) | (d >> b * 3 & 0x60);
+                for (int j = 0; j < 3 && n < count; j++, n++)
+                    out[n] = (IntSeqData){ d >> (mq[j] + b * j) & mask, QuintsTable[j][x] };
+            }
+        } else {
+            for (int i = 0, p = offset; i < block_count; i++, p += block_size) {
+                uint64_t d = getbits64(buf, p, (i < block_count - 1) ? block_size : last_block_size);
+                int x = (d >> b & 7) | (d >> b * 2 & 0x18) | (d >> b * 3 & 0x60);
+                for (int j = 0; j < 3 && n < count; j++, n++)
+                    out[n] = (IntSeqData){ d >> (mq[j] + b * j) & mask, QuintsTable[j][x] };
+            }
+        }
+    } else {
+        if (reverse)
+            for (int p = offset - b; n < count; n++, p -= b)
+                out[n] = (IntSeqData){ bit_reverse_u8(getbits(buf, p, b), b), 0 };
+        else
+            for (int p = offset; n < count; n++, p += b)
+                out[n] = (IntSeqData){ getbits(buf, p, b), 0 };
+    }
+}
+
+void decode_block_params(const uint8_t *buf, BlockData *block_data) {
+    block_data->dual_plane = (buf[1] & 4) >> 2;
+    block_data->weight_range = (buf[0] >> 4 & 1) | (buf[1] << 2 & 8);
+
+    if (buf[0] & 3) {
+        block_data->weight_range |= buf[0] << 1 & 6;
+        switch (buf[0] & 0xc) {
+            case 0:
+                block_data->width = (*(uint16_t*)buf >> 7 & 3) + 4;
+                block_data->height = (buf[0] >> 5 & 3) + 2;
+                break;
+            case 4:
+                block_data->width = (*(uint16_t*)buf >> 7 & 3) + 8;
+                block_data->height = (buf[0] >> 5 & 3) + 2;
+                break;
+            case 8:
+                block_data->width = (buf[0] >> 5 & 3) + 2;
+                block_data->height = (*(uint16_t*)buf >> 7 & 3) + 8;
+                break;
+            case 12:
+                if (buf[1] & 1) {
+                    block_data->width = (buf[0] >> 7 & 1) + 2;
+                    block_data->height = (buf[0] >> 5 & 3) + 2;
+                } else {
+                    block_data->width = (buf[0] >> 5 & 3) + 2;
+                    block_data->height = (buf[0] >> 7 & 1) + 6;
+                }
+                break;
+        }
+    } else {
+        block_data->weight_range |= buf[0] >> 1 & 6;
+        switch ((*(int*)buf) & 0x180) {
+            case 0:
+                block_data->width = 12;
+                block_data->height = (buf[0] >> 5 & 3) + 2;
+                break;
+            case 0x80:
+                block_data->width = (buf[0] >> 5 & 3) + 2;
+                block_data->height = 12;
+                break;
+            case 0x100:
+                block_data->width = (buf[0] >> 5 & 3) + 6;
+                block_data->height = (buf[1] >> 1 & 3) + 6;
+                block_data->dual_plane = 0;
+                block_data->weight_range &= 7;
+                break;
+            case 0x180:
+                block_data->width = (buf[0] & 0x20) ? 10 : 6;
+                block_data->height = (buf[0] & 0x20) ? 6 : 10;
+                break;
+        }
+    }
+
+    block_data->part_num = (buf[1] >> 3 & 3) + 1;
+
+    block_data->weight_num = block_data->width * block_data->height;
+    if (block_data->dual_plane)
+        block_data->weight_num *= 2;
+
+    int weight_bits, config_bits, cem_base = 0;
+
+    switch (WeightPrecTableA[block_data->weight_range]) {
+        case 3:
+            weight_bits = block_data->weight_num * WeightPrecTableB[block_data->weight_range] + (block_data->weight_num * 8 + 4) / 5;
+            break;
+        case 5:
+            weight_bits = block_data->weight_num * WeightPrecTableB[block_data->weight_range] + (block_data->weight_num * 7 + 2) / 3;
+            break;
+        default:
+            weight_bits = block_data->weight_num * WeightPrecTableB[block_data->weight_range];
+    }
+
+    if (block_data->part_num == 1) {
+        block_data->cem[0] = *(int*)(buf + 1) >> 5 & 0xf;
+        config_bits = 17;
+    } else {
+        cem_base = *(int*)(buf + 2) >> 7 & 3;
+        if (cem_base == 0) {
+            int cem = buf[3] >> 1 & 0xf;
+            for (int i = 0; i < block_data->part_num; i++)
+                block_data->cem[i] = cem;
+            config_bits = 29;
+        } else {
+            for (int i = 0; i < block_data->part_num; i++)
+                block_data->cem[i] = ((buf[3] >> (i + 1) & 1) + cem_base - 1) << 2;
+            switch (block_data->part_num) {
+                case 2:
+                    block_data->cem[0] |= buf[3] >> 3 & 3;
+                    block_data->cem[1] |= getbits(buf, 126 - weight_bits, 2);
+                    break;
+                case 3:
+                    block_data->cem[0] |= buf[3] >> 4 & 1;
+                    block_data->cem[0] |= getbits(buf, 122 - weight_bits, 2) & 2;
+                    block_data->cem[1] |= getbits(buf, 124 - weight_bits, 2);
+                    block_data->cem[2] |= getbits(buf, 126 - weight_bits, 2);
+                    break;
+                case 4:
+                    for (int i = 0; i < 4; i++)
+                        block_data->cem[i] |= getbits(buf, 120 + i * 2 - weight_bits, 2);
+                    break;
+                default:
+                    assert(0);
+            }
+            config_bits = 25 + block_data->part_num * 3;
+        }
+    }
+
+    if (block_data->dual_plane) {
+        config_bits += 2;
+        block_data->plane_selector = getbits(buf, cem_base ? 130 - weight_bits - block_data->part_num * 3 : 126 - weight_bits, 2);
+    }
+
+    int remain_bits = 128 - config_bits - weight_bits;
+
+    block_data->endpoint_value_num = 0;
+    for (int i = 0; i < block_data->part_num; i++)
+        block_data->endpoint_value_num += (block_data->cem[i] >> 1 & 6) + 2;
+
+    for (int i = 0, endpoint_bits; i < (int)(sizeof(CemTableA) / sizeof(int)); i++) {
+        switch (CemTableA[i]) {
+            case 3:
+                endpoint_bits = block_data->endpoint_value_num * CemTableB[i] + (block_data->endpoint_value_num * 8 + 4) / 5;
+                break;
+            case 5:
+                endpoint_bits = block_data->endpoint_value_num * CemTableB[i] + (block_data->endpoint_value_num * 7 + 2) / 3;
+                break;
+            default:
+                endpoint_bits = block_data->endpoint_value_num * CemTableB[i];
+        }
+
+        if (endpoint_bits <= remain_bits) {
+            block_data->cem_range = i;
+            break;
+        }
+    }
+}
+
+void decode_endpoints(const uint8_t *buf, BlockData *data) {
+    static int TritsTable[] = { 0, 204, 93, 44, 22, 11, 5 };
+    static int QuintsTable[] = { 0, 113, 54, 26, 13, 6 };
+    IntSeqData seq[32];
+    int ev[32];
+    decode_intseq(buf, data->part_num == 1 ? 17 : 29, CemTableA[data->cem_range], CemTableB[data->cem_range], data->endpoint_value_num, 0, seq);
+
+    switch (CemTableA[data->cem_range]) {
+        case 3:
+            for (int i = 0, b, c = TritsTable[CemTableB[data->cem_range]]; i < data->endpoint_value_num; i++) {
+                int a = (seq[i].bits & 1) * 0x1ff;
+                int x = seq[i].bits >> 1;
+                switch (CemTableB[data->cem_range]) {
+                    case 1:
+                        b = 0;
+                        break;
+                    case 2:
+                        b = 0b100010110 * x;
+                        break;
+                    case 3:
+                        b = x << 7 | x << 2 | x;
+                        break;
+                    case 4:
+                        b = x << 6 | x;
+                        break;
+                    case 5:
+                        b = x << 5 | x >> 2;
+                        break;
+                    case 6:
+                        b = x << 4 | x >> 4;
+                        break;
+                }
+                ev[i] = (a & 0x80) | ((seq[i].nonbits * c + b) ^ a) >> 2;
+            }
+            break;
+        case 5:
+            for (int i = 0, b, c = QuintsTable[CemTableB[data->cem_range]]; i < data->endpoint_value_num; i++) {
+                int a = (seq[i].bits & 1) * 0x1ff;
+                int x = seq[i].bits >> 1;
+                switch (CemTableB[data->cem_range]) {
+                    case 1:
+                        b = 0;
+                        break;
+                    case 2:
+                        b = 0b100001100 * x;
+                        break;
+                    case 3:
+                        b = x << 7 | x << 1 | x >> 1;
+                        break;
+                    case 4:
+                        b = x << 6 | x >> 1;
+                        break;
+                    case 5:
+                        b = x << 5 | x >> 3;
+                        break;
+                }
+                ev[i] = (a & 0x80) | ((seq[i].nonbits * c + b) ^ a) >> 2;
+            }
+            break;
+        default:
+            switch (CemTableB[data->cem_range]) {
+                case 1:
+                    for (int i = 0; i < data->endpoint_value_num; i++)
+                        ev[i] = seq[i].bits * 0xff;
+                    break;
+                case 2:
+                    for (int i = 0; i < data->endpoint_value_num; i++)
+                        ev[i] = seq[i].bits * 0x55;
+                    break;
+                case 3:
+                    for (int i = 0; i < data->endpoint_value_num; i++)
+                        ev[i] = seq[i].bits << 5 | seq[i].bits << 2 | seq[i].bits >> 1;
+                    break;
+                case 4:
+                    for (int i = 0; i < data->endpoint_value_num; i++)
+                        ev[i] = seq[i].bits << 4 | seq[i].bits;
+                    break;
+                case 5:
+                    for (int i = 0; i < data->endpoint_value_num; i++)
+                        ev[i] = seq[i].bits << 3 | seq[i].bits >> 2;
+                    break;
+                case 6:
+                    for (int i = 0; i < data->endpoint_value_num; i++)
+                        ev[i] = seq[i].bits << 2 | seq[i].bits >> 4;
+                    break;
+                case 7:
+                    for (int i = 0; i < data->endpoint_value_num; i++)
+                        ev[i] = seq[i].bits << 1 | seq[i].bits >> 6;
+                    break;
+                case 8:
+                    for (int i = 0; i < data->endpoint_value_num; i++)
+                        ev[i] = seq[i].bits;
+                    break;
+            }
+    }
+
+    int *v = ev;
+    for (int cem = 0; cem < data->part_num; v += (data->cem[cem] / 4 + 1) * 2, cem++) {
+        switch (data->cem[cem]) {
+            case 0:
+                set_endpoint(data->endpoints[cem], v[0], v[0], v[0], 255, v[1], v[1], v[1], 255);
+                break;
+            case 1:
+                {
+                    uint8_t l0 = (v[0] >> 2) | (v[1] & 0xc0);
+                    uint8_t l1 = clamp(l0 + (v[1] & 0x3f));
+                    set_endpoint(data->endpoints[cem], l0, l0, l0, 255, l1, l1, l1, 255);
+                }
+                break;
+            case 4:
+                set_endpoint(data->endpoints[cem], v[0], v[0], v[0], v[2], v[1], v[1], v[1], v[3]);
+                break;
+            case 5:
+                bit_transfer_signed(&v[1], &v[0]);
+                bit_transfer_signed(&v[3], &v[2]);
+                v[1] += v[0];
+                set_endpoint_clamp(data->endpoints[cem], v[0], v[0], v[0], v[2], v[1], v[1], v[1], v[2] + v[3]);
+                break;
+            case 6:
+                set_endpoint(data->endpoints[cem], v[0] * v[3] >> 8, v[1] * v[3] >> 8, v[2] * v[3] >> 8, 255, v[0], v[1], v[2], 255);
+                break;
+            case 8:
+                if (v[0] + v[2] + v[4] <= v[1] + v[3] + v[5])
+                    set_endpoint(data->endpoints[cem], v[0], v[2], v[4], 255, v[1], v[3], v[5], 255);
+                else
+                    set_endpoint_blue(data->endpoints[cem], v[1], v[3], v[5], 255, v[0], v[2], v[4], 255);
+                break;
+            case 9:
+                bit_transfer_signed(&v[1], &v[0]);
+                bit_transfer_signed(&v[3], &v[2]);
+                bit_transfer_signed(&v[5], &v[4]);
+                if (v[1] + v[3] + v[5] >= 0)
+                    set_endpoint_clamp(data->endpoints[cem], v[0], v[2], v[4], 255, v[0] + v[1], v[2] + v[3], v[4] + v[5], 255);
+                else
+                    set_endpoint_blue_clamp(data->endpoints[cem], v[0] + v[1], v[2] + v[3], v[4] + v[5], 255, v[0], v[2], v[4], 255);
+                break;
+            case 10:
+                set_endpoint(data->endpoints[cem], v[0] * v[3] >> 8, v[1] * v[3] >> 8, v[2] * v[3] >> 8, v[4], v[0], v[1], v[2], v[5]);
+                break;
+            case 12:
+                if (v[0] + v[2] + v[4] <= v[1] + v[3] + v[5])
+                    set_endpoint(data->endpoints[cem], v[0], v[2], v[4], v[6], v[1], v[3], v[5], v[7]);
+                else
+                    set_endpoint_blue(data->endpoints[cem], v[1], v[3], v[5], v[7], v[0], v[2], v[4], v[6]);
+                break;
+            case 13:
+                bit_transfer_signed(&v[1], &v[0]);
+                bit_transfer_signed(&v[3], &v[2]);
+                bit_transfer_signed(&v[5], &v[4]);
+                bit_transfer_signed(&v[7], &v[6]);
+                if (v[1] + v[3] + v[5] >= 0)
+                    set_endpoint_clamp(data->endpoints[cem], v[0], v[2], v[4], v[6], v[0] + v[1], v[2] + v[3], v[4] + v[5], v[6] + v[7]);
+                else
+                    set_endpoint_blue_clamp(data->endpoints[cem], v[0] + v[1], v[2] + v[3], v[4] + v[5], v[6] + v[7], v[0], v[2], v[4], v[6]);
+                break;
+            default:
+                assert(0);
+        }
+    }
+}
+
+void decode_weights(const uint8_t *buf, BlockData *data) {
+    IntSeqData seq[128];
+    int wv[128] = {};
+    decode_intseq(buf, 128, WeightPrecTableA[data->weight_range], WeightPrecTableB[data->weight_range], data->weight_num, 1, seq);
+
+    if (WeightPrecTableA[data->weight_range] == 0) {
+        switch (WeightPrecTableB[data->weight_range]) {
+            case 1:
+                for (int i = 0; i < data->weight_num; i++)
+                    wv[i] = seq[i].bits ? 63 : 0;
+                break;
+            case 2:
+                for (int i = 0; i < data->weight_num; i++)
+                    wv[i] = seq[i].bits << 4 | seq[i].bits << 2 | seq[i].bits;
+                break;
+            case 3:
+                for (int i = 0; i < data->weight_num; i++)
+                    wv[i] = seq[i].bits << 3 | seq[i].bits;
+                break;
+            case 4:
+                for (int i = 0; i < data->weight_num; i++)
+                    wv[i] = seq[i].bits << 2 | seq[i].bits >> 2;
+                break;
+            case 5:
+                for (int i = 0; i < data->weight_num; i++)
+                    wv[i] = seq[i].bits << 1 | seq[i].bits >> 4;
+                break;
+        }
+        for (int i = 0; i < data->weight_num; i++)
+            if (wv[i] > 32)
+                ++wv[i];
+    } else if (WeightPrecTableB[data->weight_range] == 0) {
+        int s = WeightPrecTableA[data->weight_range] == 3 ? 32 : 16;
+        for (int i = 0; i < data->weight_num; i++)
+            wv[i] = seq[i].nonbits * s;
+    } else {
+        if (WeightPrecTableA[data->weight_range] == 3) {
+            switch (WeightPrecTableB[data->weight_range]) {
+                case 1:
+                    for (int i = 0; i < data->weight_num; i++)
+                        wv[i] = seq[i].nonbits * 50;
+                    break;
+                case 2:
+                    for (int i = 0; i < data->weight_num; i++) {
+                        wv[i] = seq[i].nonbits * 23;
+                        if (seq[i].bits & 2)
+                            wv[i] += 0b1000101;
+                    }
+                    break;
+                case 3:
+                    for (int i = 0; i < data->weight_num; i++)
+                        wv[i] = seq[i].nonbits * 11 + ((seq[i].bits << 4 | seq[i].bits >> 1) & 0b1100011);
+                    break;
+            }
+        } else if (WeightPrecTableA[data->weight_range] == 5) {
+            switch (WeightPrecTableB[data->weight_range]) {
+                case 1:
+                    for (int i = 0; i < data->weight_num; i++)
+                        wv[i] = seq[i].nonbits * 28;
+                    break;
+                case 2:
+                    for (int i = 0; i < data->weight_num; i++) {
+                        wv[i] = seq[i].nonbits * 13;
+                        if (seq[i].bits & 2)
+                            wv[i] += 0b1000010;
+                    }
+                    break;
+            }
+        }
+        for (int i = 0; i < data->weight_num; i++) {
+            int a = (seq[i].bits & 1) * 0x7f;
+            wv[i] = (a & 0x20) | ((wv[i] ^ a) >> 2);
+            if (wv[i] > 32)
+                ++wv[i];
+        }
+    }
+
+    int ds = (1024 + data->bw / 2) / (data->bw - 1);
+    int dt = (1024 + data->bh / 2) / (data->bh - 1);
+    int pn = data->dual_plane ? 2 : 1;
+
+    for (int t = 0, i = 0; t < data->bh; t++) {
+        for (int s = 0; s < data->bw; s++, i++) {
+            int gs = (ds * s * (data->width - 1) + 32) >> 6;
+            int gt = (dt * t * (data->height - 1) + 32) >> 6;
+            int fs = gs & 0xf;
+            int ft = gt & 0xf;
+            int v = (gs >> 4) + (gt >> 4) * data->width;
+            int w11 = (fs * ft + 8) >> 4;
+            int w10 = ft - w11;
+            int w01 = fs - w11;
+            int w00 = 16 - fs - ft + w11;
+
+            for (int p = 0; p < pn; p++) {
+                int p00 = wv[v * pn + p];
+                int p01 = wv[(v + 1) * pn + p];
+                int p10 = wv[(v + data->width) * pn + p];
+                int p11 = wv[(v + data->width + 1) * pn + p];
+                data->weights[i][p] = (p00 * w00 + p01 * w01 + p10 * w10 + p11 * w11 + 8) >> 4;
+            }
+        }
+    }
+}
+
+void select_partition(const uint8_t *buf, BlockData *data) {
+    int small_block = data->bw * data->bh < 31;
+    int seed = (*(int*)buf >> 13 & 0x3ff) | (data->part_num - 1) << 10;
+
+    uint32_t rnum = seed;
+    rnum ^= rnum >> 15;
+    rnum -= rnum << 17;
+    rnum += rnum << 7;
+    rnum += rnum << 4;
+    rnum ^= rnum >> 5;
+    rnum += rnum << 16;
+    rnum ^= rnum >> 7;
+    rnum ^= rnum >> 3;
+    rnum ^= rnum << 6;
+    rnum ^= rnum >> 17;
+
+    uint8_t seeds[8];
+    for (int i = 0; i < 8; i++) {
+        seeds[i] = (rnum >> (i * 4)) & 0xF;
+        seeds[i] *= seeds[i];
+    }
+
+    int sh[2] = { seed & 2 ? 4 : 5, data->part_num == 3 ? 6 : 5 };
+
+    if (seed & 1)
+        for (int i = 0; i < 8; i++)
+            seeds[i] >>= sh[i % 2];
+    else
+        for (int i = 0; i < 8; i++)
+            seeds[i] >>= sh[1 - i % 2];
+
+    if (small_block) {
+        for (int t = 0, i = 0; t < data->bh; t++) {
+            for (int s = 0; s < data->bw; s++, i++) {
+                int x = s << 1;
+                int y = t << 1;
+                int a = (seeds[0] * x + seeds[1] * y + (rnum >> 14)) & 0x3f;
+                int b = (seeds[2] * x + seeds[3] * y + (rnum >> 10)) & 0x3f;
+                int c = data->part_num < 3 ? 0 : (seeds[4] * x + seeds[5] * y + (rnum >> 6)) & 0x3f;
+                int d = data->part_num < 4 ? 0 : (seeds[6] * x + seeds[7] * y + (rnum >> 2)) & 0x3f;
+                data->partition[i] = (a >= b && a >= c && a >= d) ? 0 : (b >= c && b >= d) ? 1 : (c >= d) ? 2 : 3;
+            }
+        }
+    } else {
+        for (int y = 0, i = 0; y < data->bh; y++) {
+            for (int x = 0; x < data->bw; x++, i++) {
+                int a = (seeds[0] * x + seeds[1] * y + (rnum >> 14)) & 0x3f;
+                int b = (seeds[2] * x + seeds[3] * y + (rnum >> 10)) & 0x3f;
+                int c = data->part_num < 3 ? 0 : (seeds[4] * x + seeds[5] * y + (rnum >> 6)) & 0x3f;
+                int d = data->part_num < 4 ? 0 : (seeds[6] * x + seeds[7] * y + (rnum >> 2)) & 0x3f;
+                data->partition[i] = (a >= b && a >= c && a >= d) ? 0 : (b >= c && b >= d) ? 1 : (c >= d) ? 2 : 3;
+            }
+        }
+    }
+}
+
+static inline uint8_t select_color(int v0, int v1, int weight) {
+    return ((((v0 << 8 | v0) * (64 - weight) + (v1 << 8 | v1) * weight + 32) >> 6) * 255 + 32768) / 65536;
+}
+
+void applicate_color(const BlockData *data, uint32_t *outbuf) {
+    if (data->dual_plane) {
+        int ps[] = { 0, 0, 0, 0 };
+        ps[data->plane_selector] = 1;
+        if (data->part_num > 1) {
+            for (int i = 0; i < data->bw * data->bh; i++) {
+                int p = data->partition[i];
+                uint8_t r = select_color(data->endpoints[p][0], data->endpoints[p][4], data->weights[i][ps[0]]);
+                uint8_t g = select_color(data->endpoints[p][1], data->endpoints[p][5], data->weights[i][ps[1]]);
+                uint8_t b = select_color(data->endpoints[p][2], data->endpoints[p][6], data->weights[i][ps[2]]);
+                uint8_t a = select_color(data->endpoints[p][3], data->endpoints[p][7], data->weights[i][ps[3]]);
+                outbuf[i] = color(r, g, b, a);
+            }
+        } else {
+            for (int i = 0; i < data->bw * data->bh; i++) {
+                uint8_t r = select_color(data->endpoints[0][0], data->endpoints[0][4], data->weights[i][ps[0]]);
+                uint8_t g = select_color(data->endpoints[0][1], data->endpoints[0][5], data->weights[i][ps[1]]);
+                uint8_t b = select_color(data->endpoints[0][2], data->endpoints[0][6], data->weights[i][ps[2]]);
+                uint8_t a = select_color(data->endpoints[0][3], data->endpoints[0][7], data->weights[i][ps[3]]);
+                outbuf[i] = color(r, g, b, a);
+            }
+        }
+    } else if (data->part_num > 1) {
+        for (int i = 0; i < data->bw * data->bh; i++) {
+            int p = data->partition[i];
+            uint8_t r = select_color(data->endpoints[p][0], data->endpoints[p][4], data->weights[i][0]);
+            uint8_t g = select_color(data->endpoints[p][1], data->endpoints[p][5], data->weights[i][0]);
+            uint8_t b = select_color(data->endpoints[p][2], data->endpoints[p][6], data->weights[i][0]);
+            uint8_t a = select_color(data->endpoints[p][3], data->endpoints[p][7], data->weights[i][0]);
+            outbuf[i] = color(r, g, b, a);
+        }
+    } else {
+        for (int i = 0; i < data->bw * data->bh; i++) {
+            uint8_t r = select_color(data->endpoints[0][0], data->endpoints[0][4], data->weights[i][0]);
+            uint8_t g = select_color(data->endpoints[0][1], data->endpoints[0][5], data->weights[i][0]);
+            uint8_t b = select_color(data->endpoints[0][2], data->endpoints[0][6], data->weights[i][0]);
+            uint8_t a = select_color(data->endpoints[0][3], data->endpoints[0][7], data->weights[i][0]);
+            outbuf[i] = color(r, g, b, a);
+        }
+    }
+}
+
+void decode_block(const uint8_t *buf, const int bw, const int bh, uint32_t *outbuf) {
+    if (buf[0] == 0xfc && (buf[1] & 1) == 1) {
+        uint32_t c = color(buf[9], buf[11], buf[13], buf[15]);
+        for (int i = 0; i < bw * bh; i++)
+            outbuf[i] = c;
+    } else {
+        BlockData block_data;
+        block_data.bw = bw;
+        block_data.bh = bh;
+        decode_block_params(buf, &block_data);
+        decode_endpoints(buf, &block_data);
+        decode_weights(buf, &block_data);
+        if (block_data.part_num > 1)
+            select_partition(buf, &block_data);
+        applicate_color(&block_data, outbuf);
+    }
+}
+
+void decode_astc(const uint8_t *data, const int w, const int h, const int bw, const int bh, uint32_t *image) {
+    int bcw = (w + bw - 1) / bw;
+    int bch = (h + bh - 1) / bh;
+    int clen_last = (w + bw - 1) % bw + 1;
+    uint32_t *buf = (uint32_t*)calloc(bw * bh, sizeof(uint32_t));
+    const uint8_t *ptr = data;
+    for (int t = 0; t < bch; t++) {
+        for (int s = 0; s < bcw; s++, ptr += 16) {
+            decode_block(ptr, bw, bh, buf);
+            int clen = (s < bcw - 1 ? bw : clen_last) * 4;
+            for (int i = 0, y = h - t * bh - 1; i < bh && y >= 0; i++, y--)
+                memcpy(image + y * w + s * bw, buf + i * bw, clen);
+        }
+    }
+    free(buf);
+}
diff --git a/ext/decoders/native/astc.h b/ext/decoders/native/astc.h
new file mode 100644
index 0000000..b5b2ee5
--- /dev/null
+++ b/ext/decoders/native/astc.h
@@ -0,0 +1,8 @@
+#ifndef ASTC_H
+#define ASTC_H
+
+#include <stdint.h>
+
+void decode_astc(const uint8_t*, const int, const int, const int, const int, uint32_t*);
+
+#endif /* end of include guard: ASTC_H */
diff --git a/ext/decoders/native/extconf.rb b/ext/decoders/native/extconf.rb
new file mode 100644
index 0000000..e93467e
--- /dev/null
+++ b/ext/decoders/native/extconf.rb
@@ -0,0 +1,7 @@
+require 'mkmf'
+
+append_cppflags('-O3')
+append_cppflags('-Wall')
+append_cppflags('-Wextra')
+append_cppflags('-Wvla')
+create_makefile('mikunyan/decoders/native')
diff --git a/ext/decoders/native/main.c b/ext/decoders/native/main.c
new file mode 100644
index 0000000..77f541f
--- /dev/null
+++ b/ext/decoders/native/main.c
@@ -0,0 +1,19 @@
+#include <stdlib.h>
+#include <stdint.h>
+#include <ruby.h>
+#include "astc.h"
+
+static VALUE rb_decode_astc(VALUE self, VALUE rb_data, VALUE w, VALUE h, VALUE bw, VALUE bh) {
+    const uint8_t *data = (uint8_t*)RSTRING_PTR(rb_data);
+    uint32_t *image = (uint32_t*)calloc(FIX2LONG(w) * FIX2LONG(h), sizeof(uint32_t));
+    decode_astc(data, FIX2INT(w), FIX2INT(h), FIX2INT(bw), FIX2INT(bh), image);
+    VALUE ret = rb_str_new((char*)image, FIX2LONG(w) * FIX2LONG(h) * sizeof(uint32_t));
+    free(image);
+    return ret;
+}
+
+void Init_native() {
+    VALUE mMikunyan = rb_define_module("Mikunyan");
+    VALUE mDecodeHelper = rb_define_module_under(mMikunyan, "DecodeHelper");
+	rb_define_module_function(mDecodeHelper, "decode_astc", rb_decode_astc, 5);
+}
diff --git a/lib/mikunyan/decoders/astc_block_decoder.rb b/lib/mikunyan/decoders/astc_block_decoder.rb
deleted file mode 100644
index 8f4a8f5..0000000
--- a/lib/mikunyan/decoders/astc_block_decoder.rb
+++ /dev/null
@@ -1,526 +0,0 @@
-require 'bin_utils'
-require 'fiddle'
-
-module Mikunyan
-    module DecodeHelper
-        # Class for decode ASTC block
-        # @attr_reader [String] data decoded data
-        class AstcBlockDecoder
-            attr_reader :data
-
-            # Decode block
-            # @param [String] bin binary
-            # @param [Integer] bw block width
-            # @param [Integer] bh block height
-            def initialize(bin, bw, bh)
-                if bin[0].ord == 0xfc && bin[1].ord % 2 == 1
-                    @data = (bin[9] + bin[11] + bin[13] + bin[15]) * bw * bh
-                else
-                    @d2 = BinUtils.get_int64_le(bin)
-                    @d1 = BinUtils.get_int64_le(bin, 8)
-                    @bw = bw
-                    @bh = bh
-
-                    decode_block_params
-                    decode_endpoints
-                    decode_weights
-                    select_partition
-                    applicate_color
-                end
-            end
-
-            private
-
-            WeightPrecTableA = [nil, nil, 0, 3, 0, 5, 3, 0, nil, nil, 5, 3, 0, 5, 3, 0]
-            WeightPrecTableB = [nil, nil, 1, 0, 2, 0, 1, 3, nil, nil, 1, 2, 4, 2, 3, 5]
-
-            CemTableA = [0, 0, 3, 0, 5, 3, 0, 5, 3, 0, 5, 3, 0, 5, 3, 0, 5, 3, 0]
-            CemTableB = [1, 2, 1, 3, 1, 2, 4, 2, 3, 5, 3, 4, 6, 4, 5, 7, 5, 6, 8]
-
-            TritsTable = [
-                [0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 1, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 1, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2, 0, 1, 2, 0, 0, 1, 2, 1, 0, 1, 2, 2, 0, 1, 2, 2],
-                [0, 0, 0, 0, 1, 1, 1, 0, 2, 2, 2, 0, 2, 2, 2, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 2, 2, 1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 2, 2, 2, 0, 2, 2, 2, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 2, 2, 1, 1, 1, 1, 0, 0, 0, 0, 0, 1, 1, 1, 0, 2, 2, 2, 0, 2, 2, 2, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 2, 2, 1, 2, 2, 2, 0, 0, 0, 0, 0, 1, 1, 1, 0, 2, 2, 2, 0, 2, 2, 2, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 2, 2, 1, 2, 2, 2, 0, 0, 0, 0, 0, 1, 1, 1, 0, 2, 2, 2, 0, 2, 2, 2, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 2, 2, 1, 0, 0, 0, 1, 0, 0, 0, 0, 1, 1, 1, 0, 2, 2, 2, 0, 2, 2, 2, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 2, 2, 1, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 0, 2, 2, 2, 0, 2, 2, 2, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 2, 2, 1, 2, 2, 2, 1, 0, 0, 0, 0, 1, 1, 1, 0, 2, 2, 2, 0, 2, 2, 2, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 2, 2, 1, 2, 2, 2, 1],
-                [0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 2, 2, 2, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 2, 2, 2, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 2, 2, 2, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 2, 2, 2, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 2, 2, 2, 2, 0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 2, 2, 2, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 2, 2, 2, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 2, 2, 2, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 0, 0, 0, 2, 0, 0, 0, 2, 0, 0, 0, 2, 2, 2, 2, 2, 1, 1, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 2, 2, 2, 2],
-                [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2],
-                [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2]
-            ]
-
-            QuintsTable = [
-                [0, 1, 2, 3, 4, 0, 4, 4, 0, 1, 2, 3, 4, 1, 4, 4, 0, 1, 2, 3, 4, 2, 4, 4, 0, 1, 2, 3, 4, 3, 4, 4, 0, 1, 2, 3, 4, 0, 4, 0, 0, 1, 2, 3, 4, 1, 4, 1, 0, 1, 2, 3, 4, 2, 4, 2, 0, 1, 2, 3, 4, 3, 4, 3, 0, 1, 2, 3, 4, 0, 2, 3, 0, 1, 2, 3, 4, 1, 2, 3, 0, 1, 2, 3, 4, 2, 2, 3, 0, 1, 2, 3, 4, 3, 2, 3, 0, 1, 2, 3, 4, 0, 0, 1, 0, 1, 2, 3, 4, 1, 0, 1, 0, 1, 2, 3, 4, 2, 0, 1, 0, 1, 2, 3, 4, 3, 0, 1],
-                [0, 0, 0, 0, 0, 4, 4, 4, 1, 1, 1, 1, 1, 4, 4, 4, 2, 2, 2, 2, 2, 4, 4, 4, 3, 3, 3, 3, 3, 4, 4, 4, 0, 0, 0, 0, 0, 4, 0, 4, 1, 1, 1, 1, 1, 4, 1, 4, 2, 2, 2, 2, 2, 4, 2, 4, 3, 3, 3, 3, 3, 4, 3, 4, 0, 0, 0, 0, 0, 4, 0, 0, 1, 1, 1, 1, 1, 4, 1, 1, 2, 2, 2, 2, 2, 4, 2, 2, 3, 3, 3, 3, 3, 4, 3, 3, 0, 0, 0, 0, 0, 4, 0, 0, 1, 1, 1, 1, 1, 4, 1, 1, 2, 2, 2, 2, 2, 4, 2, 2, 3, 3, 3, 3, 3, 4, 3, 3],
-                [0, 0, 0, 0, 0, 0, 0, 4, 0, 0, 0, 0, 0, 0, 1, 4, 0, 0, 0, 0, 0, 0, 2, 4, 0, 0, 0, 0, 0, 0, 3, 4, 1, 1, 1, 1, 1, 1, 4, 4, 1, 1, 1, 1, 1, 1, 4, 4, 1, 1, 1, 1, 1, 1, 4, 4, 1, 1, 1, 1, 1, 1, 4, 4, 2, 2, 2, 2, 2, 2, 4, 4, 2, 2, 2, 2, 2, 2, 4, 4, 2, 2, 2, 2, 2, 2, 4, 4, 2, 2, 2, 2, 2, 2, 4, 4, 3, 3, 3, 3, 3, 3, 4, 4, 3, 3, 3, 3, 3, 3, 4, 4, 3, 3, 3, 3, 3, 3, 4, 4, 3, 3, 3, 3, 3, 3, 4, 4]
-            ]
-
-            def [](i, j = 1)
-                if j < 1
-                    0
-                elsif j == 1
-                    i < 64 ? @d2[i] : @d1[i - 64]
-                else
-                    if i + j <= 64
-                        @d2 >> i & (1 << j) - 1
-                    elsif i >= 64
-                        @d1 >> (i - 64) & (1 << j) - 1
-                    else
-                        @d2 >> i | (@d1 & (1 << i + j - 64) - 1) << 64 - i
-                    end
-                end
-            end
-
-            def decode_block_params
-                # Block Mode
-                @weight_range = @d2 >> 4 & 1 | @d2 >> 6 & 8
-                @dual_plane = @d2 & 0x400 == 0x400
-                if @d2 & 0x3 != 0
-                    @weight_range |= @d2 << 1 & 6
-                    case @d2 & 0xc
-                    when 0
-                        @width = (@d2 >> 7 & 3) + 4
-                        @height = (@d2 >> 5 & 3) + 2
-                    when 0x4
-                        @width = (@d2 >> 7 & 3) + 8
-                        @height = (@d2 >> 5 & 3) + 2
-                    when 0x8
-                        @width = (@d2 >> 5 & 3) + 2
-                        @height = (@d2 >> 7 & 3) + 8
-                    else # 0xc
-                        if @d2 & 0x100 == 0
-                            @width = (@d2 >> 5 & 3) + 2
-                            @height = @d2[7] + 6
-                        else
-                            @width = @d2[7] + 2
-                            @height = (@d2 >> 5 & 3) + 2
-                        end
-                    end
-                else
-                    @weight_range |= @d2 >> 1 & 6
-                    case @d2 & 0x180
-                    when 0
-                        @width = 12
-                        @height = (@d2 >> 5 & 3) + 2
-                    when 0x80
-                        @width = (@d2 >> 5 & 3) + 2
-                        @height = 12
-                    when 0x180
-                        @width = (@d2 & 0x20 == 0) ? 6 : 10
-                        @height = 16 - @width
-                    else # 0x100
-                        @width = (@d2 >> 5 & 3) + 6
-                        @height = (@d2 >> 9 & 3) + 6
-                        @dual_plane = false
-                        @weight_range &= 7
-                    end
-                end
-
-                # Count Partitions
-                @part_num = (@d2 >> 11 & 3) + 1
-
-                # Count Weight Bits
-                @weight_num = @width * @height
-                @weight_num *= 2 if @dual_plane
-                case WeightPrecTableA[@weight_range]
-                when 3
-                    @weight_bit = @weight_num * WeightPrecTableB[@weight_range] + (@weight_num * 8 + 4) / 5
-                when 5
-                    @weight_bit = @weight_num * WeightPrecTableB[@weight_range] + (@weight_num * 7 + 2) / 3
-                else # 0
-                    @weight_bit = @weight_num * WeightPrecTableB[@weight_range]
-                end
-
-                # CEM
-                if @part_num == 1
-                    @cem = [@d2 >> 13 & 0xf]
-                    config_bit = 17
-                else
-                    cembase = @d2 >> 23 & 3
-                    if cembase == 0
-                        @cem = Array.new(@part_num, @d2 >> 25 & 0xf)
-                        config_bit = 29
-                    else
-                        @cem = (0...@part_num).map{|i| ((@d2 >> (25 + i) & 1) + cembase - 1) << 2}
-
-                        case @part_num
-                        when 2
-                            @cem[0] |= @d2 >> 27 & 3
-                            @cem[1] |= self[126 - @weight_bit, 2]
-                        when 3
-                            @cem[0] |= @d2[28]
-                            @cem[0] |= self[123 - @weight_bit] << 1
-                            @cem[1] |= self[124 - @weight_bit, 2]
-                            @cem[2] |= self[126 - @weight_bit, 2]
-                        else # 4
-                            4.times do |i|
-                                @cem[i] |= self[120 + 2 * i - @weight_bit, 2]
-                            end
-                        end
-
-                        config_bit = 25 + @part_num * 3
-                    end
-                end
-
-                # Count Color Endpoint Bits
-                config_bit += 2 if @dual_plane
-                remain_bit = 128 - config_bit - @weight_bit
-                @cem_num = @cem.map{|i| (i >> 1 & 6) + 2}.inject(:+)
-
-                CemTableA.count.times do |n|
-                    i = CemTableA.count - n - 1
-                    case CemTableA[i]
-                    when 3
-                        @cem_bit = @cem_num * CemTableB[i] + (@cem_num * 8 + 4) / 5
-                    when 5
-                        @cem_bit = @cem_num * CemTableB[i] + (@cem_num * 7 + 2) / 3
-                    else # 0
-                        @cem_bit = @cem_num * CemTableB[i]
-                    end
-
-                    if @cem_bit <= remain_bit
-                        @cem_range = i
-                        break
-                    end
-                end
-
-                if @dual_plane
-                    if @part_num == 1 || cembase == 0
-                        @plane_selector = self[126 - @weight_bit, 2]
-                    else
-                        @plane_selector = self[130 - @weight_bit - @part_num * 3, 2]
-                    end
-                end
-            end
-
-            def decode_endpoints
-                values = decode_intseq_raw(self[@part_num == 1 ? 17 : 29, @cem_bit], CemTableA[@cem_range], CemTableB[@cem_range], @cem_num).map do |e|
-                    unquantize_endpoint(CemTableA[@cem_range], CemTableB[@cem_range], e[0], e[1])
-                end
-
-                @endpoint = @cem.map do |cem|
-                    v = values.slice!(0, (cem >> 1 & 6) + 2)
-                    case cem
-                    when 0
-                        [v[0], v[0], v[0], 255, v[1], v[1], v[1], 255]
-                    when 1
-                        l0 = (v[0] >> 2) | (v[1] & 0xc0)
-                        l1 = (l0 + (v[1] & 0x3f)).clamp(0, 255)
-                        [l0, l0, l0, 255, l1, l1, l1, 255]
-                    when 4
-                        [v[0], v[0], v[0], v[2], v[1], v[1], v[1], v[3]]
-                    when 5
-                        v[1], v[0] = bit_transfer_signed(v[1], v[0])
-                        v[3], v[2] = bit_transfer_signed(v[3], v[2])
-                        [v[0], v[0], v[0], v[2], v[0] + v[1], v[0] + v[1], v[0] + v[1], v[2] + v[3]].map{|i| i.clamp(0, 255)}
-                    when 6
-                        [v[0] * v[3] >> 8, v[1] * v[3] >> 8, v[2] * v[3] >> 8, 255, v[0], v[1], v[2], 255]
-                    when 8
-                        if v[0] + v[2] + v[4] <= v[1] + v[3] + v[5]
-                            [v[0], v[2], v[4], 255, v[1], v[3], v[5], 255]
-                        else
-                            blue_contract(v[1], v[3], v[5], 255, v[0], v[2], v[4], 255)
-                        end
-                    when 9
-                        v[1], v[0] = bit_transfer_signed(v[1], v[0])
-                        v[3], v[2] = bit_transfer_signed(v[3], v[2])
-                        v[5], v[4] = bit_transfer_signed(v[5], v[4])
-                        if v[1] + v[3] + v[5] >= 0
-                            [v[0], v[2], v[4], 255, v[0] + v[1], v[2] + v[3], v[4] + v[5], 255].map{|i| i.clamp(0, 255)}
-                        else
-                            blue_contract(v[0] + v[1], v[2] + v[3], v[4] + v[5], 255, v[0], v[2], v[4], 255).map{|i| i.clamp(0, 255)}
-                        end
-                    when 10
-                        [v[0] * v[3] >> 8, v[1] * v[3] >> 8, v[2] * v[3] >> 8, v[4], v[0], v[1], v[2], v[5]]
-                    when 12
-                        if v[0] + v[2] + v[4] <= v[1] + v[3] + v[5]
-                            [v[0], v[2], v[4], v[6], v[1], v[3], v[5], v[7]]
-                        else
-                            blue_contract(v[1], v[3], v[5], v[7], v[0], v[2], v[4], v[6])
-                        end
-                    when 13
-                        v[1], v[0] = bit_transfer_signed(v[1], v[0])
-                        v[3], v[2] = bit_transfer_signed(v[3], v[2])
-                        v[5], v[4] = bit_transfer_signed(v[5], v[4])
-                        v[7], v[6] = bit_transfer_signed(v[7], v[6])
-                        if v[1] + v[3] + v[5] >= 0
-                            [v[0], v[2], v[4], v[6], v[0] + v[1], v[2] + v[3], v[4] + v[5], v[6] + v[7]].map{|i| i.clamp(0, 255)}
-                        else
-                            blue_contract(v[0] + v[1], v[2] + v[3], v[4] + v[5], v[6] + v[7], v[0], v[2], v[4], v[6]).map{|i| i.clamp(0, 255)}
-                        end
-                    else
-                        throw NotImplementedError.new("HDR image is not supported. (CEM: #{cem})")
-                    end
-                end
-            end
-
-            def bit_transfer_signed(a, b)
-                b = (b >> 1) | (a & 0x80)
-                a = (a >> 1) & 0x3f
-                [a[5] == 1 ? a - 0x40 : a, b]
-            end
-
-            def blue_contract(r1, g1, b1, a1, r2, g2, b2, a2)
-                [(r1 + b1) >> 1, (g1 + b1) >> 1, b1, a1, (r2 + b2) >> 1, (g2 + b2) >> 1, b2, a2]
-            end
-
-            def decode_weights
-                data = (0...(@weight_bit + 7) / 8).map{|i| (((self[120 - i * 8, 8]) * 0x80200802) & 0x0884422110) * 0x0101010101 >> 32 & 0xff}
-                    .map.with_index{|e, i| e << i * 8}.inject(:|) & (1 << @weight_bit) - 1
-
-                weight_point = decode_intseq_raw(data, WeightPrecTableA[@weight_range], WeightPrecTableB[@weight_range], @weight_num).map do |e|
-                    unquantize_weight(WeightPrecTableA[@weight_range], WeightPrecTableB[@weight_range], e[0], e[1])
-                end
-
-                ds = (1024 + @bw / 2) / (@bw - 1)
-                dt = (1024 + @bh / 2) / (@bh - 1)
-
-                if @dual_plane
-                    @weight0 = Fiddle::Pointer.malloc(@bw * @bh)
-                    @weight1 = Fiddle::Pointer.malloc(@bw * @bh)
-
-                    for t in 0...@bh
-                        for s in 0...@bw
-                            gs = (ds * s * (@width - 1) + 32) >> 6
-                            gt = (dt * t * (@height - 1) + 32) >> 6
-                            fs = gs & 0xf
-                            ft = gt & 0xf
-                            v = (gs >> 4) + (gt >> 4) * @width
-                            w11 = (fs * ft + 8) >> 4
-                            w10 = ft - w11
-                            w01 = fs - w11
-                            w00 = 16 - fs - ft + w11
-
-                            p00 = weight_point[v * 2] || 0
-                            p01 = weight_point[(v + 1) * 2] || 0
-                            p10 = weight_point[(v + @width) * 2] || 0
-                            p11 = weight_point[(v + @width + 1) * 2] || 0
-                            @weight0[s + t * @bw] = (p00 * w00 + p01 * w01 + p10 * w10 + p11 * w11 + 8) >> 4
-
-                            p00 = weight_point[v * 2 + 1] || 0
-                            p01 = weight_point[(v + 1) * 2 + 1] || 0
-                            p10 = weight_point[(v + @width) * 2 + 1] || 0
-                            p11 = weight_point[(v + @width + 1) * 2 + 1] || 0
-                            @weight1[s + t * @bw] = (p00 * w00 + p01 * w01 + p10 * w10 + p11 * w11 + 8) >> 4
-                        end
-                    end
-                else
-                    @weight = Fiddle::Pointer.malloc(@bw * @bh)
-
-                    for t in 0...@bh
-                        for s in 0...@bw
-                            gs = (ds * s * (@width - 1) + 32) >> 6
-                            gt = (dt * t * (@height - 1) + 32) >> 6
-                            fs = gs & 0xf
-                            ft = gt & 0xf
-                            v = (gs >> 4) + (gt >> 4) * @width
-                            w11 = (fs * ft + 8) >> 4
-
-                            p00 = weight_point[v] || 0
-                            p01 = weight_point[v + 1] || 0
-                            p10 = weight_point[v + @width] || 0
-                            p11 = weight_point[v + @width + 1] || 0
-                            @weight[s + t * @bw] = (p00 * (16 - fs - ft + w11) + p01 * (fs - w11) + p10 * (ft - w11) + p11 * w11 + 8) >> 4
-                        end
-                    end
-                end
-            end
-
-            def select_partition
-                if @part_num > 1
-                    small_block = @bw * @bh < 31
-
-                    seed = (@d2 >> 13 & 0x3ff) | ((@part_num - 1) << 10)
-
-                    rnum = seed
-                    rnum ^= (rnum >> 15)
-                    rnum = (rnum - (rnum << 17)) & 0xffffffff
-                    rnum = (rnum + (rnum << 7)) & 0xffffffff
-                    rnum = (rnum + (rnum << 4)) & 0xffffffff
-                    rnum ^= rnum >> 5
-                    rnum = (rnum + (rnum << 16)) & 0xffffffff
-                    rnum ^= rnum >> 7
-                    rnum ^= rnum >> 3
-                    rnum = (rnum ^ (rnum << 6)) & 0xffffffff
-                    rnum = rnum ^ (rnum >> 17)
-
-                    seeds = [0, 4, 8, 12, 16, 20, 24, 28].map{|i| (rnum >> i) & 0xf}.map!{|e| e * e}
-                    sh = [seed & 2 == 2 ? 4 : 5, @part_num == 3 ? 6 : 5]
-                    sh.reverse! if seed & 1 == 0
-                    seeds.map!.with_index{|e, i| e >> sh[i % 2]}
-
-                    @partition = (0...@bw * @bh).map do |i|
-                        x = i % @bw
-                        y = i / @bw
-                        if small_block
-                            x <<= 1
-                            y <<= 1
-                        end
-
-                        a = (seeds[0] * x + seeds[1] * y + (rnum >> 14)) & 0x3f
-                        b = (seeds[2] * x + seeds[3] * y + (rnum >> 10)) & 0x3f
-                        c = @part_num < 3 ? 0 : (seeds[4] * x + seeds[5] * y + (rnum >> 6)) & 0x3f
-                        d = @part_num < 4 ? 0 : (seeds[6] * x + seeds[7] * y + (rnum >> 2)) & 0x3f
-
-                        3 - [d, c, b, a].each_with_index.max[1]
-                    end
-                end
-            end
-
-            def applicate_color
-                @data = String.new(capacity: @bw * @bh * 4)
-
-                if @dual_plane
-                    plane_arr = [0, 1, 2, 3]
-                    plane_arr.delete_at(@plane_selector)
-
-                    if @partition
-                        (@bw * @bh).times do |i|
-                            part = @partition[i]
-                            4.times do |c|
-                                BinUtils.append_int8!(@data, select_color(@endpoint[part][c], @endpoint[part][4 + c], c == @plane_selector ? @weight1[i] : @weight0[i]))
-                            end
-                        end
-                    else
-                        (@bw * @bh).times do |i|
-                            4.times do |c|
-                                BinUtils.append_int8!(@data, select_color(@endpoint[0][c], @endpoint[0][4 + c], c == @plane_selector ? @weight1[i] : @weight0[i]))
-                            end
-                        end
-                    end
-                elsif @partition
-                    (@bw * @bh).times do |i|
-                        part = @partition[i]
-                        4.times{|c| BinUtils.append_int8!(@data, select_color(@endpoint[part][c], @endpoint[part][4 + c], @weight[i]))}
-                    end
-                else
-                    (@bw * @bh).times do |i|
-                        4.times{|c| BinUtils.append_int8!(@data, select_color(@endpoint[0][c], @endpoint[0][4 + c], @weight[i]))}
-                    end
-                end
-            end
-
-            def select_color(v0, v1, weight)
-                v0 |= v0 << 8
-                v1 |= v1 << 8
-                v = (v0 * (64 - weight) + v1 * weight + 32) >> 6
-                (v * 255 + 32768) / 65536
-            end
-
-            def decode_intseq_raw(data, a, b, count)
-                mask = (1 << b) - 1
-                case a
-                when 3
-                    rc = (count + 4) / 5
-                    ret = Array.new(rc * 5)
-                    m = [0, 2 + b, 4 + b * 2, 5 + b * 3, 7 + b * 4]
-                    rc.times do |i|
-                        t = (data >> b & 3) | (data >> b * 2 & 0xc) | (data >> b * 3 & 0x10) | (data >> b * 4 & 0x60) | (data >> b * 5 & 0x80)
-                        5.times do |j|
-                            ret[i * 5 + j] = [data >> m[j] & mask, TritsTable[j][t]]
-                        end
-                        data >>= b * 5 + 8
-                    end
-                    ret[0, count]
-                when 5
-                    rc = (count + 2) / 3
-                    ret = Array.new(rc * 3)
-                    m = [0, 3 + b, 5 + b * 2]
-                    rc.times do |i|
-                        q = (data >> b & 7) | (data >> b * 2 & 0x18) | (data >> b * 3 & 0x60)
-                        3.times do |j|
-                            ret[i * 3 + j] = [data >> m[j] & mask, QuintsTable[j][q]]
-                        end
-                        data >>= b * 3 + 7
-                    end
-                    ret[0, count]
-                else # 0
-                    (0...count).map do |i|
-                        [data >> b * i & mask, 0]
-                    end
-                end
-            end
-
-            def unquantize_endpoint(a, b, bit, val_d)
-                if a == 0
-                    case b
-                    when 1
-                        bit * 0xff
-                    when 2
-                        bit * 0x55
-                    when 3
-                        bit << 5 | bit << 2 | bit >> 1
-                    when 4
-                        bit << 4 | bit
-                    when 5
-                        bit << 3 | bit >> 2
-                    when 6
-                        bit << 2 | bit >> 4
-                    when 7
-                        bit << 1 | bit >> 6
-                    else # 8
-                        bit
-                    end
-                else
-                    val_a = (bit & 1) * 0x1ff
-                    tmp_b = bit >> 1
-                    case b
-                    when 1
-                        val_b = 0
-                        val_c = a == 3 ? 204 : 113
-                    when 2
-                        val_b = a == 3 ? (0b100010110) * tmp_b : (0b100001100) * tmp_b
-                        val_c = a == 3 ? 93 : 54
-                    when 3
-                        val_b = a == 3 ? tmp_b << 7 | tmp_b << 2 | tmp_b : tmp_b << 7 | tmp_b << 1 | tmp_b >> 1
-                        val_c = a == 3 ? 44 : 26
-                    when 4
-                        val_b = tmp_b << 6 | tmp_b >> (a == 3 ? 0 : 1)
-                        val_c = a == 3 ? 22 : 13
-                    when 5
-                        val_b = tmp_b << 5 | tmp_b >> (a == 3 ? 2 : 3)
-                        val_c = a == 3 ? 11 : 6
-                    else # 6
-                        val_b = tmp_b << 4 | tmp_b >> 4
-                        val_c = 5
-                    end
-                    t = val_d * val_c + val_b
-                    t ^= val_a
-                    (val_a & 0x80) | (t >> 2)
-                end
-            end
-
-            def unquantize_weight(a, b, bit, val_d)
-                if a == 0
-                    case b
-                    when 1
-                        t = bit == 1 ? 63 : 0
-                    when 2
-                        t = bit << 4 | bit << 2 | bit
-                    when 3
-                        t = bit << 3 | bit
-                    when 4
-                        t = bit << 2 | bit >> 2
-                    else # 5
-                        t = bit << 1 | bit >> 4
-                    end
-                elsif b == 0
-                    t = (a == 3 ? [0, 32, 63] : [0, 16, 32, 47, 63])[val_d]
-                else
-                    val_a = (bit & 1) * 0x7f
-                    case b
-                    when 1
-                        val_b = 0
-                        val_c = a == 3 ? 50 : 28
-                    when 2
-                        val_b = (a == 3 ? 0b1000101 : 0b1000010) * bit[1]
-                        val_c = a == 3 ? 23 : 13
-                    else # 3
-                        val_b = (bit << 4 | bit >> 1) & 0b1100011
-                        val_c = 11
-                    end
-                    t = val_d * val_c + val_b
-                    t ^= val_a
-                    t = (val_a & 0x20) | (t >> 2)
-                end
-                t > 32 ? t + 1 : t
-            end
-        end
-    end
-end
diff --git a/lib/mikunyan/decoders/image_decoder.rb b/lib/mikunyan/decoders/image_decoder.rb
index dcf91d5..f551729 100644
--- a/lib/mikunyan/decoders/image_decoder.rb
+++ b/lib/mikunyan/decoders/image_decoder.rb
@@ -1,6 +1,6 @@
 begin; require 'oily_png'; rescue LoadError; require 'chunky_png'; end
 require 'bin_utils'
-require 'mikunyan/decoders/astc_block_decoder'
+require 'mikunyan/decoders/native'
 require 'mikunyan/decoders/dxtc_block_decoder'
 
 module Mikunyan
@@ -455,16 +455,7 @@ module Mikunyan
         # @param [String] bin binary to decode
         # @return [ChunkyPNG::Image] decoded image
         def self.decode_astc(width, height, blocksize, bin)
-            bw = (width + blocksize - 1) / blocksize
-            bh = (height + blocksize - 1) / blocksize
-            ret = ChunkyPNG::Image.new(bw * blocksize, bh * blocksize)
-            bh.times do |by|
-                bw.times do |bx|
-                    block = DecodeHelper::AstcBlockDecoder.new(bin.byteslice((by * bw + bx) * 16, 16), blocksize, blocksize).data
-                    ret.replace!(ChunkyPNG::Image.from_rgba_stream(blocksize, blocksize, block), bx * blocksize, by * blocksize)
-                end
-            end
-            ret.crop(0, 0, width, height).flip
+            ChunkyPNG::Image.from_rgba_stream(width, height, DecodeHelper.decode_astc(bin, width, height, blocksize, blocksize))
         end
 
         # Create ASTC file data from ObjectValue
diff --git a/mikunyan.gemspec b/mikunyan.gemspec
index 75b2944..8f4586a 100644
--- a/mikunyan.gemspec
+++ b/mikunyan.gemspec
@@ -20,6 +20,7 @@ Gem::Specification.new do |spec|
   spec.bindir        = "exe"
   spec.executables   = spec.files.grep(%r{^exe/}) { |f| File.basename(f) }
   spec.require_paths = ["lib"]
+  spec.extensions    = ["ext/decoders/native/extconf.rb"]
 
   spec.add_dependency 'extlz4', '~> 0'
   spec.add_dependency 'bin_utils', '~> 0'