All wasm code was tested except for NLT.

Author: aous72

src/core/codestream/ojph_codestream_wasm.cpp

@@ -87,8 +87,8 @@ namespace ojph {
       v128_t zero = wasm_i32x4_splat(0);
       v128_t one = wasm_i32x4_splat(1);
       v128_t tmax = wasm_v128_load(max_val);
-      v128_t *p = (v128_t*)sp;
-      for (ui32 i = 0; i < count; i += 4, p += 1, dp += 4)
+      si32 *p = (si32*)sp;
+      for (ui32 i = 0; i < count; i += 4, p += 4, dp += 4)
       {
         v128_t v = wasm_v128_load(p);
         v128_t sign = wasm_i32x4_lt(v, zero);
@@ -192,7 +192,7 @@ namespace ojph {
       si64 *p = (si64*)sp;
       for (ui32 i = 0; i < count; i += 2, p += 2, dp += 2)
       {
-        v128_t v = wasm_v128_load((v128_t*)sp);
+        v128_t v = wasm_v128_load(p);
         v128_t sign = wasm_i64x2_lt(v, zero);
         v128_t val = wasm_v128_xor(v, sign); // negate 1's complement
         v128_t ones = wasm_v128_and(sign, one);
@@ -204,7 +204,7 @@ namespace ojph {
         wasm_v128_store(dp, val);
       }
       wasm_v128_store(max_val, tmax);
-    }
+    }   
 
     //////////////////////////////////////////////////////////////////////////
     void wasm_rev_tx_from_cb64(const ui64 *sp, void *dp, ui32 K_max, 

src/core/common/ojph_arch.h

@@ -286,9 +286,15 @@ namespace ojph {
   ////////////////////////////////////////////////////////////////////////////
   // constants
   ////////////////////////////////////////////////////////////////////////////
-  const ui32 byte_alignment = 64; // 64 bytes == 512 bits
-  const ui32 log_byte_alignment = 31 - count_leading_zeros(byte_alignment);
-  const ui32 object_alignment = 8;
+  #ifndef OJPH_EMSCRIPTEN
+    const ui32 byte_alignment = 64; // 64 bytes == 512 bits
+    const ui32 log_byte_alignment = 31 - count_leading_zeros(byte_alignment);
+    const ui32 object_alignment = 8;
+  #else
+    const ui32 byte_alignment = 16; // 16 bytes == 128 bits
+    const ui32 log_byte_alignment = 31 - count_leading_zeros(byte_alignment);
+    const ui32 object_alignment = 8;
+    #endif
 
   ////////////////////////////////////////////////////////////////////////////
   // templates for alignment

src/core/transform/ojph_colour_wasm.cpp

@@ -83,7 +83,7 @@ namespace ojph {
 
             t = wasm_i64x2_extend_high_i32x4(s);
             t = wasm_i64x2_add(t, sh);
-            wasm_v128_store(dp + 1, t);
+            wasm_v128_store(dp + 2, t);
           }            
         }
       }
@@ -99,7 +99,7 @@ namespace ojph {
           v128_t s0, s1;
           s0 = wasm_v128_load(sp);
           s0 = wasm_i64x2_add(s0, sh);
-          s1 = wasm_v128_load(sp + 1);
+          s1 = wasm_v128_load(sp + 2);
           s1 = wasm_i64x2_add(s1, sh);
           s0 = wasm_i32x4_shuffle(s0, s1, 0, 2, 4 + 0, 4 + 2);
           wasm_v128_store(dp, s0);
@@ -160,7 +160,7 @@ namespace ojph {
             u = wasm_v128_andnot(c, u);        // keep only +ve or 0
             u = wasm_v128_or(u, v_m_sh);       // combine
 
-            wasm_v128_store(dp + 1, u);
+            wasm_v128_store(dp + 2, u);
           }
         }
       }
@@ -184,7 +184,7 @@ namespace ojph {
           p = wasm_v128_andnot(m, s);   // +ve
           t0 = wasm_v128_or(n, p);
 
-          s = wasm_v128_load(sp + 1);
+          s = wasm_v128_load(sp + 2);
           m = wasm_i64x2_lt(s, zero);   // 64b -1 for -ve value
           tm = wasm_i64x2_sub(sh, s);   // - shift - value
           n = wasm_v128_and(m, tm);     // -ve

src/core/transform/ojph_transform_wasm.cpp

@@ -108,7 +108,7 @@ namespace ojph {
         v128_t a = wasm_v128_load(spl);
         v128_t b = wasm_v128_load(sph);
         v128_t c = wasm_i64x2_shuffle(a, b, 0, 2 + 0);
-        v128_t d = wasm_i64x2_shuffle(a, b, 2, 2 + 2);
+        v128_t d = wasm_i64x2_shuffle(a, b, 1, 2 + 1);
         wasm_v128_store(dp, c);
         wasm_v128_store(dp + 2, d);
       }

tests/CMakeLists.txt

@@ -3,7 +3,7 @@
 include(FetchContent)
 FetchContent_Declare(
   googletest
-  URL https://github.com/google/googletest/archive/refs/tags/v1.13.0.tar.gz
+  URL https://github.com/google/googletest/archive/refs/tags/v1.14.0.tar.gz
   EXCLUDE_FROM_ALL
 )
 # For Windows: Prevent overriding the parent project's compiler/linker settings

tests/test_executables.cpp

@@ -109,6 +109,12 @@ int execute(const std::string& cmd, std::string& result)
 #define COMPARE_FILES_PATH  "./compare_files"
 #define EXPAND_EXECUTABLE "./ojph_expand"
 #define COMPRESS_EXECUTABLE "./ojph_compress"
+//#define EXPAND_EXECUTABLE "20.18.0_64bit/bin/node ./ojph_expand.js"
+//#define COMPRESS_EXECUTABLE "20.18.0_64bit/bin/node ./ojph_compress.js"
+//#define EXPAND_EXECUTABLE "node-v18.7.0-linux-x64/bin/node ./ojph_expand_simd.js"
+//#define COMPRESS_EXECUTABLE "node-v18.7.0-linux-x64/bin/node ./ojph_compress_simd.js"
+//#define EXPAND_EXECUTABLE "./../../../sde/sde64 -skx -- ./ojph_expand"
+//#define COMPRESS_EXECUTABLE "./../../../sde/sde64 -skx -- ./ojph_compress"
 #endif
 #define TOL_DOUBLE 0.01
 #define TOL_INTEGER 1