Fix glUniformMatrix4fv in CAN_ADDRESS_2GB mode (#21264)

Author: sbc100

.circleci/config.yml

@@ -818,10 +818,7 @@ jobs:
             browser_2gb.test_fetch_to_memory
             browser_2gb.test_emscripten_animate_canvas_element_size_manual_css
             browser_2gb.test_fulles2_sdlproc
-            browser_2gb.test_cubegeom
-            browser_2gb.test_cubegeom_normal
-            browser_2gb.test_cubegeom_normal_dap
-            browser_2gb.test_cubegeom_normal_dap_far
+            browser_2gb.test_cubegeom*
             "
   test-browser-chrome-wasm64-4gb:
     executor: bionic

src/library_webgl.js

@@ -2911,14 +2911,14 @@ for (/**@suppress{duplicate}*/var i = 0; i < {{{ GL_POOL_TEMP_BUFFERS_SIZE }}};
 #if GL_ASSERTIONS
     assert(GL.currentContext.version >= 2);
 #endif
-    count && GLctx.uniformMatrix4fv(webglGetUniformLocation(location), !!transpose, HEAPF32, value>>2, count*16);
+    count && GLctx.uniformMatrix4fv(webglGetUniformLocation(location), !!transpose, HEAPF32, {{{ getHeapOffset('value', 'float') }}}, count*16);
 #else
 
 #if MAX_WEBGL_VERSION >= 2
     // WebGL 2 provides new garbage-free entry points to call to WebGL. Use
     // those always when possible.
     if ({{{ isCurrentContextWebGL2() }}}) {
-      count && GLctx.uniformMatrix4fv(webglGetUniformLocation(location), !!transpose, HEAPF32, value>>2, count*16);
+      count && GLctx.uniformMatrix4fv(webglGetUniformLocation(location), !!transpose, HEAPF32, {{{ getHeapOffset('value', 'float') }}}, count*16);
       return;
     }
 #endif
@@ -2929,7 +2929,7 @@ for (/**@suppress{duplicate}*/var i = 0; i < {{{ GL_POOL_TEMP_BUFFERS_SIZE }}};
       var view = miniTempWebGLFloatBuffers[16*count-1];
       // hoist the heap out of the loop for size and for pthreads+growth.
       var heap = HEAPF32;
-      value >>= 2;
+      value = {{{ getHeapOffset('value', 'float') }}};
       for (var i = 0; i < 16 * count; i += 16) {
         var dst = value + i;
         view[i] = heap[dst];