some changes

tests/cfor-gs-improve-multidim-struct-1.ispc

@@ -4,19 +4,27 @@ export uniform int width() { return programCount; }
 
 
 struct Foo {
-    uniform float udx[25][25];
+    uniform float udx[32][32];
 };
 
 export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
     float a = aFOO[programIndex]; 
+#ifndef __NVPTX__ 
     uniform Foo f[5];
+#else __NVPTX__  /* too much shared memory allocated, nvcc fails to link */
+    uniform Foo * uniform f = uniform new uniform Foo[5];
+#define _UNMALLOC
+#endif
     cfor (uniform int i = 0; i < 5; ++i)
-        cfor (uniform int j = 0; j < 25; ++j)
-            cfor (uniform int k = 0; k < 25; ++k)
+        cfor (uniform int j = 0; j < 32; ++j)
+            cfor (uniform int k = 0; k < 32; ++k)
                 f[i].udx[j][k] = 1000*i+100*j+k;
 
     int x = 1;
     RET[programIndex] = f[x+1].udx[b-4][programIndex];
+#ifdef _UNMALLOC
+    delete f;
+#endif
 }
 
 export void result(uniform float RET[]) { RET[programIndex] = 2100 +programIndex; }