diff --git a/builtins/target-nvptx.ll b/builtins/target-nvptx.ll
index 1b5a6392..3f609b85 100644
--- a/builtins/target-nvptx.ll
+++ b/builtins/target-nvptx.ll
@@ -1404,3 +1404,63 @@ define <1 x i8*> @__insert_void(<1 x i8*>, i32,
   ret <1 x i8*> %insert
 }
 
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;; assert
+
+declare void @__assertfail(i64,i64,i32,i64,i64) noreturn;
+declare i32 @vprintf(i64,i64)
+define i32 @__puts_nvptx(i8*) alwaysinline
+{
+  %str   = ptrtoint i8* %0 to i64
+  %parm  = or i64 0, 0
+  %call  = call i32 @vprintf(i64 %str, i64 %parm)
+  %cr    = alloca <2 x i8>
+  store <2 x i8> <i8 10, i8 0>, <2 x i8>* %cr
+  %cr1   = ptrtoint <2 x i8>* %cr to i64
+  %call1 = call i32 @vprintf(i64 %cr1, i64 %parm)
+  ret i32 %call1;
+}
+define void @__abort_nvptx(i8* %str) noreturn
+{
+  %tmp1 = alloca <3 x i8>
+  store <3 x i8> <i8 58, i8 58, i8 0>, <3 x i8>* %tmp1
+  %tmp2 = alloca <2 x i8>
+  store <2 x i8> <i8 0, i8 0>, <2 x i8>* %tmp2
+
+  %param1 = ptrtoint <2 x i8>* %tmp2 to i64
+  %param3 = or i32 0, 0
+  %string = ptrtoint i8* %str to i64
+  %param4 = ptrtoint <3 x i8>* %tmp1 to i64
+  %param5 = or i64 1, 1
+  call void @__assertfail(i64 %param1, i64 %string, i32 %param3, i64 %param4, i64 %param5);
+  ret void
+}
+
+define void @__do_assert_uniform(i8 *%str, i1 %test, <WIDTH x MASK> %mask) {
+  br i1 %test, label %ok, label %fail
+
+fail:
+  call void @__abort_nvptx(i8* %str) noreturn
+  unreachable
+
+ok:
+  ret void
+}
+
+
+define void @__do_assert_varying(i8 *%str, <WIDTH x MASK> %test,
+                                 <WIDTH x MASK> %mask) {
+  %nottest = xor <WIDTH x MASK> %test,
+                 < forloop(i, 1, eval(WIDTH-1), `MASK -1, ') MASK -1 >
+  %nottest_and_mask = and <WIDTH x MASK> %nottest, %mask
+  %mm = call i64 @__movmsk(<WIDTH x MASK> %nottest_and_mask)
+  %all_ok = icmp eq i64 %mm, 0
+  br i1 %all_ok, label %ok, label %fail
+
+fail:
+  call void @__abort_nvptx(i8* %str) noreturn
+  unreachable
+
+ok:
+  ret void
+}
diff --git a/builtins/util-nvptx.m4 b/builtins/util-nvptx.m4
index 804c07e3..65577454 100644
--- a/builtins/util-nvptx.m4
+++ b/builtins/util-nvptx.m4
@@ -2064,42 +2064,6 @@ define void @__memset64(i8 * %dst, i8 %val, i64 %len) alwaysinline {
     ret void
 }
 
-;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
-;; assert
-
-declare i32 @puts(i8*)
-declare void @abort() noreturn
-
-define void @__do_assert_uniform(i8 *%str, i1 %test, <WIDTH x MASK> %mask) {
-  br i1 %test, label %ok, label %fail
-
-fail:
-  %call = call i32 @puts(i8* %str)
-  call void @abort() noreturn
-  unreachable
-
-ok:
-  ret void
-}
-
-
-define void @__do_assert_varying(i8 *%str, <WIDTH x MASK> %test,
-                                 <WIDTH x MASK> %mask) {
-  %nottest = xor <WIDTH x MASK> %test,
-                 < forloop(i, 1, eval(WIDTH-1), `MASK -1, ') MASK -1 >
-  %nottest_and_mask = and <WIDTH x MASK> %nottest, %mask
-  %mm = call i64 @__movmsk(<WIDTH x MASK> %nottest_and_mask)
-  %all_ok = icmp eq i64 %mm, 0
-  br i1 %all_ok, label %ok, label %fail
-
-fail:
-  %call = call i32 @puts(i8* %str)
-  call void @abort() noreturn
-  unreachable
-
-ok:
-  ret void
-}
 
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;; new/delete
diff --git a/tests/cfor-gs-improve-multidim-struct-1.ispc b/tests/cfor-gs-improve-multidim-struct-1.ispc
index ad4053dc..0d682f9a 100644
--- a/tests/cfor-gs-improve-multidim-struct-1.ispc
+++ b/tests/cfor-gs-improve-multidim-struct-1.ispc
@@ -11,7 +11,7 @@ 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 */
+#else     /* too much shared memory allocated, nvcc fails to link */
     uniform Foo * uniform f = uniform new uniform Foo[5];
 #define _UNMALLOC
 #endif
diff --git a/tests/soa-17.ispc b/tests/soa-17.ispc
index 897840f3..5dc9ea2f 100644
--- a/tests/soa-17.ispc
+++ b/tests/soa-17.ispc
@@ -39,9 +39,7 @@ export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
     float a = aFOO[programIndex]; 
     a *= -1;
     Point vp = { a, { 2*a, 3*a, 4*a }, {5*a} };
-#ifndef __NVPTX__
     assert(2+programIndex < 160);
-#endif
     pts[2+programIndex] = vp;
 
     RET[programIndex] = pts[programIndex].y[2];
diff --git a/tests/soa-22.ispc b/tests/soa-22.ispc
index 4d8fd299..ba3ffa0c 100644
--- a/tests/soa-22.ispc
+++ b/tests/soa-22.ispc
@@ -26,9 +26,7 @@ export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
         }
     }
    
-#ifndef __NVPTX__ 
     assert(programIndex < 80);
-#endif
     RET[programIndex] = pts[programIndex].pts[programIndex % 3][programIndex % 4].z;
 }