+checkpoint: fix .ispc kernels for compilation with llc->ptx

examples_ptx/aobench/ao.ispc

@@ -81,6 +81,9 @@ static inline void vnormalize(vec &v) {
 }
 
 
+#if 1
+inline
+#endif
 static void
 ray_plane_intersect(Isect &isect, Ray &ray, const uniform Plane &plane) {
     float d = -dot(plane.p, plane.n);
@@ -123,6 +126,9 @@ ray_sphere_intersect(Isect &isect, Ray &ray, const uniform Sphere &sphere) {
 }
 
 
+#if 1
+inline
+#endif
 static void
 orthoBasis(vec basis[3], vec n) {
     basis[2] = n;
@@ -146,6 +152,9 @@ orthoBasis(vec basis[3], vec n) {
 }
 
 
+#if 1
+inline
+#endif
 static float
 ambient_occlusion(Isect &isect, const uniform Plane &plane, const uniform Sphere spheres[3],
                   RNGState &rngstate) {

examples_ptx/common_gpu.mk

@@ -108,8 +108,8 @@ objs_gpu/%_ispc.h objs_gpu/%_ispc.bc: %.ispc
 objs_gpu/%_llvm_ispc.ptx: objs_gpu/%_ispc.bc
 	$(LLC) $(LLC_FLAGS) -o $@ $<
 objs_gpu/%_nvvm_ispc.ptx: objs_gpu/%_ispc.bc
-	$(LLVM32DIS) $< -o objs/$*_ispc-ll32.ll
-	$(PTXGEN) objs/$*_ispc-ll32.ll > $@
+	$(LLVM32DIS) $< -o objs_gpu/$*_ispc-ll32.ll
+	$(PTXGEN) objs_gpu/$*_ispc-ll32.ll > $@
 
 # generate an object file from PTX
 objs_gpu/%_ispc.o: objs_gpu/%_ispc.ptx

examples_ptx/rt/rt.ispc

@@ -80,6 +80,9 @@ static inline float Dot(const float3 a, const float3 b) {
 }
 
 
+#if 1
+inline
+#endif
 static void generateRay(uniform const float raster2camera[4][4], 
                         uniform const float camera2world[4][4],
                         float x, float y, Ray &ray) {
@@ -116,6 +119,9 @@ static void generateRay(uniform const float raster2camera[4][4],
 }
 
 
+#if 1
+inline
+#endif
 static bool_t BBoxIntersect(const uniform float bounds[2][3], 
                           const Ray &ray) {
     uniform float3 bounds0 = { bounds[0][0], bounds[0][1], bounds[0][2] };
@@ -155,6 +161,9 @@ static bool_t BBoxIntersect(const uniform float bounds[2][3],
 
 
 
+#if 1
+inline
+#endif
 static bool_t TriIntersect(const uniform Triangle &tri, Ray &ray) {
     uniform float3 p0 = { tri.p[0][0], tri.p[0][1], tri.p[0][2] };
     uniform float3 p1 = { tri.p[1][0], tri.p[1][1], tri.p[1][2] };
@@ -195,6 +204,9 @@ static bool_t TriIntersect(const uniform Triangle &tri, Ray &ray) {
 }
 
 
+#if 1
+inline
+#endif
 bool_t
 BVHIntersect(const uniform LinearBVHNode nodes[], 
                   const uniform Triangle tris[], Ray &r) {
@@ -253,6 +265,9 @@ BVHIntersect(const uniform LinearBVHNode nodes[],
 }
 
 
+#if 1
+inline
+#endif
 static void raytrace_tile(uniform int x0, uniform int x1,
                           uniform int y0, uniform int y1, 
                           uniform int width, uniform int height,