Cycles: Fix crash with split branched path tracing
authorMai Lavelle <mai.lavelle@gmail.com>
Thu, 16 Nov 2017 08:32:02 +0000 (03:32 -0500)
committerMai Lavelle <mai.lavelle@gmail.com>
Thu, 16 Nov 2017 09:59:31 +0000 (04:59 -0500)
ShaderData memory was getting clobbered in the branched path code paths.

Was caused by 087331c495b04ebd37903c0dc0e46262354cf026

intern/cycles/kernel/split/kernel_branched.h
intern/cycles/kernel/split/kernel_next_iteration_setup.h
intern/cycles/kernel/split/kernel_split_data.h
intern/cycles/kernel/split/kernel_split_data_types.h
intern/cycles/kernel/split/kernel_subsurface_scatter.h

index 6456636..368a439 100644 (file)
@@ -33,9 +33,9 @@ ccl_device_inline void kernel_split_branched_path_indirect_loop_init(KernelGloba
        BRANCHED_STORE(isect);
        BRANCHED_STORE(ray_state);
 
-       branched_state->sd = *kernel_split_sd(sd, ray_index);
-       for(int i = 0; i < branched_state->sd.num_closure; i++) {
-               branched_state->sd.closure[i] = kernel_split_sd(sd, ray_index)->closure[i];
+       *kernel_split_sd(branched_state_sd, ray_index) = *kernel_split_sd(sd, ray_index);
+       for(int i = 0; i < kernel_split_sd(branched_state_sd, ray_index)->num_closure; i++) {
+               kernel_split_sd(branched_state_sd, ray_index)->closure[i] = kernel_split_sd(sd, ray_index)->closure[i];
        }
 
 #undef BRANCHED_STORE
@@ -60,9 +60,9 @@ ccl_device_inline void kernel_split_branched_path_indirect_loop_end(KernelGlobal
        BRANCHED_RESTORE(isect);
        BRANCHED_RESTORE(ray_state);
 
-       *kernel_split_sd(sd, ray_index) = branched_state->sd;
-       for(int i = 0; i < branched_state->sd.num_closure; i++) {
-               kernel_split_sd(sd, ray_index)->closure[i] = branched_state->sd.closure[i];
+       *kernel_split_sd(sd, ray_index) = *kernel_split_sd(branched_state_sd, ray_index);
+       for(int i = 0; i < kernel_split_sd(branched_state_sd, ray_index)->num_closure; i++) {
+               kernel_split_sd(sd, ray_index)->closure[i] = kernel_split_sd(branched_state_sd, ray_index)->closure[i];
        }
 
 #undef BRANCHED_RESTORE
@@ -83,10 +83,17 @@ ccl_device_inline bool kernel_split_branched_indirect_start_shared(KernelGlobals
        }
 
 #define SPLIT_DATA_ENTRY(type, name, num) \
-               kernel_split_state.name[inactive_ray] = kernel_split_state.name[ray_index];
+               if(num) { \
+                       kernel_split_state.name[inactive_ray] = kernel_split_state.name[ray_index]; \
+               }
        SPLIT_DATA_ENTRIES_BRANCHED_SHARED
 #undef SPLIT_DATA_ENTRY
 
+       *kernel_split_sd(sd, inactive_ray) = *kernel_split_sd(sd, ray_index);
+       for(int i = 0; i < kernel_split_sd(sd, ray_index)->num_closure; i++) {
+               kernel_split_sd(sd, inactive_ray)->closure[i] = kernel_split_sd(sd, ray_index)->closure[i];
+       }
+
        kernel_split_state.branched_state[inactive_ray].shared_sample_count = 0;
        kernel_split_state.branched_state[inactive_ray].original_ray = ray_index;
        kernel_split_state.branched_state[inactive_ray].waiting_on_shared_samples = false;
index bb6bf1c..75a0af7 100644 (file)
@@ -145,7 +145,7 @@ ccl_device void kernel_next_iteration_setup(KernelGlobals *kg,
                        if(kernel_split_branched_path_surface_indirect_light_iter(kg,
                                                                                  ray_index,
                                                                                  1.0f,
-                                                                                 &kernel_split_state.branched_state[ray_index].sd,
+                                                                                 kernel_split_sd(branched_state_sd, ray_index),
                                                                                  true,
                                                                                  true))
                        {
@@ -190,7 +190,7 @@ ccl_device void kernel_next_iteration_setup(KernelGlobals *kg,
                if(kernel_split_branched_path_surface_indirect_light_iter(kg,
                                                                          ray_index,
                                                                          1.0f,
-                                                                         &kernel_split_state.branched_state[ray_index].sd,
+                                                                         kernel_split_sd(branched_state_sd, ray_index),
                                                                          true,
                                                                          true))
                {
index fa2f0b2..9297e1e 100644 (file)
@@ -34,7 +34,7 @@ ccl_device_inline uint64_t split_data_buffer_size(KernelGlobals *kg, size_t num_
        uint64_t closure_size = sizeof(ShaderClosure) * (kernel_data.integrator.max_closures-1);
 
 #ifdef __BRANCHED_PATH__
-       size += align_up(closure_size * num_elements, 16);
+       size += align_up(num_elements * (sizeof(ShaderData) + closure_size), 16);
 #endif
 
        size += align_up(num_elements * (sizeof(ShaderData) + closure_size), 16);
@@ -60,7 +60,8 @@ ccl_device_inline void split_data_init(KernelGlobals *kg,
        uint64_t closure_size = sizeof(ShaderClosure) * (kernel_data.integrator.max_closures-1);
 
 #ifdef __BRANCHED_PATH__
-       p += align_up(closure_size * num_elements, 16);
+       split_data->_branched_state_sd = (ShaderData*)p;
+       p += align_up(num_elements * (sizeof(ShaderData) + closure_size), 16);
 #endif
 
        split_data->_sd = (ShaderData*)p;
index 9ac3f90..5f40fdc 100644 (file)
@@ -76,13 +76,11 @@ typedef ccl_global struct SplitBranchedState {
        int shared_sample_count; /* number of branched samples shared with other threads */
        int original_ray; /* index of original ray when sharing branched samples */
        bool waiting_on_shared_samples;
-
-       /* Must be last in to allow for dynamic size of closures */
-       struct ShaderData sd;
 } SplitBranchedState;
 
 #define SPLIT_DATA_BRANCHED_ENTRIES \
-       SPLIT_DATA_ENTRY( SplitBranchedState, branched_state, 1)
+       SPLIT_DATA_ENTRY( SplitBranchedState, branched_state, 1) \
+       SPLIT_DATA_ENTRY(ShaderData, _branched_state_sd, 0)
 #else
 #define SPLIT_DATA_BRANCHED_ENTRIES
 #endif  /* __BRANCHED_PATH__ */
index 887c3e3..5bf7483 100644 (file)
@@ -37,7 +37,7 @@ ccl_device_noinline bool kernel_split_branched_path_subsurface_indirect_light_it
 {
        SplitBranchedState *branched_state = &kernel_split_state.branched_state[ray_index];
 
-       ShaderData *sd = &branched_state->sd;
+       ShaderData *sd = kernel_split_sd(branched_state_sd, ray_index);
        PathRadiance *L = &kernel_split_state.path_radiance[ray_index];
        ShaderData *emission_sd = AS_SHADER_DATA(&kernel_split_state.sd_DL_shadow[ray_index]);