[Bf-blender-cvs] SVN commit: /data/svn/bf-blender [57317] trunk/blender/intern/cycles/kernel : Fix #35665: cycles CUDA crash after recent changes.
Brecht Van Lommel
brechtvanlommel at pandora.be
Sun Jun 9 18:37:05 CEST 2013
Revision: 57317
http://projects.blender.org/scm/viewvc.php?view=rev&root=bf-blender&revision=57317
Author: blendix
Date: 2013-06-09 16:37:04 +0000 (Sun, 09 Jun 2013)
Log Message:
-----------
Fix #35665: cycles CUDA crash after recent changes. This works around a compiler
bug in CUDA 4.2 (solved in 5.5) with typedef'd function parameters.
Modified Paths:
--------------
trunk/blender/intern/cycles/kernel/kernel_path.h
trunk/blender/intern/cycles/kernel/kernel_random.h
Modified: trunk/blender/intern/cycles/kernel/kernel_path.h
===================================================================
--- trunk/blender/intern/cycles/kernel/kernel_path.h 2013-06-09 16:18:23 UTC (rev 57316)
+++ trunk/blender/intern/cycles/kernel/kernel_path.h 2013-06-09 16:37:04 UTC (rev 57317)
@@ -233,7 +233,7 @@
return result;
}
-__device float4 kernel_path_progressive(KernelGlobals *kg, RNG rng, int sample, Ray ray, __global float *buffer)
+__device float4 kernel_path_progressive(KernelGlobals *kg, RNG *rng, int sample, Ray ray, __global float *buffer)
{
/* initialize */
PathRadiance L;
@@ -271,7 +271,7 @@
}
extmax = kernel_data.curve_kernel_data.maximum_width;
- lcg_state = lcg_init(rng + rng_offset + sample*0x51633e2d);
+ lcg_state = lcg_init(*rng + rng_offset + sample*0x51633e2d);
}
bool hit = scene_intersect(kg, &ray, visibility, &isect, &lcg_state, difl, extmax);
@@ -399,7 +399,7 @@
/* do bssrdf scatter step if we picked a bssrdf closure */
if(sc) {
- uint lcg_state = lcg_init(rng + rng_offset + sample*0x68bc21eb);
+ uint lcg_state = lcg_init(*rng + rng_offset + sample*0x68bc21eb);
subsurface_scatter_step(kg, &sd, state.flag, sc, &lcg_state, false);
}
}
@@ -538,7 +538,7 @@
#ifdef __NON_PROGRESSIVE__
-__device void kernel_path_indirect(KernelGlobals *kg, RNG rng, int sample, Ray ray, __global float *buffer,
+__device void kernel_path_indirect(KernelGlobals *kg, RNG *rng, int sample, Ray ray, __global float *buffer,
float3 throughput, int num_samples, int num_total_samples,
float min_ray_pdf, float ray_pdf, PathState state, int rng_offset, PathRadiance *L)
{
@@ -644,7 +644,7 @@
/* do bssrdf scatter step if we picked a bssrdf closure */
if(sc) {
- uint lcg_state = lcg_init(rng + rng_offset + sample*0x68bc21eb);
+ uint lcg_state = lcg_init(*rng + rng_offset + sample*0x68bc21eb);
subsurface_scatter_step(kg, &sd, state.flag, sc, &lcg_state, false);
}
}
@@ -767,7 +767,7 @@
}
}
-__device_noinline void kernel_path_non_progressive_lighting(KernelGlobals *kg, RNG rng, int sample,
+__device_noinline void kernel_path_non_progressive_lighting(KernelGlobals *kg, RNG *rng, int sample,
ShaderData *sd, float3 throughput, float num_samples_adjust,
float min_ray_pdf, float ray_pdf, PathState state,
int rng_offset, PathRadiance *L, __global float *buffer)
@@ -830,13 +830,13 @@
for(int i = 0; i < kernel_data.integrator.num_all_lights; i++) {
int num_samples = ceil_to_int(num_samples_adjust*light_select_num_samples(kg, i));
float num_samples_inv = num_samples_adjust/(num_samples*kernel_data.integrator.num_all_lights);
- RNG lamp_rng = cmj_hash(rng, i);
+ RNG lamp_rng = cmj_hash(*rng, i);
if(kernel_data.integrator.pdf_triangles != 0.0f)
num_samples_inv *= 0.5f;
for(int j = 0; j < num_samples; j++) {
- float2 light_uv = path_rng_2D(kg, lamp_rng, sample*num_samples + j, aa_samples*num_samples, rng_offset + PRNG_LIGHT_U);
+ float2 light_uv = path_rng_2D(kg, &lamp_rng, sample*num_samples + j, aa_samples*num_samples, rng_offset + PRNG_LIGHT_U);
float light_u = light_uv.x;
float light_v = light_uv.y;
@@ -905,7 +905,7 @@
num_samples = ceil_to_int(num_samples_adjust*num_samples);
float num_samples_inv = num_samples_adjust/num_samples;
- RNG bsdf_rng = cmj_hash(rng, i);
+ RNG bsdf_rng = cmj_hash(*rng, i);
for(int j = 0; j < num_samples; j++) {
/* sample BSDF */
@@ -913,7 +913,7 @@
BsdfEval bsdf_eval;
float3 bsdf_omega_in;
differential3 bsdf_domega_in;
- float2 bsdf_uv = path_rng_2D(kg, bsdf_rng, sample*num_samples + j, aa_samples*num_samples, rng_offset + PRNG_BSDF_U);
+ float2 bsdf_uv = path_rng_2D(kg, &bsdf_rng, sample*num_samples + j, aa_samples*num_samples, rng_offset + PRNG_BSDF_U);
float bsdf_u = bsdf_uv.x;
float bsdf_v = bsdf_uv.y;
int label;
@@ -964,7 +964,7 @@
}
}
-__device float4 kernel_path_non_progressive(KernelGlobals *kg, RNG rng, int sample, Ray ray, __global float *buffer)
+__device float4 kernel_path_non_progressive(KernelGlobals *kg, RNG *rng, int sample, Ray ray, __global float *buffer)
{
/* initialize */
PathRadiance L;
@@ -997,7 +997,7 @@
}
extmax = kernel_data.curve_kernel_data.maximum_width;
- lcg_state = lcg_init(rng + rng_offset + sample*0x51633e2d);
+ lcg_state = lcg_init(*rng + rng_offset + sample*0x51633e2d);
}
if(!scene_intersect(kg, &ray, visibility, &isect, &lcg_state, difl, extmax)) {
@@ -1090,7 +1090,7 @@
continue;
/* set up random number generator */
- uint lcg_state = lcg_init(rng + rng_offset + sample*0x68bc21eb);
+ uint lcg_state = lcg_init(*rng + rng_offset + sample*0x68bc21eb);
int num_samples = kernel_data.integrator.subsurface_samples;
float num_samples_inv = 1.0f/num_samples;
@@ -1163,7 +1163,7 @@
float lens_u = 0.0f, lens_v = 0.0f;
if(kernel_data.cam.aperturesize > 0.0f) {
- float2 lens_uv = path_rng_2D(kg, rng, sample, num_samples, PRNG_LENS_U);
+ float2 lens_uv = path_rng_2D(kg, &rng, sample, num_samples, PRNG_LENS_U);
lens_u = lens_uv.x;
lens_v = lens_uv.y;
}
@@ -1172,7 +1172,7 @@
#ifdef __CAMERA_MOTION__
if(kernel_data.cam.shuttertime != -1.0f)
- time = path_rng_1D(kg, rng, sample, num_samples, PRNG_TIME);
+ time = path_rng_1D(kg, &rng, sample, num_samples, PRNG_TIME);
#endif
camera_sample(kg, x, y, filter_u, filter_v, lens_u, lens_v, time, &ray);
@@ -1184,10 +1184,10 @@
#ifdef __NON_PROGRESSIVE__
if(kernel_data.integrator.progressive)
#endif
- L = kernel_path_progressive(kg, rng, sample, ray, buffer);
+ L = kernel_path_progressive(kg, &rng, sample, ray, buffer);
#ifdef __NON_PROGRESSIVE__
else
- L = kernel_path_non_progressive(kg, rng, sample, ray, buffer);
+ L = kernel_path_non_progressive(kg, &rng, sample, ray, buffer);
#endif
}
else
Modified: trunk/blender/intern/cycles/kernel/kernel_random.h
===================================================================
--- trunk/blender/intern/cycles/kernel/kernel_random.h 2013-06-09 16:18:23 UTC (rev 57316)
+++ trunk/blender/intern/cycles/kernel/kernel_random.h 2013-06-09 16:37:04 UTC (rev 57317)
@@ -102,10 +102,10 @@
return index;
}
-__device_inline float path_rng(KernelGlobals *kg, RNG rng, int sample, int dimension)
+__device_inline float path_rng(KernelGlobals *kg, RNG *rng, int sample, int dimension)
{
#ifdef __SOBOL_FULL_SCREEN__
- uint result = sobol_dimension(kg, rng, dimension);
+ uint result = sobol_dimension(kg, *rng, dimension);
float r = (float)result * (1.0f/(float)0xFFFFFFFF);
return r;
#else
@@ -117,20 +117,20 @@
float shift;
if(dimension & 1)
- shift = (rng >> 16)*(1.0f/(float)0xFFFF);
+ shift = (*rng >> 16)*(1.0f/(float)0xFFFF);
else
- shift = (rng & 0xFFFF)*(1.0f/(float)0xFFFF);
+ shift = (*rng & 0xFFFF)*(1.0f/(float)0xFFFF);
return r + shift - floorf(r + shift);
#endif
}
-__device_inline float path_rng_1D(KernelGlobals *kg, RNG rng, int sample, int num_samples, int dimension)
+__device_inline float path_rng_1D(KernelGlobals *kg, RNG *rng, int sample, int num_samples, int dimension)
{
#ifdef __CMJ__
if(kernel_data.integrator.sampling_pattern == SAMPLING_PATTERN_CMJ) {
/* correlated multi-jittered */
- int p = rng + dimension;
+ int p = *rng + dimension;
return cmj_sample_1D(sample, num_samples, p);
}
#endif
@@ -139,12 +139,12 @@
return path_rng(kg, rng, sample, dimension);
}
-__device_inline float2 path_rng_2D(KernelGlobals *kg, RNG rng, int sample, int num_samples, int dimension)
+__device_inline float2 path_rng_2D(KernelGlobals *kg, RNG *rng, int sample, int num_samples, int dimension)
{
#ifdef __CMJ__
if(kernel_data.integrator.sampling_pattern == SAMPLING_PATTERN_CMJ) {
/* correlated multi-jittered */
- int p = rng + dimension;
+ int p = *rng + dimension;
return cmj_sample_2D(sample, num_samples, p);
}
#endif
@@ -184,7 +184,7 @@
*fy = 0.5f;
}
else {
- float2 fxy = path_rng_2D(kg, *rng, sample, num_samples, PRNG_FILTER_U);
+ float2 fxy = path_rng_2D(kg, rng, sample, num_samples, PRNG_FILTER_U);
*fx = fxy.x;
*fy = fxy.y;
More information about the Bf-blender-cvs
mailing list