sandey/blender
1
0
Fork 0
forked from bartvdbraak/blender
Code Pull Requests Activity

Code refactor: use KernelShader and KernelParticle instead of float arrays.

Browse Source 
Original patch by Stefan with modifications by Brecht.
This commit is contained in:
Stefan Werner 2018-03-08 00:35:24 +01:00 committed by Brecht Van Lommel
parent fa9175ff02
commit f3010e98c3
12 changed files with 72 additions and 72 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
intern/cycles/kernel/bvh/bvh_shadow_all.h
View File

@ -276,7 +276,7 @@ bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
shader = __float_as_int(str.z); shader = __float_as_int(str.z);
} }
#endif #endif
int flag = kernel_tex_fetch(__shader_flag, (shader & SHADER_MASK)*SHADER_SIZE); int flag = kernel_tex_fetch(__shaders, (shader & SHADER_MASK)).flags;
/* if no transparent shadows, all light is blocked */ /* if no transparent shadows, all light is blocked */
if(!(flag & SD_HAS_TRANSPARENT_SHADOW)) { if(!(flag & SD_HAS_TRANSPARENT_SHADOW)) {

2
intern/cycles/kernel/bvh/qbvh_shadow_all.h
View File

@ -358,7 +358,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
shader = __float_as_int(str.z); shader = __float_as_int(str.z);
} }
#endif #endif
int flag = kernel_tex_fetch(__shader_flag, (shader & SHADER_MASK)*SHADER_SIZE); int flag = kernel_tex_fetch(__shaders, (shader & SHADER_MASK)).flags;
/* if no transparent shadows, all light is blocked */ /* if no transparent shadows, all light is blocked */
if(!(flag & SD_HAS_TRANSPARENT_SHADOW)) { if(!(flag & SD_HAS_TRANSPARENT_SHADOW)) {

36
intern/cycles/kernel/geom/geom_object.h
View File

@ -322,67 +322,49 @@ ccl_device_inline uint object_patch_map_offset(KernelGlobals *kg, int object)
ccl_device int shader_pass_id(KernelGlobals *kg, const ShaderData *sd) ccl_device int shader_pass_id(KernelGlobals *kg, const ShaderData *sd)
{ {
return kernel_tex_fetch(__shader_flag, (sd->shader & SHADER_MASK)*SHADER_SIZE + 1); return kernel_tex_fetch(__shaders, (sd->shader & SHADER_MASK)).pass_id;
} }
/* Particle data from which object was instanced */ /* Particle data from which object was instanced */
ccl_device_inline uint particle_index(KernelGlobals *kg, int particle) ccl_device_inline uint particle_index(KernelGlobals *kg, int particle)
{ {
int offset = particle*PARTICLE_SIZE; return kernel_tex_fetch(__particles, particle).index;
float4 f = kernel_tex_fetch(__particles, offset + 0);
return __float_as_uint(f.x);
} }
ccl_device float particle_age(KernelGlobals *kg, int particle) ccl_device float particle_age(KernelGlobals *kg, int particle)
{ {
int offset = particle*PARTICLE_SIZE; return kernel_tex_fetch(__particles, particle).age;
float4 f = kernel_tex_fetch(__particles, offset + 0);
return f.y;
} }
ccl_device float particle_lifetime(KernelGlobals *kg, int particle) ccl_device float particle_lifetime(KernelGlobals *kg, int particle)
{ {
int offset = particle*PARTICLE_SIZE; return kernel_tex_fetch(__particles, particle).lifetime;
float4 f = kernel_tex_fetch(__particles, offset + 0);
return f.z;
} }
ccl_device float particle_size(KernelGlobals *kg, int particle) ccl_device float particle_size(KernelGlobals *kg, int particle)
{ {
int offset = particle*PARTICLE_SIZE; return kernel_tex_fetch(__particles, particle).size;
float4 f = kernel_tex_fetch(__particles, offset + 0);
return f.w;
} }
ccl_device float4 particle_rotation(KernelGlobals *kg, int particle) ccl_device float4 particle_rotation(KernelGlobals *kg, int particle)
{ {
int offset = particle*PARTICLE_SIZE; return kernel_tex_fetch(__particles, particle).rotation;
float4 f = kernel_tex_fetch(__particles, offset + 1);
return f;
} }
ccl_device float3 particle_location(KernelGlobals *kg, int particle) ccl_device float3 particle_location(KernelGlobals *kg, int particle)
{ {
int offset = particle*PARTICLE_SIZE; return float4_to_float3(kernel_tex_fetch(__particles, particle).location);
float4 f = kernel_tex_fetch(__particles, offset + 2);
return make_float3(f.x, f.y, f.z);
} }
ccl_device float3 particle_velocity(KernelGlobals *kg, int particle) ccl_device float3 particle_velocity(KernelGlobals *kg, int particle)
{ {
int offset = particle*PARTICLE_SIZE; return float4_to_float3(kernel_tex_fetch(__particles, particle).velocity);
float4 f2 = kernel_tex_fetch(__particles, offset + 2);
float4 f3 = kernel_tex_fetch(__particles, offset + 3);
return make_float3(f2.w, f3.x, f3.y);
} }
ccl_device float3 particle_angular_velocity(KernelGlobals *kg, int particle) ccl_device float3 particle_angular_velocity(KernelGlobals *kg, int particle)
{ {
int offset = particle*PARTICLE_SIZE; return float4_to_float3(kernel_tex_fetch(__particles, particle).angular_velocity);
float4 f3 = kernel_tex_fetch(__particles, offset + 3);
float4 f4 = kernel_tex_fetch(__particles, offset + 4);
return make_float3(f3.z, f3.w, f4.x);
} }
/* Object intersection in BVH */ /* Object intersection in BVH */

8
intern/cycles/kernel/kernel_emission.h
View File

@ -29,7 +29,7 @@ ccl_device_noinline float3 direct_emissive_eval(KernelGlobals *kg,
/* setup shading at emitter */ /* setup shading at emitter */
float3 eval; float3 eval;
int shader_flag = kernel_tex_fetch(__shader_flag, (ls->shader & SHADER_MASK)*SHADER_SIZE); int shader_flag = kernel_tex_fetch(__shaders, (ls->shader & SHADER_MASK)).flags;
#ifdef __BACKGROUND_MIS__ #ifdef __BACKGROUND_MIS__
if(ls->type == LIGHT_BACKGROUND) { if(ls->type == LIGHT_BACKGROUND) {
@ -51,9 +51,9 @@ ccl_device_noinline float3 direct_emissive_eval(KernelGlobals *kg,
#endif #endif
if(shader_flag & SD_HAS_CONSTANT_EMISSION) if(shader_flag & SD_HAS_CONSTANT_EMISSION)
{ {
eval.x = __int_as_float(kernel_tex_fetch(__shader_flag, (ls->shader & SHADER_MASK)*SHADER_SIZE + 2)); eval.x = kernel_tex_fetch(__shaders, (ls->shader & SHADER_MASK)).constant_emission[0];
eval.y = __int_as_float(kernel_tex_fetch(__shader_flag, (ls->shader & SHADER_MASK)*SHADER_SIZE + 3)); eval.y = kernel_tex_fetch(__shaders, (ls->shader & SHADER_MASK)).constant_emission[1];
eval.z = __int_as_float(kernel_tex_fetch(__shader_flag, (ls->shader & SHADER_MASK)*SHADER_SIZE + 4)); eval.z = kernel_tex_fetch(__shaders, (ls->shader & SHADER_MASK)).constant_emission[2];
if((ls->prim != PRIM_NONE) && dot(ls->Ng, I) < 0.0f) { if((ls->prim != PRIM_NONE) && dot(ls->Ng, I) < 0.0f) {
ls->Ng = -ls->Ng; ls->Ng = -ls->Ng;
} }

12
intern/cycles/kernel/kernel_shader.h
View File

@ -114,7 +114,7 @@ ccl_device_noinline void shader_setup_from_ray(KernelGlobals *kg,
sd->I = -ray->D; sd->I = -ray->D;
sd->flag |= kernel_tex_fetch(__shader_flag, (sd->shader & SHADER_MASK)*SHADER_SIZE); sd->flag |= kernel_tex_fetch(__shaders, (sd->shader & SHADER_MASK)).flags;
#ifdef __INSTANCING__ #ifdef __INSTANCING__
if(isect->object != OBJECT_NONE) { if(isect->object != OBJECT_NONE) {
@ -199,7 +199,7 @@ void shader_setup_from_subsurface(
motion_triangle_shader_setup(kg, sd, isect, ray, true); motion_triangle_shader_setup(kg, sd, isect, ray, true);
} }
sd->flag |= kernel_tex_fetch(__shader_flag, (sd->shader & SHADER_MASK)*SHADER_SIZE); sd->flag |= kernel_tex_fetch(__shaders, (sd->shader & SHADER_MASK)).flags;
# ifdef __INSTANCING__ # ifdef __INSTANCING__
if(isect->object != OBJECT_NONE) { if(isect->object != OBJECT_NONE) {
@ -276,7 +276,7 @@ ccl_device_inline void shader_setup_from_sample(KernelGlobals *kg,
sd->time = time; sd->time = time;
sd->ray_length = t; sd->ray_length = t;
sd->flag = kernel_tex_fetch(__shader_flag, (sd->shader & SHADER_MASK)*SHADER_SIZE); sd->flag = kernel_tex_fetch(__shaders, (sd->shader & SHADER_MASK)).flags;
sd->object_flag = 0; sd->object_flag = 0;
if(sd->object != OBJECT_NONE) { if(sd->object != OBJECT_NONE) {
sd->object_flag |= kernel_tex_fetch(__object_flag, sd->object_flag |= kernel_tex_fetch(__object_flag,
@ -386,7 +386,7 @@ ccl_device_inline void shader_setup_from_background(KernelGlobals *kg, ShaderDat
sd->Ng = -ray->D; sd->Ng = -ray->D;
sd->I = -ray->D; sd->I = -ray->D;
sd->shader = kernel_data.background.surface_shader; sd->shader = kernel_data.background.surface_shader;
sd->flag = kernel_tex_fetch(__shader_flag, (sd->shader & SHADER_MASK)*SHADER_SIZE); sd->flag = kernel_tex_fetch(__shaders, (sd->shader & SHADER_MASK)).flags;
sd->object_flag = 0; sd->object_flag = 0;
sd->time = ray->time; sd->time = ray->time;
sd->ray_length = 0.0f; sd->ray_length = 0.0f;
@ -1181,7 +1181,7 @@ ccl_device_inline void shader_eval_volume(KernelGlobals *kg,
sd->shader = stack[i].shader; sd->shader = stack[i].shader;
sd->flag &= ~SD_SHADER_FLAGS; sd->flag &= ~SD_SHADER_FLAGS;
sd->flag |= kernel_tex_fetch(__shader_flag, (sd->shader & SHADER_MASK)*SHADER_SIZE); sd->flag |= kernel_tex_fetch(__shaders, (sd->shader & SHADER_MASK)).flags;
sd->object_flag &= ~SD_OBJECT_FLAGS; sd->object_flag &= ~SD_OBJECT_FLAGS;
if(sd->object != OBJECT_NONE) { if(sd->object != OBJECT_NONE) {
@ -1254,7 +1254,7 @@ ccl_device bool shader_transparent_shadow(KernelGlobals *kg, Intersection *isect
shader = __float_as_int(str.z); shader = __float_as_int(str.z);
} }
#endif #endif
int flag = kernel_tex_fetch(__shader_flag, (shader & SHADER_MASK)*SHADER_SIZE); int flag = kernel_tex_fetch(__shaders, (shader & SHADER_MASK)).flags;
return (flag & SD_HAS_TRANSPARENT_SHADOW) != 0; return (flag & SD_HAS_TRANSPARENT_SHADOW) != 0;
} }

4
intern/cycles/kernel/kernel_textures.h
View File

@ -61,11 +61,11 @@ KERNEL_TEX(float2, __light_background_marginal_cdf)
KERNEL_TEX(float2, __light_background_conditional_cdf) KERNEL_TEX(float2, __light_background_conditional_cdf)
/* particles */ /* particles */
KERNEL_TEX(float4, __particles) KERNEL_TEX(KernelParticle, __particles)
/* shaders */ /* shaders */
KERNEL_TEX(uint4, __svm_nodes) KERNEL_TEX(uint4, __svm_nodes)
KERNEL_TEX(uint, __shader_flag) KERNEL_TEX(KernelShader, __shaders)
KERNEL_TEX(uint, __object_flag) KERNEL_TEX(uint, __object_flag)
/* lookup tables */ /* lookup tables */

27
intern/cycles/kernel/kernel_types.h
View File

@ -39,8 +39,6 @@ CCL_NAMESPACE_BEGIN
#define FILTER_TABLE_SIZE 1024 #define FILTER_TABLE_SIZE 1024
#define RAMP_TABLE_SIZE 256 #define RAMP_TABLE_SIZE 256
#define SHUTTER_TABLE_SIZE 256 #define SHUTTER_TABLE_SIZE 256
#define PARTICLE_SIZE 5
#define SHADER_SIZE 5
#define BSSRDF_MIN_RADIUS 1e-8f #define BSSRDF_MIN_RADIUS 1e-8f
#define BSSRDF_MAX_HITS 4 #define BSSRDF_MAX_HITS 4
@ -923,7 +921,7 @@ enum ShaderDataFlag {
SD_HAS_BUMP = (1 << 25), SD_HAS_BUMP = (1 << 25),
/* Has true displacement. */ /* Has true displacement. */
SD_HAS_DISPLACEMENT = (1 << 26), SD_HAS_DISPLACEMENT = (1 << 26),
/* Has constant emission (value stored in __shader_flag) */ /* Has constant emission (value stored in __shaders) */
SD_HAS_CONSTANT_EMISSION = (1 << 27), SD_HAS_CONSTANT_EMISSION = (1 << 27),
/* Needs to access attributes */ /* Needs to access attributes */
SD_NEED_ATTRIBUTES = (1 << 28), SD_NEED_ATTRIBUTES = (1 << 28),
@ -1511,6 +1509,29 @@ typedef struct KernelLightDistribution {
} KernelLightDistribution; } KernelLightDistribution;
static_assert_align(KernelLightDistribution, 16); static_assert_align(KernelLightDistribution, 16);
typedef struct KernelParticle {
int index;
float age;
float lifetime;
float size;
float4 rotation;
/* Only xyz are used of the following. float4 instead of float3 are used
* to ensure consistent padding/alignment across devices. */
float4 location;
float4 velocity;
float4 angular_velocity;
} KernelParticle;
static_assert_align(KernelParticle, 16);
typedef struct KernelShader {
float constant_emission[3];
float pad1;
int flags;
int pass_id;
int pad2, pad3;
} KernelShader;
static_assert_align(KernelShader, 16);
/* Declarations required for split kernel */ /* Declarations required for split kernel */
/* Macro for queues */ /* Macro for queues */

4
intern/cycles/kernel/kernel_volume.h
View File

@ -104,7 +104,7 @@ ccl_device float kernel_volume_channel_get(float3 value, int channel)
ccl_device bool volume_stack_is_heterogeneous(KernelGlobals *kg, ccl_addr_space VolumeStack *stack) ccl_device bool volume_stack_is_heterogeneous(KernelGlobals *kg, ccl_addr_space VolumeStack *stack)
{ {
for(int i = 0; stack[i].shader != SHADER_NONE; i++) { for(int i = 0; stack[i].shader != SHADER_NONE; i++) {
int shader_flag = kernel_tex_fetch(__shader_flag, (stack[i].shader & SHADER_MASK)*SHADER_SIZE); int shader_flag = kernel_tex_fetch(__shaders, (stack[i].shader & SHADER_MASK)).flags;
if(shader_flag & SD_HETEROGENEOUS_VOLUME) { if(shader_flag & SD_HETEROGENEOUS_VOLUME) {
return true; return true;
@ -134,7 +134,7 @@ ccl_device int volume_stack_sampling_method(KernelGlobals *kg, VolumeStack *stac
int method = -1; int method = -1;
for(int i = 0; stack[i].shader != SHADER_NONE; i++) { for(int i = 0; stack[i].shader != SHADER_NONE; i++) {
int shader_flag = kernel_tex_fetch(__shader_flag, (stack[i].shader & SHADER_MASK)*SHADER_SIZE); int shader_flag = kernel_tex_fetch(__shaders, (stack[i].shader & SHADER_MASK)).flags;
if(shader_flag & SD_VOLUME_MIS) { if(shader_flag & SD_VOLUME_MIS) {
return SD_VOLUME_MIS; return SD_VOLUME_MIS;

22
intern/cycles/render/particles.cpp
View File

@ -62,14 +62,10 @@ void ParticleSystemManager::device_update_particles(Device *, DeviceScene *dscen
for(size_t j = 0; j < scene->particle_systems.size(); j++) for(size_t j = 0; j < scene->particle_systems.size(); j++)
num_particles += scene->particle_systems[j]->particles.size(); num_particles += scene->particle_systems[j]->particles.size();
float4 *particles = dscene->particles.alloc(PARTICLE_SIZE*num_particles); KernelParticle *kparticles = dscene->particles.alloc(num_particles);
/* dummy particle */ /* dummy particle */
particles[0] = make_float4(0.0f, 0.0f, 0.0f, 0.0f); memset(kparticles, 0, sizeof(KernelParticle));
particles[1] = make_float4(0.0f, 0.0f, 0.0f, 0.0f);
particles[2] = make_float4(0.0f, 0.0f, 0.0f, 0.0f);
particles[3] = make_float4(0.0f, 0.0f, 0.0f, 0.0f);
particles[4] = make_float4(0.0f, 0.0f, 0.0f, 0.0f);
int i = 1; int i = 1;
for(size_t j = 0; j < scene->particle_systems.size(); j++) { for(size_t j = 0; j < scene->particle_systems.size(); j++) {
@ -78,13 +74,15 @@ void ParticleSystemManager::device_update_particles(Device *, DeviceScene *dscen
for(size_t k = 0; k < psys->particles.size(); k++) { for(size_t k = 0; k < psys->particles.size(); k++) {
/* pack in texture */ /* pack in texture */
Particle& pa = psys->particles[k]; Particle& pa = psys->particles[k];
int offset = i*PARTICLE_SIZE;
particles[offset] = make_float4(__uint_as_float(pa.index), pa.age, pa.lifetime, pa.size); kparticles[i].index = pa.index;
particles[offset+1] = pa.rotation; kparticles[i].age = pa.age;
particles[offset+2] = make_float4(pa.location.x, pa.location.y, pa.location.z, pa.velocity.x); kparticles[i].lifetime = pa.lifetime;
particles[offset+3] = make_float4(pa.velocity.y, pa.velocity.z, pa.angular_velocity.x, pa.angular_velocity.y); kparticles[i].size = pa.size;
particles[offset+4] = make_float4(pa.angular_velocity.z, 0.0f, 0.0f, 0.0f); kparticles[i].rotation = pa.rotation;
kparticles[i].location = float3_to_float4(pa.location);
kparticles[i].velocity = float3_to_float4(pa.velocity);
kparticles[i].angular_velocity = float3_to_float4(pa.angular_velocity);
i++; i++;

2
intern/cycles/render/scene.cpp
View File

@ -71,7 +71,7 @@ DeviceScene::DeviceScene(Device *device)
light_background_conditional_cdf(device, "__light_background_conditional_cdf", MEM_TEXTURE), light_background_conditional_cdf(device, "__light_background_conditional_cdf", MEM_TEXTURE),
particles(device, "__particles", MEM_TEXTURE), particles(device, "__particles", MEM_TEXTURE),
svm_nodes(device, "__svm_nodes", MEM_TEXTURE), svm_nodes(device, "__svm_nodes", MEM_TEXTURE),
shader_flag(device, "__shader_flag", MEM_TEXTURE), shaders(device, "__shaders", MEM_TEXTURE),
object_flag(device, "__object_flag", MEM_TEXTURE), object_flag(device, "__object_flag", MEM_TEXTURE),
lookup_table(device, "__lookup_table", MEM_TEXTURE), lookup_table(device, "__lookup_table", MEM_TEXTURE),
sobol_directions(device, "__sobol_directions", MEM_TEXTURE) sobol_directions(device, "__sobol_directions", MEM_TEXTURE)

4
intern/cycles/render/scene.h
View File

@ -102,11 +102,11 @@ public:
device_vector<float2> light_background_conditional_cdf; device_vector<float2> light_background_conditional_cdf;
/* particles */ /* particles */
device_vector<float4> particles; device_vector<KernelParticle> particles;
/* shaders */ /* shaders */
device_vector<int4> svm_nodes; device_vector<int4> svm_nodes;
device_vector<uint> shader_flag; device_vector<KernelShader> shaders;
device_vector<uint> object_flag; device_vector<uint> object_flag;
/* lookup tables */ /* lookup tables */

21
intern/cycles/render/shader.cpp
View File

@ -432,14 +432,12 @@ void ShaderManager::device_update_common(Device *device,
Scene *scene, Scene *scene,
Progress& /*progress*/) Progress& /*progress*/)
{ {
dscene->shader_flag.free(); dscene->shaders.free();
if(scene->shaders.size() == 0) if(scene->shaders.size() == 0)
return; return;
uint shader_flag_size = scene->shaders.size()*SHADER_SIZE; KernelShader *kshader = dscene->shaders.alloc(scene->shaders.size());
uint *shader_flag = dscene->shader_flag.alloc(shader_flag_size);
uint i = 0;
bool has_volumes = false; bool has_volumes = false;
bool has_transparent_shadow = false; bool has_transparent_shadow = false;
@ -487,16 +485,17 @@ void ShaderManager::device_update_common(Device *device,
flag |= SD_HAS_CONSTANT_EMISSION; flag |= SD_HAS_CONSTANT_EMISSION;
/* regular shader */ /* regular shader */
shader_flag[i++] = flag; kshader->flags = flag;
shader_flag[i++] = shader->pass_id; kshader->pass_id = shader->pass_id;
shader_flag[i++] = __float_as_int(constant_emission.x); kshader->constant_emission[0] = constant_emission.x;
shader_flag[i++] = __float_as_int(constant_emission.y); kshader->constant_emission[1] = constant_emission.y;
shader_flag[i++] = __float_as_int(constant_emission.z); kshader->constant_emission[2] = constant_emission.z;
kshader++;
has_transparent_shadow |= (flag & SD_HAS_TRANSPARENT_SHADOW) != 0; has_transparent_shadow |= (flag & SD_HAS_TRANSPARENT_SHADOW) != 0;
} }
dscene->shader_flag.copy_to_device(); dscene->shaders.copy_to_device();
/* lookup tables */ /* lookup tables */
KernelTables *ktables = &dscene->data.tables; KernelTables *ktables = &dscene->data.tables;
@ -525,7 +524,7 @@ void ShaderManager::device_free_common(Device *, DeviceScene *dscene, Scene *sce
{ {
scene->lookup_tables->remove_table(&beckmann_table_offset); scene->lookup_tables->remove_table(&beckmann_table_offset);
dscene->shader_flag.free(); dscene->shaders.free();
} }
void ShaderManager::add_default(Scene *scene) void ShaderManager::add_default(Scene *scene)