griefith/test
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

EEVEE Next: Optimize closure light stack array access

Browse Source 
This PR ensures the ClosureLightStack is only accessed
with constant indices. This reduces compiler spill and allows
furhter constant folding, improving the deferred lighting
evaluation performance by 30%.

Authored by Apple: Michael Parkin-White

Pull Request: https://projects.blender.org/blender/blender/pulls/121124
This commit is contained in:
Jason Fielder
2024-05-10 19:13:46 +02:00
committed by Clément Foucault
parent 008483fe8a
commit aa994f3cba
3 changed files with 83 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files

69
source/blender/draw/engines/eevee_next/shaders/eevee_deferred_light_frag.glsl
View File

@@ -15,6 +15,34 @@
#pragma BLENDER_REQUIRE(eevee_subsurface_lib.glsl)
#pragma BLENDER_REQUIRE(eevee_thickness_lib.glsl)
void write_radiance_direct(int layer_index, ivec2 texel, vec3 radiance)
{
/* TODO(fclem): Layered texture. */
if (layer_index == 0) {
imageStore(direct_radiance_1_img, texel, vec4(radiance, 1.0));
}
else if (layer_index == 1) {
imageStore(direct_radiance_2_img, texel, vec4(radiance, 1.0));
}
else if (layer_index == 2) {
imageStore(direct_radiance_3_img, texel, vec4(radiance, 1.0));
}
}
void write_radiance_indirect(int layer_index, ivec2 texel, vec3 radiance)
{
/* TODO(fclem): Layered texture. */
if (layer_index == 0) {
imageStore(indirect_radiance_1_img, texel, vec4(radiance, 1.0));
}
else if (layer_index == 1) {
imageStore(indirect_radiance_2_img, texel, vec4(radiance, 1.0));
}
else if (layer_index == 2) {
imageStore(indirect_radiance_3_img, texel, vec4(radiance, 1.0));
}
}
void main()
{
ivec2 texel = ivec2(gl_FragCoord.xy);
@@ -33,8 +61,9 @@ void main()
float vPz = dot(drw_view_forward(), P) - dot(drw_view_forward(), drw_view_position());
ClosureLightStack stack;
/* Unroll light stack array assignments to avoid non-constant indexing. */
for (int i = 0; i < LIGHT_CLOSURE_EVAL_COUNT && i < gbuf.closure_count; i++) {
stack.cl[i] = closure_light_new(gbuffer_closure_get(gbuf, i), V);
closure_light_set(stack, i, closure_light_new(gbuffer_closure_get(gbuf, i), V));
}
/* TODO(fclem): If transmission (no SSS) is present, we could reduce LIGHT_CLOSURE_EVAL_COUNT
@@ -80,8 +109,8 @@ void main()
vec3 radiance_shadowed = vec3(0);
vec3 radiance_unshadowed = vec3(0);
for (int i = 0; i < LIGHT_CLOSURE_EVAL_COUNT && i < gbuf.closure_count; i++) {
radiance_shadowed += stack.cl[i].light_shadowed;
radiance_unshadowed += stack.cl[i].light_unshadowed;
radiance_shadowed += closure_light_get(stack, i).light_shadowed;
radiance_unshadowed += closure_light_get(stack, i).light_unshadowed;
}
/* TODO(fclem): Change shadow pass to be colored. */
vec3 shadows = radiance_shadowed * safe_rcp(radiance_unshadowed);
@@ -98,36 +127,22 @@ void main()
ClosureUndetermined cl = gbuffer_closure_get(gbuf, i);
vec3 indirect_light = lightprobe_eval(samp, cl, P, V, gbuf.thickness);
int layer_index = gbuffer_closure_get_bin_index(gbuf, i);
vec3 direct_light = closure_light_get(stack, i).light_shadowed;
if (use_split_indirect) {
int layer_index = gbuffer_closure_get_bin_index(gbuf, i);
/* TODO(fclem): Layered texture. */
if (layer_index == 0) {
imageStore(indirect_radiance_1_img, texel, vec4(indirect_light, 1.0));
}
else if (layer_index == 1) {
imageStore(indirect_radiance_2_img, texel, vec4(indirect_light, 1.0));
}
else if (layer_index == 2) {
imageStore(indirect_radiance_3_img, texel, vec4(indirect_light, 1.0));
}
write_radiance_indirect(layer_index, texel, indirect_light);
write_radiance_direct(layer_index, texel, direct_light);
}
else {
stack.cl[i].light_shadowed += indirect_light;
write_radiance_direct(layer_index, texel, direct_light + indirect_light);
}
}
}
for (int i = 0; i < LIGHT_CLOSURE_EVAL_COUNT && i < gbuf.closure_count; i++) {
int layer_index = gbuffer_closure_get_bin_index(gbuf, i);
/* TODO(fclem): Layered texture. */
if (layer_index == 0) {
imageStore(direct_radiance_1_img, texel, vec4(stack.cl[i].light_shadowed, 1.0));
}
else if (layer_index == 1) {
imageStore(direct_radiance_2_img, texel, vec4(stack.cl[i].light_shadowed, 1.0));
}
else if (layer_index == 2) {
imageStore(direct_radiance_3_img, texel, vec4(stack.cl[i].light_shadowed, 1.0));
else {
for (int i = 0; i < LIGHT_CLOSURE_EVAL_COUNT && i < gbuf.closure_count; i++) {
int layer_index = gbuffer_closure_get_bin_index(gbuf, i);
vec3 direct_light = closure_light_get(stack, i).light_shadowed;
write_radiance_direct(layer_index, texel, direct_light);
}
}
}

2
source/blender/draw/engines/eevee_next/shaders/eevee_forward_lib.glsl
View File

@@ -80,7 +80,7 @@ void forward_lighting_eval(float thickness, out vec3 radiance, out vec3 transmit
for (int i = 0; i < LIGHT_CLOSURE_EVAL_COUNT; i++) {
ClosureUndetermined cl = g_closure_get(i);
if (cl.weight > 1e-5) {
vec3 direct_light = stack.cl[i].light_shadowed;
vec3 direct_light = closure_light_get(stack, i).light_shadowed;
vec3 indirect_light = lightprobe_eval(samp, cl, g_data.P, V, thickness);
if ((cl.type == CLOSURE_BSDF_TRANSLUCENT_ID ||

40
source/blender/draw/engines/eevee_next/shaders/eevee_light_eval_lib.glsl
View File

@@ -47,6 +47,46 @@ struct ClosureLightStack {
ClosureLight cl[LIGHT_CLOSURE_EVAL_COUNT];
};
ClosureLight closure_light_get(ClosureLightStack stack, int index)
{
switch (index) {
case 0:
return stack.cl[0];
#if LIGHT_CLOSURE_EVAL_COUNT > 1
case 1:
return stack.cl[1];
#endif
#if LIGHT_CLOSURE_EVAL_COUNT > 2
case 2:
return stack.cl[2];
#endif
#if LIGHT_CLOSURE_EVAL_COUNT > 3
# error
#endif
}
ClosureLight closure_null;
return closure_null;
}
void closure_light_set(inout ClosureLightStack stack, int index, ClosureLight cl_light)
{
switch (index) {
case 0:
stack.cl[0] = cl_light;
#if LIGHT_CLOSURE_EVAL_COUNT > 1
case 1:
stack.cl[1] = cl_light;
#endif
#if LIGHT_CLOSURE_EVAL_COUNT > 2
case 2:
stack.cl[2] = cl_light;
#endif
#if LIGHT_CLOSURE_EVAL_COUNT > 3
# error
#endif
}
}
ClosureLight closure_light_new_ex(ClosureUndetermined cl,
vec3 V,
float thickness,