Cr8-xyz/blender
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Cycles: reduce closure memory usage for emission/shadow shader data.

Browse Source 
With a Titan Xp, reduces path trace local memory from 1092MB to 840MB.
Benchmark performance was within 1% with both RX 480 and Titan Xp.

Original patch was implemented by Sergey.

Differential Revision: https://developer.blender.org/D2249
This commit is contained in:
Brecht Van Lommel
2017-11-01 21:02:28 +01:00
parent c571be4e05
commit 8a72be7697
20 changed files with 80 additions and 61 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
intern/cycles/kernel/closure/alloc.h
View File

@@ -20,17 +20,16 @@ ccl_device ShaderClosure *closure_alloc(ShaderData *sd, int size, ClosureType ty
{
kernel_assert(size <= sizeof(ShaderClosure));
int num_closure = sd->num_closure;
int num_closure_extra = sd->num_closure_extra;
if(num_closure + num_closure_extra >= MAX_CLOSURE)
if(sd->num_closure_left == 0)
return NULL;
ShaderClosure *sc = &sd->closure[num_closure];
ShaderClosure *sc = &sd->closure[sd->num_closure];
sc->type = type;
sc->weight = weight;
sd->num_closure++;
sd->num_closure_left--;
return sc;
}
@@ -44,18 +43,16 @@ ccl_device ccl_addr_space void *closure_alloc_extra(ShaderData *sd, int size)
* This lets us keep the same fast array iteration over closures, as we
* found linked list iteration and iteration with skipping to be slower. */
int num_extra = ((size + sizeof(ShaderClosure) - 1) / sizeof(ShaderClosure));
int num_closure = sd->num_closure;
int num_closure_extra = sd->num_closure_extra + num_extra;
if(num_closure + num_closure_extra > MAX_CLOSURE) {
if(num_extra > sd->num_closure_left) {
/* Remove previous closure. */
sd->num_closure--;
sd->num_closure_extra++;
sd->num_closure_left++;
return NULL;
}
sd->num_closure_extra = num_closure_extra;
return (ccl_addr_space void*)(sd->closure + MAX_CLOSURE - num_closure_extra);
sd->num_closure_left -= num_extra;
return (ccl_addr_space void*)(sd->closure + sd->num_closure + sd->num_closure_left);
}
ccl_device_inline ShaderClosure *bsdf_alloc(ShaderData *sd, int size, float3 weight)