diff options
author | Sergey Sharybin <sergey.vfx@gmail.com> | 2015-04-27 22:13:03 +0300 |
---|---|---|
committer | Sergey Sharybin <sergey.vfx@gmail.com> | 2015-04-27 22:38:32 +0300 |
commit | ae7d84dbc1936ef7ddd00f9c22d074389f97f04f (patch) | |
tree | 7103596bc78785ed769e90c1570d654cc14a839d /intern/cycles/kernel/svm/svm_closure.h | |
parent | 278ff15c7f3c4a91ba049b7a13ecf8a4183ebf67 (diff) |
Cycles: Use native saturate function for CUDA
This more a workaround for CUDA optimizer which can't optimize clamp(x, 0, 1)
into a single instruction and uses 4 instructions instead.
Original patch by @lockal with own modification:
Don't make changes outside of the kernel. They don't make any difference
anyway and term saturate() has a bit different meaning outside of kernel.
This gives around 2% of speedup in Barcelona file, but in more complex shader
setups with lots of math nodes with clamping speedup could be much nicer.
Subscribers: dingto
Projects: #cycles
Differential Revision: https://developer.blender.org/D1224
Diffstat (limited to 'intern/cycles/kernel/svm/svm_closure.h')
-rw-r--r-- | intern/cycles/kernel/svm/svm_closure.h | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/intern/cycles/kernel/svm/svm_closure.h b/intern/cycles/kernel/svm/svm_closure.h index 8afdb73fb38..0d2d155f827 100644 --- a/intern/cycles/kernel/svm/svm_closure.h +++ b/intern/cycles/kernel/svm/svm_closure.h @@ -347,7 +347,7 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float * sc->N = N; /* sigma */ - sc->data0 = clamp(param1, 0.0f, 1.0f); + sc->data0 = saturate(param1); sc->data1 = 0.0f; sc->data2 = 0.0f; sd->flag |= bsdf_ashikhmin_velvet_setup(sc); @@ -655,7 +655,7 @@ ccl_device void svm_node_mix_closure(ShaderData *sd, float *stack, uint4 node) decode_node_uchar4(node.y, &weight_offset, &in_weight_offset, &weight1_offset, &weight2_offset); float weight = stack_load_float(stack, weight_offset); - weight = clamp(weight, 0.0f, 1.0f); + weight = saturate(weight); float in_weight = (stack_valid(in_weight_offset))? stack_load_float(stack, in_weight_offset): 1.0f; |