Cycles: Tweak inline policy for some functions

The goal is to make Experimental kernel closer in performance to the
official kernel, avoiding spills and such.

There should not be big impact on official kernel, own tests showed
few percent performance drop on laptop's GPU. CPU was always the
same speed on AVX, AVX2 and SSE4.1 CPUs i've been testing here.

This seems to be the last essential step before we can get rid of
Experimental kernel and enable SSS officially on GPU without causing
some major performance issues.

Surely some more tweaks are possibly required, but that we can do
for until cows go home anyway.

6 files changed, 42 insertions, 21 deletions

diff --git a/intern/cycles/kernel/kernel_light.h b/intern/cycles/kernel/kernel_light.h
index 7590ec2d706..ef098aeb050 100644
--- a/intern/cycles/kernel/kernel_light.h
+++ b/intern/cycles/kernel/kernel_light.h
@@ -845,7 +845,14 @@ ccl_device bool light_select_reached_max_bounces(KernelGlobals *kg, int index, i
 	return (bounce > __float_as_int(data4.x));
 }
 
-ccl_device void light_sample(KernelGlobals *kg, float randt, float randu, float randv, float time, float3 P, int bounce, LightSample *ls)
+ccl_device_noinline void light_sample(KernelGlobals *kg,
+                                      float randt,
+                                      float randu,
+                                      float randv,
+                                      float time,
+                                      float3 P,
+                                      int bounce,
+                                      LightSample *ls)
 {
 	/* sample index */
 	int index = light_distribution_sample(kg, randt);
diff --git a/intern/cycles/kernel/kernel_path.h b/intern/cycles/kernel/kernel_path.h
index fc32201596f..650e3b047f0 100644
--- a/intern/cycles/kernel/kernel_path.h
+++ b/intern/cycles/kernel/kernel_path.h
@@ -391,7 +391,12 @@ ccl_device void kernel_path_indirect(KernelGlobals *kg,
 	}
 }
 
-ccl_device void kernel_path_ao(KernelGlobals *kg, ShaderData *sd, PathRadiance *L, PathState *state, RNG *rng, float3 throughput)
+ccl_device_noinline void kernel_path_ao(KernelGlobals *kg,
+                                        ShaderData *sd,
+                                        PathRadiance *L,
+                                        PathState *state,
+                                        RNG *rng,
+                                        float3 throughput)
 {
 	/* todo: solve correlation */
 	float bsdf_u, bsdf_v;
@@ -586,7 +591,11 @@ ccl_device void kernel_path_subsurface_setup_indirect(
 
 #endif  /* __SUBSURFACE__ */
 
-ccl_device float4 kernel_path_integrate(KernelGlobals *kg, RNG *rng, int sample, Ray ray, ccl_global float *buffer)
+ccl_device_inline float4 kernel_path_integrate(KernelGlobals *kg,
+                                               RNG *rng,
+                                               int sample,
+                                               Ray ray,
+                                               ccl_global float *buffer)
 {
 	/* initialize */
 	PathRadiance L;
diff --git a/intern/cycles/kernel/kernel_shader.h b/intern/cycles/kernel/kernel_shader.h
index 63e5ace8f94..9a4d2372075 100644
--- a/intern/cycles/kernel/kernel_shader.h
+++ b/intern/cycles/kernel/kernel_shader.h
@@ -48,8 +48,10 @@ ccl_device void shader_setup_object_transforms(KernelGlobals *kg, ShaderData *sd
 }
 #endif
 
-ccl_device void shader_setup_from_ray(KernelGlobals *kg, ShaderData *sd,
-	const Intersection *isect, const Ray *ray)
+ccl_device_noinline void shader_setup_from_ray(KernelGlobals *kg,
+                                               ShaderData *sd,
+                                               const Intersection *isect,
+                                               const Ray *ray)
 {
 #ifdef __INSTANCING__
 	ccl_fetch(sd, object) = (isect->object == PRIM_NONE)? kernel_tex_fetch(__prim_object, isect->prim): isect->object;
diff --git a/intern/cycles/kernel/kernel_shadow.h b/intern/cycles/kernel/kernel_shadow.h
index cb5fcf813f3..e86051095a5 100644
--- a/intern/cycles/kernel/kernel_shadow.h
+++ b/intern/cycles/kernel/kernel_shadow.h
@@ -183,11 +183,14 @@ ccl_device_inline bool shadow_blocked(KernelGlobals *kg, PathState *state, Ray *
  * potentially transparent, and only in that case start marching. this gives
  * one extra ray cast for the cases were we do want transparency. */
 
-ccl_device_inline bool shadow_blocked(KernelGlobals *kg, ccl_addr_space PathState *state, ccl_addr_space Ray *ray_input, float3 *shadow
+ccl_device_noinline bool shadow_blocked(KernelGlobals *kg,
+                                        ccl_addr_space PathState *state,
+                                        ccl_addr_space Ray *ray_input,
+                                        float3 *shadow
 #ifdef __SPLIT_KERNEL__
-                                      , ShaderData *sd_mem, Intersection *isect_mem
+                                        , ShaderData *sd_mem, Intersection *isect_mem
 #endif
-                                      )
+                                       )
 {
 	*shadow = make_float3(1.0f, 1.0f, 1.0f);
 
diff --git a/intern/cycles/kernel/svm/svm_mix.h b/intern/cycles/kernel/svm/svm_mix.h
index 6111214acba..7cbda111d81 100644
--- a/intern/cycles/kernel/svm/svm_mix.h
+++ b/intern/cycles/kernel/svm/svm_mix.h
@@ -261,7 +261,7 @@ ccl_device float3 svm_mix_clamp(float3 col)
 	return outcol;
 }
 
-ccl_device float3 svm_mix(NodeMix type, float fac, float3 c1, float3 c2)
+ccl_device_noinline float3 svm_mix(NodeMix type, float fac, float3 c1, float3 c2)
 {
 	float t = saturate(fac);
 
diff --git a/intern/cycles/kernel/svm/svm_tex_coord.h b/intern/cycles/kernel/svm/svm_tex_coord.h
index eebd9bee420..27fed89fdf7 100644
--- a/intern/cycles/kernel/svm/svm_tex_coord.h
+++ b/intern/cycles/kernel/svm/svm_tex_coord.h
@@ -99,12 +99,12 @@ ccl_device void svm_node_tex_coord(KernelGlobals *kg,
 	stack_store_float3(stack, out_offset, data);
 }
 
-ccl_device void svm_node_tex_coord_bump_dx(KernelGlobals *kg,
-                                           ShaderData *sd,
-                                           int path_flag,
-                                           float *stack,
-                                           uint4 node,
-                                           int *offset)
+ccl_device_inline void svm_node_tex_coord_bump_dx(KernelGlobals *kg,
+                                                  ShaderData *sd,
+                                                  int path_flag,
+                                                  float *stack,
+                                                  uint4 node,
+                                                  int *offset)
 {
 #ifdef __RAY_DIFFERENTIALS__
 	float3 data;
@@ -184,12 +184,12 @@ ccl_device void svm_node_tex_coord_bump_dx(KernelGlobals *kg,
 #endif
 }
 
-ccl_device void svm_node_tex_coord_bump_dy(KernelGlobals *kg,
-                                           ShaderData *sd,
-                                           int path_flag,
-                                           float *stack,
-                                           uint4 node,
-                                           int *offset)
+ccl_device_inline void svm_node_tex_coord_bump_dy(KernelGlobals *kg,
+                                                  ShaderData *sd,
+                                                  int path_flag,
+                                                  float *stack,
+                                                  uint4 node,
+                                                  int *offset)
 {
 #ifdef __RAY_DIFFERENTIALS__
 	float3 data;