Cycles: inline more functions on the GPU

This makes little difference for CUDA and OpenCL, but will be helpful
for Optix.

7 files changed, 32 insertions, 41 deletions

diff --git a/intern/cycles/kernel/svm/svm_attribute.h b/intern/cycles/kernel/svm/svm_attribute.h
index 341eaad28ee..eaee0f9e4ee 100644
--- a/intern/cycles/kernel/svm/svm_attribute.h
+++ b/intern/cycles/kernel/svm/svm_attribute.h
@@ -80,13 +80,7 @@ ccl_device void svm_node_attr(KernelGlobals *kg, ShaderData *sd, float *stack, u
   }
 }
 
-#ifndef __KERNEL_CUDA__
-ccl_device
-#else
-ccl_device_noinline
-#endif
-    void
-    svm_node_attr_bump_dx(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
+ccl_device void svm_node_attr_bump_dx(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
 {
   NodeAttributeType type = NODE_ATTR_FLOAT;
   uint out_offset = 0;
@@ -125,13 +119,7 @@ ccl_device_noinline
   }
 }
 
-#ifndef __KERNEL_CUDA__
-ccl_device
-#else
-ccl_device_noinline
-#endif
-    void
-    svm_node_attr_bump_dy(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
+ccl_device void svm_node_attr_bump_dy(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
 {
   NodeAttributeType type = NODE_ATTR_FLOAT;
   uint out_offset = 0;
diff --git a/intern/cycles/kernel/svm/svm_brick.h b/intern/cycles/kernel/svm/svm_brick.h
index f1d74b7df96..6984afa30a5 100644
--- a/intern/cycles/kernel/svm/svm_brick.h
+++ b/intern/cycles/kernel/svm/svm_brick.h
@@ -27,16 +27,16 @@ ccl_device_inline float brick_noise(uint n) /* fast integer noise */
   return 0.5f * ((float)nn / 1073741824.0f);
 }
 
-ccl_device_noinline float2 svm_brick(float3 p,
-                                     float mortar_size,
-                                     float mortar_smooth,
-                                     float bias,
-                                     float brick_width,
-                                     float row_height,
-                                     float offset_amount,
-                                     int offset_frequency,
-                                     float squash_amount,
-                                     int squash_frequency)
+ccl_device_noinline_cpu float2 svm_brick(float3 p,
+                                         float mortar_size,
+                                         float mortar_smooth,
+                                         float bias,
+                                         float brick_width,
+                                         float row_height,
+                                         float offset_amount,
+                                         int offset_frequency,
+                                         float squash_amount,
+                                         int squash_frequency)
 {
   int bricknum, rownum;
   float offset = 0.0f;
diff --git a/intern/cycles/kernel/svm/svm_color_util.h b/intern/cycles/kernel/svm/svm_color_util.h
index 12b59d2616b..3a6a5ba782f 100644
--- a/intern/cycles/kernel/svm/svm_color_util.h
+++ b/intern/cycles/kernel/svm/svm_color_util.h
@@ -264,7 +264,7 @@ ccl_device float3 svm_mix_clamp(float3 col)
   return outcol;
 }
 
-ccl_device_noinline float3 svm_mix(NodeMix type, float fac, float3 c1, float3 c2)
+ccl_device_noinline_cpu float3 svm_mix(NodeMix type, float fac, float3 c1, float3 c2)
 {
   float t = saturate(fac);
 
diff --git a/intern/cycles/kernel/svm/svm_magic.h b/intern/cycles/kernel/svm/svm_magic.h
index 6ba1a5817ad..9c160e6d8cc 100644
--- a/intern/cycles/kernel/svm/svm_magic.h
+++ b/intern/cycles/kernel/svm/svm_magic.h
@@ -18,7 +18,7 @@ CCL_NAMESPACE_BEGIN
 
 /* Magic */
 
-ccl_device_noinline float3 svm_magic(float3 p, int n, float distortion)
+ccl_device_noinline_cpu float3 svm_magic(float3 p, int n, float distortion)
 {
   float x = sinf((p.x + p.y + p.z) * 5.0f);
   float y = cosf((-p.x + p.y - p.z) * 5.0f);
diff --git a/intern/cycles/kernel/svm/svm_musgrave.h b/intern/cycles/kernel/svm/svm_musgrave.h
index 298155b795f..9291c7e7295 100644
--- a/intern/cycles/kernel/svm/svm_musgrave.h
+++ b/intern/cycles/kernel/svm/svm_musgrave.h
@@ -25,7 +25,10 @@ CCL_NAMESPACE_BEGIN
  * from "Texturing and Modelling: A procedural approach"
  */
 
-ccl_device_noinline float noise_musgrave_fBm(float3 p, float H, float lacunarity, float octaves)
+ccl_device_noinline_cpu float noise_musgrave_fBm(float3 p,
+                                                 float H,
+                                                 float lacunarity,
+                                                 float octaves)
 {
   float rmd;
   float value = 0.0f;
@@ -53,10 +56,10 @@ ccl_device_noinline float noise_musgrave_fBm(float3 p, float H, float lacunarity
  * octaves: number of frequencies in the fBm
  */
 
-ccl_device_noinline float noise_musgrave_multi_fractal(float3 p,
-                                                       float H,
-                                                       float lacunarity,
-                                                       float octaves)
+ccl_device_noinline_cpu float noise_musgrave_multi_fractal(float3 p,
+                                                           float H,
+                                                           float lacunarity,
+                                                           float octaves)
 {
   float rmd;
   float value = 1.0f;
@@ -85,7 +88,7 @@ ccl_device_noinline float noise_musgrave_multi_fractal(float3 p,
  * offset: raises the terrain from `sea level'
  */
 
-ccl_device_noinline float noise_musgrave_hetero_terrain(
+ccl_device_noinline_cpu float noise_musgrave_hetero_terrain(
     float3 p, float H, float lacunarity, float octaves, float offset)
 {
   float value, increment, rmd;
@@ -121,7 +124,7 @@ ccl_device_noinline float noise_musgrave_hetero_terrain(
  * offset: raises the terrain from `sea level'
  */
 
-ccl_device_noinline float noise_musgrave_hybrid_multi_fractal(
+ccl_device_noinline_cpu float noise_musgrave_hybrid_multi_fractal(
     float3 p, float H, float lacunarity, float octaves, float offset, float gain)
 {
   float result, signal, weight, rmd;
@@ -159,7 +162,7 @@ ccl_device_noinline float noise_musgrave_hybrid_multi_fractal(
  * offset: raises the terrain from `sea level'
  */
 
-ccl_device_noinline float noise_musgrave_ridged_multi_fractal(
+ccl_device_noinline_cpu float noise_musgrave_ridged_multi_fractal(
     float3 p, float H, float lacunarity, float octaves, float offset, float gain)
 {
   float result, signal, weight;
diff --git a/intern/cycles/kernel/svm/svm_noise.h b/intern/cycles/kernel/svm/svm_noise.h
index 0bf3dfda4df..dd375af27e5 100644
--- a/intern/cycles/kernel/svm/svm_noise.h
+++ b/intern/cycles/kernel/svm/svm_noise.h
@@ -182,7 +182,7 @@ ccl_device_inline ssef scale3_sse(const ssef &result)
 #endif
 
 #ifndef __KERNEL_SSE2__
-ccl_device_noinline float perlin(float x, float y, float z)
+ccl_device_noinline_cpu float perlin(float x, float y, float z)
 {
   int X;
   float fx = floorfrac(x, &X);
diff --git a/intern/cycles/kernel/svm/svm_wave.h b/intern/cycles/kernel/svm/svm_wave.h
index 03b7f330970..baaa89ab0cb 100644
--- a/intern/cycles/kernel/svm/svm_wave.h
+++ b/intern/cycles/kernel/svm/svm_wave.h
@@ -18,12 +18,12 @@ CCL_NAMESPACE_BEGIN
 
 /* Wave */
 
-ccl_device_noinline float svm_wave(NodeWaveType type,
-                                   NodeWaveProfile profile,
-                                   float3 p,
-                                   float detail,
-                                   float distortion,
-                                   float dscale)
+ccl_device_noinline_cpu float svm_wave(NodeWaveType type,
+                                       NodeWaveProfile profile,
+                                       float3 p,
+                                       float detail,
+                                       float distortion,
+                                       float dscale)
 {
   float n;