Cycles: avoid using float3 in kernel constant memory, just so we're sure alignment

is working compatible between cpu and gpu.

5 files changed, 18 insertions, 20 deletions

diff --git a/intern/cycles/kernel/kernel_camera.h b/intern/cycles/kernel/kernel_camera.h
index 9cdc2f1f865..2dbdd076891 100644
--- a/intern/cycles/kernel/kernel_camera.h
+++ b/intern/cycles/kernel/kernel_camera.h
@@ -74,8 +74,8 @@ __device void camera_sample_perspective(KernelGlobals *kg, float raster_x, float
 	ray->dP.dx = make_float3(0.0f, 0.0f, 0.0f);
 	ray->dP.dy = make_float3(0.0f, 0.0f, 0.0f);
 
-	ray->dD.dx = normalize(Ddiff + kernel_data.cam.dx) - normalize(Ddiff);
-	ray->dD.dy = normalize(Ddiff + kernel_data.cam.dy) - normalize(Ddiff);
+	ray->dD.dx = normalize(Ddiff + float4_to_float3(kernel_data.cam.dx)) - normalize(Ddiff);
+	ray->dD.dy = normalize(Ddiff + float4_to_float3(kernel_data.cam.dy)) - normalize(Ddiff);
 #endif
 
 #ifdef __CAMERA_CLIPPING__
@@ -107,8 +107,8 @@ __device void camera_sample_orthographic(KernelGlobals *kg, float raster_x, floa
 
 #ifdef __RAY_DIFFERENTIALS__
 	/* ray differential */
-	ray->dP.dx = kernel_data.cam.dx;
-	ray->dP.dy = kernel_data.cam.dy;
+	ray->dP.dx = float4_to_float3(kernel_data.cam.dx);
+	ray->dP.dy = float4_to_float3(kernel_data.cam.dy);
 
 	ray->dD.dx = make_float3(0.0f, 0.0f, 0.0f);
 	ray->dD.dy = make_float3(0.0f, 0.0f, 0.0f);
diff --git a/intern/cycles/kernel/kernel_types.h b/intern/cycles/kernel/kernel_types.h
index 72ebfefbd90..ea73f87a8a5 100644
--- a/intern/cycles/kernel/kernel_types.h
+++ b/intern/cycles/kernel/kernel_types.h
@@ -295,7 +295,11 @@ typedef struct ShaderData {
 #endif
 } ShaderData;
 
-/* Constrant Kernel Data */
+/* Constrant Kernel Data
+ *
+ * These structs are passed from CPU to various devices, and the struct layout
+ * must match exactly. Structs are padded to ensure 16 byte alignment, and we
+ * do not use float3 because its size may not be the same on all devices. */
 
 typedef struct KernelCamera {
 	/* type */
@@ -307,14 +311,8 @@ typedef struct KernelCamera {
 	Transform rastertocamera;
 
 	/* differentials */
-	float3 dx;
-#ifndef WITH_OPENCL
-	float pad1;
-#endif
-	float3 dy;
-#ifndef WITH_OPENCL
-	float pad2;
-#endif
+	float4 dx;
+	float4 dy;
 
 	/* depth of field */
 	float aperturesize;
@@ -355,10 +353,6 @@ typedef struct KernelBackground {
 typedef struct KernelSunSky {
 	/* sun direction in spherical and cartesian */
 	float theta, phi, pad3, pad4;
-	float3 dir;
-#ifndef WITH_OPENCL
-	float pad;
-#endif
 
 	/* perez function parameters */
 	float zenith_Y, zenith_x, zenith_y, pad2;
diff --git a/intern/cycles/render/camera.cpp b/intern/cycles/render/camera.cpp
index d5fca87491d..a83ae81844c 100644
--- a/intern/cycles/render/camera.cpp
+++ b/intern/cycles/render/camera.cpp
@@ -150,8 +150,8 @@ void Camera::device_update(Device *device, DeviceScene *dscene)
 	kcam->ortho = ortho;
 
 	/* store differentials */
-	kcam->dx = dx;
-	kcam->dy = dy;
+	kcam->dx = float3_to_float4(dx);
+	kcam->dy = float3_to_float4(dy);
 
 	/* clipping */
 	kcam->nearclip = nearclip;
diff --git a/intern/cycles/render/nodes.cpp b/intern/cycles/render/nodes.cpp
index 7d873221cd6..81d156a079d 100644
--- a/intern/cycles/render/nodes.cpp
+++ b/intern/cycles/render/nodes.cpp
@@ -273,7 +273,6 @@ static void sky_texture_precompute(KernelSunSky *ksunsky, float3 dir, float turb
 
 	ksunsky->theta = theta;
 	ksunsky->phi = phi;
-	ksunsky->dir = dir;
 
 	float theta2 = theta*theta;
 	float theta3 = theta*theta*theta;
diff --git a/intern/cycles/util/util_math.h b/intern/cycles/util/util_math.h
index 7c56f0fbb12..0a1d8ff4555 100644
--- a/intern/cycles/util/util_math.h
+++ b/intern/cycles/util/util_math.h
@@ -536,6 +536,11 @@ __device_inline float3 float4_to_float3(const float4 a)
 	return make_float3(a.x, a.y, a.z);
 }
 
+__device_inline float4 float3_to_float4(const float3 a)
+{
+	return make_float4(a.x, a.y, a.z, 1.0f);
+}
+
 #ifndef __KERNEL_GPU__
 
 __device_inline void print_float3(const char *label, const float3& a)