kannonier8/blender
1
0
Fork 0
forked from bartvdbraak/blender
Code Pull Requests Activity

Cycles: fix for CUDA build.

Browse Source
This commit is contained in:
Brecht Van Lommel 2012-04-28 09:10:20 +00:00
parent 2c592b0d96
commit 44924a2e5e
1 changed files with 19 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

38
intern/cycles/util/util_math.h
View File

@ -515,15 +515,6 @@ __device_inline void print_float3(const char *label, const float3& a)
printf("%s: %.8f %.8f %.8f\n", label, a.x, a.y, a.z);
}
__device_inline float reduce_add(const float3& a)
{
#ifdef __KERNEL_SSE__
return (a.x + a.y + a.z);
#else
return (a.x + a.y + a.z);
#endif
}
__device_inline float3 rcp(const float3& a)
{
#ifdef __KERNEL_SSE__
@ -550,6 +541,15 @@ __device_inline bool is_zero(const float3 a)
#endif
}
__device_inline float reduce_add(const float3& a)
{
#ifdef __KERNEL_SSE__
return (a.x + a.y + a.z);
#else
return (a.x + a.y + a.z);
#endif
}
__device_inline float average(const float3 a)
{
return reduce_add(a)*(1.0f/3.0f);
@ -783,16 +783,6 @@ __device_inline float4 reduce_add(const float4& a)
}
#endif
__device_inline float reduce_add(const float4& a)
{
#ifdef __KERNEL_SSE__
float4 h = shuffle<1,0,3,2>(a) + a;
return _mm_cvtss_f32(shuffle<2,3,0,1>(h) + h); /* todo: efficiency? */
#else
return ((a.x + a.y) + (a.z + a.w));
#endif
}
__device_inline void print_float4(const char *label, const float4& a)
{
printf("%s: %.8f %.8f %.8f %.8f\n", label, a.x, a.y, a.z, a.w);
@ -811,6 +801,16 @@ __device_inline bool is_zero(const float4& a)
#endif
}
__device_inline float reduce_add(const float4& a)
{
#ifdef __KERNEL_SSE__
float4 h = shuffle<1,0,3,2>(a) + a;
return _mm_cvtss_f32(shuffle<2,3,0,1>(h) + h); /* todo: efficiency? */
#else
return ((a.x + a.y) + (a.z + a.w));
#endif
}
__device_inline float average(const float4& a)
{
return reduce_add(a) * 0.25f;