gl: Improve CS throughput
- Avoids making too many invocations, especially given the 1D nature of some GPU dispatch handlers
This commit is contained in:
@@ -21,6 +21,8 @@ namespace gl
|
|||||||
optimal_group_size = 128;
|
optimal_group_size = 128;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
optimal_kernel_size = 256 / optimal_group_size;
|
||||||
|
|
||||||
glGetIntegeri_v(GL_MAX_COMPUTE_WORK_GROUP_COUNT, 0, reinterpret_cast<GLint*>(&max_invocations_x));
|
glGetIntegeri_v(GL_MAX_COMPUTE_WORK_GROUP_COUNT, 0, reinterpret_cast<GLint*>(&max_invocations_x));
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|||||||
Reference in New Issue
Block a user