kernel.h File Reference

#include "kernel/types.h"
#include "util/half.h"
#include "kernel/device/cpu/kernel_arch.h"

Go to the source code of this file.

Macros
#define	KERNEL_NAME_JOIN(x, y, z)
#define	KERNEL_NAME_EVAL(arch, name)
#define	KERNEL_FUNCTION_FULL_NAME(name)
#define	KERNEL_ARCH   cpu
#define	KERNEL_ARCH   cpu_avx2

Functions
KernelGlobalsCPU *	kernel_globals_create ()
void	kernel_globals_free (KernelGlobalsCPU *kg)
void *	kernel_osl_memory (const KernelGlobalsCPU *kg)
bool	kernel_osl_use (const KernelGlobalsCPU *kg)
void	kernel_const_copy (KernelGlobalsCPU *kg, const char *name, void *host, const size_t size)
void	kernel_global_memory_copy (KernelGlobalsCPU *kg, const char *name, void *mem, const size_t size)

Macro Definition Documentation

KERNEL_ARCH [1/2]

#define KERNEL_ARCH   cpu_avx2

Definition at line 35 of file kernel/device/cpu/kernel.h.

KERNEL_ARCH [2/2]

#define KERNEL_ARCH   cpu

Definition at line 35 of file kernel/device/cpu/kernel.h.

KERNEL_FUNCTION_FULL_NAME

#define KERNEL_FUNCTION_FULL_NAME

(

name

)

Value:

KERNEL_NAME_EVAL(KERNEL_ARCH, name)

Definition at line 17 of file kernel/device/cpu/kernel.h.

Referenced by shader_eval_displace().

KERNEL_NAME_EVAL

#define KERNEL_NAME_EVAL	(		arch,
			name )

Value:

KERNEL_NAME_JOIN(kernel, arch, name)

Definition at line 16 of file kernel/device/cpu/kernel.h.

KERNEL_NAME_JOIN

#define KERNEL_NAME_JOIN	(		x,
			y,
			z )

Value:

x##_##y##_##z

Definition at line 15 of file kernel/device/cpu/kernel.h.

Function Documentation

kernel_const_copy()

void kernel_const_copy	(	KernelGlobalsCPU *	kg,
		const char *	name,
		void *	host,
		const size_t	size )

Definition at line 56 of file kernel/device/cpu/kernel.cpp.

References assert, KernelGlobalsCPU::data, and name.

Referenced by CPUDevice::const_copy_to().

kernel_global_memory_copy()

void kernel_global_memory_copy	(	KernelGlobalsCPU *	kg,
		const char *	name,
		void *	mem,
		const size_t	size )

RGBE format represents an RGB value with 4 bytes.

The original implementation by Greg Ward uses 8 bits for RGB each, plus 8 bits shared exponent. It has the same relative precision as the 0 to 255 range of standard 24-bit image formats, but offers extended intensity range: https://www.graphics.cornell.edu/~bjw/rgbe.html GL_EXT_texture_shared_exponent uses 9 bits for RGB and 5 bits for exponent instead, with smaller range but higher precision: https://registry.khronos.org/OpenGL/extensions/EXT/EXT_texture_shared_exponent.txt

Our implementation is mostly based on GL_EXT_texture_shared_exponent, but uses 8 bits for RGB each, and adds 3 sign bits to represent negative values. The memory layout is as follows:

   xxxxxxxx  xxxxxxxx  xxxxxxxx  xxx  xxxxx
     m(R)      m(G)      m(B)    sgn   exp

Each float component is interpreted as sgn exp - bias f = (-1) * 0.m * 2

We choose a bias of 15, so that the largest representable value is RGBE_MAX = 0.11111111 * 2^(31 - 15) = 65280, and the smallest positive representable value is RGBE_MIN = 0.00000001 * 2^(0 - 15) = 1.1920929e-7

Definition at line 66 of file kernel/device/cpu/kernel.cpp.

References assert, name, and size().

Referenced by CPUDevice::global_alloc().

kernel_globals_create()

KernelGlobalsCPU * kernel_globals_create

(

)

kernel_globals_free()

void kernel_globals_free

(

KernelGlobalsCPU *

kg

)

kernel_osl_memory()

void * kernel_osl_memory

(

const KernelGlobalsCPU *

kg

)

kernel_osl_use()

bool kernel_osl_use

(

const KernelGlobalsCPU *

kg

)

References name, and size().