Differential D2856 Diff 9351 intern/cycles/kernel/kernel_passes.h

Changeset View

Standalone View

intern/cycles/kernel/kernel_passes.h

	Show All 10 Lines
	* distributed under the License is distributed on an "AS IS" BASIS,			* distributed under the License is distributed on an "AS IS" BASIS,
	* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.			* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
	* See the License for the specific language governing permissions and			* See the License for the specific language governing permissions and
	* limitations under the License.			* limitations under the License.
	*/			*/

	CCL_NAMESPACE_BEGIN			CCL_NAMESPACE_BEGIN

				#if defined(__SPLIT_KERNEL__) \|\| defined(__KERNEL_CUDA__)
				#define __ATOMIC_PASS_WRITE__
				#endif

	ccl_device_inline void kernel_write_pass_float(ccl_global float *buffer, float value)			ccl_device_inline void kernel_write_pass_float(ccl_global float *buffer, float value)
	{			{
	ccl_global float *buf = buffer;			ccl_global float *buf = buffer;
	#if defined(__SPLIT_KERNEL__)			#ifdef __ATOMIC_PASS_WRITE__
	atomic_add_and_fetch_float(buf, value);			atomic_add_and_fetch_float(buf, value);
	#else			#else
	*buf += value;			*buf += value;
	#endif /* __SPLIT_KERNEL__ */			#endif
	}			}

	ccl_device_inline void kernel_write_pass_float3(ccl_global float *buffer, float3 value)			ccl_device_inline void kernel_write_pass_float3(ccl_global float *buffer, float3 value)
	{			{
	#if defined(__SPLIT_KERNEL__)			#ifdef __ATOMIC_PASS_WRITE__
	ccl_global float *buf_x = buffer + 0;			ccl_global float *buf_x = buffer + 0;
	ccl_global float *buf_y = buffer + 1;			ccl_global float *buf_y = buffer + 1;
	ccl_global float *buf_z = buffer + 2;			ccl_global float *buf_z = buffer + 2;

	atomic_add_and_fetch_float(buf_x, value.x);			atomic_add_and_fetch_float(buf_x, value.x);
	atomic_add_and_fetch_float(buf_y, value.y);			atomic_add_and_fetch_float(buf_y, value.y);
	atomic_add_and_fetch_float(buf_z, value.z);			atomic_add_and_fetch_float(buf_z, value.z);
	#else			#else
	ccl_global float3 buf = (ccl_global float3)buffer;			ccl_global float3 buf = (ccl_global float3)buffer;
	*buf += value;			*buf += value;
	#endif /* __SPLIT_KERNEL__ */			#endif
	}			}

	ccl_device_inline void kernel_write_pass_float4(ccl_global float *buffer, float4 value)			ccl_device_inline void kernel_write_pass_float4(ccl_global float *buffer, float4 value)
	{			{
	#if defined(__SPLIT_KERNEL__)			#ifdef __ATOMIC_PASS_WRITE__
	ccl_global float *buf_x = buffer + 0;			ccl_global float *buf_x = buffer + 0;
	ccl_global float *buf_y = buffer + 1;			ccl_global float *buf_y = buffer + 1;
	ccl_global float *buf_z = buffer + 2;			ccl_global float *buf_z = buffer + 2;
	ccl_global float *buf_w = buffer + 3;			ccl_global float *buf_w = buffer + 3;

	atomic_add_and_fetch_float(buf_x, value.x);			atomic_add_and_fetch_float(buf_x, value.x);
	atomic_add_and_fetch_float(buf_y, value.y);			atomic_add_and_fetch_float(buf_y, value.y);
	atomic_add_and_fetch_float(buf_z, value.z);			atomic_add_and_fetch_float(buf_z, value.z);
	atomic_add_and_fetch_float(buf_w, value.w);			atomic_add_and_fetch_float(buf_w, value.w);
	#else			#else
	ccl_global float4 buf = (ccl_global float4)buffer;			ccl_global float4 buf = (ccl_global float4)buffer;
	*buf += value;			*buf += value;
	#endif /* __SPLIT_KERNEL__ */			#endif
	}			}

	#ifdef __DENOISING_FEATURES__			#ifdef __DENOISING_FEATURES__
	ccl_device_inline void kernel_write_pass_float_variance(ccl_global float *buffer, float value)			ccl_device_inline void kernel_write_pass_float_variance(ccl_global float *buffer, float value)
	{			{
	kernel_write_pass_float(buffer, value);			kernel_write_pass_float(buffer, value);

	/* The online one-pass variance update that's used for the megakernel can't easily be implemented			/* The online one-pass variance update that's used for the megakernel can't easily be implemented
	* with atomics, so for the split kernel the E[x^2] - 1/N * (E[x])^2 fallback is used. */			* with atomics, so for the split kernel the E[x^2] - 1/N * (E[x])^2 fallback is used. */
	kernel_write_pass_float(buffer+1, value*value);			kernel_write_pass_float(buffer+1, value*value);
	}			}

	# if defined(__SPLIT_KERNEL__)			# ifdef __ATOMIC_PASS_WRITE__
	# define kernel_write_pass_float3_unaligned kernel_write_pass_float3			# define kernel_write_pass_float3_unaligned kernel_write_pass_float3
	# else			# else
	ccl_device_inline void kernel_write_pass_float3_unaligned(ccl_global float *buffer, float3 value)			ccl_device_inline void kernel_write_pass_float3_unaligned(ccl_global float *buffer, float3 value)
	{			{
	buffer[0] += value.x;			buffer[0] += value.x;
	buffer[1] += value.y;			buffer[1] += value.y;
	buffer[2] += value.z;			buffer[2] += value.z;
	}			}
	▲ Show 20 Lines • Show All 291 Lines • Show Last 20 Lines