Changeset View

Standalone View

	Show First 20 Lines • Show All 55 Lines • ▼ Show 20 Lines
	#endif			#endif

	struct half4 {			struct half4 {
	half x, y, z, w;			half x, y, z, w;
	};			};

	#if defined(__KERNEL_CUDA__) \|\| defined(__KERNEL_HIP__)			#if defined(__KERNEL_CUDA__) \|\| defined(__KERNEL_HIP__)

	ccl_device_inline void float4_store_half(half *h, float4 f)			ccl_device_inline void float4_store_half(ccl_private half *h, float4 f)
	{			{
	h[0] = __float2half(f.x);			h[0] = __float2half(f.x);
	h[1] = __float2half(f.y);			h[1] = __float2half(f.y);
	h[2] = __float2half(f.z);			h[2] = __float2half(f.z);
	h[3] = __float2half(f.w);			h[3] = __float2half(f.w);
	}			}

	#else			#else

	ccl_device_inline void float4_store_half(half *h, float4 f)			ccl_device_inline void float4_store_half(ccl_private half *h, float4 f)
	{			{

	# ifndef __KERNEL_SSE2__			# ifndef __KERNEL_SSE2__
	for (int i = 0; i < 4; i++) {			for (int i = 0; i < 4; i++) {
	/* optimized float to half for pixels:			/* optimized float to half for pixels:
	* assumes no negative, no nan, no inf, and sets denormal to 0 */			* assumes no negative, no nan, no inf, and sets denormal to 0 */
	union {			union {
	uint i;			uint i;
	▲ Show 20 Lines • Show All 97 Lines • Show Last 20 Lines