Cr8-xyz/blender
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Cycles: Pack kernel textures into buffers for OpenCL

Browse Source 
Image textures were being packed into a single buffer for OpenCL, which
limited the amount of memory available for images to the size of one
buffer (usually 4gb on AMD hardware). By packing textures into multiple
buffers that limit is removed, while simultaneously reducing the number
of buffers that need to be passed to each kernel.

Benchmarks were within 2%.

Fixes T51554.

Differential Revision: https://developer.blender.org/D2745
This commit is contained in:
Mai Lavelle
2017-08-08 07:12:04 -04:00
parent b53e35c655
commit ec8ae4d5e9
25 changed files with 685 additions and 328 deletions
Download Patch File Download Diff File Expand all files Collapse all files

66
intern/cycles/kernel/kernel_image_opencl.h
View File

@@ -15,30 +15,42 @@
*/
/* For OpenCL all images are packed in a single array, and we do manual lookup
* and interpolation. */
/* For OpenCL we do manual lookup and interpolation. */
ccl_device_inline ccl_global tex_info_t* kernel_tex_info(KernelGlobals *kg, uint id) {
const uint tex_offset = id
#define KERNEL_TEX(type, ttype, name) + 1
#include "kernel/kernel_textures.h"
;
return &((ccl_global tex_info_t*)kg->buffers[0])[tex_offset];
}
#define tex_fetch(type, info, index) ((ccl_global type*)(kg->buffers[info->buffer] + info->offset))[(index)]
ccl_device_inline float4 svm_image_texture_read(KernelGlobals *kg, int id, int offset)
{
const ccl_global tex_info_t *info = kernel_tex_info(kg, id);
const int texture_type = kernel_tex_type(id);
/* Float4 */
if(texture_type == IMAGE_DATA_TYPE_FLOAT4) {
return kernel_tex_fetch(__tex_image_float4_packed, offset);
return tex_fetch(float4, info, offset);
}
/* Byte4 */
else if(texture_type == IMAGE_DATA_TYPE_BYTE4) {
uchar4 r = kernel_tex_fetch(__tex_image_byte4_packed, offset);
uchar4 r = tex_fetch(uchar4, info, offset);
float f = 1.0f/255.0f;
return make_float4(r.x*f, r.y*f, r.z*f, r.w*f);
}
/* Float */
else if(texture_type == IMAGE_DATA_TYPE_FLOAT) {
float f = kernel_tex_fetch(__tex_image_float_packed, offset);
float f = tex_fetch(float, info, offset);
return make_float4(f, f, f, 1.0f);
}
/* Byte */
else {
uchar r = kernel_tex_fetch(__tex_image_byte_packed, offset);
uchar r = tex_fetch(uchar, info, offset);
float f = r * (1.0f/255.0f);
return make_float4(f, f, f, 1.0f);
}
@@ -64,17 +76,17 @@ ccl_device_inline float svm_image_texture_frac(float x, int *ix)
return x - (float)i;
}
ccl_device_inline uint kernel_decode_image_interpolation(uint4 info)
ccl_device_inline uint kernel_decode_image_interpolation(uint info)
{
return (info.w & (1 << 0)) ? INTERPOLATION_CLOSEST : INTERPOLATION_LINEAR;
return (info & (1 << 0)) ? INTERPOLATION_CLOSEST : INTERPOLATION_LINEAR;
}
ccl_device_inline uint kernel_decode_image_extension(uint4 info)
ccl_device_inline uint kernel_decode_image_extension(uint info)
{
if(info.w & (1 << 1)) {
if(info & (1 << 1)) {
return EXTENSION_REPEAT;
}
else if(info.w & (1 << 2)) {
else if(info & (1 << 2)) {
return EXTENSION_EXTEND;
}
else {
@@ -84,13 +96,16 @@ ccl_device_inline uint kernel_decode_image_extension(uint4 info)
ccl_device float4 kernel_tex_image_interp(KernelGlobals *kg, int id, float x, float y)
{
uint4 info = kernel_tex_fetch(__tex_image_packed_info, id*2);
uint width = info.x;
uint height = info.y;
uint offset = info.z;
const ccl_global tex_info_t *info = kernel_tex_info(kg, id);
uint width = info->width;
uint height = info->height;
uint offset = 0;
/* Decode image options. */
uint interpolation = kernel_decode_image_interpolation(info);
uint extension = kernel_decode_image_extension(info);
uint interpolation = kernel_decode_image_interpolation(info->options);
uint extension = kernel_decode_image_extension(info->options);
/* Actual sampling. */
float4 r;
int ix, iy, nix, niy;
@@ -150,14 +165,17 @@ ccl_device float4 kernel_tex_image_interp(KernelGlobals *kg, int id, float x, fl
ccl_device float4 kernel_tex_image_interp_3d(KernelGlobals *kg, int id, float x, float y, float z)
{
uint4 info = kernel_tex_fetch(__tex_image_packed_info, id*2);
uint width = info.x;
uint height = info.y;
uint offset = info.z;
uint depth = kernel_tex_fetch(__tex_image_packed_info, id*2+1).x;
const ccl_global tex_info_t *info = kernel_tex_info(kg, id);
uint width = info->width;
uint height = info->height;
uint offset = 0;
uint depth = info->depth;
/* Decode image options. */
uint interpolation = kernel_decode_image_interpolation(info);
uint extension = kernel_decode_image_extension(info);
uint interpolation = kernel_decode_image_interpolation(info->options);
uint extension = kernel_decode_image_extension(info->options);
/* Actual sampling. */
float4 r;
int ix, iy, iz, nix, niy, niz;