opencl.c File Reference

#include "common/opencl.h"
#include "common/paths.h"
#include "common/startup_progress.h"
#include "common/utility.h"
#include "system/capabilities.h"
#include "system/sys_resources.h"
#include "pixel/bilateralcl.h"
#include "darktable.h"
#include "common/dlopencl.h"
#include "pixel/dwt.h"
#include "common/file_location.h"
#include "pixel/gaussian.h"
#include "pixel/guided_filter.h"
#include "pixel/heal.h"
#include "pixel/interpolation.h"
#include "pixel/locallaplaciancl.h"
#include "system/nvidia_gpus.h"
#include <fcntl.h>
#include "system/opencl_drivers_blacklist.h"
#include "common/conf.h"
#include "develop/blend.h"
#include "develop/pixelpipe.h"
#include "caches/pixelpipe_cache.h"
#include "common/times.h"
#include <assert.h>
#include <locale.h>
#include <stdio.h>
#include <string.h>
#include <strings.h>
#include <ctype.h>
#include <errno.h>
#include <glib/gstdio.h>
#include <sys/stat.h>
#include <zlib.h>

Include dependency graph for opencl.c:

Go to the source code of this file.

Data Structures
struct	dt_opencl_t
struct	_gpu_runtime_t
struct	dt_opencl_mem_record_t

Typedefs
typedef struct dt_opencl_t	dt_opencl_t
typedef struct _gpu_runtime_t	_gpu_runtime_t
typedef struct dt_opencl_mem_record_t	dt_opencl_mem_record_t

Functions
static void	dt_opencl_cleanup_device (dt_opencl_t *cl, int i)
static void	_opencl_splash_update_compile (const char *programname)
static const char *	dt_opencl_get_vendor_by_id (unsigned int id)
static const char *	_gpu_runtime_of_platform (const char *platform_name)
static int	_gpu_runtime_crash_streak (const char *runtime)
static void	_gpu_runtime_set_crash_streak (const char *runtime, int value)
static char *	_ascii_str_canonical (const char *in, char *out, int maxlen)
static void	dt_opencl_priority_parse (dt_opencl_t *cl, char *configstr, int *priority_list, int *mandatory)
static void	dt_opencl_update_priorities ()
static void	dt_opencl_apply_scheduling_profile ()
static void	dt_opencl_set_synchronization_timeout (int value)
static int	dt_opencl_get_device_info (dt_opencl_t *cl, cl_device_id device, cl_device_info param_name, void **param_value, size_t *param_value_size)
int	dt_opencl_avoid_atomics (const int devid)
int	dt_opencl_micro_nap (const int devid)
gboolean	dt_opencl_use_pinned_memory (const int devid)
gboolean	dt_opencl_is_pinned_memory (cl_mem mem)
void	dt_opencl_write_device_config (const int devid)
static int	_dt_opencl_get_conf_int (const gchar *key_device, const gchar *conf_name, gboolean *safety_ok)
gboolean	dt_opencl_read_device_config (const int devid)
int	dt_opencl_get_detected_device_count (void)
const dt_opencl_detected_device_t *	dt_opencl_get_detected_device (const int detected)
gboolean	dt_opencl_detected_device_enabled (const int detected)
int	dt_opencl_set_detected_device_enabled (const int detected, const gboolean enabled)
gboolean	dt_opencl_detected_device_pinned_memory (const int detected)
int	dt_opencl_set_detected_device_pinned_memory (const int detected, const gboolean enabled)
size_t	dt_opencl_detected_device_headroom (const int detected)
int	dt_opencl_set_detected_device_headroom (const int detected, const size_t headroom)
static int	dt_opencl_device_init (dt_opencl_t *cl, const int dev, cl_device_id *devices, const int k)
static void	_opencl_driver_crash_marker_path (char *buf, size_t buf_len)
void	dt_opencl_note_crash_backtrace (const char *backtrace, size_t backtrace_len)
	Tell the OpenCL module that the process crashed, and let it judge whether the crash was its driver's doing.
static void	_opencl_fold_driver_crashes (void)
void	dt_opencl_init (const gboolean exclude_opencl, const gboolean print_statistics)
void	dt_opencl_clear_driver_crash_streak (void)
void	dt_opencl_cleanup (void)
gboolean	dt_opencl_finish (const int devid)
int	dt_opencl_enqueue_barrier (const int devid)
static int	_take_from_list (int *list, int value)
static int	_device_by_cname (const char *name)
int	dt_opencl_reserve_device_for_pipe (const int pipetype)
	Reserve a device for a pipe run: choose a free one by the pipe's priority list and take its lock. Blocks or gives up per the opencl_mandatory_timeout conf key.
void	dt_opencl_release_device (const int devid)
	Release a device reserved by either reserve function.
void	dt_opencl_reserve_device_by_id (const int devid)
	Reserve a device the caller has already identified, blocking until it is free.
int	dt_opencl_try_reserve_device_by_id (const int devid)
	Reserve a device only if it is free right now.
int	dt_opencl_get_num_devices (void)
	Number of usable OpenCL devices; 0 when OpenCL is unavailable.
const char *	dt_opencl_get_device_name (const int devid)
	Human-readable device name, owned by the OpenCL module.
gboolean	dt_opencl_get_device_max_image_size (const int devid, int *width, int *height)
	Largest 2D image the device will accept, which is what bounds tile size.
size_t	dt_opencl_get_device_max_global_mem (const int devid)
	Total device memory in bytes, or 0 for an out-of-range id.
size_t	dt_opencl_get_device_memory_in_use (const int devid)
	Bytes Ansel holds on the device by its own accounting (dt_opencl_memory_statistics()), or 0 for an out-of-range id. What other applications hold there is not in it.
size_t	dt_opencl_get_device_headroom (const int devid)
	Bytes of the device left to other applications (its forced_headroom, from its own config key or memory_opencl_headroom), or 0 for an out-of-range id. Ansel plans against the device's total memory less this.
int	dt_opencl_report_pipe_error (void)
	Record that a pipe run failed on OpenCL, and decide whether to give up on it.
static FILE *	fopen_stat (const char *filename, struct stat *st)
void	dt_opencl_md5sum (const char **files, char **md5sums)
int	dt_opencl_load_program (const int dev, const int prog, const char *filename, const char *binname, const char *cachedir, char *md5sum, char **includemd5, int *loaded_cached)
int	dt_opencl_build_program (const int dev, const int prog, const char *binname, const char *cachedir, char *md5sum, int loaded_cached)
int	dt_opencl_create_kernel (const int prog, const char *name)
void	dt_opencl_free_kernel (const int kernel)
int	dt_opencl_get_max_work_item_sizes (const int dev, size_t *sizes)
int	dt_opencl_get_work_group_limits (const int dev, size_t *sizes, size_t *workgroupsize, unsigned long *localmemsize)
int	dt_opencl_get_kernel_work_group_size (const int dev, const int kernel, size_t *kernelworkgroupsize)
int	dt_opencl_set_kernel_arg (const int dev, const int kernel, const int num, const size_t size, const void *arg)
int	dt_opencl_enqueue_kernel_2d (const int dev, const int kernel, const size_t *sizes)
int	dt_opencl_enqueue_kernel_2d_with_local (const int dev, const int kernel, const size_t *sizes, const size_t *local)
int	dt_opencl_copy_device_to_host (const int devid, void *host, void *device, const int width, const int height, const int bpp)
int	dt_opencl_read_host_from_device (const int devid, void *host, void *device, const int width, const int height, const int bpp)
int	dt_opencl_read_host_from_device_rowpitch (const int devid, void *host, void *device, const int width, const int height, const int rowpitch)
int	dt_opencl_read_host_from_device_non_blocking (const int devid, void *host, void *device, const int width, const int height, const int bpp)
int	dt_opencl_read_host_from_device_rowpitch_non_blocking (const int devid, void *host, void *device, const int width, const int height, const int rowpitch)
int	dt_opencl_read_host_from_device_raw (const int devid, void *host, void *device, const size_t *origin, const size_t *region, const int rowpitch, const int blocking)
int	dt_opencl_write_host_to_device (const int devid, void *host, void *device, const int width, const int height, const int bpp)
int	dt_opencl_write_host_to_device_rowpitch (const int devid, void *host, void *device, const int width, const int height, const int rowpitch)
int	dt_opencl_write_host_to_device_non_blocking (const int devid, void *host, void *device, const int width, const int height, const int bpp)
int	dt_opencl_write_host_to_device_rowpitch_non_blocking (const int devid, void *host, void *device, const int width, const int height, const int rowpitch)
int	dt_opencl_write_host_to_device_raw (const int devid, const void *host, void *device, const size_t *origin, const size_t *region, const int rowpitch, const int blocking)
int	dt_opencl_enqueue_copy_image (const int devid, cl_mem src, cl_mem dst, size_t *orig_src, size_t *orig_dst, size_t *region)
int	dt_opencl_enqueue_copy_image_to_buffer (const int devid, cl_mem src_image, cl_mem dst_buffer, size_t *origin, size_t *region, size_t offset)
int	dt_opencl_enqueue_copy_buffer_to_image (const int devid, cl_mem src_buffer, cl_mem dst_image, size_t offset, size_t *origin, size_t *region)
int	dt_opencl_enqueue_copy_buffer_to_buffer (const int devid, cl_mem src_buffer, cl_mem dst_buffer, size_t srcoffset, size_t dstoffset, size_t size)
int	dt_opencl_read_buffer_from_device (const int devid, void *host, void *device, const size_t offset, const size_t size, const int blocking)
int	dt_opencl_write_buffer_to_device (const int devid, void *host, void *device, const size_t offset, const size_t size, const int blocking)
void *	dt_opencl_copy_host_to_device_constant (const int devid, const size_t size, void *host)
void *	dt_opencl_copy_host_to_device (const int devid, void *host, const int width, const int height, const int bpp)
void *	dt_opencl_copy_host_to_device_rowpitch (const int devid, void *host, const int width, const int height, const int bpp, const int rowpitch)
void	dt_opencl_release_mem_object (cl_mem mem)
void *	dt_opencl_map_buffer (const int devid, cl_mem buffer, const int blocking, const int flags, size_t offset, size_t size)
void *	dt_opencl_map_image (const int devid, cl_mem buffer, const int blocking, const int flags, size_t width, size_t height, int bpp)
int	dt_opencl_unmap_mem_object (const int devid, cl_mem mem_object, void *mapped_ptr)
static void *	_dt_opencl_alloc_image2d (const int devid, const int width, const int height, const size_t bytes, const cl_mem_flags flags, const cl_image_format fmt, void *host, const char *const context)
void *	dt_opencl_alloc_device (const int devid, const int width, const int height, const int bpp)
void *	dt_opencl_alloc_device_use_host_pointer (const int devid, const int width, const int height, const int bpp, void *host, const int flags)
void *	dt_opencl_alloc_device_buffer_with_flags (const int devid, const size_t size, const int flags, void *host_ptr)
void *	dt_opencl_alloc_device_buffer (const int devid, const size_t size)
size_t	dt_opencl_get_mem_object_size (cl_mem mem)
int	dt_opencl_get_mem_context_id (cl_mem mem)
cl_mem_flags	dt_opencl_get_mem_flags (cl_mem mem)
int	dt_opencl_get_image_width (cl_mem mem)
int	dt_opencl_get_image_height (cl_mem mem)
int	dt_opencl_get_image_element_size (cl_mem mem)
void	dt_opencl_memory_statistics (int devid, cl_mem mem, size_t size, dt_opencl_memory_t action)
gboolean	dt_opencl_get_mem_record (cl_mem mem, int *devid, size_t *size)
void	dt_opencl_check_tuning (const int devid)
cl_ulong	dt_opencl_get_device_available (const int devid)
static cl_ulong	_opencl_get_device_memalloc (const int devid)
cl_ulong	dt_opencl_get_device_memalloc (const int devid)
dt_opencl_fit_reason_t	dt_opencl_image_fits_device_reason (const int devid, const size_t width, const size_t height, const unsigned bpp, const float factor, const size_t overhead, size_t *needed, size_t *limit)
gboolean	dt_opencl_image_fits_device (const int devid, const size_t width, const size_t height, const unsigned bpp, const float factor, const size_t overhead)
int	dt_opencl_dev_roundup_width (int size, const int devid)
int	dt_opencl_dev_roundup_height (int size, const int devid)
int	dt_opencl_is_inited (void)
int	dt_opencl_is_enabled (void)
void	dt_opencl_disable (void)
int	dt_opencl_update_settings (void)
cl_event *	dt_opencl_events_get_slot (const int devid, const char *tag)
void	dt_opencl_events_reset (const int devid)
void	dt_opencl_events_wait_for (const int devid)
cl_int	dt_opencl_events_flush (const int devid, const int reset)
void	dt_opencl_events_profiling (const int devid, const int aggregated)
static int	nextpow2 (int n)
int	dt_opencl_local_buffer_opt (const int devid, const int kernel, dt_opencl_local_buffer_t *factors)

Variables
static dt_opencl_t *	_opencl = NULL
static const _gpu_runtime_t	_gpu_runtimes []
static char	_driver_crash_marker [DT_PATH_MAX] = { 0 }

Typedef Documentation

_gpu_runtime_t

typedef struct _gpu_runtime_t _gpu_runtime_t

dt_opencl_mem_record_t

typedef struct dt_opencl_mem_record_t dt_opencl_mem_record_t

dt_opencl_t

typedef struct dt_opencl_t dt_opencl_t

Function Documentation

_ascii_str_canonical()

static char * _ascii_str_canonical ( const char *  in, char *  out, int  maxlen  )

static

Definition at line 1772 of file opencl.c.

References IS_NULL_PTR, n, and out.

Referenced by _device_by_cname(), and dt_opencl_device_init().

_device_by_cname()

static int _device_by_cname ( const char *  name )

static

Definition at line 1750 of file opencl.c.

References _ascii_str_canonical(), _opencl, dt_opencl_device_t::cname, dt_opencl_t::dev, i, name, and dt_opencl_t::num_devs.

Referenced by dt_opencl_priority_parse().

_dt_opencl_alloc_image2d()

static void * _dt_opencl_alloc_image2d ( const int  devid, const int  width, const int  height, const size_t  bytes, const cl_mem_flags  flags, const cl_image_format  fmt, void *  host, const char *const  context  )

inlinestatic

Definition at line 2921 of file opencl.c.

References _opencl, dt_opencl_device_t::context, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clCreateImage2D, DT_DEBUG_OPENCL, dt_dev_pixelpipe_cache_flush_clmem(), dt_opencl_memory_statistics(), dt_print(), flags, height, dt_opencl_t::inited, OPENCL_MEMORY_ADD, dt_dlopencl_t::symbols, and width.

Referenced by dt_opencl_alloc_device(), and dt_opencl_alloc_device_use_host_pointer().

_dt_opencl_get_conf_int()

static int _dt_opencl_get_conf_int ( const gchar *  key_device, const gchar *  conf_name, gboolean *  safety_ok  )

static

Definition at line 277 of file opencl.c.

References dt_conf_get_int(), dt_conf_key_not_empty(), DT_DEBUG_OPENCL, dt_free, dt_print(), FALSE, key, and res.

Referenced by dt_opencl_read_device_config().

_gpu_runtime_crash_streak()

static int _gpu_runtime_crash_streak ( const char *  runtime )

static

Definition at line 1213 of file opencl.c.

References dt_conf_get_int(), IS_NULL_PTR, and key.

Referenced by _opencl_fold_driver_crashes(), dt_opencl_clear_driver_crash_streak(), and dt_opencl_device_init().

_gpu_runtime_of_platform()

static const char * _gpu_runtime_of_platform ( const char *  platform_name )

static

per-runtime GPU-driver crash streak; defined with the runtime table further down, declared here because the per-device setup above it consults them

Definition at line 1197 of file opencl.c.

References _gpu_runtimes, _gpu_runtime_t::conf_suffix, i, and IS_NULL_PTR.

Referenced by dt_opencl_device_init().

_gpu_runtime_set_crash_streak()

static void _gpu_runtime_set_crash_streak ( const char *  runtime, int  value  )

static

Definition at line 1221 of file opencl.c.

References dt_conf_set_int(), IS_NULL_PTR, key, and value.

Referenced by _opencl_fold_driver_crashes(), and dt_opencl_clear_driver_crash_streak().

_opencl_driver_crash_marker_path()

static void _opencl_driver_crash_marker_path ( char *  buf, size_t  buf_len  )

static

Definition at line 1234 of file opencl.c.

References dt_loc_get_user_cache_dir(), and DT_PATH_MAX.

Referenced by dt_opencl_init().

_opencl_fold_driver_crashes()

static void _opencl_fold_driver_crashes ( void  )

static

Definition at line 1273 of file opencl.c.

References _driver_crash_marker, _gpu_runtime_crash_streak(), _gpu_runtime_set_crash_streak(), _gpu_runtimes, _gpu_runtime_t::conf_suffix, i, and IS_NULL_PTR.

Referenced by dt_opencl_init().

_opencl_get_device_memalloc()

static cl_ulong _opencl_get_device_memalloc ( const int  devid )

static

Definition at line 3248 of file opencl.c.

References _opencl, dt_opencl_t::dev, and dt_opencl_device_t::max_mem_alloc.

Referenced by dt_opencl_get_device_memalloc(), and dt_opencl_image_fits_device_reason().

_opencl_splash_update_compile()

static void _opencl_splash_update_compile ( const char *  programname )

inlinestatic

Definition at line 136 of file opencl.c.

References dt_startup_progress_report(), and IS_NULL_PTR.

Referenced by dt_opencl_device_init().

_take_from_list()

static int _take_from_list ( int *  list, int  value  )

static

Definition at line 1733 of file opencl.c.

References value.

Referenced by dt_opencl_priority_parse().

dt_opencl_alloc_device()

void * dt_opencl_alloc_device	(	const int	devid,
		const int	width,
		const int	height,
		const int	bpp
	)

Definition at line 2952 of file opencl.c.

References _dt_opencl_alloc_image2d(), bpp, DT_OPENCL_BPP_DECODE, DT_OPENCL_BPP_IS_RGBA8, height, and width.

Referenced by _blend_layer_over_input_cl(), _cf_joint_stage_cl_selftest(), _cf_stage_cl(), _default_process_tiling_cl_ptp(), _default_process_tiling_cl_roi(), _downsample_guided_laplacian_postfilter_cl(), _drawlayer_acquire_layer_image(), _hf_stage_cl(), _hl_knee_estimate_cl(), _refine_with_detail_mask_cl(), _region_blur_cl_selftest(), _retouch_blur_cl(), box_max_cl(), box_min_cl(), color_smoothing_cl(), dt_bilateral_slice_to_output_cl(), dt_dev_pixelpipe_cache_alloc_cl_device_buffer(), dt_develop_blend_process_cl(), dt_ioppr_transform_image_colorspace_rgb_cl(), dt_local_laplacian_init_cl(), green_equilibration_cl(), guided_filter_cl_impl(), process_cl(), process_cl(), process_cl(), process_default_cl(), process_harmonic_cl(), process_laplacian_bayer_cl(), process_laplacian_passthrough_cl(), process_laplacian_xtrans_cl(), process_markesteijn_cl(), process_nlmeans_cl(), process_rcd_cl(), process_vng_cl(), process_wavelets_cl(), and reconstruct_highlights_cl().

dt_opencl_alloc_device_buffer()

void * dt_opencl_alloc_device_buffer	(	const int	devid,
		const size_t	size
	)

Definition at line 3028 of file opencl.c.

References dt_opencl_alloc_device_buffer_with_flags(), and size.

Referenced by _aniso_stage_cl_selftest(), _cf_harmonic_fill_cl_n(), _cf_harmonic_fill_cl_selftest(), _cf_joint_fit_cl(), _cf_joint_stage_cl(), _cf_joint_stage_cl_selftest(), _cf_pair_stage_cl(), _cf_stage_cl(), _cf_stage_cl_selftest(), _chromaticity_gradient_stage_cl_selftest(), _harmonic_reconstruct_cl(), _hf_stage_cl(), _hf_stage_cl_selftest(), _hl_cl_count_clipped(), _hl_knee_estimate_cl(), _joint_core_stage_cl_selftest(), _knee_cl_selftest(), _refine_with_detail_mask_cl(), _region_guided_filter_cl_selftest(), _retouch_clone_cl(), _retouch_heal_cl(), _selfdome_stage_cl_selftest(), _sp_cl_upload(), _upload_layer_bank(), _weights_cl(), dt_bilateral_init_cl(), dt_gaussian_init_cl(), dt_iop_colorreconstruct_bilateral_init_cl(), dwt_wavelet_decompose_cl(), green_equilibration_cl(), main(), nlmeans_cl_run(), process_cl(), process_harmonic_cl(), process_laplacian_bayer_cl(), process_laplacian_passthrough_cl(), process_laplacian_xtrans_cl(), process_markesteijn_cl(), process_nlmeans_cl(), process_rcd_cl(), process_wavelets_cl(), and rt_upload_scaled_mask_cl().

dt_opencl_alloc_device_buffer_with_flags()

void * dt_opencl_alloc_device_buffer_with_flags	(	const int	devid,
		const size_t	size,
		const int	flags,
		void *	host_ptr
	)

Definition at line 2998 of file opencl.c.

References _opencl, dt_opencl_device_t::context, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clCreateBuffer, DT_DEBUG_OPENCL, dt_dev_pixelpipe_cache_flush_clmem(), dt_opencl_memory_statistics(), dt_print(), flags, dt_opencl_t::inited, OPENCL_MEMORY_ADD, size, and dt_dlopencl_t::symbols.

Referenced by dt_opencl_alloc_device_buffer().

dt_opencl_alloc_device_use_host_pointer()

void * dt_opencl_alloc_device_use_host_pointer	(	const int	devid,
		const int	width,
		const int	height,
		const int	bpp,
		void *	host,
		const int	flags
	)

Definition at line 2975 of file opencl.c.

References _dt_opencl_alloc_image2d(), bpp, DT_OPENCL_BPP_DECODE, DT_OPENCL_BPP_IS_RGBA8, flags, height, and width.

Referenced by _drawlayer_acquire_source_image(), dt_dev_pixelpipe_cache_get_cl_buffer(), and dt_dev_pixelpipe_cache_get_pinned_image().

dt_opencl_apply_scheduling_profile()

static void dt_opencl_apply_scheduling_profile ( )

static

adjust opencl subsystem according to scheduling profile

Definition at line 3379 of file opencl.c.

References _opencl, dt_conf_get_int(), dt_opencl_set_synchronization_timeout(), dt_opencl_update_priorities(), dt_pthread_mutex_lock(), dt_pthread_mutex_unlock(), and dt_opencl_t::lock.

Referenced by dt_opencl_init().

dt_opencl_avoid_atomics()

int dt_opencl_avoid_atomics

(

const int

devid

)

Definition at line 219 of file opencl.c.

References _opencl, dt_opencl_device_t::avoid_atomics, dt_opencl_t::dev, and dt_opencl_t::inited.

Referenced by commit_params().

dt_opencl_build_program()

int dt_opencl_build_program	(	const int	dev,
		const int	prog,
		const char *	binname,
		const char *	cachedir,
		char *	md5sum,
		int	loaded_cached
	)

builds the given program.

defined(_WIN32)

Definition at line 2362 of file opencl.c.

References _opencl, dt_opencl_t::dev, dt_opencl_device_t::devid, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clBuildProgram, dt_dlopencl_symbols_t::dt_clGetProgramBuildInfo, dt_dlopencl_symbols_t::dt_clGetProgramInfo, DT_DEBUG_OPENCL, dt_free, DT_OPENCL_MAX_PROGRAMS, DT_PATH_MAX, dt_print(), dt_vprint(), f, i, IS_NULL_PTR, dt_opencl_device_t::options, dt_opencl_device_t::program, and dt_dlopencl_t::symbols.

Referenced by dt_opencl_device_init().

dt_opencl_check_tuning()

void dt_opencl_check_tuning

(

const int

devid

)

check tuning settings and available memory for the device

Definition at line 3186 of file opencl.c.

References _opencl, dt_opencl_t::dev, DT_DEBUG_MEMORY, DT_DEBUG_OPENCL, dt_print(), dt_opencl_device_t::forced_headroom, dt_opencl_t::inited, MAX, dt_opencl_device_t::max_global_mem, dt_opencl_device_t::name, and dt_opencl_device_t::used_available.

Referenced by dt_dev_pixelpipe_process(), and dt_opencl_set_detected_device_headroom().

dt_opencl_cleanup()

void dt_opencl_cleanup

(

void

)

cleans up the opencl subsystem. Touches no conf: conf may already be gone.

Definition at line 1642 of file opencl.c.

References _opencl, dt_opencl_detected_device_t::cname, dt_opencl_t::detected_devs, dt_opencl_t::dev, dt_opencl_t::dev_priority_export, dt_opencl_t::dev_priority_image, dt_opencl_t::dev_priority_preview, dt_opencl_t::dev_priority_thumbnail, dt_opencl_t::dlocl, dt_bilateral_free_cl_global(), dt_colorspaces_free_cl_global(), dt_develop_blend_free_cl_global(), dt_dwt_free_cl_global(), dt_free, dt_gaussian_free_cl_global(), dt_guided_filter_free_cl_global(), dt_heal_free_cl_global(), dt_interpolation_free_cl_global(), dt_local_laplacian_free_cl_global(), dt_opencl_cleanup_device(), dt_pthread_mutex_destroy(), i, dt_opencl_t::inited, IS_NULL_PTR, dt_dlopencl_t::library, dt_opencl_t::lock, dt_opencl_t::mem_sizes, dt_opencl_t::mem_sizes_lock, dt_opencl_detected_device_t::name, dt_opencl_t::num_detected_devs, dt_opencl_t::num_devs, and dt_dlopencl_t::symbols.

Referenced by dt_cleanup().

dt_opencl_cleanup_device()

static void dt_opencl_cleanup_device ( dt_opencl_t *  cl, int  i  )

static

Definition at line 1567 of file opencl.c.

References dt_opencl_device_t::cmd_queue, dt_opencl_device_t::cname, dt_opencl_device_t::context, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clReleaseCommandQueue, dt_dlopencl_symbols_t::dt_clReleaseContext, dt_dlopencl_symbols_t::dt_clReleaseKernel, dt_dlopencl_symbols_t::dt_clReleaseProgram, DT_DEBUG_MEMORY, DT_DEBUG_OPENCL, dt_free, dt_get_debug_flags(), dt_opencl_events_reset(), DT_OPENCL_MAX_KERNELS, DT_OPENCL_MAX_PROGRAMS, dt_print_nts(), dt_pthread_mutex_destroy(), dt_opencl_device_t::eventlist, dt_opencl_device_t::eventtags, i, IS_NULL_PTR, k, dt_opencl_device_t::kernel, dt_opencl_device_t::kernel_used, dt_opencl_device_t::lock, dt_opencl_device_t::maxeventslot, dt_opencl_device_t::name, dt_opencl_device_t::options, dt_opencl_device_t::options_md5, dt_opencl_device_t::peak_memory, dt_opencl_t::print_statistics, dt_opencl_device_t::program, dt_opencl_device_t::program_used, dt_dlopencl_t::symbols, dt_opencl_device_t::totalevents, dt_opencl_device_t::totallost, dt_opencl_device_t::totalsuccess, dt_opencl_device_t::use_events, and dt_opencl_device_t::vendor.

Referenced by dt_opencl_cleanup(), and dt_opencl_init().

dt_opencl_clear_driver_crash_streak()

void dt_opencl_clear_driver_crash_streak

(

void

)

Clears the persisted driver-crash streak after a session in which OpenCL ran to a clean shutdown. Reads and writes conf: call it from the shutdown sequence BEFORE dt_conf_cleanup(), and before dt_opencl_cleanup().

Definition at line 1616 of file opencl.c.

References _gpu_runtime_crash_streak(), _gpu_runtime_set_crash_streak(), _opencl, dt_opencl_t::dev, dt_opencl_device_t::disabled, dt_opencl_t::enabled, i, dt_opencl_t::inited, IS_NULL_PTR, dt_opencl_t::num_devs, and dt_opencl_device_t::runtime_id.

Referenced by dt_cleanup().

dt_opencl_copy_device_to_host()

int dt_opencl_copy_device_to_host	(	const int	devid,
		void *	host,
		void *	device,
		const int	width,
		const int	height,
		const int	bpp
	)

HAVE_OPENCL mode only: copy and alloc buffers.

Definition at line 2639 of file opencl.c.

References bpp, dt_opencl_read_host_from_device(), height, and width.

Referenced by _harmonic_cl_roundtrip(), _region_blur_cl_selftest(), dt_colorspaces_apply_profile_cl(), dt_develop_blend_process_cl(), dt_ioppr_transform_image_colorspace_rgb_cl(), process_cl(), and process_harmonic_cl().

dt_opencl_copy_host_to_device()

void * dt_opencl_copy_host_to_device	(	const int	devid,
		void *	host,
		const int	width,
		const int	height,
		const int	bpp
	)

Definition at line 2823 of file opencl.c.

References bpp, dt_opencl_copy_host_to_device_rowpitch(), height, and width.

Referenced by _drawlayer_acquire_source_image(), dt_colorspaces_apply_profile_cl(), dt_ioppr_build_iccprofile_params_cl(), dt_ioppr_transform_image_colorspace_rgb_cl(), and process_cl().

dt_opencl_copy_host_to_device_constant()

void * dt_opencl_copy_host_to_device_constant	(	const int	devid,
		const size_t	size,
		void *	host
	)

Definition at line 2808 of file opencl.c.

References _opencl, dt_opencl_device_t::context, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clCreateBuffer, DT_DEBUG_OPENCL, dt_opencl_memory_statistics(), dt_print(), dt_opencl_t::inited, OPENCL_MEMORY_ADD, size, and dt_dlopencl_t::symbols.

Referenced by _hl_cl_count_clipped(), _hl_cl_lch_xtrans(), _hl_cl_visualize(), _refine_with_detail_mask_cl(), _retouch_fill_cl(), _upload_layer_bank(), apply_global_distortion_map_cl(), dt_colorspaces_apply_profile_cl(), dt_develop_blend_process_cl(), dt_interpolation_resample_cl(), dt_ioppr_build_iccprofile_params_cl(), dt_ioppr_transform_image_colorspace_rgb_cl(), dual_demosaic_cl(), process_cl(), process_cl(), process_harmonic_cl(), process_laplacian_bayer_cl(), process_laplacian_passthrough_cl(), process_laplacian_xtrans_cl(), process_markesteijn_cl(), process_vng_cl(), process_wavelets_cl(), rt_copy_image_masked_cl(), rt_copy_in_to_out_cl(), and rt_copy_mask_to_alpha_cl().

dt_opencl_copy_host_to_device_rowpitch()

void * dt_opencl_copy_host_to_device_rowpitch	(	const int	devid,
		void *	host,
		const int	width,
		const int	height,
		const int	bpp,
		const int	rowpitch
	)

Todo:

: if fmt = uint16_t, blow up to 4xuint16_t and copy manually!

Definition at line 2829 of file opencl.c.

References _opencl, bpp, dt_opencl_device_t::context, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clCreateImage2D, DT_DEBUG_OPENCL, dt_opencl_memory_statistics(), dt_print(), height, dt_opencl_t::inited, OPENCL_MEMORY_ADD, dt_dlopencl_t::symbols, and width.

Referenced by dt_opencl_copy_host_to_device().

dt_opencl_create_kernel()

int dt_opencl_create_kernel	(	const int	program,
		const char *	name
	)

inits a kernel. returns the index or -1 if fail.

Definition at line 2506 of file opencl.c.

References _opencl, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clCreateKernel, DT_DEBUG_OPENCL, DT_OPENCL_MAX_KERNELS, DT_OPENCL_MAX_PROGRAMS, dt_print(), dt_pthread_mutex_lock(), dt_pthread_mutex_unlock(), dt_vprint(), error(), dt_opencl_t::inited, k, dt_opencl_device_t::kernel, dt_opencl_device_t::kernel_used, dt_opencl_t::lock, name, dt_opencl_t::num_devs, dt_opencl_device_t::program, and dt_dlopencl_t::symbols.

Referenced by dt_bilateral_init_cl_global(), dt_colorspaces_init_cl_global(), dt_develop_blend_init_cl_global(), dt_dwt_init_cl_global(), dt_gaussian_init_cl_global(), dt_guided_filter_init_cl_global(), dt_interpolation_init_cl_global(), dt_local_laplacian_init_cl_global(), dt_nn_cl_create(), init_global(), and init_global().

dt_opencl_detected_device_enabled()

gboolean dt_opencl_detected_device_enabled

(

const int

detected

)

Return whether a detected GPU is enabled by its per-device OpenCL preference.

Definition at line 415 of file opencl.c.

References dt_opencl_detected_device_t::cname, dt_opencl_detected_device_t::config_id, dt_opencl_detected_device_t::disabled, DT_CLDEVICE_HEAD, dt_conf_get_int(), dt_conf_key_not_empty(), dt_opencl_get_detected_device(), FALSE, IS_NULL_PTR, and key.

Referenced by dt_gui_preferences_show(), and dt_opencl_set_detected_device_enabled().

dt_opencl_detected_device_headroom()

size_t dt_opencl_detected_device_headroom

(

const int

detected

)

Return the per-device GPU vRAM headroom in MiB.

Definition at line 494 of file opencl.c.

References dt_opencl_detected_device_t::cname, dt_opencl_detected_device_t::config_id, DT_CLDEVICE_HEAD, dt_conf_get_int(), dt_conf_key_not_empty(), dt_opencl_get_detected_device(), dt_opencl_detected_device_t::forced_headroom, IS_NULL_PTR, and key.

Referenced by dt_gui_preferences_show().

dt_opencl_detected_device_pinned_memory()

gboolean dt_opencl_detected_device_pinned_memory

(

const int

detected

)

Return whether pinned memory transfer is enabled for a detected GPU.

Definition at line 460 of file opencl.c.

References dt_opencl_detected_device_t::cname, dt_opencl_detected_device_t::config_id, DT_CLDEVICE_HEAD, dt_conf_get_int(), dt_conf_key_not_empty(), dt_opencl_get_detected_device(), DT_OPENCL_PINNING_ON, FALSE, IS_NULL_PTR, key, and dt_opencl_detected_device_t::pinned_memory.

Referenced by dt_gui_preferences_show().

dt_opencl_dev_roundup_height()

int dt_opencl_dev_roundup_height	(	int	size,
		const int	devid
	)

Definition at line 3321 of file opencl.c.

References _opencl, dt_opencl_device_t::clroundup_ht, dt_opencl_t::dev, and size.

dt_opencl_dev_roundup_width()

int dt_opencl_dev_roundup_width	(	int	size,
		const int	devid
	)

round size to a multiple of the value given in the device specifig config parameter clroundup_wd/ht

Definition at line 3316 of file opencl.c.

References _opencl, dt_opencl_device_t::clroundup_wd, dt_opencl_t::dev, and size.

dt_opencl_device_init()

static int dt_opencl_device_init ( dt_opencl_t *  cl, const int  dev, cl_device_id *  devices, const int  k  )

static

Todo:

: compile as a job?

Definition at line 532 of file opencl.c.

References _ascii_str_canonical(), _gpu_runtime_crash_streak(), _gpu_runtime_of_platform(), _opencl_splash_update_compile(), dt_opencl_device_t::avoid_atomics, dt_opencl_device_t::clroundup_ht, dt_opencl_device_t::clroundup_wd, dt_opencl_device_t::cltype, dt_opencl_detected_device_t::cltype, dt_opencl_device_t::cmd_queue, dt_opencl_device_t::cname, dt_opencl_detected_device_t::cname, dt_opencl_detected_device_t::config_id, dt_opencl_device_t::context, dt_opencl_t::crc, dt_opencl_t::detected_devs, dt_opencl_t::dev, dt_opencl_device_t::devid, dt_opencl_device_t::disabled, dt_opencl_detected_device_t::disabled, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clCreateCommandQueue, dt_dlopencl_symbols_t::dt_clCreateContext, DT_CLDEVICE_HEAD, dt_dlopencl_symbols_t::dt_clGetDeviceInfo, dt_dlopencl_symbols_t::dt_clGetPlatformInfo, dt_dlopencl_symbols_t::dt_clReleaseCommandQueue, dt_dlopencl_symbols_t::dt_clReleaseContext, dt_dlopencl_symbols_t::dt_clReleaseKernel, dt_dlopencl_symbols_t::dt_clReleaseProgram, dt_concat_path_file(), dt_conf_flush(), dt_conf_get_bool(), dt_conf_get_string_const(), dt_conf_key_exists(), dt_conf_set_string(), DT_DEBUG_ALWAYS, DT_DEBUG_OPENCL, DT_DEBUG_PERF, dt_free, dt_get_debug_flags(), dt_get_wtime(), dt_loc_get_kerneldir(), dt_loc_get_user_cache_dir(), dt_nvidia_gpu_supports_sm_20(), dt_opencl_build_program(), DT_OPENCL_CBUFFSIZE, dt_opencl_check_driver_blacklist(), DT_OPENCL_DEFAULT_COMPILE, DT_OPENCL_DEFAULT_COMPILE_AMD, DT_OPENCL_DEFAULT_COMPILE_INTEL, DT_OPENCL_DEFAULT_COMPILE_NVIDIA, dt_opencl_get_device_info(), dt_opencl_get_vendor_by_id(), dt_opencl_load_program(), DT_OPENCL_MAX_INCLUDES, DT_OPENCL_MAX_KERNELS, DT_OPENCL_MAX_PROGRAMS, dt_opencl_md5sum(), DT_OPENCL_PINNING_ON, dt_opencl_read_device_config(), DT_OPENCL_VENDOR_AMD, DT_OPENCL_VENDOR_INTEL, DT_OPENCL_VENDOR_NVIDIA, dt_opencl_write_device_config(), DT_PATH_MAX, dt_print(), dt_print_nts(), dt_pthread_mutex_destroy(), dt_pthread_mutex_init(), dt_util_str_replace(), dt_vprint(), dt_opencl_device_t::event_handles, dt_opencl_device_t::eventlist, dt_opencl_device_t::eventsconsolidated, dt_opencl_device_t::eventtags, f, FALSE, dt_opencl_device_t::forced_headroom, dt_opencl_detected_device_t::forced_headroom, dt_opencl_device_t::host_unified_memory, dt_opencl_detected_device_t::host_unified_memory, i, IS_NULL_PTR, k, dt_opencl_device_t::kernel, dt_opencl_device_t::kernel_used, dt_opencl_device_t::lock, dt_opencl_device_t::lostevents, dt_opencl_device_t::max_global_mem, dt_opencl_device_t::max_image_height, dt_opencl_device_t::max_image_width, dt_opencl_device_t::max_mem_alloc, dt_opencl_device_t::maxevents, dt_opencl_device_t::maxeventslot, dt_opencl_device_t::memory_in_use, dt_opencl_device_t::micro_nap, MIN, n, dt_opencl_device_t::name, dt_opencl_detected_device_t::name, dt_opencl_t::num_detected_devs, dt_opencl_device_t::numevents, dt_opencl_device_t::nvidia_sm_20, dt_opencl_device_t::options, dt_opencl_device_t::options_md5, dt_opencl_device_t::peak_memory, dt_opencl_device_t::pinned_memory, dt_opencl_detected_device_t::pinned_memory, dt_opencl_device_t::program, dt_opencl_device_t::program_used, res, dt_opencl_device_t::runtime_error, dt_opencl_device_t::runtime_id, dt_opencl_device_t::summary, dt_dlopencl_t::symbols, dt_opencl_device_t::totalevents, dt_opencl_device_t::totallost, dt_opencl_device_t::totalsuccess, TRUE, type, dt_opencl_device_t::use_events, dt_opencl_device_t::used_available, dt_opencl_device_t::used_global_mem, dt_opencl_device_t::vendor, and dt_opencl_device_t::vendor_id.

Referenced by dt_opencl_init().

dt_opencl_disable()

void dt_opencl_disable

(

void

)

disable opencl

Definition at line 3343 of file opencl.c.

References _opencl, dt_conf_set_bool(), dt_opencl_t::enabled, FALSE, and dt_opencl_t::inited.

dt_opencl_enqueue_barrier()

int dt_opencl_enqueue_barrier

(

const int

devid

)

enqueues a synchronization point.

Definition at line 1726 of file opencl.c.

References _opencl, dt_opencl_device_t::cmd_queue, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clEnqueueBarrier, dt_opencl_t::inited, and dt_dlopencl_t::symbols.

dt_opencl_enqueue_copy_buffer_to_buffer()

int dt_opencl_enqueue_copy_buffer_to_buffer	(	const int	devid,
		cl_mem	src_buffer,
		cl_mem	dst_buffer,
		size_t	srcoffset,
		size_t	dstoffset,
		size_t	size
	)

Definition at line 2772 of file opencl.c.

References _opencl, dt_opencl_device_t::cmd_queue, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clEnqueueCopyBuffer, DT_DEBUG_OPENCL, dt_opencl_events_get_slot(), dt_print(), dt_opencl_t::inited, size, and dt_dlopencl_t::symbols.

Referenced by _simulate_channel_cl(), _unet_forward_cl(), dt_bilateral_blur_cl(), dt_iop_colorreconstruct_bilateral_blur_cl(), dwt_get_image_layer_cl(), process_cl(), and process_markesteijn_cl().

dt_opencl_enqueue_copy_buffer_to_image()

int dt_opencl_enqueue_copy_buffer_to_image	(	const int	devid,
		cl_mem	src_buffer,
		cl_mem	dst_image,
		size_t	offset,
		size_t *	origin,
		size_t *	region
	)

Definition at line 2760 of file opencl.c.

References _opencl, dt_opencl_device_t::cmd_queue, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clEnqueueCopyBufferToImage, DT_DEBUG_OPENCL, dt_opencl_events_get_slot(), dt_print(), dt_opencl_t::inited, and dt_dlopencl_t::symbols.

Referenced by _cf_stage_cl(), _region_edge_blur_cl(), dt_gaussian_blur_cl(), process_cl(), and process_harmonic_cl().

dt_opencl_enqueue_copy_image()

int dt_opencl_enqueue_copy_image	(	const int	devid,
		cl_mem	src,
		cl_mem	dst,
		size_t *	orig_src,
		size_t *	orig_dst,
		size_t *	region
	)

Definition at line 2737 of file opencl.c.

References _opencl, dt_opencl_device_t::cmd_queue, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clEnqueueCopyImage, DT_DEBUG_OPENCL, dt_opencl_events_get_slot(), dt_print(), dt_opencl_t::inited, and dt_dlopencl_t::symbols.

Referenced by _blend_layer_over_input_cl(), _drawlayer_copy_or_resample_layer_roi(), color_smoothing_cl(), dt_bilateral_slice_to_output_cl(), dt_develop_blend_process_cl(), dt_interpolation_resample_cl(), dt_ioppr_transform_image_colorspace_rgb_cl(), process_cl(), process_cl(), process_markesteijn_cl(), process_vng_cl(), process_wavelets_cl(), and reconstruct_highlights_cl().

dt_opencl_enqueue_copy_image_to_buffer()

int dt_opencl_enqueue_copy_image_to_buffer	(	const int	devid,
		cl_mem	src_image,
		cl_mem	dst_buffer,
		size_t *	origin,
		size_t *	region,
		size_t	offset
	)

Definition at line 2748 of file opencl.c.

References _opencl, dt_opencl_device_t::cmd_queue, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clEnqueueCopyImageToBuffer, DT_DEBUG_OPENCL, dt_opencl_events_get_slot(), dt_print(), dt_opencl_t::inited, and dt_dlopencl_t::symbols.

Referenced by _cf_stage_cl(), _region_edge_blur_cl(), dt_gaussian_blur_cl(), process_cl(), and process_harmonic_cl().

dt_opencl_enqueue_kernel_2d()

int dt_opencl_enqueue_kernel_2d	(	const int	dev,
		const int	kernel,
		const size_t *	sizes
	)

launch kernel!

Definition at line 2612 of file opencl.c.

References dt_opencl_enqueue_kernel_2d_with_local(), and kernel().

Referenced by _cf_harmonic_fill_cl_n(), _cf_joint_fit_cl(), _cf_joint_stage_cl(), _cf_joint_stage_cl_selftest(), _cf_pair_stage_cl(), _cf_stage_cl(), _downsample_guided_laplacian_postfilter_cl(), _drawlayer_run_premult_over_kernel(), _harmonic_reconstruct_cl(), _hf_stage_cl(), _hl_cl_clip(), _hl_cl_lch_bayer(), _hl_cl_visualize(), _hl_knee_apply_cfa_cl(), _hl_knee_estimate_cl(), _refine_with_detail_mask_cl(), _region_edge_blur_cl(), _retouch_blur_cl(), _retouch_fill_cl(), _simulate_channel_cl(), _sp_chol_factor_cl(), _upsample_cl(), apply_global_distortion_map_cl(), box_max_cl(), box_min_cl(), cl_box_mean(), cl_covariances(), cl_generate_result(), cl_solve(), cl_split_rgb(), cl_update_covariance(), cl_variances(), dehaze_cl(), dt_bilateral_blur_cl(), dt_bilateral_init_cl(), dt_bilateral_slice_cl(), dt_bilateral_slice_to_output_cl(), dt_colorspaces_apply_profile_cl(), dt_develop_blend_process_cl(), dt_gaussian_blur_cl(), dt_iop_colorreconstruct_bilateral_blur_cl(), dt_iop_colorreconstruct_bilateral_init_cl(), dt_iop_colorreconstruct_bilateral_slice_cl(), dt_ioppr_transform_image_colorspace_rgb_cl(), dt_local_laplacian_cl(), dual_demosaic_cl(), dwt_add_layer_cl(), dwt_subtract_layer_cl(), dwt_wavelet_decompose_cl(), green_equilibration_cl(), nlmeans_cl_accu(), nlmeans_cl_init(), nlmeans_cl_run(), process_cl(), process_cl(), process_cl(), process_default_cl(), process_harmonic_cl(), process_laplacian_bayer_cl(), process_laplacian_passthrough_cl(), process_laplacian_xtrans_cl(), process_markesteijn_cl(), process_nlmeans_cl(), process_rcd_cl(), process_vng_cl(), process_wavelets_cl(), reconstruct_highlights_cl(), rt_copy_image_masked_cl(), rt_copy_in_to_out_cl(), rt_copy_mask_to_alpha_cl(), transition_map_cl(), wavelets_process_cl(), and wavelets_process_cl().

dt_opencl_enqueue_kernel_2d_with_local()

int dt_opencl_enqueue_kernel_2d_with_local	(	const int	dev,
		const int	kernel,
		const size_t *	sizes,
		const size_t *	local
	)

launch kernel with defined local size!

Definition at line 2618 of file opencl.c.

References _opencl, dt_opencl_device_t::cmd_queue, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clEnqueueNDRangeKernel, dt_dlopencl_symbols_t::dt_clGetKernelInfo, DT_DEBUG_OPENCL, dt_get_debug_flags(), dt_opencl_events_get_slot(), DT_OPENCL_MAX_KERNELS, dt_print(), dt_opencl_t::inited, kernel(), dt_opencl_device_t::kernel, and dt_dlopencl_t::symbols.

Referenced by _cf_harmonic_fill_cl_n(), _conv_cl(), _downsample_guided_laplacian_postfilter_cl(), _hl_cl_count_clipped(), _hl_cl_lch_xtrans(), _sp_chol_factor_cl(), _sp_chol_solve_cl(), color_smoothing_cl(), dt_bilateral_splat_cl(), dt_gaussian_blur_cl(), dt_interpolation_resample_cl(), dt_iop_colorreconstruct_bilateral_splat_cl(), dt_opencl_enqueue_kernel_2d(), green_equilibration_cl(), nlmeans_cl_horiz(), nlmeans_cl_run(), process_cl(), process_default_cl(), process_laplacian_bayer_cl(), process_laplacian_passthrough_cl(), process_laplacian_xtrans_cl(), process_markesteijn_cl(), process_nlmeans_cl(), process_rcd_cl(), process_vng_cl(), process_wavelets_cl(), reconstruct_highlights_cl(), wavelets_process_cl(), and wavelets_process_cl().

dt_opencl_events_flush()

cl_int dt_opencl_events_flush	(	const int	devid,
		const int	reset
	)

Wait for events in eventlist to terminate, check for return status and profiling info of events. If "reset" is TRUE report summary info (would be CL_COMPLETE or last error code) and print profiling info if needed. If "reset" is FALSE just store info (success value, profiling) from terminated events and release events for re-use by OpenCL driver.

Definition at line 3584 of file opencl.c.

References _opencl, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clGetEventInfo, dt_dlopencl_symbols_t::dt_clGetEventProfilingInfo, dt_dlopencl_symbols_t::dt_clReleaseEvent, DT_DEBUG_OPENCL, DT_DEBUG_PERF, dt_get_debug_flags(), dt_opencl_events_profiling(), dt_opencl_events_reset(), dt_opencl_events_wait_for(), dt_print(), dt_opencl_device_t::eventlist, dt_opencl_device_t::eventsconsolidated, dt_opencl_device_t::eventtags, FALSE, dt_opencl_t::inited, IS_NULL_PTR, k, dt_opencl_device_t::lostevents, dt_opencl_device_t::numevents, reset(), dt_opencl_device_t::summary, dt_dlopencl_t::symbols, dt_opencl_device_t::totalsuccess, and dt_opencl_device_t::use_events.

Referenced by dt_dev_pixelpipe_process(), dt_opencl_events_get_slot(), and dt_opencl_finish().

dt_opencl_events_get_slot()

cl_event * dt_opencl_events_get_slot	(	const int	devid,
		const char *	tag
	)

the following eventlist functions assume that affected structures are locked upstream get next free slot in eventlist (and manage size of eventlist)

Definition at line 3391 of file opencl.c.

References _opencl, dt_opencl_t::dev, DT_DEBUG_OPENCL, dt_free, DT_OPENCL_EVENTLISTSIZE, DT_OPENCL_EVENTNAMELENGTH, dt_opencl_events_flush(), dt_print(), dt_opencl_device_t::event_handles, dt_opencl_device_t::eventlist, dt_opencl_device_t::eventsconsolidated, dt_opencl_device_t::eventtags, dt_opencl_t::inited, IS_NULL_PTR, dt_opencl_device_t::lostevents, MAX, dt_opencl_device_t::maxevents, dt_opencl_device_t::maxeventslot, dt_opencl_device_t::numevents, dt_opencl_device_t::totalevents, dt_opencl_device_t::totallost, dt_opencl_device_t::use_events, and void().

Referenced by dt_opencl_enqueue_copy_buffer_to_buffer(), dt_opencl_enqueue_copy_buffer_to_image(), dt_opencl_enqueue_copy_image(), dt_opencl_enqueue_copy_image_to_buffer(), dt_opencl_enqueue_kernel_2d_with_local(), dt_opencl_map_buffer(), dt_opencl_map_image(), dt_opencl_read_buffer_from_device(), dt_opencl_read_host_from_device_raw(), dt_opencl_unmap_mem_object(), dt_opencl_write_buffer_to_device(), and dt_opencl_write_host_to_device_raw().

dt_opencl_events_profiling()

void dt_opencl_events_profiling	(	const int	devid,
		const int	aggregated
	)

display OpenCL profiling information. If "aggregated" is TRUE, try to generate summarized info for each kernel

Definition at line 3687 of file opencl.c.

References _opencl, dt_opencl_t::dev, DT_DEBUG_OPENCL, dt_free, DT_OPENCL_EVENTNAMELENGTH, dt_print(), dt_opencl_device_t::eventlist, dt_opencl_device_t::eventsconsolidated, dt_opencl_device_t::eventtags, i, dt_opencl_t::inited, IS_NULL_PTR, items, k, dt_opencl_device_t::lostevents, dt_opencl_device_t::name, dt_opencl_device_t::numevents, and dt_opencl_device_t::use_events.

Referenced by dt_opencl_events_flush().

dt_opencl_events_reset()

void dt_opencl_events_reset

(

const int

devid

)

reset eventlist to empty state

Definition at line 3489 of file opencl.c.

References _opencl, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clReleaseEvent, dt_opencl_device_t::eventlist, dt_opencl_device_t::eventsconsolidated, dt_opencl_device_t::eventtags, dt_opencl_t::inited, IS_NULL_PTR, k, dt_opencl_device_t::lostevents, dt_opencl_device_t::maxevents, dt_opencl_device_t::numevents, dt_opencl_device_t::summary, dt_dlopencl_t::symbols, and dt_opencl_device_t::use_events.

Referenced by dt_dev_pixelpipe_process(), dt_opencl_cleanup_device(), and dt_opencl_events_flush().

dt_opencl_events_wait_for()

void dt_opencl_events_wait_for

(

const int

devid

)

Wait for events in eventlist to terminate -> this is a blocking synchronization point! Does not flush eventlist. Side effect: might adjust numevents.

Todo:

: run clWaitForEvents in separate thread and implement watchdog timer

Definition at line 3527 of file opencl.c.

References _opencl, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clWaitForEvents, DT_DEBUG_OPENCL, dt_vprint(), dt_opencl_device_t::eventlist, dt_opencl_device_t::eventsconsolidated, dt_opencl_t::inited, IS_NULL_PTR, k, dt_opencl_device_t::lostevents, dt_opencl_device_t::numevents, dt_dlopencl_t::symbols, dt_opencl_device_t::totallost, and dt_opencl_device_t::use_events.

Referenced by dt_dev_pixelpipe_cache_flush_clmem(), dt_dev_pixelpipe_cache_flush_host_pinned_image(), dt_dev_pixelpipe_cache_prepare_cl_input(), and dt_opencl_events_flush().

dt_opencl_finish()

gboolean dt_opencl_finish

(

const int

devid

)

cleans up the i-th device in the cl->dev list both finish functions return TRUE in case of success cleans up command queue.

Definition at line 1712 of file opencl.c.

References _opencl, dt_opencl_device_t::cmd_queue, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clFinish, dt_opencl_events_flush(), FALSE, dt_opencl_t::inited, and dt_dlopencl_t::symbols.

Referenced by _blend_layer_over_input_cl(), _default_process_tiling_cl_ptp(), _default_process_tiling_cl_roi(), _free_cache_entry(), _harmonic_reconstruct_cl(), _sp_chol_factor_cl(), _unet_forward_cl(), dt_cleanup(), nlmeans_cl_run(), pixelpipe_process_on_GPU(), process_cl(), process_harmonic_cl(), process_laplacian_bayer_cl(), process_laplacian_passthrough_cl(), process_laplacian_xtrans_cl(), and process_nlmeans_cl().

dt_opencl_free_kernel()

void dt_opencl_free_kernel

(

const int

kernel

)

releases kernel resources again.

Definition at line 2549 of file opencl.c.

References _opencl, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clReleaseKernel, DT_OPENCL_MAX_KERNELS, dt_pthread_mutex_lock(), dt_pthread_mutex_unlock(), dt_opencl_t::inited, kernel(), dt_opencl_device_t::kernel, dt_opencl_device_t::kernel_used, dt_opencl_t::lock, dt_opencl_t::num_devs, and dt_dlopencl_t::symbols.

Referenced by cleanup_global(), cleanup_global(), dt_bilateral_free_cl_global(), dt_colorspaces_free_cl_global(), dt_develop_blend_free_cl_global(), dt_dwt_free_cl_global(), dt_gaussian_free_cl_global(), dt_guided_filter_free_cl_global(), dt_interpolation_free_cl_global(), dt_local_laplacian_free_cl_global(), and dt_nn_cl_destroy().

dt_opencl_get_detected_device()

const dt_opencl_detected_device_t * dt_opencl_get_detected_device

(

const int

detected

)

Return metadata for a detected GPU, or NULL when the detected-device index is invalid.

Definition at line 407 of file opencl.c.

References _opencl, dt_opencl_t::detected_devs, IS_NULL_PTR, and dt_opencl_t::num_detected_devs.

Referenced by dt_gui_preferences_show(), dt_opencl_detected_device_enabled(), dt_opencl_detected_device_headroom(), dt_opencl_detected_device_pinned_memory(), dt_opencl_set_detected_device_enabled(), dt_opencl_set_detected_device_headroom(), and dt_opencl_set_detected_device_pinned_memory().

dt_opencl_get_detected_device_count()

int dt_opencl_get_detected_device_count

(

void

)

Number of GPU devices detected during OpenCL probing, including devices disabled by user preference.

Definition at line 399 of file opencl.c.

References _opencl, IS_NULL_PTR, and dt_opencl_t::num_detected_devs.

Referenced by dt_gui_preferences_show().

dt_opencl_get_device_available()

cl_ulong dt_opencl_get_device_available

(

const int

devid

)

get available memory for the device

Definition at line 3203 of file opencl.c.

References _opencl, dt_opencl_t::dev, dt_get_system_available_mem(), dt_opencl_device_t::forced_headroom, dt_opencl_device_t::host_unified_memory, dt_opencl_t::inited, dt_opencl_device_t::max_mem_alloc, dt_opencl_device_t::memory_in_use, MIN, and dt_opencl_device_t::used_available.

Referenced by _default_process_tiling_cl_ptp(), _default_process_tiling_cl_roi(), dt_opencl_image_fits_device_reason(), and pixelpipe_process_on_GPU().

dt_opencl_get_device_headroom()

size_t dt_opencl_get_device_headroom

(

const int

devid

)

Bytes of the device left to other applications (its forced_headroom, from its own config key or memory_opencl_headroom), or 0 for an out-of-range id. Ansel plans against the device's total memory less this.

Definition at line 2084 of file opencl.c.

References _opencl, dt_opencl_t::dev, dt_opencl_device_t::forced_headroom, dt_opencl_t::inited, IS_NULL_PTR, and dt_opencl_t::num_devs.

Referenced by _add_device_section().

dt_opencl_get_device_info()

static int dt_opencl_get_device_info ( dt_opencl_t *  cl, cl_device_id  device, cl_device_info  param_name, void **  param_value, size_t *  param_value_size  )

static

Todo:

: anything better?

Definition at line 162 of file opencl.c.

References dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clGetDeviceInfo, DT_DEBUG_OPENCL, dt_free, dt_print(), error(), IS_NULL_PTR, and dt_dlopencl_t::symbols.

Referenced by dt_opencl_device_init().

dt_opencl_get_device_max_global_mem()

size_t dt_opencl_get_device_max_global_mem

(

const int

devid

)

Total device memory in bytes, or 0 for an out-of-range id.

Definition at line 2068 of file opencl.c.

References _opencl, dt_opencl_t::dev, dt_opencl_t::inited, IS_NULL_PTR, dt_opencl_device_t::max_global_mem, and dt_opencl_t::num_devs.

Referenced by _add_device_section().

dt_opencl_get_device_max_image_size()

gboolean dt_opencl_get_device_max_image_size	(	const int	devid,
		int *	width,
		int *	height
	)

Largest 2D image the device will accept, which is what bounds tile size.

Parameters

devid	device id.
width	receives the maximum width. Not written for an out-of-range id.
height	receives the maximum height. Same.

Returns

TRUE when both were written.

Definition at line 2058 of file opencl.c.

References _opencl, dt_opencl_t::dev, FALSE, height, dt_opencl_t::inited, IS_NULL_PTR, dt_opencl_device_t::max_image_height, dt_opencl_device_t::max_image_width, dt_opencl_t::num_devs, TRUE, and width.

Referenced by _default_process_tiling_cl_ptp(), and _default_process_tiling_cl_roi().

dt_opencl_get_device_memalloc()

cl_ulong dt_opencl_get_device_memalloc

(

const int

devid

)

get size of allocatable single buffer

Definition at line 3253 of file opencl.c.

References _opencl, _opencl_get_device_memalloc(), and dt_opencl_t::inited.

Referenced by _default_process_tiling_cl_ptp(), and _default_process_tiling_cl_roi().

dt_opencl_get_device_memory_in_use()

size_t dt_opencl_get_device_memory_in_use

(

const int

devid

)

Bytes Ansel holds on the device by its own accounting (dt_opencl_memory_statistics()), or 0 for an out-of-range id. What other applications hold there is not in it.

Definition at line 2076 of file opencl.c.

References _opencl, dt_opencl_t::dev, dt_opencl_t::inited, IS_NULL_PTR, dt_opencl_device_t::memory_in_use, and dt_opencl_t::num_devs.

Referenced by _add_device_section().

dt_opencl_get_device_name()

const char * dt_opencl_get_device_name

(

const int

devid

)

Human-readable device name, owned by the OpenCL module.

Parameters

devid

device id.

Returns

the name, or NULL for an out-of-range id or a device with no name. Valid until cleanup; do not free.

Definition at line 2050 of file opencl.c.

References _opencl, dt_opencl_t::dev, dt_opencl_t::inited, IS_NULL_PTR, dt_opencl_device_t::name, and dt_opencl_t::num_devs.

Referenced by _add_device_section().

dt_opencl_get_image_element_size()

int dt_opencl_get_image_element_size

(

cl_mem

mem

)

Definition at line 3093 of file opencl.c.

References _opencl, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clGetImageInfo, IS_NULL_PTR, size, and dt_dlopencl_t::symbols.

Referenced by _cache_entry_materialize_host_data_locked(), _pixel_cache_clmem_get(), ambient_light_cl(), and dt_dev_pixelpipe_cache_borrow_cl_payload().

dt_opencl_get_image_height()

int dt_opencl_get_image_height

(

cl_mem

mem

)

Definition at line 3082 of file opencl.c.

References _opencl, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clGetImageInfo, IS_NULL_PTR, size, and dt_dlopencl_t::symbols.

Referenced by _cache_entry_materialize_host_data_locked(), _pixel_cache_clmem_get(), ambient_light_cl(), box_max_cl(), box_min_cl(), dehaze_cl(), dt_dev_pixelpipe_cache_borrow_cl_payload(), and transition_map_cl().

dt_opencl_get_image_width()

int dt_opencl_get_image_width

(

cl_mem

mem

)

Definition at line 3071 of file opencl.c.

References _opencl, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clGetImageInfo, IS_NULL_PTR, size, and dt_dlopencl_t::symbols.

Referenced by _cache_entry_materialize_host_data_locked(), _pixel_cache_clmem_get(), ambient_light_cl(), box_max_cl(), box_min_cl(), dehaze_cl(), dt_dev_pixelpipe_cache_borrow_cl_payload(), and transition_map_cl().

dt_opencl_get_kernel_work_group_size()

int dt_opencl_get_kernel_work_group_size	(	const int	dev,
		const int	kernel,
		size_t *	kernelworkgroupsize
	)

return max workgroup size for a specific kernel

Definition at line 2591 of file opencl.c.

References _opencl, dt_opencl_t::dev, dt_opencl_device_t::devid, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clGetKernelWorkGroupInfo, DT_OPENCL_MAX_KERNELS, dt_opencl_t::inited, kernel(), dt_opencl_device_t::kernel, and dt_dlopencl_t::symbols.

Referenced by dt_opencl_local_buffer_opt().

dt_opencl_get_max_work_item_sizes()

int dt_opencl_get_max_work_item_sizes	(	const int	dev,
		size_t *	sizes
	)

return max size in sizes[3].

Definition at line 2563 of file opencl.c.

References _opencl, dt_opencl_t::dev, dt_opencl_device_t::devid, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clGetDeviceInfo, dt_opencl_t::inited, and dt_dlopencl_t::symbols.

Referenced by dt_opencl_get_work_group_limits().

dt_opencl_get_mem_context_id()

int dt_opencl_get_mem_context_id

(

cl_mem

mem

)

Definition at line 3044 of file opencl.c.

References _opencl, dt_opencl_device_t::context, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clGetMemObjectInfo, IS_NULL_PTR, dt_opencl_t::num_devs, and dt_dlopencl_t::symbols.

Referenced by _cache_entry_clmem_flush_device(), _cache_entry_clmem_flush_host_pinned_locked(), _cache_entry_clmem_has_host_pinned_locked(), _cache_entry_materialize_host_data_locked(), _cache_try_restore_device_payload(), _free_cache_entry(), _non_threadsafe_stats_append(), _pixel_cache_clmem_get(), _pixel_cache_clmem_put(), and dt_dev_pixelpipe_cache_borrow_cl_payload().

dt_opencl_get_mem_flags()

cl_mem_flags dt_opencl_get_mem_flags

(

cl_mem

mem

)

Definition at line 3062 of file opencl.c.

References _opencl, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clGetMemObjectInfo, flags, dt_opencl_t::inited, IS_NULL_PTR, and dt_dlopencl_t::symbols.

Referenced by dt_opencl_is_pinned_memory().

dt_opencl_get_mem_object_size()

size_t dt_opencl_get_mem_object_size

(

cl_mem

mem

)

Definition at line 3034 of file opencl.c.

References _opencl, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clGetMemObjectInfo, IS_NULL_PTR, size, and dt_dlopencl_t::symbols.

Referenced by _non_threadsafe_stats_append().

dt_opencl_get_mem_record()

gboolean dt_opencl_get_mem_record	(	cl_mem	mem,
		int *	devid,
		size_t *	size
	)

Definition at line 3171 of file opencl.c.

References _opencl, dt_opencl_mem_record_t::devid, dt_pthread_mutex_lock(), dt_pthread_mutex_unlock(), FALSE, dt_opencl_t::inited, IS_NULL_PTR, dt_opencl_t::mem_sizes, dt_opencl_t::mem_sizes_lock, size, and dt_opencl_mem_record_t::size.

Referenced by _non_threadsafe_stats_append().

dt_opencl_get_num_devices()

int dt_opencl_get_num_devices

(

void

)

Number of usable OpenCL devices; 0 when OpenCL is unavailable.

Definition at line 2043 of file opencl.c.

References _opencl, dt_opencl_t::inited, and dt_opencl_t::num_devs.

Referenced by _rebuild_memory(), dt_cleanup(), and dt_dev_pixelpipe_cache_get_entries_stats().

dt_opencl_get_vendor_by_id()

static const char * dt_opencl_get_vendor_by_id ( unsigned int  id )

static

Definition at line 1690 of file opencl.c.

References DT_OPENCL_VENDOR_AMD, DT_OPENCL_VENDOR_INTEL, and DT_OPENCL_VENDOR_NVIDIA.

Referenced by dt_opencl_device_init().

dt_opencl_get_work_group_limits()

int dt_opencl_get_work_group_limits	(	const int	dev,
		size_t *	sizes,
		size_t *	workgroupsize,
		unsigned long *	localmemsize
	)

return max size per dimension in sizes[3] and max total size in workgroupsize

Definition at line 2571 of file opencl.c.

References _opencl, dt_opencl_t::dev, dt_opencl_device_t::devid, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clGetDeviceInfo, dt_opencl_get_max_work_item_sizes(), dt_opencl_t::inited, and dt_dlopencl_t::symbols.

Referenced by dt_opencl_local_buffer_opt().

dt_opencl_image_fits_device()

gboolean dt_opencl_image_fits_device	(	const int	devid,
		const size_t	width,
		const size_t	height,
		const unsigned	bpp,
		const float	factor,
		const size_t	overhead
	)

check if image size fit into limits given by OpenCL runtime

Definition at line 3308 of file opencl.c.

References bpp, DT_OPENCL_FIT_OK, dt_opencl_image_fits_device_reason(), factor, height, and width.

Referenced by _harmonic_reconstruct_cl(), and guided_filter_cl().

dt_opencl_image_fits_device_reason()

dt_opencl_fit_reason_t dt_opencl_image_fits_device_reason	(	const int	devid,
		const size_t	width,
		const size_t	height,
		const unsigned	bpp,
		const float	factor,
		const size_t	overhead,
		size_t *	needed,
		size_t *	limit
	)

Like dt_opencl_image_fits_device() but also reports why the image (didn't) fit. needed and limit are filled with the two byte quantities that were actually compared for the returned reason (both 0 for DIMENSION/UNINITED), so a caller can log a message consistent with the real check instead of guessing a limit. Either out-param may be NULL.

Definition at line 3259 of file opencl.c.

References _opencl, _opencl_get_device_memalloc(), bpp, dt_opencl_t::dev, DT_DEBUG_OPENCL, DT_OPENCL_FIT_ALLOC_LIMIT, DT_OPENCL_FIT_AVAILABLE, DT_OPENCL_FIT_DIMENSION, DT_OPENCL_FIT_OK, DT_OPENCL_FIT_UNINITED, dt_opencl_get_device_available(), dt_print(), factor, height, dt_opencl_t::inited, dt_opencl_device_t::max_image_height, dt_opencl_device_t::max_image_width, n, and width.

Referenced by dt_opencl_image_fits_device(), and pixelpipe_process_on_GPU().

dt_opencl_init()

void dt_opencl_init	(	const gboolean	exclude_opencl,
		const gboolean	print_statistics
	)

Definition at line 1301 of file opencl.c.

References _driver_crash_marker, _opencl, _opencl_driver_crash_marker_path(), _opencl_fold_driver_crashes(), dt_opencl_t::crc, dt_opencl_t::detected_devs, dt_opencl_t::dev, dt_opencl_t::dev_priority_export, dt_opencl_t::dev_priority_image, dt_opencl_t::dev_priority_preview, dt_opencl_t::dev_priority_thumbnail, dt_opencl_t::dlocl, dt_bilateral_init_cl_global(), dt_capabilities_add(), dt_dlopencl_symbols_t::dt_clGetDeviceIDs, dt_dlopencl_symbols_t::dt_clGetPlatformIDs, dt_dlopencl_symbols_t::dt_clGetPlatformInfo, dt_colorspaces_init_cl_global(), dt_conf_get_bool(), dt_conf_get_int(), dt_conf_get_string_const(), DT_DEBUG_OPENCL, dt_develop_blend_init_cl_global(), dt_dlopencl_init(), dt_dwt_init_cl_global(), dt_free, dt_gaussian_init_cl_global(), dt_guided_filter_init_cl_global(), dt_heal_init_cl_global(), dt_interpolation_init_cl_global(), dt_local_laplacian_init_cl_global(), dt_opencl_apply_scheduling_profile(), DT_OPENCL_CBUFFSIZE, dt_opencl_cleanup_device(), dt_opencl_device_init(), DT_OPENCL_MAX_PLATFORMS, dt_print_nts(), dt_pthread_mutex_init(), dt_opencl_t::enabled, dt_opencl_t::error_count, i, dt_opencl_t::inited, IS_NULL_PTR, k, dt_dlopencl_t::library, dt_opencl_t::lock, dt_opencl_t::mandatory, dt_opencl_t::mem_sizes, dt_opencl_t::mem_sizes_lock, n, dt_opencl_device_t::name, dt_opencl_t::num_detected_devs, dt_opencl_t::num_devs, dt_opencl_t::print_statistics, res, dt_opencl_t::stopped, and dt_dlopencl_t::symbols.

Referenced by dt_init().

dt_opencl_is_enabled()

int dt_opencl_is_enabled

(

void

)

check if opencl is enabled

Definition at line 3335 of file opencl.c.

References _opencl, dt_opencl_t::enabled, FALSE, and dt_opencl_t::inited.

Referenced by _default_size(), _expand_rating_and_metadata(), and _rebuild_memory().

dt_opencl_is_inited()

int dt_opencl_is_inited

(

void

)

check if opencl is inited

Definition at line 3328 of file opencl.c.

References _opencl, and dt_opencl_t::inited.

Referenced by _free_cache_entry(), _iop_load_module_static(), _is_opencl_supported(), _seal_opencl_cache_policy(), dt_cleanup(), and dt_dev_pixelpipe_cache_flush_clmem_for_pipe().

dt_opencl_is_pinned_memory()

gboolean dt_opencl_is_pinned_memory

(

cl_mem

mem

)

Definition at line 238 of file opencl.c.

References dt_opencl_get_mem_flags(), and flags.

Referenced by _cache_entry_materialize_host_data_locked(), _drawlayer_sync_host_image_to_device(), dt_dev_pixelpipe_cache_get_pinned_image(), dt_dev_pixelpipe_cache_prepare_cl_input(), and dt_dev_pixelpipe_cache_sync_cl_buffer().

dt_opencl_load_program()

int dt_opencl_load_program	(	const int	dev,
		const int	prog,
		const char *	filename,
		const char *	binname,
		const char *	cachedir,
		char *	md5sum,
		char **	includemd5,
		int *	loaded_cached
	)

loads the given .cl file and returns a reference to an internal program.

defined(_WIN32)

Definition at line 2180 of file opencl.c.

References _opencl, dt_opencl_device_t::context, dt_opencl_t::dev, dt_opencl_device_t::devid, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clCreateProgramWithBinary, dt_dlopencl_symbols_t::dt_clCreateProgramWithSource, dt_dlopencl_symbols_t::dt_clGetDeviceInfo, dt_dlopencl_symbols_t::dt_clGetPlatformInfo, dt_concat_path_file(), DT_DEBUG_OPENCL, dt_free, DT_OPENCL_MAX_INCLUDES, DT_OPENCL_MAX_PROGRAMS, DT_PATH_MAX, dt_print(), dt_vprint(), f, fopen_stat(), IS_NULL_PTR, n, dt_opencl_device_t::options, dt_opencl_device_t::options_md5, dt_opencl_device_t::program, dt_opencl_device_t::program_used, and dt_dlopencl_t::symbols.

Referenced by dt_opencl_device_init().

dt_opencl_local_buffer_opt()

int dt_opencl_local_buffer_opt	(	const int	devid,
		const int	kernel,
		dt_opencl_local_buffer_t *	factors
	)

utility function to calculate optimal work group dimensions for a given kernel

Definition at line 3786 of file opencl.c.

References _opencl, dt_opencl_local_buffer_t::cellsize, DT_DEBUG_OPENCL, dt_opencl_get_kernel_work_group_size(), dt_opencl_get_work_group_limits(), dt_print(), FALSE, dt_opencl_t::inited, kernel(), nextpow2(), dt_opencl_local_buffer_t::overhead, dt_opencl_local_buffer_t::sizex, dt_opencl_local_buffer_t::sizey, TRUE, dt_opencl_local_buffer_t::xfactor, dt_opencl_local_buffer_t::xoffset, dt_opencl_local_buffer_t::yfactor, and dt_opencl_local_buffer_t::yoffset.

Referenced by _downsample_guided_laplacian_postfilter_cl(), _hl_cl_lch_xtrans(), color_smoothing_cl(), dt_bilateral_init_cl(), dt_gaussian_init_cl(), dt_interpolation_resample_cl(), dt_iop_colorreconstruct_bilateral_init_cl(), get_blocksizes(), green_equilibration_cl(), process_cl(), process_default_cl(), process_laplacian_bayer_cl(), process_laplacian_passthrough_cl(), process_laplacian_xtrans_cl(), process_markesteijn_cl(), process_nlmeans_cl(), process_rcd_cl(), process_vng_cl(), process_wavelets_cl(), reconstruct_highlights_cl(), wavelets_process_cl(), and wavelets_process_cl().

dt_opencl_map_buffer()

void * dt_opencl_map_buffer	(	const int	devid,
		cl_mem	buffer,
		const int	blocking,
		const int	flags,
		size_t	offset,
		size_t	size
	)

Definition at line 2876 of file opencl.c.

References _opencl, dt_opencl_device_t::cmd_queue, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clEnqueueMapBuffer, DT_DEBUG_OPENCL, dt_opencl_events_get_slot(), dt_print(), flags, dt_opencl_t::inited, size, and dt_dlopencl_t::symbols.

dt_opencl_map_image()

void * dt_opencl_map_image	(	const int	devid,
		cl_mem	buffer,
		const int	blocking,
		const int	flags,
		size_t	width,
		size_t	height,
		int	bpp
	)

Definition at line 2890 of file opencl.c.

References _opencl, dt_opencl_device_t::cmd_queue, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clEnqueueMapImage, DT_DEBUG_OPENCL, dt_opencl_events_get_slot(), dt_print(), flags, height, dt_opencl_t::inited, dt_dlopencl_t::symbols, and width.

Referenced by _cache_entry_materialize_host_data_locked(), _drawlayer_sync_host_image_to_device(), dt_dev_pixelpipe_cache_get_pinned_image(), and dt_dev_pixelpipe_cache_sync_cl_buffer().

dt_opencl_md5sum()

void dt_opencl_md5sum	(	const char **	files,
		char **	md5sums
	)

calculates md5sums for a list of CL include files.

Definition at line 2126 of file opencl.c.

References dt_concat_path_file(), DT_DEBUG_OPENCL, dt_free, dt_loc_get_kerneldir(), DT_OPENCL_MAX_INCLUDES, DT_PATH_MAX, dt_print(), f, fopen_stat(), IS_NULL_PTR, and n.

Referenced by dt_opencl_device_init().

dt_opencl_memory_statistics()

void dt_opencl_memory_statistics	(	int	devid,
		cl_mem	mem,
		size_t	size,
		dt_opencl_memory_t	action
	)

Definition at line 3111 of file opencl.c.

References _opencl, dt_opencl_t::dev, dt_opencl_mem_record_t::devid, DT_DEBUG_MEMORY, DT_DEBUG_OPENCL, dt_get_debug_flags(), dt_print(), dt_pthread_mutex_lock(), dt_pthread_mutex_unlock(), IS_NULL_PTR, MAX, dt_opencl_t::mem_sizes, dt_opencl_t::mem_sizes_lock, dt_opencl_device_t::memory_in_use, OPENCL_MEMORY_ADD, dt_opencl_device_t::peak_memory, size, and dt_opencl_mem_record_t::size.

Referenced by _dt_opencl_alloc_image2d(), dt_opencl_alloc_device_buffer_with_flags(), dt_opencl_copy_host_to_device_constant(), dt_opencl_copy_host_to_device_rowpitch(), and dt_opencl_release_mem_object().

dt_opencl_micro_nap()

int dt_opencl_micro_nap

(

const int

devid

)

Definition at line 225 of file opencl.c.

References _opencl, dt_opencl_t::dev, dt_opencl_t::inited, and dt_opencl_device_t::micro_nap.

Referenced by nlmeans_cl_run().

dt_opencl_note_crash_backtrace()

void dt_opencl_note_crash_backtrace	(	const char *	backtrace,
		size_t	backtrace_len
	)

Tell the OpenCL module that the process crashed, and let it judge whether the crash was its driver's doing.

internally calls dt_clGetDeviceInfo, and takes care of memory allocation afterwards, *param_value will point to memory block of size at least *param_value which needs to be g_free()'d manually inits the opencl subsystem.

Vendor OpenCL runtimes fault on their own threads, with no frame of ours on the stack, and no code here can prevent that. What this module can do is notice it happening repeatedly and stop enabling OpenCL on the next start. It recognises its own drivers by name in the backtrace; a crash anywhere else is recorded as nothing.

Sentry captures the backtrace and knows nothing about GPUs; this module knows about GPUs and nothing about crash reporting. dt_init() connects the two, so neither has to include the other.

Parameters

backtrace	NUL-terminated backtrace text, valid only for the duration of the call.
backtrace_len	its length in bytes.

Warning

Runs in a signal handler. Uses only open/write/close – no allocation, no locks, no conf – and does nothing at all before dt_opencl_init() has built the record's path.

Definition at line 1243 of file opencl.c.

References _driver_crash_marker, _gpu_runtimes, i, IS_NULL_PTR, void(), w1, and w2.

Referenced by dt_init().

dt_opencl_priority_parse()

static void dt_opencl_priority_parse ( dt_opencl_t *  cl, char *  configstr, int *  priority_list, int *  mandatory  )

static

parse a single token of priority string and store priorities in priority_list

Definition at line 1798 of file opencl.c.

References _device_by_cname(), _take_from_list(), dt_free, i, IS_NULL_PTR, and dt_opencl_t::num_devs.

Referenced by dt_opencl_update_priorities().

dt_opencl_read_buffer_from_device()

int dt_opencl_read_buffer_from_device	(	const int	devid,
		void *	host,
		void *	device,
		const size_t	offset,
		const size_t	size,
		const int	blocking
	)

Definition at line 2785 of file opencl.c.

References _opencl, dt_opencl_device_t::cmd_queue, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clEnqueueReadBuffer, dt_opencl_events_get_slot(), dt_opencl_t::inited, size, and dt_dlopencl_t::symbols.

Referenced by _aniso_stage_cl_selftest(), _cf_harmonic_fill_cl_selftest(), _cf_joint_stage_cl_selftest(), _cf_stage_cl_selftest(), _chromaticity_gradient_stage_cl_selftest(), _harmonic_reconstruct_cl(), _hf_stage_cl_selftest(), _hl_cl_count_clipped(), _hl_knee_estimate_cl(), _joint_core_stage_cl_selftest(), _knee_cl_selftest(), _region_guided_filter_cl_selftest(), _selfdome_stage_cl_selftest(), _sp_chol_cl_selftest(), dt_heal_cl(), dt_iop_colorreconstruct_bilateral_freeze_cl(), green_equilibration_cl(), main(), process_cl(), process_harmonic_cl(), process_wavelets_cl(), rt_adjust_levels_cl(), and rt_process_stats_cl().

dt_opencl_read_device_config()

gboolean dt_opencl_read_device_config

(

const int

devid

)

Definition at line 294 of file opencl.c.

References _dt_opencl_get_conf_int(), _opencl, dt_opencl_device_t::avoid_atomics, dt_opencl_device_t::clroundup_ht, dt_opencl_device_t::clroundup_wd, dt_opencl_device_t::cname, dt_opencl_t::dev, dt_opencl_device_t::disabled, DT_CLDEVICE_HEAD, dt_conf_get_int(), dt_conf_get_int64(), dt_conf_key_exists(), dt_conf_key_not_empty(), dt_conf_set_bool(), DT_DEBUG_OPENCL, DT_OPENCL_PINNING_DISABLED, DT_OPENCL_PINNING_OFF, DT_OPENCL_PINNING_ON, dt_opencl_write_device_config(), dt_print(), dt_opencl_device_t::event_handles, FALSE, dt_opencl_device_t::forced_headroom, dt_opencl_device_t::host_unified_memory, MAX, dt_opencl_device_t::micro_nap, dt_opencl_device_t::pinned_memory, TRUE, and dt_opencl_device_t::use_events.

Referenced by dt_opencl_device_init().

dt_opencl_read_host_from_device()

int dt_opencl_read_host_from_device	(	const int	devid,
		void *	host,
		void *	device,
		const int	width,
		const int	height,
		const int	bpp
	)

Definition at line 2645 of file opencl.c.

References bpp, dt_opencl_read_host_from_device_rowpitch(), height, and width.

Referenced by _cache_entry_materialize_host_data_locked(), _refine_with_detail_mask_cl(), ambient_light_cl(), dt_dev_pixelpipe_cache_sync_cl_buffer(), dt_opencl_copy_device_to_host(), and guided_filter_cl_fallback().

dt_opencl_read_host_from_device_non_blocking()

int dt_opencl_read_host_from_device_non_blocking	(	const int	devid,
		void *	host,
		void *	device,
		const int	width,
		const int	height,
		const int	bpp
	)

Definition at line 2661 of file opencl.c.

References bpp, dt_opencl_read_host_from_device_rowpitch_non_blocking(), height, and width.

dt_opencl_read_host_from_device_raw()

int dt_opencl_read_host_from_device_raw	(	const int	devid,
		void *	host,
		void *	device,
		const size_t *	origin,
		const size_t *	region,
		const int	rowpitch,
		const int	blocking
	)

Definition at line 2680 of file opencl.c.

References _opencl, dt_opencl_device_t::cmd_queue, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clEnqueueReadImage, dt_opencl_events_get_slot(), dt_opencl_t::inited, and dt_dlopencl_t::symbols.

Referenced by _default_process_tiling_cl_ptp(), _default_process_tiling_cl_roi(), dt_opencl_read_host_from_device_rowpitch(), and dt_opencl_read_host_from_device_rowpitch_non_blocking().

dt_opencl_read_host_from_device_rowpitch()

int dt_opencl_read_host_from_device_rowpitch	(	const int	devid,
		void *	host,
		void *	device,
		const int	width,
		const int	height,
		const int	rowpitch
	)

Definition at line 2651 of file opencl.c.

References _opencl, dt_opencl_read_host_from_device_raw(), height, dt_opencl_t::inited, and width.

Referenced by dt_opencl_read_host_from_device().

dt_opencl_read_host_from_device_rowpitch_non_blocking()

int dt_opencl_read_host_from_device_rowpitch_non_blocking	(	const int	devid,
		void *	host,
		void *	device,
		const int	width,
		const int	height,
		const int	rowpitch
	)

Definition at line 2668 of file opencl.c.

References _opencl, dt_opencl_read_host_from_device_raw(), height, dt_opencl_t::inited, and width.

Referenced by dt_opencl_read_host_from_device_non_blocking().

dt_opencl_release_device()

void dt_opencl_release_device

(

const int

devid

)

Release a device reserved by either reserve function.

done with your command queue.

Definition at line 2013 of file opencl.c.

References _opencl, dt_opencl_t::dev, dt_pthread_mutex_BAD_unlock(), dt_opencl_t::inited, dt_opencl_device_t::lock, and dt_opencl_t::num_devs.

Referenced by _free_cache_entry(), dt_dev_pixelpipe_cache_flush_clmem_for_pipe(), dt_dev_pixelpipe_process(), and main().

dt_opencl_release_mem_object()

void dt_opencl_release_mem_object

(

cl_mem

mem

)

Definition at line 2863 of file opencl.c.

References _opencl, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clReleaseMemObject, dt_opencl_memory_statistics(), dt_opencl_t::inited, IS_NULL_PTR, OPENCL_MEMORY_SUB, and dt_dlopencl_t::symbols.

Referenced by _aniso_stage_cl_selftest(), _blend_layer_over_input_cl(), _cache_entry_clmem_flush_device(), _cache_entry_clmem_flush_host_pinned_locked(), _cf_harmonic_fill_cl_n(), _cf_harmonic_fill_cl_selftest(), _cf_joint_fit_cl(), _cf_joint_stage_cl(), _cf_joint_stage_cl_selftest(), _cf_pair_stage_cl(), _cf_stage_cl(), _cf_stage_cl_selftest(), _chromaticity_gradient_stage_cl_selftest(), _default_process_tiling_cl_ptp(), _default_process_tiling_cl_roi(), _downsample_guided_laplacian_postfilter_cl(), _drawlayer_acquire_source_image(), _harmonic_reconstruct_cl(), _hf_stage_cl(), _hf_stage_cl_selftest(), _hl_cl_count_clipped(), _hl_cl_lch_xtrans(), _hl_cl_visualize(), _hl_knee_apply_cfa_cl(), _hl_knee_estimate_cl(), _joint_core_stage_cl_selftest(), _knee_cl_selftest(), _pixel_cache_clmem_put(), _refine_with_detail_mask_cl(), _region_blur_cl_selftest(), _region_guided_filter_cl_selftest(), _retouch_blur_cl(), _retouch_clone_cl(), _retouch_fill_cl(), _retouch_heal_cl(), _selfdome_stage_cl_selftest(), _simulate_channel_cl(), _sp_chol_cl_free(), _sp_chol_cl_selftest(), _sp_cl_upload(), _unet_forward_cl(), _upload_layer_bank(), _weights_cl(), apply_global_distortion_map_cl(), box_max_cl(), box_min_cl(), color_smoothing_cl(), dt_bilateral_free_cl(), dt_bilateral_slice_to_output_cl(), dt_colorspaces_apply_profile_cl(), dt_dev_pixelpipe_cache_flush_entry_clmem(), dt_dev_pixelpipe_cache_get_pinned_image(), dt_dev_pixelpipe_cache_release_cl_buffer(), dt_develop_blend_process_cl(), dt_gaussian_free_cl(), dt_gaussian_init_cl(), dt_interpolation_resample_cl(), dt_iop_colorreconstruct_bilateral_free_cl(), dt_ioppr_free_iccprofile_params_cl(), dt_ioppr_transform_image_colorspace_rgb_cl(), dt_local_laplacian_free_cl(), dt_nn_model_free_cl(), dual_demosaic_cl(), dwt_wavelet_decompose_cl(), green_equilibration_cl(), guided_filter_cl_impl(), main(), nlmeans_cl_run(), process_cl(), process_cl(), process_cl(), process_default_cl(), process_harmonic_cl(), process_laplacian_bayer_cl(), process_laplacian_passthrough_cl(), process_laplacian_xtrans_cl(), process_markesteijn_cl(), process_nlmeans_cl(), process_rcd_cl(), process_vng_cl(), process_wavelets_cl(), reconstruct_highlights_cl(), rt_copy_image_masked_cl(), rt_copy_in_to_out_cl(), rt_copy_mask_to_alpha_cl(), rt_process_forms_cl(), and rt_upload_scaled_mask_cl().

dt_opencl_report_pipe_error()

int dt_opencl_report_pipe_error

(

void

)

Record that a pipe run failed on OpenCL, and decide whether to give up on it.

The count and the give-up threshold belong to the OpenCL module, not to the pipeline: the pipeline's job is to report the failure and be told what to do next. Crossing the threshold also drops the "opencl" capability, so this is the one place that decides OpenCL is finished for the session.

Returns

1 for "this run failed, retry on CPU", 2 for "too many failures, OpenCL is off for the rest of the session".

Definition at line 2093 of file opencl.c.

References _opencl, dt_capabilities_remove(), DT_OPENCL_MAX_ERRORS, dt_opencl_t::error_count, and dt_opencl_t::stopped.

Referenced by dt_dev_pixelpipe_process().

dt_opencl_reserve_device_by_id()

void dt_opencl_reserve_device_by_id

(

const int

devid

)

Reserve a device the caller has already identified, blocking until it is free.

Parameters

devid

device id; out-of-range ids are ignored.

Definition at line 2024 of file opencl.c.

References _opencl, dt_opencl_t::dev, dt_pthread_mutex_BAD_lock(), dt_opencl_t::inited, dt_opencl_device_t::lock, and dt_opencl_t::num_devs.

Referenced by dt_dev_pixelpipe_cache_flush_clmem_for_pipe().

dt_opencl_reserve_device_for_pipe()

int dt_opencl_reserve_device_for_pipe

(

const int

pipetype

)

Reserve a device for a pipe run: choose a free one by the pipe's priority list and take its lock. Blocks or gives up per the opencl_mandatory_timeout conf key.

locks a device for your thread's exclusive use

Parameters

pipetype

a dt_dev_pixelpipe_type_t; selects which priority list applies.

Returns

the reserved device id, or -1 if none could be had. Release with dt_opencl_release_device().

Note

There is ONE lock per device and this is it – reserving a device and locking it are the same act. dt_opencl_reserve_device_by_id() takes the same lock when the caller already knows which device it needs. This used to be called dt_opencl_lock_device(), which took a PIPE TYPE while its dt_opencl_unlock_device() counterpart took a DEVICE ID – two different ints, one of them silently wrong if the pair were ever read as symmetric. The old names are gone rather than redefined, so a stale caller fails to compile.

Definition at line 1928 of file opencl.c.

References _opencl, dt_opencl_t::dev, dt_opencl_t::dev_priority_export, dt_opencl_t::dev_priority_image, dt_opencl_t::dev_priority_preview, dt_opencl_t::dev_priority_thumbnail, dt_conf_get_int(), DT_DEBUG_OPENCL, DT_DEV_PIXELPIPE_EXPORT, DT_DEV_PIXELPIPE_FULL, DT_DEV_PIXELPIPE_PREVIEW, DT_DEV_PIXELPIPE_THUMBNAIL, dt_free, dt_iop_nap(), dt_print(), dt_pthread_mutex_BAD_trylock(), dt_pthread_mutex_lock(), dt_pthread_mutex_unlock(), dt_opencl_t::inited, dt_opencl_t::lock, dt_opencl_device_t::lock, dt_opencl_t::mandatory, MAX, n, and dt_opencl_t::num_devs.

Referenced by dt_dev_pixelpipe_process(), and main().

dt_opencl_set_detected_device_enabled()

int dt_opencl_set_detected_device_enabled	(	const int	detected,
		const gboolean	enabled
	)

Persist the per-device OpenCL preference and derive the global OpenCL switch from all GPU states.

Definition at line 428 of file opencl.c.

References _opencl, dt_opencl_detected_device_t::cname, dt_opencl_detected_device_t::config_id, dt_opencl_t::detected_devs, dt_opencl_detected_device_t::disabled, DT_CLDEVICE_HEAD, dt_conf_set_bool(), dt_conf_set_int(), dt_opencl_detected_device_enabled(), dt_opencl_get_detected_device(), enabled, IS_NULL_PTR, key, and TRUE.

Referenced by _opencl_device_enabled_callback().

dt_opencl_set_detected_device_headroom()

int dt_opencl_set_detected_device_headroom	(	const int	detected,
		const size_t	headroom
	)

Persist the per-device GPU vRAM headroom in MiB.

Definition at line 506 of file opencl.c.

References _opencl, dt_opencl_detected_device_t::cname, dt_opencl_detected_device_t::config_id, dt_opencl_t::detected_devs, dt_opencl_t::dev, DT_CLDEVICE_HEAD, dt_conf_set_int(), dt_opencl_check_tuning(), dt_opencl_get_detected_device(), dt_opencl_device_t::forced_headroom, dt_opencl_detected_device_t::forced_headroom, IS_NULL_PTR, key, MIN, and dt_opencl_t::num_devs.

Referenced by _opencl_device_headroom_callback().

dt_opencl_set_detected_device_pinned_memory()

int dt_opencl_set_detected_device_pinned_memory	(	const int	detected,
		const gboolean	enabled
	)

Persist the per-device pinned memory transfer preference.

Definition at line 473 of file opencl.c.

References _opencl, dt_opencl_detected_device_t::cname, dt_opencl_detected_device_t::config_id, dt_opencl_t::detected_devs, dt_opencl_t::dev, DT_CLDEVICE_HEAD, dt_conf_set_int(), dt_opencl_get_detected_device(), DT_OPENCL_PINNING_OFF, DT_OPENCL_PINNING_ON, enabled, IS_NULL_PTR, key, dt_opencl_t::num_devs, dt_opencl_device_t::pinned_memory, and dt_opencl_detected_device_t::pinned_memory.

Referenced by _opencl_device_pinned_memory_callback().

dt_opencl_set_kernel_arg()

int dt_opencl_set_kernel_arg	(	const int	dev,
		const int	kernel,
		const int	num,
		const size_t	size,
		const void *	arg
	)

attach arg.

Definition at line 2603 of file opencl.c.

References _opencl, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clSetKernelArg, DT_OPENCL_MAX_KERNELS, dt_opencl_t::inited, kernel(), dt_opencl_device_t::kernel, size, and dt_dlopencl_t::symbols.

Referenced by _cf_harmonic_fill_cl_n(), _cf_joint_fit_cl(), _cf_joint_stage_cl(), _cf_joint_stage_cl_selftest(), _cf_pair_stage_cl(), _cf_stage_cl(), _conv_cl(), _downsample_guided_laplacian_postfilter_cl(), _drawlayer_run_premult_over_kernel(), _harmonic_reconstruct_cl(), _hf_stage_cl(), _hl_cl_clip(), _hl_cl_count_clipped(), _hl_cl_lch_bayer(), _hl_cl_lch_xtrans(), _hl_cl_visualize(), _hl_knee_apply_cfa_cl(), _hl_knee_estimate_cl(), _refine_with_detail_mask_cl(), _region_edge_blur_cl(), _retouch_blur_cl(), _retouch_fill_cl(), _simulate_channel_cl(), _sp_chol_factor_cl(), _sp_chol_solve_cl(), _upsample_cl(), apply_global_distortion_map_cl(), box_max_cl(), box_min_cl(), cl_box_mean(), cl_covariances(), cl_generate_result(), cl_solve(), cl_split_rgb(), cl_update_covariance(), cl_variances(), color_smoothing_cl(), dehaze_cl(), dt_bilateral_blur_cl(), dt_bilateral_init_cl(), dt_bilateral_slice_cl(), dt_bilateral_slice_to_output_cl(), dt_bilateral_splat_cl(), dt_colorspaces_apply_profile_cl(), dt_develop_blend_process_cl(), dt_gaussian_blur_cl(), dt_interpolation_resample_cl(), dt_iop_colorreconstruct_bilateral_blur_cl(), dt_iop_colorreconstruct_bilateral_init_cl(), dt_iop_colorreconstruct_bilateral_slice_cl(), dt_iop_colorreconstruct_bilateral_splat_cl(), dt_ioppr_transform_image_colorspace_rgb_cl(), dt_local_laplacian_cl(), dual_demosaic_cl(), dwt_add_layer_cl(), dwt_subtract_layer_cl(), dwt_wavelet_decompose_cl(), green_equilibration_cl(), nlmeans_cl_accu(), nlmeans_cl_horiz(), nlmeans_cl_init(), nlmeans_cl_run(), process_cl(), process_cl(), process_cl(), process_default_cl(), process_harmonic_cl(), process_laplacian_bayer_cl(), process_laplacian_passthrough_cl(), process_laplacian_xtrans_cl(), process_markesteijn_cl(), process_nlmeans_cl(), process_rcd_cl(), process_vng_cl(), process_wavelets_cl(), reconstruct_highlights_cl(), rt_copy_image_masked_cl(), rt_copy_in_to_out_cl(), rt_copy_mask_to_alpha_cl(), transition_map_cl(), wavelets_process_cl(), and wavelets_process_cl().

dt_opencl_set_synchronization_timeout()

static void dt_opencl_set_synchronization_timeout ( int  value )

static

set opencl specific synchronization timeout

Definition at line 3372 of file opencl.c.

References _opencl, DT_DEBUG_OPENCL, dt_print_nts(), dt_opencl_t::opencl_synchronization_timeout, and value.

Referenced by dt_opencl_apply_scheduling_profile().

dt_opencl_try_reserve_device_by_id()

int dt_opencl_try_reserve_device_by_id

(

const int

devid

)

Reserve a device only if it is free right now.

Parameters

devid

device id.

Returns

0 when the device was reserved (pthread convention), non-zero when it was busy or the id was out of range. Never waits – callers use this on paths that must not block behind a pipe that is mid-run.

Definition at line 2035 of file opencl.c.

References _opencl, dt_opencl_t::dev, dt_pthread_mutex_BAD_trylock(), dt_opencl_t::inited, dt_opencl_device_t::lock, and dt_opencl_t::num_devs.

Referenced by _free_cache_entry().

dt_opencl_unmap_mem_object()

int dt_opencl_unmap_mem_object	(	const int	devid,
		cl_mem	mem_object,
		void *	mapped_ptr
	)

Definition at line 2910 of file opencl.c.

References _opencl, dt_opencl_device_t::cmd_queue, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clEnqueueUnmapMemObject, DT_DEBUG_OPENCL, dt_opencl_events_get_slot(), dt_print(), dt_opencl_t::inited, and dt_dlopencl_t::symbols.

Referenced by _cache_entry_materialize_host_data_locked(), _drawlayer_sync_host_image_to_device(), dt_dev_pixelpipe_cache_get_pinned_image(), and dt_dev_pixelpipe_cache_sync_cl_buffer().

dt_opencl_update_priorities()

static void dt_opencl_update_priorities ( )

static

set device priorities according to config string

Definition at line 1893 of file opencl.c.

References _opencl, dt_opencl_t::dev_priority_export, dt_opencl_t::dev_priority_image, dt_opencl_t::dev_priority_preview, dt_opencl_t::dev_priority_thumbnail, dt_conf_get_string(), DT_DEBUG_OPENCL, dt_free, dt_opencl_priority_parse(), dt_print_nts(), i, dt_opencl_t::inited, dt_opencl_t::mandatory, dt_opencl_t::num_devs, and preview.

Referenced by dt_opencl_apply_scheduling_profile().

dt_opencl_update_settings()

int dt_opencl_update_settings

(

void

)

update enabled flag and profile with value from preferences, returns enabled flag

Todo:

: This pulls in prefs every time the pixelpipe runs. Instead have a callback for DT_SIGNAL_PREFERENCES_CHANGE?

Definition at line 3352 of file opencl.c.

References _opencl, dt_conf_get_bool(), DT_DEBUG_OPENCL, dt_print(), dt_opencl_t::enabled, dt_opencl_t::error_count, FALSE, dt_opencl_t::inited, and dt_opencl_t::stopped.

Referenced by dt_dev_pixelpipe_process().

dt_opencl_use_pinned_memory()

gboolean dt_opencl_use_pinned_memory

(

const int

devid

)

Definition at line 231 of file opencl.c.

References _opencl, dt_opencl_t::dev, DT_OPENCL_PINNING_ON, FALSE, dt_opencl_t::inited, and dt_opencl_device_t::pinned_memory.

Referenced by dt_dev_pixelpipe_cache_get_cl_buffer(), and dt_dev_pixelpipe_cache_get_pinned_image().

dt_opencl_write_buffer_to_device()

int dt_opencl_write_buffer_to_device	(	const int	devid,
		void *	host,
		void *	device,
		const size_t	offset,
		const size_t	size,
		const int	blocking
	)

Definition at line 2796 of file opencl.c.

References _opencl, dt_opencl_device_t::cmd_queue, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clEnqueueWriteBuffer, dt_opencl_events_get_slot(), dt_opencl_t::inited, size, and dt_dlopencl_t::symbols.

Referenced by _aniso_stage_cl_selftest(), _cf_harmonic_fill_cl_selftest(), _cf_joint_stage_cl_selftest(), _cf_stage_cl_selftest(), _chromaticity_gradient_stage_cl_selftest(), _harmonic_reconstruct_cl(), _hf_stage_cl_selftest(), _hl_knee_estimate_cl(), _joint_core_stage_cl_selftest(), _knee_cl_selftest(), _region_guided_filter_cl_selftest(), _selfdome_stage_cl_selftest(), _sp_cl_upload(), _upload_layer_bank(), _weights_cl(), dt_heal_cl(), main(), process_cl(), rt_adjust_levels_cl(), rt_process_stats_cl(), and rt_upload_scaled_mask_cl().

dt_opencl_write_device_config()

void dt_opencl_write_device_config

(

const int

devid

)

utility functions handling device specific properties

Definition at line 244 of file opencl.c.

References _opencl, dt_opencl_device_t::avoid_atomics, dt_opencl_device_t::clroundup_ht, dt_opencl_device_t::clroundup_wd, dt_opencl_device_t::cname, dt_opencl_t::dev, dt_opencl_device_t::disabled, DT_CLDEVICE_HEAD, dt_conf_set_int(), DT_OPENCL_PINNING_DISABLED, DT_OPENCL_PINNING_ON, dt_opencl_device_t::event_handles, dt_opencl_device_t::forced_headroom, dt_opencl_device_t::micro_nap, and dt_opencl_device_t::pinned_memory.

Referenced by dt_opencl_device_init(), and dt_opencl_read_device_config().

dt_opencl_write_host_to_device()

int dt_opencl_write_host_to_device	(	const int	devid,
		void *	host,
		void *	device,
		const int	width,
		const int	height,
		const int	bpp
	)

Definition at line 2692 of file opencl.c.

References bpp, dt_opencl_write_host_to_device_rowpitch(), height, and width.

Referenced by _drawlayer_sync_host_image_to_device(), _harmonic_cl_roundtrip(), _refine_with_detail_mask_cl(), _region_blur_cl_selftest(), dt_colorspaces_apply_profile_cl(), dt_dev_pixelpipe_cache_get_pinned_image(), dt_dev_pixelpipe_cache_prepare_cl_input(), dt_dev_pixelpipe_cache_sync_cl_buffer(), dt_develop_blend_process_cl(), dt_ioppr_transform_image_colorspace_rgb_cl(), guided_filter_cl_fallback(), process_cl(), and process_harmonic_cl().

dt_opencl_write_host_to_device_non_blocking()

int dt_opencl_write_host_to_device_non_blocking	(	const int	devid,
		void *	host,
		void *	device,
		const int	width,
		const int	height,
		const int	bpp
	)

Definition at line 2708 of file opencl.c.

References bpp, dt_opencl_write_host_to_device_rowpitch_non_blocking(), height, and width.

dt_opencl_write_host_to_device_raw()

int dt_opencl_write_host_to_device_raw	(	const int	devid,
		const void *	host,
		void *	device,
		const size_t *	origin,
		const size_t *	region,
		const int	rowpitch,
		const int	blocking
	)

Definition at line 2725 of file opencl.c.

References _opencl, dt_opencl_device_t::cmd_queue, dt_opencl_t::dev, dt_opencl_t::dlocl, dt_dlopencl_symbols_t::dt_clEnqueueWriteImage, dt_opencl_events_get_slot(), dt_opencl_t::inited, and dt_dlopencl_t::symbols.

Referenced by _default_process_tiling_cl_ptp(), _default_process_tiling_cl_roi(), _drawlayer_sync_host_image_to_device(), dt_opencl_write_host_to_device_rowpitch(), dt_opencl_write_host_to_device_rowpitch_non_blocking(), and process_cl().

dt_opencl_write_host_to_device_rowpitch()

int dt_opencl_write_host_to_device_rowpitch	(	const int	devid,
		void *	host,
		void *	device,
		const int	width,
		const int	height,
		const int	rowpitch
	)

Definition at line 2698 of file opencl.c.

References _opencl, dt_opencl_write_host_to_device_raw(), height, dt_opencl_t::inited, and width.

Referenced by dt_opencl_write_host_to_device().

dt_opencl_write_host_to_device_rowpitch_non_blocking()

int dt_opencl_write_host_to_device_rowpitch_non_blocking	(	const int	devid,
		void *	host,
		void *	device,
		const int	width,
		const int	height,
		const int	rowpitch
	)

Definition at line 2714 of file opencl.c.

References _opencl, dt_opencl_write_host_to_device_raw(), height, dt_opencl_t::inited, and width.

Referenced by dt_opencl_write_host_to_device_non_blocking().

fopen_stat()

static FILE * fopen_stat ( const char *  filename, struct stat *  st  )

static

Definition at line 2107 of file opencl.c.

References DT_DEBUG_OPENCL, dt_print(), f, and IS_NULL_PTR.

Referenced by dt_opencl_load_program(), and dt_opencl_md5sum().

nextpow2()

static int nextpow2 ( int  n )

static

Definition at line 3776 of file opencl.c.

References k, and n.

Referenced by dt_opencl_local_buffer_opt().

Variable Documentation

_driver_crash_marker

char _driver_crash_marker[DT_PATH_MAX] = { 0 }

static

Definition at line 1232 of file opencl.c.

Referenced by _opencl_fold_driver_crashes(), dt_opencl_init(), and dt_opencl_note_crash_backtrace().

_gpu_runtimes

const _gpu_runtime_t _gpu_runtimes[]

static

Initial value:

= {
  { "RusticlOpenCL",    "rusticl",         "rusticl" },  
  { "libMesaOpenCL",    "Clover",          "clover" },   
  { "amdocl",           "AMD",             "amd" },      
  { "libigdfcl",        "Intel(R) OpenCL", "intel" },    
  { "libigdrcl",        "Intel(R) OpenCL", "intel" },
  { "intelocl",         "Intel(R) OpenCL", "intel" },
  { "libnvidia-opencl", "NVIDIA",          "nvidia" },
  { "nvopencl",         "NVIDIA",          "nvidia" },
  { NULL, NULL, NULL }
}

Definition at line 1183 of file opencl.c.

Referenced by _gpu_runtime_of_platform(), _opencl_fold_driver_crashes(), and dt_opencl_note_crash_backtrace().

_opencl

dt_opencl_t* _opencl = NULL

static

Definition at line 131 of file opencl.c.

Referenced by _device_by_cname(), _dt_opencl_alloc_image2d(), _opencl_get_device_memalloc(), dt_opencl_alloc_device_buffer_with_flags(), dt_opencl_apply_scheduling_profile(), dt_opencl_avoid_atomics(), dt_opencl_build_program(), dt_opencl_check_tuning(), dt_opencl_cleanup(), dt_opencl_clear_driver_crash_streak(), dt_opencl_copy_host_to_device_constant(), dt_opencl_copy_host_to_device_rowpitch(), dt_opencl_create_kernel(), dt_opencl_dev_roundup_height(), dt_opencl_dev_roundup_width(), dt_opencl_disable(), dt_opencl_enqueue_barrier(), dt_opencl_enqueue_copy_buffer_to_buffer(), dt_opencl_enqueue_copy_buffer_to_image(), dt_opencl_enqueue_copy_image(), dt_opencl_enqueue_copy_image_to_buffer(), dt_opencl_enqueue_kernel_2d_with_local(), dt_opencl_events_flush(), dt_opencl_events_get_slot(), dt_opencl_events_profiling(), dt_opencl_events_reset(), dt_opencl_events_wait_for(), dt_opencl_finish(), dt_opencl_free_kernel(), dt_opencl_get_detected_device(), dt_opencl_get_detected_device_count(), dt_opencl_get_device_available(), dt_opencl_get_device_headroom(), dt_opencl_get_device_max_global_mem(), dt_opencl_get_device_max_image_size(), dt_opencl_get_device_memalloc(), dt_opencl_get_device_memory_in_use(), dt_opencl_get_device_name(), dt_opencl_get_image_element_size(), dt_opencl_get_image_height(), dt_opencl_get_image_width(), dt_opencl_get_kernel_work_group_size(), dt_opencl_get_max_work_item_sizes(), dt_opencl_get_mem_context_id(), dt_opencl_get_mem_flags(), dt_opencl_get_mem_object_size(), dt_opencl_get_mem_record(), dt_opencl_get_num_devices(), dt_opencl_get_work_group_limits(), dt_opencl_image_fits_device_reason(), dt_opencl_init(), dt_opencl_is_enabled(), dt_opencl_is_inited(), dt_opencl_load_program(), dt_opencl_local_buffer_opt(), dt_opencl_map_buffer(), dt_opencl_map_image(), dt_opencl_memory_statistics(), dt_opencl_micro_nap(), dt_opencl_read_buffer_from_device(), dt_opencl_read_device_config(), dt_opencl_read_host_from_device_raw(), dt_opencl_read_host_from_device_rowpitch(), dt_opencl_read_host_from_device_rowpitch_non_blocking(), dt_opencl_release_device(), dt_opencl_release_mem_object(), dt_opencl_report_pipe_error(), dt_opencl_reserve_device_by_id(), dt_opencl_reserve_device_for_pipe(), dt_opencl_set_detected_device_enabled(), dt_opencl_set_detected_device_headroom(), dt_opencl_set_detected_device_pinned_memory(), dt_opencl_set_kernel_arg(), dt_opencl_set_synchronization_timeout(), dt_opencl_try_reserve_device_by_id(), dt_opencl_unmap_mem_object(), dt_opencl_update_priorities(), dt_opencl_update_settings(), dt_opencl_use_pinned_memory(), dt_opencl_write_buffer_to_device(), dt_opencl_write_device_config(), dt_opencl_write_host_to_device_raw(), dt_opencl_write_host_to_device_rowpitch(), and dt_opencl_write_host_to_device_rowpitch_non_blocking().