diff options
Diffstat (limited to 'source/blender/gpu/intern')
| -rw-r--r-- | source/blender/gpu/intern/gpu_codegen.c | 2 | ||||
| -rw-r--r-- | source/blender/gpu/intern/gpu_debug.c | 6 | ||||
| -rw-r--r-- | source/blender/gpu/intern/gpu_select.c | 272 | ||||
| -rw-r--r-- | source/blender/gpu/intern/gpu_select_pick.c | 744 | ||||
| -rw-r--r-- | source/blender/gpu/intern/gpu_select_private.h | 53 | ||||
| -rw-r--r-- | source/blender/gpu/intern/gpu_select_sample_query.c | 209 |
6 files changed, 1129 insertions, 157 deletions
diff --git a/source/blender/gpu/intern/gpu_codegen.c b/source/blender/gpu/intern/gpu_codegen.c index 211394e7932..c3896fbd659 100644 --- a/source/blender/gpu/intern/gpu_codegen.c +++ b/source/blender/gpu/intern/gpu_codegen.c @@ -360,7 +360,7 @@ static void codegen_print_datatype(DynStr *ds, const GPUType type, float *data) BLI_dynstr_appendf(ds, "%s(", GPU_DATATYPE_STR[type]); for (i = 0; i < type; i++) { - BLI_dynstr_appendf(ds, "%f", data[i]); + BLI_dynstr_appendf(ds, "%.12f", data[i]); if (i == type - 1) BLI_dynstr_append(ds, ")"); else diff --git a/source/blender/gpu/intern/gpu_debug.c b/source/blender/gpu/intern/gpu_debug.c index d632e767ca9..ba68d1a6a0f 100644 --- a/source/blender/gpu/intern/gpu_debug.c +++ b/source/blender/gpu/intern/gpu_debug.c @@ -161,7 +161,7 @@ const char *gpuErrorString(GLenum err) #endif -static const char* source_name(GLenum source) +static const char *source_name(GLenum source) { switch (source) { case GL_DEBUG_SOURCE_API: return "API"; @@ -174,7 +174,7 @@ static const char* source_name(GLenum source) } } -static const char* message_type_name(GLenum message) +static const char *message_type_name(GLenum message) { switch (message) { case GL_DEBUG_TYPE_ERROR: return "error"; @@ -188,7 +188,7 @@ static const char* message_type_name(GLenum message) } } -static const char* category_name_amd(GLenum category) +static const char *category_name_amd(GLenum category) { switch (category) { case GL_DEBUG_CATEGORY_API_ERROR_AMD: return "API error"; diff --git a/source/blender/gpu/intern/gpu_select.c b/source/blender/gpu/intern/gpu_select.c index 58582232cd5..9496ff137dc 100644 --- a/source/blender/gpu/intern/gpu_select.c +++ b/source/blender/gpu/intern/gpu_select.c @@ -29,109 +29,86 @@ * Interface for accessing gpu-related methods for selection. The semantics will be * similar to glRenderMode(GL_SELECT) since the goal is to maintain compatibility. */ +#include <stdlib.h> + #include "GPU_select.h" #include "GPU_extensions.h" #include "GPU_glew.h" - + #include "MEM_guardedalloc.h" #include "DNA_userdef_types.h" #include "BLI_utildefines.h" -/* Ad hoc number of queries to allocate to skip doing many glGenQueries */ -#define ALLOC_QUERIES 200 - -typedef struct GPUQueryState { +#include "gpu_select_private.h" + +/* Internal algorithm used */ +enum { + /** GL_SELECT, legacy OpenGL selection */ + ALGO_GL_LEGACY = 1, + /** glBegin/EndQuery(GL_SAMPLES_PASSED... ), `gpu_select_query.c` + * Only sets 4th component (ID) correctly. */ + ALGO_GL_QUERY = 2, + /** Read depth buffer for every drawing pass and extract depths, `gpu_select_pick.c` + * Only sets 4th component (ID) correctly. */ + ALGO_GL_PICK = 3, +}; + +typedef struct GPUSelectState { /* To ignore selection id calls when not initialized */ bool select_is_active; - /* Tracks whether a query has been issued so that gpu_load_id can end the previous one */ - bool query_issued; - /* array holding the OpenGL query identifiers */ - unsigned int *queries; - /* array holding the id corresponding to each query */ - unsigned int *id; - /* number of queries in *queries and *id */ - unsigned int num_of_queries; - /* index to the next query to start */ - unsigned int active_query; /* flag to cache user preference for occlusion based selection */ bool use_gpu_select; - /* cache on initialization */ - unsigned int *buffer; - /* buffer size (stores number of integers, for actual size multiply by sizeof integer)*/ - unsigned int bufsize; /* mode of operation */ char mode; - unsigned int index; - int oldhits; -} GPUQueryState; + /* internal algorithm for selection */ + char algorithm; + /* allow GPU_select_begin/end without drawing */ + bool use_cache; +} GPUSelectState; -static GPUQueryState g_query_state = {0}; +static GPUSelectState g_select_state = {0}; /** * initialize and provide buffer for results */ -void GPU_select_begin(unsigned int *buffer, unsigned int bufsize, rctf *input, char mode, int oldhits) +void GPU_select_begin(unsigned int *buffer, unsigned int bufsize, const rcti *input, char mode, int oldhits) { - g_query_state.select_is_active = true; - g_query_state.query_issued = false; - g_query_state.active_query = 0; - g_query_state.use_gpu_select = GPU_select_query_check_active(); - g_query_state.num_of_queries = 0; - g_query_state.bufsize = bufsize; - g_query_state.buffer = buffer; - g_query_state.mode = mode; - g_query_state.index = 0; - g_query_state.oldhits = oldhits; + g_select_state.select_is_active = true; + g_select_state.use_gpu_select = GPU_select_query_check_active(); + g_select_state.mode = mode; - if (!g_query_state.use_gpu_select) { - glSelectBuffer(bufsize, (GLuint *)buffer); - glRenderMode(GL_SELECT); - glInitNames(); - glPushName(-1); + if (ELEM(g_select_state.mode, GPU_SELECT_PICK_ALL, GPU_SELECT_PICK_NEAREST)) { + g_select_state.algorithm = ALGO_GL_PICK; + } + else if (!g_select_state.use_gpu_select) { + g_select_state.algorithm = ALGO_GL_LEGACY; } else { - float viewport[4]; - - g_query_state.num_of_queries = ALLOC_QUERIES; - - g_query_state.queries = MEM_mallocN(g_query_state.num_of_queries * sizeof(*g_query_state.queries), "gpu selection queries"); - g_query_state.id = MEM_mallocN(g_query_state.num_of_queries * sizeof(*g_query_state.id), "gpu selection ids"); - glGenQueries(g_query_state.num_of_queries, g_query_state.queries); - - glPushAttrib(GL_DEPTH_BUFFER_BIT | GL_VIEWPORT_BIT); - /* disable writing to the framebuffer */ - glColorMask(GL_FALSE, GL_FALSE, GL_FALSE, GL_FALSE); - - /* In order to save some fill rate we minimize the viewport using rect. - * We need to get the region of the scissor so that our geometry doesn't - * get rejected before the depth test. Should probably cull rect against - * scissor for viewport but this is a rare case I think */ - glGetFloatv(GL_SCISSOR_BOX, viewport); - if (!input || input->xmin == input->xmax) { - glViewport(viewport[0], viewport[1], 24, 24); - } - else { - glViewport(viewport[0], viewport[1], (int)(input->xmax - input->xmin), (int)(input->ymax - input->ymin)); - } + g_select_state.algorithm = ALGO_GL_QUERY; + } - /* occlusion queries operates on fragments that pass tests and since we are interested on all - * objects in the view frustum independently of their order, we need to disable the depth test */ - if (mode == GPU_SELECT_ALL) { - glDisable(GL_DEPTH_TEST); - glDepthMask(GL_FALSE); + switch (g_select_state.algorithm) { + case ALGO_GL_LEGACY: + { + g_select_state.use_cache = false; + glSelectBuffer(bufsize, (GLuint *)buffer); + glRenderMode(GL_SELECT); + glInitNames(); + glPushName(-1); + break; } - else if (mode == GPU_SELECT_NEAREST_FIRST_PASS) { - glClear(GL_DEPTH_BUFFER_BIT); - glEnable(GL_DEPTH_TEST); - glDepthMask(GL_TRUE); - glDepthFunc(GL_LEQUAL); + case ALGO_GL_QUERY: + { + g_select_state.use_cache = false; + gpu_select_query_begin((unsigned int (*)[4])buffer, bufsize / 4, input, mode, oldhits); + break; } - else if (mode == GPU_SELECT_NEAREST_SECOND_PASS) { - glEnable(GL_DEPTH_TEST); - glDepthMask(GL_FALSE); - glDepthFunc(GL_EQUAL); + default: /* ALGO_GL_PICK */ + { + gpu_select_pick_begin((unsigned int (*)[4])buffer, bufsize / 4, input, mode); + break; } } } @@ -146,41 +123,24 @@ void GPU_select_begin(unsigned int *buffer, unsigned int bufsize, rctf *input, c bool GPU_select_load_id(unsigned int id) { /* if no selection mode active, ignore */ - if (!g_query_state.select_is_active) + if (!g_select_state.select_is_active) return true; - if (!g_query_state.use_gpu_select) { - glLoadName(id); - } - else { - if (g_query_state.query_issued) { - glEndQuery(GL_SAMPLES_PASSED); + switch (g_select_state.algorithm) { + case ALGO_GL_LEGACY: + { + glLoadName(id); + return true; } - /* if required, allocate extra queries */ - if (g_query_state.active_query == g_query_state.num_of_queries) { - g_query_state.num_of_queries += ALLOC_QUERIES; - g_query_state.queries = MEM_reallocN(g_query_state.queries, g_query_state.num_of_queries * sizeof(*g_query_state.queries)); - g_query_state.id = MEM_reallocN(g_query_state.id, g_query_state.num_of_queries * sizeof(*g_query_state.id)); - glGenQueries(ALLOC_QUERIES, &g_query_state.queries[g_query_state.active_query]); + case ALGO_GL_QUERY: + { + return gpu_select_query_load_id(id); } - - glBeginQuery(GL_SAMPLES_PASSED, g_query_state.queries[g_query_state.active_query]); - g_query_state.id[g_query_state.active_query] = id; - g_query_state.active_query++; - g_query_state.query_issued = true; - - if (g_query_state.mode == GPU_SELECT_NEAREST_SECOND_PASS && g_query_state.index < g_query_state.oldhits) { - if (g_query_state.buffer[g_query_state.index * 4 + 3] == id) { - g_query_state.index++; - return true; - } - else { - return false; - } + default: /* ALGO_GL_PICK */ + { + return gpu_select_pick_load_id(id); } } - - return true; } /** @@ -191,59 +151,27 @@ bool GPU_select_load_id(unsigned int id) unsigned int GPU_select_end(void) { unsigned int hits = 0; - if (!g_query_state.use_gpu_select) { - glPopName(); - hits = glRenderMode(GL_RENDER); - } - else { - int i; - if (g_query_state.query_issued) { - glEndQuery(GL_SAMPLES_PASSED); + switch (g_select_state.algorithm) { + case ALGO_GL_LEGACY: + { + glPopName(); + hits = glRenderMode(GL_RENDER); + break; } - - for (i = 0; i < g_query_state.active_query; i++) { - unsigned int result; - glGetQueryObjectuiv(g_query_state.queries[i], GL_QUERY_RESULT, &result); - if (result > 0) { - if (g_query_state.mode != GPU_SELECT_NEAREST_SECOND_PASS) { - int maxhits = g_query_state.bufsize / 4; - - if (hits < maxhits) { - g_query_state.buffer[hits * 4] = 1; - g_query_state.buffer[hits * 4 + 1] = 0xFFFF; - g_query_state.buffer[hits * 4 + 2] = 0xFFFF; - g_query_state.buffer[hits * 4 + 3] = g_query_state.id[i]; - - hits++; - } - else { - hits = -1; - break; - } - } - else { - int j; - /* search in buffer and make selected object first */ - for (j = 0; j < g_query_state.oldhits; j++) { - if (g_query_state.buffer[j * 4 + 3] == g_query_state.id[i]) { - g_query_state.buffer[j * 4 + 1] = 0; - g_query_state.buffer[j * 4 + 2] = 0; - } - } - break; - } - } + case ALGO_GL_QUERY: + { + hits = gpu_select_query_end(); + break; + } + default: /* ALGO_GL_PICK */ + { + hits = gpu_select_pick_end(); + break; } - - glDeleteQueries(g_query_state.num_of_queries, g_query_state.queries); - MEM_freeN(g_query_state.queries); - MEM_freeN(g_query_state.id); - glPopAttrib(); - glColorMask(GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE); } - g_query_state.select_is_active = false; + g_select_state.select_is_active = false; return hits; } @@ -260,3 +188,41 @@ bool GPU_select_query_check_active(void) GPU_type_matches(GPU_DEVICE_NVIDIA, GPU_OS_UNIX, GPU_DRIVER_OPENSOURCE)))); } + +/* ---------------------------------------------------------------------------- + * Caching + * + * Support multiple begin/end's as long as they are within the initial region. + * Currently only used by ALGO_GL_PICK. + */ + +void GPU_select_cache_begin(void) +{ + /* validate on GPU_select_begin, clear if not supported */ + BLI_assert(g_select_state.use_cache == false); + g_select_state.use_cache = true; + if (g_select_state.algorithm == ALGO_GL_PICK) { + gpu_select_pick_cache_begin(); + } +} + +void GPU_select_cache_load_id(void) +{ + BLI_assert(g_select_state.use_cache == true); + if (g_select_state.algorithm == ALGO_GL_PICK) { + gpu_select_pick_cache_load_id(); + } +} + +void GPU_select_cache_end(void) +{ + if (g_select_state.algorithm == ALGO_GL_PICK) { + gpu_select_pick_cache_end(); + } + g_select_state.use_cache = false; +} + +bool GPU_select_is_cached(void) +{ + return g_select_state.use_cache && gpu_select_pick_is_cached(); +} diff --git a/source/blender/gpu/intern/gpu_select_pick.c b/source/blender/gpu/intern/gpu_select_pick.c new file mode 100644 index 00000000000..0a77420fa25 --- /dev/null +++ b/source/blender/gpu/intern/gpu_select_pick.c @@ -0,0 +1,744 @@ +/* + * ***** BEGIN GPL LICENSE BLOCK ***** + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public License + * as published by the Free Software Foundation; either version 2 + * of the License, or (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software Foundation, + * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. + * + * The Original Code is Copyright (C) 2017 Blender Foundation. + * All rights reserved. + * + * ***** END GPL LICENSE BLOCK ***** + */ + +/** \file blender/gpu/intern/gpu_select_pick.c + * \ingroup gpu + * + * Custom select code for picking small regions (not efficient for large regions). + * `gpu_select_pick_*` API. + */ +#include <string.h> +#include <stdlib.h> +#include <float.h> + +#include "GPU_select.h" +#include "GPU_extensions.h" +#include "GPU_glew.h" + +#include "MEM_guardedalloc.h" + +#include "BLI_rect.h" +#include "BLI_listbase.h" +#include "BLI_math_vector.h" +#include "BLI_utildefines.h" + +#include "gpu_select_private.h" + +#include "BLI_strict_flags.h" + +/* #define DEBUG_PRINT */ + +/* Alloc number for depths */ +#define ALLOC_DEPTHS 200 + +/* Z-depth of cleared depth buffer */ +#define DEPTH_MAX 0xffffffff + +/* ---------------------------------------------------------------------------- + * SubRectStride + */ + +/* For looping over a sub-region of a rect, could be moved into 'rct.c'*/ +typedef struct SubRectStride { + unsigned int start; /* start here */ + unsigned int span; /* read these */ + unsigned int span_len; /* len times (read span 'len' times). */ + unsigned int skip; /* skip those */ +} SubRectStride; + +/* we may want to change back to float if uint isn't well supported */ +typedef unsigned int depth_t; + +/** + * Calculate values needed for looping over a sub-region (smaller buffer within a larger buffer). + * + * 'src' must be bigger than 'dst'. + */ +static void rect_subregion_stride_calc(const rcti *src, const rcti *dst, SubRectStride *r_sub) +{ + const int src_x = BLI_rcti_size_x(src); + // const int src_y = BLI_rcti_size_y(src); + const int dst_x = BLI_rcti_size_x(dst); + const int dst_y = BLI_rcti_size_y(dst); + const int x = dst->xmin - src->xmin; + const int y = dst->ymin - src->ymin; + + BLI_assert(src->xmin <= dst->xmin && src->ymin <= dst->ymin && + src->ymax >= dst->ymax && src->ymax >= dst->ymax); + BLI_assert(x >= 0 && y >= 0); + + r_sub->start = (unsigned int)((src_x * y) + x); + r_sub->span = (unsigned int)dst_x; + r_sub->span_len = (unsigned int)dst_y; + r_sub->skip = (unsigned int)(src_x - dst_x); +} + +/** + * Ignore depth clearing as a change, + * only check if its been changed _and_ filled in (ignore clearing since XRAY does this). + */ +BLI_INLINE bool depth_is_filled(const depth_t *prev, const depth_t *curr) +{ + return (*prev != *curr) && (*curr != DEPTH_MAX); +} + +/* ---------------------------------------------------------------------------- + * DepthBufCache + * + * Result of reading glReadPixels, + * use for both cache and non-cached storage. + */ + +/* store result of glReadPixels */ +typedef struct DepthBufCache { + struct DepthBufCache *next, *prev; + unsigned int id; + depth_t buf[0]; +} DepthBufCache; + +static DepthBufCache *depth_buf_malloc(unsigned int rect_len) +{ + DepthBufCache *rect = MEM_mallocN(sizeof(DepthBufCache) + sizeof(depth_t) * rect_len, __func__); + rect->id = SELECT_ID_NONE; + return rect; +} + +static bool depth_buf_rect_depth_any( + const DepthBufCache *rect_depth, + unsigned int rect_len) +{ + const depth_t *curr = rect_depth->buf; + for (unsigned int i = 0; i < rect_len; i++, curr++) { + if (*curr != DEPTH_MAX) { + return true; + } + } + return false; +} + +static bool depth_buf_subrect_depth_any( + const DepthBufCache *rect_depth, + const SubRectStride *sub_rect) +{ + const depth_t *curr = rect_depth->buf + sub_rect->start; + for (unsigned int i = 0; i < sub_rect->span_len; i++) { + const depth_t *curr_end = curr + sub_rect->span; + for (; curr < curr_end; curr++, curr++) { + if (*curr != DEPTH_MAX) { + return true; + } + } + curr += sub_rect->skip; + } + return false; +} + +static bool depth_buf_rect_depth_any_filled( + const DepthBufCache *rect_prev, const DepthBufCache *rect_curr, + unsigned int rect_len) +{ +#if 0 + return memcmp(rect_depth_a->buf, rect_depth_b->buf, rect_len * sizeof(depth_t)) != 0; +#else + const depth_t *prev = rect_prev->buf; + const depth_t *curr = rect_curr->buf; + for (unsigned int i = 0; i < rect_len; i++, curr++, prev++) { + if (depth_is_filled(prev, curr)) { + return true; + } + } + return false; +#endif +} + +/** + * Both buffers are the same size, just check if the sub-rect contains any differences. + */ +static bool depth_buf_subrect_depth_any_filled( + const DepthBufCache *rect_src, const DepthBufCache *rect_dst, + const SubRectStride *sub_rect) +{ + /* same as above but different rect sizes */ + const depth_t *prev = rect_src->buf + sub_rect->start; + const depth_t *curr = rect_dst->buf + sub_rect->start; + for (unsigned int i = 0; i < sub_rect->span_len; i++) { + const depth_t *curr_end = curr + sub_rect->span; + for (; curr < curr_end; prev++, curr++) { + if (depth_is_filled(prev, curr)) { + return true; + } + } + prev += sub_rect->skip; + curr += sub_rect->skip; + } + return false; +} + +/* ---------------------------------------------------------------------------- + * DepthID + * + * Internal structure for storing hits. + */ + +typedef struct DepthID { + unsigned int id; + depth_t depth; +} DepthID; + +static int depth_id_cmp(const void *v1, const void *v2) +{ + const DepthID *d1 = v1, *d2 = v2; + if (d1->id < d2->id) { + return -1; + } + else if (d1->id > d2->id) { + return 1; + } + else { + return 0; + } +} + +static int depth_cmp(const void *v1, const void *v2) +{ + const DepthID *d1 = v1, *d2 = v2; + if (d1->depth < d2->depth) { + return -1; + } + else if (d1->depth > d2->depth) { + return 1; + } + else { + return 0; + } +} + +/* depth sorting */ +typedef struct GPUPickState { + /* cache on initialization */ + unsigned int (*buffer)[4]; + + /* buffer size (stores number of integers, for actual size multiply by sizeof integer)*/ + unsigned int bufsize; + /* mode of operation */ + char mode; + + /* OpenGL drawing, never use when (is_cached == true). */ + struct { + /* The current depth, accumulated as we draw */ + DepthBufCache *rect_depth; + /* Scratch buffer, avoid allocs every time (when not caching) */ + DepthBufCache *rect_depth_test; + + /* Pass to glReadPixels (x, y, w, h) */ + int clip_readpixels[4]; + + /* Set after first draw */ + bool is_init; + unsigned int prev_id; + } gl; + + /* src: data stored in 'cache' and 'gl', + * dst: use when cached region is smaller (where src -> dst isn't 1:1) */ + struct { + rcti clip_rect; + unsigned int rect_len; + } src, dst; + + /* Store cache between `GPU_select_cache_begin/end` */ + bool use_cache; + bool is_cached; + struct { + /* Cleanup used for iterating over both source and destination buffers: + * src.clip_rect -> dst.clip_rect */ + SubRectStride sub_rect; + + /* List of DepthBufCache, sized of 'src.clip_rect' */ + ListBase bufs; + } cache; + + /* Pickign methods */ + union { + /* GPU_SELECT_PICK_ALL */ + struct { + DepthID *hits; + unsigned int hits_len; + unsigned int hits_len_alloc; + } all; + + /* GPU_SELECT_PICK_NEAREST */ + struct { + unsigned int *rect_id; + } nearest; + }; +} GPUPickState; + + +static GPUPickState g_pick_state = {0}; + +void gpu_select_pick_begin( + unsigned int (*buffer)[4], unsigned int bufsize, + const rcti *input, char mode) +{ + GPUPickState *ps = &g_pick_state; + +#ifdef DEBUG_PRINT + printf("%s: mode=%d, use_cache=%d, is_cache=%d\n", __func__, mode, ps->use_cache, ps->is_cached); +#endif + + ps->bufsize = bufsize; + ps->buffer = buffer; + ps->mode = mode; + + const unsigned int rect_len = (unsigned int)(BLI_rcti_size_x(input) * BLI_rcti_size_y(input)); + ps->dst.clip_rect = *input; + ps->dst.rect_len = rect_len; + + /* Restrict OpenGL operations for when we don't have cache */ + if (ps->is_cached == false) { + + glPushAttrib(GL_DEPTH_BUFFER_BIT | GL_VIEWPORT_BIT); + /* disable writing to the framebuffer */ + glColorMask(GL_FALSE, GL_FALSE, GL_FALSE, GL_FALSE); + + glEnable(GL_DEPTH_TEST); + glDepthMask(GL_TRUE); + + if (mode == GPU_SELECT_PICK_ALL) { + glDepthFunc(GL_ALWAYS); + } + else { + glDepthFunc(GL_LEQUAL); + } + + /* set just in case */ + glPixelTransferf(GL_DEPTH_BIAS, 0.0); + glPixelTransferf(GL_DEPTH_SCALE, 1.0); + + float viewport[4]; + glGetFloatv(GL_SCISSOR_BOX, viewport); + + ps->src.clip_rect = *input; + ps->src.rect_len = rect_len; + + ps->gl.clip_readpixels[0] = (int)viewport[0]; + ps->gl.clip_readpixels[1] = (int)viewport[1]; + ps->gl.clip_readpixels[2] = BLI_rcti_size_x(&ps->src.clip_rect); + ps->gl.clip_readpixels[3] = BLI_rcti_size_y(&ps->src.clip_rect); + + glViewport(UNPACK4(ps->gl.clip_readpixels)); + + /* It's possible we don't want to clear depth buffer, + * so existing elements are masked by current z-buffer. */ + glClear(GL_DEPTH_BUFFER_BIT); + + /* scratch buffer (read new values here) */ + ps->gl.rect_depth_test = depth_buf_malloc(rect_len); + ps->gl.rect_depth = depth_buf_malloc(rect_len); + + /* set initial 'far' value */ +#if 0 + glReadPixels(UNPACK4(ps->gl.clip_readpixels), GL_DEPTH_COMPONENT, GL_UNSIGNED_INT, ps->gl.rect_depth->buf); +#else + for (unsigned int i = 0; i < rect_len; i++) { + ps->gl.rect_depth->buf[i] = DEPTH_MAX; + } +#endif + + ps->gl.is_init = false; + ps->gl.prev_id = 0; + } + else { + /* Using cache (ps->is_cached == true) */ + /* src.clip_rect -> dst.clip_rect */ + rect_subregion_stride_calc(&ps->src.clip_rect, &ps->dst.clip_rect, &ps->cache.sub_rect); + BLI_assert(ps->gl.rect_depth == NULL); + BLI_assert(ps->gl.rect_depth_test == NULL); + } + + if (mode == GPU_SELECT_PICK_ALL) { + ps->all.hits = MEM_mallocN(sizeof(*ps->all.hits) * ALLOC_DEPTHS, __func__); + ps->all.hits_len = 0; + ps->all.hits_len_alloc = ALLOC_DEPTHS; + } + else { + /* Set to 0xff for SELECT_ID_NONE */ + ps->nearest.rect_id = MEM_mallocN(sizeof(unsigned int) * ps->dst.rect_len, __func__); + memset(ps->nearest.rect_id, 0xff, sizeof(unsigned int) * ps->dst.rect_len); + } +} + +/** + * Given 2x depths, we know are different - update the depth information + * use for both cached/uncached depth buffers. + */ +static void gpu_select_load_id_pass_all(const DepthBufCache *rect_curr) +{ + GPUPickState *ps = &g_pick_state; + const unsigned int id = rect_curr->id; + /* find the best depth for this pass and store in 'all.hits' */ + depth_t depth_best = DEPTH_MAX; + +#define EVAL_TEST() \ + if (depth_best > *curr) { \ + depth_best = *curr; \ + } ((void)0) + + if (ps->is_cached == false) { + const depth_t *curr = rect_curr->buf; + BLI_assert(ps->src.rect_len == ps->dst.rect_len); + const unsigned int rect_len = ps->src.rect_len; + for (unsigned int i = 0; i < rect_len; i++, curr++) { + EVAL_TEST(); + } + } + else { + /* same as above but different rect sizes */ + const depth_t *curr = rect_curr->buf + ps->cache.sub_rect.start; + for (unsigned int i = 0; i < ps->cache.sub_rect.span_len; i++) { + const depth_t *curr_end = curr + ps->cache.sub_rect.span; + for (; curr < curr_end; curr++) { + EVAL_TEST(); + } + curr += ps->cache.sub_rect.skip; + } + } + +#undef EVAL_TEST + + /* ensure enough space */ + if (UNLIKELY(ps->all.hits_len == ps->all.hits_len_alloc)) { + ps->all.hits_len_alloc += ALLOC_DEPTHS; + ps->all.hits = MEM_reallocN(ps->all.hits, ps->all.hits_len_alloc * sizeof(*ps->all.hits)); + } + DepthID *d = &ps->all.hits[ps->all.hits_len++]; + d->id = id; + d->depth = depth_best; +} + +static void gpu_select_load_id_pass_nearest(const DepthBufCache *rect_prev, const DepthBufCache *rect_curr) +{ + GPUPickState *ps = &g_pick_state; + const unsigned int id = rect_curr->id; + /* keep track each pixels ID in 'nearest.rect_id' */ + if (id != SELECT_ID_NONE) { + unsigned int *id_ptr = ps->nearest.rect_id; + + /* Check against DEPTH_MAX because XRAY will clear the buffer, + * so previously set values will become unset. + * In this case just leave those id's left as-is. */ +#define EVAL_TEST() \ + if (depth_is_filled(prev, curr)) { \ + *id_ptr = id; \ + } ((void)0) + + if (ps->is_cached == false) { + const depth_t *prev = rect_prev->buf; + const depth_t *curr = rect_curr->buf; + BLI_assert(ps->src.rect_len == ps->dst.rect_len); + const unsigned int rect_len = ps->src.rect_len; + for (unsigned int i = 0; i < rect_len; i++, curr++, prev++, id_ptr++) { + EVAL_TEST(); + } + } + else { + /* same as above but different rect sizes */ + const depth_t *prev = rect_prev->buf + ps->cache.sub_rect.start; + const depth_t *curr = rect_curr->buf + ps->cache.sub_rect.start; + for (unsigned int i = 0; i < ps->cache.sub_rect.span_len; i++) { + const depth_t *curr_end = curr + ps->cache.sub_rect.span; + for (; curr < curr_end; prev++, curr++, id_ptr++) { + EVAL_TEST(); + } + prev += ps->cache.sub_rect.skip; + curr += ps->cache.sub_rect.skip; + } + } + +#undef EVAL_TEST + } +} + + +bool gpu_select_pick_load_id(unsigned int id) +{ + GPUPickState *ps = &g_pick_state; + if (ps->gl.is_init) { + const unsigned int rect_len = ps->src.rect_len; + glReadPixels(UNPACK4(ps->gl.clip_readpixels), GL_DEPTH_COMPONENT, GL_UNSIGNED_INT, ps->gl.rect_depth_test->buf); + /* perform initial check since most cases the array remains unchanged */ + + bool do_pass = false; + if (g_pick_state.mode == GPU_SELECT_PICK_ALL) { + if (depth_buf_rect_depth_any(ps->gl.rect_depth_test, rect_len)) { + ps->gl.rect_depth_test->id = ps->gl.prev_id; + gpu_select_load_id_pass_all(ps->gl.rect_depth_test); + do_pass = true; + } + } + else { + if (depth_buf_rect_depth_any_filled(ps->gl.rect_depth, ps->gl.rect_depth_test, rect_len)) { + ps->gl.rect_depth_test->id = ps->gl.prev_id; + gpu_select_load_id_pass_nearest(ps->gl.rect_depth, ps->gl.rect_depth_test); + do_pass = true; + } + } + + if (do_pass) { + /* Store depth in cache */ + if (ps->use_cache) { + BLI_addtail(&ps->cache.bufs, ps->gl.rect_depth); + ps->gl.rect_depth = depth_buf_malloc(ps->src.rect_len); + } + + SWAP(DepthBufCache *, ps->gl.rect_depth, ps->gl.rect_depth_test); + + if (g_pick_state.mode == GPU_SELECT_PICK_ALL) { + /* we want new depths every time */ + glClear(GL_DEPTH_BUFFER_BIT); + } + } + } + + ps->gl.is_init = true; + ps->gl.prev_id = id; + + return true; +} + +unsigned int gpu_select_pick_end(void) +{ + GPUPickState *ps = &g_pick_state; + +#ifdef DEBUG_PRINT + printf("%s\n", __func__); +#endif + + if (ps->is_cached == false) { + if (ps->gl.is_init) { + /* force finishing last pass */ + gpu_select_pick_load_id(ps->gl.prev_id); + } + + glPopAttrib(); + glColorMask(GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE); + } + + /* assign but never free directly since it may be in cache */ + DepthBufCache *rect_depth_final; + + /* Store depth in cache */ + if (ps->use_cache && !ps->is_cached) { + BLI_addtail(&ps->cache.bufs, ps->gl.rect_depth); + ps->gl.rect_depth = NULL; + rect_depth_final = ps->cache.bufs.last; + } + else if (ps->is_cached) { + rect_depth_final = ps->cache.bufs.last; + } + else { + /* common case, no cache */ + rect_depth_final = ps->gl.rect_depth; + } + + unsigned int maxhits = g_pick_state.bufsize; + DepthID *depth_data; + unsigned int depth_data_len = 0; + + if (g_pick_state.mode == GPU_SELECT_PICK_ALL) { + depth_data = ps->all.hits; + depth_data_len = ps->all.hits_len; + /* move ownership */ + ps->all.hits = NULL; + ps->all.hits_len = 0; + ps->all.hits_len_alloc = 0; + } + else { + /* GPU_SELECT_PICK_NEAREST */ + + /* Over alloc (unlikely we have as many depths as pixels) */ + unsigned int depth_data_len_first_pass = 0; + depth_data = MEM_mallocN(ps->dst.rect_len * sizeof(*depth_data), __func__); + + /* Partially de-duplicating copy, + * when contiguous ID's are found - update their closest depth. + * This isn't essential but means there is less data to sort. */ + +#define EVAL_TEST(i_src, i_dst) \ + { \ + const unsigned int id = ps->nearest.rect_id[i_dst]; \ + if (id != SELECT_ID_NONE) { \ + const depth_t depth = rect_depth_final->buf[i_src]; \ + if (depth_last == NULL || depth_last->id != id) { \ + DepthID *d = &depth_data[depth_data_len_first_pass++]; \ + d->id = id; \ + d->depth = depth; \ + } \ + else if (depth_last->depth > depth) { \ + depth_last->depth = depth; \ + } \ + } \ + } ((void)0) + + { + DepthID *depth_last = NULL; + if (ps->is_cached == false) { + for (unsigned int i = 0; i < ps->src.rect_len; i++) { + EVAL_TEST(i, i); + } + } + else { + /* same as above but different rect sizes */ + unsigned int i_src = ps->cache.sub_rect.start, i_dst = 0; + for (unsigned int j = 0; j < ps->cache.sub_rect.span_len; j++) { + const unsigned int i_src_end = i_src + ps->cache.sub_rect.span; + for (; i_src < i_src_end; i_src++, i_dst++) { + EVAL_TEST(i_src, i_dst); + } + i_src += ps->cache.sub_rect.skip; + } + } + } + +#undef EVAL_TEST + + qsort(depth_data, depth_data_len_first_pass, sizeof(DepthID), depth_id_cmp); + + /* Sort by ID's then keep the best depth for each ID */ + depth_data_len = 0; + { + DepthID *depth_last = NULL; + for (unsigned int i = 0; i < depth_data_len_first_pass; i++) { + if (depth_last == NULL || depth_last->id != depth_data[i].id) { + depth_last = &depth_data[depth_data_len++]; + *depth_last = depth_data[i]; + } + else if (depth_last->depth > depth_data[i].depth) { + depth_last->depth = depth_data[i].depth; + } + } + } + } + + /* Finally sort each unique (id, depth) pair by depth + * so the final hit-list is sorted by depth (nearest first) */ + unsigned int hits = 0; + + if (depth_data_len > maxhits) { + hits = (unsigned int)-1; + } + else { + /* leave sorting up to the caller */ + qsort(depth_data, depth_data_len, sizeof(DepthID), depth_cmp); + + for (unsigned int i = 0; i < depth_data_len; i++) { +#ifdef DEBUG_PRINT + printf(" hit: %u: depth %u\n", depth_data[i].id, depth_data[i].depth); +#endif + /* first 3 are dummy values */ + g_pick_state.buffer[hits][0] = 1; + g_pick_state.buffer[hits][1] = 0x0; /* depth_data[i].depth; */ /* unused */ + g_pick_state.buffer[hits][2] = 0x0; /* z-far is currently never used. */ + g_pick_state.buffer[hits][3] = depth_data[i].id; + hits++; + } + BLI_assert(hits < maxhits); + } + + MEM_freeN(depth_data); + + MEM_SAFE_FREE(ps->gl.rect_depth); + MEM_SAFE_FREE(ps->gl.rect_depth_test); + + if (g_pick_state.mode == GPU_SELECT_PICK_ALL) { + /* 'hits' already freed as 'depth_data' */ + } + else { + MEM_freeN(ps->nearest.rect_id); + ps->nearest.rect_id = NULL; + } + + if (ps->use_cache) { + ps->is_cached = true; + } + + return hits; +} + +/* ---------------------------------------------------------------------------- + * Caching + * + * Support multiple begin/end's reusing depth buffers. + */ + +void gpu_select_pick_cache_begin(void) +{ + BLI_assert(g_pick_state.use_cache == false); +#ifdef DEBUG_PRINT + printf("%s\n", __func__); +#endif + g_pick_state.use_cache = true; + g_pick_state.is_cached = false; +} + +void gpu_select_pick_cache_end(void) +{ +#ifdef DEBUG_PRINT + printf("%s: with %d buffers\n", __func__, BLI_listbase_count(&g_pick_state.cache.bufs)); +#endif + g_pick_state.use_cache = false; + g_pick_state.is_cached = false; + + BLI_freelistN(&g_pick_state.cache.bufs); +} + +/* is drawing needed? */ +bool gpu_select_pick_is_cached(void) +{ + return g_pick_state.is_cached; +} + +void gpu_select_pick_cache_load_id(void) +{ + BLI_assert(g_pick_state.is_cached == true); + GPUPickState *ps = &g_pick_state; +#ifdef DEBUG_PRINT + printf("%s (building depth from cache)\n", __func__); +#endif + for (DepthBufCache *rect_depth = ps->cache.bufs.first; rect_depth; rect_depth = rect_depth->next) { + if (rect_depth->next != NULL) { + /* we know the buffers differ, but this sub-region may not. + * double check before adding an id-pass */ + if (g_pick_state.mode == GPU_SELECT_PICK_ALL) { + if (depth_buf_subrect_depth_any(rect_depth->next, &ps->cache.sub_rect)) { + gpu_select_load_id_pass_all(rect_depth->next); + } + } + else { + if (depth_buf_subrect_depth_any_filled(rect_depth, rect_depth->next, &ps->cache.sub_rect)) { + gpu_select_load_id_pass_nearest(rect_depth, rect_depth->next); + } + } + } + } +} diff --git a/source/blender/gpu/intern/gpu_select_private.h b/source/blender/gpu/intern/gpu_select_private.h new file mode 100644 index 00000000000..8935bd7b253 --- /dev/null +++ b/source/blender/gpu/intern/gpu_select_private.h @@ -0,0 +1,53 @@ +/* + * ***** BEGIN GPL LICENSE BLOCK ***** + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public License + * as published by the Free Software Foundation; either version 2 + * of the License, or (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software Foundation, + * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. + * + * The Original Code is Copyright (C) 2014 Blender Foundation. + * All rights reserved. + * + * Contributor(s): Antony Riakiotakis. + * + * ***** END GPL LICENSE BLOCK ***** + */ + +/** \file blender/gpu/intern/gpu_select_private.h + * \ingroup gpu + * + * Selection implementations. + */ + +#ifndef __GPU_SELECT_PRIVATE_H__ +#define __GPU_SELECT_PRIVATE_H__ + +/* gpu_select_pick */ +void gpu_select_pick_begin(unsigned int (*buffer)[4], unsigned int bufsize, const rcti *input, char mode); +bool gpu_select_pick_load_id(unsigned int id); +unsigned int gpu_select_pick_end(void); + +void gpu_select_pick_cache_begin(void); +void gpu_select_pick_cache_end(void); +bool gpu_select_pick_is_cached(void); +void gpu_select_pick_cache_load_id(void); + +/* gpu_select_sample_query */ +void gpu_select_query_begin(unsigned int (*buffer)[4], unsigned int bufsize, const rcti *input, char mode, int oldhits); +bool gpu_select_query_load_id(unsigned int id); +unsigned int gpu_select_query_end(void); + + +#define SELECT_ID_NONE ((unsigned int)0xffffffff) + +#endif /* __GPU_SELECT_PRIVATE_H__ */ diff --git a/source/blender/gpu/intern/gpu_select_sample_query.c b/source/blender/gpu/intern/gpu_select_sample_query.c new file mode 100644 index 00000000000..ba5fefc5227 --- /dev/null +++ b/source/blender/gpu/intern/gpu_select_sample_query.c @@ -0,0 +1,209 @@ +/* + * ***** BEGIN GPL LICENSE BLOCK ***** + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public License + * as published by the Free Software Foundation; either version 2 + * of the License, or (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software Foundation, + * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. + * + * The Original Code is Copyright (C) 2014 Blender Foundation. + * All rights reserved. + * + * Contributor(s): Antony Riakiotakis. + * + * ***** END GPL LICENSE BLOCK ***** + */ + +/** \file blender/gpu/intern/gpu_select_sample_query.c + * \ingroup gpu + * + * Interface for accessing gpu-related methods for selection. The semantics will be + * similar to glRenderMode(GL_SELECT) since the goal is to maintain compatibility. + */ + +#include <stdlib.h> + +#include "GPU_select.h" +#include "GPU_extensions.h" +#include "GPU_glew.h" + +#include "MEM_guardedalloc.h" + +#include "BLI_rect.h" + +#include "BLI_utildefines.h" + +#include "gpu_select_private.h" + + +/* Ad hoc number of queries to allocate to skip doing many glGenQueries */ +#define ALLOC_QUERIES 200 + +typedef struct GPUQueryState { + /* Tracks whether a query has been issued so that gpu_load_id can end the previous one */ + bool query_issued; + /* array holding the OpenGL query identifiers */ + unsigned int *queries; + /* array holding the id corresponding to each query */ + unsigned int *id; + /* number of queries in *queries and *id */ + unsigned int num_of_queries; + /* index to the next query to start */ + unsigned int active_query; + /* cache on initialization */ + unsigned int (*buffer)[4]; + /* buffer size (stores number of integers, for actual size multiply by sizeof integer)*/ + unsigned int bufsize; + /* mode of operation */ + char mode; + unsigned int index; + int oldhits; +} GPUQueryState; + +static GPUQueryState g_query_state = {0}; + + +void gpu_select_query_begin( + unsigned int (*buffer)[4], unsigned int bufsize, + const rcti *input, char mode, + int oldhits) +{ + float viewport[4]; + + g_query_state.query_issued = false; + g_query_state.active_query = 0; + g_query_state.num_of_queries = 0; + g_query_state.bufsize = bufsize; + g_query_state.buffer = buffer; + g_query_state.mode = mode; + g_query_state.index = 0; + g_query_state.oldhits = oldhits; + + g_query_state.num_of_queries = ALLOC_QUERIES; + + g_query_state.queries = MEM_mallocN(g_query_state.num_of_queries * sizeof(*g_query_state.queries), "gpu selection queries"); + g_query_state.id = MEM_mallocN(g_query_state.num_of_queries * sizeof(*g_query_state.id), "gpu selection ids"); + glGenQueries(g_query_state.num_of_queries, g_query_state.queries); + + glPushAttrib(GL_DEPTH_BUFFER_BIT | GL_VIEWPORT_BIT); + /* disable writing to the framebuffer */ + glColorMask(GL_FALSE, GL_FALSE, GL_FALSE, GL_FALSE); + + /* In order to save some fill rate we minimize the viewport using rect. + * We need to get the region of the scissor so that our geometry doesn't + * get rejected before the depth test. Should probably cull rect against + * scissor for viewport but this is a rare case I think */ + glGetFloatv(GL_SCISSOR_BOX, viewport); + glViewport(viewport[0], viewport[1], BLI_rcti_size_x(input), BLI_rcti_size_y(input)); + + /* occlusion queries operates on fragments that pass tests and since we are interested on all + * objects in the view frustum independently of their order, we need to disable the depth test */ + if (mode == GPU_SELECT_ALL) { + glDisable(GL_DEPTH_TEST); + glDepthMask(GL_FALSE); + } + else if (mode == GPU_SELECT_NEAREST_FIRST_PASS) { + glClear(GL_DEPTH_BUFFER_BIT); + glEnable(GL_DEPTH_TEST); + glDepthMask(GL_TRUE); + glDepthFunc(GL_LEQUAL); + } + else if (mode == GPU_SELECT_NEAREST_SECOND_PASS) { + glEnable(GL_DEPTH_TEST); + glDepthMask(GL_FALSE); + glDepthFunc(GL_EQUAL); + } +} + +bool gpu_select_query_load_id(unsigned int id) +{ + if (g_query_state.query_issued) { + glEndQuery(GL_SAMPLES_PASSED); + } + /* if required, allocate extra queries */ + if (g_query_state.active_query == g_query_state.num_of_queries) { + g_query_state.num_of_queries += ALLOC_QUERIES; + g_query_state.queries = MEM_reallocN(g_query_state.queries, g_query_state.num_of_queries * sizeof(*g_query_state.queries)); + g_query_state.id = MEM_reallocN(g_query_state.id, g_query_state.num_of_queries * sizeof(*g_query_state.id)); + glGenQueries(ALLOC_QUERIES, &g_query_state.queries[g_query_state.active_query]); + } + + glBeginQuery(GL_SAMPLES_PASSED, g_query_state.queries[g_query_state.active_query]); + g_query_state.id[g_query_state.active_query] = id; + g_query_state.active_query++; + g_query_state.query_issued = true; + + if (g_query_state.mode == GPU_SELECT_NEAREST_SECOND_PASS && g_query_state.index < g_query_state.oldhits) { + if (g_query_state.buffer[g_query_state.index][3] == id) { + g_query_state.index++; + return true; + } + else { + return false; + } + } + + return true; +} + +unsigned int gpu_select_query_end(void) +{ + int i; + + unsigned int hits = 0; + const unsigned int maxhits = g_query_state.bufsize; + + if (g_query_state.query_issued) { + glEndQuery(GL_SAMPLES_PASSED); + } + + for (i = 0; i < g_query_state.active_query; i++) { + unsigned int result; + glGetQueryObjectuiv(g_query_state.queries[i], GL_QUERY_RESULT, &result); + if (result > 0) { + if (g_query_state.mode != GPU_SELECT_NEAREST_SECOND_PASS) { + + if (hits < maxhits) { + g_query_state.buffer[hits][0] = 1; + g_query_state.buffer[hits][1] = 0xFFFF; + g_query_state.buffer[hits][2] = 0xFFFF; + g_query_state.buffer[hits][3] = g_query_state.id[i]; + + hits++; + } + else { + hits = -1; + break; + } + } + else { + int j; + /* search in buffer and make selected object first */ + for (j = 0; j < g_query_state.oldhits; j++) { + if (g_query_state.buffer[j][3] == g_query_state.id[i]) { + g_query_state.buffer[j][1] = 0; + g_query_state.buffer[j][2] = 0; + } + } + break; + } + } + } + + glDeleteQueries(g_query_state.num_of_queries, g_query_state.queries); + MEM_freeN(g_query_state.queries); + MEM_freeN(g_query_state.id); + glPopAttrib(); + glColorMask(GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE); + + return hits; +} |