Welcome to mirror list, hosted at ThFree Co, Russian Federation.

git.blender.org/blender.git - Unnamed repository; edit this file 'description' to name the repository.
summaryrefslogtreecommitdiff
path: root/source/blender/draw/intern/draw_hair.cc
diff options
context:
space:
mode:
Diffstat (limited to 'source/blender/draw/intern/draw_hair.cc')
-rw-r--r--source/blender/draw/intern/draw_hair.cc264
1 files changed, 151 insertions, 113 deletions
diff --git a/source/blender/draw/intern/draw_hair.cc b/source/blender/draw/intern/draw_hair.cc
index dc791314333..08e5d780cba 100644
--- a/source/blender/draw/intern/draw_hair.cc
+++ b/source/blender/draw/intern/draw_hair.cc
@@ -22,6 +22,7 @@
#include "GPU_batch.h"
#include "GPU_capabilities.h"
#include "GPU_compute.h"
+#include "GPU_context.h"
#include "GPU_material.h"
#include "GPU_shader.h"
#include "GPU_texture.h"
@@ -33,25 +34,17 @@
#include "draw_shader.h"
#include "draw_shader_shared.h"
-#ifndef __APPLE__
-# define USE_TRANSFORM_FEEDBACK
-# define USE_COMPUTE_SHADERS
-#endif
-
BLI_INLINE eParticleRefineShaderType drw_hair_shader_type_get()
{
-#ifdef USE_COMPUTE_SHADERS
if (GPU_compute_shader_support() && GPU_shader_storage_buffer_objects_support()) {
return PART_REFINE_SHADER_COMPUTE;
}
-#endif
-#ifdef USE_TRANSFORM_FEEDBACK
- return PART_REFINE_SHADER_TRANSFORM_FEEDBACK;
-#endif
+ if (GPU_transform_feedback_support()) {
+ return PART_REFINE_SHADER_TRANSFORM_FEEDBACK;
+ }
return PART_REFINE_SHADER_TRANSFORM_FEEDBACK_WORKAROUND;
}
-#ifndef USE_TRANSFORM_FEEDBACK
struct ParticleRefineCall {
struct ParticleRefineCall *next;
GPUVertBuf *vbo;
@@ -63,11 +56,10 @@ static ParticleRefineCall *g_tf_calls = nullptr;
static int g_tf_id_offset;
static int g_tf_target_width;
static int g_tf_target_height;
-#endif
static GPUVertBuf *g_dummy_vbo = nullptr;
static GPUTexture *g_dummy_texture = nullptr;
-static DRWPass *g_tf_pass; /* XXX can be a problem with multiple DRWManager in the future */
+static DRWPass *g_tf_pass; /* XXX can be a problem with multiple #DRWManager in the future */
static blender::draw::UniformBuffer<CurvesInfos> *g_dummy_curves_info = nullptr;
static GPUShader *hair_refine_shader_get(ParticleRefineShader refinement)
@@ -77,23 +69,25 @@ static GPUShader *hair_refine_shader_get(ParticleRefineShader refinement)
void DRW_hair_init(void)
{
-#if defined(USE_TRANSFORM_FEEDBACK) || defined(USE_COMPUTE_SHADERS)
- g_tf_pass = DRW_pass_create("Update Hair Pass", DRW_STATE_NO_DRAW);
-#else
- g_tf_pass = DRW_pass_create("Update Hair Pass", DRW_STATE_WRITE_COLOR);
-#endif
+ if (GPU_transform_feedback_support() || GPU_compute_shader_support()) {
+ g_tf_pass = DRW_pass_create("Update Hair Pass", DRW_STATE_NO_DRAW);
+ }
+ else {
+ g_tf_pass = DRW_pass_create("Update Hair Pass", DRW_STATE_WRITE_COLOR);
+ }
if (g_dummy_vbo == nullptr) {
/* initialize vertex format */
GPUVertFormat format = {0};
uint dummy_id = GPU_vertformat_attr_add(&format, "dummy", GPU_COMP_F32, 4, GPU_FETCH_FLOAT);
- g_dummy_vbo = GPU_vertbuf_create_with_format(&format);
+ g_dummy_vbo = GPU_vertbuf_create_with_format_ex(
+ &format, GPU_USAGE_STATIC | GPU_USAGE_FLAG_BUFFER_TEXTURE_ONLY);
const float vert[4] = {0.0f, 0.0f, 0.0f, 0.0f};
GPU_vertbuf_data_alloc(g_dummy_vbo, 1);
GPU_vertbuf_attr_fill(g_dummy_vbo, dummy_id, vert);
- /* Create vbo immediately to bind to texture buffer. */
+ /* Create VBO immediately to bind to texture buffer. */
GPU_vertbuf_use(g_dummy_vbo);
g_dummy_texture = GPU_texture_create_from_vertbuf("hair_dummy_attr", g_dummy_vbo);
@@ -111,9 +105,9 @@ static void drw_hair_particle_cache_shgrp_attach_resources(DRWShadingGroup *shgr
ParticleHairCache *cache,
const int subdiv)
{
- DRW_shgroup_uniform_texture(shgrp, "hairPointBuffer", cache->point_tex);
- DRW_shgroup_uniform_texture(shgrp, "hairStrandBuffer", cache->strand_tex);
- DRW_shgroup_uniform_texture(shgrp, "hairStrandSegBuffer", cache->strand_seg_tex);
+ DRW_shgroup_buffer_texture(shgrp, "hairPointBuffer", cache->proc_point_buf);
+ DRW_shgroup_buffer_texture(shgrp, "hairStrandBuffer", cache->proc_strand_buf);
+ DRW_shgroup_buffer_texture(shgrp, "hairStrandSegBuffer", cache->proc_strand_seg_buf);
DRW_shgroup_uniform_int(shgrp, "hairStrandsRes", &cache->final[subdiv].strands_res, 1);
}
@@ -146,22 +140,25 @@ static void drw_hair_particle_cache_update_transform_feedback(ParticleHairCache
if (final_points_len > 0) {
GPUShader *tf_shader = hair_refine_shader_get(PART_REFINE_CATMULL_ROM);
-#ifdef USE_TRANSFORM_FEEDBACK
- DRWShadingGroup *tf_shgrp = DRW_shgroup_transform_feedback_create(
- tf_shader, g_tf_pass, cache->final[subdiv].proc_buf);
-#else
- DRWShadingGroup *tf_shgrp = DRW_shgroup_create(tf_shader, g_tf_pass);
-
- ParticleRefineCall *pr_call = (ParticleRefineCall *)MEM_mallocN(sizeof(*pr_call), __func__);
- pr_call->next = g_tf_calls;
- pr_call->vbo = cache->final[subdiv].proc_buf;
- pr_call->shgrp = tf_shgrp;
- pr_call->vert_len = final_points_len;
- g_tf_calls = pr_call;
- DRW_shgroup_uniform_int(tf_shgrp, "targetHeight", &g_tf_target_height, 1);
- DRW_shgroup_uniform_int(tf_shgrp, "targetWidth", &g_tf_target_width, 1);
- DRW_shgroup_uniform_int(tf_shgrp, "idOffset", &g_tf_id_offset, 1);
-#endif
+ DRWShadingGroup *tf_shgrp = nullptr;
+ if (GPU_transform_feedback_support()) {
+ tf_shgrp = DRW_shgroup_transform_feedback_create(
+ tf_shader, g_tf_pass, cache->final[subdiv].proc_buf);
+ }
+ else {
+ tf_shgrp = DRW_shgroup_create(tf_shader, g_tf_pass);
+
+ ParticleRefineCall *pr_call = (ParticleRefineCall *)MEM_mallocN(sizeof(*pr_call), __func__);
+ pr_call->next = g_tf_calls;
+ pr_call->vbo = cache->final[subdiv].proc_buf;
+ pr_call->shgrp = tf_shgrp;
+ pr_call->vert_len = final_points_len;
+ g_tf_calls = pr_call;
+ DRW_shgroup_uniform_int(tf_shgrp, "targetHeight", &g_tf_target_height, 1);
+ DRW_shgroup_uniform_int(tf_shgrp, "targetWidth", &g_tf_target_width, 1);
+ DRW_shgroup_uniform_int(tf_shgrp, "idOffset", &g_tf_id_offset, 1);
+ }
+ BLI_assert(tf_shgrp != nullptr);
drw_hair_particle_cache_shgrp_attach_resources(tf_shgrp, cache, subdiv);
DRW_shgroup_call_procedural_points(tf_shgrp, nullptr, final_points_len);
@@ -205,8 +202,8 @@ GPUVertBuf *DRW_hair_pos_buffer_get(Object *object, ParticleSystem *psys, Modifi
}
void DRW_hair_duplimat_get(Object *object,
- ParticleSystem *UNUSED(psys),
- ModifierData *UNUSED(md),
+ ParticleSystem * /*psys*/,
+ ModifierData * /*md*/,
float (*dupli_mat)[4])
{
Object *dupli_parent = DRW_object_get_dupli_parent(object);
@@ -250,7 +247,7 @@ DRWShadingGroup *DRW_shgroup_hair_create_sub(Object *object,
DRWShadingGroup *shgrp = DRW_shgroup_create_sub(shgrp_parent);
- /* TODO: optimize this. Only bind the ones GPUMaterial needs. */
+ /* TODO: optimize this. Only bind the ones #GPUMaterial needs. */
for (int i = 0; i < hair_cache->num_uv_layers; i++) {
for (int n = 0; n < MAX_LAYER_NAME_CT && hair_cache->uv_layer_names[i][n][0] != '\0'; n++) {
DRW_shgroup_uniform_texture(shgrp, hair_cache->uv_layer_names[i][n], hair_cache->uv_tex[i]);
@@ -283,9 +280,9 @@ DRWShadingGroup *DRW_shgroup_hair_create_sub(Object *object,
float hair_rad_tip = part->rad_tip * part->rad_scale * 0.5f;
bool hair_close_tip = (part->shape_flag & PART_SHAPE_CLOSE_TIP) != 0;
- DRW_shgroup_uniform_texture(shgrp, "hairPointBuffer", hair_cache->final[subdiv].proc_tex);
- if (hair_cache->length_tex) {
- DRW_shgroup_uniform_texture(shgrp, "l", hair_cache->length_tex);
+ DRW_shgroup_buffer_texture(shgrp, "hairPointBuffer", hair_cache->final[subdiv].proc_buf);
+ if (hair_cache->proc_length_buf) {
+ DRW_shgroup_buffer_texture(shgrp, "l", hair_cache->proc_length_buf);
}
DRW_shgroup_uniform_block(shgrp, "drw_curves", *g_dummy_curves_info);
@@ -296,6 +293,11 @@ DRWShadingGroup *DRW_shgroup_hair_create_sub(Object *object,
DRW_shgroup_uniform_float_copy(shgrp, "hairRadRoot", hair_rad_root);
DRW_shgroup_uniform_float_copy(shgrp, "hairRadTip", hair_rad_tip);
DRW_shgroup_uniform_bool_copy(shgrp, "hairCloseTip", hair_close_tip);
+ if (gpu_material) {
+ /* \note: This needs to happen before the drawcall to allow correct attribute extraction.
+ * (see T101896) */
+ DRW_shgroup_add_material_resources(shgrp, gpu_material);
+ }
/* TODO(fclem): Until we have a better way to cull the hair and render with orco, bypass
* culling test. */
GPUBatch *geom = hair_cache->final[subdiv].proc_hairs[thickness_res - 1];
@@ -306,81 +308,117 @@ DRWShadingGroup *DRW_shgroup_hair_create_sub(Object *object,
void DRW_hair_update()
{
-#ifndef USE_TRANSFORM_FEEDBACK
- /**
- * Workaround to transform feedback not working on mac.
- * On some system it crashes (see T58489) and on some other it renders garbage (see T60171).
- *
- * So instead of using transform feedback we render to a texture,
- * read back the result to system memory and re-upload as VBO data.
- * It is really not ideal performance wise, but it is the simplest
- * and the most local workaround that still uses the power of the GPU.
- */
-
- if (g_tf_calls == nullptr) {
- return;
- }
+ if (!GPU_transform_feedback_support()) {
+ /**
+ * Workaround to transform feedback not working on mac.
+ * On some system it crashes (see T58489) and on some other it renders garbage (see T60171).
+ *
+ * So instead of using transform feedback we render to a texture,
+ * read back the result to system memory and re-upload as VBO data.
+ * It is really not ideal performance wise, but it is the simplest
+ * and the most local workaround that still uses the power of the GPU.
+ */
+
+ if (g_tf_calls == nullptr) {
+ return;
+ }
- /* Search ideal buffer size. */
- uint max_size = 0;
- for (ParticleRefineCall *pr_call = g_tf_calls; pr_call; pr_call = pr_call->next) {
- max_size = max_ii(max_size, pr_call->vert_len);
- }
+ /* Search ideal buffer size. */
+ uint max_size = 0;
+ for (ParticleRefineCall *pr_call = g_tf_calls; pr_call; pr_call = pr_call->next) {
+ max_size = max_ii(max_size, pr_call->vert_len);
+ }
+
+ /* Create target Texture / Frame-buffer */
+ /* Don't use max size as it can be really heavy and fail.
+ * Do chunks of maximum 2048 * 2048 hair points. */
+ int width = 2048;
+ int height = min_ii(width, 1 + max_size / width);
+ GPUTexture *tex = DRW_texture_pool_query_2d(
+ width, height, GPU_RGBA32F, (DrawEngineType *)DRW_hair_update);
+ g_tf_target_height = height;
+ g_tf_target_width = width;
+
+ GPUFrameBuffer *fb = nullptr;
+ GPU_framebuffer_ensure_config(&fb,
+ {
+ GPU_ATTACHMENT_NONE,
+ GPU_ATTACHMENT_TEXTURE(tex),
+ });
+
+ float *data = (float *)MEM_mallocN(sizeof(float[4]) * width * height, "tf fallback buffer");
+
+ GPU_framebuffer_bind(fb);
+ while (g_tf_calls != nullptr) {
+ ParticleRefineCall *pr_call = g_tf_calls;
+ g_tf_calls = g_tf_calls->next;
+
+ g_tf_id_offset = 0;
+ while (pr_call->vert_len > 0) {
+ int max_read_px_len = min_ii(width * height, pr_call->vert_len);
+
+ DRW_draw_pass_subset(g_tf_pass, pr_call->shgrp, pr_call->shgrp);
+ /* Read back result to main memory. */
+ GPU_framebuffer_read_color(fb, 0, 0, width, height, 4, 0, GPU_DATA_FLOAT, data);
+ /* Upload back to VBO. */
+ GPU_vertbuf_use(pr_call->vbo);
+ GPU_vertbuf_update_sub(pr_call->vbo,
+ sizeof(float[4]) * g_tf_id_offset,
+ sizeof(float[4]) * max_read_px_len,
+ data);
+
+ g_tf_id_offset += max_read_px_len;
+ pr_call->vert_len -= max_read_px_len;
+ }
- /* Create target Texture / Frame-buffer */
- /* Don't use max size as it can be really heavy and fail.
- * Do chunks of maximum 2048 * 2048 hair points. */
- int width = 2048;
- int height = min_ii(width, 1 + max_size / width);
- GPUTexture *tex = DRW_texture_pool_query_2d(
- width, height, GPU_RGBA32F, (DrawEngineType *)DRW_hair_update);
- g_tf_target_height = height;
- g_tf_target_width = width;
-
- GPUFrameBuffer *fb = nullptr;
- GPU_framebuffer_ensure_config(&fb,
- {
- GPU_ATTACHMENT_NONE,
- GPU_ATTACHMENT_TEXTURE(tex),
- });
-
- float *data = (float *)MEM_mallocN(sizeof(float[4]) * width * height, "tf fallback buffer");
-
- GPU_framebuffer_bind(fb);
- while (g_tf_calls != nullptr) {
- ParticleRefineCall *pr_call = g_tf_calls;
- g_tf_calls = g_tf_calls->next;
-
- g_tf_id_offset = 0;
- while (pr_call->vert_len > 0) {
- int max_read_px_len = min_ii(width * height, pr_call->vert_len);
-
- DRW_draw_pass_subset(g_tf_pass, pr_call->shgrp, pr_call->shgrp);
- /* Read back result to main memory. */
- GPU_framebuffer_read_color(fb, 0, 0, width, height, 4, 0, GPU_DATA_FLOAT, data);
- /* Upload back to VBO. */
- GPU_vertbuf_use(pr_call->vbo);
- GPU_vertbuf_update_sub(pr_call->vbo,
- sizeof(float[4]) * g_tf_id_offset,
- sizeof(float[4]) * max_read_px_len,
- data);
-
- g_tf_id_offset += max_read_px_len;
- pr_call->vert_len -= max_read_px_len;
+ MEM_freeN(pr_call);
}
- MEM_freeN(pr_call);
+ MEM_freeN(data);
+ GPU_framebuffer_free(fb);
}
+ else {
+ /* NOTE(Metal): If compute is not supported, bind a temporary frame-buffer to avoid
+ * side-effects from rendering in the active buffer.
+ * We also need to guarantee that a frame-buffer is active to perform any rendering work,
+ * even if there is no output. */
+ GPUFrameBuffer *temp_fb = nullptr;
+ GPUFrameBuffer *prev_fb = nullptr;
+ if (GPU_type_matches_ex(GPU_DEVICE_ANY, GPU_OS_MAC, GPU_DRIVER_ANY, GPU_BACKEND_METAL)) {
+ if (!GPU_compute_shader_support()) {
+ prev_fb = GPU_framebuffer_active_get();
+ char errorOut[256];
+ /* if the frame-buffer is invalid we need a dummy frame-buffer to be bound. */
+ if (!GPU_framebuffer_check_valid(prev_fb, errorOut)) {
+ int width = 64;
+ int height = 64;
+ GPUTexture *tex = DRW_texture_pool_query_2d(
+ width, height, GPU_DEPTH_COMPONENT32F, (DrawEngineType *)DRW_hair_update);
+ g_tf_target_height = height;
+ g_tf_target_width = width;
+
+ GPU_framebuffer_ensure_config(&temp_fb, {GPU_ATTACHMENT_TEXTURE(tex)});
+
+ GPU_framebuffer_bind(temp_fb);
+ }
+ }
+ }
- MEM_freeN(data);
- GPU_framebuffer_free(fb);
-#else
- /* Just render the pass when using compute shaders or transform feedback. */
- DRW_draw_pass(g_tf_pass);
- if (drw_hair_shader_type_get() == PART_REFINE_SHADER_COMPUTE) {
- GPU_memory_barrier(GPU_BARRIER_SHADER_STORAGE);
+ /* Just render the pass when using compute shaders or transform feedback. */
+ DRW_draw_pass(g_tf_pass);
+ if (drw_hair_shader_type_get() == PART_REFINE_SHADER_COMPUTE) {
+ GPU_memory_barrier(GPU_BARRIER_SHADER_STORAGE);
+ }
+
+ /* Release temporary frame-buffer. */
+ if (temp_fb != nullptr) {
+ GPU_framebuffer_free(temp_fb);
+ }
+ /* Rebind existing frame-buffer */
+ if (prev_fb != nullptr) {
+ GPU_framebuffer_bind(prev_fb);
+ }
}
-#endif
}
void DRW_hair_free(void)
generated by cgit v1.2.3 (git 2.25.1) at 2024-06-16 18:15:46 +0300