VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
/* SPDX-FileCopyrightText: 2012-2024 Blender Authors
|
2023-05-31 16:19:06 +02:00
|
|
|
*
|
|
|
|
|
* SPDX-License-Identifier: GPL-2.0-or-later */
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2019-02-18 08:08:12 +11:00
|
|
|
/** \file
|
|
|
|
|
* \ingroup bke
|
2012-08-19 15:41:56 +00:00
|
|
|
*/
|
|
|
|
|
|
2023-07-22 11:27:25 +10:00
|
|
|
#include <cstddef>
|
|
|
|
|
#include <cstring>
|
2012-08-19 15:41:56 +00:00
|
|
|
|
|
|
|
|
#include "MEM_guardedalloc.h"
|
|
|
|
|
|
2024-09-19 18:14:49 +02:00
|
|
|
#include "BLI_array.hh"
|
2012-08-19 15:41:56 +00:00
|
|
|
#include "BLI_listbase.h"
|
2024-09-11 12:42:03 +02:00
|
|
|
#include "BLI_math_geom.h"
|
2024-09-19 18:14:49 +02:00
|
|
|
#include "BLI_math_vector.hh"
|
2012-08-19 15:41:56 +00:00
|
|
|
#include "BLI_string.h"
|
2023-10-18 17:15:30 +02:00
|
|
|
#include "BLI_string_utils.hh"
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
#include "BLI_task.hh"
|
2012-08-19 15:41:56 +00:00
|
|
|
#include "BLI_utildefines.h"
|
|
|
|
|
|
2024-02-09 18:59:42 +01:00
|
|
|
#include "BLT_translation.hh"
|
2013-03-25 08:29:06 +00:00
|
|
|
|
2019-08-15 20:23:13 +02:00
|
|
|
#include "DNA_mask_types.h"
|
2015-12-28 11:55:14 +01:00
|
|
|
#include "DNA_scene_types.h"
|
2020-03-19 09:33:03 +01:00
|
|
|
#include "DNA_sequence_types.h"
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2023-12-21 10:10:53 +01:00
|
|
|
#include "BKE_colortools.hh"
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2024-01-18 22:50:23 +02:00
|
|
|
#include "IMB_colormanagement.hh"
|
|
|
|
|
#include "IMB_imbuf.hh"
|
|
|
|
|
#include "IMB_imbuf_types.hh"
|
2020-11-01 21:03:31 +01:00
|
|
|
|
2023-11-02 01:05:06 +01:00
|
|
|
#include "SEQ_modifier.hh"
|
|
|
|
|
#include "SEQ_render.hh"
|
|
|
|
|
#include "SEQ_sound.hh"
|
2024-05-07 19:37:24 +02:00
|
|
|
#include "SEQ_time.hh"
|
|
|
|
|
#include "SEQ_utils.hh"
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2023-08-28 15:01:05 +02:00
|
|
|
#include "BLO_read_write.hh"
|
2020-11-06 14:27:51 +01:00
|
|
|
|
2023-11-02 01:05:06 +01:00
|
|
|
#include "render.hh"
|
2020-10-26 00:47:06 +01:00
|
|
|
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
using namespace blender;
|
|
|
|
|
|
2012-08-19 15:41:56 +00:00
|
|
|
static SequenceModifierTypeInfo *modifiersTypes[NUM_SEQUENCE_MODIFIER_TYPES];
|
2014-03-20 15:45:20 +06:00
|
|
|
static bool modifierTypesInit = false;
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2020-06-16 12:32:42 +10:00
|
|
|
/* -------------------------------------------------------------------- */
|
|
|
|
|
/** \name Modifier Multi-Threading Utilities
|
|
|
|
|
* \{ */
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2023-07-24 11:04:19 +10:00
|
|
|
using modifier_apply_threaded_cb = void (*)(int width,
|
|
|
|
|
int height,
|
|
|
|
|
uchar *rect,
|
|
|
|
|
float *rect_float,
|
|
|
|
|
uchar *mask_rect,
|
|
|
|
|
const float *mask_rect_float,
|
|
|
|
|
void *data_v);
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2023-07-20 09:46:24 +02:00
|
|
|
struct ModifierInitData {
|
2012-08-19 15:41:56 +00:00
|
|
|
ImBuf *ibuf;
|
|
|
|
|
ImBuf *mask;
|
|
|
|
|
void *user_data;
|
|
|
|
|
|
|
|
|
|
modifier_apply_threaded_cb apply_callback;
|
2023-07-20 09:46:24 +02:00
|
|
|
};
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2023-07-20 09:46:24 +02:00
|
|
|
struct ModifierThread {
|
2012-08-19 15:41:56 +00:00
|
|
|
int width, height;
|
|
|
|
|
|
2022-09-25 17:04:52 +10:00
|
|
|
uchar *rect, *mask_rect;
|
2012-08-19 15:41:56 +00:00
|
|
|
float *rect_float, *mask_rect_float;
|
|
|
|
|
|
|
|
|
|
void *user_data;
|
|
|
|
|
|
|
|
|
|
modifier_apply_threaded_cb apply_callback;
|
2023-07-20 09:46:24 +02:00
|
|
|
};
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2020-11-05 13:33:27 +01:00
|
|
|
/**
|
2020-11-06 14:10:59 +01:00
|
|
|
* \a timeline_frame is offset by \a fra_offset only in case we are using a real mask.
|
2020-11-05 13:33:27 +01:00
|
|
|
*/
|
|
|
|
|
static ImBuf *modifier_render_mask_input(const SeqRenderData *context,
|
|
|
|
|
int mask_input_type,
|
|
|
|
|
Sequence *mask_sequence,
|
|
|
|
|
Mask *mask_id,
|
2020-11-06 14:10:59 +01:00
|
|
|
int timeline_frame,
|
2020-11-05 13:33:27 +01:00
|
|
|
int fra_offset,
|
|
|
|
|
bool make_float)
|
|
|
|
|
{
|
2023-07-20 09:46:24 +02:00
|
|
|
ImBuf *mask_input = nullptr;
|
2020-11-05 13:33:27 +01:00
|
|
|
|
|
|
|
|
if (mask_input_type == SEQUENCE_MASK_INPUT_STRIP) {
|
|
|
|
|
if (mask_sequence) {
|
|
|
|
|
SeqRenderState state;
|
|
|
|
|
|
2020-11-06 14:10:59 +01:00
|
|
|
mask_input = seq_render_strip(context, &state, mask_sequence, timeline_frame);
|
2020-11-05 13:33:27 +01:00
|
|
|
|
|
|
|
|
if (make_float) {
|
2023-05-18 10:19:01 +02:00
|
|
|
if (!mask_input->float_buffer.data) {
|
2020-11-05 13:33:27 +01:00
|
|
|
IMB_float_from_rect(mask_input);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
else {
|
2023-05-18 10:19:01 +02:00
|
|
|
if (!mask_input->byte_buffer.data) {
|
2020-11-05 13:33:27 +01:00
|
|
|
IMB_rect_from_float(mask_input);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
else if (mask_input_type == SEQUENCE_MASK_INPUT_ID) {
|
2020-11-06 14:10:59 +01:00
|
|
|
mask_input = seq_render_mask(context, mask_id, timeline_frame - fra_offset, make_float);
|
2020-11-05 13:33:27 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return mask_input;
|
|
|
|
|
}
|
|
|
|
|
|
2015-03-20 12:39:25 +01:00
|
|
|
static ImBuf *modifier_mask_get(SequenceModifierData *smd,
|
|
|
|
|
const SeqRenderData *context,
|
2020-11-06 14:10:59 +01:00
|
|
|
int timeline_frame,
|
2015-03-20 12:39:25 +01:00
|
|
|
int fra_offset,
|
|
|
|
|
bool make_float)
|
2012-08-19 15:41:56 +00:00
|
|
|
{
|
2020-11-05 13:33:27 +01:00
|
|
|
return modifier_render_mask_input(context,
|
|
|
|
|
smd->mask_input_type,
|
|
|
|
|
smd->mask_sequence,
|
|
|
|
|
smd->mask_id,
|
2020-11-06 14:10:59 +01:00
|
|
|
timeline_frame,
|
2020-11-05 13:33:27 +01:00
|
|
|
fra_offset,
|
|
|
|
|
make_float);
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void modifier_init_handle(void *handle_v, int start_line, int tot_line, void *init_data_v)
|
|
|
|
|
{
|
|
|
|
|
ModifierThread *handle = (ModifierThread *)handle_v;
|
|
|
|
|
ModifierInitData *init_data = (ModifierInitData *)init_data_v;
|
|
|
|
|
ImBuf *ibuf = init_data->ibuf;
|
|
|
|
|
ImBuf *mask = init_data->mask;
|
|
|
|
|
|
|
|
|
|
int offset = 4 * start_line * ibuf->x;
|
|
|
|
|
|
|
|
|
|
memset(handle, 0, sizeof(ModifierThread));
|
|
|
|
|
|
|
|
|
|
handle->width = ibuf->x;
|
|
|
|
|
handle->height = tot_line;
|
|
|
|
|
handle->apply_callback = init_data->apply_callback;
|
|
|
|
|
handle->user_data = init_data->user_data;
|
|
|
|
|
|
2023-05-18 10:19:01 +02:00
|
|
|
if (ibuf->byte_buffer.data) {
|
|
|
|
|
handle->rect = ibuf->byte_buffer.data + offset;
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2023-05-18 10:19:01 +02:00
|
|
|
if (ibuf->float_buffer.data) {
|
|
|
|
|
handle->rect_float = ibuf->float_buffer.data + offset;
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
2012-08-19 15:41:56 +00:00
|
|
|
|
|
|
|
|
if (mask) {
|
2023-05-18 10:19:01 +02:00
|
|
|
if (mask->byte_buffer.data) {
|
|
|
|
|
handle->mask_rect = mask->byte_buffer.data + offset;
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2023-05-18 10:19:01 +02:00
|
|
|
if (mask->float_buffer.data) {
|
|
|
|
|
handle->mask_rect_float = mask->float_buffer.data + offset;
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
else {
|
2023-07-20 09:46:24 +02:00
|
|
|
handle->mask_rect = nullptr;
|
|
|
|
|
handle->mask_rect_float = nullptr;
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void *modifier_do_thread(void *thread_data_v)
|
|
|
|
|
{
|
|
|
|
|
ModifierThread *td = (ModifierThread *)thread_data_v;
|
|
|
|
|
|
|
|
|
|
td->apply_callback(td->width,
|
|
|
|
|
td->height,
|
|
|
|
|
td->rect,
|
|
|
|
|
td->rect_float,
|
|
|
|
|
td->mask_rect,
|
|
|
|
|
td->mask_rect_float,
|
|
|
|
|
td->user_data);
|
|
|
|
|
|
2023-07-20 09:46:24 +02:00
|
|
|
return nullptr;
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void modifier_apply_threaded(ImBuf *ibuf,
|
|
|
|
|
ImBuf *mask,
|
|
|
|
|
modifier_apply_threaded_cb apply_callback,
|
|
|
|
|
void *user_data)
|
|
|
|
|
{
|
|
|
|
|
ModifierInitData init_data;
|
|
|
|
|
|
|
|
|
|
init_data.ibuf = ibuf;
|
|
|
|
|
init_data.mask = mask;
|
|
|
|
|
init_data.user_data = user_data;
|
|
|
|
|
|
|
|
|
|
init_data.apply_callback = apply_callback;
|
|
|
|
|
|
|
|
|
|
IMB_processor_apply_threaded(
|
|
|
|
|
ibuf->y, sizeof(ModifierThread), &init_data, modifier_init_handle, modifier_do_thread);
|
|
|
|
|
}
|
|
|
|
|
|
2020-06-16 12:32:42 +10:00
|
|
|
/** \} */
|
|
|
|
|
|
|
|
|
|
/* -------------------------------------------------------------------- */
|
|
|
|
|
/** \name Color Balance Modifier
|
|
|
|
|
* \{ */
|
2012-08-19 15:41:56 +00:00
|
|
|
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
static StripColorBalance calc_cb_lgg(const StripColorBalance *cb_)
|
2020-11-05 13:33:27 +01:00
|
|
|
{
|
|
|
|
|
StripColorBalance cb = *cb_;
|
|
|
|
|
int c;
|
|
|
|
|
|
|
|
|
|
for (c = 0; c < 3; c++) {
|
|
|
|
|
cb.lift[c] = 2.0f - cb.lift[c];
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (cb.flag & SEQ_COLOR_BALANCE_INVERSE_LIFT) {
|
|
|
|
|
for (c = 0; c < 3; c++) {
|
|
|
|
|
/* tweak to give more subtle results
|
|
|
|
|
* values above 1.0 are scaled */
|
|
|
|
|
if (cb.lift[c] > 1.0f) {
|
|
|
|
|
cb.lift[c] = pow(cb.lift[c] - 1.0f, 2.0) + 1.0;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
cb.lift[c] = 2.0f - cb.lift[c];
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (cb.flag & SEQ_COLOR_BALANCE_INVERSE_GAIN) {
|
|
|
|
|
for (c = 0; c < 3; c++) {
|
|
|
|
|
if (cb.gain[c] != 0.0f) {
|
|
|
|
|
cb.gain[c] = 1.0f / cb.gain[c];
|
|
|
|
|
}
|
|
|
|
|
else {
|
|
|
|
|
cb.gain[c] = 1000000; /* should be enough :) */
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (!(cb.flag & SEQ_COLOR_BALANCE_INVERSE_GAMMA)) {
|
|
|
|
|
for (c = 0; c < 3; c++) {
|
|
|
|
|
if (cb.gamma[c] != 0.0f) {
|
|
|
|
|
cb.gamma[c] = 1.0f / cb.gamma[c];
|
|
|
|
|
}
|
|
|
|
|
else {
|
|
|
|
|
cb.gamma[c] = 1000000; /* should be enough :) */
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return cb;
|
|
|
|
|
}
|
|
|
|
|
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
static StripColorBalance calc_cb_sop(const StripColorBalance *cb_)
|
2021-09-30 21:09:47 +02:00
|
|
|
{
|
|
|
|
|
StripColorBalance cb = *cb_;
|
|
|
|
|
int c;
|
|
|
|
|
|
|
|
|
|
for (c = 0; c < 3; c++) {
|
|
|
|
|
if (cb.flag & SEQ_COLOR_BALANCE_INVERSE_SLOPE) {
|
|
|
|
|
if (cb.slope[c] != 0.0f) {
|
|
|
|
|
cb.slope[c] = 1.0f / cb.slope[c];
|
|
|
|
|
}
|
|
|
|
|
else {
|
|
|
|
|
cb.slope[c] = 1000000;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (cb.flag & SEQ_COLOR_BALANCE_INVERSE_OFFSET) {
|
|
|
|
|
cb.offset[c] = -1.0f * (cb.offset[c] - 1.0f);
|
|
|
|
|
}
|
|
|
|
|
else {
|
|
|
|
|
cb.offset[c] = cb.offset[c] - 1.0f;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (!(cb.flag & SEQ_COLOR_BALANCE_INVERSE_POWER)) {
|
|
|
|
|
if (cb.power[c] != 0.0f) {
|
|
|
|
|
cb.power[c] = 1.0f / cb.power[c];
|
|
|
|
|
}
|
|
|
|
|
else {
|
|
|
|
|
cb.power[c] = 1000000;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return cb;
|
|
|
|
|
}
|
|
|
|
|
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
static StripColorBalance calc_cb(const StripColorBalance *cb_)
|
2021-09-30 21:09:47 +02:00
|
|
|
{
|
|
|
|
|
if (cb_->method == SEQ_COLOR_BALANCE_METHOD_LIFTGAMMAGAIN) {
|
|
|
|
|
return calc_cb_lgg(cb_);
|
|
|
|
|
}
|
2021-10-01 16:20:31 +10:00
|
|
|
/* `cb_->method == SEQ_COLOR_BALANCE_METHOD_SLOPEOFFSETPOWER`. */
|
|
|
|
|
return calc_cb_sop(cb_);
|
2021-09-30 21:09:47 +02:00
|
|
|
}
|
|
|
|
|
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
/* Lift-Gamma-Gain math. NOTE: lift is actually (2-lift). */
|
|
|
|
|
static float color_balance_lgg(
|
2020-11-05 13:33:27 +01:00
|
|
|
float in, const float lift, const float gain, const float gamma, const float mul)
|
|
|
|
|
{
|
|
|
|
|
float x = (((in - 1.0f) * lift) + 1.0f) * gain;
|
|
|
|
|
|
|
|
|
|
/* prevent NaN */
|
2020-11-06 11:25:27 +11:00
|
|
|
if (x < 0.0f) {
|
|
|
|
|
x = 0.0f;
|
2020-11-05 13:33:27 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
x = powf(x, gamma) * mul;
|
|
|
|
|
CLAMP(x, FLT_MIN, FLT_MAX);
|
|
|
|
|
return x;
|
|
|
|
|
}
|
|
|
|
|
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
/* Slope-Offset-Power (ASC CDL) math, see https://en.wikipedia.org/wiki/ASC_CDL */
|
|
|
|
|
static float color_balance_sop(
|
|
|
|
|
float in, const float slope, const float offset, const float power, float mul)
|
2020-11-05 13:33:27 +01:00
|
|
|
{
|
2021-09-30 21:09:47 +02:00
|
|
|
float x = in * slope + offset;
|
|
|
|
|
|
|
|
|
|
/* prevent NaN */
|
|
|
|
|
if (x < 0.0f) {
|
|
|
|
|
x = 0.0f;
|
|
|
|
|
}
|
2020-11-05 13:33:27 +01:00
|
|
|
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
x = powf(x, power);
|
2021-09-30 21:09:47 +02:00
|
|
|
x *= mul;
|
|
|
|
|
CLAMP(x, FLT_MIN, FLT_MAX);
|
|
|
|
|
return x;
|
|
|
|
|
}
|
|
|
|
|
|
2024-09-13 10:56:26 +10:00
|
|
|
/**
|
|
|
|
|
* Use a larger lookup table than 256 possible byte values: due to alpha
|
|
|
|
|
* pre-multiplication, dark values with low alphas might need more precision.
|
|
|
|
|
*/
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
static constexpr int CB_TABLE_SIZE = 1024;
|
2021-09-30 21:09:47 +02:00
|
|
|
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
static void make_cb_table_lgg(
|
|
|
|
|
float lift, float gain, float gamma, float mul, float r_table[CB_TABLE_SIZE])
|
2021-09-30 21:09:47 +02:00
|
|
|
{
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
for (int i = 0; i < CB_TABLE_SIZE; i++) {
|
|
|
|
|
float x = float(i) * (1.0f / (CB_TABLE_SIZE - 1.0f));
|
|
|
|
|
r_table[i] = color_balance_lgg(x, lift, gain, gamma, mul);
|
2020-11-05 13:33:27 +01:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
static void make_cb_table_sop(
|
|
|
|
|
float slope, float offset, float power, float mul, float r_table[CB_TABLE_SIZE])
|
2020-11-05 13:33:27 +01:00
|
|
|
{
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
for (int i = 0; i < CB_TABLE_SIZE; i++) {
|
|
|
|
|
float x = float(i) * (1.0f / (CB_TABLE_SIZE - 1.0f));
|
|
|
|
|
r_table[i] = color_balance_sop(x, slope, offset, power, mul);
|
2020-11-05 13:33:27 +01:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
static void color_balance_byte(const float cb_tab[3][CB_TABLE_SIZE],
|
|
|
|
|
uchar *rect,
|
|
|
|
|
const uchar *mask_rect,
|
|
|
|
|
int width,
|
|
|
|
|
int height)
|
2020-11-05 13:33:27 +01:00
|
|
|
{
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
uchar *ptr = rect;
|
2024-09-15 23:14:09 +10:00
|
|
|
const uchar *ptr_end = ptr + int64_t(width) * height * 4;
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
const uchar *mask_ptr = mask_rect;
|
|
|
|
|
|
|
|
|
|
if (mask_ptr != nullptr) {
|
|
|
|
|
/* Mask is used.*/
|
|
|
|
|
while (ptr < ptr_end) {
|
|
|
|
|
float pix[4];
|
|
|
|
|
straight_uchar_to_premul_float(pix, ptr);
|
|
|
|
|
|
|
|
|
|
int p0 = int(pix[0] * (CB_TABLE_SIZE - 1.0f) + 0.5f);
|
|
|
|
|
int p1 = int(pix[1] * (CB_TABLE_SIZE - 1.0f) + 0.5f);
|
|
|
|
|
int p2 = int(pix[2] * (CB_TABLE_SIZE - 1.0f) + 0.5f);
|
|
|
|
|
const float t[3] = {mask_ptr[0] / 255.0f, mask_ptr[1] / 255.0f, mask_ptr[2] / 255.0f};
|
|
|
|
|
|
|
|
|
|
pix[0] = pix[0] * (1.0f - t[0]) + t[0] * cb_tab[0][p0];
|
|
|
|
|
pix[1] = pix[1] * (1.0f - t[1]) + t[1] * cb_tab[1][p1];
|
|
|
|
|
pix[2] = pix[2] * (1.0f - t[2]) + t[2] * cb_tab[2][p2];
|
|
|
|
|
|
|
|
|
|
premul_float_to_straight_uchar(ptr, pix);
|
|
|
|
|
ptr += 4;
|
|
|
|
|
mask_ptr += 4;
|
2021-09-30 21:09:47 +02:00
|
|
|
}
|
2020-11-05 13:33:27 +01:00
|
|
|
}
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
else {
|
|
|
|
|
/* No mask. */
|
|
|
|
|
while (ptr < ptr_end) {
|
|
|
|
|
float pix[4];
|
|
|
|
|
straight_uchar_to_premul_float(pix, ptr);
|
|
|
|
|
|
|
|
|
|
int p0 = int(pix[0] * (CB_TABLE_SIZE - 1.0f) + 0.5f);
|
|
|
|
|
int p1 = int(pix[1] * (CB_TABLE_SIZE - 1.0f) + 0.5f);
|
|
|
|
|
int p2 = int(pix[2] * (CB_TABLE_SIZE - 1.0f) + 0.5f);
|
|
|
|
|
pix[0] = cb_tab[0][p0];
|
|
|
|
|
pix[1] = cb_tab[1][p1];
|
|
|
|
|
pix[2] = cb_tab[2][p2];
|
|
|
|
|
premul_float_to_straight_uchar(ptr, pix);
|
|
|
|
|
ptr += 4;
|
2020-11-05 13:33:27 +01:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
static void color_balance_float(const StripColorBalance *cb,
|
|
|
|
|
float *rect_float,
|
|
|
|
|
const float *mask_rect_float,
|
|
|
|
|
int width,
|
|
|
|
|
int height,
|
|
|
|
|
float mul)
|
2020-11-05 13:33:27 +01:00
|
|
|
{
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
float *ptr = rect_float;
|
|
|
|
|
const float *ptr_end = rect_float + int64_t(width) * height * 4;
|
|
|
|
|
const float *mask_ptr = mask_rect_float;
|
|
|
|
|
|
|
|
|
|
if (cb->method == SEQ_COLOR_BALANCE_METHOD_LIFTGAMMAGAIN) {
|
|
|
|
|
/* Lift/Gamma/Gain */
|
|
|
|
|
const float3 lift = cb->lift;
|
|
|
|
|
const float3 gain = cb->gain;
|
|
|
|
|
const float3 gamma = cb->gamma;
|
|
|
|
|
while (ptr < ptr_end) {
|
|
|
|
|
float t0 = color_balance_lgg(ptr[0], lift.x, gain.x, gamma.x, mul);
|
|
|
|
|
float t1 = color_balance_lgg(ptr[1], lift.y, gain.y, gamma.y, mul);
|
|
|
|
|
float t2 = color_balance_lgg(ptr[2], lift.z, gain.z, gamma.z, mul);
|
|
|
|
|
if (mask_ptr) {
|
|
|
|
|
ptr[0] = ptr[0] * (1.0f - mask_ptr[0]) + t0 * mask_ptr[0];
|
|
|
|
|
ptr[1] = ptr[1] * (1.0f - mask_ptr[1]) + t1 * mask_ptr[1];
|
|
|
|
|
ptr[2] = ptr[2] * (1.0f - mask_ptr[2]) + t2 * mask_ptr[2];
|
2021-09-30 21:09:47 +02:00
|
|
|
}
|
|
|
|
|
else {
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
ptr[0] = t0;
|
|
|
|
|
ptr[1] = t1;
|
|
|
|
|
ptr[2] = t2;
|
2021-09-30 21:09:47 +02:00
|
|
|
}
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
ptr += 4;
|
|
|
|
|
if (mask_ptr) {
|
|
|
|
|
mask_ptr += 4;
|
2020-11-05 13:33:27 +01:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
else {
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
/* Slope/Offset/Power */
|
|
|
|
|
const float3 slope = cb->slope;
|
|
|
|
|
const float3 offset = cb->offset;
|
|
|
|
|
const float3 power = cb->power;
|
|
|
|
|
while (ptr < ptr_end) {
|
|
|
|
|
float t0 = color_balance_sop(ptr[0], slope.x, offset.x, power.x, mul);
|
|
|
|
|
float t1 = color_balance_sop(ptr[1], slope.y, offset.y, power.y, mul);
|
|
|
|
|
float t2 = color_balance_sop(ptr[2], slope.z, offset.z, power.z, mul);
|
|
|
|
|
if (mask_ptr) {
|
|
|
|
|
ptr[0] = ptr[0] * (1.0f - mask_ptr[0]) + t0 * mask_ptr[0];
|
|
|
|
|
ptr[1] = ptr[1] * (1.0f - mask_ptr[1]) + t1 * mask_ptr[1];
|
|
|
|
|
ptr[2] = ptr[2] * (1.0f - mask_ptr[2]) + t2 * mask_ptr[2];
|
|
|
|
|
}
|
|
|
|
|
else {
|
|
|
|
|
ptr[0] = t0;
|
|
|
|
|
ptr[1] = t1;
|
|
|
|
|
ptr[2] = t2;
|
|
|
|
|
}
|
|
|
|
|
ptr += 4;
|
|
|
|
|
if (mask_ptr) {
|
|
|
|
|
mask_ptr += 4;
|
|
|
|
|
}
|
|
|
|
|
}
|
2020-11-05 13:33:27 +01:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2012-09-15 01:52:28 +00:00
|
|
|
static void colorBalance_init_data(SequenceModifierData *smd)
|
2012-08-19 15:41:56 +00:00
|
|
|
{
|
|
|
|
|
ColorBalanceModifierData *cbmd = (ColorBalanceModifierData *)smd;
|
|
|
|
|
|
|
|
|
|
cbmd->color_multiply = 1.0f;
|
2021-09-30 21:09:47 +02:00
|
|
|
cbmd->color_balance.method = 0;
|
2012-08-19 15:41:56 +00:00
|
|
|
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
for (int c = 0; c < 3; c++) {
|
2012-08-19 15:41:56 +00:00
|
|
|
cbmd->color_balance.lift[c] = 1.0f;
|
|
|
|
|
cbmd->color_balance.gamma[c] = 1.0f;
|
|
|
|
|
cbmd->color_balance.gain[c] = 1.0f;
|
2021-09-30 21:09:47 +02:00
|
|
|
cbmd->color_balance.slope[c] = 1.0f;
|
|
|
|
|
cbmd->color_balance.offset[c] = 1.0f;
|
|
|
|
|
cbmd->color_balance.power[c] = 1.0f;
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2024-09-11 12:42:03 +02:00
|
|
|
static void colorBalance_apply(const StripScreenQuad & /*quad*/,
|
|
|
|
|
SequenceModifierData *smd,
|
|
|
|
|
ImBuf *ibuf,
|
|
|
|
|
ImBuf *mask)
|
2020-11-05 13:33:27 +01:00
|
|
|
{
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
const ColorBalanceModifierData *cbmd = (const ColorBalanceModifierData *)smd;
|
2020-11-05 13:33:27 +01:00
|
|
|
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
const StripColorBalance cb = calc_cb(&cbmd->color_balance);
|
|
|
|
|
const float mul = cbmd->color_multiply;
|
2020-11-05 13:33:27 +01:00
|
|
|
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
/* When working on non-float image, precalculate CB LUTs. */
|
|
|
|
|
float cb_tab[3][CB_TABLE_SIZE];
|
|
|
|
|
if (ibuf->float_buffer.data == nullptr) {
|
|
|
|
|
for (int c = 0; c < 3; c++) {
|
|
|
|
|
if (cb.method == SEQ_COLOR_BALANCE_METHOD_LIFTGAMMAGAIN) {
|
|
|
|
|
make_cb_table_lgg(cb.lift[c], cb.gain[c], cb.gamma[c], mul, cb_tab[c]);
|
|
|
|
|
}
|
|
|
|
|
else {
|
|
|
|
|
make_cb_table_sop(cb.slope[c], cb.offset[c], cb.power[c], mul, cb_tab[c]);
|
|
|
|
|
}
|
|
|
|
|
}
|
2020-11-05 13:33:27 +01:00
|
|
|
}
|
|
|
|
|
|
VSE: Optimize the Color Balance modifier
Speedup the Color Balance VSE strip modifier, with two things:
- Generally, use a much lower overhead parallel_for, also with
lower grain size (32 image rows, instead of 64 that were used
before). This is what makes the "float" variant faster.
- For "byte" variant, create a precalculated lookup table instead
of doing all the math per-pixel. This was *almost* done in
existing code, except it was put into the code path that was
never-ever used. However, since this is all done on premultiplied
values, I'm using lookup table size of 1024 instead of 256, so
that semitransparent pixels get some more precision for
"in-between values". This LUT is what results in the main speedup
of "byte" variant.
Calculating Color Balance at 4K resolution, times in milliseconds:
- PC (Ryzen 5950X), PNG (byte): 22.2 -> 2.9 ms, EXR (float): 20.1 -> 15.2 ms
- Mac (M1 Max), PNG (byte): 28.9 -> 7.5 ms, EXR (float): 21.8 -> 8.5 ms
More timing details in PR.
Pull Request: https://projects.blender.org/blender/blender/pulls/127121
2024-09-05 19:21:35 +02:00
|
|
|
threading::parallel_for(IndexRange(ibuf->y), 32, [&](const IndexRange y_range) {
|
|
|
|
|
const int64_t offset = y_range.first() * ibuf->x * 4;
|
|
|
|
|
const int y_size = int(y_range.size());
|
|
|
|
|
if (ibuf->float_buffer.data != nullptr) {
|
|
|
|
|
/* Float pixels. */
|
|
|
|
|
color_balance_float(&cb,
|
|
|
|
|
ibuf->float_buffer.data + offset,
|
|
|
|
|
mask ? mask->float_buffer.data + offset : nullptr,
|
|
|
|
|
ibuf->x,
|
|
|
|
|
y_size,
|
|
|
|
|
mul);
|
|
|
|
|
}
|
|
|
|
|
else {
|
|
|
|
|
/* Byte pixels. */
|
|
|
|
|
color_balance_byte(cb_tab,
|
|
|
|
|
ibuf->byte_buffer.data + offset,
|
|
|
|
|
mask ? mask->byte_buffer.data + offset : nullptr,
|
|
|
|
|
ibuf->x,
|
|
|
|
|
y_size);
|
|
|
|
|
}
|
|
|
|
|
});
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static SequenceModifierTypeInfo seqModifier_ColorBalance = {
|
2023-01-16 12:41:11 +11:00
|
|
|
/*name*/ CTX_N_(BLT_I18NCONTEXT_ID_SEQUENCE, "Color Balance"),
|
|
|
|
|
/*struct_name*/ "ColorBalanceModifierData",
|
|
|
|
|
/*struct_size*/ sizeof(ColorBalanceModifierData),
|
|
|
|
|
/*init_data*/ colorBalance_init_data,
|
2023-07-20 09:46:24 +02:00
|
|
|
/*free_data*/ nullptr,
|
|
|
|
|
/*copy_data*/ nullptr,
|
2023-01-16 12:41:11 +11:00
|
|
|
/*apply*/ colorBalance_apply,
|
2012-08-19 15:41:56 +00:00
|
|
|
};
|
|
|
|
|
|
2020-06-16 12:32:42 +10:00
|
|
|
/** \} */
|
|
|
|
|
|
|
|
|
|
/* -------------------------------------------------------------------- */
|
|
|
|
|
/** \name White Balance Modifier
|
|
|
|
|
* \{ */
|
2015-12-28 11:55:14 +01:00
|
|
|
|
|
|
|
|
static void whiteBalance_init_data(SequenceModifierData *smd)
|
|
|
|
|
{
|
|
|
|
|
WhiteBalanceModifierData *cbmd = (WhiteBalanceModifierData *)smd;
|
|
|
|
|
copy_v3_fl(cbmd->white_value, 1.0f);
|
|
|
|
|
}
|
|
|
|
|
|
2023-07-20 09:46:24 +02:00
|
|
|
struct WhiteBalanceThreadData {
|
2015-12-28 11:55:14 +01:00
|
|
|
float white[3];
|
2023-07-20 09:46:24 +02:00
|
|
|
};
|
2015-12-28 11:55:14 +01:00
|
|
|
|
|
|
|
|
static void whiteBalance_apply_threaded(int width,
|
|
|
|
|
int height,
|
2022-09-25 17:04:52 +10:00
|
|
|
uchar *rect,
|
2015-12-28 11:55:14 +01:00
|
|
|
float *rect_float,
|
2022-09-25 17:04:52 +10:00
|
|
|
uchar *mask_rect,
|
2020-07-13 12:48:46 +02:00
|
|
|
const float *mask_rect_float,
|
2015-12-28 11:55:14 +01:00
|
|
|
void *data_v)
|
|
|
|
|
{
|
|
|
|
|
int x, y;
|
|
|
|
|
float multiplier[3];
|
|
|
|
|
|
|
|
|
|
WhiteBalanceThreadData *data = (WhiteBalanceThreadData *)data_v;
|
|
|
|
|
|
2016-02-23 14:50:30 +11:00
|
|
|
multiplier[0] = (data->white[0] != 0.0f) ? 1.0f / data->white[0] : FLT_MAX;
|
|
|
|
|
multiplier[1] = (data->white[1] != 0.0f) ? 1.0f / data->white[1] : FLT_MAX;
|
|
|
|
|
multiplier[2] = (data->white[2] != 0.0f) ? 1.0f / data->white[2] : FLT_MAX;
|
2015-12-28 11:55:14 +01:00
|
|
|
|
|
|
|
|
for (y = 0; y < height; y++) {
|
|
|
|
|
for (x = 0; x < width; x++) {
|
|
|
|
|
int pixel_index = (y * width + x) * 4;
|
2017-05-19 11:29:18 +02:00
|
|
|
float rgba[4], result[4], mask[3] = {1.0f, 1.0f, 1.0f};
|
2015-12-28 11:55:14 +01:00
|
|
|
|
2015-12-29 01:08:13 +11:00
|
|
|
if (rect_float) {
|
2017-05-19 11:29:18 +02:00
|
|
|
copy_v3_v3(rgba, rect_float + pixel_index);
|
2015-12-29 01:08:13 +11:00
|
|
|
}
|
|
|
|
|
else {
|
2017-05-19 11:29:18 +02:00
|
|
|
straight_uchar_to_premul_float(rgba, rect + pixel_index);
|
2015-12-28 11:55:14 +01:00
|
|
|
}
|
|
|
|
|
|
2017-05-19 11:29:18 +02:00
|
|
|
copy_v4_v4(result, rgba);
|
2018-10-19 09:07:40 +11:00
|
|
|
#if 0
|
|
|
|
|
mul_v3_v3(result, multiplier);
|
|
|
|
|
#else
|
2016-02-23 14:50:30 +11:00
|
|
|
/* similar to division without the clipping */
|
|
|
|
|
for (int i = 0; i < 3; i++) {
|
2017-05-19 11:29:18 +02:00
|
|
|
result[i] = 1.0f - powf(1.0f - rgba[i], multiplier[i]);
|
2016-02-23 14:50:30 +11:00
|
|
|
}
|
2018-10-19 09:07:40 +11:00
|
|
|
#endif
|
2015-12-28 11:55:14 +01:00
|
|
|
|
2016-01-11 14:59:18 +11:00
|
|
|
if (mask_rect_float) {
|
2015-12-28 11:55:14 +01:00
|
|
|
copy_v3_v3(mask, mask_rect_float + pixel_index);
|
2016-01-11 14:59:18 +11:00
|
|
|
}
|
|
|
|
|
else if (mask_rect) {
|
2015-12-28 11:55:14 +01:00
|
|
|
rgb_uchar_to_float(mask, mask_rect + pixel_index);
|
2016-01-11 14:59:18 +11:00
|
|
|
}
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2017-05-19 11:29:18 +02:00
|
|
|
result[0] = rgba[0] * (1.0f - mask[0]) + result[0] * mask[0];
|
|
|
|
|
result[1] = rgba[1] * (1.0f - mask[1]) + result[1] * mask[1];
|
|
|
|
|
result[2] = rgba[2] * (1.0f - mask[2]) + result[2] * mask[2];
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2016-01-11 14:59:18 +11:00
|
|
|
if (rect_float) {
|
2015-12-28 11:55:14 +01:00
|
|
|
copy_v3_v3(rect_float + pixel_index, result);
|
2016-01-11 14:59:18 +11:00
|
|
|
}
|
|
|
|
|
else {
|
|
|
|
|
premul_float_to_straight_uchar(rect + pixel_index, result);
|
|
|
|
|
}
|
2015-12-29 01:08:13 +11:00
|
|
|
}
|
2015-12-28 11:55:14 +01:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2024-09-11 12:42:03 +02:00
|
|
|
static void whiteBalance_apply(const StripScreenQuad & /*quad*/,
|
|
|
|
|
SequenceModifierData *smd,
|
|
|
|
|
ImBuf *ibuf,
|
|
|
|
|
ImBuf *mask)
|
2015-12-28 11:55:14 +01:00
|
|
|
{
|
|
|
|
|
WhiteBalanceThreadData data;
|
|
|
|
|
WhiteBalanceModifierData *wbmd = (WhiteBalanceModifierData *)smd;
|
|
|
|
|
|
|
|
|
|
copy_v3_v3(data.white, wbmd->white_value);
|
|
|
|
|
|
|
|
|
|
modifier_apply_threaded(ibuf, mask, whiteBalance_apply_threaded, &data);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static SequenceModifierTypeInfo seqModifier_WhiteBalance = {
|
2023-01-16 12:41:11 +11:00
|
|
|
/*name*/ CTX_N_(BLT_I18NCONTEXT_ID_SEQUENCE, "White Balance"),
|
|
|
|
|
/*struct_name*/ "WhiteBalanceModifierData",
|
|
|
|
|
/*struct_size*/ sizeof(WhiteBalanceModifierData),
|
|
|
|
|
/*init_data*/ whiteBalance_init_data,
|
2023-07-20 09:46:24 +02:00
|
|
|
/*free_data*/ nullptr,
|
|
|
|
|
/*copy_data*/ nullptr,
|
2023-01-16 12:41:11 +11:00
|
|
|
/*apply*/ whiteBalance_apply,
|
2015-12-28 11:55:14 +01:00
|
|
|
};
|
|
|
|
|
|
2020-06-16 12:32:42 +10:00
|
|
|
/** \} */
|
|
|
|
|
|
|
|
|
|
/* -------------------------------------------------------------------- */
|
|
|
|
|
/** \name Curves Modifier
|
|
|
|
|
* \{ */
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2012-09-15 01:52:28 +00:00
|
|
|
static void curves_init_data(SequenceModifierData *smd)
|
2012-08-19 15:41:56 +00:00
|
|
|
{
|
|
|
|
|
CurvesModifierData *cmd = (CurvesModifierData *)smd;
|
|
|
|
|
|
2023-08-30 22:36:36 +02:00
|
|
|
BKE_curvemapping_set_defaults(&cmd->curve_mapping, 4, 0.0f, 0.0f, 1.0f, 1.0f, HD_AUTO);
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
|
2012-09-15 01:52:28 +00:00
|
|
|
static void curves_free_data(SequenceModifierData *smd)
|
2012-08-19 15:41:56 +00:00
|
|
|
{
|
|
|
|
|
CurvesModifierData *cmd = (CurvesModifierData *)smd;
|
|
|
|
|
|
2019-08-07 03:21:55 +10:00
|
|
|
BKE_curvemapping_free_data(&cmd->curve_mapping);
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
|
2012-09-15 01:52:28 +00:00
|
|
|
static void curves_copy_data(SequenceModifierData *target, SequenceModifierData *smd)
|
2012-08-19 15:41:56 +00:00
|
|
|
{
|
|
|
|
|
CurvesModifierData *cmd = (CurvesModifierData *)smd;
|
|
|
|
|
CurvesModifierData *cmd_target = (CurvesModifierData *)target;
|
|
|
|
|
|
2019-08-07 03:21:55 +10:00
|
|
|
BKE_curvemapping_copy_data(&cmd_target->curve_mapping, &cmd->curve_mapping);
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
|
2012-09-15 01:52:28 +00:00
|
|
|
static void curves_apply_threaded(int width,
|
|
|
|
|
int height,
|
2022-09-25 17:04:52 +10:00
|
|
|
uchar *rect,
|
2012-09-15 01:52:28 +00:00
|
|
|
float *rect_float,
|
2022-09-25 17:04:52 +10:00
|
|
|
uchar *mask_rect,
|
2020-07-13 11:27:09 +02:00
|
|
|
const float *mask_rect_float,
|
2012-09-15 01:52:28 +00:00
|
|
|
void *data_v)
|
2012-08-19 15:41:56 +00:00
|
|
|
{
|
|
|
|
|
CurveMapping *curve_mapping = (CurveMapping *)data_v;
|
|
|
|
|
int x, y;
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-19 15:41:56 +00:00
|
|
|
for (y = 0; y < height; y++) {
|
|
|
|
|
for (x = 0; x < width; x++) {
|
|
|
|
|
int pixel_index = (y * width + x) * 4;
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-19 15:41:56 +00:00
|
|
|
if (rect_float) {
|
|
|
|
|
float *pixel = rect_float + pixel_index;
|
|
|
|
|
float result[3];
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2019-08-07 03:21:55 +10:00
|
|
|
BKE_curvemapping_evaluate_premulRGBF(curve_mapping, result, pixel);
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-19 15:41:56 +00:00
|
|
|
if (mask_rect_float) {
|
2014-04-27 00:20:13 +10:00
|
|
|
const float *m = mask_rect_float + pixel_index;
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-19 15:41:56 +00:00
|
|
|
pixel[0] = pixel[0] * (1.0f - m[0]) + result[0] * m[0];
|
|
|
|
|
pixel[1] = pixel[1] * (1.0f - m[1]) + result[1] * m[1];
|
|
|
|
|
pixel[2] = pixel[2] * (1.0f - m[2]) + result[2] * m[2];
|
|
|
|
|
}
|
|
|
|
|
else {
|
|
|
|
|
pixel[0] = result[0];
|
|
|
|
|
pixel[1] = result[1];
|
|
|
|
|
pixel[2] = result[2];
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
if (rect) {
|
2022-09-25 17:04:52 +10:00
|
|
|
uchar *pixel = rect + pixel_index;
|
2012-12-31 13:52:13 +00:00
|
|
|
float result[3], tempc[4];
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-12-31 13:52:13 +00:00
|
|
|
straight_uchar_to_premul_float(tempc, pixel);
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2019-08-07 03:21:55 +10:00
|
|
|
BKE_curvemapping_evaluate_premulRGBF(curve_mapping, result, tempc);
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-19 15:41:56 +00:00
|
|
|
if (mask_rect) {
|
|
|
|
|
float t[3];
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-19 15:41:56 +00:00
|
|
|
rgb_uchar_to_float(t, mask_rect + pixel_index);
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2013-02-27 09:58:40 +00:00
|
|
|
tempc[0] = tempc[0] * (1.0f - t[0]) + result[0] * t[0];
|
|
|
|
|
tempc[1] = tempc[1] * (1.0f - t[1]) + result[1] * t[1];
|
|
|
|
|
tempc[2] = tempc[2] * (1.0f - t[2]) + result[2] * t[2];
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
else {
|
2012-12-31 13:52:13 +00:00
|
|
|
tempc[0] = result[0];
|
|
|
|
|
tempc[1] = result[1];
|
|
|
|
|
tempc[2] = result[2];
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-12-31 13:52:13 +00:00
|
|
|
premul_float_to_straight_uchar(pixel, tempc);
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2024-09-11 12:42:03 +02:00
|
|
|
static void curves_apply(const StripScreenQuad & /*quad*/,
|
|
|
|
|
SequenceModifierData *smd,
|
|
|
|
|
ImBuf *ibuf,
|
|
|
|
|
ImBuf *mask)
|
2012-08-19 15:41:56 +00:00
|
|
|
{
|
|
|
|
|
CurvesModifierData *cmd = (CurvesModifierData *)smd;
|
|
|
|
|
|
2020-08-07 22:36:11 +10:00
|
|
|
const float black[3] = {0.0f, 0.0f, 0.0f};
|
|
|
|
|
const float white[3] = {1.0f, 1.0f, 1.0f};
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2020-08-01 13:02:21 +10:00
|
|
|
BKE_curvemapping_init(&cmd->curve_mapping);
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2023-07-22 11:36:59 +10:00
|
|
|
BKE_curvemapping_premultiply(&cmd->curve_mapping, false);
|
2019-08-07 03:21:55 +10:00
|
|
|
BKE_curvemapping_set_black_white(&cmd->curve_mapping, black, white);
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2012-09-04 16:55:12 +00:00
|
|
|
modifier_apply_threaded(ibuf, mask, curves_apply_threaded, &cmd->curve_mapping);
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2023-07-22 11:36:59 +10:00
|
|
|
BKE_curvemapping_premultiply(&cmd->curve_mapping, true);
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static SequenceModifierTypeInfo seqModifier_Curves = {
|
2023-01-16 12:41:11 +11:00
|
|
|
/*name*/ CTX_N_(BLT_I18NCONTEXT_ID_SEQUENCE, "Curves"),
|
|
|
|
|
/*struct_name*/ "CurvesModifierData",
|
|
|
|
|
/*struct_size*/ sizeof(CurvesModifierData),
|
|
|
|
|
/*init_data*/ curves_init_data,
|
|
|
|
|
/*free_data*/ curves_free_data,
|
|
|
|
|
/*copy_data*/ curves_copy_data,
|
|
|
|
|
/*apply*/ curves_apply,
|
2012-08-19 15:41:56 +00:00
|
|
|
};
|
|
|
|
|
|
2020-06-16 12:32:42 +10:00
|
|
|
/** \} */
|
|
|
|
|
|
|
|
|
|
/* -------------------------------------------------------------------- */
|
|
|
|
|
/** \name Hue Correct Modifier
|
|
|
|
|
* \{ */
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2012-09-15 01:52:28 +00:00
|
|
|
static void hue_correct_init_data(SequenceModifierData *smd)
|
2012-08-19 15:41:56 +00:00
|
|
|
{
|
|
|
|
|
HueCorrectModifierData *hcmd = (HueCorrectModifierData *)smd;
|
|
|
|
|
int c;
|
|
|
|
|
|
2023-08-30 22:36:36 +02:00
|
|
|
BKE_curvemapping_set_defaults(&hcmd->curve_mapping, 1, 0.0f, 0.0f, 1.0f, 1.0f, HD_AUTO);
|
Compositor: Make Hue Correct node wrap
Compositor: Make Hue Correct node wrap
This patch makes the Hue Correct node as well as the Hue Correct VSE
modifiers to wrap, such that no discontinuities occur for the red hue.
Since it now wraps, the default curve preset now exempts the last point
of the curve.
A new CUMA_USE_WRAPPING flag was added to specify wrapping for curve
maps. The implementation works by adding two virtual points before and
after the terminal points in the curve map, such that their handles
match, and would then produce a continues curve.
This is a breaking change, since existing curves were also adjusted
using versioning. However, the change will not be significant, since in
most realistic cases, the terminal points will be close to each other,
and even with wrapping, the connection will be very sharp, almost
matching the old behavior.
Pull Request: https://projects.blender.org/blender/blender/pulls/117114
2024-03-21 15:35:02 +01:00
|
|
|
hcmd->curve_mapping.preset = CURVE_PRESET_MID8;
|
2012-08-19 15:41:56 +00:00
|
|
|
|
|
|
|
|
for (c = 0; c < 3; c++) {
|
|
|
|
|
CurveMap *cuma = &hcmd->curve_mapping.cm[c];
|
2019-08-07 03:21:55 +10:00
|
|
|
BKE_curvemap_reset(
|
2012-08-19 15:41:56 +00:00
|
|
|
cuma, &hcmd->curve_mapping.clipr, hcmd->curve_mapping.preset, CURVEMAP_SLOPE_POSITIVE);
|
|
|
|
|
}
|
Compositor: Make Hue Correct node wrap
Compositor: Make Hue Correct node wrap
This patch makes the Hue Correct node as well as the Hue Correct VSE
modifiers to wrap, such that no discontinuities occur for the red hue.
Since it now wraps, the default curve preset now exempts the last point
of the curve.
A new CUMA_USE_WRAPPING flag was added to specify wrapping for curve
maps. The implementation works by adding two virtual points before and
after the terminal points in the curve map, such that their handles
match, and would then produce a continues curve.
This is a breaking change, since existing curves were also adjusted
using versioning. However, the change will not be significant, since in
most realistic cases, the terminal points will be close to each other,
and even with wrapping, the connection will be very sharp, almost
matching the old behavior.
Pull Request: https://projects.blender.org/blender/blender/pulls/117114
2024-03-21 15:35:02 +01:00
|
|
|
/* use wrapping for all hue correct modifiers */
|
|
|
|
|
hcmd->curve_mapping.flag |= CUMA_USE_WRAPPING;
|
2012-08-19 15:41:56 +00:00
|
|
|
/* default to showing Saturation */
|
|
|
|
|
hcmd->curve_mapping.cur = 1;
|
|
|
|
|
}
|
|
|
|
|
|
2012-09-15 01:52:28 +00:00
|
|
|
static void hue_correct_free_data(SequenceModifierData *smd)
|
2012-08-19 15:41:56 +00:00
|
|
|
{
|
|
|
|
|
HueCorrectModifierData *hcmd = (HueCorrectModifierData *)smd;
|
|
|
|
|
|
2019-08-07 03:21:55 +10:00
|
|
|
BKE_curvemapping_free_data(&hcmd->curve_mapping);
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
|
2012-09-15 01:52:28 +00:00
|
|
|
static void hue_correct_copy_data(SequenceModifierData *target, SequenceModifierData *smd)
|
2012-08-19 15:41:56 +00:00
|
|
|
{
|
|
|
|
|
HueCorrectModifierData *hcmd = (HueCorrectModifierData *)smd;
|
|
|
|
|
HueCorrectModifierData *hcmd_target = (HueCorrectModifierData *)target;
|
|
|
|
|
|
2019-08-07 03:21:55 +10:00
|
|
|
BKE_curvemapping_copy_data(&hcmd_target->curve_mapping, &hcmd->curve_mapping);
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
|
2012-09-15 01:52:28 +00:00
|
|
|
static void hue_correct_apply_threaded(int width,
|
|
|
|
|
int height,
|
2022-09-25 17:04:52 +10:00
|
|
|
uchar *rect,
|
2012-09-15 01:52:28 +00:00
|
|
|
float *rect_float,
|
2022-09-25 17:04:52 +10:00
|
|
|
uchar *mask_rect,
|
2020-07-13 12:48:46 +02:00
|
|
|
const float *mask_rect_float,
|
2012-08-19 15:41:56 +00:00
|
|
|
void *data_v)
|
|
|
|
|
{
|
|
|
|
|
CurveMapping *curve_mapping = (CurveMapping *)data_v;
|
|
|
|
|
int x, y;
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-19 15:41:56 +00:00
|
|
|
for (y = 0; y < height; y++) {
|
|
|
|
|
for (x = 0; x < width; x++) {
|
|
|
|
|
int pixel_index = (y * width + x) * 4;
|
|
|
|
|
float pixel[3], result[3], mask[3] = {1.0f, 1.0f, 1.0f};
|
|
|
|
|
float hsv[3], f;
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2019-04-22 09:39:35 +10:00
|
|
|
if (rect_float) {
|
2012-08-19 15:41:56 +00:00
|
|
|
copy_v3_v3(pixel, rect_float + pixel_index);
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
|
|
|
|
else {
|
2012-08-19 15:41:56 +00:00
|
|
|
rgb_uchar_to_float(pixel, rect + pixel_index);
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-19 15:41:56 +00:00
|
|
|
rgb_to_hsv(pixel[0], pixel[1], pixel[2], hsv, hsv + 1, hsv + 2);
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-19 15:41:56 +00:00
|
|
|
/* adjust hue, scaling returned default 0.5 up to 1 */
|
2019-08-07 03:21:55 +10:00
|
|
|
f = BKE_curvemapping_evaluateF(curve_mapping, 0, hsv[0]);
|
2012-08-19 15:41:56 +00:00
|
|
|
hsv[0] += f - 0.5f;
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-19 15:41:56 +00:00
|
|
|
/* adjust saturation, scaling returned default 0.5 up to 1 */
|
2019-08-07 03:21:55 +10:00
|
|
|
f = BKE_curvemapping_evaluateF(curve_mapping, 1, hsv[0]);
|
2012-08-19 15:41:56 +00:00
|
|
|
hsv[1] *= (f * 2.0f);
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-19 15:41:56 +00:00
|
|
|
/* adjust value, scaling returned default 0.5 up to 1 */
|
2019-08-07 03:21:55 +10:00
|
|
|
f = BKE_curvemapping_evaluateF(curve_mapping, 2, hsv[0]);
|
2020-11-06 11:25:27 +11:00
|
|
|
hsv[2] *= (f * 2.0f);
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-19 15:41:56 +00:00
|
|
|
hsv[0] = hsv[0] - floorf(hsv[0]); /* mod 1.0 */
|
|
|
|
|
CLAMP(hsv[1], 0.0f, 1.0f);
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-19 15:41:56 +00:00
|
|
|
/* convert back to rgb */
|
|
|
|
|
hsv_to_rgb(hsv[0], hsv[1], hsv[2], result, result + 1, result + 2);
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2019-04-22 09:39:35 +10:00
|
|
|
if (mask_rect_float) {
|
2012-11-09 09:33:28 +00:00
|
|
|
copy_v3_v3(mask, mask_rect_float + pixel_index);
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
|
|
|
|
else if (mask_rect) {
|
2012-11-09 09:33:28 +00:00
|
|
|
rgb_uchar_to_float(mask, mask_rect + pixel_index);
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-19 15:41:56 +00:00
|
|
|
result[0] = pixel[0] * (1.0f - mask[0]) + result[0] * mask[0];
|
|
|
|
|
result[1] = pixel[1] * (1.0f - mask[1]) + result[1] * mask[1];
|
|
|
|
|
result[2] = pixel[2] * (1.0f - mask[2]) + result[2] * mask[2];
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2019-04-22 09:39:35 +10:00
|
|
|
if (rect_float) {
|
2012-08-19 15:41:56 +00:00
|
|
|
copy_v3_v3(rect_float + pixel_index, result);
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
|
|
|
|
else {
|
2012-08-19 15:41:56 +00:00
|
|
|
rgb_float_to_uchar(rect + pixel_index, result);
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2024-09-11 12:42:03 +02:00
|
|
|
static void hue_correct_apply(const StripScreenQuad & /*quad*/,
|
|
|
|
|
SequenceModifierData *smd,
|
|
|
|
|
ImBuf *ibuf,
|
|
|
|
|
ImBuf *mask)
|
2012-08-19 15:41:56 +00:00
|
|
|
{
|
|
|
|
|
HueCorrectModifierData *hcmd = (HueCorrectModifierData *)smd;
|
|
|
|
|
|
2020-08-01 13:02:21 +10:00
|
|
|
BKE_curvemapping_init(&hcmd->curve_mapping);
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2012-09-04 16:55:12 +00:00
|
|
|
modifier_apply_threaded(ibuf, mask, hue_correct_apply_threaded, &hcmd->curve_mapping);
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static SequenceModifierTypeInfo seqModifier_HueCorrect = {
|
2023-01-16 12:41:11 +11:00
|
|
|
/*name*/ CTX_N_(BLT_I18NCONTEXT_ID_SEQUENCE, "Hue Correct"),
|
|
|
|
|
/*struct_name*/ "HueCorrectModifierData",
|
|
|
|
|
/*struct_size*/ sizeof(HueCorrectModifierData),
|
|
|
|
|
/*init_data*/ hue_correct_init_data,
|
|
|
|
|
/*free_data*/ hue_correct_free_data,
|
|
|
|
|
/*copy_data*/ hue_correct_copy_data,
|
|
|
|
|
/*apply*/ hue_correct_apply,
|
2012-08-19 15:41:56 +00:00
|
|
|
};
|
|
|
|
|
|
2020-06-16 12:32:42 +10:00
|
|
|
/** \} */
|
|
|
|
|
|
|
|
|
|
/* -------------------------------------------------------------------- */
|
2023-04-11 21:32:29 +02:00
|
|
|
/** \name Brightness/Contrast Modifier
|
2020-06-16 12:32:42 +10:00
|
|
|
* \{ */
|
2012-08-24 09:07:04 +00:00
|
|
|
|
2023-07-20 09:46:24 +02:00
|
|
|
struct BrightContrastThreadData {
|
2012-08-24 09:07:04 +00:00
|
|
|
float bright;
|
|
|
|
|
float contrast;
|
2023-07-20 09:46:24 +02:00
|
|
|
};
|
2012-08-24 09:07:04 +00:00
|
|
|
|
2012-09-15 01:52:28 +00:00
|
|
|
static void brightcontrast_apply_threaded(int width,
|
|
|
|
|
int height,
|
2022-09-26 10:56:05 +10:00
|
|
|
uchar *rect,
|
2012-09-15 01:52:28 +00:00
|
|
|
float *rect_float,
|
2022-09-26 10:56:05 +10:00
|
|
|
uchar *mask_rect,
|
2020-07-13 11:27:09 +02:00
|
|
|
const float *mask_rect_float,
|
2012-09-15 01:52:28 +00:00
|
|
|
void *data_v)
|
2012-08-24 09:07:04 +00:00
|
|
|
{
|
|
|
|
|
BrightContrastThreadData *data = (BrightContrastThreadData *)data_v;
|
|
|
|
|
int x, y;
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-24 09:07:04 +00:00
|
|
|
float i;
|
|
|
|
|
int c;
|
|
|
|
|
float a, b, v;
|
2024-09-16 11:39:02 +10:00
|
|
|
const float brightness = data->bright / 100.0f;
|
|
|
|
|
const float contrast = data->contrast;
|
2012-08-24 09:07:04 +00:00
|
|
|
float delta = contrast / 200.0f;
|
|
|
|
|
/*
|
|
|
|
|
* The algorithm is by Werner D. Streidt
|
|
|
|
|
* (http://visca.com/ffactory/archives/5-99/msg00021.html)
|
2022-09-21 11:57:20 +10:00
|
|
|
* Extracted of OpenCV `demhist.c`.
|
2012-08-24 09:07:04 +00:00
|
|
|
*/
|
|
|
|
|
if (contrast > 0) {
|
2019-08-01 18:17:56 +02:00
|
|
|
a = 1.0f - delta * 2.0f;
|
|
|
|
|
a = 1.0f / max_ff(a, FLT_EPSILON);
|
2012-08-24 09:07:04 +00:00
|
|
|
b = a * (brightness - delta);
|
|
|
|
|
}
|
|
|
|
|
else {
|
|
|
|
|
delta *= -1;
|
2019-08-01 18:17:56 +02:00
|
|
|
a = max_ff(1.0f - delta * 2.0f, 0.0f);
|
|
|
|
|
b = a * brightness + delta;
|
2012-08-24 09:07:04 +00:00
|
|
|
}
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-24 09:07:04 +00:00
|
|
|
for (y = 0; y < height; y++) {
|
|
|
|
|
for (x = 0; x < width; x++) {
|
|
|
|
|
int pixel_index = (y * width + x) * 4;
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-24 09:07:04 +00:00
|
|
|
if (rect) {
|
2022-09-26 10:56:05 +10:00
|
|
|
uchar *pixel = rect + pixel_index;
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-24 09:07:04 +00:00
|
|
|
for (c = 0; c < 3; c++) {
|
2023-07-20 20:11:08 +10:00
|
|
|
i = float(pixel[c]) / 255.0f;
|
2012-08-24 09:07:04 +00:00
|
|
|
v = a * i + b;
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-24 09:07:04 +00:00
|
|
|
if (mask_rect) {
|
2024-09-16 11:39:02 +10:00
|
|
|
const uchar *m = mask_rect + pixel_index;
|
|
|
|
|
const float t = float(m[c]) / 255.0f;
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2023-07-20 20:11:08 +10:00
|
|
|
v = float(pixel[c]) / 255.0f * (1.0f - t) + v * t;
|
2012-08-24 09:07:04 +00:00
|
|
|
}
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2018-05-07 17:31:28 +02:00
|
|
|
pixel[c] = unit_float_to_uchar_clamp(v);
|
2012-08-24 09:07:04 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
else if (rect_float) {
|
|
|
|
|
float *pixel = rect_float + pixel_index;
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-24 09:07:04 +00:00
|
|
|
for (c = 0; c < 3; c++) {
|
|
|
|
|
i = pixel[c];
|
|
|
|
|
v = a * i + b;
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-24 09:07:04 +00:00
|
|
|
if (mask_rect_float) {
|
2014-04-27 00:20:13 +10:00
|
|
|
const float *m = mask_rect_float + pixel_index;
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-24 09:07:04 +00:00
|
|
|
pixel[c] = pixel[c] * (1.0f - m[c]) + v * m[c];
|
|
|
|
|
}
|
2019-04-22 09:39:35 +10:00
|
|
|
else {
|
2012-08-24 09:07:04 +00:00
|
|
|
pixel[c] = v;
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
2012-08-24 09:07:04 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2024-09-11 12:42:03 +02:00
|
|
|
static void brightcontrast_apply(const StripScreenQuad & /*quad*/,
|
|
|
|
|
SequenceModifierData *smd,
|
|
|
|
|
ImBuf *ibuf,
|
|
|
|
|
ImBuf *mask)
|
2012-08-24 09:07:04 +00:00
|
|
|
{
|
2024-09-15 23:14:09 +10:00
|
|
|
const BrightContrastModifierData *bcmd = (BrightContrastModifierData *)smd;
|
2012-08-24 09:07:04 +00:00
|
|
|
BrightContrastThreadData data;
|
|
|
|
|
|
|
|
|
|
data.bright = bcmd->bright;
|
|
|
|
|
data.contrast = bcmd->contrast;
|
|
|
|
|
|
2012-09-04 16:55:12 +00:00
|
|
|
modifier_apply_threaded(ibuf, mask, brightcontrast_apply_threaded, &data);
|
2012-08-24 09:07:04 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static SequenceModifierTypeInfo seqModifier_BrightContrast = {
|
2023-04-11 21:32:29 +02:00
|
|
|
/*name*/ CTX_N_(BLT_I18NCONTEXT_ID_SEQUENCE, "Brightness/Contrast"),
|
2023-01-16 12:41:11 +11:00
|
|
|
/*struct_name*/ "BrightContrastModifierData",
|
|
|
|
|
/*struct_size*/ sizeof(BrightContrastModifierData),
|
2023-07-20 09:46:24 +02:00
|
|
|
/*init_data*/ nullptr,
|
|
|
|
|
/*free_data*/ nullptr,
|
|
|
|
|
/*copy_data*/ nullptr,
|
2023-01-16 12:41:11 +11:00
|
|
|
/*apply*/ brightcontrast_apply,
|
2012-08-24 09:07:04 +00:00
|
|
|
};
|
|
|
|
|
|
2020-06-16 12:32:42 +10:00
|
|
|
/** \} */
|
|
|
|
|
|
|
|
|
|
/* -------------------------------------------------------------------- */
|
|
|
|
|
/** \name Mask Modifier
|
|
|
|
|
* \{ */
|
2013-05-08 14:20:57 +00:00
|
|
|
|
|
|
|
|
static void maskmodifier_apply_threaded(int width,
|
|
|
|
|
int height,
|
2022-09-26 10:56:05 +10:00
|
|
|
uchar *rect,
|
2013-05-08 14:20:57 +00:00
|
|
|
float *rect_float,
|
2022-09-26 10:56:05 +10:00
|
|
|
uchar *mask_rect,
|
2020-07-13 11:27:09 +02:00
|
|
|
const float *mask_rect_float,
|
2023-07-20 09:46:24 +02:00
|
|
|
void * /*data_v*/)
|
2013-05-08 14:20:57 +00:00
|
|
|
{
|
|
|
|
|
int x, y;
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2019-04-22 09:39:35 +10:00
|
|
|
if (rect && !mask_rect) {
|
2013-05-08 14:20:57 +00:00
|
|
|
return;
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2019-04-22 09:39:35 +10:00
|
|
|
if (rect_float && !mask_rect_float) {
|
2013-05-08 14:20:57 +00:00
|
|
|
return;
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2013-05-08 14:20:57 +00:00
|
|
|
for (y = 0; y < height; y++) {
|
|
|
|
|
for (x = 0; x < width; x++) {
|
2024-09-16 11:39:02 +10:00
|
|
|
const int pixel_index = (y * width + x) * 4;
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2013-05-08 14:20:57 +00:00
|
|
|
if (rect) {
|
2024-09-16 11:39:02 +10:00
|
|
|
const uchar *mask_pixel = mask_rect + pixel_index;
|
|
|
|
|
const uchar mask = min_iii(mask_pixel[0], mask_pixel[1], mask_pixel[2]);
|
2022-09-26 10:56:05 +10:00
|
|
|
uchar *pixel = rect + pixel_index;
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2013-05-08 14:20:57 +00:00
|
|
|
/* byte buffer is straight, so only affect on alpha itself,
|
|
|
|
|
* this is the only way to alpha-over byte strip after
|
|
|
|
|
* applying mask modifier.
|
|
|
|
|
*/
|
2023-07-20 20:11:08 +10:00
|
|
|
pixel[3] = float(pixel[3] * mask) / 255.0f;
|
2013-05-08 14:20:57 +00:00
|
|
|
}
|
|
|
|
|
else if (rect_float) {
|
2014-04-27 00:20:13 +10:00
|
|
|
const float *mask_pixel = mask_rect_float + pixel_index;
|
2024-09-16 11:39:02 +10:00
|
|
|
const float mask = min_fff(mask_pixel[0], mask_pixel[1], mask_pixel[2]);
|
|
|
|
|
float *pixel = rect_float + pixel_index;
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2013-05-08 14:20:57 +00:00
|
|
|
/* float buffers are premultiplied, so need to premul color
|
|
|
|
|
* as well to make it easy to alpha-over masted strip.
|
|
|
|
|
*/
|
2024-09-16 11:39:02 +10:00
|
|
|
for (int c = 0; c < 4; c++) {
|
2013-05-08 14:20:57 +00:00
|
|
|
pixel[c] = pixel[c] * mask;
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
2013-05-08 14:20:57 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2024-09-11 12:42:03 +02:00
|
|
|
static void maskmodifier_apply(const StripScreenQuad & /*quad*/,
|
|
|
|
|
SequenceModifierData * /*smd*/,
|
|
|
|
|
ImBuf *ibuf,
|
|
|
|
|
ImBuf *mask)
|
2013-05-08 14:20:57 +00:00
|
|
|
{
|
2015-03-19 01:09:41 +11:00
|
|
|
// SequencerMaskModifierData *bcmd = (SequencerMaskModifierData *)smd;
|
2013-05-08 14:20:57 +00:00
|
|
|
|
2023-07-20 09:46:24 +02:00
|
|
|
modifier_apply_threaded(ibuf, mask, maskmodifier_apply_threaded, nullptr);
|
2021-12-06 15:20:15 +01:00
|
|
|
ibuf->planes = R_IMF_PLANES_RGBA;
|
2013-05-08 14:20:57 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static SequenceModifierTypeInfo seqModifier_Mask = {
|
2023-01-16 12:41:11 +11:00
|
|
|
/*name*/ CTX_N_(BLT_I18NCONTEXT_ID_SEQUENCE, "Mask"),
|
|
|
|
|
/*struct_name*/ "SequencerMaskModifierData",
|
|
|
|
|
/*struct_size*/ sizeof(SequencerMaskModifierData),
|
2023-07-20 09:46:24 +02:00
|
|
|
/*init_data*/ nullptr,
|
|
|
|
|
/*free_data*/ nullptr,
|
|
|
|
|
/*copy_data*/ nullptr,
|
2023-01-16 12:41:11 +11:00
|
|
|
/*apply*/ maskmodifier_apply,
|
2013-05-08 14:20:57 +00:00
|
|
|
};
|
|
|
|
|
|
2020-06-16 12:32:42 +10:00
|
|
|
/** \} */
|
|
|
|
|
|
|
|
|
|
/* -------------------------------------------------------------------- */
|
|
|
|
|
/** \name Tonemap Modifier
|
|
|
|
|
* \{ */
|
2016-01-19 15:53:43 +01:00
|
|
|
|
2023-07-20 09:46:24 +02:00
|
|
|
struct AvgLogLum {
|
2024-09-19 18:14:49 +02:00
|
|
|
const SequencerTonemapModifierData *tmmd;
|
2016-01-19 15:53:43 +01:00
|
|
|
float al;
|
|
|
|
|
float auto_key;
|
|
|
|
|
float lav;
|
2024-09-19 18:14:49 +02:00
|
|
|
float3 cav;
|
2016-01-19 15:53:43 +01:00
|
|
|
float igm;
|
2023-07-20 09:46:24 +02:00
|
|
|
};
|
2016-01-19 15:53:43 +01:00
|
|
|
|
|
|
|
|
static void tonemapmodifier_init_data(SequenceModifierData *smd)
|
|
|
|
|
{
|
|
|
|
|
SequencerTonemapModifierData *tmmd = (SequencerTonemapModifierData *)smd;
|
2022-09-16 18:13:19 +10:00
|
|
|
/* Same as tone-map compositor node. */
|
2016-01-19 15:53:43 +01:00
|
|
|
tmmd->type = SEQ_TONEMAP_RD_PHOTORECEPTOR;
|
|
|
|
|
tmmd->key = 0.18f;
|
|
|
|
|
tmmd->offset = 1.0f;
|
|
|
|
|
tmmd->gamma = 1.0f;
|
|
|
|
|
tmmd->intensity = 0.0f;
|
|
|
|
|
tmmd->contrast = 0.0f;
|
|
|
|
|
tmmd->adaptation = 1.0f;
|
|
|
|
|
tmmd->correction = 0.0f;
|
|
|
|
|
}
|
|
|
|
|
|
2024-09-19 18:14:49 +02:00
|
|
|
/* Convert chunk of float image pixels to scene linear space, in-place. */
|
|
|
|
|
static void pixels_to_scene_linear_float(ColorSpace *colorspace, float4 *pixels, int64_t count)
|
2016-01-19 15:53:43 +01:00
|
|
|
{
|
2024-09-19 18:14:49 +02:00
|
|
|
IMB_colormanagement_colorspace_to_scene_linear(
|
|
|
|
|
(float *)(pixels), int(count), 1, 4, colorspace, false);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Convert chunk of byte image pixels to scene linear space, into a destination array. */
|
|
|
|
|
static void pixels_to_scene_linear_byte(ColorSpace *colorspace,
|
|
|
|
|
const uchar *pixels,
|
|
|
|
|
float4 *dst,
|
|
|
|
|
int64_t count)
|
|
|
|
|
{
|
|
|
|
|
const uchar *bptr = pixels;
|
|
|
|
|
float4 *dst_ptr = dst;
|
|
|
|
|
for (int64_t i = 0; i < count; i++) {
|
|
|
|
|
straight_uchar_to_premul_float(*dst_ptr, bptr);
|
|
|
|
|
bptr += 4;
|
|
|
|
|
dst_ptr++;
|
|
|
|
|
}
|
|
|
|
|
IMB_colormanagement_colorspace_to_scene_linear(
|
|
|
|
|
(float *)dst, int(count), 1, 4, colorspace, false);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void scene_linear_to_image_chunk_float(ImBuf *ibuf, IndexRange range)
|
|
|
|
|
{
|
|
|
|
|
ColorSpace *colorspace = ibuf->float_buffer.colorspace;
|
|
|
|
|
float4 *fptr = reinterpret_cast<float4 *>(ibuf->float_buffer.data);
|
|
|
|
|
IMB_colormanagement_scene_linear_to_colorspace(
|
|
|
|
|
(float *)(fptr + range.first()), int(range.size()), 1, 4, colorspace);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void scene_linear_to_image_chunk_byte(float4 *src, ImBuf *ibuf, IndexRange range)
|
|
|
|
|
{
|
|
|
|
|
ColorSpace *colorspace = ibuf->byte_buffer.colorspace;
|
|
|
|
|
IMB_colormanagement_scene_linear_to_colorspace(
|
|
|
|
|
(float *)src, int(range.size()), 1, 4, colorspace);
|
|
|
|
|
const float4 *src_ptr = src;
|
|
|
|
|
uchar *bptr = ibuf->byte_buffer.data;
|
|
|
|
|
for (const int64_t idx : range) {
|
|
|
|
|
premul_float_to_straight_uchar(bptr + idx * 4, *src_ptr);
|
|
|
|
|
src_ptr++;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void tonemap_simple(float4 *scene_linear,
|
|
|
|
|
ImBuf *mask,
|
|
|
|
|
IndexRange range,
|
|
|
|
|
const AvgLogLum &avg)
|
|
|
|
|
{
|
|
|
|
|
const float4 *mask_float = mask != nullptr ? (const float4 *)mask->float_buffer.data : nullptr;
|
|
|
|
|
const uchar4 *mask_byte = mask != nullptr ? (const uchar4 *)mask->byte_buffer.data : nullptr;
|
|
|
|
|
|
|
|
|
|
int64_t index = 0;
|
|
|
|
|
for (const int64_t pixel_index : range) {
|
|
|
|
|
float4 input = scene_linear[index];
|
|
|
|
|
|
|
|
|
|
/* Apply correction. */
|
|
|
|
|
float3 pixel = input.xyz() * avg.al;
|
|
|
|
|
float3 d = pixel + avg.tmmd->offset;
|
|
|
|
|
pixel.x /= (d.x == 0.0f) ? 1.0f : d.x;
|
|
|
|
|
pixel.y /= (d.y == 0.0f) ? 1.0f : d.y;
|
|
|
|
|
pixel.z /= (d.z == 0.0f) ? 1.0f : d.z;
|
|
|
|
|
const float igm = avg.igm;
|
|
|
|
|
if (igm != 0.0f) {
|
|
|
|
|
pixel.x = powf(math::max(pixel.x, 0.0f), igm);
|
|
|
|
|
pixel.y = powf(math::max(pixel.y, 0.0f), igm);
|
|
|
|
|
pixel.z = powf(math::max(pixel.z, 0.0f), igm);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Apply mask. */
|
|
|
|
|
if (mask != nullptr) {
|
|
|
|
|
float3 msk(1.0f);
|
|
|
|
|
if (mask_float != nullptr) {
|
|
|
|
|
msk = mask_float[pixel_index].xyz();
|
2016-01-19 15:53:43 +01:00
|
|
|
}
|
2024-09-19 18:14:49 +02:00
|
|
|
else if (mask_byte != nullptr) {
|
|
|
|
|
rgb_uchar_to_float(msk, mask_byte[pixel_index]);
|
2016-01-19 15:53:43 +01:00
|
|
|
}
|
2024-09-19 18:14:49 +02:00
|
|
|
pixel = math::interpolate(input.xyz(), pixel, msk);
|
2016-01-19 15:53:43 +01:00
|
|
|
}
|
2024-09-19 18:14:49 +02:00
|
|
|
|
|
|
|
|
scene_linear[index] = float4(pixel.x, pixel.y, pixel.z, input.w);
|
|
|
|
|
index++;
|
2016-01-19 15:53:43 +01:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2024-09-19 18:14:49 +02:00
|
|
|
static void tonemap_rd_photoreceptor(float4 *scene_linear,
|
|
|
|
|
ImBuf *mask,
|
|
|
|
|
IndexRange range,
|
|
|
|
|
const AvgLogLum &avg)
|
2016-01-19 15:53:43 +01:00
|
|
|
{
|
2024-09-19 18:14:49 +02:00
|
|
|
const float4 *mask_float = mask != nullptr ? (const float4 *)mask->float_buffer.data : nullptr;
|
|
|
|
|
const uchar4 *mask_byte = mask != nullptr ? (const uchar4 *)mask->byte_buffer.data : nullptr;
|
|
|
|
|
|
|
|
|
|
const float f = expf(-avg.tmmd->intensity);
|
|
|
|
|
const float m = (avg.tmmd->contrast > 0.0f) ? avg.tmmd->contrast :
|
|
|
|
|
(0.3f + 0.7f * powf(avg.auto_key, 1.4f));
|
|
|
|
|
const float ic = 1.0f - avg.tmmd->correction, ia = 1.0f - avg.tmmd->adaptation;
|
|
|
|
|
|
|
|
|
|
int64_t index = 0;
|
|
|
|
|
for (const int64_t pixel_index : range) {
|
|
|
|
|
float4 input = scene_linear[index];
|
|
|
|
|
|
|
|
|
|
/* Apply correction. */
|
|
|
|
|
float3 pixel = input.xyz();
|
|
|
|
|
const float L = IMB_colormanagement_get_luminance(pixel);
|
|
|
|
|
float I_l = pixel.x + ic * (L - pixel.x);
|
|
|
|
|
float I_g = avg.cav.x + ic * (avg.lav - avg.cav.x);
|
|
|
|
|
float I_a = I_l + ia * (I_g - I_l);
|
|
|
|
|
pixel.x /= std::max(pixel.x + powf(f * I_a, m), 1.0e-30f);
|
|
|
|
|
I_l = pixel.y + ic * (L - pixel.y);
|
|
|
|
|
I_g = avg.cav.y + ic * (avg.lav - avg.cav.y);
|
|
|
|
|
I_a = I_l + ia * (I_g - I_l);
|
|
|
|
|
pixel.y /= std::max(pixel.y + powf(f * I_a, m), 1.0e-30f);
|
|
|
|
|
I_l = pixel.z + ic * (L - pixel.z);
|
|
|
|
|
I_g = avg.cav.z + ic * (avg.lav - avg.cav.z);
|
|
|
|
|
I_a = I_l + ia * (I_g - I_l);
|
|
|
|
|
pixel.z /= std::max(pixel.z + powf(f * I_a, m), 1.0e-30f);
|
|
|
|
|
|
|
|
|
|
/* Apply mask. */
|
|
|
|
|
if (mask != nullptr) {
|
|
|
|
|
float3 msk(1.0f);
|
|
|
|
|
if (mask_float != nullptr) {
|
|
|
|
|
msk = mask_float[pixel_index].xyz();
|
2016-01-19 15:53:43 +01:00
|
|
|
}
|
2024-09-19 18:14:49 +02:00
|
|
|
else if (mask_byte != nullptr) {
|
|
|
|
|
rgb_uchar_to_float(msk, mask_byte[pixel_index]);
|
2016-01-19 15:53:43 +01:00
|
|
|
}
|
2024-09-19 18:14:49 +02:00
|
|
|
pixel = math::interpolate(input.xyz(), pixel, msk);
|
2016-01-19 15:53:43 +01:00
|
|
|
}
|
2024-09-19 18:14:49 +02:00
|
|
|
|
|
|
|
|
scene_linear[index] = float4(pixel.x, pixel.y, pixel.z, input.w);
|
|
|
|
|
index++;
|
2016-01-19 15:53:43 +01:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2024-09-11 12:42:03 +02:00
|
|
|
static bool is_point_inside_quad(const StripScreenQuad &quad, int x, int y)
|
|
|
|
|
{
|
2024-09-26 10:17:04 +03:00
|
|
|
float2 pt(x + 0.5f, y + 0.5f);
|
2024-09-11 12:42:03 +02:00
|
|
|
return isect_point_quad_v2(pt, quad.v0, quad.v1, quad.v2, quad.v3);
|
|
|
|
|
}
|
|
|
|
|
|
2024-09-19 18:14:49 +02:00
|
|
|
struct AreaLuminance {
|
|
|
|
|
int64_t pixel_count = 0;
|
|
|
|
|
double sum = 0.0f;
|
|
|
|
|
float3 color_sum = {0, 0, 0};
|
|
|
|
|
double log_sum = 0.0;
|
|
|
|
|
float min = FLT_MAX;
|
|
|
|
|
float max = -FLT_MAX;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
static void tonemap_calc_chunk_luminance(const StripScreenQuad &quad,
|
|
|
|
|
const bool all_pixels_inside_quad,
|
|
|
|
|
const int width,
|
|
|
|
|
const IndexRange y_range,
|
|
|
|
|
const float4 *scene_linear,
|
|
|
|
|
AreaLuminance &r_lum)
|
2016-01-19 15:53:43 +01:00
|
|
|
{
|
2024-09-19 18:14:49 +02:00
|
|
|
for (const int y : y_range) {
|
|
|
|
|
for (int x = 0; x < width; x++) {
|
|
|
|
|
if (all_pixels_inside_quad || is_point_inside_quad(quad, x, y)) {
|
|
|
|
|
float4 pixel = *scene_linear;
|
|
|
|
|
r_lum.pixel_count++;
|
|
|
|
|
float L = IMB_colormanagement_get_luminance(pixel);
|
|
|
|
|
r_lum.sum += L;
|
|
|
|
|
r_lum.color_sum.x += pixel.x;
|
|
|
|
|
r_lum.color_sum.y += pixel.y;
|
|
|
|
|
r_lum.color_sum.z += pixel.z;
|
|
|
|
|
r_lum.log_sum += logf(math::max(L, 0.0f) + 1e-5f);
|
|
|
|
|
r_lum.max = math::max(r_lum.max, L);
|
|
|
|
|
r_lum.min = math::min(r_lum.min, L);
|
|
|
|
|
}
|
|
|
|
|
scene_linear++;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
2024-09-11 12:42:03 +02:00
|
|
|
|
2024-09-19 18:14:49 +02:00
|
|
|
static AreaLuminance tonemap_calc_input_luminance(const StripScreenQuad &quad, const ImBuf *ibuf)
|
|
|
|
|
{
|
2024-09-11 12:42:03 +02:00
|
|
|
/* Pixels outside the pre-transform strip area are ignored for luminance calculations.
|
|
|
|
|
* If strip area covers whole image, we can trivially accept all pixels. */
|
|
|
|
|
const bool all_pixels_inside_quad = is_point_inside_quad(quad, 0, 0) &&
|
|
|
|
|
is_point_inside_quad(quad, ibuf->x - 1, 0) &&
|
|
|
|
|
is_point_inside_quad(quad, 0, ibuf->y - 1) &&
|
|
|
|
|
is_point_inside_quad(quad, ibuf->x - 1, ibuf->y - 1);
|
2024-09-19 18:14:49 +02:00
|
|
|
|
|
|
|
|
AreaLuminance lum;
|
|
|
|
|
lum = threading::parallel_reduce(
|
|
|
|
|
IndexRange(ibuf->y),
|
|
|
|
|
32,
|
|
|
|
|
lum,
|
|
|
|
|
/* Calculate luminance for a chunk. */
|
|
|
|
|
[&](const IndexRange y_range, const AreaLuminance &init) {
|
|
|
|
|
AreaLuminance lum = init;
|
|
|
|
|
const int64_t chunk_size = y_range.size() * ibuf->x;
|
2024-09-20 13:14:57 +10:00
|
|
|
/* For float images, convert to scene-linear in place. The rest
|
|
|
|
|
* of tone-mapper can then continue with scene-linear values. */
|
2024-09-19 18:14:49 +02:00
|
|
|
if (ibuf->float_buffer.data != nullptr) {
|
|
|
|
|
float4 *fptr = reinterpret_cast<float4 *>(ibuf->float_buffer.data);
|
|
|
|
|
fptr += y_range.first() * ibuf->x;
|
|
|
|
|
pixels_to_scene_linear_float(ibuf->float_buffer.colorspace, fptr, chunk_size);
|
|
|
|
|
tonemap_calc_chunk_luminance(quad, all_pixels_inside_quad, ibuf->x, y_range, fptr, lum);
|
2024-09-11 12:42:03 +02:00
|
|
|
}
|
|
|
|
|
else {
|
2024-09-19 18:14:49 +02:00
|
|
|
const uchar *bptr = ibuf->byte_buffer.data + y_range.first() * ibuf->x * 4;
|
|
|
|
|
Array<float4> scene_linear(chunk_size);
|
|
|
|
|
pixels_to_scene_linear_byte(
|
|
|
|
|
ibuf->byte_buffer.colorspace, bptr, scene_linear.data(), chunk_size);
|
|
|
|
|
tonemap_calc_chunk_luminance(
|
|
|
|
|
quad, all_pixels_inside_quad, ibuf->x, y_range, scene_linear.data(), lum);
|
2024-09-11 12:42:03 +02:00
|
|
|
}
|
2024-09-19 18:14:49 +02:00
|
|
|
return lum;
|
|
|
|
|
},
|
|
|
|
|
/* Reduce luminance results. */
|
|
|
|
|
[&](const AreaLuminance &a, const AreaLuminance &b) {
|
|
|
|
|
AreaLuminance res;
|
|
|
|
|
res.pixel_count = a.pixel_count + b.pixel_count;
|
|
|
|
|
res.sum = a.sum + b.sum;
|
|
|
|
|
res.color_sum = a.color_sum + b.color_sum;
|
|
|
|
|
res.log_sum = a.log_sum + b.log_sum;
|
|
|
|
|
res.min = math::min(a.min, b.min);
|
|
|
|
|
res.max = math::max(a.max, b.max);
|
|
|
|
|
return res;
|
|
|
|
|
});
|
|
|
|
|
return lum;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void tonemapmodifier_apply(const StripScreenQuad &quad,
|
|
|
|
|
SequenceModifierData *smd,
|
|
|
|
|
ImBuf *ibuf,
|
|
|
|
|
ImBuf *mask)
|
|
|
|
|
{
|
|
|
|
|
const SequencerTonemapModifierData *tmmd = (const SequencerTonemapModifierData *)smd;
|
|
|
|
|
|
|
|
|
|
AreaLuminance lum = tonemap_calc_input_luminance(quad, ibuf);
|
|
|
|
|
if (lum.pixel_count == 0) {
|
2024-09-11 12:42:03 +02:00
|
|
|
return; /* Strip is zero size or off-screen. */
|
|
|
|
|
}
|
2024-09-19 18:14:49 +02:00
|
|
|
|
|
|
|
|
AvgLogLum data;
|
|
|
|
|
data.tmmd = tmmd;
|
|
|
|
|
data.lav = lum.sum / lum.pixel_count;
|
|
|
|
|
data.cav.x = lum.color_sum.x / lum.pixel_count;
|
|
|
|
|
data.cav.y = lum.color_sum.y / lum.pixel_count;
|
|
|
|
|
data.cav.z = lum.color_sum.z / lum.pixel_count;
|
|
|
|
|
float maxl = log(double(lum.max) + 1e-5f);
|
|
|
|
|
float minl = log(double(lum.min) + 1e-5f);
|
|
|
|
|
float avl = lum.log_sum / lum.pixel_count;
|
2016-01-19 15:53:43 +01:00
|
|
|
data.auto_key = (maxl > minl) ? ((maxl - avl) / (maxl - minl)) : 1.0f;
|
2024-09-19 18:14:49 +02:00
|
|
|
float al = exp(double(avl));
|
2016-01-19 15:53:43 +01:00
|
|
|
data.al = (al == 0.0f) ? 0.0f : (tmmd->key / al);
|
|
|
|
|
data.igm = (tmmd->gamma == 0.0f) ? 1.0f : (1.0f / tmmd->gamma);
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2024-09-19 18:14:49 +02:00
|
|
|
threading::parallel_for(
|
|
|
|
|
IndexRange(int64_t(ibuf->x) * ibuf->y), 64 * 1024, [&](IndexRange range) {
|
|
|
|
|
if (ibuf->float_buffer.data != nullptr) {
|
|
|
|
|
/* Float pixels: no need for temporary storage. Luminance calculation already converted
|
|
|
|
|
* data to scene linear. */
|
|
|
|
|
float4 *pixels = (float4 *)(ibuf->float_buffer.data) + range.first();
|
|
|
|
|
if (tmmd->type == SEQ_TONEMAP_RD_PHOTORECEPTOR) {
|
|
|
|
|
tonemap_rd_photoreceptor(pixels, mask, range, data);
|
|
|
|
|
}
|
|
|
|
|
else {
|
|
|
|
|
BLI_assert(tmmd->type == SEQ_TONEMAP_RH_SIMPLE);
|
|
|
|
|
tonemap_simple(pixels, mask, range, data);
|
|
|
|
|
}
|
|
|
|
|
scene_linear_to_image_chunk_float(ibuf, range);
|
|
|
|
|
}
|
|
|
|
|
else {
|
|
|
|
|
/* Byte pixels: temporary storage for scene linear pixel values. */
|
|
|
|
|
Array<float4> scene_linear(range.size());
|
|
|
|
|
pixels_to_scene_linear_byte(ibuf->byte_buffer.colorspace,
|
|
|
|
|
ibuf->byte_buffer.data + range.first() * 4,
|
|
|
|
|
scene_linear.data(),
|
|
|
|
|
range.size());
|
|
|
|
|
if (tmmd->type == SEQ_TONEMAP_RD_PHOTORECEPTOR) {
|
|
|
|
|
tonemap_rd_photoreceptor(scene_linear.data(), mask, range, data);
|
|
|
|
|
}
|
|
|
|
|
else {
|
|
|
|
|
BLI_assert(tmmd->type == SEQ_TONEMAP_RH_SIMPLE);
|
|
|
|
|
tonemap_simple(scene_linear.data(), mask, range, data);
|
|
|
|
|
}
|
|
|
|
|
scene_linear_to_image_chunk_byte(scene_linear.data(), ibuf, range);
|
|
|
|
|
}
|
|
|
|
|
});
|
2016-01-19 15:53:43 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static SequenceModifierTypeInfo seqModifier_Tonemap = {
|
2023-01-16 12:41:11 +11:00
|
|
|
/*name*/ CTX_N_(BLT_I18NCONTEXT_ID_SEQUENCE, "Tonemap"),
|
|
|
|
|
/*struct_name*/ "SequencerTonemapModifierData",
|
|
|
|
|
/*struct_size*/ sizeof(SequencerTonemapModifierData),
|
|
|
|
|
/*init_data*/ tonemapmodifier_init_data,
|
2023-07-20 09:46:24 +02:00
|
|
|
/*free_data*/ nullptr,
|
|
|
|
|
/*copy_data*/ nullptr,
|
2023-01-16 12:41:11 +11:00
|
|
|
/*apply*/ tonemapmodifier_apply,
|
2016-01-19 15:53:43 +01:00
|
|
|
};
|
|
|
|
|
|
2023-08-30 22:36:36 +02:00
|
|
|
static SequenceModifierTypeInfo seqModifier_SoundEqualizer = {
|
2024-08-19 11:37:20 +10:00
|
|
|
/*name*/ CTX_N_(BLT_I18NCONTEXT_ID_SEQUENCE, "Equalizer"),
|
|
|
|
|
/*struct_name*/ "SoundEqualizerModifierData",
|
|
|
|
|
/*struct_size*/ sizeof(SoundEqualizerModifierData),
|
|
|
|
|
/*init_data*/ SEQ_sound_equalizermodifier_init_data,
|
|
|
|
|
/*free_data*/ SEQ_sound_equalizermodifier_free,
|
|
|
|
|
/*copy_data*/ SEQ_sound_equalizermodifier_copy_data,
|
|
|
|
|
/*apply*/ nullptr,
|
2023-08-30 22:36:36 +02:00
|
|
|
};
|
2020-06-16 12:32:42 +10:00
|
|
|
/** \} */
|
|
|
|
|
|
|
|
|
|
/* -------------------------------------------------------------------- */
|
|
|
|
|
/** \name Public Modifier Functions
|
|
|
|
|
* \{ */
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2023-07-20 20:11:08 +10:00
|
|
|
static void sequence_modifier_type_info_init()
|
2012-08-19 15:41:56 +00:00
|
|
|
{
|
|
|
|
|
#define INIT_TYPE(typeName) (modifiersTypes[seqModifierType_##typeName] = &seqModifier_##typeName)
|
|
|
|
|
|
|
|
|
|
INIT_TYPE(ColorBalance);
|
|
|
|
|
INIT_TYPE(Curves);
|
|
|
|
|
INIT_TYPE(HueCorrect);
|
2012-08-24 09:07:04 +00:00
|
|
|
INIT_TYPE(BrightContrast);
|
2013-05-08 14:20:57 +00:00
|
|
|
INIT_TYPE(Mask);
|
2015-12-28 11:55:14 +01:00
|
|
|
INIT_TYPE(WhiteBalance);
|
2016-01-19 15:53:43 +01:00
|
|
|
INIT_TYPE(Tonemap);
|
2023-08-30 22:36:36 +02:00
|
|
|
INIT_TYPE(SoundEqualizer);
|
2012-08-19 15:41:56 +00:00
|
|
|
|
|
|
|
|
#undef INIT_TYPE
|
|
|
|
|
}
|
|
|
|
|
|
2020-12-19 05:57:27 +01:00
|
|
|
const SequenceModifierTypeInfo *SEQ_modifier_type_info_get(int type)
|
2012-08-19 15:41:56 +00:00
|
|
|
{
|
|
|
|
|
if (!modifierTypesInit) {
|
|
|
|
|
sequence_modifier_type_info_init();
|
2014-03-20 15:45:20 +06:00
|
|
|
modifierTypesInit = true;
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return modifiersTypes[type];
|
|
|
|
|
}
|
|
|
|
|
|
2020-12-19 05:57:27 +01:00
|
|
|
SequenceModifierData *SEQ_modifier_new(Sequence *seq, const char *name, int type)
|
2012-08-19 15:41:56 +00:00
|
|
|
{
|
|
|
|
|
SequenceModifierData *smd;
|
2020-12-19 05:57:27 +01:00
|
|
|
const SequenceModifierTypeInfo *smti = SEQ_modifier_type_info_get(type);
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2023-07-20 09:46:24 +02:00
|
|
|
smd = static_cast<SequenceModifierData *>(MEM_callocN(smti->struct_size, "sequence modifier"));
|
2012-08-19 15:41:56 +00:00
|
|
|
|
|
|
|
|
smd->type = type;
|
|
|
|
|
smd->flag |= SEQUENCE_MODIFIER_EXPANDED;
|
|
|
|
|
|
2019-04-22 09:39:35 +10:00
|
|
|
if (!name || !name[0]) {
|
2023-05-09 12:50:37 +10:00
|
|
|
STRNCPY(smd->name, smti->name);
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
|
|
|
|
else {
|
2023-05-09 12:50:37 +10:00
|
|
|
STRNCPY(smd->name, name);
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
2012-08-19 15:41:56 +00:00
|
|
|
|
|
|
|
|
BLI_addtail(&seq->modifiers, smd);
|
|
|
|
|
|
2020-12-19 05:57:27 +01:00
|
|
|
SEQ_modifier_unique_name(seq, smd);
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2019-04-22 09:39:35 +10:00
|
|
|
if (smti->init_data) {
|
2012-08-19 15:41:56 +00:00
|
|
|
smti->init_data(smd);
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
2012-08-20 10:15:32 +00:00
|
|
|
|
|
|
|
|
return smd;
|
|
|
|
|
}
|
|
|
|
|
|
2020-12-19 05:57:27 +01:00
|
|
|
bool SEQ_modifier_remove(Sequence *seq, SequenceModifierData *smd)
|
2012-08-20 10:15:32 +00:00
|
|
|
{
|
2019-04-22 09:39:35 +10:00
|
|
|
if (BLI_findindex(&seq->modifiers, smd) == -1) {
|
2014-03-20 15:45:20 +06:00
|
|
|
return false;
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
2012-08-20 10:15:32 +00:00
|
|
|
|
|
|
|
|
BLI_remlink(&seq->modifiers, smd);
|
2020-12-19 05:57:27 +01:00
|
|
|
SEQ_modifier_free(smd);
|
2012-08-20 10:15:32 +00:00
|
|
|
|
2014-03-20 15:45:20 +06:00
|
|
|
return true;
|
2012-08-20 10:15:32 +00:00
|
|
|
}
|
|
|
|
|
|
2020-12-19 05:57:27 +01:00
|
|
|
void SEQ_modifier_clear(Sequence *seq)
|
2012-08-20 10:15:32 +00:00
|
|
|
{
|
|
|
|
|
SequenceModifierData *smd, *smd_next;
|
|
|
|
|
|
2023-07-20 09:46:24 +02:00
|
|
|
for (smd = static_cast<SequenceModifierData *>(seq->modifiers.first); smd; smd = smd_next) {
|
2012-08-20 10:15:32 +00:00
|
|
|
smd_next = smd->next;
|
2020-12-19 05:57:27 +01:00
|
|
|
SEQ_modifier_free(smd);
|
2012-08-20 10:15:32 +00:00
|
|
|
}
|
|
|
|
|
|
2014-02-08 06:07:10 +11:00
|
|
|
BLI_listbase_clear(&seq->modifiers);
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
|
2020-12-19 05:57:27 +01:00
|
|
|
void SEQ_modifier_free(SequenceModifierData *smd)
|
2012-08-19 15:41:56 +00:00
|
|
|
{
|
2020-12-19 05:57:27 +01:00
|
|
|
const SequenceModifierTypeInfo *smti = SEQ_modifier_type_info_get(smd->type);
|
2012-08-19 15:41:56 +00:00
|
|
|
|
|
|
|
|
if (smti && smti->free_data) {
|
|
|
|
|
smti->free_data(smd);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
MEM_freeN(smd);
|
|
|
|
|
}
|
|
|
|
|
|
2020-12-19 05:57:27 +01:00
|
|
|
void SEQ_modifier_unique_name(Sequence *seq, SequenceModifierData *smd)
|
2012-08-19 15:41:56 +00:00
|
|
|
{
|
2020-12-19 05:57:27 +01:00
|
|
|
const SequenceModifierTypeInfo *smti = SEQ_modifier_type_info_get(smd->type);
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2015-08-16 17:32:01 +10:00
|
|
|
BLI_uniquename(&seq->modifiers,
|
|
|
|
|
smd,
|
|
|
|
|
CTX_DATA_(BLT_I18NCONTEXT_ID_SEQUENCE, smti->name),
|
|
|
|
|
'.',
|
2013-03-25 08:29:06 +00:00
|
|
|
offsetof(SequenceModifierData, name),
|
|
|
|
|
sizeof(smd->name));
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
|
2020-12-19 05:57:27 +01:00
|
|
|
SequenceModifierData *SEQ_modifier_find_by_name(Sequence *seq, const char *name)
|
2012-08-19 15:41:56 +00:00
|
|
|
{
|
2023-07-20 09:46:24 +02:00
|
|
|
return static_cast<SequenceModifierData *>(
|
|
|
|
|
BLI_findstring(&(seq->modifiers), name, offsetof(SequenceModifierData, name)));
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
|
2024-05-07 19:37:24 +02:00
|
|
|
static bool skip_modifier(Scene *scene, const SequenceModifierData *smd, int timeline_frame)
|
|
|
|
|
{
|
|
|
|
|
using namespace blender::seq;
|
|
|
|
|
|
|
|
|
|
if (smd->mask_sequence == nullptr) {
|
|
|
|
|
return false;
|
|
|
|
|
}
|
|
|
|
|
const bool strip_has_ended_skip = smd->mask_input_type == SEQUENCE_MASK_INPUT_STRIP &&
|
|
|
|
|
smd->mask_time == SEQUENCE_MASK_TIME_RELATIVE &&
|
|
|
|
|
!SEQ_time_strip_intersects_frame(
|
|
|
|
|
scene, smd->mask_sequence, timeline_frame);
|
|
|
|
|
const bool missing_data_skip = !SEQ_sequence_has_valid_data(smd->mask_sequence) ||
|
|
|
|
|
media_presence_is_missing(scene, smd->mask_sequence);
|
|
|
|
|
|
|
|
|
|
return strip_has_ended_skip || missing_data_skip;
|
|
|
|
|
}
|
|
|
|
|
|
2024-09-09 14:55:24 +02:00
|
|
|
void SEQ_modifier_apply_stack(const SeqRenderData *context,
|
|
|
|
|
const Sequence *seq,
|
|
|
|
|
ImBuf *ibuf,
|
|
|
|
|
int timeline_frame)
|
2012-08-19 15:41:56 +00:00
|
|
|
{
|
2024-09-11 12:42:03 +02:00
|
|
|
const StripScreenQuad quad = get_strip_screen_quad(context, seq);
|
|
|
|
|
|
Color Management, Stage 2: Switch color pipeline to use OpenColorIO
Replace old color pipeline which was supporting linear/sRGB color spaces
only with OpenColorIO-based pipeline.
This introduces two configurable color spaces:
- Input color space for images and movie clips. This space is used to convert
images/movies from color space in which file is saved to Blender's linear
space (for float images, byte images are not internally converted, only input
space is stored for such images and used later).
This setting could be found in image/clip data block settings.
- Display color space which defines space in which particular display is working.
This settings could be found in scene's Color Management panel.
When render result is being displayed on the screen, apart from converting image
to display space, some additional conversions could happen.
This conversions are:
- View, which defines tone curve applying before display transformation.
These are different ways to view the image on the same display device.
For example it could be used to emulate film view on sRGB display.
- Exposure affects on image exposure before tone map is applied.
- Gamma is post-display gamma correction, could be used to match particular
display gamma.
- RGB curves are user-defined curves which are applying before display
transformation, could be used for different purposes.
All this settings by default are only applying on render result and does not
affect on other images. If some particular image needs to be affected by this
transformation, "View as Render" setting of image data block should be set to
truth. Movie clips are always affected by all display transformations.
This commit also introduces configurable color space in which sequencer is
working. This setting could be found in scene's Color Management panel and
it should be used if such stuff as grading needs to be done in color space
different from sRGB (i.e. when Film view on sRGB display is use, using VD16
space as sequencer's internal space would make grading working in space
which is close to the space using for display).
Some technical notes:
- Image buffer's float buffer is now always in linear space, even if it was
created from 16bit byte images.
- Space of byte buffer is stored in image buffer's rect_colorspace property.
- Profile of image buffer was removed since it's not longer meaningful.
- OpenGL and GLSL is supposed to always work in sRGB space. It is possible
to support other spaces, but it's quite large project which isn't so
much important.
- Legacy Color Management option disabled is emulated by using None display.
It could have some regressions, but there's no clear way to avoid them.
- If OpenColorIO is disabled on build time, it should make blender behaving
in the same way as previous release with color management enabled.
More details could be found at this page (more details would be added soon):
http://wiki.blender.org/index.php/Dev:Ref/Release_Notes/2.64/Color_Management
--
Thanks to Xavier Thomas, Lukas Toene for initial work on OpenColorIO
integration and to Brecht van Lommel for some further development and code/
usecase review!
2012-09-15 10:05:07 +00:00
|
|
|
if (seq->modifiers.first && (seq->flag & SEQ_USE_LINEAR_MODIFIERS)) {
|
2024-09-09 14:55:24 +02:00
|
|
|
SEQ_render_imbuf_from_sequencer_space(context->scene, ibuf);
|
Color Management, Stage 2: Switch color pipeline to use OpenColorIO
Replace old color pipeline which was supporting linear/sRGB color spaces
only with OpenColorIO-based pipeline.
This introduces two configurable color spaces:
- Input color space for images and movie clips. This space is used to convert
images/movies from color space in which file is saved to Blender's linear
space (for float images, byte images are not internally converted, only input
space is stored for such images and used later).
This setting could be found in image/clip data block settings.
- Display color space which defines space in which particular display is working.
This settings could be found in scene's Color Management panel.
When render result is being displayed on the screen, apart from converting image
to display space, some additional conversions could happen.
This conversions are:
- View, which defines tone curve applying before display transformation.
These are different ways to view the image on the same display device.
For example it could be used to emulate film view on sRGB display.
- Exposure affects on image exposure before tone map is applied.
- Gamma is post-display gamma correction, could be used to match particular
display gamma.
- RGB curves are user-defined curves which are applying before display
transformation, could be used for different purposes.
All this settings by default are only applying on render result and does not
affect on other images. If some particular image needs to be affected by this
transformation, "View as Render" setting of image data block should be set to
truth. Movie clips are always affected by all display transformations.
This commit also introduces configurable color space in which sequencer is
working. This setting could be found in scene's Color Management panel and
it should be used if such stuff as grading needs to be done in color space
different from sRGB (i.e. when Film view on sRGB display is use, using VD16
space as sequencer's internal space would make grading working in space
which is close to the space using for display).
Some technical notes:
- Image buffer's float buffer is now always in linear space, even if it was
created from 16bit byte images.
- Space of byte buffer is stored in image buffer's rect_colorspace property.
- Profile of image buffer was removed since it's not longer meaningful.
- OpenGL and GLSL is supposed to always work in sRGB space. It is possible
to support other spaces, but it's quite large project which isn't so
much important.
- Legacy Color Management option disabled is emulated by using None display.
It could have some regressions, but there's no clear way to avoid them.
- If OpenColorIO is disabled on build time, it should make blender behaving
in the same way as previous release with color management enabled.
More details could be found at this page (more details would be added soon):
http://wiki.blender.org/index.php/Dev:Ref/Release_Notes/2.64/Color_Management
--
Thanks to Xavier Thomas, Lukas Toene for initial work on OpenColorIO
integration and to Brecht van Lommel for some further development and code/
usecase review!
2012-09-15 10:05:07 +00:00
|
|
|
}
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2023-08-04 08:51:13 +10:00
|
|
|
LISTBASE_FOREACH (SequenceModifierData *, smd, &seq->modifiers) {
|
2020-12-19 05:57:27 +01:00
|
|
|
const SequenceModifierTypeInfo *smti = SEQ_modifier_type_info_get(smd->type);
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-19 15:41:56 +00:00
|
|
|
/* could happen if modifier is being removed or not exists in current version of blender */
|
2019-04-22 09:39:35 +10:00
|
|
|
if (!smti) {
|
2012-08-19 15:41:56 +00:00
|
|
|
continue;
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2012-08-19 15:41:56 +00:00
|
|
|
/* modifier is muted, do nothing */
|
2019-04-22 09:39:35 +10:00
|
|
|
if (smd->flag & SEQUENCE_MODIFIER_MUTE) {
|
2012-08-19 15:41:56 +00:00
|
|
|
continue;
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2024-05-07 19:37:24 +02:00
|
|
|
if (smti->apply && !skip_modifier(context->scene, smd, timeline_frame)) {
|
2016-01-25 11:16:49 +01:00
|
|
|
int frame_offset;
|
|
|
|
|
if (smd->mask_time == SEQUENCE_MASK_TIME_RELATIVE) {
|
|
|
|
|
frame_offset = seq->start;
|
|
|
|
|
}
|
2021-06-26 21:35:18 +10:00
|
|
|
else /* if (smd->mask_time == SEQUENCE_MASK_TIME_ABSOLUTE) */ {
|
2019-11-11 10:52:26 -03:00
|
|
|
frame_offset = smd->mask_id ? ((Mask *)smd->mask_id)->sfra : 0;
|
2016-01-25 11:16:49 +01:00
|
|
|
}
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2020-11-06 14:10:59 +01:00
|
|
|
ImBuf *mask = modifier_mask_get(
|
2023-07-20 09:46:24 +02:00
|
|
|
smd, context, timeline_frame, frame_offset, ibuf->float_buffer.data != nullptr);
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2024-09-11 12:42:03 +02:00
|
|
|
smti->apply(quad, smd, ibuf, mask);
|
2019-04-17 06:17:24 +02:00
|
|
|
|
2019-04-22 09:39:35 +10:00
|
|
|
if (mask) {
|
2012-08-19 15:41:56 +00:00
|
|
|
IMB_freeImBuf(mask);
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
2019-04-17 06:17:24 +02:00
|
|
|
}
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
2019-04-17 06:17:24 +02:00
|
|
|
|
Color Management, Stage 2: Switch color pipeline to use OpenColorIO
Replace old color pipeline which was supporting linear/sRGB color spaces
only with OpenColorIO-based pipeline.
This introduces two configurable color spaces:
- Input color space for images and movie clips. This space is used to convert
images/movies from color space in which file is saved to Blender's linear
space (for float images, byte images are not internally converted, only input
space is stored for such images and used later).
This setting could be found in image/clip data block settings.
- Display color space which defines space in which particular display is working.
This settings could be found in scene's Color Management panel.
When render result is being displayed on the screen, apart from converting image
to display space, some additional conversions could happen.
This conversions are:
- View, which defines tone curve applying before display transformation.
These are different ways to view the image on the same display device.
For example it could be used to emulate film view on sRGB display.
- Exposure affects on image exposure before tone map is applied.
- Gamma is post-display gamma correction, could be used to match particular
display gamma.
- RGB curves are user-defined curves which are applying before display
transformation, could be used for different purposes.
All this settings by default are only applying on render result and does not
affect on other images. If some particular image needs to be affected by this
transformation, "View as Render" setting of image data block should be set to
truth. Movie clips are always affected by all display transformations.
This commit also introduces configurable color space in which sequencer is
working. This setting could be found in scene's Color Management panel and
it should be used if such stuff as grading needs to be done in color space
different from sRGB (i.e. when Film view on sRGB display is use, using VD16
space as sequencer's internal space would make grading working in space
which is close to the space using for display).
Some technical notes:
- Image buffer's float buffer is now always in linear space, even if it was
created from 16bit byte images.
- Space of byte buffer is stored in image buffer's rect_colorspace property.
- Profile of image buffer was removed since it's not longer meaningful.
- OpenGL and GLSL is supposed to always work in sRGB space. It is possible
to support other spaces, but it's quite large project which isn't so
much important.
- Legacy Color Management option disabled is emulated by using None display.
It could have some regressions, but there's no clear way to avoid them.
- If OpenColorIO is disabled on build time, it should make blender behaving
in the same way as previous release with color management enabled.
More details could be found at this page (more details would be added soon):
http://wiki.blender.org/index.php/Dev:Ref/Release_Notes/2.64/Color_Management
--
Thanks to Xavier Thomas, Lukas Toene for initial work on OpenColorIO
integration and to Brecht van Lommel for some further development and code/
usecase review!
2012-09-15 10:05:07 +00:00
|
|
|
if (seq->modifiers.first && (seq->flag & SEQ_USE_LINEAR_MODIFIERS)) {
|
2024-09-09 14:55:24 +02:00
|
|
|
seq_imbuf_to_sequencer_space(context->scene, ibuf, false);
|
Color Management, Stage 2: Switch color pipeline to use OpenColorIO
Replace old color pipeline which was supporting linear/sRGB color spaces
only with OpenColorIO-based pipeline.
This introduces two configurable color spaces:
- Input color space for images and movie clips. This space is used to convert
images/movies from color space in which file is saved to Blender's linear
space (for float images, byte images are not internally converted, only input
space is stored for such images and used later).
This setting could be found in image/clip data block settings.
- Display color space which defines space in which particular display is working.
This settings could be found in scene's Color Management panel.
When render result is being displayed on the screen, apart from converting image
to display space, some additional conversions could happen.
This conversions are:
- View, which defines tone curve applying before display transformation.
These are different ways to view the image on the same display device.
For example it could be used to emulate film view on sRGB display.
- Exposure affects on image exposure before tone map is applied.
- Gamma is post-display gamma correction, could be used to match particular
display gamma.
- RGB curves are user-defined curves which are applying before display
transformation, could be used for different purposes.
All this settings by default are only applying on render result and does not
affect on other images. If some particular image needs to be affected by this
transformation, "View as Render" setting of image data block should be set to
truth. Movie clips are always affected by all display transformations.
This commit also introduces configurable color space in which sequencer is
working. This setting could be found in scene's Color Management panel and
it should be used if such stuff as grading needs to be done in color space
different from sRGB (i.e. when Film view on sRGB display is use, using VD16
space as sequencer's internal space would make grading working in space
which is close to the space using for display).
Some technical notes:
- Image buffer's float buffer is now always in linear space, even if it was
created from 16bit byte images.
- Space of byte buffer is stored in image buffer's rect_colorspace property.
- Profile of image buffer was removed since it's not longer meaningful.
- OpenGL and GLSL is supposed to always work in sRGB space. It is possible
to support other spaces, but it's quite large project which isn't so
much important.
- Legacy Color Management option disabled is emulated by using None display.
It could have some regressions, but there's no clear way to avoid them.
- If OpenColorIO is disabled on build time, it should make blender behaving
in the same way as previous release with color management enabled.
More details could be found at this page (more details would be added soon):
http://wiki.blender.org/index.php/Dev:Ref/Release_Notes/2.64/Color_Management
--
Thanks to Xavier Thomas, Lukas Toene for initial work on OpenColorIO
integration and to Brecht van Lommel for some further development and code/
usecase review!
2012-09-15 10:05:07 +00:00
|
|
|
}
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
|
2020-12-19 05:57:27 +01:00
|
|
|
void SEQ_modifier_list_copy(Sequence *seqn, Sequence *seq)
|
2012-08-19 15:41:56 +00:00
|
|
|
{
|
2023-08-04 08:51:13 +10:00
|
|
|
LISTBASE_FOREACH (SequenceModifierData *, smd, &seq->modifiers) {
|
2012-08-19 15:41:56 +00:00
|
|
|
SequenceModifierData *smdn;
|
2020-12-19 05:57:27 +01:00
|
|
|
const SequenceModifierTypeInfo *smti = SEQ_modifier_type_info_get(smd->type);
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2023-07-20 09:46:24 +02:00
|
|
|
smdn = static_cast<SequenceModifierData *>(MEM_dupallocN(smd));
|
2012-08-19 15:41:56 +00:00
|
|
|
|
2019-04-22 09:39:35 +10:00
|
|
|
if (smti && smti->copy_data) {
|
2012-08-19 15:41:56 +00:00
|
|
|
smti->copy_data(smdn, smd);
|
2019-04-22 09:39:35 +10:00
|
|
|
}
|
2012-08-19 15:41:56 +00:00
|
|
|
|
|
|
|
|
BLI_addtail(&seqn->modifiers, smdn);
|
2023-08-28 18:56:25 +10:00
|
|
|
BLI_uniquename(&seqn->modifiers,
|
|
|
|
|
smdn,
|
|
|
|
|
"Strip Modifier",
|
|
|
|
|
'.',
|
|
|
|
|
offsetof(SequenceModifierData, name),
|
|
|
|
|
sizeof(SequenceModifierData::name));
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2020-12-19 05:57:27 +01:00
|
|
|
int SEQ_sequence_supports_modifiers(Sequence *seq)
|
2012-08-19 15:41:56 +00:00
|
|
|
{
|
2023-08-15 14:49:11 +02:00
|
|
|
return (seq->type != SEQ_TYPE_SOUND_RAM);
|
2012-08-19 15:41:56 +00:00
|
|
|
}
|
2020-06-18 12:40:38 +10:00
|
|
|
|
|
|
|
|
/** \} */
|
2020-11-06 14:27:51 +01:00
|
|
|
|
|
|
|
|
/* -------------------------------------------------------------------- */
|
|
|
|
|
/** \name .blend File I/O
|
|
|
|
|
* \{ */
|
|
|
|
|
|
2020-12-19 05:57:27 +01:00
|
|
|
void SEQ_modifier_blend_write(BlendWriter *writer, ListBase *modbase)
|
2020-11-06 14:27:51 +01:00
|
|
|
{
|
|
|
|
|
LISTBASE_FOREACH (SequenceModifierData *, smd, modbase) {
|
2020-12-19 05:57:27 +01:00
|
|
|
const SequenceModifierTypeInfo *smti = SEQ_modifier_type_info_get(smd->type);
|
2020-11-06 14:27:51 +01:00
|
|
|
|
|
|
|
|
if (smti) {
|
|
|
|
|
BLO_write_struct_by_name(writer, smti->struct_name, smd);
|
|
|
|
|
|
|
|
|
|
if (smd->type == seqModifierType_Curves) {
|
|
|
|
|
CurvesModifierData *cmd = (CurvesModifierData *)smd;
|
|
|
|
|
|
|
|
|
|
BKE_curvemapping_blend_write(writer, &cmd->curve_mapping);
|
|
|
|
|
}
|
|
|
|
|
else if (smd->type == seqModifierType_HueCorrect) {
|
|
|
|
|
HueCorrectModifierData *hcmd = (HueCorrectModifierData *)smd;
|
|
|
|
|
|
|
|
|
|
BKE_curvemapping_blend_write(writer, &hcmd->curve_mapping);
|
|
|
|
|
}
|
2023-08-30 22:36:36 +02:00
|
|
|
else if (smd->type == seqModifierType_SoundEqualizer) {
|
|
|
|
|
SoundEqualizerModifierData *semd = (SoundEqualizerModifierData *)smd;
|
|
|
|
|
LISTBASE_FOREACH (EQCurveMappingData *, eqcmd, &semd->graphics) {
|
|
|
|
|
BLO_write_struct_by_name(writer, "EQCurveMappingData", eqcmd);
|
|
|
|
|
BKE_curvemapping_blend_write(writer, &eqcmd->curve_mapping);
|
|
|
|
|
}
|
|
|
|
|
}
|
2020-11-06 14:27:51 +01:00
|
|
|
}
|
|
|
|
|
else {
|
|
|
|
|
BLO_write_struct(writer, SequenceModifierData, smd);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2020-12-19 05:57:27 +01:00
|
|
|
void SEQ_modifier_blend_read_data(BlendDataReader *reader, ListBase *lb)
|
2020-11-06 14:27:51 +01:00
|
|
|
{
|
2024-04-24 17:01:22 +02:00
|
|
|
BLO_read_struct_list(reader, SequenceModifierData, lb);
|
2020-11-06 14:27:51 +01:00
|
|
|
|
|
|
|
|
LISTBASE_FOREACH (SequenceModifierData *, smd, lb) {
|
|
|
|
|
if (smd->mask_sequence) {
|
2024-04-24 17:01:22 +02:00
|
|
|
BLO_read_struct(reader, Sequence, &smd->mask_sequence);
|
2020-11-06 14:27:51 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (smd->type == seqModifierType_Curves) {
|
|
|
|
|
CurvesModifierData *cmd = (CurvesModifierData *)smd;
|
|
|
|
|
|
|
|
|
|
BKE_curvemapping_blend_read(reader, &cmd->curve_mapping);
|
|
|
|
|
}
|
|
|
|
|
else if (smd->type == seqModifierType_HueCorrect) {
|
|
|
|
|
HueCorrectModifierData *hcmd = (HueCorrectModifierData *)smd;
|
|
|
|
|
|
|
|
|
|
BKE_curvemapping_blend_read(reader, &hcmd->curve_mapping);
|
|
|
|
|
}
|
2023-08-30 22:36:36 +02:00
|
|
|
else if (smd->type == seqModifierType_SoundEqualizer) {
|
|
|
|
|
SoundEqualizerModifierData *semd = (SoundEqualizerModifierData *)smd;
|
2024-04-24 17:01:22 +02:00
|
|
|
BLO_read_struct_list(reader, EQCurveMappingData, &semd->graphics);
|
2023-08-30 22:36:36 +02:00
|
|
|
LISTBASE_FOREACH (EQCurveMappingData *, eqcmd, &semd->graphics) {
|
|
|
|
|
BKE_curvemapping_blend_read(reader, &eqcmd->curve_mapping);
|
|
|
|
|
}
|
|
|
|
|
}
|
2020-11-06 14:27:51 +01:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/** \} */
|