griefith/test2
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

VSE: speedup Luma Waveform display output mode

Browse Source 
Use multithreading, plus make the non-float / non-separate luma
calculations faster by avoiding byte->float->byte conversions
back and forth.

On a 4K resolution sequencer display, time taken to calculate the
waveform (Windows, Ryzen 5950X):

- regular images: 127.0ms -> 6.4ms
- regular images, separate colors: 160.3ms -> 13.1ms
- float images: 86.2ms -> 11.1ms
- float images, separate colors: 162.9ms -> 17.4ms

This also fixes curious "one black pixel row" in the middle of the
waveform, which was caused by the code spreading 256 possible luma
values over "off by one" vertical range:

Pull Request: https://projects.blender.org/blender/blender/pulls/115579
This commit is contained in:
Aras Pranckevicius
2023-11-30 18:19:40 +01:00
committed by Aras Pranckevicius
parent da83b2951d
commit 93ca2788ff
1 changed files with 110 additions and 69 deletions
Download Patch File Download Diff File Expand all files Collapse all files

179
source/blender/editors/space_sequencer/sequencer_scopes.cc
View File

@@ -10,6 +10,7 @@
#include <cstring>
#include "BLI_task.h"
#include "BLI_task.hh"
#include "BLI_utildefines.h"
#include "IMB_colormanagement.h"
@@ -18,6 +19,12 @@
#include "sequencer_intern.hh"
// #define DEBUG_TIME
#ifdef DEBUG_TIME
# include "BLI_timeit.hh"
#endif
/* XXX(@ideasman42): why is this function better than BLI_math version?
* only difference is it does some normalize after, need to double check on this. */
static void rgb_to_yuv_normalized(const float rgb[3], float yuv[3])
@@ -125,8 +132,11 @@ static void wform_put_grid(uchar *tgt, int w, int h)
static ImBuf *make_waveform_view_from_ibuf_byte(ImBuf *ibuf)
{
#ifdef DEBUG_TIME
SCOPED_TIMER_AVERAGED(__func__);
#endif
using namespace blender;
ImBuf *rval = IMB_allocImBuf(ibuf->x + 3, 515, 32, IB_rect);
int x, y;
const uchar *src = ibuf->byte_buffer.data;
uchar *tgt = rval->byte_buffer.data;
int w = ibuf->x + 3;
@@ -137,37 +147,53 @@ static ImBuf *make_waveform_view_from_ibuf_byte(ImBuf *ibuf)
wform_put_grid(tgt, w, h);
wform_put_border(tgt, w, h);
for (x = 0; x < 256; x++) {
for (int x = 0; x < 256; x++) {
wtable[x] = uchar(pow((float(x) + 1.0f) / 256.0f, waveform_gamma) * 255.0f);
}
for (y = 0; y < ibuf->y; y++) {
uchar *last_p = nullptr;
/* IMB_colormanagement_get_luminance_byte for each pixel is quite a lot of
* overhead, so instead get luma coefficients as 16-bit integers. */
float coeffs[3];
IMB_colormanagement_get_luminance_coefficients(coeffs);
int muls[3] = {int(coeffs[0] * 65535), int(coeffs[1] * 65535), int(coeffs[2] * 65535)};
for (x = 0; x < ibuf->x; x++) {
const uchar *rgb = src + 4 * (ibuf->x * y + x);
float v = float(IMB_colormanagement_get_luminance_byte(rgb)) / 255.0f;
uchar *p = tgt;
p += 4 * (w * (int(v * (h - 3)) + 1) + x + 1);
/* Parallel over x, since each column is easily independent from others. */
threading::parallel_for(IndexRange(ibuf->x), 16, [&](IndexRange x_range) {
for (int y = 0; y < ibuf->y; y++) {
uchar *last_p = nullptr;
scope_put_pixel(wtable, p);
p += 4 * w;
scope_put_pixel(wtable, p);
for (const int x : x_range) {
const uchar *rgb = src + 4 * (ibuf->x * y + x);
/* +1 is "Sree's solution" from http://stereopsis.com/doubleblend.html */
int rgb0 = rgb[0] + 1;
int rgb1 = rgb[1] + 1;
int rgb2 = rgb[2] + 1;
int luma = (rgb0 * muls[0] + rgb1 * muls[1] + rgb2 * muls[2]) >> 16;
int luma_y = clamp_i(luma, 0, 255) * 2;
uchar *p = tgt + 4 * (w * (luma_y + 1) + x + 1);
if (last_p != nullptr) {
wform_put_line(w, last_p, p);
scope_put_pixel(wtable, p);
p += 4 * w;
scope_put_pixel(wtable, p);
if (last_p != nullptr) {
wform_put_line(w, last_p, p);
}
last_p = p;
}
last_p = p;
}
}
});
return rval;
}
static ImBuf *make_waveform_view_from_ibuf_float(ImBuf *ibuf)
{
#ifdef DEBUG_TIME
SCOPED_TIMER_AVERAGED(__func__);
#endif
using namespace blender;
ImBuf *rval = IMB_allocImBuf(ibuf->x + 3, 515, 32, IB_rect);
int x, y;
const float *src = ibuf->float_buffer.data;
uchar *tgt = rval->byte_buffer.data;
int w = ibuf->x + 3;
@@ -177,32 +203,35 @@ static ImBuf *make_waveform_view_from_ibuf_float(ImBuf *ibuf)
wform_put_grid(tgt, w, h);
for (x = 0; x < 256; x++) {
for (int x = 0; x < 256; x++) {
wtable[x] = uchar(pow((float(x) + 1.0f) / 256.0f, waveform_gamma) * 255.0f);
}
for (y = 0; y < ibuf->y; y++) {
uchar *last_p = nullptr;
/* Parallel over x, since each column is easily independent from others. */
threading::parallel_for(IndexRange(ibuf->x), 16, [&](IndexRange x_range) {
for (int y = 0; y < ibuf->y; y++) {
uchar *last_p = nullptr;
for (x = 0; x < ibuf->x; x++) {
const float *rgb = src + 4 * (ibuf->x * y + x);
float v = IMB_colormanagement_get_luminance(rgb);
uchar *p = tgt;
for (const int x : x_range) {
const float *rgb = src + 4 * (ibuf->x * y + x);
float v = IMB_colormanagement_get_luminance(rgb);
uchar *p = tgt;
CLAMP(v, 0.0f, 1.0f);
CLAMP(v, 0.0f, 1.0f);
p += 4 * (w * (int(v * (h - 3)) + 1) + x + 1);
p += 4 * (w * (int(v * (h - 3)) + 1) + x + 1);
scope_put_pixel(wtable, p);
p += 4 * w;
scope_put_pixel(wtable, p);
scope_put_pixel(wtable, p);
p += 4 * w;
scope_put_pixel(wtable, p);
if (last_p != nullptr) {
wform_put_line(w, last_p, p);
if (last_p != nullptr) {
wform_put_line(w, last_p, p);
}
last_p = p;
}
last_p = p;
}
}
});
wform_put_border(tgt, w, h);
@@ -219,8 +248,11 @@ ImBuf *make_waveform_view_from_ibuf(ImBuf *ibuf)
static ImBuf *make_sep_waveform_view_from_ibuf_byte(ImBuf *ibuf)
{
#ifdef DEBUG_TIME
SCOPED_TIMER_AVERAGED(__func__);
#endif
using namespace blender;
ImBuf *rval = IMB_allocImBuf(ibuf->x + 3, 515, 32, IB_rect);
int x, y;
const uchar *src = ibuf->byte_buffer.data;
uchar *tgt = rval->byte_buffer.data;
int w = ibuf->x + 3;
@@ -231,31 +263,34 @@ static ImBuf *make_sep_waveform_view_from_ibuf_byte(ImBuf *ibuf)
wform_put_grid(tgt, w, h);
for (x = 0; x < 256; x++) {
for (int x = 0; x < 256; x++) {
wtable[x] = uchar(pow((float(x) + 1.0f) / 256.0f, waveform_gamma) * 255.0f);
}
for (y = 0; y < ibuf->y; y++) {
uchar *last_p[3] = {nullptr, nullptr, nullptr};
/* Parallel over x, since each column is easily independent from others. */
threading::parallel_for(IndexRange(ibuf->x), 16, [&](IndexRange x_range) {
for (int y = 0; y < ibuf->y; y++) {
uchar *last_p[3] = {nullptr, nullptr, nullptr};
for (x = 0; x < ibuf->x; x++) {
int c;
const uchar *rgb = src + 4 * (ibuf->x * y + x);
for (c = 0; c < 3; c++) {
uchar *p = tgt;
p += 4 * (w * ((rgb[c] * (h - 3)) / 255 + 1) + c * sw + x / 3 + 1);
for (const int x : x_range) {
int c;
const uchar *rgb = src + 4 * (ibuf->x * y + x);
for (c = 0; c < 3; c++) {
uchar *p = tgt;
p += 4 * (w * (rgb[c] * 2 + 1) + c * sw + x / 3 + 1);
scope_put_pixel_single(wtable, p, c);
p += 4 * w;
scope_put_pixel_single(wtable, p, c);
scope_put_pixel_single(wtable, p, c);
p += 4 * w;
scope_put_pixel_single(wtable, p, c);
if (last_p[c] != nullptr) {
wform_put_line_single(w, last_p[c], p, c);
if (last_p[c] != nullptr) {
wform_put_line_single(w, last_p[c], p, c);
}
last_p[c] = p;
}
last_p[c] = p;
}
}
}
});
wform_put_border(tgt, w, h);
@@ -264,8 +299,11 @@ static ImBuf *make_sep_waveform_view_from_ibuf_byte(ImBuf *ibuf)
static ImBuf *make_sep_waveform_view_from_ibuf_float(ImBuf *ibuf)
{
#ifdef DEBUG_TIME
SCOPED_TIMER_AVERAGED(__func__);
#endif
using namespace blender;
ImBuf *rval = IMB_allocImBuf(ibuf->x + 3, 515, 32, IB_rect);
int x, y;
const float *src = ibuf->float_buffer.data;
uchar *tgt = rval->byte_buffer.data;
int w = ibuf->x + 3;
@@ -276,35 +314,38 @@ static ImBuf *make_sep_waveform_view_from_ibuf_float(ImBuf *ibuf)
wform_put_grid(tgt, w, h);
for (x = 0; x < 256; x++) {
for (int x = 0; x < 256; x++) {
wtable[x] = uchar(pow((float(x) + 1.0f) / 256.0f, waveform_gamma) * 255.0f);
}
for (y = 0; y < ibuf->y; y++) {
uchar *last_p[3] = {nullptr, nullptr, nullptr};
/* Parallel over x, since each column is easily independent from others. */
threading::parallel_for(IndexRange(ibuf->x), 16, [&](IndexRange x_range) {
for (int y = 0; y < ibuf->y; y++) {
uchar *last_p[3] = {nullptr, nullptr, nullptr};
for (x = 0; x < ibuf->x; x++) {
int c;
const float *rgb = src + 4 * (ibuf->x * y + x);
for (c = 0; c < 3; c++) {
uchar *p = tgt;
float v = rgb[c];
for (const int x : x_range) {
int c;
const float *rgb = src + 4 * (ibuf->x * y + x);
for (c = 0; c < 3; c++) {
uchar *p = tgt;
float v = rgb[c];
CLAMP(v, 0.0f, 1.0f);
CLAMP(v, 0.0f, 1.0f);
p += 4 * (w * (int(v * (h - 3)) + 1) + c * sw + x / 3 + 1);
p += 4 * (w * (int(v * (h - 3)) + 1) + c * sw + x / 3 + 1);
scope_put_pixel_single(wtable, p, c);
p += 4 * w;
scope_put_pixel_single(wtable, p, c);
scope_put_pixel_single(wtable, p, c);
p += 4 * w;
scope_put_pixel_single(wtable, p, c);
if (last_p[c] != nullptr) {
wform_put_line_single(w, last_p[c], p, c);
if (last_p[c] != nullptr) {
wform_put_line_single(w, last_p[c], p, c);
}
last_p[c] = p;
}
last_p[c] = p;
}
}
}
});
wform_put_border(tgt, w, h);