[Bf-blender-cvs] [9b89071] master: Multi-thread generated image creation
Sergey Sharybin
noreply at git.blender.org
Thu May 5 23:47:28 CEST 2016
Commit: 9b89071c9cdd2a341ba16db44cd4625303fb6e2f
Author: Sergey Sharybin
Date: Thu May 5 23:33:40 2016 +0200
Branches: master
https://developer.blender.org/rB9b89071c9cdd2a341ba16db44cd4625303fb6e2f
Multi-thread generated image creation
Gives about 2x speedup on laptop when creating new hires generated image,
regardless of it's type (color, color grid, uv grid).
===================================================================
M source/blender/blenkernel/intern/image_gen.c
M source/blender/imbuf/IMB_imbuf.h
M source/blender/imbuf/intern/divers.c
===================================================================
diff --git a/source/blender/blenkernel/intern/image_gen.c b/source/blender/blenkernel/intern/image_gen.c
index 650ccd4..2c8399a 100644
--- a/source/blender/blenkernel/intern/image_gen.c
+++ b/source/blender/blenkernel/intern/image_gen.c
@@ -38,7 +38,17 @@
#include "BLF_api.h"
-void BKE_image_buf_fill_color(unsigned char *rect, float *rect_float, int width, int height, const float color[4])
+typedef struct FillColorThreadData {
+ unsigned char *rect;
+ float *rect_float;
+ int width;
+ float color[4];
+} FillColorThreadData;
+
+static void image_buf_fill_color_slice(unsigned char *rect,
+ float *rect_float,
+ int width, int height,
+ const float color[4])
{
int x, y;
@@ -53,15 +63,12 @@ void BKE_image_buf_fill_color(unsigned char *rect, float *rect_float, int width,
}
}
}
-
+
if (rect) {
unsigned char ccol[4];
-
rgba_float_to_uchar(ccol, color);
-
for (y = 0; y < height; y++) {
for (x = 0; x < width; x++) {
-
rect[0] = ccol[0];
rect[1] = ccol[1];
rect[2] = ccol[2];
@@ -72,19 +79,55 @@ void BKE_image_buf_fill_color(unsigned char *rect, float *rect_float, int width,
}
}
+static void image_buf_fill_color_thread_do(void *data_v,
+ int start_scanline,
+ int num_scanlines)
+{
+ FillColorThreadData *data = (FillColorThreadData *)data_v;
+ size_t offset = ((size_t)start_scanline) * data->width * 4;
+ unsigned char *rect = (data->rect != NULL) ? (data->rect + offset) : NULL;
+ float *rect_float = (data->rect_float != NULL) ? (data->rect_float + offset) : NULL;
+ image_buf_fill_color_slice(rect,
+ rect_float,
+ data->width,
+ num_scanlines,
+ data->color);
+}
-void BKE_image_buf_fill_checker(unsigned char *rect, float *rect_float, int width, int height)
+void BKE_image_buf_fill_color(unsigned char *rect,
+ float *rect_float,
+ int width, int height,
+ const float color[4])
{
- /* these two passes could be combined into one, but it's more readable and
+ if (((size_t)width) * height < 64 * 64) {
+ image_buf_fill_color_slice(rect, rect_float, width, height, color);
+ }
+ else {
+ FillColorThreadData data;
+ data.rect = rect;
+ data.rect_float = rect_float;
+ data.width = width;
+ copy_v4_v4(data.color, color);
+ IMB_processor_apply_threaded_scanlines(
+ height, image_buf_fill_color_thread_do, &data);
+ }
+}
+
+static void image_buf_fill_checker_slice(unsigned char *rect,
+ float *rect_float,
+ int width, int height,
+ int offset)
+{
+ /* these two passes could be combined into one, but it's more readable and
* easy to tweak like this, speed isn't really that much of an issue in this situation... */
-
+
int checkerwidth = 32, dark = 1;
int x, y;
unsigned char *rect_orig = rect;
float *rect_float_orig = rect_float;
-
+
float h = 0.0, hoffs = 0.0;
float hsv[3] = {0.0f, 0.9f, 0.9f};
float rgb[3];
@@ -96,12 +139,12 @@ void BKE_image_buf_fill_checker(unsigned char *rect, float *rect_float, int widt
}
/* checkers */
- for (y = 0; y < height; y++) {
+ for (y = offset; y < height + offset; y++) {
dark = powf(-1.0f, floorf(y / checkerwidth));
-
+
for (x = 0; x < width; x++) {
if (x % checkerwidth == 0) dark = -dark;
-
+
if (rect_float) {
if (dark > 0) {
rect_float[0] = rect_float[1] = rect_float[2] = dark_linear_color;
@@ -131,12 +174,12 @@ void BKE_image_buf_fill_checker(unsigned char *rect, float *rect_float, int widt
rect_float = rect_float_orig;
/* 2nd pass, colored + */
- for (y = 0; y < height; y++) {
+ for (y = offset; y < height + offset; y++) {
hoffs = 0.125f * floorf(y / checkerwidth);
-
+
for (x = 0; x < width; x++) {
h = 0.125f * floorf(x / checkerwidth);
-
+
if ((abs((x % checkerwidth) - (checkerwidth / 2)) < 4) &&
(abs((y % checkerwidth) - (checkerwidth / 2)) < 4))
{
@@ -145,14 +188,14 @@ void BKE_image_buf_fill_checker(unsigned char *rect, float *rect_float, int widt
{
hsv[0] = fmodf(fabsf(h - hoffs), 1.0f);
hsv_to_rgb_v(hsv, rgb);
-
+
if (rect) {
rect[0] = (char)(rgb[0] * 255.0f);
rect[1] = (char)(rgb[1] * 255.0f);
rect[2] = (char)(rgb[2] * 255.0f);
rect[3] = 255;
}
-
+
if (rect_float) {
srgb_to_linearrgb_v3_v3(rect_float, rgb);
rect_float[3] = 1.0f;
@@ -166,13 +209,55 @@ void BKE_image_buf_fill_checker(unsigned char *rect, float *rect_float, int widt
}
}
+typedef struct FillCheckerThreadData {
+ unsigned char *rect;
+ float *rect_float;
+ int width;
+} FillCheckerThreadData;
+
+static void image_buf_fill_checker_thread_do(void *data_v,
+ int start_scanline,
+ int num_scanlines)
+{
+ FillCheckerThreadData *data = (FillCheckerThreadData *)data_v;
+ size_t offset = ((size_t)start_scanline) * data->width * 4;
+ unsigned char *rect = (data->rect != NULL) ? (data->rect + offset) : NULL;
+ float *rect_float = (data->rect_float != NULL) ? (data->rect_float + offset) : NULL;
+ image_buf_fill_checker_slice(rect,
+ rect_float,
+ data->width,
+ num_scanlines,
+ start_scanline);
+}
+
+void BKE_image_buf_fill_checker(unsigned char *rect,
+ float *rect_float,
+ int width, int height)
+{
+ if (((size_t)width) * height < 64 * 64) {
+ image_buf_fill_checker_slice(rect, rect_float, width, height, 0);
+ }
+ else {
+ FillCheckerThreadData data;
+ data.rect = rect;
+ data.rect_float = rect_float;
+ data.width = width;
+ IMB_processor_apply_threaded_scanlines(
+ height, image_buf_fill_checker_thread_do, &data);
+ }
+}
/* Utility functions for BKE_image_buf_fill_checker_color */
#define BLEND_FLOAT(real, add) (real + add <= 1.0f) ? (real + add) : 1.0f
#define BLEND_CHAR(real, add) ((real + (char)(add * 255.0f)) <= 255) ? (real + (char)(add * 255.0f)) : 255
-static void checker_board_color_fill(unsigned char *rect, float *rect_float, int width, int height)
+static void checker_board_color_fill(unsigned char *rect,
+ float *rect_float,
+ int width,
+ int height,
+ int offset,
+ int total_height)
{
int hue_step, y, x;
float hsv[3], rgb[3];
@@ -182,9 +267,9 @@ static void checker_board_color_fill(unsigned char *rect, float *rect_float, int
hue_step = power_of_2_max_i(width / 8);
if (hue_step < 8) hue_step = 8;
- for (y = 0; y < height; y++) {
+ for (y = offset; y < height + offset; y++) {
- hsv[2] = 0.1 + (y * (0.4 / height)); /* use a number lower then 1.0 else its too bright */
+ hsv[2] = 0.1 + (y * (0.4 / total_height)); /* use a number lower then 1.0 else its too bright */
for (x = 0; x < width; x++) {
hsv[0] = (float)((double)(x / hue_step) * 1.0 / width * hue_step);
hsv_to_rgb_v(hsv, rgb);
@@ -194,7 +279,7 @@ static void checker_board_color_fill(unsigned char *rect, float *rect_float, int
rect[1] = (char)(rgb[1] * 255.0f);
rect[2] = (char)(rgb[2] * 255.0f);
rect[3] = 255;
-
+
rect += 4;
}
@@ -203,27 +288,35 @@ static void checker_board_color_fill(unsigned char *rect, float *rect_float, int
rect_float[1] = rgb[1];
rect_float[2] = rgb[2];
rect_float[3] = 1.0f;
-
+
rect_float += 4;
}
}
}
}
-static void checker_board_color_tint(unsigned char *rect, float *rect_float, int width, int height, int size, float blend)
+static void checker_board_color_tint(unsigned char *rect,
+ float *rect_float,
+ int width,
+ int height,
+ int size,
+ float blend,
+ int offset)
{
int x, y;
float blend_half = blend * 0.5f;
- for (y = 0; y < height; y++) {
+ for (y = offset; y < height + offset; y++) {
for (x = 0; x < width; x++) {
- if (((y / size) % 2 == 1 && (x / size) % 2 == 1) || ( (y / size) % 2 == 0 && (x / size) % 2 == 0)) {
+ if (((y / size) % 2 == 1 && (x / size) % 2 == 1) ||
+ ((y / size) % 2 == 0 && (x / size) % 2 == 0))
+ {
if (rect) {
rect[0] = (char)BLEND_CHAR(rect[0], blend);
rect[1] = (char)BLEND_CHAR(rect[1], blend);
rect[2] = (char)BLEND_CHAR(rect[2], blend);
rect[3] = 255;
-
+
rect += 4;
}
if (rect_float) {
@@ -231,7 +324,7 @@ static void checker_board_color_tint(unsigned char *rect, float *rect_float, int
rect_float[1] = BLEND_FLOAT(rect_float[1], blend);
rect_float[2] = BLEND_FLOAT(rect_float[2], blend);
rect_float[3] = 1.0f;
-
+
rect_float += 4;
}
}
@@ -241,7 +334,7 @@ static void checker_board_color_tint(unsigned char *rect, float *rect_float, int
rect[1] = (char)BLEND_CHAR(rect[1], blend_half);
rect[2] = (char)BLEND_CHAR(rect[2], blend_half);
rect[3] = 255;
-
+
rect += 4;
}
if (rect_float) {
@@ -249,19 +342,24 @@ static void checker_board_color_tint(unsigned char *rect, float *rect_float, int
rect_float[1] = BLEND_FLOAT(rect_float[1], blend_half);
rect_float[2] = BLEND_FLOAT(rect_float[2], blend_half);
rect_float[3] = 1.0f;
-
+
rect_float += 4;
}
}
-
+
}
}
}
-static void checker_board_grid_fill(unsigned char *rect, float *rect_float, int width, int height, float blend)
+static void checker_board_grid_fill(unsigned char *rect,
+ f
@@ Diff output truncated at 10240 characters. @@
More information about the Bf-blender-cvs
mailing list