- Inkscape Devel - Inkscape Mailman

Multithreaded blur using OpenMP
by Jasper van de Gronde 25 Nov '08

25 Nov '08

Using GCC's support of OpenMP (now also in MingW!) I had a stab at making the outer loop of blurring parallel (the loops are typical DSP loops and shouldn't even throw an exception, so it should be perfectly safe). On my dual-core system blurring (a large image) is approximately twice as fast, so it seems to scale really well. As I don't have access to a Linux system at the moment I haven't modified the Makefiles to add support, but it shouldn't be too difficult (see the changes to build.xml). Also, I've made it write a log to be able to analyze the performance, but it uses Win32 specific functions for timing. So, if anyone would like to have a go at making it work on Linux and/or testing it that would be great :) The number of threads defaults to whatever omp_get_num_procs() returns, but on my system this doesn't work... So to use it, set the number of threads in the preferences, using: <group id="threading" numthreads="1" /> in the "options" group (so it's /options/threading/numthreads). Obviously this is not the most efficient way to do this, as new threads are created for each blur. But so far the overhead doesn't appear to be too bad, compared with the actual blurring operation. And using OpenMP the impact on the code is minimal, I only had to put two #pragma's before the blur loops, allocate the temporary data it needs for all threads and add some preference for the number of threads. So if this works out it might provide a relatively easy way to start making use of all these dual- and quad-core machines people have lying around doing almost nothing these days :) Patch attached, if wanted I can also provide a binary for Windows. This version writes a log file (blurlog.txt) that lists how long each blur takes. Note that inkscape also needs mingwm10.dll to run with this patch. As it isn't included with devlibs at the moment you'll have to copy it to Inkscape's directory manually (it's in the MingW bin directory). Index: build.xml =================================================================== --- build.xml (revision 20219) +++ build.xml (working copy) @@ -375,6 +375,7 @@ -Wall -Wformat -Werror=format-security -W -Wpointer-arith -Wcast-align -Wsign-compare -Woverloaded-virtual -Wswitch -O2 -mms-bitfields + -fopenmp </flags> <defines> -DVERSION=\"${version}\" @@ -476,6 +477,7 @@ objcopycommand="${archutil}objcopy"> <flags> -mwindows + -mthreads </flags> <fileset dir="${build}"> <include name="inkres.o"/> @@ -501,6 +503,7 @@ -lpng -ljpeg.dll -ltiff.dll -lpopt ${devlibs}/lib/zdll.lib -lgc -lws2_32 -lintl -lgdi32 -lcomdlg32 -lm + -lgomp -lpthreadGC2 </libs> </link> </target> @@ -530,6 +533,7 @@ objcopycommand="${archutil}objcopy"> <flags> -mwindows + -mthreads </flags> <fileset dir="${build}"> <include name="inkviewres.o"/> @@ -553,6 +557,7 @@ -lpng -ljpeg.dll -ltiff.dll -lpopt ${devlibs}/lib/zdll.lib -lgc -lws2_32 -lintl -lgdi32 -lcomdlg32 -lm + -lgomp -lpthreadGC2 </libs> </link> </target> @@ -572,6 +577,7 @@ stripcommand="${archutil}strip" objcopycommand="${archutil}objcopy"> <flags> + -mthreads </flags> <fileset dir="${build}"> <include name="obj/test-main.o"/> @@ -602,6 +608,7 @@ -lpng -ljpeg.dll -ltiff.dll -lpopt ${devlibs}/lib/zdll.lib -lgc -lws2_32 -lintl -lgdi32 -lcomdlg32 -lm + -lgomp -lpthreadGC2 </libs> </link> </target> @@ -667,6 +674,7 @@ <copy todir="${dist}" file="${devlibs}/bin/libpopt-0.dll"/> <copy todir="${dist}" file="${devlibs}/bin/liblcms-1.dll"/> <copy todir="${dist}" file="${devlibs}/bin/intl.dll"/> + <copy todir="${dist}" file="${devlibs}/bin/pthreadGC2.dll"/> <copy file="${devlibs}/bin/intl.dll" tofile="${dist}/libintl-2.dll"/>  Index: src/display/nr-filter-gaussian.cpp =================================================================== --- src/display/nr-filter-gaussian.cpp (revision 20219) +++ src/display/nr-filter-gaussian.cpp (working copy) @@ -16,9 +16,11 @@ #include <algorithm> #include <cmath> #include <complex> +#include <cstdlib> #include <glib.h> -#include <cstdlib> #include <limits> +#include <omp.h> +#include <windows.h> // For performance logging #include "2geom/isnan.h" @@ -268,9 +270,11 @@ filter2D_IIR(PT *const dest, int const dstr1, int const dstr2, PT const *const src, int const sstr1, int const sstr2, int const n1, int const n2, IIRValue const b[N+1], double const M[N*N], - IIRValue *const tmpdata) + IIRValue *const tmpdata[], int const num_threads) { +#pragma omp parallel for num_threads(num_threads) for ( int c2 = 0 ; c2 < n2 ; c2++ ) { + unsigned int tid = omp_get_thread_num(); // corresponding line in the source and output buffer PT const * srcimg = src + c2*sstr2; PT * dstimg = dest + c2*dstr2 + n1*dstr1; @@ -288,7 +292,7 @@ for(unsigned int i=1; i<N+1; i++) { for(unsigned int c=0; c<PC; c++) u[0][c] += u[i][c]*b[i]; } - copy_n(u[0], PC, tmpdata+c1*PC); + copy_n(u[0], PC, tmpdata[tid]+c1*PC); } // Backward pass IIRValue v[N+1][PC]; @@ -303,7 +307,7 @@ int c1=n1-1; while(c1-->0) { for(unsigned int i=N; i>0; i--) copy_n(v[i-1], PC, v[i]); - copy_n(tmpdata+c1*PC, PC, v[0]); + copy_n(tmpdata[tid]+c1*PC, PC, v[0]); for(unsigned int c=0; c<PC; c++) v[0][c] *= b[0]; for(unsigned int i=1; i<N+1; i++) { for(unsigned int c=0; c<PC; c++) v[0][c] += v[i][c]*b[i]; @@ -326,11 +330,12 @@ static void filter2D_FIR(PT *const dst, int const dstr1, int const dstr2, PT const *const src, int const sstr1, int const sstr2, - int const n1, int const n2, FIRValue const *const kernel, int const scr_len) + int const n1, int const n2, FIRValue const *const kernel, int const scr_len, int const num_threads) { // Past pixels seen (to enable in-place operation) PT history[scr_len+1][PC]; +#pragma omp parallel for num_threads(num_threads) private(history) for ( int c2 = 0 ; c2 < n2 ; c2++ ) { // corresponding line in the source buffer @@ -539,13 +544,14 @@ } // Some common constants + Inkscape::Preferences *prefs = Inkscape::Preferences::get(); int const width_org = in->area.x1-in->area.x0, height_org = in->area.y1-in->area.y0; double const deviation_x_org = _deviation_x * NR::expansionX(trans); double const deviation_y_org = _deviation_y * NR::expansionY(trans); int const PC = NR_PIXBLOCK_BPP(in); + int const NTHREADS = std::max(1,std::min(8,prefs->getInt("/options/threading/numthreads",omp_get_num_procs()))); // Subsampling constants - Inkscape::Preferences *prefs = Inkscape::Preferences::get(); int const quality = prefs->getInt("/options/blurquality/value"); int const x_step_l2 = _effect_subsample_step_log2(deviation_x_org, quality); int const y_step_l2 = _effect_subsample_step_log2(deviation_y_org, quality); @@ -567,6 +573,11 @@ bool const use_IIR_x = deviation_x > 3; bool const use_IIR_y = deviation_y > 3; + // Temporary performance logging + LARGE_INTEGER startTime, endTime, timeFrequency; + QueryPerformanceFrequency(&timeFrequency); + QueryPerformanceCounter(&startTime); + // new buffer for the subsampled output NRPixBlock *out = new NRPixBlock; nr_pixblock_setup_fast(out, in->mode, in->area.x0/x_step, in->area.y0/y_step, @@ -577,13 +588,19 @@ } // Temporary storage for IIR filter // NOTE: This can be eliminated, but it reduces the precision a bit - IIRValue * tmpdata = 0; + IIRValue * tmpdata[NTHREADS]; + std::fill_n(tmpdata, NTHREADS, (IIRValue*)0); if ( use_IIR_x || use_IIR_y ) { - tmpdata = new IIRValue[std::max(width,height)*PC]; - if (tmpdata == NULL) { - nr_pixblock_release(out); - delete out; - return 0; + for(int i=0; i<NTHREADS; i++) { + tmpdata[i] = new IIRValue[std::max(width,height)*PC]; + if (tmpdata[i] == NULL) { + nr_pixblock_release(out); + while(i-->0) { + delete[] tmpdata[i]; + } + delete out; + return 0; + } } } NRPixBlock *ssin = in; @@ -629,16 +646,16 @@ // Filter (x) switch(in->mode) { case NR_PIXBLOCK_MODE_A8: ///< Grayscale - filter2D_IIR<unsigned char,1,false>(NR_PIXBLOCK_PX(out), 1, out->rs, NR_PIXBLOCK_PX(ssin), 1, ssin->rs, width, height, b, M, tmpdata); + filter2D_IIR<unsigned char,1,false>(NR_PIXBLOCK_PX(out), 1, out->rs, NR_PIXBLOCK_PX(ssin), 1, ssin->rs, width, height, b, M, tmpdata, NTHREADS); break; case NR_PIXBLOCK_MODE_R8G8B8: ///< 8 bit RGB - filter2D_IIR<unsigned char,3,false>(NR_PIXBLOCK_PX(out), 3, out->rs, NR_PIXBLOCK_PX(ssin), 3, ssin->rs, width, height, b, M, tmpdata); + filter2D_IIR<unsigned char,3,false>(NR_PIXBLOCK_PX(out), 3, out->rs, NR_PIXBLOCK_PX(ssin), 3, ssin->rs, width, height, b, M, tmpdata, NTHREADS); break; case NR_PIXBLOCK_MODE_R8G8B8A8N: ///< Normal 8 bit RGBA - filter2D_IIR<unsigned char,4,false>(NR_PIXBLOCK_PX(out), 4, out->rs, NR_PIXBLOCK_PX(ssin), 4, ssin->rs, width, height, b, M, tmpdata); + filter2D_IIR<unsigned char,4,false>(NR_PIXBLOCK_PX(out), 4, out->rs, NR_PIXBLOCK_PX(ssin), 4, ssin->rs, width, height, b, M, tmpdata, NTHREADS); break; case NR_PIXBLOCK_MODE_R8G8B8A8P: ///< Premultiplied 8 bit RGBA - filter2D_IIR<unsigned char,4,true >(NR_PIXBLOCK_PX(out), 4, out->rs, NR_PIXBLOCK_PX(ssin), 4, ssin->rs, width, height, b, M, tmpdata); + filter2D_IIR<unsigned char,4,true >(NR_PIXBLOCK_PX(out), 4, out->rs, NR_PIXBLOCK_PX(ssin), 4, ssin->rs, width, height, b, M, tmpdata, NTHREADS); break; default: assert(false); @@ -651,16 +668,16 @@ // Filter (x) switch(in->mode) { case NR_PIXBLOCK_MODE_A8: ///< Grayscale - filter2D_FIR<unsigned char,1>(NR_PIXBLOCK_PX(out), 1, out->rs, NR_PIXBLOCK_PX(ssin), 1, ssin->rs, width, height, kernel, scr_len_x); + filter2D_FIR<unsigned char,1>(NR_PIXBLOCK_PX(out), 1, out->rs, NR_PIXBLOCK_PX(ssin), 1, ssin->rs, width, height, kernel, scr_len_x, NTHREADS); break; case NR_PIXBLOCK_MODE_R8G8B8: ///< 8 bit RGB - filter2D_FIR<unsigned char,3>(NR_PIXBLOCK_PX(out), 3, out->rs, NR_PIXBLOCK_PX(ssin), 3, ssin->rs, width, height, kernel, scr_len_x); + filter2D_FIR<unsigned char,3>(NR_PIXBLOCK_PX(out), 3, out->rs, NR_PIXBLOCK_PX(ssin), 3, ssin->rs, width, height, kernel, scr_len_x, NTHREADS); break; case NR_PIXBLOCK_MODE_R8G8B8A8N: ///< Normal 8 bit RGBA - filter2D_FIR<unsigned char,4>(NR_PIXBLOCK_PX(out), 4, out->rs, NR_PIXBLOCK_PX(ssin), 4, ssin->rs, width, height, kernel, scr_len_x); + filter2D_FIR<unsigned char,4>(NR_PIXBLOCK_PX(out), 4, out->rs, NR_PIXBLOCK_PX(ssin), 4, ssin->rs, width, height, kernel, scr_len_x, NTHREADS); break; case NR_PIXBLOCK_MODE_R8G8B8A8P: ///< Premultiplied 8 bit RGBA - filter2D_FIR<unsigned char,4>(NR_PIXBLOCK_PX(out), 4, out->rs, NR_PIXBLOCK_PX(ssin), 4, ssin->rs, width, height, kernel, scr_len_x); + filter2D_FIR<unsigned char,4>(NR_PIXBLOCK_PX(out), 4, out->rs, NR_PIXBLOCK_PX(ssin), 4, ssin->rs, width, height, kernel, scr_len_x, NTHREADS); break; default: assert(false); @@ -688,16 +705,16 @@ // Filter (y) switch(in->mode) { case NR_PIXBLOCK_MODE_A8: ///< Grayscale - filter2D_IIR<unsigned char,1,false>(NR_PIXBLOCK_PX(out), out->rs, 1, NR_PIXBLOCK_PX(out), out->rs, 1, height, width, b, M, tmpdata); + filter2D_IIR<unsigned char,1,false>(NR_PIXBLOCK_PX(out), out->rs, 1, NR_PIXBLOCK_PX(out), out->rs, 1, height, width, b, M, tmpdata, NTHREADS); break; case NR_PIXBLOCK_MODE_R8G8B8: ///< 8 bit RGB - filter2D_IIR<unsigned char,3,false>(NR_PIXBLOCK_PX(out), out->rs, 3, NR_PIXBLOCK_PX(out), out->rs, 3, height, width, b, M, tmpdata); + filter2D_IIR<unsigned char,3,false>(NR_PIXBLOCK_PX(out), out->rs, 3, NR_PIXBLOCK_PX(out), out->rs, 3, height, width, b, M, tmpdata, NTHREADS); break; case NR_PIXBLOCK_MODE_R8G8B8A8N: ///< Normal 8 bit RGBA - filter2D_IIR<unsigned char,4,false>(NR_PIXBLOCK_PX(out), out->rs, 4, NR_PIXBLOCK_PX(out), out->rs, 4, height, width, b, M, tmpdata); + filter2D_IIR<unsigned char,4,false>(NR_PIXBLOCK_PX(out), out->rs, 4, NR_PIXBLOCK_PX(out), out->rs, 4, height, width, b, M, tmpdata, NTHREADS); break; case NR_PIXBLOCK_MODE_R8G8B8A8P: ///< Premultiplied 8 bit RGBA - filter2D_IIR<unsigned char,4,true >(NR_PIXBLOCK_PX(out), out->rs, 4, NR_PIXBLOCK_PX(out), out->rs, 4, height, width, b, M, tmpdata); + filter2D_IIR<unsigned char,4,true >(NR_PIXBLOCK_PX(out), out->rs, 4, NR_PIXBLOCK_PX(out), out->rs, 4, height, width, b, M, tmpdata, NTHREADS); break; default: assert(false); @@ -710,24 +727,32 @@ // Filter (y) switch(in->mode) { case NR_PIXBLOCK_MODE_A8: ///< Grayscale - filter2D_FIR<unsigned char,1>(NR_PIXBLOCK_PX(out), out->rs, 1, NR_PIXBLOCK_PX(out), out->rs, 1, height, width, kernel, scr_len_y); + filter2D_FIR<unsigned char,1>(NR_PIXBLOCK_PX(out), out->rs, 1, NR_PIXBLOCK_PX(out), out->rs, 1, height, width, kernel, scr_len_y, NTHREADS); break; case NR_PIXBLOCK_MODE_R8G8B8: ///< 8 bit RGB - filter2D_FIR<unsigned char,3>(NR_PIXBLOCK_PX(out), out->rs, 3, NR_PIXBLOCK_PX(out), out->rs, 3, height, width, kernel, scr_len_y); + filter2D_FIR<unsigned char,3>(NR_PIXBLOCK_PX(out), out->rs, 3, NR_PIXBLOCK_PX(out), out->rs, 3, height, width, kernel, scr_len_y, NTHREADS); break; case NR_PIXBLOCK_MODE_R8G8B8A8N: ///< Normal 8 bit RGBA - filter2D_FIR<unsigned char,4>(NR_PIXBLOCK_PX(out), out->rs, 4, NR_PIXBLOCK_PX(out), out->rs, 4, height, width, kernel, scr_len_y); + filter2D_FIR<unsigned char,4>(NR_PIXBLOCK_PX(out), out->rs, 4, NR_PIXBLOCK_PX(out), out->rs, 4, height, width, kernel, scr_len_y, NTHREADS); break; case NR_PIXBLOCK_MODE_R8G8B8A8P: ///< Premultiplied 8 bit RGBA - filter2D_FIR<unsigned char,4>(NR_PIXBLOCK_PX(out), out->rs, 4, NR_PIXBLOCK_PX(out), out->rs, 4, height, width, kernel, scr_len_y); + filter2D_FIR<unsigned char,4>(NR_PIXBLOCK_PX(out), out->rs, 4, NR_PIXBLOCK_PX(out), out->rs, 4, height, width, kernel, scr_len_y, NTHREADS); break; default: assert(false); }; } - delete[] tmpdata; // deleting a nullptr has no effect, so this is save + for(int i=0; i<NTHREADS; i++) { + delete[] tmpdata[i]; // deleting a nullptr has no effect, so this is safe + } + // Temporary performance logging + QueryPerformanceCounter(&endTime); + FILE* logfile = fopen("blurlog.txt", "at"); + fprintf(logfile, "image size: %dx%d, threads: %d, time: %.3gs\n", width, height, NTHREADS, static_cast<double>(endTime.QuadPart-startTime.QuadPart)/timeFrequency.QuadPart); + fclose(logfile); + if ( !resampling ) { // No upsampling needed out->empty = FALSE;

10 26

Re: [Inkscape-devel] build error
by Alexandre Prokoudine 25 Nov '08

25 Nov '08

2008/11/25 Ted Gould <ted@...11...>: >> $ ls -al filters/.deps/.dirstamp >> ls: cannot access filters/.deps/.dirstamp: No such file or directory > > Hmm, that's odd. But the file is really only there to mark when the > directory was last read. It would seem that: > > touch filters/.deps/.dirstamp > > Would fix the problem, but I have no idea why it happened. $ touch src/filters/.deps/.dirstamp touch: cannot touch `src/filters/.deps/.dirstamp': Permission denied $ ls -lh src/filters/.deps/ total 80K -rw-r--r-- 1 root root 8 Nov 25 16:36 blend.Po etc. All files have root access permission. I'm quite puzzled. These are really two different machines with two different checkouts, and the last successful build was surely after November 8, just like the one for the laptop at home. Alexandre

2 1

build error
by Alexandre Prokoudine 25 Nov '08

25 Nov '08

Hi, I'm seing this error on two different systems: rm -f extension/script/libscript.a ar cru extension/script/libscript.a extension/script/InkscapeScript.o ranlib extension/script/libscript.a /bin/sh: filters/.deps/.dirstamp: Access denied Anyone? :-) Alexandre

2 1

cut/crop marks extensions
by Alexandre Prokoudine 25 Nov '08

25 Nov '08

Hi, Recently a Perl extension called "Cut marks" was added to SVN. I must admit I didn't have very good feelings about this, because Windows users who are majority of our user base or at least very close to it will have an extra step of installing ActivePerl to be able to use it. IMO, having several programming languages supported is fine as long as extensions work everywhere out of box. We already have functionality available on one platform and not available on the other and it probably wouldn't be wise pushing this even further. And now here comes a similar extension written in Python: https://bugs.launchpad.net/inkscape/+bug/301776 So what do we do? Alexandre

1 0

lpe-hatches
by jf barraud 24 Nov '08

24 Nov '08

Hi all, I've just added a new lpe to fill a shape with rough hatches. It's still at an experimental state. As usual, I'm in trouble with the parameters. They are purely technical atm, i.e. they are the onesI used in the code to control the shape of the hatches. But what is technical-oriented is generally not user friendly (in particular, I was not able to supply tips for some of them). Please try it, and suggest user firendly parameters (from which the technical ones would be deduced). Ah, and yes, bending needs improvement, and growth has no effect atm... Anyway, I hope you'll like it ;-) jf.

7 13

Number in circle effect
by Mihaela 23 Nov '08

23 Nov '08

1 0

verbs and the undo stack
by Felipe Sanches 23 Nov '08

23 Nov '08

I am not so sure about the mechanics of the undo stack and the meaning of verbs. I have coded an improvement to the user interface for embedding color profiles in a document and I have used the sp_document_done call. The patch is attached. Could anybody review it? It adds a "remove" context menu option in the "Embedded color profiles" box that appears in the "Color Management" tab in "document settings" dialog. Undoing and redoing works ok, but I am not sure aobut the verbs. I expected it to call the update() method onde the user press CTRL+Z or SHIFT+CTRL+Z so that my widgets would be properly updated. Juca

2 3

[PATCH] JavaFX exporting
by Silveira Neto 21 Nov '08

21 Nov '08

Changelog: * Public attributes instead of private functions, so can externally use them. * Sanatize also '#' character. * Removed a compilation warning on sanatize(). Todo: * There's some wrong strings for i18n. * The attribute 'name' from the class JavaFXOutput. * Strange black stroke to be fixed. * Gradients are not been correctly applied.

1 1

Craft Robo can cut directly from inkscape?
by stuart 21 Nov '08

21 Nov '08

Hi... Been talking to Lucille across the pond and understand that people are cutting directly from inkscape to Craft Robo paper cutters using (I believe) the Craft Robo 5.1 software package. As I have not heard anything about this here (support for non standard Graphtec cutter protocols) I assume the Craft Robo 5.1 software package uses a Craft Robo USB driver that identifies and interprets the Post Script output (nuts, should have ask Lucille how inkscape was setup) of inkscape. Anyhow, if true, this would be great for Craft Robo users. However, I doubt that Wishblades (North American versions of Craft Robo) cutters will benefit. The old blue and new pink Wishblades identify them selves differently and have not (in the past) worked with the Craft Robo software. Doesn't seam fair as the old blue Wishblades are no longer supported by Xyron going forward into Windows Vista from what I understand. Yeah, I'm going to buy more stuff from Xyron, NOT! You can still cut from inkscape to a Wishblade but you will need to in Linux, have Python and Glade installed and use that Python script we've talked about here on this list (I'll have to look up the URL if people are interested). So, are there still people interested in porting that Python script to windows? I've tried, but haven't been able to figure out how to create a USB driver under Windows. From all that I have read, Windows is a real mess when it comes to creating USB drivers (i.e. you can find companies that will do it but they charge $1000s, ouch...)

1 0

Marker direction regression
by Jasper van de Gronde 18 Nov '08

18 Nov '08

Running the rendering tests this week I encountered a regression in how markers are oriented. If anyone touched this code lately, perhaps it's good to go back and have a look. The relevant bug report: https://bugs.launchpad.net/inkscape/+bug/299414 Overview of rendering tests: http://home.hccnet.nl/th.v.d.gronde/inkscape/ResultViewer.html

2 1