d3631c6 | Tobias Ellinghaus | 06 November 2016, 11:54:59 UTC | Don't remove RELEASE_NOTES from release tarballs | 06 November 2016, 11:54:59 UTC |
b0c0210 | Tobias Ellinghaus | 06 November 2016, 11:50:10 UTC | release notes: Add new/updated translations | 06 November 2016, 11:50:10 UTC |
c6bb04b | Roman Lebedev | 05 November 2016, 21:14:56 UTC | Update RELEASE NOTES - camera support | 05 November 2016, 21:14:56 UTC |
198e925 | Ulrich Pegelow | 04 November 2016, 19:33:58 UTC | usermanual: keybinding updates | 04 November 2016, 19:33:58 UTC |
588903a | Roman Lebedev | 03 November 2016, 17:01:23 UTC | Update RELEASE_NOTES | 03 November 2016, 17:01:23 UTC |
3e949ca | Roman Lebedev | 03 November 2016, 14:40:00 UTC | Re-deduplicate git shortlog (update .mailmap) | 03 November 2016, 14:40:00 UTC |
7f56e6f | Roman Lebedev | 03 November 2016, 13:33:25 UTC | Drop darktable-viewer tool. Use slideshow mode. acked by jo. | 03 November 2016, 13:38:24 UTC |
90a56dd | Roman Lebedev | 03 November 2016, 13:20:41 UTC | Update RELEASE_NOTES | 03 November 2016, 13:20:41 UTC |
f4c5785 | Roman Lebedev | 02 November 2016, 21:36:33 UTC | Noiseprofiles schema: deduplicate models. It is not documented what happens when there is more than one model entry. Current dt_iop_denoiseprofile_get_auto_profile() would use the non-skipped entries from the first entry. Fixed that by moving all the duplicate models (i.e. the ones that are located after the first one) into the first one, and prepending them with "skip": true | 02 November 2016, 21:38:09 UTC |
9763365 | Roman Lebedev | 02 November 2016, 21:18:57 UTC | Noiseprofiles: drop complete copy-paste duplicate "olympus e-m10 by polarfox" | 02 November 2016, 21:38:09 UTC |
a331326 | Roman Lebedev | 02 November 2016, 21:17:38 UTC | Noiseprofiles: drop complete copy-paste duplicate "additional data for the D5000 by jcsogo (red channel underexposed)" | 02 November 2016, 21:38:09 UTC |
b08a329 | Roman Lebedev | 02 November 2016, 21:16:12 UTC | Noiseprofiles: drop complete copy-paste duplicate "nikon d5000, contributed by Raphaël Monnard" | 02 November 2016, 21:38:09 UTC |
9ceddfe | Roman Lebedev | 02 November 2016, 21:14:50 UTC | Noiseprofiles schema: deduplicate makers. It is not documented what happens when there is more than one model entry. Current dt_iop_denoiseprofile_get_auto_profile() would use the non-skipped entries from the first entry. Also, better not have more than one maker entry, for same reasons. | 02 November 2016, 21:38:09 UTC |
62d8013 | Roman Lebedev | 02 November 2016, 19:57:22 UTC | Rawspeed: NIKON D5100: add tmp mode="14bit-uncompressed" entry. Refs #11268 At this time, i have no better idea how to handle this. I suppose, NefDecoder::getMode() needs to be changed, to not take into account the COMPRESSION tag, but use NEFIsUncompressed() heuristic. | 02 November 2016, 21:38:09 UTC |
b344113 | Ulrich Pegelow | 02 November 2016, 19:58:36 UTC | usermanual: mention presets in the more modules panel | 02 November 2016, 19:58:36 UTC |
2e35846 | Ulrich Pegelow | 02 November 2016, 19:07:48 UTC | usermanual: on comparative and range operators in the collect images panel | 02 November 2016, 19:07:48 UTC |
26eaec4 | Ulrich Pegelow | 02 November 2016, 17:35:34 UTC | lens: disable tiling in the lens correction module This is a quick fix for bug #11279. The underlying cause of that bug is a region of interest mismatch during tiling, thus affecting mostly file export with OpenCL and low to medium memory GPUs. A real fix would require a rework of the interplay of lens.c:modify_roi_in() and the corresponding tiling routines and could not be implemented without heavy testing. As we are shortly before a release the quick fix disables tiling in that module. This will give a certain slow-down on the mentioned low to medium memory GPUs but only during export which is tolerable. | 02 November 2016, 17:35:34 UTC |
119afd7 | Tobias Ellinghaus | 02 November 2016, 17:18:20 UTC | Show CII Best Practice progress https://bestpractices.coreinfrastructure.org/projects/470 | 02 November 2016, 17:18:20 UTC |
03d7adc | Tobias Ellinghaus | 02 November 2016, 16:10:33 UTC | Fix typo in release notes draft | 02 November 2016, 16:10:33 UTC |
82d7ce3 | Tobias Ellinghaus | 02 November 2016, 15:31:16 UTC | Start working on 220 release notes | 02 November 2016, 15:31:16 UTC |
3c9b03d | Roman Lebedev | 01 November 2016, 20:55:19 UTC | dt_imageio_open_rawspeed_sraw(): don't check getBpp(). It means nothing here, we only want to check the type of the buffer, not the actual original bitness of data. Seems to fix some sRaw's. | 01 November 2016, 20:55:19 UTC |
b42ddf6 | hanatos | 01 November 2016, 13:51:56 UTC | Merge pull request #1337 from schenlap/noise_profile Noise profile, better readable outputs | 01 November 2016, 13:51:56 UTC |
f101153 | Ulrich Pegelow | 01 November 2016, 06:48:45 UTC | usermanual: some more 2.2 updates (and docbook source beautify) | 01 November 2016, 06:48:45 UTC |
b55a4ed | Roman Lebedev | 31 October 2016, 13:15:50 UTC | Rawspeed imageio: add some more safety checks. | 31 October 2016, 13:15:50 UTC |
a6a6e8e | Roman Lebedev | 31 October 2016, 13:14:17 UTC | Rawspeed imageio: if HDR, set processed_maximum to 1.0; Fixes #11254 | 31 October 2016, 13:14:17 UTC |
15d23c5 | Steffen Binder | 30 October 2016, 19:18:33 UTC | Wb presets: add finetuned WB presets for Nikon D7100. Fixes #11277 | 31 October 2016, 12:38:54 UTC |
50ebe5a | Roman Lebedev | 30 October 2016, 17:50:25 UTC | Travis CI: check validity of usermanual. (#1309) Until we figure out how to make actually building usermanual PDF faster (and figure out uploading), this is good-enough. | 30 October 2016, 17:50:25 UTC |
e90e242 | Roman Lebedev | 30 October 2016, 17:08:40 UTC | Dockerfile: add docbook, docbook-xml packages. [skip ci] | 30 October 2016, 17:08:40 UTC |
a66657d | Tobias Ellinghaus | 30 October 2016, 15:17:42 UTC | Properly ellipsize bauhaus comboboxen | 30 October 2016, 15:17:42 UTC |
7d51e1c | Tobias Ellinghaus | 30 October 2016, 15:14:16 UTC | Add option to left align bauhaus combobox entries ... instead of using ugly inband signalling. Also fix a small memleak. | 30 October 2016, 15:14:16 UTC |
2086bce | Tobias Ellinghaus | 30 October 2016, 15:00:01 UTC | masks: don't add spaces in iop's masks dropdown It seems they are only there to differentiate between own masks and those from other IOPs, but as there is already a line of text saying that inbetween they are only visual clutter. | 30 October 2016, 15:00:01 UTC |
706985e | Tobias Ellinghaus | 30 October 2016, 11:59:29 UTC | Fix #11274: Sort user's color profiles by name This merges profiles from the home dir witht hose from the global config dir. Should that not be what we want it's easy to just add an intermediate sort + concat + NULLing step. | 30 October 2016, 11:59:29 UTC |
e112e8d | Roman Lebedev | 30 October 2016, 10:49:05 UTC | Color picker: cleanup. Reading spec helps :) There is no need to write separate paths for omp and seq computations, because OpenMP3.1 support if() clause in parallel Construct, which sets num of threads to 1. It still does run code parallel, but just by one thread. Also, let's try-out separate parallel and for. | 30 October 2016, 10:49:05 UTC |
dc26486 | Roman Lebedev | 30 October 2016, 10:15:29 UTC | Lens iop: modify_roi_in(): gcc-4.7 is required anyway. Also, let's try-out separate parallel and for. | 30 October 2016, 10:28:08 UTC |
c635a6f | Roman Lebedev | 30 October 2016, 09:22:54 UTC | rs-identify: openmp | 30 October 2016, 10:28:08 UTC |
5f98414 | Roman Lebedev | 30 October 2016, 08:58:44 UTC | rs-identify: print filename, no cpp(?). | 30 October 2016, 10:28:08 UTC |
7d93759 | Roman Lebedev | 30 October 2016, 08:27:23 UTC | Rawspeed: BitPumpPlain: TiffEntryBE: don't load type into enum first. Fixes following complaints: /home/lebedevri/darktable/src/external/rawspeed/RawSpeed/TiffEntryBE.cpp:39:7: runtime error: load of value 26624, which is not a valid value for type 'TiffDataType' /home/lebedevri/darktable/src/external/rawspeed/RawSpeed/TiffEntryBE.cpp:40:78: runtime error: load of value 26624, which is not a valid value for type 'TiffDataType' | 30 October 2016, 10:28:08 UTC |
dffa475 | Roman Lebedev | 30 October 2016, 08:23:50 UTC | Rawspeed: BitPumpPlain: TiffEntry: don't load type into enum first. Fixes following complaints: /home/lebedevri/darktable/src/external/rawspeed/RawSpeed/TiffEntryBE.cpp:39:7: runtime error: load of value 26624, which is not a valid value for type 'TiffDataType' /home/lebedevri/darktable/src/external/rawspeed/RawSpeed/TiffEntryBE.cpp:40:78: runtime error: load of value 26624, which is not a valid value for type 'TiffDataType' | 30 October 2016, 10:28:08 UTC |
b20c9c2 | Roman Lebedev | 30 October 2016, 08:08:39 UTC | Rawspeed: BitPumpPlain: getBits(): don't load uint32 from misaligned addresses. When loading ARW: /home/lebedevri/darktable/src/external/rawspeed/RawSpeed/BitPumpPlain.cpp:48:14: runtime error: load of misaligned address 0x7f27d5c3e811 for type 'uint32', which requires 4 byte alignment 0x7f27d5c3e811: note: pointer points here 00 00 00 c2 21 cd 93 c1 e2 b0 58 24 0c 0e 81 61 90 50 18 d0 23 1d 34 c4 e4 c1 88 08 0a 07 c2 41 ^ | 30 October 2016, 10:28:08 UTC |
7522acf | Roman Lebedev | 30 October 2016, 07:42:28 UTC | Rawspeed: BitPumpMSB: peekByteNoFill(): don't load uint32 from misaligned addresses. When loading NEF: /home/lebedevri/darktable/src/external/rawspeed/RawSpeed/BitPumpMSB.h:88:50: runtime error: load of misaligned address 0x7ffc93af4e4b for type 'uint32', which requires 4 byte alignment 0x7ffc93af4e4b: note: pointer points here d6 f2 ea fb ea 00 00 00 00 10 24 21 f3 45 7f 00 00 f4 e0 ff 00 60 55 00 00 0c 00 00 00 00 00 00 ^ | 30 October 2016, 10:28:07 UTC |
26427d1 | Roman Lebedev | 30 October 2016, 07:41:23 UTC | Rawspeed: BitPumpMSB: peekBitsNoFill(): don't load uint32 from misaligned addresses. When loading NEF: 0x7ffc93af4e49: note: pointer points here 5c 33 d6 f2 ea fb ea 00 00 00 00 10 24 21 f3 45 7f 00 00 f4 e0 ff 00 60 55 00 00 0c 00 00 00 00 ^ /home/lebedevri/darktable/src/external/rawspeed/RawSpeed/BitPumpMSB.h:53:51: runtime error: load of misaligned address 0x7ffc93af4e4a for type 'uint32', which requires 4 byte alignment 0x7ffc93af4e4a: note: pointer points here 33 d6 f2 ea fb ea 00 00 00 00 10 24 21 f3 45 7f 00 00 f4 e0 ff 00 60 55 00 00 0c 00 00 00 00 00 ^ | 30 October 2016, 10:28:06 UTC |
83ab18a | Roman Lebedev | 29 October 2016, 17:56:00 UTC | Pixelpipe hb: dt_dev_pixelpipe_process_rec(): check for NAN before demosaic too | 29 October 2016, 18:05:41 UTC |
614f59e | Roman Lebedev | 29 October 2016, 17:28:15 UTC | Rawspeed: BitPumpJPEG: peekBitsNoFill(): don't load uint32 from misaligned addresses. When loading CR2: /home/lebedevri/darktable/src/external/rawspeed/RawSpeed/BitPumpJPEG.h:53:51: runtime error: load of misaligned address 0x557522ca95aa for type 'uint32', which requires 4 byte alignment 0x557522ca95aa: note: pointer points here 87 7c fc cb 1d ff 00 00 00 00 2a ef 62 01 60 00 00 00 0d 00 00 00 00 00 00 00 e2 22 22 22 22 22 ^ /home/lebedevri/darktable/src/external/rawspeed/RawSpeed/BitPumpJPEG.h:53:51: runtime error: load of misaligned address 0x557522ca95a9 for type 'uint32', which requires 4 byte alignment 0x557522ca95a9: note: pointer points here bc 87 7c fc cb 1d ff 00 00 00 00 2a ef 62 01 57 00 00 00 0d 00 00 00 00 00 00 00 e2 22 22 22 22 ^ | 29 October 2016, 18:05:41 UTC |
c716d1d | johannes hanika | 29 October 2016, 17:40:06 UTC | noiseprofile: sony slt-a37 by steve | 29 October 2016, 17:40:43 UTC |
b3e2ad3 | Pascal Obry | 29 October 2016, 15:50:05 UTC | Round values for pipe_rect, avoid off-by-one issue. For #11271. | 29 October 2016, 16:35:37 UTC |
8034495 | Roman Lebedev | 29 October 2016, 15:20:19 UTC | Workaround travis-ci/travis-ci#329 by encrypting the notification settings Courtesy of mozilla/build-tools@c13a297 | 29 October 2016, 15:20:19 UTC |
bb95ff4 | Roman Lebedev | 29 October 2016, 13:49:15 UTC | Ashift iop: lsd: nfa(): let's try to dyn-allocate double inv[100000] Using _attribute__((constructor)) and __attribute__((destructor)) Seems to work. Is it supported everywhere though? | 29 October 2016, 13:49:15 UTC |
47e32f7 | johannes hanika | 29 October 2016, 13:13:35 UTC | basecurve: fix legacy param update | 29 October 2016, 13:13:54 UTC |
3979c4f | Pascal Obry | 29 October 2016, 09:44:09 UTC | Fix crash in undo/redo for deleted instances. Properly pass the module as user data to the history_invalidate_cb. This is a bug introduced during the last code refactoring to move undo/redo history specific code out of the generic undo/redo history. Fixes #11265. | 29 October 2016, 09:54:16 UTC |
71c74aa | Roman Lebedev | 29 October 2016, 09:09:44 UTC | CMake: xmllint: pass --nonet too | 29 October 2016, 09:09:44 UTC |
a1b1494 | Roman Lebedev | 29 October 2016, 08:42:04 UTC | CMake: xsltproc: ALWAYS pass --nonet. It is easier to install the DTDs, entities, documents and stylesheets locally than it is to understand that you need to install them locally. Without this change anyway. | 29 October 2016, 08:45:18 UTC |
3b6938f | Roman Lebedev | 29 October 2016, 08:36:06 UTC | CMake: Usermanual: actually use xsltproc, xml2po you just found. | 29 October 2016, 08:45:18 UTC |
19c3415 | sbraitbart | 28 October 2016, 19:32:09 UTC | Fixes in Hebrew translation (#1335) | 28 October 2016, 19:32:09 UTC |
f8f2af3 | Ulrich Pegelow | 28 October 2016, 18:53:29 UTC | usermanual: updates on the history stack panel in lighttable | 28 October 2016, 18:53:29 UTC |
d88ed2b | Roman Lebedev | 28 October 2016, 18:41:54 UTC | Colorchecker iop: gui_update(): always queue redraw. fixes update after reset. Since this module is pretty much completely drawn by draw(), it needs to be redrawn after every params change, not just when the param count changes. | 28 October 2016, 18:41:54 UTC |
b1a4a90 | Roman Lebedev | 28 October 2016, 18:10:57 UTC | Colorchecker iop: mess with dsvd(). Seems to fix #11267. | 28 October 2016, 18:10:57 UTC |
4817cbf | johannes hanika | 28 October 2016, 17:00:14 UTC | basecurve: proper fix for ui glitch | 28 October 2016, 17:03:17 UTC |
23032ef | Roman Lebedev | 28 October 2016, 15:08:24 UTC | Sort includes and fix build. Undo: correct includes. echo "DisableFormat: true" > .clang-format && tools/beautify_style.sh Same as f2ef4cb49e052a9f1a36eef9539b57cc39dd0bae. | 28 October 2016, 15:08:24 UTC |
86f9d88 | Roman Lebedev | 28 October 2016, 14:40:41 UTC | ci-script.sh: attempt to use Ninja as make. Seems to be a bit faster. (#1334) | 28 October 2016, 14:40:41 UTC |
97a6182 | Roman Lebedev | 28 October 2016, 13:26:50 UTC | Dockerfile: add missing \\ :( | 28 October 2016, 13:26:50 UTC |
f5ca028 | Roman Lebedev | 28 October 2016, 13:14:16 UTC | Dockerfile: add ninja-build | 28 October 2016, 13:14:59 UTC |
b98116f | Ulrich Pegelow | 28 October 2016, 13:13:14 UTC | opencl: fix bug in colorpicker | 28 October 2016, 13:13:14 UTC |
4ec9973 | Roman Lebedev | 28 October 2016, 13:09:20 UTC | Poison: don't do it in C++ | 28 October 2016, 13:09:20 UTC |
7cbc9ec | Roman Lebedev | 28 October 2016, 11:31:38 UTC | Merge pull request #1330 from LebedevRI/asan-poison Caches: poison | 28 October 2016, 11:31:38 UTC |
e547d9a | sbraitbart | 28 October 2016, 11:29:00 UTC | Updates and fixes in Hebrew translation (#1333) Updated Hebrew translation | 28 October 2016, 11:29:00 UTC |
128a1fa | Ulrich Pegelow | 28 October 2016, 09:50:36 UTC | opencl: fix "false color" mode in rawoverexposed | 28 October 2016, 09:50:36 UTC |
9eecebc | Ulrich Pegelow | 28 October 2016, 09:40:49 UTC | usermanual: on the raw overexposed indicator | 28 October 2016, 09:40:49 UTC |
7bbcf10 | Roman Lebedev | 27 October 2016, 21:48:47 UTC | amaze_demosaic: reorder includes, again. maybe fixes build? | 27 October 2016, 21:48:47 UTC |
11e6ad3 | Tobias Ellinghaus | 27 October 2016, 21:11:27 UTC | Maybe fix a build error | 27 October 2016, 21:11:27 UTC |
2431985 | Roman Lebedev | 27 October 2016, 20:51:59 UTC | Colorchecker iop: commit_params(): do not use p->num_patches | 27 October 2016, 21:02:19 UTC |
cb5df9e | Roman Lebedev | 27 October 2016, 20:17:45 UTC | Colorchecker iop: commit_params(): do MIN(MAX_PATCHES, p->num_patches); Avoids crashes later on. | 27 October 2016, 21:02:19 UTC |
4b2969d | Roman Lebedev | 27 October 2016, 20:15:27 UTC | Colorchecker iop: checker_button_press(): don't add more than max patches | 27 October 2016, 21:02:19 UTC |
4347cd7 | Roman Lebedev | 27 October 2016, 20:14:57 UTC | Colorchecker iop: gui callbacks(): check for out-of-patch in more places | 27 October 2016, 21:02:18 UTC |
25dff0b | Roman Lebedev | 27 October 2016, 20:01:03 UTC | Colorchecker iop: checker_motion_notify(): don't explode after deleting last patch | 27 October 2016, 21:02:18 UTC |
e43056a | Tobias Ellinghaus | 27 October 2016, 19:30:05 UTC | Guard headers with "pragma once" - It looks cleaner - We had a couple of guard defines of the __FOO_H__ style which is technically not to be used | 27 October 2016, 19:30:05 UTC |
b37b1a9 | schenlap | 27 October 2016, 17:48:56 UTC | Nikon Df: Reenable 12Bit support, fixes #11231 (#1332) | 27 October 2016, 17:48:56 UTC |
a5fafb6 | Roman Lebedev | 27 October 2016, 17:44:23 UTC | Update CONTRIBUTING.md: clang-format | 27 October 2016, 17:44:23 UTC |
3f7c469 | Stefan Schöfegger | 26 October 2016, 18:06:55 UTC | gen-profile: create a single result pdf | 27 October 2016, 17:22:24 UTC |
32caf45 | johannes hanika | 27 October 2016, 15:15:37 UTC | noiseprofiles: fuji x-t2. fixes #11151 | 27 October 2016, 15:16:40 UTC |
71d18ac | Ulrich Pegelow | 27 October 2016, 14:28:57 UTC | opencl: fix bug in white balance (prevented rawoverexposed from working) | 27 October 2016, 14:28:57 UTC |
f41c5cf | Tobias Ellinghaus | 27 October 2016, 10:00:22 UTC | [l10n] Update German translation | 27 October 2016, 10:00:22 UTC |
2cfdad0 | Ger Siemerink | 26 October 2016, 18:55:47 UTC | Updated Dutch translation | 26 October 2016, 18:55:47 UTC |
f00dad5 | Ger Siemerink | 26 October 2016, 18:50:50 UTC | Updated Dutch translation | 26 October 2016, 18:50:50 UTC |
8c50aa3 | Roman Lebedev | 26 October 2016, 18:12:45 UTC | Cache: dt_cache_release_with_caller(): if no other readers, poison. We are using pthread's rwlock. At any given point in time, there can be either one(!) writer, or many readers. If we are unlocking writer, then there are no other clients (with read lock). If we are unlocking reader, there may be other readers. So if we are unlocking the last read lock, only then can we poison the memory. Else, we will get false use-after-user-poison. FIXME: yes, this is *HIGHLY* unportable and is accessing implementation details. Once we migrate to CK, this should probably get better. Since this is only needed when compiling with ASAN, whoever wants to compile with ASAN with different pthread implementation shall simply change 1 to 0 in #if | 26 October 2016, 18:12:45 UTC |
e290288 | Roman Lebedev | 18 October 2016, 18:32:16 UTC | Cache: dt_cache_{test,}get(): 'w': do poison. If we have just acquired 'w' lock, it means there is no thing else holding a lock, so we can safely poison the memory, just in case. | 26 October 2016, 17:55:17 UTC |
a678025 | Roman Lebedev | 18 October 2016, 17:34:30 UTC | Cache: dt_cache_get_with_caller(): after new alloc, poison. It is up to the caller to decide how much of the buffer it actually wants to be accessible. | 26 October 2016, 17:55:07 UTC |
a7d0fd4 | Roman Lebedev | 19 October 2016, 16:32:23 UTC | dt_mipmap_cache_get_with_caller(): unpoison the actual buffer. | 26 October 2016, 17:54:55 UTC |
db1e14b | Roman Lebedev | 26 October 2016, 17:21:27 UTC | dt_mipmap_cache_get_with_caller(): unpoison dsc. | 26 October 2016, 17:54:43 UTC |
9857e1c | Roman Lebedev | 26 October 2016, 17:19:03 UTC | Mipmap cache: dt_mipmap_cache_remove(): do unpoison. | 26 October 2016, 17:54:32 UTC |
0c3371f | Roman Lebedev | 18 October 2016, 17:59:32 UTC | dt_mipmap_cache_alloc(): poison the whole buf, and unpoison needed part. This function is directly called by various code that needs MIP_FULL buffer. Since we keep MIP_FULL buffer, and only increase it, we need to make sure that if the requested size is less than the whole buffer size, only the requested size is actually used. | 26 October 2016, 17:54:22 UTC |
c025747 | Roman Lebedev | 26 October 2016, 17:54:08 UTC | Mipmap cache: struct dt_mipmap_buffer_dsc: add char[16] at the end. Since we 'prepend' this scruct at the beginning of the actual data to avoid doing 2 alloc's, we need to manually make sure that there is a redzone for ASAN to detect buffer under/over-runs. Also, to not calculate exactly how much of the dsc we need to unpoison, i'm adding dt_mipmap_buffer_dsc_size. | 26 October 2016, 17:54:08 UTC |
12ed170 | Roman Lebedev | 18 October 2016, 14:44:48 UTC | Image cache: dt_image_cache_{,test}get(): unpoison before return. | 26 October 2016, 17:46:34 UTC |
3db7793 | Roman Lebedev | 18 October 2016, 14:30:09 UTC | Cache: dt_cache_{remove,cleanup,gc}(): unpoison before calling cleanup_cb(). We do not really know what will that cleanup_cb() do, and since it is supposed to free() the buf, we should not worry whether to unpoison all the buf, or just the part. | 26 October 2016, 17:46:34 UTC |
0adae46 | Roman Lebedev | 18 October 2016, 17:31:31 UTC | Cache: dt_cache_{test,}get(): add a note about not unpoisoning. Why? Because we do not know how much to unpoison. E.g. mipmap cache MIP_FULL only increases sizes of the buffers, so with time, it may be using bigger buffers than it needs. | 26 October 2016, 17:46:33 UTC |
71682a0 | Roman Lebedev | 26 October 2016, 17:46:24 UTC | dt_mipmap_cache_alloc(): cleanup re-allocation a bit. | 26 October 2016, 17:46:24 UTC |
188a655 | Stefan Schöfegger | 26 October 2016, 17:34:11 UTC | gen-profile: add lables to charts, add ideal curve | 26 October 2016, 17:34:11 UTC |
1cf3131 | Roman Lebedev | 18 October 2016, 13:47:14 UTC | dt_mipmap_cache_alloc(): factor out dt_cache_entry_t *entry variable. Multiple pointer dereferences in one expression look so wrong. | 26 October 2016, 17:32:06 UTC |
e54436b | Roman Lebedev | 18 October 2016, 13:00:40 UTC | dt_mipmap_cache_allocate_dynamic(): cleanup new entry data allocation a bit. In particular, use new entry->data_size directly. | 26 October 2016, 17:32:06 UTC |
cdbe175 | Roman Lebedev | 18 October 2016, 12:49:42 UTC | dt_cache_get_with_caller(): cleanup new entry data allocation a bit. In particular, use new entry->data_size directly. | 26 October 2016, 17:32:06 UTC |
6b65e4b | Roman Lebedev | 18 October 2016, 12:48:00 UTC | dt_cache_get_with_caller(): new entry: copy data_size from cache. Should any cache need anything else, it'll need to modify entry->data_size in allocate_cb. | 26 October 2016, 17:32:06 UTC |