BuddhaShader/BuddhaTest/Shaders, branch v0.0.3½-speedFirst steps with OpenGL: A compute shader that renders a BuddhaBrot to an SSBO. Can also save the result to PNG for future viewing pleasure. https://git.grois.info/BuddhaShader/atom/BuddhaTest/Shaders?h=v0.0.3%C2%BD-speed2018-03-18T20:11:39ZUndo splitting of buffers in three. Makes buffer access much faster2018-03-18T20:11:39ZAndreas Groisandi@grois.info2018-03-18T20:11:39Zurn:sha1:dc193156a82a80cc8d2321cf065019035855516f
Not kidding here. For the default parameters this increased the maximum pixel brightness in the image by a factor of 2.25.
This has the drawback that the maximum possible image size got reduced, but hey, there's a command line switch to ignore all size limits ;-)
Move totaliterations calculation to CPU. This frees 4 VGPRs.2018-03-18T19:45:26ZAndreas Groisandi@grois.info2018-03-18T19:45:26Zurn:sha1:db2ea861c688aadf59fc35ff6cfca69149a1a94c
Now the code is back to the limit of 32 VGPRs without the need to directly operate on the state buffer.
This means that now it's bound by memory access.
Undo change to directly work on buffer. Causes about 30% slowdown...2018-03-18T16:19:33ZAndreas Groisandi@grois.info2018-03-18T16:19:33Zurn:sha1:c6ba30d76aca396ce5a3975f4625e05e8cf915eeAdd option to skip short orbits. Add option to print debug output.2018-03-18T15:24:50ZAndreas Groisandi@grois.info2018-03-18T15:24:50Zurn:sha1:385dd0dcf8fd003db604355341b00baa1fe1c1a0Remove layout specifier from struct. Seems it's not allowed after all.2018-03-17T19:52:50ZAndreas Groisandi@grois.info2018-03-17T19:52:50Zurn:sha1:c2018354129a4a6eff1649eaba4d99756b537ed3Fix memory allocation for state buffer. How could that ever work?2018-03-17T18:35:55ZAndreas Groisandi@grois.info2018-03-17T18:35:55Zurn:sha1:873409ffdc8f172fb9fe6c46b6ddbad9239cf02dCorrect line endings.2018-03-17T16:51:19ZAndreas Groisandi@grois.info2018-03-17T16:51:19Zurn:sha1:b05e87dc5004f5badc02508570ddc55f31b451ccAgain, cleaned code a bit2018-03-17T16:48:12ZAndreas Groisandi@grois.info2018-03-17T16:48:12Zurn:sha1:5a9c8599daee7d1d569ade3647d0078e56123c7aWorkers now directly operate on their state buffer.2018-03-17T16:28:40ZAndreas Groisandi@grois.info2018-03-17T16:28:40Zurn:sha1:3ff4226638aceaf179ce27ea9d0fd63327caa53b
To quote the documentation:
First, within a single shader invocation, if you perform an incoherent memory write, the value written will always be visible for reading. But only through that particular variable and only within the shader invocation that issued
the write. You need not do anything special to make this happen.
This means there is nothing stopping us from just reading and writing in the buffer as if the values were local variables, as they are specific to a given invocation. Also, buffer access happens infrequently, and furthermore
there's nothing stopping the GPU from caching the values, as memory access is incoherent anyhow.
Long story short: Now the code uses only 28 VGPRs, meaning that the AMD GCN hardware ideal of 32 is again achieved.
Fix y-value being always the same if only one worker is used2018-03-17T16:00:53ZAndreas Groisandi@grois.info2018-03-17T16:00:53Zurn:sha1:c93afe46650cca3d133dfb49642db2fac2edee0d