Re: [Mesa-dev] shader-db, and justifying an i965 compiler optimization.
On Wed, May 18, 2011 at 12:23:40PM -0700, Eric Anholt wrote: > On Wed, 18 May 2011 09:00:09 +0200, Ian Romanick wrote: > > -BEGIN PGP SIGNED MESSAGE- > > Hash: SHA1 > > > > On 05/18/2011 05:22 AM, Eric Anholt wrote: > > > One of the pain points of working on compiler optimizations has been > > > justifying them -- sometimes I come up with something I think is > > > useful and spend a day or two on it, but the value doesn't show up as > > > fps in the application that suggested the optimization to me. Then I > > > wonder if this transformation of the code is paying off in general, > > > and thus if I should push it. If I don't push it, I end up bringing > > > that patch out on every application I look at that it could affect, to > > > see if now I finally have justification to get it out of a private > > > branch. > > > > > > At a conference this week, we heard about how another team is are > > > using a database of (assembly) shaders, which they run through their > > > compiler and count resulting instructions for testing purposes. This > > > sounded like a fun idea, so I threw one together. Patch #1 is good in > > > > This is one of those ideas that seems so obvious after you hear about it > > that you can't believe you hadn't thought of it years ago. This seems > > like something we'd want in piglit, but I'm not sure how that would look. > > Incidentally, Tom Stellard has apparently been doing this across piglit > already. This makes me think that maybe I want to just roll the > captured open-source shaders into glslparsertest, and just use the > analysis stuff on piglit. > I use this piglit patch to help capture shader stats: http://lists.freedesktop.org/archives/piglit/2010-December/000189.html It redirects any line of output that begins with ~ to a stats file. Then I use sdiff to compare stats files from different piglit runs. The output looks like this: shaders/glsl-orangebook-ch06-bump FRAGMENT PROGRAM ~~~ ~ 25 Instructions ~ 25 Vector Instructions (RGB) ~ 4 Scalar Instructions (Alpha) ~ 0 Flow Control Instructions ~ 0 Texture Instructions ~ 2 Presub Operations ~ 6 Temporary Registers ~~ END ~~ This patch is probably a little overkill, though, because as Marek pointed out, the same thing could be accomplished by grep'ing the raw output from piglit. This has been useful for testing compiler optimizations, but it would be much better if there were some real world shaders in piglit. Also, the glslparsertest hack isn't working on r300g, because shaders don't get compiled in the r300 backend until the first time they are used. It's done this way so the driver can emulate things like shadow samplers in the shader. I'm not sure what the best solution is for this. Maybe we could add an environment variable to force compilation at link time. -Tom ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
Re: [Mesa-dev] shader-db, and justifying an i965 compiler optimization.
On Wed, May 18, 2011 at 3:16 PM, Eric Anholt wrote: > On Wed, 18 May 2011 11:05:39 -0400, Jerome Glisse wrote: >> On Tue, May 17, 2011 at 11:22 PM, Eric Anholt wrote: >> > One of the pain points of working on compiler optimizations has been >> > justifying them -- sometimes I come up with something I think is >> > useful and spend a day or two on it, but the value doesn't show up as >> > fps in the application that suggested the optimization to me. Then I >> > wonder if this transformation of the code is paying off in general, >> > and thus if I should push it. If I don't push it, I end up bringing >> > that patch out on every application I look at that it could affect, to >> > see if now I finally have justification to get it out of a private >> > branch. >> > >> > At a conference this week, we heard about how another team is are >> > using a database of (assembly) shaders, which they run through their >> > compiler and count resulting instructions for testing purposes. This >> > sounded like a fun idea, so I threw one together. Patch #1 is good in >> > general (hey, link errors, finally!), but also means that a quick hack >> > to glslparsertest makes it link a passing compile shader and therefore >> > generate assembly that gets dumped under INTEL_DEBUG=wm. Patch #2 I >> > used for automatic scraping of shaders in every application I could >> > find on my system at the time. The open-source ones I pushed to: >> > >> > http://cgit.freedesktop.org/~anholt/shader-db >> > >> > And finally, patch #3 is something I built before but couldn't really >> > justify until now. However, given that it reduced fragment shader >> > instructions 0.3% across 831 shaders (affecting 52 of them including >> > yofrankie, warsow, norsetto, and gstreamer) and didn't increase >> > instructions anywhere, I'm a lot happier now. >> > >> > Hopefully we hook up EXT_timer_query to apitrace soon so I can do more >> > targeted optimizations and need this less :) In the meantime, I hope >> > this can prove useful to others -- if you want to contribute >> > appropriately-licensed shaders to the database so we track those, or >> > if you want to make the analysis work on your hardware backend, feel >> > free. >> > >> >> I have been thinking at doing somethings slightly different. Sadly >> instruction count is not necesarily the best metric to evaluate >> optimization performed by shader compiler. Hidding texture fetch >> latency of a shader can improve performance a lot more than saving 2 >> instructions. So my idea was to do a gl app that render into >> framebuffer thousand time the same shader. The use of fbo is to avoid >> to have things like swapbuffer or a like to play a role while we are >> solely interested in shader performance. Also use an fbo as big as >> possible so fragment shader has a lot of pixel to go through and i >> believe disabling things like blending, zbuffer ... so no other part >> of the pipeline impact in anyway the shader. > > You might take a look at mesa-demos/src/perf for that. I haven't had > success using them for performance work due to the noisiness of the > results. > > More generally, imo, the problem with that plan is you have to build the > shaders yourself and justify to yourself why that shader you wrote is > representative, and you spend all your time on building the tests when > you just wanted to know if an instruction-reduction optimization did > anything. shader-db took me one evening to build and collect for all > applications I had (I've got a personal branch for all the closed-source > stuff :/ ) Shader is a bunch of input, so for each shader collected the issue is to provide proper input, texture could use dummy texture unless the shader have some dependency on the texture data (like if the texture fetched data determine the number of iteration or is use to kill a fragment, ...). Well it's all about going through know shader and building a reasonable set of input for each of them, it's time consuming but i believe it brings a lot more for testing point of view. > For actual performance testing of apps without idsoftware-style > timedemos, I'm way more excited by the potential of using apitrace with > EXT_timer_query to decide which shaders I should be analyzing, and then > I'd know afterward whether I impacted a real application by replaying > the trace. That is, assuming I didn't increase CPU costs in the > process, which is where an apitrace replay would not be representative. > > Our perspective is: if we are driving the hardware anywhere below what > is possible, that is a bug that we should fix. Analyzing the costs of > instructions, scheduling impacts, CPU overhead impacts, etc. may be out > of scope for shader-db, but does make some types of analysis quick and > easy (test all shaders you have ever seen of in a couple minutes). I agree that shader-db provide a usefull tools, i am just convinced that number of instruction in complex shader is a bad metric especialy wh
Re: [Mesa-dev] shader-db, and justifying an i965 compiler optimization.
On Wed, 18 May 2011 11:05:39 -0400, Jerome Glisse wrote: > On Tue, May 17, 2011 at 11:22 PM, Eric Anholt wrote: > > One of the pain points of working on compiler optimizations has been > > justifying them -- sometimes I come up with something I think is > > useful and spend a day or two on it, but the value doesn't show up as > > fps in the application that suggested the optimization to me. Then I > > wonder if this transformation of the code is paying off in general, > > and thus if I should push it. If I don't push it, I end up bringing > > that patch out on every application I look at that it could affect, to > > see if now I finally have justification to get it out of a private > > branch. > > > > At a conference this week, we heard about how another team is are > > using a database of (assembly) shaders, which they run through their > > compiler and count resulting instructions for testing purposes. This > > sounded like a fun idea, so I threw one together. Patch #1 is good in > > general (hey, link errors, finally!), but also means that a quick hack > > to glslparsertest makes it link a passing compile shader and therefore > > generate assembly that gets dumped under INTEL_DEBUG=wm. Patch #2 I > > used for automatic scraping of shaders in every application I could > > find on my system at the time. The open-source ones I pushed to: > > > > http://cgit.freedesktop.org/~anholt/shader-db > > > > And finally, patch #3 is something I built before but couldn't really > > justify until now. However, given that it reduced fragment shader > > instructions 0.3% across 831 shaders (affecting 52 of them including > > yofrankie, warsow, norsetto, and gstreamer) and didn't increase > > instructions anywhere, I'm a lot happier now. > > > > Hopefully we hook up EXT_timer_query to apitrace soon so I can do more > > targeted optimizations and need this less :) In the meantime, I hope > > this can prove useful to others -- if you want to contribute > > appropriately-licensed shaders to the database so we track those, or > > if you want to make the analysis work on your hardware backend, feel > > free. > > > > I have been thinking at doing somethings slightly different. Sadly > instruction count is not necesarily the best metric to evaluate > optimization performed by shader compiler. Hidding texture fetch > latency of a shader can improve performance a lot more than saving 2 > instructions. So my idea was to do a gl app that render into > framebuffer thousand time the same shader. The use of fbo is to avoid > to have things like swapbuffer or a like to play a role while we are > solely interested in shader performance. Also use an fbo as big as > possible so fragment shader has a lot of pixel to go through and i > believe disabling things like blending, zbuffer ... so no other part > of the pipeline impact in anyway the shader. You might take a look at mesa-demos/src/perf for that. I haven't had success using them for performance work due to the noisiness of the results. More generally, imo, the problem with that plan is you have to build the shaders yourself and justify to yourself why that shader you wrote is representative, and you spend all your time on building the tests when you just wanted to know if an instruction-reduction optimization did anything. shader-db took me one evening to build and collect for all applications I had (I've got a personal branch for all the closed-source stuff :/ ) For actual performance testing of apps without idsoftware-style timedemos, I'm way more excited by the potential of using apitrace with EXT_timer_query to decide which shaders I should be analyzing, and then I'd know afterward whether I impacted a real application by replaying the trace. That is, assuming I didn't increase CPU costs in the process, which is where an apitrace replay would not be representative. Our perspective is: if we are driving the hardware anywhere below what is possible, that is a bug that we should fix. Analyzing the costs of instructions, scheduling impacts, CPU overhead impacts, etc. may be out of scope for shader-db, but does make some types of analysis quick and easy (test all shaders you have ever seen of in a couple minutes). pgpeSdItkPTWG.pgp Description: PGP signature ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
Re: [Mesa-dev] shader-db, and justifying an i965 compiler optimization.
On Wed, 18 May 2011 09:00:09 +0200, Ian Romanick wrote: > -BEGIN PGP SIGNED MESSAGE- > Hash: SHA1 > > On 05/18/2011 05:22 AM, Eric Anholt wrote: > > One of the pain points of working on compiler optimizations has been > > justifying them -- sometimes I come up with something I think is > > useful and spend a day or two on it, but the value doesn't show up as > > fps in the application that suggested the optimization to me. Then I > > wonder if this transformation of the code is paying off in general, > > and thus if I should push it. If I don't push it, I end up bringing > > that patch out on every application I look at that it could affect, to > > see if now I finally have justification to get it out of a private > > branch. > > > > At a conference this week, we heard about how another team is are > > using a database of (assembly) shaders, which they run through their > > compiler and count resulting instructions for testing purposes. This > > sounded like a fun idea, so I threw one together. Patch #1 is good in > > This is one of those ideas that seems so obvious after you hear about it > that you can't believe you hadn't thought of it years ago. This seems > like something we'd want in piglit, but I'm not sure how that would look. Incidentally, Tom Stellard has apparently been doing this across piglit already. This makes me think that maybe I want to just roll the captured open-source shaders into glslparsertest, and just use the analysis stuff on piglit. > The first problem is, obviously, using INTEL_DEBUG=wm to get the > instruction counts won't work. :) Perhaps we could extend some of the > existing assembly program queries (e.g., > GL_PROGRAM_NATIVE_INSTRUCTIONS_ARB) to GLSL. That would help even if we > didn't incorporate this into piglit. You say it won't work, but I'm using it and it is working :) Oh, you mean you want a clean solution and not a dirty hack? Yeah, I'd really like to have an interface for apps (read: shader debuggers) to get our annotated assembly out. > > And finally, patch #3 is something I built before but couldn't really > > justify until now. However, given that it reduced fragment shader > > instructions 0.3% across 831 shaders (affecting 52 of them including > > yofrankie, warsow, norsetto, and gstreamer) and didn't increase > > instructions anywhere, I'm a lot happier now. > > We'll probably want to be able to disable this once we have some sort of > CSE on the low-level IR. This sort of optimization can cause problems > for CSE in cases where the same register is a source and a destination. > Imagine something like > > z = sqrt(x) + y; > z = z * w; > q = sqrt(x) + y; > > If the result of the first 'sqrt(x) + y' is written directly to z, the > value is "gone" when the second 'sqrt(x) + y' is executed. If that > result is written to a temporary register that is then copied to z, the > value is still around at the second instance. > > Since we don't have any CSE, this doesn't matter now. However, it's > something to keep in mind. I think for CSE on 965 LIR, we'll want to be aggressive, and just consider whether the RHS values are still around, so we can execute to a temp and reuse it on math instructions. Otherwise, you end up with weird ordering requirements on the optimization passes to ensure that register coalescing doesn't kill these CSE opportunities. pgpVDvsrSbsMC.pgp Description: PGP signature ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
Re: [Mesa-dev] shader-db, and justifying an i965 compiler optimization.
On Tue, May 17, 2011 at 11:22 PM, Eric Anholt wrote: > One of the pain points of working on compiler optimizations has been > justifying them -- sometimes I come up with something I think is > useful and spend a day or two on it, but the value doesn't show up as > fps in the application that suggested the optimization to me. Then I > wonder if this transformation of the code is paying off in general, > and thus if I should push it. If I don't push it, I end up bringing > that patch out on every application I look at that it could affect, to > see if now I finally have justification to get it out of a private > branch. > > At a conference this week, we heard about how another team is are > using a database of (assembly) shaders, which they run through their > compiler and count resulting instructions for testing purposes. This > sounded like a fun idea, so I threw one together. Patch #1 is good in > general (hey, link errors, finally!), but also means that a quick hack > to glslparsertest makes it link a passing compile shader and therefore > generate assembly that gets dumped under INTEL_DEBUG=wm. Patch #2 I > used for automatic scraping of shaders in every application I could > find on my system at the time. The open-source ones I pushed to: > > http://cgit.freedesktop.org/~anholt/shader-db > > And finally, patch #3 is something I built before but couldn't really > justify until now. However, given that it reduced fragment shader > instructions 0.3% across 831 shaders (affecting 52 of them including > yofrankie, warsow, norsetto, and gstreamer) and didn't increase > instructions anywhere, I'm a lot happier now. > > Hopefully we hook up EXT_timer_query to apitrace soon so I can do more > targeted optimizations and need this less :) In the meantime, I hope > this can prove useful to others -- if you want to contribute > appropriately-licensed shaders to the database so we track those, or > if you want to make the analysis work on your hardware backend, feel > free. > I have been thinking at doing somethings slightly different. Sadly instruction count is not necesarily the best metric to evaluate optimization performed by shader compiler. Hidding texture fetch latency of a shader can improve performance a lot more than saving 2 instructions. So my idea was to do a gl app that render into framebuffer thousand time the same shader. The use of fbo is to avoid to have things like swapbuffer or a like to play a role while we are solely interested in shader performance. Also use an fbo as big as possible so fragment shader has a lot of pixel to go through and i believe disabling things like blending, zbuffer ... so no other part of the pipeline impact in anyway the shader. Others things might play a role, for instance if we provide small dummy texture we might just hide the gain texture fetch optimization might give, as the GPU might be able to have the texture in cache and thus have very low latency on each texture fetch. Same if we are using same texture for all unit, texture cache might hide latency that real application might otherwise face. So i think we need to have big enough dummy texture like 512*512 and different one for each unit, also try to provide random u,v for texture fetch so that texture cache doesn't hide too much of the latency. I am sure i am missing other factor that we should try to diminish while testing for shader performance. I think such things isn't a good fit for piglit but it can still be added as a subtools (so that we don't add yet another repository) Thanks a lot for extracting all those shader, i am sure we can get some people to write us shader with some what advance math under acceptable license. Cheers, Jerome ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
Re: [Mesa-dev] shader-db, and justifying an i965 compiler optimization.
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 On 05/18/2011 05:22 AM, Eric Anholt wrote: > One of the pain points of working on compiler optimizations has been > justifying them -- sometimes I come up with something I think is > useful and spend a day or two on it, but the value doesn't show up as > fps in the application that suggested the optimization to me. Then I > wonder if this transformation of the code is paying off in general, > and thus if I should push it. If I don't push it, I end up bringing > that patch out on every application I look at that it could affect, to > see if now I finally have justification to get it out of a private > branch. > > At a conference this week, we heard about how another team is are > using a database of (assembly) shaders, which they run through their > compiler and count resulting instructions for testing purposes. This > sounded like a fun idea, so I threw one together. Patch #1 is good in This is one of those ideas that seems so obvious after you hear about it that you can't believe you hadn't thought of it years ago. This seems like something we'd want in piglit, but I'm not sure how that would look. The first problem is, obviously, using INTEL_DEBUG=wm to get the instruction counts won't work. :) Perhaps we could extend some of the existing assembly program queries (e.g., GL_PROGRAM_NATIVE_INSTRUCTIONS_ARB) to GLSL. That would help even if we didn't incorporate this into piglit. The other problem is what the test would report for a result. Hmm... > general (hey, link errors, finally!), but also means that a quick hack > to glslparsertest makes it link a passing compile shader and therefore > generate assembly that gets dumped under INTEL_DEBUG=wm. Patch #2 I > used for automatic scraping of shaders in every application I could > find on my system at the time. The open-source ones I pushed to: > > http://cgit.freedesktop.org/~anholt/shader-db > > And finally, patch #3 is something I built before but couldn't really > justify until now. However, given that it reduced fragment shader > instructions 0.3% across 831 shaders (affecting 52 of them including > yofrankie, warsow, norsetto, and gstreamer) and didn't increase > instructions anywhere, I'm a lot happier now. We'll probably want to be able to disable this once we have some sort of CSE on the low-level IR. This sort of optimization can cause problems for CSE in cases where the same register is a source and a destination. Imagine something like z = sqrt(x) + y; z = z * w; q = sqrt(x) + y; If the result of the first 'sqrt(x) + y' is written directly to z, the value is "gone" when the second 'sqrt(x) + y' is executed. If that result is written to a temporary register that is then copied to z, the value is still around at the second instance. Since we don't have any CSE, this doesn't matter now. However, it's something to keep in mind. > Hopefully we hook up EXT_timer_query to apitrace soon so I can do more > targeted optimizations and need this less :) In the meantime, I hope > this can prove useful to others -- if you want to contribute > appropriately-licensed shaders to the database so we track those, or > if you want to make the analysis work on your hardware backend, feel > free. -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.11 (GNU/Linux) Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org/ iEYEARECAAYFAk3TbnkACgkQX1gOwKyEAw96twCfcEHQaQMe4HtpLar6zAFxj9Ww i/wAnRfQCSlN5E5vCIyE7t3Ep7EfXuL0 =aVeT -END PGP SIGNATURE- ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
[Mesa-dev] shader-db, and justifying an i965 compiler optimization.
One of the pain points of working on compiler optimizations has been justifying them -- sometimes I come up with something I think is useful and spend a day or two on it, but the value doesn't show up as fps in the application that suggested the optimization to me. Then I wonder if this transformation of the code is paying off in general, and thus if I should push it. If I don't push it, I end up bringing that patch out on every application I look at that it could affect, to see if now I finally have justification to get it out of a private branch. At a conference this week, we heard about how another team is are using a database of (assembly) shaders, which they run through their compiler and count resulting instructions for testing purposes. This sounded like a fun idea, so I threw one together. Patch #1 is good in general (hey, link errors, finally!), but also means that a quick hack to glslparsertest makes it link a passing compile shader and therefore generate assembly that gets dumped under INTEL_DEBUG=wm. Patch #2 I used for automatic scraping of shaders in every application I could find on my system at the time. The open-source ones I pushed to: http://cgit.freedesktop.org/~anholt/shader-db And finally, patch #3 is something I built before but couldn't really justify until now. However, given that it reduced fragment shader instructions 0.3% across 831 shaders (affecting 52 of them including yofrankie, warsow, norsetto, and gstreamer) and didn't increase instructions anywhere, I'm a lot happier now. Hopefully we hook up EXT_timer_query to apitrace soon so I can do more targeted optimizations and need this less :) In the meantime, I hope this can prove useful to others -- if you want to contribute appropriately-licensed shaders to the database so we track those, or if you want to make the analysis work on your hardware backend, feel free. ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev