<html><head><meta http-equiv="content-type" content="text/html; charset=utf-8"></head><body dir="auto"><div><div style="direction: inherit;"><br></div><br>Sent from my iPhone</div><div><br>On Jan 31, 2017, at 7:27 PM, Tian, Xinmin <<a href="mailto:xinmin.tian@intel.com">xinmin.tian@intel.com</a>> wrote:<br><br></div><blockquote type="cite"><div><span>Remember that, the prepare-phase is invoked in the FE or right after FE, inlining is not happening, that is why we don't call it "pass". Chandler made a good point for this case a long time back. </span><br><div style="direction: inherit;"><br></div></div></blockquote><div style="direction: inherit;"><br></div><div style="direction: inherit;">What I was describing is the inlining in the optimizer pipeline.</div><div style="direction: inherit;"><br></div><div style="direction: inherit;"><br></div><div style="direction: inherit;"><br></div><blockquote type="cite"><div><span></span><span>Hoisting alloca is totally ok. A new alloca is generated during outlining later on for anything marked as "private" (so long the "private" information is saves in the tag). I thought we talked this in an early email. </span><br></div></blockquote><div style="direction: inherit;"><br></div><div style="direction: inherit;"><br></div><div style="direction: inherit;">Can you describe how (and at which point) you get the private for "var" added to the tag?</div><div style="direction: inherit;"><br></div><div style="direction: inherit;"><br></div><div style="direction: inherit;">-- </div><div style="direction: inherit;">Mehdi</div><div style="direction: inherit;"><br></div><blockquote type="cite"><div><span></span><br><span>By the way, all concerns you have are all valid, we had worked on resolving these issues 10+ years back when we did similar things in our compilers. I wouldn't claim we have perfect solutions, but we do reasonable good solutions for handling general directives and openmp directives. </span></div></blockquote><blockquote type="cite"><div><div style="direction: inherit;"><br></div><span></span><span>Xinmin</span><br><span></span><br><span>-----Original Message-----</span><br><span>From: <a href="mailto:mehdi.amini@apple.com">mehdi.amini@apple.com</a> [<a href="mailto:mehdi.amini@apple.com">mailto:mehdi.amini@apple.com</a>] </span><br><span>Sent: Tuesday, January 31, 2017 7:08 PM</span><br><span>To: Tian, Xinmin <<a href="mailto:xinmin.tian@intel.com">xinmin.tian@intel.com</a>></span><br><span>Cc: Sanjoy Das <<a href="mailto:sanjoy@playingwithpointers.com">sanjoy@playingwithpointers.com</a>>; Adve, Vikram Sadanand <<a href="mailto:vadve@illinois.edu">vadve@illinois.edu</a>>; <a href="mailto:llvm-dev@lists.llvm.org">llvm-dev@lists.llvm.org</a>; <a href="mailto:llvm-dev-request@lists.llvm.org">llvm-dev-request@lists.llvm.org</a></span><br><span>Subject: Re: [llvm-dev] [RFC] IR-level Region Annotations</span><br><span></span><br><span></span><br><blockquote type="cite"><span>On Jan 31, 2017, at 6:48 PM, Tian, Xinmin <<a href="mailto:xinmin.tian@intel.com">xinmin.tian@intel.com</a>> wrote:</span><br></blockquote><blockquote type="cite"><span></span><br></blockquote><blockquote type="cite"><span>Let me try this. </span><br></blockquote><blockquote type="cite"><span></span><br></blockquote><blockquote type="cite"><span>You can simply consider the prepare-phase (e.g. pre-privatization) were done in FE (actually a library can be used by multiple FEs at LLVM IR level), the region is run with 1 thread, region annotation (scope, single-entry-single-exit) as memory barrier conservatively for now (instead of checking individual memory dependency, aliasing via tags which is the actual implementation is done) marked with region intrinsic functions. What optimization will mess up with this region-annotation? </span><br></blockquote><span></span><br><span>The first thing that comes to my mind is inlining that can put the IR in a form that breaks the invariant you tried to enforce with your "prepare-phase” (for example by hoisting an allocas).</span><br><span></span><br><span>— </span><br><span>Mehdi</span><br><span></span><br><span></span><br><blockquote type="cite"><span></span><br></blockquote><blockquote type="cite"><span></span><br></blockquote><blockquote type="cite"><span>-----Original Message-----</span><br></blockquote><blockquote type="cite"><span>From: <a href="mailto:mehdi.amini@apple.com">mehdi.amini@apple.com</a> [<a href="mailto:mehdi.amini@apple.com">mailto:mehdi.amini@apple.com</a>] </span><br></blockquote><blockquote type="cite"><span>Sent: Tuesday, January 31, 2017 5:47 PM</span><br></blockquote><blockquote type="cite"><span>To: Tian, Xinmin <<a href="mailto:xinmin.tian@intel.com">xinmin.tian@intel.com</a>></span><br></blockquote><blockquote type="cite"><span>Cc: Sanjoy Das <<a href="mailto:sanjoy@playingwithpointers.com">sanjoy@playingwithpointers.com</a>>; Adve, Vikram Sadanand <<a href="mailto:vadve@illinois.edu">vadve@illinois.edu</a>>; <a href="mailto:llvm-dev@lists.llvm.org">llvm-dev@lists.llvm.org</a>; <a href="mailto:llvm-dev-request@lists.llvm.org">llvm-dev-request@lists.llvm.org</a></span><br></blockquote><blockquote type="cite"><span>Subject: Re: [llvm-dev] [RFC] IR-level Region Annotations</span><br></blockquote><blockquote type="cite"><span></span><br></blockquote><blockquote type="cite"><span></span><br></blockquote><blockquote type="cite"><blockquote type="cite"><span>On Jan 31, 2017, at 5:38 PM, Tian, Xinmin <<a href="mailto:xinmin.tian@intel.com">xinmin.tian@intel.com</a>> wrote:</span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Ok, but this looks like a “workaround" for your specific use-case, I don’t see how it can scale as a model-agnostic and general-purpose region semantic.</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span>I would say it is a design trade-off.</span><br></blockquote></blockquote><blockquote type="cite"><span></span><br></blockquote><blockquote type="cite"><span>I’m not sure if we’re talking about the same thing here: my understanding at this point is that the design trading-off you’re making “simplicity” by scarifying “correctness”.</span><br></blockquote><blockquote type="cite"><span></span><br></blockquote><blockquote type="cite"><span>Requiring the IR to stay in what you’re calling a “canonical” form in your answer to Sanjoy in order to not miscompile a program is not an approach that seems compatible with how we deal with the IR usually.</span><br></blockquote><blockquote type="cite"><span></span><br></blockquote><blockquote type="cite"><blockquote type="cite"><span>Regardless it is a new instruction or an intrinsics with token/tag, it will consist of model-agnostic part and model-non-agnostic part. The package comes with a framework for parsing and using these intrinsics. See the reply I had for Sanjoy's email. </span><br></blockquote></blockquote><blockquote type="cite"><span></span><br></blockquote><blockquote type="cite"><span>The answer to Sanjoy is not really helpful to clarify anything to me. At this point I still don’t understand how this is supposed to be correct in general.</span><br></blockquote><blockquote type="cite"><span></span><br></blockquote><blockquote type="cite"><span>It would be helpful to have a LangRef patch that describes the semantic associated to your region intrinsics. Then we may be able to process some examples through the formalized description.</span><br></blockquote><blockquote type="cite"><span></span><br></blockquote><blockquote type="cite"><span>— </span><br></blockquote><blockquote type="cite"><span>Mehdi</span><br></blockquote><blockquote type="cite"><span></span><br></blockquote><blockquote type="cite"><span></span><br></blockquote><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span>-----Original Message-----</span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span>From: <a href="mailto:mehdi.amini@apple.com">mehdi.amini@apple.com</a> [<a href="mailto:mehdi.amini@apple.com">mailto:mehdi.amini@apple.com</a>] </span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span>Sent: Saturday, January 21, 2017 1:57 PM</span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span>To: Tian, Xinmin <<a href="mailto:xinmin.tian@intel.com">xinmin.tian@intel.com</a>></span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span>Cc: Sanjoy Das <<a href="mailto:sanjoy@playingwithpointers.com">sanjoy@playingwithpointers.com</a>>; Adve, Vikram Sadanand <<a href="mailto:vadve@illinois.edu">vadve@illinois.edu</a>>; <a href="mailto:llvm-dev@lists.llvm.org">llvm-dev@lists.llvm.org</a>; <a href="mailto:llvm-dev-request@lists.llvm.org">llvm-dev-request@lists.llvm.org</a></span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span>Subject: Re: [llvm-dev] [RFC] IR-level Region Annotations</span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>On Jan 20, 2017, at 11:17 AM, Tian, Xinmin <<a href="mailto:xinmin.tian@intel.com">xinmin.tian@intel.com</a>> wrote:</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>This means that the optimizer has to be aware of it, I’m missing the magic here?</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>This is one option. </span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>The another option is that, as I mentioned in our LLVM-HPC paper in our implementation. We have a "prepare phase for pre-privatization" can be invoked by both Clang FE and Fortran FE right after LLVM IR is generated. So, in this way, we are able to minimize the optimizations impact for the original val and I</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span>Ok, but this looks like a “workaround" for your specific use-case, I don’t see how it can scale as a model-agnostic and general-purpose region semantic.</span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span>The fact that you needed this pre-step in the first place seems to indicate to me that it confirms what multiple people expressed in this thread, for example what Daniel wrote here: <a href="http://lists.llvm.org/pipermail/llvm-dev/2017-January/108997.html">http://lists.llvm.org/pipermail/llvm-dev/2017-January/108997.html</a></span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span>— </span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span>Mehdi</span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>{ void main() {</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>i32* val = alloca i32</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>i32* I = alloca 32</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>i32* priv_val = alloca i32</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>i32* priv_i alloca 32</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>tok = llvm.experimental.intrinsic_a()[ "DIR.PARALLEL"(),"QUAL.PRIVATE"(i32* %priv_val, i32 %priv_i), "QUAL.NUM_THREADS"(i32 4)]</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>%priv_i = omp_get_thread_num();</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>compute_something_into_val(%priv_val, %priv_i); </span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>a[priv_i] = %priv_val;</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>llvm.experimental.intrinsic_b(tok)["DIR.END.PARALLEL"()];</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>....</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>I = </span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Val =</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Foo(val, i). </span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>}</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>"Prepare phase" is our way of minimizing the impact to existing optimizations. </span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Xinmin</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>-----Original Message-----</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>From: <a href="mailto:mehdi.amini@apple.com">mehdi.amini@apple.com</a> [<a href="mailto:mehdi.amini@apple.com">mailto:mehdi.amini@apple.com</a>] </span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Sent: Friday, January 20, 2017 10:54 AM</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>To: Tian, Xinmin</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Cc: Sanjoy Das; Adve, Vikram Sadanand; <a href="mailto:llvm-dev@lists.llvm.org">llvm-dev@lists.llvm.org</a>; <a href="mailto:llvm-dev-request@lists.llvm.org">llvm-dev-request@lists.llvm.org</a></span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Subject: Re: [llvm-dev] [RFC] IR-level Region Annotations</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>On Jan 20, 2017, at 10:44 AM, Tian, Xinmin via llvm-dev <<a href="mailto:llvm-dev@lists.llvm.org">llvm-dev@lists.llvm.org</a>> wrote:</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Sanjoy, the IR would be like something below. It is ok to hoist alloca instruction outside the region. There are some small changes in optimizer to understand region-annotation intrinsic. </span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>{ void main() {</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>i32* val = alloca i32</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>tok = llvm.experimental.intrinsic_a()[ "DIR.PARALLEL"(), </span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>"QUAL.PRIVATE"(i32* val), "QUAL.NUM_THREADS"(i32 4)]</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>int i = omp_get_thread_num();</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>compute_something_into_val(val, i);</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>a[i] = val;</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>llvm.experimental.intrinsic_b(tok)["DIR.END.PARALLEL"()];</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>}</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>With above representation, we can do privatization and outlining as </span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>below</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>{ void main() {</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>i32* val = alloca i32</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>i32* I = alloca 32</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>tok = llvm.experimental.intrinsic_a()[ "DIR.PARALLEL"(), </span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>"QUAL.PRIVATE"(i32* %val, i32 %i), "QUAL.NUM_THREADS"(i32 4)]</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>%ii = omp_get_thread_num();</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>compute_something_into_val(%val, %i); a[i] = %val;</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>llvm.experimental.intrinsic_b(tok)["DIR.END.PARALLEL"()];</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>}</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Here we come to the interesting part: the hoisting of "i32* I = alloca 32” above the intrinsics required to update the intrinsics information “QUAL.PRIVATE”. </span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>This means that the optimizer has to be aware of it, I’m missing the magic here?</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>I understand that an openmp specific optimization can do it, the question is how it an openmp agnostic supposed to behave in face of llvm.experimental.intrinsic_a?</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>—</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Mehdi</span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>1. create i32* priv_val = alloca i32 %priv_i = ...in the region, and replace all %val with %prv_val in the region. </span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>2. perform outlining. </span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Caller code</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>....</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>omp_push_num_threads(4)</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>omp_fork_call( .... outline_par_region....) ....</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Callee code: </span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Outlined_par_rgion {</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>I32* priv_val = alloca 32</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>I32* priv_i = ....</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Ret</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>}</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>For OpenMP, we do support it at -O0, -O1, -O2 and -O3. We had to make sure it runs correctly w/ and w/o optimizations and advanced analysis. So we need to preserve all source information for BE. </span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>You can take a look our LLVM-HPC paper for a bit some details. There are still tons of work to be done. Thanks. </span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Xinmin </span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>-----Original Message-----</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>From: llvm-dev [<a href="mailto:llvm-dev-bounces@lists.llvm.org">mailto:llvm-dev-bounces@lists.llvm.org</a>] On Behalf Of Sanjoy Das via llvm-dev</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Sent: Thursday, January 19, 2017 10:13 PM</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>To: Adve, Vikram Sadanand <<a href="mailto:vadve@illinois.edu">vadve@illinois.edu</a>></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Cc: llvm-dev <<a href="mailto:llvm-dev@lists.llvm.org">llvm-dev@lists.llvm.org</a>>; <a href="mailto:llvm-dev-request@lists.llvm.org">llvm-dev-request@lists.llvm.org</a></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Subject: Re: [llvm-dev] [RFC] IR-level Region Annotations</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Hi Vikram,</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>On Thu, Jan 19, 2017 at 9:27 PM, Adve, Vikram Sadanand <<a href="mailto:vadve@illinois.edu">vadve@illinois.edu</a>> wrote:</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Hi Sanjoy,</span><br></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Yes, that's exactly what we have been looking at recently here, but </span><br></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>the region tags seem to make it possible to express the control flow </span><br></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>as well, so I think we could start with reg ions+metadata, as Hal and</span><br></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>I'm not yet convinced that region tags are sufficient to model exotic control flow.</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>(I don't know OpenMP so this is a copy-pasted-edited example)</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Say we have:</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>void main() {</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>#pragma omp parallel num_threads(4)</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>{</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>int i = omp_get_thread_num();</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>int val;</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>compute_something_into_val(&val, i);</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>a[i] = val;</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>}</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>}</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>I presume the (eventual) intended lowering is something like this (if the intended lowering is different than this, and avoids the issue I'm trying to highlight then my point is moot):</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>void main() {</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>tok = llvm.experimental.intrinsic_a();</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>int i = omp_get_thread_num();</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>i32* val = alloca i32</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>compute_something_into_val(val, i);</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>a[i] = val;</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>llvm.experimental.intrinsic_b(tok);</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>}</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>However, LLVM is free to hoist the alloca to the entry block:</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>void main() {</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>i32* val = alloca i32</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>tok = llvm.experimental.intrinsic_a();</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>int i = omp_get_thread_num();</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>compute_something_into_val(val, i);</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>a[i] = val;</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>llvm.experimental.intrinsic_b(tok);</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>}</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>and now you have a race between the four parallel forks.</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>The problem here is that nothing in the IR expresses that we have four copies of the region running "at the same time". In fact, such a control flow is alien to LLVM today.</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>For instance, another evil optimization may turn:</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>void main() {</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>int a[4];</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>#pragma omp parallel num_threads(4)</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>{</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>int i = omp_get_thread_num();</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>int val = compute_something_into_val(i);</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>a[i] = val;</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>}</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>return a[0] + a[1];</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>}</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>to</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>void main() {</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>int a[4];</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>#pragma omp parallel num_threads(4)</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>{</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>int i = omp_get_thread_num();</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>int val = compute_something_into_val(i);</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>a[i] = val;</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>}</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>return undef;</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>}</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>since a[i] = val could have initialized at most one element in a.</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Now you could say that the llvm.experimental.intrinsic_a and llvm.experimental.intrinsic_b intrinsics are magic, and even such "obvious" optimizations are not allowed to happen across them; but then calls to these intrinsics is pretty fundamentally different from "normal" calls, and are probably best modeled as new instructions.</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>You're going to have to do the same kind of auditing of passes either way, and the only extra cost of a new instruction is the extra bitcode reading / writing code.</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>I hope I made sense.</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>-- Sanjoy</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Xinmin proposed, and then figure out what needs to be first class </span><br></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>instructions.</span><br></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>--Vikram Adve</span><br></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>On Jan 19, 2017, at 11:03 PM, Sanjoy Das <<a href="mailto:sanjoy@playingwithpointers.com">sanjoy@playingwithpointers.com</a>> wrote:</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Hi,</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>My bias is to use both (b) and (d), since they have complementary </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>strengths. We should use (b) for expressing concepts that can't be </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>semantically modeled as a call or invoke (this branch takes both its </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>successors), and (d) for expressing things that can be (this call may </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>never return), and annotation like things (this region (denoted by </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>def-use of a token) is a reduction).</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>I don't grok OpenMP, but perhaps we can come with one or two </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>"generalized control flow"-type instructions that can be used to </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>model the non-call/invoke like semantics we'd like LLVM to know </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>about, and model the rest with (d)?</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>-- Sanjoy</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>On Thu, Jan 19, 2017 at 8:28 PM, Hal Finkel via llvm-dev </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span><<a href="mailto:llvm-dev@lists.llvm.org">llvm-dev@lists.llvm.org</a>> wrote:</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>On 01/19/2017 03:36 PM, Mehdi Amini via llvm-dev wrote:</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>On Jan 19, 2017, at 1:32 PM, Daniel Berlin <<a href="mailto:dberlin@dberlin.org">dberlin@dberlin.org</a>> wrote:</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>On Thu, Jan 19, 2017 at 1:12 PM, Mehdi Amini <<a href="mailto:mehdi.amini@apple.com">mehdi.amini@apple.com</a>> wrote:</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>On Jan 19, 2017, at 12:04 PM, Daniel Berlin <<a href="mailto:dberlin@dberlin.org">dberlin@dberlin.org</a>> wrote:</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>On Thu, Jan 19, 2017 at 11:46 AM, Mehdi Amini via llvm-dev </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span><<a href="mailto:llvm-dev@lists.llvm.org">llvm-dev@lists.llvm.org</a>> wrote:</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>On Jan 19, 2017, at 11:36 AM, Adve, Vikram Sadanand via llvm-dev </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span><<a href="mailto:llvm-dev@lists.llvm.org">llvm-dev@lists.llvm.org</a>> wrote:</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Hi Johannes,</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>I am especially curious where you get your data from. Tapir [0] </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>(and to some degree PIR [1]) have shown that, </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>counterintuitively, only a few changes to LLVM passes are </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>needed. Tapir was recently used in an MIT class with a lot of </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>students and it seemed to work well with only minimal changes to </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>analysis and especially transformation passes.</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>TAPIR is an elegant, small extension and, in particular, I think </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>the idea of asymmetric parallel tasks and control flow is a </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>clever way to express parallelism with serial semantics, as in </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Cilk. Encoding the control flow extensions as explicit </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>instructions is orthogonal to that, though arguably more elegant than using region tags + metadata.</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>However, Cilk is a tiny language compared with the full </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>complexity of other languages, like OpenMP. To take just one </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>example, TAPIR cannot express the ORDERED construct of OpenMP. A </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>more serious concern, IMO, is that TAPIR (like Cilk) requires </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>serial semantics, whereas there are many parallel languages, OpenMP included, that do not obey that restriction.</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Third, OpenMP has *numerous* clauses, e.g., REDUCTION or PRIVATE, </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>that are needed because without that, you’d be dependent on </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>fundamentally hard compiler analyses to extract the same </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>information for satisfactory parallel performance; realistic </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>applications cannot depend on the success of such analyses.</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>I agree with this, but I’m also wondering if it needs to be first </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>class in the IR?</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>For example we know our alias analysis is very basic, and C/C++ </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>have a higher constraint thanks to their type system, but we </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>didn’t inject this higher level information that helps the </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>optimizer as first class IR constructs.</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>FWIW, while i agree with the general point, i wouldn't use this example.</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Because we pretty much still suffer to this day because of it (both </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>in AA, and devirt, and ...) :) We can't always even tell fields </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>apart</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Is it inherent to the infrastructure, i.e. using metadata instead </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>of first class IR construct or is it just a “quality of implementation” issue?</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Not to derail this conversation:</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>IMHO, At some point there is no real difference :)</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Because otherwise, everything is a QOI issue.</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>IE if it's super tricky to get metadata that works well and works </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>right, doesn't get lost, etc, and that's inherent to using metadata, </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>that to me is not a QOI issue.</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>So could it be done with metadata? Probably?</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>But at the same time, if it had been done with more first class </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>constructs, it would have happened years ago and been much lower cost.</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>This is what I meant by “inherent to the infrastructure”, thanks for </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>clarifying.</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>To clarify, we were proposing metadata that is used as arguments to </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>the region-annotation intrinsics. This metadata has the nice </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>property that it does not get dropped (so it is just being used as a </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>way of encoding whatever data structures are necessary without predefining a syntactic schema).</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>-Hal</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>—</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Mehdi</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>_______________________________________________</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>LLVM Developers mailing list</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span><a href="mailto:llvm-dev@lists.llvm.org">llvm-dev@lists.llvm.org</a></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span><a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev">http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev</a></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>--</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Hal Finkel</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Lead, Compiler Technology and Programming Languages Leadership </span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Computing Facility Argonne National Laboratory</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>_______________________________________________</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>LLVM Developers mailing list</span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span><a href="mailto:llvm-dev@lists.llvm.org">llvm-dev@lists.llvm.org</a></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span><a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev">http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev</a></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>_______________________________________________</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>LLVM Developers mailing list</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span><a href="mailto:llvm-dev@lists.llvm.org">llvm-dev@lists.llvm.org</a></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span><a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev">http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev</a></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>_______________________________________________</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>LLVM Developers mailing list</span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span><a href="mailto:llvm-dev@lists.llvm.org">llvm-dev@lists.llvm.org</a></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span><a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev">http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev</a></span><br></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote><blockquote type="cite"><span></span><br></blockquote><span></span><br></div></blockquote></body></html>