<html><head><meta http-equiv="Content-Type" content="text/html charset=iso-8859-1"></head><body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; "><br><div><blockquote type="cite"><div><br></div><div>I am the code owner for NVPTX and I approve this patch.</div><div class="gmail_extra"><br><br><div class="gmail_quote">On Fri, Nov 16, 2012 at 10:27 PM, Pawel Wodnicki <span dir="ltr"><<a href="mailto:pawel@32bitmicro.com" target="_blank">pawel@32bitmicro.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> Justin,<br>
<div class="im"><br>
> One more for 3.2. :)<br>
<br>
</div>I will queue it up but merging into the 3.2 is for now suspended<br>
as per:<br>
<br>
<a href="http://lists.cs.uiuc.edu/pipermail/llvmdev/2012-November/055895.html" target="_blank">http://lists.cs.uiuc.edu/pipermail/llvmdev/2012-November/055895.html</a><br>
<span class="HOEnZb"><font color="#888888"><br>
Pawel<br>
</font></span><div class="HOEnZb"><div class="h5"><br>
<br>
><br>
><br>
> On Fri, Nov 16, 2012 at 4:03 PM, Justin Holewinski<br>
> <<a href="mailto:jholewinski@nvidia.com">jholewinski@nvidia.com</a>>wrote:<br>
><br>
>> Author: jholewinski<br>
>> Date: Fri Nov 16 15:03:51 2012<br>
>> New Revision: 168198<br>
>><br>
>> URL: <a href="http://llvm.org/viewvc/llvm-project?rev=168198&view=rev" target="_blank">http://llvm.org/viewvc/llvm-project?rev=168198&view=rev</a><br>
>> Log:<br>
>> [NVPTX] Order global variables in def-use order before emiting them in the<br>
>> final assembly<br>
>><br>
>> Added:<br>
>> llvm/trunk/test/CodeGen/NVPTX/global-ordering.ll<br>
>> Modified:<br>
>> llvm/trunk/lib/Target/NVPTX/NVPTXAsmPrinter.cpp<br>
>><br>
>> Modified: llvm/trunk/lib/Target/NVPTX/NVPTXAsmPrinter.cpp<br>
>> URL:<br>
>> <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/NVPTX/NVPTXAsmPrinter.cpp?rev=168198&r1=168197&r2=168198&view=diff" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/NVPTX/NVPTXAsmPrinter.cpp?rev=168198&r1=168197&r2=168198&view=diff</a><br>
>><br>
>> ==============================================================================<br>
>> --- llvm/trunk/lib/Target/NVPTX/NVPTXAsmPrinter.cpp (original)<br>
>> +++ llvm/trunk/lib/Target/NVPTX/NVPTXAsmPrinter.cpp Fri Nov 16 15:03:51<br>
>> 2012<br>
>> @@ -68,7 +68,54 @@<br>
>><br>
>> cl::location(llvm::InterleaveSrcInPtx));<br>
>><br>
>><br>
>> +namespace {<br>
>> +/// DiscoverDependentGlobals - Return a set of GlobalVariables on which<br>
>> \p V<br>
>> +/// depends.<br>
>> +void DiscoverDependentGlobals(Value *V,<br>
>> + DenseSet<GlobalVariable*> &Globals) {<br>
>> + if (GlobalVariable *GV = dyn_cast<GlobalVariable>(V))<br>
>> + Globals.insert(GV);<br>
>> + else {<br>
>> + if (User *U = dyn_cast<User>(V)) {<br>
>> + for (unsigned i = 0, e = U->getNumOperands(); i != e; ++i) {<br>
>> + DiscoverDependentGlobals(U->getOperand(i), Globals);<br>
>> + }<br>
>> + }<br>
>> + }<br>
>> +}<br>
>><br>
>> +/// VisitGlobalVariableForEmission - Add \p GV to the list of<br>
>> GlobalVariable<br>
>> +/// instances to be emitted, but only after any dependents have been added<br>
>> +/// first.<br>
>> +void VisitGlobalVariableForEmission(GlobalVariable *GV,<br>
>> + SmallVectorImpl<GlobalVariable*><br>
>> &Order,<br>
>> + DenseSet<GlobalVariable*> &Visited,<br>
>> + DenseSet<GlobalVariable*> &Visiting) {<br>
>> + // Have we already visited this one?<br>
>> + if (Visited.count(GV)) return;<br>
>> +<br>
>> + // Do we have a circular dependency?<br>
>> + if (Visiting.count(GV))<br>
>> + report_fatal_error("Circular dependency found in global variable<br>
>> set");<br>
>> +<br>
>> + // Start visiting this global<br>
>> + Visiting.insert(GV);<br>
>> +<br>
>> + // Make sure we visit all dependents first<br>
>> + DenseSet<GlobalVariable*> Others;<br>
>> + for (unsigned i = 0, e = GV->getNumOperands(); i != e; ++i)<br>
>> + DiscoverDependentGlobals(GV->getOperand(i), Others);<br>
>> +<br>
>> + for (DenseSet<GlobalVariable*>::iterator I = Others.begin(),<br>
>> + E = Others.end(); I != E; ++I)<br>
>> + VisitGlobalVariableForEmission(*I, Order, Visited, Visiting);<br>
>> +<br>
>> + // Now we can visit ourself<br>
>> + Order.push_back(GV);<br>
>> + Visited.insert(GV);<br>
>> + Visiting.erase(GV);<br>
>> +}<br>
>> +}<br>
>><br>
>> // @TODO: This is a copy from AsmPrinter.cpp. The function is static, so<br>
>> we<br>
>> // cannot just link to the existing version.<br>
>> @@ -893,10 +940,27 @@<br>
>><br>
>> emitDeclarations(M, OS2);<br>
>><br>
>> - // Print out module-level global variables here.<br>
>> + // As ptxas does not support forward references of globals, we need to<br>
>> first<br>
>> + // sort the list of module-level globals in def-use order. We visit each<br>
>> + // global variable in order, and ensure that we emit it *after* its<br>
>> dependent<br>
>> + // globals. We use a little extra memory maintaining both a set and a<br>
>> list to<br>
>> + // have fast searches while maintaining a strict ordering.<br>
>> + SmallVector<GlobalVariable*,8> Globals;<br>
>> + DenseSet<GlobalVariable*> GVVisited;<br>
>> + DenseSet<GlobalVariable*> GVVisiting;<br>
>> +<br>
>> + // Visit each global variable, in order<br>
>> for (Module::global_iterator I = M.global_begin(), E = M.global_end();<br>
>> - I != E; ++I)<br>
>> - printModuleLevelGV(I, OS2);<br>
>> + I != E; ++I)<br>
>> + VisitGlobalVariableForEmission(I, Globals, GVVisited, GVVisiting);<br>
>> +<br>
>> + assert(GVVisited.size() == M.getGlobalList().size() &&<br>
>> + "Missed a global variable");<br>
>> + assert(GVVisiting.size() == 0 && "Did not fully process a global<br>
>> variable");<br>
>> +<br>
>> + // Print out module-level global variables in proper order<br>
>> + for (unsigned i = 0, e = Globals.size(); i != e; ++i)<br>
>> + printModuleLevelGV(Globals[i], OS2);<br>
>><br>
>> OS2 << '\n';<br>
>><br>
>><br>
>> Added: llvm/trunk/test/CodeGen/NVPTX/global-ordering.ll<br>
>> URL:<br>
>> <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/NVPTX/global-ordering.ll?rev=168198&view=auto" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/NVPTX/global-ordering.ll?rev=168198&view=auto</a><br>
>><br>
>> ==============================================================================<br>
>> --- llvm/trunk/test/CodeGen/NVPTX/global-ordering.ll (added)<br>
>> +++ llvm/trunk/test/CodeGen/NVPTX/global-ordering.ll Fri Nov 16 15:03:51<br>
>> 2012<br>
>> @@ -0,0 +1,20 @@<br>
>> +; RUN: llc < %s -march=nvptx -mcpu=sm_20 | FileCheck %s<br>
>> --check-prefix=PTX32<br>
>> +; RUN: llc < %s -march=nvptx64 -mcpu=sm_20 | FileCheck %s<br>
>> --check-prefix=PTX64<br>
>> +<br>
>> +; Make sure we emit these globals in def-use order<br>
>> +<br>
>> +<br>
>> +; PTX32: .visible .global .align 1 .u8 a = 2;<br>
>> +; PTX32-NEXT: .visible .global .align 4 .u32 a2 = a;<br>
>> +; PTX64: .visible .global .align 1 .u8 a = 2;<br>
>> +; PTX64-NEXT: .visible .global .align 8 .u64 a2 = a;<br>
>> +@a2 = addrspace(1) global i8 addrspace(1)* @a<br>
>> +@a = addrspace(1) global i8 2<br>
>> +<br>
>> +<br>
>> +; PTX32: .visible .global .align 1 .u8 b = 1;<br>
>> +; PTX32-NEXT: .visible .global .align 4 .u32 b2[2] = {b, b};<br>
>> +; PTX64: .visible .global .align 1 .u8 b = 1;<br>
>> +; PTX64-NEXT: .visible .global .align 8 .u64 b2[2] = {b, b};<br>
>> +@b2 = addrspace(1) global [2 x i8 addrspace(1)*] [i8 addrspace(1)* @b, i8<br>
>> addrspace(1)* @b]<br>
>> +@b = addrspace(1) global i8 1<br>
>><br>
>><br>
>> _______________________________________________<br>
>> llvm-commits mailing list<br>
>> <a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a><br>
>> <a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits</a><br>
>><br>
><br>
><br>
><br>
<br>
</div></div></blockquote></div><br><br clear="all"><div><br></div>-- <br><br><div>Thanks,</div><div><br></div><div>Justin Holewinski</div><br>
</div>
_______________________________________________<br>llvm-commits mailing list<br><a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a><br>http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits<br></blockquote></div><br></body></html>