This has been driving me crazy. In order to JIT stuff in a simple REPL I'm trying to wrap them in anonymous functions and then evaluate them (as per recommendation of the Kaleidoscope guide). With simple expressions it works fine. Like if I try and add 4 and 5, I get this from the bare llvm::Value dump: double 9.000000e+00 and this after it's wrapped in a function: define double @0() { entry: ret double 9.000000e+00 } Then I define a function (ignores its one parameter and returns 4): define double @test(double %x) { entry: ret double 4.000000e+00 } And get the dump from the llvm::Value call to it (passing 10 as the arg) %calltmp = call double @test(double 1.000000e+01) ; <double> [#uses=1] the dump from the call wrapped in a function: define double @1() { entry: ret double %calltmp } but when I try to verify the wrapping function, I get: Instruction does not dominate all uses! %calltmp = call double @test(double 1.000000e+01) ; <double> [#uses=1] ret double %calltmp The code that I'm using to create the wrapping functions: http://pastebin.com/1Jzm8LgP What's going on here? From what I understand , "Instruction does not dominate all uses!" has to do with using a variable before its definition? I don't see where/how that's happening here. -- View this message in context: http://old.nabble.com/Instruction-does-not-dominate-all-uses--tp29422733p29422733.html Sent from the LLVM - Dev mailing list archive at Nabble.com.
You need to insert v into the basic block before the return. Note that %calltmp is absent from the dump of the function. The verifier is complaining because it found this pointer to %calltmp, but it didn't find the definition anywhere above its use. Reid On Thu, Aug 12, 2010 at 1:04 PM, alecbenzer <alecbenzer at gmail.com> wrote:> > This has been driving me crazy. In order to JIT stuff in a simple REPL I'm > trying to wrap them in anonymous functions and then evaluate them (as per > recommendation of the Kaleidoscope guide). > > With simple expressions it works fine. Like if I try and add 4 and 5, I get > this from the bare llvm::Value dump: > > double 9.000000e+00 > > and this after it's wrapped in a function: > > define double @0() { > entry: > ret double 9.000000e+00 > } > > > Then I define a function (ignores its one parameter and returns 4): > > define double @test(double %x) { > entry: > ret double 4.000000e+00 > } > > And get the dump from the llvm::Value call to it (passing 10 as the arg) > > %calltmp = call double @test(double 1.000000e+01) ; <double> [#uses=1] > > the dump from the call wrapped in a function: > > define double @1() { > entry: > ret double %calltmp > } > > but when I try to verify the wrapping function, I get: > > Instruction does not dominate all uses! > %calltmp = call double @test(double 1.000000e+01) ; <double> [#uses=1] > ret double %calltmp > > The code that I'm using to create the wrapping functions: > http://pastebin.com/1Jzm8LgP > > What's going on here? From what I understand , "Instruction does not > dominate all uses!" has to do with using a variable before its definition? I > don't see where/how that's happening here. > > -- > View this message in context: http://old.nabble.com/Instruction-does-not-dominate-all-uses--tp29422733p29422733.html > Sent from the LLVM - Dev mailing list archive at Nabble.com. > > _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev >
Ah, ok, so the call to builder.CreateCall needs to come after builder.SetInsertPoint for the wrapping function to work properly? I think I get it now. On Thu, Aug 12, 2010 at 4:26 PM, Reid Kleckner <reid.kleckner at gmail.com>wrote:> You need to insert v into the basic block before the return. Note > that %calltmp is absent from the dump of the function. The verifier > is complaining because it found this pointer to %calltmp, but it > didn't find the definition anywhere above its use. > > Reid > > On Thu, Aug 12, 2010 at 1:04 PM, alecbenzer <alecbenzer at gmail.com> wrote: > > > > This has been driving me crazy. In order to JIT stuff in a simple REPL > I'm > > trying to wrap them in anonymous functions and then evaluate them (as per > > recommendation of the Kaleidoscope guide). > > > > With simple expressions it works fine. Like if I try and add 4 and 5, I > get > > this from the bare llvm::Value dump: > > > > double 9.000000e+00 > > > > and this after it's wrapped in a function: > > > > define double @0() { > > entry: > > ret double 9.000000e+00 > > } > > > > > > Then I define a function (ignores its one parameter and returns 4): > > > > define double @test(double %x) { > > entry: > > ret double 4.000000e+00 > > } > > > > And get the dump from the llvm::Value call to it (passing 10 as the arg) > > > > %calltmp = call double @test(double 1.000000e+01) ; <double> [#uses=1] > > > > the dump from the call wrapped in a function: > > > > define double @1() { > > entry: > > ret double %calltmp > > } > > > > but when I try to verify the wrapping function, I get: > > > > Instruction does not dominate all uses! > > %calltmp = call double @test(double 1.000000e+01) ; <double> [#uses=1] > > ret double %calltmp > > > > The code that I'm using to create the wrapping functions: > > http://pastebin.com/1Jzm8LgP > > > > What's going on here? From what I understand , "Instruction does not > > dominate all uses!" has to do with using a variable before its > definition? I > > don't see where/how that's happening here. > > > > -- > > View this message in context: > http://old.nabble.com/Instruction-does-not-dominate-all-uses--tp29422733p29422733.html > > Sent from the LLVM - Dev mailing list archive at Nabble.com. > > > > _______________________________________________ > > LLVM Developers mailing list > > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev > > >-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20100812/b7cc44fb/attachment.html>
Maybe Matching Threads
- [LLVMdev] Instruction does not dominate all uses?
- Kaleidoscope tutorial: comments, corrections and Windows support
- [LLVMdev] Optimization of calls to functions without side effects (from Kaleidoscope example)
- [LLVMdev] Optimization of calls to functions without side effects (from Kaleidoscope example)
- [LLVMdev] Kaleidoscope doesn't work properly