search for: i163

...d.i135.i = phi float [ %tmp43.i131.i, %cond.then.i132.i ], [ %tmp38.i125.i, %phi.exit69.i ] > %call.i156.i = tail call float @fabs(float %tmp11.i29.i) nounwind > %tmp5.i161.i = fmul float %call.i156.i, 0x3FCDA67120000000 > %tmp6.i162.i = fadd float %tmp5.i161.i, 1.000000e+00 > %tmp7.i163.i = fdiv float 1.000000e+00, %tmp6.i162.i > %call16.i172.i = tail call float @llvm.exp.f32(float %tmp15.i33.i) nounwind > %tmp17.i173.i = fmul float %call16.i172.i, 0x3FD9884540000000 > %tmp19.i175.i = fmul float %tmp17.i173.i, %tmp7.i163.i > %tmp29.i185.i = fmul float %tmp7.i163.i,...

Hello everybody, I have stumbled upon a test case (the attached module is a slightly reduced version) that shows extremely reduced performance on linux compared to windows when executed using LLVM's JIT. We narrowed the problem down to the actual code being generated, the source IR on both systems is the same. Try compiling the attached module: llc -O3 -filetype=asm -o BAD.s BAD.ll Under

..., i32 4 %incdec.ptr.4.i157.i.i.i.i = getelementptr inbounds [13 x i16]* %bi.i.i.i.i, i32 0, i32 5 %incdec.ptr.5.i159.i.i.i.i = getelementptr inbounds [13 x i16]* %bi.i.i.i.i, i32 0, i32 6 %incdec.ptr.6.i161.i.i.i.i = getelementptr inbounds [13 x i16]* %bi.i.i.i.i, i32 0, i32 7 %incdec.ptr.7.i163.i.i.i.i = getelementptr inbounds [13 x i16]* %bi.i.i.i.i, i32 0, i32 8 %incdec.ptr.8.i165.i.i.i.i = getelementptr inbounds [13 x i16]* %bi.i.i.i.i, i32 0, i32 9 %incdec.ptr.9.i167.i.i.i.i = getelementptr inbounds [13 x i16]* %bi.i.i.i.i, i32 0, i32 10 %incdec.ptr.10.i169.i.i.i.i = getelementp...