<html><head><meta http-equiv="Content-Type" content="text/html charset=windows-1252"></head><body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space;">I think ARMCC::HS is the wrong condition code for USUBO. I stumbled over the same problem with ARM64.<div>-Juergen</div><div><br><div><div>On May 7, 2014, at 3:13 PM, Louis Gerbarg <<a href="mailto:lgg@apple.com">lgg@apple.com</a>> wrote:</div><br class="Apple-interchange-newline"><blockquote type="cite"><meta http-equiv="Content-Type" content="text/html charset=us-ascii"><div style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space;"><div>Okay, here is a newer revision of the patch. Changes:</div><div><br></div><div>1) I only registered for i32s. I think there were probably some issues with i64s and i32 is going to be the more common case on ARM.</div><div>2) I updated all the tests to use regexps and track variables.</div><div>3) I spent some time trying to track through and get it to go to ADDS/SUBS, but in the end that proved to be more work than I thought it would be due to the differences between ARM and ARM64 in how the CPSR is handled, and the ARM models CMPs directly (where as ARM64 converts them to SUBS), and the fact that ARM backend only has working plumbing for CMP, not CMN. This patch is substantially better behavior than we currently have, and I would rather get it in now than wait trying to eek out that last bit of optimization.</div><div><br></div><div>Louis</div><div><br></div></div><span><0001-Add-custom-lowering-for-add-sub-with-overflow-intrin.patch></span><meta http-equiv="Content-Type" content="text/html charset=windows-1252"><div style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space;"><br><div><div>On May 6, 2014, at 1:46 PM, Louis Gerbarg <<a href="mailto:lgg@apple.com">lgg@apple.com</a>> wrote:</div><br class="Apple-interchange-newline"><blockquote type="cite"><div style="font-size: 12px; font-style: normal; font-variant: normal; font-weight: normal; letter-spacing: normal; line-height: normal; orphans: auto; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; widows: auto; word-spacing: 0px; -webkit-text-stroke-width: 0px;"><br>On May 2, 2014, at 11:40 PM, Tim Northover <<a href="mailto:t.p.northover@gmail.com">t.p.northover@gmail.com</a>> wrote:<br><br><blockquote type="cite">Hi Louis,<br><br>On 3 May 2014 01:43, Louis Gerbarg <<a href="mailto:lgg@apple.com">lgg@apple.com</a>> wrote:<br><blockquote type="cite">The attached patch provides custom lowering of overflow intrinsics for the ARM architecture.<br></blockquote><br>As Pete said, a very good idea!<br><br>I think I'd prefer it if the tests also tracked data dependencies as<br>well as just the instructions used though. If you swapped some<br>operands around the results might be disastrous.<br><br><blockquote type="cite"> add r2, r0, r1<br> mov r1, #1<br> cmp r2, r0<br> movvc r1, #0<br></blockquote><br>Couldn't this be simplified to:<br><br> adds r2, r0, r1<br> mov r1, #1<br> movvc r1, #0<br><br>? This would probably involve adding ARMISD::ADDS and ARMISD::SUBS<br>nodes (see ARM64 for reference, it already has them).<br></blockquote><br>Yes, this can be simplified further. I took another look at this morning. The issue that prevented me from doing it in the first place is that the arm backend handles the CPSR somewhat differently than ARM64. There are already somewhat analogous nodes (ARMISD::ADDC and ARMISD::SUBC), but they decide whether or not to set flags based on the liveness of the CPSR, so I need to figure out how to get that wired up. I don’t think adding explicit ADDS/SUBS nodes would make that any simpler because I suspect they would have to deal with equivalent CPSR issues in the tablegen files.<br><br>Louis<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">http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits</a></div></blockquote></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></div></body></html>