Summary: | wrong code at -Oz due to sign extension | ||
---|---|---|---|
Product: | gcc | Reporter: | Zdenek Sojka <zsojka> |
Component: | target | Assignee: | Roger Sayle <roger> |
Status: | RESOLVED FIXED | ||
Severity: | normal | CC: | jakub, marxin, roger |
Priority: | P3 | Keywords: | wrong-code |
Version: | 12.0 | ||
Target Milestone: | 12.0 | ||
See Also: | https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103785 | ||
Host: | x86_64-pc-linux-gnu | Target: | x86_64-pc-linux-gnu |
Build: | Known to work: | ||
Known to fail: | 12.0 | Last reconfirmed: | 2021-12-20 00:00:00 |
Attachments: | reduced testcase |
Description
Zdenek Sojka
2021-12-20 12:47:26 UTC
Sorry. Mine. Bootstrapping and regression testing a fix now. Also: int a; void foo (void) { a = -42; } which corrupts some random unrelated memory. For the TARGET_64BIT movsi case, it needs to be done only if the destination is a register rather than MEM. Mostly unrelated, I wonder if we shouldn't have a separate TYPE_PUSHPOP, decide in the (set (attr "type") and then just emit it and get e.g. the "length" attribute right for it. (In reply to Roger Sayle from comment #3) > Patch proposed > https://gcc.gnu.org/pipermail/gcc-patches/2021-December/587258.html Thank you for the patch. Does it work correctly with the red zone? I am asking since the "mov" instruction normally doesn't affect stack, and I don't see any indication that the "push" alternative does. But I definitely do not have enough knowledge about the GCC internals. The master branch has been updated by Roger Sayle <sayle@gcc.gnu.org>: https://gcc.gnu.org/g:ef26c151c14a87177d46fd3d725e7f82e040e89f commit r12-6106-gef26c151c14a87177d46fd3d725e7f82e040e89f Author: Roger Sayle <roger@nextmovesoftware.com> Date: Thu Dec 23 12:33:07 2021 +0000 x86: PR target/103773: Fix wrong-code with -Oz from pop to memory. This is a fix to PR target/103773 where -Oz shouldn't use push/pop on x86 to shrink writing small integer constants to memory. Instead clang uses "andl $0, mem" for writing zero, and "orl $-1, mem" when writing -1 to memory when using -Oz. This patch implements this via peephole2 where we can confirm that its ok to clobber the flags. 2021-12-23 Roger Sayle <roger@nextmovesoftware.com> Uroš Bizjak <ubizjak@gmail.com> gcc/ChangeLog PR target/103773 * config/i386/i386.md (*mov<mode>_and): New define_insn for writing a zero to memory using AND. (*mov<mode>_or): Extend to allow memory destination and HImode. (*movdi_internal): Remove -Oz push/pop optimization from here. (*movsi_internal): Likewise. (peephole2): Perform -Oz push/pop optimization here, only for register destinations, values other than zero, and in functions that don't used the red zone. (peephole2): With -Oz, convert writes of 0 or -1 to memory into their clobber forms, i.e. *mov<mode>_and and *mov<mode>_or resp. gcc/testsuite/ChangeLog PR target/103773 * gcc.target/i386/pr103773-2.c: New test case. * gcc.target/i386/pr103773.c: New test case. This shoud now be fixed on mainline. |