https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105325
--- Comment #17 from CVS Commits <cvs-commit at gcc dot gnu.org> --- The master branch has been updated by Michael Meissner <meiss...@gcc.gnu.org>: https://gcc.gnu.org/g:370de1488a9a49956c47e5ec8c8f1489b4314a34 commit r14-2049-g370de1488a9a49956c47e5ec8c8f1489b4314a34 Author: Michael Meissner <meiss...@linux.ibm.com> Date: Fri Jun 23 11:32:39 2023 -0400 Fix power10 fusion bug with prefixed loads, PR target/105325 This changes fixes PR target/105325. PR target/105325 is a bug where an invalid lwa instruction is generated due to power10 fusion of a load instruction to a GPR and an compare immediate instruction with the immediate being -1, 0, or 1. In some cases, when the load instruction is done, the GCC compiler would generate a load instruction with an offset that was too large to fit into the normal load instruction. In particular, loads from the stack might originally have a small offset, so that the load is not a prefixed load. However, after the stack is set up, and register allocation has been done, the offset now is large enough that we would have to use a prefixed load instruction. The support for prefixed loads did not consider that patterns with a fused load and compare might have a prefixed address. Without this support, the proper prefixed load won't be generated. In the original code, when the split2 pass is run after reload has finished the ds_form_mem_operand predicate that was used for lwa and ld no longer returns true. When the pattern was created, ds_form_mem_operand recognized the insn as being valid since the offset was small. But after register allocation, ds_form_mem_operand did not return true. Because it didn't return true, the insn could not be split. Since the insn was not split and the prefix support did not indicate a prefixed instruction was used, the wrong load is generated. The solution involves: 1) Don't use ds_form_mem_operand for ld and lwa, always use non_update_memory_operand. 2) Delete ds_form_mem_operand since it is no longer used. 3) Use the "YZ" constraints for ld/lwa instead of "m". 4) If we don't need to sign extend the lwa, convert it to lwz, and use cmpwi instead of cmpdi. Adjust the insn name to reflect the code generate. 5) Insure that the insn using lwa will be recognized as having a prefixed operand (and hence the insn length will be 16 bytes instead of 8 bytes). 5a) Set the prefixed and maybe_prefix attributes to know that fused_load_cmpi are also load insns; 5b) In the case where we are just setting CC and not using the memory afterward, set the clobber to use a DI register, and put an explicit sign_extend operation in the split; 5c) Set the sign_extend attribute to "yes" for lwa. 5d) 5a-5c are the things that prefixed_load_p in rs6000.cc checks to ensure that lwa is treated as a ds-form instruction and not as a d-form instruction (i.e. lwz). 6) Add a new test case for this case. 7) Adjust the insn counts in fusion-p10-ldcmpi.c. Because we are no longer using ds_form_mem_operand, the ld and lwa instructions will fuse x-form (reg+reg) addresses in addition ds-form (reg+offset or reg). 2023-06-23 Michael Meissner <meiss...@linux.ibm.com> gcc/ PR target/105325 * config/rs6000/genfusion.pl (gen_ld_cmpi_p10_one): Fix problems that allowed prefixed lwa to be generated. * config/rs6000/fusion.md: Regenerate. * config/rs6000/predicates.md (ds_form_mem_operand): Delete. * config/rs6000/rs6000.md (prefixed attribute): Add support for load plus compare immediate fused insns. (maybe_prefixed): Likewise. gcc/testsuite/ PR target/105325 * g++.target/powerpc/pr105325.C: New test. * gcc.target/powerpc/fusion-p10-ldcmpi.c: Update insn counts. Co-Authored-By: Aaron Sawdey <acsaw...@linux.ibm.com>
