https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119084
--- Comment #7 from GCC Commits <cvs-commit at gcc dot gnu.org> --- The master branch has been updated by Xi Ruoyao <xry...@gcc.gnu.org>: https://gcc.gnu.org/g:4856292f7a680ec478e7607f1b71781996d7d542 commit r15-7821-g4856292f7a680ec478e7607f1b71781996d7d542 Author: Xi Ruoyao <xry...@xry111.site> Date: Sun Mar 2 19:02:50 2025 +0800 LoongArch: Fix incorrect reorder of __lsx_vldx and __lasx_xvldx [PR119084] They could be incorrectly reordered with store instructions like st.b because the RTL expression does not have a memory_operand or a (mem) expression. The incorrect reorder has been observed in openh264 LTO build. Expand them to a (mem) expression instead of unspec to fix the issue. Then we need to make loongarch_address_insns return 1 for ADDRESS_REG_REG because the constraint "R" expects this behavior, or the vldx instruction will be considered invalid by the register allocate pass and turned to add.d + vld. Apply the ADDRESS_REG_REG penalty in loongarch_address_cost instead, loongarch_rtx_costs should also call loongarch_address_cost instead of loongarch_address_insns then. Closes: https://github.com/cisco/openh264/issues/3857 gcc/ChangeLog: PR target/119084 * config/loongarch/lasx.md (UNSPEC_LASX_XVLDX): Remove. (lasx_xvldx): Remove. * config/loongarch/lsx.md (UNSPEC_LSX_VLDX): Remove. (lsx_vldx): Remove. * config/loongarch/simd.md (QIVEC): New define_mode_iterator. (<simd_isa>_<x>vldx): New define_expand. * config/loongarch/loongarch.cc (loongarch_address_insns_1): New static function with most logic factored out from ... (loongarch_address_insns): ... here. Call loongarch_address_insns_1 with reg_reg_cost = 1. (loongarch_address_cost): Call loongarch_address_insns_1 with reg_reg_cost = la_addr_reg_reg_cost. gcc/testsuite/ChangeLog: PR target/119084 * gcc.target/loongarch/pr119084.c: New test.
