early-remat.c:process_block wasn't handling insns that set multiple candidate registers, which led to an assertion failure at the end of the main loop.
Instructions that set two pseudos aren't rematerialisation candidates in themselves, but we still need to track them if another instruction that sets the same register is a rematerialisation candidate. Tested on aarch64-linux-gnu, aarch64_be-elf and x86_64-linux-gnu. Pushed as obvious. I'll backport to GCC 9 and 8 too. Richard 2020-04-16 Richard Sandiford <richard.sandif...@arm.com> gcc/ PR rtl-optimization/94605 * early-remat.c (early_remat::process_block): Handle insns that set multiple candidate registers. gcc/testsuite/ PR rtl-optimization/94605 * gcc.target/aarch64/sve/pr94605.c: New test. --- gcc/early-remat.c | 2 +- gcc/testsuite/gcc.target/aarch64/sve/pr94605.c | 12 ++++++++++++ 2 files changed, 13 insertions(+), 1 deletion(-) create mode 100644 gcc/testsuite/gcc.target/aarch64/sve/pr94605.c diff --git a/gcc/early-remat.c b/gcc/early-remat.c index 80672cca241..9f5f8541644 100644 --- a/gcc/early-remat.c +++ b/gcc/early-remat.c @@ -2020,7 +2020,7 @@ early_remat::process_block (basic_block bb) } /* Now process definitions. */ - if (next_def && insn == next_def->insn) + while (next_def && insn == next_def->insn) { unsigned int gen = canon_candidate (next_candidate); diff --git a/gcc/testsuite/gcc.target/aarch64/sve/pr94605.c b/gcc/testsuite/gcc.target/aarch64/sve/pr94605.c new file mode 100644 index 00000000000..593e959e292 --- /dev/null +++ b/gcc/testsuite/gcc.target/aarch64/sve/pr94605.c @@ -0,0 +1,12 @@ +/* { dg-options "-O2 -msve-vector-bits=256" } */ + +typedef int v8si __attribute__((vector_size(32))); +int g (v8si, v8si); + +void +f (void) +{ + v8si x = {}, y = {}; + while (g (x, y)) + asm ("" : "+w" (x), "+w" (y)); +} -- 2.17.1