On 6 September 2012 18:42, Richard Earnshaw <rearn...@arm.com> wrote: > On 06/09/12 17:07, Christophe Lyon wrote: >> >> But why are the thumb1_XXX patterns still necessary? >> I tried removing them, but compiling the testcase with -march=armv6 >> -mthumb makes the compiler fail (internal compiler error: >> output_operand: invalid %-code) >> > > They probably aren't necessary. It should be possible to combine these > patterns into > > (define_insn "*arm_revsh" > [(set (match_operand:SI 0 "s_register_operand" "=l,r") > (sign_extend:SI (bswap:HI (match_operand:HI 1 > "s_register_operand" "l,r"))))] > "arm_arch6" > "revsh%?\t%0, %1" > [(set_attr "predicable" "yes") > (set_attr "arch" "t,32") > (set_attr "length" "2,4")] > >
Thanks for showing me the right "arch" value. The problem with this pattern if I delete the *thumb1_revsh one, is that %? is not accepted as a valid punctuation indicator for an operand when in thumb1 mode. In particular, arm_print_operand_punct_valid_p would return true in thumb1 if the punctuation character was '_'. However, I failed to find a '%_' operand in the ARM description: is arm_print_operand_punct_valid_p still accurate? Or can I replace '_' by '?' in this function? Thanks, Christophe.
