RE: [PATCH] RL78 movdi improvement

2018-01-11 Thread Sebastian Perta
lso later on (in 290r.peephole2) we see that instruction 41 is eliminated as well: DCE: Deleting insn 41 deleting insn with uid = 41. Best Regards, Sebastian > -Original Message- > From: DJ Delorie [mailto:d...@redhat.com] > Sent: 09 January 2018 22:55 > To: Sebastian Perta

Re: [PATCH] RL78 movdi improvement

2018-01-09 Thread DJ Delorie
I saw one regression with this patch: PASS-FAIL: gcc.dg/torture/vshuf-v8qi.c -O2 execution test Could you confirm this? Note: I'm testing with your other (pre-2018 at least) patches applied at the same time (smax etc, anddi, bswaphi) so it might be an interaction, but the code looks like a

[PATCH] RL78 movdi improvement

2017-12-11 Thread Sebastian Perta
Hello, The following patch improves 64 bit operations by instructing GCC to use 16 bit movw instead of 8 bit mov. On the following test case the patch reduces the code size from 323 bytes to 245 bytes. unsigned long long my_anddi3(unsigned long long x, unsigned long long y){ return x & y; } I