The following testcase ICEs on s390x with -march=z196 -O2 (or any -O+) since r10-3720-gac87f0f3459a57f03503e51aeffc54bb6ef36b90 : unsigned foo (unsigned short a) { a = a - (a >> 1 & 21845); a = (a & 13107) + (a >> 2 & 13107); return (unsigned short) ((a + (a >> 4) & 3855) * 257) >> 8; } The problem is in the popcounthi2_z196 expander that emits invalid RTL.
diff --git a/gcc/config/s390/s390.md b/gcc/config/s390/s390.md index 2959d8c0f17..e37ba49444a 100644 --- a/gcc/config/s390/s390.md +++ b/gcc/config/s390/s390.md @@ -11670,21 +11670,28 @@ (define_expand "popcountsi2" }) (define_expand "popcounthi2_z196" - [; popcnt op0, op1 - (parallel [(set (match_operand:HI 0 "register_operand" "") + [; popcnt op2, op1 + (parallel [(set (match_dup 2) (unspec:HI [(match_operand:HI 1 "register_operand")] UNSPEC_POPCNT)) (clobber (reg:CC CC_REGNUM))]) - ; sllk op2, op0, 8 - (set (match_dup 2) - (ashift:SI (match_dup 0) (const_int 8))) - ; ar op0, op2 - (parallel [(set (match_dup 0) (plus:SI (match_dup 0) (match_dup 2))) + ; lr op3, op2 + (set (match_dup 3) (subreg:SI (match_dup 2) 0)) + ; srl op4, op3, 8 + (set (match_dup 4) (lshiftrt:SI (match_dup 3) (const_int 8))) + ; ar op3, op4 + (parallel [(set (match_dup 3) (plus:SI (match_dup 3) (match_dup 4))) (clobber (reg:CC CC_REGNUM))]) - ; srl op0, op0, 8 - (set (match_dup 0) (lshiftrt:HI (match_dup 0) (const_int 8)))] + ; llgc op0, op3 + (parallel [(set (match_operand:HI 0 "register_operand" "") + (and:HI (subreg:HI (match_dup 3) 2) (const_int 255))) + (clobber (reg:CC CC_REGNUM))])] "TARGET_Z196" - "operands[2] = gen_reg_rtx (SImode);") +{ + operands[2] = gen_reg_rtx (HImode); + operands[3] = gen_reg_rtx (SImode); + operands[4] = gen_reg_rtx (SImode); +}) (define_expand "popcounthi2" [(set (match_dup 2) seems to work for me.
Created attachment 47761 [details] gcc10-pr93533.patch Full untested patch.
The master branch has been updated by Jakub Jelinek <jakub@gcc.gnu.org>: https://gcc.gnu.org/g:f626ae5478887b0cec886160dcfc4d59bf6fda07 commit r10-6400-gf626ae5478887b0cec886160dcfc4d59bf6fda07 Author: Jakub Jelinek <jakub@redhat.com> Date: Mon Feb 3 09:00:19 2020 +0100 s390x: Fix popcounthi2_z196 expander [PR93533] The following testcase started to ICE when .POPCOUNT matching has been added to match.pd; we had __builtin_popcount*, but nothing would use the popcounthi2 expander before. The problem is that the popcounthi2_z196 expander doesn't emit valid RTL: error: unrecognizable insn: (insn 138 137 139 27 (set (reg:SI 190) (ashift:SI (reg:HI 95 [ _105 ]) (const_int 8 [0x8]))) -1 (nil)) during RTL pass: vregs The following patch is an attempt to fix that, furthermore I've tried to slightly simplify it as well, it makes no sense to me to perform (x + (x << 8)) >> 8 when we need to either zero extend or mask the result at the end in order to avoid bits from above HImode to affect it, when we can do (x + (x >> 8)) & 0xff (or zero extension). 2020-02-03 Jakub Jelinek <jakub@redhat.com> PR target/93533 * config/s390/s390.md (popcounthi2_z196): Fix up expander to emit valid RTL to sum up the lowest and second lowest bytes of the popcnt result. * gcc.c-torture/compile/pr93533.c: New test. * gcc.target/s390/pr93533.c: New test.
Fixed.