Hi!
The following testcase is miscompiled on x86_64-linux.
expand_compound_operation is called on
(zero_extract:DI (mem/c:TI (reg/f:DI 16 argp) [3 i+0 S16 A128])
(const_int 16 [0x10])
(const_int 63 [0x3f]))
so mode is DImode, inner_mode is TImode, pos 63, len 16 and modewidth 64.
A couple of lines above the problematic spot we have:
if (modewidth >= pos + len)
{
tem = gen_lowpart (mode, XEXP (x, 0));
where the code uses gen_lowpart and then shift left/right to extract it
in mode. But the guarding condition is false - 64 >= 63 + 16
and so we enter the next condition, where the code shifts XEXP (x, 0)
right by pos and then adds AND. It does so incorrectly though.
Given the modewidth < pos + len, inner_mode must be necessarily larger
than mode and XEXP (x, 0) has the innermode, but it was calling
simplify_shift_const with mode rather than inner_mode, which meant
inconsistent arguments to simplify_shift_const and in this case made
a DImode MEM shift out of it.
The following patch fixes it, by doing the shift in inner_mode properly
and then after the shift doing the lowpart subreg and masking already
in mode.
Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
2021-04-08 Jakub Jelinek
PR rtl-optimization/99905
* combine.c (expand_compound_operation): If pos + len > modewidth,
perform the right shift by pos in inner_mode and then convert to mode,
instead of trying to simplify a shift of rtx with inner_mode by pos
as if it was a shift in mode.
* gcc.target/i386/pr99905.c: New test.
--- gcc/combine.c.jj2021-01-04 10:25:39.127230495 +0100
+++ gcc/combine.c 2021-04-07 11:33:13.442626682 +0200
@@ -7409,11 +7409,15 @@ expand_compound_operation (rtx x)
mode, tem, modewidth - len);
}
else if (unsignedp && len < HOST_BITS_PER_WIDE_INT)
-tem = simplify_and_const_int (NULL_RTX, mode,
- simplify_shift_const (NULL_RTX, LSHIFTRT,
- mode, XEXP (x, 0),
- pos),
- (HOST_WIDE_INT_1U << len) - 1);
+{
+ tem = simplify_shift_const (NULL_RTX, LSHIFTRT, inner_mode,
+ XEXP (x, 0), pos);
+ tem = gen_lowpart (mode, tem);
+ if (!tem || GET_CODE (tem) == CLOBBER)
+ return x;
+ tem = simplify_and_const_int (NULL_RTX, mode, tem,
+ (HOST_WIDE_INT_1U << len) - 1);
+}
else
/* Any other cases we can't handle. */
return x;
--- gcc/testsuite/gcc.target/i386/pr99905.c.jj 2021-04-07 11:42:40.954292535
+0200
+++ gcc/testsuite/gcc.target/i386/pr99905.c 2021-04-07 11:42:06.659675315
+0200
@@ -0,0 +1,33 @@
+/* PR rtl-optimization/99905 */
+/* { dg-do run { target int128 } } */
+/* { dg-options "-Os -mno-mmx -mno-sse" } */
+
+typedef unsigned char U;
+typedef unsigned char __attribute__((__vector_size__ (8))) A;
+typedef unsigned char __attribute__((__vector_size__ (16))) B;
+typedef unsigned char __attribute__((__vector_size__ (32))) C;
+typedef unsigned int __attribute__((__vector_size__ (8))) D;
+typedef unsigned long long __attribute__((__vector_size__ (8))) E;
+typedef unsigned __int128 I;
+typedef unsigned long long L;
+
+D gv;
+I gi;
+
+L __attribute__((__noipa__))
+foo (int ua, int ub, int uc, int ud, E ue, I i)
+{
+ D d = (U) __builtin_bswap16 (i >> 63) + gv;
+ B y = ((union { C a; B b[2];}) (C){ }).b[0] + (B) gi;
+ A z = ((union { B a; A b[2];}) y).b[0] + (A) d;
+ return (L)z;
+}
+
+int
+main ()
+{
+ L x = foo (0, 0, 0, 0, (E) { }, (I) 0x100 << 63);
+ if (x != 0x10001)
+__builtin_abort ();
+ return 0;
+}
Jakub