* [PATCH] i386: Fix up replacement of registers in certain peephole2s [PR108832]
@ 2023-02-18 10:35 Jakub Jelinek
2023-02-18 11:22 ` Uros Bizjak
0 siblings, 1 reply; 2+ messages in thread
From: Jakub Jelinek @ 2023-02-18 10:35 UTC (permalink / raw)
To: Uros Bizjak; +Cc: gcc-patches
Hi!
As mentioned in the PR, replace_rtx has 2 modes, one that only replaces
x == from with to, the other which i386.md uses which also replaces
REGNO (x) == REGNO (from) with to if both are REGs, but assert they have
the same mode. This is reasonable behavior if one replaces from with
some other expression, say constant etc., but ICEs whenever the register
appears in a different mode, which happens e.g. on the following testcase,
where from/to has DImode but inside of the operands we have SImode of
the from register. replace_rtx also does some limited simplifications
(though far less than simplify_replace_fn_rtx), which is needed if
from a REG is replaced say with CONST_INT, but the peephole2s that use
this only replace one REG with another one.
The following patch introduces a new backend function for this, avoids doing
any simplifications and just replaces the REGs, for safety on a copy of
the expression if any changes will be needed.
Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
2023-02-18 Jakub Jelinek <jakub@redhat.com>
PR target/108832
* config/i386/i386-protos.h (ix86_replace_reg_with_reg): Declare.
* config/i386/i386-expand.cc (ix86_replace_reg_with_reg): New
function.
* config/i386/i386.md: Replace replace_rtx calls in all peephole2s
with ix86_replace_reg_with_reg.
* gcc.target/i386/pr108832.c: New test.
--- gcc/config/i386/i386-protos.h.jj 2023-01-16 11:52:15.960735877 +0100
+++ gcc/config/i386/i386-protos.h 2023-02-17 14:32:27.426259498 +0100
@@ -168,6 +168,7 @@ extern void ix86_split_lshr (rtx *, rtx,
extern void ix86_expand_v1ti_shift (enum rtx_code, rtx[]);
extern void ix86_expand_v1ti_rotate (enum rtx_code, rtx[]);
extern void ix86_expand_v1ti_ashiftrt (rtx[]);
+extern rtx ix86_replace_reg_with_reg (rtx, rtx, rtx);
extern rtx ix86_find_base_term (rtx);
extern bool ix86_check_movabs (rtx, int);
extern bool ix86_check_no_addr_space (rtx);
--- gcc/config/i386/i386-expand.cc.jj 2023-01-31 10:12:12.636168637 +0100
+++ gcc/config/i386/i386-expand.cc 2023-02-17 14:32:56.462839237 +0100
@@ -7093,6 +7093,37 @@ ix86_expand_v1ti_ashiftrt (rtx operands[
}
}
+/* Replace all occurrences of REG FROM with REG TO in X, including
+ occurrences with different modes. */
+
+rtx
+ix86_replace_reg_with_reg (rtx x, rtx from, rtx to)
+{
+ gcc_checking_assert (REG_P (from)
+ && REG_P (to)
+ && GET_MODE (from) == GET_MODE (to));
+ if (!reg_overlap_mentioned_p (from, x))
+ return x;
+ rtx ret = copy_rtx (x);
+ subrtx_ptr_iterator::array_type array;
+ FOR_EACH_SUBRTX_PTR (iter, array, &ret, NONCONST)
+ {
+ rtx *loc = *iter;
+ x = *loc;
+ if (REG_P (x) && REGNO (x) == REGNO (from))
+ {
+ if (x == from)
+ *loc = to;
+ else
+ {
+ gcc_checking_assert (REG_NREGS (x) == 1);
+ *loc = gen_rtx_REG (GET_MODE (x), REGNO (to));
+ }
+ }
+ }
+ return ret;
+}
+
/* Return mode for the memcpy/memset loop counter. Prefer SImode over
DImode for constant loop counts. */
--- gcc/config/i386/i386.md.jj 2023-02-16 10:13:23.705210608 +0100
+++ gcc/config/i386/i386.md 2023-02-17 14:34:22.606592399 +0100
@@ -22037,8 +22037,10 @@ (define_peephole2
(match_dup 0)))]
{
operands[7] = SET_DEST (XVECEXP (PATTERN (peep2_next_insn (1)), 0, 0));
- operands[8] = replace_rtx (operands[5], operands[0], operands[1], true);
- operands[9] = replace_rtx (operands[6], operands[0], operands[1], true);
+ operands[8]
+ = ix86_replace_reg_with_reg (operands[5], operands[0], operands[1]);
+ operands[9]
+ = ix86_replace_reg_with_reg (operands[6], operands[0], operands[1]);
})
;; Eliminate a reg-reg mov by inverting the condition of a cmov (#2).
@@ -22070,8 +22072,10 @@ (define_peephole2
(match_dup 0)))]
{
operands[7] = SET_DEST (XVECEXP (PATTERN (peep2_next_insn (2)), 0, 0));
- operands[8] = replace_rtx (operands[5], operands[0], operands[1], true);
- operands[9] = replace_rtx (operands[6], operands[0], operands[1], true);
+ operands[8]
+ = ix86_replace_reg_with_reg (operands[5], operands[0], operands[1]);
+ operands[9]
+ = ix86_replace_reg_with_reg (operands[6], operands[0], operands[1]);
})
(define_insn "movhf_mask"
@@ -23210,7 +23214,10 @@ (define_peephole2
(parallel [(set (match_dup 0)
(match_op_dup 3 [(match_dup 0) (match_dup 1)]))
(clobber (reg:CC FLAGS_REG))])]
- "operands[4] = replace_rtx (operands[2], operands[0], operands[1], true);")
+{
+ operands[4]
+ = ix86_replace_reg_with_reg (operands[2], operands[0], operands[1]);
+})
(define_peephole2
[(set (match_operand 0 "mmx_reg_operand")
--- gcc/testsuite/gcc.target/i386/pr108832.c.jj 2023-02-17 14:40:39.568136274 +0100
+++ gcc/testsuite/gcc.target/i386/pr108832.c 2023-02-17 14:39:36.846044106 +0100
@@ -0,0 +1,19 @@
+/* PR target/108832 */
+/* { dg-do compile } */
+/* { dg-options "-O2 -funroll-loops" } */
+
+unsigned int m;
+short int n;
+
+long int
+bar (unsigned int x)
+{
+ return x ? x : 1;
+}
+
+__attribute__ ((simd)) void
+foo (void)
+{
+ int a = m / bar (3);
+ n = 1 % bar (a << 1);
+}
Jakub
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] i386: Fix up replacement of registers in certain peephole2s [PR108832]
2023-02-18 10:35 [PATCH] i386: Fix up replacement of registers in certain peephole2s [PR108832] Jakub Jelinek
@ 2023-02-18 11:22 ` Uros Bizjak
0 siblings, 0 replies; 2+ messages in thread
From: Uros Bizjak @ 2023-02-18 11:22 UTC (permalink / raw)
To: Jakub Jelinek; +Cc: gcc-patches
On Sat, Feb 18, 2023 at 11:35 AM Jakub Jelinek <jakub@redhat.com> wrote:
>
> Hi!
>
> As mentioned in the PR, replace_rtx has 2 modes, one that only replaces
> x == from with to, the other which i386.md uses which also replaces
> REGNO (x) == REGNO (from) with to if both are REGs, but assert they have
> the same mode. This is reasonable behavior if one replaces from with
> some other expression, say constant etc., but ICEs whenever the register
> appears in a different mode, which happens e.g. on the following testcase,
> where from/to has DImode but inside of the operands we have SImode of
> the from register. replace_rtx also does some limited simplifications
> (though far less than simplify_replace_fn_rtx), which is needed if
> from a REG is replaced say with CONST_INT, but the peephole2s that use
> this only replace one REG with another one.
>
> The following patch introduces a new backend function for this, avoids doing
> any simplifications and just replaces the REGs, for safety on a copy of
> the expression if any changes will be needed.
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
>
> 2023-02-18 Jakub Jelinek <jakub@redhat.com>
>
> PR target/108832
> * config/i386/i386-protos.h (ix86_replace_reg_with_reg): Declare.
> * config/i386/i386-expand.cc (ix86_replace_reg_with_reg): New
> function.
> * config/i386/i386.md: Replace replace_rtx calls in all peephole2s
> with ix86_replace_reg_with_reg.
>
> * gcc.target/i386/pr108832.c: New test.
OK.
Thanks,
Uros.
>
> --- gcc/config/i386/i386-protos.h.jj 2023-01-16 11:52:15.960735877 +0100
> +++ gcc/config/i386/i386-protos.h 2023-02-17 14:32:27.426259498 +0100
> @@ -168,6 +168,7 @@ extern void ix86_split_lshr (rtx *, rtx,
> extern void ix86_expand_v1ti_shift (enum rtx_code, rtx[]);
> extern void ix86_expand_v1ti_rotate (enum rtx_code, rtx[]);
> extern void ix86_expand_v1ti_ashiftrt (rtx[]);
> +extern rtx ix86_replace_reg_with_reg (rtx, rtx, rtx);
> extern rtx ix86_find_base_term (rtx);
> extern bool ix86_check_movabs (rtx, int);
> extern bool ix86_check_no_addr_space (rtx);
> --- gcc/config/i386/i386-expand.cc.jj 2023-01-31 10:12:12.636168637 +0100
> +++ gcc/config/i386/i386-expand.cc 2023-02-17 14:32:56.462839237 +0100
> @@ -7093,6 +7093,37 @@ ix86_expand_v1ti_ashiftrt (rtx operands[
> }
> }
>
> +/* Replace all occurrences of REG FROM with REG TO in X, including
> + occurrences with different modes. */
> +
> +rtx
> +ix86_replace_reg_with_reg (rtx x, rtx from, rtx to)
> +{
> + gcc_checking_assert (REG_P (from)
> + && REG_P (to)
> + && GET_MODE (from) == GET_MODE (to));
> + if (!reg_overlap_mentioned_p (from, x))
> + return x;
> + rtx ret = copy_rtx (x);
> + subrtx_ptr_iterator::array_type array;
> + FOR_EACH_SUBRTX_PTR (iter, array, &ret, NONCONST)
> + {
> + rtx *loc = *iter;
> + x = *loc;
> + if (REG_P (x) && REGNO (x) == REGNO (from))
> + {
> + if (x == from)
> + *loc = to;
> + else
> + {
> + gcc_checking_assert (REG_NREGS (x) == 1);
> + *loc = gen_rtx_REG (GET_MODE (x), REGNO (to));
> + }
> + }
> + }
> + return ret;
> +}
> +
> /* Return mode for the memcpy/memset loop counter. Prefer SImode over
> DImode for constant loop counts. */
>
> --- gcc/config/i386/i386.md.jj 2023-02-16 10:13:23.705210608 +0100
> +++ gcc/config/i386/i386.md 2023-02-17 14:34:22.606592399 +0100
> @@ -22037,8 +22037,10 @@ (define_peephole2
> (match_dup 0)))]
> {
> operands[7] = SET_DEST (XVECEXP (PATTERN (peep2_next_insn (1)), 0, 0));
> - operands[8] = replace_rtx (operands[5], operands[0], operands[1], true);
> - operands[9] = replace_rtx (operands[6], operands[0], operands[1], true);
> + operands[8]
> + = ix86_replace_reg_with_reg (operands[5], operands[0], operands[1]);
> + operands[9]
> + = ix86_replace_reg_with_reg (operands[6], operands[0], operands[1]);
> })
>
> ;; Eliminate a reg-reg mov by inverting the condition of a cmov (#2).
> @@ -22070,8 +22072,10 @@ (define_peephole2
> (match_dup 0)))]
> {
> operands[7] = SET_DEST (XVECEXP (PATTERN (peep2_next_insn (2)), 0, 0));
> - operands[8] = replace_rtx (operands[5], operands[0], operands[1], true);
> - operands[9] = replace_rtx (operands[6], operands[0], operands[1], true);
> + operands[8]
> + = ix86_replace_reg_with_reg (operands[5], operands[0], operands[1]);
> + operands[9]
> + = ix86_replace_reg_with_reg (operands[6], operands[0], operands[1]);
> })
>
> (define_insn "movhf_mask"
> @@ -23210,7 +23214,10 @@ (define_peephole2
> (parallel [(set (match_dup 0)
> (match_op_dup 3 [(match_dup 0) (match_dup 1)]))
> (clobber (reg:CC FLAGS_REG))])]
> - "operands[4] = replace_rtx (operands[2], operands[0], operands[1], true);")
> +{
> + operands[4]
> + = ix86_replace_reg_with_reg (operands[2], operands[0], operands[1]);
> +})
>
> (define_peephole2
> [(set (match_operand 0 "mmx_reg_operand")
> --- gcc/testsuite/gcc.target/i386/pr108832.c.jj 2023-02-17 14:40:39.568136274 +0100
> +++ gcc/testsuite/gcc.target/i386/pr108832.c 2023-02-17 14:39:36.846044106 +0100
> @@ -0,0 +1,19 @@
> +/* PR target/108832 */
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -funroll-loops" } */
> +
> +unsigned int m;
> +short int n;
> +
> +long int
> +bar (unsigned int x)
> +{
> + return x ? x : 1;
> +}
> +
> +__attribute__ ((simd)) void
> +foo (void)
> +{
> + int a = m / bar (3);
> + n = 1 % bar (a << 1);
> +}
>
> Jakub
>
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2023-02-18 11:22 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-02-18 10:35 [PATCH] i386: Fix up replacement of registers in certain peephole2s [PR108832] Jakub Jelinek
2023-02-18 11:22 ` Uros Bizjak
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).