public inbox for gcc-cvs@sourceware.org
help / color / mirror / Atom feed
* [gcc r13-5098] xtensa: Make instruction cost estimation for size more accurate
@ 2023-01-11  5:54 Max Filippov
  0 siblings, 0 replies; only message in thread
From: Max Filippov @ 2023-01-11  5:54 UTC (permalink / raw)
  To: gcc-cvs

https://gcc.gnu.org/g:b399afd22c6ea50722bbff3247d52f3bd14bf2b6

commit r13-5098-gb399afd22c6ea50722bbff3247d52f3bd14bf2b6
Author: Takayuki 'January June' Suwa <jjsuwa_sys3175@yahoo.co.jp>
Date:   Tue Jan 10 12:34:01 2023 +0900

    xtensa: Make instruction cost estimation for size more accurate
    
    Until now, we applied COSTS_N_INSNS() (multiplying by 4) after dividing
    the instruction length by 3, so we couldn't express the difference less
    than modulo 3 in insn cost for size (e.g. 11 Bytes and 12 bytes cost the
    same).
    
    This patch fixes that.
    
    ;; 2 bytes
    addi.n  a2, a2, -1      ; cost 3
    
    ;; 3 bytes
    addmi   a2, a2, 1024    ; cost 4
    
    ;; 4 bytes
    movi.n  a3, 80          ; cost 5
    bnez.n  a2, a3, .L4
    
    ;; 5 bytes
    srli    a2, a3, 1       ; cost 7
    add.n   a2, a2, a2
    
    ;; 6 bytes
    ssai    8               ; cost 8
    src     a4, a2, a3
    
    :: 3 + 4 bytes
    l32r    a2, .L5         ; cost 9
    
    ;; 11 bytes             ; cost 15
    ;; 12 bytes             ; cost 16
    
    gcc/ChangeLog:
    
            * config/xtensa/xtensa.cc (xtensa_insn_cost):
            Let insn cost for size be obtained by applying COSTS_N_INSNS()
            to instruction length and then dividing by 3.

Diff:
---
 gcc/config/xtensa/xtensa.cc | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)

diff --git a/gcc/config/xtensa/xtensa.cc b/gcc/config/xtensa/xtensa.cc
index a1f184950ae..6cf6b35399a 100644
--- a/gcc/config/xtensa/xtensa.cc
+++ b/gcc/config/xtensa/xtensa.cc
@@ -4519,13 +4519,15 @@ xtensa_insn_cost (rtx_insn *insn, bool speed)
 {
   if (!(recog_memoized (insn) < 0))
     {
-      int len = get_attr_length (insn), n = (len + 2) / 3;
+      int len = get_attr_length (insn);
 
       if (len == 0)
 	return COSTS_N_INSNS (0);
 
       if (speed)  /* For speed cost.  */
 	{
+	  int n = (len + 2) / 3;
+
 	  /* "L32R" may be particular slow (implementation-dependent).  */
 	  if (xtensa_is_insn_L32R_p (insn))
 	    return COSTS_N_INSNS (1 + xtensa_extra_l32r_costs);
@@ -4572,10 +4574,11 @@ xtensa_insn_cost (rtx_insn *insn, bool speed)
 	    {
 	      /* "L32R" itself plus constant in litpool.  */
 	      if (xtensa_is_insn_L32R_p (insn))
-		return COSTS_N_INSNS (2) + 1;
+		len = 3 + 4;
 
-	      /* Consider ".n" short instructions.  */
-	      return COSTS_N_INSNS (n) - (n * 3 - len);
+	      /* Consider fractional instruction length (for example, ".n"
+		 short instructions or "L32R" litpool constants.  */
+	      return (COSTS_N_INSNS (len) + 1) / 3;
 	    }
 	}
     }

^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2023-01-11  5:54 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-01-11  5:54 [gcc r13-5098] xtensa: Make instruction cost estimation for size more accurate Max Filippov

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).