procyberian/gcc
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

CRIS: Add new peephole2 "lra_szext_decomposed_indir_plus"

Browse source 
Exposed when running the test-suite with -flate-combine-instructions.

	* config/cris/cris.md (lra_szext_decomposed_indir_plus): New
	peephole2 pattern.
This commit is contained in:
Hans-Peter Nilsson 2024-07-08 03:59:55 +02:00 • committed by Hans-Peter Nilsson
parent 9ea9d05908
commit 62dd893ff8
1 changed files with 45 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

45
gcc/config/cris/cris.md
View file

@ -3024,6 +3024,7 @@
;; Re-compose a decomposed "indirect offset" address for a szext
;; operation. The non-clobbering "addi" is generated by LRA.
;; This and lra_szext_decomposed is covered by cris/rld-legit1.c.
;; (Unfortunately not true when enabling late-combine.)
(define_peephole2 ; lra_szext_decomposed_indirect_with_offset
[(parallel
[(set (match_operand:SI 0 "register_operand")
@ -3046,6 +3047,50 @@
(mem:BW2 (plus:SI (szext:SI (mem:BW (match_dup 1))) (match_dup 2)))))
(clobber (reg:CC CRIS_CC0_REGNUM))])])
;; When enabling late-combine, we get a slightly changed register
;; allocation. The two allocations for the pseudo-registers involved
;; in the matching pattern get "swapped" and the (plus ...) in the
;; pattern above is now a load from a stack-slot. If peephole2 is
;; disabled, we see that the original sequence is actually improved;
;; one less incoming instruction, a load. We need to "undo" that
;; improvement a bit and move that load "back" to before the sequence
;; we combine in lra_szext_decomposed_indirect_with_offset. But that
;; changed again, so there's no define_peephole2 for that sequence
;; here, because it'd be hard or impossible to write a matching
;; test-case. A few commits later, the incoming pattern sequence has
;; changed again: back to the original but with the (plus...) part of
;; the address inside the second memory reference.
;; Coverage: cris/rld-legit1.c@r15-1880-gce34fcc572a0dc or
;; r15-3386-gaf1500dd8c00 when adding -flate-combine-instructions.
(define_peephole2 ; lra_szext_decomposed_indir_plus
[(parallel
[(set (match_operand:SI 0 "register_operand")
(sign_extend:SI (mem:BW (match_operand:SI 1 "register_operand"))))
(clobber (reg:CC CRIS_CC0_REGNUM))])
(parallel
[(set (match_operand:SI 3 "register_operand")
(szext:SI (mem:BW2 (plus:SI
(match_operand:SI 4 "register_operand")
(match_operand:SI 2 "register_operand")))))
(clobber (reg:CC CRIS_CC0_REGNUM))])]
"(REGNO (operands[0]) == REGNO (operands[3])
|| peep2_reg_dead_p (3, operands[0]))
&& (REGNO (operands[0]) == REGNO (operands[1])
|| peep2_reg_dead_p (3, operands[0]))
&& (rtx_equal_p (operands[2], operands[0])
|| rtx_equal_p (operands[4], operands[0]))"
[(parallel
[(set
(match_dup 3)
(szext:SI
(mem:BW2 (plus:SI (szext:SI (mem:BW (match_dup 1))) (match_dup 2)))))
(clobber (reg:CC CRIS_CC0_REGNUM))])]
{
if (! rtx_equal_p (operands[4], operands[0]))
operands[2] = operands[4];
})
;; Add operations with similar or same decomposed addresses here, when
;; encountered - but only when covered by mentioned test-cases for at
;; least one of the cases generalized in the pattern.