http://gcc.gnu.org/ml/gcc-patches/2011-07/msg02186.html
Richard Henderson wrote:
> On 07/25/2011 10:30 AM, Georg-Johann Lay wrote:
>> PR target/29560
>> * config/avr/avr.md: Add peephole2 to map ashlhi3 to ashlqi3 if
>> high part of shift target is unused.
>
> Ok.
>
> r~
Here is a more complete version that transforms earlier
and increases the number of cases covered.
Moreover, the original peep2 is not fully correct because it
maps a 16-bit shift to a 8-bit one. The correct mapping is
(set (match_dup 2)
(subreg:QI (ashift:HI (zero_extend:HI (match_dup 2))
(match_dup 1))
0))
instead of
(set (match_dup 2)
(ashift:QI (match_dup 2)
(match_dup 1)))
I don't think it makes a difference that late in the
compilation process, yet I prefer correct semantics.
The new iterators are already seen in the widening
multiply patch.
Ok?
Johann
PR target/29560
* config/avr/avr.md (any_extend): New code interator.
(extend_prefix): New code attribute.
(*ashl<extend_prefix>qihiqi3): New insns.
(*ashl<extend_prefix>qihiqi3.mem): New insn-and-splits.
(*ashlhiqi3): New insn-and-split.
Add peephole2 to map ashlhi3 to ashlqi3 if high part of
shift target is unused.
Index: config/avr/avr.md
===================================================================
--- config/avr/avr.md (revision 176624)
+++ config/avr/avr.md (working copy)
@@ -132,6 +132,15 @@ (define_mode_iterator QIDI [(QI "") (HI
(define_mode_iterator HIDI [(HI "") (SI "") (DI "")])
(define_mode_iterator HISI [(HI "") (SI "")])
+;; Define code iterators
+(define_code_iterator any_extend [sign_extend zero_extend])
+
+;; Define code attributes
+(define_code_attr extend_prefix
+ [(sign_extend "s")
+ (zero_extend "u")])
+
+
;;========================================================================
;; The following is used by nonlocal_goto and setjmp.
;; The receiver pattern will create no instructions since internally
@@ -1993,6 +2002,89 @@ (define_insn "ashlhi3"
[(set_attr "length" "6,0,2,2,4,10,10")
(set_attr "cc" "clobber,none,set_n,clobber,set_n,clobber,clobber")])
+
+;; Insn like the following are generated when (implicitly) extending 8-bit shifts
+;; like char1 = char2 << char3. Only the low-byte is needed in that situation.
+;; The output is the same as of *ashlqi3. Don't, however, map this to *ashlqi3
+;; to preserve the semantics of ashift:HI which is different to ashift:QI.
+
+;; "*ashluqihiqi3"
+;; "*ashlsqihiqi3"
+(define_insn "*ashl<extend_prefix>qihiqi3"
+ [(set (match_operand:QI 0 "register_operand" "=r")
+ (subreg:QI (ashift:HI (any_extend:HI (match_operand:QI 1 "register_operand" "0"))
+ (match_operand:QI 2 "register_operand" "r"))
+ 0))]
+ ""
+ {
+ return ashlqi3_out (insn, operands, NULL);
+ }
+ [(set_attr "length" "5")
+ (set_attr "cc" "clobber")])
+
+;; Map to above.
+
+;; "*ashluqihiqi3.mem"
+;; "*ashlsqihiqi3.mem"
+(define_insn_and_split "*ashl<extend_prefix>qihiqi3.mem"
+ [(set (match_operand:QI 0 "memory_operand" "=m")
+ (subreg:QI (ashift:HI (any_extend:HI (match_operand:QI 1 "register_operand" "r"))
+ (match_operand:QI 2 "register_operand" "r"))
+ 0))]
+ "!reload_completed"
+ { gcc_unreachable(); }
+ "&& 1"
+ [(set (match_dup 3)
+ (subreg:QI (ashift:HI (any_extend:HI (match_dup 1))
+ (match_dup 2)) 0))
+ (set (match_dup 0)
+ (match_dup 3))]
+ {
+ operands[3] = force_reg (QImode, operands[0]);
+ })
+
+;; Ditto.
+
+(define_insn_and_split "*ashlhiqi3"
+ [(set (match_operand:QI 0 "nonimmediate_operand" "=r")
+ (subreg:QI (ashift:HI (match_operand:HI 1 "register_operand" "0")
+ (match_operand:QI 2 "register_operand" "r")) 0))]
+ "!reload_completed"
+ { gcc_unreachable(); }
+ "&& 1"
+ [(set (match_dup 4)
+ (subreg:QI (ashift:HI (zero_extend:HI (match_dup 3))
+ (match_dup 2)) 0))
+ (set (match_dup 0)
+ (match_dup 4))]
+ {
+ operands[3] = simplify_gen_subreg (QImode, operands[1], HImode, 0);
+ operands[4] = force_reg (QImode, operands[0]);
+ })
+
+;; High part of 16-bit shift is unused after the instruction:
+;; No need to compute it, map to 8-bit shift.
+
+(define_peephole2
+ [(set (match_operand:HI 0 "register_operand" "")
+ (ashift:HI (match_dup 0)
+ (match_operand:QI 1 "register_operand" "")))]
+ ""
+ [(set (match_dup 2)
+ (subreg:QI (ashift:HI (zero_extend:HI (match_dup 2))
+ (match_dup 1))
+ 0))
+ (clobber (match_dup 3))]
+ {
+ operands[3] = simplify_gen_subreg (QImode, operands[0], HImode, 1);
+
+ if (!peep2_reg_dead_p (1, operands[3]))
+ FAIL;
+
+ operands[2] = simplify_gen_subreg (QImode, operands[0], HImode, 0);
+ })
+
+
(define_insn "ashlsi3"
[(set (match_operand:SI 0 "register_operand" "=r,r,r,r,r,r,r")
(ashift:SI (match_operand:SI 1 "register_operand" "0,0,0,r,0,0,0")
