This patch would like to combine the vec_duplicate + vwaddu.wv to the
vwaddu.wx. From example as below code. The related pattern will depend
on the cost of vec_duplicate from GR2VR. Then the late-combine will
take action if the cost of GR2VR is zero, and reject the combination
if the GR2VR cost is greater than zero.
Assume we have asm code like below, GR2VR cost is 0.
Before this patch:
11 beq a3,zero,.L8
12 vsetvli a5,zero,e32,m1,ta,ma
13 vmv.v.x v2,a2
...
16 .L3:
17 vsetvli a5,a3,e32,m1,ta,ma
...
22 vwaddu.wv v1,v2,v3
...
25 bne a3,zero,.L3
After this patch:
11 beq a3,zero,.L8
...
14 .L3:
15 vsetvli a5,a3,e32,m1,ta,ma
...
20 vwaddu.wx v1,a2,v3
...
23 bne a3,zero,.L3
Unfortunately, and similar as vwaddu.vv, only widening from uint32_t to
uint64_t has the necessary zero-extend during combine, we loss the extend
op after expand for any other types.
gcc/ChangeLog:
* config/riscv/autovec-opt.md (*widen_waddu_wx_<mode>): Add new
pattern to match vwaddu.wx.
Signed-off-by: Pan Li <pan2.li@intel.com>
}
[(set_attr "type" "viwalu")])
+(define_insn_and_split "*widen_waddu_wx_<mode>"
+ [(set (match_operand:VWEXTI_D 0 "register_operand")
+ (any_widen_binop:VWEXTI_D
+ (vec_duplicate:VWEXTI_D
+ (any_extend:<VEL>
+ (match_operand:<VSUBEL> 2 "register_operand")))
+ (match_operand:VWEXTI_D 1 "register_operand")))]
+ "TARGET_VECTOR && TARGET_64BIT && can_create_pseudo_p ()"
+ "#"
+ "&& 1"
+ [(const_int 0)]
+ {
+ insn_code icode = code_for_pred_single_widen_scalar (PLUS, ZERO_EXTEND,
+ <MODE>mode);
+ riscv_vector::emit_vlmax_insn (icode, riscv_vector::BINARY_OP, operands);
+
+ DONE;
+ }
+ [(set_attr "type" "viwalu")])
+
;; =============================================================================
;; Combine vec_duplicate + op.vv to op.vf
;; Include