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

aarch64: Merge sqdmlal2 and sqdmlsl2 expanders

Browse source 
The various sqdmlal2 and sqdmlsl2 expanders perform almost identical functions and can be
merged using code iterators and attributes to reduce the code in the MD file.
No behavioural change is expected.

gcc/ChangeLog:

	* config/aarch64/aarch64-simd.md (aarch64_sqdmlal2<mode>): Merge into...
	(aarch64_sqdml<SBINQOPS:as>l2<mode>): ... This.
	(aarch64_sqdmlsl2<mode>): Delete.
	(aarch64_sqdmlal2_lane<mode>): Merge this...
	(aarch64_sqdmlsl2_lane<mode>): ... And this...
	(aarch64_sqdml<SBINQOPS:as>l2_lane<mode>): ... Into this.
	(aarch64_sqdmlal2_laneq<mode>): Merge this...
	(aarch64_sqdmlsl2_laneq<mode>): ... And this...
	(aarch64_sqdml<SBINQOPS:as>l2_laneq<mode>): ... Into this.
	(aarch64_sqdmlal2_n<mode>): Merge this...
	(aarch64_sqdmlsl2_n<mode>): ... And this...
	(aarch64_sqdml<SBINQOPS:as>l2_n<mode>): ... Into this.
This commit is contained in:
Kyrylo Tkachov 2021-05-12 10:52:51 +01:00
parent 3489257a33
commit 543c0cbca0
1 changed files with 28 additions and 76 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

104
gcc/config/aarch64/aarch64-simd.md
View file

@ -5356,29 +5356,19 @@
[(set_attr "type" "neon_sat_mla_<Vetype>_scalar_long")]
)
(define_expand "aarch64_sqdmlal2<mode>"
(define_expand "aarch64_sqdml<SBINQOPS:as>l2<mode>"
[(match_operand:<VWIDE> 0 "register_operand")
(match_operand:<VWIDE> 1 "register_operand")
(SBINQOPS:<VWIDE>
(match_operand:<VWIDE> 1 "register_operand")
(match_dup 1))
(match_operand:VQ_HSI 2 "register_operand")
(match_operand:VQ_HSI 3 "register_operand")]
"TARGET_SIMD"
{
rtx p = aarch64_simd_vect_par_cnst_half (<MODE>mode, <nunits>, true);
emit_insn (gen_aarch64_sqdmlal2<mode>_internal (operands[0], operands[1],
operands[2], operands[3], p));
DONE;
})
(define_expand "aarch64_sqdmlsl2<mode>"
[(match_operand:<VWIDE> 0 "register_operand")
(match_operand:<VWIDE> 1 "register_operand")
(match_operand:VQ_HSI 2 "register_operand")
(match_operand:VQ_HSI 3 "register_operand")]
"TARGET_SIMD"
{
rtx p = aarch64_simd_vect_par_cnst_half (<MODE>mode, <nunits>, true);
emit_insn (gen_aarch64_sqdmlsl2<mode>_internal (operands[0], operands[1],
operands[2], operands[3], p));
emit_insn (gen_aarch64_sqdml<SBINQOPS:as>l2<mode>_internal (operands[0],
operands[1], operands[2],
operands[3], p));
DONE;
})
@ -5436,63 +5426,37 @@
[(set_attr "type" "neon_sat_mla_<Vetype>_scalar_long")]
)
(define_expand "aarch64_sqdmlal2_lane<mode>"
(define_expand "aarch64_sqdml<SBINQOPS:as>l2_lane<mode>"
[(match_operand:<VWIDE> 0 "register_operand")
(match_operand:<VWIDE> 1 "register_operand")
(SBINQOPS:<VWIDE>
(match_operand:<VWIDE> 1 "register_operand")
(match_dup 1))
(match_operand:VQ_HSI 2 "register_operand")
(match_operand:<VCOND> 3 "register_operand")
(match_operand:SI 4 "immediate_operand")]
"TARGET_SIMD"
{
rtx p = aarch64_simd_vect_par_cnst_half (<MODE>mode, <nunits>, true);
emit_insn (gen_aarch64_sqdmlal2_lane<mode>_internal (operands[0], operands[1],
operands[2], operands[3],
operands[4], p));
emit_insn (gen_aarch64_sqdml<SBINQOPS:as>l2_lane<mode>_internal (operands[0],
operands[1], operands[2],
operands[3], operands[4], p));
DONE;
})
(define_expand "aarch64_sqdmlal2_laneq<mode>"
(define_expand "aarch64_sqdml<SBINQOPS:as>l2_laneq<mode>"
[(match_operand:<VWIDE> 0 "register_operand")
(match_operand:<VWIDE> 1 "register_operand")
(SBINQOPS:<VWIDE>
(match_operand:<VWIDE> 1 "register_operand")
(match_dup 1))
(match_operand:VQ_HSI 2 "register_operand")
(match_operand:<VCONQ> 3 "register_operand")
(match_operand:SI 4 "immediate_operand")]
"TARGET_SIMD"
{
rtx p = aarch64_simd_vect_par_cnst_half (<MODE>mode, <nunits>, true);
emit_insn (gen_aarch64_sqdmlal2_laneq<mode>_internal (operands[0], operands[1],
operands[2], operands[3],
operands[4], p));
DONE;
})
(define_expand "aarch64_sqdmlsl2_lane<mode>"
[(match_operand:<VWIDE> 0 "register_operand")
(match_operand:<VWIDE> 1 "register_operand")
(match_operand:VQ_HSI 2 "register_operand")
(match_operand:<VCOND> 3 "register_operand")
(match_operand:SI 4 "immediate_operand")]
"TARGET_SIMD"
{
rtx p = aarch64_simd_vect_par_cnst_half (<MODE>mode, <nunits>, true);
emit_insn (gen_aarch64_sqdmlsl2_lane<mode>_internal (operands[0], operands[1],
operands[2], operands[3],
operands[4], p));
DONE;
})
(define_expand "aarch64_sqdmlsl2_laneq<mode>"
[(match_operand:<VWIDE> 0 "register_operand")
(match_operand:<VWIDE> 1 "register_operand")
(match_operand:VQ_HSI 2 "register_operand")
(match_operand:<VCONQ> 3 "register_operand")
(match_operand:SI 4 "immediate_operand")]
"TARGET_SIMD"
{
rtx p = aarch64_simd_vect_par_cnst_half (<MODE>mode, <nunits>, true);
emit_insn (gen_aarch64_sqdmlsl2_laneq<mode>_internal (operands[0], operands[1],
operands[2], operands[3],
operands[4], p));
emit_insn (gen_aarch64_sqdml<SBINQOPS:as>l2_laneq<mode>_internal (operands[0],
operands[1], operands[2],
operands[3], operands[4], p));
DONE;
})
@ -5515,31 +5479,19 @@
[(set_attr "type" "neon_sat_mla_<Vetype>_scalar_long")]
)
(define_expand "aarch64_sqdmlal2_n<mode>"
(define_expand "aarch64_sqdml<SBINQOPS:as>l2_n<mode>"
[(match_operand:<VWIDE> 0 "register_operand")
(match_operand:<VWIDE> 1 "register_operand")
(SBINQOPS:<VWIDE>
(match_operand:<VWIDE> 1 "register_operand")
(match_dup 1))
(match_operand:VQ_HSI 2 "register_operand")
(match_operand:<VEL> 3 "register_operand")]
"TARGET_SIMD"
{
rtx p = aarch64_simd_vect_par_cnst_half (<MODE>mode, <nunits>, true);
emit_insn (gen_aarch64_sqdmlal2_n<mode>_internal (operands[0], operands[1],
operands[2], operands[3],
p));
DONE;
})
(define_expand "aarch64_sqdmlsl2_n<mode>"
[(match_operand:<VWIDE> 0 "register_operand")
(match_operand:<VWIDE> 1 "register_operand")
(match_operand:VQ_HSI 2 "register_operand")
(match_operand:<VEL> 3 "register_operand")]
"TARGET_SIMD"
{
rtx p = aarch64_simd_vect_par_cnst_half (<MODE>mode, <nunits>, true);
emit_insn (gen_aarch64_sqdmlsl2_n<mode>_internal (operands[0], operands[1],
operands[2], operands[3],
p));
emit_insn (gen_aarch64_sqdml<SBINQOPS:as>l2_n<mode>_internal (operands[0],
operands[1], operands[2],
operands[3], p));
DONE;
})