diff options
| author | Kyrylo Tkachov <kyrylo.tkachov@arm.com> | 2023-04-19 15:43:49 +0100 |
|---|---|---|
| committer | Kyrylo Tkachov <kyrylo.tkachov@arm.com> | 2023-04-19 15:43:49 +0100 |
| commit | a30078d5d974c3b2c784c522a84fd12df74767dd (patch) | |
| tree | d984a14b21c52978bf1e33256bb1a4a150c28f1e | |
| parent | 57aecdbc118d4c1900d651cb3ada2c9632a67ad0 (diff) | |
| download | gcc-a30078d5d974c3b2c784c522a84fd12df74767dd.zip gcc-a30078d5d974c3b2c784c522a84fd12df74767dd.tar.gz gcc-a30078d5d974c3b2c784c522a84fd12df74767dd.tar.bz2 | |
aarch64: Factorise widening add/sub high-half expanders with iterators
I noticed these define_expand are almost identical modulo some string substitutions.
This patch compresses them together with a couple of code iterators.
No functional change intended.
Bootstrapped and tested on aarch64-none-linux-gnu.
gcc/ChangeLog:
* config/aarch64/aarch64-simd.md (aarch64_saddw2<mode>): Delete.
(aarch64_uaddw2<mode>): Delete.
(aarch64_ssubw2<mode>): Delete.
(aarch64_usubw2<mode>): Delete.
(aarch64_<ANY_EXTEND:su><ADDSUB:optab>w2<mode>): New define_expand.
| -rw-r--r-- | gcc/config/aarch64/aarch64-simd.md | 66 |
1 files changed, 20 insertions, 46 deletions
diff --git a/gcc/config/aarch64/aarch64-simd.md b/gcc/config/aarch64/aarch64-simd.md index de2b738..1bed244 100644 --- a/gcc/config/aarch64/aarch64-simd.md +++ b/gcc/config/aarch64/aarch64-simd.md @@ -4713,52 +4713,26 @@ [(set_attr "type" "neon_add_widen")] ) -(define_expand "aarch64_saddw2<mode>" - [(match_operand:<VWIDE> 0 "register_operand") - (match_operand:<VWIDE> 1 "register_operand") - (match_operand:VQW 2 "register_operand")] - "TARGET_SIMD" -{ - rtx p = aarch64_simd_vect_par_cnst_half (<MODE>mode, <nunits>, true); - emit_insn (gen_aarch64_saddw2<mode>_internal (operands[0], operands[1], - operands[2], p)); - DONE; -}) - -(define_expand "aarch64_uaddw2<mode>" - [(match_operand:<VWIDE> 0 "register_operand") - (match_operand:<VWIDE> 1 "register_operand") - (match_operand:VQW 2 "register_operand")] - "TARGET_SIMD" -{ - rtx p = aarch64_simd_vect_par_cnst_half (<MODE>mode, <nunits>, true); - emit_insn (gen_aarch64_uaddw2<mode>_internal (operands[0], operands[1], - operands[2], p)); - DONE; -}) - - -(define_expand "aarch64_ssubw2<mode>" - [(match_operand:<VWIDE> 0 "register_operand") - (match_operand:<VWIDE> 1 "register_operand") - (match_operand:VQW 2 "register_operand")] - "TARGET_SIMD" -{ - rtx p = aarch64_simd_vect_par_cnst_half (<MODE>mode, <nunits>, true); - emit_insn (gen_aarch64_ssubw2<mode>_internal (operands[0], operands[1], - operands[2], p)); - DONE; -}) - -(define_expand "aarch64_usubw2<mode>" - [(match_operand:<VWIDE> 0 "register_operand") - (match_operand:<VWIDE> 1 "register_operand") - (match_operand:VQW 2 "register_operand")] - "TARGET_SIMD" -{ - rtx p = aarch64_simd_vect_par_cnst_half (<MODE>mode, <nunits>, true); - emit_insn (gen_aarch64_usubw2<mode>_internal (operands[0], operands[1], - operands[2], p)); +(define_expand "aarch64_<ANY_EXTEND:su><ADDSUB:optab>w2<mode>" + [(set (match_operand:<VWIDE> 0 "register_operand") + (ADDSUB:<VWIDE> + (ANY_EXTEND:<VWIDE> + (vec_select:<VHALF> + (match_operand:VQW 2 "register_operand") + (match_dup 3))) + (match_operand:<VWIDE> 1 "register_operand")))] + "TARGET_SIMD" +{ + /* We still do an emit_insn rather than relying on the pattern above + because for the MINUS case the operands would need to be swapped + around. */ + operands[3] + = aarch64_simd_vect_par_cnst_half (<MODE>mode, <nunits>, true); + emit_insn (gen_aarch64_<ANY_EXTEND:su><ADDSUB:optab>w2<mode>_internal( + operands[0], + operands[1], + operands[2], + operands[3])); DONE; }) |