From 89edc9866776ba7805b40b592a2bb91252f8e1b5 Mon Sep 17 00:00:00 2001 From: Alan Lawrence Date: Tue, 4 Nov 2014 12:05:52 +0000 Subject: [PATCH] [ARM] Migrate to new reduc_plus_scal_optab config/arm/neon.md (reduc_plus_*): Rename to... (reduc_plus_scal_*): ...this; reduce to temp and extract scalar result. From-SVN: r217079 --- gcc/ChangeLog | 5 +++++ gcc/config/arm/neon.md | 43 ++++++++++++++++++++++-------------------- 2 files changed, 28 insertions(+), 20 deletions(-) diff --git a/gcc/ChangeLog b/gcc/ChangeLog index 5eb313d252c..a2568e47b5f 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,3 +1,8 @@ +2014-11-04 Alan Lawrence + + config/arm/neon.md (reduc_plus_*): Rename to... + (reduc_plus_scal_*): ...this; reduce to temp and extract scalar result. + 2014-11-04 Michael Collison * config/aarch64/iterators.md (lconst_atomic): New mode attribute diff --git a/gcc/config/arm/neon.md b/gcc/config/arm/neon.md index a0ec446071a..48270157715 100644 --- a/gcc/config/arm/neon.md +++ b/gcc/config/arm/neon.md @@ -1349,33 +1349,47 @@ ;; Reduction operations -(define_expand "reduc_splus_" - [(match_operand:VD 0 "s_register_operand" "") +(define_expand "reduc_plus_scal_" + [(match_operand: 0 "nonimmediate_operand" "") (match_operand:VD 1 "s_register_operand" "")] "TARGET_NEON && (! || flag_unsafe_math_optimizations)" { - neon_pairwise_reduce (operands[0], operands[1], mode, + rtx vec = gen_reg_rtx (mode); + neon_pairwise_reduce (vec, operands[1], mode, &gen_neon_vpadd_internal); + /* The same result is actually computed into every element. */ + emit_insn (gen_vec_extract (operands[0], vec, const0_rtx)); DONE; }) -(define_expand "reduc_splus_" - [(match_operand:VQ 0 "s_register_operand" "") +(define_expand "reduc_plus_scal_" + [(match_operand: 0 "nonimmediate_operand" "") (match_operand:VQ 1 "s_register_operand" "")] "TARGET_NEON && (! || flag_unsafe_math_optimizations) && !BYTES_BIG_ENDIAN" { rtx step1 = gen_reg_rtx (mode); - rtx res_d = gen_reg_rtx (mode); emit_insn (gen_quad_halves_plus (step1, operands[1])); - emit_insn (gen_reduc_splus_ (res_d, step1)); - emit_insn (gen_move_lo_quad_ (operands[0], res_d)); + emit_insn (gen_reduc_plus_scal_ (operands[0], step1)); DONE; }) -(define_insn "reduc_splus_v2di" +(define_expand "reduc_plus_scal_v2di" + [(match_operand:DI 0 "nonimmediate_operand" "=w") + (match_operand:V2DI 1 "s_register_operand" "")] + "TARGET_NEON && !BYTES_BIG_ENDIAN" +{ + rtx vec = gen_reg_rtx (V2DImode); + + emit_insn (gen_arm_reduc_plus_internal_v2di (vec, operands[1])); + emit_insn (gen_vec_extractv2di (operands[0], vec, const0_rtx)); + + DONE; +}) + +(define_insn "arm_reduc_plus_internal_v2di" [(set (match_operand:V2DI 0 "s_register_operand" "=w") (unspec:V2DI [(match_operand:V2DI 1 "s_register_operand" "w")] UNSPEC_VPADD))] @@ -1384,17 +1398,6 @@ [(set_attr "type" "neon_add_q")] ) -;; NEON does not distinguish between signed and unsigned addition except on -;; widening operations. -(define_expand "reduc_uplus_" - [(match_operand:VDQI 0 "s_register_operand" "") - (match_operand:VDQI 1 "s_register_operand" "")] - "TARGET_NEON && ( || !BYTES_BIG_ENDIAN)" -{ - emit_insn (gen_reduc_splus_ (operands[0], operands[1])); - DONE; -}) - (define_expand "reduc_smin_" [(match_operand:VD 0 "s_register_operand" "") (match_operand:VD 1 "s_register_operand" "")]