Add support for MAC and DMPY instructions. ARCv2 Only. gcc/ 2016-12-08 Claudiu Zissulescu <claz...@synopsys.com>
* config/arc/arc.c (arc_conditional_register_usage): Handle ACCL, ACCH registers. * config/arc/arc.md (mulsidi3): Use advanced mpy instructions when available. (umulsidi3): Likewise. (mulsidi3_700): Disable this pattern when we have advanced mpy instructions. (umulsidi3_700): Likewise. (maddsidi4): New pattern. (macd, mac, mac_r, umaddsidi4, macdu, macu, macu_r): Likewise. (mpyd_arcv2hs, mpyd_imm_arcv2hs, mpydu_arcv2hs): Likewise. (mpydu_imm_arcv2hs): Likewise. * config/arc/predicates.md (accl_operand): New predicate. --- gcc/config/arc/arc.c | 5 +- gcc/config/arc/arc.md | 266 ++++++++++++++++++++++++++++++++++++++++++- gcc/config/arc/predicates.md | 5 + 3 files changed, 272 insertions(+), 4 deletions(-) diff --git a/gcc/config/arc/arc.c b/gcc/config/arc/arc.c index 170b8dd..238244b 100644 --- a/gcc/config/arc/arc.c +++ b/gcc/config/arc/arc.c @@ -1789,8 +1789,9 @@ arc_conditional_register_usage (void) arc_regno_reg_class[PROGRAM_COUNTER_REGNO] = GENERAL_REGS; /*ARCV2 Accumulator. */ - if (TARGET_V2 - && (TARGET_FP_DP_FUSED || TARGET_FP_SP_FUSED)) + if ((TARGET_V2 + && (TARGET_FP_DP_FUSED || TARGET_FP_SP_FUSED)) + || TARGET_PLUS_DMPY) { arc_regno_reg_class[ACCL_REGNO] = WRITABLE_CORE_REGS; arc_regno_reg_class[ACCH_REGNO] = WRITABLE_CORE_REGS; diff --git a/gcc/config/arc/arc.md b/gcc/config/arc/arc.md index ffab390..db5867c 100644 --- a/gcc/config/arc/arc.md +++ b/gcc/config/arc/arc.md @@ -2157,6 +2157,18 @@ "TARGET_ANY_MPY" " { + if (TARGET_PLUS_MACD) + { + if (CONST_INT_P (operands[2])) + { + emit_insn (gen_mpyd_imm_arcv2hs (operands[0], operands[1], operands[2])); + } + else + { + emit_insn (gen_mpyd_arcv2hs (operands[0], operands[1], operands[2])); + } + DONE; + } if (TARGET_MPY) { operands[2] = force_reg (SImode, operands[2]); @@ -2237,7 +2249,7 @@ [(set (match_operand:DI 0 "register_operand" "=&r") (mult:DI (sign_extend:DI (match_operand:SI 1 "register_operand" "%c")) (sign_extend:DI (match_operand:SI 2 "extend_operand" "cL"))))] - "TARGET_MPY" + "TARGET_MPY && !TARGET_PLUS_MACD" "#" "&& reload_completed" [(const_int 0)] @@ -2390,6 +2402,18 @@ (zero_extend:DI (match_operand:SI 2 "nonmemory_operand" ""))))] "" { + if (TARGET_PLUS_MACD) + { + if (CONST_INT_P (operands[2])) + { + emit_insn (gen_mpydu_imm_arcv2hs (operands[0], operands[1], operands[2])); + } + else + { + emit_insn (gen_mpydu_arcv2hs (operands[0], operands[1], operands[2])); + } + DONE; + } if (TARGET_MPY) { operands[2] = force_reg (SImode, operands[2]); @@ -2480,7 +2504,7 @@ [(set (match_operand:DI 0 "dest_reg_operand" "=&r") (mult:DI (zero_extend:DI (match_operand:SI 1 "register_operand" "%c")) (zero_extend:DI (match_operand:SI 2 "extend_operand" "cL"))))] - "TARGET_MPY" + "TARGET_MPY && !TARGET_PLUS_MACD" "#" "reload_completed" [(const_int 0)] @@ -6215,6 +6239,244 @@ "" [(set_attr "length" "0")]) +;; MAC and DMPY instructions +(define_insn_and_split "maddsidi4" + [(set (match_operand:DI 0 "register_operand" "=r") + (plus:DI + (mult:DI + (sign_extend:DI (match_operand:SI 1 "register_operand" "%r")) + (sign_extend:DI (match_operand:SI 2 "extend_operand" "ri"))) + (match_operand:DI 3 "register_operand" "r")))] + "TARGET_PLUS_DMPY" + "#" + "TARGET_PLUS_DMPY && reload_completed" + [(const_int 0)] + "{ + rtx acc_reg = gen_rtx_REG (DImode, ACC_REG_FIRST); + emit_move_insn (acc_reg, operands[3]); + if (TARGET_PLUS_MACD) + emit_insn (gen_macd (operands[0], operands[1], operands[2])); + else + { + emit_insn (gen_mac (operands[1], operands[2])); + emit_move_insn (operands[0], acc_reg); + } + DONE; + }" + [(set_attr "type" "multi") + (set_attr "length" "36")]) + +(define_insn "macd" + [(set (match_operand:DI 0 "even_register_operand" "=Rcr,r,r") + (plus:DI + (mult:DI + (sign_extend:DI (match_operand:SI 1 "register_operand" "%0,c,c")) + (sign_extend:DI (match_operand:SI 2 "extend_operand" " c,cI,Cal"))) + (reg:DI ARCV2_ACC))) + (set (reg:DI ARCV2_ACC) + (plus:DI + (mult:DI (sign_extend:DI (match_dup 1)) + (sign_extend:DI (match_dup 2))) + (reg:DI ARCV2_ACC)))] + "TARGET_PLUS_MACD" + "macd %0,%1,%2" + [(set_attr "length" "4,4,8") + (set_attr "type" "multi") + (set_attr "predicable" "yes,no,no") + (set_attr "cond" "canuse,nocond,nocond")]) + +(define_insn "mac" + [(set (reg:DI ARCV2_ACC) + (plus:DI + (mult:DI (sign_extend:DI (match_operand:SI 0 "register_operand" "%r,r")) + (sign_extend:DI (match_operand:SI 1 "extend_operand" "rI,i"))) + (reg:DI ARCV2_ACC)))] + "TARGET_PLUS_DMPY" + "mac 0,%0,%1" + [(set_attr "length" "4,8") + (set_attr "type" "multi") + (set_attr "predicable" "no") + (set_attr "cond" "nocond")]) + +(define_peephole2 + [(set (reg:DI ARCV2_ACC) + (plus:DI + (mult:DI (sign_extend:DI (match_operand:SI 0 "register_operand" "")) + (sign_extend:DI (match_operand:SI 1 "extend_operand" ""))) + (reg:DI ARCV2_ACC))) + (set (match_operand:SI 2 "register_operand" "") + (match_operand:SI 3 "accl_operand" ""))] + "TARGET_PLUS_DMPY" + [(const_int 0)] + { + emit_insn (gen_mac_r (operands[2], operands[0], operands[1])); + DONE; + }) + +(define_insn "mac_r" + [(set (match_operand:SI 0 "register_operand" "=r,r") + (truncate:SI + (plus:DI + (mult:DI (sign_extend:DI (match_operand:SI 1 "register_operand" "%r,r")) + (sign_extend:DI (match_operand:SI 2 "extend_operand" "rI,i"))) + (reg:DI ARCV2_ACC)))) + (clobber (reg:DI ARCV2_ACC))] + "TARGET_PLUS_DMPY" + "mac %0,%1,%2" + [(set_attr "length" "4,8") + (set_attr "type" "multi") + (set_attr "predicable" "no") + (set_attr "cond" "nocond")]) + +(define_insn_and_split "umaddsidi4" + [(set (match_operand:DI 0 "register_operand" "=r") + (plus:DI + (mult:DI + (zero_extend:DI (match_operand:SI 1 "register_operand" "%r")) + (zero_extend:DI (match_operand:SI 2 "extend_operand" "ri"))) + (match_operand:DI 3 "register_operand" "r")))] + "TARGET_PLUS_DMPY" + "#" + "TARGET_PLUS_DMPY && reload_completed" + [(const_int 0)] + "{ + rtx acc_reg = gen_rtx_REG (DImode, ACC_REG_FIRST); + emit_move_insn (acc_reg, operands[3]); + if (TARGET_PLUS_MACD) + emit_insn (gen_macdu (operands[0], operands[1], operands[2])); + else + { + emit_insn (gen_macu (operands[1], operands[2])); + emit_move_insn (operands[0], acc_reg); + } + DONE; + }" + [(set_attr "type" "multi") + (set_attr "length" "36")]) + +(define_insn "macdu" + [(set (match_operand:DI 0 "even_register_operand" "=Rcr,r,r") + (plus:DI + (mult:DI + (zero_extend:DI (match_operand:SI 1 "register_operand" "%0,c,c")) + (zero_extend:DI (match_operand:SI 2 "extend_operand" " c,cI,i"))) + (reg:DI ARCV2_ACC))) + (set (reg:DI ARCV2_ACC) + (plus:DI + (mult:DI (zero_extend:DI (match_dup 1)) + (zero_extend:DI (match_dup 2))) + (reg:DI ARCV2_ACC)))] + "TARGET_PLUS_MACD" + "macdu %0,%1,%2" + [(set_attr "length" "4,4,8") + (set_attr "type" "multi") + (set_attr "predicable" "yes,no,no") + (set_attr "cond" "canuse,nocond,nocond")]) + +(define_insn "macu" + [(set (reg:DI ARCV2_ACC) + (plus:DI + (mult:DI (zero_extend:DI (match_operand:SI 0 "register_operand" "%r,r")) + (zero_extend:DI (match_operand:SI 1 "extend_operand" "rI,i"))) + (reg:DI ARCV2_ACC)))] + "TARGET_PLUS_DMPY" + "macu 0,%0,%1" + [(set_attr "length" "4,8") + (set_attr "type" "multi") + (set_attr "predicable" "no") + (set_attr "cond" "nocond")]) + +(define_peephole2 + [(set (reg:DI ARCV2_ACC) + (plus:DI + (mult:DI (zero_extend:DI (match_operand:SI 0 "register_operand" "")) + (zero_extend:DI (match_operand:SI 1 "extend_operand" ""))) + (reg:DI ARCV2_ACC))) + (set (match_operand:SI 2 "register_operand" "") + (match_operand:SI 3 "accl_operand" ""))] + "TARGET_PLUS_DMPY" + [(const_int 0)] + { + emit_insn (gen_macu_r (operands[2], operands[0], operands[1])); + DONE; + }) + +(define_insn "macu_r" + [(set (match_operand:SI 0 "register_operand" "=r,r") + (truncate:SI + (plus:DI + (mult:DI (zero_extend:DI (match_operand:SI 1 "register_operand" "%r,r")) + (zero_extend:DI (match_operand:SI 2 "extend_operand" "rI,i"))) + (reg:DI ARCV2_ACC)))) + (clobber (reg:DI ARCV2_ACC))] + "TARGET_PLUS_DMPY" + "macu %0,%1,%2" + [(set_attr "length" "4,8") + (set_attr "type" "multi") + (set_attr "predicable" "no") + (set_attr "cond" "nocond")]) + +(define_insn "mpyd_arcv2hs" + [(set (match_operand:DI 0 "even_register_operand" "=Rcr, r") + (mult:DI (sign_extend:DI (match_operand:SI 1 "register_operand" " 0, c")) + (sign_extend:DI (match_operand:SI 2 "register_operand" " c, c")))) + (set (reg:DI ARCV2_ACC) + (mult:DI + (sign_extend:DI (match_dup 1)) + (sign_extend:DI (match_dup 2))))] + "TARGET_PLUS_MACD" + "mpyd%? %0,%1,%2" + [(set_attr "length" "4,4") + (set_attr "iscompact" "false") + (set_attr "type" "multi") + (set_attr "predicable" "yes,no") + (set_attr "cond" "canuse,nocond")]) + +(define_insn "mpyd_imm_arcv2hs" + [(set (match_operand:DI 0 "even_register_operand" "=Rcr, r,r,Rcr, r") + (mult:DI (sign_extend:DI (match_operand:SI 1 "register_operand" " 0, c,0, 0, c")) + (match_operand 2 "immediate_operand" " L, L,I,Cal,Cal"))) + (set (reg:DI ARCV2_ACC) + (mult:DI (sign_extend:DI (match_dup 1)) + (match_dup 2)))] + "TARGET_PLUS_MACD" + "mpyd%? %0,%1,%2" + [(set_attr "length" "4,4,4,8,8") + (set_attr "iscompact" "false") + (set_attr "type" "multi") + (set_attr "predicable" "yes,no,no,yes,no") + (set_attr "cond" "canuse,nocond,nocond,canuse_limm,nocond")]) + +(define_insn "mpydu_arcv2hs" + [(set (match_operand:DI 0 "even_register_operand" "=Rcr, r") + (mult:DI (zero_extend:DI (match_operand:SI 1 "register_operand" " 0, c")) + (zero_extend:DI (match_operand:SI 2 "register_operand" " c, c")))) + (set (reg:DI ARCV2_ACC) + (mult:DI (zero_extend:DI (match_dup 1)) + (zero_extend:DI (match_dup 2))))] + "TARGET_PLUS_MACD" + "mpydu%? %0,%1,%2" + [(set_attr "length" "4,4") + (set_attr "iscompact" "false") + (set_attr "type" "multi") + (set_attr "predicable" "yes,no") + (set_attr "cond" "canuse,nocond")]) + +(define_insn "mpydu_imm_arcv2hs" + [(set (match_operand:DI 0 "even_register_operand" "=Rcr, r,r,Rcr, r") + (mult:DI (zero_extend:DI (match_operand:SI 1 "register_operand" " 0, c,0, 0, c")) + (match_operand 2 "immediate_operand" " L, L,I,Cal,Cal"))) + (set (reg:DI ARCV2_ACC) + (mult:DI (zero_extend:DI (match_dup 1)) + (match_dup 2)))] + "TARGET_PLUS_MACD" + "mpydu%? %0,%1,%2" + [(set_attr "length" "4,4,4,8,8") + (set_attr "iscompact" "false") + (set_attr "type" "multi") + (set_attr "predicable" "yes,no,no,yes,no") + (set_attr "cond" "canuse,nocond,nocond,canuse_limm,nocond")]) + ;; include the arc-FPX instructions (include "fpx.md") diff --git a/gcc/config/arc/predicates.md b/gcc/config/arc/predicates.md index f4c2a80..7ddec91 100644 --- a/gcc/config/arc/predicates.md +++ b/gcc/config/arc/predicates.md @@ -697,6 +697,11 @@ (and (match_code "reg") (match_test "REGNO (op) == (TARGET_BIG_ENDIAN ? 58 : 59)"))) +(define_predicate "accl_operand" + (and (match_code "reg") + (match_test "REGNO (op) == (TARGET_BIG_ENDIAN ? 59 : 58)") + (match_test "TARGET_V2"))) + ; Unfortunately, we can not allow a const_int_operand before reload, because ; reload needs a non-void mode to guide it how to reload the inside of a ; {sign_}extend. -- 1.9.1