From: rth Date: Tue, 9 Dec 1997 05:11:34 +0000 (+0000) Subject: * alpha.c (REG_RA, alpha_return_addr, output_epilog): X-Git-Url: http://git.sourceforge.jp/view?a=commitdiff_plain;h=d2832bd8a772fd3c047f863fc3970e9a132d72f2;p=pf3gnuchains%2Fgcc-fork.git * alpha.c (REG_RA, alpha_return_addr, output_epilog): Fix merge problems. * alpha.c (override_options): Don't know about scheduling for EV6. * alpha.md (ev5 function units): Don't overload as ev6. * alpha.c (alpha_adjust_cost): Simplify. Fix typo in ev5 mult case. * alpha.md (define_attr type): Add mvi. (ev5_e0): Define sceduling parameters for it. (TARGET_MAX insns): Type is mvi not shift. git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@17020 138bc75d-0d04-0410-961f-82ee72b054a4 --- diff --git a/gcc/ChangeLog b/gcc/ChangeLog index e584f737f8c..579746261c0 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,3 +1,16 @@ +Mon Dec 8 21:03:28 1997 Richard Henderson + + * alpha.c (REG_RA, alpha_return_addr, output_epilog): + Fix merge problems. + + * alpha.c (override_options): Don't know about scheduling for EV6. + * alpha.md (ev5 function units): Don't overload as ev6. + + * alpha.c (alpha_adjust_cost): Simplify. Fix typo in ev5 mult case. + * alpha.md (define_attr type): Add mvi. + (ev5_e0): Define sceduling parameters for it. + (TARGET_MAX insns): Type is mvi not shift. + Mon Dec 8 18:15:00 1997 Richard Henderson * alpha/win-nt.h (TRAMPOLINE_TEMPLATE): Fix backported gcc-2.8 bug. diff --git a/gcc/config/alpha/alpha.c b/gcc/config/alpha/alpha.c index 91c068c7dc3..0bf28c4db40 100644 --- a/gcc/config/alpha/alpha.c +++ b/gcc/config/alpha/alpha.c @@ -106,14 +106,23 @@ static void alpha_sa_mask PROTO((unsigned long *imaskP, #else #define NUM_ARGS current_function_args_info #endif + +#if OPEN_VMS +#define REG_PV 27 +#define REG_RA 26 +#else +#define REG_RA 26 +#endif /* Parse target option strings. */ void override_options () { + /* 971208 -- EV6 scheduling parameters are still secret, so don't even + pretend and just schedule for an EV5 for now. -- r~ */ alpha_cpu - = TARGET_CPU_DEFAULT & MASK_CPU_EV6 ? PROCESSOR_EV6 + = TARGET_CPU_DEFAULT & MASK_CPU_EV6 ? PROCESSOR_EV5 : (TARGET_CPU_DEFAULT & MASK_CPU_EV5 ? PROCESSOR_EV5 : PROCESSOR_EV4); if (alpha_cpu_string) @@ -138,7 +147,8 @@ override_options () target_flags &= ~ (MASK_CIX | MASK_MAX); } else if (! strcmp (alpha_cpu_string, "pca56") - || ! strcmp (alpha_cpu_string, "21164PC")) + || ! strcmp (alpha_cpu_string, "21164PC") + || ! strcmp (alpha_cpu_string, "21164pc")) { alpha_cpu = PROCESSOR_EV5; target_flags |= MASK_BWX | MASK_MAX; @@ -147,7 +157,7 @@ override_options () else if (! strcmp (alpha_cpu_string, "ev6") || ! strcmp (alpha_cpu_string, "21264")) { - alpha_cpu = PROCESSOR_EV6; + alpha_cpu = PROCESSOR_EV5; target_flags |= MASK_BWX | MASK_CIX | MASK_MAX; } else @@ -1191,6 +1201,7 @@ alpha_adjust_cost (insn, link, dep_insn, cost) int cost; { rtx set, set_src; + enum attr_type insn_type, dep_insn_type; /* If the dependence is an anti-dependence, there is no cost. For an output dependence, there is sometimes a cost, but it doesn't seem @@ -1199,17 +1210,22 @@ alpha_adjust_cost (insn, link, dep_insn, cost) if (REG_NOTE_KIND (link) != 0) return 0; + /* If we can't recognize the insns, we can't really do anything. */ + if (recog_memoized (insn) < 0 || recog_memoized (dep_insn) < 0) + return cost; + + insn_type = get_attr_type (insn); + dep_insn_type = get_attr_type (dep_insn); + if (alpha_cpu == PROCESSOR_EV5) { /* And the lord DEC saith: "A special bypass provides an effective latency of 0 cycles for an ICMP or ILOG insn producing the test operand of an IBR or CMOV insn." */ - if (recog_memoized (dep_insn) >= 0 - && (get_attr_type (dep_insn) == TYPE_ICMP - || get_attr_type (dep_insn) == TYPE_ILOG) - && recog_memoized (insn) >= 0 - && (get_attr_type (insn) == TYPE_IBR - || (get_attr_type (insn) == TYPE_CMOV + if ((dep_insn_type == TYPE_ICMP + || dep_insn_type == TYPE_ILOG) + && (insn_type == TYPE_IBR + || (insn_type == TYPE_CMOV && !((set = single_set (dep_insn)) != 0 && GET_CODE (PATTERN (insn)) == SET && (set_src = SET_SRC (PATTERN (insn)), @@ -1219,14 +1235,15 @@ alpha_adjust_cost (insn, link, dep_insn, cost) || rtx_equal_p (set, XEXP (set_src, 2))))))) return 0; - /* On EV5 it takes longer to get data to the multiplier than to - anywhere else, so increase costs. */ + /* "The multiplier is unable to receive data from IEU bypass paths. + The instruction issues at the expected time, but its latency is + increased by the time it takes for the input data to become + available to the multiplier" -- which happens in pipeline stage + six, when results are comitted to the register file. */ - if (recog_memoized (insn) >= 0 - && recog_memoized (dep_insn) >= 0 - && (get_attr_type (insn) == TYPE_IMULL - || get_attr_type (insn) == TYPE_IMULQ - || get_attr_type (insn) == TYPE_IMULH) + if ((insn_type == TYPE_IMULL + || insn_type == TYPE_IMULQ + || insn_type == TYPE_IMULH) && (set = single_set (dep_insn)) != 0 && GET_CODE (PATTERN (insn)) == SET && (set_src = SET_SRC (PATTERN (insn)), @@ -1235,19 +1252,19 @@ alpha_adjust_cost (insn, link, dep_insn, cost) rtx_equal_p (set, XEXP (set_src, 0)) || rtx_equal_p (set, XEXP (set_src, 1)))) { - switch (get_attr_type (insn)) + switch (dep_insn_type) { + /* These insns produce their results in pipeline stage five. */ case TYPE_LD: case TYPE_CMOV: case TYPE_IMULL: case TYPE_IMULQ: case TYPE_IMULH: + case TYPE_MVI: return cost + 1; - case TYPE_JSR: - case TYPE_IADD: - case TYPE_ILOG: - case TYPE_SHIFT: - case TYPE_ICMP: + + /* Other integer insns produce results in pipeline stage four. */ + default: return cost + 2; } } @@ -1257,12 +1274,12 @@ alpha_adjust_cost (insn, link, dep_insn, cost) /* On EV4, if INSN is a store insn and DEP_INSN is setting the data being stored, we can sometimes lower the cost. */ - if (recog_memoized (insn) >= 0 && get_attr_type (insn) == TYPE_ST + if (insn_type == TYPE_ST && (set = single_set (dep_insn)) != 0 && GET_CODE (PATTERN (insn)) == SET && rtx_equal_p (SET_DEST (set), SET_SRC (PATTERN (insn)))) { - switch (get_attr_type (dep_insn)) + switch (dep_insn_type) { case TYPE_LD: /* No savings here. */ @@ -1285,11 +1302,9 @@ alpha_adjust_cost (insn, link, dep_insn, cost) two in the MD file. The only case that it is actually two is for the address in loads and stores. */ - if (recog_memoized (dep_insn) >= 0 - && (get_attr_type (dep_insn) == TYPE_IADD - || get_attr_type (dep_insn) == TYPE_ILOG)) + if (dep_insn_type == TYPE_IADD || dep_insn_type == TYPE_ILOG) { - switch (get_attr_type (insn)) + switch (insn_type) { case TYPE_LD: case TYPE_ST: @@ -1302,10 +1317,7 @@ alpha_adjust_cost (insn, link, dep_insn, cost) /* The final case is when a compare feeds into an integer branch; the cost is only one cycle in that case. */ - if (recog_memoized (dep_insn) >= 0 - && get_attr_type (dep_insn) == TYPE_ICMP - && recog_memoized (insn) >= 0 - && get_attr_type (insn) == TYPE_IBR) + if (dep_insn_type == TYPE_ICMP && insn_type == TYPE_IBR) return 1; } @@ -1781,39 +1793,6 @@ alpha_builtin_saveregs (arglist) } } -#if OPEN_VMS -#define REG_PV 27 -#define REG_RA 26 -#else -#define REG_RA 26 -#endif - -/* Find the current function's return address. - - ??? It would be better to arrange things such that if we would ordinarily - have been a leaf function and we didn't spill the hard reg that we - wouldn't have to save the register in the prolog. But it's not clear - how to get the right information at the right time. */ - -static rtx alpha_return_addr_rtx; - -rtx -alpha_return_addr () -{ - rtx ret; - - if ((ret = alpha_return_addr_rtx) == NULL) - { - alpha_return_addr_rtx = ret = gen_reg_rtx (Pmode); - - emit_insn_after (gen_rtx (SET, VOIDmode, ret, - gen_rtx (REG, Pmode, REG_RA)), - get_insns ()); - } - - return ret; -} - /* This page contains routines that are used to determine what the function prologue and epilogue code will do and write them out. */ @@ -2725,8 +2704,6 @@ output_epilog (file, size) /* Show that we know this function if it is called again. */ SYMBOL_REF_FLAG (XEXP (DECL_RTL (current_function_decl), 0)) = 1; - - alpha_return_addr_rtx = 0; } #endif /* !OPEN_VMS */ diff --git a/gcc/config/alpha/alpha.md b/gcc/config/alpha/alpha.md index 2e6b7b73c30..6e24c31bb0b 100644 --- a/gcc/config/alpha/alpha.md +++ b/gcc/config/alpha/alpha.md @@ -24,7 +24,7 @@ ;; Processor type -- this attribute must exactly match the processor_type ;; enumeration in alpha.h. -(define_attr "cpu" "ev4,ev5,ev6" +(define_attr "cpu" "ev4,ev5" (const (symbol_ref "alpha_cpu"))) ;; Define an insn type attribute. This is used in function unit delay @@ -33,7 +33,7 @@ ;; separately. (define_attr "type" - "ld,st,ibr,fbr,jsr,iadd,ilog,shift,cmov,icmp,imull,imulq,imulh,fadd,fmul,fcpys,fdivs,fdivt,ldsym,misc" + "ld,st,ibr,fbr,jsr,iadd,ilog,shift,cmov,icmp,imull,imulq,imulh,fadd,fmul,fcpys,fdivs,fdivt,ldsym,misc,mvi" (const_string "iadd")) ;; The TRAP_TYPE attribute marks instructions that may generate traps @@ -122,60 +122,66 @@ ;; those as well. (define_function_unit "ev5_ebox" 2 0 - (and (eq_attr "cpu" "ev5,ev6") - (eq_attr "type" "iadd,ilog,icmp,st,shift,imull,imulq,imulh")) + (and (eq_attr "cpu" "ev5") + (eq_attr "type" "iadd,ilog,icmp,st,shift,imull,imulq,imulh,mvi")) 1 1) ;; Memory takes at least 2 clocks, and load cannot dual issue with stores. (define_function_unit "ev5_ebox" 2 0 - (and (eq_attr "cpu" "ev5,ev6") + (and (eq_attr "cpu" "ev5") (eq_attr "type" "ld,ldsym")) 2 1) (define_function_unit "ev5_e0" 1 0 - (and (eq_attr "cpu" "ev5,ev6") + (and (eq_attr "cpu" "ev5") (eq_attr "type" "ld,ldsym")) 0 1 [(eq_attr "type" "st")]) ;; Conditional moves always take 2 ticks. (define_function_unit "ev5_ebox" 2 0 - (and (eq_attr "cpu" "ev5,ev6") + (and (eq_attr "cpu" "ev5") (eq_attr "type" "cmov")) 2 1) -;; Stores, shifts, and multiplies can only issue to E0 +;; Stores, shifts, multiplies can only issue to E0 (define_function_unit "ev5_e0" 1 0 - (and (eq_attr "cpu" "ev5,ev6") + (and (eq_attr "cpu" "ev5") (eq_attr "type" "st")) 1 1) +;; Motion video insns also issue only to E0, and take two ticks. +(define_function_unit "ev5_e0" 1 0 + (and (eq_attr "cpu" "ev5") + (eq_attr "type" "mvi")) + 2 1) + ;; But shifts and multiplies don't conflict with loads. (define_function_unit "ev5_e0" 1 0 - (and (eq_attr "cpu" "ev5,ev6") - (eq_attr "type" "shift,imull,imulq,imulh")) + (and (eq_attr "cpu" "ev5") + (eq_attr "type" "shift,imull,imulq,imulh,mvi")) 1 1 - [(eq_attr "type" "st,shift,imull,imulq,imulh")]) + [(eq_attr "type" "st,shift,imull,imulq,imulh,mvi")]) ;; Branches can only issue to E1 (define_function_unit "ev5_e1" 1 0 - (and (eq_attr "cpu" "ev5,ev6") + (and (eq_attr "cpu" "ev5") (eq_attr "type" "ibr,jsr")) 1 1) ;; Multiplies also use the integer multiplier. (define_function_unit "ev5_imult" 1 0 - (and (eq_attr "cpu" "ev5,ev6") + (and (eq_attr "cpu" "ev5") (eq_attr "type" "imull")) 8 4) (define_function_unit "ev5_imult" 1 0 - (and (eq_attr "cpu" "ev5,ev6") + (and (eq_attr "cpu" "ev5") (eq_attr "type" "imulq")) 12 8) (define_function_unit "ev5_imult" 1 0 - (and (eq_attr "cpu" "ev5,ev6") + (and (eq_attr "cpu" "ev5") (eq_attr "type" "imulh")) 14 8) @@ -183,33 +189,33 @@ ;; on either so we have to play the game again. (define_function_unit "ev5_fpu" 2 0 - (and (eq_attr "cpu" "ev5,ev6") + (and (eq_attr "cpu" "ev5") (eq_attr "type" "fadd,fmul,fcpys,fbr,fdivs,fdivt")) 4 1) ;; Multiplies (resp. adds) also use the fmul (resp. fadd) units. (define_function_unit "ev5_fm" 1 0 - (and (eq_attr "cpu" "ev5,ev6") + (and (eq_attr "cpu" "ev5") (eq_attr "type" "fmul")) 4 1) (define_function_unit "ev5_fa" 1 0 - (and (eq_attr "cpu" "ev5,ev6") + (and (eq_attr "cpu" "ev5") (eq_attr "type" "fadd")) 4 1) (define_function_unit "ev5_fa" 1 0 - (and (eq_attr "cpu" "ev5,ev6") + (and (eq_attr "cpu" "ev5") (eq_attr "type" "fbr")) 1 1) (define_function_unit "ev5_fa" 1 0 - (and (eq_attr "cpu" "ev5,ev6") + (and (eq_attr "cpu" "ev5") (eq_attr "type" "fdivs")) 15 1) (define_function_unit "ev5_fa" 1 0 - (and (eq_attr "cpu" "ev5,ev6") + (and (eq_attr "cpu" "ev5") (eq_attr "type" "fdivt")) 22 1) @@ -2069,7 +2075,7 @@ (match_operand:QI 2 "reg_or_8bit_operand" "rI")))] "TARGET_MAX" "minsb8 %r1,%2,%0" - [(set_attr "type" "shift")]) + [(set_attr "type" "mvi")]) (define_insn "uminqi3" [(set (match_operand:QI 0 "register_operand" "=r") @@ -2077,7 +2083,7 @@ (match_operand:QI 2 "reg_or_8bit_operand" "rI")))] "TARGET_MAX" "minub8 %r1,%2,%0" - [(set_attr "type" "shift")]) + [(set_attr "type" "mvi")]) (define_insn "smaxqi3" [(set (match_operand:QI 0 "register_operand" "=r") @@ -2085,7 +2091,7 @@ (match_operand:QI 2 "reg_or_8bit_operand" "rI")))] "TARGET_MAX" "maxsb8 %r1,%2,%0" - [(set_attr "type" "shift")]) + [(set_attr "type" "mvi")]) (define_insn "umaxqi3" [(set (match_operand:QI 0 "register_operand" "=r") @@ -2093,7 +2099,7 @@ (match_operand:QI 2 "reg_or_8bit_operand" "rI")))] "TARGET_MAX" "maxub8 %r1,%2,%0" - [(set_attr "type" "shift")]) + [(set_attr "type" "mvi")]) (define_insn "sminhi3" [(set (match_operand:HI 0 "register_operand" "=r") @@ -2101,7 +2107,7 @@ (match_operand:HI 2 "reg_or_8bit_operand" "rI")))] "TARGET_MAX" "minsw4 %r1,%2,%0" - [(set_attr "type" "shift")]) + [(set_attr "type" "mvi")]) (define_insn "uminhi3" [(set (match_operand:HI 0 "register_operand" "=r") @@ -2109,7 +2115,7 @@ (match_operand:HI 2 "reg_or_8bit_operand" "rI")))] "TARGET_MAX" "minuw4 %r1,%2,%0" - [(set_attr "type" "shift")]) + [(set_attr "type" "mvi")]) (define_insn "smaxhi3" [(set (match_operand:HI 0 "register_operand" "=r") @@ -2117,7 +2123,7 @@ (match_operand:HI 2 "reg_or_8bit_operand" "rI")))] "TARGET_MAX" "maxsw4 %r1,%2,%0" - [(set_attr "type" "shift")]) + [(set_attr "type" "mvi")]) (define_insn "umaxhi3" [(set (match_operand:HI 0 "register_operand" "=r")