There are 3 patches left in the basic IEEE 128-bit floating point support for the compiler. I will submit these at the same time. They are split to make the review process similar. Patch #5 and #6 are indpendent of each other and can be applied in either order. Patch #7 assumes that patches 1-6 have been applied.
Patch #5 adds the following: * Support for the reload handlers that will be enabled in patch #7. * Adds IFmode/KFmode to other iterators as appropriate. * Adds the basic negate, absolute value, and negative absolute value support. * Adds the insns for the 128-bit pack/unpack routines. Patch #6 adds the following: * Adds support for comparisons. * Updates the cannot change mode support. Patch #7 finishes up the initial basic support. * It defines macros for IEEE 128-bit floating point users. * It defines the basic move support. * It sets up the calling sequence. * It registers the __float128 and __ibm128 keywords. * It sets up the various handler functions. * It adds 'q' and 'Q' as the suffix for IEEE 128-bit floating point. * It adds target attribute/pragma support for the IEEE 128-bit options. * It treats IEEE 128-bit in VSX register modes as vector. * It uses a unique mangling for IEEE 128-bit in VSX registers. * It moves vector modes tieable above scalar floating point. * It adds a simple minded test to make sure IEEE args are passed as vectors. Things to be done: * Work with GDB to add debug support. * Work with GLIBC to add basic software emulation support. * Work with GLIBC on other IEEE 128-bit support. * Look into Complex support. * Look into libquadmath support. * Enable -mfloat128-software if -mvsx. * Add more tests. * Fix bugs that show up if -mabi=ieeelongdouble is used. Each patch bootstraps without error and has no regressions. Are they ok to install in the trunk? This is patch #5: 2015-08-13 Michael Meissner <meiss...@linux.vnet.ibm.com> * config/rs6000/predicates.md (int_reg_operand_not_pseudo): New predicate for only GPR hard registers. * config/rs6000/rs6000.md (FP): Add IEEE 128-bit floating point modes to iterators. Add new iterators for moving 128-bit values in scalar FPR registers and VSX registers. (FMOVE128): Likewise. (FMOVE128_FPR): Likewise. (FMOVE128_GPR): Likewise. (FMOVE128_VSX): Likewise. (FLOAT128_SFDFTF): New iterators for IEEE 128-bit floating point in VSX registers. (IFKF): Likewise. (IBM128): Likewise. (TFIFKF): Likewise. (RELOAD): Add IEEE 128-bit floating point modes. (signbittf2): Convert TF insns to add support for new IEEE 128-bit floating point in VSX registers modes. (signbit<mode>2, IBM128 iterator): Likewise. (mov<mode>_64bit_dm, FMOVE128_FPR iterator): Likewise. (mov<mode>_32bit, FMOVE128_FPR iterator): Likewise. (negtf2): Likewise. (neg<mode>2, TFIFKF iterator): Likewise. (negtf2_internal): Likewise. (abstf2): Likewise. (abs<mode>2, TFIFKF iterator): Likewise. (ieee_128bit_negative_zero): New IEEE 128-bit floating point in VSX insn support for negate, absolute value, and negative absolute value. (ieee_128bit_vsx_neg<mode>2): Likewise. (ieee_128bit_vsx_neg<mode>2_internal): Likewise. (ieee_128bit_vsx_abs<mode>2): Likewise. (ieee_128bit_vsx_abs<mode>2_internal): Likewise. (ieee_128bit_vsx_nabs<mode>2): Likewise. (ieee_128bit_vsx_nabs<mode>2_internal): Likewise. (FP128_64): Update pack/unpack 128-bit insns for IEEE 128-bit floating point in VSX registers. (unpack<mode>_dm): Likewise. (unpack<mode>_nodm): Likewise. (pack<mode>): Likewise. (unpackv1ti): Likewise. (unpack<mode>, FMOVE128_VSX iterator): Likewise. (packv1ti): Likewise. (pack<mode>, FMOVE128_VSX iterator): Likewise. -- Michael Meissner, IBM IBM, M/S 2506R, 550 King Street, Littleton, MA 01460-6245, USA email: meiss...@linux.vnet.ibm.com, phone: +1 (978) 899-4797
Index: gcc/config/rs6000/predicates.md =================================================================== --- gcc/config/rs6000/predicates.md (revision 226869) +++ gcc/config/rs6000/predicates.md (working copy) @@ -239,6 +239,25 @@ (define_predicate "int_reg_operand" return INT_REGNO_P (REGNO (op)); }) +;; Like int_reg_operand, but don't return true for pseudo registers +(define_predicate "int_reg_operand_not_pseudo" + (match_operand 0 "register_operand") +{ + if ((TARGET_E500_DOUBLE || TARGET_SPE) && invalid_e500_subreg (op, mode)) + return 0; + + if (GET_CODE (op) == SUBREG) + op = SUBREG_REG (op); + + if (!REG_P (op)) + return 0; + + if (REGNO (op) >= FIRST_PSEUDO_REGISTER) + return 0; + + return INT_REGNO_P (REGNO (op)); +}) + ;; Like int_reg_operand, but only return true for base registers (define_predicate "base_reg_operand" (match_operand 0 "int_reg_operand") Index: gcc/config/rs6000/rs6000.md =================================================================== --- gcc/config/rs6000/rs6000.md (revision 226869) +++ gcc/config/rs6000/rs6000.md (working copy) @@ -348,6 +348,8 @@ (define_mode_iterator FP [ && TARGET_HARD_FLOAT && (TARGET_FPRS || TARGET_E500_DOUBLE) && TARGET_LONG_DOUBLE_128") + (IF "TARGET_FLOAT128") + (KF "TARGET_FLOAT128") (DD "TARGET_DFP") (TD "TARGET_DFP")]) @@ -365,9 +367,14 @@ (define_mode_iterator FMA_F [ (define_mode_iterator FMOVE32 [SF SD]) (define_mode_iterator FMOVE64 [DF DD]) (define_mode_iterator FMOVE64X [DI DF DD]) -(define_mode_iterator FMOVE128 [(TF "!TARGET_IEEEQUAD && TARGET_LONG_DOUBLE_128") +(define_mode_iterator FMOVE128 [(TF "TARGET_LONG_DOUBLE_128") + (IF "TARGET_LONG_DOUBLE_128") (TD "TARGET_HARD_FLOAT && TARGET_FPRS")]) +(define_mode_iterator FMOVE128_FPR [(TF "FLOAT128_2REG_P (TFmode)") + (IF "FLOAT128_2REG_P (IFmode)") + (TD "TARGET_HARD_FLOAT && TARGET_FPRS")]) + ; Iterators for 128 bit types for direct move (define_mode_iterator FMOVE128_GPR [(TI "TARGET_VSX_TIMODE") (V16QI "") @@ -376,7 +383,13 @@ (define_mode_iterator FMOVE128_GPR [(TI (V4SF "") (V2DI "") (V2DF "") - (V1TI "")]) + (V1TI "") + (KF "") + (TF "") + (IF "")]) + +; Iterator for 128-bit VSX types for pack/unpack +(define_mode_iterator FMOVE128_VSX [V1TI KF]) ; Whether a floating point move is ok, don't allow SD without hardware FP (define_mode_attr fmove_ok [(SF "") @@ -432,6 +445,25 @@ (define_mode_iterator RECIPF [SF DF V4SF ; Iterator for just SF/DF (define_mode_iterator SFDF [SF DF]) +; Iterator for float128 floating conversions +(define_mode_iterator FLOAT128_SFDFTF [ + (SF "TARGET_FLOAT128") + (DF "TARGET_FLOAT128") + (TF "FLOAT128_IBM_P (TFmode)") + (IF "TARGET_FLOAT128")]) + +; Iterator for special 128-bit floating point. This is for non-default +; conversions, so TFmode is not used here. +(define_mode_iterator IFKF [IF KF]) + +; Iterator for 128-bit floating point that uses the IBM double-double format +(define_mode_iterator IBM128 [IF TF]) + +; Iterator for 128-bit floating point +(define_mode_iterator TFIFKF [(KF "TARGET_FLOAT128") + (IF "TARGET_FLOAT128") + (TF "TARGET_LONG_DOUBLE_128")]) + ; SF/DF suffix for traditional floating instructions (define_mode_attr Ftrad [(SF "s") (DF "")]) @@ -596,7 +628,7 @@ (define_mode_attr BOOL_REGS_UNARY [(TI " ;; Reload iterator for creating the function to allocate a base register to ;; supplement addressing modes. (define_mode_iterator RELOAD [V16QI V8HI V4SI V2DI V4SF V2DF V1TI - SF SD SI DF DD DI TI PTI]) + SF SD SI DF DD DI TI PTI KF IF TF]) ;; Start with fixed-point load and store insns. Here we put only the more @@ -4216,19 +4248,18 @@ (define_insn "*truncdfsf2_fpr" ;; This expander is here to avoid FLOAT_WORDS_BIGENDIAN tests in ;; builtins.c and optabs.c that are not correct for IBM long double ;; when little-endian. -(define_expand "signbittf2" +(define_expand "signbit<mode>2" [(set (match_dup 2) - (float_truncate:DF (match_operand:TF 1 "gpc_reg_operand" ""))) + (float_truncate:DF (match_operand:IBM128 1 "gpc_reg_operand" ""))) (set (match_dup 3) (subreg:DI (match_dup 2) 0)) (set (match_dup 4) (match_dup 5)) (set (match_operand:SI 0 "gpc_reg_operand" "") (match_dup 6))] - "!TARGET_IEEEQUAD + "FLOAT128_IBM_P (<MODE>mode) && TARGET_HARD_FLOAT - && (TARGET_FPRS || TARGET_E500_DOUBLE) - && TARGET_LONG_DOUBLE_128" + && (TARGET_FPRS || TARGET_E500_DOUBLE)" { operands[2] = gen_reg_rtx (DFmode); operands[3] = gen_reg_rtx (DImode); @@ -6402,9 +6433,10 @@ (define_expand "mov<mode>" ;; problematical. Don't allow direct move for this case. (define_insn_and_split "*mov<mode>_64bit_dm" - [(set (match_operand:FMOVE128 0 "nonimmediate_operand" "=m,d,d,ws,Y,r,r,r,wm") - (match_operand:FMOVE128 1 "input_operand" "d,m,d,j,r,jY,r,wm,r"))] + [(set (match_operand:FMOVE128_FPR 0 "nonimmediate_operand" "=m,d,d,ws,Y,r,r,r,wm") + (match_operand:FMOVE128_FPR 1 "input_operand" "d,m,d,j,r,jY,r,wm,r"))] "TARGET_HARD_FLOAT && TARGET_FPRS && TARGET_POWERPC64 + && FLOAT128_2REG_P (<MODE>mode) && (<MODE>mode != TDmode || WORDS_BIG_ENDIAN) && (gpc_reg_operand (operands[0], <MODE>mode) || gpc_reg_operand (operands[1], <MODE>mode))" @@ -6427,9 +6459,12 @@ (define_insn_and_split "*movtd_64bit_nod [(set_attr "length" "8,8,8,8,12,12,8")]) (define_insn_and_split "*mov<mode>_32bit" - [(set (match_operand:FMOVE128 0 "nonimmediate_operand" "=m,d,d,ws,Y,r,r") - (match_operand:FMOVE128 1 "input_operand" "d,m,d,j,r,jY,r"))] + [(set (match_operand:FMOVE128_FPR 0 "nonimmediate_operand" "=m,d,d,ws,Y,r,r") + (match_operand:FMOVE128_FPR 1 "input_operand" "d,m,d,j,r,jY,r"))] "TARGET_HARD_FLOAT && TARGET_FPRS && !TARGET_POWERPC64 + && (FLOAT128_2REG_P (<MODE>mode) + || int_reg_operand_not_pseudo (operands[0], <MODE>mode) + || int_reg_operand_not_pseudo (operands[1], <MODE>mode)) && (gpc_reg_operand (operands[0], <MODE>mode) || gpc_reg_operand (operands[1], <MODE>mode))" "#" @@ -6663,20 +6698,37 @@ (define_insn_and_split "*fix_trunctfsi2_ DONE; }) -(define_expand "negtf2" - [(set (match_operand:TF 0 "gpc_reg_operand" "") - (neg:TF (match_operand:TF 1 "gpc_reg_operand" "")))] - "!TARGET_IEEEQUAD - && TARGET_HARD_FLOAT - && (TARGET_FPRS || TARGET_E500_DOUBLE) - && TARGET_LONG_DOUBLE_128" - "") +(define_expand "neg<mode>2" + [(set (match_operand:TFIFKF 0 "gpc_reg_operand" "") + (neg:TFIFKF (match_operand:TFIFKF 1 "gpc_reg_operand" "")))] + "FLOAT128_IEEE_P (<MODE>mode) + || (FLOAT128_IBM_P (<MODE>mode) + && TARGET_HARD_FLOAT + && (TARGET_FPRS || TARGET_E500_DOUBLE))" + " +{ + if (FLOAT128_IEEE_P (<MODE>mode)) + { + if (TARGET_FLOAT128) + emit_insn (gen_ieee_128bit_vsx_neg<mode>2 (operands[0], operands[1])); + else + { + rtx libfunc = optab_libfunc (neg_optab, <MODE>mode); + rtx target = emit_library_call_value (libfunc, operands[0], LCT_CONST, + <MODE>mode, 1, + operands[1], <MODE>mode); + + if (target && !rtx_equal_p (target, operands[0])) + emit_move_insn (operands[0], target); + } + DONE; + } +}") (define_insn "negtf2_internal" [(set (match_operand:TF 0 "gpc_reg_operand" "=d") (neg:TF (match_operand:TF 1 "gpc_reg_operand" "d")))] - "!TARGET_IEEEQUAD - && TARGET_HARD_FLOAT && TARGET_FPRS && TARGET_LONG_DOUBLE_128" + "TARGET_HARD_FLOAT && TARGET_FPRS && FLOAT128_IBM_P (TFmode)" "* { if (REGNO (operands[0]) == REGNO (operands[1]) + 1) @@ -6687,16 +6739,29 @@ (define_insn "negtf2_internal" [(set_attr "type" "fp") (set_attr "length" "8")]) -(define_expand "abstf2" - [(set (match_operand:TF 0 "gpc_reg_operand" "") - (abs:TF (match_operand:TF 1 "gpc_reg_operand" "")))] - "!TARGET_IEEEQUAD - && TARGET_HARD_FLOAT - && (TARGET_FPRS || TARGET_E500_DOUBLE) - && TARGET_LONG_DOUBLE_128" +(define_expand "abs<mode>2" + [(set (match_operand:TFIFKF 0 "gpc_reg_operand" "") + (abs:TFIFKF (match_operand:TFIFKF 1 "gpc_reg_operand" "")))] + "FLOAT128_IEEE_P (<MODE>mode) + || (FLOAT128_IBM_P (<MODE>mode) + && TARGET_HARD_FLOAT + && (TARGET_FPRS || TARGET_E500_DOUBLE))" " { - rtx label = gen_label_rtx (); + rtx label; + + if (FLOAT128_IEEE_P (<MODE>mode)) + { + if (TARGET_FLOAT128) + { + emit_insn (gen_ieee_128bit_vsx_abs<mode>2 (operands[0], operands[1])); + DONE; + } + else + FAIL; + } + + label = gen_label_rtx (); if (TARGET_E500_DOUBLE) { if (flag_finite_math_only && !flag_trapping_math) @@ -6732,6 +6797,129 @@ (define_expand "abstf2_internal" operands[5] = simplify_gen_subreg (DFmode, operands[0], TFmode, hi_word); operands[6] = simplify_gen_subreg (DFmode, operands[0], TFmode, lo_word); }") + + +;; Generate IEEE 128-bit -0.0 (0x80000000000000000000000000000000) in a vector +;; register + +(define_expand "ieee_128bit_negative_zero" + [(set (match_operand:V16QI 0 "register_operand" "") (match_dup 1))] + "TARGET_FLOAT128" +{ + rtvec v = rtvec_alloc (16); + int i, high; + + for (i = 0; i < 16; i++) + RTVEC_ELT (v, i) = const0_rtx; + + high = (BYTES_BIG_ENDIAN) ? 0 : 15; + RTVEC_ELT (v, high) = GEN_INT (0x80); + + rs6000_expand_vector_init (operands[0], gen_rtx_PARALLEL (V16QImode, v)); + DONE; +}) + +;; IEEE 128-bit negate + +;; We have 2 insns here for negate and absolute value. The first uses +;; match_scratch so that phases like combine can recognize neg/abs as generic +;; insns, and second insn after the first split pass loads up the bit to +;; twiddle the sign bit. Later GCSE passes can then combine multiple uses of +;; neg/abs to create the constant just once. + +(define_insn_and_split "ieee_128bit_vsx_neg<mode>2" + [(set (match_operand:TFIFKF 0 "register_operand" "=wa") + (neg:TFIFKF (match_operand:TFIFKF 1 "register_operand" "wa"))) + (clobber (match_scratch:V16QI 2 "=v"))] + "TARGET_FLOAT128 && FLOAT128_IEEE_P (<MODE>mode)" + "#" + "&& 1" + [(parallel [(set (match_dup 0) + (neg:TFIFKF (match_dup 1))) + (use (match_dup 2))])] +{ + if (GET_CODE (operands[2]) == SCRATCH) + operands[2] = gen_reg_rtx (V16QImode); + + operands[3] = gen_reg_rtx (V16QImode); + emit_insn (gen_ieee_128bit_negative_zero (operands[2])); +} + [(set_attr "length" "8") + (set_attr "type" "vecsimple")]) + +(define_insn "*ieee_128bit_vsx_neg<mode>2_internal" + [(set (match_operand:TFIFKF 0 "register_operand" "=wa") + (neg:TFIFKF (match_operand:TFIFKF 1 "register_operand" "wa"))) + (use (match_operand:V16QI 2 "register_operand" "=v"))] + "TARGET_FLOAT128" + "xxlxor %x0,%x1,%x2" + [(set_attr "length" "4") + (set_attr "type" "vecsimple")]) + +;; IEEE 128-bit absolute value +(define_insn_and_split "ieee_128bit_vsx_abs<mode>2" + [(set (match_operand:TFIFKF 0 "register_operand" "=wa") + (abs:TFIFKF (match_operand:TFIFKF 1 "register_operand" "wa"))) + (clobber (match_scratch:V16QI 2 "=v"))] + "TARGET_FLOAT128 && FLOAT128_IEEE_P (<MODE>mode)" + "#" + "&& 1" + [(parallel [(set (match_dup 0) + (abs:TFIFKF (match_dup 1))) + (use (match_dup 2))])] +{ + if (GET_CODE (operands[2]) == SCRATCH) + operands[2] = gen_reg_rtx (V16QImode); + + operands[3] = gen_reg_rtx (V16QImode); + emit_insn (gen_ieee_128bit_negative_zero (operands[2])); +} + [(set_attr "length" "8") + (set_attr "type" "vecsimple")]) + +(define_insn "*ieee_128bit_vsx_abs<mode>2_internal" + [(set (match_operand:TFIFKF 0 "register_operand" "=wa") + (abs:TFIFKF (match_operand:TFIFKF 1 "register_operand" "wa"))) + (use (match_operand:V16QI 2 "register_operand" "=v"))] + "TARGET_FLOAT128" + "xxlandc %x0,%x1,%x2" + [(set_attr "length" "4") + (set_attr "type" "vecsimple")]) + +;; IEEE 128-bit negative absolute value +(define_insn_and_split "*ieee_128bit_vsx_nabs<mode>2" + [(set (match_operand:TFIFKF 0 "register_operand" "=wa") + (neg:TFIFKF + (abs:TFIFKF + (match_operand:TFIFKF 1 "register_operand" "wa")))) + (clobber (match_scratch:V16QI 2 "=v"))] + "TARGET_FLOAT128 && FLOAT128_IEEE_P (<MODE>mode)" + "#" + "&& 1" + [(parallel [(set (match_dup 0) + (abs:TFIFKF (match_dup 1))) + (use (match_dup 2))])] +{ + if (GET_CODE (operands[2]) == SCRATCH) + operands[2] = gen_reg_rtx (V16QImode); + + operands[3] = gen_reg_rtx (V16QImode); + emit_insn (gen_ieee_128bit_negative_zero (operands[2])); +} + [(set_attr "length" "8") + (set_attr "type" "vecsimple")]) + +(define_insn "*ieee_128bit_vsx_nabs<mode>2_internal" + [(set (match_operand:TFIFKF 0 "register_operand" "=wa") + (neg:TFIFKF + (abs:TFIFKF + (match_operand:TFIFKF 1 "register_operand" "wa")))) + (use (match_operand:V16QI 2 "register_operand" "=v"))] + "TARGET_FLOAT128" + "xxlor %x0,%x1,%x2" + [(set_attr "length" "4") + (set_attr "type" "vecsimple")]) + ;; Reload helper functions used by rs6000_secondary_reload. The patterns all ;; must have 3 arguments, and scratch register constraint must be a single @@ -12134,7 +12322,10 @@ (define_insn "div<div_extend>_<mode>" ;; Pack/unpack 128-bit floating point types that take 2 scalar registers ; Type of the 64-bit part when packing/unpacking 128-bit floating point types -(define_mode_attr FP128_64 [(TF "DF") (TD "DI")]) +(define_mode_attr FP128_64 [(TF "DF") + (IF "DF") + (TD "DI") + (KF "DI")]) (define_expand "unpack<mode>" [(set (match_operand:<FP128_64> 0 "nonimmediate_operand" "") @@ -12142,7 +12333,7 @@ (define_expand "unpack<mode>" [(match_operand:FMOVE128 1 "register_operand" "") (match_operand:QI 2 "const_0_to_1_operand" "")] UNSPEC_UNPACK_128BIT))] - "" + "FLOAT128_2REG_P (<MODE>mode)" "") (define_insn_and_split "unpack<mode>_dm" @@ -12151,7 +12342,7 @@ (define_insn_and_split "unpack<mode>_dm" [(match_operand:FMOVE128 1 "register_operand" "d,d,r,d,r") (match_operand:QI 2 "const_0_to_1_operand" "i,i,i,i,i")] UNSPEC_UNPACK_128BIT))] - "TARGET_POWERPC64 && TARGET_DIRECT_MOVE" + "TARGET_POWERPC64 && TARGET_DIRECT_MOVE && FLOAT128_2REG_P (<MODE>mode)" "#" "&& reload_completed" [(set (match_dup 0) (match_dup 3))] @@ -12175,7 +12366,7 @@ (define_insn_and_split "unpack<mode>_nod [(match_operand:FMOVE128 1 "register_operand" "d,d") (match_operand:QI 2 "const_0_to_1_operand" "i,i")] UNSPEC_UNPACK_128BIT))] - "!TARGET_POWERPC64 || !TARGET_DIRECT_MOVE" + "(!TARGET_POWERPC64 || !TARGET_DIRECT_MOVE) && FLOAT128_2REG_P (<MODE>mode)" "#" "&& reload_completed" [(set (match_dup 0) (match_dup 3))] @@ -12199,7 +12390,7 @@ (define_insn_and_split "pack<mode>" [(match_operand:<FP128_64> 1 "register_operand" "0,d") (match_operand:<FP128_64> 2 "register_operand" "d,d")] UNSPEC_PACK_128BIT))] - "" + "FLOAT128_2REG_P (<MODE>mode)" "@ fmr %L0,%2 #" @@ -12219,12 +12410,12 @@ (define_insn_and_split "pack<mode>" [(set_attr "type" "fp,fp") (set_attr "length" "4,8")]) -(define_insn "unpackv1ti" +(define_insn "unpack<mode>" [(set (match_operand:DI 0 "register_operand" "=d,d") - (unspec:DI [(match_operand:V1TI 1 "register_operand" "0,wa") + (unspec:DI [(match_operand:FMOVE128_VSX 1 "register_operand" "0,wa") (match_operand:QI 2 "const_0_to_1_operand" "O,i")] UNSPEC_UNPACK_128BIT))] - "TARGET_VSX" + "VECTOR_MEM_ALTIVEC_OR_VSX_P (<MODE>mode)" { if (REGNO (operands[0]) == REGNO (operands[1]) && INTVAL (operands[2]) == 0) return ASM_COMMENT_START " xxpermdi to same register"; @@ -12235,9 +12426,9 @@ (define_insn "unpackv1ti" [(set_attr "type" "vecperm") (set_attr "length" "4")]) -(define_insn "packv1ti" - [(set (match_operand:V1TI 0 "register_operand" "=wa") - (unspec:V1TI +(define_insn "pack<mode>" + [(set (match_operand:FMOVE128_VSX 0 "register_operand" "=wa") + (unspec:FMOVE128_VSX [(match_operand:DI 1 "register_operand" "d") (match_operand:DI 2 "register_operand" "d")] UNSPEC_PACK_128BIT))]