From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: by sourceware.org (Postfix, from userid 1005) id 696D63858C83; Sat, 29 Oct 2022 03:16:51 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 696D63858C83 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org; s=default; t=1667013411; bh=nHB51Qw4kdc+i4/DzzRF3h+Ml8u/4UsDShvFjYeCSMk=; h=From:To:Subject:Date:From; b=rduEy/kmudWx1BdbpEz5+d0hh3XAj1SMO/y9q4Qj1sgG+4qj52ohztOeZEczwBIwu uVOHDaIF9JA10C8lEeWmgCCYK8ntoOMgsRshnu3GcEDlCf57FYPG9zjNSPpyxAaexS nasKPPl6MyzzHgepiMu8NnyTocuGmXn7qKyY9Ypc= Content-Type: text/plain; charset="us-ascii" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit From: Michael Meissner To: gcc-cvs@gcc.gnu.org Subject: [gcc(refs/users/meissner/heads/work101)] Update float 128-bit conversions. X-Act-Checkin: gcc X-Git-Author: Michael Meissner X-Git-Refname: refs/users/meissner/heads/work101 X-Git-Oldrev: 02b4f8e5f236b645649c6bf53e8b3d45a05a2ce1 X-Git-Newrev: e2e9dc6129014a166cf67ab08366497c8f7a0595 Message-Id: <20221029031651.696D63858C83@sourceware.org> Date: Sat, 29 Oct 2022 03:16:51 +0000 (GMT) List-Id: https://gcc.gnu.org/g:e2e9dc6129014a166cf67ab08366497c8f7a0595 commit e2e9dc6129014a166cf67ab08366497c8f7a0595 Author: Michael Meissner Date: Fri Oct 28 23:16:17 2022 -0400 Update float 128-bit conversions. This patch is a rewrite of the patch submitted on August 18th: | https://gcc.gnu.org/pipermail/gcc-patches/2022-August/599988.html This patch reworks the conversions between 128-bit binary floating point types. Previously, we would call rs6000_expand_float128_convert to do all conversions. Now, we only define the conversions between the same representation that turn into a NOP. The appropriate extend or truncate insn is generated, and after register allocation, it is converted to a move. This patch also fixes two places where we want to override the external name for the conversion function, and the wrong optab was used. Previously, rs6000_expand_float128_convert would handle the move or generate the call as needed. Now, it lets the machine independent code generate the call. But if we use the machine independent code to generate the call, we need to update the name for two optabs where a truncate would be used in terms of converting between the modes. This patch updates those two optabs. While I know you feel the whole area needs to be rewritten, I would think it is better to make things work incrementally rather than waiting for some grand rewrite (that may or may not occur). With the current sources, we don't yet need this patch. But we will need this patch when a future patch is submitted that will change the internal __float128 type to use the _Float128 type when long double is IEEE 128-bit. I'm trying to break out the smaller patches that each can stand alone, without having a single larger patch. This future patch will fix various testsuite issues with signalling NaNs when long double is IEEE 128-bit. I tested this patch on: 1) LE Power10 using --with-cpu=power10 --with-long-double-format=ieee 2) LE Power10 using --with-cpu=power9 --with-long-double-format=ibm 3) LE Power10 using --with-cpu=power8 --with-long-double-format=ibm 4) LE Power10 using --with-cpu=power10 --with-long-double-format=ibm 5) LE Power9 using --with-cpu=power9 --with-long-double-format=ibm 6) BE Power7 using --with-cpu=power7 --with-long-double-format=ibm There were no regressions in the bootstrap process or running the tests. Can I check this patch into the trunk? 2022-10-28 Michael Meissner gcc/ * config/rs6000/rs6000.cc (init_float128_ieee): Use the correct float_extend or float_truncate optab based on how the machine converts between IEEE 128-bit and IBM 128-bit. * config/rs6000/rs6000.md (IFKF): Delete. (IFKF_reg): Delete. (extendiftf2): Rewrite to be a move if IFmode and TFmode are both IBM 128-bit. Do not run if TFmode is IEEE 128-bit. (extendifkf2): Delete. (extendtfkf2): Delete. (extendtfif2): Delete. (trunciftf2): Delete. (truncifkf2): Delete. (trunckftf2): Delete. (extendkftf2): Implement conversion of IEEE 128-bit types as a move. (trunctfif2): Delete. (trunctfkf2): Implement conversion of IEEE 128-bit types as a move. (extendtf2_internal): Delete. (extendtf2_internal): Delete. Diff: --- gcc/config/rs6000/rs6000.cc | 4 +- gcc/config/rs6000/rs6000.md | 177 ++++++++++++-------------------------------- 2 files changed, 50 insertions(+), 131 deletions(-) diff --git a/gcc/config/rs6000/rs6000.cc b/gcc/config/rs6000/rs6000.cc index 8a8357512c0..9a5907c7130 100644 --- a/gcc/config/rs6000/rs6000.cc +++ b/gcc/config/rs6000/rs6000.cc @@ -11156,11 +11156,11 @@ init_float128_ieee (machine_mode mode) set_conv_libfunc (trunc_optab, SFmode, mode, "__trunckfsf2"); set_conv_libfunc (trunc_optab, DFmode, mode, "__trunckfdf2"); - set_conv_libfunc (sext_optab, mode, IFmode, "__trunctfkf2"); + set_conv_libfunc (trunc_optab, mode, IFmode, "__trunctfkf2"); if (mode != TFmode && FLOAT128_IBM_P (TFmode)) set_conv_libfunc (sext_optab, mode, TFmode, "__trunctfkf2"); - set_conv_libfunc (trunc_optab, IFmode, mode, "__extendkftf2"); + set_conv_libfunc (sext_optab, IFmode, mode, "__extendkftf2"); if (mode != TFmode && FLOAT128_IBM_P (TFmode)) set_conv_libfunc (trunc_optab, TFmode, mode, "__extendkftf2"); diff --git a/gcc/config/rs6000/rs6000.md b/gcc/config/rs6000/rs6000.md index 3bae303086b..4880df5c51c 100644 --- a/gcc/config/rs6000/rs6000.md +++ b/gcc/config/rs6000/rs6000.md @@ -543,12 +543,6 @@ ; Iterator for 128-bit VSX types for pack/unpack (define_mode_iterator FMOVE128_VSX [V1TI KF]) -; Iterators for converting to/from TFmode -(define_mode_iterator IFKF [IF KF]) - -; Constraints for moving IF/KFmode. -(define_mode_attr IFKF_reg [(IF "d") (KF "wa")]) - ; Whether a floating point move is ok, don't allow SD without hardware FP (define_mode_attr fmove_ok [(SF "") (DF "") @@ -9096,106 +9090,65 @@ "xxlor %x0,%x1,%x2" [(set_attr "type" "veclogical")]) -;; Float128 conversion functions. These expand to library function calls. -;; We use expand to convert from IBM double double to IEEE 128-bit -;; and trunc for the opposite. -(define_expand "extendiftf2" - [(set (match_operand:TF 0 "gpc_reg_operand") - (float_extend:TF (match_operand:IF 1 "gpc_reg_operand")))] - "TARGET_FLOAT128_TYPE" -{ - rs6000_expand_float128_convert (operands[0], operands[1], false); - DONE; -}) - -(define_expand "extendifkf2" - [(set (match_operand:KF 0 "gpc_reg_operand") - (float_extend:KF (match_operand:IF 1 "gpc_reg_operand")))] - "TARGET_FLOAT128_TYPE" -{ - rs6000_expand_float128_convert (operands[0], operands[1], false); - DONE; -}) - -(define_expand "extendtfkf2" - [(set (match_operand:KF 0 "gpc_reg_operand") - (float_extend:KF (match_operand:TF 1 "gpc_reg_operand")))] - "TARGET_FLOAT128_TYPE" -{ - rs6000_expand_float128_convert (operands[0], operands[1], false); - DONE; -}) - -(define_expand "extendtfif2" - [(set (match_operand:IF 0 "gpc_reg_operand") - (float_extend:IF (match_operand:TF 1 "gpc_reg_operand")))] - "TARGET_FLOAT128_TYPE" -{ - rs6000_expand_float128_convert (operands[0], operands[1], false); - DONE; -}) - -(define_expand "trunciftf2" - [(set (match_operand:TF 0 "gpc_reg_operand") - (float_truncate:TF (match_operand:IF 1 "gpc_reg_operand")))] - "TARGET_FLOAT128_TYPE" -{ - rs6000_expand_float128_convert (operands[0], operands[1], false); - DONE; -}) - -(define_expand "truncifkf2" - [(set (match_operand:KF 0 "gpc_reg_operand") - (float_truncate:KF (match_operand:IF 1 "gpc_reg_operand")))] - "TARGET_FLOAT128_TYPE" -{ - rs6000_expand_float128_convert (operands[0], operands[1], false); - DONE; -}) - -(define_expand "trunckftf2" - [(set (match_operand:TF 0 "gpc_reg_operand") - (float_truncate:TF (match_operand:KF 1 "gpc_reg_operand")))] - "TARGET_FLOAT128_TYPE" +;; Float128 conversion functions. We only define the 'conversions' between two +;; formats that use the same representation. We call the library function to +;; convert between IEEE 128-bit and IBM 128-bit. We can't do these moves by +;; using a SUBREG before register allocation. We set up the moves to prefer +;; the output register being the same as the input register, which would enable +;; the move to be deleted completely. +(define_insn_and_split "extendkftf2" + [(set (match_operand:TF 0 "gpc_reg_operand" "=wa,wa") + (float_extend:TF (match_operand:KF 1 "gpc_reg_operand" "0,wa")))] + "TARGET_FLOAT128_TYPE && FLOAT128_IEEE_P (TFmode)" + "#" + "&& reload_completed" + [(set (match_dup 0) + (match_dup 2))] { - rs6000_expand_float128_convert (operands[0], operands[1], false); - DONE; -}) + operands[2] = gen_lowpart (TFmode, operands[1]); +} + [(set_attr "type" "veclogical")]) -(define_expand "trunctfif2" - [(set (match_operand:IF 0 "gpc_reg_operand") - (float_truncate:IF (match_operand:TF 1 "gpc_reg_operand")))] - "TARGET_FLOAT128_TYPE" +(define_insn_and_split "trunctfkf2" + [(set (match_operand:KF 0 "gpc_reg_operand" "=wa,wa") + (float_truncate:KF (match_operand:TF 1 "gpc_reg_operand" "0,wa")))] + "TARGET_FLOAT128_TYPE && FLOAT128_IEEE_P (TFmode)" + "#" + "&& reload_completed" + [(set (match_dup 0) + (match_dup 2))] { - rs6000_expand_float128_convert (operands[0], operands[1], false); - DONE; -}) + operands[2] = gen_lowpart (KFmode, operands[1]); +} + [(set_attr "type" "veclogical")]) -(define_insn_and_split "*extendtf2_internal" - [(set (match_operand:TF 0 "gpc_reg_operand" "=") - (float_extend:TF - (match_operand:IFKF 1 "gpc_reg_operand" "")))] - "TARGET_FLOAT128_TYPE - && FLOAT128_IBM_P (TFmode) == FLOAT128_IBM_P (mode)" +(define_insn_and_split "extendtfif2" + [(set (match_operand:IF 0 "gpc_reg_operand" "=wa,wa,r,r") + (float_extend:IF (match_operand:TF 1 "gpc_reg_operand" "0,wa,0,r")))] + "TARGET_HARD_FLOAT && FLOAT128_IBM_P (TFmode)" "#" "&& reload_completed" - [(set (match_dup 0) (match_dup 2))] + [(set (match_dup 0) + (match_dup 2))] { - operands[2] = gen_rtx_REG (TFmode, REGNO (operands[1])); -}) + operands[2] = gen_lowpart (IFmode, operands[1]); +} + [(set_attr "num_insns" "2") + (set_attr "length" "8")]) -(define_insn_and_split "*extendtf2_internal" - [(set (match_operand:IFKF 0 "gpc_reg_operand" "=") - (float_extend:IFKF - (match_operand:TF 1 "gpc_reg_operand" "")))] - "TARGET_FLOAT128_TYPE - && FLOAT128_IBM_P (TFmode) == FLOAT128_IBM_P (mode)" +(define_insn_and_split "extendiftf2" + [(set (match_operand:TF 0 "gpc_reg_operand" "=wa,wa,r,r") + (float_extend:TF (match_operand:IF 1 "gpc_reg_operand" "0,wa,0,r")))] + "TARGET_HARD_FLOAT && FLOAT128_IBM_P (TFmode)" "#" "&& reload_completed" - [(set (match_dup 0) (match_dup 2))] + [(set (match_dup 0) + (match_dup 2))] { - operands[2] = gen_rtx_REG (mode, REGNO (operands[1])); -}) + operands[2] = gen_lowpart (TFmode, operands[1]); +} + [(set_attr "num_insns" "2") + (set_attr "length" "8")]) ;; Reload helper functions used by rs6000_secondary_reload. The patterns all @@ -14909,40 +14862,6 @@ [(set_attr "type" "vecfloat") (set_attr "size" "128")]) -;; Conversion between KFmode and TFmode if TFmode is ieee 128-bit floating -;; point is a simple copy. -(define_insn_and_split "extendkftf2" - [(set (match_operand:TF 0 "vsx_register_operand" "=wa,?wa") - (float_extend:TF (match_operand:KF 1 "vsx_register_operand" "0,wa")))] - "TARGET_FLOAT128_TYPE && TARGET_IEEEQUAD" - "@ - # - xxlor %x0,%x1,%x1" - "&& reload_completed && REGNO (operands[0]) == REGNO (operands[1])" - [(const_int 0)] -{ - emit_note (NOTE_INSN_DELETED); - DONE; -} - [(set_attr "type" "*,veclogical") - (set_attr "length" "0,4")]) - -(define_insn_and_split "trunctfkf2" - [(set (match_operand:KF 0 "vsx_register_operand" "=wa,?wa") - (float_extend:KF (match_operand:TF 1 "vsx_register_operand" "0,wa")))] - "TARGET_FLOAT128_TYPE && TARGET_IEEEQUAD" - "@ - # - xxlor %x0,%x1,%x1" - "&& reload_completed && REGNO (operands[0]) == REGNO (operands[1])" - [(const_int 0)] -{ - emit_note (NOTE_INSN_DELETED); - DONE; -} - [(set_attr "type" "*,veclogical") - (set_attr "length" "0,4")]) - (define_insn "truncdf2_hw" [(set (match_operand:DF 0 "altivec_register_operand" "=v") (float_truncate:DF