From patchwork Thu Jul 27 17:13:13 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Jakub Jelinek X-Patchwork-Id: 127073 Return-Path: Delivered-To: ouuuleilei@gmail.com Received: by 2002:a59:a985:0:b0:3e4:2afc:c1 with SMTP id t5csp1246573vqo; Thu, 27 Jul 2023 10:16:54 -0700 (PDT) X-Google-Smtp-Source: APBJJlEcqEJFL2DCQBF72YSG6VQeJCXsNBINOoYyCTVqXfo9MA6eYE4xrsRWsH91nzSBMqbfTX4v X-Received: by 2002:adf:ea8f:0:b0:314:3985:b291 with SMTP id s15-20020adfea8f000000b003143985b291mr2728369wrm.15.1690478213776; Thu, 27 Jul 2023 10:16:53 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1690478213; cv=none; d=google.com; s=arc-20160816; b=BNu5RmglLzI6mLJI7diyHIpMXW/f33HBHKeCj9HAnSz61cQkJJgzTu8+WSS9albBl9 QJcYVvRz1D5T7t4weLeYtUOBiW+IYICN1BvUA4zrZV0pbkwRr4T7l9XZzm+xjzXhmIMR GZaRdoLjj2Wi25A2hVw+YrH2uyQj0WMOicXzjPXqqf3Lg2gJ6ZKRA30GzfPAtajXI8mM qIsrJ2R1vdMY1qNmU9NQQ8HYeYqgaOM96OX2VpnfXkOhvFz5vrMFA2sxbMkJdEZP+4Cs gM/Cc4QSsJQMtaSm4LaQicT0sqLJGEKN/QHuapKmF29OeeoOSnoQwoNlwbhprEOtWHOs MChQ== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:reply-to:from:list-subscribe:list-help:list-post :list-archive:list-unsubscribe:list-id:precedence :content-disposition:mime-version:message-id:subject:cc:to:date :dmarc-filter:delivered-to:dkim-signature:dkim-filter; bh=3LkAcL5WDoisXxMs58ZE/LbLC4KvGrMSeKIVLh2dy9Q=; fh=WLIzpkiNYT+sSgm6PnMRUugV0RZtfwj0ju7DyNqWOiY=; b=NmpySVoevzOQlqwrR8bIYldufrIjm3SrBKIIFQq7L9nq9ZsPLs9I9x1TuYUr6LNBL3 OtyLMCwi2xADyWzucuIPnbqXl6Dp/HnNyIg+P6J2B1w3vJs0AhUEIGMztvO1V5r1YSq5 mAp3sl1U9ahyUTe5kGmiKBg//Ay+AYMNRQuJ/voMVx4pVfF6gePkq3KSzxK2/BMmUOBs ZsVlnogoPTfoxeJ9GSRkYDf0u9WynqKpYxCP4ouXpyBSmEK4Ir3l9vs9vbChlClVPq4r q4QeBRGfwPUeRdwSCIwGIy6GXGoue99DD/T1hjVUB0emLchwLiBi4LOcQ4qNrfJ51TFW zs2g== ARC-Authentication-Results: i=1; mx.google.com; dkim=pass header.i=@gcc.gnu.org header.s=default header.b=PDqIjjAU; spf=pass (google.com: domain of gcc-patches-bounces+ouuuleilei=gmail.com@gcc.gnu.org designates 2620:52:3:1:0:246e:9693:128c as permitted sender) smtp.mailfrom="gcc-patches-bounces+ouuuleilei=gmail.com@gcc.gnu.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=gnu.org Received: from server2.sourceware.org (server2.sourceware.org. [2620:52:3:1:0:246e:9693:128c]) by mx.google.com with ESMTPS id i12-20020a170906a28c00b00997d315d050si1252030ejz.84.2023.07.27.10.16.53 for (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Thu, 27 Jul 2023 10:16:53 -0700 (PDT) Received-SPF: pass (google.com: domain of gcc-patches-bounces+ouuuleilei=gmail.com@gcc.gnu.org designates 2620:52:3:1:0:246e:9693:128c as permitted sender) client-ip=2620:52:3:1:0:246e:9693:128c; Authentication-Results: mx.google.com; dkim=pass header.i=@gcc.gnu.org header.s=default header.b=PDqIjjAU; spf=pass (google.com: domain of gcc-patches-bounces+ouuuleilei=gmail.com@gcc.gnu.org designates 2620:52:3:1:0:246e:9693:128c as permitted sender) smtp.mailfrom="gcc-patches-bounces+ouuuleilei=gmail.com@gcc.gnu.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=gnu.org Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id C7AE13854C58 for ; Thu, 27 Jul 2023 17:16:05 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org C7AE13854C58 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org; s=default; t=1690478165; bh=3LkAcL5WDoisXxMs58ZE/LbLC4KvGrMSeKIVLh2dy9Q=; h=Date:To:Cc:Subject:List-Id:List-Unsubscribe:List-Archive: List-Post:List-Help:List-Subscribe:From:Reply-To:From; b=PDqIjjAUjCznjKVA1EXARBNQaGeAqf974NZ5v83KCCtWhsCJxwxIjqv8PethoiHQh uGt0sVvWal+5R2JU3m9sGpMFovzb7Kxz4PGB71ZpsLqikeGsvEIsgQ8gTEcJQoS4Ie EtNAdUC3ZXph9csaxc31gQ7a8ew6cfWPuYK/y4Kc= X-Original-To: gcc-patches@gcc.gnu.org Delivered-To: gcc-patches@gcc.gnu.org Received: from us-smtp-delivery-124.mimecast.com (us-smtp-delivery-124.mimecast.com [170.10.133.124]) by sourceware.org (Postfix) with ESMTPS id 0F8343854822 for ; Thu, 27 Jul 2023 17:13:23 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org 0F8343854822 Received: from mimecast-mx02.redhat.com (mimecast-mx02.redhat.com [66.187.233.88]) by relay.mimecast.com with ESMTP with STARTTLS (version=TLSv1.2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id us-mta-70-qRvCplv4MWaX6wAQjnf8Ew-1; Thu, 27 Jul 2023 13:13:19 -0400 X-MC-Unique: qRvCplv4MWaX6wAQjnf8Ew-1 Received: from smtp.corp.redhat.com (int-mx06.intmail.prod.int.rdu2.redhat.com [10.11.54.6]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by mimecast-mx02.redhat.com (Postfix) with ESMTPS id B49C1805951; Thu, 27 Jul 2023 17:13:18 +0000 (UTC) Received: from tucnak.zalov.cz (unknown [10.45.224.18]) by smtp.corp.redhat.com (Postfix) with ESMTPS id 138942166B27; Thu, 27 Jul 2023 17:13:17 +0000 (UTC) Received: from tucnak.zalov.cz (localhost [127.0.0.1]) by tucnak.zalov.cz (8.17.1/8.17.1) with ESMTPS id 36RHDFO82823055 (version=TLSv1.3 cipher=TLS_AES_256_GCM_SHA384 bits=256 verify=NOT); Thu, 27 Jul 2023 19:13:15 +0200 Received: (from jakub@localhost) by tucnak.zalov.cz (8.17.1/8.17.1/Submit) id 36RHDDwW2823054; Thu, 27 Jul 2023 19:13:13 +0200 Date: Thu, 27 Jul 2023 19:13:13 +0200 To: gcc-patches@gcc.gnu.org Cc: "Joseph S. Myers" , Uros Bizjak , Ian Lance Taylor Subject: [PATCH 2/5] libgcc _BitInt support [PR102989] Message-ID: MIME-Version: 1.0 X-Scanned-By: MIMEDefang 3.1 on 10.11.54.6 X-Mimecast-Spam-Score: 0 X-Mimecast-Originator: redhat.com Content-Disposition: inline X-Spam-Status: No, score=-14.9 required=5.0 tests=BAYES_00, DKIM_INVALID, DKIM_SIGNED, KAM_DMARC_NONE, KAM_DMARC_STATUS, KAM_SHORT, RCVD_IN_DNSWL_NONE, RCVD_IN_MSPIKE_H4, RCVD_IN_MSPIKE_WL, SPF_HELO_NONE, SPF_NONE, TXREP, T_SCC_BODY_TEXT_LINE autolearn=no autolearn_force=no version=3.4.6 X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on server2.sourceware.org X-BeenThere: gcc-patches@gcc.gnu.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Gcc-patches mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-Patchwork-Original-From: Jakub Jelinek via Gcc-patches From: Jakub Jelinek Reply-To: Jakub Jelinek Errors-To: gcc-patches-bounces+ouuuleilei=gmail.com@gcc.gnu.org Sender: "Gcc-patches" X-getmail-retrieved-from-mailbox: INBOX X-GMAIL-THRID: 1772594883464710519 X-GMAIL-MSGID: 1772594883464710519 Hi! This patch adds the library helpers for multiplication, division + modulo and casts from and to floating point. As described in the intro, the first step is try to reduce further the passed in precision by skipping over most significant limbs with just zeros or sign bit copies. For multiplication and division I've implemented a simple algorithm, using something smarter like Karatsuba or Toom N-Way might be faster for very large _BitInts (which we don't support right now anyway), but could mean more code in libgcc, which maybe isn't what people are willing to accept. For the to/from floating point conversions the patch uses soft-fp, because it already has tons of handy macros which can be used for that. In theory it could be implemented using {,unsigned} long long or {,unsigned} __int128 to/from floating point conversions with some frexp before/after, but at that point we already need to force it into integer registers and analyze it anyway. Plus, for 32-bit arches there is no __int128 that could be used for XF/TF mode stuff. I know that soft-fp is owned by glibc and I think the op-common.h change should be propagated there, but the bitint stuff is really GCC specific and IMHO doesn't belong into the glibc copy. 2023-07-27 Jakub Jelinek PR c/102989 libgcc/ * config/aarch64/t-softfp (softfp_extras): Use += rather than :=. * config/i386/64/t-softfp (softfp_extras): Likewise. * config/i386/libgcc-glibc.ver (GCC_14.0.0): Export _BitInt support routines. * config/i386/t-softfp (softfp_extras): Add fixxfbitint and bf, hf and xf mode floatbitint. (CFLAGS-floatbitintbf.c, CFLAGS-floatbitinthf.c): Add -msse2. * config/riscv/t-softfp32 (softfp_extras): Use += rather than :=. * config/rs6000/t-e500v1-fp (softfp_extras): Likewise. * config/rs6000/t-e500v2-fp (softfp_extras): Likewise. * config/t-softfp (softfp_floatbitint_funcs): New. (softfp_func_list): Add sf and df mode from and to _BitInt libcalls. * config/t-softfp-sfdftf (softfp_extras): Add fixtfbitint and floatbitinttf. * config/t-softfp-tf (softfp_extras): Likewise. * libgcc2.c (bitint_reduce_prec): New inline function. (BITINT_INC, BITINT_END): Define. (bitint_mul_1, bitint_addmul_1): New helper functions. (__mulbitint3): New function. (bitint_negate, bitint_submul_1): New helper functions. (__divmodbitint4): New function. * libgcc2.h (LIBGCC2_UNITS_PER_WORD): When building _BitInt support libcalls, redefine depending on __LIBGCC_BITINT_LIMB_WIDTH__. (__mulbitint3, __divmodbitint4): Declare. * libgcc-std.ver.in (GCC_14.0.0): Export _BitInt support routines. * Makefile.in (lib2funcs): Add _mulbitint3. (LIB2_DIVMOD_FUNCS): Add _divmodbitint4. * soft-fp/bitint.h: New file. * soft-fp/fixdfbitint.c: New file. * soft-fp/fixsfbitint.c: New file. * soft-fp/fixtfbitint.c: New file. * soft-fp/fixxfbitint.c: New file. * soft-fp/floatbitintbf.c: New file. * soft-fp/floatbitintdf.c: New file. * soft-fp/floatbitinthf.c: New file. * soft-fp/floatbitintsf.c: New file. * soft-fp/floatbitinttf.c: New file. * soft-fp/floatbitintxf.c: New file. * soft-fp/op-common.h (_FP_FROM_INT): Add support for rsize up to 4 * _FP_W_TYPE_SIZE rather than just 2 * _FP_W_TYPE_SIZE. Jakub --- libgcc/config/aarch64/t-softfp.jj 2023-03-13 00:11:52.330213322 +0100 +++ libgcc/config/aarch64/t-softfp 2023-07-14 12:38:30.764869473 +0200 @@ -3,7 +3,7 @@ softfp_int_modes := si di ti softfp_extensions := sftf dftf hftf bfsf softfp_truncations := tfsf tfdf tfhf tfbf dfbf sfbf hfbf softfp_exclude_libgcc2 := n -softfp_extras := fixhfti fixunshfti floattihf floatuntihf \ +softfp_extras += fixhfti fixunshfti floattihf floatuntihf \ floatdibf floatundibf floattibf floatuntibf TARGET_LIBGCC2_CFLAGS += -Wno-missing-prototypes --- libgcc/config/i386/64/t-softfp.jj 2023-03-10 20:39:43.849687830 +0100 +++ libgcc/config/i386/64/t-softfp 2023-07-14 12:37:55.422344930 +0200 @@ -1,4 +1,4 @@ -softfp_extras := fixhfti fixunshfti floattihf floatuntihf \ +softfp_extras += fixhfti fixunshfti floattihf floatuntihf \ floattibf floatuntibf CFLAGS-fixhfti.c += -msse2 --- libgcc/config/i386/libgcc-glibc.ver.jj 2023-07-11 13:39:49.760107863 +0200 +++ libgcc/config/i386/libgcc-glibc.ver 2023-07-17 09:45:43.128281615 +0200 @@ -226,3 +226,13 @@ GCC_13.0.0 { __truncxfbf2 __trunchfbf2 } + +%inherit GCC_14.0.0 GCC_13.0.0 +GCC_14.0.0 { + __PFX__fixxfbitint + __PFX__fixtfbitint + __PFX__floatbitintbf + __PFX__floatbitinthf + __PFX__floatbitintxf + __PFX__floatbitinttf +} --- libgcc/config/i386/t-softfp.jj 2022-10-14 09:35:56.268989311 +0200 +++ libgcc/config/i386/t-softfp 2023-07-17 09:38:43.575980078 +0200 @@ -10,7 +10,7 @@ softfp_extensions := hfsf hfdf hftf hfxf softfp_truncations := tfhf xfhf dfhf sfhf tfsf dfsf tfdf tfxf \ tfbf xfbf dfbf sfbf hfbf -softfp_extras += eqhf2 +softfp_extras += eqhf2 fixxfbitint $(foreach m,hf bf xf,floatbitint$(m)) CFLAGS-extendhfsf2.c += -msse2 CFLAGS-extendhfdf2.c += -msse2 @@ -28,6 +28,9 @@ CFLAGS-truncxfbf2.c += -msse2 CFLAGS-trunctfbf2.c += -msse2 CFLAGS-trunchfbf2.c += -msse2 +CFLAGS-floatbitintbf.c += -msse2 +CFLAGS-floatbitinthf.c += -msse2 + CFLAGS-eqhf2.c += -msse2 CFLAGS-_divhc3.c += -msse2 CFLAGS-_mulhc3.c += -msse2 --- libgcc/config/riscv/t-softfp32.jj 2022-08-17 16:59:09.291544033 +0200 +++ libgcc/config/riscv/t-softfp32 2023-07-14 12:36:57.895118834 +0200 @@ -13,7 +13,7 @@ softfp_extensions := sftf dftf softfp_truncations := tfsf tfdf # Enable divide routines to make -mno-fdiv work. -softfp_extras := divsf3 divdf3 +softfp_extras += divsf3 divdf3 else # !ABI_DOUBLE @@ -28,7 +28,7 @@ else # ABI_SINGLE # Enable divide routines to make -mno-fdiv work. -softfp_extras := divsf3 +softfp_extras += divsf3 endif @@ -38,7 +38,7 @@ else # ABI_QUAD # Enable divide routines to make -mno-fdiv work. -softfp_extras := divsf3 divdf3 divtf3 +softfp_extras += divsf3 divdf3 divtf3 endif --- libgcc/config/rs6000/t-e500v1-fp.jj 2023-01-16 11:52:16.774723882 +0100 +++ libgcc/config/rs6000/t-e500v1-fp 2023-07-14 12:39:03.906423712 +0200 @@ -29,4 +29,4 @@ softfp_int_modes := si di softfp_extensions := sfdf softfp_truncations := dfsf softfp_exclude_libgcc2 := n -softfp_extras := unordsf2 +softfp_extras += unordsf2 --- libgcc/config/rs6000/t-e500v2-fp.jj 2023-01-16 11:52:16.774723882 +0100 +++ libgcc/config/rs6000/t-e500v2-fp 2023-07-14 12:39:09.924342992 +0200 @@ -23,4 +23,4 @@ softfp_int_modes := softfp_extensions := softfp_truncations := softfp_exclude_libgcc2 := n -softfp_extras := unordsf2 unorddf2 +softfp_extras += unordsf2 unorddf2 --- libgcc/config/t-softfp.jj 2023-01-16 11:52:16.778723823 +0100 +++ libgcc/config/t-softfp 2023-07-17 09:35:07.802912565 +0200 @@ -64,12 +64,14 @@ softfp_float_funcs = add$(m)3 div$(m)3 e neg$(m)2 sub$(m)3 unord$(m)2 softfp_floatint_funcs = fix$(m)$(i) fixuns$(m)$(i) \ float$(i)$(m) floatun$(i)$(m) +softfp_floatbitint_funcs = fix$(m)bitint floatbitint$(m) softfp_func_list := \ $(foreach m,$(softfp_float_modes), \ $(softfp_float_funcs) \ $(foreach i,$(softfp_int_modes), \ $(softfp_floatint_funcs))) \ + $(foreach m,sf df,$(softfp_floatbitint_funcs)) \ $(foreach e,$(softfp_extensions),extend$(e)2) \ $(foreach t,$(softfp_truncations),trunc$(t)2) \ $(softfp_extras) --- libgcc/config/t-softfp-sfdftf.jj 2020-01-12 11:54:38.691379040 +0100 +++ libgcc/config/t-softfp-sfdftf 2023-07-17 09:35:59.283212918 +0200 @@ -2,4 +2,5 @@ softfp_float_modes := sf df tf softfp_int_modes := si di softfp_extensions := sfdf sftf dftf xftf softfp_truncations := dfsf tfsf tfdf tfxf +softfp_extras += fixtfbitint floatbitinttf softfp_exclude_libgcc2 := n --- libgcc/config/t-softfp-tf.jj 2020-01-12 11:54:38.691379040 +0100 +++ libgcc/config/t-softfp-tf 2023-07-17 09:35:18.190771387 +0200 @@ -2,4 +2,5 @@ softfp_float_modes := tf softfp_int_modes := si di ti softfp_extensions := sftf dftf xftf softfp_truncations := tfsf tfdf tfxf +softfp_extras += fixtfbitint floatbitinttf softfp_exclude_libgcc2 := n --- libgcc/libgcc2.c.jj 2023-07-11 13:39:49.777107635 +0200 +++ libgcc/libgcc2.c 2023-07-22 12:18:43.743740024 +0200 @@ -1301,6 +1301,687 @@ __udivdi3 (UDWtype n, UDWtype d) } #endif +#if (defined(__BITINT_MAXWIDTH__) \ + && (defined(L_mulbitint3) || defined(L_divmodbitint4))) +/* _BitInt support. */ + +/* If *P is zero or sign extended (the latter only for PREC < 0) from + some narrower _BitInt value, reduce precision. */ + +static inline __attribute__((__always_inline__)) SItype +bitint_reduce_prec (const UWtype **p, SItype prec) +{ + UWtype mslimb; + SItype i; + if (prec < 0) + { +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ + i = 0; +#else + i = ((USItype) -1 - prec) / W_TYPE_SIZE; +#endif + mslimb = (*p)[i]; + if (mslimb & ((UWtype) 1 << (((USItype) -1 - prec) % W_TYPE_SIZE))) + { + SItype n = ((USItype) -prec) % W_TYPE_SIZE; + if (n) + { + mslimb |= ((UWtype) -1 << (((USItype) -1 - prec) % W_TYPE_SIZE)); + if (mslimb == (UWtype) -1) + { + prec += n; + if (prec >= -1) + return -2; +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ + ++p; +#else + --i; +#endif + mslimb = (*p)[i]; + n = 0; + } + } + while (mslimb == (UWtype) -1) + { + prec += W_TYPE_SIZE; + if (prec >= -1) + return -2; +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ + ++p; +#else + --i; +#endif + mslimb = (*p)[i]; + } + if (n == 0) + { + if ((Wtype) mslimb >= 0) + { +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ + --p; +#endif + return prec - 1; + } + } + return prec; + } + else + prec = -prec; + } + else + { +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ + i = 0; +#else + i = ((USItype) prec - 1) / W_TYPE_SIZE; +#endif + mslimb = (*p)[i]; + } + SItype n = ((USItype) prec) % W_TYPE_SIZE; + if (n) + { + mslimb &= ((UWtype) 1 << (((USItype) prec) % W_TYPE_SIZE)) - 1; + if (mslimb == 0) + { + prec -= n; + if (prec == 0) + return 1; +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ + ++p; +#else + --i; +#endif + mslimb = (*p)[i]; + } + } + while (mslimb == 0) + { + prec -= W_TYPE_SIZE; + if (prec == 0) + return 1; +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ + ++p; +#else + --i; +#endif + mslimb = (*p)[i]; + } + return prec; +} + +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ +# define BITINT_INC -1 +# define BITINT_END(be, le) (be) +#else +# define BITINT_INC 1 +# define BITINT_END(be, le) (le) +#endif + +#ifdef L_mulbitint3 +/* D = S * L. */ + +static UWtype +bitint_mul_1 (UWtype *d, const UWtype *s, UWtype l, SItype n) +{ + UWtype sv, hi, lo, c = 0; + do + { + sv = *s; + s += BITINT_INC; + umul_ppmm (hi, lo, sv, l); + c = __builtin_add_overflow (lo, c, &lo) + hi; + *d = lo; + d += BITINT_INC; + } + while (--n); + return c; +} + +/* D += S * L. */ + +static UWtype +bitint_addmul_1 (UWtype *d, const UWtype *s, UWtype l, SItype n) +{ + UWtype sv, hi, lo, c = 0; + do + { + sv = *s; + s += BITINT_INC; + umul_ppmm (hi, lo, sv, l); + hi += __builtin_add_overflow (lo, *d, &lo); + c = __builtin_add_overflow (lo, c, &lo) + hi; + *d = lo; + d += BITINT_INC; + } + while (--n); + return c; +} + +/* If XPREC is positive, it is precision in bits + of an unsigned _BitInt operand (which has XPREC/W_TYPE_SIZE + full limbs and if Xprec%W_TYPE_SIZE one partial limb. + If Xprec is negative, -XPREC is precision in bits + of a signed _BitInt operand. RETPREC should be always + positive. */ + +void +__mulbitint3 (UWtype *ret, SItype retprec, + const UWtype *u, SItype uprec, + const UWtype *v, SItype vprec) +{ + uprec = bitint_reduce_prec (&u, uprec); + vprec = bitint_reduce_prec (&v, vprec); + USItype auprec = uprec < 0 ? -uprec : uprec; + USItype avprec = vprec < 0 ? -vprec : vprec; + + /* Prefer non-negative U. + Otherwise make sure V doesn't have higher precision than U. */ + if ((uprec < 0 && vprec >= 0) + || (avprec > auprec && !(uprec >= 0 && vprec < 0))) + { + SItype p; + const UWtype *t; + p = uprec; uprec = vprec; vprec = p; + p = auprec; auprec = avprec; avprec = p; + t = u; u = v; v = t; + } + + USItype un = auprec / W_TYPE_SIZE; + USItype un2 = (auprec + W_TYPE_SIZE - 1) / W_TYPE_SIZE; + USItype vn = avprec / W_TYPE_SIZE; + USItype vn2 = (avprec + W_TYPE_SIZE - 1) / W_TYPE_SIZE; + USItype retn = ((USItype) retprec + W_TYPE_SIZE - 1) / W_TYPE_SIZE; + USItype retidx, uidx, vidx; + UWtype vv; + /* Indexes of least significant limb. */ +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ + retidx = retn - 1; + uidx = un2 - 1; + vidx = vn2 - 1; +#else + retidx = 0; + uidx = 0; + vidx = 0; +#endif + if (__builtin_expect (auprec <= W_TYPE_SIZE, 0) && vprec < 0) + { + UWtype uu = u[uidx]; + if (__builtin_expect (auprec < W_TYPE_SIZE, 0)) + uu &= ((UWtype) 1 << (auprec % W_TYPE_SIZE)) - 1; + if (uu == 0) + { + /* 0 * negative would be otherwise mishandled below, so + handle it specially. */ + __builtin_memset (ret, 0, retn * sizeof (UWtype)); + return; + } + } + vv = v[vidx]; + if (__builtin_expect (avprec < W_TYPE_SIZE, 0)) + { + if (vprec > 0) + vv &= ((UWtype) 1 << (avprec % W_TYPE_SIZE)) - 1; + else + vv |= (UWtype) -1 << (avprec % W_TYPE_SIZE); + } + + USItype n = un > retn ? retn : un; + USItype n2 = n; + USItype retidx2 = retidx + n * BITINT_INC; + UWtype c = 0, uv = 0; + if (n) + c = bitint_mul_1 (ret + retidx, u + uidx, vv, n); + if (retn > un && un2 != un) + { + UWtype hi, lo; + uv = u[uidx + n * BITINT_INC]; + if (uprec > 0) + uv &= ((UWtype) 1 << (auprec % W_TYPE_SIZE)) - 1; + else + uv |= (UWtype) -1 << (auprec % W_TYPE_SIZE); + umul_ppmm (hi, lo, uv, vv); + c = __builtin_add_overflow (lo, c, &lo) + hi; + ret[retidx2] = lo; + retidx2 += BITINT_INC; + ++n2; + } + if (retn > un2) + { + if (uprec < 0) + { + while (n2 < retn) + { + if (n2 >= un2 + vn2) + break; + UWtype hi, lo; + umul_ppmm (hi, lo, (UWtype) -1, vv); + c = __builtin_add_overflow (lo, c, &lo) + hi; + ret[retidx2] = lo; + retidx2 += BITINT_INC; + ++n2; + } + } + else + { + ret[retidx2] = c; + retidx2 += BITINT_INC; + ++n2; + } + /* If RET has more limbs than U after precision reduction, + fill in the remaining limbs. */ + while (n2 < retn) + { + if (n2 < un2 + vn2 || (uprec ^ vprec) >= 0) + c = 0; + else + c = (UWtype) -1; + ret[retidx2] = c; + retidx2 += BITINT_INC; + ++n2; + } + } + /* N is now number of possibly non-zero limbs in RET (ignoring + limbs above UN2 + VN2 which if any have been finalized already). */ + USItype end = vprec < 0 ? un2 + vn2 : vn2; + if (retn > un2 + vn2) retn = un2 + vn2; + if (end > retn) end = retn; + for (USItype m = 1; m < end; ++m) + { + retidx += BITINT_INC; + vidx += BITINT_INC; + if (m < vn2) + { + vv = v[vidx]; + if (__builtin_expect (m == vn, 0)) + { + if (vprec > 0) + vv &= ((UWtype) 1 << (avprec % W_TYPE_SIZE)) - 1; + else + vv |= (UWtype) -1 << (avprec % W_TYPE_SIZE); + } + } + else + vv = (UWtype) -1; + if (m + n > retn) + n = retn - m; + c = 0; + if (n) + c = bitint_addmul_1 (ret + retidx, u + uidx, vv, n); + n2 = m + n; + retidx2 = retidx + n * BITINT_INC; + if (n2 < retn && un2 != un) + { + UWtype hi, lo; + umul_ppmm (hi, lo, uv, vv); + hi += __builtin_add_overflow (lo, ret[retidx2], &lo); + c = __builtin_add_overflow (lo, c, &lo) + hi; + ret[retidx2] = lo; + retidx2 += BITINT_INC; + ++n2; + } + if (uprec < 0) + while (n2 < retn) + { + UWtype hi, lo; + umul_ppmm (hi, lo, (UWtype) -1, vv); + hi += __builtin_add_overflow (lo, ret[retidx2], &lo); + c = __builtin_add_overflow (lo, c, &lo) + hi; + ret[retidx2] = lo; + retidx2 += BITINT_INC; + ++n2; + } + else if (n2 < retn) + { + ret[retidx2] = c; + retidx2 += BITINT_INC; + } + } +} +#endif + +#ifdef L_divmodbitint4 +static void +bitint_negate (UWtype *d, const UWtype *s, SItype n) +{ + UWtype c = 1; + do + { + UWtype sv = *s, lo; + s += BITINT_INC; + c = __builtin_add_overflow (~sv, c, &lo); + *d = lo; + d += BITINT_INC; + } + while (--n); +} + +/* D -= S * L. */ + +static UWtype +bitint_submul_1 (UWtype *d, const UWtype *s, UWtype l, SItype n) +{ + UWtype sv, hi, lo, c = 0; + do + { + sv = *s; + s += BITINT_INC; + umul_ppmm (hi, lo, sv, l); + hi += __builtin_sub_overflow (*d, lo, &lo); + c = __builtin_sub_overflow (lo, c, &lo) + hi; + *d = lo; + d += BITINT_INC; + } + while (--n); + return c; +} + +/* If XPREC is positive, it is precision in bits + of an unsigned _BitInt operand (which has XPREC/W_TYPE_SIZE + full limbs and if Xprec%W_TYPE_SIZE one partial limb. + If Xprec is negative, -XPREC is precision in bits + of a signed _BitInt operand. QPREC and RPREC should be + always non-negative. If either Q or R is NULL (at least + one should be non-NULL), then corresponding QPREC or RPREC + should be 0. */ + +void +__divmodbitint4 (UWtype *q, SItype qprec, + UWtype *r, SItype rprec, + const UWtype *u, SItype uprec, + const UWtype *v, SItype vprec) +{ + uprec = bitint_reduce_prec (&u, uprec); + vprec = bitint_reduce_prec (&v, vprec); + USItype auprec = uprec < 0 ? -uprec : uprec; + USItype avprec = vprec < 0 ? -vprec : vprec; + USItype un = (auprec + W_TYPE_SIZE - 1) / W_TYPE_SIZE; + USItype vn = (avprec + W_TYPE_SIZE - 1) / W_TYPE_SIZE; + USItype qn = ((USItype) qprec + W_TYPE_SIZE - 1) / W_TYPE_SIZE; + USItype rn = ((USItype) rprec + W_TYPE_SIZE - 1) / W_TYPE_SIZE; + USItype up = auprec % W_TYPE_SIZE; + USItype vp = avprec % W_TYPE_SIZE; + if (__builtin_expect (un < vn, 0)) + { + /* If abs(v) > abs(u), then q is 0 and r is u. */ + if (q) + __builtin_memset (q, 0, qn * sizeof (UWtype)); + if (r == NULL) + return; +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ + r += rn - 1; + u += un - 1; +#endif + if (up) + --un; + if (rn < un) + un = rn; + for (rn -= un; un; --un) + { + *r = *u; + r += BITINT_INC; + u += BITINT_INC; + } + if (!rn) + return; + if (up) + { + if (uprec > 0) + *r = *u & (((UWtype) 1 << up) - 1); + else + *r = *u | ((UWtype) -1 << up); + r += BITINT_INC; + if (!--rn) + return; + } + UWtype c = uprec < 0 ? (UWtype) -1 : (UWtype) 0; + for (; rn; --rn) + { + *r = c; + r += BITINT_INC; + } + return; + } + USItype qn2 = un - vn + 1; + if (qn >= qn2) + qn2 = 0; + USItype sz = un + 1 + vn + qn2; + UWtype *buf = __builtin_alloca (sz * sizeof (UWtype)); + USItype uidx, vidx; +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ + uidx = un - 1; + vidx = vn - 1; +#else + uidx = 0; + vidx = 0; +#endif + if (uprec < 0) + bitint_negate (buf + BITINT_END (uidx + 1, 0), u + uidx, un); + else + __builtin_memcpy (buf + BITINT_END (1, 0), u, un * sizeof (UWtype)); + if (up) + buf[BITINT_END (1, un - 1)] &= (((UWtype) 1 << up) - 1); + if (vprec < 0) + bitint_negate (buf + un + 1 + vidx, v + vidx, vn); + else + __builtin_memcpy (buf + un + 1, v, vn * sizeof (UWtype)); + if (vp) + buf[un + 1 + BITINT_END (0, vn - 1)] &= (((UWtype) 1 << vp) - 1); + UWtype *u2 = buf; + UWtype *v2 = u2 + un + 1; + UWtype *q2 = v2 + vn; + if (!qn2) + q2 = q + BITINT_END (qn - (un - vn + 1), 0); + + /* Knuth's algorithm. See also ../gcc/wide-int.cc (divmod_internal_2). */ + +#ifndef UDIV_NEEDS_NORMALIZATION + /* Handle single limb divisor first. */ + if (vn == 1) + { + UWtype vv = v2[0]; + if (vv == 0) + vv = 1 / vv; /* Divide intentionally by zero. */ + UWtype k = 0; +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ + for (SItype i = 0; i <= un - 1; ++i) +#else + for (SItype i = un - 1; i >= 0; --i) +#endif + udiv_qrnnd (q2[i], k, k, u2[BITINT_END (i + 1, i)], vv); + if (r != NULL) + r[BITINT_END (rn - 1, 0)] = k; + } + else +#endif + { + SItype s; +#ifdef UDIV_NEEDS_NORMALIZATION + if (vn == 1 && v2[0] == 0) + s = 0; + else +#endif + if (sizeof (0U) == sizeof (UWtype)) + s = __builtin_clz (v2[BITINT_END (0, vn - 1)]); + else if (sizeof (0UL) == sizeof (UWtype)) + s = __builtin_clzl (v2[BITINT_END (0, vn - 1)]); + else + s = __builtin_clzll (v2[BITINT_END (0, vn - 1)]); + if (s) + { + /* Normalize by shifting v2 left so that it has msb set. */ + const SItype n = sizeof (UWtype) * __CHAR_BIT__; +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ + for (SItype i = 0; i < vn - 1; ++i) +#else + for (SItype i = vn - 1; i > 0; --i) +#endif + v2[i] = (v2[i] << s) | (v2[i - BITINT_INC] >> (n - s)); + v2[vidx] = v2[vidx] << s; + /* And shift u2 left by the same amount. */ + u2[BITINT_END (0, un)] = u2[BITINT_END (1, un - 1)] >> (n - s); +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ + for (SItype i = 1; i < un; ++i) +#else + for (SItype i = un - 1; i > 0; --i) +#endif + u2[i] = (u2[i] << s) | (u2[i - BITINT_INC] >> (n - s)); + u2[BITINT_END (un, 0)] = u2[BITINT_END (un, 0)] << s; + } + else + u2[BITINT_END (0, un)] = 0; +#ifdef UDIV_NEEDS_NORMALIZATION + /* Handle single limb divisor first. */ + if (vn == 1) + { + UWtype vv = v2[0]; + if (vv == 0) + vv = 1 / vv; /* Divide intentionally by zero. */ + UWtype k = u2[BITINT_END (0, un)]; +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ + for (SItype i = 0; i <= un - 1; ++i) +#else + for (SItype i = un - 1; i >= 0; --i) +#endif + udiv_qrnnd (q2[i], k, k, u2[BITINT_END (i + 1, i)], vv); + if (r != NULL) + r[BITINT_END (rn - 1, 0)] = k >> s; + } + else +#endif + { + UWtype vv1 = v2[BITINT_END (0, vn - 1)]; + UWtype vv0 = v2[BITINT_END (1, vn - 2)]; + /* Main loop. */ + for (SItype j = un - vn; j >= 0; --j) + { + /* Compute estimate in qhat. */ + UWtype uv1 = u2[BITINT_END (un - j - vn, j + vn)]; + UWtype uv0 = u2[BITINT_END (un - j - vn + 1, j + vn - 1)]; + UWtype qhat, rhat, hi, lo, c; + if (uv1 >= vv1) + { + /* udiv_qrnnd doesn't support quotients which don't + fit into UWtype, so subtract from uv1:uv0 vv1 + first. */ + uv1 -= vv1 + __builtin_sub_overflow (uv0, vv1, &uv0); + udiv_qrnnd (qhat, rhat, uv1, uv0, vv1); + if (!__builtin_add_overflow (rhat, vv1, &rhat)) + goto again; + } + else + { + udiv_qrnnd (qhat, rhat, uv1, uv0, vv1); + again: + umul_ppmm (hi, lo, qhat, vv0); + if (hi > rhat + || (hi == rhat + && lo > u2[BITINT_END (un - j - vn + 2, + j + vn - 2)])) + { + --qhat; + if (!__builtin_add_overflow (rhat, vv1, &rhat)) + goto again; + } + } + + c = bitint_submul_1 (u2 + BITINT_END (un - j, j), + v2 + BITINT_END (vn - 1, 0), qhat, vn); + u2[BITINT_END (un - j - vn, j + vn)] -= c; + /* If we've subtracted too much, decrease qhat and + and add back. */ + if ((Wtype) u2[BITINT_END (un - j - vn, j + vn)] < 0) + { + --qhat; + c = 0; + for (USItype i = 0; i < vn; ++i) + { + UWtype s = v2[BITINT_END (vn - 1 - i, i)]; + UWtype d = u2[BITINT_END (un - i - j, i + j)]; + UWtype c1 = __builtin_add_overflow (d, s, &d); + UWtype c2 = __builtin_add_overflow (d, c, &d); + c = c1 + c2; + u2[BITINT_END (un - i - j, i + j)] = d; + } + u2[BITINT_END (un - j - vn, j + vn)] += c; + } + q2[BITINT_END (un - vn - j, j)] = qhat; + } + if (r != NULL) + { + if (s) + { + const SItype n = sizeof (UWtype) * __CHAR_BIT__; + /* Unnormalize remainder. */ + USItype i; + for (i = 0; i < vn && i < rn; ++i) + r[BITINT_END (rn - 1 - i, i)] + = ((u2[BITINT_END (un - i, i)] >> s) + | (u2[BITINT_END (un - i - 1, i + 1)] << (n - s))); + if (i < rn) + r[BITINT_END (rn - vn, vn - 1)] + = u2[BITINT_END (un - vn + 1, vn - 1)] >> s; + } + else if (rn > vn) + __builtin_memcpy (&r[BITINT_END (rn - vn, 0)], + &u2[BITINT_END (un + 1 - vn, 0)], + vn * sizeof (UWtype)); + else + __builtin_memcpy (&r[0], &u2[BITINT_END (un + 1 - rn, 0)], + rn * sizeof (UWtype)); + } + } + } + if (q != NULL) + { + if ((uprec < 0) ^ (vprec < 0)) + { + /* Negative quotient. */ + USItype n; + if (un - vn + 1 > qn) + n = qn; + else + n = un - vn + 1; + bitint_negate (q + BITINT_END (qn - 1, 0), + q2 + BITINT_END (un - vn, 0), n); + if (qn > n) + __builtin_memset (q + BITINT_END (0, n), -1, + (qn - n) * sizeof (UWtype)); + } + else + { + /* Positive quotient. */ + if (qn2) + __builtin_memcpy (q, q2 + BITINT_END (un - vn + 1 - qn, 0), + qn * sizeof (UWtype)); + else if (qn > un - vn + 1) + __builtin_memset (q + BITINT_END (0, un - vn + 1), 0, + (qn - (un - vn + 1)) * sizeof (UWtype)); + } + } + if (r != NULL) + { + if (uprec < 0) + { + /* Negative remainder. */ + bitint_negate (r + BITINT_END (rn - 1, 0), + r + BITINT_END (rn - 1, 0), + rn > vn ? vn : rn); + if (rn > vn) + __builtin_memset (r + BITINT_END (0, vn), -1, + (rn - vn) * sizeof (UWtype)); + } + else + { + /* Positive remainder. */ + if (rn > vn) + __builtin_memset (r + BITINT_END (0, vn), 0, + (rn - vn) * sizeof (UWtype)); + } + } +} +#endif +#endif + #ifdef L_cmpdi2 cmp_return_type __cmpdi2 (DWtype a, DWtype b) --- libgcc/libgcc2.h.jj 2023-07-11 13:39:49.817107100 +0200 +++ libgcc/libgcc2.h 2023-07-12 14:21:56.996358364 +0200 @@ -181,6 +181,12 @@ typedef int shift_count_type __attribute #define float bogus_type #define double bogus_type +#if (defined(__BITINT_MAXWIDTH__) \ + && (defined(L_mulbitint3) || defined(L_divmodbitint4))) +#undef LIBGCC2_UNITS_PER_WORD +#define LIBGCC2_UNITS_PER_WORD (__LIBGCC_BITINT_LIMB_WIDTH__ / __CHAR_BIT__) +#endif + /* Versions prior to 3.4.4 were not taking into account the word size for the 5 trapping arithmetic functions absv, addv, subv, mulv and negv. As a consequence, the si and di variants were always and the only ones emitted. @@ -390,6 +396,15 @@ extern DWtype __divmoddi4 (DWtype, DWtyp extern UDWtype __udivmoddi4 (UDWtype, UDWtype, UDWtype *); #endif +#if (defined(__BITINT_MAXWIDTH__) \ + && (defined(L_mulbitint3) || defined(L_divmodbitint4))) +/* _BitInt support. */ +extern void __mulbitint3 (UWtype *, SItype, const UWtype *, SItype, + const UWtype *, SItype); +extern void __divmodbitint4 (UWtype *, SItype, UWtype *, SItype, + const UWtype *, SItype, const UWtype *, SItype); +#endif + /* __negdi2 is static inline when building other libgcc2 portions. */ #if !defined(L_divdi3) && !defined(L_moddi3) extern DWtype __negdi2 (DWtype); --- libgcc/libgcc-std.ver.in.jj 2023-07-11 13:39:49.775107662 +0200 +++ libgcc/libgcc-std.ver.in 2023-07-17 09:45:50.463182114 +0200 @@ -1944,3 +1944,13 @@ GCC_7.0.0 { __PFX__divmoddi4 __PFX__divmodti4 } + +%inherit GCC_14.0.0 GCC_7.0.0 +GCC_14.0.0 { + __PFX__mulbitint3 + __PFX__divmodbitint4 + __PFX__fixsfbitint + __PFX__fixdfbitint + __PFX__floatbitintsf + __PFX__floatbitintdf +} --- libgcc/Makefile.in.jj 2023-07-11 13:39:49.722108372 +0200 +++ libgcc/Makefile.in 2023-07-14 12:32:32.134694076 +0200 @@ -446,7 +446,7 @@ lib2funcs = _muldi3 _negdi2 _lshrdi3 _as _paritysi2 _paritydi2 _powisf2 _powidf2 _powixf2 _powitf2 \ _mulhc3 _mulsc3 _muldc3 _mulxc3 _multc3 _divhc3 _divsc3 \ _divdc3 _divxc3 _divtc3 _bswapsi2 _bswapdi2 _clrsbsi2 \ - _clrsbdi2 + _clrsbdi2 _mulbitint3 # The floating-point conversion routines that involve a single-word integer. # XX stands for the integer mode. @@ -466,7 +466,8 @@ endif # These might cause a divide overflow trap and so are compiled with # unwinder info. LIB2_DIVMOD_FUNCS = _divdi3 _moddi3 _divmoddi4 \ - _udivdi3 _umoddi3 _udivmoddi4 _udiv_w_sdiv + _udivdi3 _umoddi3 _udivmoddi4 _udiv_w_sdiv \ + _divmodbitint4 # Remove any objects from lib2funcs and LIB2_DIVMOD_FUNCS that are # defined as optimized assembly code in LIB1ASMFUNCS or as C code --- libgcc/soft-fp/bitint.h.jj 2023-07-13 12:45:21.000663607 +0200 +++ libgcc/soft-fp/bitint.h 2023-07-14 23:35:51.539466644 +0200 @@ -0,0 +1,306 @@ +/* Software floating-point emulation. + Definitions for _BitInt implementation details. + + Copyright (C) 2023 Free Software Foundation, Inc. + +This file is part of GCC. + +GCC is free software; you can redistribute it and/or modify it under +the terms of the GNU General Public License as published by the Free +Software Foundation; either version 3, or (at your option) any later +version. + +GCC is distributed in the hope that it will be useful, but WITHOUT ANY +WARRANTY; without even the implied warranty of MERCHANTABILITY or +FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +for more details. + +Under Section 7 of GPL version 3, you are granted additional +permissions described in the GCC Runtime Library Exception, version +3.1, as published by the Free Software Foundation. + +You should have received a copy of the GNU General Public License and +a copy of the GCC Runtime Library Exception along with this program; +see the files COPYING3 and COPYING.RUNTIME respectively. If not, see +. */ + +#ifndef GCC_SOFT_FP_BITINT_H +#define GCC_SOFT_FP_BITINT_H + +#ifdef __BITINT_MAXWIDTH__ +#define BIL_UNITS_PER_WORD (__LIBGCC_BITINT_LIMB_WIDTH__ / __CHAR_BIT__) + +#if BIL_UNITS_PER_WORD == 8 +#define BIL_TYPE_SIZE (8 * __CHAR_BIT__) +#define BILtype DItype +#define UBILtype UDItype +#elif BIL_UNITS_PER_WORD == 4 +#define BIL_TYPE_SIZE (4 * __CHAR_BIT__) +#define BILtype SItype +#define UBILtype USItype +#elif BIL_UNITS_PER_WORD == 2 +#define BIL_TYPE_SIZE (2 * __CHAR_BIT__) +#define BILtype HItype +#define UBILtype UHItype +#else +#define BIL_TYPE_SIZE __CHAR_BIT__ +#define BILtype QItype +#define UBILtype UQItype +#endif + +/* If *P is zero or sign extended (the latter only for PREC < 0) from + some narrower _BitInt value, reduce precision. */ + +static inline __attribute__((__always_inline__)) SItype +bitint_reduce_prec (const UBILtype **p, SItype prec) +{ + UBILtype mslimb; + SItype i; + if (prec < 0) + { +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ + i = 0; +#else + i = ((USItype) -1 - prec) / BIL_TYPE_SIZE; +#endif + mslimb = (*p)[i]; + if (mslimb & ((UBILtype) 1 << (((USItype) -1 - prec) % BIL_TYPE_SIZE))) + { + SItype n = ((USItype) -prec) % BIL_TYPE_SIZE; + if (n) + { + mslimb |= ((UBILtype) -1 << (((USItype) -1 - prec) % BIL_TYPE_SIZE)); + if (mslimb == (UBILtype) -1) + { + prec += n; + if (prec >= -1) + return -2; +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ + ++p; +#else + --i; +#endif + mslimb = (*p)[i]; + n = 0; + } + } + while (mslimb == (UBILtype) -1) + { + prec += BIL_TYPE_SIZE; + if (prec >= -1) + return -2; +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ + ++p; +#else + --i; +#endif + mslimb = (*p)[i]; + } + if (n == 0) + { + if ((BILtype) mslimb >= 0) + { +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ + --p; +#endif + return prec - 1; + } + } + return prec; + } + else + prec = -prec; + } + else + { +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ + i = 0; +#else + i = ((USItype) prec - 1) / BIL_TYPE_SIZE; +#endif + mslimb = (*p)[i]; + } + SItype n = ((USItype) prec) % BIL_TYPE_SIZE; + if (n) + { + mslimb &= ((UBILtype) 1 << (((USItype) prec) % BIL_TYPE_SIZE)) - 1; + if (mslimb == 0) + { + prec -= n; + if (prec == 0) + return 1; +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ + ++p; +#else + --i; +#endif + mslimb = (*p)[i]; + } + } + while (mslimb == 0) + { + prec -= BIL_TYPE_SIZE; + if (prec == 0) + return 1; +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ + ++p; +#else + --i; +#endif + mslimb = (*p)[i]; + } + return prec; +} + +#if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ +# define BITINT_INC -1 +# define BITINT_END(be, le) (be) +#else +# define BITINT_INC 1 +# define BITINT_END(be, le) (le) +#endif + +#define FP_TO_BITINT(r, rn, arprec, shift, rv, rsize, rsigned, ovf, DI) \ + if (ovf) \ + { \ + if ((rv & 1) != 0) \ + __builtin_memset (r, -1, rn * sizeof (UBILtype)); \ + else \ + __builtin_memset (r, 0, rn * sizeof (UBILtype)); \ + if (rv & (((U##DI##type) 1) << (rsize - 1))) \ + r[BITINT_END (0, rn - 1)] \ + |= (UBILtype) -1 << ((arprec - 1) % BIL_TYPE_SIZE); \ + else \ + r[BITINT_END (0, rn - 1)] \ + &= ~((UBILtype) -1 << ((arprec - 1) % BIL_TYPE_SIZE)); \ + } \ + else \ + { \ + USItype shiftl = shift / BIL_TYPE_SIZE; \ + rsize = DI##_BITS; \ + if (rsigned && (DI##type) rv >= 0) \ + rsigned = 0; \ + if (shift + DI##_BITS > arprec) \ + rsize = arprec - shift; \ + USItype shiftr = shift % BIL_TYPE_SIZE; \ + if (shiftl) \ + __builtin_memset (r + BITINT_END (rn - shiftl, 0), 0, \ + shiftl * sizeof (UBILtype)); \ + USItype idx = BITINT_END (rn - shiftl - 1, shiftl); \ + DI##type rvs = rv; \ + if (shiftr) \ + { \ + r[idx] = (rsigned ? (UBILtype) rvs : (UBILtype) rv) << shiftr;\ + idx += BITINT_INC; \ + if (rsize > BIL_TYPE_SIZE - shiftr) \ + { \ + rv >>= BIL_TYPE_SIZE - shiftr; \ + rvs >>= BIL_TYPE_SIZE - shiftr; \ + rsize -= BIL_TYPE_SIZE - shiftr; \ + } \ + else \ + rsize = 0; \ + } \ + while (rsize) \ + { \ + r[idx] = rsigned ? (UBILtype) rvs : (UBILtype) rv; \ + idx += BITINT_INC; \ + if (rsize <= BIL_TYPE_SIZE) \ + break; \ + rv >>= (DI##_BITS > BIL_TYPE_SIZE ? BIL_TYPE_SIZE : 0); \ + rvs >>= (DI##_BITS > BIL_TYPE_SIZE ? BIL_TYPE_SIZE : 0); \ + rsize -= BIL_TYPE_SIZE; \ + } \ + if (idx < rn) \ + __builtin_memset (r + BITINT_END (0, idx), rsigned ? -1 : 0, \ + BITINT_END (idx + 1, rn - idx) \ + * sizeof (UBILtype)); \ + } + +#define FP_FROM_BITINT(i, iprec, iv, shift, DI) \ + do \ + { \ + iprec = bitint_reduce_prec (&i, iprec); \ + USItype aiprec = iprec < 0 ? -iprec : iprec; \ + USItype in = (aiprec + BIL_TYPE_SIZE - 1) / BIL_TYPE_SIZE; \ + USItype idx = BITINT_END (0, in - 1); \ + UBILtype msb = i[idx]; \ + SItype n = 0; \ + if (aiprec % BIL_TYPE_SIZE) \ + { \ + if (iprec > 0) \ + msb &= ((UBILtype) 1 << (aiprec % BIL_TYPE_SIZE)) - 1; \ + else \ + msb |= (UBILtype) -1 << (aiprec % BIL_TYPE_SIZE); \ + } \ + if (iprec < 0) \ + { \ + n = sizeof (0ULL) * __CHAR_BIT__ + 1 - __builtin_clzll (~msb);\ + if (BIL_TYPE_SIZE > DI##_BITS && n > DI##_BITS) \ + { \ + iv = msb >> (n - DI##_BITS); \ + shift = n - DI##_BITS; \ + n = 0; \ + } \ + else \ + { \ + iv = (BILtype) msb; \ + n = DI##_BITS - n; \ + } \ + } \ + /* bitint_reduce_prec guarantees that if msb is 0, then whole \ + i must be zero, otherwise it would have reduced the \ + precision. */ \ + else if (msb == 0) \ + iv = 0; \ + else \ + { \ + n = sizeof (0ULL) * __CHAR_BIT__ - __builtin_clzll (msb); \ + if (BIL_TYPE_SIZE >= DI##_BITS && n >= DI##_BITS) \ + { \ + iv = msb >> (n - DI##_BITS + 1); \ + shift = n - DI##_BITS + 1; \ + n = 0; \ + } \ + else \ + { \ + iv = msb; \ + n = DI##_BITS - 1 - n; \ + } \ + } \ + while (n && BITINT_END (idx < in - 1, idx)) \ + { \ + idx -= BITINT_INC; \ + msb = i[idx]; \ + if (BIL_TYPE_SIZE < DI##_BITS && n >= BIL_TYPE_SIZE) \ + { \ + iv = (U##DI##type) iv << (BIL_TYPE_SIZE < DI##_BITS \ + ? BIL_TYPE_SIZE : 0); \ + iv |= msb; \ + n -= BIL_TYPE_SIZE; \ + } \ + else \ + { \ + iv = (U##DI##type) iv << n; \ + iv |= msb >> (BIL_TYPE_SIZE - n); \ + shift = BIL_TYPE_SIZE - n; \ + break; \ + } \ + } \ + \ + UBILtype low_bits = 0; \ + if (shift) \ + low_bits = msb & (((UBILtype) 1 << shift) - 1); \ + shift += BITINT_END (in - 1 - idx, idx) * BIL_TYPE_SIZE; \ + while (!low_bits && BITINT_END (idx < in - 1, idx)) \ + { \ + idx -= BITINT_INC; \ + low_bits |= i[idx]; \ + } \ + iv |= (low_bits != 0); \ + } \ + while (0) + +#endif /* __BITINT_MAXWIDTH__ */ + +#endif /* GCC_SOFT_FP_BITINT_H */ --- libgcc/soft-fp/fixdfbitint.c.jj 2023-07-13 12:50:52.639175463 +0200 +++ libgcc/soft-fp/fixdfbitint.c 2023-07-14 09:11:13.705084773 +0200 @@ -0,0 +1,71 @@ +/* Software floating-point emulation. + Convert IEEE double to signed or unsigned _BitInt. + + Copyright (C) 2023 Free Software Foundation, Inc. + +This file is part of GCC. + +GCC is free software; you can redistribute it and/or modify it under +the terms of the GNU General Public License as published by the Free +Software Foundation; either version 3, or (at your option) any later +version. + +GCC is distributed in the hope that it will be useful, but WITHOUT ANY +WARRANTY; without even the implied warranty of MERCHANTABILITY or +FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +for more details. + +Under Section 7 of GPL version 3, you are granted additional +permissions described in the GCC Runtime Library Exception, version +3.1, as published by the Free Software Foundation. + +You should have received a copy of the GNU General Public License and +a copy of the GCC Runtime Library Exception along with this program; +see the files COPYING3 and COPYING.RUNTIME respectively. If not, see +. */ + +#include "soft-fp.h" +#include "double.h" +#include "bitint.h" + +#ifdef __BITINT_MAXWIDTH__ +void +__fixdfbitint (UBILtype *r, SItype rprec, DFtype a) +{ + FP_DECL_EX; + FP_DECL_D (A); + USItype arprec = rprec < 0 ? -rprec : rprec; + USItype rn = ((USItype) arprec + BIL_TYPE_SIZE - 1) / BIL_TYPE_SIZE; + UDItype rv; + USItype rsize = arprec > DI_BITS ? DI_BITS : arprec; + USItype rsigned = rprec < 0; + USItype ovf = 0; + USItype shift = 0; + + FP_INIT_EXCEPTIONS; + FP_UNPACK_RAW_D (A, a); + if (arprec > DI_BITS) + { + if (A_e < _FP_EXPBIAS_D || (A_s && !rsigned)) + ovf = 1; + else if (A_e >= (_FP_EXPMAX_D < _FP_EXPBIAS_D + arprec + ? _FP_EXPMAX_D + : _FP_EXPBIAS_D + arprec - rsigned)) + { + ovf = 1; + if (A_s + && A_e == _FP_EXPBIAS_D + arprec - 1 + && A_e < _FP_EXPMAX_D) + A_e -= arprec - DI_BITS; + } + else if (A_e >= _FP_EXPBIAS_D + DI_BITS - rsigned) + { + shift = A_e - (_FP_EXPBIAS_D + DI_BITS - rsigned - 1); + A_e -= shift; + } + } + FP_TO_INT_D (rv, A, rsize, rsigned); + FP_HANDLE_EXCEPTIONS; + FP_TO_BITINT (r, rn, arprec, shift, rv, rsize, rsigned, ovf, DI); +} +#endif --- libgcc/soft-fp/fixsfbitint.c.jj 2023-07-14 09:17:15.812063032 +0200 +++ libgcc/soft-fp/fixsfbitint.c 2023-07-14 09:22:00.174119081 +0200 @@ -0,0 +1,71 @@ +/* Software floating-point emulation. + Convert IEEE single to signed or unsigned _BitInt. + + Copyright (C) 2023 Free Software Foundation, Inc. + +This file is part of GCC. + +GCC is free software; you can redistribute it and/or modify it under +the terms of the GNU General Public License as published by the Free +Software Foundation; either version 3, or (at your option) any later +version. + +GCC is distributed in the hope that it will be useful, but WITHOUT ANY +WARRANTY; without even the implied warranty of MERCHANTABILITY or +FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +for more details. + +Under Section 7 of GPL version 3, you are granted additional +permissions described in the GCC Runtime Library Exception, version +3.1, as published by the Free Software Foundation. + +You should have received a copy of the GNU General Public License and +a copy of the GCC Runtime Library Exception along with this program; +see the files COPYING3 and COPYING.RUNTIME respectively. If not, see +. */ + +#include "soft-fp.h" +#include "single.h" +#include "bitint.h" + +#ifdef __BITINT_MAXWIDTH__ +void +__fixsfbitint (UBILtype *r, SItype rprec, SFtype a) +{ + FP_DECL_EX; + FP_DECL_S (A); + USItype arprec = rprec < 0 ? -rprec : rprec; + USItype rn = ((USItype) arprec + BIL_TYPE_SIZE - 1) / BIL_TYPE_SIZE; + USItype rv; + USItype rsize = arprec > SI_BITS ? SI_BITS : arprec; + USItype rsigned = rprec < 0; + USItype ovf = 0; + USItype shift = 0; + + FP_INIT_EXCEPTIONS; + FP_UNPACK_RAW_S (A, a); + if (arprec > SI_BITS) + { + if (A_e < _FP_EXPBIAS_S || (A_s && !rsigned)) + ovf = 1; + else if (A_e >= (_FP_EXPMAX_S < _FP_EXPBIAS_S + arprec + ? _FP_EXPMAX_S + : _FP_EXPBIAS_S + arprec - rsigned)) + { + ovf = 1; + if (A_s + && A_e == _FP_EXPBIAS_S + arprec - 1 + && A_e < _FP_EXPMAX_S) + A_e -= arprec - SI_BITS; + } + else if (A_e >= _FP_EXPBIAS_S + SI_BITS - rsigned) + { + shift = A_e - (_FP_EXPBIAS_S + SI_BITS - rsigned - 1); + A_e -= shift; + } + } + FP_TO_INT_S (rv, A, rsize, rsigned); + FP_HANDLE_EXCEPTIONS; + FP_TO_BITINT (r, rn, arprec, shift, rv, rsize, rsigned, ovf, SI); +} +#endif --- libgcc/soft-fp/fixtfbitint.c.jj 2023-07-14 12:24:54.024856956 +0200 +++ libgcc/soft-fp/fixtfbitint.c 2023-07-14 09:34:17.091902919 +0200 @@ -0,0 +1,81 @@ +/* Software floating-point emulation. + Convert IEEE quad to signed or unsigned _BitInt. + + Copyright (C) 2023 Free Software Foundation, Inc. + +This file is part of GCC. + +GCC is free software; you can redistribute it and/or modify it under +the terms of the GNU General Public License as published by the Free +Software Foundation; either version 3, or (at your option) any later +version. + +GCC is distributed in the hope that it will be useful, but WITHOUT ANY +WARRANTY; without even the implied warranty of MERCHANTABILITY or +FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +for more details. + +Under Section 7 of GPL version 3, you are granted additional +permissions described in the GCC Runtime Library Exception, version +3.1, as published by the Free Software Foundation. + +You should have received a copy of the GNU General Public License and +a copy of the GCC Runtime Library Exception along with this program; +see the files COPYING3 and COPYING.RUNTIME respectively. If not, see +. */ + +#include "soft-fp.h" +#include "quad.h" +#include "bitint.h" + +#ifdef __BITINT_MAXWIDTH__ + +#ifndef TI_BITS +/* As mantissa is 112 bits + 1 implicit bit, we need 128-bit + type, but on most 32-bit architectures TImode isn't supported. + Use _BitInt(128) instead. */ +typedef _BitInt(128) TItype; +typedef unsigned _BitInt(128) UTItype; +#define TI_BITS 128 +#endif + +void +__fixtfbitint (UBILtype *r, SItype rprec, TFtype a) +{ + FP_DECL_EX; + FP_DECL_Q (A); + USItype arprec = rprec < 0 ? -rprec : rprec; + USItype rn = ((USItype) arprec + BIL_TYPE_SIZE - 1) / BIL_TYPE_SIZE; + UTItype rv; + USItype rsize = arprec > TI_BITS ? TI_BITS : arprec; + USItype rsigned = rprec < 0; + USItype ovf = 0; + USItype shift = 0; + + FP_INIT_EXCEPTIONS; + FP_UNPACK_RAW_Q (A, a); + if (arprec > TI_BITS) + { + if (A_e < _FP_EXPBIAS_Q || (A_s && !rsigned)) + ovf = 1; + else if (A_e >= (_FP_EXPMAX_Q < _FP_EXPBIAS_Q + arprec + ? _FP_EXPMAX_Q + : _FP_EXPBIAS_Q + arprec - rsigned)) + { + ovf = 1; + if (A_s + && A_e == _FP_EXPBIAS_Q + arprec - 1 + && A_e < _FP_EXPMAX_Q) + A_e -= arprec - TI_BITS; + } + else if (A_e >= _FP_EXPBIAS_Q + TI_BITS - rsigned) + { + shift = A_e - (_FP_EXPBIAS_Q + TI_BITS - rsigned - 1); + A_e -= shift; + } + } + FP_TO_INT_Q (rv, A, rsize, rsigned); + FP_HANDLE_EXCEPTIONS; + FP_TO_BITINT (r, rn, arprec, shift, rv, rsize, rsigned, ovf, TI); +} +#endif --- libgcc/soft-fp/fixxfbitint.c.jj 2023-07-14 09:19:04.908549924 +0200 +++ libgcc/soft-fp/fixxfbitint.c 2023-07-14 10:10:22.936886921 +0200 @@ -0,0 +1,82 @@ +/* Software floating-point emulation. + Convert IEEE extended to signed or unsigned _BitInt. + + Copyright (C) 2023 Free Software Foundation, Inc. + +This file is part of GCC. + +GCC is free software; you can redistribute it and/or modify it under +the terms of the GNU General Public License as published by the Free +Software Foundation; either version 3, or (at your option) any later +version. + +GCC is distributed in the hope that it will be useful, but WITHOUT ANY +WARRANTY; without even the implied warranty of MERCHANTABILITY or +FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +for more details. + +Under Section 7 of GPL version 3, you are granted additional +permissions described in the GCC Runtime Library Exception, version +3.1, as published by the Free Software Foundation. + +You should have received a copy of the GNU General Public License and +a copy of the GCC Runtime Library Exception along with this program; +see the files COPYING3 and COPYING.RUNTIME respectively. If not, see +. */ + +#include "soft-fp.h" +#include "extended.h" +#include "bitint.h" + +#ifdef __BITINT_MAXWIDTH__ + +#ifndef TI_BITS +/* While mantissa is 64 bits including 1 explicit bit, extended.h uses + op-2.h for W_TYPE_SIZE 64 and op-4.h for W_TYPE_SIZE 32, so we have + to use 128-bit type here. On most 32-bit architectures TImode isn't + supported, so use _BitInt(128) instead. */ +typedef _BitInt(128) TItype; +typedef unsigned _BitInt(128) UTItype; +#define TI_BITS 128 +#endif + +void +__fixxfbitint (UBILtype *r, SItype rprec, XFtype a) +{ + FP_DECL_EX; + FP_DECL_E (A); + USItype arprec = rprec < 0 ? -rprec : rprec; + USItype rn = ((USItype) arprec + BIL_TYPE_SIZE - 1) / BIL_TYPE_SIZE; + UTItype rv; + USItype rsize = arprec > TI_BITS ? TI_BITS : arprec; + USItype rsigned = rprec < 0; + USItype ovf = 0; + USItype shift = 0; + + FP_INIT_EXCEPTIONS; + FP_UNPACK_RAW_E (A, a); + if (arprec > TI_BITS) + { + if (A_e < _FP_EXPBIAS_E || (A_s && !rsigned)) + ovf = 1; + else if (A_e >= (_FP_EXPMAX_E < _FP_EXPBIAS_E + arprec + ? _FP_EXPMAX_E + : _FP_EXPBIAS_E + arprec - rsigned)) + { + ovf = 1; + if (A_s + && A_e == _FP_EXPBIAS_E + arprec - 1 + && A_e < _FP_EXPMAX_E) + A_e -= arprec - TI_BITS; + } + else if (A_e >= _FP_EXPBIAS_E + TI_BITS - rsigned) + { + shift = A_e - (_FP_EXPBIAS_E + TI_BITS - rsigned - 1); + A_e -= shift; + } + } + FP_TO_INT_E (rv, A, rsize, rsigned); + FP_HANDLE_EXCEPTIONS; + FP_TO_BITINT (r, rn, arprec, shift, rv, rsize, rsigned, ovf, TI); +} +#endif --- libgcc/soft-fp/floatbitintbf.c.jj 2023-07-14 23:42:07.422424792 +0200 +++ libgcc/soft-fp/floatbitintbf.c 2023-07-17 09:48:02.380392453 +0200 @@ -0,0 +1,59 @@ +/* Software floating-point emulation. + Convert a _BitInt to bfloat16. + + Copyright (C) 2023 Free Software Foundation, Inc. + +This file is part of GCC. + +GCC is free software; you can redistribute it and/or modify it under +the terms of the GNU General Public License as published by the Free +Software Foundation; either version 3, or (at your option) any later +version. + +GCC is distributed in the hope that it will be useful, but WITHOUT ANY +WARRANTY; without even the implied warranty of MERCHANTABILITY or +FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +for more details. + +Under Section 7 of GPL version 3, you are granted additional +permissions described in the GCC Runtime Library Exception, version +3.1, as published by the Free Software Foundation. + +You should have received a copy of the GNU General Public License and +a copy of the GCC Runtime Library Exception along with this program; +see the files COPYING3 and COPYING.RUNTIME respectively. If not, see +. */ + +#include "soft-fp.h" +#include "brain.h" +#include "bitint.h" + +#ifdef __BITINT_MAXWIDTH__ +BFtype +__floatbitintbf (const UBILtype *i, SItype iprec) +{ + SItype iv; + USItype shift = 0; + FP_DECL_EX; + FP_DECL_B (A); + BFtype a; + + FP_FROM_BITINT (i, iprec, iv, shift, SI); + FP_INIT_ROUNDMODE; + FP_FROM_INT_B (A, iv, SI_BITS, USItype); + if (shift) + { + A_e += shift; + if (A_e >= _FP_EXPMAX_B) + { + /* Exponent too big; overflow to infinity. */ + _FP_OVERFLOW_SEMIRAW (B, 1, A); + _FP_PACK_SEMIRAW (B, 1, A); + } + } + FP_PACK_RAW_B (a, A); + FP_HANDLE_EXCEPTIONS; + + return a; +} +#endif --- libgcc/soft-fp/floatbitintdf.c.jj 2023-07-14 13:47:58.082710779 +0200 +++ libgcc/soft-fp/floatbitintdf.c 2023-07-17 09:48:02.380392453 +0200 @@ -0,0 +1,64 @@ +/* Software floating-point emulation. + Convert a _BitInt to IEEE double. + + Copyright (C) 2023 Free Software Foundation, Inc. + +This file is part of GCC. + +GCC is free software; you can redistribute it and/or modify it under +the terms of the GNU General Public License as published by the Free +Software Foundation; either version 3, or (at your option) any later +version. + +GCC is distributed in the hope that it will be useful, but WITHOUT ANY +WARRANTY; without even the implied warranty of MERCHANTABILITY or +FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +for more details. + +Under Section 7 of GPL version 3, you are granted additional +permissions described in the GCC Runtime Library Exception, version +3.1, as published by the Free Software Foundation. + +You should have received a copy of the GNU General Public License and +a copy of the GCC Runtime Library Exception along with this program; +see the files COPYING3 and COPYING.RUNTIME respectively. If not, see +. */ + +#include "soft-fp.h" +#include "double.h" +#include "bitint.h" + +#ifdef __BITINT_MAXWIDTH__ +DFtype +__floatbitintdf (const UBILtype *i, SItype iprec) +{ + DItype iv; + USItype shift = 0; + FP_DECL_EX; + FP_DECL_D (A); + DFtype a; + + FP_FROM_BITINT (i, iprec, iv, shift, DI); + FP_INIT_ROUNDMODE; + FP_FROM_INT_D (A, iv, DI_BITS, UDItype); + if (shift) + { + A_e += shift; + if (A_e >= _FP_EXPMAX_D) + { + /* Exponent too big; overflow to infinity. */ +#if _FP_W_TYPE_SIZE < 64 + _FP_OVERFLOW_SEMIRAW (D, 2, A); + _FP_PACK_SEMIRAW (D, 2, A); +#else + _FP_OVERFLOW_SEMIRAW (D, 1, A); + _FP_PACK_SEMIRAW (D, 1, A); +#endif + } + } + FP_PACK_RAW_D (a, A); + FP_HANDLE_EXCEPTIONS; + + return a; +} +#endif --- libgcc/soft-fp/floatbitinthf.c.jj 2023-07-14 23:39:20.033670033 +0200 +++ libgcc/soft-fp/floatbitinthf.c 2023-07-17 09:48:02.380392453 +0200 @@ -0,0 +1,59 @@ +/* Software floating-point emulation. + Convert a _BitInt to IEEE half. + + Copyright (C) 2023 Free Software Foundation, Inc. + +This file is part of GCC. + +GCC is free software; you can redistribute it and/or modify it under +the terms of the GNU General Public License as published by the Free +Software Foundation; either version 3, or (at your option) any later +version. + +GCC is distributed in the hope that it will be useful, but WITHOUT ANY +WARRANTY; without even the implied warranty of MERCHANTABILITY or +FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +for more details. + +Under Section 7 of GPL version 3, you are granted additional +permissions described in the GCC Runtime Library Exception, version +3.1, as published by the Free Software Foundation. + +You should have received a copy of the GNU General Public License and +a copy of the GCC Runtime Library Exception along with this program; +see the files COPYING3 and COPYING.RUNTIME respectively. If not, see +. */ + +#include "soft-fp.h" +#include "half.h" +#include "bitint.h" + +#ifdef __BITINT_MAXWIDTH__ +HFtype +__floatbitinthf (const UBILtype *i, SItype iprec) +{ + SItype iv; + USItype shift = 0; + FP_DECL_EX; + FP_DECL_H (A); + HFtype a; + + FP_FROM_BITINT (i, iprec, iv, shift, SI); + FP_INIT_ROUNDMODE; + FP_FROM_INT_H (A, iv, SI_BITS, USItype); + if (shift) + { + A_e += shift; + if (A_e >= _FP_EXPMAX_H) + { + /* Exponent too big; overflow to infinity. */ + _FP_OVERFLOW_SEMIRAW (H, 1, A); + _FP_PACK_SEMIRAW (H, 1, A); + } + } + FP_PACK_RAW_H (a, A); + FP_HANDLE_EXCEPTIONS; + + return a; +} +#endif --- libgcc/soft-fp/floatbitintsf.c.jj 2023-07-14 23:37:18.464300690 +0200 +++ libgcc/soft-fp/floatbitintsf.c 2023-07-17 09:48:02.380392453 +0200 @@ -0,0 +1,59 @@ +/* Software floating-point emulation. + Convert a _BitInt to IEEE single. + + Copyright (C) 2023 Free Software Foundation, Inc. + +This file is part of GCC. + +GCC is free software; you can redistribute it and/or modify it under +the terms of the GNU General Public License as published by the Free +Software Foundation; either version 3, or (at your option) any later +version. + +GCC is distributed in the hope that it will be useful, but WITHOUT ANY +WARRANTY; without even the implied warranty of MERCHANTABILITY or +FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +for more details. + +Under Section 7 of GPL version 3, you are granted additional +permissions described in the GCC Runtime Library Exception, version +3.1, as published by the Free Software Foundation. + +You should have received a copy of the GNU General Public License and +a copy of the GCC Runtime Library Exception along with this program; +see the files COPYING3 and COPYING.RUNTIME respectively. If not, see +. */ + +#include "soft-fp.h" +#include "single.h" +#include "bitint.h" + +#ifdef __BITINT_MAXWIDTH__ +SFtype +__floatbitintsf (const UBILtype *i, SItype iprec) +{ + SItype iv; + USItype shift = 0; + FP_DECL_EX; + FP_DECL_S (A); + SFtype a; + + FP_FROM_BITINT (i, iprec, iv, shift, SI); + FP_INIT_ROUNDMODE; + FP_FROM_INT_S (A, iv, SI_BITS, USItype); + if (shift) + { + A_e += shift; + if (A_e >= _FP_EXPMAX_S) + { + /* Exponent too big; overflow to infinity. */ + _FP_OVERFLOW_SEMIRAW (S, 1, A); + _FP_PACK_SEMIRAW (S, 1, A); + } + } + FP_PACK_RAW_S (a, A); + FP_HANDLE_EXCEPTIONS; + + return a; +} +#endif --- libgcc/soft-fp/floatbitinttf.c.jj 2023-07-17 09:32:55.202714681 +0200 +++ libgcc/soft-fp/floatbitinttf.c 2023-07-17 09:49:00.734600781 +0200 @@ -0,0 +1,73 @@ +/* Software floating-point emulation. + Convert a _BitInt to IEEE quad. + + Copyright (C) 2023 Free Software Foundation, Inc. + +This file is part of GCC. + +GCC is free software; you can redistribute it and/or modify it under +the terms of the GNU General Public License as published by the Free +Software Foundation; either version 3, or (at your option) any later +version. + +GCC is distributed in the hope that it will be useful, but WITHOUT ANY +WARRANTY; without even the implied warranty of MERCHANTABILITY or +FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +for more details. + +Under Section 7 of GPL version 3, you are granted additional +permissions described in the GCC Runtime Library Exception, version +3.1, as published by the Free Software Foundation. + +You should have received a copy of the GNU General Public License and +a copy of the GCC Runtime Library Exception along with this program; +see the files COPYING3 and COPYING.RUNTIME respectively. If not, see +. */ + +#include "soft-fp.h" +#include "quad.h" +#include "bitint.h" + +#ifdef __BITINT_MAXWIDTH__ +#ifndef TI_BITS +/* As mantissa is 112 bits + 1 implicit bit, we need 128-bit + type, but on most 32-bit architectures TImode isn't supported. + Use _BitInt(128) instead. */ +typedef _BitInt(128) TItype; +typedef unsigned _BitInt(128) UTItype; +#define TI_BITS 128 +#endif + +TFtype +__floatbitinttf (const UBILtype *i, SItype iprec) +{ + TItype iv; + USItype shift = 0; + FP_DECL_EX; + FP_DECL_Q (A); + TFtype a; + + FP_FROM_BITINT (i, iprec, iv, shift, TI); + FP_INIT_ROUNDMODE; + FP_FROM_INT_Q (A, iv, TI_BITS, UTItype); + if (shift) + { + A_e += shift; + if (A_e >= _FP_EXPMAX_Q) + { + /* Exponent too big; overflow to infinity. */ +#if _FP_W_TYPE_SIZE < 64 + _FP_OVERFLOW_SEMIRAW (Q, 4, A); + _FP_PACK_SEMIRAW (Q, 4, A); +#else + _FP_OVERFLOW_SEMIRAW (Q, 2, A); + _FP_PACK_SEMIRAW (Q, 2, A); +#endif + } + } + FP_PACK_RAW_Q (a, A); + FP_HANDLE_EXCEPTIONS; + + return a; +} +#endif --- libgcc/soft-fp/floatbitintxf.c.jj 2023-07-17 09:28:59.311920588 +0200 +++ libgcc/soft-fp/floatbitintxf.c 2023-07-17 09:48:02.381392439 +0200 @@ -0,0 +1,74 @@ +/* Software floating-point emulation. + Convert a _BitInt to IEEE extended. + + Copyright (C) 2023 Free Software Foundation, Inc. + +This file is part of GCC. + +GCC is free software; you can redistribute it and/or modify it under +the terms of the GNU General Public License as published by the Free +Software Foundation; either version 3, or (at your option) any later +version. + +GCC is distributed in the hope that it will be useful, but WITHOUT ANY +WARRANTY; without even the implied warranty of MERCHANTABILITY or +FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +for more details. + +Under Section 7 of GPL version 3, you are granted additional +permissions described in the GCC Runtime Library Exception, version +3.1, as published by the Free Software Foundation. + +You should have received a copy of the GNU General Public License and +a copy of the GCC Runtime Library Exception along with this program; +see the files COPYING3 and COPYING.RUNTIME respectively. If not, see +. */ + +#include "soft-fp.h" +#include "extended.h" +#include "bitint.h" + +#ifdef __BITINT_MAXWIDTH__ +#ifndef TI_BITS +/* While mantissa is 64 bits including 1 explicit bit, extended.h uses + op-2.h for W_TYPE_SIZE 64 and op-4.h for W_TYPE_SIZE 32, so we have + to use 128-bit type here. On most 32-bit architectures TImode isn't + supported, so use _BitInt(128) instead. */ +typedef _BitInt(128) TItype; +typedef unsigned _BitInt(128) UTItype; +#define TI_BITS 128 +#endif + +XFtype +__floatbitintxf (const UBILtype *i, SItype iprec) +{ + TItype iv; + USItype shift = 0; + FP_DECL_EX; + FP_DECL_E (A); + XFtype a; + + FP_FROM_BITINT (i, iprec, iv, shift, TI); + FP_INIT_ROUNDMODE; + FP_FROM_INT_E (A, iv, TI_BITS, UTItype); + if (shift) + { + A_e += shift; + if (A_e >= _FP_EXPMAX_E) + { + /* Exponent too big; overflow to infinity. */ +#if _FP_W_TYPE_SIZE < 64 + _FP_OVERFLOW_SEMIRAW (E, 4, A); + _FP_PACK_SEMIRAW (E, 4, A); +#else + _FP_OVERFLOW_SEMIRAW (E, 2, A); + _FP_PACK_SEMIRAW (E, 2, A); +#endif + } + } + FP_PACK_RAW_E (a, A); + FP_HANDLE_EXCEPTIONS; + + return a; +} +#endif --- libgcc/soft-fp/op-common.h.jj 2022-08-17 16:59:09.303543880 +0200 +++ libgcc/soft-fp/op-common.h 2023-07-17 12:02:29.369681164 +0200 @@ -1800,7 +1800,7 @@ if ((X##_s = ((r) < 0))) \ _FP_FROM_INT_ur = -_FP_FROM_INT_ur; \ \ - _FP_STATIC_ASSERT ((rsize) <= 2 * _FP_W_TYPE_SIZE, \ + _FP_STATIC_ASSERT ((rsize) <= 4 * _FP_W_TYPE_SIZE, \ "rsize too large"); \ (void) (((rsize) <= _FP_W_TYPE_SIZE) \ ? ({ \ @@ -1810,13 +1810,38 @@ X##_e = (_FP_EXPBIAS_##fs + _FP_W_TYPE_SIZE - 1 \ - _FP_FROM_INT_lz); \ }) \ - : ({ \ + : ((rsize) <= 2 * _FP_W_TYPE_SIZE) \ + ? ({ \ int _FP_FROM_INT_lz; \ __FP_CLZ_2 (_FP_FROM_INT_lz, \ (_FP_W_TYPE) (_FP_FROM_INT_ur \ >> _FP_W_TYPE_SIZE), \ (_FP_W_TYPE) _FP_FROM_INT_ur); \ - X##_e = (_FP_EXPBIAS_##fs + 2 * _FP_W_TYPE_SIZE - 1 \ + X##_e = (_FP_EXPBIAS_##fs \ + + 2 * _FP_W_TYPE_SIZE - 1 \ + - _FP_FROM_INT_lz); \ + }) \ + : ({ \ + int _FP_FROM_INT_lz; \ + if (_FP_FROM_INT_ur >> (2 * _FP_W_TYPE_SIZE)) \ + { \ + rtype _FP_FROM_INT_uru \ + = _FP_FROM_INT_ur >> (2 * _FP_W_TYPE_SIZE); \ + __FP_CLZ_2 (_FP_FROM_INT_lz, \ + (_FP_W_TYPE) (_FP_FROM_INT_uru \ + >> _FP_W_TYPE_SIZE),\ + (_FP_W_TYPE) _FP_FROM_INT_uru); \ + } \ + else \ + { \ + __FP_CLZ_2 (_FP_FROM_INT_lz, \ + (_FP_W_TYPE) (_FP_FROM_INT_ur \ + >> _FP_W_TYPE_SIZE),\ + (_FP_W_TYPE) _FP_FROM_INT_ur); \ + _FP_FROM_INT_lz += 2 * _FP_W_TYPE_SIZE; \ + } \ + X##_e = (_FP_EXPBIAS_##fs \ + + 4 * _FP_W_TYPE_SIZE - 1 \ - _FP_FROM_INT_lz); \ })); \ \