From patchwork Tue Sep 14 14:28:11 2010 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Patchwork-Submitter: "H.J. Lu" X-Patchwork-Id: 64703 Return-Path: X-Original-To: incoming@patchwork.ozlabs.org Delivered-To: patchwork-incoming@bilbo.ozlabs.org Received: from sourceware.org (server1.sourceware.org [209.132.180.131]) by ozlabs.org (Postfix) with SMTP id 8140EB6F0D for ; Wed, 15 Sep 2010 00:28:27 +1000 (EST) Received: (qmail 5512 invoked by alias); 14 Sep 2010 14:28:25 -0000 Received: (qmail 5476 invoked by uid 22791); 14 Sep 2010 14:28:22 -0000 X-SWARE-Spam-Status: No, hits=-1.8 required=5.0 tests=AWL, BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, FREEMAIL_FROM, RCVD_IN_DNSWL_NONE X-Spam-Check-By: sourceware.org Received: from mail-vw0-f47.google.com (HELO mail-vw0-f47.google.com) (209.85.212.47) by sourceware.org (qpsmtpd/0.43rc1) with ESMTP; Tue, 14 Sep 2010 14:28:13 +0000 Received: by vws9 with SMTP id 9so6073066vws.20 for ; Tue, 14 Sep 2010 07:28:11 -0700 (PDT) MIME-Version: 1.0 Received: by 10.220.47.163 with SMTP id n35mr4641vcf.32.1284474491523; Tue, 14 Sep 2010 07:28:11 -0700 (PDT) Received: by 10.220.78.193 with HTTP; Tue, 14 Sep 2010 07:28:11 -0700 (PDT) In-Reply-To: <20100914132037.GR16898@codesourcery.com> References: <20100913222732.GQ16898@codesourcery.com> <20100914132037.GR16898@codesourcery.com> Date: Tue, 14 Sep 2010 07:28:11 -0700 Message-ID: Subject: Re: PATCH: Fix bootstrap on x86 From: "H.J. Lu" To: Nathan Froyd Cc: Andrew Pinski , Uros Bizjak , gcc-patches@gcc.gnu.org X-IsSubscribed: yes Mailing-List: contact gcc-patches-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Id: List-Unsubscribe: List-Archive: List-Post: List-Help: Sender: gcc-patches-owner@gcc.gnu.org Delivered-To: mailing list gcc-patches@gcc.gnu.org On Tue, Sep 14, 2010 at 6:20 AM, Nathan Froyd wrote: > On Mon, Sep 13, 2010 at 03:34:43PM -0700, H.J. Lu wrote: >> On Mon, Sep 13, 2010 at 3:27 PM, Nathan Froyd wrote: >> > If this patch eliminates all instances of UNITS_PER_SIMD_WORD, you >> > should also poison that identifier in system.h. >> >> Here is the additional patch. > > A minor nit, but it'd be nice if the identifier was in the poison pragma > headed by: > > /* Old target macros that have moved to the target hooks structure.  */ > > earlier in the file. > Here is the updated patch. If this isn't reviewed in the next few hours, I will check in this simple patch: http://gcc.gnu.org/ml/gcc-patches/2010-09/msg01135.html to restore bootstrap on x86. Sorry for the breakage. Thanks. diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c index 1d547b0..6f260ec 100644 --- a/gcc/config/arm/arm.c +++ b/gcc/config/arm/arm.c @@ -236,6 +236,7 @@ static rtx arm_trampoline_adjust_address (rtx); static rtx arm_pic_static_addr (rtx orig, rtx reg); static bool cortex_a9_sched_adjust_cost (rtx, rtx, rtx, int *); static bool xscale_sched_adjust_cost (rtx, rtx, rtx, int *); +static unsigned int arm_units_per_simd_word (enum machine_mode); /* Table of machine attributes. */ @@ -363,6 +364,8 @@ static const struct attribute_spec arm_attribute_table[] = #define TARGET_SHIFT_TRUNCATION_MASK arm_shift_truncation_mask #undef TARGET_VECTOR_MODE_SUPPORTED_P #define TARGET_VECTOR_MODE_SUPPORTED_P arm_vector_mode_supported_p +#undef TARGET_VECTORIZE_UNITS_PER_SIMD_WORD +#define TARGET_VECTORIZE_UNITS_PER_SIMD_WORD arm_units_per_simd_word #undef TARGET_MACHINE_DEPENDENT_REORG #define TARGET_MACHINE_DEPENDENT_REORG arm_reorg @@ -21869,6 +21872,17 @@ arm_vector_mode_supported_p (enum machine_mode mode) return false; } +/* Use the option -mvectorize-with-neon-quad to override the use of doubleword + registers when autovectorizing for Neon, at least until multiple vector + widths are supported properly by the middle-end. */ + +static unsigned int +arm_units_per_simd_word (enum machine_mode mode ATTRIBUTE_UNUSED) +{ + return (TARGET_NEON + ? (TARGET_NEON_VECTORIZE_QUAD ? 16 : 8) : UNITS_PER_WORD); +} + /* Implements target hook small_register_classes_for_mode_p. */ bool arm_small_register_classes_for_mode_p (enum machine_mode mode ATTRIBUTE_UNUSED) diff --git a/gcc/config/arm/arm.h b/gcc/config/arm/arm.h index 5b66510..36b618a 100644 --- a/gcc/config/arm/arm.h +++ b/gcc/config/arm/arm.h @@ -580,12 +580,6 @@ extern int arm_arch_hwdiv; #define UNITS_PER_WORD 4 -/* Use the option -mvectorize-with-neon-quad to override the use of doubleword - registers when autovectorizing for Neon, at least until multiple vector - widths are supported properly by the middle-end. */ -#define UNITS_PER_SIMD_WORD(MODE) \ - (TARGET_NEON ? (TARGET_NEON_VECTORIZE_QUAD ? 16 : 8) : UNITS_PER_WORD) - /* True if natural alignment is used for doubleword types. */ #define ARM_DOUBLEWORD_ALIGN TARGET_AAPCS_BASED diff --git a/gcc/config/i386/i386-protos.h b/gcc/config/i386/i386-protos.h index c4c61f7..900b424 100644 --- a/gcc/config/i386/i386-protos.h +++ b/gcc/config/i386/i386-protos.h @@ -50,7 +50,6 @@ extern bool x86_extended_QIreg_mentioned_p (rtx); extern bool x86_extended_reg_mentioned_p (rtx); extern bool x86_maybe_negate_const_int (rtx *, enum machine_mode); extern enum machine_mode ix86_cc_mode (enum rtx_code, rtx, rtx); -extern unsigned int ix86_units_per_simd_word (enum machine_mode); extern int avx_vpermilp_parallel (rtx par, enum machine_mode mode); extern int avx_vperm2f128_parallel (rtx par, enum machine_mode mode); diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c index a1bc205..1d4e4ae 100644 --- a/gcc/config/i386/i386.c +++ b/gcc/config/i386/i386.c @@ -32341,7 +32341,7 @@ has_dispatch (rtx insn, int action) /* ??? No autovectorization into MMX or 3DNOW until we can reliably place emms and femms instructions. */ -unsigned int +static unsigned int ix86_units_per_simd_word (enum machine_mode mode) { /* Disable double precision vectorizer if needed. */ @@ -32609,6 +32609,9 @@ ix86_units_per_simd_word (enum machine_mode mode) #undef TARGET_VECTORIZE_BUILTIN_VEC_PERM_OK #define TARGET_VECTORIZE_BUILTIN_VEC_PERM_OK \ ix86_vectorize_builtin_vec_perm_ok +#undef TARGET_VECTORIZE_UNITS_PER_SIMD_WORD +#define TARGET_VECTORIZE_UNITS_PER_SIMD_WORD \ + ix86_units_per_simd_word #undef TARGET_SET_CURRENT_FUNCTION #define TARGET_SET_CURRENT_FUNCTION ix86_set_current_function diff --git a/gcc/config/i386/i386.h b/gcc/config/i386/i386.h index 6fc2249..fe48efd 100644 --- a/gcc/config/i386/i386.h +++ b/gcc/config/i386/i386.h @@ -1033,8 +1033,6 @@ enum target_cpu_default || (MODE) == V2SImode || (MODE) == SImode \ || (MODE) == V4HImode || (MODE) == V8QImode) -#define UNITS_PER_SIMD_WORD(MODE) ix86_units_per_simd_word (MODE) - #define VALID_DFP_MODE_P(MODE) \ ((MODE) == SDmode || (MODE) == DDmode || (MODE) == TDmode) diff --git a/gcc/config/mips/mips.c b/gcc/config/mips/mips.c index 20b63c7..945ee42 100644 --- a/gcc/config/mips/mips.c +++ b/gcc/config/mips/mips.c @@ -11140,6 +11140,14 @@ mips_scalar_mode_supported_p (enum machine_mode mode) return default_scalar_mode_supported_p (mode); } +/* Implement TARGET_VECTORIZE_UNITS_PER_SIMD_WORD. */ + +static bool +mips_units_per_simd_word (enum machine_mode mode ATTRIBUTE_UNUSED) +{ + return TARGET_PAIRED_SINGLE_FLOAT ? 8 : UNITS_PER_WORD; +} + /* Implement TARGET_INIT_LIBFUNCS. */ #include "config/gofast.h" @@ -16506,6 +16514,9 @@ mips_shift_truncation_mask (enum machine_mode mode) #undef TARGET_SCALAR_MODE_SUPPORTED_P #define TARGET_SCALAR_MODE_SUPPORTED_P mips_scalar_mode_supported_p +#undef TARGET_VECTORIZE_UNITS_PER_SIMD_WORD +#define TARGET_VECTORIZE_UNITS_PER_SIMD_WORD mips_units_per_simd_word + #undef TARGET_INIT_BUILTINS #define TARGET_INIT_BUILTINS mips_init_builtins #undef TARGET_EXPAND_BUILTIN diff --git a/gcc/config/mips/mips.h b/gcc/config/mips/mips.h index 74bf3d2..4f44130 100644 --- a/gcc/config/mips/mips.h +++ b/gcc/config/mips/mips.h @@ -1328,9 +1328,6 @@ enum mips_code_readable_setting { /* The number of bytes in a double. */ #define UNITS_PER_DOUBLE (TYPE_PRECISION (double_type_node) / BITS_PER_UNIT) -#define UNITS_PER_SIMD_WORD(MODE) \ - (TARGET_PAIRED_SINGLE_FLOAT ? 8 : UNITS_PER_WORD) - /* Set the sizes of the core types. */ #define SHORT_TYPE_SIZE 16 #define INT_TYPE_SIZE 32 diff --git a/gcc/config/rs6000/rs6000.c b/gcc/config/rs6000/rs6000.c index ab5da58..2ebe484 100644 --- a/gcc/config/rs6000/rs6000.c +++ b/gcc/config/rs6000/rs6000.c @@ -3570,6 +3570,18 @@ rs6000_builtin_vectorization_cost (enum vect_cost_for_stmt type_of_cost, } } +/* Implement targetm.vectorize.units_per_simd_word. */ + +static int +rs6000_units_per_simd_word (enum machine_mode mode ATTRIBUTE_UNUSED) +{ + return (TARGET_VSX ? UNITS_PER_VSX_WORD + : (TARGET_ALTIVEC ? UNITS_PER_ALTIVEC_WORD + : (TARGET_SPE ? UNITS_PER_SPE_WORD + : (TARGET_PAIRED_FLOAT ? UNITS_PER_PAIRED_WORD + : UNITS_PER_WORD)))); +} + /* Handle generic options of the form -mfoo=yes/no. NAME is the option name. VALUE is the option value. diff --git a/gcc/config/rs6000/rs6000.h b/gcc/config/rs6000/rs6000.h index 63f1bba..5b0e4c7 100644 --- a/gcc/config/rs6000/rs6000.h +++ b/gcc/config/rs6000/rs6000.h @@ -1150,13 +1150,6 @@ extern unsigned rs6000_pointer_size; #define PAIRED_VECTOR_MODE(MODE) \ ((MODE) == V2SFmode) -#define UNITS_PER_SIMD_WORD(MODE) \ - (TARGET_VSX ? UNITS_PER_VSX_WORD \ - : (TARGET_ALTIVEC ? UNITS_PER_ALTIVEC_WORD \ - : (TARGET_SPE ? UNITS_PER_SPE_WORD \ - : (TARGET_PAIRED_FLOAT ? UNITS_PER_PAIRED_WORD \ - : UNITS_PER_WORD)))) - /* Value is TRUE if hard register REGNO can hold a value of machine-mode MODE. */ #define HARD_REGNO_MODE_OK(REGNO, MODE) \ diff --git a/gcc/config/sparc/sparc.c b/gcc/config/sparc/sparc.c index ba034e6..71c94e1 100644 --- a/gcc/config/sparc/sparc.c +++ b/gcc/config/sparc/sparc.c @@ -434,6 +434,7 @@ static bool sparc_can_eliminate (const int, const int); static const char *sparc_mangle_type (const_tree); #endif static void sparc_trampoline_init (rtx, tree, rtx); +static bool sparc_units_per_simd_word (enum machine_mode); #ifdef SUBTARGET_ATTRIBUTE_TABLE /* Table of valid machine attributes. */ @@ -571,6 +572,9 @@ static bool fpu_option_set = false; #undef TARGET_VECTOR_MODE_SUPPORTED_P #define TARGET_VECTOR_MODE_SUPPORTED_P sparc_vector_mode_supported_p +#undef TARGET_VECTORIZE_UNITS_PER_SIMD_WORD +#define TARGET_VECTORIZE_UNITS_PER_SIMD_WORD sparc_units_per_simd_word + #undef TARGET_DWARF_HANDLE_FRAME_UNSPEC #define TARGET_DWARF_HANDLE_FRAME_UNSPEC sparc_dwarf_handle_frame_unspec @@ -6233,6 +6237,14 @@ sparc_vector_mode_supported_p (enum machine_mode mode) return TARGET_VIS && VECTOR_MODE_P (mode) ? true : false; } +/* Implement the TARGET_VECTORIZE_UNITS_PER_SIMD_WORD target hook. */ + +static bool +sparc_units_per_simd_word (enum machine_mode mode ATTRIBUTE_UNUSED) +{ + return TARGET_VIS ? 8 : UNITS_PER_WORD; +} + /* Return the string to output an unconditional branch to LABEL, which is the operand number of the label. diff --git a/gcc/config/sparc/sparc.h b/gcc/config/sparc/sparc.h index eae3959..21694eb 100644 --- a/gcc/config/sparc/sparc.h +++ b/gcc/config/sparc/sparc.h @@ -607,8 +607,6 @@ extern struct sparc_cpu_select sparc_select[]; #define MIN_UNITS_PER_WORD 4 #endif -#define UNITS_PER_SIMD_WORD(MODE) (TARGET_VIS ? 8 : UNITS_PER_WORD) - /* Now define the sizes of the C data types. */ #define SHORT_TYPE_SIZE 16 diff --git a/gcc/defaults.h b/gcc/defaults.h index 62b8228..a50052e 100644 --- a/gcc/defaults.h +++ b/gcc/defaults.h @@ -949,12 +949,6 @@ see the files COPYING3 and COPYING.RUNTIME respectively. If not, see #define HAS_LONG_UNCOND_BRANCH 0 #endif -/* By default, only attempt to parallelize bitwise operations, and - possibly adds/subtracts using bit-twiddling. */ -#ifndef UNITS_PER_SIMD_WORD -#define UNITS_PER_SIMD_WORD(MODE) UNITS_PER_WORD -#endif - /* Determine whether __cxa_atexit, rather than atexit, is used to register C++ destructors for local statics and global objects. */ #ifndef DEFAULT_USE_CXA_ATEXIT diff --git a/gcc/doc/tm.texi b/gcc/doc/tm.texi index 91fbaf7..8e66da5 100644 --- a/gcc/doc/tm.texi +++ b/gcc/doc/tm.texi @@ -995,13 +995,6 @@ Minimum number of units in a word. If this is undefined, the default is smallest value that @code{UNITS_PER_WORD} can have at run-time. @end defmac -@defmac UNITS_PER_SIMD_WORD (@var{mode}) -Number of units in the vectors that the vectorizer can produce for -scalar mode @var{mode}. The default is equal to @code{UNITS_PER_WORD}, -because the vectorizer can do some transformations even in absence of -specialized @acronym{SIMD} hardware. -@end defmac - @defmac POINTER_SIZE Width of a pointer, in bits. You must specify a value no wider than the width of @code{Pmode}. If it is not equal to the width of @code{Pmode}, @@ -5765,6 +5758,13 @@ the elements in the vectors should be of type @var{type}. @var{is_packed} parameter is true if the memory access is defined in a packed struct. @end deftypefn +@deftypefn {Target Hook} {unsigned int} TARGET_VECTORIZE_UNITS_PER_SIMD_WORD (enum machine_mode @var{mode}) +This hook should return th number of units in the vectors that the +vectorizer can produce for scalar mode @var{mode}. The default is +equal to @code{UNITS_PER_WORD}, because the vectorizer can do some +transformations even in absence of specialized @acronym{SIMD} hardware. +@end deftypefn + @node Anchored Addresses @section Anchored Addresses @cindex anchored addresses diff --git a/gcc/doc/tm.texi.in b/gcc/doc/tm.texi.in index a180a56..d25ead1 100644 --- a/gcc/doc/tm.texi.in +++ b/gcc/doc/tm.texi.in @@ -995,13 +995,6 @@ Minimum number of units in a word. If this is undefined, the default is smallest value that @code{UNITS_PER_WORD} can have at run-time. @end defmac -@defmac UNITS_PER_SIMD_WORD (@var{mode}) -Number of units in the vectors that the vectorizer can produce for -scalar mode @var{mode}. The default is equal to @code{UNITS_PER_WORD}, -because the vectorizer can do some transformations even in absence of -specialized @acronym{SIMD} hardware. -@end defmac - @defmac POINTER_SIZE Width of a pointer, in bits. You must specify a value no wider than the width of @code{Pmode}. If it is not equal to the width of @code{Pmode}, @@ -5765,6 +5758,13 @@ the elements in the vectors should be of type @var{type}. @var{is_packed} parameter is true if the memory access is defined in a packed struct. @end deftypefn +@hook TARGET_VECTORIZE_UNITS_PER_SIMD_WORD +This hook should return th number of units in the vectors that the +vectorizer can produce for scalar mode @var{mode}. The default is +equal to @code{UNITS_PER_WORD}, because the vectorizer can do some +transformations even in absence of specialized @acronym{SIMD} hardware. +@end deftypefn + @node Anchored Addresses @section Anchored Addresses @cindex anchored addresses diff --git a/gcc/system.h b/gcc/system.h index a5cfaba..35f9ac8 100644 --- a/gcc/system.h +++ b/gcc/system.h @@ -715,7 +715,7 @@ extern void fancy_abort (const char *, int, const char *) ATTRIBUTE_NORETURN; ALLOCATE_INITIAL_VALUE LEGITIMIZE_ADDRESS FRAME_POINTER_REQUIRED \ CAN_ELIMINATE TRAMPOLINE_TEMPLATE INITIALIZE_TRAMPOLINE \ TRAMPOLINE_ADJUST_ADDRESS STATIC_CHAIN STATIC_CHAIN_INCOMING \ - RETURN_POPS_ARGS + RETURN_POPS_ARGS UNITS_PER_SIMD_WORD /* Other obsolete target macros, or macros that used to be in target headers and were not used, and may be obsolete or may never have diff --git a/gcc/target.def b/gcc/target.def index cecf882..155f0fd 100644 --- a/gcc/target.def +++ b/gcc/target.def @@ -880,6 +880,14 @@ DEFHOOK (enum machine_mode mode, const_tree type, int misalignment, bool is_packed), default_builtin_support_vector_misalignment) +/* Return units per SIMD word. */ +DEFHOOK +(units_per_simd_word, + "", + unsigned int, + (enum machine_mode mode), + default_units_per_simd_word) + HOOK_VECTOR_END (vectorize) #undef HOOK_PREFIX diff --git a/gcc/targhooks.c b/gcc/targhooks.c index a0e24c0..046c07f 100644 --- a/gcc/targhooks.c +++ b/gcc/targhooks.c @@ -984,6 +984,15 @@ default_builtin_support_vector_misalignment (enum machine_mode mode, return false; } +/* By default, only attempt to parallelize bitwise operations, and + possibly adds/subtracts using bit-twiddling. */ + +unsigned int +default_units_per_simd_word (enum machine_mode mode ATTRIBUTE_UNUSED) +{ + return UNITS_PER_WORD; +} + /* Determine whether or not a pointer mode is valid. Assume defaults of ptr_mode or Pmode - can be overridden. */ bool diff --git a/gcc/targhooks.h b/gcc/targhooks.h index 4073cbe..a6231f9 100644 --- a/gcc/targhooks.h +++ b/gcc/targhooks.h @@ -86,6 +86,7 @@ extern bool default_builtin_support_vector_misalignment (enum machine_mode mode, const_tree, int, bool); +extern unsigned int default_units_per_simd_word (enum machine_mode mode); /* These are here, and not in hooks.[ch], because not all users of hooks.h include tm.h, and thus we don't have CUMULATIVE_ARGS. */ diff --git a/gcc/tree-vect-loop.c b/gcc/tree-vect-loop.c index 2ed8c76..5c35080 100644 --- a/gcc/tree-vect-loop.c +++ b/gcc/tree-vect-loop.c @@ -126,9 +126,10 @@ along with GCC; see the file COPYING3. If not see Target modeling: ================= Currently the only target specific information that is used is the - size of the vector (in bytes) - "UNITS_PER_SIMD_WORD". Targets that can - support different sizes of vectors, for now will need to specify one value - for "UNITS_PER_SIMD_WORD". More flexibility will be added in the future. + size of the vector (in bytes) - "TARGET_VECTORIZE_UNITS_PER_SIMD_WORD". + Targets that can support different sizes of vectors, for now will need + to specify one value for "TARGET_VECTORIZE_UNITS_PER_SIMD_WORD". More + flexibility will be added in the future. Since we only vectorize operations which vector form can be expressed using existing tree codes, to verify that an operation is diff --git a/gcc/tree-vect-stmts.c b/gcc/tree-vect-stmts.c index 6d15bda..9ae79a8 100644 --- a/gcc/tree-vect-stmts.c +++ b/gcc/tree-vect-stmts.c @@ -4675,7 +4675,8 @@ get_vectype_for_scalar_type (tree scalar_type) int nunits; tree vectype; - if (nbytes == 0 || nbytes >= UNITS_PER_SIMD_WORD (inner_mode)) + if (nbytes == 0 + || (nbytes >= targetm.vectorize.units_per_simd_word (inner_mode))) return NULL_TREE; /* We can't build a vector type of elements with alignment bigger than @@ -4691,9 +4692,9 @@ get_vectype_for_scalar_type (tree scalar_type) && GET_MODE_BITSIZE (inner_mode) != TYPE_PRECISION (scalar_type)) return NULL_TREE; - /* FORNOW: Only a single vector size per mode (UNITS_PER_SIMD_WORD) - is expected. */ - nunits = UNITS_PER_SIMD_WORD (inner_mode) / nbytes; + /* FORNOW: Only a single vector size per mode + (TARGET_VECTORIZE_UNITS_PER_SIMD_WORD) is expected. */ + nunits = targetm.vectorize.units_per_simd_word (inner_mode) / nbytes; vectype = build_vector_type (scalar_type, nunits); if (vect_print_dump_info (REPORT_DETAILS))