From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mgamail.intel.com (mgamail.intel.com [192.198.163.19]) by sourceware.org (Postfix) with ESMTPS id 3DAB93858416 for ; Thu, 30 May 2024 05:52:24 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org 3DAB93858416 Authentication-Results: sourceware.org; dmarc=pass (p=none dis=none) header.from=intel.com Authentication-Results: sourceware.org; spf=pass smtp.mailfrom=intel.com ARC-Filter: OpenARC Filter v1.0.0 sourceware.org 3DAB93858416 Authentication-Results: server2.sourceware.org; arc=none smtp.remote-ip=192.198.163.19 ARC-Seal: i=1; a=rsa-sha256; d=sourceware.org; s=key; t=1717048348; cv=none; b=J8eE9NYhciO5EffIO3PwQy/UDIx/ifoYhKTbzDcZv0j9FeqBOWoGFl9LG4euAfOdQS4m14aow5jhfdXafAa33II9Lh4f2DG8uuYR/8CM6T2pnSpIauj7kBBQC3mnJiYxs9qJ8eDUZMwT2RDATXYv+do7E8tVpQYMrWfZJFEHt5w= ARC-Message-Signature: i=1; a=rsa-sha256; d=sourceware.org; s=key; t=1717048348; c=relaxed/simple; bh=lFJBY+qr3a3JL20VPz4QXE9aiyLjxCILtH4ARvuAb0E=; h=DKIM-Signature:From:To:Subject:Date:Message-Id:MIME-Version; b=PplRBAxmfgyxUgVBvMQn2OXcv6TSckO68gI43Rf85DkCW71vIDljDu5hzuOCJzmZ2IeRoDNmzF9FGfZxAd7aShVkjHXjQzK+iSEgmy9Zu2DnqMYKatN/4Ak6eOs9JcvbJi1JTrVLll/79hdH82GNZI+6nWF0JJcO2P/1JgCxq4w= ARC-Authentication-Results: i=1; server2.sourceware.org DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1717048344; x=1748584344; h=from:to:cc:subject:date:message-id:mime-version: content-transfer-encoding; bh=lFJBY+qr3a3JL20VPz4QXE9aiyLjxCILtH4ARvuAb0E=; b=FGoxaMDiO0H/uRPgwvcntPRv+IOPsrncj9auPjqb8xS4y1xl7DrGspAI 3xIZAZz0SfFK0nPUeqsHRcpMuD9zSGl4naGIgkRVEACE4aOnQUfUp7mF2 q3UBSjhBw3ehM//S14OsuR/Q84D9ghKMSjV4bYLROXbAFUhGzRpvU52CG 4vR1QQFhxHlKcLcVT/D8UlDJ4YcbJtPaCFRxWZ4AmTESGdkXXMIqrWOwv PFPXsqQG/MBUFuDTOlnc02Nl3gzH/trTQ651Iax8DPimSjsyxZUCf5BrM JoH147r4VmrporkWrypKmX6PAdeD8gAsQj507radTo9wquhfM1Ci7IhyP g==; X-CSE-ConnectionGUID: o0hawA3STkSKBK1yQR4G6w== X-CSE-MsgGUID: X6KZ4zOZQ2ScFphxGzqHcQ== X-IronPort-AV: E=McAfee;i="6600,9927,11087"; a="13320295" X-IronPort-AV: E=Sophos;i="6.08,199,1712646000"; d="scan'208";a="13320295" Received: from orviesa010.jf.intel.com ([10.64.159.150]) by fmvoesa113.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 29 May 2024 22:52:23 -0700 X-CSE-ConnectionGUID: UPDLAdizQjaA++CMHQ7i5w== X-CSE-MsgGUID: IfUTTz+CRJqmozzmbXVNfg== X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="6.08,199,1712646000"; d="scan'208";a="35646558" Received: from shvmail03.sh.intel.com ([10.239.245.20]) by orviesa010.jf.intel.com with ESMTP; 29 May 2024 22:52:20 -0700 Received: from shliclel4217.sh.intel.com (shliclel4217.sh.intel.com [10.239.240.127]) by shvmail03.sh.intel.com (Postfix) with ESMTP id 0A9C5100568E; Thu, 30 May 2024 13:52:19 +0800 (CST) From: "Hu, Lin1" To: gcc-patches@gcc.gnu.org Cc: hongtao.liu@intel.com, ubizjak@gmail.com Subject: [PATCH] i386: Handle target of __builtin_ia32_cmp[p|s][s|d] from avx into sse/sse2/avx Date: Thu, 30 May 2024 13:52:19 +0800 Message-Id: <20240530055219.3477674-1-lin1.hu@intel.com> X-Mailer: git-send-email 2.31.1 MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Spam-Status: No, score=-10.6 required=5.0 tests=BAYES_00,DKIMWL_WL_HIGH,DKIM_SIGNED,DKIM_VALID,DKIM_VALID_AU,DKIM_VALID_EF,GIT_PATCH_0,KAM_SHORT,SPF_HELO_NONE,SPF_NONE,TXREP,T_SCC_BODY_TEXT_LINE autolearn=ham autolearn_force=no version=3.4.6 X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on server2.sourceware.org List-Id: Hi, all This patch aims to extend __builtin_ia32_cmp[p|s][s|d] from avx to sse/sse2/avx, where its immediate is in range of [0, 7]. Bootstrapped and regtested on x86_64-pc-linux-gnu, OK for trunk? BRs, Lin gcc/ChangeLog: * config/i386/avxintrin.h: Move cmp[p|s][s|d] to [e|x]mmintrin.h, and move macros to xmmintrin.h * config/i386/emmintrin.h: Add cmp[p|s]s intrins. * config/i386/i386-builtin.def: Modify __builtin_ia32_cmp[p|s][s|d]. * config/i386/i386-expand.cc (ix86_expand_args_builtin): Raise error when imm is in range of [8, 32] without avx. * config/i386/sse.md (avx_cmp3): Modefy define_insn. (avx_vmcmp3): Ditto. * config/i386/xmmintrin.h (_CMP_EQ_OQ): New macro for sse/sse2. (_CMP_LT_OS): Ditto (_CMP_LE_OS): Ditto (_CMP_UNORD_Q): Ditto (_CMP_NEQ_UQ): Ditto (_CMP_NLT_US): Ditto (_CMP_NLE_US): Ditto (_CMP_ORD_Q): Ditto (_mm_cmp_ps): Move intrin from avxintrin.h to xmmintrin.h (_mm_cmp_ss): Ditto. gcc/testsuite/ChangeLog: * gcc.target/i386/sse-cmp-1.c: New test. * gcc.target/i386/sse-cmp-2.c: Ditto. * gcc.target/i386/sse-cmp-error-1.c: Ditto. --- gcc/config/i386/avxintrin.h | 56 ----------- gcc/config/i386/emmintrin.h | 22 +++++ gcc/config/i386/i386-builtin.def | 10 +- gcc/config/i386/i386-expand.cc | 6 ++ gcc/config/i386/predicates.md | 5 + gcc/config/i386/sse.md | 42 ++++---- gcc/config/i386/xmmintrin.h | 41 ++++++++ gcc/testsuite/gcc.target/i386/sse-cmp-1.c | 20 ++++ gcc/testsuite/gcc.target/i386/sse-cmp-2.c | 96 +++++++++++++++++++ gcc/testsuite/gcc.target/i386/sse-cmp-error.c | 16 ++++ 10 files changed, 236 insertions(+), 78 deletions(-) create mode 100644 gcc/testsuite/gcc.target/i386/sse-cmp-1.c create mode 100644 gcc/testsuite/gcc.target/i386/sse-cmp-2.c create mode 100644 gcc/testsuite/gcc.target/i386/sse-cmp-error.c diff --git a/gcc/config/i386/avxintrin.h b/gcc/config/i386/avxintrin.h index 80214540888..ec9b9905b5f 100644 --- a/gcc/config/i386/avxintrin.h +++ b/gcc/config/i386/avxintrin.h @@ -72,22 +72,6 @@ typedef double __m256d_u __attribute__ ((__vector_size__ (32), /* Compare predicates for scalar and packed compare intrinsics. */ -/* Equal (ordered, non-signaling) */ -#define _CMP_EQ_OQ 0x00 -/* Less-than (ordered, signaling) */ -#define _CMP_LT_OS 0x01 -/* Less-than-or-equal (ordered, signaling) */ -#define _CMP_LE_OS 0x02 -/* Unordered (non-signaling) */ -#define _CMP_UNORD_Q 0x03 -/* Not-equal (unordered, non-signaling) */ -#define _CMP_NEQ_UQ 0x04 -/* Not-less-than (unordered, signaling) */ -#define _CMP_NLT_US 0x05 -/* Not-less-than-or-equal (unordered, signaling) */ -#define _CMP_NLE_US 0x06 -/* Ordered (nonsignaling) */ -#define _CMP_ORD_Q 0x07 /* Equal (unordered, non-signaling) */ #define _CMP_EQ_UQ 0x08 /* Not-greater-than-or-equal (unordered, signaling) */ @@ -381,18 +365,6 @@ _mm256_xor_ps (__m256 __A, __m256 __B) } #ifdef __OPTIMIZE__ -extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) -_mm_cmp_pd (__m128d __X, __m128d __Y, const int __P) -{ - return (__m128d) __builtin_ia32_cmppd ((__v2df)__X, (__v2df)__Y, __P); -} - -extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) -_mm_cmp_ps (__m128 __X, __m128 __Y, const int __P) -{ - return (__m128) __builtin_ia32_cmpps ((__v4sf)__X, (__v4sf)__Y, __P); -} - extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) _mm256_cmp_pd (__m256d __X, __m256d __Y, const int __P) { @@ -406,27 +378,7 @@ _mm256_cmp_ps (__m256 __X, __m256 __Y, const int __P) return (__m256) __builtin_ia32_cmpps256 ((__v8sf)__X, (__v8sf)__Y, __P); } - -extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) -_mm_cmp_sd (__m128d __X, __m128d __Y, const int __P) -{ - return (__m128d) __builtin_ia32_cmpsd ((__v2df)__X, (__v2df)__Y, __P); -} - -extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) -_mm_cmp_ss (__m128 __X, __m128 __Y, const int __P) -{ - return (__m128) __builtin_ia32_cmpss ((__v4sf)__X, (__v4sf)__Y, __P); -} #else -#define _mm_cmp_pd(X, Y, P) \ - ((__m128d) __builtin_ia32_cmppd ((__v2df)(__m128d)(X), \ - (__v2df)(__m128d)(Y), (int)(P))) - -#define _mm_cmp_ps(X, Y, P) \ - ((__m128) __builtin_ia32_cmpps ((__v4sf)(__m128)(X), \ - (__v4sf)(__m128)(Y), (int)(P))) - #define _mm256_cmp_pd(X, Y, P) \ ((__m256d) __builtin_ia32_cmppd256 ((__v4df)(__m256d)(X), \ (__v4df)(__m256d)(Y), (int)(P))) @@ -434,14 +386,6 @@ _mm_cmp_ss (__m128 __X, __m128 __Y, const int __P) #define _mm256_cmp_ps(X, Y, P) \ ((__m256) __builtin_ia32_cmpps256 ((__v8sf)(__m256)(X), \ (__v8sf)(__m256)(Y), (int)(P))) - -#define _mm_cmp_sd(X, Y, P) \ - ((__m128d) __builtin_ia32_cmpsd ((__v2df)(__m128d)(X), \ - (__v2df)(__m128d)(Y), (int)(P))) - -#define _mm_cmp_ss(X, Y, P) \ - ((__m128) __builtin_ia32_cmpss ((__v4sf)(__m128)(X), \ - (__v4sf)(__m128)(Y), (int)(P))) #endif extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) diff --git a/gcc/config/i386/emmintrin.h b/gcc/config/i386/emmintrin.h index fa301103daf..c10fc3433af 100644 --- a/gcc/config/i386/emmintrin.h +++ b/gcc/config/i386/emmintrin.h @@ -1390,6 +1390,28 @@ _mm_cmpgt_epi32 (__m128i __A, __m128i __B) return (__m128i) ((__v4si)__A > (__v4si)__B); } +#ifdef __OPTIMIZE__ +extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) +_mm_cmp_pd (__m128d __X, __m128d __Y, const int __P) +{ + return (__m128d) __builtin_ia32_cmppd ((__v2df)__X, (__v2df)__Y, __P); +} + +extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) +_mm_cmp_sd (__m128d __X, __m128d __Y, const int __P) +{ + return (__m128d) __builtin_ia32_cmpsd ((__v2df)__X, (__v2df)__Y, __P); +} +#else +#define _mm_cmp_pd(X, Y, P) \ + ((__m128d) __builtin_ia32_cmppd ((__v2df)(__m128d)(X), \ + (__v2df)(__m128d)(Y), (int)(P))) + +#define _mm_cmp_sd(X, Y, P) \ + ((__m128d) __builtin_ia32_cmpsd ((__v2df)(__m128d)(X), \ + (__v2df)(__m128d)(Y), (int)(P))) +#endif + #ifdef __OPTIMIZE__ extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) _mm_extract_epi16 (__m128i const __A, int const __N) diff --git a/gcc/config/i386/i386-builtin.def b/gcc/config/i386/i386-builtin.def index a28c48c7566..85b01592d7e 100644 --- a/gcc/config/i386/i386-builtin.def +++ b/gcc/config/i386/i386-builtin.def @@ -671,6 +671,9 @@ BDESC (OPTION_MASK_ISA_SSE, 0, CODE_FOR_sse_vmmaskcmpv4sf3, "__builtin_ia32_cmpn BDESC (OPTION_MASK_ISA_SSE, 0, CODE_FOR_sse_vmmaskcmpv4sf3, "__builtin_ia32_cmpnless", IX86_BUILTIN_CMPNLESS, UNGT, (int) V4SF_FTYPE_V4SF_V4SF) BDESC (OPTION_MASK_ISA_SSE, 0, CODE_FOR_sse_vmmaskcmpv4sf3, "__builtin_ia32_cmpordss", IX86_BUILTIN_CMPORDSS, ORDERED, (int) V4SF_FTYPE_V4SF_V4SF) +BDESC (OPTION_MASK_ISA_SSE, 0, CODE_FOR_avx_vmcmpv4sf3, "__builtin_ia32_cmpss", IX86_BUILTIN_CMPSS, UNKNOWN, (int) V4SF_FTYPE_V4SF_V4SF_INT) +BDESC (OPTION_MASK_ISA_SSE, 0, CODE_FOR_avx_cmpv4sf3, "__builtin_ia32_cmpps", IX86_BUILTIN_CMPPS, UNKNOWN, (int) V4SF_FTYPE_V4SF_V4SF_INT) + BDESC (OPTION_MASK_ISA_SSE, 0, CODE_FOR_sminv4sf3, "__builtin_ia32_minps", IX86_BUILTIN_MINPS, UNKNOWN, (int) V4SF_FTYPE_V4SF_V4SF) BDESC (OPTION_MASK_ISA_SSE, 0, CODE_FOR_smaxv4sf3, "__builtin_ia32_maxps", IX86_BUILTIN_MAXPS, UNKNOWN, (int) V4SF_FTYPE_V4SF_V4SF) BDESC (OPTION_MASK_ISA_SSE, 0, CODE_FOR_sse_vmsminv4sf3, "__builtin_ia32_minss", IX86_BUILTIN_MINSS, UNKNOWN, (int) V4SF_FTYPE_V4SF_V4SF) @@ -827,6 +830,9 @@ BDESC (OPTION_MASK_ISA_SSE2, 0, CODE_FOR_nothing, "__builtin_ia32_pcmpgtb128", I BDESC (OPTION_MASK_ISA_SSE2, 0, CODE_FOR_nothing, "__builtin_ia32_pcmpgtw128", IX86_BUILTIN_PCMPGTW128, UNKNOWN, (int) V8HI_FTYPE_V8HI_V8HI) BDESC (OPTION_MASK_ISA_SSE2, 0, CODE_FOR_nothing, "__builtin_ia32_pcmpgtd128", IX86_BUILTIN_PCMPGTD128, UNKNOWN, (int) V4SI_FTYPE_V4SI_V4SI ) +BDESC (OPTION_MASK_ISA_SSE2, 0, CODE_FOR_avx_vmcmpv2df3, "__builtin_ia32_cmpsd", IX86_BUILTIN_CMPSD, UNKNOWN, (int) V2DF_FTYPE_V2DF_V2DF_INT) +BDESC (OPTION_MASK_ISA_SSE2, 0, CODE_FOR_avx_cmpv2df3, "__builtin_ia32_cmppd", IX86_BUILTIN_CMPPD, UNKNOWN, (int) V2DF_FTYPE_V2DF_V2DF_INT) + BDESC (OPTION_MASK_ISA_SSE2, 0, CODE_FOR_umaxv16qi3, "__builtin_ia32_pmaxub128", IX86_BUILTIN_PMAXUB128, UNKNOWN, (int) V16QI_FTYPE_V16QI_V16QI) BDESC (OPTION_MASK_ISA_SSE2, 0, CODE_FOR_smaxv8hi3, "__builtin_ia32_pmaxsw128", IX86_BUILTIN_PMAXSW128, UNKNOWN, (int) V8HI_FTYPE_V8HI_V8HI) BDESC (OPTION_MASK_ISA_SSE2, 0, CODE_FOR_uminv16qi3, "__builtin_ia32_pminub128", IX86_BUILTIN_PMINUB128, UNKNOWN, (int) V16QI_FTYPE_V16QI_V16QI) @@ -1077,10 +1083,6 @@ BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_blendvps256, "__builtin_ia32_blendvp BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_dpps256, "__builtin_ia32_dpps256", IX86_BUILTIN_DPPS256, UNKNOWN, (int) V8SF_FTYPE_V8SF_V8SF_INT) BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_shufpd256, "__builtin_ia32_shufpd256", IX86_BUILTIN_SHUFPD256, UNKNOWN, (int) V4DF_FTYPE_V4DF_V4DF_INT) BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_shufps256, "__builtin_ia32_shufps256", IX86_BUILTIN_SHUFPS256, UNKNOWN, (int) V8SF_FTYPE_V8SF_V8SF_INT) -BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_vmcmpv2df3, "__builtin_ia32_cmpsd", IX86_BUILTIN_CMPSD, UNKNOWN, (int) V2DF_FTYPE_V2DF_V2DF_INT) -BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_vmcmpv4sf3, "__builtin_ia32_cmpss", IX86_BUILTIN_CMPSS, UNKNOWN, (int) V4SF_FTYPE_V4SF_V4SF_INT) -BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_cmpv2df3, "__builtin_ia32_cmppd", IX86_BUILTIN_CMPPD, UNKNOWN, (int) V2DF_FTYPE_V2DF_V2DF_INT) -BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_cmpv4sf3, "__builtin_ia32_cmpps", IX86_BUILTIN_CMPPS, UNKNOWN, (int) V4SF_FTYPE_V4SF_V4SF_INT) BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_cmpv4df3, "__builtin_ia32_cmppd256", IX86_BUILTIN_CMPPD256, UNKNOWN, (int) V4DF_FTYPE_V4DF_V4DF_INT) BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_cmpv8sf3, "__builtin_ia32_cmpps256", IX86_BUILTIN_CMPPS256, UNKNOWN, (int) V8SF_FTYPE_V8SF_V8SF_INT) BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_vextractf128v4df, "__builtin_ia32_vextractf128_pd256", IX86_BUILTIN_EXTRACTF128PD256, UNKNOWN, (int) V2DF_FTYPE_V4DF_INT) diff --git a/gcc/config/i386/i386-expand.cc b/gcc/config/i386/i386-expand.cc index 215a998fc26..537f349ed30 100644 --- a/gcc/config/i386/i386-expand.cc +++ b/gcc/config/i386/i386-expand.cc @@ -11833,6 +11833,12 @@ ix86_expand_args_builtin (const struct builtin_description *d, case CODE_FOR_avx_vmcmpv4sf3: case CODE_FOR_avx_cmpv2df3: case CODE_FOR_avx_cmpv4sf3: + if (CONST_INT_P (op) && IN_RANGE (INTVAL (op), 8, 31)) + { + error ("'%s' needs isa option %s", d->name, "-mavx"); + return const0_rtx; + } + /* FALLTHRU */ case CODE_FOR_avx_cmpv4df3: case CODE_FOR_avx_cmpv8sf3: case CODE_FOR_avx512f_cmpv8df3_mask: diff --git a/gcc/config/i386/predicates.md b/gcc/config/i386/predicates.md index 2a97776fc32..29e26f56dde 100644 --- a/gcc/config/i386/predicates.md +++ b/gcc/config/i386/predicates.md @@ -1056,6 +1056,11 @@ (define_predicate "const_28_to_31_operand" (and (match_code "const_int") (match_test "IN_RANGE (INTVAL (op), 28, 31)"))) +(define_predicate "cmpps_imm_operand" + (ior (match_operand 0 "const_0_to_7_operand") + (and (match_test "TARGET_AVX") + (match_operand 0 "const_0_to_31_operand")))) + ;; True if this is a constant appropriate for an increment or decrement. (define_predicate "incdec_operand" (match_code "const_int") diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md index 0f4fbcb2c5d..542d032ff63 100644 --- a/gcc/config/i386/sse.md +++ b/gcc/config/i386/sse.md @@ -3641,18 +3641,21 @@ (define_expand "reduc__scal_" ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; (define_insn "avx_cmp3" - [(set (match_operand:VF_128_256 0 "register_operand" "=x") + [(set (match_operand:VF_128_256 0 "register_operand" "=x,x") (unspec:VF_128_256 - [(match_operand:VF_128_256 1 "register_operand" "x") - (match_operand:VF_128_256 2 "nonimmediate_operand" "xjm") - (match_operand:SI 3 "const_0_to_31_operand")] + [(match_operand:VF_128_256 1 "register_operand" "0,x") + (match_operand:VF_128_256 2 "nonimmediate_operand" "xm,xjm") + (match_operand:SI 3 "cmpps_imm_operand")] UNSPEC_PCMP))] - "TARGET_AVX" - "vcmp\t{%3, %2, %1, %0|%0, %1, %2, %3}" - [(set_attr "type" "ssecmp") - (set_attr "addr" "gpr16") + "TARGET_SSE" + "@ + cmp\t{%3, %2, %0|%0, %2, %3} + vcmp\t{%3, %2, %1, %0|%0, %1, %2, %3}" + [(set_attr "isa" "noavx,avx") + (set_attr "type" "ssecmp") + (set_attr "addr" "*,gpr16") (set_attr "length_immediate" "1") - (set_attr "prefix" "vex") + (set_attr "prefix" "orig,vex") (set_attr "mode" "")]) (define_insn_and_split "*avx_cmp3_1" @@ -3852,21 +3855,24 @@ (define_insn_and_split "*avx_cmp3_ltint_not" }) (define_insn "avx_vmcmp3" - [(set (match_operand:VF_128 0 "register_operand" "=x") + [(set (match_operand:VF_128 0 "register_operand" "=x,x") (vec_merge:VF_128 (unspec:VF_128 - [(match_operand:VF_128 1 "register_operand" "x") - (match_operand:VF_128 2 "nonimmediate_operand" "xjm") - (match_operand:SI 3 "const_0_to_31_operand")] + [(match_operand:VF_128 1 "register_operand" "0,x") + (match_operand:VF_128 2 "nonimmediate_operand" "xm,xjm") + (match_operand:SI 3 "cmpps_imm_operand")] UNSPEC_PCMP) (match_dup 1) (const_int 1)))] - "TARGET_AVX" - "vcmp\t{%3, %2, %1, %0|%0, %1, %2, %3}" - [(set_attr "type" "ssecmp") - (set_attr "addr" "gpr16") + "TARGET_SSE" + "@ + cmp\t{%3, %2, %0|%0, %2, %3} + vcmp\t{%3, %2, %1, %0|%0, %1, %2, %3}" + [(set_attr "isa" "noavx,avx") + (set_attr "type" "ssecmp") + (set_attr "addr" "*,gpr16") (set_attr "length_immediate" "1") - (set_attr "prefix" "vex") + (set_attr "prefix" "orig,vex") (set_attr "mode" "")]) (define_insn "*_maskcmp3_comm" diff --git a/gcc/config/i386/xmmintrin.h b/gcc/config/i386/xmmintrin.h index 87515ecb218..4b6266c5cde 100644 --- a/gcc/config/i386/xmmintrin.h +++ b/gcc/config/i386/xmmintrin.h @@ -108,6 +108,25 @@ typedef float __v4sf __attribute__ ((__vector_size__ (16))); #define _MM_FLUSH_ZERO_ON 0x8000 #define _MM_FLUSH_ZERO_OFF 0x0000 +/* Compare predicates for scalar and packed compare intrinsics. */ + +/* Equal (ordered, non-signaling) */ +#define _CMP_EQ_OQ 0x00 +/* Less-than (ordered, signaling) */ +#define _CMP_LT_OS 0x01 +/* Less-than-or-equal (ordered, signaling) */ +#define _CMP_LE_OS 0x02 +/* Unordered (non-signaling) */ +#define _CMP_UNORD_Q 0x03 +/* Not-equal (unordered, non-signaling) */ +#define _CMP_NEQ_UQ 0x04 +/* Not-less-than (unordered, signaling) */ +#define _CMP_NLT_US 0x05 +/* Not-less-than-or-equal (unordered, signaling) */ +#define _CMP_NLE_US 0x06 +/* Ordered (nonsignaling) */ +#define _CMP_ORD_Q 0x07 + /* Create an undefined vector. */ extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) _mm_undefined_ps (void) @@ -434,6 +453,28 @@ _mm_cmpunord_ps (__m128 __A, __m128 __B) return (__m128) __builtin_ia32_cmpunordps ((__v4sf)__A, (__v4sf)__B); } +#ifdef __OPTIMIZE__ +extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) +_mm_cmp_ps (__m128 __X, __m128 __Y, const int __P) +{ + return (__m128) __builtin_ia32_cmpps ((__v4sf)__X, (__v4sf)__Y, __P); +} + +extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) +_mm_cmp_ss (__m128 __X, __m128 __Y, const int __P) +{ + return (__m128) __builtin_ia32_cmpss ((__v4sf)__X, (__v4sf)__Y, __P); +} +#else +#define _mm_cmp_ps(X, Y, P) \ + ((__m128) __builtin_ia32_cmpps ((__v4sf)(__m128)(X), \ + (__v4sf)(__m128)(Y), (int)(P))) + +#define _mm_cmp_ss(X, Y, P) \ + ((__m128) __builtin_ia32_cmpss ((__v4sf)(__m128)(X), \ + (__v4sf)(__m128)(Y), (int)(P))) +#endif + /* Compare the lower SPFP values of A and B and return 1 if true and 0 if false. */ diff --git a/gcc/testsuite/gcc.target/i386/sse-cmp-1.c b/gcc/testsuite/gcc.target/i386/sse-cmp-1.c new file mode 100644 index 00000000000..eff90d4790e --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/sse-cmp-1.c @@ -0,0 +1,20 @@ +/* { dg-do compile } */ +/* { dg-options "-O0 -msse2" } */ +/* { dg-final { scan-assembler-times "cmpsd" 1 } } */ +/* { dg-final { scan-assembler-times "cmpss" 1 } } */ +/* { dg-final { scan-assembler-times "cmppd" 1 } } */ +/* { dg-final { scan-assembler-times "cmpps" 1 } } */ + +#include + +__m128 a1, a2, a3, a4, a5, a6; +__m128d d1, d2, d3, d4, d5, d6; + +void +test (void) +{ + d1 = _mm_cmp_sd (d2, d3, 1); + a1 = _mm_cmp_ss (a2, a3, 2); + d1 = _mm_cmp_pd (d2, d3, 3); + a1 = _mm_cmp_ps (a2, a3, 4); +} diff --git a/gcc/testsuite/gcc.target/i386/sse-cmp-2.c b/gcc/testsuite/gcc.target/i386/sse-cmp-2.c new file mode 100644 index 00000000000..77c05c484b6 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/sse-cmp-2.c @@ -0,0 +1,96 @@ +/* { dg-do run } */ +/* { dg-require-effective-target sse2 } */ +/* { dg-require-effective-target c99_runtime } */ +/* { dg-options "-O2 -msse2 -std=c99" } */ + +#include "sse2-check.h" +#include "emmintrin.h" +#include + +double sd1[2]={2134.3343,6678.346}; +double sd2[2]={41124.234,6678.346}; + +float ss1[4]={2134.3343,6678.346,453.345635,54646.464356}; +float ss2[4]={41124.234,6678.346,8653.65635,856.43576}; + +union +{ + double x[2]; + long long a[2]; +}dd, ed; + +union +{ + float x[4]; + int a[4]; +}ds, es; + +#define CHECK(INTSIZE, SIZE, NUNITS, SUFFIX) \ +void check##SUFFIX(unsigned imm, char *id) \ +{ \ + if(checkV##INTSIZE(d##SIZE.a, e##SIZE.a, NUNITS)){ \ + printf("mm_cmp_" #SUFFIX "(%s: 0x%x) FAILED\n", id, imm);\ + abort(); \ + } \ +} + +CHECK (l, d, 2, pd) +CHECK (i, s, 4, ps) +CHECK (l, d, 2, sd) +CHECK (i, s, 4, ss) + +#define CMP(imm, rel, SIZE, NUNITS, SUFFIX) \ + for (i = 0; i < NUNITS; i++) e##SIZE.a[i] = rel ? -1 : 0; \ + source##SIZE##1 = _mm_loadu_p##SIZE(s##SIZE##1); \ + source##SIZE##2 = _mm_loadu_p##SIZE(s##SIZE##2); \ + dest##SIZE = _mm_cmp_##SUFFIX(source##SIZE##1, source##SIZE##2, imm); \ + _mm_storeu_p##SIZE(d##SIZE.x, dest##SIZE); \ + check##SUFFIX(imm, "" #imm ""); + +static void +sse2_test () +{ + __m128d sourced1, sourced2, destd; + __m128 sources1, sources2, dests; + int i; + + CMP(_CMP_EQ_OQ, !isunordered(sd1[i], sd2[i]) && sd1[i] == sd2[i], d, 2, pd); + CMP(_CMP_LT_OS, !isunordered(sd1[i], sd2[i]) && sd1[i] < sd2[i], d, 2, pd); + CMP(_CMP_LE_OS, !isunordered(sd1[i], sd2[i]) && sd1[i] <= sd2[i], d, 2, pd); + CMP(_CMP_UNORD_Q, isunordered(sd1[i], sd2[i]), d, 2, pd); + CMP(_CMP_NEQ_UQ, isunordered(sd1[i], sd2[i]) || sd1[i] != sd2[i], d, 2, pd); + CMP(_CMP_NLT_US, isunordered(sd1[i], sd2[i]) || sd1[i] >= sd2[i], d, 2, pd); + CMP(_CMP_NLE_US, isunordered(sd1[i], sd2[i]) || sd1[i] > sd2[i], d, 2, pd); + CMP(_CMP_ORD_Q, !isunordered(sd1[i], sd2[i]), d, 2, pd); + + CMP(_CMP_EQ_OQ, !isunordered(ss1[i], ss2[i]) && ss1[i] == ss2[i], s, 4, ps); + CMP(_CMP_LT_OS, !isunordered(ss1[i], ss2[i]) && ss1[i] < ss2[i], s, 4, ps); + CMP(_CMP_LE_OS, !isunordered(ss1[i], ss2[i]) && ss1[i] <= ss2[i], s, 4, ps); + CMP(_CMP_UNORD_Q, isunordered(ss1[i], ss2[i]), s, 4, ps); + CMP(_CMP_NEQ_UQ, isunordered(ss1[i], ss2[i]) || ss1[i] != ss2[i], s, 4, ps); + CMP(_CMP_NLT_US, isunordered(ss1[i], ss2[i]) || ss1[i] >= ss2[i], s, 4, ps); + CMP(_CMP_NLE_US, isunordered(ss1[i], ss2[i]) || ss1[i] > ss2[i], s, 4, ps); + CMP(_CMP_ORD_Q, !isunordered(ss1[i], ss2[i]), s, 4, ps); + + dd.x[1] = ed.x[1] = sd1[1]; + + CMP(_CMP_EQ_OQ, !isunordered(sd1[0], sd2[0]) && sd1[0] == sd2[0], d, 1, sd); + CMP(_CMP_LT_OS, !isunordered(sd1[0], sd2[0]) && sd1[0] < sd2[0], d, 1, sd); + CMP(_CMP_LE_OS, !isunordered(sd1[0], sd2[0]) && sd1[0] <= sd2[0], d, 1, sd); + CMP(_CMP_UNORD_Q, isunordered(sd1[0], sd2[0]), d, 1, sd); + CMP(_CMP_NEQ_UQ, isunordered(sd1[0], sd2[0]) || sd1[0] != sd2[0], d, 1, sd); + CMP(_CMP_NLT_US, isunordered(sd1[0], sd2[0]) || sd1[0] >= sd2[0], d, 1, sd); + CMP(_CMP_NLE_US, isunordered(sd1[0], sd2[0]) || sd1[0] > sd2[0], d, 1, sd); + CMP(_CMP_ORD_Q, !isunordered(sd1[0], sd2[0]), d, 1, sd); + + for(i = 1; i < 4; i++) ds.x[i] = es.x[i] = ss1[i]; + + CMP(_CMP_EQ_OQ, !isunordered(ss1[0], ss2[0]) && ss1[0] == ss2[0], s, 1, ss); + CMP(_CMP_LT_OS, !isunordered(ss1[0], ss2[0]) && ss1[0] < ss2[0], s, 1, ss); + CMP(_CMP_LE_OS, !isunordered(ss1[0], ss2[0]) && ss1[0] <= ss2[0], s, 1, ss); + CMP(_CMP_UNORD_Q, isunordered(ss1[0], ss2[0]), s, 1, ss); + CMP(_CMP_NEQ_UQ, isunordered(ss1[0], ss2[0]) || ss1[0] != ss2[0], s, 1, ss); + CMP(_CMP_NLT_US, isunordered(ss1[0], ss2[0]) || ss1[0] >= ss2[0], s, 1, ss); + CMP(_CMP_NLE_US, isunordered(ss1[0], ss2[0]) || ss1[0] > ss2[0], s, 1, ss); + CMP(_CMP_ORD_Q, !isunordered(ss1[0], ss2[0]), s, 1, ss); +} diff --git a/gcc/testsuite/gcc.target/i386/sse-cmp-error.c b/gcc/testsuite/gcc.target/i386/sse-cmp-error.c new file mode 100644 index 00000000000..79633b94e2f --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/sse-cmp-error.c @@ -0,0 +1,16 @@ +/* { dg-do compile } */ +/* { dg-options "-O0 -msse2" } */ + +#include + +volatile __m128 a1, a2, a3; +volatile __m128d d1, d2, d3; + +void +test (void) +{ + d1 = _mm_cmp_sd (d2, d3, 8); /* { dg-error "'__builtin_ia32_cmpsd' needs isa option -mavx" } */ + a1 = _mm_cmp_ss (a2, a3, 8); /* { dg-error "'__builtin_ia32_cmpss' needs isa option -mavx" } */ + d1 = _mm_cmp_pd (d2, d3, 8); /* { dg-error "'__builtin_ia32_cmppd' needs isa option -mavx" } */ + a1 = _mm_cmp_ps (a2, a3, 8); /* { dg-error "'__builtin_ia32_cmpps' needs isa option -mavx" } */ +} -- 2.31.1