From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <hjl.tools@gmail.com>
Received: from mail-qk1-x72d.google.com (mail-qk1-x72d.google.com [IPv6:2607:f8b0:4864:20::72d])
	by sourceware.org (Postfix) with ESMTPS id 0018B3858C52
	for <libc-alpha@sourceware.org>; Fri, 14 Oct 2022 18:02:51 +0000 (GMT)
DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 0018B3858C52
Authentication-Results: sourceware.org; dmarc=pass (p=none dis=none) header.from=gmail.com
Authentication-Results: sourceware.org; spf=pass smtp.mailfrom=gmail.com
Received: by mail-qk1-x72d.google.com with SMTP id o22so2966439qkl.8
        for <libc-alpha@sourceware.org>; Fri, 14 Oct 2022 11:02:51 -0700 (PDT)
DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed;
        d=gmail.com; s=20210112;
        h=cc:to:subject:message-id:date:from:in-reply-to:references
         :mime-version:from:to:cc:subject:date:message-id:reply-to;
        bh=mil1s+kbhKnqDOMez7HNonfUdIeKCQc2suOGcuW6uZI=;
        b=U6Hk45q2AnkwFDRHbP/2mFbBW+JVekbbVEZpPL7vGKFX7KIFpGuWJ2v05Na2RMscNq
         Fx8sQtyqg7+H/9VG5eJGoQxyumc0P1Xe/y02C256iKbCnjV2BZdxUGqGiPRb+hIeXLkf
         tOxPi4lxNmA8pqQVD0Tykr+54jKFs6VJQBoBNXDLtleYtcjnG5yAeTZd2t0wlKNd0Q3N
         9EhqyJIak+X1D7Oj9S96HeVwTR7AlSBY0UOWWIQYHERz1Gq6aWRLSyJIRclEmACm2nAK
         kZbYf88dFWLgG3/noubTG5yOIgkCZcGUmSPwRvXYZdqBAXnjytroAki/aRUFGo0NxwRd
         gNMQ==
X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed;
        d=1e100.net; s=20210112;
        h=cc:to:subject:message-id:date:from:in-reply-to:references
         :mime-version:x-gm-message-state:from:to:cc:subject:date:message-id
         :reply-to;
        bh=mil1s+kbhKnqDOMez7HNonfUdIeKCQc2suOGcuW6uZI=;
        b=CnEE7/dcLdnU+yc0G/m5ZxKMHl4m7UK/XkU9SY/1nhApW+NRRQ7BiFrEoFoeE4EDWx
         3Bpnhhe9COrvCLL6DkJEOI3NQG+ZkctJPfxWLPR5ligxAWDg6ZMjwN6NFd3qcL/kCQSS
         gHQdmVq4tiGacMzXAnfmXjFpwWjmGzvcUJJdBAEq+B7ogqKNgyh6FRLpdqaFkdmBjOt8
         8Y1Hp3FNSqFC7uUGWXByjafrYLr8q6uBvbYIDA+fM+MOPbmEjkyQ7u+BSm6nZ5B2+73G
         tWhs8Mwm2GD1Iw5LD0pUERpT/PMvf6+wRyEVX04JYPe93knhtzhwzv3Z92OhBf1JBDaE
         klJw==
X-Gm-Message-State: ACrzQf0ZAhOSn1gdW02HFgtT4eDGwjUr/SG0Aj/uszwS4MkjBdbOUNjF
	H58CBmgpIexhALYNOmgOgGLYceDu+DH9aEQyBdw=
X-Google-Smtp-Source: AMsMyM73l3eVVOY6ZmmZYm68WDCKzLJniTQGADWM5cEACPs3fLnC0W0eAUrllsf4GlqYkiOPNTA+f/z08hjGR3WGJWo=
X-Received: by 2002:a05:620a:13f3:b0:6ee:92b9:fe2f with SMTP id
 h19-20020a05620a13f300b006ee92b9fe2fmr4570427qkl.477.1665770571087; Fri, 14
 Oct 2022 11:02:51 -0700 (PDT)
MIME-Version: 1.0
References: <20221014164008.1325863-1-goldstein.w.n@gmail.com> <20221014164008.1325863-2-goldstein.w.n@gmail.com>
In-Reply-To: <20221014164008.1325863-2-goldstein.w.n@gmail.com>
From: "H.J. Lu" <hjl.tools@gmail.com>
Date: Fri, 14 Oct 2022 11:02:14 -0700
Message-ID: <CAMe9rOpwo+wAOSDQKFZTa40OYKOc+NBQuGF9voSpjaoXdfkXKQ@mail.gmail.com>
Subject: Re: [PATCH v1 2/3] x86: Add macros for GPRs / mask insn based on VEC_SIZE
To: Noah Goldstein <goldstein.w.n@gmail.com>
Cc: libc-alpha@sourceware.org, carlos@systemhalted.org
Content-Type: text/plain; charset="UTF-8"
X-Spam-Status: No, score=-3023.1 required=5.0 tests=BAYES_00,DKIM_SIGNED,DKIM_VALID,DKIM_VALID_AU,DKIM_VALID_EF,FREEMAIL_FROM,GIT_PATCH_0,KAM_SHORT,RCVD_IN_DNSWL_NONE,SPF_HELO_NONE,SPF_PASS,TXREP,URIBL_BLACK 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: <libc-alpha.sourceware.org>

 On Fri, Oct 14, 2022 at 9:40 AM Noah Goldstein <goldstein.w.n@gmail.com> wrote:
>
> This is to make it easier to do think like:
> ```
> vpcmpb %VEC(0), %VEC(1), %k0
> kmov{d|q} %k0, %{eax|rax}
> test %{eax|rax}
> ```

Since all these register macros are based on VEC_SIZE which is either 32
bytes or 64 bytes, only 32-bit or 64-bit integer and mask register macros are
needed.  8-bit and 16-bit macros aren't needed.

> It adds macro s.t any GPR can get the proper width with:
>     `V{upper_case_GPR_name}`
>
> and any mask insn can get the proper width with:
>     `{mask_insn_without_postfix}V`

All macros should be in upper cases.

> This commit does not change libc.so
>
> Tested build on x86-64
> ---
>  sysdeps/x86_64/multiarch/reg-macros.h         | 337 ++++++++++++++++++
>  .../multiarch/scripts/gen-reg-map-macros.py   |  97 +++++
>  2 files changed, 434 insertions(+)
>  create mode 100644 sysdeps/x86_64/multiarch/reg-macros.h
>  create mode 100644 sysdeps/x86_64/multiarch/scripts/gen-reg-map-macros.py
>
> diff --git a/sysdeps/x86_64/multiarch/reg-macros.h b/sysdeps/x86_64/multiarch/reg-macros.h
> new file mode 100644
> index 0000000000..c4d7f57b66
> --- /dev/null
> +++ b/sysdeps/x86_64/multiarch/reg-macros.h

vreg-macros.h to indicate macros based on vector size.   Please
add comments to indicate that register macros are expanded based
on vector size.

> @@ -0,0 +1,337 @@
> +/* This file was generated by: sysdeps/x86_64/multiarch/scripts/gen-reg-map-macros.py.
> +
> +   Copyright (C) 2022 Free Software Foundation, Inc.
> +   This file is part of the GNU C Library.
> +
> +   The GNU C Library is free software; you can redistribute it and/or
> +   modify it under the terms of the GNU Lesser General Public
> +   License as published by the Free Software Foundation; either
> +   version 2.1 of the License, or (at your option) any later version.
> +
> +   The GNU C Library 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
> +   Lesser General Public License for more details.
> +
> +   You should have received a copy of the GNU Lesser General Public
> +   License along with the GNU C Library; if not, see
> +   <https://www.gnu.org/licenses/>.  */
> +
> +#ifndef _REG_MAP_MACROS_H
> +#define _REG_MAP_MACROS_H      1
> +
> +#define rax_8  al
> +#define eax_8  al
> +#define ax_8   al
> +#define al_8   al
> +#define rax_16 ax
> +#define eax_16 ax
> +#define ax_16  ax
> +#define al_16  ax
> +#define rax_32 eax
> +#define eax_32 eax
> +#define ax_32  eax
> +#define al_32  eax
> +#define rax_64 rax
> +#define eax_64 rax
> +#define ax_64  rax
> +#define al_64  rax

Only rax_32 and rax_64 are needed.

> +#define rbx_8  bl
> +#define ebx_8  bl
> +#define bx_8   bl
> +#define bl_8   bl
> +#define rbx_16 bx
> +#define ebx_16 bx
> +#define bx_16  bx
> +#define bl_16  bx
> +#define rbx_32 ebx
> +#define ebx_32 ebx
> +#define bx_32  ebx
> +#define bl_32  ebx
> +#define rbx_64 rbx
> +#define ebx_64 rbx
> +#define bx_64  rbx
> +#define bl_64  rbx
> +#define rcx_8  cl
> +#define ecx_8  cl
> +#define cx_8   cl
> +#define cl_8   cl
> +#define rcx_16 cx
> +#define ecx_16 cx
> +#define cx_16  cx
> +#define cl_16  cx
> +#define rcx_32 ecx
> +#define ecx_32 ecx
> +#define cx_32  ecx
> +#define cl_32  ecx
> +#define rcx_64 rcx
> +#define ecx_64 rcx
> +#define cx_64  rcx
> +#define cl_64  rcx
> +#define rdx_8  dl
> +#define edx_8  dl
> +#define dx_8   dl
> +#define dl_8   dl
> +#define rdx_16 dx
> +#define edx_16 dx
> +#define dx_16  dx
> +#define dl_16  dx
> +#define rdx_32 edx
> +#define edx_32 edx
> +#define dx_32  edx
> +#define dl_32  edx
> +#define rdx_64 rdx
> +#define edx_64 rdx
> +#define dx_64  rdx
> +#define dl_64  rdx
> +#define rbp_8  bpl
> +#define ebp_8  bpl
> +#define bp_8   bpl
> +#define bpl_8  bpl
> +#define rbp_16 bp
> +#define ebp_16 bp
> +#define bp_16  bp
> +#define bpl_16 bp
> +#define rbp_32 ebp
> +#define ebp_32 ebp
> +#define bp_32  ebp
> +#define bpl_32 ebp
> +#define rbp_64 rbp
> +#define ebp_64 rbp
> +#define bp_64  rbp
> +#define bpl_64 rbp
> +#define rsp_8  spl
> +#define esp_8  spl
> +#define sp_8   spl
> +#define spl_8  spl
> +#define rsp_16 sp
> +#define esp_16 sp
> +#define sp_16  sp
> +#define spl_16 sp
> +#define rsp_32 esp
> +#define esp_32 esp
> +#define sp_32  esp
> +#define spl_32 esp
> +#define rsp_64 rsp
> +#define esp_64 rsp
> +#define sp_64  rsp
> +#define spl_64 rsp
> +#define rsi_8  sil
> +#define esi_8  sil
> +#define si_8   sil
> +#define sil_8  sil
> +#define rsi_16 si
> +#define esi_16 si
> +#define si_16  si
> +#define sil_16 si
> +#define rsi_32 esi
> +#define esi_32 esi
> +#define si_32  esi
> +#define sil_32 esi
> +#define rsi_64 rsi
> +#define esi_64 rsi
> +#define si_64  rsi
> +#define sil_64 rsi
> +#define rdi_8  dil
> +#define edi_8  dil
> +#define di_8   dil
> +#define dil_8  dil
> +#define rdi_16 di
> +#define edi_16 di
> +#define di_16  di
> +#define dil_16 di
> +#define rdi_32 edi
> +#define edi_32 edi
> +#define di_32  edi
> +#define dil_32 edi
> +#define rdi_64 rdi
> +#define edi_64 rdi
> +#define di_64  rdi
> +#define dil_64 rdi
> +#define r8_8   r8b
> +#define r8d_8  r8b
> +#define r8w_8  r8b
> +#define r8b_8  r8b
> +#define r8_16  r8w
> +#define r8d_16 r8w
> +#define r8w_16 r8w
> +#define r8b_16 r8w
> +#define r8_32  r8d
> +#define r8d_32 r8d
> +#define r8w_32 r8d
> +#define r8b_32 r8d
> +#define r8_64  r8
> +#define r8d_64 r8
> +#define r8w_64 r8
> +#define r8b_64 r8
> +#define r9_8   r9b
> +#define r9d_8  r9b
> +#define r9w_8  r9b
> +#define r9b_8  r9b
> +#define r9_16  r9w
> +#define r9d_16 r9w
> +#define r9w_16 r9w
> +#define r9b_16 r9w
> +#define r9_32  r9d
> +#define r9d_32 r9d
> +#define r9w_32 r9d
> +#define r9b_32 r9d
> +#define r9_64  r9
> +#define r9d_64 r9
> +#define r9w_64 r9
> +#define r9b_64 r9
> +#define r10_8  r10b
> +#define r10d_8 r10b
> +#define r10w_8 r10b
> +#define r10b_8 r10b
> +#define r10_16 r10w
> +#define r10d_16        r10w
> +#define r10w_16        r10w
> +#define r10b_16        r10w
> +#define r10_32 r10d
> +#define r10d_32        r10d
> +#define r10w_32        r10d
> +#define r10b_32        r10d
> +#define r10_64 r10
> +#define r10d_64        r10
> +#define r10w_64        r10
> +#define r10b_64        r10
> +#define r11_8  r11b
> +#define r11d_8 r11b
> +#define r11w_8 r11b
> +#define r11b_8 r11b
> +#define r11_16 r11w
> +#define r11d_16        r11w
> +#define r11w_16        r11w
> +#define r11b_16        r11w
> +#define r11_32 r11d
> +#define r11d_32        r11d
> +#define r11w_32        r11d
> +#define r11b_32        r11d
> +#define r11_64 r11
> +#define r11d_64        r11
> +#define r11w_64        r11
> +#define r11b_64        r11
> +#define r12_8  r12b
> +#define r12d_8 r12b
> +#define r12w_8 r12b
> +#define r12b_8 r12b
> +#define r12_16 r12w
> +#define r12d_16        r12w
> +#define r12w_16        r12w
> +#define r12b_16        r12w
> +#define r12_32 r12d
> +#define r12d_32        r12d
> +#define r12w_32        r12d
> +#define r12b_32        r12d
> +#define r12_64 r12
> +#define r12d_64        r12
> +#define r12w_64        r12
> +#define r12b_64        r12
> +#define r13_8  r13b
> +#define r13d_8 r13b
> +#define r13w_8 r13b
> +#define r13b_8 r13b
> +#define r13_16 r13w
> +#define r13d_16        r13w
> +#define r13w_16        r13w
> +#define r13b_16        r13w
> +#define r13_32 r13d
> +#define r13d_32        r13d
> +#define r13w_32        r13d
> +#define r13b_32        r13d
> +#define r13_64 r13
> +#define r13d_64        r13
> +#define r13w_64        r13
> +#define r13b_64        r13
> +#define r14_8  r14b
> +#define r14d_8 r14b
> +#define r14w_8 r14b
> +#define r14b_8 r14b
> +#define r14_16 r14w
> +#define r14d_16        r14w
> +#define r14w_16        r14w
> +#define r14b_16        r14w
> +#define r14_32 r14d
> +#define r14d_32        r14d
> +#define r14w_32        r14d
> +#define r14b_32        r14d
> +#define r14_64 r14
> +#define r14d_64        r14
> +#define r14w_64        r14
> +#define r14b_64        r14
> +#define r15_8  r15b
> +#define r15d_8 r15b
> +#define r15w_8 r15b
> +#define r15b_8 r15b
> +#define r15_16 r15w
> +#define r15d_16        r15w
> +#define r15w_16        r15w
> +#define r15b_16        r15w
> +#define r15_32 r15d
> +#define r15d_32        r15d
> +#define r15w_32        r15d
> +#define r15b_32        r15d
> +#define r15_64 r15
> +#define r15d_64        r15
> +#define r15w_64        r15
> +#define r15b_64        r15
> +
> +#define VRAX   VGPR(rax)
> +#define VRBX   VGPR(rbx)
> +#define VRCX   VGPR(rcx)
> +#define VRDX   VGPR(rdx)
> +#define VRBP   VGPR(rbp)
> +#define VRSP   VGPR(rsp)
> +#define VRSI   VGPR(rsi)
> +#define VRDI   VGPR(rdi)
> +#define VR8    VGPR(r8)
> +#define VR9    VGPR(r9)
> +#define VR10   VGPR(r10)
> +#define VR11   VGPR(r11)
> +#define VR12   VGPR(r12)
> +#define VR13   VGPR(r13)
> +#define VR14   VGPR(r14)
> +#define VR15   VGPR(r15)
> +
> +#define kmov_8 kmovb
> +#define kmov_16        kmovw
> +#define kmov_32        kmovd
> +#define kmov_64        kmovq

Only 32 and 64 are needed.

> +#define kortest_8      kortestb
> +#define kortest_16     kortestw
> +#define kortest_32     kortestd
> +#define kortest_64     kortestq
> +#define kor_8  korb
> +#define kor_16 korw
> +#define kor_32 kord
> +#define kor_64 korq
> +#define ktest_8        ktestb
> +#define ktest_16       ktestw
> +#define ktest_32       ktestd
> +#define ktest_64       ktestq
> +#define kand_8 kandb
> +#define kand_16        kandw
> +#define kand_32        kandd
> +#define kand_64        kandq
> +#define kxor_8 kxorb
> +#define kxor_16        kxorw
> +#define kxor_32        kxord
> +#define kxor_64        kxorq
> +
> +#define kmovV  VKINSN_SZ(kmov, REG_WIDTH)
> +#define kortestV       VKINSN_SZ(kortest, REG_WIDTH)
> +#define korV   VKINSN_SZ(kor, REG_WIDTH)
> +#define ktestV         VKINSN_SZ(ktest, REG_WIDTH)
> +#define kandV  VKINSN_SZ(kand, REG_WIDTH)
> +#define kxorV  VKINSN_SZ(kxor, REG_WIDTH)

#define VKINSN(op) VKINSN_SZ(op, REG_WIDTH)

> +
> +#ifndef REG_WIDTH
> +#define REG_WIDTH VEC_SIZE

Since REG_WIDTH must be the same as VEC_SIZE, REG_WIDTH
can be dropped.

> +#endif
> +#define PRIM_VGPR_SZ(reg_name, reg_size)       reg_name##_##reg_size

This is used for both register and instruction.  How about

#define VPASTER(x,y) x##_##y


> +#define VGPR_SZ(reg_name, reg_size)    PRIM_VGPR_SZ(reg_name, reg_size)
> +#define VGPR(reg_name) VGPR_SZ(reg_name, REG_WIDTH)
> +#define VKINSN_SZ(insn, reg_size)      PRIM_VGPR_SZ(insn, reg_size)

No need for both VGPR_SZ and VKINSN_SZ.  How about

#define VEVALUATOR(x,y) VPASTER(x,y)

> +
> +#endif
> diff --git a/sysdeps/x86_64/multiarch/scripts/gen-reg-map-macros.py b/sysdeps/x86_64/multiarch/scripts/gen-reg-map-macros.py
> new file mode 100644
> index 0000000000..5b04e89ecb
> --- /dev/null
> +++ b/sysdeps/x86_64/multiarch/scripts/gen-reg-map-macros.py
> @@ -0,0 +1,97 @@
> +#!/usr/bin/python3
> +# Copyright (C) 2022 Free Software Foundation, Inc.
> +# This file is part of the GNU C Library.
> +#
> +# The GNU C Library is free software; you can redistribute it and/or
> +# modify it under the terms of the GNU Lesser General Public
> +# License as published by the Free Software Foundation; either
> +# version 2.1 of the License, or (at your option) any later version.
> +#
> +# The GNU C Library 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
> +# Lesser General Public License for more details.
> +#
> +# You should have received a copy of the GNU Lesser General Public
> +# License along with the GNU C Library; if not, see
> +# <https://www.gnu.org/licenses/>.
> +"""Generate macros for getting GPR name of a certain size
> +
> +Inputs: None
> +Output: Prints header fill to stdout
> +
> +API:
> +    VGPR(reg_name)
> +        - Get register name VEC_SIZE component of `reg_name`
> +    VGPR_SZ(reg_name, reg_size)
> +        - Get register name `reg_size` component of `reg_name`
> +"""
> +
> +import sys
> +from datetime import datetime
> +
> +registers = [["rax", "eax", "ax", "al"], ["rbx", "ebx", "bx", "bl"],
> +             ["rcx", "ecx", "cx", "cl"], ["rdx", "edx", "dx", "dl"],
> +             ["rbp", "ebp", "bp", "bpl"], ["rsp", "esp", "sp", "spl"],
> +             ["rsi", "esi", "si", "sil"], ["rdi", "edi", "di", "dil"],
> +             ["r8", "r8d", "r8w", "r8b"], ["r9", "r9d", "r9w", "r9b"],
> +             ["r10", "r10d", "r10w", "r10b"], ["r11", "r11d", "r11w", "r11b"],
> +             ["r12", "r12d", "r12w", "r12b"], ["r13", "r13d", "r13w", "r13b"],
> +             ["r14", "r14d", "r14w", "r14b"], ["r15", "r15d", "r15w", "r15b"]]
> +
> +mask_insns = ["kmov", "kortest", "kor", "ktest", "kand", "kxor"]
> +mask_insns_ext = ["b", "w", "d", "q"]
> +
> +cr = """
> +   Copyright (C) {} Free Software Foundation, Inc.
> +   This file is part of the GNU C Library.
> +
> +   The GNU C Library is free software; you can redistribute it and/or
> +   modify it under the terms of the GNU Lesser General Public
> +   License as published by the Free Software Foundation; either
> +   version 2.1 of the License, or (at your option) any later version.
> +
> +   The GNU C Library 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
> +   Lesser General Public License for more details.
> +
> +   You should have received a copy of the GNU Lesser General Public
> +   License along with the GNU C Library; if not, see
> +   <https://www.gnu.org/licenses/>.  */
> +"""
> +
> +print("/* This file was generated by: {}.".format(sys.argv[0]))
> +print(cr.format(datetime.today().year))
> +
> +print("#ifndef _REG_MAP_MACROS_H")
> +print("#define _REG_MAP_MACROS_H\t1\n")
> +for reg in registers:
> +    for i in range(0, 4):
> +        for j in range(0, 4):
> +            print("#define {}_{}\t{}".format(reg[j], 8 << i, reg[3 - i]))
> +
> +print("")
> +for reg in registers:
> +    print("#define V{}\tVGPR({})".format(reg[0].upper(), reg[0]))
> +
> +print("")
> +for mask_insn in mask_insns:
> +    for i in range(0, 4):
> +        print("#define {}_{}\t{}{}".format(mask_insn, 8 << i, mask_insn,
> +                                           mask_insns_ext[i]))
> +
> +print("")
> +for mask_insn in mask_insns:
> +    print("#define {}V \tVKINSN_SZ({}, REG_WIDTH)".format(mask_insn, mask_insn))
> +print("")
> +
> +print("#ifndef REG_WIDTH")
> +print("#define REG_WIDTH VEC_SIZE")
> +print("#endif")
> +print("#define PRIM_VGPR_SZ(reg_name, reg_size)\treg_name##_##reg_size")
> +print("#define VGPR_SZ(reg_name, reg_size)\tPRIM_VGPR_SZ(reg_name, reg_size)")
> +print("#define VGPR(reg_name)\tVGPR_SZ(reg_name, REG_WIDTH)")
> +print("#define VKINSN_SZ(insn, reg_size)\tPRIM_VGPR_SZ(insn, reg_size)")
> +
> +print("\n#endif")
> --
> 2.34.1
>


-- 
H.J.