From a89a14ef5da44684a16b204e7a70460cc8c4922a Mon Sep 17 00:00:00 2001 From: Thomas Voss Date: Fri, 21 Jun 2024 23:36:36 +0200 Subject: Basic constant folding implementation --- vendor/gmp-6.3.0/doc/isa_abi_headache | 128 ++++++++++++++++++++++++++++++++++ 1 file changed, 128 insertions(+) create mode 100644 vendor/gmp-6.3.0/doc/isa_abi_headache (limited to 'vendor/gmp-6.3.0/doc/isa_abi_headache') diff --git a/vendor/gmp-6.3.0/doc/isa_abi_headache b/vendor/gmp-6.3.0/doc/isa_abi_headache new file mode 100644 index 0000000..7e1430d --- /dev/null +++ b/vendor/gmp-6.3.0/doc/isa_abi_headache @@ -0,0 +1,128 @@ +Copyright 2000 Free Software Foundation, Inc. + +This file is part of the GNU MP Library. + +The GNU MP Library is free software; you can redistribute it and/or modify +it under the terms of either: + + * the GNU Lesser General Public License as published by the Free + Software Foundation; either version 3 of the License, or (at your + option) any later version. + +or + + * the GNU General Public License as published by the Free Software + Foundation; either version 2 of the License, or (at your option) any + later version. + +or both in parallel, as here. + +The GNU MP 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 General Public License +for more details. + +You should have received copies of the GNU General Public License and the +GNU Lesser General Public License along with the GNU MP Library. If not, +see https://www.gnu.org/licenses/. + + + + +Terms Used In This Document: + ISA = Instruction Set Architecture. The instructions the current + processor provides. + ABI = Application Binary Interface. Specifies calling convention, + type sizes, etc. + AR64 = Arithmetic operations are 64-bit using 64-bit instructions + (E.g., addition, subtraction, load, store, of 64-bit integer types + are done with single instructions, not 32 bits at a time.) + Environment = The operating system and compiler. + +GMP is a very complex package to build since its speed is very +sensitive to the ISA and ABI. For example, if the ISA provides 64-bit +instructions, it is crucial that GMP is configured to use them. + +Most environments that run on a 64-bit ISA provide more than one ABI. +Typically one of the supported ABI's is a backward compatible 32-bit +ABI, and one ABI provides 64-bit addressing and `long' (sometimes +known as LP64). But a few environments (IRIX, HP-UX) provide +intermediate ABI's using 32-bit addressing but allow efficient 64-bit +operations through a `long long' type. For the latter to be useful to +GMP, the ABI must allow operations using the native 64-bit +instructions provided by the ISA, and allow passing of 64-bit +quantities atomically. + +The ABI is typically chosen by means of command line options to the +compiler tools (gcc, cc, c89, nm, ar, ld, as). Different environments +use different defaults, but as of this writing (May 2000) the +dominating default is to the plain 32-bit ABI in its most arcane form. + +The GMP 3.0.x approach was to compile using the ABI that gives the +best performance. That places the burden on users to pass special +options to the compiler when they compile their GMP applications. +That approach has its advantages and disadvantages. The main +advantage is that users don't unknowingly get bad GMP performance. +The main disadvantage is that users' compiles (actually links) will +fail unless they pass special compiler options. + +** SPARC + +System vendors often confuse ABI, ISA, and implementation. The worst +case is Solaris, were the unbundled compiler confuses ISA and ABI, and +the options have very confusing names. + + option interpretation + ====== ============== +cc -xarch=v8plus ISA=sparcv9, ABI=V8plus (PTR=32, see below) +gcc -mv8plus ISA=sparcv9, ABI=V8plus (see below) +cc -xarch=v9 ISA=sparcv9, ABI=V9 (implying AR=64, PTR=64) + +It's hard to believe, but the option v8plus really means ISA=V9! + +Solaris releases prior to version 7 running on a V9 CPU fails to +save/restore the upper 32 bits of the `i' and `l' registers. The +`v8plus' option generates code that use as many V9 features as +possible under such circumstances. + +** MIPS + +The IRIX 6 compilers gets things right. They have a clear +understanding of the differences between ABI and ISA. The option +names are descriptive. + + option interpretation + ====== ============== +cc -n32 ABI=n32 (implying AR=64, PTR=32) +gcc -mabi=n32 ABI=n32 (implying AR=64, PTR=32) +cc -64 ABI=64 (implying AR=64, PTR=64) +gcc -mabi=64 ABI=64 (implying AR=64, PTR=64) +cc -mips3 ISA=mips3 +gcc -mips3 ISA=mips3 +cc -mips4 ISA=mips4 +gcc -mips4 ISA=mips4 + +** HP-PA + +HP-UX is somewhat weird, but not as broken as Solaris. + + option interpretation + ====== ============== +cc +DA2.0 ABI=32bit (implying AR=64, PTR=32) +cc +DD64 ABI=64bit (implying AR=64, PTR=64) + +Code performing 64-bit arithmetic in the HP-UX 32-bit is not +compatible with the 64-bit ABI; the former has a calling convention +that passes/returns 64-bit integer quantities as two 32-bit chunks. + +** PowerPC + +While the PowerPC ABI's are capable of supporting 64-bit +registers/operations, the compilers under AIX are similar to Solaris' +cc in that they don't currently provide any 32-bit addressing with +64-bit arithmetic. + + option interpretation + ====== ============== +cc -q64 ABI=64bit (implying AR=64, PTR=64) +gcc -maix64 -mpowerpc64 ABI=64bit (implying AR=64, PTR=64) -- cgit v1.2.3