From 4b4c5f463c188cb8fece81a3a9d74e90de3ffa00 Mon Sep 17 00:00:00 2001
From: Oksana Shadura <oksana.shadura@cern.ch>
Date: Wed, 28 Mar 2018 18:16:46 +0200
Subject: [PATCH] Reverting CRC32-SSE4.2 implementation, ROOT is not using
CRC32 but Adler32 instead The problem was in crc32-pclmul_asm.S that is
licensed under GPL, while ROOT has LGPL. ROOT compression performance stay
the same.
---
builtins/zlib/CMakeLists.txt | 9 -
.../zlib/contrib/amd64/crc32-pclmul_asm.S | 258 ------------------
builtins/zlib/crc32.c | 67 -----
3 files changed, 334 deletions(-)
delete mode 100644 builtins/zlib/contrib/amd64/crc32-pclmul_asm.S
diff --git a/builtins/zlib/CMakeLists.txt b/builtins/zlib/CMakeLists.txt
index ac424bfd08f..95429540499 100644
--- a/builtins/zlib/CMakeLists.txt
+++ b/builtins/zlib/CMakeLists.txt
@@ -1,14 +1,5 @@
project(ZLIB C)
-set(CMAKE_ASM_OUTPUT_EXTENSION ".o")
-set(CMAKE_ASM_OUTPUT_EXTENSION_REPLACE 1)
-set(CMAKE_ASM_COMPILE_OBJECT "<CMAKE_ASM_COMPILER> <FLAGS> -o <OBJECT> <SOURCE>")
-set(CMAKE_ASM_SOURCE_FILE_EXTENSIONS s;asm;msa)
-if((CMAKE_SYSTEM_PROCESSOR MATCHES "amd64|x86_64|AMD64|X86_64") AND (CMAKE_SYSTEM_NAME MATCHES "Linux"))
- set(ZLIB_ASMS contrib/amd64/crc32-pclmul_asm.S)
- set_source_files_properties(${ZLIB_ASMS} PROPERTIES LANGUAGE C)
-endif()
-
root_check_assembler()
set(ZLIB_PUBLIC_HEADERS
diff --git a/builtins/zlib/contrib/amd64/crc32-pclmul_asm.S b/builtins/zlib/contrib/amd64/crc32-pclmul_asm.S
deleted file mode 100644
index ed23699a052..00000000000
--- a/builtins/zlib/contrib/amd64/crc32-pclmul_asm.S
+++ /dev/null
@@ -1,258 +0,0 @@
-/* GPL HEADER START
- *
- * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
- *
- * This program is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License version 2 only,
- * as published by the Free Software Foundation.
- *
- * This program 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 version 2 for more details (a copy is included
- * in the LICENSE file that accompanied this code).
- *
- * You should have received a copy of the GNU General Public License
- * version 2 along with this program; If not, see http://www.gnu.org/licenses
- *
- * Please visit http://www.xyratex.com/contact if you need additional
- * information or have any questions.
- *
- * GPL HEADER END
- */
-
-/*
- * Copyright 2012 Xyratex Technology Limited
- *
- * Using hardware provided PCLMULQDQ instruction to accelerate the CRC32
- * calculation.
- * CRC32 polynomial:0x04c11db7(BE)/0xEDB88320(LE)
- * PCLMULQDQ is a new instruction in Intel SSE4.2, the reference can be found
- * at:
- * http://www.intel.com/products/processor/manuals/
- * Intel(R) 64 and IA-32 Architectures Software Developer's Manual
- * Volume 2B: Instruction Set Reference, N-Z
- *
- * Authors: Gregory Prestas <Gregory_Prestas@us.xyratex.com>
- * Alexander Boyko <Alexander_Boyko@xyratex.com>
- */
-
-/* This file is "stolen" from linux kernel 3.14 with following minor changes to
- * make it self-contained:
- * - remove all header files it includes
- * - define ENTRY and ENDPROC macros
- * - prepend '$' to some immediate operands to make assembler happy.
- */
-
-#define ENTRY(name) \
-.globl name; \
-.hidden name; \
-.type name, @function; \
-name:
-
-#define ENDPROC(name) \
-.size name, .-name
-
-.align 16
-/*
- * [x4*128+32 mod P(x) << 32)]' << 1 = 0x154442bd4
- * #define CONSTANT_R1 0x154442bd4LL
- *
- * [(x4*128-32 mod P(x) << 32)]' << 1 = 0x1c6e41596
- * #define CONSTANT_R2 0x1c6e41596LL
- */
-.Lconstant_R2R1:
- .octa 0x00000001c6e415960000000154442bd4
-/*
- * [(x128+32 mod P(x) << 32)]' << 1 = 0x1751997d0
- * #define CONSTANT_R3 0x1751997d0LL
- *
- * [(x128-32 mod P(x) << 32)]' << 1 = 0x0ccaa009e
- * #define CONSTANT_R4 0x0ccaa009eLL
- */
-.Lconstant_R4R3:
- .octa 0x00000000ccaa009e00000001751997d0
-/*
- * [(x64 mod P(x) << 32)]' << 1 = 0x163cd6124
- * #define CONSTANT_R5 0x163cd6124LL
- */
-.Lconstant_R5:
- .octa 0x00000000000000000000000163cd6124
-.Lconstant_mask32:
- .octa 0x000000000000000000000000FFFFFFFF
-/*
- * #define CRCPOLY_TRUE_LE_FULL 0x1DB710641LL
- *
- * Barrett Reduction constant (u64`) = u` = (x**64 / P(x))` = 0x1F7011641LL
- * #define CONSTANT_RU 0x1F7011641LL
- */
-.Lconstant_RUpoly:
- .octa 0x00000001F701164100000001DB710641
-
-#define CONSTANT %xmm0
-
-#ifdef __x86_64__
-#define BUF %rdi
-#define LEN %rsi
-#define CRC %edx
-#else
-#define BUF %eax
-#define LEN %edx
-#define CRC %ecx
-#endif
-
-
-
-.text
-/**
- * Calculate crc32
- * BUF - buffer (16 bytes aligned)
- * LEN - sizeof buffer (16 bytes aligned), LEN should be grater than 63
- * CRC - initial crc32
- * return %eax crc32
- * uint crc32_pclmul_le_16(unsigned char const *buffer,
- * size_t len, uint crc32)
- */
-
-ENTRY(crc32_pclmul_le_16) /* buffer and buffer size are 16 bytes aligned */
- movdqa (BUF), %xmm1
- movdqa 0x10(BUF), %xmm2
- movdqa 0x20(BUF), %xmm3
- movdqa 0x30(BUF), %xmm4
- movd CRC, CONSTANT
- pxor CONSTANT, %xmm1
- sub $0x40, LEN
- add $0x40, BUF
-#ifndef __x86_64__
- /* This is for position independent code(-fPIC) support for 32bit */
- call delta
-delta:
- pop %ecx
-#endif
- cmp $0x40, LEN
- jb less_64
-
-#ifdef __x86_64__
- movdqa .Lconstant_R2R1(%rip), CONSTANT
-#else
- movdqa .Lconstant_R2R1 - delta(%ecx), CONSTANT
-#endif
-
-loop_64:/* 64 bytes Full cache line folding */
- prefetchnta 0x40(BUF)
- movdqa %xmm1, %xmm5
- movdqa %xmm2, %xmm6
- movdqa %xmm3, %xmm7
-#ifdef __x86_64__
- movdqa %xmm4, %xmm8
-#endif
- PCLMULQDQ $00, CONSTANT, %xmm1
- PCLMULQDQ $00, CONSTANT, %xmm2
- PCLMULQDQ $00, CONSTANT, %xmm3
-#ifdef __x86_64__
- PCLMULQDQ $00, CONSTANT, %xmm4
-#endif
- PCLMULQDQ $0x11, CONSTANT, %xmm5
- PCLMULQDQ $0x11, CONSTANT, %xmm6
- PCLMULQDQ $0x11, CONSTANT, %xmm7
-#ifdef __x86_64__
- PCLMULQDQ $0x11, CONSTANT, %xmm8
-#endif
- pxor %xmm5, %xmm1
- pxor %xmm6, %xmm2
- pxor %xmm7, %xmm3
-#ifdef __x86_64__
- pxor %xmm8, %xmm4
-#else
- /* xmm8 unsupported for x32 */
- movdqa %xmm4, %xmm5
- PCLMULQDQ $00, CONSTANT, %xmm4
- PCLMULQDQ $0x11, CONSTANT, %xmm5
- pxor %xmm5, %xmm4
-#endif
-
- pxor (BUF), %xmm1
- pxor 0x10(BUF), %xmm2
- pxor 0x20(BUF), %xmm3
- pxor 0x30(BUF), %xmm4
-
- sub $0x40, LEN
- add $0x40, BUF
- cmp $0x40, LEN
- jge loop_64
-less_64:/* Folding cache line into 128bit */
-#ifdef __x86_64__
- movdqa .Lconstant_R4R3(%rip), CONSTANT
-#else
- movdqa .Lconstant_R4R3 - delta(%ecx), CONSTANT
-#endif
- prefetchnta (BUF)
-
- movdqa %xmm1, %xmm5
- PCLMULQDQ $0x00, CONSTANT, %xmm1
- PCLMULQDQ $0x11, CONSTANT, %xmm5
- pxor %xmm5, %xmm1
- pxor %xmm2, %xmm1
-
- movdqa %xmm1, %xmm5
- PCLMULQDQ $0x00, CONSTANT, %xmm1
- PCLMULQDQ $0x11, CONSTANT, %xmm5
- pxor %xmm5, %xmm1
- pxor %xmm3, %xmm1
-
- movdqa %xmm1, %xmm5
- PCLMULQDQ $0x00, CONSTANT, %xmm1
- PCLMULQDQ $0x11, CONSTANT, %xmm5
- pxor %xmm5, %xmm1
- pxor %xmm4, %xmm1
-
- cmp $0x10, LEN
- jb fold_64
-loop_16:/* Folding rest buffer into 128bit */
- movdqa %xmm1, %xmm5
- PCLMULQDQ $0x00, CONSTANT, %xmm1
- PCLMULQDQ $0x11, CONSTANT, %xmm5
- pxor %xmm5, %xmm1
- pxor (BUF), %xmm1
- sub $0x10, LEN
- add $0x10, BUF
- cmp $0x10, LEN
- jge loop_16
-
-fold_64:
- /* perform the last 64 bit fold, also adds 32 zeroes
- * to the input stream */
- PCLMULQDQ $0x01, %xmm1, CONSTANT /* R4 * xmm1.low */
- psrldq $0x08, %xmm1
- pxor CONSTANT, %xmm1
-
- /* final 32-bit fold */
- movdqa %xmm1, %xmm2
-#ifdef __x86_64__
- movdqa .Lconstant_R5(%rip), CONSTANT
- movdqa .Lconstant_mask32(%rip), %xmm3
-#else
- movdqa .Lconstant_R5 - delta(%ecx), CONSTANT
- movdqa .Lconstant_mask32 - delta(%ecx), %xmm3
-#endif
- psrldq $0x04, %xmm2
- pand %xmm3, %xmm1
- PCLMULQDQ $0x00, CONSTANT, %xmm1
- pxor %xmm2, %xmm1
-
- /* Finish up with the bit-reversed barrett reduction 64 ==> 32 bits */
-#ifdef __x86_64__
- movdqa .Lconstant_RUpoly(%rip), CONSTANT
-#else
- movdqa .Lconstant_RUpoly - delta(%ecx), CONSTANT
-#endif
- movdqa %xmm1, %xmm2
- pand %xmm3, %xmm1
- PCLMULQDQ $0x10, CONSTANT, %xmm1
- pand %xmm3, %xmm1
- PCLMULQDQ $0x00, CONSTANT, %xmm1
- pxor %xmm2, %xmm1
- PEXTRD $0x01, %xmm1, %eax
-
- ret
-ENDPROC(crc32_pclmul_le_16)
diff --git a/builtins/zlib/crc32.c b/builtins/zlib/crc32.c
index 254d8aba544..acb6c50845b 100644
--- a/builtins/zlib/crc32.c
+++ b/builtins/zlib/crc32.c
@@ -273,72 +273,6 @@ local unsigned long crc32_generic(crc, buf, len)
return crc ^ 0xffffffffUL;
}
-#if defined (__x86_64__) && defined (__linux__)
-
-/* Function stolen from linux kernel 3.14. It computes the CRC over the given
- * buffer with initial CRC value <crc32>. The buffer is <len> byte in length,
- * and must be 16-byte aligned.
- */
-extern uint crc32_pclmul_le_16(unsigned char const *buffer,
- size_t len, uInt crc32);
-
-uLong crc32_pclmul(uLong, const Bytef *, uInt) __attribute__ ((__target__ ("sse4.2,pclmul")));
-
-uLong crc32_pclmul(crc, buf, len)
- uLong crc;
- const Bytef *buf;
- uInt len;
-{
-#define PCLMUL_MIN_LEN 64
-#define PCLMUL_ALIGN 16
-#define PCLMUL_ALIGN_MASK 15
-
- if (len < PCLMUL_MIN_LEN + PCLMUL_ALIGN - 1)
- return crc32_generic(crc, buf, len);
-
- /* Handle the leading patial chunk */
- uInt misalign = PCLMUL_ALIGN_MASK & ((unsigned long)buf);
- uInt sz = (PCLMUL_ALIGN - misalign) % PCLMUL_ALIGN;
- if (sz) {
- crc = crc32_generic(crc, buf, sz);
- buf += sz;
- len -= sz;
- }
-
- /* Go over 16-byte chunks */
- crc = crc32_pclmul_le_16(buf, (len & ~PCLMUL_ALIGN_MASK),
- crc ^ 0xffffffffUL);
- crc = crc ^ 0xffffffffUL;
-
- /* Handle the trailing partial chunk */
- sz = len & PCLMUL_ALIGN_MASK;
- if (sz) {
- crc = crc32_generic(crc, buf + len - sz, sz);
- }
-
- return crc;
-
-#undef PCLMUL_MIN_LEN
-#undef PCLMUL_ALIGN
-#undef PCLMUL_ALIGN_MASK
-}
-
-void *resolve_crc32(void)
-{
- unsigned int eax, ebx, ecx, edx;
- if (!__get_cpuid (1, &eax, &ebx, &ecx, &edx))
- return crc32_generic;
- /* We need SSE4.2 and PCLMUL ISA support */
- if (!((ecx & bit_SSE4_2) && (ecx & bit_PCLMUL)))
- return crc32_generic;
- return crc32_pclmul;
-}
-
-/* This function needs to be resolved at load time */
-uLong crc32(unsigned long, const unsigned char FAR *, unsigned) __attribute__ ((ifunc ("resolve_crc32")));
-
-#else // if not x86_64
-
uLong crc32(crc, buf, len)
uLong crc;
const Bytef *buf;
@@ -346,7 +280,6 @@ uLong crc32(crc, buf, len)
{
return crc32_generic(crc, buf, len);
}
-#endif
#ifdef BYFOUR
--
GitLab