From 96d2287c722e4566fd30271c76869f83831e86c2 Mon Sep 17 00:00:00 2001 From: Mark Crichton Date: Sat, 5 Mar 2011 22:22:57 -0500 Subject: [PATCH] X86_64 SSE2 support for Linux --- Makefile.am | 7 +- configure.ac | 33 ++++++ cpu-miner.c | 19 ++++ miner.h | 9 ++ sha256_sse2_amd64.c | 129 +++++++++++++++++++++ x86_64/.gitignore | 1 + x86_64/Makefile.am | 8 ++ x86_64/sha256_xmm_amd64.asm | 219 ++++++++++++++++++++++++++++++++++++ 8 files changed, 424 insertions(+), 1 deletion(-) create mode 100644 sha256_sse2_amd64.c create mode 100644 x86_64/.gitignore create mode 100644 x86_64/Makefile.am create mode 100644 x86_64/sha256_xmm_amd64.asm diff --git a/Makefile.am b/Makefile.am index 3cf7c00..a56fd5c 100644 --- a/Makefile.am +++ b/Makefile.am @@ -14,7 +14,12 @@ INCLUDES = $(PTHREAD_FLAGS) -fno-strict-aliasing $(JANSSON_INCLUDES) bin_PROGRAMS = minerd minerd_SOURCES = cpu-miner.c sha256_generic.c sha256_4way.c sha256_via.c \ - sha256_cryptopp.c util.c miner.h compat.h + sha256_cryptopp.c sha256_sse2_amd64.c util.c miner.h compat.h minerd_LDFLAGS = $(PTHREAD_FLAGS) minerd_LDADD = @LIBCURL@ @JANSSON_LIBS@ @PTHREAD_LIBS@ +if HAS_YASM +SUBDIRS += x86_64 +minerd_LDADD += x86_64/libx8664.a +AM_CFLAGS = -DHAS_YASM +endif diff --git a/configure.ac b/configure.ac index 25ee2fa..c0d3dd9 100644 --- a/configure.ac +++ b/configure.ac @@ -42,6 +42,38 @@ else JANSSON_LIBS=-ljansson fi +dnl Find YASM +has_yasm=false +AC_PATH_PROG(YASM,[yasm],[true yasm]) +if test "$YASM" = "false" ; then + AC_MSG_WARN([yasm is required for the x86_64 SSE2 code, but it was not found]) +fi + +AC_MSG_CHECKING([if yasm version is greater than 1.0.1]) +yasmver=`yasm --version | head -1 | cut -d\ -f2` +yamajor=`echo $yasmver | cut -d. -f1` +yaminor=`echo $yasmver | cut -d. -f2` +yamini=`echo $yasmver | cut -d. -f3` +if test "$yamajor" -ge "1" ; then + if test "$yamajor" -eq "1" ; then + if test "$yaminor" -ge "0" ; then + if test "$yamini" -ge "1"; then + has_yasm=true + fi + fi + fi +else +has_yasm=false +fi + +if test "x$has_yasm" != "x" ; then + AC_MSG_RESULT([yes]) +else + AC_MSG_RESULT([no]) +fi + +AM_CONDITIONAL([HAS_YASM], [test x$has_yasm = xtrue]) + PKG_PROG_PKG_CONFIG() LIBCURL_CHECK_CONFIG(, 7.10.1, , @@ -55,6 +87,7 @@ AC_CONFIG_FILES([ Makefile compat/Makefile compat/jansson/Makefile + x86_64/Makefile ]) AC_OUTPUT diff --git a/cpu-miner.c b/cpu-miner.c index dffc7ed..fd64a0c 100644 --- a/cpu-miner.c +++ b/cpu-miner.c @@ -38,6 +38,7 @@ enum sha256_algos { ALGO_VIA, /* VIA padlock */ ALGO_CRYPTOPP, /* Crypto++ (C) */ ALGO_CRYPTOPP_ASM32, /* Crypto++ 32-bit assembly */ + ALGO_SSE2_64, /* SSE2 for x86_64 */ }; static const char *algo_names[] = { @@ -52,6 +53,9 @@ static const char *algo_names[] = { #ifdef WANT_CRYPTOPP_ASM32 [ALGO_CRYPTOPP_ASM32] = "cryptopp_asm32", #endif +#ifdef WANT_X8664_SSE2 + [ALGO_SSE2_64] = "sse2_64", +#endif }; bool opt_debug = false; @@ -93,6 +97,9 @@ static struct option_help options_help[] = { "\n\tcryptopp\tCrypto++ C/C++ implementation" #ifdef WANT_CRYPTOPP_ASM32 "\n\tcryptopp_asm32\tCrypto++ 32-bit assembler implementation" +#endif +#ifdef WANT_X8664_SSE2 + "\n\tsse2_64\t\tSSE2 implementation for x86_64 machines" #endif }, @@ -331,6 +338,18 @@ static void *miner_thread(void *thr_id_int) max_nonce, &hashes_done); break; +#ifdef WANT_X8664_SSE2 + case ALGO_SSE2_64: { + unsigned int rc5 = + scanhash_sse2_64(work.midstate, work.data + 64, + work.hash1, work.hash, + work.target, + max_nonce, &hashes_done); + rc = (rc5 == -1) ? false : true; + } + break; +#endif + #ifdef WANT_SSE2_4WAY case ALGO_4WAY: { unsigned int rc4 = diff --git a/miner.h b/miner.h index 1c73114..3468501 100644 --- a/miner.h +++ b/miner.h @@ -15,6 +15,10 @@ #define WANT_VIA_PADLOCK 1 #endif +#if defined(__x86_64__) && defined(__SSE2__) && defined(HAS_YASM) +#define WANT_X8664_SSE2 1 +#endif + #if ((__GNUC__ > 4) || (__GNUC__ == 4 && __GNUC_MINOR__ >= 3)) #define WANT_BUILTIN_BSWAP #else @@ -74,6 +78,11 @@ extern unsigned int ScanHash_4WaySSE2(const unsigned char *pmidstate, const unsigned char *ptarget, uint32_t max_nonce, unsigned long *nHashesDone); +extern unsigned int scanhash_sse2_amd64(const unsigned char *pmidstate, + unsigned char *pdata, unsigned char *phash1, unsigned char *phash, + const unsigned char *ptarget, + uint32_t max_nonce, unsigned long *nHashesDone); + extern bool scanhash_via(unsigned char *data_inout, const unsigned char *target, uint32_t max_nonce, unsigned long *hashes_done); diff --git a/sha256_sse2_amd64.c b/sha256_sse2_amd64.c new file mode 100644 index 0000000..87c6993 --- /dev/null +++ b/sha256_sse2_amd64.c @@ -0,0 +1,129 @@ +/* + * SHA-256 driver for ASM routine for x86_64 on Linux + * Copyright (c) Mark Crichton + * + * This program is free software; you can redistribute it and/or modify it + * under the terms of 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. + * + */ + +#include "miner.h" + +#ifdef WANT_X8664_SSE2 + +#include +#include + +#include +#include +#include + +extern void CalcSha256_x64(__m128i *res, __m128i *data, uint32_t init[8]); + +uint32_t g_sha256_k[] = { + 0x428a2f98, 0x71374491, 0xb5c0fbcf, 0xe9b5dba5, /* 0 */ + 0x3956c25b, 0x59f111f1, 0x923f82a4, 0xab1c5ed5, + 0xd807aa98, 0x12835b01, 0x243185be, 0x550c7dc3, /* 8 */ + 0x72be5d74, 0x80deb1fe, 0x9bdc06a7, 0xc19bf174, + 0xe49b69c1, 0xefbe4786, 0x0fc19dc6, 0x240ca1cc, /* 16 */ + 0x2de92c6f, 0x4a7484aa, 0x5cb0a9dc, 0x76f988da, + 0x983e5152, 0xa831c66d, 0xb00327c8, 0xbf597fc7, /* 24 */ + 0xc6e00bf3, 0xd5a79147, 0x06ca6351, 0x14292967, + 0x27b70a85, 0x2e1b2138, 0x4d2c6dfc, 0x53380d13, /* 32 */ + 0x650a7354, 0x766a0abb, 0x81c2c92e, 0x92722c85, + 0xa2bfe8a1, 0xa81a664b, 0xc24b8b70, 0xc76c51a3, /* 40 */ + 0xd192e819, 0xd6990624, 0xf40e3585, 0x106aa070, + 0x19a4c116, 0x1e376c08, 0x2748774c, 0x34b0bcb5, /* 48 */ + 0x391c0cb3, 0x4ed8aa4a, 0x5b9cca4f, 0x682e6ff3, + 0x748f82ee, 0x78a5636f, 0x84c87814, 0x8cc70208, /* 56 */ + 0x90befffa, 0xa4506ceb, 0xbef9a3f7, 0xc67178f2 +}; + + +uint32_t g_sha256_hinit[8] = +{0x6a09e667, 0xbb67ae85, 0x3c6ef372, 0xa54ff53a, 0x510e527f, 0x9b05688c, 0x1f83d9ab, 0x5be0cd19}; + +__m128i g_4sha256_k[64]; + +int scanhash_sse2_64(const unsigned char *pmidstate, unsigned char *pdata, + unsigned char *phash1, unsigned char *phash, + const unsigned char *ptarget, + uint32_t max_nonce, unsigned long *nHashesDone) +{ + uint32_t *nNonce_p = (uint32_t *)(pdata + 12); + uint32_t nonce = 0; + uint32_t m_midstate[8], m_w[16], m_w1[16]; + __m128i m_4w[64], m_4hash[64], m_4hash1[64]; + __m128i offset; + __m128i g_4sha256_hinit[8]; + int i; + + /* For debugging */ + union { + __m128i m; + uint32_t i[4]; + } mi; + + /* Message expansion */ + memcpy(m_midstate, pmidstate, sizeof(m_midstate)); + memcpy(m_w, pdata, sizeof(m_w)); /* The 2nd half of the data */ + memcpy(m_w1, phash1, sizeof(m_w1)); + memset(m_4hash, 0, sizeof(m_4hash)); + + /* Transmongrify */ + for (i = 0; i < 16; i++) + m_4w[i] = _mm_set1_epi32(m_w[i]); + + for (i = 0; i < 16; i++) + m_4hash1[i] = _mm_set1_epi32(m_w1[i]); + + for (i = 0; i < 64; i++) + g_4sha256_k[i] = _mm_set1_epi32(g_sha256_k[i]); + + offset = _mm_set_epi32(0x3, 0x2, 0x1, 0x0); + + for (;;) + { + int j; + + m_4w[3] = _mm_add_epi32(offset, _mm_set1_epi32(nonce)); + + /* Some optimization can be done here W.R.T. precalculating some hash */ + CalcSha256_x64(m_4hash1, m_4w, m_midstate); + CalcSha256_x64(m_4hash, m_4hash1, g_sha256_hinit); + + for (j = 0; j < 4; j++) { + mi.m = m_4hash[7]; + if (mi.i[j] == 0) + break; + } + + /* If j = true, we found a hit...so check it */ + /* Use the C version for a check... */ + if (j != 4) { + for (i = 0; i < 8; i++) { + mi.m = m_4hash[i]; + *(uint32_t *)&(phash)[i*4] = mi.i[j]; + } + + if (fulltest(phash, ptarget)) { + *nHashesDone = nonce; + *nNonce_p = nonce + j; + return nonce + j; + } + } + + nonce += 4; + + if (nonce >= max_nonce) + { + *nHashesDone = nonce; + return -1; + } + } +} + +#endif /* WANT_X8664_SSE2 */ + diff --git a/x86_64/.gitignore b/x86_64/.gitignore new file mode 100644 index 0000000..a966652 --- /dev/null +++ b/x86_64/.gitignore @@ -0,0 +1 @@ +libx8664.a diff --git a/x86_64/Makefile.am b/x86_64/Makefile.am new file mode 100644 index 0000000..c74ddd2 --- /dev/null +++ b/x86_64/Makefile.am @@ -0,0 +1,8 @@ +noinst_LIBRARIES = libx8664.a + +SUFFIXES = .asm + +libx8664_a_SOURCES = sha256_xmm_amd64.asm + +.asm.o: + $(YASM) -f elf64 $< diff --git a/x86_64/sha256_xmm_amd64.asm b/x86_64/sha256_xmm_amd64.asm new file mode 100644 index 0000000..4fa0ea9 --- /dev/null +++ b/x86_64/sha256_xmm_amd64.asm @@ -0,0 +1,219 @@ +;; SHA-256 for X86-64 for Linux, based off of: + +; (c) Ufasoft 2011 http://ufasoft.com mailto:support@ufasoft.com +; Version 2011 +; This software is Public Domain + +; SHA-256 CPU SSE cruncher for Bitcoin Miner + +ALIGN 32 +BITS 64 + +%define hash rdi +%define data rsi +%define init rdx + +extern g_4sha256_k + +global CalcSha256_x64 +; CalcSha256 hash(rdi), data(rsi), init(rdx) +CalcSha256_x64: + + push rbx + +LAB_NEXT_NONCE: + mov r11, data +; mov rax, pnonce +; mov eax, [rax] +; mov [rbx+3*16], eax +; inc eax +; mov [rbx+3*16+4], eax +; inc eax +; mov [rbx+3*16+8], eax +; inc eax +; mov [rbx+3*16+12], eax + + mov rcx, 64*4 ;rcx is # of SHA-2 rounds + mov rax, 16*4 ;rax is where we expand to + +LAB_SHA: + push rcx + lea rcx, qword [r11+rcx*4] + lea r11, qword [r11+rax*4] +LAB_CALC: + movdqa xmm0, [r11-15*16] + movdqa xmm2, xmm0 ; (Rotr32(w_15, 7) ^ Rotr32(w_15, 18) ^ (w_15 >> 3)) + psrld xmm0, 3 + movdqa xmm1, xmm0 + pslld xmm2, 14 + psrld xmm1, 4 + pxor xmm0, xmm1 + pxor xmm0, xmm2 + pslld xmm2, 11 + psrld xmm1, 11 + pxor xmm0, xmm1 + pxor xmm0, xmm2 + + paddd xmm0, [r11-16*16] + + movdqa xmm3, [r11-2*16] + movdqa xmm2, xmm3 ; (Rotr32(w_2, 17) ^ Rotr32(w_2, 19) ^ (w_2 >> 10)) + psrld xmm3, 10 + movdqa xmm1, xmm3 + pslld xmm2, 13 + psrld xmm1, 7 + pxor xmm3, xmm1 + pxor xmm3, xmm2 + pslld xmm2, 2 + psrld xmm1, 2 + pxor xmm3, xmm1 + pxor xmm3, xmm2 + paddd xmm0, xmm3 + + paddd xmm0, [r11-7*16] + movdqa [r11], xmm0 + add r11, 16 + cmp r11, rcx + jb LAB_CALC + pop rcx + + mov rax, 0 + +; Load the init values of the message into the hash. + + movd xmm0, dword [rdx+4*4] ; xmm0 == e + pshufd xmm0, xmm0, 0 + movd xmm3, dword [rdx+3*4] ; xmm3 == d + pshufd xmm3, xmm3, 0 + movd xmm4, dword [rdx+2*4] ; xmm4 == c + pshufd xmm4, xmm4, 0 + movd xmm5, dword [rdx+1*4] ; xmm5 == b + pshufd xmm5, xmm5, 0 + movd xmm7, dword [rdx+0*4] ; xmm7 == a + pshufd xmm7, xmm7, 0 + movd xmm8, dword [rdx+5*4] ; xmm8 == f + pshufd xmm8, xmm8, 0 + movd xmm9, dword [rdx+6*4] ; xmm9 == g + pshufd xmm9, xmm9, 0 + movd xmm10, dword [rdx+7*4] ; xmm10 == h + pshufd xmm10, xmm10, 0 + +LAB_LOOP: + +;; T t1 = h + (Rotr32(e, 6) ^ Rotr32(e, 11) ^ Rotr32(e, 25)) + ((e & f) ^ AndNot(e, g)) + Expand32(g_sha256_k[j]) + w[j] + + movdqa xmm6, [rsi+rax*4] + paddd xmm6, g_4sha256_k[rax*4] + add rax, 4 + + paddd xmm6, xmm10 ; +h + + movdqa xmm1, xmm0 + movdqa xmm2, xmm9 + pandn xmm1, xmm2 ; ~e & g + + movdqa xmm10, xmm2 ; h = g + movdqa xmm2, xmm8 ; f + movdqa xmm9, xmm2 ; g = f + + pand xmm2, xmm0 ; e & f + pxor xmm1, xmm2 ; (e & f) ^ (~e & g) + movdqa xmm8, xmm0 ; f = e + + paddd xmm6, xmm1 ; Ch + h + w[i] + k[i] + + movdqa xmm1, xmm0 + psrld xmm0, 6 + movdqa xmm2, xmm0 + pslld xmm1, 7 + psrld xmm2, 5 + pxor xmm0, xmm1 + pxor xmm0, xmm2 + pslld xmm1, 14 + psrld xmm2, 14 + pxor xmm0, xmm1 + pxor xmm0, xmm2 + pslld xmm1, 5 + pxor xmm0, xmm1 ; Rotr32(e, 6) ^ Rotr32(e, 11) ^ Rotr32(e, 25) + paddd xmm6, xmm0 ; xmm6 = t1 + + movdqa xmm0, xmm3 ; d + paddd xmm0, xmm6 ; e = d+t1 + + movdqa xmm1, xmm5 ; =b + movdqa xmm3, xmm4 ; d = c + movdqa xmm2, xmm4 ; c + pand xmm2, xmm5 ; b & c + pand xmm4, xmm7 ; a & c + pand xmm1, xmm7 ; a & b + pxor xmm1, xmm4 + movdqa xmm4, xmm5 ; c = b + movdqa xmm5, xmm7 ; b = a + pxor xmm1, xmm2 ; (a & c) ^ (a & d) ^ (c & d) + paddd xmm6, xmm1 ; t1 + ((a & c) ^ (a & d) ^ (c & d)) + + movdqa xmm2, xmm7 + psrld xmm7, 2 + movdqa xmm1, xmm7 + pslld xmm2, 10 + psrld xmm1, 11 + pxor xmm7, xmm2 + pxor xmm7, xmm1 + pslld xmm2, 9 + psrld xmm1, 9 + pxor xmm7, xmm2 + pxor xmm7, xmm1 + pslld xmm2, 11 + pxor xmm7, xmm2 + paddd xmm7, xmm6 ; a = t1 + (Rotr32(a, 2) ^ Rotr32(a, 13) ^ Rotr32(a, 22)) + ((a & c) ^ (a & d) ^ (c & d)); + + cmp rax, rcx + jb LAB_LOOP + +; Finished the 64 rounds, calculate hash and save + + movd xmm1, dword [rdx+0*4] + pshufd xmm1, xmm1, 0 + paddd xmm7, xmm1 + + movd xmm1, dword [rdx+1*4] + pshufd xmm1, xmm1, 0 + paddd xmm5, xmm1 + + movd xmm1, dword [rdx+2*4] + pshufd xmm1, xmm1, 0 + paddd xmm4, xmm1 + + movd xmm1, dword [rdx+3*4] + pshufd xmm1, xmm1, 0 + paddd xmm3, xmm1 + + movd xmm1, dword [rdx+4*4] + pshufd xmm1, xmm1, 0 + paddd xmm0, xmm1 + + movd xmm1, dword [rdx+5*4] + pshufd xmm1, xmm1, 0 + paddd xmm8, xmm1 + + movd xmm1, dword [rdx+6*4] + pshufd xmm1, xmm1, 0 + paddd xmm9, xmm1 + + movd xmm1, dword [rdx+7*4] + pshufd xmm1, xmm1, 0 + paddd xmm10, xmm1 + +debug_me: + movdqa [rdi+0*16], xmm7 + movdqa [rdi+1*16], xmm5 + movdqa [rdi+2*16], xmm4 + movdqa [rdi+3*16], xmm3 + movdqa [rdi+4*16], xmm0 + movdqa [rdi+5*16], xmm8 + movdqa [rdi+6*16], xmm9 + movdqa [rdi+7*16], xmm10 + +LAB_RET: + pop rbx + ret