From f45bd686f23c192af450aa42a59c10fcb179d8e2 Mon Sep 17 00:00:00 2001 From: bg Date: Wed, 29 Apr 2009 18:50:37 +0000 Subject: [PATCH] faster smaller better --- Makefile | 12 ++-- avr-makefile.inc | 17 +++--- bmw_large.c | 147 ++++++++++++++++++++++++++++++++++------------- bmw_small.c | 94 +++++++++++++++++++++++------- 4 files changed, 199 insertions(+), 71 deletions(-) diff --git a/Makefile b/Makefile index e730a4c..f233faf 100644 --- a/Makefile +++ b/Makefile @@ -267,7 +267,7 @@ clean: rm -rf $(BIN_DIR)*.o *.o $(TESTBIN_DIR)*.elf $(TESTBIN_DIR)* *.elf *.eps *.png *.pdf *.bak rm -rf *.lst *.map $(EXTRA_CLEAN_FILES) $(STAT_DIR)$(SIZESTAT_FILE) $(STAT_DIR)*_size.txt xclean: clean - rm -rf $(DEP_DIR)*.d + rm -rf $(DEP_DIR)*.d $(AUTOASM_DIR)*.s docu: doxygen @@ -317,10 +317,14 @@ make-info: make.png make.svg @echo "[objcopy]: $@" @$(OBJCOPY) -j .eeprom --change-section-lma .eeprom=0 -O binary $< $@ - -# Every thing below here is used by avr-libc's build system and can be ignored -# by the casual user. +#------------------------------------------------------------------------------- +$(AUTOASM_DIR)%.s: %.c + $(CC) $(CFLAGS) $(AUTOASM_OPT) -o $@ $< +%.s: %.c + $(CC) $(CFLAGS) $(AUTOASM_OPT) -o $@ $< + +#------------------------------------------------------------------------------- FIG2DEV = fig2dev EXTRA_CLEAN_FILES = *.hex *.bin *.srec diff --git a/avr-makefile.inc b/avr-makefile.inc index 95e0228..ee46746 100644 --- a/avr-makefile.inc +++ b/avr-makefile.inc @@ -17,6 +17,8 @@ TESTLOG_DIR = testlog/ TESTPREFIX = nessie- LIST_DIR = listings/ STAT_DIR = stats/ +AUTOASM_DIR = autoasm/ +AUTOASM_OPT = -S CC = avr-gcc CSTD = c99 @@ -27,11 +29,12 @@ override ASFLAGS = -mmcu=$(MCU_TARGET) -Wa,--gdwarf-2 SIZESTAT_FILE = sizestats.txt -OBJCOPY = avr-objcopy -OBJDUMP = avr-objdump -SIZE = avr-size -RUBY = ruby -GET_TEST = host/get_test.rb -MAKE = make +OBJCOPY = avr-objcopy +OBJDUMP = avr-objdump +SIZE = avr-size +READELF = readelf +RUBY = ruby +GET_TEST = host/get_test.rb +MAKE = make MAKE2GRAPH = ~/bin/make2graph.rb -TWOPI = twopi +TWOPI = twopi diff --git a/bmw_large.c b/bmw_large.c index e48da3a..454c81a 100644 --- a/bmw_large.c +++ b/bmw_large.c @@ -27,6 +27,7 @@ #include #include +#include #include "bmw_large.h" #define SHL64(a,n) ((a)<<(n)) @@ -35,7 +36,7 @@ #define ROTR64(a,n) (((a)>>(n))|((a)<<(64-(n)))) #define BUG24 1 - +#define F0_HACK 1 #define DEBUG 0 #if DEBUG #include "cli.h" @@ -164,72 +165,140 @@ uint64_t bmw_large_r7(uint64_t x){ return r; } +#define K 0x0555555555555555LL +static +uint64_t k_lut[] PROGMEM = { + 16LL*K, 17LL*K, 18LL*K, 19LL*K, 20LL*K, 21LL*K, 22LL*K, 23LL*K, + 24LL*K, 25LL*K, 26LL*K, 27LL*K, 28LL*K, 29LL*K, 30LL*K, 31LL*K }; + + uint64_t bmw_large_expand1(uint8_t j, const uint64_t* q, const void* m){ uint64_t(*s[])(uint64_t) = {bmw_large_s1, bmw_large_s2, bmw_large_s3, bmw_large_s0}; - uint64_t r; + union{ + uint64_t v64; + uint32_t v32[2]; + } r; uint8_t i; - r = 0x0555555555555555LL*(j+16); + /* r = 0x0555555555555555LL*(j+16); */ + r.v32[0] = pgm_read_dword(((uint8_t*)k_lut+8*j)); + r.v32[1] = pgm_read_dword(((uint8_t*)k_lut+8*j+4)); for(i=0; i<16; ++i){ - r += s[i%4](q[j+i]); + r.v64 += s[i%4](q[j+i]); } - r += ((uint64_t*)m)[j]; - r += ((uint64_t*)m)[j+3]; - r -= ((uint64_t*)m)[j+10]; - return r; + r.v64 += ((uint64_t*)m)[j]; + r.v64 += ((uint64_t*)m)[j+3]; + r.v64 -= ((uint64_t*)m)[j+10]; + return r.v64; } uint64_t bmw_large_expand2(uint8_t j, const uint64_t* q, const void* m){ uint64_t(*rf[])(uint64_t) = {bmw_large_r1, bmw_large_r2, bmw_large_r3, bmw_large_r4, bmw_large_r5, bmw_large_r6, bmw_large_r7}; - uint64_t r; + union{ + uint64_t v64; + uint32_t v32[2]; + } r; uint8_t i; - r = 0x0555555555555555LL*(j+16); + /* r = 0x0555555555555555LL*(j+16); */ + r.v32[0] = pgm_read_dword(((uint8_t*)k_lut+8*j)); + r.v32[1] = pgm_read_dword(((uint8_t*)k_lut+8*j+4)); for(i=0; i<14; i+=2){ - r += q[j+i]; + r.v64 += q[j+i]; } for(i=0; i<14; i+=2){ - r += rf[i/2](q[j+i+1]); + r.v64 += rf[i/2](q[j+i+1]); } - r += bmw_large_s5(q[j+14]); - r += bmw_large_s4(q[j+15]); - r += ((uint64_t*)m)[j]; - r += ((uint64_t*)m)[(j+3)%16]; - r -= ((uint64_t*)m)[(j+10)%16]; - return r; + r.v64 += bmw_large_s5(q[j+14]); + r.v64 += bmw_large_s4(q[j+15]); + r.v64 += ((uint64_t*)m)[j]; + r.v64 += ((uint64_t*)m)[(j+3)%16]; + r.v64 -= ((uint64_t*)m)[(j+10)%16]; + return r.v64; } -void bmw_large_f0(uint64_t* q, const uint64_t* h, const void* m){ - uint64_t t[16]; - uint8_t i; +#if F0_HACK +static +uint8_t f0_lut[] PROGMEM ={ + 5<<1, ( 7<<1)+1, (10<<1)+0, (13<<1)+0, (14<<1)+0, + 6<<1, ( 8<<1)+1, (11<<1)+0, (14<<1)+0, (15<<1)+1, + 0<<1, ( 7<<1)+0, ( 9<<1)+0, (12<<1)+1, (15<<1)+0, + 0<<1, ( 1<<1)+1, ( 8<<1)+0, (10<<1)+1, (13<<1)+0, + 1<<1, ( 2<<1)+0, ( 9<<1)+0, (11<<1)+1, (14<<1)+1, + 3<<1, ( 2<<1)+1, (10<<1)+0, (12<<1)+1, (15<<1)+0, + 4<<1, ( 0<<1)+1, ( 3<<1)+1, (11<<1)+1, (13<<1)+0, + 1<<1, ( 4<<1)+1, ( 5<<1)+1, (12<<1)+1, (14<<1)+1, + 2<<1, ( 5<<1)+1, ( 6<<1)+1, (13<<1)+0, (15<<1)+1, + 0<<1, ( 3<<1)+1, ( 6<<1)+0, ( 7<<1)+1, (14<<1)+0, + 8<<1, ( 1<<1)+1, ( 4<<1)+1, ( 7<<1)+1, (15<<1)+0, + 8<<1, ( 0<<1)+1, ( 2<<1)+1, ( 5<<1)+1, ( 9<<1)+0, + 1<<1, ( 3<<1)+0, ( 6<<1)+1, ( 9<<1)+1, (10<<1)+0, + 2<<1, ( 4<<1)+0, ( 7<<1)+0, (10<<1)+0, (11<<1)+0, + 3<<1, ( 5<<1)+1, ( 8<<1)+0, (11<<1)+1, (12<<1)+1, + 12<<1, ( 4<<1)+1, ( 6<<1)+1, ( 9<<1)+1, (13<<1)+0 +}; + +void bmw_large_f0(uint64_t* q, uint64_t* h, const void* m){ + uint8_t i,j=0,v,sign,l=4; uint64_t(*s[])(uint64_t)={ bmw_large_s0, bmw_large_s1, bmw_large_s2, bmw_large_s3, bmw_large_s4 }; for(i=0; i<16; ++i){ - t[i] = h[i] ^ ((uint64_t*)m)[i]; + h[i] ^= ((uint64_t*)m)[i]; + } + dump_x(h, 16, 'T'); + memset(q, 0, 4*16); + for(i=0; i<5*16; ++i){ + v = pgm_read_byte(f0_lut+i); + sign = v&1; + v >>=1; + if(sign){ + q[j] -= h[v]; + }else{ + q[j] += h[v]; + } + if(i==l){ + j++; + l+=5; + } } - dump_x(t, 16, 'T'); - q[ 0] = (t[ 5] - t[ 7] + t[10] + t[13] + t[14]); - q[ 1] = (t[ 6] - t[ 8] + t[11] + t[14] - t[15]); - q[ 2] = (t[ 0] + t[ 7] + t[ 9] - t[12] + t[15]); - q[ 3] = (t[ 0] - t[ 1] + t[ 8] - t[10] + t[13]); - q[ 4] = (t[ 1] + t[ 2] + t[ 9] - t[11] - t[14]); - q[ 5] = (t[ 3] - t[ 2] + t[10] - t[12] + t[15]); - q[ 6] = (t[ 4] - t[ 0] - t[ 3] - t[11] + t[13]); - q[ 7] = (t[ 1] - t[ 4] - t[ 5] - t[12] - t[14]); - q[ 8] = (t[ 2] - t[ 5] - t[ 6] + t[13] - t[15]); - q[ 9] = (t[ 0] - t[ 3] + t[ 6] - t[ 7] + t[14]); - q[10] = (t[ 8] - t[ 1] - t[ 4] - t[ 7] + t[15]); - q[11] = (t[ 8] - t[ 0] - t[ 2] - t[ 5] + t[ 9]); - q[12] = (t[ 1] + t[ 3] - t[ 6] - t[ 9] + t[10]); - q[13] = (t[ 2] + t[ 4] + t[ 7] + t[10] + t[11]); - q[14] = (t[ 3] - t[ 5] + t[ 8] - t[11] - t[12]); - q[15] = (t[12] - t[ 4] - t[ 6] - t[ 9] + t[13]); dump_x(q, 16, 'W'); for(i=0; i<16; ++i){ q[i] = s[i%5](q[i]); } } +#else +void bmw_large_f0(uint64_t* q, uint64_t* h, const void* m){ + uint8_t i; + uint64_t(*s[])(uint64_t)={ bmw_large_s0, bmw_large_s1, bmw_large_s2, + bmw_large_s3, bmw_large_s4 }; + for(i=0; i<16; ++i){ + h[i] ^= ((uint64_t*)m)[i]; + } + dump_x(t, 16, 'T'); + q[ 0] = (h[ 5] - h[ 7] + h[10] + h[13] + h[14]); + q[ 1] = (h[ 6] - h[ 8] + h[11] + h[14] - h[15]); + q[ 2] = (h[ 0] + h[ 7] + h[ 9] - h[12] + h[15]); + q[ 3] = (h[ 0] - h[ 1] + h[ 8] - h[10] + h[13]); + q[ 4] = (h[ 1] + h[ 2] + h[ 9] - h[11] - h[14]); + q[ 5] = (h[ 3] - h[ 2] + h[10] - h[12] + h[15]); + q[ 6] = (h[ 4] - h[ 0] - h[ 3] - h[11] + h[13]); + q[ 7] = (h[ 1] - h[ 4] - h[ 5] - h[12] - h[14]); + q[ 8] = (h[ 2] - h[ 5] - h[ 6] + h[13] - h[15]); + q[ 9] = (h[ 0] - h[ 3] + h[ 6] - h[ 7] + h[14]); + q[10] = (h[ 8] - h[ 1] - h[ 4] - h[ 7] + h[15]); + q[11] = (h[ 8] - h[ 0] - h[ 2] - h[ 5] + h[ 9]); + q[12] = (h[ 1] + h[ 3] - h[ 6] - h[ 9] + h[10]); + q[13] = (h[ 2] + h[ 4] + h[ 7] + h[10] + h[11]); + q[14] = (h[ 3] - h[ 5] + h[ 8] - h[11] - h[12]); + q[15] = (h[12] - h[ 4] - h[ 6] - h[ 9] + h[13]); + dump_x(q, 16, 'W'); + for(i=0; i<16; ++i){ + q[i] = s[i%5](q[i]); + } +} +#endif + void bmw_large_f1(uint64_t* q, const void* m){ uint8_t i; q[16] = bmw_large_expand1(0, q, m); diff --git a/bmw_small.c b/bmw_small.c index e1ddeff..006f1e0 100644 --- a/bmw_small.c +++ b/bmw_small.c @@ -37,6 +37,7 @@ #define ROTR32(a,n) (((a)>>(n))|((a)<<(32-(n)))) #define BUG24 1 +#define F0_HACK 1 #define DEBUG 0 #if DEBUG @@ -167,6 +168,7 @@ uint32_t bmw_small_r7(uint32_t x){ } #define K 0x05555555L +static uint32_t k_lut[] PROGMEM = { 16L*K, 17L*K, 18L*K, 19L*K, 20L*K, 21L*K, 22L*K, 23L*K, 24L*K, 25L*K, 26L*K, 27L*K, 28L*K, 29L*K, 30L*K, 31L*K @@ -208,38 +210,88 @@ uint32_t bmw_small_expand2(uint8_t j, const uint32_t* q, const void* m){ r -= ((uint32_t*)m)[(j+10)%16]; return r; } +#if F0_HACK +static +uint8_t f0_lut[] PROGMEM = { + 5<<1, ( 7<<1)+1, (10<<1)+0, (13<<1)+0, (14<<1)+0, + 6<<1, ( 8<<1)+1, (11<<1)+0, (14<<1)+0, (15<<1)+1, + 0<<1, ( 7<<1)+0, ( 9<<1)+0, (12<<1)+1, (15<<1)+0, + 0<<1, ( 1<<1)+1, ( 8<<1)+0, (10<<1)+1, (13<<1)+0, + 1<<1, ( 2<<1)+0, ( 9<<1)+0, (11<<1)+1, (14<<1)+1, + 3<<1, ( 2<<1)+1, (10<<1)+0, (12<<1)+1, (15<<1)+0, + 4<<1, ( 0<<1)+1, ( 3<<1)+1, (11<<1)+1, (13<<1)+0, + 1<<1, ( 4<<1)+1, ( 5<<1)+1, (12<<1)+1, (14<<1)+1, + 2<<1, ( 5<<1)+1, ( 6<<1)+1, (13<<1)+0, (15<<1)+1, + 0<<1, ( 3<<1)+1, ( 6<<1)+0, ( 7<<1)+1, (14<<1)+0, + 8<<1, ( 1<<1)+1, ( 4<<1)+1, ( 7<<1)+1, (15<<1)+0, + 8<<1, ( 0<<1)+1, ( 2<<1)+1, ( 5<<1)+1, ( 9<<1)+0, + 1<<1, ( 3<<1)+0, ( 6<<1)+1, ( 9<<1)+1, (10<<1)+0, + 2<<1, ( 4<<1)+0, ( 7<<1)+0, (10<<1)+0, (11<<1)+0, + 3<<1, ( 5<<1)+1, ( 8<<1)+0, (11<<1)+1, (12<<1)+1, + 12<<1, ( 4<<1)+1, ( 6<<1)+1, ( 9<<1)+1, (13<<1)+0 +}; -void bmw_small_f0(uint32_t* q, const uint32_t* h, const void* m){ - uint32_t t[16]; - uint8_t i; +void bmw_small_f0(uint32_t* q, uint32_t* h, const void* m){ + uint8_t i,j=0,v,sign,l=4; uint32_t(*s[])(uint32_t)={ bmw_small_s0, bmw_small_s1, bmw_small_s2, bmw_small_s3, bmw_small_s4 }; for(i=0; i<16; ++i){ - t[i] = h[i] ^ ((uint32_t*)m)[i]; + h[i] ^= ((uint32_t*)m)[i]; + } + dump_x(h, 16, 'T'); + memset(q, 0, 4*16); + for(i=0; i<5*16; ++i){ + v = pgm_read_byte(f0_lut+i); + sign = v&1; + v >>=1; + if(sign){ + q[j] -= h[v]; + }else{ + q[j] += h[v]; + } + if(i==l){ + j++; + l+=5; + } } - dump_x(t, 16, 'T'); - q[ 0] = (t[ 5] - t[ 7] + t[10] + t[13] + t[14]); - q[ 1] = (t[ 6] - t[ 8] + t[11] + t[14] - t[15]); - q[ 2] = (t[ 0] + t[ 7] + t[ 9] - t[12] + t[15]); - q[ 3] = (t[ 0] - t[ 1] + t[ 8] - t[10] + t[13]); - q[ 4] = (t[ 1] + t[ 2] + t[ 9] - t[11] - t[14]); - q[ 5] = (t[ 3] - t[ 2] + t[10] - t[12] + t[15]); - q[ 6] = (t[ 4] - t[ 0] - t[ 3] - t[11] + t[13]); - q[ 7] = (t[ 1] - t[ 4] - t[ 5] - t[12] - t[14]); - q[ 8] = (t[ 2] - t[ 5] - t[ 6] + t[13] - t[15]); - q[ 9] = (t[ 0] - t[ 3] + t[ 6] - t[ 7] + t[14]); - q[10] = (t[ 8] - t[ 1] - t[ 4] - t[ 7] + t[15]); - q[11] = (t[ 8] - t[ 0] - t[ 2] - t[ 5] + t[ 9]); - q[12] = (t[ 1] + t[ 3] - t[ 6] - t[ 9] + t[10]); - q[13] = (t[ 2] + t[ 4] + t[ 7] + t[10] + t[11]); - q[14] = (t[ 3] - t[ 5] + t[ 8] - t[11] - t[12]); - q[15] = (t[12] - t[ 4] - t[ 6] - t[ 9] + t[13]); dump_x(q, 16, 'W'); for(i=0; i<16; ++i){ q[i] = s[i%5](q[i]); } } +#else +void bmw_small_f0(uint32_t* q, uint32_t* h, const void* m){ + uint8_t i; + uint32_t(*s[])(uint32_t)={ bmw_small_s0, bmw_small_s1, bmw_small_s2, + bmw_small_s3, bmw_small_s4 }; + for(i=0; i<16; ++i){ + h[i] ^= ((uint32_t*)m)[i]; + } + dump_x(h, 16, 'T'); + q[ 0] = (h[ 5] - h[ 7] + h[10] + h[13] + h[14]); + q[ 1] = (h[ 6] - h[ 8] + h[11] + h[14] - h[15]); + q[ 2] = (h[ 0] + h[ 7] + h[ 9] - h[12] + h[15]); + q[ 3] = (h[ 0] - h[ 1] + h[ 8] - h[10] + h[13]); + q[ 4] = (h[ 1] + h[ 2] + h[ 9] - h[11] - h[14]); + q[ 5] = (h[ 3] - h[ 2] + h[10] - h[12] + h[15]); + q[ 6] = (h[ 4] - h[ 0] - h[ 3] - h[11] + h[13]); + q[ 7] = (h[ 1] - h[ 4] - h[ 5] - h[12] - h[14]); + q[ 8] = (h[ 2] - h[ 5] - h[ 6] + h[13] - h[15]); + q[ 9] = (h[ 0] - h[ 3] + h[ 6] - h[ 7] + h[14]); + q[10] = (h[ 8] - h[ 1] - h[ 4] - h[ 7] + h[15]); + q[11] = (h[ 8] - h[ 0] - h[ 2] - h[ 5] + h[ 9]); + q[12] = (h[ 1] + h[ 3] - h[ 6] - h[ 9] + h[10]); + q[13] = (h[ 2] + h[ 4] + h[ 7] + h[10] + h[11]); + q[14] = (h[ 3] - h[ 5] + h[ 8] - h[11] - h[12]); + q[15] = (h[12] - h[ 4] - h[ 6] - h[ 9] + h[13]); + dump_x(q, 16, 'W'); + for(i=0; i<16; ++i){ + q[i] = s[i%5](q[i]); + } +} +#endif + void bmw_small_f1(uint32_t* q, const void* m){ uint8_t i; q[16] = bmw_small_expand1(0, q, m);