From 97034edb35e92361daaa24512989d00f6c3fd517 Mon Sep 17 00:00:00 2001 From: Paul Selkirk Date: Mon, 26 Nov 2018 17:26:55 -0500 Subject: Add loop unrolling to bring the profilable mem* functions closer to newlib, because memset is called a LOT in the course of RSA signing, and we need to understand how much time we're actually spending there. --- libraries/libprof/Makefile | 6 ++- libraries/libprof/memfunc.c | 127 ++++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 132 insertions(+), 1 deletion(-) create mode 100644 libraries/libprof/memfunc.c (limited to 'libraries/libprof') diff --git a/libraries/libprof/Makefile b/libraries/libprof/Makefile index 4fe5fb4..37b9a23 100644 --- a/libraries/libprof/Makefile +++ b/libraries/libprof/Makefile @@ -1,12 +1,16 @@ LIB = libprof.a -OBJS = gmon.o profil.o profiler.o +OBJS = gmon.o profil.o profiler.o memfunc.o # Don't profile the profiling code, because that way lies madness (and recursion). CFLAGS := $(subst -pg,,$(CFLAGS)) all: $(LIB) +# But do profile the mem functions +memfunc.o: memfunc.c + $(CC) $(CFLAGS) -pg -c -o $@ $< + %.o : %.c $(CC) $(CFLAGS) -c -o $@ $< diff --git a/libraries/libprof/memfunc.c b/libraries/libprof/memfunc.c new file mode 100644 index 0000000..fc908e1 --- /dev/null +++ b/libraries/libprof/memfunc.c @@ -0,0 +1,127 @@ +#include +#include + +/* + * Profilable substitutes for mem*(), lacking libc_p.a + * + * This code was written with reference to newlib, and was recently + * brought closer into line with newlib, to make profiling more accurate. + * + * Newlib is maintained by Cygwin, which is Red Hat. There is no copyright + * statement in the corresponding newlib source files, nor is there a + * COPYING file in newlib/libc/string or newlib/libc. Consider this file + * to be covered under one or more of the 50 copyright notices in + * newlib/COPYING, most of which are BSD. In any case, this file is only + * used for profiling, and is not used in production builds. + */ + +#define is_word_aligned(x) (((size_t)(x) & 3) == 0) + +void *memcpy(void *dst, const void *src, size_t n) +{ + uint8_t *d8 = (uint8_t *)dst; + uint8_t *s8 = (uint8_t *)src; + + if (n >= sizeof(uint32_t) && is_word_aligned(src) && is_word_aligned(dst)) { + uint32_t *d32 = (uint32_t *)dst; + uint32_t *s32 = (uint32_t *)src; + while (n >= 4 * sizeof(uint32_t)) { + *d32++ = *s32++; + *d32++ = *s32++; + *d32++ = *s32++; + *d32++ = *s32++; + n -= 4 * sizeof(uint32_t); + } + while (n >= sizeof(uint32_t)) { + *d32++ = *s32++; + n -= sizeof(uint32_t); + } + d8 = (uint8_t *)d32; + s8 = (uint8_t *)s32; + } + while (n-- > 0) { + *d8++ = *s8++; + } + + return dst; +} + +void *memset(void *dst, int c, size_t n) +{ + uint8_t *d8 = (uint8_t *)dst; + uint8_t c8 = (uint8_t)c; + + while (!is_word_aligned(d8)) { + if (n--) + *d8++ = c8; + else + return dst; + } + if (n >= sizeof(uint32_t)) { + uint32_t *d32 = (uint32_t *)d8; + uint32_t c32 = (c8 << 24) | (c8 << 16) | (c8 << 8) | (c8); + while (n >= 4 * sizeof(uint32_t)) { + *d32++ = c32; + *d32++ = c32; + *d32++ = c32; + *d32++ = c32; + n -= 4 * sizeof(uint32_t); + } + while (n >= sizeof(uint32_t)) { + *d32++ = c32; + n -= sizeof(uint32_t); + } + d8 = (uint8_t *)d32; + } + while (n-- > 0) { + *d8++ = c8; + } + + return dst; +} + +int memcmp(const void *dst, const void *src, size_t n) +{ + uint8_t *d8 = (uint8_t *)dst; + uint8_t *s8 = (uint8_t *)src; + + if (n >= sizeof(uint32_t) && is_word_aligned(src) && is_word_aligned(dst)) { + uint32_t *d32 = (uint32_t *)dst; + uint32_t *s32 = (uint32_t *)src; + while (n >= sizeof(uint32_t)) { + if (*d32 != *s32) + break; + d32++; + s32++; + n -= sizeof(uint32_t); + } + d8 = (uint8_t *)d32; + s8 = (uint8_t *)s32; + } + while (n-- > 0) { + if (*d8 != *s8) + return (*d8 - *s8); + d8++; + s8++; + } + + return 0; +} + +void *memmove(void *dst, const void *src, size_t n) +{ + uint8_t *d8 = (uint8_t *)dst; + uint8_t *s8 = (uint8_t *)src; + + if ((s8 < d8) && (d8 < s8 + n)) { + /* Destructive overlap...have to copy backwards */ + s8 += n; + d8 += n; + while (n-- > 0) { + *--d8 = *--s8; + } + return dst; + } + + return memcpy(dst, src, n); +} -- cgit v1.2.3 From e99737c12cac1fcc8604ac89a14dac5b2606a42d Mon Sep 17 00:00:00 2001 From: Paul Selkirk Date: Sun, 2 Dec 2018 17:19:44 -0500 Subject: Clean up the profiling code to the point where I stand a chance of understanding it 6 months from now. While I'm at it, try to make it a little more efficient (because 50-60% of time in a typical profiling run is spent in the function-entry counting), and collapse profil.c into gmon.c. --- libraries/libprof/Makefile | 2 +- libraries/libprof/gmon.c | 427 +++++++++++++++++++-------------------------- libraries/libprof/gmon.h | 57 ++---- libraries/libprof/profil.c | 96 ---------- libraries/libprof/profil.h | 60 ------- 5 files changed, 197 insertions(+), 445 deletions(-) delete mode 100644 libraries/libprof/profil.c delete mode 100644 libraries/libprof/profil.h (limited to 'libraries/libprof') diff --git a/libraries/libprof/Makefile b/libraries/libprof/Makefile index 37b9a23..28bedea 100644 --- a/libraries/libprof/Makefile +++ b/libraries/libprof/Makefile @@ -1,6 +1,6 @@ LIB = libprof.a -OBJS = gmon.o profil.o profiler.o memfunc.o +OBJS = gmon.o profiler.o memfunc.o # Don't profile the profiling code, because that way lies madness (and recursion). CFLAGS := $(subst -pg,,$(CFLAGS)) diff --git a/libraries/libprof/gmon.c b/libraries/libprof/gmon.c index 317a173..a34f1a2 100644 --- a/libraries/libprof/gmon.c +++ b/libraries/libprof/gmon.c @@ -36,289 +36,226 @@ #include #include #include -#include -#include "gmon.h" -#include "profil.h" #include +#include "gmon.h" #define bzero(ptr,size) memset (ptr, 0, size); #define ERR(s) write(2, s, sizeof(s)) -struct gmonparam _gmonparam = { GMON_PROF_OFF, NULL, 0, NULL, 0, NULL, 0, 0L, 0, 0, 0}; -static char already_setup = 0; /* flag to indicate if we need to init */ -static int s_scale; -/* see profil(2) where this is described (incorrectly) */ -#define SCALE_1_TO_1 0x10000L +/* profiling frequency. (No larger than 1000) */ +/* Note this doesn't set the frequency, but merely describes it. */ +#define PROF_HZ 1000 -static void moncontrol(int mode); +struct gmonparam _gmonparam = { off, NULL, 0, NULL, 0, NULL, 0, 0, 0, 0, 0}; -void monstartup (size_t lowpc, size_t highpc) { - register size_t o; - char *cp; - struct gmonparam *p = &_gmonparam; +void monstartup(size_t lowpc, size_t highpc) +{ + static char already_setup = 0; + struct gmonparam *p = &_gmonparam; - if (already_setup) { - /* zero out cp as value will be added there */ - bzero(p->tos, p->kcountsize + p->fromssize + p->tossize); - moncontrol(1); /* start */ - return; - } - already_setup = 1; + if (already_setup) { + /* reinitialize counters and arcs */ + bzero(p->kcount, p->kcountsize); + bzero(p->froms, p->fromssize); + bzero(p->tos, p->tossize); + p->state = on; + return; + } + already_setup = 1; - /* enable semihosting, for eventual output */ - extern void initialise_monitor_handles(void); - initialise_monitor_handles(); + /* enable semihosting, for eventual output */ + extern void initialise_monitor_handles(void); + initialise_monitor_handles(); - /* - * round lowpc and highpc to multiples of the density we're using - * so the rest of the scaling (here and in gprof) stays in ints. - */ - p->lowpc = ROUNDDOWN(lowpc, HISTFRACTION * sizeof(HISTCOUNTER)); - p->highpc = ROUNDUP(highpc, HISTFRACTION * sizeof(HISTCOUNTER)); - p->textsize = p->highpc - p->lowpc + 0x20; - p->kcountsize = p->textsize / HISTFRACTION; - p->fromssize = p->textsize / HASHFRACTION; - p->tolimit = p->textsize * ARCDENSITY / 100; - if (p->tolimit < MINARCS) { - p->tolimit = MINARCS; - } else if (p->tolimit > MAXARCS) { - p->tolimit = MAXARCS; - } - p->tossize = p->tolimit * sizeof(struct tostruct); + /* + * round lowpc and highpc to multiples of the density we're using + * so the rest of the scaling (here and in gprof) stays in ints. + */ + p->lowpc = ROUNDDOWN(lowpc, HISTFRACTION * sizeof(HISTCOUNTER)); + p->highpc = ROUNDUP(highpc, HISTFRACTION * sizeof(HISTCOUNTER)); + p->textsize = p->highpc - p->lowpc; + p->kcountsize = p->textsize / HISTFRACTION; + p->fromssize = p->textsize / HASHFRACTION; + p->tolimit = p->textsize * ARCDENSITY / 100; + if (p->tolimit < MINARCS) { + p->tolimit = MINARCS; + } else if (p->tolimit > MAXARCS) { + p->tolimit = MAXARCS; + } + p->tossize = p->tolimit * sizeof(struct tostruct); - extern void *hal_allocate_static_memory(const size_t size); - cp = hal_allocate_static_memory(p->kcountsize + p->fromssize + p->tossize); - if (cp == NULL) { - ERR("monstartup: out of memory\n"); - return; - } + extern void *hal_allocate_static_memory(const size_t size); + void *cp = hal_allocate_static_memory(p->kcountsize + p->fromssize + p->tossize); + if (cp == NULL) { + ERR("monstartup: out of memory\n"); + return; + } - /* zero out cp as value will be added there */ - bzero(cp, p->kcountsize + p->fromssize + p->tossize); + bzero(cp, p->kcountsize + p->fromssize + p->tossize); + p->kcount = (unsigned short *)cp; cp += p->kcountsize; + p->froms = (unsigned short *)cp; cp += p->fromssize; + p->tos = (struct tostruct *)cp; - p->tos = (struct tostruct *)cp; - cp += p->tossize; - p->kcount = (unsigned short *)cp; - cp += p->kcountsize; - p->froms = (unsigned short *)cp; + p->state = on; +} - p->tos[0].link = 0; +void _mcleanup(void) +{ + static const char gmon_out[] = "gmon.out"; + int fd; + struct gmonparam *p = &_gmonparam; - o = p->highpc - p->lowpc; - if (p->kcountsize < o) { -#ifndef notdef - s_scale = ((float)p->kcountsize / o ) * SCALE_1_TO_1; -#else /* avoid floating point */ - int quot = o / p->kcountsize; + if (p->state == err) { + ERR("_mcleanup: tos overflow\n"); + } - if (quot >= 0x10000) - s_scale = 1; - else if (quot >= 0x100) - s_scale = 0x10000 / quot; - else if (o >= 0x800000) - s_scale = 0x1000000 / (o / (p->kcountsize >> 8)); - else - s_scale = 0x1000000 / ((o << 8) / p->kcountsize); -#endif - } else { - s_scale = SCALE_1_TO_1; - } - moncontrol(1); /* start */ -} + fd = open(gmon_out , O_CREAT|O_TRUNC|O_WRONLY|O_BINARY, 0666); + if (fd < 0) { + perror( gmon_out ); + return; + } -void _mcleanup(void) { - static const char gmon_out[] = "gmon.out"; - int fd; - int fromindex; - int endfrom; - size_t frompc; - int toindex; - struct rawarc rawarc; - struct gmonparam *p = &_gmonparam; - struct gmonhdr gmonhdr = {0}, *hdr; - const char *proffile; -#ifdef DEBUG - int log, len; - char dbuf[200]; -#endif + struct gmonhdr hdr = { + .lpc = p->lowpc, + .hpc = p->highpc, + .ncnt = p->kcountsize + sizeof(struct gmonhdr), + .version = GMONVERSION, + .profrate = PROF_HZ + }; + write(fd, &hdr, sizeof(hdr)); - if (p->state == GMON_PROF_ERROR) { - ERR("_mcleanup: tos overflow\n"); - } - moncontrol(0); /* stop */ - proffile = gmon_out; - fd = open(proffile , O_CREAT|O_TRUNC|O_WRONLY|O_BINARY, 0666); - if (fd < 0) { - perror( proffile ); - return; - } -#ifdef DEBUG - log = open("gmon.log", O_CREAT|O_TRUNC|O_WRONLY, 0664); - if (log < 0) { - perror("mcount: gmon.log"); - return; - } - len = sprintf(dbuf, "[mcleanup1] kcount 0x%x ssiz %d\n", - (unsigned int)p->kcount, p->kcountsize); - write(log, dbuf, len); -#endif - hdr = (struct gmonhdr *)&gmonhdr; - hdr->lpc = p->lowpc; - hdr->hpc = p->highpc; - hdr->ncnt = p->kcountsize + sizeof(gmonhdr); - hdr->version = GMONVERSION; - hdr->profrate = PROF_HZ; - hdr->spare[0] = hdr->spare[1] = hdr->spare[2] = 0; - write(fd, (char *)hdr, sizeof *hdr); - write(fd, p->kcount, p->kcountsize); - endfrom = p->fromssize / sizeof(*p->froms); - for (fromindex = 0; fromindex < endfrom; fromindex++) { - if (p->froms[fromindex] == 0) { - continue; - } - frompc = p->lowpc; - frompc += fromindex * HASHFRACTION * sizeof(*p->froms); - for (toindex = p->froms[fromindex]; toindex != 0; toindex = p->tos[toindex].link) { -#ifdef DEBUG - len = sprintf(dbuf, - "[mcleanup2] frompc 0x%x selfpc 0x%x count %ld\n" , - frompc, p->tos[toindex].selfpc, - p->tos[toindex].count); - write(log, dbuf, len); -#endif - rawarc.raw_frompc = frompc; - rawarc.raw_selfpc = p->tos[toindex].selfpc; - rawarc.raw_count = p->tos[toindex].count; - write(fd, &rawarc, sizeof rawarc); - } - } - close(fd); -} + write(fd, p->kcount, p->kcountsize); -/* - * Control profiling - * profiling is what mcount checks to see if - * all the data structures are ready. - */ -static void moncontrol(int mode) { - struct gmonparam *p = &_gmonparam; + for (size_t fromindex = 0; fromindex < p->fromssize / sizeof(*p->froms); fromindex++) { + size_t frompc = p->lowpc + fromindex * HASHFRACTION * sizeof(*p->froms); + for (size_t toindex = p->froms[fromindex]; toindex != 0; toindex = p->tos[toindex].next) { + struct rawarc arc = { + .frompc = frompc, + .selfpc = p->tos[toindex].selfpc, + .count = p->tos[toindex].count + }; + write(fd, &arc, sizeof(arc)); + } + } - if (mode) { - /* start */ - profil((char *)p->kcount, p->kcountsize, p->lowpc, s_scale); - p->state = GMON_PROF_ON; - } else { - /* stop */ - profil((char *)0, 0, 0, 0); - p->state = GMON_PROF_OFF; - } + close(fd); } -void _mcount_internal(uint32_t *frompcindex, uint32_t *selfpc) { - register struct tostruct *top; - register struct tostruct *prevtop; - register long toindex; +void _mcount_internal(size_t frompc, size_t selfpc) +{ + register unsigned short *fromptr; + register struct tostruct *top; + register unsigned short toindex; struct gmonparam *p = &_gmonparam; /* - * check that we are profiling - * and that we aren't recursively invoked. - */ - if (p->state!=GMON_PROF_ON) { - goto out; - } - p->state++; - /* - * check that frompcindex is a reasonable pc value. - * for example: signal catchers get called from the stack, - * not from text space. too bad. + * check that we are profiling and that we aren't recursively invoked. + * check that frompc is a reasonable pc value. */ - frompcindex = (uint32_t*)((long)frompcindex - (long)p->lowpc); - if ((unsigned long)frompcindex > p->textsize) { - goto done; + if (p->state != on || (frompc -= p->lowpc) > p->textsize) { + return; } - frompcindex = (uint32_t*)&p->froms[((long)frompcindex) / (HASHFRACTION * sizeof(*p->froms))]; - toindex = *((unsigned short*)frompcindex); /* get froms[] value */ - if (toindex == 0) { - /* - * first time traversing this arc - */ - toindex = ++p->tos[0].link; /* the link of tos[0] points to the last used record in the array */ + + fromptr = &p->froms[frompc / (HASHFRACTION * sizeof(*p->froms))]; + toindex = *fromptr; /* get froms[] value */ + + if (toindex == 0) { /* we haven't seen this caller before */ + toindex = ++p->tos[0].next; /* index of the last used record in the array */ if (toindex >= p->tolimit) { /* more tos[] entries than we can handle! */ - goto overflow; - } - *((unsigned short*)frompcindex) = (unsigned short)toindex; /* store new 'to' value into froms[] */ + overflow: + p->state = err; /* halt further profiling */ +#define TOLIMIT "mcount: tos overflow\n" + write (2, TOLIMIT, sizeof(TOLIMIT)); + return; + } + *fromptr = toindex; /* store new 'to' value into froms[] */ top = &p->tos[toindex]; - top->selfpc = (size_t)selfpc; + top->selfpc = selfpc; top->count = 1; - top->link = 0; - goto done; - } - top = &p->tos[toindex]; - if (top->selfpc == (size_t)selfpc) { - /* - * arc at front of chain; usual case. - */ - top->count++; - goto done; + top->next = 0; } - /* - * have to go looking down chain for it. - * top points to what we are looking at, - * prevtop points to previous top. - * we know it is not at the head of the chain. - */ - for (; /* goto done */; ) { - if (top->link == 0) { + + else { /* we've seen this caller before */ + top = &p->tos[toindex]; + if (top->selfpc == selfpc) { /* - * top is end of the chain and none of the chain - * had top->selfpc == selfpc. - * so we allocate a new tostruct - * and link it to the head of the chain. + * arc at front of chain; usual case. */ - toindex = ++p->tos[0].link; - if (toindex >= p->tolimit) { - goto overflow; - } - top = &p->tos[toindex]; - top->selfpc = (size_t)selfpc; - top->count = 1; - top->link = *((unsigned short*)frompcindex); - *(unsigned short*)frompcindex = (unsigned short)toindex; - goto done; + top->count++; } - /* - * otherwise, check the next arc on the chain. - */ - prevtop = top; - top = &p->tos[top->link]; - if (top->selfpc == (size_t)selfpc) { + + else { /* - * there it is. - * increment its count - * move it to the head of the chain. + * have to go looking down chain for it. + * top points to what we are looking at, + * prevtop points to previous top. + * we know it is not at the head of the chain. */ - top->count++; - toindex = prevtop->link; - prevtop->link = top->link; - top->link = *((unsigned short*)frompcindex); - *((unsigned short*)frompcindex) = (unsigned short)toindex; - goto done; + while (1) { + if (top->next == 0) { + /* + * top is end of the chain and none of the chain + * had top->selfpc == selfpc. + * so we allocate a new tostruct + * and put it at the head of the chain. + */ + toindex = ++p->tos[0].next; + if (toindex >= p->tolimit) { + goto overflow; + } + top = &p->tos[toindex]; + top->selfpc = selfpc; + top->count = 1; + top->next = *fromptr; + *fromptr = toindex; + break; + } + + else { + /* + * otherwise, check the next arc on the chain. + */ + register struct tostruct *prevtop = top; + top = &p->tos[top->next]; + if (top->selfpc == selfpc) { + /* + * there it is. + * increment its count + * move it to the head of the chain. + */ + top->count++; + toindex = prevtop->next; + prevtop->next = top->next; + top->next = *fromptr; + *fromptr = toindex; + break; + } + } + } } } - done: - p->state--; - /* and fall through */ - out: - return; /* normal return restores saved registers */ - overflow: - p->state++; /* halt further profiling */ - #define TOLIMIT "mcount: tos overflow\n" - write (2, TOLIMIT, sizeof(TOLIMIT)); - goto out; + + return; /* normal return restores saved registers */ } -void _monInit(void) { - _gmonparam.state = GMON_PROF_OFF; - already_setup = 0; +#include +#include "stm32f4xx_hal.h" /* __get_MSP */ + +/* called from the SysTick handler */ +void profil_callback(void) +{ + struct gmonparam *p = &_gmonparam; + + if (p->state == on) { + /* The interrupt mechanism pushes xPSR, PC, LR, R12, and R3-R0 onto the + * stack, so PC is the 6th word from the top at that point. However, the + * normal function entry code pushes registers as well, so the stack + * offset right now depends on the call tree that got us here. + */ + size_t pc = (size_t)((uint32_t *)__get_MSP())[6 + 6]; + if ((pc -= p->lowpc) < p->textsize) { + size_t idx = pc / (HISTFRACTION * sizeof(*p->kcount)); + p->kcount[idx]++; + } + } } diff --git a/libraries/libprof/gmon.h b/libraries/libprof/gmon.h index 8b5ecf0..9016502 100644 --- a/libraries/libprof/gmon.h +++ b/libraries/libprof/gmon.h @@ -40,28 +40,6 @@ #ifndef _SYS_GMON_H_ #define _SYS_GMON_H_ -#ifndef __P -#define __P(x) x -#endif - -/* On POSIX systems, profile.h is a KRB5 header. To avoid collisions, just - pull in profile.h's content here. The profile.h header won't be provided - by Mingw-w64 anymore at one point. */ -#if 0 -#include -#else -#ifndef _WIN64 -#define _MCOUNT_CALL __attribute__ ((regparm (2))) -extern void _mcount(void); -#else -#define _MCOUNT_CALL -extern void mcount(void); -#endif -#define _MCOUNT_DECL __attribute__((gnu_inline)) __inline__ \ - void _MCOUNT_CALL _mcount_private -#define MCOUNT -#endif - /* * Structure prepended to gmon.out profiling data file. */ @@ -83,7 +61,8 @@ struct gmonhdr { /* * fraction of text space to allocate for histogram counters here, 1/2 */ -#define HISTFRACTION 2 +//#define HISTFRACTION 2 +#define HISTFRACTION 1 /* * Fraction of text space to allocate for from hash buckets. @@ -113,7 +92,8 @@ struct gmonhdr { * profiling data structures without (in practice) sacrificing * any granularity. */ -#define HASHFRACTION 2 +//#define HASHFRACTION 2 +#define HASHFRACTION 1 /* * percent of text space to allocate for tostructs with a minimum. @@ -123,10 +103,10 @@ struct gmonhdr { #define MAXARCS ((1 << (8 * sizeof(HISTCOUNTER))) - 2) struct tostruct { - size_t selfpc; /* callee address/program counter. The caller address is in froms[] array which points to tos[] array */ - long count; /* how many times it has been called */ - unsigned short link; /* link to next entry in hash table. For tos[0] this points to the last used entry */ - unsigned short pad; /* additional padding bytes, to have entries 4byte aligned */ + size_t selfpc; /* callee address. The caller address is in froms[] array which points to tos[] array */ + unsigned long count; /* how many times it has been called */ + unsigned short next; /* next entry in hash table. For tos[0] this is the index of the last used entry */ + unsigned short pad; /* additional padding bytes, to have entries 4byte aligned */ }; /* @@ -134,9 +114,9 @@ struct tostruct { * the called site and a count. */ struct rawarc { - size_t raw_frompc; - size_t raw_selfpc; - long raw_count; + size_t frompc; + size_t selfpc; + long count; }; /* @@ -149,29 +129,20 @@ struct rawarc { * The profiling data structures are housed in this structure. */ struct gmonparam { - int state; + enum { off, on, err } state; unsigned short *kcount; /* histogram PC sample array */ size_t kcountsize; /* size of kcount[] array in bytes */ unsigned short *froms; /* array of hashed 'from' addresses. The 16bit value is an index into the tos[] array */ size_t fromssize; /* size of froms[] array in bytes */ - struct tostruct *tos; /* to struct, contains histogram counter */ + struct tostruct *tos; /* to struct, contains arc counters */ size_t tossize; /* size of tos[] array in bytes */ - long tolimit; + size_t tolimit; size_t lowpc; /* low program counter of area */ size_t highpc; /* high program counter */ size_t textsize; /* code size */ }; extern struct gmonparam _gmonparam; -/* - * Possible states of profiling. - */ -#define GMON_PROF_ON 0 -#define GMON_PROF_BUSY 1 -#define GMON_PROF_ERROR 2 -#define GMON_PROF_OFF 3 - void _mcleanup(void); /* routine to be called to write gmon.out file */ -void _monInit(void); /* initialization routine */ #endif /* !_SYS_GMONH_ */ diff --git a/libraries/libprof/profil.c b/libraries/libprof/profil.c deleted file mode 100644 index b0d8d55..0000000 --- a/libraries/libprof/profil.c +++ /dev/null @@ -1,96 +0,0 @@ -/* profil.c -- win32 profil.c equivalent - - Copyright 1998, 1999, 2000, 2001, 2002 Red Hat, Inc. - - This file is part of Cygwin. - - This software is a copyrighted work licensed under the terms of the - Cygwin license. Please consult the file "CYGWIN_LICENSE" for - details. */ - -/* - * This file is taken from Cygwin distribution, adapted to be used for bare embedded targets. - */ -#include -#include -#include -#include -#include "profil.h" -#include - -#include "stm32f4xx_hal.h" /* __get_MSP */ - -/* global profinfo for profil() call */ -static struct profinfo prof = { - PROFILE_NOT_INIT, 0, 0, 0, 0 -}; - -/* sample the current program counter */ -void profil_callback(void) { - if (prof.state == PROFILE_ON) { - /* The interrupt mechanism pushes xPSR, PC, LR, R12, and R3-R0 onto the - * stack, so PC is the 6th word from the top at that point. However, the - * normal function entry code pushes registers as well, so the stack - * offset right now depends on the call tree that got us here. - */ - size_t pc = (size_t)((uint32_t *)__get_MSP())[6 + 6]; - if (pc >= prof.lowpc && pc < prof.highpc) { - size_t idx = PROFIDX (pc, prof.lowpc, prof.scale); - prof.counter[idx]++; - } - } -} - -/* Stop profiling to the profiling buffer pointed to by p. */ -static int profile_off (struct profinfo *p) { - p->state = PROFILE_OFF; - return 0; -} - -/* Create a timer thread and pass it a pointer P to the profiling buffer. */ -static int profile_on (struct profinfo *p) { - p->state = PROFILE_ON; - return 0; /* ok */ -} - -/* - * start or stop profiling - * - * profiling goes into the SAMPLES buffer of size SIZE (which is treated - * as an array of unsigned shorts of size size/2) - * - * each bin represents a range of pc addresses from OFFSET. The number - * of pc addresses in a bin depends on SCALE. (A scale of 65536 maps - * each bin to two addresses, A scale of 32768 maps each bin to 4 addresses, - * a scale of 1 maps each bin to 128k address). Scale may be 1 - 65536, - * or zero to turn off profiling - */ -int profile_ctl (struct profinfo *p, char *samples, size_t size, size_t offset, unsigned int scale) { - size_t maxbin; - - if (scale > 65536) { - errno = EINVAL; - return -1; - } - profile_off(p); - if (scale) { - memset(samples, 0, size); - memset(p, 0, sizeof *p); - maxbin = size >> 1; - prof.counter = (unsigned short*)samples; - prof.lowpc = offset; - prof.highpc = PROFADDR(maxbin, offset, scale); - prof.scale = scale; - return profile_on(p); - } - return 0; -} - -/* Equivalent to unix profil() - Every SLEEPTIME interval, the user's program counter (PC) is examined: - offset is subtracted and the result is multiplied by scale. - The word pointed to by this address is incremented. */ -int profil (char *samples, size_t size, size_t offset, unsigned int scale) { - return profile_ctl (&prof, samples, size, offset, scale); -} - diff --git a/libraries/libprof/profil.h b/libraries/libprof/profil.h deleted file mode 100644 index c72dc00..0000000 --- a/libraries/libprof/profil.h +++ /dev/null @@ -1,60 +0,0 @@ -/* profil.h: gprof profiling header file - - Copyright 1998, 1999, 2000, 2001, 2002 Red Hat, Inc. - -This file is part of Cygwin. - -This software is a copyrighted work licensed under the terms of the -Cygwin license. Please consult the file "CYGWIN_LICENSE" for -details. */ - -/* - * This file is taken from Cygwin distribution. Please keep it in sync. - * The differences should be within __MINGW32__ guard. - */ - -#ifndef __PROFIL_H__ -#define __PROFIL_H__ - -/* profiling frequency. (No larger than 1000) */ -#define PROF_HZ 1000 - -/* convert an addr to an index */ -#define PROFIDX(pc, base, scale) \ - ({ \ - size_t i = (pc - base) / 2; \ - if (sizeof (unsigned long long int) > sizeof (size_t)) \ - i = (unsigned long long int) i * scale / 65536; \ - else \ - i = i / 65536 * scale + i % 65536 * scale / 65536; \ - i; \ - }) - -/* convert an index into an address */ -#define PROFADDR(idx, base, scale) \ - ((base) \ - + ((((unsigned long long)(idx) << 16) \ - / (unsigned long long)(scale)) << 1)) - -/* convert a bin size into a scale */ -#define PROFSCALE(range, bins) (((bins) << 16) / ((range) >> 1)) - -typedef void *_WINHANDLE; - -typedef enum { - PROFILE_NOT_INIT = 0, - PROFILE_ON, - PROFILE_OFF -} PROFILE_State; - -struct profinfo { - PROFILE_State state; /* profiling state */ - unsigned short *counter; /* profiling counters */ - size_t lowpc, highpc; /* range to be profiled */ - unsigned int scale; /* scale value of bins */ -}; - -int profile_ctl(struct profinfo *, char *, size_t, size_t, unsigned int); -int profil(char *, size_t, size_t, unsigned int); - -#endif /* __PROFIL_H__ */ -- cgit v1.2.3