phys_size_t initdram (int board_type)
{
volatile immap_t *immap = (immap_t *) CONFIG_SYS_IMMR;
volatile memctl8260_t *memctl = &immap->im_memctl;
#ifndef CONFIG_SYS_RAMBOOT
ulong size8, size9;
#endif
ulong psize = 32 * 1024 * 1024;
memctl->memc_psrt = CONFIG_SYS_PSRT;
memctl->memc_mptpr = CONFIG_SYS_MPTPR;
#ifndef CONFIG_SYS_RAMBOOT
size8 = try_init (memctl, CONFIG_SYS_PSDMR_8COL, CONFIG_SYS_OR2_8COL,
(uchar *) CONFIG_SYS_SDRAM_BASE);
size9 = try_init (memctl, CONFIG_SYS_PSDMR_9COL, CONFIG_SYS_OR2_9COL,
(uchar *) CONFIG_SYS_SDRAM_BASE);
if (size8 < size9) {
psize = size9;
printf ("(60x:9COL) ");
} else {
psize = try_init (memctl, CONFIG_SYS_PSDMR_8COL, CONFIG_SYS_OR2_8COL,
(uchar *) CONFIG_SYS_SDRAM_BASE);
printf ("(60x:8COL) ");
}
#endif
return (psize);
}
phys_size_t initdram (int board_type)
{
volatile immap_t *immap = (immap_t *) CONFIG_SYS_IMMR;
volatile memctl8260_t *memctl = &immap->im_memctl;
long psize;
#ifndef CONFIG_SYS_RAMBOOT
long sizelittle, sizebig;
#endif
memctl->memc_psrt = CONFIG_SYS_PSRT;
memctl->memc_mptpr = CONFIG_SYS_MPTPR;
#ifndef CONFIG_SYS_RAMBOOT
/* 60x SDRAM setup:
*/
sizelittle = try_init (memctl, CONFIG_SYS_PSDMR_LITTLE, CONFIG_SYS_OR1_LITTLE,
(uchar *) CONFIG_SYS_SDRAM_BASE);
sizebig = try_init (memctl, CONFIG_SYS_PSDMR_BIG, CONFIG_SYS_OR1_BIG,
(uchar *) CONFIG_SYS_SDRAM_BASE);
if (sizelittle < sizebig) {
psize = sizebig;
} else {
psize = try_init (memctl, CONFIG_SYS_PSDMR_LITTLE, CONFIG_SYS_OR1_LITTLE,
(uchar *) CONFIG_SYS_SDRAM_BASE);
}
#endif /* CONFIG_SYS_RAMBOOT */
icache_enable ();
return (psize);
}
static int do_init(FILE *f)
{
assert(con_mode == CON_INIT);
try_init(&console_none,f);
assert(con != me);
assert(con == &console_none);
assert(con_mode == CON_NONE);
if (opt_console == CON_NONE)
return con_mode;
if (!acc_isatty(STDIN_FILENO) || !acc_isatty(STDOUT_FILENO) || !acc_isatty(STDERR_FILENO))
return con_mode;
#if defined(USE_ANSI)
try_init(&console_ansi_mono,f);
try_init(&console_ansi_color,f);
#endif
#if defined(USE_SCREEN)
try_init(&console_screen,f);
#endif
#if defined(USE_AALIB)
try_init(&console_aalib,f);
#endif
return con_mode;
}
void swanLoadProgramFromSource( const char *module, const unsigned char *ptx, size_t len , int devtype ) {
int i=0;
CUresult err;
try_init();
// let's see whether this module is already loaded
for( i=0; i < state.num_mods; i++ ) {
if( !strcmp( state.mod_names[i], module ) ) {
return; // already loaded
}
}
if( ptx == NULL || len == 0 ) {
fprintf ( stderr, "SWAN : Module load failure [%s]. No source \n", module );
error( "Module source invalid" );
}
i = state.num_mods;
state.num_mods++;
state.mods = (CUmodule*) realloc( state.mods, state.num_mods * sizeof(CUmodule) );
state.mod_names = (char**) realloc( state.mod_names, state.num_mods * sizeof(char*) );
state.mod_names[i] = (char*) malloc( strlen( module ) + 1 );
strcpy( state.mod_names[i], module );
// now load the PTX into a module
err = cuModuleLoadData( &state.mods[i], ptx );
if( err != CUDA_SUCCESS ) {
fprintf ( stderr, "SWAN : Module load result [%s] [%d]\n", module, err );
error( "Module load failed\n" );
}
}
void *swanMalloc( size_t len ) {
void *ptr;
CUdeviceptr dptr;
CUresult err;
try_init();
if( len == 0 ) {
// printf("SWAN: WARNING - swnaMalloc() called with 0\n");
return NULL;
}
err = cuMemAlloc( (CUdeviceptr*) &dptr, len );
ptr = (void*)dptr;
if ( err != CUDA_SUCCESS ) {
printf("Attempted to allocate %lu bytes (%lu already allocated)\n", len, state.bytes_allocated );
abort();
error("swanMalloc failed\n" );
}
state.bytes_allocated += len;
// MJH likes his memory clean
swanMemset( ptr, 0, len );
return ptr;
}
size_t swanMemAvailable( void ) {
size_t free, total;
try_init();
cuMemGetInfo( &free, &total );
return free;
}
long int initdram (int board_type)
{
volatile immap_t *immap = (immap_t *) CFG_IMMR;
volatile memctl8260_t *memctl = &immap->im_memctl;
#ifndef CFG_RAMBOOT
ulong size8, size9, size10;
#endif
long psize;
psize = 32 * 1024 * 1024;
memctl->memc_mptpr = CFG_MPTPR;
memctl->memc_psrt = CFG_PSRT;
#ifndef CFG_RAMBOOT
/* 60x SDRAM setup:
*/
size8 = try_init (memctl, CFG_PSDMR_8COL, CFG_OR2_8COL,
(uchar *) CFG_SDRAM_BASE);
size9 = try_init (memctl, CFG_PSDMR_9COL, CFG_OR2_9COL,
(uchar *) CFG_SDRAM_BASE);
size10 = try_init (memctl, CFG_PSDMR_10COL, CFG_OR2_10COL,
(uchar *) CFG_SDRAM_BASE);
psize = max(size8,max(size9,size10));
if (psize == size8) {
psize = try_init (memctl, CFG_PSDMR_8COL, CFG_OR2_8COL,
(uchar *) CFG_SDRAM_BASE);
printf ("(60x:8COL) ");
} else if (psize == size9){
psize = try_init (memctl, CFG_PSDMR_9COL, CFG_OR2_9COL,
(uchar *) CFG_SDRAM_BASE);
printf ("(60x:9COL) ");
} else
printf ("(60x:10COL) ");
#endif /* CFG_RAMBOOT */
icache_enable ();
return (psize);
}
modland_data_t *modland_lookup(const char *md5) {
try_init();
if (!initialized) {
return NULL;
}
return modland_internal_lookup(md5);
}
static long probe_sdram(memctl8260_t *memctl)
{
int n = 0;
long psize = 0;
for (n = 0; n < ARRAY_SIZE(sdram_conf); psize = 0, n++) {
psize = try_init(memctl,
CONFIG_SYS_PSDMR | sdram_conf[n].psdmr,
CONFIG_SYS_OR1 | sdram_conf[n].or1,
(uchar *) CONFIG_SYS_SDRAM_BASE);
debug("Probing %ld bytes returned %ld\n",
sdram_conf[n].size, psize);
if (psize == sdram_conf[n].size)
break;
}
return psize;
}
void *swanMallocHost( size_t len ) {
CUresult err;
void *ptr;
try_init();
#ifdef DEV_EXTENSIONS
err= cuMemHostAlloc( &ptr, len, CU_MEMHOSTALLOC_PORTABLE ); //| CU_MEMHOSTALLOC_DEVICEMAP ); //| CU_MEMHOSTALLOC_WRITECOMBINED );
#else
err = cuMemAllocHost( &ptr, len );
#endif
if ( err != CUDA_SUCCESS ) {
fprintf( stderr, "swanMallocHost error: %d\n", err );
error("swanMallocHost failed\n" );
}
// printf("MallocHost %p\n", ptr );
memset( ptr, 0, len );
return ptr;
}
phys_size_t initdram (int board_type)
{
volatile immap_t *immap = (immap_t *) CONFIG_SYS_IMMR;
volatile memctl8260_t *memctl = &immap->im_memctl;
long psize;
memctl->memc_psrt = CONFIG_SYS_PSRT;
memctl->memc_mptpr = CONFIG_SYS_MPTPR;
#ifndef CONFIG_SYS_RAMBOOT
/* 60x SDRAM setup:
*/
psize = try_init (memctl, CONFIG_SYS_PSDMR, CONFIG_SYS_OR1,
(uchar *) CONFIG_SYS_SDRAM_BASE);
#endif /* CONFIG_SYS_RAMBOOT */
icache_enable ();
return (psize);
}
static long probe_sdram(memctl8260_t *memctl)
{
return try_init(memctl, CONFIG_SYS_PSDMR, CONFIG_SYS_OR1,
(uchar *) CONFIG_SYS_SDRAM_BASE);
}
void swanInit(void) {
try_init();
}
void swanRunKernelAsync( const char *kernel, block_config_t grid , block_config_t block, size_t shmem, int flags, void *ptrs[], int *types ) {
// find the kernel
if( !grid.x || !grid.y || !grid.z || !block.x || !block.y || !block.z ) { return; } // suppress launch of kernel if any of the launch dims are 0
CUfunction f = NULL;
int i;
int offset = 0;
CUresult err;
int type;
int idx=0;
try_init();
for( i=0; i < state.num_funcs; i++ ) {
if( !strcmp( state.func_names[i], kernel ) ) {
f = state.funcs[i];
break;
}
}
if( f == NULL ) {
for( i=0; i < state.num_mods; i++ ) {
cuModuleGetFunction( &f, state.mods[i], kernel );
if( f!= NULL ) {
// found a kernel. store it for future use
int j = state.num_funcs;
state.num_funcs++;
state.funcs = (CUfunction*) realloc( state.funcs, sizeof(CUfunction) * state.num_funcs );
state.funcs[j] = f;
state.func_names = (char**) realloc( state.func_names, sizeof(char*) * state.num_funcs );
state.func_names[j] = (char*) malloc( strlen(kernel) + 1 );
strcpy( state.func_names[j], kernel );
break;
}
}
}
if( f== NULL ) {
fprintf(stderr, "Error running kernel [%s] : \n", kernel );
error( "No kernel found" );
}
if( grid.z != 1 ) {
printf("Kernel [%s] launched with (%d %d %d)(%d %d %d)\n", kernel, grid.x, grid.y, grid.z, block.x, block.y, block.z );
error( "grid.z needs to be 1" );
}
//printf("Running kernel [%s]\n", kernel );
type = types[idx];
while( type != SWAN_END ) {
void *ptr = ptrs[idx];
switch( type ) {
// DEBLOCK( SWAN_uchar, uchar, 1 );
DEBLOCK( SWAN_uchar2, uchar2, 2 );
DEBLOCK( SWAN_uchar3, uchar3, 1 );
DEBLOCK( SWAN_uchar4, uchar4, 4 );
DEBLOCK( SWAN_char , int, 1 );
// DEBLOCK( SWAN_char1 , char1, 1 );
DEBLOCK( SWAN_char2 , char2, 2 );
DEBLOCK( SWAN_char3 , char3, 1 );
DEBLOCK( SWAN_char4 , char4, 4 );
DEBLOCK( SWAN_int, int, 4 );
// DEBLOCK( SWAN_int1, int1, 4 );
DEBLOCK( SWAN_int2, int2, 8 );
DEBLOCK( SWAN_int3, int3, 4 );
DEBLOCK( SWAN_int4, int4, 16 );
// DEBLOCK( SWAN_float, double, 4 );
// DEBLOCK( SWAN_float1, float1, 4 );
DEBLOCK( SWAN_float2, float2, 8 );
DEBLOCK( SWAN_float3, float3, 4 );
DEBLOCK( SWAN_float4, float4, 16 );
DEBLOCK( SWAN_uint, uint, 4 );
DEBLOCK( SWAN_uint2, uint2, 8 );
DEBLOCK( SWAN_uint3, uint3, 4 );
DEBLOCK( SWAN_uint4, uint4, 16 );
DEBLOCK( SWAN_float, float, 4 );
//#define DEBLOCK(swan_type,type,OFFSET)
#if ( CUDA_MAJOR == 3 && CUDA_MINOR >= 2 ) || CUDA_MAJOR >= 4
case SWAN_PTR:
{
//printf("PTR as NATIVE\n");
ALIGN_UP( offset, (sizeof(void*)));
cuParamSetv( f, offset, ptr, sizeof(void*) );
offset += sizeof(void*); }
break;
#else
case SWAN_PTR:
{
//printf("PTR as INT\n");
ALIGN_UP( offset, (sizeof(int)));
cuParamSetv( f, offset, ptr, sizeof(int) );
offset += sizeof(int); }
break;
#endif
default:
printf("%d\n", type );
error("Parameter type not handled\n");
}
idx++;
type = types[idx];
}
//printf("Launching kernel [%s] [%X] with (%d %d %d) (%d %d %d)\n", kernel, f, grid.x, grid.y, grid.z, block.x, block.y, block.z );
//printf(" TOTAL OFFSET %d\n", offset );
CU_SAFE_CALL_NO_SYNC( cuParamSetSize( f, offset ) );
CU_SAFE_CALL_NO_SYNC( cuFuncSetBlockShape( f, block.x, block.y, block.z ) );
CU_SAFE_CALL_NO_SYNC( cuFuncSetSharedSize( f, shmem ) );
#if (CUDA_MAJOR ==3 && CUDA_MINOR >=1 ) || CUDA_MAJOR>=4
cuFuncSetCacheConfig( f, CU_FUNC_CACHE_PREFER_SHARED ); // This seems to be better in every case for acemd
#endif
err = cuLaunchGridAsync( f, grid.x, grid.y, NULL ) ; //state.stream ) ;
if( err != CUDA_SUCCESS ) {
fprintf( stderr , "SWAN : FATAL : Failure executing kernel [%s] [%d] [%d,%d,%d][%d,%d,%d]\n", kernel, err, grid.x ,grid.y, grid.z, block.x, block.y, block.z );
assert(0);
exit(-99);
}
//printf("Kernel completed\n" );
}
int swanDeviceVersion( void ) {
try_init();
return state.device_version;
}
int swanGetNumberOfComputeElements( void ) {
try_init();
// struct cudaDeviceProp prop;
// cudaGetDeviceProperties( &prop, state.device );
return state.multiProcessorCount;
}
void SIM_init(){
sky_pref_t* pref;
char* welcome_str = get_front_message();
/*
* get the corrent_config_file and do some initialization
*/
skyeye_config_t* config = get_current_config();
skyeye_option_init(config);
/*
* initialization of callback data structure, it needs to
* be initialized at very beginning.
*/
init_callback();
/*
* initilize the data structure for command
* register some default built-in command
*/
init_command_list();
init_stepi();
/*
* initialization of module manangement
*/
init_module_list();
/*
* initialization of architecture and cores
*/
init_arch();
/*
* initialization of bus and memory module
*/
init_bus();
/*
* initialization of machine module
*/
init_mach();
/*
* initialization of breakpoint, that depends on callback module.
*/
init_bp();
/*
* get the current preference for simulator
*/
pref = get_skyeye_pref();
/*
* loading all the modules in search directory
*/
if(!pref->module_search_dir)
pref->module_search_dir = skyeye_strdup(default_lib_dir);
SKY_load_all_modules(pref->module_search_dir, NULL);
//skyeye_config_t *config;
//config = malloc(sizeof(skyeye_config_t));
if(try_init() == No_exp){
if(pref->autoboot == True){
SIM_run();
}
}
/*
* if we run simulator in GUI or external IDE, we do not need to
* launch our CLI.
*/
if(pref->interactive_mode == True){
SIM_cli();
}
}