rtems_task Floating_point_task_2(
rtems_task_argument argument
)
{
Thread_Control *executing;
FP_DECLARE;
context_switch_save_restore_idle_time = benchmark_timer_read();
executing = _Thread_Executing;
_Thread_Executing =
(Thread_Control *) _Thread_Ready_chain[FP1_PRIORITY].first;
FP_LOAD( 1.0 );
/* do not force context switch */
_Context_Switch_necessary = false;
_Thread_Disable_dispatch();
benchmark_timer_initialize();
#if (CPU_HARDWARE_FP == 1) || (CPU_SOFTWARE_FP == 1)
_Context_Save_fp( &executing->fp_context );
_Context_Restore_fp( &_Thread_Executing->fp_context );
#endif
_Context_Switch( &executing->Registers, &_Thread_Executing->Registers );
/* switch to Floating_point_task_1 */
context_switch_save_restore_initted_time = benchmark_timer_read();
complete_test();
}
rtems_task Floating_point_task_2(
rtems_task_argument argument
)
{
Scheduler_priority_Context *scheduler_context =
_Scheduler_priority_Get_context( _Scheduler_Get( _Thread_Get_executing() ) );
Thread_Control *executing;
FP_DECLARE;
context_switch_save_restore_idle_time = benchmark_timer_read();
executing = _Thread_Get_executing();
set_thread_executing(
(Thread_Control *) _Chain_First(&scheduler_context->Ready[FP1_PRIORITY])
);
FP_LOAD( 1.0 );
benchmark_timer_initialize();
#if (CPU_HARDWARE_FP == 1) || (CPU_SOFTWARE_FP == 1)
_Context_Save_fp( &executing->fp_context );
_Context_Restore_fp( &_Thread_Get_executing()->fp_context );
#endif
_Context_Switch(
&executing->Registers,
&_Thread_Get_executing()->Registers
);
/* switch to Floating_point_task_1 */
context_switch_save_restore_initted_time = benchmark_timer_read();
complete_test();
}
rtems_task Floating_point_task_1(
rtems_task_argument argument
)
{
Chain_Control *ready_queues;
Thread_Control *executing;
FP_DECLARE;
context_switch_restore_1st_fp_time = benchmark_timer_read();
executing = _Thread_Executing;
ready_queues = (Chain_Control *) _Scheduler.information;
_Thread_Executing =
(Thread_Control *) _Chain_First(&ready_queues[FP2_PRIORITY]);
/* do not force context switch */
_Thread_Dispatch_necessary = false;
_Thread_Disable_dispatch();
benchmark_timer_initialize();
#if (CPU_HARDWARE_FP == 1) || (CPU_SOFTWARE_FP == 1)
_Context_Save_fp( &executing->fp_context );
_Context_Restore_fp( &_Thread_Executing->fp_context );
#endif
_Context_Switch( &executing->Registers, &_Thread_Executing->Registers );
/* switch to Floating_point_task_2 */
context_switch_save_idle_restore_initted_time = benchmark_timer_read();
FP_LOAD( 1.0 );
executing = _Thread_Executing;
ready_queues = (Chain_Control *) _Scheduler.information;
_Thread_Executing =
(Thread_Control *) _Chain_First(&ready_queues[FP2_PRIORITY]);
/* do not force context switch */
_Thread_Dispatch_necessary = false;
_Thread_Disable_dispatch();
benchmark_timer_initialize();
#if (CPU_HARDWARE_FP == 1) || (CPU_SOFTWARE_FP == 1)
_Context_Save_fp( &executing->fp_context );
_Context_Restore_fp( &_Thread_Executing->fp_context );
#endif
_Context_Switch( &executing->Registers, &_Thread_Executing->Registers );
/* switch to Floating_point_task_2 */
}
static inline void qpx_data_fn_strided_c0 (void * target,
void * src,
size_t bytes,
void * cookie,
size_t c0)
{
if (bytes < 512) {
data_fn_strided(target, src, bytes, cookie);
return;
}
double *s0 = (double *) src;
double *s1;
double *d0;
register size_t inc = 0;
size_t offset = 0;
size_t i = 0, j;
A1PAMI_Info_t *ainfo = (A1PAMI_Info_t *) cookie;
size_t sy, ey;
size_t stride = ainfo->dstride;
size_t n = c0 / 32;
size_t poffset = ((size_t)target - (size_t)ainfo->dstaddr);
sy = poffset / c0;
ey = (poffset+512) / c0;
register double f0 FP_REG(0);
register double f1 FP_REG(1);
register double f2 FP_REG(2);
register double f4 FP_REG(4);
//f2 = ainfo->scale;
double *scale_ptr = &(ainfo->scale);
FP_LOAD(scale_ptr, inc, f2);
inc = 32;
s0 -= 4;
for (j = sy; j < ey; ++j) {
offset = (j*stride) / sizeof(double);
d0 = (double*)((char*)target + offset);
d0 -= 4;
s1 = d0;
for (i = 0; i < n; ++i) {
//d[offset + i] += scale * s[nd++];
QPX_LOAD(s0,inc,f0);
QPX_LOAD(s1,inc,f1);
ASM("qvfmadd 4, 0, 2, 1" ::: FP_REG1(4) );
QPX_STORE(d0,inc,f4);
}
}
A1PAMI_Assert ((size_t)s0 == (size_t)((char*)src+512-32));
}
rtems_task First_FP_task(
rtems_task_argument argument
)
{
rtems_status_code status;
rtems_id tid;
rtems_time_of_day time;
uint32_t task_index;
INTEGER_DECLARE;
FP_DECLARE;
status = rtems_task_ident( RTEMS_SELF, RTEMS_SEARCH_ALL_NODES, &tid );
directive_failed( status, "rtems_task_ident" );
task_index = task_number( tid );
INTEGER_LOAD( INTEGER_factors[ task_index ] );
FP_LOAD( FP_factors[ task_index ] );
put_name( Task_name[ task_index ], FALSE );
printf(
" - integer base = (0x%" PRIx32 ")\n",
INTEGER_factors[ task_index ]
);
put_name( Task_name[ task_index ], FALSE );
#if ( RTEMS_HAS_HARDWARE_FP == 1 )
printf( " - float base = (%g)\n", FP_factors[ task_index ] );
#else
printf( " - float base = (NA)\n" );
#endif
if ( argument == 0 ) {
status = rtems_task_restart( RTEMS_SELF, 1 );
directive_failed( status, "rtems_task_restart of RTEMS_SELF" );
} else {
build_time( &time, 12, 31, 1988, 9, 0, 0, 0 );
status = rtems_clock_set( &time );
directive_failed( status, "rtems_clock_set" );
status = rtems_task_delete( RTEMS_SELF );
directive_failed( status, "rtems_task_delete of RTEMS_SELF" );
}
}
static inline void qpx_data_fn_sx (void * target,
void * src,
size_t bytes,
double * scale)
{
double *src0 = (double *) src;
double *src1 = (double *) target;
double *dst = (double *) target;
unsigned num_dbls = 64;
register double f0 FP_REG(0);
register double f1 FP_REG(1);
register double f2 FP_REG(2);
register double f3 FP_REG(3);
register double f4 FP_REG(4);
register double f5 FP_REG(5);
register double f6 FP_REG(6);
register double f7 FP_REG(7);
register double f8 FP_REG(8);
register double f9 FP_REG(9);
register double f10 FP_REG(10);
register double f11 FP_REG(11);
register double f12 FP_REG(12);
/*f12 = *scale;*/ /* compiler may optimize this away*/
register int inc = 0;
FP_LOAD(scale, inc, f12);
unsigned niter = num_dbls >> 5;
niter --;
double *src0_1, *src1_1;
double *src0_2, *src1_2;
double *dst_1, *dst_2;
src0_1 = src0 -8;
src1_1 = src1 -8;
src0_2 = src0 -4;
src1_2 = src1 -4;
dst_1 = dst -8;
dst_2 = dst -4;
inc=64;
QPX_LOAD(src0_1,inc,f0);
QPX_LOAD(src1_1,inc,f1);
QPX_LOAD(src0_1,inc,f2);
QPX_LOAD(src1_1,inc,f3);
QPX_LOAD(src0_1,inc,f4);
QPX_LOAD(src1_1,inc,f5);
QPX_LOAD(src0_1,inc,f6);
QPX_LOAD(src1_1,inc,f7);
ASM("qvfmadd 8, 0, 12, 1" ::: FP_REG1(8) );
QPX_LOAD(src0_2,inc,f0);
QPX_LOAD(src1_2,inc,f1);
ASM("qvfmadd 9, 2, 12, 3" ::: FP_REG1(9) );
QPX_LOAD(src0_2,inc,f2);
QPX_LOAD(src1_2,inc,f3);
ASM("qvfmadd 10, 4, 12, 5" ::: FP_REG1(10) );
QPX_LOAD(src0_2,inc,f4);
QPX_LOAD(src1_2,inc,f5);
ASM("qvfmadd 11, 6, 12, 7" ::: FP_REG1(11) );
QPX_LOAD(src0_2,inc,f6);
QPX_LOAD(src1_2,inc,f7);
while (niter --) {
QPX_STORE(dst_1,inc,f8);
ASM("qvfmadd 8, 0, 12, 1" ::: FP_REG1(8) );
QPX_LOAD(src0_1,inc,f0);
QPX_LOAD(src1_1,inc,f1);
QPX_STORE(dst_1,inc,f9);
ASM("qvfmadd 9, 2, 12, 3" ::: FP_REG1(9) );
QPX_LOAD(src0_1,inc,f2);
QPX_LOAD(src1_1,inc,f3);
QPX_STORE(dst_1,inc,f10);
ASM("qvfmadd 10, 4, 12, 5" ::: FP_REG1(10) );
QPX_LOAD(src0_1,inc,f4);
QPX_LOAD(src1_1,inc,f5);
QPX_STORE(dst_1,inc,f11);
ASM("qvfmadd 11, 6, 12, 7" ::: FP_REG1(11) );
QPX_LOAD(src0_1,inc,f6);
QPX_LOAD(src1_1,inc,f7);
QPX_STORE(dst_2,inc,f8);
ASM("qvfmadd 8, 0, 12, 1" ::: FP_REG1(8) );
QPX_LOAD(src0_2,inc,f0);
QPX_LOAD(src1_2,inc,f1);
QPX_STORE(dst_2,inc,f9);
ASM("qvfmadd 9, 2, 12, 3" ::: FP_REG1(9) );
QPX_LOAD(src0_2,inc,f2);
QPX_LOAD(src1_2,inc,f3);
QPX_STORE(dst_2,inc,f10);
ASM("qvfmadd 10, 4, 12, 5" ::: FP_REG1(10) );
QPX_LOAD(src0_2,inc,f4);
QPX_LOAD(src1_2,inc,f5);
QPX_STORE(dst_2,inc,f11);
ASM("qvfmadd 11, 6, 12, 7" ::: FP_REG1(11) );
QPX_LOAD(src0_2,inc,f6);
QPX_LOAD(src1_2,inc,f7);
}
QPX_STORE(dst_1,inc,f8);
ASM("qvfmadd 8, 0, 12, 1" ::: FP_REG1(8) );
QPX_STORE(dst_1,inc,f9);
ASM("qvfmadd 9, 2, 12, 3" ::: FP_REG1(9) );
QPX_STORE(dst_1,inc,f10);
ASM("qvfmadd 10, 4, 12, 5" ::: FP_REG1(10) );
QPX_STORE(dst_1,inc,f11);
ASM("qvfmadd 11, 6, 12, 7" ::: FP_REG1(11) );
QPX_STORE(dst_2,inc,f8);
QPX_STORE(dst_2,inc,f9);
QPX_STORE(dst_2,inc,f10);
QPX_STORE(dst_2,inc,f11);
}
rtems_task FP_task(
rtems_task_argument argument
)
{
rtems_status_code status;
rtems_id tid;
rtems_time_of_day time;
uint32_t task_index;
uint32_t previous_seconds;
INTEGER_DECLARE;
FP_DECLARE;
status = rtems_task_ident( RTEMS_SELF, RTEMS_SEARCH_ALL_NODES, &tid );
directive_failed( status, "rtems_task_ident of self" );
task_index = task_number( tid );
INTEGER_LOAD( INTEGER_factors[ task_index ] );
FP_LOAD( FP_factors[ task_index ] );
put_name( Task_name[ task_index ], FALSE );
printf( " - integer base = (0x%x)\n", INTEGER_factors[ task_index ] );
put_name( Task_name[ task_index ], FALSE );
#if ( RTEMS_HAS_HARDWARE_FP == 1 )
printf( " - float base = (%g)\n", FP_factors[ task_index ] );
#else
printf( " - float base = (NA)\n" );
#endif
previous_seconds = -1;
while( FOREVER ) {
status = rtems_clock_get( RTEMS_CLOCK_GET_TOD, &time );
directive_failed( status, "rtems_clock_get" );
if ( time.second >= 16 ) {
if ( task_number( tid ) == 4 ) {
puts( "TA4 - rtems_task_delete - self" );
status = rtems_task_delete( RTEMS_SELF );
directive_failed( status, "rtems_task_delete of TA4" );
}
puts( "TA5 - rtems_task_delete - TA3" );
status = rtems_task_delete( Task_id[ 3 ] );
directive_failed( status, "rtems_task_delete of TA3" );
puts( "*** END OF TEST 19 *** " );
rtems_test_exit( 0 );
}
if (previous_seconds != time.second)
{
put_name( Task_name[ task_index ], FALSE );
print_time( " - rtems_clock_get - ", &time, "\n" );
previous_seconds = time.second;
}
INTEGER_CHECK( INTEGER_factors[ task_index ] );
FP_CHECK( FP_factors[ task_index ] );
/* for the first 4 seconds we spin as fast as possible
* so that we likely are interrupted
* After that, we go to sleep for a second at a time
*/
if (time.second >= 4)
{
status = rtems_task_wake_after( TICKS_PER_SECOND );
directive_failed( status, "rtems_task_wake_after" );
}
}
}
