/*
* A \in M(m, n)
* B \in M(n, p)
* C \in M(m, p)
*/
VOID_TASK_8(rec_matmul, REAL*, A, REAL*, B, REAL*, C, int, m, int, n, int, p, int, ld, int, add)
{
if ((m + n + p) <= 64) {
int i, j, k;
/* base case */
if (add) {
for (i = 0; i < m; i++)
for (k = 0; k < p; k++) {
REAL c = 0.0;
for (j = 0; j < n; j++)
c += A[i * ld + j] * B[j * ld + k];
C[i * ld + k] += c;
}
} else {
for (i = 0; i < m; i++)
for (k = 0; k < p; k++) {
REAL c = 0.0;
for (j = 0; j < n; j++)
c += A[i * ld + j] * B[j * ld + k];
C[i * ld + k] = c;
}
}
} else if (m >= n && n >= p) {
int m1 = m >> 1;
SPAWN(rec_matmul, A, B, C, m1, n, p, ld, add);
CALL(rec_matmul, A + m1 * ld, B, C + m1 * ld, m - m1, n, p, ld, add);
SYNC(rec_matmul);
} else if (n >= m && n >= p) {
TASK_3(BDD, union_par, BDD*, arr, int, first, int, last)
{
if (first == last) return arr[first];
BDD left, right;
int mid = (first+last)/2;
// check: last == first + 1
if (first == mid) { // we have 2
return CALL(sylvan_ite, arr[first], sylvan_true, arr[last], 0); // or
}
if (mid+1 == last) { // we have 3
left = sylvan_ref(CALL(sylvan_ite, arr[first], sylvan_true, arr[mid], 0)); // or
BDD result = CALL(sylvan_ite, left, sylvan_true, arr[last], 0);
sylvan_deref(left);
return result;
}
SPAWN(union_par, arr, first, mid);
right = sylvan_ref(CALL(union_par, arr, mid+1, last));
left = sylvan_ref(SYNC(union_par));
BDD result = CALL(sylvan_ite, left, sylvan_true, right, 0);
sylvan_deref(left);
sylvan_deref(right);
return result;
}
/*
* Run a command. The "cmd" is a pointer to a command string, or NULL if you
* want to run a copy of DCL in the subjob (this is how the standard routine
* LIB$SPAWN works. You have to do wierd stuff with the terminal on the way in
* and the way out, because DCL does not want the channel to be in raw mode.
*/
int sys(char *cmd)
{
struct dsc$descriptor cdsc;
struct dsc$descriptor *cdscp;
long status;
long substatus;
long iosb[2];
status = SYS$QIOW(EFN, iochan, IO$_SETMODE, iosb, 0, 0,
oldmode, sizeof(oldmode), 0, 0, 0, 0);
if (status != SS$_NORMAL || (iosb[0] & 0xFFFF) != SS$_NORMAL)
return FALSE;
cdscp = NULL; /* Assume DCL. */
if (cmd != NULL) { /* Build descriptor. */
cdsc.dsc$a_pointer = cmd;
cdsc.dsc$w_length = strlen(cmd);
cdsc.dsc$b_dtype = DSC$K_DTYPE_T;
cdsc.dsc$b_class = DSC$K_CLASS_S;
cdscp = &cdsc;
}
status = LIB$SPAWN(cdscp, 0, 0, 0, 0, 0, &substatus, 0, 0, 0);
if (status != SS$_NORMAL)
substatus = status;
status = SYS$QIOW(EFN, iochan, IO$_SETMODE, iosb, 0, 0,
newmode, sizeof(newmode), 0, 0, 0, 0);
if (status != SS$_NORMAL || (iosb[0] & 0xFFFF) != SS$_NORMAL)
return FALSE;
if ((substatus & STS$M_SUCCESS) == 0) /* Command failed. */
return FALSE;
return TRUE;
}
TASK_1(int, pfib, int, n)
{
if( n < 2 ) {
return n;
} else {
int m,k;
SPAWN( pfib, n-1 );
k = CALL( pfib, n-2 );
m = SYNC( pfib );
return m+k;
}
}
globle void VMSSystem(
char *cmd)
{
long status, complcode;
struct dsc$descriptor_s cmd_desc;
cmd_desc.dsc$w_length = strlen(cmd);
cmd_desc.dsc$a_pointer = cmd;
cmd_desc.dsc$b_class = DSC$K_CLASS_S;
cmd_desc.dsc$b_dtype = DSC$K_DTYPE_T;
status = LIB$SPAWN(&cmd_desc,0,0,0,0,0,&complcode,0,0,0);
}
static void spawn_authorize(char *expected_file)
{
static $DESCRIPTOR(cmd_dx, "MCR SYS$SYSTEM:AUTHORIZE list/brief *");
int status, flags;
printf("Spawning authorize command...\n");
flags = 2 + 8; /* noclisym + nokeypad */
status = LIB$SPAWN(&cmd_dx, 0, 0, &flags, 0, 0, 0, 0, 0, 0, 0, 0, 0);
if ((status & 1) == 0) {
printf("Error spawning spawn\n");
exit(status);
}
}
MVMint64 MVM_proc_shell(MVMThreadContext *tc, MVMString *cmd, MVMString *cwd, MVMObject *env) {
MVMint64 result = 0, spawn_result = 0;
uv_process_t *process = calloc(1, sizeof(uv_process_t));
uv_process_options_t process_options = {0};
uv_stdio_container_t process_stdio[3];
int i;
char * const cmdin = MVM_string_utf8_encode_C_string(tc, cmd);
char * const _cwd = MVM_string_utf8_encode_C_string(tc, cwd);
const MVMuint64 size = MVM_repr_elems(tc, env);
MVMIter * const iter = (MVMIter *)MVM_iter(tc, env);
char **_env = malloc((size + 1) * sizeof(char *));
#ifdef _WIN32
const MVMuint16 acp = GetACP(); /* We should get ACP at runtime. */
char * const _cmd = ANSIToUTF8(acp, getenv("ComSpec"));
char *args[3];
args[0] = "/c";
{
MVMint64 len = strlen(cmdin);
MVMint64 i;
for (i = 0; i < len; i++)
if (cmdin[i] == '/')
cmdin[i] = '\\';
}
args[1] = cmdin;
args[2] = NULL;
#else
char * const _cmd = "/bin/sh";
char *args[4];
args[0] = "/bin/sh";
args[1] = "-c";
args[2] = cmdin;
args[3] = NULL;
#endif
INIT_ENV();
SPAWN(_cmd);
FREE_ENV();
free(_cwd);
#ifdef _WIN32
free(_cmd);
#endif
free(cmdin);
return result;
}
void CpuGMTask::draw() {
SkBitmap bitmap;
SetupBitmap(fColorType, fGM.get(), &bitmap);
SkCanvas canvas(bitmap);
canvas.concat(fGM->getInitialTransform());
fGM->draw(&canvas);
canvas.flush();
#define SPAWN(ChildTask, ...) this->spawnChild(SkNEW_ARGS(ChildTask, (*this, __VA_ARGS__)))
SPAWN(ExpectationsTask, fExpectations, bitmap);
SPAWN(PipeTask, fGMFactory(NULL), bitmap, false, false);
SPAWN(PipeTask, fGMFactory(NULL), bitmap, true, false);
SPAWN(PipeTask, fGMFactory(NULL), bitmap, true, true);
SPAWN(QuiltTask, fGMFactory(NULL), bitmap);
SPAWN(RecordTask, fGMFactory(NULL), bitmap);
SPAWN(ReplayTask, fGMFactory(NULL), bitmap, false);
SPAWN(ReplayTask, fGMFactory(NULL), bitmap, true);
SPAWN(SerializeTask, fGMFactory(NULL), bitmap);
SPAWN(WriteTask, bitmap);
#undef SPAWN
}
TASK_3(int, nqueens, int, n , int, j, char*, a) {
#else
int nqueens(int n, int j, char *a) {
#endif
// No more queens to place.
if(n == j) {
return 1;
}
// Try each possible position for queen <j>
int solutions = 0;
for(int i = 0; i < n; i++) {
#ifdef WOOL
SPAWN(nqueens_next, n, j, a, i);
#else
a[j] = i;
if(ok(j + 1, a)) {
solutions += nqueens(n, j + 1, a);
}
#endif
}
#ifdef WOOL
for(int i = 0; i < n; i++) {
solutions += SYNC(nqueens_next);
}
#endif
return solutions;
}
#ifdef WOOL
TASK_IMPL_4(int, nqueens_next, int, n, int, j, char*, a, int, i) {
char *b = (char *) alloca((j + 1) * sizeof(char));
memcpy(b, a, j * sizeof(char));
b[j] = i;
if(ok(j + 1, b)) {
return CALL(nqueens, n, j + 1, b);
} else {
return 0;
}
}
TASK_3( int, tree, int, d, int, s, int, n )
{
if( d>0 ) {
int r = 1, l = 1, a, b;
if( s<0 ) {
r = -s;
} else {
l = s;
}
SPAWN( tree, d-r, s, n );
a = CALL( tree, d-l, s, n);
b = SYNC( tree );
return a+b;
} else {
loop( n );
return 1;
}
}
TASK_3(MTBDD, bdd_from_ldd, MDD, dd, MDD, bits_dd, uint32_t, firstvar)
{
/* simple for leaves */
if (dd == lddmc_false) return mtbdd_false;
if (dd == lddmc_true) return mtbdd_true;
MTBDD result;
/* get from cache */
/* note: some assumptions about the encoding... */
if (cache_get3(bdd_from_ldd_id, dd, bits_dd, firstvar, &result)) return result;
mddnode_t n = LDD_GETNODE(dd);
mddnode_t nbits = LDD_GETNODE(bits_dd);
int bits = (int)mddnode_getvalue(nbits);
/* spawn right, same bits_dd and firstvar */
mtbdd_refs_spawn(SPAWN(bdd_from_ldd, mddnode_getright(n), bits_dd, firstvar));
/* call down, with next bits_dd and firstvar */
MTBDD down = CALL(bdd_from_ldd, mddnode_getdown(n), mddnode_getdown(nbits), firstvar + 2*bits);
/* encode current value */
uint32_t val = mddnode_getvalue(n);
for (int i=0; i<bits; i++) {
/* encode with high bit first */
int bit = bits-i-1;
if (val & (1LL<<i)) down = mtbdd_makenode(firstvar + 2*bit, mtbdd_false, down);
else down = mtbdd_makenode(firstvar + 2*bit, down, mtbdd_false);
}
/* sync right */
mtbdd_refs_push(down);
MTBDD right = mtbdd_refs_sync(SYNC(bdd_from_ldd));
/* take union of current and right */
mtbdd_refs_push(right);
result = sylvan_or(down, right);
mtbdd_refs_pop(2);
/* put in cache */
cache_put3(bdd_from_ldd_id, dd, bits_dd, firstvar, result);
return result;
}
/*
* return the optimal solution for n items (first is e) and
* capacity c. Value so far is v.
*/
TASK_4(int, knapsack, struct item *, e, int, c, int, n, int, v)
{
int with, without, best;
double ub;
/* base case: full knapsack or no items */
if (c < 0)
return INT_MIN;
if (n == 0 || c == 0)
return v; /* feasible solution, with value v */
ub = (double) v + c * e->value / e->weight;
if (ub < best_so_far) {
/* prune ! */
//return INT_MIN;
}
/*
* compute the best solution without the current item in the knapsack
*/
SPAWN(knapsack, e + 1, c, n - 1, v);
/* compute the best solution with the current item in the knapsack */
with = CALL(knapsack, e + 1, c - e->weight, n - 1, v + e->value);
without = SYNC(knapsack);
best = with > without ? with : without;
/*
* notice the race condition here. The program is still
* correct, in the sense that the best solution so far
* is at least best_so_far. Moreover best_so_far gets updated
* when returning, so eventually it should get the right
* value. The program is highly non-deterministic.
*/
//if (best > best_so_far)
// best_so_far = best;
return best;
}
VOID_TASK_3(compute_highest_action, MDD, dd, MDD, meta, uint32_t*, target)
{
if (dd == lddmc_true || dd == lddmc_false) return;
if (meta == lddmc_true) return;
uint64_t result = 1;
if (cache_get3(compute_highest_action_id, dd, meta, 0, &result)) return;
cache_put3(compute_highest_action_id, dd, meta, 0, result);
/* meta:
* 0 is skip
* 1 is read
* 2 is write
* 3 is only-read
* 4 is only-write
* 5 is action label (at end, before -1)
* -1 is end
*/
const mddnode_t n = LDD_GETNODE(dd);
const mddnode_t nmeta = LDD_GETNODE(meta);
const uint32_t vmeta = mddnode_getvalue(nmeta);
if (vmeta == (uint32_t)-1) return;
SPAWN(compute_highest_action, mddnode_getright(n), meta, target);
CALL(compute_highest_action, mddnode_getdown(n), mddnode_getdown(nmeta), target);
SYNC(compute_highest_action);
if (vmeta == 5) {
has_actions = 1;
const uint32_t v = mddnode_getvalue(n);
while (1) {
const uint32_t cur = *(volatile uint32_t*)target;
if (v <= cur) break;
if (__sync_bool_compare_and_swap(target, cur, v)) break;
}
}
}
VOID_TASK_2(compute_highest, MDD, dd, uint32_t*, arr)
{
if (dd == lddmc_true || dd == lddmc_false) return;
uint64_t result = 1;
if (cache_get3(compute_highest_id, dd, 0, 0, &result)) return;
cache_put3(compute_highest_id, dd, 0, 0, result);
mddnode_t n = LDD_GETNODE(dd);
SPAWN(compute_highest, mddnode_getright(n), arr);
CALL(compute_highest, mddnode_getdown(n), arr+1);
SYNC(compute_highest);
if (!mddnode_getcopy(n)) {
const uint32_t v = mddnode_getvalue(n);
while (1) {
const uint32_t cur = *(volatile uint32_t*)arr;
if (v <= cur) break;
if (__sync_bool_compare_and_swap(arr, cur, v)) break;
}
}
}
TASK_2(Result, parTreeSearch, int, depth, Node *, parent) {
int numChildren, childType;
counter_t parentHeight = parent->height;
Result r = { depth, 1, 0 };
numChildren = uts_numChildren(parent);
childType = uts_childType(parent);
// record number of children in parent
parent->numChildren = numChildren;
// Recurse on the children
if (numChildren > 0) {
int i, j;
for (i = 0; i < numChildren; i++) {
Node *child = (Node*)alloca(sizeof(Node));
child->type = childType;
child->height = parentHeight + 1;
child->numChildren = -1; // not yet determined
for (j = 0; j < computeGranularity; j++) {
rng_spawn(parent->state.state, child->state.state, i);
}
SPAWN(parTreeSearch, depth+1, child);
}
/* Wait a bit */
struct timespec tim = (struct timespec){0, 100L*numChildren};
nanosleep(&tim, NULL);
for (i = 0; i < numChildren; i++) {
Result c = SYNC(parTreeSearch);
if (c.maxdepth>r.maxdepth) r.maxdepth = c.maxdepth;
r.size += c.size;
r.leaves += c.leaves;
}
} else {
MVMint64 MVM_proc_spawn(MVMThreadContext *tc, MVMObject *argv, MVMString *cwd, MVMObject *env) {
MVMint64 result = 0, spawn_result = 0;
uv_process_t *process = calloc(1, sizeof(uv_process_t));
uv_process_options_t process_options = {0};
uv_stdio_container_t process_stdio[3];
int i;
char * const _cwd = MVM_string_utf8_encode_C_string(tc, cwd);
const MVMuint64 size = MVM_repr_elems(tc, env);
MVMIter * const iter = (MVMIter *)MVM_iter(tc, env);
char **_env = malloc((size + 1) * sizeof(char *));
const MVMuint64 arg_size = MVM_repr_elems(tc, argv);
char **args = malloc((arg_size + 1) * sizeof(char *));
MVMRegister reg;
i = 0;
while(i < arg_size) {
REPR(argv)->pos_funcs.at_pos(tc, STABLE(argv), argv, OBJECT_BODY(argv), i, ®, MVM_reg_obj);
args[i++] = MVM_string_utf8_encode_C_string(tc, MVM_repr_get_str(tc, reg.o));
}
args[arg_size] = NULL;
INIT_ENV();
SPAWN(arg_size ? args[0] : NULL);
FREE_ENV();
free(_cwd);
i = 0;
while(args[i])
free(args[i++]);
free(args);
return result;
}
/*****************************************************************************
**
** OptimizedStrassenMultiply
**
** For large matrices A, B, and C of size MatrixSize * MatrixSize this
** function performs the operation C = A x B efficiently.
**
** INPUT:
** C = (*C WRITE) Address of top left element of matrix C.
** A = (*A IS READ ONLY) Address of top left element of matrix A.
** B = (*B IS READ ONLY) Address of top left element of matrix B.
** MatrixSize = Size of matrices (for n*n matrix, MatrixSize = n)
** RowWidthA = Number of elements in memory between A[x,y] and A[x,y+1]
** RowWidthB = Number of elements in memory between B[x,y] and B[x,y+1]
** RowWidthC = Number of elements in memory between C[x,y] and C[x,y+1]
**
** OUTPUT:
** C = (*C WRITE) Matrix C contains A x B. (Initial value of *C undefined.)
**
*****************************************************************************/
VOID_TASK_7(OptimizedStrassenMultiply, REAL *, C, REAL *, A, REAL *, B,
unsigned, MatrixSize,
unsigned, RowWidthC,
unsigned, RowWidthA,
unsigned, RowWidthB
)
{
unsigned QuadrantSize = MatrixSize >> 1; /* MatixSize / 2 */
unsigned QuadrantSizeInBytes = sizeof(REAL) * QuadrantSize * QuadrantSize
+ 32;
unsigned Column, Row;
/************************************************************************
** For each matrix A, B, and C, we'll want pointers to each quandrant
** in the matrix. These quandrants will be addressed as follows:
** -- --
** | A11 A12 |
** | |
** | A21 A22 |
** -- --
************************************************************************/
REAL /* *A11, *B11, *C11, */ *A12, *B12, *C12,
*A21, *B21, *C21, *A22, *B22, *C22;
REAL *S1,*S2,*S3,*S4,*S5,*S6,*S7,*S8,*M2,*M5,*T1sMULT;
#define T2sMULT C22
#define NumberOfVariables 11
PTR TempMatrixOffset = 0;
PTR MatrixOffsetA = 0;
PTR MatrixOffsetB = 0;
char *Heap;
void *StartHeap;
/* Distance between the end of a matrix row and the start of the next row */
PTR RowIncrementA = ( RowWidthA - QuadrantSize ) << 3;
PTR RowIncrementB = ( RowWidthB - QuadrantSize ) << 3;
PTR RowIncrementC = ( RowWidthC - QuadrantSize ) << 3;
if (MatrixSize <= SizeAtWhichDivideAndConquerIsMoreEfficient) {
MultiplyByDivideAndConquer(C, A, B,
MatrixSize,
RowWidthC,
RowWidthA,
RowWidthB,
0);
return;
}
/* Initialize quandrant matrices */
#define A11 A
#define B11 B
#define C11 C
A12 = A11 + QuadrantSize;
B12 = B11 + QuadrantSize;
C12 = C11 + QuadrantSize;
A21 = A + (RowWidthA * QuadrantSize);
B21 = B + (RowWidthB * QuadrantSize);
C21 = C + (RowWidthC * QuadrantSize);
A22 = A21 + QuadrantSize;
B22 = B21 + QuadrantSize;
C22 = C21 + QuadrantSize;
/* Allocate Heap Space Here */
StartHeap = Heap = malloc(QuadrantSizeInBytes * NumberOfVariables);
/* ensure that heap is on cache boundary */
if ( ((PTR) Heap) & 31)
Heap = (char*) ( ((PTR) Heap) + 32 - ( ((PTR) Heap) & 31) );
/* Distribute the heap space over the variables */
S1 = (REAL*) Heap; Heap += QuadrantSizeInBytes;
S2 = (REAL*) Heap; Heap += QuadrantSizeInBytes;
S3 = (REAL*) Heap; Heap += QuadrantSizeInBytes;
S4 = (REAL*) Heap; Heap += QuadrantSizeInBytes;
S5 = (REAL*) Heap; Heap += QuadrantSizeInBytes;
S6 = (REAL*) Heap; Heap += QuadrantSizeInBytes;
S7 = (REAL*) Heap; Heap += QuadrantSizeInBytes;
S8 = (REAL*) Heap; Heap += QuadrantSizeInBytes;
M2 = (REAL*) Heap; Heap += QuadrantSizeInBytes;
M5 = (REAL*) Heap; Heap += QuadrantSizeInBytes;
T1sMULT = (REAL*) Heap; Heap += QuadrantSizeInBytes;
/***************************************************************************
** Step through all columns row by row (vertically)
** (jumps in memory by RowWidth => bad locality)
** (but we want the best locality on the innermost loop)
***************************************************************************/
for (Row = 0; Row < QuadrantSize; Row++) {
/*************************************************************************
** Step through each row horizontally (addressing elements in each column)
** (jumps linearly througn memory => good locality)
*************************************************************************/
for (Column = 0; Column < QuadrantSize; Column++) {
/***********************************************************
** Within this loop, the following holds for MatrixOffset:
** MatrixOffset = (Row * RowWidth) + Column
** (note: that the unit of the offset is number of reals)
***********************************************************/
/* Element of Global Matrix, such as A, B, C */
#define E(Matrix) (* (REAL*) ( ((PTR) Matrix) + TempMatrixOffset ) )
#define EA(Matrix) (* (REAL*) ( ((PTR) Matrix) + MatrixOffsetA ) )
#define EB(Matrix) (* (REAL*) ( ((PTR) Matrix) + MatrixOffsetB ) )
/* FIXME - may pay to expand these out - got higher speed-ups below */
/* S4 = A12 - ( S2 = ( S1 = A21 + A22 ) - A11 ) */
E(S4) = EA(A12) - ( E(S2) = ( E(S1) = EA(A21) + EA(A22) ) - EA(A11) );
/* S8 = (S6 = B22 - ( S5 = B12 - B11 ) ) - B21 */
E(S8) = ( E(S6) = EB(B22) - ( E(S5) = EB(B12) - EB(B11) ) ) - EB(B21);
/* S3 = A11 - A21 */
E(S3) = EA(A11) - EA(A21);
/* S7 = B22 - B12 */
E(S7) = EB(B22) - EB(B12);
TempMatrixOffset += sizeof(REAL);
MatrixOffsetA += sizeof(REAL);
MatrixOffsetB += sizeof(REAL);
} /* end row loop*/
MatrixOffsetA += RowIncrementA;
MatrixOffsetB += RowIncrementB;
} /* end column loop */
/* M2 = A11 x B11 */
SPAWN(OptimizedStrassenMultiply, M2, A11, B11, QuadrantSize,
QuadrantSize, RowWidthA, RowWidthB);
/* M5 = S1 * S5 */
SPAWN(OptimizedStrassenMultiply, M5, S1, S5, QuadrantSize,
QuadrantSize, QuadrantSize, QuadrantSize);
/* Step 1 of T1 = S2 x S6 + M2 */
SPAWN(OptimizedStrassenMultiply, T1sMULT, S2, S6, QuadrantSize,
QuadrantSize, QuadrantSize, QuadrantSize);
/* Step 1 of T2 = T1 + S3 x S7 */
SPAWN(OptimizedStrassenMultiply, C22, S3, S7, QuadrantSize,
RowWidthC /*FIXME*/, QuadrantSize, QuadrantSize);
/* Step 1 of C11 = M2 + A12 * B21 */
SPAWN(OptimizedStrassenMultiply, C11, A12, B21, QuadrantSize,
RowWidthC, RowWidthA, RowWidthB);
/* Step 1 of C12 = S4 x B22 + T1 + M5 */
SPAWN(OptimizedStrassenMultiply, C12, S4, B22, QuadrantSize,
RowWidthC, QuadrantSize, RowWidthB);
/* Step 1 of C21 = T2 - A22 * S8 */
SPAWN(OptimizedStrassenMultiply, C21, A22, S8, QuadrantSize,
RowWidthC, RowWidthA, QuadrantSize);
/**********************************************
** Synchronization Point
**********************************************/
SYNC(OptimizedStrassenMultiply);
SYNC(OptimizedStrassenMultiply);
SYNC(OptimizedStrassenMultiply);
SYNC(OptimizedStrassenMultiply);
SYNC(OptimizedStrassenMultiply);
SYNC(OptimizedStrassenMultiply);
SYNC(OptimizedStrassenMultiply);
/***************************************************************************
** Step through all columns row by row (vertically)
** (jumps in memory by RowWidth => bad locality)
** (but we want the best locality on the innermost loop)
***************************************************************************/
for (Row = 0; Row < QuadrantSize; Row++) {
/*************************************************************************
** Step through each row horizontally (addressing elements in each column)
** (jumps linearly througn memory => good locality)
*************************************************************************/
for (Column = 0; Column < QuadrantSize; Column += 4) {
REAL LocalM5_0 = *(M5);
REAL LocalM5_1 = *(M5+1);
REAL LocalM5_2 = *(M5+2);
REAL LocalM5_3 = *(M5+3);
REAL LocalM2_0 = *(M2);
REAL LocalM2_1 = *(M2+1);
REAL LocalM2_2 = *(M2+2);
REAL LocalM2_3 = *(M2+3);
REAL T1_0 = *(T1sMULT) + LocalM2_0;
REAL T1_1 = *(T1sMULT+1) + LocalM2_1;
REAL T1_2 = *(T1sMULT+2) + LocalM2_2;
REAL T1_3 = *(T1sMULT+3) + LocalM2_3;
REAL T2_0 = *(C22) + T1_0;
REAL T2_1 = *(C22+1) + T1_1;
REAL T2_2 = *(C22+2) + T1_2;
REAL T2_3 = *(C22+3) + T1_3;
(*(C11)) += LocalM2_0;
(*(C11+1)) += LocalM2_1;
(*(C11+2)) += LocalM2_2;
(*(C11+3)) += LocalM2_3;
(*(C12)) += LocalM5_0 + T1_0;
(*(C12+1)) += LocalM5_1 + T1_1;
(*(C12+2)) += LocalM5_2 + T1_2;
(*(C12+3)) += LocalM5_3 + T1_3;
(*(C22)) = LocalM5_0 + T2_0;
(*(C22+1)) = LocalM5_1 + T2_1;
(*(C22+2)) = LocalM5_2 + T2_2;
(*(C22+3)) = LocalM5_3 + T2_3;
(*(C21 )) = (- *(C21 )) + T2_0;
(*(C21+1)) = (- *(C21+1)) + T2_1;
(*(C21+2)) = (- *(C21+2)) + T2_2;
(*(C21+3)) = (- *(C21+3)) + T2_3;
M5 += 4;
M2 += 4;
T1sMULT += 4;
C11 += 4;
C12 += 4;
C21 += 4;
C22 += 4;
}
C11 = (REAL*) ( ((PTR) C11 ) + RowIncrementC);
C12 = (REAL*) ( ((PTR) C12 ) + RowIncrementC);
C21 = (REAL*) ( ((PTR) C21 ) + RowIncrementC);
C22 = (REAL*) ( ((PTR) C22 ) + RowIncrementC);
}
free(StartHeap);
}
int
main (int argc, char **argv)
{
int i;
char cwdev [128], *devptr;
int devlen;
char *cwd = getcwd (0, 1024);
devptr = strchr (cwd, ':');
devlen = (devptr - cwd) + 1;
strncpy (cwdev, cwd, devlen);
cwdev [devlen] = '\0';
search_dirs = xstrdup (system_search_dirs);
defines = xstrdup (default_defines);
addarg ("cc");
preprocess_args (&argc , argv);
process_args (&argc , argv);
if (strlen (search_dirs) > 0)
{
addarg ("/include=(");
addarg (search_dirs);
addarg (")");
}
if (strlen (defines) > 0)
{
addarg ("/define=(");
addarg (defines);
addarg (")");
}
for (i = 1; i < argc; i++)
{
int arg_len = strlen (argv[i]);
if (strcmp (argv[i], "-o") == 0)
i++;
else if (strcmp (argv[i], "-v" ) == 0
|| strcmp (argv[i], "-E") == 0
|| strcmp (argv[i], "-c") == 0
|| strncmp (argv[i], "-g", 2 ) == 0
|| strncmp (argv[i], "-O", 2 ) == 0
|| strcmp (argv[i], "-save-temps") == 0
|| (arg_len > 2 && strncmp (argv[i], "-I", 2) == 0)
|| (arg_len > 2 && strncmp (argv[i], "-D", 2) == 0))
;
/* Unix style file specs and VMS style switches look alike, so assume
an arg consisting of one and only one slash, and that being first, is
really a switch. */
else if ((argv[i][0] == '/') && (strchr (&argv[i][1], '/') == 0))
addarg (argv[i]);
else
{
/* Assume filename arg */
char buff [256], *ptr;
ptr = to_host_file_spec (argv[i]);
arg_len = strlen (ptr);
if (ptr[0] == '[')
sprintf (buff, "%s%s", cwdev, ptr);
else if (strchr (ptr, ':'))
sprintf (buff, "%s", ptr);
else
sprintf (buff, "%s%s", cwd, ptr);
ptr = xstrdup (buff);
addarg (ptr);
}
}
addarg (NULL);
if (verbose)
{
int i;
for (i = 0; i < comp_arg_index; i++)
printf ("%s ", comp_args [i]);
putchar ('\n');
}
{
int i;
int len = 0;
for (i = 0; comp_args[i]; i++)
len = len + strlen (comp_args[i]) + 1;
{
char *allargs = (char *) alloca (len + 1);
Descr cmd;
int status;
int status1 = 1;
for (i = 0; i < len + 1; i++)
allargs [i] = 0;
for (i = 0; comp_args [i]; i++)
{
strcat (allargs, comp_args [i]);
strcat (allargs, " ");
}
cmd.adr = allargs;
cmd.len = len;
cmd.mbz = 0;
i = LIB$SPAWN (&cmd, 0, 0, 0, 0, 0, &status);
if ((i & 1) != 1)
{
LIB$SIGNAL (i);
exit (1);
}
if ((status & 1) == 1 && (status1 & 1) == 1)
exit (0);
exit (1);
}
}
}
int
pload (int tbl)
{
int c = 0, i, status;
if (verbose > 0)
{
fprintf (stderr, "Starting %d children to load %s",
children, tdefs[tbl].comment);
}
for (c = 0; c < children; c++)
{
pids[c] = SPAWN ();
if (pids[c] == -1)
{
perror ("Child loader not created");
kill_load ();
exit (-1);
}
else if (pids[c] == 0) /* CHILD */
{
SET_HANDLER (stop_proc);
verbose = 0;
partial (tbl, c+1);
exit (0);
}
else if (verbose > 0) /* PARENT */
fprintf (stderr, ".");
}
if (verbose > 0)
fprintf (stderr, "waiting...");
c = children;
while (c)
{
i = WAIT (&status, pids[c - 1]);
if (i == -1 && children)
{
if (errno == ECHILD)
fprintf (stderr, "\nCould not wait on pid %d\n", pids[c - 1]);
else if (errno == EINTR)
fprintf (stderr, "\nProcess %d stopped abnormally\n", pids[c - 1]);
else if (errno == EINVAL)
fprintf (stderr, "\nProgram bug\n");
}
if (! WIFEXITED(status)) {
(void) fprintf(stderr, "\nProcess %d: ", i);
if (WIFSIGNALED(status)) {
(void) fprintf(stderr, "rcvd signal %d\n",
WTERMSIG(status));
} else if (WIFSTOPPED(status)) {
(void) fprintf(stderr, "stopped, signal %d\n",
WSTOPSIG(status));
}
}
c--;
}
if (verbose > 0)
fprintf (stderr, "done\n");
return (0);
}
TASK_4(MTBDD, bdd_from_ldd_rel, MDD, dd, MDD, bits_dd, uint32_t, firstvar, MDD, meta)
{
if (dd == lddmc_false) return mtbdd_false;
if (dd == lddmc_true) return mtbdd_true;
assert(meta != lddmc_false && meta != lddmc_true);
/* meta:
* -1 is end
* 0 is skip
* 1 is read
* 2 is write
* 3 is only-read
* 4 is only-write
*/
MTBDD result;
/* note: assumptions */
if (cache_get4(bdd_from_ldd_rel_id, dd, bits_dd, firstvar, meta, &result)) return result;
const mddnode_t n = LDD_GETNODE(dd);
const mddnode_t nmeta = LDD_GETNODE(meta);
const mddnode_t nbits = LDD_GETNODE(bits_dd);
const int bits = (int)mddnode_getvalue(nbits);
const uint32_t vmeta = mddnode_getvalue(nmeta);
assert(vmeta != (uint32_t)-1);
if (vmeta == 0) {
/* skip level */
result = bdd_from_ldd_rel(dd, mddnode_getdown(nbits), firstvar + 2*bits, mddnode_getdown(nmeta));
} else if (vmeta == 1) {
/* read level */
assert(!mddnode_getcopy(n)); // do not process read copy nodes for now
assert(mddnode_getright(n) != mtbdd_true);
/* spawn right */
mtbdd_refs_spawn(SPAWN(bdd_from_ldd_rel, mddnode_getright(n), bits_dd, firstvar, meta));
/* compute down with same bits / firstvar */
MTBDD down = bdd_from_ldd_rel(mddnode_getdown(n), bits_dd, firstvar, mddnode_getdown(nmeta));
mtbdd_refs_push(down);
/* encode read value */
uint32_t val = mddnode_getvalue(n);
MTBDD part = mtbdd_true;
for (int i=0; i<bits; i++) {
/* encode with high bit first */
int bit = bits-i-1;
if (val & (1LL<<i)) part = mtbdd_makenode(firstvar + 2*bit, mtbdd_false, part);
else part = mtbdd_makenode(firstvar + 2*bit, part, mtbdd_false);
}
/* intersect read value with down result */
mtbdd_refs_push(part);
down = sylvan_and(part, down);
mtbdd_refs_pop(2);
/* sync right */
mtbdd_refs_push(down);
MTBDD right = mtbdd_refs_sync(SYNC(bdd_from_ldd_rel));
/* take union of current and right */
mtbdd_refs_push(right);
result = sylvan_or(down, right);
mtbdd_refs_pop(2);
} else if (vmeta == 2 || vmeta == 4) {
/* write or only-write level */
/* spawn right */
assert(mddnode_getright(n) != mtbdd_true);
mtbdd_refs_spawn(SPAWN(bdd_from_ldd_rel, mddnode_getright(n), bits_dd, firstvar, meta));
/* get recursive result */
MTBDD down = CALL(bdd_from_ldd_rel, mddnode_getdown(n), mddnode_getdown(nbits), firstvar + 2*bits, mddnode_getdown(nmeta));
if (mddnode_getcopy(n)) {
/* encode a copy node */
for (int i=0; i<bits; i++) {
int bit = bits-i-1;
MTBDD low = mtbdd_makenode(firstvar + 2*bit + 1, down, mtbdd_false);
mtbdd_refs_push(low);
MTBDD high = mtbdd_makenode(firstvar + 2*bit + 1, mtbdd_false, down);
mtbdd_refs_pop(1);
down = mtbdd_makenode(firstvar + 2*bit, low, high);
}
} else {
/* encode written value */
uint32_t val = mddnode_getvalue(n);
for (int i=0; i<bits; i++) {
/* encode with high bit first */
int bit = bits-i-1;
if (val & (1LL<<i)) down = mtbdd_makenode(firstvar + 2*bit + 1, mtbdd_false, down);
else down = mtbdd_makenode(firstvar + 2*bit + 1, down, mtbdd_false);
}
}
/* sync right */
mtbdd_refs_push(down);
MTBDD right = mtbdd_refs_sync(SYNC(bdd_from_ldd_rel));
/* take union of current and right */
mtbdd_refs_push(right);
result = sylvan_or(down, right);
mtbdd_refs_pop(2);
} else if (vmeta == 3) {
/* only-read level */
assert(!mddnode_getcopy(n)); // do not process read copy nodes
/* spawn right */
mtbdd_refs_spawn(SPAWN(bdd_from_ldd_rel, mddnode_getright(n), bits_dd, firstvar, meta));
/* get recursive result */
MTBDD down = CALL(bdd_from_ldd_rel, mddnode_getdown(n), mddnode_getdown(nbits), firstvar + 2*bits, mddnode_getdown(nmeta));
/* encode read value */
uint32_t val = mddnode_getvalue(n);
for (int i=0; i<bits; i++) {
/* encode with high bit first */
int bit = bits-i-1;
/* only-read, so write same value */
if (val & (1LL<<i)) down = mtbdd_makenode(firstvar + 2*bit + 1, mtbdd_false, down);
else down = mtbdd_makenode(firstvar + 2*bit + 1, down, mtbdd_false);
if (val & (1LL<<i)) down = mtbdd_makenode(firstvar + 2*bit, mtbdd_false, down);
else down = mtbdd_makenode(firstvar + 2*bit, down, mtbdd_false);
}
/* sync right */
mtbdd_refs_push(down);
MTBDD right = mtbdd_refs_sync(SYNC(bdd_from_ldd_rel));
/* take union of current and right */
mtbdd_refs_push(right);
result = sylvan_or(down, right);
mtbdd_refs_pop(2);
} else if (vmeta == 5) {
assert(!mddnode_getcopy(n)); // not allowed!
/* we assume this is the last value */
result = mtbdd_true;
/* encode action value */
uint32_t val = mddnode_getvalue(n);
for (int i=0; i<actionbits; i++) {
/* encode with high bit first */
int bit = actionbits-i-1;
/* only-read, so write same value */
if (val & (1LL<<i)) result = mtbdd_makenode(1000000 + bit, mtbdd_false, result);
else result = mtbdd_makenode(1000000 + bit, result, mtbdd_false);
}
} else {
assert(vmeta <= 5);
}
cache_put4(bdd_from_ldd_rel_id, dd, bits_dd, firstvar, meta, result);
return result;
}
VOID_TASK_2(quicksort, int *, array, int, length) {
#else
void quicksort(int *array, int length) {
#endif
if(length < 2) {
return;
}
int pivot = array[length / 2];
int left = 0;
int right = length - 1;
while(left <= right) {
int left_data = array[left];
if(left_data < pivot) {
left++;
continue;
}
int right_data = array[right];
if(pivot < right_data) {
right--;
continue;
}
array[left++] = right_data;
array[right--] = left_data;
}
#ifdef WOOL
SPAWN(quicksort, array, right + 1);
CALL(quicksort, array + left, length - left);
SYNC(quicksort);
#else
quicksort(array, right + 1);
quicksort(array + left, length - left);
#endif
}
int main(int argc, char **argv) {
#ifdef WOOL
argc = wool_init(argc, argv);
#endif
if(argc < 2) {
#ifdef WOOL
fprintf(stderr, "Usage: quicksort <woolopt> <n>\n");
#else
fprintf(stderr, "Usage: quicksort <n>\n");
#endif
return 1;
}
int n = atoi(argv[1]);
int *data = (int *) malloc(n * sizeof(int));
for(int i = 0; i < n; i++) {
data[i] = rand();
}
#ifdef WOOL
CALL(quicksort, data, n);
wool_fini();
#else
quicksort(data, n);
#endif
printf("[ %d, %d, %d, %d, %d ... %d, %d, %d, %d, %d ] (len=%d)\n",
data[0], data[1], data[2], data[3], data[4],
data[n-5], data[n-4], data[n-3], data[n-2], data[n-1],
n);
return 0;
}
void CpuGMTask::draw() {
SkBitmap bitmap;
AllocatePixels(fColorType, fGM->getISize().width(), fGM->getISize().height(), &bitmap);
SkCanvas canvas(bitmap);
canvas.concat(fGM->getInitialTransform());
fGM->draw(&canvas);
canvas.flush();
#define SPAWN(ChildTask, ...) this->spawnChild(SkNEW_ARGS(ChildTask, (*this, __VA_ARGS__)))
SPAWN(ExpectationsTask, fExpectations, bitmap);
SPAWN(PipeTask, fGMFactory(NULL), bitmap, PipeTask::kInProcess_Mode);
SPAWN(PipeTask, fGMFactory(NULL), bitmap, PipeTask::kCrossProcess_Mode);
SPAWN(PipeTask, fGMFactory(NULL), bitmap, PipeTask::kSharedAddress_Mode);
SPAWN(QuiltTask, fGMFactory(NULL), bitmap);
SPAWN(RecordTask, fGMFactory(NULL), bitmap, RecordTask::kOptimize_Mode);
SPAWN(RecordTask, fGMFactory(NULL), bitmap, RecordTask::kNoOptimize_Mode);
SPAWN(ReplayTask, fGMFactory(NULL), bitmap, ReplayTask::kNormal_Mode);
SPAWN(ReplayTask, fGMFactory(NULL), bitmap, ReplayTask::kRTree_Mode);
SPAWN(SerializeTask, fGMFactory(NULL), bitmap);
SPAWN(WriteTask, bitmap);
#undef SPAWN
}