Skip to content

Use 32-bit addressing in take1 when possible #166

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
May 12, 2016
Merged

Use 32-bit addressing in take1 when possible #166

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
8713cfd
Add the option to do addressing in 32 bits since it is 3 times slower…
abergeron May 9, 2016
df7632b
Fix some mistakes following review.
abergeron May 10, 2016
d22c478
Fix the casting.
abergeron May 10, 2016
d32cd66
Fix the deref in kernel code.
abergeron May 11, 2016
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
87 changes: 46 additions & 41 deletions src/gpuarray_array.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -311,10 +311,11 @@ int GpuArray_index(GpuArray *r, const GpuArray *a, const ssize_t *starts,
static int gen_take1_kernel(GpuKernel *k, const gpuarray_buffer_ops *ops,
void *ctx, char **err_str,
GpuArray *a, const GpuArray *v,
const GpuArray *ind) {
const GpuArray *ind, int addr32) {
strb sb = STRB_STATIC_INIT;
int *atypes;
size_t nargs, apos;
char *sz, *ssz;
unsigned int i, i2;
int flags = GA_USE_CLUDA;
int res;
Expand All @@ -325,6 +326,14 @@ static int gen_take1_kernel(GpuKernel *k, const gpuarray_buffer_ops *ops,
if (atypes == NULL)
return GA_MEMORY_ERROR;

if (addr32) {
sz = "ga_uint";
ssz = "ga_int";
} else {
sz = "ga_size";
ssz = "ga_ssize";
}

apos = 0;
strb_appendf(&sb, "KERNEL void take1(GLOBAL_MEM %s *r, "
"GLOBAL_MEM const %s *v, ga_size off,",
Expand All @@ -338,44 +347,43 @@ static int gen_take1_kernel(GpuKernel *k, const gpuarray_buffer_ops *ops,
atypes[apos++] = GA_SSIZE;
atypes[apos++] = GA_SIZE;
}
strb_appends(&sb, " GLOBAL_MEM const ga_ssize *ind, ga_size n0, ga_size n1,"
" GLOBAL_MEM int* err) {\n");
atypes[apos++] = GA_BUFFER;
atypes[apos++] = GA_SIZE;
atypes[apos++] = GA_SIZE;
atypes[apos++] = GA_BUFFER;
assert(apos == nargs);
strb_appends(&sb, " GLOBAL_MEM const ga_size *ind, ga_size n0, ga_size n1,"
" GLOBAL_MEM int* err) {\n"
" const ga_size idx0 = LDIM_0 * GID_0 + LID_0;\n"
" const ga_size numThreads0 = LDIM_0 * GDIM_0;\n"
" const ga_size idx1 = LDIM_1 * GID_1 + LID_1;\n"
" const ga_size numThreads1 = LDIM_1 * GDIM_1;\n"
" ga_size i0, i1;\n"
" for (i0 = idx0; i0 < n0; i0 += numThreads0) {\n"
" ga_ssize ii0 = ind[i0];\n"
" ga_size pos0 = off;\n"
strb_appendf(&sb, " const %s idx0 = LDIM_0 * GID_0 + LID_0;\n"
" const %s numThreads0 = LDIM_0 * GDIM_0;\n"
" const %s idx1 = LDIM_1 * GID_1 + LID_1;\n"
" const %s numThreads1 = LDIM_1 * GDIM_1;\n"
" %s i0, i1;\n", sz, sz, sz, sz, sz);
strb_appendf(&sb, " for (i0 = idx0; i0 < n0; i0 += numThreads0) {\n"
" %s ii0 = ind[i0];\n"
" %s pos0 = off;\n"
" if (ii0 < 0) ii0 += d0;\n"
" if ((ii0 < 0) || ((ga_size)ii0 >= d0)) {\n"
" if ((ii0 < 0) || (ii0 >= d0)) {\n"
" *err = -1;\n"
" continue;\n"
" }\n"
" pos0 += ii0 * s0;\n"
" pos0 += ii0 * (%s)s0;\n"
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What the d in this line and the one bellow mean? Is there a special meaning for strb_appendf?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No, it will just be a literal 'd' followed by the formatted number. It's the name of one of the inputs.

" for (i1 = idx1; i1 < n1; i1 += numThreads1) {\n"
" GLOBAL_MEM ga_byte *p = (GLOBAL_MEM ga_byte *)v;\n"
" p += pos0;\n");
" %s p = pos0;\n", ssz, sz, sz, sz);
if (v->nd > 1) {
strb_appends(&sb, " ga_size pos, ii = i1;\n");
strb_appendf(&sb, " %s pos, ii = i1;\n", sz);
for (i2 = v->nd; i2 > 1; i2--) {
i = i2 - 1;
if (i > 1)
strb_appendf(&sb, " pos = ii %% d%u;\n"
" ii /= d%u;\n", i, i);
strb_appendf(&sb, " pos = ii %% (%s)d%u;\n"
" ii /= (%s)d%u;\n", sz, i, sz, i);
else
strb_appends(&sb, " pos = ii;\n");
strb_appendf(&sb, " p += pos * s%u;\n", i);
strb_appendf(&sb, " p += pos * (%s)s%u;\n", ssz, i);
}
}
strb_appendf(&sb, " r[i0*n1 + i1] = ((GLOBAL_MEM %s *)p)[0];\n",
gpuarray_get_type(v->typecode)->cluda_name);
strb_appendf(&sb, " r[i0*((%s)n1) + i1] = *((GLOBAL_MEM %s *)(((GLOBAL_MEM char *)v) + p));\n",
sz, gpuarray_get_type(v->typecode)->cluda_name);
strb_appends(&sb, " }\n"
" }\n"
"}\n");
Expand All @@ -400,11 +408,11 @@ int GpuArray_take1(GpuArray *a, const GpuArray *v, const GpuArray *i,
#if DEBUG
char *errstr = NULL;
#endif
void **args = NULL;
size_t argp;
GpuKernel k;
unsigned int j;
int err, kerr = 0;
int addr32 = 0;

if (a->ops != v->ops || a->ops != i->ops)
return GA_INVALID_ERROR;
Expand Down Expand Up @@ -434,6 +442,10 @@ int GpuArray_take1(GpuArray *a, const GpuArray *v, const GpuArray *i,
n[1] *= v->dimensions[j];
}

if (n[0] * n[1] < SADDR32_MAX) {
addr32 = 1;
}

err = v->ops->property(NULL, v->data, NULL, GA_CTX_PROP_ERRBUF, &errbuf);
if (err != GA_NO_ERROR)
return err;
Expand All @@ -444,7 +456,7 @@ int GpuArray_take1(GpuArray *a, const GpuArray *v, const GpuArray *i,
#else
NULL,
#endif
a, v, i);
a, v, i, addr32);
#if DEBUG
if (errstr != NULL) {
fprintf(stderr, "%s\n", errstr);
Expand All @@ -469,27 +481,20 @@ int GpuArray_take1(GpuArray *a, const GpuArray *v, const GpuArray *i,
}
gs[0] = 1;

args = calloc(7 + 2 * v->nd, sizeof(void *));
if (args == NULL) {
err = GA_MEMORY_ERROR;
goto out;
}

argp = 0;
args[argp++] = a->data;
args[argp++] = v->data;
args[argp++] = (void *)&v->offset;
GpuKernel_setarg(&k, argp++, a->data);
GpuKernel_setarg(&k, argp++, v->data);
GpuKernel_setarg(&k, argp++, (void *)&v->offset);
for (j = 0; j < v->nd; j++) {
args[argp++] = &v->strides[j];
args[argp++] = &v->dimensions[j];
GpuKernel_setarg(&k, argp++, &v->strides[j]);
GpuKernel_setarg(&k, argp++, &v->dimensions[j]);
}
args[argp++] = i->data;
args[argp++] = &n[0];
args[argp++] = &n[1];
args[argp++] = errbuf;
GpuKernel_setarg(&k, argp++, i->data);
GpuKernel_setarg(&k, argp++, &n[0]);
GpuKernel_setarg(&k, argp++, &n[1]);
GpuKernel_setarg(&k, argp++, errbuf);

err = GpuKernel_call(&k, 2, ls, gs, 0, args);
free(args);
err = GpuKernel_call(&k, 2, ls, gs, 0, NULL);
if (check_error && err == GA_NO_ERROR) {
err = v->ops->buffer_read(&kerr, errbuf, 0, sizeof(int));
if (err == GA_NO_ERROR && kerr != 0) {
Expand Down Expand Up @@ -1102,7 +1107,7 @@ int GpuArray_fdump(FILE *fd, const GpuArray *a) {
int GpuArray_is_c_contiguous(const GpuArray *a) {
size_t size = GpuArray_ITEMSIZE(a);
int i;

for (i = a->nd - 1; i >= 0; i--) {
if (a->dimensions[i] != 1 && a->strides[i] != size)
return 0;
Expand Down
4 changes: 0 additions & 4 deletions src/gpuarray_elemwise.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,10 +21,6 @@ struct _GpuElemwise {

#define GEN_ADDR32 0x1

#define ADDR32_MAX 4294967295
#define SADDR32_MIN -2147483648
#define SADDR32_MAX 2147483647

#define is_array(a) (ISCLR((a).flags, GE_SCALAR))

static inline const char *ctype(int typecode) {
Expand Down
4 changes: 4 additions & 0 deletions src/private.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,6 +21,10 @@ extern "C" {
}
#endif

#define ADDR32_MAX 4294967295
#define SADDR32_MIN -2147483648
#define SADDR32_MAX 2147483647

static inline void *memdup(const void *p, size_t s) {
void *res = malloc(s);
if (res != NULL)
Expand Down