END_TEST
START_TEST(test_basic_simple)
{
GpuArray a;
GpuArray b;
GpuArray c;
GpuElemwise *ge;
static const uint32_t data1[3] = {1, 2, 3};
static const uint32_t data2[3] = {4, 5, 6};
uint32_t data3[3] = {0};
size_t dims[2];
gpuelemwise_arg args[3] = {{0}};
void *rargs[3];
dims[0] = 1;
dims[1] = 3;
ga_assert_ok(GpuArray_empty(&a, ctx, GA_UINT, 2, dims, GA_C_ORDER));
ga_assert_ok(GpuArray_write(&a, data1, sizeof(data1)));
ga_assert_ok(GpuArray_empty(&b, ctx, GA_UINT, 2, dims, GA_F_ORDER));
ga_assert_ok(GpuArray_write(&b, data2, sizeof(data2)));
ga_assert_ok(GpuArray_empty(&c, ctx, GA_UINT, 2, dims, GA_C_ORDER));
args[0].name = "a";
args[0].typecode = GA_UINT;
args[0].flags = GE_READ;
args[1].name = "b";
args[1].typecode = GA_UINT;
args[1].flags = GE_READ;
args[2].name = "c";
args[2].typecode = GA_UINT;
args[2].flags = GE_WRITE;
ge = GpuElemwise_new(ctx, "", "c = a + b", 3, args, 2, 0);
ck_assert_ptr_ne(ge, NULL);
rargs[0] = &a;
rargs[1] = &b;
rargs[2] = &c;
ga_assert_ok(GpuElemwise_call(ge, rargs, GE_NOCOLLAPSE));
ga_assert_ok(GpuArray_read(data3, sizeof(data3), &c));
ck_assert_int_eq(data3[0], 5);
ck_assert_int_eq(data3[1], 7);
ck_assert_int_eq(data3[2], 9);
}
static int ga_extcopy(GpuArray *dst, const GpuArray *src) {
struct extcopy_args a, *aa;
gpucontext *ctx = gpudata_context(dst->data);
GpuElemwise *k = NULL;
void *args[2];
if (ctx != gpudata_context(src->data))
return GA_INVALID_ERROR;
a.itype = src->typecode;
a.otype = dst->typecode;
if (ctx->extcopy_cache != NULL)
k = cache_get(ctx->extcopy_cache, &a);
if (k == NULL) {
gpuelemwise_arg gargs[2];
gargs[0].name = "src";
gargs[0].typecode = src->typecode;
gargs[0].flags = GE_READ;
gargs[1].name = "dst";
gargs[1].typecode = dst->typecode;
gargs[1].flags = GE_WRITE;
k = GpuElemwise_new(ctx, "", "dst = src", 2, gargs, 0, 0);
if (k == NULL)
return GA_MISC_ERROR;
aa = memdup(&a, sizeof(a));
if (aa == NULL) {
GpuElemwise_free(k);
return GA_MEMORY_ERROR;
}
if (ctx->extcopy_cache == NULL)
ctx->extcopy_cache = cache_twoq(4, 8, 8, 2, extcopy_eq, extcopy_hash,
extcopy_free,
(cache_freev_fn)GpuElemwise_free,
ctx->err);
if (ctx->extcopy_cache == NULL)
return GA_MISC_ERROR;
if (cache_add(ctx->extcopy_cache, aa, k) != 0)
return GA_MISC_ERROR;
}
args[0] = (void *)src;
args[1] = (void *)dst;
return GpuElemwise_call(k, args, GE_BROADCAST);
}
END_TEST
START_TEST(test_basic_neg_strides)
{
GpuArray a;
GpuArray b;
GpuArray c;
GpuElemwise *ge;
static const uint32_t data1[6] = {1, 2, 3, 4, 5, 6};
static const uint32_t data2[6] = {7, 8, 9, 10, 11, 12};
uint32_t data3[6] = {0};
size_t dims[1];
gpuelemwise_arg args[3] = {{0}};
void *rargs[3];
ssize_t starts[1];
ssize_t stops[1];
ssize_t steps[1];
dims[0] = 6;
ga_assert_ok(GpuArray_empty(&a, ctx, GA_UINT, 1, dims, GA_C_ORDER));
ga_assert_ok(GpuArray_write(&a, data1, sizeof(data1)));
ga_assert_ok(GpuArray_empty(&b, ctx, GA_UINT, 1, dims, GA_C_ORDER));
ga_assert_ok(GpuArray_write(&b, data2, sizeof(data2)));
starts[0] = 5;
stops[0] = -1;
steps[0] = -1;
ga_assert_ok(GpuArray_index_inplace(&b, starts, stops, steps));
ga_assert_ok(GpuArray_empty(&c, ctx, GA_UINT, 1, dims, GA_C_ORDER));
args[0].name = "a";
args[0].typecode = GA_UINT;
args[0].flags = GE_READ;
args[1].name = "b";
args[1].typecode = GA_UINT;
args[1].flags = GE_READ;
args[2].name = "c";
args[2].typecode = GA_UINT;
args[2].flags = GE_WRITE;
ge = GpuElemwise_new(ctx, "", "c = a + b", 3, args, 1, 0);
ck_assert_ptr_ne(ge, NULL);
rargs[0] = &a;
rargs[1] = &b;
rargs[2] = &c;
ga_assert_ok(GpuElemwise_call(ge, rargs, 0));
ga_assert_ok(GpuArray_read(data3, sizeof(data3), &c));
ck_assert_int_eq(data3[0], 13);
ck_assert_int_eq(data3[1], 13);
ck_assert_int_eq(data3[2], 13);
ck_assert_int_eq(data3[3], 13);
ck_assert_int_eq(data3[4], 13);
ck_assert_int_eq(data3[5], 13);
}
END_TEST
START_TEST(test_basic_remove1)
{
GpuArray a;
GpuArray b;
GpuArray c;
GpuElemwise *ge;
static const uint32_t data1[6] = {1, 2, 3, 4, 5, 6};
static const uint32_t data2[6] = {7, 8, 9, 10, 11, 12};
uint32_t data3[6] = {0};
size_t dims[4];
gpuelemwise_arg args[3] = {{0}};
void *rargs[3];
dims[0] = 1;
dims[1] = 3;
dims[2] = 2;
dims[3] = 1;
ga_assert_ok(GpuArray_empty(&a, ctx, GA_UINT, 4, dims, GA_C_ORDER));
ga_assert_ok(GpuArray_write(&a, data1, sizeof(data1)));
ga_assert_ok(GpuArray_empty(&b, ctx, GA_UINT, 4, dims, GA_F_ORDER));
ga_assert_ok(GpuArray_write(&b, data2, sizeof(data2)));
ga_assert_ok(GpuArray_empty(&c, ctx, GA_UINT, 4, dims, GA_C_ORDER));
args[0].name = "a";
args[0].typecode = GA_UINT;
args[0].flags = GE_READ;
args[1].name = "b";
args[1].typecode = GA_UINT;
args[1].flags = GE_READ;
args[2].name = "c";
args[2].typecode = GA_UINT;
args[2].flags = GE_WRITE;
ge = GpuElemwise_new(ctx, "", "c = a + b", 3, args, 0, 0);
ck_assert_ptr_ne(ge, NULL);
rargs[0] = &a;
rargs[1] = &b;
rargs[2] = &c;
ga_assert_ok(GpuElemwise_call(ge, rargs, 0));
ga_assert_ok(GpuArray_read(data3, sizeof(data3), &c));
ck_assert_int_eq(data3[0], 8);
ck_assert_int_eq(data3[1], 12);
ck_assert_int_eq(data3[2], 11);
ck_assert_int_eq(data3[3], 15);
ck_assert_int_eq(data3[4], 14);
ck_assert_int_eq(data3[5], 18);
}