Skip to content

LLM inference in C/C++

C++ ggml

ggml : add a set of functions for checking contiguity of inner tensor dimensions (#25650)

fairydreaminga7312ae94f80
diff --git a/ggml/include/ggml.h b/ggml/include/ggml.h
index b2859ebe7d..92ba65222a 100644
--- a/ggml/include/ggml.h
+++ b/ggml/include/ggml.h
@@ -780,6 +780,10 @@ extern "C" {
GGML_API bool ggml_is_contiguous_1(const struct ggml_tensor * tensor); // contiguous for dims >= 1
GGML_API bool ggml_is_contiguous_2(const struct ggml_tensor * tensor); // contiguous for dims >= 2
+ GGML_API bool ggml_is_contiguous_to_1(const struct ggml_tensor * tensor); // contiguous for dims < 1
+ GGML_API bool ggml_is_contiguous_to_2(const struct ggml_tensor * tensor); // contiguous for dims < 2
+ GGML_API bool ggml_is_contiguous_to_3(const struct ggml_tensor * tensor); // contiguous for dims < 3
+
// returns whether the tensor elements are allocated as one contiguous block of memory (no gaps, but permutation ok)
GGML_API bool ggml_is_contiguously_allocated(const struct ggml_tensor * tensor);
diff --git a/ggml/src/ggml.c b/ggml/src/ggml.c
index f9cc7465e8..086b6ab084 100644
--- a/ggml/src/ggml.c
+++ b/ggml/src/ggml.c
@@ -1464,14 +1464,14 @@ bool ggml_is_transposed(const struct ggml_tensor * tensor) {
return tensor->nb[0] > tensor->nb[1];
}
-static bool ggml_is_contiguous_n(const struct ggml_tensor * tensor, int n) {
+static bool ggml_is_contiguous_m_n(const struct ggml_tensor * tensor, int m, int n) {
size_t next_nb = ggml_type_size(tensor->type);
if (tensor->ne[0] != ggml_blck_size(tensor->type) && tensor->nb[0] != next_nb) {
return false;
}
next_nb *= tensor->ne[0]/ggml_blck_size(tensor->type);
- for (int i = 1; i < GGML_MAX_DIMS; i++) {
- if (i > n) {
+ for (int i = 1; i < n; i++) {
+ if (i > m) {
if (tensor->ne[i] != 1 && tensor->nb[i] != next_nb) {
return false;
}
@@ -1489,15 +1489,27 @@ bool ggml_is_contiguous(const struct ggml_tensor * tensor) {
}
bool ggml_is_contiguous_0(const struct ggml_tensor * tensor) {
- return ggml_is_contiguous_n(tensor, 0);
+ return ggml_is_contiguous_m_n(tensor, 0, GGML_MAX_DIMS);
}
bool ggml_is_contiguous_1(const struct ggml_tensor * tensor) {
- return ggml_is_contiguous_n(tensor, 1);
+ return ggml_is_contiguous_m_n(tensor, 1, GGML_MAX_DIMS);
}
bool ggml_is_contiguous_2(const struct ggml_tensor * tensor) {
- return ggml_is_contiguous_n(tensor, 2);
+ return ggml_is_contiguous_m_n(tensor, 2, GGML_MAX_DIMS);
+}
+
+bool ggml_is_contiguous_to_1(const struct ggml_tensor * tensor) {
+ return ggml_is_contiguous_m_n(tensor, 0, 1);
+}
+
+bool ggml_is_contiguous_to_2(const struct ggml_tensor * tensor) {
+ return ggml_is_contiguous_m_n(tensor, 0, 2);
+}
+
+bool ggml_is_contiguous_to_3(const struct ggml_tensor * tensor) {
+ return ggml_is_contiguous_m_n(tensor, 0, 3);
}
bool ggml_is_contiguously_allocated(const struct ggml_tensor * tensor) {