ggml : fix cont with transposed tensors when one dimension is 1 (ggml/934)

* ggml_cont: fix issue with transposed tensors when one dimension is 1

when using multiple threads, it is not enough
to check for the tensors to be contiguous for
ggml_compute_forward_dup_same_cont to work correctly.
The tensors strides also need to match.

Signed-off-by: Salvatore Mesoraca <s.mesoraca16@gmail.com>

* Add ggml_cont tests

Signed-off-by: Salvatore Mesoraca <s.mesoraca16@gmail.com>

* Remove dead code

it isn't possible to reach this code because
all these functions are invoked by ggml_compute_forward_dup
if and only if src0->type != dst->type

Signed-off-by: Salvatore Mesoraca <s.mesoraca16@gmail.com>

* Make ggml_compute_forward_dup_same_cont work with contiguous tensors

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Signed-off-by: Salvatore Mesoraca <s.mesoraca16@gmail.com>

---------

Signed-off-by: Salvatore Mesoraca <s.mesoraca16@gmail.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

ggml/src/ggml.c

@@ -8322,8 +8322,7 @@ static void ggml_compute_forward_dup_same_cont(
     GGML_ASSERT(ggml_is_contiguous(dst) && ggml_is_contiguous(src0));
     GGML_ASSERT(src0->type == dst->type);
 
-    const size_t nb00 = src0->nb[0];
-    const size_t nb0 = dst->nb[0];
+    const size_t nb0 = ggml_type_size(src0->type);
 
     const int ith = params->ith; // thread index
     const int nth = params->nth; // number of threads
@@ -8337,8 +8336,8 @@ static void ggml_compute_forward_dup_same_cont(
     if (ie0 < ie1) {
         memcpy(
             ((char *)  dst->data + ie0*nb0),
-            ((char *) src0->data + ie0*nb00),
-            (ie1 - ie0) * ggml_type_size(src0->type));
+            ((char *) src0->data + ie0*nb0),
+            (ie1 - ie0) * nb0);
     }
 }
 
@@ -8355,11 +8354,6 @@ static void ggml_compute_forward_dup_f16(
     const int ith = params->ith; // thread index
     const int nth = params->nth; // number of threads
 
-    if (ggml_is_contiguous(src0) && ggml_is_contiguous(dst) && src0->type == dst->type) {
-        ggml_compute_forward_dup_same_cont(params, dst);
-        return;
-    }
-
     // parallelize by rows
     const int nr = ne01;
     // number of rows per thread
@@ -8624,11 +8618,6 @@ static void ggml_compute_forward_dup_bf16(
     const int ith = params->ith; // thread index
     const int nth = params->nth; // number of threads
 
-    if (ggml_is_contiguous(src0) && ggml_is_contiguous(dst) && src0->type == dst->type) {
-        ggml_compute_forward_dup_same_cont(params, dst);
-        return;
-    }
-
     // parallelize by rows
     const int nr = ne01;
     // number of rows per thread
@@ -8980,11 +8969,6 @@ static void ggml_compute_forward_dup_f32(
     const int ith = params->ith; // thread index
     const int nth = params->nth; // number of threads
 
-    if (ggml_is_contiguous(src0) && ggml_is_contiguous(dst) && src0->type == dst->type) {
-        ggml_compute_forward_dup_same_cont(params, dst);
-        return;
-    }
-
     // parallelize by rows
     const int nr = ne01;
     // number of rows per thread
@@ -9294,13 +9278,13 @@ static void ggml_compute_forward_dup_bytes(
     GGML_ASSERT(ggml_nelements(dst) == ggml_nelements(src0));
     GGML_ASSERT(src0->type == dst->type);
 
+    GGML_TENSOR_UNARY_OP_LOCALS;
+
     if (ggml_is_contiguous(src0) && ggml_is_contiguous(dst)) {
         ggml_compute_forward_dup_same_cont(params, dst);
         return;
     }
 
-    GGML_TENSOR_UNARY_OP_LOCALS;
-
     const size_t type_size = ggml_type_size(src0->type);
     const int ith = params->ith; // thread index
     const int nth = params->nth; // number of threads

tests/test-backend-ops.cpp

@@ -2322,6 +2322,15 @@ static bool test_backend(ggml_backend_t backend, test_mode mode, const char * op
     }
 
     test_cases.emplace_back(new test_cont());
+    test_cases.emplace_back(new test_cont(GGML_TYPE_F32, {2, 1, 1 ,1}));
+    test_cases.emplace_back(new test_cont(GGML_TYPE_F32, {2, 1, 3 ,5}));
+    test_cases.emplace_back(new test_cont(GGML_TYPE_F32, {2, 3, 5 ,7}));
+    test_cases.emplace_back(new test_cont(GGML_TYPE_F16, {2, 1, 1 ,1}));
+    test_cases.emplace_back(new test_cont(GGML_TYPE_F16, {2, 1, 3 ,5}));
+    test_cases.emplace_back(new test_cont(GGML_TYPE_F16, {2, 3, 5 ,7}));
+    test_cases.emplace_back(new test_cont(GGML_TYPE_BF16, {2, 1, 1 ,1}));
+    test_cases.emplace_back(new test_cont(GGML_TYPE_BF16, {2, 1, 3 ,5}));
+    test_cases.emplace_back(new test_cont(GGML_TYPE_BF16, {2, 3, 5 ,7}));
 
     auto add_test_bin_bcast = [&](ggml_type type, std::array<int64_t, 4> ne, std::array<int, 4> nr) {
         for (auto op : {ggml_add, ggml_mul, ggml_div}) {