ggml : extend bin bcast for permuted src1 (#19484)
* tests : extend bin bcast for permuted src1 * cont : extend bin support * cont : s0 is always 1 * tests : simplify
This commit is contained in:
@@ -59,11 +59,7 @@ static void apply_binary_op(const ggml_compute_params * params, ggml_tensor * ds
|
||||
GGML_ASSERT(nb00 == sizeof(src0_t));
|
||||
|
||||
const auto [ir0, ir1] = get_thread_range(params, src0);
|
||||
const bool is_src1_contiguous = (nb10 == sizeof(src1_t));
|
||||
|
||||
if (!is_src1_contiguous) { // broadcast not implemented yet for non-contiguous
|
||||
GGML_ASSERT(ggml_are_same_shape(src0, src1));
|
||||
}
|
||||
const bool is_src1_contiguous_rows = ggml_is_contiguous_rows(src1);
|
||||
|
||||
#ifdef GGML_USE_ACCELERATE
|
||||
vDSP_fn_t vDSP_op = nullptr;
|
||||
@@ -94,7 +90,7 @@ static void apply_binary_op(const ggml_compute_params * params, ggml_tensor * ds
|
||||
const src0_t * src0_ptr = (const src0_t *) ((const char *) src0->data + i03*nb03 + i02*nb02 + i01*nb01);
|
||||
const src1_t * src1_ptr = (const src1_t *) ((const char *) src1->data + i13*nb13 + i12*nb12 + i11*nb11);
|
||||
|
||||
if (is_src1_contiguous) {
|
||||
if (is_src1_contiguous_rows) {
|
||||
// src1 is broadcastable across src0 and dst in i1, i2, i3
|
||||
const int64_t nr0 = ne00 / ne10;
|
||||
|
||||
|
||||
Reference in New Issue
Block a user