ggml : add ggml_row_size() (fixes llama out of space) (#4461)

* Fixes "Not enough space in the context's memory pool" encountered on certain models, which seems to be caused by some imprecision related to the automatic casting of floating point values

* do not cast to size_t, instead just use doubles

* ggml : add ggml_row_size(), deprecate ggml_type_sizef()

* ggml : fix row size compute to avoid overflows

* tests : fix sizey -> sizez

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

1 files changed, 7 insertions, 7 deletions

diff --git a/examples/benchmark/benchmark-matmult.cpp b/examples/benchmark/benchmark-matmult.cpp
index 284733b1..434e1d6b 100644
--- a/examples/benchmark/benchmark-matmult.cpp
+++ b/examples/benchmark/benchmark-matmult.cpp
@@ -129,13 +129,13 @@ int main(int argc, char ** argv)  {
     const ggml_type qtype = GGML_TYPE_Q4_1;
 
     size_t ctx_size = 0;
-    ctx_size += sizex*sizey*ggml_type_sizef(GGML_TYPE_F32);
-    ctx_size += sizex*sizey*ggml_type_sizef(GGML_TYPE_F32);
-    ctx_size += sizex*sizez*ggml_type_sizef(GGML_TYPE_F32);
-    ctx_size += sizex*sizey*ggml_type_sizef(qtype);
-    ctx_size += sizex*sizey*ggml_type_sizef(qtype);
-    ctx_size += sizex*sizey*ggml_type_sizef(GGML_TYPE_F32); // BLAS
-    ctx_size += sizex*sizey*ggml_type_sizef(GGML_TYPE_F32); // BLAS
+    ctx_size += ggml_row_size(GGML_TYPE_F32, sizex*sizey);
+    ctx_size += ggml_row_size(GGML_TYPE_F32, sizex*sizey);
+    ctx_size += ggml_row_size(GGML_TYPE_F32, sizex*sizez);
+    ctx_size += ggml_row_size(qtype,         sizex*sizey);
+    ctx_size += ggml_row_size(qtype,         sizex*sizey);
+    ctx_size += ggml_row_size(GGML_TYPE_F32, sizex*sizey); // BLAS
+    ctx_size += ggml_row_size(GGML_TYPE_F32, sizex*sizey); // BLAS
     ctx_size += 1024*1024*16;
 
     printf("Allocating Memory of size %zi bytes, %zi MB\n",ctx_size, (ctx_size/1024/1024));