ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

ggml : reduce hash table reset cost (#8698)

Browse source 
* ggml : reduce hash table reset cost

* fix unreachable code warnings after GGML_ASSERT(false)

* GGML_ASSERT(false) -> GGML_ABORT("fatal error")

* GGML_ABORT use format string
This commit is contained in:
slaren 2024-07-27 04:41:55 +02:00 • committed by GitHub
parent 01245f5b16
commit 2b1f616b20
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
46 changed files with 851 additions and 754 deletions
Download patch file Download diff file Expand all files Collapse all files

6
ggml/src/ggml-cuda.cu
View file

@ -98,7 +98,7 @@ void ggml_cuda_error(const char * stmt, const char * func, const char * file, in
GGML_CUDA_LOG_ERROR(" current device: %d, in function %s at %s:%d\n", id, func, file, line);
GGML_CUDA_LOG_ERROR(" %s\n", stmt);
// abort with GGML_ASSERT to get a stack trace
GGML_ASSERT(!"CUDA error");
GGML_ABORT("CUDA error");
}
// this is faster on Windows
@ -1596,7 +1596,7 @@ static void ggml_cuda_op_mul_mat(
CUDA_CHECK(ggml_cuda_cpy_tensor_2d(
src1_ddf_i, src1, i03, i02, src1_col_0, src1_col_0+src1_ncols, stream));
} else {
GGML_ASSERT(false);
GGML_ABORT("fatal error");
}
if (quantize_src1 && !src1_is_contiguous) {
@ -2945,7 +2945,7 @@ static void ggml_backend_cuda_event_wait(ggml_backend_t backend, ggml_backend_ev
CUDA_CHECK(cudaLaunchHostFunc(cuda_ctx->stream(), wait_fn, event));
#endif
GGML_ASSERT(false);
GGML_ABORT("fatal error");
}
}