From 17366df842e358768c0df7024484fffecfc7865b Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Johannes=20G=C3=A4=C3=9Fler?= Date: Tue, 6 Jun 2023 21:33:23 +0200 Subject: Multi GPU support, CUDA refactor, CUDA scratch buffer (#1703) * CUDA multi GPU + scratch ggml_cuda_compute_forward Tensor parallelism ggml_cuda_add ggml_cuda_rms_norm ggml_cuda_silu CUDA scratch buffer --main-gpu CLI option --- ggml-cuda.h | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-) (limited to 'ggml-cuda.h') diff --git a/ggml-cuda.h b/ggml-cuda.h index 6a04dde6..3b74e32e 100644 --- a/ggml-cuda.h +++ b/ggml-cuda.h @@ -1,10 +1,19 @@ +#pragma once + #include "ggml.h" #ifdef __cplusplus extern "C" { #endif +#define GGML_CUDA_MAX_DEVICES 16 + +struct ggml_tensor_extra_gpu { + void * data_device[GGML_CUDA_MAX_DEVICES]; // 1 pointer for each device for split tensors +}; + void ggml_init_cublas(void); +void ggml_cuda_set_tensor_split(const float * tensor_split); void ggml_cuda_mul(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * dst); bool ggml_cuda_can_mul_mat(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * dst); @@ -15,8 +24,12 @@ void ggml_cuda_mul_mat(const struct ggml_tensor * src0, const struct ggml_tens void * ggml_cuda_host_malloc(size_t size); void ggml_cuda_host_free(void * ptr); -void ggml_cuda_transform_tensor(struct ggml_tensor * tensor); -void ggml_cuda_load_data(const char * fname, struct ggml_tensor * tensors, size_t offset); +void ggml_cuda_load_data(const char * fname, struct ggml_tensor * tensors, size_t offset); +void ggml_cuda_free_data(struct ggml_tensor * tensor); +void ggml_cuda_assign_buffers(struct ggml_tensor * tensor); +void ggml_cuda_set_main_device(int main_device); +void ggml_cuda_set_scratch_size(size_t scratch_size); +bool ggml_cuda_compute_forward(struct ggml_compute_params * params, struct ggml_tensor * tensor); #ifdef __cplusplus } -- cgit v1.2.3