ExternalVendorCode/whisper.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/whisper.cpp.git synced 2025-06-17 06:18:08 +00:00
Code Issues Actions 4 Packages Projects Releases Wiki Activity

whisper : disable CUDA mel + fix FFMPEG

Browse Source
This commit is contained in:
Georgi Gerganov
2024-06-26 20:11:38 +03:00
parent 3efedb9511
commit dc8cc2dd6f
6 changed files with 68 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files

53
src/CMakeLists.txt
View File

@ -77,27 +77,27 @@ if (WHISPER_OPENVINO)
set_target_properties(${TARGET} PROPERTIES FOLDER "libs")
endif()
if (GGML_CUDA)
cmake_minimum_required(VERSION 3.18) # for CMAKE_CUDA_ARCHITECTURES
find_package(CUDAToolkit)
if (CUDAToolkit_FOUND)
message(STATUS "CUDA found")
if (NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
# 52 == lowest CUDA 12 standard
# 60 == f16 CUDA intrinsics
# 61 == integer CUDA intrinsics
# 70 == compute capability at which unrolling a loop in mul_mat_q kernels is faster
set(CMAKE_CUDA_ARCHITECTURES "52;61;70") # lowest CUDA 12 standard + lowest for integer intrinsics
endif()
message(STATUS "Using CUDA architectures: ${CMAKE_CUDA_ARCHITECTURES}")
enable_language(CUDA)
else()
message(WARNING "CUDA not found")
endif()
endif()
#if (GGML_CUDA)
# cmake_minimum_required(VERSION 3.18) # for CMAKE_CUDA_ARCHITECTURES
#
# find_package(CUDAToolkit)
# if (CUDAToolkit_FOUND)
# message(STATUS "CUDA found")
#
# if (NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
# # 52 == lowest CUDA 12 standard
# # 60 == f16 CUDA intrinsics
# # 61 == integer CUDA intrinsics
# # 70 == compute capability at which unrolling a loop in mul_mat_q kernels is faster
# set(CMAKE_CUDA_ARCHITECTURES "52;61;70") # lowest CUDA 12 standard + lowest for integer intrinsics
# endif()
# message(STATUS "Using CUDA architectures: ${CMAKE_CUDA_ARCHITECTURES}")
#
# enable_language(CUDA)
# else()
# message(WARNING "CUDA not found")
# endif()
#endif()
# whisper
@ -107,11 +107,12 @@ add_library(whisper
whisper-mel.hpp
)
if (GGML_CUDA)
target_sources(whisper PRIVATE whisper-mel-cuda.cu)
target_link_libraries(whisper PRIVATE CUDA::cufft)
endif()
# TODO: disabled because it relies on ggml internals that are no longer accessible (ggml-backend-impl.h, ggml-cuda/common.cuh, ..)
#if (GGML_CUDA)
# target_sources(whisper PRIVATE whisper-mel-cuda.cu)
#
# target_link_libraries(whisper PRIVATE CUDA::cufft)
#endif()
# Set the version numbers
set_target_properties(whisper PROPERTIES

3
src/whisper-mel-cuda.cu
View File

@ -2,8 +2,7 @@
#include "whisper-mel-cuda.hpp"
#include "whisper.h"
#include <ggml-cuda/common.cuh>
#include <ggml-backend-impl.h>
#include <ggml-backend.h>
#include <cuda.h>
#include <cuda_runtime.h>

4
src/whisper.cpp
View File

@ -3215,7 +3215,9 @@ struct mel_calc_cpu : public whisper_mel_calc {
}
static whisper_mel_calc * whisper_mel_calc_create(ggml_backend_t backend, const whisper_filters & filters) {
#if defined(GGML_USE_CUDA) && !defined(GGML_USE_HIPBLAS)
// TODO: disabled because it relies on ggml internals that are no longer accessible (ggml-backend-impl.h, ggml-cuda/common.cuh, ..)
//#if defined(GGML_USE_CUDA) && !defined(GGML_USE_HIPBLAS)
#if 0
if (ggml_backend_is_cuda(backend)) {
auto ret = whisper_mel_calc_create_cuda(backend, filters);
if (ret) {