extern/whisper.cpp
1
0
Fork 1
mirror of https://github.com/ggerganov/whisper.cpp.git synced 2024-11-07 08:34:37 +01:00
Code Issues Packages Projects Releases Wiki Activity

whisper : revert mel-related changes (#0)

Browse Source 
too much extra logic and complexity for small benefit
This commit is contained in:
Georgi Gerganov 2024-10-05 14:29:45 +03:00
parent 941912467d
commit 396089f3cf
8 changed files with 143 additions and 676 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
.gitignore vendored
View File

@ -9,6 +9,7 @@
.DS_Store
.vimspector.json
/CMakeSettings.json
/talk-llama.dSYM/
build/
build-*/

8
Makefile
View File

@ -512,9 +512,6 @@ ifdef GGML_CUDA
OBJ_GGML += ggml/src/ggml-cuda.o
OBJ_GGML += $(patsubst %.cu,%.o,$(wildcard ggml/src/ggml-cuda/*.cu))
OBJ_GGML += $(OBJ_CUDA_TMPL)
#OBJ_WHISPER += src/whisper-mel-cuda.o
ifdef WHISPER_FATAL_WARNINGS
MK_NVCCFLAGS += -Werror all-warnings
endif # WHISPER_FATAL_WARNINGS
@ -623,10 +620,6 @@ ggml/src/ggml-cuda.o: \
ggml/src/ggml-common.h \
$(wildcard ggml/src/ggml-cuda/*.cuh)
$(NVCC_COMPILE)
#src/whisper-mel-cuda.o: src/whisper-mel-cuda.cu src/whisper-mel-cuda.hpp
# $(NVCC) $(NVCCFLAGS) $(CPPFLAGS) -Xcompiler "$(CUDA_CXXFLAGS)" -c $< -o $@
endif # GGML_CUDA
ifdef GGML_VULKAN
@ -955,7 +948,6 @@ $(LIB_GGML_S): \
src/whisper.o: \
src/whisper.cpp \
src/whisper-mel.hpp \
include/whisper.h \
ggml/include/ggml.h \
ggml/include/ggml-alloc.h \

1
bindings/ruby/ext/extconf.rb
View File

@ -1,7 +1,6 @@
require 'mkmf'
system("cp #{File.join(File.dirname(__FILE__),'..','..','..','whisper.cpp')} .")
system("cp #{File.join(File.dirname(__FILE__),'..','..','..','whisper.h')} .")
system("cp #{File.join(File.dirname(__FILE__),'..','..','..','whisper-mel.hpp')} .")
system("cp #{File.join(File.dirname(__FILE__),'..','..','..','ggml.h')} .")
system("cp #{File.join(File.dirname(__FILE__),'..','..','..','ggml.c')} .")
system("cp #{File.join(File.dirname(__FILE__),'..','..','..','ggml-impl.h')} .")

30
src/CMakeLists.txt
View File

@ -78,43 +78,13 @@ 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()
# whisper
add_library(whisper
../include/whisper.h
whisper.cpp
whisper-mel.hpp
)
# 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
VERSION ${PROJECT_VERSION}

363
src/whisper-mel-cuda.cu
View File

@ -1,363 +0,0 @@
#define CUB_IGNORE_DEPRECATED_CPP_DIALECT
#include "whisper-mel-cuda.hpp"
#include "whisper.h"
#include <ggml-backend.h>
#include <cuda.h>
#include <cuda_runtime.h>
#include <cufft.h>
#include <cublas_v2.h>
#include <cuComplex.h>
#include <cub/device/device_reduce.cuh>
#include <device_launch_parameters.h>
#include <algorithm>
#if defined(_MSC_VER)
#pragma warning(disable: 4324) // added padding
#endif
namespace {
static const char* cufftGetErrorString(cufftResult_t res) {
switch (res) {
case CUFFT_SUCCESS: return "The cuFFT operation was successful";
case CUFFT_INVALID_PLAN: return "cuFFT was passed an invalid plan handle";
case CUFFT_ALLOC_FAILED: return "cuFFT failed to allocate GPU or CPU memory";
case CUFFT_INVALID_TYPE: return "No longer used";
case CUFFT_INVALID_VALUE: return "User specified an invalid pointer or parameter";
case CUFFT_INTERNAL_ERROR: return "Driver or internal cuFFT library error";
case CUFFT_EXEC_FAILED: return "Failed to execute an FFT on the GPU";
case CUFFT_SETUP_FAILED: return "The cuFFT library failed to initialize";
case CUFFT_INVALID_SIZE: return "User specified an invalid transform size";
case CUFFT_UNALIGNED_DATA: return "No longer used";
case CUFFT_INCOMPLETE_PARAMETER_LIST: return "Missing parameters in call";
case CUFFT_INVALID_DEVICE: return "Execution of a plan was on different GPU than plan creation";
case CUFFT_PARSE_ERROR: return "Internal plan database error";
case CUFFT_NO_WORKSPACE: return "No workspace has been provided prior to plan execution";
case CUFFT_NOT_IMPLEMENTED: return "Function does not implement functionality for parameters given.";
case CUFFT_LICENSE_ERROR: return "Used in previous versions.";
case CUFFT_NOT_SUPPORTED: return "Operation is not supported for parameters given.";
default: return "Unknown error";
}
}
#define CUFFT_CHECK(err) CUDA_CHECK_GEN(err, CUFFT_SUCCESS, cufftGetErrorString)
__global__ void k_fill_stft_input(
const float * padded_samples,
const int n_frames,
const float * hann_window,
float * stft_in
) {
auto y = blockIdx.y * blockDim.y + threadIdx.y;
// if (y >= n_frames) return;
auto x = blockIdx.x * blockDim.x + threadIdx.x;
// if (x >= WHISPER_N_FFT) return;
auto line = padded_samples + y * WHISPER_HOP_LENGTH;
auto outLine = stft_in + y * WHISPER_N_FFT;
outLine[x] = line[x] * hann_window[x];
}
__global__ void k_calc_magnitudes(
const cuComplex * stft_out,
const int n_frames,
float * magnitudes
) {
auto y = blockIdx.y * blockDim.y + threadIdx.y;
// if (y >= n_frames) return;
auto x = blockIdx.x * blockDim.x + threadIdx.x;
// if (x >= WHISPER_N_FFT_HALF) return;
auto idx = y * WHISPER_N_FFT_HALF + x;
auto r = stft_out[idx].x;
auto i = stft_out[idx].y;
magnitudes[idx] = r * r + i * i;
}
__global__ void k_calc_log_mel(
const float * mel_data,
const int n_mel,
const float * max_val,
float * log_mel
) {
auto x = blockIdx.x * blockDim.x + threadIdx.x;
if (x >= n_mel) return;
float val = mel_data[x];
constexpr float e = 1e-10f;
if (val < e) val = e;
val = log10(val);
const float max = log10(*max_val) - 8.f;
if (val < max) val = max;
log_mel[x] = (val + 4) / 4;
}
static void fill_stft_input(
const float * padded_samples,
int n_frames,
const float * hann_window,
float * stft_in,
cudaStream_t stream
) {
dim3 block(WHISPER_N_FFT, 1);
dim3 grid(1, n_frames);
k_fill_stft_input<<<grid, block, 0, stream>>>(padded_samples, n_frames, hann_window, stft_in);
}
static void calc_magnitudes(
const cuComplex * stft_out,
int n_frames,
float * magnitudes,
cudaStream_t stream
) {
dim3 block(WHISPER_N_FFT_HALF, 1);
dim3 grid(1, n_frames);
k_calc_magnitudes<<<grid, block, 0, stream>>>(stft_out, n_frames, magnitudes);
}
constexpr auto LOG_MEL_PREFIX_SIZE = 256;
static void calc_log_mel(
const float * mel_data,
int n_mel,
void * tempStorage,
int tempStorageSize,
float * log_mel,
cudaStream_t stream
) {
float * max_val = reinterpret_cast<float *>(tempStorage);
void * maxTemp = reinterpret_cast<char*>(tempStorage) + LOG_MEL_PREFIX_SIZE;
size_t nbytes = size_t(tempStorageSize - LOG_MEL_PREFIX_SIZE);
cub::DeviceReduce::Max(maxTemp, nbytes, mel_data, max_val, n_mel, stream);
int block = 256;
int grid = (n_mel + block - 1) / block;
k_calc_log_mel<<<grid, block, 0, stream>>>(mel_data, n_mel, max_val, log_mel);
}
class mel_calc_cuda : public whisper_mel_calc {
const int m_n_mel;
ggml_backend_t m_backend = nullptr;
int m_device = -1;
cudaStream_t m_stream = nullptr;
cublasHandle_t m_cublas_handle = nullptr;
float * m_hann_window = nullptr;
float * m_filters = nullptr;
// max samples for which we have allocated memory for the temp working areas below (cufft, log_mel)
int m_n_max_samples = 0;
size_t m_cufft_workspace_size = 0;
void * m_cufft_workspace = nullptr;
size_t m_log_mel_temp_storage_size = 0;
void * m_log_mel_temp_storage = nullptr;
public:
mel_calc_cuda(ggml_backend_t backend, const whisper_filters & filters)
: m_n_mel(filters.n_mel)
, m_backend(backend)
{
ggml_backend_cuda_context* cuda_ctx = (ggml_backend_cuda_context*)m_backend->context;
m_device = cuda_ctx->device;
if (ggml_cuda_info().devices[m_device].cc < 600) {
// we've only tesed on 6.0 and higher and we've had reports of crashes on 5.0:
// https://github.com/ggerganov/whisper.cpp/issues/2230
// to be safe forbid anything below 6.0
throw std::runtime_error("CUDA compute capability 6.0 or higher is required");
}
ggml_cuda_set_device(m_device);
if (filters.n_fft != WHISPER_N_FFT_HALF) {
throw std::invalid_argument("MelFilters n_frames must be WHISPER_N_FFT_HALF");
}
assert(filters.data.size() == filters.n_mel * WHISPER_N_FFT_HALF);
CUDA_CHECK(cudaStreamCreate(&m_stream));
CUBLAS_CHECK(cublasCreate(&m_cublas_handle));
CUBLAS_CHECK(cublasSetMathMode(m_cublas_handle, CUBLAS_TF32_TENSOR_OP_MATH));
CUBLAS_CHECK(cublasSetStream(m_cublas_handle, m_stream));
// create Hann window
{
auto hw = whisper_mel_calc::hann_window();
CUDA_CHECK(cudaMallocAsync(&m_hann_window, hw.len * sizeof(float), m_stream));
CUDA_CHECK(cudaMemcpyAsync(m_hann_window, hw.data, hw.len * sizeof(float), cudaMemcpyHostToDevice, m_stream));
}
// fill filters
{
auto& f = filters.data;
CUDA_CHECK(cudaMallocAsync(&m_filters, f.size() * sizeof(float), m_stream));
CUDA_CHECK(cudaMemcpyAsync(m_filters, f.data(), f.size() * sizeof(float), cudaMemcpyHostToDevice, m_stream));
}
// preallocate working areas enough for the most common cases (<= 30s)
ensure_working_areas(WHISPER_N_SAMPLES);
}
~mel_calc_cuda() {
ggml_cuda_set_device(m_device);
CUDA_CHECK(cudaStreamSynchronize(m_stream));
CUDA_CHECK(cudaStreamDestroy(m_stream));
CUDA_CHECK(cudaFree(m_hann_window));
CUDA_CHECK(cudaFree(m_cufft_workspace));
CUDA_CHECK(cudaFree(m_filters));
CUDA_CHECK(cudaFree(m_log_mel_temp_storage));
}
void ensure_working_areas(int n_samples) {
if (n_samples <= m_n_max_samples) {
return;
}
const auto max_padded_samples = n_samples + WHISPER_N_SAMPLES + WHISPER_N_FFT;
const auto max_frames = 1 + (max_padded_samples - WHISPER_N_FFT) / WHISPER_HOP_LENGTH;
// cufft workspace
{
if (m_cufft_workspace) {
CUDA_CHECK(cudaFree(m_cufft_workspace));
m_cufft_workspace_size = 0;
m_cufft_workspace = nullptr;
}
CUFFT_CHECK(cufftEstimate1d(WHISPER_N_FFT, CUFFT_R2C, max_frames, &m_cufft_workspace_size));
CUDA_CHECK(cudaMallocAsync(&m_cufft_workspace, m_cufft_workspace_size, m_stream));
}
// device reduce working area
{
if (m_log_mel_temp_storage) {
CUDA_CHECK(cudaFree(m_log_mel_temp_storage));
m_log_mel_temp_storage_size = 0;
m_log_mel_temp_storage = nullptr;
}
const auto max_mels = 160;
size_t nbytes = 0;
float* temp = nullptr;
cub::DeviceReduce::Max(nullptr, nbytes, temp, temp, max_frames * max_mels);
m_log_mel_temp_storage_size = nbytes + LOG_MEL_PREFIX_SIZE;
CUDA_CHECK(cudaMallocAsync(&m_log_mel_temp_storage, m_log_mel_temp_storage_size, m_stream));
}
m_n_max_samples = n_samples;
}
virtual whisper_mel calculate(whisper_span<const float> samples, int /*n_threads*/) override {
ggml_cuda_set_device(m_device);
ensure_working_areas(samples.len);
const size_t mirror_pad = WHISPER_N_FFT / 2;
const size_t padded_size = samples.len + WHISPER_N_SAMPLES + WHISPER_N_FFT;
// pad
std::vector<float> padded_samples(padded_size);
std::reverse_copy(samples.data + 1, samples.data + 1 + mirror_pad, padded_samples.begin()); // reflect
std::copy(samples.data, samples.data + samples.len, padded_samples.begin() + mirror_pad); // copy
// fill the rest of the data
// it should canonically be mirrored at the end as well,
// but we just assume the last MEL_FRAME_SIZE/2 samples are zeros
std::fill(padded_samples.begin() + mirror_pad + samples.len, padded_samples.end(), 0.f);
const auto n_frames = 1 + (padded_samples.size() - WHISPER_N_FFT) / WHISPER_HOP_LENGTH;
float * cu_padded_samples = nullptr;
CUDA_CHECK(cudaMallocAsync(&cu_padded_samples, padded_samples.size() * sizeof(float), m_stream));
CUDA_CHECK(cudaMemcpyAsync(cu_padded_samples, padded_samples.data(), padded_samples.size() * sizeof(float), cudaMemcpyHostToDevice, m_stream));
float * stft_in = nullptr; // contiguous buffer for stft input
CUDA_CHECK(cudaMallocAsync(&stft_in, n_frames * WHISPER_N_FFT * sizeof(float), m_stream));
fill_stft_input(cu_padded_samples, int(n_frames), m_hann_window, stft_in, m_stream);
cufftComplex* stft_out;
CUDA_CHECK(cudaMallocAsync(&stft_out, n_frames * WHISPER_N_FFT_HALF * sizeof(cufftComplex), m_stream));
cufftHandle plan;
CUFFT_CHECK(cufftCreate(&plan));
CUFFT_CHECK(cufftSetAutoAllocation(plan, 0));
{
size_t waSize;
CUFFT_CHECK(cufftMakePlan1d(plan, WHISPER_N_FFT, CUFFT_R2C, int(n_frames), &waSize));
assert(waSize <= m_cufft_workspace_size);
CUFFT_CHECK(cufftSetWorkArea(plan, m_cufft_workspace));
CUFFT_CHECK(cufftSetStream(plan, m_stream));
}
CUFFT_CHECK(cufftExecR2C(plan, stft_in, stft_out));
const auto n_mag_frames = n_frames - 1; // drop last frame
float * magnitudes;
CUDA_CHECK(cudaMallocAsync(&magnitudes, n_mag_frames * WHISPER_N_FFT_HALF * sizeof(float), m_stream));
calc_magnitudes(stft_out, int(n_mag_frames), magnitudes, m_stream);
float * mel_data = nullptr;
CUDA_CHECK(cudaMallocAsync(&mel_data, m_n_mel * n_mag_frames * sizeof(float), m_stream));
const float fone = 1.0f, fzero = 0.0f;
CUBLAS_CHECK(cublasSgemm(m_cublas_handle, CUBLAS_OP_T, CUBLAS_OP_N,
int(n_mag_frames), m_n_mel, WHISPER_N_FFT_HALF,
&fone,
magnitudes, WHISPER_N_FFT_HALF,
m_filters, WHISPER_N_FFT_HALF,
&fzero,
mel_data, int(n_mag_frames)));
whisper_mel ret;
// Calculate semi-padded sample length to ensure compatibility
int n_len_org = 1 + int(samples.len + mirror_pad - WHISPER_N_FFT) / WHISPER_HOP_LENGTH;
whisper_mel_init(ret, m_backend, int(n_mag_frames), n_len_org, m_n_mel);
assert(ggml_nbytes(ret.tensor) == m_n_mel * n_mag_frames * sizeof(float));
float* log_mels = reinterpret_cast<float*>(ret.tensor->data);
calc_log_mel(
mel_data, int(m_n_mel * n_mag_frames),
m_log_mel_temp_storage , int(m_log_mel_temp_storage_size),
log_mels, m_stream);
CUDA_CHECK(cudaStreamSynchronize(m_stream));
// cleanup
CUFFT_CHECK(cufftDestroy(plan));
CUDA_CHECK(cudaFreeAsync(mel_data, m_stream));
CUDA_CHECK(cudaFreeAsync(magnitudes, m_stream));
CUDA_CHECK(cudaFreeAsync(stft_out, m_stream));
CUDA_CHECK(cudaFreeAsync(stft_in, m_stream));
CUDA_CHECK(cudaFreeAsync(cu_padded_samples, m_stream));
return ret;
}
};
}
whisper_mel_calc * whisper_mel_calc_create_cuda(ggml_backend_t backend, const whisper_filters & filters) {
try {
return new mel_calc_cuda(backend, filters);
}
catch (...) {
// TODO: log error (but for this we would have to expose the log state to be accessible here)
return nullptr;
}
}

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

@ -1,3 +0,0 @@
#include "whisper-mel.hpp"
whisper_mel_calc * whisper_mel_calc_create_cuda(ggml_backend_t backend, const whisper_filters & filters);

34
src/whisper-mel.hpp
View File

@ -1,34 +0,0 @@
#pragma once
#include "ggml-backend.h"
#include <vector>
struct whisper_mel {
int n_len_org = 0;
ggml_context * ctx = nullptr;
ggml_tensor * tensor = nullptr;
ggml_backend_buffer_t buffer = nullptr;
};
void whisper_mel_init(whisper_mel & mel, ggml_backend_t backend, int n_len, int n_len_org, int n_mel);
void whisper_mel_free(whisper_mel & mel);
struct whisper_filters {
int32_t n_mel;
int32_t n_fft;
std::vector<float> data;
};
template <typename T>
struct whisper_span {
T * data;
int len;
};
struct whisper_mel_calc {
virtual ~whisper_mel_calc();
virtual whisper_mel calculate(whisper_span<const float> samples, int n_threads) = 0;
static whisper_span<const float> hann_window();
};

379
src/whisper.cpp
View File

@ -10,7 +10,6 @@
#ifdef GGML_USE_CUDA
#include "ggml-cuda.h"
#include "whisper-mel-cuda.hpp"
#endif
#ifdef GGML_USE_SYCL
@ -37,8 +36,6 @@
#include "ggml-alloc.h"
#include "ggml-backend.h"
#include "whisper-mel.hpp"
#include <atomic>
#include <algorithm>
#include <cassert>
@ -401,6 +398,21 @@ static const std::map<whisper_alignment_heads_preset, whisper_aheads> g_aheads {
static std::vector<uint32_t> get_alignment_heads_by_layer(const whisper_context_params & cparams, int il, int32_t n_text_layer, int32_t n_head);
struct whisper_mel {
int n_len;
int n_len_org;
int n_mel;
std::vector<float> data;
};
struct whisper_filters {
int32_t n_mel;
int32_t n_fft;
std::vector<float> data;
};
struct whisper_vocab {
using id = int32_t;
using token = std::string;
@ -830,8 +842,6 @@ struct whisper_state {
whisper_kv_cache kv_pad;
whisper_mel mel;
whisper_mel_calc * mel_calc = nullptr;
whisper_mel_calc * mel_calc_fallback = nullptr;
whisper_batch batch;
@ -850,6 +860,7 @@ struct whisper_state {
struct ggml_tensor * embd_enc = nullptr;
// helpers for GPU offloading
std::vector<float> inp_mel;
std::vector<float> inp_mask;
// decode output (2-dimensional array: [n_tokens][n_vocab])
@ -1912,8 +1923,7 @@ static bool whisper_encode_external(const whisper_state & wstate) {
static struct ggml_cgraph * whisper_build_graph_conv(
whisper_context & wctx,
whisper_state & wstate,
const int mel_offset) {
whisper_state & wstate) {
const auto & model = wctx.model;
const auto & hparams = model.hparams;
@ -1932,35 +1942,9 @@ static struct ggml_cgraph * whisper_build_graph_conv(
ggml_cgraph * gf = ggml_new_graph(ctx0);
GGML_ASSERT(wstate.mel.tensor);
ggml_tensor * mel_inp = wstate.mel.tensor;
ggml_set_input(mel_inp);
ggml_tensor * mel;
if (ggml_nelements(mel_inp) > 0) {
const int n_len = int(mel_inp->ne[0]);
const int out_s = 2 * n_ctx;
const int i0 = std::min(mel_offset, n_len);
const int i1 = std::min(mel_offset + out_s, n_len);
const int mel_s = i1 - i0;
assert(mel_inp->type == GGML_TYPE_F32);
assert(mel_inp->ne[1] == n_mels);
ggml_tensor * cur = ggml_view_2d(ctx0, mel_inp, out_s, n_mels, mel_inp->nb[1], ggml_row_size(mel_inp->type, i0));
if (mel_s < out_s) {
mel = ggml_pad(ctx0, cur, out_s - mel_s, 0, 0, 0);
} else {
mel = ggml_cont(ctx0, cur);
}
} else {
// empty mel - just create a dummy tensor with the correct size
mel = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, 2*n_ctx, n_mels);
}
struct ggml_tensor * mel = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, 2*n_ctx, n_mels);
ggml_set_name(mel, "mel");
ggml_set_input(mel);
struct ggml_tensor * cur = nullptr;
@ -2332,21 +2316,45 @@ static bool whisper_encode_internal(
{
auto & sched = wstate.sched_conv.sched;
ggml_cgraph * gf = whisper_build_graph_conv(wctx, wstate, mel_offset);
ggml_cgraph * gf = whisper_build_graph_conv(wctx, wstate);
if (!ggml_backend_sched_alloc_graph(sched, gf)) {
// should never happen as we pre-allocate the memory
return false;
}
if (!ggml_graph_compute_helper(sched, gf, n_threads)) {
return false;
struct ggml_tensor * mel = ggml_graph_get_tensor(gf, "mel");
// set the input
{
const auto & mel_inp = wstate.mel;
const int n_ctx = wstate.exp_n_audio_ctx > 0 ? wstate.exp_n_audio_ctx : wctx.model.hparams.n_audio_ctx;
assert(mel->type == GGML_TYPE_F32);
assert(mel_inp.n_mel == wctx.model.hparams.n_mels);
wstate.inp_mel.resize(ggml_nelements(mel));
float * dst = wstate.inp_mel.data();
memset(dst, 0, ggml_nbytes(mel));
const int i0 = std::min(mel_offset, mel_inp.n_len);
const int i1 = std::min(mel_offset + 2*n_ctx, mel_inp.n_len);
for (int j = 0; j < mel_inp.n_mel; ++j) {
for (int i = i0; i < i1; ++i) {
dst[j*2*n_ctx + (i - i0)] = mel_inp.data[j*mel_inp.n_len + i];
}
}
ggml_backend_tensor_set(mel, wstate.inp_mel.data(), 0, ggml_nelements(mel)*sizeof(float));
}
if (whisper_encode_external(wstate)) {
ggml_tensor * mel = ggml_graph_get_tensor(gf, "mel");
assert(mel->ne[1] == wctx.model.hparams.n_mels);
GGML_UNUSED(mel);
if (!whisper_encode_external(wstate)) {
if (!ggml_graph_compute_helper(sched, gf, n_threads)) {
return false;
}
} else {
#if defined(WHISPER_USE_COREML)
whisper_coreml_encode(wstate.ctx_coreml, mel->ne[0], mel->ne[1], (float *) mel->data, (float *) wstate.embd_enc->data);
#elif defined(WHISPER_USE_OPENVINO)
@ -2970,35 +2978,6 @@ struct whisper_global_cache {
} global_cache;
}
// Mel spectrogram
void whisper_mel_init(whisper_mel & mel, ggml_backend_t backend, int n_len, int n_len_org, int n_mel) {
//WHISPER_LOG_INFO("%s: n_len = %d, n_len_org = %d, n_mel = %d\n", __func__, n_len, n_len_org, n_mel);
mel.n_len_org = n_len_org;
assert(!mel.ctx);
mel.ctx = ggml_init({ggml_tensor_overhead(), nullptr, true});
mel.tensor = ggml_new_tensor_2d(mel.ctx, GGML_TYPE_F32, n_len, n_mel);
mel.buffer = ggml_backend_alloc_buffer(backend, ggml_nbytes(mel.tensor) + ggml_backend_get_alignment(backend));
auto alloc = ggml_tallocr_new(mel.buffer);
ggml_tallocr_alloc(&alloc, mel.tensor);
}
void whisper_mel_free(whisper_mel & mel) {
ggml_free(mel.ctx);
ggml_backend_buffer_free(mel.buffer);
mel.n_len_org = 0;
mel.ctx = nullptr;
mel.tensor = nullptr;
mel.buffer = nullptr;
}
whisper_mel_calc::~whisper_mel_calc() = default; // export vtable
whisper_span<const float> whisper_mel_calc::hann_window() {
return {global_cache.hann_window, WHISPER_N_FFT};
}
// naive Discrete Fourier Transform
// input is real-valued
// output is complex-valued
@ -3068,22 +3047,12 @@ static void fft(float* in, int N, float* out) {
}
}
namespace {
struct whisper_mel_data {
int n_len;
int n_len_org;
int n_mel;
float * data;
};
void log_mel_spectrogram_worker_thread(int ith, const float * hann, const std::vector<float> & samples,
int n_samples, int n_threads,
const whisper_filters & filters, whisper_mel_data & mel) {
const auto frame_size = WHISPER_N_FFT;
const auto frame_step = WHISPER_HOP_LENGTH;
static void log_mel_spectrogram_worker_thread(int ith, const float * hann, const std::vector<float> & samples,
int n_samples, int frame_size, int frame_step, int n_threads,
const whisper_filters & filters, whisper_mel & mel) {
std::vector<float> fft_in(frame_size * 2, 0.0);
std::vector<float> fft_out(frame_size * 2 * 2 * 2);
int n_fft = filters.n_fft;
int i = ith;
@ -3098,6 +3067,7 @@ void log_mel_spectrogram_worker_thread(int ith, const float * hann, const std::v
for (int j = 0; j < std::min(frame_size, n_samples - offset); j++) {
fft_in[j] = hann[j] * samples[offset + j];
}
// fill the rest with zeros
if (n_samples - offset < frame_size) {
std::fill(fft_in.begin() + (n_samples - offset), fft_in.end(), 0.0);
@ -3115,7 +3085,6 @@ void log_mel_spectrogram_worker_thread(int ith, const float * hann, const std::v
// mel spectrogram
for (int j = 0; j < mel.n_mel; j++) {
double sum = 0.0;
// unroll loop (suggested by GH user @lunixbochs)
int k = 0;
for (k = 0; k < n_fft - 3; k += 4) {
@ -3125,14 +3094,11 @@ void log_mel_spectrogram_worker_thread(int ith, const float * hann, const std::v
fft_out[k + 2] * filters.data[j * n_fft + k + 2] +
fft_out[k + 3] * filters.data[j * n_fft + k + 3];
}
// handle n_fft remainder
for (; k < n_fft; k++) {
sum += fft_out[k] * filters.data[j * n_fft + k];
}
sum = log10(std::max(sum, 1e-10));
mel.data[j * mel.n_len + i] = sum;
}
}
@ -3146,116 +3112,97 @@ void log_mel_spectrogram_worker_thread(int ith, const float * hann, const std::v
}
}
struct mel_calc_cpu : public whisper_mel_calc {
ggml_backend_t m_backend;
const whisper_filters & m_filters;
mel_calc_cpu(ggml_backend_t backend, const whisper_filters & filters) : m_backend(backend), m_filters(filters) {}
// ref: https://github.com/openai/whisper/blob/main/whisper/audio.py#L110-L157
static bool log_mel_spectrogram(
whisper_state & wstate,
const float * samples,
const int n_samples,
const int /*sample_rate*/,
const int frame_size,
const int frame_step,
const int n_mel,
const int n_threads,
const whisper_filters & filters,
const bool debug,
whisper_mel & mel) {
const int64_t t_start_us = ggml_time_us();
// ref: https://github.com/openai/whisper/blob/main/whisper/audio.py#L110-L157
whisper_mel calculate(whisper_span<const float> ssamples, int n_threads) override {
// Hann window
const float * hann = global_cache.hann_window;
// Hann window
WHISPER_ASSERT(frame_size == WHISPER_N_FFT && "Unsupported frame_size");
const float * hann = global_cache.hann_window;
// Calculate the length of padding
int64_t stage_1_pad = WHISPER_SAMPLE_RATE * 30;
int64_t stage_2_pad = WHISPER_N_FFT / 2;
// Calculate the length of padding
int64_t stage_1_pad = WHISPER_SAMPLE_RATE * 30;
int64_t stage_2_pad = frame_size / 2;
const int n_samples = int(ssamples.len);
const float * samples = ssamples.data;
// Initialize a vector and copy data from C array to it.
std::vector<float> samples_padded;
samples_padded.resize(n_samples + stage_1_pad + stage_2_pad * 2);
std::copy(samples, samples + n_samples, samples_padded.begin() + stage_2_pad);
// Initialize a vector and copy data from C array to it.
std::vector<float> samples_padded;
samples_padded.resize(n_samples + stage_1_pad + stage_2_pad * 2);
std::copy(samples, samples + n_samples, samples_padded.begin() + stage_2_pad);
// pad 30 seconds of zeros at the end of audio (480,000 samples) + reflective pad 200 samples at the end of audio
std::fill(samples_padded.begin() + n_samples + stage_2_pad, samples_padded.begin() + n_samples + stage_1_pad + 2 * stage_2_pad, 0);
// pad 30 seconds of zeros at the end of audio (480,000 samples) + reflective pad 200 samples at the end of audio
std::fill(samples_padded.begin() + n_samples + stage_2_pad, samples_padded.begin() + n_samples + stage_1_pad + 2 * stage_2_pad, 0);
// reflective pad 200 samples at the beginning of audio
std::reverse_copy(samples + 1, samples + 1 + stage_2_pad, samples_padded.begin());
// reflective pad 200 samples at the beginning of audio
std::reverse_copy(samples + 1, samples + 1 + stage_2_pad, samples_padded.begin());
mel.n_mel = n_mel;
// https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/native/SpectralOps.cpp#L936
// Calculate number of frames + remove the last frame
mel.n_len = (samples_padded.size() - frame_size) / frame_step;
// Calculate semi-padded sample length to ensure compatibility
mel.n_len_org = 1 + (n_samples + stage_2_pad - frame_size) / frame_step;
mel.data.resize(mel.n_mel * mel.n_len);
whisper_mel_data mel;
mel.n_mel = m_filters.n_mel;
// https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/native/SpectralOps.cpp#L936
// Calculate number of frames + remove the last frame
mel.n_len = (samples_padded.size() - WHISPER_N_FFT) / WHISPER_HOP_LENGTH;
// Calculate semi-padded sample length to ensure compatibility
mel.n_len_org = 1 + (n_samples + stage_2_pad - WHISPER_N_FFT) / WHISPER_HOP_LENGTH;
std::vector<float> host_mel_data;
whisper_mel ret;
whisper_mel_init(ret, m_backend, mel.n_len, mel.n_len_org, mel.n_mel);
if (ggml_backend_buffer_is_host(ret.buffer)) {
mel.data = reinterpret_cast<float*>(ret.tensor->data);
} else {
host_mel_data.resize(mel.n_len * mel.n_mel);
mel.data = host_mel_data.data();
{
std::vector<std::thread> workers(n_threads - 1);
for (int iw = 0; iw < n_threads - 1; ++iw) {
workers[iw] = std::thread(
log_mel_spectrogram_worker_thread, iw + 1, hann, samples_padded,
n_samples + stage_2_pad, frame_size, frame_step, n_threads,
std::cref(filters), std::ref(mel));
}
{
std::vector<std::thread> workers(n_threads - 1);
for (int iw = 0; iw < n_threads - 1; ++iw) {
workers[iw] = std::thread(
log_mel_spectrogram_worker_thread, iw + 1, hann, samples_padded,
n_samples + stage_2_pad, n_threads,
std::cref(m_filters), std::ref(mel));
}
// main thread
log_mel_spectrogram_worker_thread(0, hann, samples_padded, n_samples + stage_2_pad, frame_size, frame_step, n_threads, filters, mel);
// main thread
log_mel_spectrogram_worker_thread(0, hann, samples_padded, n_samples + stage_2_pad, n_threads, m_filters, mel);
for (int iw = 0; iw < n_threads - 1; ++iw) {
workers[iw].join();
}
}
// clamping and normalization
double mmax = -1e20;
for (int i = 0; i < mel.n_mel*mel.n_len; i++) {
if (mel.data[i] > mmax) {
mmax = mel.data[i];
}
}
mmax -= 8.0;
for (int i = 0; i < mel.n_mel*mel.n_len; i++) {
if (mel.data[i] < mmax) {
mel.data[i] = mmax;
}
mel.data[i] = (mel.data[i] + 4.0)/4.0;
}
if (!host_mel_data.empty()) {
// the ret buffer is not host-accessible so we used this temporary buffer and now we need to upload it
ggml_backend_tensor_set(ret.tensor, host_mel_data.data(), 0, ggml_nbytes(ret.tensor));
}
return ret;
}
};
}
static whisper_mel_calc * whisper_mel_calc_create(ggml_backend_t backend, const whisper_filters & filters) {
// 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) {
// run a warmup to avoid the first kernel launch overhead (thus we get the best perf even on the first run)
const float warmup[256] = { 0 };
ret->calculate({ warmup, 256 }, 1);
return ret;
for (int iw = 0; iw < n_threads - 1; ++iw) {
workers[iw].join();
}
}
#endif
// a specialized mel_calc could not be created
// fall back to CPU
return new mel_calc_cpu(backend, filters);
// clamping and normalization
double mmax = -1e20;
for (int i = 0; i < mel.n_mel*mel.n_len; i++) {
if (mel.data[i] > mmax) {
mmax = mel.data[i];
}
}
mmax -= 8.0;
for (int i = 0; i < mel.n_mel*mel.n_len; i++) {
if (mel.data[i] < mmax) {
mel.data[i] = mmax;
}
mel.data[i] = (mel.data[i] + 4.0)/4.0;
}
wstate.t_mel_us += ggml_time_us() - t_start_us;
// Dump log_mel_spectrogram
if (debug) {
std::ofstream outFile("log_mel_spectrogram.json");
outFile << "[";
for (uint64_t i = 0; i < mel.data.size() - 1; i++) {
outFile << mel.data[i] << ", ";
}
outFile << mel.data[mel.data.size() - 1] << "]";
outFile.close();
}
return true;
}
// split text into tokens
@ -3380,17 +3327,6 @@ struct whisper_state * whisper_init_state(whisper_context * ctx) {
return nullptr;
}
state->mel_calc = whisper_mel_calc_create(state->backends[0], ctx->model.filters);
// init 60s of random mel data
{
const int n_len = 2*100*WHISPER_CHUNK_SIZE;
const int n_mel = ctx->model.filters.n_mel;
whisper_mel_free(state->mel);
whisper_mel_init(state->mel, state->backends[0], n_len, n_len, n_mel);
}
// at this point, we don't know yet how many decoders will be used
// later during decoding, if more decoders are used, we will recreate the KV cache respectively
state->kv_self_n_dec = 1;
@ -3483,7 +3419,7 @@ struct whisper_state * whisper_init_state(whisper_context * ctx) {
{
bool ok = whisper_sched_graph_init(state->sched_conv, state->backends,
[&]() {
return whisper_build_graph_conv(*ctx, *state, 0);
return whisper_build_graph_conv(*ctx, *state);
});
if (!ok) {
@ -3805,13 +3741,6 @@ void whisper_free_state(struct whisper_state * state) {
whisper_kv_cache_free(state->kv_cross);
whisper_kv_cache_free(state->kv_pad);
whisper_mel_free(state->mel);
delete state->mel_calc;
state->mel_calc = nullptr;
delete state->mel_calc_fallback;
state->mel_calc_fallback = nullptr;
#ifdef WHISPER_USE_COREML
if (state->ctx_coreml != nullptr) {
whisper_coreml_free(state->ctx_coreml);
@ -3869,37 +3798,11 @@ void whisper_free_params(struct whisper_full_params * params) {
}
int whisper_pcm_to_mel_with_state(struct whisper_context * ctx, struct whisper_state * state, const float * samples, int n_samples, int n_threads) {
const int64_t t_start_us = ggml_time_us();
whisper_mel_free(state->mel);
if (n_samples <= 5 * 60 * WHISPER_SAMPLE_RATE) {
// calculate mel spectrogram for lengths up to 5 minutes on the most optimal mel calculator
state->mel = state->mel_calc->calculate({samples, n_samples}, n_threads);
} else {
// calcuate mel spectrogram for longer audios on the CPU
// 1. gpu calculations may use hundreds of megabytes of memory for longer audios so we're being conservative
// with our gpu demands
// 2. the time to transcribe audios this long will be dominated by the decoding time, so the mel calculation
// taking longer is not a major concern
if (!state->mel_calc_fallback) {
state->mel_calc_fallback = new mel_calc_cpu(state->backends[0], ctx->model.filters);
}
state->mel = state->mel_calc_fallback->calculate({samples, n_samples}, n_threads);
if (!log_mel_spectrogram(*state, samples, n_samples, WHISPER_SAMPLE_RATE, WHISPER_N_FFT, WHISPER_HOP_LENGTH, ctx->model.filters.n_mel, n_threads, ctx->model.filters, false, state->mel)) {
WHISPER_LOG_ERROR("%s: failed to compute mel spectrogram\n", __func__);
return -1;
}
state->t_mel_us += ggml_time_us() - t_start_us;
// Dump log_mel_spectrogram
//{
// auto& mel = state->mel;
// std::ofstream outFile("log_mel_spectrogram.json");
// outFile << "[";
// for (uint64_t i = 0; i < mel.data.size() - 1; i++) {
// outFile << mel.data[i] << ", ";
// }
// outFile << mel.data[mel.data.size() - 1] << "]";
// outFile.close();
//}
return 0;
}
@ -3918,10 +3821,12 @@ int whisper_set_mel_with_state(
return -1;
}
whisper_mel_free(state->mel);
whisper_mel_init(state->mel, state->backends[0], n_len, n_len, n_mel);
state->mel.n_len = n_len;
state->mel.n_len_org = n_len;
state->mel.n_mel = n_mel;
ggml_backend_tensor_set(state->mel.tensor, data, 0, ggml_nbytes(state->mel.tensor));
state->mel.data.resize(n_len*n_mel);
memcpy(state->mel.data.data(), data, n_len*n_mel*sizeof(float));
return 0;
}