sean-m
/
kendryte-freertos-sdk
mirror of https://github.com/kendryte/kendryte-freertos-sdk.git
1
0
Fork 0
Code Issues Releases Wiki Activity
kendryte-freertos-sdk/lib/bsp/device/kpu.cpp

1073 lines
38 KiB
C++
Executable File
Raw Blame History

/* Copyright 2018 Canaan Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include <stdlib.h>
#include <string.h>
#include <FreeRTOS.h>
#include <task.h>
#include <dmac.h>
#include <hal.h>
#include <kernel/driver_impl.hpp>
#include <kpu.h>
#include <sysctl.h>
#include <math.h>
#include <float.h>
#include <time.h>
#include <sys/time.h>
#include <assert.h>
#include <iomem.h>
#include <printf.h>
using namespace sys;
#define KPU_DEBUG 0
#define NNCASE_DEBUG 0
#define USE_CACHED_AI_RAM 0
#define min(a, b) (((a) < (b)) ? (a) : (b))
#define max(a, b) (((a) > (b)) ? (a) : (b))
#define ALIGN_UP(x, align) ((x + (align - 1)) & (~(align - 1)))
#define COMMON_ENTRY \
semaphore_lock locker(free_mutex_);
class k_model_context : public heap_object, public free_object_access
{
public:
k_model_context(uint8_t *buffer)
{
#if FIX_CACHE
configASSERT(is_memory_cache((uintptr_t)buffer));
#endif
uintptr_t base_addr = (uintptr_t)buffer;
const kpu_model_header_t *header = (const kpu_model_header_t *)buffer;
if (header->version == 3 && header->arch == 0)
{
model_buffer_ = buffer;
output_count_ = header->output_count;
outputs_ = (const kpu_model_output_t *)(base_addr + sizeof(kpu_model_header_t));
layer_headers_ = (const kpu_model_layer_header_t *)((uintptr_t)outputs_ + sizeof(kpu_model_output_t) * output_count_);
layers_length_ = header->layers_length;
body_start_ = (const uint8_t *)((uintptr_t)layer_headers_ + sizeof(kpu_model_layer_header_t) * header->layers_length);
uint32_t body_size = 0;
for(int i=0; i<layers_length_; i++)
{
const kpu_model_layer_header_t *cnt_layer_header = layer_headers_ + i;
body_size += cnt_layer_header->body_size;
}
uint8_t *body_start_iomem = (uint8_t *)((uintptr_t)body_start_ - IOMEM);
const uint8_t *body_start_cache = body_start_;
memcpy(body_start_iomem, body_start_cache, body_size);
storage_ = std::make_unique<uint8_t[]>(header->main_mem_usage);
main_buffer_ = { storage_.get(), ptrdiff_t(header->main_mem_usage) };
}
else
{
throw std::runtime_error("Cannot load kmodel.");
}
}
void get(kpu_model_context_t *ctx)
{
ctx->body_start = body_start_;
ctx->model_buffer = model_buffer_;
ctx->main_buffer = main_buffer_.data();
ctx->layer_headers = layer_headers_;
ctx->layers_length = layers_length_;
ctx->output_count = output_count_;
ctx->outputs = outputs_;
}
private:
const uint8_t *model_buffer_;
const kpu_model_layer_header_t *layer_headers_;
const uint8_t *body_start_;
uint32_t layers_length_;
uint32_t output_count_;
const kpu_model_output_t * outputs_;
gsl::span<uint8_t> main_buffer_;
std::unique_ptr<uint8_t[]> storage_;
};
class k_kpu_driver : public kpu_driver, public static_object, public free_object_access
{
public:
k_kpu_driver(uintptr_t base_addr, sysctl_clock_t clock, sysctl_dma_select_t dma_req)
: kpu_(*reinterpret_cast<volatile kpu_config_t *>(base_addr)), clock_(clock), dma_req_(dma_req)
{
completion_event_ = xSemaphoreCreateBinary();
}
virtual void install() override
{
free_mutex_ = xSemaphoreCreateMutex();
sysctl_clock_disable(clock_);
}
virtual void on_first_open() override
{
sysctl_clock_enable(clock_);
dma_ch_ = dma_open_free();
}
virtual void on_last_close() override
{
sysctl_clock_disable(clock_);
dma_close(dma_ch_);
}
virtual handle_t model_load_from_buffer(uint8_t *buffer) override
{
return system_alloc_handle(make_accessor(make_object<k_model_context>(buffer)));
}
virtual int run(handle_t context, const uint8_t *src) override
{
COMMON_ENTRY;
auto model_context = system_handle_to_object(context).as<k_model_context>();
model_context->get(&ctx_);
ctx_.current_layer = 0;
ctx_.current_body = ctx_.body_start;
kpu_model_header_t *header = (kpu_model_header_t *)ctx_.model_buffer;
kpu_.interrupt_clear.reg = 7;
kpu_.fifo_threshold.reg = 0x1a;
kpu_.eight_bit_mode.reg = header->flags & 1;
kpu_.interrupt_mask.reg = 0b110;
pic_set_irq_priority(IRQN_AI_INTERRUPT, 2);
pic_set_irq_handler(IRQN_AI_INTERRUPT, kpu_isr_handle, this);
pic_set_irq_enable(IRQN_AI_INTERRUPT, 1);
const kpu_model_layer_header_t *first_layer_header = ctx_.layer_headers;
if (first_layer_header->type != KL_K210_CONV)
return -1;
const kpu_model_conv_layer_argument_t *first_layer = (const kpu_model_conv_layer_argument_t *)ctx_.body_start;
kpu_layer_argument_t layer_arg = *(kpu_layer_argument_t *)(ctx_.model_buffer + first_layer->layer_offset);
#if KPU_DEBUG
gettimeofday(&last_time_, NULL);
#endif
if ((layer_arg.image_size.data.i_row_wid + 1) % 64 != 0)
{
kpu_input_with_padding(&layer_arg, src);
xSemaphoreGive(completion_event_);
}
else
{
kpu_input_dma(&layer_arg, src);
}
while (!done_flag_)
{
if(xSemaphoreTake(completion_event_, 200) == pdTRUE)
{
if(mem_out_flag_)
{
memcpy(dest_kpu_, dest_io_, dest_len_);
mem_out_flag_ = 0;
}
if (ctx_.current_layer != ctx_.layers_length)
{
while(ai_step() == 1)
;
}
else
{
kpu_done();
}
}
}
done_flag_ = 0;
return 0;
}
virtual int get_output(handle_t context, uint32_t index, uint8_t **data, size_t *size) override
{
COMMON_ENTRY;
auto model_context = system_handle_to_object(context).as<k_model_context>();
model_context->get(&ctx_);
if (index >= ctx_.output_count)
return -1;
const kpu_model_output_t *output = ctx_.outputs + index;
*data = ctx_.main_buffer + output->address;
*size = output->size;
return 0;
}
private:
static void kpu_isr_handle(void *userdata)
{
auto &driver = *reinterpret_cast<k_kpu_driver *>(userdata);
driver.kpu_.interrupt_clear.reg = 0b111;
driver.kpu_.interrupt_mask.reg = 0b111;
BaseType_t xHigherPriorityTaskWoken = pdFALSE;
xSemaphoreGiveFromISR(driver.completion_event_, &xHigherPriorityTaskWoken);
if (xHigherPriorityTaskWoken)
{
portYIELD_FROM_ISR();
}
}
void kpu_flush_cache(uint32_t addr, size_t lines)
{
size_t line;
for (line = 0; line < lines; line++)
{
const uint64_t *src = (const uint64_t *)(AI_RAM_BASE_ADDR + (addr + line) * 64);
uint64_t *dest = (uint64_t *)(AI_IO_BASE_ADDR + (addr + line) * 64);
size_t i;
for (i = 0; i < 8; i++)
dest[i] = src[i];
}
}
void kpu_send_layer(const kpu_layer_argument_t *layer)
{
kpu_.layer_argument_fifo = layer->interrupt_enabe.reg;
kpu_.layer_argument_fifo = layer->image_addr.reg;
kpu_.layer_argument_fifo = layer->image_channel_num.reg;
kpu_.layer_argument_fifo = layer->image_size.reg;
kpu_.layer_argument_fifo = layer->kernel_pool_type_cfg.reg;
kpu_.layer_argument_fifo = layer->kernel_load_cfg.reg;
kpu_.layer_argument_fifo = layer->kernel_offset.reg;
kpu_.layer_argument_fifo = layer->kernel_calc_type_cfg.reg;
kpu_.layer_argument_fifo = layer->write_back_cfg.reg;
kpu_.layer_argument_fifo = layer->conv_value.reg;
kpu_.layer_argument_fifo = layer->conv_value2.reg;
kpu_.layer_argument_fifo = layer->dma_parameter.reg;
}
void kpu_upload_core(size_t width, size_t height, size_t channels, const uint8_t *src, uint32_t kpu_addr)
{
uint8_t *dest = (uint8_t *)AI_IO_BASE_ADDR + kpu_addr * 64;
size_t oc, y, x;
uint32_t row_padding;
uint32_t row_group;
uint32_t row_length;
if (width <= 16)
{
row_padding = 16;
row_group = 4;
row_length = 1;
}
else if (width <= 32)
{
row_padding = 32;
row_group = 2;
row_length = 1;
}
else
{
row_padding = 64;
row_group = 1;
row_length = (width + 63) / 64;
}
if ((uintptr_t)src % 8 == 0 && width % 8 == 0)
{
#define UPLOAD_BEGIN() \
for (oc = 0; oc < channels; oc++) \
{ \
uint8_t* channel_origin = dest + oc / row_group * row_length * height * 64 + oc % row_group * row_padding; \
for (y = 0; y < height; y++) \
{ \
uint64_t *y_origin = (uint64_t *)channel_origin + y * row_length * 64; \
#define UPLOAD_END() \
} \
}
width /= 8;
const uint64_t *u64_src = (const uint64_t *)src;
if (width == 1)
{
UPLOAD_BEGIN()
y_origin[0] = *u64_src++;
UPLOAD_END()
}
else if (width == 2)
{
UPLOAD_BEGIN()
{
y_origin[0] = *u64_src++;
y_origin[1] = *u64_src++;
}
UPLOAD_END()
}
else if (width == 4)
{
UPLOAD_BEGIN()
{
y_origin[0] = *u64_src++;
y_origin[1] = *u64_src++;
y_origin[2] = *u64_src++;
y_origin[3] = *u64_src++;
}
UPLOAD_END()
}
else
{
UPLOAD_BEGIN()
for (x = 0; x < width; x++)
y_origin[x] = *u64_src++;
UPLOAD_END()
}
}
else
{
for (oc = 0; oc < channels; oc++)
{
uint8_t *channel_origin = dest + oc / row_group * row_length * height * 64 + oc % row_group * row_padding;
for (y = 0; y < height; y++)
{
uint8_t *y_origin = channel_origin + y * row_length * 64;
for (x = 0; x < width; x++)
y_origin[x] = *src++;
}
}
}
}
void kpu_input_dma(const kpu_layer_argument_t *layer, const uint8_t *src)
{
uint64_t input_size = layer->kernel_calc_type_cfg.data.channel_switch_addr * 64 * (layer->image_channel_num.data.i_ch_num + 1);
dma_set_request_source(dma_ch_, dma_req_);
dma_transmit_async(dma_ch_, src, (void *)(uintptr_t)((uint8_t *)AI_IO_BASE_ADDR + layer->image_addr.data.image_src_addr * 64), 1, 1, sizeof(uint64_t), input_size / 8, 16, completion_event_);
}
void kpu_input_with_padding(const kpu_layer_argument_t *layer, const uint8_t *src)
{
size_t width = layer->image_size.data.i_row_wid + 1;
size_t height = layer->image_size.data.i_col_high + 1;
size_t channels = layer->image_channel_num.data.i_ch_num + 1;
kpu_upload_core(width, height, channels, src, layer->image_addr.data.image_src_addr);
}
void kpu_add(const kpu_model_add_layer_argument_t *arg)
{
const float *src_a = (const float *)(ctx_.main_buffer + arg->main_mem_in_a_address);
const float *src_b = (const float *)(ctx_.main_buffer + arg->main_mem_in_b_address);
float *dest = (float *)(ctx_.main_buffer + arg->main_mem_out_address);
size_t i, count = arg->count;
for (i = 0; i < count; i++)
dest[i] = src_a[i] + src_b[i];
}
void kpu_quantized_add(const kpu_model_quant_add_layer_argument_t *arg)
{
const uint8_t *src_a = (const uint8_t *)(ctx_.main_buffer + arg->main_mem_in_a_address);
const uint8_t *src_b = (const uint8_t *)(ctx_.main_buffer + arg->main_mem_in_b_address);
size_t count = ALIGN_UP(arg->count, 8) / 8;
int64_t off_a = arg->in_a_offset, mul_a = arg->in_a_mul, sh_a = arg->in_a_shift;
int64_t off_b = arg->in_b_offset, mul_b = arg->in_b_mul, sh_b = arg->in_b_shift;
int64_t off_o = arg->out_offset, mul_o = arg->out_mul, sh_o = arg->out_shift;
uint8_t *dest = (uint8_t *)(ctx_.main_buffer + arg->main_mem_out_address);
size_t i;
if (sh_a == sh_b)
{
#define QADD_UNROLL_1(x) \
int64_t a##x = *src_a++; \
int64_t b##x = *src_b++;
#define QADD_UNROLL_2(x) \
a##x += off_a; \
b##x += off_b;
#define QADD_UNROLL_3(x) \
a##x *= mul_a; \
b##x *= mul_b;
#define QADD_UNROLL_4(x) \
int64_t v##x = a##x + b##x;
#define QADD_UNROLL_5(x) \
v##x >>= sh_a;
#define QADD_UNROLL_6(x) \
v##x *= mul_o;
#define QADD_UNROLL_7(x) \
v##x >>= sh_o;
#define QADD_UNROLL_8(x) \
v##x += off_o;
#define QADD_UNROLL_9(x) \
v##x = min(0xFF, max(0, v##x));
#define QADD_UNROLL_10(x) \
*dest++ = v##x;
#define QADD_UNROLL_S(x) \
QADD_UNROLL_##x(0) \
QADD_UNROLL_##x(1) \
QADD_UNROLL_##x(2) \
QADD_UNROLL_##x(3) \
QADD_UNROLL_##x(4) \
QADD_UNROLL_##x(5) \
QADD_UNROLL_##x(6) \
QADD_UNROLL_##x(7)
for (i = 0; i < count; i++)
{
QADD_UNROLL_S(1);
QADD_UNROLL_S(2);
QADD_UNROLL_S(3);
QADD_UNROLL_S(4);
QADD_UNROLL_S(5);
QADD_UNROLL_S(6);
QADD_UNROLL_S(7);
QADD_UNROLL_S(8);
QADD_UNROLL_S(9);
QADD_UNROLL_S(10);
}
}
else
{
#undef QADD_UNROLL_1
#define QADD_UNROLL_1(x) \
int64_t a##x = *src_a++; \
int64_t b##x = *src_b++;
#undef QADD_UNROLL_2
#define QADD_UNROLL_2(x) \
a##x += off_a; \
b##x += off_b;
#undef QADD_UNROLL_3
#define QADD_UNROLL_3(x) \
a##x *= mul_a; \
b##x *= mul_b;
#undef QADD_UNROLL_4
#define QADD_UNROLL_4(x) \
a##x >>= sh_a; \
b##x >>= sh_b;
#undef QADD_UNROLL_5
#define QADD_UNROLL_5(x) \
int64_t v##x = a##x + b##x;
#undef QADD_UNROLL_6
#define QADD_UNROLL_6(x) \
v##x *= mul_o;
#undef QADD_UNROLL_7
#define QADD_UNROLL_7(x) \
v##x >>= sh_o;
#undef QADD_UNROLL_8
#define QADD_UNROLL_8(x) \
v##x += off_o;
#undef QADD_UNROLL_9
#define QADD_UNROLL_9(x) \
v##x = min(0xFF, max(0, v##x));
#undef QADD_UNROLL_10
#define QADD_UNROLL_10(x) \
*dest++ = v##x;
#undef QADD_UNROLL_S
#define QADD_UNROLL_S(x) \
QADD_UNROLL_##x(0) \
QADD_UNROLL_##x(1) \
QADD_UNROLL_##x(2) \
QADD_UNROLL_##x(3) \
QADD_UNROLL_##x(4) \
QADD_UNROLL_##x(5) \
QADD_UNROLL_##x(6) \
QADD_UNROLL_##x(7)
for (i = 0; i < count; i++)
{
QADD_UNROLL_S(1);
QADD_UNROLL_S(2);
QADD_UNROLL_S(3);
QADD_UNROLL_S(4);
QADD_UNROLL_S(5);
QADD_UNROLL_S(6);
QADD_UNROLL_S(7);
QADD_UNROLL_S(8);
QADD_UNROLL_S(9);
QADD_UNROLL_S(10);
}
}
}
void kpu_global_average_pool2d(const kpu_model_gap2d_layer_argument_t *arg)
{
const float *src = (const float *)(ctx_.main_buffer + arg->main_mem_in_address);
float *dest = (float *)(ctx_.main_buffer + arg->main_mem_out_address);
size_t oc, channels = arg->channels, kernel_size = arg->kernel_size;
for (oc = 0; oc < channels; oc++)
{
float sum = 0.f;
size_t i;
for (i = 0; i < kernel_size; i++)
sum += *src++;
dest[oc] = sum / kernel_size;
}
}
void kpu_quantized_max_pool2d(const kpu_model_quant_max_pool2d_layer_argument_t *arg)
{
const uint8_t *src = (const uint8_t *)(ctx_.main_buffer + arg->main_mem_in_address);
uint8_t *dest = (uint8_t *)(ctx_.main_buffer + arg->main_mem_out_address);
kpu_model_shape_t in_shape = arg->in_shape, out_shape = arg->out_shape;
uint32_t kernel_width = arg->kernel_width, kernel_height = arg->kernel_height;
uint32_t stride_width = arg->stride_width, stride_height = arg->stride_height;
uint32_t padding_width = arg->padding_width, padding_height = arg->padding_height;
uint32_t out_y, out_x, oc;
for (oc = 0; oc < out_shape.channels; oc++)
{
const uint8_t *channel_src = src + in_shape.width * in_shape.height * oc;
for (out_y = 0; out_y < out_shape.height; out_y++)
{
for (out_x = 0; out_x < out_shape.width; out_x++)
{
int32_t in_x_origin = (int32_t)(out_x * stride_width) - padding_width;
int32_t in_y_origin = (int32_t)(out_y * stride_height) - padding_height;
int32_t kernel_x_start = max(0, -in_x_origin);
int32_t kernel_x_end = min(kernel_width, in_shape.width - in_x_origin);
int32_t kernel_y_start = max(0, -in_y_origin);
int32_t kernel_y_end = min(kernel_height, in_shape.height - in_y_origin);
uint8_t value = 0;
int32_t kernel_y, kernel_x;
for (kernel_y = kernel_y_start; kernel_y < kernel_y_end; kernel_y++)
{
for (kernel_x = kernel_x_start; kernel_x < kernel_x_end; kernel_x++)
{
int32_t in_x = in_x_origin + kernel_x;
int32_t in_y = in_y_origin + kernel_y;
value = max(value, channel_src[in_y * in_shape.width + in_x]);
}
}
*dest++ = value;
}
}
}
}
void kpu_average_pool2d(const kpu_model_ave_pool2d_layer_argument_t *arg)
{
const float *src = (const float *)(ctx_.main_buffer + arg->main_mem_in_address);
float *dest = (float *)(ctx_.main_buffer + arg->main_mem_out_address);
kpu_model_shape_t in_shape = arg->in_shape, out_shape = arg->out_shape;
uint32_t kernel_width = arg->kernel_width, kernel_height = arg->kernel_height;
uint32_t stride_width = arg->stride_width, stride_height = arg->stride_height;
uint32_t padding_width = arg->padding_width, padding_height = arg->padding_height;
uint32_t out_y, out_x, oc;
for (oc = 0; oc < out_shape.channels; oc++)
{
const float *channel_src = src + in_shape.width * in_shape.height * oc;
for (out_y = 0; out_y < out_shape.height; out_y++)
{
for (out_x = 0; out_x < out_shape.width; out_x++)
{
int32_t in_x_origin = (int32_t)(out_x * stride_width) - padding_width;
int32_t in_y_origin = (int32_t)(out_y * stride_height) - padding_height;
int32_t kernel_x_start = max(0, -in_x_origin);
int32_t kernel_x_end = min(kernel_width, in_shape.width - in_x_origin);
int32_t kernel_y_start = max(0, -in_y_origin);
int32_t kernel_y_end = min(kernel_height, in_shape.height - in_y_origin);
float value = 0;
float kernel_count = 0;
int32_t kernel_y, kernel_x;
for (kernel_y = kernel_y_start; kernel_y < kernel_y_end; kernel_y++)
{
for (kernel_x = kernel_x_start; kernel_x < kernel_x_end; kernel_x++)
{
int32_t in_x = in_x_origin + kernel_x;
int32_t in_y = in_y_origin + kernel_y;
value += channel_src[in_y * in_shape.width + in_x];
kernel_count++;
}
}
*dest++ = value / kernel_count;
}
}
}
}
void kpu_quantize(const kpu_model_quantize_layer_argument_t *arg)
{
size_t count = arg->count;
const float *src = (const float *)(ctx_.main_buffer + arg->main_mem_in_address);
kpu_model_quant_param_t q = arg->quant_param;
float scale = 1.f / q.scale;
uint8_t *dest = (uint8_t *)(ctx_.main_buffer + arg->mem_out_address);
size_t i;
for (i = 0; i < count; i++)
{
int value = (*src++ - q.bias) * scale;
if (value < 0) value = 0;
if (value > 0xFF) value = 0xFF;
*dest++ = (uint8_t)value;
}
}
void kpu_dequantize(const kpu_model_dequantize_layer_argument_t *arg)
{
const uint8_t *src = (const uint8_t *)(ctx_.main_buffer + arg->main_mem_in_address);
float *dest = (float *)(ctx_.main_buffer + arg->main_mem_out_address);
size_t oc, count = arg->count;
kpu_model_quant_param_t q = arg->quant_param;
for (oc = 0; oc < count; oc++)
dest[oc] = *src++ * q.scale + q.bias;
}
void kpu_requantize(const kpu_model_requantize_layer_argument_t *arg)
{
const uint8_t *src = (const uint8_t *)(ctx_.main_buffer + arg->main_mem_in_address);
uint8_t *dest = (uint8_t *)(ctx_.main_buffer + arg->main_mem_out_address);
size_t oc, count = ALIGN_UP(arg->count, 8) / 8;
const uint8_t *table = arg->table;
for (oc = 0; oc < count;)
{
dest[oc++] = table[*src++];
dest[oc++] = table[*src++];
dest[oc++] = table[*src++];
dest[oc++] = table[*src++];
dest[oc++] = table[*src++];
dest[oc++] = table[*src++];
dest[oc++] = table[*src++];
dest[oc++] = table[*src++];
}
}
void kpu_l2_normalization(const kpu_model_l2_norm_layer_argument_t *arg)
{
const float *src = (const float *)(ctx_.main_buffer + arg->main_mem_in_address);
float *dest = (float *)(ctx_.main_buffer + arg->main_mem_out_address);
size_t oc, channels = arg->channels;
float sum = 0.f;
const float epsilon = 1e-10f;
for (oc = 0; oc < channels; oc++)
sum += src[oc] * src[oc];
if (sum < epsilon)
sum = epsilon;
sum = 1.f / sqrtf(sum);
for (oc = 0; oc < channels; oc++)
dest[oc] = src[oc] * sum;
}
void kpu_softmax(const kpu_model_softmax_layer_argument_t *arg)
{
const float *src = (const float *)(ctx_.main_buffer + arg->main_mem_in_address);
float *dest = (float *)(ctx_.main_buffer + arg->main_mem_out_address);
size_t oc, channels = arg->channels;
float max = FLT_MIN;
for (oc = 0; oc < channels; oc++)
max = fmaxf(max, src[oc]);
float sum = 0.f;
for (oc = 0; oc < channels; oc++)
{
float value = expf(src[oc] - max);
sum += value;
dest[oc] = value;
}
for (oc = 0; oc < channels; oc++)
dest[oc] /= sum;
}
void kpu_concat(const kpu_model_concat_layer_argument_t *arg)
{
uint8_t *dest = (uint8_t *)(ctx_.main_buffer + arg->main_mem_out_address);
uint32_t count = arg->input_count, i;
for (i = 0; i < count; i++)
{
kpu_model_memory_range_t input = arg->inputs_mem[i];
const uint8_t *src = (const uint8_t *)(ctx_.main_buffer + input.start);
memcpy(dest, src, input.size);
dest += input.size;
}
}
void kpu_fully_connected(const kpu_model_fully_connected_layer_argument_t *arg)
{
const float *src = (const float *)(ctx_.main_buffer + arg->main_mem_in_address);
float *dest = (float *)(ctx_.main_buffer + arg->main_mem_out_address);
uint32_t in_channels = arg->in_channels, out_channels = arg->out_channels, ic, oc;
float *weights = (float *)malloc(out_channels * in_channels);
float *bias = (float *)malloc(out_channels);
memcpy(weights, arg->weights, out_channels * in_channels * sizeof(float));
memcpy(bias, arg->weights + in_channels * out_channels, out_channels * sizeof(float));
for (oc = 0; oc < out_channels; oc++)
{
const float *c_weights = weights + oc * in_channels;
float sum = 0.0f;
for (ic = 0; ic < in_channels; ic++)
sum += src[ic] * c_weights[ic];
dest[oc] = sum + bias[oc];
}
free(weights);
free(bias);
}
void kpu_tf_flatten(const kpu_model_tf_flatten_layer_argument_t *arg)
{
const float *src = (const float *)(ctx_.main_buffer + arg->main_mem_in_address);
float *dest = (float *)(ctx_.main_buffer + arg->main_mem_out_address);
kpu_model_shape_t in_shape = arg->shape;
uint32_t oc, oy, ox;
for (oy = 0; oy < in_shape.height; oy++)
for (ox = 0; ox < in_shape.width; ox++)
for (oc = 0; oc < in_shape.channels; oc++)
*dest++ = src[(oc * in_shape.height + oy) * in_shape.width + ox];
}
void kpu_resize_nearest_neighbor(const kpu_model_resize_nearest_neighbor_layer_argument_t *arg)
{
const float *src = (const float *)(ctx_.main_buffer + arg->main_mem_in_address);
float *dest = (float *)(ctx_.main_buffer + arg->main_mem_out_address);
kpu_model_shape_t in_shape = arg->in_shape;
uint32_t out_width = arg->out_width, out_height = arg->out_height;
uint32_t oc, oy, ox;
float height_scale = (float)in_shape.height / out_height;
float width_scale = (float)in_shape.width / out_width;
for (oc = 0; oc < in_shape.channels; oc++)
{
const float *channel_src = src + in_shape.width * in_shape.height * oc;
for (oy = 0; oy <out_height; oy++)
{
uint32_t in_y = (uint32_t)min(floorf(oy * height_scale), in_shape.height - 1);
const float *y_origin = channel_src + in_y * in_shape.width;
for (ox = 0; ox < out_width; ox++)
{
uint32_t in_x = (uint32_t)min(floorf(ox * width_scale), in_shape.width - 1);
*dest++ = y_origin[in_x];
}
}
}
}
void kpu_conv(const kpu_model_conv_layer_argument_t *arg)
{
volatile kpu_layer_argument_t layer = *(kpu_layer_argument_t *)(ctx_.model_buffer + arg->layer_offset);
layer.kernel_load_cfg.data.para_start_addr = (uintptr_t)(ctx_.model_buffer + arg->weights_offset) - IOMEM;
layer.kernel_pool_type_cfg.data.bwsx_base_addr = (uintptr_t)(ctx_.model_buffer + arg->bn_offset) - IOMEM;
layer.kernel_calc_type_cfg.data.active_addr = (uintptr_t)(ctx_.model_buffer + arg->act_offset) - IOMEM;
if (arg->flags & KLF_MAIN_MEM_OUT)
{
mem_out_flag_ = 1;
kpu_.interrupt_clear.reg = 0b111;
kpu_.interrupt_mask.reg = 0b111;
layer.dma_parameter.data.send_data_out = 1;
dma_set_request_source(dma_ch_, dma_req_);
dest_len_ = (layer.dma_parameter.data.dma_total_byte + 8) / 8 * sizeof(uint64_t);
dest_kpu_ = ctx_.main_buffer + arg->main_mem_out_address;
if(dest_len_ > max_len_)
{
max_len_ = dest_len_;
iomem_free(dest_io_);
dest_io_ = (uint8_t *)iomem_malloc(dest_len_);
}
dma_transmit_async(dma_ch_, (void *)(&kpu_.fifo_data_out), (void *)dest_io_, 0, 1, sizeof(uint64_t), (layer.dma_parameter.data.dma_total_byte + 8) / 8, 8, completion_event_);
}
else
{
kpu_.interrupt_clear.reg = 0b111;
kpu_.interrupt_mask.reg = 0b110;
layer.interrupt_enabe.data.int_en = 1;
}
kpu_send_layer((const kpu_layer_argument_t *)&layer);
}
void kpu_add_padding(const kpu_model_add_padding_layer_argument_t *arg)
{
const uint8_t *src = (const uint8_t *)(ctx_.main_buffer + arg->main_mem_in_address);
#if USE_CACHED_AI_RAM
uint8_t *dest = (uint8_t *)AI_RAM_BASE_ADDR + arg->kpu_mem_out_address * 64;
#else
uint8_t *dest = (uint8_t *)AI_IO_BASE_ADDR + arg->kpu_mem_out_address * 64;
#endif
uint32_t row_padding = 16;
uint32_t row_group = 4;
uint32_t row_length = 1;
uint32_t height = 4;
uint32_t oc, x, y, channels = arg->channels;
for (oc = 0; oc < channels; oc++)
{
uint8_t *channel_origin = dest + oc / row_group * row_length * height * 64 + oc % row_group * row_padding;
for (y = 0; y < 1; y++)
{
uint8_t *y_origin = channel_origin + y * row_length * 64;
for (x = 0; x < 1; x++)
y_origin[x] = *src++;
}
}
#if USE_CACHED_AI_RAM
uint32_t lines = row_length * height * channels / row_group;
kpu_flush_cache(arg->kpu_mem_out_address, lines);
#endif
}
void kpu_remove_padding(const kpu_model_remove_padding_layer_argument_t *arg)
{
const uint8_t *src = (const uint8_t *)(ctx_.main_buffer + arg->main_mem_in_address);
uint8_t *dest = (uint8_t *)(ctx_.main_buffer + arg->main_mem_out_address);
uint32_t oc, channels = arg->channels;
for (oc = 0; oc < channels; oc++)
*dest++ = src[oc * 16];
}
void kpu_upload(const kpu_model_upload_layer_argument_t *arg)
{
size_t width = arg->width;
size_t height = arg->height;
size_t channels = arg->channels;
kpu_upload_core(width, height, channels, ctx_.main_buffer + arg->main_mem_in_address, arg->kpu_mem_out_address);
}
#if KPU_DEBUG
const char *str_layer_type(uint32_t type)
{
switch (type)
{
case KL_ADD:
return "Add";
case KL_QUANTIZED_ADD:
return "QuantAdd";
case KL_GLOBAL_AVERAGE_POOL2D:
return "GAP";
case KL_QUANTIZED_MAX_POOL2D:
return "QuantMaxPool2d";
case KL_AVERAGE_POOL2D:
return "AveragePool2d";
case KL_QUANTIZE:
return "Quantize";
case KL_DEQUANTIZE:
return "Dequantize";
case KL_REQUANTIZE:
return "Requantize";
case KL_L2_NORMALIZATION:
return "L2Norm";
case KL_SOFTMAX:
return "Softmax";
case KL_CONCAT:
return "Concat";
case KL_QUANTIZED_CONCAT:
return "QuantConcat";
case KL_FULLY_CONNECTED:
return "FullyConnected";
case KL_TENSORFLOW_FLATTEN:
return "TFFlatten";
case KL_RESIZE_NEAREST_NEIGHBOR:
return "ResizeNearestNeighbor";
case KL_K210_CONV:
return "K210Conv";
case KL_K210_ADD_PADDING:
return "K210AddPad";
case KL_K210_REMOVE_PADDING:
return "K210RemovePad";
case KL_K210_UPLOAD:
return "K210Upload";
default:
return "Unknown";
}
}
#endif
int kpu_done()
{
kpu_.interrupt_clear.reg = 0b111;
kpu_.interrupt_mask.reg = 0b111;
#if KPU_DEBUG
uint32_t cnt_layer_id = ctx_.current_layer - 1;
gettimeofday(&time_, NULL);
if (total_time_ != 0)
{
uint64_t layer_time = (time_.tv_sec -last_time_.tv_sec) * 1000*1000 + (time_.tv_usec - last_time_.tv_usec);
printf("layer %d [%s]: %f ms\n", cnt_layer_id, str_layer_type(last_layer_type_), layer_time / 1000.0);
total_time_ += layer_time;
}
printf("Model: %f ms\n", total_time_ / 1000.0);
#endif
done_flag_ = 1;
return 0;
}
int ai_step()
{
uint32_t cnt_layer_id = ctx_.current_layer++;
const uint8_t *layer_body = ctx_.current_body;
const kpu_model_layer_header_t *cnt_layer_header = ctx_.layer_headers + cnt_layer_id;
ctx_.current_body += cnt_layer_header->body_size;
#if KPU_DEBUG
uint64_t layer_time;
gettimeofday(&time_, NULL);
layer_time = (time_.tv_sec -last_time_.tv_sec) * 1000*1000 + (time_.tv_usec - last_time_.tv_usec);
if(total_time_ == 0)
printf("DMA INPUT: %f ms\n", layer_time / 1000.0);
else
printf("layer %d [%s]: %f ms\n", cnt_layer_id - 1, str_layer_type(last_layer_type_), layer_time / 1000.0);
total_time_ += layer_time;
last_layer_type_ = cnt_layer_header->type;
gettimeofday(&last_time_, NULL);
#endif
switch (cnt_layer_header->type)
{
case KL_ADD:
kpu_add((const kpu_model_add_layer_argument_t *)layer_body);
break;
case KL_QUANTIZED_ADD:
kpu_quantized_add((const kpu_model_quant_add_layer_argument_t *)layer_body);
break;
case KL_GLOBAL_AVERAGE_POOL2D:
kpu_global_average_pool2d((const kpu_model_gap2d_layer_argument_t *)layer_body);
break;
case KL_QUANTIZED_MAX_POOL2D:
kpu_quantized_max_pool2d((const kpu_model_quant_max_pool2d_layer_argument_t *)layer_body);
break;
case KL_AVERAGE_POOL2D:
kpu_average_pool2d((const kpu_model_ave_pool2d_layer_argument_t *)layer_body);
break;
case KL_QUANTIZE:
kpu_quantize((const kpu_model_quantize_layer_argument_t *)layer_body);
break;
case KL_DEQUANTIZE:
kpu_dequantize((const kpu_model_dequantize_layer_argument_t *)layer_body);
break;
case KL_REQUANTIZE:
kpu_requantize((const kpu_model_requantize_layer_argument_t *)layer_body);
break;
case KL_L2_NORMALIZATION:
kpu_l2_normalization((const kpu_model_l2_norm_layer_argument_t *)layer_body);
break;
case KL_SOFTMAX:
kpu_softmax((const kpu_model_softmax_layer_argument_t *)layer_body);
break;
case KL_CONCAT:
case KL_QUANTIZED_CONCAT:
kpu_concat((const kpu_model_concat_layer_argument_t *)layer_body);
break;
case KL_FULLY_CONNECTED:
kpu_fully_connected((const kpu_model_fully_connected_layer_argument_t *)layer_body);
break;
case KL_TENSORFLOW_FLATTEN:
kpu_tf_flatten((const kpu_model_tf_flatten_layer_argument_t *)layer_body);
break;
case KL_RESIZE_NEAREST_NEIGHBOR:
kpu_resize_nearest_neighbor((const kpu_model_resize_nearest_neighbor_layer_argument_t *)layer_body);
break;
case KL_K210_CONV:
kpu_conv((const kpu_model_conv_layer_argument_t *)layer_body);
return 0;
case KL_K210_ADD_PADDING:
kpu_add_padding((const kpu_model_add_padding_layer_argument_t *)layer_body);
break;
case KL_K210_REMOVE_PADDING:
kpu_remove_padding((const kpu_model_remove_padding_layer_argument_t *)layer_body);
break;
case KL_K210_UPLOAD:
kpu_upload((const kpu_model_upload_layer_argument_t *)layer_body);
break;
default:
assert(!"Layer is not supported.");
}
if (cnt_layer_id != (ctx_.layers_length - 1))
{
return 1;
}
else
{
kpu_done();
return 0;
}
}
private:
volatile kpu_config_t &kpu_;
sysctl_clock_t clock_;
sysctl_dma_select_t dma_req_;
SemaphoreHandle_t free_mutex_;
uintptr_t dma_ch_;
SemaphoreHandle_t completion_event_;
uint8_t done_flag_ = 0;
kpu_model_context_t ctx_;
uint8_t *dest_kpu_;
uint8_t *dest_io_;
size_t dest_len_;
size_t max_len_;
uint8_t mem_out_flag_;
#if KPU_DEBUG
struct timeval time_;
struct timeval last_time_;
uint64_t total_time_ = 0;
uint32_t last_layer_type_;
#endif
};
static k_kpu_driver dev0_driver(AI_BASE_ADDR, SYSCTL_CLOCK_AI, SYSCTL_DMA_SELECT_AI_RX_REQ);
driver &g_kpu_driver_kpu0 = dev0_driver;
Reference in New Issue View Git Blame Copy Permalink