vortex/runtime/xrt/vortex.cpp
2025-01-14 02:21:17 -08:00

722 lines
No EOL
19 KiB
C++

// Copyright © 2019-2023
//
// 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 <common.h>
#ifdef SCOPE
#include "scope.h"
#endif
// XRT includes
#ifdef XRTSIM
#include <xrt_c.h>
#else
#include "experimental/xrt_bo.h"
#include "experimental/xrt_device.h"
#include "experimental/xrt_error.h"
#include "experimental/xrt_ip.h"
#include "experimental/xrt_kernel.h"
#include "experimental/xrt_xclbin.h"
#endif
#include <limits>
#include <stdarg.h>
#include <string>
#include <unordered_map>
#include <util.h>
#include <vector>
using namespace vortex;
#ifndef XRTSIM
#define CPP_API
#endif
// #define BANK_INTERLEAVE
#define MMIO_CTL_ADDR 0x00
#define MMIO_DEV_ADDR 0x10
#define MMIO_ISA_ADDR 0x18
#define MMIO_DCR_ADDR 0x20
#define MMIO_SCP_ADDR 0x28
#define MMIO_MEM_ADDR 0x30
#define CTL_AP_START (1 << 0)
#define CTL_AP_DONE (1 << 1)
#define CTL_AP_IDLE (1 << 2)
#define CTL_AP_READY (1 << 3)
#define CTL_AP_RESET (1 << 4)
#define CTL_AP_RESTART (1 << 7)
#ifdef CPP_API
typedef xrt::device xrt_device_t;
typedef xrt::ip xrt_kernel_t;
typedef xrt::bo xrt_buffer_t;
#else
typedef xrtDeviceHandle xrt_device_t;
typedef xrtKernelHandle xrt_kernel_t;
typedef xrtBufferHandle xrt_buffer_t;
#endif
#define DEFAULT_DEVICE_INDEX 0
#define DEFAULT_XCLBIN_PATH "vortex_afu.xclbin"
#define KERNEL_NAME "vortex_afu"
#define CHECK_HANDLE(handle, _expr, _cleanup) \
auto handle = _expr; \
if (handle == nullptr) { \
printf("[VXDRV] Error: '%s' returned NULL!\n", #_expr); \
_cleanup \
}
#ifndef CPP_API
static void dump_xrt_error(xrtDeviceHandle xrtDevice, xrtErrorCode err) {
size_t len = 0;
xrtErrorGetString(xrtDevice, err, nullptr, 0, &len);
std::vector<char> buf(len);
xrtErrorGetString(xrtDevice, err, buf.data(), buf.size(), nullptr);
printf("[VXDRV] detail: %s!\n", buf.data());
}
#endif
///////////////////////////////////////////////////////////////////////////////
class vx_device {
public:
vx_device()
: global_mem_(ALLOC_BASE_ADDR,
GLOBAL_MEM_SIZE - ALLOC_BASE_ADDR,
RAM_PAGE_SIZE,
CACHE_BLOCK_SIZE)
#ifndef CPP_API
, xrtDevice_(nullptr)
, xrtKernel_(nullptr)
#endif
{}
~vx_device() {
#ifdef SCOPE
vx_scope_stop(this);
#endif
#ifndef CPP_API
for (auto &entry : xrtBuffers_) {
#ifdef BANK_INTERLEAVE
xrtBOFree(entry);
#else
xrtBOFree(entry.second.xrtBuffer);
#endif
}
if (xrtKernel_) {
xrtKernelClose(xrtKernel_);
}
if (xrtDevice_) {
xrtDeviceClose(xrtDevice_);
}
#endif
}
int init() {
int device_index = DEFAULT_DEVICE_INDEX;
const char *device_index_s = getenv("XRT_DEVICE_INDEX");
if (device_index_s != nullptr) {
device_index = atoi(device_index_s);
}
const char *xlbin_path_s = getenv("XRT_XCLBIN_PATH");
if (xlbin_path_s == nullptr) {
xlbin_path_s = DEFAULT_XCLBIN_PATH;
}
#ifdef CPP_API
auto xrtDevice = xrt::device(device_index);
auto uuid = xrtDevice.load_xclbin(xlbin_path_s);
auto xrtKernel = xrt::ip(xrtDevice, uuid, KERNEL_NAME);
auto xclbin = xrt::xclbin(xlbin_path_s);
auto device_name = xrtDevice.get_info<xrt::info::device::name>();
#else
CHECK_HANDLE(xrtDevice, xrtDeviceOpen(device_index), {
return -1;
});
#ifndef XRTSIM
CHECK_ERR(xrtDeviceLoadXclbinFile(xrtDevice, xlbin_path_s), {
dump_xrt_error(xrtDevice, err);
xrtDeviceClose(xrtDevice);
return err;
});
xuid_t uuid;
CHECK_ERR(xrtDeviceGetXclbinUUID(xrtDevice, uuid), {
dump_xrt_error(xrtDevice, err);
xrtDeviceClose(xrtDevice);
return err;
});
CHECK_HANDLE(xrtKernel, xrtPLKernelOpenExclusive(xrtDevice, uuid, KERNEL_NAME), {
xrtDeviceClose(xrtDevice);
return -1;
});
#else
xrtKernelHandle xrtKernel = xrtDevice;
#endif
// get device name
int device_name_size;
xrtXclbinGetXSAName(xrtDevice, nullptr, 0, &device_name_size);
std::vector<char> sz_device_name(device_name_size);
xrtXclbinGetXSAName(xrtDevice, sz_device_name.data(), device_name_size, nullptr);
std::string device_name(sz_device_name.data(), device_name_size);
#endif
xrtDevice_ = xrtDevice;
xrtKernel_ = xrtKernel;
CHECK_ERR(this->write_register(MMIO_CTL_ADDR, CTL_AP_RESET), {
return err;
});
CHECK_ERR(this->read_register(MMIO_DEV_ADDR, (uint32_t *)&dev_caps_), {
return err;
});
CHECK_ERR(this->read_register(MMIO_DEV_ADDR + 4, (uint32_t *)&dev_caps_ + 1), {
return err;
});
CHECK_ERR(this->read_register(MMIO_ISA_ADDR, (uint32_t *)&isa_caps_), {
return err;
});
CHECK_ERR(this->read_register(MMIO_ISA_ADDR + 4, (uint32_t *)&isa_caps_ + 1), {
return err;
});
uint64_t num_banks;
this->get_caps(VX_CAPS_NUM_MEM_BANKS, &num_banks);
lg2_num_banks_ = log2ceil(num_banks);
uint64_t bank_size;
this->get_caps(VX_CAPS_MEM_BANK_SIZE, &bank_size);
lg2_bank_size_ = log2ceil(bank_size);
global_mem_size_ = num_banks * bank_size;
printf("info: device name=%s, memory_capacity=0x%lx bytes, memory_banks=%ld.\n", device_name.c_str(), global_mem_size_, num_banks);
#ifdef BANK_INTERLEAVE
xrtBuffers_.reserve(num_banks);
for (uint32_t i = 0; i < num_banks; ++i) {
#ifdef CPP_API
xrtBuffers_.emplace_back(xrtDevice_, bank_size, xrt::bo::flags::normal, i);
#else
CHECK_HANDLE(xrtBuffer, xrtBOAlloc(xrtDevice_, bank_size, XRT_BO_FLAGS_NONE, i), {
return -1;
});
xrtBuffers_.push_back(xrtBuffer);
#endif
printf("*** allocated bank%u/%u, size=%lu\n", i, num_banks, bank_size);
}
#endif
#ifdef SCOPE
{
scope_callback_t callback;
callback.registerWrite = [](vx_device_h hdevice, uint64_t value) -> int {
auto device = (vx_device *)hdevice;
uint32_t value_lo = (uint32_t)(value);
uint32_t value_hi = (uint32_t)(value >> 32);
CHECK_ERR(device->write_register(MMIO_SCP_ADDR, value_lo), {
return err;
});
CHECK_ERR(device->write_register(MMIO_SCP_ADDR + 4, value_hi), {
return err;
});
return 0;
};
callback.registerRead = [](vx_device_h hdevice, uint64_t *value) -> int {
auto device = (vx_device *)hdevice;
uint32_t value_lo, value_hi;
CHECK_ERR(device->read_register(MMIO_SCP_ADDR, &value_lo), {
return err;
});
CHECK_ERR(device->read_register(MMIO_SCP_ADDR + 4, &value_hi), {
return err;
});
*value = (((uint64_t)value_hi) << 32) | value_lo;
return 0;
};
CHECK_ERR(vx_scope_start(&callback, this, -1, -1), {
return err;
});
}
#endif
#ifdef CHIPSCOPE
std::cout << "\nPress ENTER to continue after setting up ILA trigger..." << std::endl;
std::cin.ignore(std::numeric_limits<std::streamsize>::max(), '\n');
#endif
return 0;
}
int get_caps(uint32_t caps_id, uint64_t *value) {
uint64_t _value;
switch (caps_id) {
case VX_CAPS_VERSION:
_value = (dev_caps_ >> 0) & 0xff;
break;
case VX_CAPS_NUM_THREADS:
_value = (dev_caps_ >> 8) & 0xff;
break;
case VX_CAPS_NUM_WARPS:
_value = (dev_caps_ >> 16) & 0xff;
break;
case VX_CAPS_NUM_CORES:
_value = (dev_caps_ >> 24) & 0xffff;
break;
case VX_CAPS_CACHE_LINE_SIZE:
_value = CACHE_BLOCK_SIZE;
break;
case VX_CAPS_GLOBAL_MEM_SIZE:
_value = global_mem_size_;
break;
case VX_CAPS_LOCAL_MEM_SIZE:
_value = 1ull << ((dev_caps_ >> 40) & 0xff);
break;
case VX_CAPS_ISA_FLAGS:
_value = isa_caps_;
break;
case VX_CAPS_NUM_MEM_BANKS:
_value = 1 << ((dev_caps_ >> 48) & 0x7);
break;
case VX_CAPS_MEM_BANK_SIZE:
_value = 1ull << (20 + ((dev_caps_ >> 51) & 0x1f));
break;
default:
fprintf(stderr, "[VXDRV] Error: invalid caps id: %d\n", caps_id);
std::abort();
return -1;
}
*value = _value;
return 0;
}
int mem_alloc(uint64_t size, int flags, uint64_t *dev_addr) {
uint64_t asize = aligned_size(size, CACHE_BLOCK_SIZE);
uint64_t addr;
CHECK_ERR(global_mem_.allocate(asize, &addr), {
return err;
});
#ifndef BANK_INTERLEAVE
uint32_t bank_id;
CHECK_ERR(this->get_bank_info(addr, &bank_id, nullptr), {
global_mem_.release(addr);
return err;
});
CHECK_ERR(get_buffer(bank_id, nullptr), {
global_mem_.release(addr);
return err;
});
#endif
CHECK_ERR(this->mem_access(addr, size, flags), {
global_mem_.release(addr);
return err;
});
*dev_addr = addr;
return 0;
}
int mem_reserve(uint64_t dev_addr, uint64_t size, int flags) {
CHECK_ERR(global_mem_.reserve(dev_addr, size), {
return err;
});
#ifndef BANK_INTERLEAVE
uint32_t bank_id;
CHECK_ERR(this->get_bank_info(dev_addr, &bank_id, nullptr), {
global_mem_.release(dev_addr);
return err;
});
CHECK_ERR(get_buffer(bank_id, nullptr), {
global_mem_.release(dev_addr);
return err;
});
#endif
CHECK_ERR(this->mem_access(dev_addr, size, flags), {
global_mem_.release(dev_addr);
return err;
});
return 0;
}
int mem_free(uint64_t dev_addr) {
CHECK_ERR(global_mem_.release(dev_addr), {
return err;
});
#ifdef BANK_INTERLEAVE
if (0 == global_mem_.allocated()) {
#ifndef CPP_API
for (auto &entry : xrtBuffers_) {
xrtBOFree(entry);
}
#endif
xrtBuffers_.clear();
}
#else
uint32_t bank_id;
CHECK_ERR(this->get_bank_info(dev_addr, &bank_id, nullptr), {
return err;
});
auto it = xrtBuffers_.find(bank_id);
if (it != xrtBuffers_.end()) {
auto count = --it->second.count;
if (0 == count) {
printf("freeing bank%d...\n", bank_id);
#ifndef CPP_API
xrtBOFree(it->second.xrtBuffer);
#endif
xrtBuffers_.erase(it);
}
} else {
fprintf(stderr, "[VXDRV] Error: invalid device memory address: 0x%lx\n",
dev_addr);
return -1;
}
#endif
return 0;
}
int mem_access(uint64_t /*dev_addr*/, uint64_t /*size*/, int /*flags*/) {
return 0;
}
int mem_info(uint64_t *mem_free, uint64_t *mem_used) const {
if (mem_free)
*mem_free = global_mem_.free();
if (mem_used)
*mem_used = global_mem_.allocated();
return 0;
}
int write_register(uint32_t addr, uint32_t value) {
#ifdef CPP_API
xrtKernel_.write_register(addr, value);
#else
CHECK_ERR(xrtKernelWriteRegister(xrtKernel_, addr, value), {
dump_xrt_error(xrtDevice_, err);
return err;
});
#endif
return 0;
}
int read_register(uint32_t addr, uint32_t *value) {
#ifdef CPP_API
*value = xrtKernel_.read_register(addr);
#else
CHECK_ERR(xrtKernelReadRegister(xrtKernel_, addr, value), {
dump_xrt_error(xrtDevice_, err);
return err;
});
#endif
return 0;
}
int upload(uint64_t dev_addr, const void *src, uint64_t size) {
auto host_ptr = (const uint8_t *)src;
// check alignment
if (!is_aligned(dev_addr, CACHE_BLOCK_SIZE))
return -1;
auto asize = aligned_size(size, CACHE_BLOCK_SIZE);
// bound checking
if (dev_addr + asize > global_mem_size_)
return -1;
for (uint64_t end = dev_addr + asize; dev_addr < end;
dev_addr += CACHE_BLOCK_SIZE, host_ptr += CACHE_BLOCK_SIZE) {
#ifdef BANK_INTERLEAVE
asize = CACHE_BLOCK_SIZE;
#else
end = 0;
#endif
uint32_t bo_index;
uint64_t bo_offset;
xrt_buffer_t xrtBuffer;
CHECK_ERR(this->get_bank_info(dev_addr, &bo_index, &bo_offset), {
return err;
});
CHECK_ERR(this->get_buffer(bo_index, &xrtBuffer), {
return err;
});
#ifdef CPP_API
xrtBuffer.write(host_ptr, size, bo_offset);
xrtBuffer.sync(XCL_BO_SYNC_BO_TO_DEVICE, size, bo_offset);
#else
CHECK_ERR(xrtBOWrite(xrtBuffer, host_ptr, size, bo_offset), {
dump_xrt_error(xrtDevice_, err);
return err;
});
CHECK_ERR(xrtBOSync(xrtBuffer, XCL_BO_SYNC_BO_TO_DEVICE, size, bo_offset), {
dump_xrt_error(xrtDevice_, err);
return err;
});
#endif
}
return 0;
}
int download(void *dest, uint64_t dev_addr, uint64_t size) {
auto host_ptr = (uint8_t *)dest;
// check alignment
if (!is_aligned(dev_addr, CACHE_BLOCK_SIZE))
return -1;
auto asize = aligned_size(size, CACHE_BLOCK_SIZE);
// bound checking
if (dev_addr + asize > global_mem_size_)
return -1;
for (uint64_t end = dev_addr + asize; dev_addr < end;
dev_addr += CACHE_BLOCK_SIZE, host_ptr += CACHE_BLOCK_SIZE) {
#ifdef BANK_INTERLEAVE
asize = CACHE_BLOCK_SIZE;
#else
end = 0;
#endif
uint32_t bo_index;
uint64_t bo_offset;
xrt_buffer_t xrtBuffer;
CHECK_ERR(this->get_bank_info(dev_addr, &bo_index, &bo_offset), {
return err;
});
CHECK_ERR(this->get_buffer(bo_index, &xrtBuffer), {
return err;
});
#ifdef CPP_API
xrtBuffer.sync(XCL_BO_SYNC_BO_FROM_DEVICE, size, bo_offset);
xrtBuffer.read(host_ptr, size, bo_offset);
#else
CHECK_ERR(xrtBOSync(xrtBuffer, XCL_BO_SYNC_BO_FROM_DEVICE, size, bo_offset), {
dump_xrt_error(xrtDevice_, err);
return err;
});
CHECK_ERR(xrtBORead(xrtBuffer, host_ptr, size, bo_offset), {
dump_xrt_error(xrtDevice_, err);
return err;
});
#endif
}
return 0;
}
int start(uint64_t krnl_addr, uint64_t args_addr) {
// set kernel info
CHECK_ERR(this->dcr_write(VX_DCR_BASE_STARTUP_ADDR0, krnl_addr & 0xffffffff), {
return err;
});
CHECK_ERR(this->dcr_write(VX_DCR_BASE_STARTUP_ADDR1, krnl_addr >> 32), {
return err;
});
CHECK_ERR(this->dcr_write(VX_DCR_BASE_STARTUP_ARG0, args_addr & 0xffffffff), {
return err;
});
CHECK_ERR(this->dcr_write(VX_DCR_BASE_STARTUP_ARG1, args_addr >> 32), {
return err;
});
// start execution
CHECK_ERR(this->write_register(MMIO_CTL_ADDR, CTL_AP_START), {
return err;
});
// clear mpm cache
mpm_cache_.clear();
return 0;
}
int ready_wait(uint64_t timeout) {
struct timespec sleep_time;
#ifndef NDEBUG
sleep_time.tv_sec = 1;
sleep_time.tv_nsec = 0;
#else
sleep_time.tv_sec = 0;
sleep_time.tv_nsec = 1000000;
#endif
// to milliseconds
uint64_t sleep_time_ms = (sleep_time.tv_sec * 1000) + (sleep_time.tv_nsec / 1000000);
for (;;) {
uint32_t status = 0;
CHECK_ERR(this->read_register(MMIO_CTL_ADDR, &status), {
return err;
});
bool is_done = (status & CTL_AP_DONE) == CTL_AP_DONE;
if (is_done)
break;
if (0 == timeout) {
return -1;
}
nanosleep(&sleep_time, nullptr);
timeout -= sleep_time_ms;
};
return 0;
}
int dcr_write(uint32_t addr, uint32_t value) {
CHECK_ERR(this->write_register(MMIO_DCR_ADDR, addr), {
return err;
});
CHECK_ERR(this->write_register(MMIO_DCR_ADDR + 4, value), {
return err;
});
dcrs_.write(addr, value);
return 0;
}
int dcr_read(uint32_t addr, uint32_t *value) const {
return dcrs_.read(addr, value);
}
int mpm_query(uint32_t addr, uint32_t core_id, uint64_t *value) {
uint32_t offset = addr - VX_CSR_MPM_BASE;
if (offset > 31)
return -1;
if (mpm_cache_.count(core_id) == 0) {
uint64_t mpm_mem_addr = IO_MPM_ADDR + core_id * 32 * sizeof(uint64_t);
CHECK_ERR(this->download(mpm_cache_[core_id].data(), mpm_mem_addr, 32 * sizeof(uint64_t)), {
return err;
});
}
*value = mpm_cache_.at(core_id).at(offset);
return 0;
}
private:
MemoryAllocator global_mem_;
xrt_device_t xrtDevice_;
xrt_kernel_t xrtKernel_;
uint64_t dev_caps_;
uint64_t isa_caps_;
uint64_t global_mem_size_;
DeviceConfig dcrs_;
std::unordered_map<uint32_t, std::array<uint64_t, 32>> mpm_cache_;
uint32_t lg2_num_banks_;
uint32_t lg2_bank_size_;
#ifdef BANK_INTERLEAVE
std::vector<xrt_buffer_t> xrtBuffers_;
int get_bank_info(uint64_t addr, uint32_t *pIdx, uint64_t *pOff) {
uint32_t num_banks = 1 << lg2_num_banks_;
uint64_t block_addr = addr / CACHE_BLOCK_SIZE;
uint32_t index = block_addr & (num_banks - 1);
uint64_t offset = (block_addr >> lg2_num_banks_) * CACHE_BLOCK_SIZE;
if (pIdx) {
*pIdx = index;
}
if (pOff) {
*pOff = offset;
}
//printf("get_bank_info(addr=0x%lx, bank=%d, offset=0x%lx\n", addr, index, offset);
return 0;
}
int get_buffer(uint32_t bank_id, xrt_buffer_t *pBuf) {
if (pBuf) {
*pBuf = xrtBuffers_.at(bank_id);
}
return 0;
}
#else
struct buf_cnt_t {
xrt_buffer_t xrtBuffer;
uint32_t count;
};
std::unordered_map<uint32_t, buf_cnt_t> xrtBuffers_;
int get_bank_info(uint64_t addr, uint32_t *pIdx, uint64_t *pOff) {
uint32_t num_banks = 1 << lg2_num_banks_;
uint64_t bank_size = 1ull << lg2_bank_size_;
uint32_t index = addr >> lg2_bank_size_;
uint64_t offset = addr & (bank_size - 1);
if (index > num_banks) {
fprintf(stderr, "[VXDRV] Error: address out of range: 0x%lx\n", addr);
return -1;
}
if (pIdx) {
*pIdx = index;
}
if (pOff) {
*pOff = offset;
}
//printf("get_bank_info(addr=0x%lx, bank=%d, offset=0x%lx\n", addr, index, offset);
return 0;
}
int get_buffer(uint32_t bank_id, xrt_buffer_t *pBuf) {
auto it = xrtBuffers_.find(bank_id);
if (it != xrtBuffers_.end()) {
if (pBuf) {
*pBuf = it->second.xrtBuffer;
} else {
printf("reusing bank%d...\n", bank_id);
++it->second.count;
}
} else {
printf("allocating bank%d...\n", bank_id);
uint64_t bank_size = 1ull << lg2_bank_size_;
#ifdef CPP_API
xrt::bo xrtBuffer(xrtDevice_, bank_size, xrt::bo::flags::normal, bank_id);
#else
CHECK_HANDLE(xrtBuffer, xrtBOAlloc(xrtDevice_, bank_size, XRT_BO_FLAGS_NONE, bank_id), {
return -1;
});
#endif
xrtBuffers_.insert({bank_id, {xrtBuffer, 1}});
if (pBuf) {
*pBuf = xrtBuffer;
}
}
return 0;
}
#endif
};
#include <callbacks.inc>