#include #include #include #include #include #include #include #include #include #include #include #define CACHE_LINESIZE 64 #define ALLOC_BASE_ADDR 0x10000000 #define LOCAL_MEM_SIZE 0xffffffff /////////////////////////////////////////////////////////////////////////////// inline size_t align_size(size_t size, size_t alignment) { assert(0 == (alignment & (alignment - 1))); return (size + alignment - 1) & ~(alignment - 1); } /////////////////////////////////////////////////////////////////////////////// class vx_device; class vx_buffer { public: vx_buffer(size_t size, vx_device* device) : size_(size) , device_(device) { auto aligned_asize = align_size(size, CACHE_LINESIZE); data_ = malloc(aligned_asize); } ~vx_buffer() { if (data_) { free(data_); } } void* data() const { return data_; } size_t size() const { return size_; } vx_device* device() const { return device_; } private: size_t size_; vx_device* device_; void* data_; }; /////////////////////////////////////////////////////////////////////////////// class vx_device { public: vx_device() { mem_allocation_ = ALLOC_BASE_ADDR; simulator_.attach_ram(&ram_); } ~vx_device() { if (future_.valid()) { future_.wait(); } } int alloc_local_mem(size_t size, size_t* dev_maddr) { auto dev_mem_size = LOCAL_MEM_SIZE; size_t asize = align_size(size, CACHE_LINESIZE); if (mem_allocation_ + asize > dev_mem_size) return -1; *dev_maddr = mem_allocation_; mem_allocation_ += asize; return 0; } int upload(void* src, size_t dest_addr, size_t size, size_t src_offset) { size_t asize = align_size(size, CACHE_LINESIZE); if (dest_addr + asize > ram_.size()) return -1; /*printf("VXDRV: upload %d bytes to 0x%x\n", size, dest_addr); for (int i = 0; i < size; i += 4) { printf("mem-write: 0x%x <- 0x%x\n", uint32_t(dest_addr + i), *(uint32_t*)((uint8_t*)src + src_offset + i)); }*/ ram_.write(dest_addr, asize, (uint8_t*)src + src_offset); return 0; } int download(const void* dest, size_t src_addr, size_t size, size_t dest_offset) { size_t asize = align_size(size, CACHE_LINESIZE); if (src_addr + asize > ram_.size()) return -1; ram_.read(src_addr, asize, (uint8_t*)dest + dest_offset); /*printf("VXDRV: download %d bytes from 0x%x\n", size, src_addr); for (int i = 0; i < size; i += 4) { printf("mem-read: 0x%x -> 0x%x\n", uint32_t(src_addr + i), *(uint32_t*)((uint8_t*)dest + dest_offset + i)); }*/ return 0; } int start() { if (future_.valid()) { future_.wait(); // ensure prior run completed } future_ = std::async(std::launch::async, [&]{ simulator_.reset(); while (simulator_.is_busy()) { simulator_.step(); } }); return 0; } int wait(long long timeout) { if (!future_.valid()) return 0; auto timeout_sec = (timeout < 0) ? timeout : (timeout / 1000); std::chrono::seconds wait_time(1); for (;;) { auto status = future_.wait_for(wait_time); // wait for 1 sec and check status if (status == std::future_status::ready || 0 == timeout_sec--) break; } return 0; } int flush_caches(size_t dev_maddr, size_t size) { if (future_.valid()) { future_.wait(); // ensure prior run completed } simulator_.flush_caches(dev_maddr, size); while (simulator_.is_busy()) { simulator_.step(); }; return 0; } int set_csr(int core_id, int addr, unsigned value) { if (future_.valid()) { future_.wait(); // ensure prior run completed } simulator_.set_csr(core_id, addr, value); while (simulator_.is_busy()) { simulator_.step(); }; return 0; } int get_csr(int core_id, int addr, unsigned *value) { if (future_.valid()) { future_.wait(); // ensure prior run completed } simulator_.get_csr(core_id, addr, value); while (simulator_.is_busy()) { simulator_.step(); }; return 0; } private: size_t mem_allocation_; RAM ram_; Simulator simulator_; std::future future_; }; /////////////////////////////////////////////////////////////////////////////// extern int vx_dev_caps(vx_device_h hdevice, unsigned caps_id, unsigned *value) { if (nullptr == hdevice) return -1; switch (caps_id) { case VX_CAPS_VERSION: *value = IMPLEMENTATION_ID; break; case VX_CAPS_MAX_CORES: *value = NUM_CORES; break; case VX_CAPS_MAX_WARPS: *value = NUM_WARPS; break; case VX_CAPS_MAX_THREADS: *value = NUM_THREADS; break; case VX_CAPS_CACHE_LINESIZE: *value = CACHE_LINESIZE; break; case VX_CAPS_LOCAL_MEM_SIZE: *value = 0xffffffff; break; case VX_CAPS_ALLOC_BASE_ADDR: *value = 0x10000000; break; case VX_CAPS_KERNEL_BASE_ADDR: *value = STARTUP_ADDR; break; default: std::cout << "invalid caps id: " << caps_id << std::endl; std::abort(); return -1; } return 0; } extern int vx_dev_open(vx_device_h* hdevice) { if (nullptr == hdevice) return -1; *hdevice = new vx_device(); return 0; } extern int vx_dev_close(vx_device_h hdevice) { if (nullptr == hdevice) return -1; vx_device *device = ((vx_device*)hdevice); #ifdef DUMP_PERF_STATS unsigned num_cores; vx_csr_get(hdevice, 0, CSR_NC, &num_cores); if (num_cores > 1) { uint64_t total_instrs = 0, total_cycles = 0; for (unsigned core_id = 0; core_id < num_cores; ++core_id) { uint64_t instrs, cycles; vx_get_perf(hdevice, core_id, &instrs, &cycles); float IPC = (float)(double(instrs) / double(cycles)); fprintf(stdout, "PERF: core%d: instrs=%ld, cycles=%ld, IPC=%f\n", core_id, instrs, cycles, IPC); total_instrs += instrs; total_cycles = std::max(total_cycles, cycles); } float IPC = (float)(double(total_instrs) / double(total_cycles)); fprintf(stdout, "PERF: instrs=%ld, cycles=%ld, IPC=%f\n", total_instrs, total_cycles, IPC); } else { uint64_t instrs, cycles; vx_get_perf(hdevice, 0, &instrs, &cycles); float IPC = (float)(double(instrs) / double(cycles)); fprintf(stdout, "PERF: instrs=%ld, cycles=%ld, IPC=%f\n", instrs, cycles, IPC); } #endif delete device; return 0; } extern int vx_alloc_dev_mem(vx_device_h hdevice, size_t size, size_t* dev_maddr) { if (nullptr == hdevice || nullptr == dev_maddr || 0 >= size) return -1; vx_device *device = ((vx_device*)hdevice); return device->alloc_local_mem(size, dev_maddr); } extern int vx_flush_caches(vx_device_h hdevice, size_t dev_maddr, size_t size) { if (nullptr == hdevice || 0 >= size) return -1; vx_device *device = ((vx_device*)hdevice); return device->flush_caches(dev_maddr, size); } extern int vx_alloc_shared_mem(vx_device_h hdevice, size_t size, vx_buffer_h* hbuffer) { if (nullptr == hdevice || 0 >= size || nullptr == hbuffer) return -1; vx_device *device = ((vx_device*)hdevice); auto buffer = new vx_buffer(size, device); if (nullptr == buffer->data()) { delete buffer; return -1; } *hbuffer = buffer; return 0; } extern volatile void* vx_host_ptr(vx_buffer_h hbuffer) { if (nullptr == hbuffer) return nullptr; vx_buffer* buffer = ((vx_buffer*)hbuffer); return buffer->data(); } extern int vx_buf_release(vx_buffer_h hbuffer) { if (nullptr == hbuffer) return -1; vx_buffer* buffer = ((vx_buffer*)hbuffer); delete buffer; return 0; } extern int vx_copy_to_dev(vx_buffer_h hbuffer, size_t dev_maddr, size_t size, size_t src_offset) { if (nullptr == hbuffer || 0 >= size) return -1; auto buffer = (vx_buffer*)hbuffer; if (size + src_offset > buffer->size()) return -1; return buffer->device()->upload(buffer->data(), dev_maddr, size, src_offset); } extern int vx_copy_from_dev(vx_buffer_h hbuffer, size_t dev_maddr, size_t size, size_t dest_offset) { if (nullptr == hbuffer || 0 >= size) return -1; auto buffer = (vx_buffer*)hbuffer; if (size + dest_offset > buffer->size()) return -1; return buffer->device()->download(buffer->data(), dev_maddr, size, dest_offset); } extern int vx_start(vx_device_h hdevice) { if (nullptr == hdevice) return -1; vx_device *device = ((vx_device*)hdevice); return device->start(); } extern int vx_ready_wait(vx_device_h hdevice, long long timeout) { if (nullptr == hdevice) return -1; vx_device *device = ((vx_device*)hdevice); return device->wait(timeout); } extern int vx_csr_set(vx_device_h hdevice, int core_id, int addr, unsigned value) { if (nullptr == hdevice) return -1; vx_device *device = ((vx_device*)hdevice); return device->set_csr(core_id, addr, value); } extern int vx_csr_get(vx_device_h hdevice, int core_id, int addr, unsigned* value) { if (nullptr == hdevice) return -1; vx_device *device = ((vx_device*)hdevice); return device->get_csr(core_id, addr, value); }