wu-arch/kernels
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

OPAE rtl fixes

Browse Source
This commit is contained in:
Blaise Tine
2020-05-08 08:28:28 -07:00
parent 13dfd5c8c7
commit c2e9240b7d
12 changed files with 208 additions and 243 deletions
Download Patch File Download Diff File Expand all files Collapse all files

37
driver/opae/vortex.cpp
View File

@@ -143,6 +143,7 @@ extern int vx_alloc_dev_mem(vx_device_h hdevice, size_t size, size_t* dev_maddr)
size_t dev_mem_size = vx_dev_caps(VX_CAPS_LOCAL_MEM_SIZE);
size_t asize = align_size(size, line_size);
if (device->mem_allocation + asize > dev_mem_size)
return -1;
@@ -189,11 +190,11 @@ extern int vx_alloc_shared_mem(vx_device_h hdevice, size_t size, vx_buffer_h* hb
return -1;
}
buffer->wsid = wsid;
buffer->wsid = wsid;
buffer->host_ptr = host_ptr;
buffer->io_addr = io_addr;
buffer->hdevice = hdevice;
buffer->size = size;
buffer->io_addr = io_addr;
buffer->hdevice = hdevice;
buffer->size = asize;
*hbuffer = buffer;
@@ -265,18 +266,18 @@ extern int vx_copy_to_dev(vx_buffer_h hbuffer, size_t dev_maddr, size_t size, si
int line_size = vx_dev_caps(VX_CAPS_CACHE_LINESIZE);
size_t dev_mem_size = vx_dev_caps(VX_CAPS_LOCAL_MEM_SIZE);
size_t asize = align_size(size, line_size);
// check alignment
if (!is_aligned(dev_maddr, line_size))
return -1;
if (!is_aligned(size, line_size))
return -1;
if (!is_aligned(buffer->io_addr + src_offset, line_size))
return -1;
// bound checking
if (size + src_offset > buffer->size)
if (src_offset + asize > buffer->size)
return -1;
if (dev_maddr + size > dev_mem_size)
if (dev_maddr + asize > dev_mem_size)
return -1;
// Ensure ready for new command
@@ -287,7 +288,7 @@ extern int vx_copy_to_dev(vx_buffer_h hbuffer, size_t dev_maddr, size_t size, si
CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_CSR_IO_ADDR, (buffer->io_addr + src_offset) >> ls_shift));
CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_CSR_MEM_ADDR, (dev_maddr >> ls_shift) ));
CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_CSR_DATA_SIZE, size >> ls_shift));
CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_CSR_DATA_SIZE, asize >> ls_shift));
CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_CSR_CMD, CMD_TYPE_WRITE));
// Wait for the write operation to finish
@@ -308,18 +309,18 @@ extern int vx_copy_from_dev(vx_buffer_h hbuffer, size_t dev_maddr, size_t size,
int line_size = vx_dev_caps(VX_CAPS_CACHE_LINESIZE);
size_t dev_mem_size = vx_dev_caps(VX_CAPS_LOCAL_MEM_SIZE);
size_t asize = align_size(size, line_size);
// check alignment
if (!is_aligned(dev_maddr, line_size))
return -1;
if (!is_aligned(size, line_size))
return -1;
if (!is_aligned(buffer->io_addr + dest_offset, line_size))
return -1;
// bound checking
if (size + dest_offset > buffer->size)
if (dest_offset + asize > buffer->size)
return -1;
if (dev_maddr + size > dev_mem_size)
if (dev_maddr + asize > dev_mem_size)
return -1;
// Ensure ready for new command
@@ -330,7 +331,7 @@ extern int vx_copy_from_dev(vx_buffer_h hbuffer, size_t dev_maddr, size_t size,
CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_CSR_IO_ADDR, (buffer->io_addr + dest_offset) >> ls_shift));
CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_CSR_MEM_ADDR, (dev_maddr) >> ls_shift));
CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_CSR_DATA_SIZE, size >> ls_shift));
CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_CSR_DATA_SIZE, asize >> ls_shift));
CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_CSR_CMD, CMD_TYPE_READ));
// Wait for the write operation to finish
@@ -347,13 +348,13 @@ extern int vx_flush_caches(vx_device_h hdevice, size_t dev_maddr, size_t size) {
vx_device_t* device = ((vx_device_t*)hdevice);
int line_size = vx_dev_caps(VX_CAPS_CACHE_LINESIZE);
int line_size = vx_dev_caps(VX_CAPS_CACHE_LINESIZE);
size_t asize = align_size(size, line_size);
// check alignment
if (!is_aligned(dev_maddr, line_size))
return -1;
if (!is_aligned(size, line_size))
return -1;
// Ensure ready for new command
if (vx_ready_wait(hdevice, -1) != 0)
@@ -362,7 +363,7 @@ extern int vx_flush_caches(vx_device_h hdevice, size_t dev_maddr, size_t size) {
auto ls_shift = (int)std::log2(line_size);
CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_CSR_MEM_ADDR, dev_maddr >> ls_shift));
CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_CSR_DATA_SIZE, size >> ls_shift));
CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_CSR_DATA_SIZE, asize >> ls_shift));
CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_CSR_CMD, CMD_TYPE_CLFLUSH));
// Wait for the write operation to finish

122
driver/tests/basic/basic.cpp
View File

@@ -4,6 +4,16 @@
int test = -1;
#define RT_CHECK(_expr) \
do { \
int _ret = _expr; \
if (0 == _ret) \
break; \
printf("Error: '%s' returned %d!\n", #_expr, (int)_ret); \
cleanup(); \
exit(-1); \
} while (false)
static void parse_args(int argc, char **argv) {
int c;
while ((c = getopt(argc, argv, "t:h?")) != -1) {
@@ -27,12 +37,27 @@ uint64_t shuffle(int i, uint64_t value) {
return (value << i) | (value & ((1 << i)-1));;
}
vx_device_h device = nullptr;
vx_buffer_h sbuf = nullptr;
vx_buffer_h dbuf = nullptr;
void cleanup() {
if (sbuf) {
vx_buf_release(sbuf);
}
if (dbuf) {
vx_buf_release(dbuf);
}
if (device) {
vx_dev_close(device);
}
}
int run_memcopy_test(vx_buffer_h sbuf,
vx_buffer_h dbuf,
uint32_t address,
uint64_t value,
int num_blocks) {
int ret;
int errors = 0;
// write sbuf data
@@ -42,15 +67,11 @@ int run_memcopy_test(vx_buffer_h sbuf,
// write buffer to local memory
std::cout << "write buffer to local memory" << std::endl;
ret = vx_copy_to_dev(sbuf, address, 64 * num_blocks, 0);
if (ret != 0)
return ret;
RT_CHECK(vx_copy_to_dev(sbuf, address, 64 * num_blocks, 0));
// read buffer from local memory
std::cout << "read buffer from local memory" << std::endl;
ret = vx_copy_from_dev(dbuf, address, 64 * num_blocks, 0);
if (ret != 0)
return ret;
RT_CHECK(vx_copy_from_dev(dbuf, address, 64 * num_blocks, 0));
// verify result
std::cout << "verify result" << std::endl;
@@ -77,7 +98,6 @@ int run_kernel_test(vx_device_h device,
vx_buffer_h sbuf,
vx_buffer_h dbuf,
const char* program) {
int ret;
int errors = 0;
uint64_t seed = 0x0badf00d40ff40ff;
@@ -93,43 +113,27 @@ int run_kernel_test(vx_device_h device,
// write buffer to local memory
std::cout << "write buffer to local memory" << std::endl;
ret = vx_copy_to_dev(sbuf, src_dev_addr, 64 * num_blocks, 0);
if (ret != 0)
return ret;
RT_CHECK(vx_copy_to_dev(sbuf, src_dev_addr, 64 * num_blocks, 0));
// upload program
std::cout << "upload program" << std::endl;
ret = vx_upload_kernel_file(device, program);
if (ret != 0) {
return ret;
}
RT_CHECK(vx_upload_kernel_file(device, program));
// start device
std::cout << "start device" << std::endl;
ret = vx_start(device);
if (ret != 0) {
return ret;
}
RT_CHECK(vx_start(device));
// wait for completion
std::cout << "wait for completion" << std::endl;
ret = vx_ready_wait(device, -1);
if (ret != 0) {
return ret;
}
RT_CHECK(vx_ready_wait(device, -1));
// flush the caches
std::cout << "flush the caches" << std::endl;
ret = vx_flush_caches(device, dest_dev_addr, 64 * num_blocks);
if (ret != 0) {
return ret;
}
RT_CHECK(vx_flush_caches(device, dest_dev_addr, 64 * num_blocks));
// read buffer from local memory
std::cout << "read buffer from local memory" << std::endl;
ret = vx_copy_from_dev(dbuf, dest_dev_addr, 64 * num_blocks, 0);
if (ret != 0)
return ret;
RT_CHECK(vx_copy_from_dev(dbuf, dest_dev_addr, 64 * num_blocks, 0));
// verify result
std::cout << "verify result" << std::endl;
@@ -152,75 +156,33 @@ int run_kernel_test(vx_device_h device,
return 0;
}
vx_device_h device = nullptr;
vx_buffer_h sbuf = nullptr;
vx_buffer_h dbuf = nullptr;
void cleanup() {
if (sbuf) {
vx_buf_release(sbuf);
}
if (dbuf) {
vx_buf_release(dbuf);
}
if (device) {
vx_dev_close(device);
}
}
int main(int argc, char *argv[]) {
int ret;
// parse command arguments
parse_args(argc, argv);
// open device connection
std::cout << "open device connection" << std::endl;
vx_device_h device;
ret = vx_dev_open(&device);
if (ret != 0)
return ret;
RT_CHECK(vx_dev_open(&device));
// create source buffer
std::cout << "create source buffer" << std::endl;
ret = vx_alloc_shared_mem(device, 4096, &sbuf);
if (ret != 0) {
cleanup();
return ret;
}
RT_CHECK(vx_alloc_shared_mem(device, 4096, &sbuf));
// create destination buffer
std::cout << "create destination buffer" << std::endl;
ret = vx_alloc_shared_mem(device, 4096, &dbuf);
if (ret != 0) {
cleanup();
return ret;
}
RT_CHECK(vx_alloc_shared_mem(device, 4096, &dbuf));
// run tests
if (0 == test || -1 == test) {
/*9if (0 == test || -1 == test) {
std::cout << "run memcopy test" << std::endl;
ret = run_memcopy_test(sbuf, dbuf, 0x10000000, 0x0badf00d00ff00ff, 1);
if (ret != 0) {
cleanup();
return ret;
}
ret = run_memcopy_test(sbuf, dbuf, 0x20000000, 0x0badf00d40ff40ff, 8);
if (ret != 0) {
cleanup();
return ret;
}
}
RT_CHECK(run_memcopy_test(sbuf, dbuf, 0x10000000, 0x0badf00d00ff00ff, 1));
RT_CHECK(run_memcopy_test(sbuf, dbuf, 0x20000000, 0x0badf00d40ff40ff, 8));
}*/
if (1 == test || -1 == test) {
std::cout << "run kernel test" << std::endl;
ret = run_kernel_test(device, sbuf, dbuf, "kernel.bin");
if (ret != 0) {
cleanup();
return ret;
}
RT_CHECK(run_kernel_test(device, sbuf, dbuf, "kernel.bin"));
}
// cleanup

119
driver/tests/demo/demo.cpp
View File

@@ -4,6 +4,16 @@
#include <vortex.h>
#include "common.h"
#define RT_CHECK(_expr) \
do { \
int _ret = _expr; \
if (0 == _ret) \
break; \
printf("Error: '%s' returned %d!\n", #_expr, (int)_ret); \
cleanup(); \
exit(-1); \
} while (false)
const char* program_file = "kernel.bin";
uint32_t data_stride = 0xffffffff;
@@ -39,40 +49,38 @@ static void parse_args(int argc, char **argv) {
}
}
vx_device_h device = nullptr;
vx_buffer_h buffer = nullptr;
void cleanup() {
if (buffer) {
vx_buf_release(buffer);
}
if (device) {
vx_dev_close(device);
}
}
int run_test(vx_device_h device,
vx_buffer_h buffer,
const kernel_arg_t& kernel_arg,
uint32_t buf_size,
uint32_t num_points) {
int ret;
// start device
std::cout << "start device" << std::endl;
ret = vx_start(device);
if (ret != 0) {
return ret;
}
RT_CHECK(vx_start(device));
// wait for completion
std::cout << "wait for completion" << std::endl;
ret = vx_ready_wait(device, -1);
if (ret != 0) {
return ret;
}
RT_CHECK(vx_ready_wait(device, -1));
// flush the destination buffer caches
std::cout << "flush the destination buffer caches" << std::endl;
ret = vx_flush_caches(device, kernel_arg.dst_ptr, buf_size);
if (ret != 0) {
return ret;
}
RT_CHECK(vx_flush_caches(device, kernel_arg.dst_ptr, buf_size));
// download destination buffer
std::cout << "download destination buffer" << std::endl;
ret = vx_copy_from_dev(buffer, kernel_arg.dst_ptr, buf_size, 0);
if (ret != 0) {
return ret;
}
RT_CHECK(vx_copy_from_dev(buffer, kernel_arg.dst_ptr, buf_size, 0));
// verify result
std::cout << "verify result" << std::endl;
@@ -96,20 +104,7 @@ int run_test(vx_device_h device,
return 0;
}
vx_device_h device = nullptr;
vx_buffer_h buffer = nullptr;
void cleanup() {
if (buffer) {
vx_buf_release(buffer);
}
if (device) {
vx_dev_close(device);
}
}
int main(int argc, char *argv[]) {
int ret;
size_t value;
kernel_arg_t kernel_arg;
@@ -132,50 +127,28 @@ int main(int argc, char *argv[]) {
// open device connection
std::cout << "open device connection" << std::endl;
ret = vx_dev_open(&device);
if (ret != 0)
return ret;
RT_CHECK(vx_dev_open(&device));
// upload program
std::cout << "upload program" << std::endl;
ret = vx_upload_kernel_file(device, program_file);
if (ret != 0) {
cleanup();
return ret;
}
RT_CHECK(vx_upload_kernel_file(device, program_file));
// allocate device memory
std::cout << "allocate device memory" << std::endl;
ret = vx_alloc_dev_mem(device, buf_size, &value);
if (ret != 0) {
cleanup();
return ret;
}
RT_CHECK(vx_alloc_dev_mem(device, buf_size, &value));
kernel_arg.src0_ptr = value;
ret = vx_alloc_dev_mem(device, buf_size, &value);
if (ret != 0) {
cleanup();
return ret;
}
RT_CHECK(vx_alloc_dev_mem(device, buf_size, &value));
kernel_arg.src1_ptr = value;
ret = vx_alloc_dev_mem(device, buf_size, &value);
if (ret != 0) {
cleanup();
return ret;
}
RT_CHECK(vx_alloc_dev_mem(device, buf_size, &value));
kernel_arg.dst_ptr = value;
// allocate shared memory
std::cout << "allocate shared memory" << std::endl;
uint32_t alloc_size = std::max<uint32_t>(buf_size, sizeof(kernel_arg_t));
ret = vx_alloc_shared_mem(device, alloc_size, &buffer);
if (ret != 0) {
cleanup();
return ret;
}
RT_CHECK(vx_alloc_shared_mem(device, alloc_size, &buffer));
// populate source buffer values
std::cout << "populate source buffer values" << std::endl;
@@ -187,19 +160,9 @@ int main(int argc, char *argv[]) {
}
// upload source buffers
std::cout << "upload source buffers" << std::endl;
ret = vx_copy_to_dev(buffer, kernel_arg.src0_ptr, buf_size, 0);
if (ret != 0) {
cleanup();
return ret;
}
ret = vx_copy_to_dev(buffer, kernel_arg.src1_ptr, buf_size, 0);
if (ret != 0) {
cleanup();
return ret;
}
std::cout << "upload source buffers" << std::endl;
RT_CHECK(vx_copy_to_dev(buffer, kernel_arg.src0_ptr, buf_size, 0));
RT_CHECK(vx_copy_to_dev(buffer, kernel_arg.src1_ptr, buf_size, 0));
// upload kernel argument
std::cout << "upload kernel argument" << std::endl;
@@ -210,20 +173,12 @@ int main(int argc, char *argv[]) {
auto buf_ptr = (int*)vx_host_ptr(buffer);
memcpy(buf_ptr, &kernel_arg, sizeof(kernel_arg_t));
ret = vx_copy_to_dev(buffer, KERNEL_ARG_DEV_MEM_ADDR, sizeof(kernel_arg_t), 0);
if (ret != 0) {
cleanup();
return ret;
}
RT_CHECK(vx_copy_to_dev(buffer, KERNEL_ARG_DEV_MEM_ADDR, sizeof(kernel_arg_t), 0));
}
// run tests
std::cout << "run tests" << std::endl;
ret = run_test(device, buffer, kernel_arg, buf_size, num_points);
if (ret != 0) {
cleanup();
return ret;
}
RT_CHECK(run_test(device, buffer, kernel_arg, buf_size, num_points));
// cleanup
std::cout << "cleanup" << std::endl;