github-mirrors/vortex
1
0
Fork 0
mirror of https://github.com/vortexgpgpu/vortex.git synced 2025-04-23 21:39:10 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge branch 'fpga_synthesis' of https://github.gatech.edu/casl/Vortex into fpga_synthesis

Browse source
This commit is contained in:
felsabbagh3 2020-03-29 21:22:08 -07:00
commit ccc65a06fe
7 changed files with 71 additions and 12 deletions
Download patch file Download diff file Expand all files Collapse all files

11
driver/sw/rtlsim/Makefile
View file

@ -1,5 +1,5 @@
CFLAGS += -std=c++11 -O3 -Wall -Wextra -pedantic -Wfatal-errors
#CFLAGS += -std=c++11 -g -O0 -Wall -Wextra -pedantic -Wfatal-errors
#CFLAGS += -std=c++11 -O3 -Wall -Wextra -pedantic -Wfatal-errors
CFLAGS += -std=c++11 -g -O0 -Wall -Wextra -pedantic -Wfatal-errors
USE_MULTICORE=1
@ -23,10 +23,9 @@ SRCS = vortex.cpp ../vx_utils.cpp ../../../rtl/simulate/$(RTL_TOP).cpp
RTL_INCLUDE = -I../../../rtl -I../../../rtl/interfaces -I../../../rtl/cache -I../../../rtl/VX_cache -I../../../rtl/shared_memory -I../../../rtl/pipe_regs -I../../../rtl/compat
THREADS ?= $(shell python3 -c 'import multiprocessing as mp; print(max(1, mp.cpu_count() // 2))')
VL_FLAGS += --threads $(THREADS)
VL_FLAGS += -Wno-UNOPTFLAT -Wno-WIDTH
# Enable Verilator multithreaded simulation
#THREADS ?= $(shell python3 -c 'import multiprocessing as mp; print(max(1, mp.cpu_count() // 2))')
#VL_FLAGS += --threads $(THREADS)
VL_FLAGS += -Wno-UNDRIVEN --Wno-PINMISSING -Wno-STMTDLY -Wno-WIDTH -Wno-UNSIGNED -Wno-UNOPTFLAT -Wno-LITENDIAN

8
driver/tests/demo/Makefile
View file

@ -43,16 +43,16 @@ $(PROJECT): $(SRCS)
$(CXX) $(CXXFLAGS) $^ $(LDFLAGS) -L../../sw/simx -lvortex -o $@
run-fpga: $(PROJECT)
LD_LIBRARY_PATH=../../sw/opae:$(LD_LIBRARY_PATH) ./$(PROJECT) -f kernel.bin
LD_LIBRARY_PATH=../../sw/opae:$(LD_LIBRARY_PATH) ./$(PROJECT) -f kernel.bin -n 16
run-ase: $(PROJECT)
LD_LIBRARY_PATH=../../sw/opae/ase:$(LD_LIBRARY_PATH) ./$(PROJECT) -f kernel.bin
LD_LIBRARY_PATH=../../sw/opae/ase:$(LD_LIBRARY_PATH) ./$(PROJECT) -f kernel.bin -n 16
run-rtlsim: $(PROJECT)
LD_LIBRARY_PATH=../../sw/rtlsim:$(LD_LIBRARY_PATH) ./$(PROJECT) -f kernel.bin
LD_LIBRARY_PATH=../../sw/rtlsim:$(LD_LIBRARY_PATH) ./$(PROJECT) -f kernel.bin -n 16
run-simx: $(PROJECT)
LD_LIBRARY_PATH=../../sw/simx:$(LD_LIBRARY_PATH) ./$(PROJECT) -f kernel.bin
LD_LIBRARY_PATH=../../sw/simx:$(LD_LIBRARY_PATH) ./$(PROJECT) -f kernel.bin -n 16
.depend: $(SRCS)
$(CXX) $(CXXFLAGS) -MM $^ > .depend;

BIN
driver/tests/demo/demo
View file

Binary file not shown.

60
driver/tests/demo/demo.cpp
View file

@ -178,6 +178,66 @@ int main(int argc, char *argv[]) {
return -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) {
cleanup();
return -1;
}
// 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) {
cleanup();
return -1;
}
// verify result
std::cout << "verify result" << std::endl;
{
auto buf_ptr = (int*)vx_host_ptr(buffer);
for (uint32_t i = 0; i < num_points; ++i) {
int ref = i * i;
int cur = buf_ptr[i];
if (cur != ref) {
++errors;
}
}
}
if (errors != 0) {
printf("Found %d errors!\n", errors);
printf("FAILED!\n");
cleanup();
return -1;
}
// start device
std::cout << "start device" << std::endl;
ret = vx_start(device);
if (ret != 0) {
cleanup();
return -1;
}
// wait for completion
std::cout << "wait for completion" << std::endl;
ret = vx_ready_wait(device, -1);
if (ret != 0) {
cleanup();
return -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) {
cleanup();
return -1;
}
// download destination buffer
std::cout << "download destination buffer" << std::endl;
ret = vx_copy_from_dev(buffer, kernel_arg.dst_ptr, buf_size, 0);

BIN
driver/tests/demo/kernel.bin
View file

Binary file not shown.

2
rtl/simulate/Vortex.cpp
View file

@ -257,7 +257,7 @@ bool Vortex::is_busy() {
void Vortex::send_snoops(uint32_t mem_addr, uint32_t size) {
// align address to LLC block boundaries
auto aligned_addr_start = GLOBAL_BLOCK_SIZE_BYTES * ((mem_addr + GLOBAL_BLOCK_SIZE_BYTES - 1) / GLOBAL_BLOCK_SIZE_BYTES);
auto aligned_addr_start = GLOBAL_BLOCK_SIZE_BYTES * (mem_addr / GLOBAL_BLOCK_SIZE_BYTES);
auto aligned_addr_end = GLOBAL_BLOCK_SIZE_BYTES * ((mem_addr + size + GLOBAL_BLOCK_SIZE_BYTES - 1) / GLOBAL_BLOCK_SIZE_BYTES);
// submit snoop requests for the needed blocks

2
rtl/simulate/Vortex_SOC.cpp
View file

@ -195,7 +195,7 @@ bool Vortex_SOC::is_busy() {
void Vortex_SOC::send_snoops(uint32_t mem_addr, uint32_t size) {
// align address to LLC block boundaries
auto aligned_addr_start = GLOBAL_BLOCK_SIZE_BYTES * ((mem_addr + GLOBAL_BLOCK_SIZE_BYTES - 1) / GLOBAL_BLOCK_SIZE_BYTES);
auto aligned_addr_start = GLOBAL_BLOCK_SIZE_BYTES * (mem_addr / GLOBAL_BLOCK_SIZE_BYTES);
auto aligned_addr_end = GLOBAL_BLOCK_SIZE_BYTES * ((mem_addr + size + GLOBAL_BLOCK_SIZE_BYTES - 1) / GLOBAL_BLOCK_SIZE_BYTES);
// submit snoop requests for the needed blocks