Skip to content

Commit

Permalink
Move towards LLVM 15 toolchain (#61)
Browse files Browse the repository at this point in the history 
* conv2d: Remove register array declarations

* traces: Use `addr2line` from software build toolchain
  • Loading branch information
@colluca
colluca authored Feb 19, 2024
1 parent d59f20b commit aa278ce
Show file tree
Hide file tree
Showing 3 changed files with 17 additions and 16 deletions.
28 changes: 14 additions & 14 deletions sw/dnn/conv2d/src/conv2d.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -341,7 +341,7 @@ void bn_relu(const float *pBuffer, const uint16_t dim_x, const uint16_t dim_y,
volatile register v2s current_kappa = ((v2s *)kappa)[co];
volatile register v2s zero = (v2s)0.0;

volatile register v2s tmp[n_unroll];
volatile v2s tmp[n_unroll];

snrt_ssr_enable();

Expand Down Expand Up @@ -422,7 +422,7 @@ void bn_relu(const float *pBuffer, const uint16_t dim_x, const uint16_t dim_y,
volatile register v2s current_kappa = ((v2s *)kappa)[co];
volatile register v2s zero = (v2s)0.0;

volatile register v2s tmp[n_unroll];
volatile v2s tmp[n_unroll];

snrt_ssr_enable();

Expand Down Expand Up @@ -653,7 +653,7 @@ static inline void conv2d_fp64(kernel_fp64 *k) {
// Output width dimension `k->dim_out_x`
for (uint32_t w = 0; w < k->dim_out_x; w++) {
// TODO: check if initialization needs to be unrolled by hand
volatile register double sum[max_unroll];
volatile double sum[max_unroll];
if (k->flag_y_accumulate_start) {
for (uint32_t i = 0; i < max_unroll; i++) {
sum[i] = 0.0;
Expand Down Expand Up @@ -718,7 +718,7 @@ static inline void conv2d_fp64(kernel_fp64 *k) {

// Output width dimension `k->dim_out_x`
for (uint32_t w = 0; w < k->dim_out_x; w++) {
volatile register double sum[max_unroll];
volatile double sum[max_unroll];
if (k->flag_y_accumulate_start) {
for (uint32_t i = 0; i < cleanup_unroll; i++) {
sum[i] = 0.0;
Expand Down Expand Up @@ -1009,8 +1009,8 @@ static inline void conv2d_fp32(kernel_fp32 *k) {
for (h0 = 0; h0 < k->dim_out_y / max_unroll; h0++) {
// Output width dimension `k->dim_out_x`
for (uint32_t w = 0; w < k->dim_out_x; w++) {
volatile register v2s sum[max_unroll];
volatile register float reduce_reg[max_unroll];
volatile v2s sum[max_unroll];
volatile float reduce_reg[max_unroll];
// pointer to output buffer location where intermediate values
// are read from and stored
float *_pOutBuffer =
Expand Down Expand Up @@ -1100,8 +1100,8 @@ static inline void conv2d_fp32(kernel_fp32 *k) {

// Output width dimension `k->dim_out_x`
for (uint32_t w = 0; w < k->dim_out_x; w++) {
volatile register v2s sum[max_unroll];
volatile register float reduce_reg[max_unroll];
volatile v2s sum[max_unroll];
volatile float reduce_reg[max_unroll];

// pointer to output buffer location where intermediate values
// are read from and stored
Expand Down Expand Up @@ -1442,7 +1442,7 @@ static inline void conv2d_dw_fp32(kernel_fp32 *k) {

// Output width dimension `k->dim_out_x`
for (uint32_t w = 0; w < k->dim_out_x; w++) {
volatile register v2s sum[max_unroll];
volatile v2s sum[max_unroll];
// pointer to output buffer location where intermediate values
// are read from and stored
v2s *_pOutBuffer =
Expand Down Expand Up @@ -1511,7 +1511,7 @@ static inline void conv2d_dw_fp32(kernel_fp32 *k) {

// Output width dimension `k->dim_out_x`
for (uint32_t w = 0; w < k->dim_out_x; w++) {
volatile register v2s sum[max_unroll];
volatile v2s sum[max_unroll];

// pointer to output buffer location where intermediate values
// are read from and stored
Expand Down Expand Up @@ -1788,8 +1788,8 @@ static inline void conv2d_chw_fp32(kernel_fp32 *k) {

// Output width dimension `k->dim_out_x`
for (uint32_t w = 0; w < k->dim_out_x; w++) {
volatile register v2s sum[max_unroll];
volatile register float reduce_reg[max_unroll];
volatile v2s sum[max_unroll];
volatile float reduce_reg[max_unroll];
// pointer to output buffer location where intermediate
// values are read from and stored
float *_pOutBuffer = &k->pOutBuffer[h * output_h_stride +
Expand Down Expand Up @@ -1884,8 +1884,8 @@ static inline void conv2d_chw_fp32(kernel_fp32 *k) {

// Output width dimension `k->dim_out_x`
for (uint32_t w = 0; w < k->dim_out_x; w++) {
volatile register v2s sum[max_unroll];
volatile register float reduce_reg[max_unroll];
volatile v2s sum[max_unroll];
volatile float reduce_reg[max_unroll];

// pointer to output buffer location where intermediate
// values are read from and stored
Expand Down
3 changes: 2 additions & 1 deletion target/common/common.mk
View file
Original file line number Diff line number Diff line change
Expand Up @@ -26,6 +26,7 @@ VSIM ?= $(QUESTA_SEPP) vsim
VOPT ?= $(QUESTA_SEPP) vopt
VLOG ?= $(QUESTA_SEPP) vlog
VLIB ?= $(QUESTA_SEPP) vlib
ADDR2LINE ?= $(LLVM_BINROOT)/llvm-addr2line

# Internal executables
GENTRACE_PY ?= $(UTIL_DIR)/trace/gen_trace.py
Expand Down Expand Up @@ -80,7 +81,7 @@ VLT_FLAGS += --unroll-count 1024
VLT_CFLAGS += -std=c++14 -pthread
VLT_CFLAGS +=-I ${VLT_BUILDDIR} -I $(VLT_ROOT)/include -I $(VLT_ROOT)/include/vltstd -I $(VLT_FESVR)/include -I $(TB_DIR) -I ${MKFILE_DIR}/test

ANNOTATE_FLAGS ?= -q --keep-time
ANNOTATE_FLAGS ?= -q --keep-time --addr2line=$(ADDR2LINE)
LAYOUT_EVENTS_FLAGS ?= --cfg=$(CFG)

# We need a recent LLVM installation (>11) to compile Verilator.
Expand Down
2 changes: 1 addition & 1 deletion util/trace/annotate.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -164,7 +164,7 @@ def dump_hunk(hunk_tstart, hunk_sstart, hunk_trace, hunk_source):


# Open ELF file for addr2line processing
elf = a2l.Elf(elf_file)
elf = a2l.Elf(elf_file, addr2line)

# core functionality
with open(trace, 'r') as f:
Expand Down

0 comments on commit aa278ce

Please sign in to comment.