Merge pull request #296 from GreenWaves-Technologies/4.8_dev

4.8 dev

configs/common.sh

@@ -139,3 +139,7 @@ else
         export XCSIM_PLATFORM=$XCSIM_PATH
     fi
 fi
+
+if [ -f "$GAP_SDK_HOME/configs/wsl.sh" ]; then
+    source $GAP_SDK_HOME/configs/wsl.sh
+fi

examples/autotiler/BilinearResize/Makefile

@@ -12,7 +12,7 @@ RESIZE_KER_PATH = $(TILER_BILINEAR_RESIZE_KERNEL_PATH)
 APP_SRCS += Bilinear_Resize.c ImgIO.c
 APP_SRCS += ResizeKernels.c $(RESIZE_KER_PATH)/ResizeBasicKernels.c
 
-APP_INC += . $(TILER_EMU_INC) $(TILER_INC) $(RESIZE_KER_PATH)
+APP_INC += . $(TILER_EMU_INC) $(TILER_INC) $(RESIZE_KER_PATH) $(TILER_DSP_KERNEL_PATH) $(TILER_CNN_KERNEL_PATH_FP16)
 
 APP_CFLAGS += -O3 -mno-memcpy -fno-tree-loop-distribute-patterns $(USER_FLAGS)
 APP_CFLAGS +=  -Wno-maybe-uninitialized -Wno-unused-but-set-variable -Wno-unused-variable

examples/nntool/mnist_gru/Makefile

@@ -64,7 +64,10 @@ PULP_APP = mnist
 APP = mnist
 APP_SRCS += $(MODEL_PREFIX).c $(MODEL_GEN_C) $(MODEL_COMMON_SRCS) $(CNN_LIB)
 
-APP_CFLAGS += -g -O3 -mno-memcpy -fno-tree-loop-distribute-patterns
+ifeq '$(TARGET_CHIP_FAMILY)' 'GAP8'
+	APP_CFLAGS += -march=rv32imcxgap8
+endif
+APP_CFLAGS += -g -O3 -mno-memcpy -fno-tree-loop-distribute-patterns 
 APP_CFLAGS += -I. -I$(MODEL_COMMON_INC) -I$(TILER_EMU_INC) -I$(TILER_INC) $(CNN_LIB_INCLUDE) -I$(MODEL_BUILD)
 APP_CFLAGS += -DPERF -DAT_MODEL_PREFIX=$(MODEL_PREFIX) $(MODEL_SIZE_CFLAGS)
 APP_CFLAGS += -DSTACK_SIZE=$(CLUSTER_STACK_SIZE) -DSLAVE_STACK_SIZE=$(CLUSTER_SLAVE_STACK_SIZE)

examples/nntool/mnist_rnn/Makefile

@@ -62,6 +62,9 @@ PULP_APP = mnist
 APP = mnist
 APP_SRCS += $(MODEL_PREFIX).c $(MODEL_GEN_C) $(MODEL_COMMON_SRCS) $(CNN_LIB)
 
+ifeq '$(TARGET_CHIP_FAMILY)' 'GAP8'
+        APP_CFLAGS += -march=rv32imcxgap8
+endif
 APP_CFLAGS += -g -O3 -mno-memcpy -fno-tree-loop-distribute-patterns
 APP_CFLAGS += -I. -I$(MODEL_COMMON_INC) -I$(TILER_EMU_INC) -I$(TILER_INC) $(CNN_LIB_INCLUDE) -I$(MODEL_BUILD)
 APP_CFLAGS += -DPERF -DAT_MODEL_PREFIX=$(MODEL_PREFIX) $(MODEL_SIZE_CFLAGS)

examples/pmsis/periph/i2s/wav_out_long/wav_out.c

@@ -50,7 +50,7 @@ static int fs_write_from_L3(void *file, void *data, int size_total, struct pi_de
         if(rest_size >= INTER_BUFF_SIZE)
             size = INTER_BUFF_SIZE;
         else
-            size = INTER_BUFF_SIZE - rest_size;
+            size = rest_size;
 
         pi_ram_read(ram, (_l3_buff+l3_index), _tmp_buffer, (uint32_t) size);
         pi_fs_write(file, _tmp_buffer, size);
@@ -59,6 +59,8 @@ static int fs_write_from_L3(void *file, void *data, int size_total, struct pi_de
         rest_size = rest_size - size;
     } while (rest_size);
 
+    pmsis_l2_malloc_free(_tmp_buffer, (uint32_t) INTER_BUFF_SIZE);
+
     return 0;
 }
 

gvsoc/gvsoc/bin/pulp-pc-info

@@ -3,6 +3,7 @@
 import argparse
 import os
 from subprocess import Popen, PIPE
+import sys
 
 
 parser = argparse.ArgumentParser(description='Generate PC debug info')
@@ -125,6 +126,11 @@ for f in functions:
 
 # And finally generate the output files
 
+if args.allFile is None and args.pcFile is None and args.debugFile is None and args.inlineFile is None:
+	for f in functions:
+		f.dumpAll(sys.stdout)
+
+
 # PC oriented file
 if args.allFile != None:
 	with open(args.allFile, 'w') as file:

gvsoc/gvsoc/engine/include/gv/gvsoc.hpp

@@ -200,7 +200,7 @@ namespace gv {
          * @param id ID of the VCD event.
          * @param value The new value.
          */
-        virtual void event_update_logical(int64_t timestamp, int id, uint8_t value) = 0;
+        virtual void event_update_logical(int64_t timestamp, int id, uint64_t value) = 0;
 
         /**
          * Called by GVSOC to update the value of a bitfield VCD event.

gvsoc/gvsoc/engine/python/gv/gvsoc.py

@@ -120,6 +120,7 @@ def gen_config(args, config):
 
         for binary in debug_binaries:
             full_config.set('**/debug_binaries', binary + '.debugInfo')
+            full_config.set('**/binaries', binary)
 
 
     gvsoc_config_path = os.path.join(config.get_str('gapy/work_dir'), 'gvsoc_config.json')

gvsoc/gvsoc/engine/src/trace/trace.cpp

@@ -83,6 +83,8 @@ void vp::component_trace::new_trace_event_string(std::string name, trace *trace)
     trace->name = name;
     trace->path = top.get_path() + "/" + name;
 
+    trace->width = 0;
+    trace->is_real = false;
     trace->is_string = true;
     trace->pending_timestamp = -1;
     trace->bytes = 0;
@@ -409,15 +411,33 @@ void vp::trace_engine::flush_event_traces(int64_t timestamp)
             }
             else if (current->is_string)
             {
-                //this->vcd_user->event_update_logical(int id, uint8_t *value, uint8_t *flags);
+                this->vcd_user->event_update_string(timestamp, current->id, (char *)current->buffer);
             }
-            else if (current->width > 1)
+            else if (current->width > 8)
             {
+                if (current->width <= 16)
+                {
+                    this->vcd_user->event_update_logical(timestamp, current->id, *(uint16_t *)current->buffer);
+                }
+                else if (current->width <= 32)
+                {
+                    this->vcd_user->event_update_logical(timestamp, current->id, *(uint32_t *)current->buffer);
+                }
+                else if (current->width <= 64)
+                {
+                    this->vcd_user->event_update_logical(timestamp, current->id, *(uint64_t *)current->buffer);
+                }
+                else
+                {
+                    // Use bitfield
+                }
 
             }
             else
             {
-                this->vcd_user->event_update_logical(timestamp, current->id, *current->buffer);
+                uint64_t value = (uint64_t)*(current->buffer);
+
+                this->vcd_user->event_update_logical(timestamp, current->id, value);
             }
         }
         else

gvsoc/gvsoc/engine/vp/trace_domain_impl.cpp

@@ -178,7 +178,7 @@ void trace_domain::check_trace_active(vp::trace *trace, int event)
     {
         for (auto &x : events_path_regex)
         {
-            if (x.second->is_path || regexec(x.second->regex, full_path.c_str(), 0, NULL, 0) == 0)
+            if ((x.second->is_path && x.second->path == full_path) || regexec(x.second->regex, full_path.c_str(), 0, NULL, 0) == 0)
             {
                 std::string file_path = x.second->file_path;                
                 vp::Event_trace *event_trace;
@@ -462,6 +462,7 @@ void trace_domain::conf_trace(int event, std::string path_str, bool enabled)
     const char *file_path = "all.vcd";
     const char *path = path_str.c_str();
     char *delim = (char *)::index(path, '@');
+
     if (delim)
     {
         *delim = 0;

gvsoc/gvsoc/models/cpu/iss/vp/include/iss_wrapper.hpp

@@ -151,6 +151,7 @@ class iss_wrapper : public vp::component, vp::Gdbserver_core
   vp::trace     inline_trace_event;
   vp::trace     line_trace_event;
   vp::trace     file_trace_event;
+  vp::trace     binaries_trace_event;
   vp::trace     pcer_trace_event[32];
   vp::trace     insn_trace_event;
 

gvsoc/gvsoc/models/cpu/iss/vp/include/platform_wrapper.hpp

@@ -170,7 +170,7 @@ static inline int iss_fetch_req_common(iss_t *_this, uint64_t addr, uint8_t *dat
   if (err != vp::IO_REQ_OK)
   {
     if (err == vp::IO_REQ_INVALID)
-      _this->trace.fatal("Invalid fetch request (addr: 0x%x, size: 0x%x)\n", addr, size);
+      _this->trace.force_warning("Invalid fetch request (addr: 0x%x, size: 0x%x)\n", addr, size);
     else
     {
       iss_exec_insn_stall(_this);

gvsoc/gvsoc/models/cpu/iss/vp/src/iss_wrapper.cpp

@@ -1300,6 +1300,7 @@ int iss_wrapper::build()
   traces.new_trace_event_string("func", &func_trace_event);
   traces.new_trace_event_string("inline_func", &inline_trace_event);
   traces.new_trace_event_string("file", &file_trace_event);
+  traces.new_trace_event_string("binaries", &binaries_trace_event);
   traces.new_trace_event("line", &line_trace_event, 32);
 
   traces.new_trace_event_real("ipc_stat", &ipc_stat_event);
@@ -1423,6 +1424,13 @@ void iss_wrapper::start()
     iss_register_debug_info(this, x->get_str().c_str());
   }
 
+  if (this->get_js_config()->get("**/binaries") != NULL)
+  {
+    for (auto x:this->get_js_config()->get("**/binaries")->get_elems())
+    {
+      this->binaries_trace_event.event_string("static enable " + x->get_str());
+    }
+  }
 
   trace.msg("ISS start (fetch: %d, is_active: %d, boot_addr: 0x%lx)\n", fetch_enable_reg.get(), is_active_reg.get(), get_config_int("boot_addr"));
 

gvsoc/gvsoc_gap/models/pulp/udma/hyper/udma_hyper_v3.cpp

@@ -83,6 +83,8 @@ Hyper_periph::Hyper_periph(udma *top, int id, int itf_id) : Udma_periph(top, id)
     this->read_req_waiting = new Udma_queue<Hyper_read_request>(-1);
     this->read_req_ready = new Udma_queue<Hyper_read_request>(-1);
 
+    memset((void *)&this->ca, 0, sizeof(this->ca));
+
 }
 
 
@@ -191,6 +193,9 @@ void Hyper_periph::reset(bool active)
 {
     Udma_periph::reset(active);
 
+    this->rx_channel->reset(active);
+    this->tx_channel->reset(active);
+
     if (!active)
     {
         this->pending_tx = false;

gvsoc/gvsoc_gap/models/pulp/udma/i2c/v4/udma_i2c.cpp

@@ -141,6 +141,9 @@ void I2c_periph::reset(bool active)
 {
     Udma_periph::reset(active);
 
+    this->foll_rx_channel->reset(active);
+    this->foll_tx_channel->reset(active);
+
     if (active)
     {
     }

gvsoc/gvsoc_gap/models/pulp/udma/i2s/udma_i2s_v3.cpp

@@ -239,29 +239,32 @@ void I2s_periph::handle_sdo(bool no_restart)
                 {
                     this->tx_wait_data_init &= ~(1 << this->active_channel);
                     data = this->tx_fifo[this->active_channel].front();
-
                     this->tx_fifo[this->active_channel].pop();
                 }
                 else
                 {
                     if (((this->tx_wait_data_init >> this->active_channel) & 1) == 0)
                     {
+                        this->trace.msg(vp::trace::LEVEL_DEBUG, "Generating TX error (slot: %d)\n", this->active_channel);
                         this->regmap.err_status.set(this->regmap.err_status.get() | (1 << (this->active_channel + 16)));
                     }
                 }
 
                 this->tx_pending_value = this->handle_tx_format(channel, data);
 
-                this->trace.msg(vp::trace::LEVEL_DEBUG, "Got new TX sample (value: 0x%x, width: %d)\n", this->tx_pending_value, this->tx_pending_bits);
-
+                this->trace.msg(vp::trace::LEVEL_DEBUG, "Got new TX sample (slot: %d, value: 0x%x, width: %d)\n", this->active_channel, this->tx_pending_value, this->tx_pending_bits);
 
                 // Now ask the next sample to the channel so that it is ready for the
                 // next frame.
                 // The channel id is pushed now since we the samples are received in order
                 // and we don't know what is the slot when we receive the sample from the channel
 
-                this->tx_fifo_slot_id.push(this->active_channel);
-                channel->get_data(channel->slot_cfg->tx_dsize_get() + 1, channel->slot_cfg->tx_id_get());
+                if (channel->is_active() || channel->slot_cfg->tx_id_get() >= 0xe0)
+                {
+                    this->tx_fifo_slot_id.push(this->active_channel);
+
+                    channel->get_data(channel->slot_cfg->tx_dsize_get() + 1, channel->slot_cfg->tx_id_get());
+                }
             }
 
             if (this->tx_pending_bits > 0)
@@ -1065,7 +1068,7 @@ void I2s_tx_channel::push_data(uint8_t *data, int size)
     uint32_t value = 0;
     memcpy((void *)&value, (void *)data, size);
 
-    this->periph->trace.msg(vp::trace::LEVEL_INFO, "Received TX sample from memory (value: 0x%x)\n", value);
+    this->periph->trace.msg(vp::trace::LEVEL_INFO, "Received TX sample from memory (slot: %d, value: 0x%x)\n", this->periph->tx_fifo_slot_id.front(), value);
 
     this->periph->tx_fifo[this->periph->tx_fifo_slot_id.front()].push(value);
     this->periph->tx_fifo_slot_id.pop();

gvsoc/gvsoc_gap/models/pulp/udma/mram/udma_mram_v2.cpp

@@ -508,6 +508,9 @@ void Mram_periph::reset(bool active)
 
     this->read_fifo->reset(active);
 
+    this->rx_channel->reset(active);
+    this->tx_channel->reset(active);
+
     if (!active)
     {
         this->pending_transfer_size = 0;

gvsoc/gvsoc_gap/models/pulp/udma/udma_v4_addrgens.cpp

@@ -75,7 +75,7 @@ void Udma_addrgen_linear::check_pending_transfer()
 {
     if (!this->active_transfer && this->nb_pending_transfers)
     {
-        trace.msg(vp::trace::LEVEL_TRACE, "Starting new buffer (addr: 0x%x, size: 0x%x)\n", this->pending_addr, this->pending_size);
+        trace.msg(vp::trace::LEVEL_TRACE, "Starting new buffer (addr: 0x%x, size: 0x%x, pending: %d)\n", this->pending_addr, this->pending_size, this->nb_pending_transfers);
 
         this->nb_pending_transfers--;
         this->set_active_transfer(true);
@@ -101,17 +101,25 @@ void Udma_addrgen_linear::cfg_ctrl_req(uint64_t reg_offset, int size, uint8_t *v
         }
         else if (this->regmap.cfg_ctrl.en_get())
         {
+            this->trace.msg(vp::trace::LEVEL_TRACE, "Enqeueing transfer (pending: %d)\n", this->nb_pending_transfers);
 
-            if (this->nb_pending_transfers > 0)
+            if (this->nb_pending_transfers == 1)
             {
-                this->remaining_size -= this->pending_size;
+                this->trace.force_warning("Trying to enqueue while alreay 2 transfers are enqueued\n");
             }
+            else
+            {
+                if (this->nb_pending_transfers > 0)
+                {
+                    this->remaining_size -= this->pending_size;
+                }
 
-            this->pending_addr = this->regmap.cfg_sa_buf0.get();
-            this->pending_size = this->regmap.cfg_size.get();
-            this->remaining_size += this->pending_size;
+                this->pending_addr = this->regmap.cfg_sa_buf0.get();
+                this->pending_size = this->regmap.cfg_size.get();
+                this->remaining_size += this->pending_size;
 
-            this->nb_pending_transfers++;
+                this->nb_pending_transfers++;
+            }
 
             this->check_pending_transfer();
 

gvsoc/gvsoc_gap/models/pulp/udma/udma_v4_impl.cpp

@@ -709,6 +709,14 @@ void udma::start()
 
 void udma::reset(bool active)
 {
+    for (auto x: this->periphs)
+    {
+        if (x)
+        {
+            x->reset(active);
+        }
+    }
+
     for (auto x: this->addrgen_linear)
     {
         x->reset(active);

gvsoc/gvsoc_gap/models/pulp/udma/udma_v4_rx_channels.cpp

@@ -186,8 +186,7 @@ void Udma_rx_channels::handle_pending(void *__this, vp::clock_event *event)
 
                 if (addr == 0)
                 {
-                    fflush(NULL);
-                    abort();
+                    _this->top->trace.fatal("UDMA trying to access NULL\n");
                 }
 
                 if (err == vp::IO_REQ_OK)

gvsoc/gvsoc_gap/models/pulp/udma/udma_v4_tx_channels.cpp

@@ -83,7 +83,7 @@ void Udma_tx_channel::get_data(int size, int channel)
 {
     this->requested_size_queue.push_back(size);
     this->requested_size += size;
-    if (this->requested_size_queue.size() == 1 && this->is_active() && !this->enqueued)
+    if (this->requested_size_queue.size() > 0 && this->is_active() && !this->enqueued)
     {
         this->top->tx_channels->push_ready_channel(this);
         this->check_state();
@@ -243,7 +243,7 @@ void Udma_tx_channels::handle_pending(void *__this, vp::clock_event *event)
             channel->requested_size_queue.push_front(requested_size_queue - size);
         }
 
-        if (channel->requested_size_queue.size() && !channel->enqueued)
+        if (channel->requested_size_queue.size() && !channel->enqueued && channel->is_active())
         {
             _this->push_ready_channel(channel);
         }
@@ -307,7 +307,6 @@ void Udma_tx_channels::check_state()
 {
     if (!this->send_reqs_event->is_enqueued())
     {
-        if (!this->pending_channels.empty() && !this->l2_free_reqs->is_empty())
         if (!this->pending_channels.empty() && !this->l2_free_reqs->is_empty() && this->pending_channels.front()->is_active())
         {
             this->top->event_enqueue(this->send_reqs_event, 1);