diff --git a/deepsocflow/c/dsf_zynq.c b/deepsocflow/c/dsf_zynq.c
index 3ffb70b..33da8f6 100644
--- a/deepsocflow/c/dsf_zynq.c
+++ b/deepsocflow/c/dsf_zynq.c
@@ -8,7 +8,7 @@
 #include <assert.h>
 #include <limits.h>
 #include <stdint.h>
-#define printf xil_printf
+#include <stdio.h>
 
 #define MEM_BASEADDR 0x20000000
 
@@ -21,20 +21,21 @@ XAxiDma dma_pixels, dma_weights, dma_output;
 XScuGic intr_controller; // Generic interrupt controller
 u32     status;
 
+
 static void start_wait_output(UINTPTR baseaddr, u32 bpt){
 	int status = XAxiDma_SimpleTransfer(&dma_output , baseaddr, bpt, XAXIDMA_DEVICE_TO_DMA);
 	if (status != XST_SUCCESS) xil_printf("S2MM transfer failed, base:%p, bpt:%d\n", baseaddr, bpt);
 	while(!done_output);
-	printf("Done output dma at :%p, bpt:%d\n", baseaddr, bpt);
+	xil_printf("Done output dma at :%p, bpt:%d\n", baseaddr, bpt);
 	Xil_DCacheFlushRange((INTPTR)baseaddr, bpt);
 	done_output = 0;
 }
 
 static void start_pixels_dma();
 
+#define printf xil_printf
 #include "runtime.h"
-
-
+#undef printf
 
 
 static void start_pixels_dma() {
@@ -84,6 +85,8 @@ static void setup_interrupt(XScuGic *p_intr_controller, u32 intr_id, Xil_Interru
 
 int main() {
   init_platform();
+
+
   xil_printf("Store wbx at: %p; y:%p; buffers {0:%p,1:%p}; debug_nhwc:%p; debug_tiled:%p \n", &mem.w, &mem.y, &mem.out_buffers[0], &mem.out_buffers[1], &mem.debug_nhwc, &mem.debug_tiled);
   print("Starting!!!\n\r");
 
@@ -127,8 +130,9 @@ int main() {
   xil_printf("Done inference: %d \n", 0);
   Xil_DCacheFlushRange((INTPTR)&mem.y, O_WORDS*sizeof(O_TYPE));  // force transfer to DDR, starting addr & length
 
-  for (int i=0; i<20; i++)
-	  xil_printf("y[%d]: %d \n", i, mem.y[i]);
+  for (int i=0; i<sizeof(mem.y)/sizeof(mem.y[0]); i++)
+	  if (bundles[N_BUNDLES-1].is_softmax) printf("y[%d]: %f \n", i, (float  )mem.y[i]);
+	  else                                 printf("y[%d]: %d \n", i, (int32_t)mem.y[i]);
 
   xil_printf("Done all: %d \n", 0);
 
diff --git a/deepsocflow/c/runtime.h b/deepsocflow/c/runtime.h
index 9ef10b8..80cf713 100644
--- a/deepsocflow/c/runtime.h
+++ b/deepsocflow/c/runtime.h
@@ -352,9 +352,7 @@ extern EXT_C void load_y (uint8_t *p_done, uint64_t *p_base_addr_next, int32_t *
                       float val = (float)out_val;
                       val = val / (float)(1 << pb->softmax_frac);
                       val = val - pb->softmax_max_f;
-#ifdef SIM
                       val = (float)exp(val);
-#endif
                       mem.y[iy_nhwc] = val;
 
                       if (i_yc == pb->co-1) {
diff --git a/deepsocflow/py/model.py b/deepsocflow/py/model.py
index 5b2eda8..0ca9d2f 100644
--- a/deepsocflow/py/model.py
+++ b/deepsocflow/py/model.py
@@ -148,7 +148,7 @@ def export_inference(self, x, hw):
 
                 out_buffer_idx = 1*(not out_buffer_idx) if ib != len(bundles)-1 else -1 # alternate between 0 and 1
 
-                ch.write(f"   {{.n={b.r.XN:<3}, .l={b.r.XL:<3}, .kw={b.r.KW:<3}, .coe={y_coe:<3}, .coe_tl={y_coe_tl:<3}, .r_ll={y_r_ll:<3}, .h={b.r.XH:<3}, .w={b.r.XW:<3}, .ci={b.r.CI:<4}, .co={b.r.CO:<3}, .w_kw2={b.r.XW-b.r.KW//2:<3}, .t={b.r.IT:<3}, .p={b.r.CP:<3}, .cm={b.r.CM:<3}, .cm_p0={b.r.CM_0:<3}, .xp_words={xp_words:<3}, ")
+                ch.write(f"   {{.n={b.r.XN:<3}, .l={b.r.XL:<3}, .kw={b.r.KW:<3}, .coe={y_coe:<3}, .coe_tl={y_coe_tl:<3}, .r_ll={y_r_ll:<3}, .h={b.r.XH:<3}, .w={b.r.XW:<3}, .ci={b.r.CI:<4}, .co={b.r.CO:<3}, .w_kw2={b.r.XW-b.r.KW//2:<3}, .t={b.r.IT:<3}, .p={b.r.CP:<3}, .cm={b.r.CM:<3}, .cm_p0={b.r.CM_0:<3}, .xp_words={xp_words:<4}, ")
                 ch.write(     f".w_bpt={w_bpt:<5}, .w_bpt_p0={w_bpt_p0:<5}, .x_bpt={x_bpt:<5}, .x_bpt_p0={x_bpt_p0:<5}, .o_words={o_words_b:<5}, .o_bytes={o_bytes_b:<5}, ")
                 ch.write(     f".out_buffer_idx={out_buffer_idx:<2}, .add_out_buffer_idx={add_out_buffer_idx:<2}, .add_in_buffer_idx={add_in_buffer_idx:<2}, ")
                 ch.write(     f".is_bias={1*(b.b is not None):<3}, .is_flatten={1*b.flatten:<3}, .is_softmax={1*b.softmax:<3}, ")