Updated PR with changes for homogeneous ADC precision, constant file

how_to_run.md

@@ -99,6 +99,11 @@ num_tile_compute = 23 # number of tiles mapped by dnn (leaving input and output
 # Do not change this - total number of tiles
 num_tile = num_node * num_tile_compute + 2 # +1 for first tile (I/O tile) - dummy, others - compute -- (Line 95)
 ```
+#### 6.2 - Setup constants file:
+
+Use the appropriate constants file from ```puma-simulator/include/example-constants/(constant file name)```
+For example: for 128x128 crossbar use ```constants-128.py```
+Copy the file to ```puma-simulator/include/``` and rename it to ```constants.py```
 
 ### 7. Run your model, in this example, the ```lstm-layer.cpp```:
 
@@ -185,3 +190,12 @@ number of tiles mapped: 23
 ### 10. To run Regression tests after running with weights for inference, go to simulator/test/val.
 
 ```python reg_test_1.py -n mlp```
+
+### 11. Quantization:
+Change the ```input_prec``` and ```weight_width``` parameters in config file to see the effects of quantization.
+
+```
+# Input and Weight parameters
+input_prec = 16
+weight_width = 16
+```

include/config.py

@@ -7,7 +7,7 @@
 xbar_record = 1
 inference = 1
 training = not(inference)
-sparse_opt = 1 # Flag for Sparsity optimisaton (Make it 0 for only dense computations)
+sparse_opt = 0 # Flag for Sparsity optimisaton (Make it 0 for only dense computations)
 
 ## Variable to define the type of MVMU
 # One of "Analog", "Digital_V1" or "Digital_V2" 
@@ -35,7 +35,7 @@
 # Fixed parameters
 addr_width = 22 # Added to address larger address space for conv layers (#TODO: Compiler needs to fix shared memory reuse)
 data_width = num_bits # (in bits)
-xbdata_width = data_width # (in bits)
+xbdata_width = data_width # (in bits), equivalent to input_prec
 instrn_width = 48 # (in bits)
 # Input and Weight parameters
 input_prec = 16
@@ -50,15 +50,19 @@
 num_adc_per_matrix = 2
 num_adc = num_adc_per_matrix * num_matrix
 
+#uncomment this line for homogeneous ADC precision
+adc_res_new ={}
+
+#uncomment adc_res_new for heterogenous adcs
 # The idea is to have different ADC resolution value for each ADC.
 # The number of ADC if defined by num_adc property. Currently it is 2 * num_matrix(2) = 4
 # NOTE: Only taking in account indexes 0 and 2, 1 and 3 are ignored, because ADCs 1 and 3 are assumed t be equal to 0 and 2. 
-adc_res_new = {
-                'matrix_adc_0' : 8,
-                'matrix_adc_1' : 4,
-                'matrix_adc_2' : 8,
-                'matrix_adc_3' : 4
-              }
+#adc_res_new = {
+#                'matrix_adc_0' : 8,
+#                'matrix_adc_1' : 4,
+#                'matrix_adc_2' : 8,
+#                'matrix_adc_3' : 4
+#              }
 
 num_ALU = num_matrix*2
 #dataMem_size = num_matrix*(6*xbar_size) # 4 for 4 input spaces within matrix (1 for f/b each, 2 for d)