Idiv lzc opt #621
Idiv lzc opt #621
Conversation
| ,.num_zero_o(clz_a_result) | ||
| ); | ||
|
|
||
| assign div_shift = (zero_divisor_i) ? width_p-1 : clz_a_result - clz_c_result; |
There was a problem hiding this comment.
Should truncate to num_bits_per_iter granularity
| @@ -22,6 +22,8 @@ module bsg_idiv_iterative_controller #(parameter width_p=32, parameter bits_per_ | |||
| ,input opA_is_neg_i | |||
| ,input opC_is_neg_i | |||
|
|
|||
| ,input [`BSG_SAFE_CLOG2(width_p)-1:0] div_shift | |||
There was a problem hiding this comment.
nit; coding convention is div_shift_i
| @@ -38,6 +40,8 @@ module bsg_idiv_iterative_controller #(parameter width_p=32, parameter bits_per_ | |||
| ,output logic latch_signed_div_o | |||
| ,output logic adder1_cin_o | |||
|
|
|||
| ,output logic [`BSG_WIDTH(width_p):0] shift_val | |||
There was a problem hiding this comment.
nit; coding convention is shift_val_o
| @@ -38,6 +40,8 @@ module bsg_idiv_iterative_controller #(parameter width_p=32, parameter bits_per_ | |||
| ,output logic latch_signed_div_o | |||
| ,output logic adder1_cin_o | |||
|
|
|||
| ,output logic [`BSG_WIDTH(width_p):0] shift_val | |||
There was a problem hiding this comment.
The max value of this seems to be 'BSG_WIDTH(width_p) as it's assigned to width_p or div_shift (which has max value of width_p). I think this port is 1b too wide (or am I misreading?)
| @@ -22,6 +22,8 @@ module bsg_idiv_iterative_controller #(parameter width_p=32, parameter bits_per_ | |||
| ,input opA_is_neg_i | |||
| ,input opC_is_neg_i | |||
|
|
|||
| ,input [`BSG_SAFE_CLOG2(width_p)-1:0] div_shift | |||
|
|
|||
There was a problem hiding this comment.
Add link to Google design doc in header comment
|
Do we have a modified schematic for the changes? I think the support for these changes should be based on an input parameter, so the user can select whether they want to pay the area overhead |
Changes made for 2b case. Cleaning up code now. Also, the testing flow is static. Working on making it easier by making makefile updates. |
This PR contains -
Working code for unsigned division with lzc optimization. Both bits_per_iter=1 and bits_per_iter=2 cases are addressed.
Updated Makefile flow:
New design doc with notes - https://docs.google.com/presentation/d/1jtqCfIULkc9yHW3kGDidr9mmkZZK64JpkwFONPGzuxM/edit#slide=id.g21dc5dfbb2f_0_521