[hardware] Add hierarchical verilation back

The feature was removed because of a custom parameter type being
passed by Ara to the Lane modules. Verilator did not like this,
so we removed hierarchical verilation. However, the github CI
cannot compile the 16-lane design anymore without running out of
memory.

hardware/Makefile

@@ -190,6 +190,7 @@ $(veril_library)/V$(veril_top): $(config_file) Makefile ../Bender.yml $(shell fi
   -GNrLanes=$(nr_lanes)                                                         \
   -GVLEN=$(vlen)                                                                \
   -O3                                                                           \
+  --hierarchical \
   -Wno-fatal                                                                    \
   -Wno-PINCONNECTEMPTY                                                          \
   -Wno-BLKANDNBLK                                                               \

hardware/include/ara/ara.svh

@@ -16,6 +16,15 @@
     logic[$bits(struct_t)-1:0]   \
   `endif
 
+  // Structs in ports of hierarchical modules are not supported in Verilator
+  // --> Flatten them for Verilator
+  `define STRUCT_PORT_BITS(bits) \
+  `ifndef VERILATOR              \
+    struct_t                     \
+  `else                          \
+    logic[bits-1:0]              \
+  `endif
+
   // Create a flattened vector of a struct. Make sure the first dimension is
   // the dimension into the vector of struct types and not the struct itself.
   `define STRUCT_VECT(struct_t, dim)  \

hardware/include/ara/ara_typedef.svh

@@ -0,0 +1,86 @@
+// Copyright 2024 ETH Zurich and University of Bologna.
+// Solderpad Hardware License, Version 0.51, see LICENSE for details.
+// SPDX-License-Identifier: SHL-0.51
+//
+// Author: Matteo Perotti <[email protected]>
+// Description:
+// Hierarchical modules cannot have parametrized data types during Verilator's
+// hierarchical verilation.
+
+typedef struct packed {
+  vid_t id; // ID of the vector instruction
+
+  ara_op_e op; // Operation
+
+  // Mask vector register operand
+  logic vm;
+  rvv_pkg::vew_e eew_vmask;
+
+  vfu_e vfu; // VFU responsible for handling this instruction
+
+  // Rescale vl taking into account the new and old EEW
+  logic scale_vl;
+
+  // The lane that provides the first element of the computation
+  logic [$clog2(MaxNrLanes)-1:0] start_lane;
+  // The lane that provides the last element of the computation
+  logic [$clog2(MaxNrLanes)-1:0] end_lane;
+
+  // 1st vector register operand
+  logic [4:0] vs1;
+  logic use_vs1;
+  opqueue_conversion_e conversion_vs1;
+  rvv_pkg::vew_e eew_vs1;
+  rvv_pkg::vew_e old_eew_vs1;
+
+  // 2nd vector register operand
+  logic [4:0] vs2;
+  logic use_vs2;
+  opqueue_conversion_e conversion_vs2;
+  rvv_pkg::vew_e eew_vs2;
+
+  // Use vd as an operand as well (e.g., vmacc)
+  logic use_vd_op;
+  rvv_pkg::vew_e eew_vd_op;
+
+  // Scalar operand
+  elen_t scalar_op;
+  logic use_scalar_op;
+
+  // If asserted: vs2 is kept in MulFPU opqueue C, and vd_op in MulFPU A
+  logic swap_vs2_vd_op;
+
+  // 2nd scalar operand: stride for constant-strided vector load/stores
+  elen_t stride;
+  logic is_stride_np2;
+
+  // Destination vector register
+  logic [4:0] vd;
+  logic use_vd;
+
+  // Effective length multiplier
+  rvv_pkg::vlmul_e emul;
+
+  // Rounding-Mode for FP operations
+  fpnew_pkg::roundmode_e fp_rm;
+  // Widen FP immediate (re-encoding)
+  logic wide_fp_imm;
+  // Resizing of FP conversions
+  resize_e cvt_resize;
+
+  // Vector machine metadata
+  vlen_t vl;
+  vlen_t vstart;
+  rvv_pkg::vtype_t vtype;
+
+  // Hazards
+  logic [NrVInsn-1:0] hazard_vs1;
+  logic [NrVInsn-1:0] hazard_vs2;
+  logic [NrVInsn-1:0] hazard_vm;
+  logic [NrVInsn-1:0] hazard_vd;
+} pe_req_t;
+
+typedef struct packed {
+  // Each set bit indicates that the corresponding vector loop has finished execution
+  logic [NrVInsn-1:0] vinsn_done;
+} pe_resp_t;

hardware/src/ara.sv

@@ -48,6 +48,7 @@ module ara import ara_pkg::*; #(
     input  axi_resp_t         axi_resp_i
   );
 
+  `include "ara/ara_typedef.svh"
   import cf_math_pkg::idx_width;
 
   ///////////////////
@@ -147,78 +148,6 @@ module ara import ara_pkg::*; #(
     vlen_t error_vl;
   } ara_resp_t;
 
-  typedef struct packed {
-    vid_t id; // ID of the vector instruction
-
-    ara_op_e op; // Operation
-
-    // Mask vector register operand
-    logic vm;
-    rvv_pkg::vew_e eew_vmask;
-
-    vfu_e vfu; // VFU responsible for handling this instruction
-
-    // Rescale vl taking into account the new and old EEW
-    logic scale_vl;
-
-    // 1st vector register operand
-    logic [4:0] vs1;
-    logic use_vs1;
-    opqueue_conversion_e conversion_vs1;
-    rvv_pkg::vew_e eew_vs1;
-
-    // 2nd vector register operand
-    logic [4:0] vs2;
-    logic use_vs2;
-    opqueue_conversion_e conversion_vs2;
-    rvv_pkg::vew_e eew_vs2;
-
-    // Use vd as an operand as well (e.g., vmacc)
-    logic use_vd_op;
-    rvv_pkg::vew_e eew_vd_op;
-
-    // Scalar operand
-    elen_t scalar_op;
-    logic use_scalar_op;
-
-    // If asserted: vs2 is kept in MulFPU opqueue C, and vd_op in MulFPU A
-    logic swap_vs2_vd_op;
-
-    // 2nd scalar operand: stride for constant-strided vector load/stores
-    elen_t stride;
-    logic is_stride_np2;
-
-    // Destination vector register
-    logic [4:0] vd;
-    logic use_vd;
-
-    // Effective length multiplier
-    rvv_pkg::vlmul_e emul;
-
-    // Rounding-Mode for FP operations
-    fpnew_pkg::roundmode_e fp_rm;
-    // Widen FP immediate (re-encoding)
-    logic wide_fp_imm;
-    // Resizing of FP conversions
-    resize_e cvt_resize;
-
-    // Vector machine metadata
-    vlen_t vl;
-    vlen_t vstart;
-    rvv_pkg::vtype_t vtype;
-
-    // Hazards
-    logic [NrVInsn-1:0] hazard_vs1;
-    logic [NrVInsn-1:0] hazard_vs2;
-    logic [NrVInsn-1:0] hazard_vm;
-    logic [NrVInsn-1:0] hazard_vd;
-  } pe_req_t;
-
-  typedef struct packed {
-    // Each set bit indicates that the corresponding vector loop has finished execution
-    logic [NrVInsn-1:0] vinsn_done;
-  } pe_resp_t;
-
   //////////////////
   //  Dispatcher  //
   //////////////////
@@ -400,8 +329,8 @@ module ara import ara_pkg::*; #(
       .FPUSupport           (FPUSupport           ),
       .FPExtSupport         (FPExtSupport         ),
       .FixPtSupport         (FixPtSupport         ),
-      .pe_req_t             (pe_req_t             ),
-      .pe_resp_t            (pe_resp_t            )
+      .pe_req_t_bits        ($bits(pe_req_t)      ),
+      .pe_resp_t_bits       ($bits(pe_resp_t)     )
     ) i_lane (
       .clk_i                           (clk_i                               ),
       .rst_ni                          (rst_ni                              ),

hardware/src/lane/lane.sv

@@ -18,8 +18,9 @@ module lane import ara_pkg::*; import rvv_pkg::*; #(
     parameter  fpext_support_e        FPExtSupport          = FPExtSupportEnable,
     // Support for fixed-point data types
     parameter  fixpt_support_e        FixPtSupport          = FixedPointEnable,
-    parameter  type                   pe_req_t              = logic,
-    parameter  type                   pe_resp_t             = logic,
+    // To please Verilator
+    parameter  int           unsigned pe_req_t_bits         = 0,
+    parameter  int           unsigned pe_resp_t_bits        = 0,
     // Dependant parameters. DO NOT CHANGE!
     // VRF Parameters
     localparam int           unsigned VLENB           = VLEN / 8,
@@ -48,11 +49,11 @@ module lane import ara_pkg::*; import rvv_pkg::*; #(
     output logic     [4:0]                                 fflags_ex_o,
     output logic                                           fflags_ex_valid_o,
     // Interface with the sequencer
-    input  `STRUCT_PORT(pe_req_t)                          pe_req_i,
+    input  `STRUCT_PORT_BITS(pe_req_t_bits)                pe_req_i,
     input  logic                                           pe_req_valid_i,
     input  logic     [NrVInsn-1:0]                         pe_vinsn_running_i,
     output logic                                           pe_req_ready_o,
-    output `STRUCT_PORT(pe_resp_t)                         pe_resp_o,
+    output `STRUCT_PORT_BITS(pe_resp_t_bits)               pe_resp_o,
     output logic                                           alu_vinsn_done_o,
     output logic                                           mfpu_vinsn_done_o,
     input  logic                [NrVInsn-1:0][NrVInsn-1:0] global_hazard_table_i,
@@ -101,6 +102,9 @@ module lane import ara_pkg::*; import rvv_pkg::*; #(
     output logic                                           mask_ready_o
   );
 
+  `include "common_cells/registers.svh"
+  `include "ara/ara_typedef.svh"
+
   ///////////////////
   //  Definitions  //
   ///////////////////