wu-arch/vortex
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

RTL code refactoring

Browse Source
This commit is contained in:
Blaise Tine
2020-04-19 09:24:04 -04:00
parent 3139d37610
commit e9dfa828fe
33 changed files with 998 additions and 1022 deletions
Download Patch File Download Diff File Expand all files Collapse all files

64
hw/rtl/VX_gpgpu_inst.v
View File

@@ -2,57 +2,57 @@
module VX_gpgpu_inst (
// Input
VX_gpu_inst_req_if vx_gpu_inst_req,
VX_gpu_inst_req_if gpu_inst_req_if,
// Output
VX_warp_ctl_if vx_warp_ctl
VX_warp_ctl_if warp_ctl_if
);
wire[`NUM_THREADS-1:0] curr_valids = vx_gpu_inst_req.valid;
wire is_split = (vx_gpu_inst_req.is_split);
wire[`NUM_THREADS-1:0] curr_valids = gpu_inst_req_if.valid;
wire is_split = (gpu_inst_req_if.is_split);
wire[`NUM_THREADS-1:0] tmc_new_mask;
wire all_threads = `NUM_THREADS < vx_gpu_inst_req.a_reg_data[0];
wire all_threads = `NUM_THREADS < gpu_inst_req_if.a_reg_data[0];
genvar curr_t;
generate
for (curr_t = 0; curr_t < `NUM_THREADS; curr_t=curr_t+1) begin : tmc_new_mask_init
assign tmc_new_mask[curr_t] = all_threads ? 1 : curr_t < vx_gpu_inst_req.a_reg_data[0];
assign tmc_new_mask[curr_t] = all_threads ? 1 : curr_t < gpu_inst_req_if.a_reg_data[0];
end
endgenerate
wire valid_inst = (|curr_valids);
assign vx_warp_ctl.warp_num = vx_gpu_inst_req.warp_num;
assign vx_warp_ctl.change_mask = (vx_gpu_inst_req.is_tmc) && valid_inst;
assign vx_warp_ctl.thread_mask = vx_gpu_inst_req.is_tmc ? tmc_new_mask : 0;
assign warp_ctl_if.warp_num = gpu_inst_req_if.warp_num;
assign warp_ctl_if.change_mask = (gpu_inst_req_if.is_tmc) && valid_inst;
assign warp_ctl_if.thread_mask = gpu_inst_req_if.is_tmc ? tmc_new_mask : 0;
// assign vx_warp_ctl.ebreak = (vx_gpu_inst_req.a_reg_data[0] == 0) && valid_inst;
assign vx_warp_ctl.ebreak = vx_warp_ctl.change_mask && (vx_warp_ctl.thread_mask == 0);
// assign warp_ctl_if.ebreak = (gpu_inst_req_if.a_reg_data[0] == 0) && valid_inst;
assign warp_ctl_if.ebreak = warp_ctl_if.change_mask && (warp_ctl_if.thread_mask == 0);
wire wspawn = vx_gpu_inst_req.is_wspawn;
wire[31:0] wspawn_pc = vx_gpu_inst_req.rd2;
wire all_active = `NUM_WARPS < vx_gpu_inst_req.a_reg_data[0];
wire wspawn = gpu_inst_req_if.is_wspawn;
wire[31:0] wspawn_pc = gpu_inst_req_if.rd2;
wire all_active = `NUM_WARPS < gpu_inst_req_if.a_reg_data[0];
wire[`NUM_WARPS-1:0] wspawn_new_active;
genvar curr_w;
generate
for (curr_w = 0; curr_w < `NUM_WARPS; curr_w=curr_w+1) begin : wspawn_new_active_init
assign wspawn_new_active[curr_w] = all_active ? 1 : curr_w < vx_gpu_inst_req.a_reg_data[0];
assign wspawn_new_active[curr_w] = all_active ? 1 : curr_w < gpu_inst_req_if.a_reg_data[0];
end
endgenerate
assign vx_warp_ctl.is_barrier = vx_gpu_inst_req.is_barrier && valid_inst;
assign vx_warp_ctl.barrier_id = vx_gpu_inst_req.a_reg_data[0];
assign warp_ctl_if.is_barrier = gpu_inst_req_if.is_barrier && valid_inst;
assign warp_ctl_if.barrier_id = gpu_inst_req_if.a_reg_data[0];
`DEBUG_BEGIN
wire[31:0] num_warps_m1 = vx_gpu_inst_req.rd2 - 1;
wire[31:0] num_warps_m1 = gpu_inst_req_if.rd2 - 1;
`DEBUG_END
assign vx_warp_ctl.num_warps = num_warps_m1[$clog2(`NUM_WARPS):0];
assign warp_ctl_if.num_warps = num_warps_m1[$clog2(`NUM_WARPS):0];
assign vx_warp_ctl.wspawn = wspawn;
assign vx_warp_ctl.wspawn_pc = wspawn_pc;
assign vx_warp_ctl.wspawn_new_active = wspawn_new_active;
assign warp_ctl_if.wspawn = wspawn;
assign warp_ctl_if.wspawn_pc = wspawn_pc;
assign warp_ctl_if.wspawn_new_active = wspawn_new_active;
wire[`NUM_THREADS-1:0] split_new_use_mask;
wire[`NUM_THREADS-1:0] split_new_later_mask;
@@ -61,7 +61,7 @@ module VX_gpgpu_inst (
genvar curr_s_t;
generate
for (curr_s_t = 0; curr_s_t < `NUM_THREADS; curr_s_t=curr_s_t+1) begin : masks_init
wire curr_bool = (vx_gpu_inst_req.a_reg_data[curr_s_t] == 32'b1);
wire curr_bool = (gpu_inst_req_if.a_reg_data[curr_s_t] == 32'b1);
assign split_new_use_mask[curr_s_t] = curr_valids[curr_s_t] & (curr_bool);
assign split_new_later_mask[curr_s_t] = curr_valids[curr_s_t] & (!curr_bool);
@@ -79,15 +79,15 @@ module VX_gpgpu_inst (
// wire[`NW_BITS-1:0] num_valids = $countones(curr_valids);
assign vx_warp_ctl.is_split = is_split && (num_valids > 1);
assign vx_warp_ctl.dont_split = vx_warp_ctl.is_split && ((split_new_use_mask == 0) || (split_new_use_mask == {`NUM_THREADS{1'b1}}));
assign vx_warp_ctl.split_new_mask = split_new_use_mask;
assign vx_warp_ctl.split_later_mask = split_new_later_mask;
assign vx_warp_ctl.split_save_pc = vx_gpu_inst_req.pc_next;
assign vx_warp_ctl.split_warp_num = vx_gpu_inst_req.warp_num;
assign warp_ctl_if.is_split = is_split && (num_valids > 1);
assign warp_ctl_if.dont_split = warp_ctl_if.is_split && ((split_new_use_mask == 0) || (split_new_use_mask == {`NUM_THREADS{1'b1}}));
assign warp_ctl_if.split_new_mask = split_new_use_mask;
assign warp_ctl_if.split_later_mask = split_new_later_mask;
assign warp_ctl_if.split_save_pc = gpu_inst_req_if.pc_next;
assign warp_ctl_if.split_warp_num = gpu_inst_req_if.warp_num;
// vx_gpu_inst_req.is_wspawn
// vx_gpu_inst_req.is_split
// vx_gpu_inst_req.is_barrier
// gpu_inst_req_if.is_wspawn
// gpu_inst_req_if.is_split
// gpu_inst_req_if.is_barrier
endmodule