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

reset network refactoring

Browse Source
This commit is contained in:
Blaise Tine
2021-07-15 11:34:55 -07:00
parent 22cf698e69
commit 7d01be367c
11 changed files with 68 additions and 137 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
hw/rtl/VX_cluster.v
View File

@@ -44,24 +44,17 @@ module VX_cluster #(
wire [`NUM_CORES-1:0] per_core_busy; wire [`NUM_CORES-1:0] per_core_busy;
wire [`NUM_CORES-1:0] core_reset;
VX_reset_relay #(
.DEPTH (`NUM_CORES > 1),
.NUM_NODES (`NUM_CORES)
) reset_relay (
.clk (clk),
.reset (reset),
.reset_o (core_reset)
);
for (genvar i = 0; i < `NUM_CORES; i++) begin for (genvar i = 0; i < `NUM_CORES; i++) begin
`RESET_RELAY (core_reset);
VX_core #( VX_core #(
.CORE_ID(i + (CLUSTER_ID * `NUM_CORES)) .CORE_ID(i + (CLUSTER_ID * `NUM_CORES))
) core ( ) core (
`SCOPE_BIND_VX_cluster_core(i) `SCOPE_BIND_VX_cluster_core(i)
.clk (clk), .clk (clk),
.reset (core_reset[i]), .reset (core_reset),
.mem_req_valid (per_core_mem_req_valid[i]), .mem_req_valid (per_core_mem_req_valid[i]),
.mem_req_rw (per_core_mem_req_rw [i]), .mem_req_rw (per_core_mem_req_rw [i]),
@@ -87,6 +80,8 @@ module VX_cluster #(
VX_perf_cache_if perf_l2cache_if(); VX_perf_cache_if perf_l2cache_if();
`endif `endif
`RESET_RELAY (l2_reset);
VX_cache #( VX_cache #(
.CACHE_ID (`L2CACHE_ID), .CACHE_ID (`L2CACHE_ID),
.CACHE_SIZE (`L2CACHE_SIZE), .CACHE_SIZE (`L2CACHE_SIZE),
@@ -107,7 +102,7 @@ module VX_cluster #(
`SCOPE_BIND_VX_cluster_l2cache `SCOPE_BIND_VX_cluster_l2cache
.clk (clk), .clk (clk),
.reset (reset), .reset (l2_reset),
`ifdef PERF_ENABLE `ifdef PERF_ENABLE
.perf_cache_if (perf_l2cache_if), .perf_cache_if (perf_l2cache_if),

15
hw/rtl/VX_execute.v
View File

@@ -43,14 +43,10 @@ module VX_execute #(
wire[`NUM_WARPS-1:0] csr_pending; wire[`NUM_WARPS-1:0] csr_pending;
wire[`NUM_WARPS-1:0] fpu_pending; wire[`NUM_WARPS-1:0] fpu_pending;
wire alu_reset, lsu_reset, csr_reset, fpu_reset, gpu_reset; `RESET_RELAY (alu_reset);
VX_reset_relay #( `RESET_RELAY (lsu_reset);
.NUM_NODES (5) `RESET_RELAY (csr_reset);
) reset_relay ( `RESET_RELAY (gpu_reset);
.clk (clk),
.reset (reset),
.reset_o ({alu_reset, lsu_reset, csr_reset, fpu_reset, gpu_reset})
);
VX_alu_unit #( VX_alu_unit #(
.CORE_ID (CORE_ID) .CORE_ID (CORE_ID)
@@ -94,6 +90,8 @@ module VX_execute #(
); );
`ifdef EXT_F_ENABLE `ifdef EXT_F_ENABLE
`RESET_RELAY (fpu_reset);
VX_fpu_unit #( VX_fpu_unit #(
.CORE_ID (CORE_ID) .CORE_ID (CORE_ID)
) fpu_unit ( ) fpu_unit (
@@ -106,7 +104,6 @@ module VX_execute #(
.pending (fpu_pending) .pending (fpu_pending)
); );
`else `else
`UNUSED_VAR (fpu_reset)
`UNUSED_VAR (csr_pending) `UNUSED_VAR (csr_pending)
`UNUSED_VAR (fpu_to_csr_if.read_frm) `UNUSED_VAR (fpu_to_csr_if.read_frm)
assign fpu_req_if.ready = 0; assign fpu_req_if.ready = 0;

14
hw/rtl/VX_mem_unit.v
View File

@@ -63,14 +63,8 @@ module VX_mem_unit # (
.CORE_TAG_WIDTH (`DCORE_TAG_WIDTH-`SM_ENABLE) .CORE_TAG_WIDTH (`DCORE_TAG_WIDTH-`SM_ENABLE)
) dcache_rsp_if(); ) dcache_rsp_if();
wire icache_reset, dcache_reset, smem_reset; `RESET_RELAY (icache_reset);
VX_reset_relay #( `RESET_RELAY (dcache_reset);
.NUM_NODES (3)
) reset_relay (
.clk (clk),
.reset (reset),
.reset_o ({icache_reset, dcache_reset, smem_reset})
);
VX_cache #( VX_cache #(
.CACHE_ID (`ICACHE_ID), .CACHE_ID (`ICACHE_ID),
@@ -214,6 +208,8 @@ module VX_mem_unit # (
.core_rsp_if (dcache_core_rsp_if) .core_rsp_if (dcache_core_rsp_if)
); );
`RESET_RELAY (smem_reset);
VX_shared_mem #( VX_shared_mem #(
.CACHE_ID (`SCACHE_ID), .CACHE_ID (`SCACHE_ID),
.CACHE_SIZE (`SMEM_SIZE), .CACHE_SIZE (`SMEM_SIZE),
@@ -249,8 +245,6 @@ module VX_mem_unit # (
.core_rsp_ready (smem_rsp_if.ready) .core_rsp_ready (smem_rsp_if.ready)
); );
end else begin end else begin
`UNUSED_VAR (smem_reset)
// core to D-cache request // core to D-cache request
assign dcache_req_if.valid = dcache_core_req_if.valid; assign dcache_req_if.valid = dcache_core_req_if.valid;
assign dcache_req_if.addr = dcache_core_req_if.addr; assign dcache_req_if.addr = dcache_core_req_if.addr;

13
hw/rtl/VX_pipeline.v
View File

@@ -132,14 +132,11 @@ module VX_pipeline #(
VX_perf_pipeline_if perf_pipeline_if(); VX_perf_pipeline_if perf_pipeline_if();
`endif `endif
wire fetch_reset, decode_reset, issue_reset, execute_reset, commit_reset; `RESET_RELAY (fetch_reset);
VX_reset_relay #( `RESET_RELAY (decode_reset);
.NUM_NODES (5) `RESET_RELAY (issue_reset);
) reset_relay ( `RESET_RELAY (execute_reset);
.clk (clk), `RESET_RELAY (commit_reset);
.reset (reset),
.reset_o ({fetch_reset, decode_reset, issue_reset, execute_reset, commit_reset})
);
VX_fetch #( VX_fetch #(
.CORE_ID(CORE_ID) .CORE_ID(CORE_ID)

8
hw/rtl/VX_platform.vh
View File

@@ -57,6 +57,14 @@
`define TRACING_ON /* verilator tracing_on */ `define TRACING_ON /* verilator tracing_on */
`define TRACING_OFF /* verilator tracing_off */ `define TRACING_OFF /* verilator tracing_off */
`define RESET_RELAY(signal) \
wire signal; \
VX_reset_relay __``signal ( \
.clk (clk), \
.reset (reset), \
.reset_o (signal) \
)
/////////////////////////////////////////////////////////////////////////////// ///////////////////////////////////////////////////////////////////////////////
`define USE_FAST_BRAM (* ramstyle = "MLAB, no_rw_check" *) `define USE_FAST_BRAM (* ramstyle = "MLAB, no_rw_check" *)

19
hw/rtl/Vortex.v
View File

@@ -42,24 +42,17 @@ module Vortex (
wire [`NUM_CLUSTERS-1:0] per_cluster_busy; wire [`NUM_CLUSTERS-1:0] per_cluster_busy;
wire [`NUM_CLUSTERS-1:0] cluster_reset;
VX_reset_relay #(
.DEPTH (`NUM_CLUSTERS > 1),
.NUM_NODES (`NUM_CLUSTERS)
) reset_relay (
.clk (clk),
.reset (reset),
.reset_o (cluster_reset)
);
for (genvar i = 0; i < `NUM_CLUSTERS; i++) begin for (genvar i = 0; i < `NUM_CLUSTERS; i++) begin
`RESET_RELAY (cluster_reset);
VX_cluster #( VX_cluster #(
.CLUSTER_ID(i) .CLUSTER_ID(i)
) cluster ( ) cluster (
`SCOPE_BIND_Vortex_cluster(i) `SCOPE_BIND_Vortex_cluster(i)
.clk (clk), .clk (clk),
.reset (cluster_reset[i]), .reset (cluster_reset),
.mem_req_valid (per_cluster_mem_req_valid [i]), .mem_req_valid (per_cluster_mem_req_valid [i]),
.mem_req_rw (per_cluster_mem_req_rw [i]), .mem_req_rw (per_cluster_mem_req_rw [i]),
@@ -85,6 +78,8 @@ module Vortex (
VX_perf_cache_if perf_l3cache_if(); VX_perf_cache_if perf_l3cache_if();
`endif `endif
`RESET_RELAY (l3_reset);
VX_cache #( VX_cache #(
.CACHE_ID (`L3CACHE_ID), .CACHE_ID (`L3CACHE_ID),
.CACHE_SIZE (`L3CACHE_SIZE), .CACHE_SIZE (`L3CACHE_SIZE),
@@ -105,7 +100,7 @@ module Vortex (
`SCOPE_BIND_Vortex_l3cache `SCOPE_BIND_Vortex_l3cache
.clk (clk), .clk (clk),
.reset (reset), .reset (l3_reset),
`ifdef PERF_ENABLE `ifdef PERF_ENABLE
.perf_cache_if (perf_l3cache_if), .perf_cache_if (perf_l3cache_if),

16
hw/rtl/fp_cores/VX_fp_div.v
View File

@@ -33,18 +33,6 @@ module VX_fp_div #(
wire stall = ~ready_out && valid_out; wire stall = ~ready_out && valid_out;
wire enable = ~stall; wire enable = ~stall;
`ifndef VERILATOR
wire [LANES-1:0] fdiv_reset;
VX_reset_relay #(
.DEPTH (LANES > 1),
.NUM_NODES (LANES)
) reset_relay (
.clk (clk),
.reset (reset),
.reset_o (fdiv_reset)
);
`endif
for (genvar i = 0; i < LANES; i++) begin for (genvar i = 0; i < LANES; i++) begin
`ifdef VERILATOR `ifdef VERILATOR
reg [31:0] r; reg [31:0] r;
@@ -67,9 +55,11 @@ module VX_fp_div #(
.data_out (result[i]) .data_out (result[i])
); );
`else `else
`RESET_RELAY (fdiv_reset);
acl_fdiv fdiv ( acl_fdiv fdiv (
.clk (clk), .clk (clk),
.areset (fdiv_reset[i]), .areset (fdiv_reset),
.en (enable), .en (enable),
.a (dataa[i]), .a (dataa[i]),
.b (datab[i]), .b (datab[i]),

16
hw/rtl/fp_cores/VX_fp_fma.v
View File

@@ -39,18 +39,6 @@ module VX_fp_fma #(
wire stall = ~ready_out && valid_out; wire stall = ~ready_out && valid_out;
wire enable = ~stall; wire enable = ~stall;
`ifndef VERILATOR
wire [LANES-1:0] fma_reset;
VX_reset_relay #(
.DEPTH (LANES > 1),
.NUM_NODES (LANES)
) reset_relay (
.clk (clk),
.reset (reset),
.reset_o (fma_reset)
);
`endif
for (genvar i = 0; i < LANES; i++) begin for (genvar i = 0; i < LANES; i++) begin
reg [31:0] a, b, c; reg [31:0] a, b, c;
@@ -96,9 +84,11 @@ module VX_fp_fma #(
.data_out (result[i]) .data_out (result[i])
); );
`else `else
`RESET_RELAY (fma_reset);
acl_fmadd fmadd ( acl_fmadd fmadd (
.clk (clk), .clk (clk),
.areset (fma_reset[i]), .areset (fma_reset),
.en (enable), .en (enable),
.a (a), .a (a),
.b (b), .b (b),

16
hw/rtl/fp_cores/VX_fp_sqrt.v
View File

@@ -32,18 +32,6 @@ module VX_fp_sqrt #(
wire stall = ~ready_out && valid_out; wire stall = ~ready_out && valid_out;
wire enable = ~stall; wire enable = ~stall;
`ifndef VERILATOR
wire [LANES-1:0] fsqrt_reset;
VX_reset_relay #(
.DEPTH (LANES > 1),
.NUM_NODES (LANES)
) reset_relay (
.clk (clk),
.reset (reset),
.reset_o (fsqrt_reset)
);
`endif
for (genvar i = 0; i < LANES; i++) begin for (genvar i = 0; i < LANES; i++) begin
`ifdef VERILATOR `ifdef VERILATOR
reg [31:0] r; reg [31:0] r;
@@ -66,9 +54,11 @@ module VX_fp_sqrt #(
.data_out (result[i]) .data_out (result[i])
); );
`else `else
`RESET_RELAY (fsqrt_reset);
acl_fsqrt fsqrt ( acl_fsqrt fsqrt (
.clk (clk), .clk (clk),
.areset (fsqrt_reset[i]), .areset (fsqrt_reset),
.en (enable), .en (enable),
.a (dataa[i]), .a (dataa[i]),
.q (result[i]) .q (result[i])

23
hw/rtl/fp_cores/VX_fpu_fpga.v
View File

@@ -71,21 +71,18 @@ module VX_fpu_fpga #(
endcase endcase
end end
wire [NUM_FPC-1:0] fpu_reset; `RESET_RELAY (fma_reset);
VX_reset_relay #( `RESET_RELAY (div_reset);
.NUM_NODES(NUM_FPC) `RESET_RELAY (sqrt_reset);
) reset_relay ( `RESET_RELAY (cvt_reset);
.clk (clk), `RESET_RELAY (ncp_reset);
.reset (reset),
.reset_o (fpu_reset)
);
VX_fp_fma #( VX_fp_fma #(
.TAGW (TAGW), .TAGW (TAGW),
.LANES(`NUM_THREADS) .LANES(`NUM_THREADS)
) fp_fma ( ) fp_fma (
.clk (clk), .clk (clk),
.reset (fpu_reset[FPU_FMA]), .reset (fma_reset),
.valid_in (valid_in && (core_select == FPU_FMA)), .valid_in (valid_in && (core_select == FPU_FMA)),
.ready_in (per_core_ready_in[FPU_FMA]), .ready_in (per_core_ready_in[FPU_FMA]),
.tag_in (tag_in), .tag_in (tag_in),
@@ -109,7 +106,7 @@ module VX_fpu_fpga #(
.LANES(`NUM_THREADS) .LANES(`NUM_THREADS)
) fp_div ( ) fp_div (
.clk (clk), .clk (clk),
.reset (fpu_reset[FPU_DIV]), .reset (div_reset),
.valid_in (valid_in && (core_select == FPU_DIV)), .valid_in (valid_in && (core_select == FPU_DIV)),
.ready_in (per_core_ready_in[FPU_DIV]), .ready_in (per_core_ready_in[FPU_DIV]),
.tag_in (tag_in), .tag_in (tag_in),
@@ -129,7 +126,7 @@ module VX_fpu_fpga #(
.LANES(`NUM_THREADS) .LANES(`NUM_THREADS)
) fp_sqrt ( ) fp_sqrt (
.clk (clk), .clk (clk),
.reset (fpu_reset[FPU_SQRT]), .reset (sqrt_reset),
.valid_in (valid_in && (core_select == FPU_SQRT)), .valid_in (valid_in && (core_select == FPU_SQRT)),
.ready_in (per_core_ready_in[FPU_SQRT]), .ready_in (per_core_ready_in[FPU_SQRT]),
.tag_in (tag_in), .tag_in (tag_in),
@@ -148,7 +145,7 @@ module VX_fpu_fpga #(
.LANES(`NUM_THREADS) .LANES(`NUM_THREADS)
) fp_cvt ( ) fp_cvt (
.clk (clk), .clk (clk),
.reset (fpu_reset[FPU_CVT]), .reset (cvt_reset),
.valid_in (valid_in && (core_select == FPU_CVT)), .valid_in (valid_in && (core_select == FPU_CVT)),
.ready_in (per_core_ready_in[FPU_CVT]), .ready_in (per_core_ready_in[FPU_CVT]),
.tag_in (tag_in), .tag_in (tag_in),
@@ -169,7 +166,7 @@ module VX_fpu_fpga #(
.LANES(`NUM_THREADS) .LANES(`NUM_THREADS)
) fp_ncomp ( ) fp_ncomp (
.clk (clk), .clk (clk),
.reset (fpu_reset[FPU_NCP]), .reset (ncp_reset),
.valid_in (valid_in && (core_select == FPU_NCP)), .valid_in (valid_in && (core_select == FPU_NCP)),
.ready_in (per_core_ready_in[FPU_NCP]), .ready_in (per_core_ready_in[FPU_NCP]),
.tag_in (tag_in), .tag_in (tag_in),

32
hw/rtl/libs/VX_reset_relay.v
View File

@@ -1,46 +1,24 @@
`include "VX_platform.vh" `include "VX_platform.vh"
module VX_reset_relay #( module VX_reset_relay #(
parameter NUM_NODES = 1,
parameter DEPTH = 1,
parameter ASYNC = 0 parameter ASYNC = 0
) ( ) (
input wire clk, input wire clk,
input wire reset, input wire reset,
output wire [NUM_NODES-1:0] reset_o output wire reset_o
); );
(* preserve *) reg reset_r;
if (DEPTH > 1) begin
`DISABLE_BRAM reg [NUM_NODES-1:0] reset_r [DEPTH-1:0];
if (ASYNC) begin if (ASYNC) begin
always @(posedge clk or posedge reset) begin always @(posedge clk or posedge reset) begin
for (integer i = DEPTH-1; i > 0; --i) reset_r <= reset;
reset_r[i] <= reset_r[i-1];
reset_r[0] <= {NUM_NODES{reset}};
end end
end else begin end else begin
always @(posedge clk) begin always @(posedge clk) begin
for (integer i = DEPTH-1; i > 0; --i) reset_r <= reset;
reset_r[i] <= reset_r[i-1];
reset_r[0] <= {NUM_NODES{reset}};
end
end
assign reset_o = reset_r[DEPTH-1];
end else if (DEPTH == 1) begin
reg [NUM_NODES-1:0] reset_r;
if (ASYNC) begin
always @(posedge clk or posedge reset) begin
reset_r <= {NUM_NODES{reset}};
end
end else begin
always @(posedge clk) begin
reset_r <= {NUM_NODES{reset}};
end end
end end
assign reset_o = reset_r; assign reset_o = reset_r;
end else begin
`UNUSED_VAR (clk)
assign reset_o = {NUM_NODES{reset}};
end
endmodule endmodule